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.pgen.1007561 | Nicotinamide-N-methyltransferase controls behavior, neurodegeneration and lifespan by regulating neuronal autophagy | Nicotinamide N-methyl-transferase (NNMT) is an essential contributor to various metabolic and epigenetic processes, including the regulating of aging, cellular stress response, and body weight gain. Epidemiological studies show that NNMT is a risk factor for psychiatric diseases like schizophrenia and neurodegeneration, especially Parkinson’s disease (PD), but its neuronal mechanisms of action remain obscure. Here, we describe the role of neuronal NNMT using C. elegans. We discovered that ANMT-1, the nematode NNMT ortholog, competes with the methyltransferase LCMT-1 for methyl groups from S—adenosyl methionine. Thereby, it regulates the catalytic capacities of LCMT-1, targeting NPRL-2, a regulator of autophagy. Autophagy is a core cellular, catabolic process for degrading cytoplasmic material, but very little is known about the regulation of autophagy during aging. We report an important role for NNMT in regulation of autophagy during aging, where high neuronal ANMT-1 activity induces autophagy via NPRL-2, which maintains neuronal function in old wild type animals and various disease models, also affecting longevity. In younger animals, however, ANMT-1 activity disturbs neuronal homeostasis and dopamine signaling, causing abnormal behavior. In summary, we provide fundamental insights into neuronal NNMT/ANMT-1 as pivotal regulator of behavior, neurodegeneration, and lifespan by controlling neuronal autophagy, potentially influencing PD and schizophrenia risk in humans.
| Neuronal disorders are a threat to human health and understanding the underlying genetic and cellular regulatory networks is a first step towards successful treatment. In this context, it has been suggested in epidemiological studies that the metabolism of nicotinamide, specifically the enzyme nicotinamide N-methyl-transferase (NNMT), could contribute to behavioral and neurodegenerative diseases such as Parkinson’s disease (PD) and schizophrenia. We used the simple nematode worm C. elegans and expressed NNMT in its dopaminergic neurons as this specific neuronal population is primarily affected by both PD and schizophrenia. We found that neuronal NNMT expression in C. elegans influences behavior, neurodegeneration, and life expectancy. NNMT activity controls the concentration of S—adenosyl methionine (SAM) in the neuronal cell. High NNMT activity leads to low SAM levels that the cell interprets as starvation signal, therefore inducing autophagy, a catabolic process important for balancing energy sources and cell homeostasis. Depending on the age of the worms, this can result in disturbed behavioral paradigms in young animals, or preserves neuronal health and decreases neurodegeneration processes in old individuals. Taken together, these identified mechanisms of NNMT action in C. elegans neurons could provide novel insights into the development of neuronal disorders in humans.
| Neurodegenerative disorders are a major health concern in all aging populations. The modifications associated with these diseases are mostly progressive and irreversible, and while several compounds have been developed that relieve symptoms in the short term, no cure has been identified for any of these conditions. Causes leading to neurodegeneration are both diverse and complex, and include various genetic, epigenetic, and environmental factors.
Parkinson’s disease (PD) is among the most prevalent neurodegenerative diseases and numbers increase steadily with age, reaching approximately 2% of all octogenarians affected worldwide. PD is characterized by dopaminergic (DA) cell death in the substantia nigra in the midbrain, leading to a variety of motor and psychiatric symptoms, such as tremor, depression, and dementia. For decades, PD was considered mostly an idiopathic and sporadic disease with only a small genetic component [1, 2]. However, genetic studies in the last decade have been instrumental in identifying the heredity basis of PD. Recent studies suggest that 27–60% of all cases might be caused by genetic factors [3–5].
The gene encoding α-synuclein, SCNA, was the first gene identified causing autosomal recessive inherited PD when mutated, followed by the discovery of PARK2 and PINK1 [6–8]. Interestingly, all of these proteins are involved in macroautophagy (referred to herein as autophagy) processes, leading to the conclusion that autophagy plays an important role in PD. Indeed, studies have shown that the autophagic flux is profoundly disrupted in PD patients, whereas it remains a matter of debate whether this is a cause or result of the disease [9]. A large meta-analysis of genome-wide association studies of PD analyzed the genome of over 13,000 PD patients and compared it to more than 95,000 controls. Many new genes causative for PD were identified from this study, including LRRK2, GBA, TMEM175, GPNMB, MAPT, SCARB2, and SREBF1, each of which is directly or indirectly involved in autophagy [10].
Autophagy is a conserved catabolic cellular process during which macromolecules, organelles, and cytosol fractions are degraded by the lysosomes, which contain acid hydrolases (such as proteases, lipases, or nucleases) that break down the internalized macromolecules. Essential components of these macromolecules can be used for energy yield and the building of new cellular material. Although initially described as a stress-induced mechanism, autophagy exerts basal activity and has a major role in the quality control of proteins to maintain cellular homeostasis. The role of autophagy for neuroprotection and neurodegeneration in general is well established, and its stringent regulation is critical for healthy neuronal homeostasis [11]. For instance, autophagy can ameliorate symptoms of PD by removing aggregated proteins, whereas excessive autophagy in contrast may contribute to DA cell death [12].
Interestingly, impaired autophagy is also linked to schizophrenia [13], a mental illness that typically commences in young adulthood with a lifetime prevalence of about 0.5% [14]. Symptoms, such as abnormal social behavior and the inability to distinguish between reality and the imaginary, are usually attributable to disturbed dopamine signaling. In contrast to PD, where the lack of dopamine is the perpetrator, high dopamine levels and dopamine hyper-responsiveness seem to be responsible for schizophrenic symptoms [15].
A potential role in both PD and schizophrenia is attributed to the enzyme nicotinamide-N-methyltransferase (NNMT), which is expressed in all body tissues including the nervous system and represents a key player in NAD+/NADH metabolism. The central coenzyme for fuel oxidation and interconversion of different classes of metabolites is NAD+, which is typically reduced to NADH during these processes. It can furthermore be used by the sirtuins, which link lysine deacetylation to the turnover of NAD+, and poly ADP-ribose polymerases (PARPs) for DNA repair. One of the products of these reactions is nicotinamide (NAM). NAM is one of three NAD+ precursor vitamins (vitamin B3) and can be salvaged and used in re-synthesis of NAD+, or converted irreversibly by NNMT to N-methylnicotinamide (MNA), using S-adenosyl methionine (SAM) as methyl group donor [16]. NNMT has been shown to play a crucial role in obesity, as limiting re-synthesis of NAD+ decreased fuel oxidation, leading to fat storage in mice [17]. In contrast, NNMT in liver improves lipid parameters via sirtuin 1 stabilization to protect from some effects of high fat diet-induced obesity [18], pointing towards significant tissue-specificity of NNMT. The enzyme has furthermore been shown to play a crucial role in reactive oxygen species signaling and aging [19], and is highly expressed in cancer cells [20, 21], where it influences epigenetic regulation [22]. Several studies have implicated NNMT in PD, schizophrenia, and other neurological disorders such as bipolar disorder and epilepsy [23–28]. Notably, both PD and schizophrenia are associated with methylation disturbances in the cell [29, 30]. Although some in vitro studies have found potential mechanisms that contribute to NNMT action in neurons [31, 32], no mechanism of action has been described in vivo so far.
Here we investigate for the first time the neuronal role of NNMT in the context of neuronal homeostasis, behavior, neurodegeneration, and lifespan in vivo using the model organism Caenorhabditis elegans. This small nematode has provided valuable insights into the cellular mechanisms of neurodegeneration, neurological control of behavior, and aging. The C. elegans nervous system is simple, yet many of its structural features and the associated cellular and biochemical processes are very similar to those of most vertebrate nervous systems. C. elegans is the only organism whose neuronal wiring has been completely documented, showing surprisingly complex neuronal circuits and behavioral plasticity. Additionally, its short lifespan of about 4 weeks allows for the study of living, aging animals, which is an important consideration since age is a major risk factor for neurodegeneration.
anmt-1, the C. elegans ortholog of human nnmt, is naturally not expressed in the nervous system of the worm. We discovered that a mild ectopic neuronal expression of anmt-1 regulates neurotransmitter production and neuronal autophagy via influencing the availability of intracellular methyl groups. Thus, ANMT-1 influences neuronal homeostasis, behavior, degeneration, and organismal health- and lifespan. ANMT-1 competes for methyl groups from SAM with another methyltransferase, LCMT-1, the worm ortholog of human LCMT1 (leucine carboxyl methyltransferase 1), as the methylation to MNA is irreversible, creating methylation drainage in the cell. Consequently, LCMT-1 activity is limited, leading to a switch in a pathway containing LET-92/PP2A (protein phosphatase 2) and NPRL-2/NPRL2 (human NPR2-like, GATOR1 complex subunit), which induces autophagy. We further show that the regulation of autophagy via ANMT-1 is beneficial in neurodegenerative disease models. Notably, in the case of neuronal autophagy dysfunction, high ANMT levels become a trigger for neurodegeneration. In summary, ANMT-1/NNMT is a pivotal element in neuronal cell metabolism that regulates neuronal homeostasis and may contribute to the prevalence of neurological disorders.
NNMT in humans is suspected of being involved in the progression of PD, which is characterized by loss of DA neurons in the substantia nigra in the brain. Also, NNMT may play a role in mental disorders such as schizophrenia and bipolar disorder that both are characterized by disturbances in dopamine levels and signaling. In this context, we investigated the influence of neuronal ANMT-1 in the DA system of C. elegans by expressing anmt-1 using the DA neuronal-specific dat-1 promoter (anmt-1dopa) and the MosSCI cloning system, which results in a mild ectopic expression in the DA neuronal tissue. anmt-1 is, in contrast to nnmt in humans, not expressed in the C. elegans nervous system [19]. To control for specific ANMT-1 activity we expressed a mutated version of anmt-1 (anmt-1dopa-MUT) in the DA neurons, which contains point mutations in 6 conserved SAM binding sites (details can be found in Methods). A mild ectopic expression of anmt-1 in the GABAergic motor neuronal system (anmt-1GABA) was used as control for DA-specific effects. S1A Fig depicts a transgenic animal that expresses GFP under the control of dat-1 (GFPdopa), which was used to visualize the DA nervous system.
As PD in humans is an age-dependent disease, we analyzed worms at day 15 after L4, which represents an old stage of life for these animals, and counted DA neuronal cell bodies. Interestingly, we found that the number of DA neurons in old anmt-1dopa is significantly higher than in wt worms of the same age (Fig 1A, see S1B Fig for DA neurons broken down into CEP, ADE, and PDE cell subclasses). We also analyzed DA neuronal morphology by checking worms for CEP dendrite dysmorphia, axonal breaks, and abnormal cell body and axon positioning as shown in different stages of neurodegeneration in S1A Fig. In agreement with the DA cell body count, 15 days old anmt-1dopa animals display a higher percentage of healthy individuals that do not show the above-mentioned morphological defects when compared to wt (Fig 1B). Analyses of neurodegeneration phenotypes at different ages revealed that a tendency towards less morphological defects could be observed as early as day 5 of adulthood (S1D Fig), whereas no differences were found for DA cell body quantity at young ages (S1C and S1E Fig), Fig 1C depicts the presence of DA neurons in anmt-1dopa and wt animals over time. Notably, the two investigated strains that express a mutated version of anmt-1 (anmt-1dopa-MUT 1 and 2) did not show decreased neurodegeneration when compared to wt at day 15 of adulthood (S1F Fig). Table 1 contains data from a pulse and chase experiment where anmt-1dopa animals with an exclusively neuronal RNAi-sensitive background were treated with RNAi against anmt-1. Worms were put on anmt-1 RNAi at different time points and microscopy for neurodegeneration was performed at day 15 of adulthood. The beneficial effect of neuronal anmt-1 expression is completely abolished when worms were fed anmt-1 RNAi beginning from the young adult stage as DA neuronal cell body quantity is indistinguishable from wt animals of the same age (S1G Fig; see Table 1 for other neurodegeneration phenotypes). Interestingly, anmt-1dopa animals that were treated with anmt-1 RNAi from hatching experience worse neurodegeneration than wt at day 15 (S1G Fig), pointing towards a critical role for anmt-1 during neurodevelopment.
The GABAergic nervous system, as visualized with an mCherry reporter (mCherryGABA; S2A Fig), was not affected by anmt-1: No differences in GABAergic cell body and commissure number (S2B and S2C Fig), and GABAergic axonal breaks (S2D Fig) were observed in anmt-1GABA and wt at day 15. We wondered whether the lack of age-dependent neurodegeneration in anmt-1dopa might also affect their longevity and found that indeed their lifespan is significantly extended compared to wt (Fig 1D), whereas lifespan analyses in anmt-1dopa-MUT 1 and 2, transgenics that have mutations potentially disturbing ANMT-1 enzymatic function, did not yield such a result (S1H Fig). anmt-1dopa with a neuronal-sensitive background for RNAi treated with anmt-1 RNAi lived significantly shorter than anmt-1dopa fed with control RNAi (S1I Fig), suggesting that indeed the ectopic anmt-1 expression in DA neurons is responsible for the observed longevity. Surprisingly, anmt-1GABA transgenics also showed a slight lifespan extension (S2E Fig), although no effect on GABAergic neurodegeneration was observed. Furthermore, fertility tests revealed a reduced brood size in anmt-1dopa (S1J Fig), but not in anmt-1GABA animals (S2F Fig). These data suggest that DA neuronal ANMT-1 signaling regulates not only neurodegeneration and aging, but also influences reproduction, suggesting that ANMT-1 may act in an endocrine or neuroendocrine manner.
Previous research by us and other laboratories has shown that some of the beneficial effects of ANMT-1/NNMT are due to elevated concentrations of MNA, the product of NAM methylation [16]. We wondered whether this is also the case in neurodegeneration and incubated wt worms with 1 μM MNA, a concentration that has been shown previously to be lifespan-extending in C. elegans [19]. At 15 days after L4, these worms showed no significant differences in DA neuron number (S1K Fig) and morphology of the DA system compared to controls (S1L Fig). Thus, we concluded that increasing MNA levels alone are insufficient for the beneficial effects of neuronal anmt-1 expression.
Since we found DA neurodegeneration modulated by ANMT-1, we wondered whether DA neuronal anmt-1 expression influences dopamine-dependent behaviors in C. elegans, possibly resembling features of schizophrenia in humans, which is characterized by dysfunctional dopamine signaling. Therefore, we tested two common dopamine-dependent behaviors in anmt-1dopa animals, namely the abilities to sense food and ethanol [33, 34]. Briefly, when C. elegans is kept without food, they survey their environment for a potential food source. The mechanic stimulus of bacteria will make them slow down and they will remain on the food rather than moving on to other areas. This so-called “basal slowing response” is mediated by DA neurons and dopamine signaling, as is the avoidance response to ethanol. At 1 and 5 days after L4, anmt-1dopa worms do not stick to a discovered food source like wt worms (Fig 2A). Furthermore, they do not actively avoid the smell of ethanol at day 5 (Fig 2B). It is interesting to note that these behavioral abnormalities manifest only in adult worms, as they are not present in L4 larvae (Fig 2A and 2B). The fact that these behavioral assays rely on movement prevented us from examining older animals that are less mobile. We therefore tested basal slowing employing a different approach according to [35]. In short, we counted body bends of worms washed free of bacteria and put either on empty NGM plates or plates seeded with OP50. Wt animals had a significantly lower number of body bends when put on OP50 than on empty plates at day 1 (S3A Fig), 5 (S3B Fig), and 10 (S3C Fig) of adulthood. This was not the case in anmtdopa transgenics as their movement was the same on empty plates and plates with bacteria, confirming the phenotype of Fig 2A. The behavior of anmt-1dopa-MUT 1 and 2, however, resembled as expected that of wt worms, because ANMT-1 is not functional (S3D Fig). Additional phenotypes related to impaired dopamine signaling, such as swimming-induced paralysis, could not be observed. C. elegans locomotion is GABA-dependent, thus we examined locomotion in anmt-1GABA transgenics, but we did not observe any differences compared to wt controls in either 5 or 10 days old worms (S2G and S2H Fig).
Abnormal dopamine-dependent behaviors have been observed previously in C. elegans with low dopamine levels, and could be rescued with external dopamine [36]. Therefore, we incubated worms with 50 mM dopamine for 4 to 6 hours before basal slowing response and ethanol avoidance assays were performed, which however did not change either of these behaviors (S3E and S3F Fig).
We then tested dopamine and GABA concentration in anmt-1dopa and anmt-1GABA, respectively. A significant increase to 8 ± 2 pmol dopamine per mg protein was found in anmt-1dopa compared to wt animals with 3.9 ± 0.9 pmol/mg (Fig 2C). These elevated levels explain why additional external dopamine could not correct the dopamine-dependent behaviors. anmt-1 overexpressed in GABAergic neurons had no impact on GABA concentration (S2I Fig).
Because of the increase dopamine levels in anmt-1dopa, we wondered whether the longevity and decreased neurodegeneration of this strain was dependent on dopamine synthesis and signaling. Therefore, we crossed anmt-1dopa animals into cat-2(n4547) worms, which have a loss-of-function deletion in the gene that encodes the tyrosine hydroxylase CAT-2, the key enzyme in dopamine synthesis [36]. cat-2(n4547);anmt-1dopa animals do not show increased lifespan compared to cat-2(n4547) controls (S3G Fig). C. elegans has at least three dopamine receptors. DOP-1 and DOP-3, homologs of mammalian D1 and D2 dopamine receptors, that work together antagonistically to regulate dopamine-dependent locomotion, whereas a function for DOP-2 has not yet been determined [35]. In contrast to disabled dopamine synthesis, anmt-1dopa animals with a deletion in dop-3 (dop-3(vs106);anmt-1dopa), hence disturbed dopamine signaling, still show lifespan extension (S3H Fig), suggesting that dopamine production and elevated levels may mediate longevity, rather than DOP-3 signaling.
We also tested neurodegeneration in old (15 d) cat-2(n4547) and dop-3(vs106) animals and found that anmt-1dopa is still able to prevent loss of DA neurons in cat-2(n4547);anmt-1dopa and dop-3(vs106);anmt-1dopa (Figs 2D and S3I), therefore acting independently of CAT-2 and DOP-3 in regards to neuroprotection.
Since neither the ANMT-1/NNMT metabolite MNA, nor dopamine signaling was responsible for the neuroprotective effect of DA neuronal anmt-1 expression, we wondered if the methylation itself that is catalyzed by ANMT-1/NNMT influences neuronal cell metabolism. ANMT-1/NNMT uses SAM as methyl group donor, and as the reaction to MNA is irreversible, the methyl groups used cannot be recycled to recreate SAM, a mechanism that is highly conserved through evolution [22]. It has been reported in yeast that decreasing SAM levels act as stress and/or starvation signal for the cell to induce autophagy [37, 38]. Autophagy is a tightly regulated catabolic process within cell metabolism to degrade misfolded proteins and damaged macromolecules, and dysregulation resulting in too high or too low levels is observed in PD and schizophrenia [9, 13]. We hypothesized that neuronal ANMT-1/NNMT activates autophagy via decreasing SAM levels. Overly active autophagy could be problematic in early life and the abnormal behavior we observed in young adult animals might be a result of excessive autophagy. In contrast, these same autophagy levels might be of advantage later, degrading misfolded proteins and dysfunctional macromolecules as they increase with age. Since the behavioral phenotypes are not accompanied by DA neurodegeneration at older stages, we hypothesized that elevated autophagy helps to maintain neuronal health and extends lifespan in anmt-1dopa animals.
We tested whether the phenotypes we observed in anmt-1dopa worms could be reversed by inhibiting neuronal autophagy. We achieved this by knocking down the neuronal expression of essential genes driving autophagy (bec-1/BECN1, atg-13/ATG13, lgg-1/MAP1LC3A) using RNAi in an exclusively neuronal RNAi-sensitive background. bec-1 and atg-13 are critical for initiating autophagy, and lgg-1 is important at the later stage of autophagosome formation [39]. Applying RNAi against each of these genes rescued the abnormal basal slowing response of anmt-1dopa animals at 5 days of adulthood (Fig 3A) and abolished the neuroprotective effect on DA cell body maintenance and morphology at day 15 (Figs 3B, 3C and S4A). Similar treatment had no effect, or was even beneficial in regards to neurodegeneration, in control animals (S4B, S4C and S4D Fig). Notably, the neuroprotective effect of anmt-1dopa is not only abolished when these worms experience a knockdown in neuronal autophagy genes, but anmt-1dopa show an increased loss of DA cells and a higher degree of morphological defects compared to wt under these circumstances (Fig 3B and 3C). The same effect in anmt-1dopa was observed at day 5 (S4E and S4F Fig), whereas the contrary could be found in wt (S4G and S4H Fig). This suggests a secondary increase in neurodegeneration due to DA neuronal anmt-1 expression when autophagy is dysfunctional, which could contribute to PD progression under these circumstances.
We wondered if these results are reflected in life expectancy and thus, analyzed lifespan of both neuronal RNAi-sensitive anmt-1dopa and wt animals treated with bec-1, atg-13 and lgg-1 RNAis. As expected, the above-mentioned RNAis decreased lifespan significantly in anmt-1dopa animals (Fig 3D), strikingly well below wt life expectancy. In wt, neuronal loss of bec-1 and atg-13 slightly extended lifespan, while lgg-1 RNAi decreased lifespan (S4I Fig).
In sum, these data suggest that the behavioral changes, neuroprotection, and lifespan extension observed in anmt-1dopa animals is dependent on neuronal autophagy. This beneficial effect on neuronal health and lifespan is reversed when autophagy is depleted, revealing a potential mechanism of how high anmt-1/NNMT expression could increase PD risk.
To investigate more directly whether ANMT-1/NNMT is able to regulate autophagy, we used a reporter strain in which LGG-1 is tagged with mCherry [Pnhx-2::mCherry::lgg-1]. LGG-1 occurs diffusely in the cytosol under non-stressed conditions. Autophagy induction leads to LGG-1 organisation around autophagosomes, which then become visible and quantifiable as puncta. To induce autophagy, we starved worms for about 24h prior quantification and observed an increase in the average number of autophagosomes/puncta per wt worm, ranging from 34.4 ± 12.9 to 110 ± 35.5 in 1 day old adults (S5A and S5B Fig). We also used pimozide as positive control, since it has been identified as autophagy inducer in C. elegans [40], and the potent autophagy inhibitor 3-methyladenine in starved worms as negative control [41] (S5A Fig). To investigate whether autophagy levels change during aging, we determined basal autophagy levels in wt animals in 5 and 10 days old adults. Puncta quantity was similar in 5 days old worms versus 1 day old, but later increased significantly to 107.7 ± 25.3 per animal at day 10 of adulthood (S5C Fig). Strong background fluorescent prevented us from quantifying puncta at later time points. Interestingly, the increase in puncta formation due to starvation appeared to be progressively dysregulated with aging, as the increase in young adults was > 3-fold when compared to fed animals, > 2-fold at day 5, and no increase at all could be detected at day 10 (S5C Fig).
Although LGG-1 puncta do occur in neurons in general, quantification in the DA neurons is difficult, as number and size of these cells are too small to gain evaluable results. We therefore decided to address this aspect in the whole animal instead of only DA neurons. A strain that overexpresses anmt-1 under the control of its endogenous promoter (Panmt-1::anmt-1; anmt-1OEx), as well as anmt-1(gk457), a loss-of-function deletion mutant of anmt-1, were crossed into the mCherry::lgg-1 autophagy reporter strain and puncta in whole worms were quantified at various time points. We found that basal levels of autophagy were significantly higher in anmt-1OEx transgenics and significantly lower in anmt-1(gk457) mutants (Fig 4A and 4B). When worms were starved for 24h to induce autophagy the same significant differences could be observed (Fig 4A and 4B), suggesting that ANMT-1 regulates autophagy proportionally to its expression.
Subsequently, we investigated whether SAM is the mediator between ANMT-1 and autophagy. Since ANMT-1 uses methyl groups from SAM, we hypothesized it regulates its cellular levels. We analyzed SAM concentrations depending on anmt-1 expression and found significantly lower SAM levels in anmt-1OEx worms, and significantly higher levels in anmt-1(gk457) mutants (Fig 4C), suggesting that ANMT-1 indeed regulates cellular SAM concentration. SAM levels were also analyzed in anmt-1dopa whole animals and lower concentrations were found compared to wt (Fig 4D), which is surprising here given that anmt-1 is overexpressed in only eight cells of the animal. When SAM is used in a metabolic reaction, it is hydrolyzed to yield homocysteine. Homocysteine is converted either to cysteine or via tetrahydrofolate into methionine, which leads to recycling of SAM [42]. When methyl groups are not limited in the cell, SAM and homocysteine reciprocally regulate each other; i.e. high SAM causes low homocysteine, and vice versa. In anmt-1(gk457) worms SAM is not used by ANMT-1 and its levels increase (Fig 4C), whereas homocysteine expectedly decreases (S5D Fig). In anmt-1OEx and anmt-1dopa, however, both SAM and homocysteine levels are decreased (Figs 4C and 4D and S5D and S5E), suggesting the cycle is out of balance and methyl groups do indeed appear lost from metabolism.
In C. elegans, the majority of SAM is synthesized from methionine by S-adenosyl methionine synthetase (SAMS-1), which is encoded by the sams-1 gene [43]. sams-1(ok3033) animals carry a major deletion in sams-1, leading to loss of function of the protein. We found that SAM concentration in sams-1(ok3033) mutants is with 2.3 nM/mg protein about 50% lower than in wt worms (S5F Fig). As expected given the low SAM levels, we found significantly increased puncta numbers in sams-1(ok3033) compared to wt, suggesting that sams-1(ok3033) mutants have increased levels of autophagy (S5G Fig). Notably, starvation did not further elevate puncta formation, consistent with a dysregulation of autophagic processes in the absence of sams-1 and reduced SAM.
To directly address whether varying SAM levels are indeed responsible for autophagy regulation due to ANMT-1, we deprived anmt-1OEx and anmt-1(gk457) animals genetically from SAM by crossing them with sams-1(ok3033) mutants. We found the regulation of autophagosome formation through anmt-1 completely abolished, as indicated by anmt-1OEx and anmt-1(gk457) mutants showing the same autophagy levels in a sams-1(ok3033) background (Figs 4E, S5H and S5I). On a side note, we observed that anmt-1(gk457);sams-1(ok3033) double mutants were completely sterile, suggesting an important developmental aspect of this pathway. Similarly, anmt-1OEx;sams-1(ok3033) animals did not have progeny that were homozygous for both the transgene and the mutation, indicating that the overexpression of anmt-1 is lethal when SAM abundance is reduced or limited.
To examine whether the phenotypes observed in anmt-1dopa are dependent on SAM, we analyzed the presence of DA neurons and morphology at day 15, as well as lifespan in anmt-1dopa;sams-1(ok3033). sams-1(ok3033) seems to be neuroprotective for DA neurons and increases lifespan compared to wt (Fig 4F, 4G and 4H). However, anmt-1dopa could not further increase these beneficial effects (Fig 4F, 4G and 4H), suggesting that the neuroprotection and longevity caused by anmt-1dopa and loss of sams-1 share a common mechanism, which include the reduction of SAM availability, hence activating autophagy.
In yeast, SAM regulates autophagy through the methyltransferase Ppm1p, an evolutionary conserved enzyme with orthologs in humans (leucine carboxyl methyltransferase, LCMT1) and C. elegans (LCMT-1). Ppm1p uses methyl groups provided by SAM to methylate and therefore activate the catalytic subunit of PP2Ap (protein phosphatase 2A in humans, LET-92 in C. elegans). Methylated PP2A then induces dephosphorylation of Npr2p (human NPR2-like, GATOR1 complex subunit and C. elegans NPRL-2), which is part of a complex that controls autophagy and cell growth via the regulation of TORC1, and potentially others [37, 44]. In C. elegans, the pathway is reportedly involved in reproduction and development [45] whereas its neuronal role has not been further investigated. However, it is known that let-92 is highly expressed in the neurons [46]. We speculate that ANMT-1 competes with LCMT-1 for methyl groups from SAM, impairing the methylation ability of LCMT-1. This leads to decreased methylation and activity of LET-92, which can no longer dephosphorylate NPRL-2, thus inducing autophagy. Indeed, neuronal-specific RNAi downregulation of the genes involved in the pathway, lcmt-1 and nprl-2, in anmt-1dopa animals led to decreased DA cell body number at day 15 (Figs 5A and S6A), and was milder at day 5 (S6B Fig). In wt, the same treatment caused no, or a slightly beneficial, effect on DA neuronal loss at day 5 and 15 (S6C and S6D Fig). Morphological damage was strongly increased by neuronal loss of lcmt-1 and nprl-2 in anmt-1dopa at day 5 (S6E Fig) and day 15 (Fig 5B), but decreased in wt at day 5 and 15 (S6F and S6G Fig).
Both neuronal lcmt-1 and nprl-2 RNAi had a slight lifespan shortening effect in anmt-1dopa animals (Fig 5C) in contrast to wt, where lcmt-1 RNAi had no effect and nprl-2 RNAi extended lifespan (S6H Fig). We were unable to test the contribution of PP2A as it is an essential gene in C. elegans, and even RNAi applied only from L4 in an exclusively neuronal-specific RNAi-sensitive background had many non-specific effects that confounded phenotypic analyses.
The data obtained from neuronal lcmt-1 and nprl-2 loss resemble the results when autophagy was blocked by bec-1, atg-13 and lgg-1 RNAi (Figs 3 and S4), suggesting that knocking down lcmt-1 and nprl-2 might indeed affect autophagy. Subsequently, the mCherry::lgg-1 autophagy reporter strain was grown on lcmt-1 and nprl-2 RNAi and a downregulation of basal autophagy under feeding conditions compared to the control RNAi was found, as puncta quantity was significantly lower between the groups (Fig 5D). In 1 day old adults, a small increase from 28.2 ± 13/28.9 ± 14 puncta per animal under feeding conditions to 37.7 ± 13.1/41.3 ± 10.8 puncta following starvation was observed when worms were grown on lcmt-1 and nprl-2 RNAi, respectively (Fig 5D). At day 5 this effect was completely abolished (S6I Fig), suggesting that lcmt-1 and nprl-2 play an important role in regulating starvation-induced autophagy. The same pattern was observed in anmt-1OEx, where RNAi against lcmt-1 and nprl-2 was able to completely abolish starvation-induced autophagy in 1 day old adults (Fig 5E). Interestingly, knocking down lcmt-1 and nprl-2 also led to downregulation of basal autophagy levels of anmt-1OEx back to wt levels (S6J Fig). Strikingly, in anmt-1(gk457) mutants starvation could still induce autophagy when grown on lcmt-1 and nprl-2 RNAi (Fig 5F), suggesting that the pathway including lcmt-1 and nprl-2 exclusively regulates autophagy in the presence of ANMT-1.
Autophagy is known to be subject to epigenetic regulation, and NNMT has been shown to regulate epigenetic processes by modifying SAM concentrations, influencing the progression of tumorigenesis [22, 47]. Whereas the epigenetic link between NNMT and autophagy is beyond the scope of this study and needs further investigation, we wondered how autophagy and sams-1 gene expression is influenced by anmt-1. Notably, besides post-translational regulation, an important role for transcriptional control of autophagy has been described in recent research [47, 48]. Thus, we analyzed gene expression of atg-13 and lgg-1 in mixed populations of anmt-1OEx, anmt-1dopa, and anmt-1(gk457) animals. We found a stable upregulation of both genes by anmt-1OEx and anmt-1dopa compared to wt, whereas anmt-1(gk457) had no effect (Fig 6A). Since ANMT-1 controls autophagy via the lcmt-1/nprl-2 pathway, we sought to test the expression of these genes and found that nprl-2 was upregulated by both anmt-1OEx and anmt-1dopa, suggesting that ANMT-1-induced autophagy is regulated both post-translationally and transcriptionally (Fig 6B). However, the results of lcmt-1 gene expression showed pronounced differences between the experiments, leading to high standard deviations (Fig 6B), which is consistent with a potential post-translational nature of LCMT-1 activity regulation via SAM. Surprisingly, we also found a strong transcriptional downregulation of sams-1 only through anmt-1dopa (Fig 6C). This could mean that lower SAM levels in anmt-1dopa are not exclusively due to metabolic consumption, but could also be regulated on a transcriptional level, potentially through epigenetic effects. Notably, all tested genes were regulated by anmt-1dopa. We hypothesize that the effect of gene regulation in only the eight DA neurons would be too small to detect in whole animals, thus again hinting towards an endocrine-like signaling of anmt-1dopa that effects not only the nervous system, but potentially the whole organism.
In summary, these data reveal an alternative pathway of SAM and autophagy regulation in addition to post-translational/metabolic regulation, involving control on the transcriptional level, possibly through epigenetic processes.
After establishing a role of ANMT-1 in regulation of autophagy, especially in the neurons, and therefore preventing age-related neurodegeneration, we wondered if anmt-1 expression was neuroprotective in disease-like states, induced by either PD-related neurotoxins or mutations. Thus, we tested a variety of compounds that have been associated with DA neurodegeneration and/or increased risk of PD onset to determine whether increased anmt-1 expression influences the neurotoxicity of these compounds. Worms were exposed to the respective substances from the egg stage and DA morphology was measured at L4, day 5 and day 10 of adulthood. β-hexachlorocyclohexane (β-HCH) is classified as persistent organic pollutant and serum levels may correlate with PD onset risk [49]. We tested the ability of β-HCH (1 mM) to damage C. elegans DA neurons and found increased morphological damage in wt as early as in the L4 stage when compared to DMSO (S7A and S7B Fig, S1 Table for statistics). Paraquat (PQ; 300 μM) and 6-OHDA (1 mM) caused a trend towards DA morphological damage at L4 and day 5 (S7A, S7B, S7C and S7D Fig) that became significant in 10-day old worms (Figs 7A and S7E) when compared to controls. Both compounds can damage DA neurons, which was reported previously in C. elegans [50, 51] and other organisms as well as humans [52–54], and are therefore suspected to cause PD. Surprisingly, anmt-1dopa worms seemed to be completely protected against the neurotoxic effects of β-HCH, PQ, and 6-OHDA (Figs 7A, S7A, S7B, S7C, S7D and S7E). While this neuroprotective effect of ANMT-1 may be due to increased autophagy, other mechanisms could contribute to this resilience, such as upregulated stress response caused by oxidative stress, which is evoked by PQ and 6-OHDA and has been reported previously for ANMT-1 [19]. However, given the structural and functional differences of the tested compounds, it is likely that the neuroprotection caused by anmt-1dopa is a general effect rather than specific to a particular family of molecules.
To investigate whether anmt-1 interacts with genetic risk factors for PD and influences their pathologies, we tested a variety of disease models. An autosomal-dominant mutation in the alpha-synuclein gene SNCA (an alanine-to-threonine substitution at position 53, A53T) was the first discovered to be responsible for heritable cases of PD [6, 55]. Shortly after, mutations in the parkin gene PARK2 were found to be responsible for autosomal-recessive juvenile PD [7]. Later, mutations in both genes have been found to contribute to sporadic PD cases as well [56]. Therefore, we employed a C. elegans model that expresses human SNCA A53T in their DA neurons (Pdat-1::SNCA-A53T; SNCA-A53Tdopa), and pdr-1(gk488), a loss of function mutant of pdr-1, the worm orthologue of human PARK2. As previously shown [57, 58], we found that both strains show degeneration of their DA neurons at day 5 of adulthood, as indicated by the loss of DA cell bodies (S8A Fig) and abnormal positioning of these cells (S8B Fig), and other morphologic anomalies of the DA system (S8C and S8D Fig). When anmt-1 was expressed in the DA neurons of these strains (SNCA-A53Tdopa;anmt-1dopa and pdr-1(gk488);anmt-1dopa), we found that the loss and abnormal positioning of DA cell bodies compared to wt was completely abolished at day 5 (S8A and S8B Fig) and day 15 (Figs 7B, 7C and S8E), and dysmorphia of CEP dendrites (S8C and S8F Fig), and axonal breaks (S8D and S8G Fig) are the same as in anmt-1dopa. SNCA tends to build aggregates, especially when mutated. It has furthermore been reported that SNCA and SNCA-A53T might diminish autophagic processes [59]. The E3 ubiquitin ligase PARK2 is an important mediator of mitophagy, which is the selective autophagic degradation of mitochondria. We speculate that the effects of SNCA A53T and pdr-1 loss on DA neurons are, at least in part, due to accumulation of aggregated SNCA and damaged mitochondria, respectively. anmt-1 expression induces autophagic processes, which could lead to reduction of these aggregates and dysfunctional organelles, restoring neuronal function.
We explored the neuronal role of ANMT-1/NNMT in vivo and found that it regulates neuronal autophagy (Fig 8) in the DA nervous system, with wide-ranging effects on neurodegeneration, behavior, fertility, and lifespan.
NNMT has been previously reported as eliciting contradictory outcomes regarding PD risk: elevated NNMT levels were found in the brains and lumbar cerebrospinal fluid of PD patients [28, 60, 61], whereas other studies in cell tissue culture found it to be neuroprotective [32, 62–64]. Our data suggest a neuroprotective role for ANMT-1/NNMT, but it cannot be ruled out that higher expression levels and/or encountering other risk factors could lead to further dysregulation and neurodegeneration. The neuroprotection results from the deprivation of SAM, which likely acts as starvation signal to the cell. It has been shown that SAM reciprocally regulates autophagy, promoting growth under high concentrations and boosting autophagy when levels decrease [37, 38]. SAMS-1, the key enzyme in SAM biosynthesis, was initially identified in an RNAi screen for positive regulators of longevity via dietary restriction [43]. sams-1 mutants show extended lifespan and mimic other phenotypes of DR worms, such as reduced brood size and delayed reproduction [65], resembling the phenotypes that we observe in anmt-1dopa worms. Reduced SAMS-1 mRNA levels as in anmt-1dopa have also been described in genetic models of dietary restriction [43]. We therefore hypothesize that high neuronal ANMT-1/NNMT activity mimics dietary restriction by reducing the availability of cellular SAM, leading to lifespan extension. Furthermore, decreased SAM, and hence reduced methylation potential, could modulate histone and DNA methylation and affect epigenetic processes [22].
In young (5 day old) anmt-1dopa individuals, however, dopamine-dependent behavior is disturbed, which could be due to increased dopamine levels in anmt-1dopa animals. It is interesting to note that schizophrenia, with a general onset age in early adulthood [66], is associated with excessive dopamine, leading to abnormal signaling and the typical behavioral outcomes. Therapy involves the use of antipsychotic drugs that block dopamine receptors, whereas drugs that drive dopamine release or increase dopamine transmission, such as amphetamines, will exacerbate psychosis in patients with schizophrenia, and can induce schizophrenia-like symptoms in otherwise healthy individuals [15]. NNMT has been associated with schizophrenia in humans [23–25], which according to our results may be due to its influence on dopamine concentration and/or signaling. Furthermore, autophagy dysregulation in the brain plays a key role in the pathology of schizophrenia [13], and an NNMT-mediated increase in autophagy could therefore also contribute to the progression of the disease. Since ANMT-1/NNMT seems to increase autophagy levels independently of age, perhaps levels are too high in young adulthood, and become beneficial only with age as the incidence of damaged macromolecules and dysfunctional organelles in the neurons increase. Increased autophagic clearance could therefore be the basis of the ANMT-1/NNMT-dependent neuroprotection and lifespan extension we observed. Autophagy has been linked to longevity in many organisms and an emerging field of investigation concerns the differential regulation of autophagy during aging, and effects on longevity [67] and neurodegeneration [68] have been reported depending on the relative age of the organism in question. We chose to examine neurodegeneration phenotypes at the age of 15 days after L4, which may resemble the human age that is most prevalent for PD onset (around 65 years) [69]. Future research could continue in this direction and establish whether an increase in autophagic processes only in older age is sufficient to mediate the beneficial effects without influencing other autophagy-sensitive diseases in younger individuals. Notably, we also found some neurodevelopmental issues in L4 larvae (S1E Fig) that were expected to not experience any neuronal loss, which could not be confirmed by our analysis. It would be interesting to further investigate whether the individuals that have neurodevelopmental problems experience an earlier onset or faster progression of neurodegeneration in older age.
Given the influence of anmt-1 expression in the DA neurons on dopamine, and the dependency of lifespan extension on dopamine production, we speculate that dopamine might act beyond its known functions and perhaps via receptors not yet described.
However, we found a greater loss of DA cells and higher morphological damage, i.e. features of PD, in anmt-1dopa than in wt animals when autophagy was abrogated. Maybe this secondary increase in neurodegeneration, when ANMT-1/NNMT levels are high and autophagy is dysfunctional, could account for the increased NNMT expression in PD patients observed in other studies [28, 60].
The LCMT-1/NPRL-2 pathway, which links ANMT-1/NNMT to autophagy regulation, also involves PP2A. We were not able to test the worm ortholog LET-92, however, given its high expression levels in the nervous system, it is likely to have an important neuronal role [46]. Recently, it has been reported that NNMT silencing is able to activate PP2A via its effects on LCMT-1 in glioblastoma cells [70]. Subsequently, this activation of PP2A lead to inactivation of serine threonine kinases (STKs). A genome-wide association study on PD in large populations of Europe and the USA found that polymorphisms in the gene encoding STK39 significantly increases the risk for PD [10]. Thus, in the context of dysfunctional autophagy, NNMT might modulate the activity of STKs such as STK39 to trigger DA neurodegeneration, while under wt conditions the overall beneficial effects of autophagy outweigh the potentially damaging effect of modulating STK39. Interestingly, the antipsychotic drug perphenazine, currently used to treat schizophrenia, activates PP2A and rescues a potential PP2A inhibition by NNMT [70].
Taken together, our research shows the contribution of NNMT to neuroprotection and its involvement in neuronal diseases and provides evidence for autophagy as underlying biochemical pathway. We have detailed this novel molecular mechanism regulating neuronal autophagy during aging and raise the possibility of the NNMT pathway as a potential target for neuroprotective interventions in PD, schizophrenia, and other neurological diseases. Further research is required to enlighten the DA-neuronal specificity of NNMT action, and to investigate how epigenetic regulation intervenes in these processes.
C. elegans were maintained as described elsewhere [71]. Briefly, worms were kept on NGM agar plates that were streaked with E. coli OP50 as food source at 15°C. All assays were performed at 20°C, and worms were grown at 20°C at least one generation before the experiment. The N2 Bristol wildtype strain (wt), as well as BZ555 (Pdat-1::GFP; GFPdopa), MT15620 (cat-2(n4547)), LX703 (dop-3(vs106)), VK1093 (vkEx1093[nhx-2p::mCherry::lgg-1]), TU3401 (sid-1(pk3321);[pCFJ90 (Pmyo-2::mCherry) + unc-119p::sid-1]), RB2240 (sams-1(ok3033)), and VC1024 (pdr-1(gk488)) were provided by the Caenorhabditis Genetics Center at the University of Minnesota. Mutant strains were outcrossed to wt at least 4 times. Strains MIR8 (Panmt-1::anmt-1::GFP; anmt-1OEx) and MIR16 (anmt-1(gk457)) were made as described previously [19]. FX14471 (tmIS904; Pdat-1::a-syn A53T) were generated by the Iwatsubo lab [72] and obtained from Dr. Shohei Mitani. Other C. elegans strains obtained by crossing and used in this study can be found in Table 2. Homozygosity of all genotypes was confirmed by PCR.
Transgenic animals were generated as follows. Plasmid DNA with the anmt-1 gene under the control of either a DA-neuronal (dat-1) or GABA-motor neuron specific promoter (unc-47) was prepared using the Gateway Cloning system and the site-specific vector pCFJ606. Transgenics were generated by microinjection of plasmid DNA and stably integrated into a defined site of the genome (locus ttTi14024, position X:22.84) using the MosSCI technique. Alternatively, the dat-1p::anmt-1 construct, or a mutated version of this construct, together with a co-injection marker were injected into BZ555, resulting in extrachromosomal expression of either wt or mutated anmt-1 in the DA nervous system. anmt-1 was mutated using site-directed mutagenesis, and two resulting lines were analyzed (anmt-1dopa-MUT 1 and anmt-1dopa-MUT 2). Strains were outcrossed 4 times to wt.
Amino acid residues of ANMT-1 that are important for SAM binding were identified using UniProt [73] (entry P34254), which provided 5 potential binding sites at positions 35 (tyrosine; Y), 40 (Y), 80 (Y), 96 (aspartic acid; D) and 101 (asparagine; N). Protein sequence alignment of C. elegans ANMT-1 and human NNMT showed high conservation between these residues. Analysing the crystal structure of human NNMT, Peng et al. reported an additional important residue at position 197 (D) [74], which is potentially conserved with a small gap, given the existence of a D in the ANMT-1 sequence at position 219 that matches the amino acid context of D197 in NNMT. All identified potentially active residues where replaced by alanine (A), resulting in the following mutations: Y35A, Y40A, Y80A, D96A, N101A, and D219A. Mutations were generated using a QuikChange II XL Site-Directed Mutagenesis Kit (Agilent Technologies).
RNAi experiments were performed according to Kamath et al. [75]. RNAi clones (E. coli HT115) of bec-1(T19E7.3), atg-13(D2007.5), lgg-1(C32D5.9), lcmt-1(B0285.4) and nprl-2(F49E8.1) were taken from the ORFeome RNAi library (Open Biosystems) and compared to an empty vector clone (L4440). Sequencing to confirm correct clones was performed for all RNAis before use. Adult worms were put on RNAi NGM plates containing 1 mM isopropyl-ß-D-thiogalactopyranoside and 50 μg/ml ampicillin and allowed to lay eggs for about 4 hours. Progeny from L4 on was transferred every two days to avoid contamination with younger generations. Neuronal RNAi experiments were performed using the respective strain crossed into TU3401 for neuronal-specific gene silencing.
Synchronized worms of different ages were placed on microscopy slides with 2% agarose pads and immobilized with 5 mM levamisole in M9 buffer. Neuronal fluorescence microscopy was conducted with a Zeiss Axio Imager M2 microscope and Zen Pro software (Carl Zeiss Canada) with 40x amplification. Present DA and GABA cell bodies and GABA commissures were counted, and worms were screened for breaks in axons and dysmorphia (breaks, punctated GFP signal, dislocation) in CEP dendrites, and abnormal DA cellular positioning. At least fifteen animals were analyzed in each of at least 3 independent experiments per condition and two-tailed t-test was performed to determine significance.
Single hermaphrodites at the L4 stage were put on NGM plates and allow to lay eggs. Parental worms were transferred to fresh plates every 12 hours for the first 3 days, then every 24 hours until they stopped laying eggs. Progeny that reached L4 per parental worm was counted. Three independent experiments were performed in quadruplicates.
Lifespan assays were performed as previously described [19]. Briefly, worms were synchronized at the egg stage (day 0). At L4, around 50 nematodes were transferred to each of 3 fresh lifespan plates per condition. After 24–48 hours, worms were transferred on plates containing 10 μM FUDR to prevent progeny contamination. FUDR was solved in water and applied on top of the grown bacteria lawn. C. elegans that did not react to repeated gentle stimulation were scored as dead. Lost animals or non-natural deaths (bagging, protrusive vulvae) were censored. JMP 11.0.0 (SAS institute Inc.) was used for statistical analyses (see Table 3).
The assay plates were prepared as follows: an about 1 cm diameter droplet of OP50 was placed on one site of a 10 cm NGM petri dish and allowed to dry. About 100 well fed worms per experiment were placed on an empty NGM plate and let crawl up to an hour to get rid of excessive bacteria. Worms were then transferred to assay plates on the opposite site of the bacterial lawn and allowed to move for 1 hour. Subsequently, worms inside and outside the bacterial lawn were counted and the basal slowing index was calculated as follows: (worms outside of lawn)-(worms in lawn)/(complete number of worms), where a result between 1 and 0 represents a healthy behavior.
For verification we used an alternative method of testing basal slowing according to [35].Worms were synchronized at the L4 stage and experiments were performed at day 1, 5, and 10 of adulthood. 35 μl of an overnight E. coli OP50 culture were spread on a 6 cm agar dish and incubated over night at 37°C. Control plates without bacteria were treated the same. Animals were washed free of bacteria with M9. After 3 min in M9, worms were allowed a 90 sec recovery period on the respective assay plate. Subsequently, body bends were counted for five consecutive 20 sec periods. A body bend was defined as a change of direction of the complete head and pharynx region relative to the vertical axis. At least 5 animals per condition were tested.
The assay plates were prepared as follows: a 10 cm NGM plate streaked with OP50 was quartered. 60 μl 96% ethanol was pipetted on two quarters of the plate and allowed to dry for about 5 mins. About 100 well fed worms per experiment were placed in the center of the plate. Worms were allowed to move freely for 1 hour. Subsequently, worms inside and outside the ethanol quadrants were counted and the chemotaxis index was calculated as follows: ((worms outside of lawn)-(worms in lawn)/(complete number of worms))*-1, where a result between 1 and 0 represents a healthy behavior.
Dopamine drug pre-treatment was performed as described previously [36]. In sum, a 50 mM dopamine hydrochloride solution in M9 buffer was prepared freshly before the assay. 400 μl of this solution were put on a 5 cm NGM plate seeded with OP50 and allowed to dry. For control, 400 μl M9 was added. Worms were put on the prepared dopamine and control plates 4 to 6 hours before the basal slowing response and ethanol avoidance assay.
In a 96-well-plate, 30 age-synchronized worms were transferred into a well filled with 100 μl M9 buffer and OP50. Swimming locomotion was automatically tracked for 10 h using a worm tracking machine (Wmicrotracker, Phylum Tech) that performs measurements as follows. Each microtiter well is crossed by two infrared light rays from top to bottom. A detector determines how often the light rays were interrupted by worms moving in the well, and the signal is used to calculate a movement score, which is the amount of animal movement in a fixed time period. All measurements were performed in triplicates in 3 independent experiments and compared to wt worms of the same age.
Worms were synchronized, grown on OP50 or RNAi bacteria from the egg stage and transferred to fresh plates every 2 days from day 1 of adulthood. Compound plates (positive/negative control plates) were poured fresh before each assay, with pimozide at a concentration of 20 μM and 3-methyladenine at 5 mM dissolved in DMSO. Assays were performed at different ages as follows. Worms were put on empty streptomycin plates for about an hour to get rid of excess bacteria. About 24 hours before the assay, fed worms went back on fresh plates with food, starved worms were placed on the same plates that were not streaked with bacteria. For RNAi experiments, the starvation period was about 16 hours. For puncta assessment, worms were put on microscope slides with 2% agarose pads and immobilized with levamisole, and assessment was performed with Zeiss Axio Imager M2 microscope and Zen Pro software at 587 nm excitation/610 nm emission for mCherry. Pictures were taken and analysed with the “Find maxima” function of ImageJ 1.49V. Heterozygous strains (used when homozygosity of genotype caused sterility) were put in lysis buffer immediately after microscopy and stored at -20°C for single worm PCR to determine genotype.
Detection of dopamine, GABA, SAM and one-carbon metabolites was performed via HPLC (high performance liquid chromatography) coupled with ESI-MS/MS (electrospray ionisation tandem mass spectrometry) detection. The method was adapted from Wojnicz et al. [76]. Metabolites were extracted by sonication in acidified water (1.89% formic acid; sonication in ice-cold water for a total of 40 sec, with pulses of 10 sec at 40% intensity using a micro tip probe) followed by acetonitrile protein precipitation and sample concentration by drying using a refrigerated CentriVap set at 10°C. Reconstituted samples kept at 4°C were injected (30μL) and separated by a Nexera X2 HPLC system (Shimadzu) using a C18-PFP column 4.6 x 150 mm, 3 μm particle size (ACE, Scotland) protected by a C18-PFP guard column 3.0 x 10mm, 3μm particle size (ACE, Scotland); column compartment set at 30°C; gradient elution at 0.6mL/min in mobile phases A (0.1% formic acid in H2Odd) and B (acetonitrile) as follows: 0 min 5% B, 2 min 5% B, 5 min 90% B, 8 min 90% B, 10 min 5% B, 14 min 5% B. Detection was performed by ESI-MS/MS in positive ion mode on a 6500 QTrap (Sciex). Transitions used were for GABA 104.0 = >87.0 (collision energy (CE):15), for d2-GABA 106.0 = >89.0 (CE:15), for dopamine 154.0 = >91.0 (CE:33), for d3-dopamine 157.0 = >93.0 (CE:46), for SAM 399.0 = >250.1 (CE:21), for d3-SAM 402.0 = >250.0 (CE:25), and 136.0 = >90.0 (CE:15) for homocysteine.
Total RNA was obtained using Trizol (Invitrogen)/chloroform extraction as described previously [19], quantified photometrically with a NanoPhotometer (Implen) and stored at -80°C until further use. cDNA from 500 ng total RNA was generated using QuantiTect reverse transcriptase (Qiagen) and diluted 1:10 to 1:1000 to determine a concentration for each gene that yielded a CT value between 15 and 25. Gene expression was analyzed using TaqMan Gene Expression Assays (Applied Biosystems) and a QuantStudio 3 Real-Time PCR System (Thermo Fisher). Data were normalized to the housekeeping gene cdc-42 and analyzed using the Δ/Δ-CT method.
Compound plates were poured fresh before each assay and streaked with OP50. We chose a concentration of each compound where there was no visible impairment of bacterial or nematode growth. Beta-chlorocyclohexane (β-HCH) was dissolved in DMSO and used at a concentration of 1 mM. Paraquat (PQ) and 6-Hydroxydopamine (6-OHDA) were dissolved in water and tested at concentrations of 300 μM and 1 mM, respectively. All compounds were obtained from Sigma. Young adult worms were allowed to lay eggs on compound plates for about 4 hours. Progeny was investigated via neuronal fluorescence microscopy at L4, day 5, and day 10, before a significant proportion of the population started to die. Worms were transferred every 2 days to fresh plates.
Morphology of the DA system was calculated from cell body count and positioning, presence of axonal breaks, and dysmorphia in CEP dendrites. The category of no degeneration (“none”) was assumed when average cell body presence was > 95%, and < 20% of animals showed axonal breaks and abnormal positioning, abnormal cell body positioning, and dysmorphia in CEP dendrites. Slight or severe degeneration was assumed when average cell body presence was > 50% or < 50%, and axonal breaks, abnormal cell body and axon positioning, and dysmorphia in CEP dendrites occurred in < 60% and > 60% of animals, respectively. At least fifteen animals were analyzed in each of at least 3 independent experiments per condition and two-tailed t-test was performed to determine significance.
|
10.1371/journal.pntd.0007043 | Potential for sylvatic and urban Aedes mosquitoes from Senegal to transmit the new emerging dengue serotypes 1, 3 and 4 in West Africa | Dengue fever (DEN) is the most common arboviral disease in the world and dengue virus (DENV) causes 390 million annual infections around the world, of which 240 million are inapparent and 96 million are symptomatic. During the past decade a changing epidemiological pattern has been observed in Africa, with DEN outbreaks reported in all regions. In Senegal, all DENV serotypes have been reported. These important changes in the epidemiological profile of DEN are occurring in a context where there is no qualified vaccine against DEN. Further there is significant gap of knowledge on the vector bionomics and transmission dynamics in the African region to effectively prevent and control epidemics. Except for DENV-2, few studies have been performed with serotypes 1, 3, and 4, so this study was undertaken to fill out this gap. We assessed the vector competence of Aedes (Diceromyia) furcifer, Ae. (Diceromyia) taylori, Ae. (Stegomyia) luteocephalus, sylvatic and urban Ae. (Stegomyia) aegypti populations from Senegal for DENV-1, DENV-3 and DENV-4 using experimental oral infection. Whole bodies and wings/legs were tested for DENV presence by cell culture assays and saliva samples were tested by real time RT-PCR to estimate infection, disseminated infection and transmission rates. Our results revealed a low capacity of sylvatic and urban Aedes mosquitoes from Senegal to transmit DENV-1, DENV-3 and DENV-4 and an impact of infection on their mortality. The highest potential transmission rate was 20% despite the high susceptibility and disseminated infection rates up to 93.7% for the 3 Ae. aegypti populations tested, and 84.6% for the sylvatic vectors Ae. furcifer, Ae. taylori and Ae. luteocephalus.
| Dengue fever remains a major public health problem in all tropical regions of the world and causes 390 million infections each year. In Africa, while dengue outbreaks were rare during the last century, recurrent urban epidemic have been reported in all regions the last decade. Serotype 3, never reported in West Africa, caused major outbreaks in 2009 in several capital cities while serotype 2, usually confined to the forest cycle, spilled over into urban areas in Senegal and Mauritania in 2014–2015. These changes are occurring in a context where vector control remains the only effective approach to prevent and control epidemics. However, the design and the implementation of efficient vector control require an accurate knowledge of the vector bionomics and competence while such data are lacking in the African region. To fill out this gap we experimentally infected domestic and wild mosquitoes from Senegal to assess their vector competence for dengue serotypes 1, 3 and 4. Finally both domestic and wild Senegalese mosquitoes showed a low ability to transmit dengue viruses.
| Dengue fever (DEN) is the most common arboviral disease in the world and is caused by four genetically distinct serotypes of virus (DENV-1, DENV-2, DENV-3, DENV-4) belonging to the genus Flavivirus of the family Flaviviridae. Among the 390 million annual infections estimated around the world, 240 million are inapparent and only 96 million are symptomatic [1]. Dengue fever causes a wide clinical spectrum similar for the four serotypes. The different clinical manifestations of DENV infection range from asymptomatic to several symptomatic forms ranging in severity from classical dengue fever, to Dengue Hemorrhagic Fever (DHF) and Dengue Shock Syndrome (DSS). Dengue viruses are transmitted to humans by mosquitoes of the genus Aedes, mainly by the peridomestic mosquito Aedes aegypti aegypti and secondarily by Ae. albopictus and other anthropophilic Aedes mosquitoes.
In Africa, the sylvatic circulation of DENV-2 appears to be predominant [2] in contrast to Asia and South America where endemic/epidemic DENV strains circulating in peridomestic cycles are most common, and a sylvatic, nonhuman primate-amplified enzootic cycle has not been identified except for in Malaysia. The first isolations of DENV-2 from naturally infected mosquitoes in Africa date to 1969 when two strains were isolated from Ibadan and Jos in Nigeria [3]. Thereafter, several epizootics of DENV-2 were reported through the periodic amplifications of the sylvatic cycle involving wild populations of mosquitoes and monkeys in several West African countries [4]. However, despite these frequent epizooties and the presence of the epidemic vector Ae. aegypti in all bioclimatic areas, only sporadic DEN cases were recorded in West Africa. This could be explained by the presence of Aedes aegypti formosus, the ancestral African sylvatic and zoophilic form that uses tree holes as its larval habitat. Indeed, both sub-species exist in Africa but the presence of Aedes aegypti aegypti (the domestic, highly anthropophilic and primarily endophilic subspecies) in West Africa remains debatable mainly because of the lack of reliable methods to distinguish the two subspecies. The first documented outbreak caused by DENV-2 in West Africa occurred in Burkina Faso in 1982 and was suspected to be triggered by an introduction from the east of an epidemic Seychelles strain [2].
Most African DEN outbreaks caused by DENV-2 have occurred in East Africa. The others DENV serotypes (1, 3 and 4) are only known from endemic-epidemic cycles in Africa with no evidence of enzootic circulation. Only DENV-1 has been found associated with Ae. aegypti. During the last century, DENV-1 epidemics were notified in South Africa in 1926–27, Sudan in 1984, and Nigeria in 1964 and 1975 while the unique DENV-3 outbreaks occurred in Mozambique in 1985 [5,6]. Serotype 4 was only reported in Senegal in contexts which still remains enigmatic [7]. Amarasinghe et al. 2011 [6] have presented an exhaustive review on dengue situation in Africa.
Over the last 2 decades a changing epidemiological pattern has been observed in Africa, with outbreaks of DEN reported in all regions and several cases exported to Europe [8].
DENV-2, responsible for several epidemics in East Africa (Somalia, Djibouti, Kenya and Tanzania) and usually circulating in a sylvatic cycle (between Aedes mosquitoes and non human primates) in West Africa, spilled over into urban areas in 2014–2015 in Senegal and Mauritania, Gabon in 2007, Angola in 2013 and Burkina Faso in 2016. Serotype 3 (DENV-3), never reported in Africa after its first emergence in 1985 in Mozambique, caused a major urban outbreak in 2009 in Cape Verde, Cote d’Ivoire, Gabon and Senegal. Since September 2017, Burkina Faso and Senegal face up to major urban outbreaks Ouagadougou and Louga respectively (S1 Table) [9,10,11,12,13,14,15,16,17,18,19,20,21,22,23]. In Senegal all DENV serotypes have been reported (S2 Table).
These important changes summarized above in the epidemiological African profile of DEN are occurring in a context where there is no vaccine against DENV recommended for all populations. Furthermore, there is a significant gap of knowledge on DENV vector bionomics and transmission dynamics in Africa to effectively prevent and control epidemics. The vector competence of mosquitoes associated with DENV in nature is poorly characterized. Except for DENV-2 [24,25] few studies have been performed with serotypes 1, 3, and 4 [26].
Following the 2009 Dakar DENV-3 epidemic, we initiated a vector competence study to evaluate the ability of Ae. aegypti populations from Dakar and Kedougou to transmit DENV-1 and -3, for which there is no evidence of enzootic, sylvatic circulation in Africa [27]; these two serotypes appear to circulate only in an endemic/epidemic cycle with peridomestic human amplification. Our prior results showed low susceptibility to DENV-3 but high infection and dissemination rates with DENV-1. However, the oral DENV doses used were low and transmission potential was not tested. Furthermore, only Ae. aegypti was tested and vector competence data for sylvatic vectors were generated for DENV-1, -3 and -4. Thereby the present study assessed the vector competence of Senegalese Ae. aegypti, Ae. furcifer, Ae. taylori and Ae. luteocephalus for DENV-1, -3 and -4.
The University of Texas Medical Branch (UTMB) Institutional Animal Care and Use Committee approved all experiments involving animal-derived cells/tissues/sera/samples under protocol 02-09-068. UTMB complies with all applicable regulatory provisions of the U.S. Department of Agriculture (USDA)-Animal Welfare Act; the National Institutes of Health (NIH), Office of Laboratory Animal Welfare-Public Health Service (PHS) Policy on Humane Care and Use of Laboratory Animals; the U.S Government Principles for the Utilization and Care of Vertebrate Animals Used in Research, Teaching, and Testing developed by the Interagency Research Animal Committee (IRAC), and other federal statutes and state regulations relating to animal research. The animal care and use program at UTMB conducts reviews involving animals in accordance with the Guide for the Care and Use of Laboratory Animals (2011) published by the National Research Council.
Mosquito species used in this study were collected from three Senegalese localities: Dakar, Saint Louis and Kedougou (Fig 1). The Table 1 describes the characteristics and geographic origins of Ae. aegypti, Ae. furcifer, Ae. taylori, and Ae. luteocephalus populations tested. The sylvatic Ae. aegypti population from Kedougou breeding in tree holes represented Ae. aegypti formosus morphologically characterised by the lack of pales scales on the first abdominal tergite and the urban populations from Dakar and Saint Louis breeding in artificial containers were consistent with Ae. aegypti aegypti contrariwise characterised by the presence of pales scales. These species were chosen based on their abundance, anthrophophilic behavior and association with DENV in nature. For each population, several larval habitats were sampled and immature stages were collected and reared in the laboratory. For Ae. furcifer, Ae. taylori and Ae. luteocephalus adult females were caught in a gallery forest at Kedougou and reared in the laboratory. Progeny of these populations were considered as the F1 generation that we used for experimental infections. Adult mosquitoes were maintained with a 10% sucrose solution at 27 °C, 75–80% relative humidity (RH), 12:12 h (Light:Dark) photoperiod.
Hosts origin, year of collection and passage histories of the virus strains used in this study are presented in Table 2. DENV-1, DENV-3 and DENV-4 strains obtained from the World Reference Center for Emerging Viruses and Arboviruses at the University of Texas Medical Branch in Galveston, Texas. For the Ae. furcifer experiment we used the DENV-4 strain from Haiti (Haiti 73), and DENV-3 strain from Barbados in the Caribbean region of North America (Carec 01–11828). For other mosquito species we used the following African strains: DENV-1 (SH 29177); DENV-3 strain (S-162 TvP-3622), and DENV-4 strain SH 38549.
An additional passage on C6/36 cells was performed for each strain to obtain the viral stock used to infect mosquitoes. Cell lines were provided by the American Type Culture Collection (Manassas, Va.), and cultured in Gibco DMEM (Dulbecco’s Modified Eagle Medium), High glucose (Gibco Cat. No. 11965–092) supplemented with 10% fetal bovine serum (FBS; Atlanta Biologicals Cat. No. S11150) heat-inactivated in 56° C water bath for 60 min, penicillin-streptomycin (Gibco Cat. No. 15140–122) and 10% of Bacto Tryptose Phosphate Broth (Becton, USA). Virus in cell culture supernatants was concentrated using Millipore UFC910024 Amicon Ultra-15 Centrifugal Filter Concentrator with Ultracel 100 Regenerated Cellulose Membrane. Concentrated viruses were collected, aliquoted and frozen at -80°C, and used as viral stocks for mosquito infection. Virus stocks were titrated using the method focus forming assays and immunostaining described below.
Virus titers for stocks and infectious blood meals after 1 hour of exposure to mosquitoes were determined by focus forming assays and immunostaining as described previously [28]. Briefly, Ten-fold serial dilutions of virus in MEM supplemented with 2% FBS and antibiotics (Invitrogen, Carlsbad, CA), were added in duplicate to confluent C6/36 cell monolayers attached to 24-well Costar (Corning, NY) plates, and incubated for 1 h with periodic gentle rocking to facilitate virus adsorption at 28 °C. Wells were then overlaid with 1 ml of 0.8% methylcellulose (Sigma-Aldrich, St. Louis) diluted in warm Optimem (Invitrogen) supplemented with 2% FBS, antibiotics and 1% (w/v) L-glutamine and incubated undisturbed for 4 days at 28 °C. Methylcellulose overlay was aspirated and cell monolayer rinsed once with phosphate buffered saline (PBS), pH 7.4 (Invitrogen) followed by fixation with a mixture of ice-cold acetone and methanol (1:1) solution and allowed to incubate for 30 min at room temperature (RT). Fixation solution was aspirated and plates were allowed to air dry. Plates were washed thrice with PBS supplemented with 3% FBS, followed by hour-long incubation with a dengue-specific hyperimmune mouse ascitic fluid. Mouse hyperimmune sera (MIAF) to DENV were prepared in adult mice; using 10% crude homogenates of DENV- infected newborn mouse brain in phosphate-buffered saline as the immunogen. The immunization schedule consisted of four intraperitoneal injections of antigen mixed with Freund’s adjuvant, given at weekly intervals. After the final immunization, mice were inoculated with sarcoma 180 cells, and the resulting immune ascitic fluids were collected. All animal work was done at UTMB under an IACUC approved animal use protocol (number 9505045). Plates were washed thrice followed by hour-long incubation with a secondary antibody, goat anti-mouse conjugated to horseradish peroxidase (HRP) (KPL, Gaithersburg, MD). Detection proceeded with the addition of aminoethylcarbazole (AEC) substrate (ENZO Life sciences, Farmingdale, CT) prepared according to vendor instructions.
Three- to 5-day-old F1 female mosquitoes were placed into 500 mL cardboard containers and sucrose-starved for 48 hours before being exposed to an infectious artificial blood meal (Hemotek Ltd, UK) using BALB/c mouse skins obtained from the University of Texas Medical Branch Animal Resource Center, as membranes. The blood meal contained a 33% volume of washed sheep erythrocytes and a 33% volume of a cell culture-derived virus stock supplemented with 21% FBS, and adenosine triphosphate (ATP) to a final concentration of 0.005 M as a phagostimulant, and sucrose at a final concentration of 10%. After feeding for up to 60 minutes, the remaining blood meal was kept at– 80 °C for virus titration using plaque assay then mosquitoes were cold-anaesthetized and fully engorged specimens were incubated with 10% sucrose at 27°±1°C, a relative humidity of 70–75% and 12:12 h (Light:Dark) photoperiod.
At 7 or 15 days post bloodmeal (dpbm), mosquitoes were cold-anaesthetized and their legs and wings were removed. The proboscis of each mosquito was then inserted into a capillary tube containing 1–2 μL of FBS for salivation for up to 30 min then expectorated saliva was collected into a tube containing 100 μL of DMEM supplemented with 5% FBS. Detection of DENV in the mosquito body but not the wings/legs indicated a non-disseminated infection (limited to the midgut), whereas the presence of virus in both the body and wings/legs indicated dissemination into the hemocoel. Mosquito bodies as well as wings/legs of infected bodies were tested for DENV after homogenization in 400 μl of MEM containing 5% of FBS, and centrifugation for 2 min at 11,500 x g at 4 °C to separate virus supernatant and debris. For each sample, 100 μl of supernatant were cultured in 24-well plates containing Vero cell monolayers and DENV was detected by focus forming assays and immunostaining described above, but without the ten-fold serial dilutions. So detection was limited to presence/absence revelation. Saliva of infected wings/legs were tested to detect DENV presence by real-time RT-PCR using an internal control of 10 no-infected mosquito saliva pooled together; 100 μl of each sample was used for RNA extraction using the QIAamp Viral RNA Extraction Kit (QIAgen, Heiden, Germany), according to the manufacturer’s protocol. Dengue virus RNAs extracted from mosquito saliva were amplified using Bio-Rad iTaq universal probes one-step kit (Cat#172–5141) following Manufacturer’s protocol. For detecting DENV-1 and DENV-3, forward primer (5’ATTAGAGAGCAGATCTCTG 3’), reverse primer (5’TGACACGCGGTTTC 3’), and Probe 5’/56-FAM/TCAATATGCTGAAACGCG/3BHQ_1/-3’ were used; for DENV-4, forward primer 5’AAT AGA GAG CAG ATC TCTG 3’ was used. The RT‐PCR was performed by Quant Studio 6 Flex instrument made from applied BioSystems by life technologies. The cycling conditions were RT step at 50.0 °C for 10 min, at 95.0 °C for 3 min, and 43 cycles of 15 s at 94.0 °C and 1 min at 55 °C.
During our experiment with Ae. furcifer, we observed 5 days after oral DENV exposure a high mortality rate. Based on this observation, we planned subsequent experiments to include a negative control cohort exposed to uninfected blood meals to assess the effect of DENV on mortality. The uninfected blood meals used as the negative control contained a 33% volume of washed sheep erythrocytes and 33% volume of cell culture media (Gibco DMEM, High glucose supplemented with 10% fetal bovine serum, penicillin-streptomycin and 10% of Bacto Tryptose Phosphate Broth) supplemented with 21% FBS, and adenosine triphosphate (ATP) to a final concentration of 0.005 M as a phagostimulant, and sucrose at a final concentration of 10%. The Table 3 showed the sample size for each virus strain and for each mosquito populations. These mosquitoes were monitored twice daily for mortality until 15 dpbm for Ae. taylori, Ae. aegypti from Kedougou and St. Louis and 20 dpbm for Ae. aegypti from Dakar, then surviving mosquitoes were tested for DENV infection as described above.
Infection (number of positive bodies/total number of engorged mosquitoes incubated and tested), disseminated infection (number of mosquitoes with positive wings-legs/ total number of engorged mosquitoes incubated and tested) and transmission (number of mosquitoes with infected saliva/ total number of engorged mosquitoes incubated and tested) rates were calculated for each species and each dpbm. The rates obtained were compared using Fisher’s exact test. For Ae. aegypti populations potential impact of the virus serotype, incubation and mosquito origin were estimated using beta regression model. A Wilcoxon test was performed to compare differences between survivals among groups. For all tests, differences were considered statistically significant at p < 0.05 using R v. 2.15.1 (R Foundation for Statistical Computing, Vienna, Austria) [29].
The titers of DENV stocks used ranged between 107 and 108 PFU/ml and the Table 4 presented the titers of the blood meals prepared from these stocks after 1-hour exposure at 37±1 °C for mosquitoes feeding in different days. These titers ranged between 1.2 x 106 and 4.7 x 107 PFU/ml. A total of 606 Ae. aegytpi (240 from Dakar, 206 from St. Louis and 160 from Kedougou), 86 Ae. taylori, 71 Ae. furcifer and 22 Ae. luteocephalus was tested after DENV exposure and incubation for 7 or 15 days. For Ae. aegypti, the minimum and maximum values of infection rates were 87.5–92.5% and 90–95% for the population from Dakar, 62.5–71.42% and 88.23–100% for the St. Louis population, and 70–80% and 87.5–100% for the Kedougou population, respectively, at 7 and 15 dpbm (Fig 2). Disseminated infection rates were 57.5–67.5% and 60–72.5% for the population from Dakar, 50–62.85% and 86.66–93.75% for the population from St. Louis and 50–56.66% and 66.66–93.33% for population from Kedougou respectively at 7 and 15 dpbm. While the infection and dissemination rates were high, the potential transmission (saliva infection) rates were globally low (0–20%), 0–5% for Dakar, 0–2.85% and 0–5.88% for St. Louis, and 0% and 0–3.33% for Kedougou, respectively at 7 and 15 dpbm.
Results showed that all species were susceptible to disseminated infection with DENV-1, -3 and -4 (Fig 2 and S1 Fig). Ae. aegytpi population from Dakar showed higher infection rates (IR) than populations from St. Louis and Kedougou for all 3 dengue serotypes at 7 dpbm. However, differences were significant only between the Dakar and St. Louis populations for DENV-1 (Fisher’s exact test: p = 0.01) and DENV-3 (Fisher’s exact test: p = 0.003). Infection rates of Ae. aegypti populations increased significantly between 7 and 15 dpbm for all 3 serotypes except for the population from Dakar. At 15 dpbm, IRs of the 3 populations did not differ significantly (Fisher’s exact test: p > 0.05). For all Ae. aegypti populations, IRs with DENV-3 were higher than those obtained with DENV-1 and DENV-4. However, the difference was statistically significant only for Ae. aegypti from Kedougou when we compare IR obtained with DENV-3 versus DENV-1 (Fisher’s exact test: p = 0.04). The minimum and maximum values of disseminated infection rates of the 3 populations of Ae. aegypti were 50–65% for DENV-1, 50–57.5% for DENV-3 and 56.66–67.5% for DENV-4 and were statistically comparable at 7 dpbm (Fisher’s exact test: p > 0.05), while at 15 dpbm Ae. aegypti from St. Louis showed significantly higher DIR than populations from Dakar (Fisher’s exact test: p = 0.001) and Kedougou (Fisher’s exact test: p = 0.005) for DENV-4. With Ae. aegypti populations from Kedougou and St. Louis the IRs and DIRs increased between 7 and 15 dpbm, however the population from Dakar were susceptible to infection and developed disseminated infection with same rates at 7 and 15 dpbm.
Among the sylvatic vectors, Ae. furcifer showed the highest IRs for DENV-3 and DENV-4 but differences were not significant (Fisher’s exact test: p > 0.05). No significant difference was observed for DENV-4 infection and dissemination among the three species despite the higher IR with Ae. furcifer and lower with Ae. luteocephalus.
The Fig 3 shows titers of the infected saliva. Globally we observed mainly for Ae. aegypti a decreasing of titer between 7 dpbm and 15 dpbm. The highest titer (29 PFU/ml) was observed with Ae. taylori. For Ae. aegypti highest titers were 15 and 16 PFU/ml for Dakar and Kedougou mosquito strains at 7 and 15 dpbm respectively.
The regression model did not reveal an effect of the virus serotype on the infection rates of Ae. aegypti populations (Table 5). However, odds ratio of mosquito strain from Saint-Louis versus mosquito strains from Dakar, decreases significantly by a factor of 0.4 (p<0.001) while relative proportion of infected mosquitoes increases by a factor of 3.5 at 15 dpbm compared to 7 dpbm (p<0.001).
For the dissemination rates, no effects of the virus serotype and mosquito origin were observed. The incubation period was the unique parameter affecting the dissemination with an odds ratio at 15 dpbm increasing by a factor of 2.59 compared to 7 dpbm.
No statistically significant relationship between transmission rate and mosquito origin, virus strains and dpbm.
When we compared survival of Ae. aegypti mosquitoes from Dakar exposed to DENV-1, -3 and -4 with infection rates of 92.68, 93.02 and 91.66%, respectively, to that of the unexposed control group, globally the difference was statistically significant (Wilcoxon test: p <0.001) (Fig 4A). For Ae. aegypti mosquitoes from St. Louis exposed to DENV-1, -3 and -4 with infection rates of 88.23, 100 and 93.75% respectively compared to unexposed group (Fig 4B), significantly higher mortality was observed (Wilcoxon test: p = 2.57x10-12). For Ae. aegypti from Kedougou (Fig 4C) exposed to DENV-1, -3, -4 with infection rates of 87.5, 100, 91.66%, respectively, mortality was also significantly higher than that of the negative controls (Wilcoxon test: p = 0.0009). The difference was also overall significant for Ae. taylori (Wilcoxon test: p = 0.01) (Fig 4D), which showed infection rates of 68, 76.66, and 83.87% for DENV-1, DENV-3 and DENV-4 respectively.
For Ae. aegypti from Kedougou and Ae. taylori, our analysis did not reveal any effect between DENV serotypes and the mosquito population survival rate (p = 0.344 and p = 0.378 respectively). But survivals were significantly different between DENV serotypes for Ae. aegypti Dakar (p <0.001) and Ae. aegypti Saint Louis (p = 0.002). For Ae. aegypti from Dakar the DENV-1 induced the highest mortality (Wilcoxon test: p<0.001).
Kaplan–Meier survival curves are shown that mortality was higher for Ae. aegypti from Dakar exposed to DENV versus unexposed. Also, survival of this population was more affected by DENV-1 than by DENV-4 and DENV-3.
Our results showed that DENV infection also affected the survival of Ae. aegypti from St. Louis. Survival of mosquitoes exposed to all three DENV was reduced compared to negative controls from the 6th dpbm.
Ae. aegypti from Kedougou and exposed to DENV-4 survived better until 9 dpbm, then mortality increased highly compared to controls from the 11th dpbm. Mosquitoes exposed to DENV-1 had reduced survival early compared to DENV-3 and DENV-4. From the 11th dpbm, survival of Ae. aegypti from Kedougou was significantly affected by all three DENV serotypes.
Survival of Ae. taylori exposed to DENV-4 or -3 was significantly lower than controls at all dpbm, but there was no significant difference between DENV serotypes. While mosquitoes exposed to DENV-1 showed reduced survival for the first 7 dpbm, no significant difference was observed later.
Our study provides important information on the vector competence of both sylvatic and domestic populations of Ae. aegypti and three sylvatic species of Aedes while some aspect could be considered as minor limitations. First, due to limited number of specimens available we were not able to test Ae. furcifer with the African DENV strains after an early experiment performed with DENV-4 and DENV-3 strains respectively from Haiti and Barbados. For the same reason we also limited the experiment with Ae. furcifer, Ae. taylori and Ae. luteocephalus at 15 dpbm only. Furthermore, only the RT-PCR was used to detect DENV genomes whether infectious particles or not for 2 reasons: i) the purpose of this article is to show the competence of the vector and we have been focused on the detection of DENV in the different compartments of the mosquitoes and in the saliva. As we have shown that the virus reached the saliva, it implies that the vector is competent ii) In our experience with other viruses, (West Nile, Usutu), we have noticed that RT-PCR and infectious viral particles are generally very consistent and concordant in their conclusions and trends [30,31]. Such a trend has also been confirmed on C6/36 cells for other virus [32]. Despite the relevance of the capillary feeding method for virus transmission assessment there is no proof of salivation activity for each tested individual. However, as the use of animals has many limitations, it is currently the best alternative technique [33,34], successfully used over years with different media like defibrinated blood [35], mineral or immersion oils [36,37], foetal bovine serum [38] to measure virus transmission. One way to assess the presence of saliva could be the detection of saliva components like protein or carbohydrates. Such an approach will require, however, a larger volume of media for saliva collection to achieve the different analysis without compromising virus detection. During our experiments, no DENV-susceptible laboratory mosquito strain was used as control. In our knowledge, there is no unique mosquito strain that can serve as a single control for each of the DENV serotypes and genotypes. Beyond the current scope of the study, future experiments would take into account this aspect by integrating at least one laboratory susceptible mosquito strain for each DENV serotype.
Our results showed high infection and disseminated infection rates with DENV serotypes 1, 3 and 4, both for Ae. aegypti populations and for the sylvatic mosquito vectors Ae. furcifer, Ae. taylori and Ae. luteocephalus. IRs with Ae. aegypti population from Dakar reached their maximum values as early as 7 dpi, while for Ae. aegypti from Kedougou and St. Louis, IRs increased between 7 dpbm and 15 dpbm to reach their maximum values later. Ae. aegypti mosquitoes from Dakar develop DENV infection earlier than populations from Kedougou and St. Louis and this could be explained by highest blood meal titer for Ae. aegypti aegypti Dakar than others populations. However differences were not significant between populations from Dakar and Kedougou for DENV-1 and -3, as observed earlier [27], but IRs and DIRs were higher in the present study. This could be explained by difference of virus strains and especially by high oral virus doses (106–107 PFU/ml) compared to those used previously (103–104 MID50/ml (Mice Infectious Dose 50)) with different Ae. aegypti populations from Thailand [39]. Moreover Ae. aegypti populations from western (Burkina Faso: Koro, Bobo and Kari mosquito strains) and eastern (Kenya: Rabai and Shimba Hills mosquito strains) Africa showed lower rates despite the same viral titers (107.3–108.1 MID50/ml). Similarly in many DEN-endemic countries infection and dissemination rates obtained [40] were lower than those showed by this study.
Comparisons between serotypes show that Senegalese Ae. aegypti were more susceptibility to DENV-3 than to other serotypes. Also, DENV-3 was detected in the saliva of all three populations; this could explain recent DENV-3 outbreaks in many African countries (S1 Table). The St. Louis population showed transmission potential for all three DENVs even if TRs were low, suggesting a salivary gland barrier within some Ae. aegypti populations. However, the population from Dakar showed higher TRs than other populations and at 7 dpbm these rates for DENV-4 reached 20% of total engorged mosquitoes and 7.5% for DENV-3. These transmission rates observed just a week after mosquito infection may explain the Dakar DENV-3 epidemic like the Cape Verde Ae. aegypti population transmitting during the 2009 outbreak while it showed infection rates of 0% at 7 dpbm and until 10 dpbm the transmission rates did not exceed 20% of only mosquitoes which disseminated the infection [41]. The potential transmission rates obtained with DENV-4 show that even if a large outbreak has not been reported, the risk of a DENV-4 epidemic is present in Dakar and St. Louis because, as noted before, even with low transmission rates a vector can cause epidemics based on its abundance, density, survival and human feeding frequency [42].
Regarding enzootic, sylvatic vectors, infection and dissemination rates were relatively high for the different serotypes tested (Fig 2 and S1 Fig). The same population of Ae. furcifer, with different DENV-2 strains, showed similar IRs (26 to 97%) and DIRs (17 to 75%) to the rates we obtained [24]. But in our study we did not detect transmission potential by Ae. furcifer. The decreasing of viral titers in mosquito saliva observed at least in Aedes aegypti between 7 and 15 dpbm may explain the low transmission rates obtained.
Vectorial capacity is the efficiency of a vector in the transmission of a pathogen due to the combined effects of many factors, both intrinsic and extrinsic. The mortality rate is an important component [43,44,45]. Even if vectors become infected after taking an infectious blood meal, if they fail to survive to bite another host, the potential for transmission by this population is low. As a result, changes in the mosquito mortality rate would directly affect transmission of the pathogen. In our study we found that the exposure of Ae. aegypti from Dakar, Kedougou and St. Louis to the 3 DENV serotypes significantly increases mortality compared to negative control cohorts exposed to uninfected blood meals. Adverse effects on the fitness of Ae. aegypti due to DENV infection were also reported previously [46]. We also found increased mortality of infected Ae. taylori mosquitoes. Several other studies showed effects of other arboviruses on the survival of mosquito vectors [47,48].
Our results showed that for all 3 populations of Ae. aegypti, DENV-1 exposure affects mosquito survival. However, for the Ae. taylori population, after 7 dpbm we no longer detected an effect on survival of mosquitoes with DENV-1 infection. This absence of effect of DENV-1 infection on Ae. taylori is surprising because this species is not normally adapted to this DENV serotype, which is not known to circulate in the forest galleries frequented by Ae. taylori in Africa [49]. Indeed, only DENV-2 has been shown to circulate regularly in a sylvatic cycle in southeastern Senegal in the Kedougou region [7,50]. In summary, our results indicate that DENV-4 exposure did not affect survival of Ae. aegypti from Kedougou before 9 dpbm but it affected early the survival of the Ae. aegypti populations from Dakar and Saint Louis and Ae. taylori. DENV-3 caused high mortality in all mosquito populations tested, mainly in Ae. taylori and Ae. aegypti from Dakar. Survival was been most affected by DENV-1 which showed the highest potential transmission rates.
|
10.1371/journal.pgen.1002901 | The PARN Deadenylase Targets a Discrete Set of mRNAs for Decay and Regulates Cell Motility in Mouse Myoblasts | PARN is one of several deadenylase enzymes present in mammalian cells, and as such the contribution it makes to the regulation of gene expression is unclear. To address this, we performed global mRNA expression and half-life analysis on mouse myoblasts depleted of PARN. PARN knockdown resulted in the stabilization of 40 mRNAs, including that encoding the mRNA decay factor ZFP36L2. Additional experiments demonstrated that PARN knockdown induced an increase in Zfp36l2 poly(A) tail length as well as increased translation. The elements responsible for PARN-dependent regulation lie within the 3′ UTR of the mRNA. Surprisingly, changes in mRNA stability showed an inverse correlation with mRNA abundance; stabilized transcripts showed either no change or a decrease in mRNA abundance. Moreover, we found that stabilized mRNAs had reduced accumulation of pre–mRNA, consistent with lower transcription rates. This presents compelling evidence for the coupling of mRNA decay and transcription to buffer mRNA abundances. Although PARN knockdown altered decay of relatively few mRNAs, there was a much larger effect on global gene expression. Many of the mRNAs whose abundance was reduced by PARN knockdown encode factors required for cell migration and adhesion. The biological relevance of this observation was demonstrated by the fact that PARN KD cells migrate faster in wound-healing assays. Collectively, these data indicate that PARN modulates decay of a defined set of mRNAs in mammalian cells and implicate this deadenylase in coordinating control of genes required for cell movement.
| Almost all cellular mRNAs terminate in a 3′ poly(A) tail, the removal of which can induce both translational silencing and mRNA decay. Mammalian cells encode many poly(A)-specific exoribonucleases, but their individual roles are poorly understood. Here, we undertook an analysis of the role of PARN deadenylase in mouse myoblasts using global measurements of mRNA decay rates. Our results reveal that a discrete set of mRNAs exhibit altered mRNA decay as a result of PARN depletion and that stabilization is associated with increased poly(A) tail length and translation efficiency. We determined that stabilization of mRNAs does not generally result in their increased abundance, supporting the idea that mRNA decay is coupled to transcription. Importantly, knockdown of PARN has wide ranging effects on gene expression that specifically impact the extracellular matrix and cell migration.
| The poly(A) tail added to mRNAs during processing in the nucleus stimulates mRNA export and translation through its association with poly(A)-binding proteins. In contrast, the removal of the poly(A) tail renders transcripts translationally silent and is also the first step in decay of the majority of transcripts in eukaryotic cells [1]. As such, the process of deadenylation has the ability to profoundly influence cellular gene expression on multiple levels. Numerous mammalian deadenylases have been identified and characterized to varying extents. They fall into two enzymatic groups; the DEDD-type (including PARN, CAF1/CNOT7 and PAN2) which bear an Asp-Glu-Asp-Asp motif in their active site, and the Exonuclease/Endonuclease/Phosphatase (EEP) type (including CCR4/CNOT6, Nocturnin (CCRN4L) and Angel proteins (ANGEL1, ANGEL2)) [2]. Of the many known poly(A) shortening enzymes, the CCR4/NOT complex and PARN are by far the best studied. CCR4/NOT represents the major cytoplasmic deadenylase in yeast where it initiates decay of the majority of mRNAs [3]. The yeast CCR4/NOT deadenylase is a large complex and contains two subunits with deadenylase activity(Ccr4p and Caf1p) as well as several other factors including the NOT proteins (Not1p-Not5p). In mammals, there are five CCR4-like proteins (CNOT6, CNOT6L, CCR4N4L/Nocturnin, ANGEL1 and ANGEL2) and three CAF1-like proteins (CNOT7, CNOT8 and CAF1Z/TOE1). Of these, CNOT6, CNOT6L, CNOT7 and CNOT8 associate with the mammalian NOT proteins to form various CCR4/NOT complexes [4]. In mammalian cells, CCR4/NOT complexes have been implicated in both miRNA-mediated and AU-rich element (ARE) mediated mRNA decay mechanisms. CCR4/NOT is recruited to mRNAs by the miRNA-associated GW182 protein [5], and by the ARE-binding protein tristetraprolin (TTP/ZFP36) [6]. Thus, for mRNAs bearing certain sequence determinants, the CCR4/NOT class of related deadenylases has an important role to play in initiating mRNA decay.
On a biochemical level, PARN is perhaps the best understood deadenylase, in part because it is the predominant activity in mammalian cell extracts [7]. PARN is unique in being able to interact with both the cap and poly(A) tail [7]–[9] and has been linked with mRNA decay mechanisms in both the nucleus [10] and the cytoplasm [7], [11]. In the cytoplasm, PARN plays an important role in controlling gene expression during the maternal-zygotic transition in Xenopus [12] and during the DNA damage response in mammalian cells [11]. In the nucleus, PARN has been linked with decay of transcripts undergoing 3′ end formation following DNA damage [10] and is important for trimming the 3′ ends of snoRNAs during their maturation [13]. In addition, PARN interacts directly with RNA-associated factors including CELF1/CUGBP1 [14], PUM2 [15] and CPEB [16]. Very few bona fide mRNA substrates of PARN in mammals have been identified to date. The Fos and Myc mRNAs exhibit increased abundance following PARN KD in HeLa cells [10] and Tnfa transcripts are deadenylated in a PARN-dependent manner in vitro [17] but to our knowledge there is no published evidence for a direct effect of PARN on mRNA stability in living mammalian cells.
Because of the large number and diversity of deadenylase activities in mammalian cells it has been challenging to discern their individual roles and their global impact on cell function. It remains unknown whether each mRNA must be targeted by a specific deadenylase to achieve appropriate control of gene expression. The impact of deadenylase activity on mRNA decay rates, mRNA abundance and translation efficiency is also not clear. Previous attempts to address these questions using RNA interference approaches have suggested partially overlapping roles for the CCR4-like (CNOT6, CNOT6L) and CAF1-like (CNOT7, CNOT8) deadenylases [18]. Surprisingly, less than 2% of all mRNAs showed changes in abundance following depletion of either CNOT6/CNOT6L or CNOT7/CNOT8 [18], implying that there is either redundancy in function between the many different deadenylase enzymes and/or that changes in mRNA abundance are not a good measure of deadenylase impact. Global measurements of mRNA decay rates following knockdown of deadenylases are necessary in order to distinguish these possibilities.
In this study we aimed to shed light on the role of PARN deadenylase in C2C12 myoblasts by directly assaying global mRNA decay rates and mRNA abundances following knockdown of PARN. We identified a relatively small set of 40 mRNAs whose decay was reduced following PARN KD and independently verified this observation for four of these transcripts. For Zfp36l2 mRNA we also showed that PARN knockdown induces elongation of the poly(A) tail and increased protein abundance. Enhanced translation efficiency in PARN KD cells was also observed for a reporter bearing the Zfp36l2 3′UTR. We conclude that PARN is directly required for deadenylation of Zfp36l2 and almost certainly other mRNAs within the stabilized set. Interestingly, slower mRNA decay did not result in the expected increases in abundance for the majority of stabilized mRNAs. We attribute this to reduced transcription rates, supporting the recently established idea of coupling between mRNA decay and transcription [19]–[22]. We also investigated the effects of PARN depletion on cellular function. We determined that loss of PARN activity decreased the abundance of transcripts encoding factors linked with cell adhesion and cell movement; processes that require extracellular matrix (ECM) interactions. This led to the discovery that depletion of PARN enhances wound healing in C2C12 myoblasts.
We used a lentiviral vector encoding an shRNA targeting the 3′UTR of murine Parn to generate a stable clonal C2C12 myoblast line with reduced expression of PARN (PARN KD). Parn mRNA and protein abundance were evaluated by qRT-PCR (Figure 1A) and western blotting (Figure 1B) in the PARN KD cell line and in a cell line generated with a control lentiviral vector lacking shRNA sequences (CTRL). The PARN KD cell line showed a robust reduction in PARN expression (Figure 1A and 1B).
We first wanted to assess mRNA decay rates in the PARN KD cells and compare them to those we obtained previously in the CTRL cell line [23]. Briefly, both cell lines were treated with Actinomycin D (Act-D) for 30 minutes and samples were collected at 0, 10, 50, 110 and 230 minutes after transcription inhibition. Total RNA was isolated from each sample and used to generate cDNA probes for hybridization to microarrays. The experiment was repeated in triplicate and three independent half-lives were generated for each transcript in each cell line by plotting the abundance at each time point and fitting to an exponential decay curve. As an example the half-lives for the Gpsm1 mRNA in the two cell lines are shown in Figure 1C. Each half-life was considered reliable if the data fit well to the curve (p<0.05) and the 95% confidence interval was less than twice the half-life. We required that the half-life met these criteria for at least two of the three replicates. Both PARN KD and CTRL cells were assayed at the same time but analysis of the results from the CTRL cells was published previously [23].
Reliable half-lives were generated for 1581 mRNAs in the PARN KD cells (Dataset S1A; GSE35944). Although this dataset is somewhat smaller than that previously obtained for the CTRL cell line [7398 mRNAs; 23], it is nevertheless large enough to be informative. Overall, we obtained half-lives in both cell lines for 1389 mRNAs (Dataset S1B; GSE35944). Comparison of half-lives in CTRL and PARN KD cells allowed us to identify 64 transcripts that showed a statistically significant difference in decay rate between the two cell lines with 40 transcripts showing stabilization and the remaining 24 being destabilized (Table 1 and Table S1, respectively). To ascertain that the microarray analysis reflected true changes in mRNA decay rates, we assayed half-lives following Act-D treatment for four of the stabilized transcripts (Adora2b, Zfp36l2, Gpsm1 and Ankrd54) by qRT-PCR (Figure 2A–2D). These transcripts were selected because they have relatively short half-lives (less than 2 hours) allowing us to assess their decay over a time frame that minimizes the toxic effects of Act-D on the cell. All four transcripts were significantly more stable following PARN knockdown, as predicted by the microarray analysis. Moreover, instability of the Zfp36l2 mRNA was restored by transfection of an expression vector encoding shRNA-resistant human PARN demonstrating that stabilization was not caused by off-target effects of the shRNA on expression of unrelated genes (Figure 2E and 2F). Thus, we conclude that the PARN deadenylase influences decay rates of a subset of mRNAs in mammalian cells.
Given that PARN is a deadenylase, we predicted that mRNAs stabilized by PARN KD would show effects on the length of their poly(A) tail. We investigated this possibility for the Zfp36l2 mRNA using an RNase H/northern blotting approach. Briefly, total RNA isolated from CTRL and PARN KD cells was treated with an oligonucleotide and RNase H to induce cleavage ∼120 nt upstream of the poly(A) tail. After separation on a polyacrylamide gel followed by electroblotting, the 3′ fragment was detected using a radiolabeled probe complementary to the 3′UTR. As shown in Figure 3A and 3B, the poly(A) tail of Zfp36l2 mRNA was clearly elongated in PARN KD cells compared to the CTRL cells. In fact, in the CTRL cells the vast majority of Zfp36l2 mRNA had a surprisingly short poly(A) tail of just 20–30 nt. In the PARN KD cells the amount of Zfp36l2 mRNA with a long poly(A) tail of up to ∼190 nt was two to three fold more than in the CTRL cells. This was not a general effect on all mRNAs as the β-Actin (Actb) mRNA showed no difference in poly(A) tail length between the two cell lines (Figure S1). Although abundance of Zfp36l2 mRNA was similar in CTRL and PARN KD cells (Figure 3C), western blotting (Figure 3D) demonstrated a small increase in abundance of ZFP36L2 protein which would be consistent with enhanced translation resulting from the elongation of the poly(A) tail. We also saw evidence for increased abundance of ZFP36L2 protein by immunofluorescence (Figure S2).
In order to determine whether the effects of PARN on the Zfp36l2 mRNA are mediated by sequences in the 3′UTR we cloned the 3′UTR into a luciferase reporter construct (Luc-36L2) and measured luciferase activity following transfection into CTRL and PARN KD cells. The empty vector (Luc) was used as a control and gave very similar activity regardless of whether expressed in the CTRL or PARN KD cells (Figure 4A). Interestingly, the Luc-36L2 reporter produced significantly less luciferase activity than the Luc reporter in the control cell line suggesting that the sequences contained therein either repress translation or promote decay of the reporter mRNA. Importantly, PARN KD cells reproducibly exhibited a two-fold higher luciferase activity than the control cells (Figure 4A) and this was also seen when PARN was knocked down with a different shRNA (Figure S3) showing that this effect is PARN-specific. Interestingly, the clear increase in luciferase activity following PARN depletion is mediated predominantly by enhanced translation as there was little effect on abundance of either reporter mRNA in PARN KD cells (Figure 4B). The increase in luciferase expression is in the same range as the increase in abundance of endogenous ZFP36L2 protein in PARN KD cells (Figure 3D). Together these results indicate that the action of PARN on the Luc-36L2 reporter results in translation repression presumably through poly(A) shortening. Moreover, factors associated specifically with the 3′UTR of Zfp36l2 mRNA are likely responsible for the effects of PARN on Zfp36l2 gene expression. At this time we do not know what factor might be responsible for recruiting PARN, but the Zfp36l2 3′UTR does have AU-rich elements like those reported to bind proteins such as TTP/ZFP36; a protein that induces PARN-mediated deadenylation in vitro [17].
The relatively small number of mRNAs affected by PARN at the level of mRNA stability precluded a meaningful analysis of Gene Ontology (GO) terms or sequences that might impacted by reduced PARN activity. Still, we did note that several of the stabilized transcripts encode proteins with roles in mRNA metabolism (Toe1/Caf1z, Edc3, Zfp36l2, Dgcr14, Nufip1) and transcription (Gata2, Zfp219, Klf14) indicating that PARN may influence gene expression at multiple levels and impact a wider range of genes. To investigate this possibility we used the 0 minute time point from the array experiments to estimate global mRNA abundances in CTRL and PARN KD cells. We found that of the 18,201 transcripts detected, 1199 showed a 1.5-fold or greater change in mRNA abundance in PARN KD cells (Dataset S2). Surprisingly, given that PARN KD was expected to increase expression of its target mRNAs, the majority (63.7%) of the affected mRNAs were down-regulated. We verified the abundance changes for several transcripts by qRT-PCR and found that of 14 mRNAs examined, all but one (Lama2) showed changes similar to those predicted by the array (Figure 5A). Moreover, there was generally a good correlation between the change predicted by the microarray and that observed by qRT-PCR in untreated cells although the qRT-PCR indicated changes of a greater magnitude than the array (Figure S4). This confirms that Act-D treatment did not globally affect our mRNA abundance measurements and that the 0 minute time point mRNA abundances are generally an acceptable indicator of relative differences in mRNA abundance between PARN KD and CTRL cell lines.
We next took advantage of the availability of both mRNA abundance and decay data to analyze the impact of changes in mRNA stability on overall mRNA levels. We were surprised to discover that for the 40 transcripts showing clear evidence for stabilization following PARN knockdown, there was generally only a small effect on mRNA abundance and in many cases abundance was reduced rather than increased (Figure 5B). There was a similar inverse correlation for the mRNAs that were destabilized (Figure 5B). In order to verify this observation, we measured the abundance of three transcripts that were stabilized by PARN depletion in proliferating myoblasts (Figure 5C). Interestingly, Adora2b mRNA (1.4-fold stabilized (Figure 5D and Figure 2A)) showed ∼2-fold reduced abundance in PARN KD cells, while Ankrd54d mRNA (1.85-fold stabilized) showed no statistically significant change in abundance (Figure 5C). In contrast, Gpsm1 mRNA (1.96-fold stabilized) did show a small increase in abundance by this assay. As described earlier (Figure 3C), there was no significant change in abundance of the Zfp36l2 transcript despite a ∼2.4-fold increase in stability. Taken together, these results strongly suggest the existence of coupling between transcription and decay for many transcripts such that changes in mRNA decay rate are compensated for by opposing effects on transcription [20]–[22].
In order to further support this idea we assessed the abundance of newly transcribed pre-mRNAs for each of the four stabilized transcripts. Briefly, C2C12 cells were labeled for a short time with 4-thiouridine (4sU) and total RNA was prepared. Newly transcribed 4sU-labeled RNAs were biotinylated and isolated on streptavidin beads. Pre-mRNAs were detected and quantified by qRT-PCR using one intronic primer and one exonic primer. As shown in Figure 5E, all four pre-mRNAs exhibited significantly reduced abundance in the PARN KD cells, consistent with slower transcription rates for these transcripts in this cell line. To summarize, each of the four mRNAs we evaluated showed increased stability following PARN KD (Figure 5D) but reduced levels of pre-mRNAs indicating reduced transcription (Figure 5E). This change in the relative balance of decay and transcription results in only small changes in mRNA abundance (Figure 5C).
GO analysis using DAVID [24] revealed that the transcripts whose expression was most affected by PARN shared some interesting features (Tables S2 and S3). In particular, amongst the down-regulated genes there was a significant enrichment of mRNAs encoding proteins required for blood vessel development, cell adhesion, cell motion and axon guidance (Table S2). This is supported by the observation that a large proportion (∼15%) of the down-regulated mRNAs encoded extracellular proteins including several collagens (Col1a1, Col1a2, Col6a1, Col6a2, Col3a1, Col12a1), biglycan (Bgn) and matrix metalloproteases (Mmp19, Mmp2). In contrast, the up-regulated mRNAs were more likely to encode components of large ribonucleoprotein complexes such as the ribosome and spliceosome (Table S3).
Our GO analysis suggested that PARN knockdown might influence cell motility as cell movement requires extensive interactions with the extracellular matrix and is required for processes such as axon guidance and blood vessel development. We used a wound healing assay to investigate the ability of CTRL and PARN KD cells to migrate. Briefly, CTRL and PARN KD cells were grown to near confluence and then deprived of serum to prevent cell division. The monolayer was scratched to remove cells and incubated for eight hours to permit cells to migrate into the wound. Wound healing was assessed by counting the number of cells present within the boundaries of the wound. There was a clear increase in the wound healing capacity of PARN KD cells compared to the CTRL cells (Figure 6A and 6B) indicating that PARN KD cells migrate more rapidly. Moreover wound healing was restored to near normal levels following transfection of a plasmid encoding human PARN (Figure 6C). We conclude that PARN modulates processes required for cell motility in C2C12 myoblasts.
In this study we identified a set of mRNAs whose decay is dependent on PARN deadenylase. For one stabilized mRNA, Zfp36l2, we demonstrated that PARN-dependent regulation is mediated through sequences in the 3′UTR and that poly(A) tail length is increased following PARN KD. Depletion of PARN leads to increased ZFP36L2 protein abundance, but has negligible effects on mRNA abundance. Unexpectedly, we found that for the majority of affected transcripts mRNA stabilization slightly reduces mRNA abundance suggesting that mRNA decay rates are coupled to transcription. Finally, abundance of mRNAs encoding extracellular factors required for cell motility and adhesion was decreased by PARN knockdown and this observation led to the discovery that PARN KD cells migrate significantly faster than control cells in wound healing assays.
To our knowledge, ours is the first study to examine the role of the PARN deadenylase in mammalian cells, and the first to examine the global impact of depletion of an mRNA decay enzyme on mRNA decay rates. Our results suggest that while PARN directly impacts decay of relatively few transcripts, it has surprisingly wide-ranging effects on expression of over 1000 genes. This could reflect that some of the genes directly regulated by PARN have important roles in regulating transcription and other cellular processes, generating a knock-on effect. In addition, PARN-mediated deadenylation also clearly regulates translation efficiency (Figure 4), perhaps in some cases without altering mRNA decay rates. Any mRNAs whose poly(A) tail length is increased without a dramatic change in mRNA decay rate in PARN KD cells would not be detected by our analysis, although downstream effects of such regulation could be picked up as mRNA abundance changes. PARN is known to induce reversible deadenylation as a means to silence translation [16], however further experimentation will be required to distinguish targets regulated in this manner.
Despite the fact that poly(A) shortening is thought to enhance decay of mRNAs, we detected 24 mRNAs that were actually less stable following PARN KD. Some of these may be direct targets; perhaps when PARN is depleted a more aggressive decay pathway substitutes. However, given the wide-ranging effects of PARN KD on gene expression, we feel it is more likely that destabilization is an indirect effect of the PARN KD mediated by a factor(s) encoded by one of the stabilized transcripts (such as ZFP36L2, CAF1Z/TOE1 or EDC3). It also remains possible that some of these mRNAs are destabilized through off-target effects of the shRNA used to deplete PARN. Future experiments will aim to distinguish between these possibilities.
PARN KD stabilizes 40 of the 1389 mRNAs (2.9%) that we generated half-lives for in both cell lines. Remarkably, stabilization resulted in a decrease in abundance, or no significant change in abundance for the majority of these affected mRNAs (Table 1, Figure 3C, Figure 5B and 5C). This was seen in both the microarray and the qRT-PCR analyses. Although this seems counterintuitive, our results are actually very similar to recent observations on mRNA stability and abundance made in two closely related yeast strains [22]. These authors determined that as many as half of the evolutionary changes in mRNA degradation rates between S. cerevisiae and S. paradoxus were coupled to opposing changes in transcription rates. It was suggested that such coupling facilitates transient responses to environmental stimuli by enabling a more rapid return to basal expression levels. Additional studies, also in yeast, have established that mRNA decay rates are dependent on events that occur at the promoter [19]–[21] demonstrating that communication between transcription and mRNA turnover pathways exists. In yeast, some of this coupling has been attributed to the Rpb4/7 subunits of RNA polymerase II and to the CCR4-NOT complex, each of which have roles in both transcription and mRNA decay [19], [22]. Our data strongly imply that the cell attempts to compensate for loss of PARN by reducing transcription to maintain appropriate mRNA abundance. Interestingly, many stabilized transcripts appear to have slightly decreased abundance in PARN KD cells (Figure 5B). This overshoot suggests that the feedback mechanism is perhaps not highly accurate, or that it may additionally compensate for increased translation efficiency. Further investigation will be required to understand the mechanisms behind this feedback as so far there is no evidence that PARN modulates transcription directly, although it has been linked with mRNA 3′end processing events [10]. It is important to note that if extensive coupling of transcription and decay exists then mRNA abundance should be considered a poor indicator of both the magnitude and direction of effects on mRNA stability. This may explain why an earlier study found that depletion of CNOT6 and CNOT7 deadenylases had a relatively minor impact on overall gene expression [18].
Cell movement requires tightly controlled interactions between the cell and the ECM coupled with dynamic changes in the cytoskeleton. It can be described in three basic phases: Protrusion of the leading edge, adherence of the leading edge to the substrate and detachment of the cell body and trailing edge from the substrate [25]. Increased migration can be achieved by increased protrusion rate, by more efficient adherence to the substrate or by more rapid detachment from the substrate. While protrusion rate is primarily dependent on cytoskeletal dynamics, adherence and attachment can be influenced by cellular proteases or by the composition of the ECM. In general, those cell types that migrate most rapidly, such as leukocytes, have weaker interactions with the substrate whereas fibroblasts and myoblasts have stronger contacts and move more slowly [26]. Thus, the increased motility of PARN KD cells could be a direct result of alterations in the ECM caused by down-regulation of collagens, biglycan and other ECM components. Alternatively, increased motility could be due to altered expression of intracellular factors, such as TRIP10 and CDK5R1. The Trip10/Cip4 mRNA (stabilized 1.44-fold by PARN KD) encodes Cdc42-interacting protein 4 which is localized to the leading edge of migrating cells and directly enhances cell motility through regulation of the actin cytoskeleton [27]. The Cdk5r1 mRNA (stabilized 1.42-fold by PARN KD) encodes p35, an activator of the CDK5 kinase required for cell migration in neuroblastoma cells [28] and for myogenic differentiation [29]. Interestingly Cdk5r1 mRNA is subject to extensive post-transcriptional control through both miRNA- and ARE-mediated mechanisms [28], [30]. Thus, increased expression of either CDK5R1 or TRIP10 proteins might directly enhance cell migration in PARN KD cells.
The fact that PARN KD enhances wound healing and cellular motility is interesting in light of previous observations that depletion of several RNA-binding proteins can affect wound healing in C2C12 cells; depletion of hnRNPD/AUF1, ELAVL1/HuR or IGF2BP2 impaired wound healing capacity and motility [31]. Another RBP, the zipcode binding protein IGF2BP1, has been implicated in regulating localized expression of mRNAs involved in cell adhesion in breast cancer cells [32]. Finally, the Drosophila 5′-3′ exoribonuclease Pacman (XRN1 in mammals) is required for normal wound healing [33]. These results suggest that factors important for cell motility may be subject to extensive post-transcriptional control.
Future studies will aim to further characterize the phenotype of PARN KD cells in order to decipher the mechanism by which cell motility is affected. It will also be interesting to determine whether PARN acts primarily on nuclear or cytoplasmic mRNAs. Finally, a high priority for future research is to uncover the mechanism by which changes in mRNA decay rates are signaled to the nucleus to influence transcription and to determine whether this is a global phenomenon in mammalian cells.
The mouse C2C12 myoblast cell line was obtained from the American Type Culture Collection (CRL1772). Two derivatives of the C2C12 cell line, CTRL and PARN KD, were cultured in Dulbecco's Modified Eagle's Medium (DMEM) containing 10% Fetal Bovine Serum (FBS), 1 µg/ml puromycin, 10 U/ml penicillin and 10 µg/ml streptomycin in 5% CO2 at 37°C. Cells were maintained at or below 70% confluency except during wound healing assays. Transfections were performed using Lipofectamine 2000 as described previously [34] and transfection efficiency was routinely in the 50–70% range. The CTRL cell line stably transfected with LKO1 vector was described previously [23]. The PARN KD clonal cell line was generated by puromycin selection following transduction with an shRNA-encoding lentivirus derived from the LKO1 vector [35]. This vector is described in more detail below.
The half-life experiment, microarray hybridization and analysis were all performed as described previously [23]. Briefly, CTRL and PARN KD cells were treated with Act-D (8 µg/ml) for 30 minutes prior to the start of the time course. Total RNA was isolated at several time points using TRIzol (Invitrogen) according to the manufacturer's directions. 300 ng of total RNA were used to generate labeled cDNA fragments for hybridization to Mouse Affymetrix Gene 1.0 ST Arrays following the manufacturer's protocol (GeneChip WT cDNA Synthesis Kit #900652 and #900720). Production of probes and hybridization was performed by the Colorado State University Genomics and Proteomics Core Facility. Half-life experiments were conducted in triplicate, with each time point hybridized to a single array.
For normalization of probe sets, we utilized the GC-bin method for background correction and applied median normalization by Affymetrix Power Tools (APT) with the ‘no adjustment’ option. Then, all probe set values were normalized to the 5th percentile value of all probe sets on the same array. Transcripts whose probe sets gave detection above background (DABG) p-value<0.05 in at least two out of three replicates at the 0 minute time point were considered expressed and used for subsequent analyses. A nonlinear least squares model [36] was used to calculate half-lives using the microarray data. A half-life measurement was considered reliable if it met both the following criteria: (i) the microarray data had a good fit to the nonlinear least squares model (p-value<0.05) and (ii) the 95% confidence interval for the half-life was less than two times the half-life. Transcripts with reliable half-lives in at least two of three replicates were selected for further analyses. The mRNAs whose half-lives were significantly different in PARN KD compared to CTRL C2C12 cells were selected based on the t-test (p-value<0.05). The datasets were deposited in the GEO database (GSE35944).
For functional analysis, lists of Gene IDs for those transcripts affected >1.5-fold were uploaded to the Database for Annotation, Visualization and Integrated Discovery (DAVID; [24]) along with the list of Gene IDs for all detected transcripts as Background. Functional clustering analysis was used to identify enriched groups of Gene Ontology (GO) terms. Clusters with enrichment scores of less than 1.3 (equivalent to a p-value greater than 0.05) were excluded.
Total RNA was isolated using the TRIzol (Invitrogen) method as recommended by the manufacturer. All samples were treated with DNase 1 to remove genomic DNA. In experiments using samples from cells transfected with luciferase plasmids an additional step was employed to ensure effective removal of plasmid DNA. After the initial DNase treatment, RNA was treated with EcoRI and EcoRV to digest residual plasmid DNA and treated a second time with DNase 1. 1 µg of total RNA was reverse transcribed in the following conditions, according to the manufacturer's instructions: 35 mM Tris-Cl pH 8.3, 50 mM NaCl, 5 mM MgCl2, 5 mM DTT, 500 ng random hexamers, 10 U RNase Inhibitor, 1 µl Improm II Reverse Transcriptase (Promega). The resulting cDNA was used for qPCR with BioRad SYBR green supermix according to the manufacturer's instructions. A two-step amplification protocol was used in either a BioRad MyIQ, or a BioRad CFX96 instrument with annealing at 60°C for 30 seconds and extension at 95°C for 30 seconds for 40 cycles. mRNA abundances were normalized to the abundance of Gapdh mRNA except for experiments using 4-sU where 7SL RNA was used as a reference. Primer sequences are listed in Table S4.
The pLightSwitch_3UTR vector was purchased from SwitchGear Genomics. The 3′UTR of Zfp36l2 was PCR amplified from C2C12 myoblast cDNA using the following oligos (5′-GCTAGCCTCTCCATCTCCGACGACTG-3′ and 5′-CTCGAGTTGGGGGAAACTACAAAAC-3′). The resulting product was digested with Xho1 and Nhe1 and ligated into pLightSwitch_3UTR digested with the same enzymes to generate pLuc-36L2. The PARN expression clone bears the open reading frame of human PARN which was amplified using primers PARN1F (5′-CATGTCGACATGGAGATAATCAGGAGCAATTTT-3′ and PARN1R (5′-CATGGTACCTTACCATGTGTCAGGAACTTCAA-3′) and cloned between the Xho1 and Kpn1 sites of pcDNA3.1(-)(Invitrogen). It is not targeted by the murine PARN shRNA. The PARN targeting shRNA vector was generated by cloning annealed and kinased oligonucleotides (5′-CCGGGCGTGTGTGTTATTAACTAATCTCGAGATTAGTTAATAACACACACGCTTTTTG-3′ and 5′-AATTCAAAAAGCGTGTGTGTTATTAACTAATCTCGAGATTAGTTAATAACACACACGC-3′) into the Age1 and EcoR1 sites of the pLKO.1puro plasmid (a gift from R. Schneider; [35]). Oligonucleotide sequences were chosen from the Broad Institute's RNAi Consortium database (http://www.broadinstitute.org/rnai/trc). This particular shRNA targets the 3′UTR of murine PARN. The template used to generate the 5S rRNA probe was previously described [37]. In order to generate templates for probes against Zfp36l2 and Actb mRNAs, total RNA was isolated from proliferating C2C12 cells, and the poly(A) tails were removed by RNase H treatment in the presence of oligo(dT)18. An RNA linker (Integrated DNA Technologies, Linker 3) was ligated to the 3′ ends of the RNAs using T4 RNA ligase treatment as described previously [38]. Ligated RNAs were subjected to reverse transcription using a specific primer complementary to the RNA linker (for details see [38]). The resulting cDNA which corresponded to the 3′ ends of the Actb and Zfp36l2 mRNAs were then PCR amplified using the a primer complementary to the linker and an upstream oligo (ActB PAT 5′-CACTCCTAAGAGGAGGATGGTCGCGTC-3′ for actin and Zfp PAT 5′-CAGTTGGAGCACCGCGTGTG-3′ for Zfp36l2) and ligated into the pGemT-Easy vector (Promega). This process generated the pGemT-Zfp36l2 and pGemT-Actin plasmids which encode the 3′-terminal 300 nt of the Actb mRNA and the 3′-terminal 183 nt of the Zfp36l2 mRNA.
Whole cell lysate was prepared by lysis of cells in RIPA buffer (50 mM Tris-HCl (pH 7.4), 150 mM NaCl, 1.0% deoxycholate, 1% Triton X-100, 1 mM EDTA, and 0.1% SDS). 40 µg of each lysate was boiled in 6× protein loading buffer, resolved on a 10% SDS-polyacrylamide gel and blotted to PVDF membrane. PARN was detected using rabbit anti-sera (1∶20,000) [14]. ZFP36L2 was detected using rabbit polyclonal antibodies (Genway GWB-C5FC76). GAPDH (Chemicon mAB374) or Tubulin (Sigma Aldrich T5168) were used as loading controls (1∶20,000). Results were visualized using a BioRad Chemidoc system and quantified using QuantityOne software (BioRad). Reported values are a measure of the pixel density of the band of interest relative to the pixel density of the loading control (GAPDH or Tubulin). These ratios were normalized relative to control samples. Reported uncertainties are standard deviations.
10 µg of total RNA was incubated with 2 µM DNA oligo (ActB RNH 5′-AAGCAATGCTGTCACCTTCC-3′ for actin and Zfp RNH 5′-CGCGGTGCTCCAACTGTACCTA-3′ for Zfp36l2), heated to 95°C for three minutes and slow cooled to 4°C over a period of 30 minutes. RNaseH (7 units) and RNase Inhibitor (20 units) were added in the supplied reaction buffer (Fermentas Cat# EN0201). For generating poly(A) tail minus (A0) controls 100 ng/µl of oligo(dT)18 was included. Reactions were incubated at 37°C for 30 minutes. RNAs were then resolved on a 5% denaturing polyacrylamide gel (7 M urea, 1× TBE), and electroblotted to nylon membrane (Hybond-XL GE Healthcare) at 700 mA for 30 minutes in 1× TBE. Nucleic acids were immobilized by UV-crosslinking (Stratalinker). Membranes were pre-hybridized for 1 hour at 60°C in 25 ml hybridization buffer (50% formamide, 750 mM NaCl, 75 mM sodium citrate, 1% SDS, 0.1 mg/ml salmon sperm DNA, 1 mg/mL polyvinylpyrrolidone, 1 mg/mL ficoll, 1 mg/mL bovine serum albumin (BSA)). Membranes were then hybridized to radio-labeled RNA probe overnight at 60°C also in hybridization buffer. Blots were washed two times in 25 ml non-stringent wash buffer (0.1% SDS, 300 mM NaCl, 30 mM sodium citrate) and two times in 25 ml stringent wash buffer (0.1% SDS, 30 mM NaCl, 3 mM sodium citrate) for 20 minutes each time at 60°C. Membranes were exposed to storage phosphor screens and imaged on the Typhoon Trio Imager (GE Healthcare). Results were analyzed using ImageQuant software (GE Healthcare). α32P-labeled RNA probes were generated by in vitro transcription reactions as described below.
Internally radio-labeled RNAs were generated by in vitro transcription reactions (20 U T7 or SP-6 RNA polymerase, 10 U RNase inhibitor, 40 mM Tris pH 7.9, 6 mM MgCl2, 10 mM DTT, 10 mM NaCl, 2 mM spermidine, 500 µM ATP, GTP, CTP, 50 µM UTP and [α-32P]-UTP(4.5 µCi/µl), 716 Ci/mmol) were carried out for 3 hours at 37°C using 1 µg of linearized plasmid DNA as template. For the RNase H/northern blot probes, the pGemT-Zfp36l2 construct was linearized with SpeI and transcribed with T7 RNA polymerase. The pGemT-Actin construct was linearized with SacII and transcribed with SP6 RNA polymerase. Transcription products were separated on a 5% polyacrylamide gel containing 7 M urea, excised and eluted overnight in 400 mM NaCl, 50 mM Tris-Cl pH 7.5, and 0.1% SDS at 22°C. RNA was precipitated and resuspended in H2O.
C2C12 cells or PARN KD cells were transfected with a mixture of pEGFP-N1 (Clontech) and either Luc, or Luc36L2 plasmids. After 24 hours, the cells were trypsinized and collected in PBS. Coelenterazine (Promega) was added to a final concentration of 3 µM. Luciferase activities were measured in a Turner TD-20e Luminometer. Error bars represent pooled standard deviations derived from at least three independent experiments.
Proliferating cultures of C2C12 myoblasts were treated with 4-thiouridine (200 µM; SIGMA) for 15 minutes. Following this labeling period, cells were collected in TRIzol and RNA was isolated according to the manufacturer's recommendation. Biotinylation and fractionation of RNAs was performed as described previously [39]. Briefly, this involved incubating 50 µg of total RNA with 100 µg of Biotin-HPDP in 100 mM Tris-Cl (pH 7.4) in the presence of 1 mM EDTA for 2 hours in the dark. An equal volume of chloroform and isoamyl alcohol (24∶1) was added to the biotinylation reaction and transferred to a Phase-Lock Gel tube (5 Prime), mixed by inversion, and centrifuged at full speed for 10 minutes at 4°C. RNA was precipitated in an equal volume of isopropanol in the presence of 0.5 M NaCl. The pellet was washed in 70% ethanol, resuspended in T.E. (10 mM Tris-Cl (pH 7.4) 1 mM EDTA), warmed to 65°C for 10 minutes and snap chilled on ice. RNA was mixed with an equal volume of streptavidin magnetic beads for 15 minutes at room temperature and loaded on an equilibrated MACS® Separation Column (Miltenyi Biotechnology). The beads were washed four times with 200 µl of wash buffer (100 mM Tris-Cl (pH 7.4) 10 mM EDTA, 1 M NaCl, and 0.1% Tween-20) at 65°C and twice more at room temperature. Labeled RNA was eluted with 100 mM DTT. Eluted RNA was precipitated as described above, washed and resuspended in 20 µl of ddH2O. 40 ng of eluted RNA was reversed transcribed using random hexamers. The resulting cDNA was used in qPCR to assess levels of pre-mRNA using primer pairs in which the reverse primer was complementary to an exon and the forward primer matched a region within the upstream intron. The pre-mRNA abundance was normalized to that of 7SL RNA.
C2C12 myoblasts (1.5×105 cells) were seeded in 12-well dishes in growth media. Once cultures approached confluency, the monolayer was scratched with a 200 µl pipette tip. Cultures were washed with PBS, switched to low serum growth media (0.1% FBS), and imaged. After eight hours, cultures were imaged again. Margins of the scratch area were determined, and the number of cells migrating to the vacated area counted and graphed. Error bars represent the standard deviation from three experiments.
|
10.1371/journal.pmed.1002108 | Pancreatic Cancer Surgical Resection Margins: Molecular Assessment by Mass Spectrometry Imaging | Surgical resection with microscopically negative margins remains the main curative option for pancreatic cancer; however, in practice intraoperative delineation of resection margins is challenging. Ambient mass spectrometry imaging has emerged as a powerful technique for chemical imaging and real-time diagnosis of tissue samples. We applied an approach combining desorption electrospray ionization mass spectrometry imaging (DESI-MSI) with the least absolute shrinkage and selection operator (Lasso) statistical method to diagnose pancreatic tissue sections and prospectively evaluate surgical resection margins from pancreatic cancer surgery.
Our methodology was developed and tested using 63 banked pancreatic cancer samples and 65 samples (tumor and specimen margins) collected prospectively during 32 pancreatectomies from February 27, 2013, to January 16, 2015. In total, mass spectra for 254,235 individual pixels were evaluated. When cross-validation was employed in the training set of samples, 98.1% agreement with histopathology was obtained. Using an independent set of samples, 98.6% agreement was achieved. We used a statistical approach to evaluate 177,727 mass spectra from samples with complex, mixed histology, achieving an agreement of 81%. The developed method showed agreement with frozen section evaluation of specimen margins in 24 of 32 surgical cases prospectively evaluated. In the remaining eight patients, margins were found to be positive by DESI-MSI/Lasso, but negative by frozen section analysis. The median overall survival after resection was only 10 mo for these eight patients as opposed to 26 mo for patients with negative margins by both techniques. This observation suggests that our method (as opposed to the standard method to date) was able to detect tumor involvement at the margin in patients who developed early recurrence. Nonetheless, a larger cohort of samples is needed to validate the findings described in this study. Careful evaluation of the long-term benefits to patients of the use of DESI-MSI for surgical margin evaluation is also needed to determine its value in clinical practice.
Our findings provide evidence that the molecular information obtained by DESI-MSI/Lasso from pancreatic tissue samples has the potential to transform the evaluation of surgical specimens. With further development, we believe the described methodology could be routinely used for intraoperative surgical margin assessment of pancreatic cancer.
| Ambient ionization mass spectrometry imaging can provide accurate diagnostic information differentiating cancerous from noncancerous tissue samples and has been recently shown to be particularly powerful in helping pathologists and surgeons determine whether cancer reaches the edge (margin) of the resection specimen in real time during surgery.
This study was performed to evaluate the feasibility and efficacy of this method during surgery for pancreatic cancer, one of the most lethal human cancers.
Our methodology was developed and tested using 63 banked pancreatic cancer samples and 65 samples (tumor and specimen margins) collected prospectively during 32 pancreatectomies performed from 2013 to 2015.
We found that desorption electrospray ionization mass spectrometry imaging (DESI-MSI) allows discrimination of normal pancreatic and pancreatic cancer tissue based on diagnostic metabolic signatures and has the potential to assist in surgical decision making by informing the surgeon whether the entire tumor has been removed or not.
These findings provide novel information on molecular markers of pancreatic cancer and showcase the value of this methodology as an adjunct to the current pathologic method (frozen section analysis) for determining the completeness of cancer surgery.
The data reported in this study could be made available during actual surgery, and allow the surgeon to extend the boundaries of surgery to remove any residual tumor discovered by DESI-MSI at the margin.
| Resection of pancreatic cancer is a complex and technically demanding surgical procedure due to the location of the pancreas adjacent to many critical organs and vascular structures, and the locally infiltrative nature of pancreatic adenocarcinoma. Surgical resection of pancreatic cancer remains the main curative option for this disease. It is well accepted that complete resection with negative surgical margins is associated with improved long-term survival [1–4]. This finding calls for careful evaluation of surgical resection margins intraoperatively. Positive margins, defined as the presence of tumor cells at the specimen edge and/or the resection bed, have been associated with increased local recurrence and decreased overall survival [1]. During surgical resection, depending on the location of the tumor (in the head or body/tail of the gland), up to five surgical margins are typically evaluated by an expert team of pathologists using histologic analysis of frozen sections. These margins include the pancreatic neck margin, the retroperitoneal/uncinate process margin, the vascular groove margin, the gastric/proximal duodenal margin, and the bile duct margin. However, in practice, the accuracy of intraoperative delineation of pancreatectomy margins can be variable and subjective, with false-negative results occurring in up to 20%–30% of pancreatic adenocarcinoma patients [5–7]. Consequently, the need exists for developing more accurate, more time-efficient, and less operator-dependent technologies to evaluate specimen margins during pancreatic cancer surgery [8]. We have recently described an approach combining desorption electrospray ionization mass spectrometry imaging (DESI-MSI) with the least absolute shrinkage and selection operator (Lasso) statistical method to classify tissue sections and evaluate surgical resection margins during gastric cancer surgery [9]. We describe herein the application of this methodology to the intraoperative evaluation of pancreatic cancer surgical margins.
In the last few years, ambient ionization mass spectrometry (MS) has become a powerful approach for tissue imaging and diagnosis [10–13]. In particular, DESI-MSI is the most extensively used ambient ionization MS technique for chemical imaging and diagnosis of tissue samples [13]. In DESI-MSI, a spray of charged droplets is directed toward a tissue sample, allowing chemicals to be dissolved at the sample surface, ionized by mechanisms similar to electrospray ionization, and transferred into a mass spectrometer, where the mass-to-charge ratios (m/z) of molecular ions and their abundances are measured [14]. By rastering the tissue sample underneath the DESI-MSI spray spot and collecting a mass spectrum at every point, it is possible to determine the distribution of numerous molecular species with a typical spatial resolution of 200 μm at an acquisition rate of 0.5 s/pixel. Besides DESI-MSI, other ambient ionization techniques including probe electrospray ionization [15], solid-probe-assisted nanoelectrospray ionization [16], touch spray [17], and rapid evaporative ionization MS [18] have been used for cancer tissue diagnosis and surgical margin evaluation [13]. Various human cancer tissues have been investigated by ambient ionization MS, including liver [19], breast [20,21], brain [22], kidney [23], prostate [24,25], bladder [26], gastric [25], colorectal [25], and ovarian [26] cancers. In this study, we tested the usefulness of DESI-MSI to classify pancreatic tissue as cancerous or benign and to evaluate surgical margins collected from pancreatic cancer surgery.
Banked tissue samples were obtained without identifiable or clinical information. Therefore, the Stanford institutional review board (IRB) determined that this portion of the research did not include human subject research and was exempt from full review. However, Stanford’s IRB ethical review committee determined that the portion of the study involving patients was human subject research, and thus IRB approval was obtained. Approval was also obtained from Stanford’s Cancer Institute Scientific Review Committee. Written informed consent was obtained for all patients recruited.
Sixty-three frozen human tissue specimens including pancreatic ductal carcinoma and benign pancreatic tissue were obtained from the Stanford Tissue Procurement Facility under approved IRB protocol. Samples were stored in a −80°C freezer until sectioned (15 μm thick) using a Leica CM1950 cryostat (Leica Microsystems). No sample size determination was done for this study. After sectioning, the glass slides were stored in a −80°C freezer. Prior to mass spectrometry imaging, the glass slides were dried in a desiccator for approximately 15 min.
Thirty-two pancreatic cancer patients scheduled to undergo pancreatectomy at Stanford University Hospital were preoperatively consented for our study. All patients gave written consent under IRB approval (protocol number 25655), approved by Stanford’s IRB committee (IRB 7). Inclusion criteria were that the patient was scheduled to undergo surgery for pancreatic cancer removal. No exclusion criteria were applied. During surgery, the neck and/or retroperitoneal/uncinate margins of the specimen were subjected to frozen section histopathologic evaluation, as is routinely performed independently of our research. In parallel to the process of frozen section, adjacent 5- and 15-μm thick tissue sections of each margin were obtained for DESI-MSI. For all but one case, a sample of the tumor was obtained in addition to the surgical margins. In total, 65 surgical tissue samples were evaluated by DESI-MSI during the period from February 27, 2013, to January 16, 2015.
The MS and histologic analyses reported were chosen during monthly meetings between the leading investigators or a subgroup of the investigators. The analysis was planned a priori and was based on a recent successful investigation that the authors had performed in gastric cancer. The statistical methods were optimized during the study based on the results obtained and the need for refinement of the methods.
A 2-D DESI-MSI source (Prosolia) coupled to an LTQ XL mass spectrometer (Thermo Scientific) was used for tissue imaging. DESI-MSI was performed in the negative ion mode from m/z 90 to m/z 1,200. The spatial resolution of the imaging experiments was 200 μm. The histologically compatible solvent system, dimethylformamide:acetonitrile 1:1 (v/v), was used for analysis at a flow rate of 0.8 μl/min. The N2 pressure was set to 175 psi. After DESI-MSI, the same tissue section was subjected to H&E staining for histopathologic evaluation by expert pathologist T. A. L. in a blind manner. For ion identification, tandem MS analyses were performed.
DESI-MSI data were collected on entire tissue sections. After DESI-MSI, the slides were stained and evaluated by the pathologist. Regions including lymphocytes, inflammation, and necrosis were observed in part of the samples, but were not selected within the pixels analyzed by statistical analysis because the goal of this study was to identify and discriminate cancer from normal pancreatic tissue, and not to characterize these other histologic features. DESI-MSI is performed as a sequence of line scans, and, thus, regions of glass slide (no tissue sample) are also analyzed. Pixels corresponding to the non-tissue area were imaged but not included in the pixels selected.
The same tissue sections analyzed by DESI-MSI were subjected afterward to standard H&E staining protocol. These sections were adjacent and serial to, but not the exact same, sections used during surgery for surgical margin assessment. Pathologic evaluation was performed using light microscopy. Regions of benign pancreatic glands, normal stroma, and pancreatic cancer were delineated on the glass slides. Tissues that contained regions of mixed histology were evaluated and given a percent composition for each cell/tissue type (benign glands, normal stroma or pancreatic cancer).
The 2-D raw data obtained by DESI-MSI were converted to text files and imported to the R package for statistical analysis. The images were plotted in R and manually segmented into regions of interest as determined by histopathologic evaluation. Intensities for a total of 13,320 m/z values were recorded in each spectrum. To reduce complexity and account for small differences in registration between spectra, these MS features were averaged in nonoverlapping bins of six m/z values to yield a total of 2,220 features per spectrum. We randomly divided the patients into one training set and two sets of test samples. Within the training set, we applied the Lasso method (multiclass logistic regression with L1 penalty) using the glmnet 2.0–2 package in the CRAN R language library [27].
The Lasso is a shrinkage and selection method for supervised learning. It minimizes the usual sum of squared errors (or negative log-likelihood) with a bound on the sum of the absolute values of the coefficients. As a result, it yields a “sparse” solution containing the most informative features for the prediction task, that is, models that involve only a subset of the variables/predictors [28]. As such, it has an advantage over methods such as support vector machines, which are not designed to yield sparse solutions. On the other hand, if interactions between features are important, methods such as random forests and boosting may yield better results. (We tried these approaches here, and they did not offer improvement over the Lasso method.)
In this application, the Lasso method yields a model with parsimonious sets of features for discriminating between pancreatic cancer, normal pancreatic glands, and normal pancreatic stromal tissue. A mathematical weight for each statistically informative feature is calculated by the Lasso depending on the importance that the mass spectral feature has in characterizing a certain class. Features that do not contribute to a class of the linear model receive a weight of zero and are disregarded. An ion whose peak height, or abundance, is important for characterizing a certain class is given a positive weight, whereas ions for which low abundance or their absence is important receive a negative weight. Because the features selected by the Lasso can occur at a valley or a shoulder of an actual peak in the mass spectrum, identification of the selected features was performed by characterizing the nearest mass peak to the statistically selected feature.
Classification was done on a pixel by pixel basis into one of three classes: (1) benign glands, (2) cancer, or (3) stroma. We employed 25-fold cross-validation (CV), leaving out one patient at a time, to select the Lasso tuning parameter and to assess the predictive accuracy within the training set. Then, the chosen model was applied to the first test set of 15 patients, from which samples were of clear diagnosis. We then applied this method and an improved approach using customized training sets [29] to an independent set of mixed samples (including banked and surgical samples).
Note that the prediction performance is of main interest when evaluating the performance of the statistical classifier. p-Values, which are commonly used to evaluate the statistical significance of defined results between groups, are not the central focus in classification approaches with a large number of features, as in our study, and thus are not applicable to our results. Survival probabilities were calculated based on the Kaplan-Meier method and compared using the log-rank test.
A total of 128 human pancreatic tissues including banked specimens and tissue prospectively collected from surgeries for our study were analyzed in the negative ion mode by DESI-MSI. Fig 1 shows a flowchart of our study design. For the majority of the samples analyzed, the 2-D ion images obtained showed high heterogeneity in the distribution of molecular ions. Most of the heterogeneity in ion distribution was assigned by histopathologic evaluation as regions of cancer, normal pancreatic glands, or normal stromal tissue. Many samples presented highly mixed regions in which all of these histologic features were simultaneously observed. In a few samples, other histologic features were observed by the study pathologist including regions with acute inflammation, lymphocytes, and necrosis. These regions were not consistently seen across samples and were not considered in our approach owing to the lack of a statistically significant number of spectra.
The mass spectra obtained for regions of normal pancreatic glands, normal stroma, and cancerous tissue presented high relative abundances of many molecular ions commonly attributed to lipid species in negative ion mode DESI-MSI mass spectra of human tissue (Fig 2). Normal pancreatic glands and cancerous regions showed high relative abundances of low m/z ions (m/z 200–400) attributed to free fatty acids (FAs), in comparison to higher m/z ions (m/z 700–1,000) attributed to glycerophospholipids (GPs). In particular, benign glands presented higher relative and total abundances of free FAs and FA dimers commonly observed in the m/z 500–600 range compared to both cancerous and stromal tissues. In normal pancreatic glands, the most abundant free FAs observed were identified as oleic acid (m/z 281.2), palmitic acid (m/z 255.3), and arachidonic acid (m/z 303.3). Dimers of these species were observed at m/z 537.0 (oleic and palmitic) and m/z 563.0 (oleic and oleic), among others. In the higher m/z range, GPs of various classes were observed including glycerophosphoinositol (PI) 38:4 at m/z 885.6, PI(36:2) at m/z 861.5, PI(34:2) at m/z 833.5, glycerophosphoglycerol (PG) 36:3 at m/z 771.5, and glycerophosphoetanolamine 37:5 at m/z 750.5. Cancerous tissue presented high relative abundances of polyunsaturated FAs including arachidonic acid (m/z 303.3) and adrenic (16-docosatetraenoic) acid (m/z 331.2). In the higher m/z range, chloride adducts of glycerophosphocholines (PCs) including PC(34:1) at m/z 794.4 and PC(34:0) at m/z 792.4, and other deprotonated GPs such as PG(36:2) at m/z 773.6 and PI(38:4) at m/z 885.6, were observed. Stromal tissue presented a characteristic lipid profile with an overall lower total lipid abundance (total mass spectrum ion counts) compared to normal pancreatic glands and cancerous tissue, and showed high relative intensities of oleic acid, palmitic acid, glycerophosphoserine (PS) 36:1 at m/z 788.8, PS(38:1) at m/z 816.5, and PI(38:4) at m/z 885.6. Fig 3 shows selected 2-D DESI-MSI ion images of the identified ions for the samples PC7817, which is composed of 90% normal pancreatic glands and 10% normal stromal tissue, PC13702, which is composed of 90% pancreatic cancer and 10% normal stromal tissue, and PC0423, which is composed of purely normal stroma.
The large amount of molecular features obtained from the 254,235 pixels analyzed makes data interpretation difficult and calls for the use of sophisticated multivariate statistical techniques [9,30,31]. First, we implemented the approach we previously developed using the Lasso method to generate a statistical prediction based on DESI-MSI data [9,28]. To build our Lasso statistical classifier, a group of 42 samples with regions of clear histologic diagnosis (>90% composition of a single tissue type) was selected and randomly divided into a training set and a validation set of samples. Mixed samples with high cellular heterogeneity were evaluated separately as a test set of samples, and the results were individually analyzed for each sample. The training set of samples consisted of 25 samples with regions of pancreatic cancer, normal pancreatic glands, or normal stroma tissue, contributing a total of 45,273 spectra. Using the training set of samples, the Lasso selected a total of 112 m/z values that are important in characterizing all classes and that yielded the fewest CV errors (Table 1). From those, 59 different m/z values were selected by the classifier as important features to characterize pancreatic cancer, and 54 m/z values and 14 m/z values were selected as important features to characterize normal pancreatic glands and normal stroma, respectively. The statistical weight for each informative feature is calculated by the Lasso depending on the importance that the mass spectral feature has in characterizing a certain class. An ion whose relative abundance is important for characterizing a certain class is given a positive weight, whereas ions for which low relative abundance or their absence is important receive a negative weight. In this way, certain m/z values were selected as important for more than one class, with different statistical weights. For example, m/z 233.7 was selected as important for characterizing both normal glands and stroma tissues, with statistical weights of −0.146 and +0.983, respectively. On the other hand, m/z 738.8 was selected as important for characterizing pancreatic cancer and normal glands, with statistical weights of +0.130 and −0.148, respectively.
The Lasso yields a classifier that predicts whether a pixel belongs to a certain class based on the highest probability assigned of being pancreatic cancer, normal pancreatic glands, or normal stroma. To test our model using the training set of samples, we performed a 25-fold leave-one-patient-out CV and evaluated the agreement between the prediction obtained by Lasso and the diagnosis obtained by histopathologic evaluation of the same tissue section, which was H&E stained after being imaged by DESI-MSI [26]. An overall agreement of 98.1% was achieved when the total of 45,273 pixels were analyzed in CV. Table 2 shows the results obtained for each class. Note that the normal pancreatic gland class showed the highest agreement (99.3%) with pathologic evaluation, followed by the cancer class (96.4%) and the stroma class (79.9%). There are two key observations that could account for the lower agreement obtained for the stroma class compared to the other classes. First, only 1,389 pixels in our classifier corresponded to pure stroma pixels, a number significantly lower than what was obtained for the other two classes (34,014 pixels of normal pancreatic glands and 9,870 pixels of pancreatic cancer). Second, stromal tissue contained the least amount of detectable ions in the mass spectra, as observed in Fig 2C and, consequentially, the lowest number of statistically significant features was selected for characterizing this class (14 m/z values). Nevertheless, when normal pancreatic glands and normal stroma were combined into one class of normal pancreatic tissue, the overall agreement rate increased to 98.7% (Table 3). When we applied our classification model to a set of 17 independent samples with regions of clear histopathologic diagnosis, an overall agreement rate of 98.6% was achieved for the 31,235 pixels considered. The results obtained for each tissue class in the validation set are shown in Table 2. Note that, similarly to the training set of samples, the lowest agreement was observed for the stroma class (83.8%), compared to the agreement obtained for the normal pancreatic gland (99.8%) and cancer (95.4%) classes. When normal stroma and normal pancreatic glands were combined into one class, a high overall agreement of 98.9% was achieved for the independent set of validation samples (Table 3).
To spatially observe the results of our classification system, two-dimensional false-color images were plotted showing the results for cancer as red pixels, normal pancreatic glands as green pixels, and normal stroma as blue pixels. These images can be directly compared to the optical images of the H&E-stained tissues. Fig 4 shows the CV classification results obtained for samples PC5756 and PC699 from the training set and for samples PC14836 and PC13702 from the validation set. Note that discrepancies in the shapes of the Lasso predicted images and the DESI-MSI images occurred for some examples due to (1) the segmentation algorithm used to select regions for statistical analysis, (2) the exclusion of pixels near the boundaries of the tissue section where background ions from glass slide (which do not contribute to the molecular information) are observed at high intensities, and (3) the suboptimal aspect ratio of the predicted Lasso images, which required rescaling and adjusting. Optical images of the H&E-stained sections with pathologic diagnosis for these samples are shown in Fig 4. As observed, the discrepancies observed between statistical results and histopathologic evaluation occur for a low number of pixels within each sample. As our approach is performed and evaluated on a pixel by pixel basis, this is not surprising: regions with a predominant histologic feature may contain few cells of a different histologic class.
A large number of the samples analyzed in our study contained regions of high cellular heterogeneity, with a mixed composition of cancer cells infiltrating normal pancreatic glands and normal stromal cells. Careful histologic evaluation was performed in order to assign and accurately describe these regions with mixed histologic features. Regions of cellular composition not accounted for when building our statistical models (such as lymphocytes, acute inflammation, and necrosis) were excluded from our statistical prediction approach. To assist with the evaluation of the DESI-MSI/Lasso results, a visual measurement of the percent cell composition was given by the pathologist for selected regions of the tissue section. For example, sample PC13336 contained a region of pure normal pancreatic glands, adjacent to a region of 15% tumor cells and 85% normal pancreatic gland cells. Sample PC12809 presented a mixed composition of 20% tumor cells within 80% stromal cells. Sample PC14851 presented 60% tumor cells mixed within 40% normal pancreatic glands. Sample PC0411 presented a region with high tumor cell composition (70%) mixed with normal stroma (30%), surrounded by normal pancreatic glands and stroma. Sample PC14132 presented a region with 60% tumor cells intermingled with normal stroma, while the remaining tissue was composed of both normal pancreatic glands and normal stromal tissues. Optical images of the H&E-stained tissue sections delineated by histopathologic evaluation are shown in Fig 5.
Due to the complex nature of these samples, the results for each sample were evaluated separately in order to assess the performance of our approach. The agreement between the DESI-MSI/Lasso prediction and the diagnosis given by histopathology was evaluated both in terms of the spatial distribution of pixels and the percent tumor/normal cell composition for the specific tissue region. For example, an excellent agreement was observed for sample PC13336, with the region of normal pancreatic glands completely assigned as normal glands, and the adjacent region of 15% tumor cell concentration and 85% normal pancreatic gland concentration assigned by the statistical approach as 12% cancer pixels within mostly normal pancreatic glands (Fig 5). Another sample in which excellent tissue composition and spatial agreement was observed was PC0411. The region with high tumor cell concentration (70%) mixed with stroma (30%) as diagnosed by histopathology was correctly classified by our approach, as well as the surrounding region of normal pancreatic glands and stroma. On the other hand, sample PC14132 was classified as containing a higher tumor cell composition than that assessed by histopathology, with regions diagnosed as normal glands assigned as containing cancer. Overall, results obtained with the developed Lasso approach showed outstanding agreement with histopathology for both cell concentration and spatial distribution for 70% of the samples evaluated. For the remaining 30% of samples, the disagreement was caused by either an overprediction of cancer cell composition (20% of samples) or false-negative prediction (5% samples). Results for the samples described above are shown in Fig 5A.
An alternative approach commonly used to evaluate results from complex samples is the “majority rule” approach, where an overall agreement is given for an entire sample based on the majority of the pixels assigned (i.e., if >50% pixels are predicted as cancer, the sample is predicted as “cancer”). Using the majority rule approach, 95% agreement was achieved for the mixed samples using our classification system. Nevertheless, because the majority rule approach disregards important spatial features of the samples, which are crucial for surgical margin evaluation, we have chosen not to evaluate our results by this method. Instead, in an effort to optimize our methods for mixed samples, we applied a novel customized training statistical method using the Lasso recently developed for MS data [29]. This customized training strategy makes predictions on the test dataset when the features of the test data are available at the time of model fitting. The data are clustered to find training points close to each test point, and then the method fits a Lasso model separately in each training cluster. This means that a customized training set is generated from the data in the training set for each test sample. The method also generates a separate, unique list of statistically significant features (m/z values) for each test sample. This procedure, by incorporating an entire dataset of training samples, is useful in situations where complex datasets have underlying structure that could account for difficulties in prediction. When using the customized training approach, our agreement increased to 81% of all mixed samples analyzed. Increased agreement was mostly observed for samples that had previously been classified by our approach as having a higher tumor cell composition than what was assessed by histopathology, with many “false positive” samples now correctly classified as only normal tissue. For example, sample PC12809, a mixed composition of 20% tumor cells within 80% stromal cells, was classified as being composed of 81% cancer by our traditional Lasso approach. Using the customized training set approach, 15% of the pixels were classified as cancer and the remaining 85% as normal stroma, a much better agreement. Results for the samples discussed above using this new customized approach are shown in Fig 5B.
To evaluate the results for the surgical cases in our study, we used the customized training set approach described above. Thirty-two pancreatic cancer patients were recruited for our study, and a total of sixty-five samples were collected (Table 4). In most cases, serial sections of at least one of the margins (neck and/or uncinate) were obtained, as well as a section of the tumor. The results from DESI-MSI/Lasso analysis were not fed back to the surgeons during the procedure, but independently evaluated post-operatively. For surgical case PCP4, for example, a sample of the cancerous tissue as well as sections of the neck and uncinate margins were obtained during surgical resection and analyzed by DESI-MSI. Histologic evaluation of the surgical margins both intraoperatively and after DESI-MSI diagnosed both margins as negative for the presence of cancer. The uncinate margin was composed of 90% normal pancreatic glands and 10% stroma, while the neck margin was composed of normal pancreatic glands only. The tumor section contained 60% tumor cells, with the remaining being composed of normal pancreatic glands and stroma. Predictions were performed using the customized training set approach, and results are shown in Fig 6. As observed, a good agreement between histopathologic evaluation and our analysis was achieved. In particular, both margins were detected as being purely normal, with stroma pixels detected by our classifier for the neck margin, whereas 55% of pixels were detected as cancerous for the cancer section, as described by histopathology. Similar results were obtained for the majority of cases with negative margins, including PCP9 (Fig 6) and PCP31.
For case PCP21, samples of tumor and neck margin were obtained. While the neck margin was diagnosed as negative by histopathology (95% normal pancreatic glands and 5% normal stroma), the tumor section contained a region with 80% tumor cell concentration within stroma cells, and the remaining tissue was diagnosed as tumor infiltrating normal glands and stroma, with a low tumor cell concentration. Results obtained by our approach are shown in Fig 6. While high spatial and tumor/normal cell composition agreement was obtained for the cancerous tissue, a very small number of pixels (2%) within the normal neck margin were detected as being cancerous by our method. Note that this error is within the error rates we obtained when developing the statistical approach. Similar results were observed for other large margin samples (over 1,000 pixels) evaluated by our classifier, including the negative neck margins from surgical cases PCP20 and PCP28. Yet an excellent agreement was observed for the cancerous tissues for both surgical cases PCP20 and PCP28.
PCP14 was the only surgical case for which a positive margin was found intraoperatively for the uncinate margin. Histologic evaluation of the serial tissue section analyzed in the laboratory by DESI-MSI also detected the presence of tumor cells within the uncinate margin, about 10% tumor cell concentration infiltrating within normal glands and normal stroma. This positive margin was also detected by our DESI-MSI/Lasso approach, with an excellent agreement of 12% pixels detected as cancer among normal glands and stroma. Overall, good agreement between histopathologic evaluation and the DESI-MSI/Lasso results was obtained in our study for 24 of the surgical cases evaluated. In four of the remaining surgical cases, a false positive was observed in surgical margins, with as many as 20% of the pixels classified as cancer by our approach while diagnosed as normal by histopathology. In the other four cases, a maximum of 2% of pixels were classified as cancerous in surgical margin tissue while diagnosed as normal by histopathology. It was interesting to note that the median survival after resection for these eight patients with false-positive margins by DESI-MSI/Lasso but not by histopathologic examination was only 10 mo, as opposed to 26 mo for patients with negative margins by both DESI-MSI/Lasso and histopathology (Fig 7). Historically, a median survival of 10 mo is what one would expect from a margin-positive pancreatic cancer resection and 26 mo from a margin-negative resection. Although this difference did not reach statistical significance (p = 0.209), likely due to small sample size, it would be intriguing to hypothesize that our method was more “sensitive” than frozen section analysis (the standard method currently) in detecting margin involvement by tumor in patients who experienced early recurrence and death. Further study is warranted to prove or disprove the aforementioned hypothesis. False-negative results were not observed by our approach when evaluating surgical margins. Excellent agreement was observed for all the cancer tissues analyzed from surgical cases.
Accurate intraoperative evaluation of resection margins in pancreatic cancer (as with any oncologic) surgery is critical to overall surgical success and patient survival. In this study, we used DESI-MSI and the Lasso method to develop an automated system to classify pancreatic tissue sections based on molecular information. In total, 254,235 individual mass spectra were considered in our approach, and classified as normal pancreatic glands, normal pancreatic stromal tissue, or pancreatic cancer. DESI-MSI molecular profiles obtained for the samples showed high relative abundances of many ions identified as FAs and GPs. The method was developed and tested using different sets of training and validation samples, and its performance was evaluated on a per pixel basis in comparison to histopathologic diagnosis. Using a set of independent validation samples with unequivocal histologic features, classification results were in agreement with pathologic diagnosis in 98.6% of the pixels evaluated. The highest error value in classification was observed in the normal stroma class, which presented low abundances of molecular species by DESI-MSI compared to the other tissue classes, and for which the lowest number of samples/pixels was obtained. A set of complex validation samples with mixed histologic features was carefully evaluated using a statistical approach that employs Lasso to generate a customized training set for each test sample considered. The results obtained were methodically compared with the histopathologic results for both spatial features and cellular composition in each sample. When judged by both criteria, 81% agreement was obtained. This is the first study to our knowledge to report the use of ambient ionization MS imaging for the diagnosis of pancreatic cancer.
We further demonstrated the value of DESI-MSI/Lasso for surgical margin evaluation in pancreatic cancer surgery using neck and/or uncinate margins and tumor tissues prospectively collected from 32 surgical procedures performed at Stanford University Hospital. Samples were imaged and classified using the customized training set approach developed for mixed samples. Using our method, we were able to correctly diagnose cancer in a case where a positive neck margin was observed by histopathology. For the remaining cases, all margins were diagnosed as negative by histopathologic analysis. In nine of the 32 cases, our method classified pixels (1%–20%) in the neck and/or uncinate margin as cancerous while histopathology did not. This disagreement in diagnosis could be attributed to the inherent error range in our analysis (~2%), especially for margin samples that are large and contain over 1,000 pixels. However, the early recurrence and death noted in patients with false-positive margins by DESI-MSI raises the question as to whether these margins were truly positive, and accurately classified by our method but not by the histopathologic analysis of frozen sections.
The classification results obtained using Lasso are similar to what has been reported for other cancers (>90% accuracy in CV) [13]. Improvements in our classification system will be sought by increasing sample size for stroma tissue, which contributes to most of the confusion in our classification system. As DESI-MSI is performed in the ambient environment with minimal sample preparation requirements, we believe this technology is attractive for routine use in clinical practice. Furthermore, as DESI-MSI evaluation is performed in real time, it is typically faster than frozen section analysis. Note, however, that in this study DESI-MSI was performed on the entire tissue section in order to unambiguously correlate and compare the molecular results with histopathologic diagnosis. In some cases, over 2 h were necessary to image a large surgical margin tissue. Thus, the timeframe involved in imaging analysis by DESI-MSI could become a limitation for its routine use in the clinic. However, we expect that in clinical practice, DESI-MSI analysis would be rapidly performed in selected regions of the tissue that present diagnostic ambiguity (<1 s/pixel), in a profiling and not imaging mode. In this way, DESI-MSI would serve as a rapid adjunct technique to the standard method of frozen section analysis, and thereby enhance intraoperative margin assessment by the pathologist.
Further limitations of incorporating DESI-MSI into clinical practice include the required staffing resources and cost. Currently, data recording, interpretation, and statistical analysis are performed after sample analysis, which requires time and expertise. Improvements in computational methods for data acquisition and processing are necessary to successfully translate the technology to clinical practice. In addition, current commercially available mass spectrometers are costly instruments (>$200,000) that require regular maintenance. Yet, as many efforts are underway to develop smaller and cheaper mass spectrometers, we expect these instruments to become more accessible to hospitals for clinical use [32,33].
In summary, this study demonstrates that DESI-MSI/Lasso can be successfully used to classify tissue as normal or pancreatic cancer. Our findings provide evidence that the molecular information obtained by DESI-MSI/Lasso from tissue samples has the potential to transform the evaluation of surgical specimens. With further development and automation, we believe the described methodology could be routinely used for surgical margin assessment of pancreatic cancer. Yet, careful evaluation of the long-term benefits to patients of the use of DESI-MSI for surgical margin evaluation is needed, using a larger cohort of cases, to determine its proper value in clinical practice.
|
10.1371/journal.pbio.1002038 | Structure-Guided Design of Selective Epac1 and Epac2 Agonists | The second messenger cAMP is known to augment glucose-induced insulin secretion. However, its downstream targets in pancreatic β-cells have not been unequivocally determined. Therefore, we designed cAMP analogues by a structure-guided approach that act as Epac2-selective agonists both in vitro and in vivo. These analogues activate Epac2 about two orders of magnitude more potently than cAMP. The high potency arises from increased affinity as well as increased maximal activation. Crystallographic studies demonstrate that this is due to unique interactions. At least one of the Epac2-specific agonists, Sp-8-BnT-cAMPS (S-220), enhances glucose-induced insulin secretion in human pancreatic cells. Selective targeting of Epac2 is thus proven possible and may be an option in diabetes treatment.
| cAMP is a small molecule produced by cells that activates proteins involved in a wide range of biological processes, including olfaction, pacemaker activity, regulation of gene expression, insulin secretion, and many others. In the case of insulin secretion, cAMP seems to impinge on different stages of the signalling cascade to regulate secretory activity in pancreatic β-cells. Here we have developed a chemically modified version of cAMP that specifically only activates Epac2, one of the cAMP-responsive proteins in this cascade. Furthermore, our cAMP analogue activates Epac2 more potently than cAMP itself does. We have determined several crystal structures of Epac2 in complex with cAMP analogues to help us explain the molecular basis of the observed selectivity and the strong activation potential. In addition, we were able to show that the analogue is able to potentiate glucose-induced secretion of insulin from human pancreatic islets. The principal challenge during this study was identifying and understanding small differences in the cAMP-binding domains of cAMP-regulated proteins and matching these differences with suitable modifications of the cAMP molecule.
| Food intake enhances insulin secretion from pancreatic β-cells in two ways. First, an elevated glucose concentration in the blood increases the availability of glucose and thus the rate of ATP formation by glycolysis in β-cells. The increased cellular ATP concentration causes the closure of ATP sensitive K+-channels and thereby depolarization of the cell [1–3]. In turn, voltage dependent Ca2+ channels open and the raise in the cellular Ca2+ concentration causes the fusion of insulin granules with the plasma membrane and thus insulin secretion. Second, this glucose-induced insulin secretion is further enhanced by the incretin hormones glucose-dependent insulinotropic peptide/gastric inhibitory peptide (GIP) and glucagon-like peptide-1 (GLP-1), which are released upon food intake by the gut. The receptors for GIP and GLP-1 are coupled to adenylyl cyclase, which mediates the formation of the second messenger cAMP in β-cells [4,5].
Generally cAMP acts via cAMP-dependent protein kinase (PKA), cyclic nucleotide regulated ion channels, and exchange protein activated by cAMP (Epac) proteins by its direct interaction with highly related cyclic nucleotide binding (CNB) domains [6,7]. Epac proteins are guanine nucleotide exchange factors for the small G-protein Rap [8,9]. Epac1 and Epac2 contain one and two N-terminal CNB domains, respectively (Fig. 1A). Only the second CNB domain of Epac2 controls the proteins exchange activity [10]. The CNB domain blocks the access of Rap to the C-terminal catalytic site in the inactive state [11] and swings away upon activation [10]. The active and the inactive conformation are in equilibrium in the ligand-free and the ligand bound state, and the binding of agonists shifts the equilibrium to various extents to the active conformation [12,13]. The activity induced by an agonist under saturating conditions is termed maximal activity (kmax) and is a measure to what extent the equilibrium is shifted to the active conformation (Fig. 1D). The natural agonist cAMP shifts the equilibrium only partially. It was shown for Epac1 that some cAMP analogues shift the equilibrium up to three times more effectively than cAMP [13].
The pathways by which cAMP augments glucose induced insulin secretion are not fully understood [5]. A function of PKA was demonstrated by selective inhibition of PKA [14]; however, other studies reported only a partial effect [15–17] or even no effect of PKA inhibition [18]. Epac2 was linked to insulin secretion first because of its ability to interact with ATP sensitive K+-channels and with Rim [19,20]. Furthermore it was proposed that Epac contributes to the release of insulin granules by increasing the cellular Ca2+ concentration [21–23], by increasing the granule density at the plasma membrane [24], and by promoting the acidification of the granules [25]. Likely, these effects are mainly mediated by Epac2, as in β-cells Epac1 is expressed at much lower levels than Epac2 [26]. Thus cAMP seems to act at various levels on the process of exocytosis and to act via PKA and Epac-dependent pathways. In agreement with this observation, a recent electrophysiological study has demonstrated that PKA- and Epac-dependent processes enhance the Ca2+-sensitivity and the rate of exocytosis, respectively [27].
Two classes of anti-diabetic drugs are suggested to impinge on Epac2. First, the potency of sulfonylureas, which act as blockers of the ATP sensitive K+-channel in β-cells, is reduced in Epac2−/− mice [28]. Even though a direct binding of sulfonylureas to Epac2 was suggested [28], this interaction could not be confirmed in independent studies [29,30]. Second, exenatide and liraglutide act as GLP-1 receptor agonists and therefore induce formation of cAMP in β-cells. The GLP-1 mediated enhancement of insulin secretion and the blood-pressure lowering effect of liraglutide are mediated at least partially by Epac2 [19,22–25,31].
Direct targeting of Epac2 may be an option for diabetes treatment. Putative rare adverse events of exenatide and liraglutide [32,33] may be circumvented by direct targeting of Epac2, in particular since Epac2 is mainly expressed in pancreatic β-cells [26]. Exenatide and liraglutide are applied by subcutaneous injection and orally applicable alternatives would ease treatment. For the exploration of Epac2 as drug target a better understanding of its signaling is key, and an Epac2 selective agonist would be a valuable tool for such an analysis and for a proof of concept. Therefore, in this study, we aimed for the design of an Epac2 selective cAMP analogue.
In biological research the cAMP analogue D-007 is used as a selective activator of Epac proteins, since it does not act on PKA or cyclic nucleotide regulated ion channels (for chemical structure and abbreviation of cAMP analogues see Fig. 2 and S1 Table) [34,35]. D-007 is modified with a para-chlorophenylthio (pCPT) substituent at the 8-position and an O-methyl substituent (O-Me) at the 2′-position (Fig. 1B). D-007 efficiently activates Epac1 with an AC50 of 1.8 μM and relative maximal activity (kmax) of 3.3. D-007 is thus a stronger agonist than cAMP (AC50 = 50 μM; kmax = 1) (Fig. 1C and 1D; Table 1) [13]. L-026, which carries only the 8-pCPT-modification, activates Epac1 with an AC50 of 0.9 μM and a relative kmax of 0.5. Z-004, which carries only the 2′-O-Me-modification, has a reduced affinity but a higher maximal activity than cAMP (AC50 > 100 μM; kmax > 2) (Table 1) [13]. Thus, the 8-pCPT-modification in D-007 and L-026 is responsible for the high affinity, and the 2′-O-Me-modification in D-007 and Z-004 is responsible for the high maximal activity.
We extended this analysis to Epac2 by using an N-terminal truncated version that lacks the “irrelevant” first CNB domain. This construct is activated by cAMP with an AC50 of 1.8 μM and a relative kmax of 1. Interestingly, Z-004 has the same kmax as cAMP. Thus the 2′-O-Me-modification does not improve the maximal activity of Epac2 contrary to the maximal activity of Epac1. Consequently, D-007 activates Epac2 with an AC50 of 3.5 μM and a relative kmax of 0.8 (Fig. 1C; Table 1). The maximal activity is thus reduced and not enhanced if compared to cAMP and therefore D-007 is a poor activator of Epac2.
The 2′-OH group of cAMP, which is replaced by the O-Me group in D-007 and Z-004, is known to be involved in critical interactions with CNB domains. In PKA and cyclic nucleotide regulated ion channels, the 2′-OH group of cAMP forms a hydrogen-bond with the side chain of a conserved glutamic acid [36,37]. Instead of glutamic acid, Epac1 contains a glutamine (Gln270) and Epac2 contains a lysine (Lys405). The consequences of these differences were analyzed by site directed mutagenesis and activity assays. Epac1Q270K, like Epac2, is poorly activated by D-007, whereas Epac2K405Q is efficiently activated by D-007 (Fig. 1C). Thus, indeed a single amino acid determines the differential response of Epac1 and Epac2 to D-007.
The structures of the complexes Epac2•cAMP•Rap, Epac2K405Q•cAMP•Rap and Epac2K405Q•D-007•Rap were determined, whereby the K405Q mutant served as a model of Epac1 (Fig. 3; Table 2). The structure of Epac2•cAMP•Rap is virtually identical to the previously determined structure of Epac2•S-000•Rap [10], where S-000 had been used as an unnecessary precaution because of its improved hydrolysis resistance. Overall all three newly determined structures are highly similar, but are distinguished by unique conformations in a short region called the hinge (Fig. 3A). The hinge rearranges upon cAMP binding and is responsible for the rigid body movement of the CNB domain, which liberates the catalytic site (Fig. 3B and 3C). In wild-type Epac2•cAMP•Rap, Lys405 points away from the cAMP molecule and interacts with the hinge (Fig. 3D). In Epac2K405Q•cAMP•Rap, Gln405 is turned towards cAMP and forms a hydrogen bond with the 2′-OH-group. The position of Gln405 allows the hinge to alter its conformation, which results in a different hydrogen bond network near Tyr480 (Fig. 3E). The conformation of the hinge in Epac2K405Q•cAMP•Rap would result in clashes with Lys405 in the wild-type protein (Fig. 3E). Overall, this situation favors the active conformation, as Epac2K405Q displays a higher maximal activity upon binding of cAMP than Epac2 (Fig. 1C).
Compared to Epac2K405Q•cAMP•Rap the 2′-O-Me group of D-007 pushes away Gln405 in Epac2K405Q•D-007•Rap (Fig. 3F). In consequence, Gln405 forces the hinge into a different and apparently more favorable conformation, in which Glu443 forms hydrogen bonds with Gln405 and Tyr480. Gln405 does not form a hydrogen bond with the 2′-O-Me group. The loss of the hydrogen bond explains the affinity reducing effect of the 2′-O-Me-group (compare Z-004 and cAMP). The aromatic ring of the 8-pCPT group is kinked perpendicular to the base and shields the binding pocket against the solvent. The transition of the aromatic ring into the hydrophobic protein environment upon binding favors the interaction, which explains the gain in affinity attributed to the 8-pCPT group (compare L-026 with cAMP).
The mechanism of efficient activation of Epac1 by D-007 is thus dependent on the unique Gln270 in Epac1. As this residue is not conserved in Epac2, D-007 is a poor activator of Epac2. This indicates that the differences between Epac1 and Epac2 are sufficient to provide a window for Epac2 selective activation.
To identify the properties of cAMP-analogues that efficiently activate Epac2, the chemical space of substitutions was systematically tested. In total, approximately 100 cAMP analogues, half of which were newly synthesized, were characterized in an iterative design process by determining their AC50 and their relative kmax (Table 1). First, cAMP-analogues modified at a single position were tested. The N-series of analogues in which the N6-position is modified shows strongly decreased maximal activities. This characteristic is in agreement with the crucial interactions of the natural NH2-group at this position of cAMP in stabilizing the active conformation of Epac [11]. In contrast to Epac1, 2′-modifications, which were tested in the Z-series, do not improve kmax values for Epac2, but result in a reduction of affinity by one or two orders of magnitude. The stereo-specific substitution of the axial oxygen in the cyclic phosphate by sulphur (Sp-cAMPS, S-000) results in a relative kmax of 2.5 without influencing the affinity (Fig. 4A left panel; Table 1). Twenty-eight different modifications at the 8-position were tested in the L-series. Several modifications like cyclohexylamino group (L-013) reduce the maximal activity. On the other hand, many modifications enhance the maximal activities, whereby relative kmax values between 1.5 and 5.1 were obtained. The improved maximal activity is accompanied either by a loss or a gain in affinity.
Since modifications of the phosphate system and the 8-position show favorable properties, 8-modified versions of S-000 were synthesized to form the S-series. The focus was on 8-modifications, which in addition to an improved maximal activity had shown a gain in affinity in the L-series. Overall the results show that the effects of both modifications on the maximal activity is “additive” resulting in double modified analogues with increased kmax values of up to 8. In addition, gains in affinity mediated by the 8-modifications are maintained. This finding is illustrated in the left panel of Fig. 4A for S-220, in which a benzylthio group (BnT) was introduced as 8-modification in S-000. S-220 is one of the most potent Epac2 activators, which activates Epac2 with an AC50 of 0.1 μM and a relative kmax of 7.7 (1.8 μM and 1 for cAMP). S-220 was selected for a more detailed analysis as it functions as an efficient activator of Epac2 and its biophysical characterization indicated it as a promising candidate to activate Epac2 selectively over Epac1 (Fig. 5A; Table 1). Additional substitutions at the benzyl ring (S-230, S-240, S-250, S-260, S-270, S-280, S-290, and S-300) do not result in improvements compared to S-220 (Table 1). Similarly, a benzylthio group was the most favorable choice if compared to benzylamino, benzyloxy, and benzylseleno groups (compare L-025 with L-015, L-023 and L-028) (Table 1).
To understand the molecular basis of efficient Epac-2 activation the crystal structures of the complexes containing S-220 and S-280 were solved (Fig. 4B; Table 2). The structure with S-280, which differs from S-220 by only one fluorine atom as a substituent at the para position in the benzyl ring, is shown in Fig. 4B, because it was solved at a higher resolution and does not differ from the Epac2 structure with S-220. The backbone conformation of these structures did not differ from that containing cAMP. Unlike in Epac2K405Q (the model of Epac1) efficient activation is thus not caused by differences in the hinge.
Also, the surroundings of the thiophosphate did not differ. The interaction of the phosphate system with the protein initiates the conformational changes leading to the movement of the CNB-domain required for activation [7,10]. Apparently, the physical properties of the sulphur in the axial position are favored over the oxygen in the active conformation of Epac2. The inability of Rp-cAMPS (R-000) (sulphur in the equatorial position of the cyclic phosphate) to activate Epac emphasizes once more the sensitivity of the cyclic phosphate to perturbations [13].
The aromatic ring of the BnT/pFBnT-group of S-220/S-280 is sandwiched between Leu379 and Lys450. Leu379 is part of the core CNB domain, and Lys450 is part of the lid. The BnT-group “glues” the lid to the core and thereby stabilizes the active conformation. Indeed, the contribution of the BnT-group to efficient activation is lost in Epac2Δ280,K450A (Fig. 4A). The kmax-value of L-025 (8-BnT-cAMP) is reduced to that of cAMP and the kmax-value of S-220 (Sp-8-BnT-cAMPS) is reduced to that of S-000 (Sp-cAMPS) (Fig. 4A). Therefore, the effects of the thiophosphate and the 8-BnT-modification are of different origin and can be separated.
For a final validation, also the activation curves of full length Epac2 (Epac2fl) were recorded in direct comparison to Epac1 (Fig. 5A). S-220 is an efficient activator of Epac2fl but a poor activator of Epac1 (Fig. 5A). The concentration of an analogue required to reach the half-maximal activity of cAMP can be defined as the activation potential (Fig. 1D). This definition reflects the effects on the maximal activity as well as on the affinity. By this definition, S-220 activates Epac2fl 200-times more potently than cAMP, but cannot activate Epac1 to the level of half-maximal cAMP-activity. Conversely, D-007 activates Epac1 100-times more potently than cAMP and reaches half-maximal cAMP activity with Epac2 only under saturating conditions.
Glu315 of Epac1 corresponds to Lys450 in Epac2. The ability of the 8-BnT-substitution to induce a high maximal activity of Epac2 is lost in Epac2K450E, which shows an activation behavior very similar to that of Epac2K450A (S1 Fig.). Thus, the preference of 8-BnT-substituted analogues for Epac2 at least partially originates from this difference.
For biological applications, Epac-selective cAMP analogues should not act on PKA. S-220 activates all four PKA isoforms in a biochemical kinase assay (Fig. 5C–5F). This is in agreement with the earlier findings that PKA is activated by several cAMP analogues with bulky modifications at the 8-position [38–40] and tolerates axial phosphorothioates [40,41]. However, PKA-Iα, PKA-Iβ, and PKA-IIα are activated with a Kact three to eight times higher than that of cAMP, whereas PKA-IIβ is activated by S-220 and cAMP with comparable affinities (Fig. 5C–5F; S2 Table).
Modifications of the 2′-OH group effectively discriminate against PKA [34,35,42]. Therefore, several substitutions of the 2′-OH group, such as hydrogen, chlorine, fluorine, or Me-O, were introduced in 8-substituted Sp-cAMPS analogues. All 2′-substitutions decreased the maximal activity and affinity of Epac2 activation compared to the corresponding un-substituted mother compound (Table 1). S-223 is the most potent 2′-substituted Epac2 activator that efficiently discriminates against Epac1 (Fig. 5A and 5B; Table 1). S-223 still activated Epac2 ten times more potently than cAMP (Fig. 5B). Conversely, the ability of S-223 to activate PKA was drastically reduced (Fig. 5C–5F). PKA-Iα and PKA-Iβ were not activated at concentrations up to 1 mM. Furthermore, the Kact of PKA-IIα and PKA-IIβ is reduced approximately 300 and 11,000 times, respectively, compared to cAMP (Fig. 5C–5F; S2 Table). Therefore, S-223 discriminates even more efficiently against PKA than D-007, whose affinity is only reduced 250 to 900 times compared to cAMP (Fig. 5C–5F; S2 Table).
A model cell-system was generated to confirm the in vitro selectivities under more physiological conditions (Fig. 6A). U2OS cells do not express Epac1 or Epac2. Therefore increased cAMP levels induced PKA but not Epac signaling. Increased cAMP levels in U2OS cell lines with a stable over-expression of Epac1 or Epac2 resulted in increased Rap•GTP levels next to enhanced PKA signaling. The phosphorylation of vasodilator-stimulated phosphoprotein (VASP) was monitored as a measure of PKA activity by a band shift.
The effect of increased cAMP levels upon the stimulation of cells could be mimicked by the nonselective analogue L-026 (Fig. 6B and 6C). D-007 strongly activated Epac1 but had only minor effects on Epac2 and PKA. S-220 efficiently activated Epac2 with only minor effects on Epac1. Interestingly, S-220-induced PKA activation was very low in the cellular model system. In biochemical assays with recombinant proteins S-220 activated PKA with a slightly lower affinity than cAMP, whereas L-026 activated PKA [40] and Epac (Table 1) at lower concentrations than cAMP. Likely, the ultimate cellular concentrations of S-220 were not sufficient to efficiently activate PKA. Unfortunately, S-223 induced neither PKA nor Epac signaling. Its potency is likely further limited by inefficient cellular uptake.
Epac is known to occur in signaling complexes with PKA-anchoring proteins [43]. To investigate a putative contribution of PKA signaling to the formation of Rap•GTP, cells were stimulated after pretreatment with a PKA inhibitor. Inhibition of PKA abolished the phosphorylation of VASP, when cAMP levels were increased by the application of forskolin and 3-isobutyl-1-methylxanthine (IBMX), but had no effect on Rap•GTP levels. Similarly, inhibition of PKA had no effect of S-220 or D-007 induced Rap activation (S2 Fig.).
To test the potential of S-220 to augment glucose-induced insulin secretion primary islets were isolated from human pancreases. The islets were stimulated either with glucose as control or with glucose and Forskolin/IBMX or S-220 (Fig. 6D). Forskolin/IBMX was used as a cAMP inducing agent and represents the maximal possible effect resulting from activation of PKA and Epac2. At concentrations of 25 to 100 μM, S-220 potentiates glucose-induced insulin secretion with similar efficiency (Fig. 6D), suggesting a strong contribution of Epac2.
Selective targeting of highly related proteins is a common challenge in drug design. Here we have targeted the CNB domain of Epac2 in an iterative design process. A comprehensive activity profile was generated by determining the affinity and the relative maximal activity of about 100 cAMP analogues (Table 1). This approach led to the identification of positions in cAMP on which modifications are tolerated by Epac proteins. For example, Epac2 does not tolerate any modification of the amino-group at the 6-position, whereas a wide variety of substituents are tolerated at the 8-position. Different substituents at the 8-position could be used to modulate the affinity and the maximal activity, whereby some modifications increased the maximal activity by a factor of 5. An independent improvement in maximal activity was obtained by introducing a sulphur atom into the axial position (Sp-isomer) of the cyclic phosphate. Interestingly, the effects of the thiophosphate and 8-substitutions are “additive” (Fig. 4A; Table 1). The crystal structure of S-220 in complex with Epac2 shows that in this case the benzyl ring of the 8-substituent stabilizes the CNB domain in the active conformation by hydrophobic interactions. Upon disruption of this interaction by site directed mutagenesis the effect of the 8-subsitution was selectively eliminated.
A similar interaction of an 8-substituent is impossible in Epac1 due to a single amino acid difference. In fact neither 8-substituted cAMP analogues nor Sp-cAMPS analogues are able to increase the maximal activity of Epac1. An increase could only be obtained by modifications of the 2′-position as for example by the 2′-O-Me group in D-007. On the other hand, it was not possible to obtain beneficial effects with 2′-substituted analogues for Epac2. Instead 2′-substitutions decreased the maximal activity of Epac2. Again it was possible to attribute these differences to a single amino acid difference between Epac1 and Epac2. Thus, even though highly related, Epac isozymes show distinct activity profiles, which originate from subtle differences in the cAMP binding site. In consequence of these differences fundamentally different effects are responsible to stabilize the CNB domain in the active conformation of Epac1 and Epac2.
For structural studies we used Epac2K405Q as a model of Epac1. Epac2K405Q mimics the response of Epac1 towards 2′-substituted cAMP analogues and is efficiently activated by D-007, while Epac2 is not (Fig. 1C). The advantage of this approach is, that any differences in the structures if compared to Epac2 must originate exclusively from the single amino acid point mutation. The 2′-O-Me-group of D-007 induces via the glutamine a different conformation in the backbone of the hinge. Thus, upon the transition from the inactive to the active state the conformation of the hinge is transformed into different, more or less favored, conformations depending on the nature of the cyclic nucleotide.
In depth characterization classified D-007 as an efficient activator of Epac1 and S-220 as an efficient activator of Epac2 (Fig. 4A and 4B). Actually, the natural agonist cAMP appears to be a rather poor activator of Epac compared to D-007 or S-220. Assuming that the most efficient activators shift the equilibrium between the inactive and the active conformation (almost) fully to the active side, about 65% and 85% of cAMP bound Epac1 and Epac2, respectively, would still be in an inactive conformation (Fig. 1D). This property of Epac eases the design of analogues, as a better activation potential can be gained by improving affinity and maximal activity. During evolution the key requirement for Epac might have been to be cAMP responsive rather than making optimal use of the catalytic potential. Alternatively, any factor that binds specifically to the active conformation of Epac would shift the equilibrium to the active side. Although the existence of such a factor is speculative, it could add an extra level of regulation by exhausting the full catalytic potential.
S-220 activates all isoforms of PKA in biophysical assays, though less efficiently than cAMP (Fig. 5C–5F). Discrimination of S-220 against PKA and Epac1 is thus not absolute. However, the capability of S-223, a variant of S-220, to activate PKA is basically absent in biophysical assays. But it must be noted that S-223 is over 20 times less potent in activating Epac2 and it turned out that S-223 is unable to induce Epac2 activation in cell culture. Interestingly, the selectivity window provided by S-220 is sufficient to cause selective activation of Epac2 in vivo. Similarly, D-007 seems to act as an Epac1 selective agonist, despite its low potential to activate Epac2 in biophysical assays (Fig. 5). The bioavailability of the analogues limits the maximal concentration that is reached in the cell. Thus the optimal compromise between bioavailability on the one site and biophysically defined selectivity and activation potential on the other site needs to be identified and validated as demonstrated in the case of S-220 and S-223. To avoid putative cross-reactivity it is in any case advisable to use the analogues at the lowest possible concentration.
D-007 was originally introduced as an Epac selective cAMP-analogue due to its poor activation potential for PKA [34,35]. However, direct data were only obtained with Epac1 [13,34] and a comprehensive biophysical analysis had introduced the concept of efficient activation based on the characterization of Epac1 [12,13]. In fact, D-007 is a poor activator of Epac2 (Figs. 1B and 4B). D-007 induces no or marginal Epac2 activation in our model system of Epac cell lines (Fig. 6B and 6C). Interestingly, several studies have used D-007 to investigate Epac2 mediated biological effects (for example [24,25,44,45]). In these studies, D-007 was frequently applied at rather high concentrations as an acetoxymethyl ester. This ester functions as a pro-drug with improved membrane permeability and releases the active mother compound in the cell. It was shown to act at 100- to 1,000-fold lower concentration in tissue culture if compared with the direct application of D-007 [46]. The acetoxymethyl ester of D-007 might therefore be capable of causing sufficient Epac2 activation if applied at high concentrations.
The combination of cAMP analogues with recently developed Epac inhibitors [47] will ease distinguishing Epac1, Epac2, and PKA-mediated effects. D-007, S-220, and N-002 act as selective agonists of Epac1, Epac2, and PKA, respectively, and can be used in direct comparison in titration experiments. N-002 was previously shown to activate PKA but not Epac1 [35], and this study demonstrates also its inability to activate Epac2 (Table 1). Inhibitors targeting the kinase domain of PKA allow selective inhibition of PKA-mediated signaling [48–51]. ESI-05 and its derivative HJC0350 are selective inhibitors of Epac2 [52–54] and CE3F4 inhibits preferentially Epac1 over Epac2 [55,56].
S-220 augments glucose induced insulin secretion from primary human islets. Selective activation of Epac2 under physiological conditions by cAMP analogues is thus feasible. S-220 showed a similar potential as agents that induce cAMP production (Fig. 6). This argues for a major role of Epac2 in mediating the effects of GIP and GLP-1 on direct insulin secretion. S-220 is therefore expected to become a valuable tool in analyzing the underlying signaling pathways in more detail. The pharmacological properties of cAMP analogues are not optimal, in particular as their membrane permeability is limited by the negative charge of the cyclic phosphate. It is, however, possible to convert cAMP analogues into pro-drugs, in which the negative charge is masked. This concept was proven by the previously mentioned acetoxymethyl ester of D-007, while Adefovir dipivoxil is an example of a related pro-drug version of a 5′-AMP analogue for the treatment of hepatitis B.
In summary, we have selectively targeted the highly related CNB domains of Epac1 and Epac2 by cAMP analogues. Several analogues are capable of activating Epac2 up to 200-fold higher potency than cAMP. The structural basis of this efficient activation and of selectivity was identified. Our data indicate that in vivo D-007 and S-220 are selective agonists of Epac1 and Epac2, respectively. S-220 potentiates insulin secretion in primary human islets, confirming a major role of Epac2 in this process. The results obtained for S-220 indicates that selective pharmacological targeting of Epac2 is possible.
Human cadaveric donor pancreases were procured via a multi-organ donation program. Islets were isolated at the Leiden University Medical Center and were used in this study if they could not be used for clinical transplantation, according to national laws, and if research consent was present.
The following constructs were used: Epac1, homo sapiens, amino acids 150–881, referred to as Epac1; Epac2, mus musculus, amino acids 1–993 and amino acids 281–993, referred to as Epac2fl and Epac2Δ280, respectively. Murine Epac2 was used as all previous structural studies were performed with it [10,11,57]. Murine Epac2 is more than 97% identical to human Epac2, and the CNB site is 100% identical.
The activity of Epac was determined by a fluorescence assay [12]. In brief, the substrate protein Rap1B was loaded with the fluorescent GDP analogue 2′-/3′-O-(N-methylanthraniloyl)-guanosine diphosphate (mGDP). The fluorescence intensity of Rap1B•mGDP is approximately twice that of free mGDP, and thus, nucleotide exchange can be observed as a decay in fluorescence upon the addition of excess unlabeled GDP. The decay is single exponential, and the observed rate constant was plotted against the concentration of cyclic nucleotide (Figs. 1C, 4A, 5A, and 5B).
The recombinant human PKA catalytic subunit (Cα1) and the four human PKA regulatory subunits (RIα, RIβ, RIIα, RIIβ) were expressed, purified, and characterized as described [58]. The PKA activity was assayed using a coupled spectrophotometric assay [59] with 260 μM Kemptide (LRRASLG; GeneCust) as a substrate and cyclic nucleotides in a range from 100 pM to 1 mM. PKA holoenzymes were formed by mixing the R- and C-subunit at a molar ratio of 1.2:1 and extensive dialysis against 20 mM MOPS (pH 7.0), 150 mM NaCl, 2.5 mM β-mercaptoethanol, supplemented with 1 mM ATP and 10 mM MgCl2 for PKA-I at 4°C overnight. PKA holoenzymes were used at about 20 nM for the activation assay. The apparent activation constants (Kact) were determined by fitting the concentration-dependent activity to a sigmoid dose-response model.
Rap•GTP levels were determined by precipitating GTP-bound Rap specifically from cell lysates and subsequent western Blotting with an α-Rap antibody (Santa Cruz Biotechnology) [46]. To generate stable cell lines, U2OS cells were transfected with pBabe-Flag-Epac1 (Epac1, homo sapiens, amino acids 1–881) or pBabe-Flag-Epac2 (Epac2, mus musculus, amino acids 1–993), selected for Epac expression and maintained under selection with 2 mg/l puromycin. The cells were cultured according to standard protocols. The activity of PKA was determined by western blotting with a monoclonal α-VASP antibody (BD Transduction Laboratories). Blots were developed by enhanced chemiluminescence (ECL) and the use of X-ray films; for quantification films were scanned. The intensities of the Rap•GTP bands and the upper band of VASP (P-VASP) were determined for each condition in ImageJ and normalized to the stimulation with forskolin and IBMX for each blot.
Epac2 proteins (Epac2, mus musculus, amino acids 306–993) were purified and crystallized as described [10].
Islets were isolated from two human donor pancreases [60]. Intact islets (n = 20 per well) were seeded on ultra-low attachment 96-well plate. After 3 days of culture, the pancreatic islets were washed two times with 115 mM NaCl, 5 mM KCl, 24 mM NaHCO3, 2.2 mM CaCl2, 1 mM MgCl2, 20 mM HEPES, and 0.2% human serum albumin (incubation buffer) and primed for 1.5 h at 37°C in incubation buffer supplemented with 2 mM glucose (low glucose). Subsequently, the cells were incubated in fresh incubation buffer supplemented with 2 mM glucose for 1 h to measure the basal insulin secretion levels. The cells were then incubated for 1 h in incubation buffer supplemented with 16.7 mM glucose (high glucose) in the presence or absence of cAMP analogues. The cell supernatants were collected immediately after incubation with low and high glucose. The insulin concentrations were measured by ELISA (Insulin ELISA Kit, Mercodia). The insulin secretion was expressed as the ratio of insulin concentration at high and low glucose concentrations for each well (insulin secretion index).
All reagents were of analytical grade or the best grade available from commercial suppliers. DMSO was stored over activated molecular sieves (3 Å) for at least two weeks before use. The UV spectra were recorded with a Helios β spectrometer (Spectronic Unicam) in aqueous phosphate buffer (pH 7.0). The mass spectra were obtained with an Esquire LC 6000 spectrometer (Bruker Daltonik) in the ESI-MS mode with 50% isopropanol/49.9% water/0.1% formic acid as matrix.
If not stated otherwise, all chromatographic operations were performed at ambient temperature. The analytical HPLC-system consisted either of a L-6200 pump, a L-4000 variable wavelength UV/Vis detector and a D-2500 GPC integrator (all Merck-Hitachi) or a LaChrom Elite instrument with a L-2130 pump, a L-2420 variable wavelength UV/Vis detector, a L-2350 column oven (set at 30°C), and EZChrom software version 3.3.1 SP1 (all VWR-Hitachi). The stationary phase was Kromasil (AkzoNobel) C 8–100, 10 μm, or YMC ODS-A 12 nm, S-11 μm (YMC), both in 250 × 4.6 mm stainless steel columns.
Preparative MPLC was accomplished with a C-605 pump (Büchi), a preparative K 2001 UV-detector (Knauer), and a L200E analog recorder (Linseis). Merck LiChroprep RP-18 6 nm, 15–25 μm (Merck-Hitachi) in a 410 × 50 mm glass column (Kronlab) was used to isolate and desalt the nucleotides. If necessary, cation exchange was carried out with Toyopearl SP-650M, 65 μm, sodium form (Tosoh Bioscience), in a 125 × 35 mm or a 250 × 50 mm glass column (Kronlab).
Preparative HPLC was performed with a LC-8A pump (Shimadzu), a preparative K 2001 UV-detector (Knauer), and a L200E analogue recorder (Linseis). YMC ODS-A 12 nm, S-11 μm (YMC) in a 250 × 20 mm or a 250 × 16 mm stainless steel column was used for purification and desalting.
Purification columns (MPLC and HPLC) were equilibrated with either 100 mM NaH2PO4 or 20 mM triethylammonium formate (TEAF) (pH 7). Subsequently, the raw products were applied and the columns were initially washed with the same buffer, followed by water. Each cyclic nucleotide was eluted with a gradient from 10% water to 20%–50% isopropanol or acetonitrile. Cation exchange to sodium was performed with compounds isolated from the purifications with TEAF buffer during the equilibration phase. The product-containing fractions were collected and evaporated under reduced pressure to obtain the target compound in the sodium form.
The typical yields of isolated cyclic nucleotides were in the range of 20%–85%. The purity of each analogue was at least >99% (by analytic HPLC at λmax). The structure of each analogue was confirmed by UV/VIS spectrometry and ESI/MS analysis. The structures of selected analogues were further confirmed by NMR (S1 Text). The nucleotides were quantified and aliquoted using the extinction coefficient at their λmax.
The following compounds were provided by BIOLOG LSI: 8-Br-2′-Cl-adenosine, 8-Br-2′-F-adenosine, cAMP, 2′-dcAMP (Z-001), 2′-F-cAMP (Z-002), 2′-NH2-cAMP (Z-003), 2′-O-Me-cAMP (Z-004), 8-Cl-cAMP (L-001), 8-Br-cAMP (L-002), 8-NH2-cAMP (L-003), 8-N3-cAMP (L-004), 8-MeA-cAMP (L-005), 8-AEA-cAMP (L-006), 8-HA-cAMP (L-009), 8-AHA-cAMP (L-010), 8-DMeA-cAMP (L-015), 8-PIP-cAMP (L-018), 8-OH-cAMP (L-021), 8-BnT-cAMP (L-025), 8-pCPT-cAMP (L-026), N6-Bn-cAMP (N-001), N6-Bnz-cAMP (N-002), N6-Phe-cAMP (N-003), 8-Br-2′-O-Me-cAMP (D-005), 8-pCPT-2′-O-Me-cAMP (D-007), 8-pOHPT-2′-O-Me-cAMP (D-008), 8-pMeOPT-2′-O-Me-cAMP (D-009), 8-OH-2′-O-Me-cAMP (D-010), Sp-cAMPS (S-000), Sp-8-Br-cAMPS (S-010), Sp-8-Br-2′-dcAMPS (S-011), Sp-8-Br-2′-O-Me-cAMPS (S-014), Sp-8-Cl-cAMPS (S-020), Sp-8-ADOA-cAMPS (S-060), Sp-8-PIP-cAMPS (S-110), Sp-8-OH-cAMPS (S-130), Sp-8-pCPT-cAMPS (S-210), Sp-8-pCPT-2′-O-Me-cAMPS (S-211), Sp-5,6-DCl-cBIMPS (S-400), Rp-cAMPS (R-000).
2′-O-Pr-cAMP (Z-005), 2′-O-Bu-cAMP (Z-006), 8-OHEA-cAMP (L-007), 8-(S-2-OHPrA)-cAMP (L-012), 8-PIA-cAMP (L-019), and Sp-8-PIA-cAMPS (S-120) were provided by B. Jastorff (Bioorganic Chemistry Unit, Department of Chemistry, University of Bremen, Bremen, Germany).
2′-O-All-cAMP (Z-007) and 2′-O-Bn-cAMP (Z-008) were prepared from cAMP by alkylation with appropriate alkylhalogenides [61].
8-Br-2′-Cl-cAMP (D-003), 8,2′-DCl-cAMP (D-006), and 8-Br-2′-F-cAMP (D-004) were prepared from 8-Br-2′-Cl-adenosine and 8-Br-2′-F-adenosine by a two-step one-pot reaction scheme with 2 equivalents (eq.) phosphoryl chloride and excess cyclisation solution consisting of 0.1% KOH in acetonitrile/water 60/40 (v:v) at room temperature [62]. 8,2′-DCl-cAMP was obtained as a side product during 8-Br-2′-Cl-cAMP production by bromine to chlorine exchange in position 8 of the adenine moiety. Chlorine was introduced in the initial phosphorylation step as verified by analytical HPLC.
Sp-8-Br-2′-Cl-cAMPS (S-013), Sp-8,2′-DCl-cAMPS (S-021), and Sp-8-Br-2′-F-cAMPS (S-012) were synthesized from Br-2′-Cl-adenosine and 8-Br-2′-F-adenosine by a related two-step one-pot reaction scheme with thiophosphoryl chloride and refluxing cyclisation solution [63]. Sp-8,2′-DCl-cAMPS was pre-formed as a side product during the thiophosphorylation of 8-Br-2′-Cl-adenosine.
8-MeSe-cAMP (L-027), 8-BnSe-cAMP (L-028), 8-BnSe-2′-O-Bn-cAMP (D-012), and 8-BnSe-2′-O-Me-cAMP (D-013) were prepared from corresponding 8-bromo nucleotides by nucleophilic substitution with sodium hydrogen selenide [64] and subsequent alkylation with methyl iodide or benzyl bromide as described [65].
cAMP analogues were dissolved in D2O to a final concentration of 10 mM. Spectra were recorded at 293 K or 298 K (S140) on a 750 MHz Bruker Avance NMR machine equipped with 5mm QXI probe (S1 Text). Reported chemical shifts are calibrated directly (1H) or indirectly (31P, 13C) with respect to DSS. Assignments of S150 and S220 are validated with 2D TOCSY (mixing times of 20 and 100 ms) and [1H;13C]-HSQC.
|
10.1371/journal.ppat.1000776 | Interaction of Cryptococcus neoformans Rim101 and Protein Kinase A Regulates Capsule | Cryptococcus neoformans is a prevalent human fungal pathogen that must survive within various tissues in order to establish a human infection. We have identified the C. neoformans Rim101 transcription factor, a highly conserved pH-response regulator in many fungal species. The rim101Δ mutant strain displays growth defects similar to other fungal species in the presence of alkaline pH, increased salt concentrations, and iron limitation. However, the rim101Δ strain is also characterized by a striking defect in capsule, an important virulence-associated phenotype. This capsular defect is likely due to alterations in polysaccharide attachment to the cell surface, not in polysaccharide biosynthesis. In contrast to many other C. neoformans capsule-defective strains, the rim101Δ mutant is hypervirulent in animal models of cryptococcosis. Whereas Rim101 activation in other fungal species occurs through the conserved Rim pathway, we demonstrate that C. neoformans Rim101 is also activated by the cAMP/PKA pathway. We report here that C. neoformans uses PKA and the Rim pathway to regulate the localization, activation, and processing of the Rim101 transcription factor. We also demonstrate specific host-relevant activating conditions for Rim101 cleavage, showing that C. neoformans has co-opted conserved signaling pathways to respond to the specific niche within the infected host. These results establish a novel mechanism for Rim101 activation and the integration of two conserved signaling cascades in response to host environmental conditions.
| Cryptococcus neoformans is an environmental fungus and an opportunistic human pathogen. Survival of this fungus within a human host depends on its ability to sense the host environment and respond with protective cellular changes. It is known that the cAMP/PKA signal transduction cascade is important for sensing host-specific environments and regulating the cellular adaptations, such as capsule and increased iron uptake, that are necessary for growth inside the infected host. Here we document that, unlike what has been described in other fungal species, a C. neoformans Rim101 homologue is directly regulated by PKA. The Rim101 signaling pathway is also involved in capsule regulation and virulence. Our study demonstrates that Rim101 integrates two conserved signal transduction cascades, and it is important in regulating microbial pathogenesis.
| All cells, including pathogenic microorganisms, must be able to sense and respond to changes in their environment. As these cells enter a human host, they need to protect themselves from the immune system and rapidly adapt to human physiologic conditions, such as low nutrient availability, varying pH, and mammalian concentrations of carbon dioxide [1]. Therefore, they must coordinate multiple signaling pathways in order to control appropriate cellular responses.
One of the most common environmental stresses for pathogenic fungi is a change in the extracellular pH. Alterations in pH can affect a large number of cellular processes including membrane and cell wall stability, morphogenesis, protein stability and function, and nutrient uptake [2]–[8]. Many of these responses to pH are regulated by the Rim101 transcription factor and its homologues (PacC in filamentous fungi). Additionally, many pathogenic fungi respond to the neutral or slightly alkaline pH of the host by inducing virulence-associated phenotypes [2], [9]–[14]. Therefore, in diverse fungi such as Aspergillus and Candida species, mutants defective in pH sensing/response no longer induce phenotypes associated with virulence in pathogenic species. For example, C. albicans rim101Δ/Δ mutants do not undergo pH-dependent dimorphic switching, do not appropriately increase uptake of iron, and do not secrete the proteases and phosphatases necessary for invasion of host tissues [3], [5], [15]–[19]. A. nidulans pacC (rim101) mutants display decreased growth, decreased secondary metabolite production, and defective invasive growth [9], [14], [20]–[22]. Although A. nidulans is non-pathogenic, these cellular processes have been associated with virulence in other Aspergillus species.
In addition to the direct effects of ambient pH on cell integrity and various metabolic processes, pH changes also affect nutrient uptake. For example, under alkaline conditions, the availability of free iron is greatly reduced as the iron equilibrium shifts from the bioavailable ferrous form to the insoluble ferric form. Studying iron flux is an important new horizon in fungal pathogenesis, as the human host keeps free iron levels at extremely low concentrations (10−18M) through constitutively expressing iron-binding proteins such as transferrin and lactoferrin. In this way, the host protects against invading microorganisms. Fungal pathogens unable to increase iron uptake in this iron-limited host environment often have severe defects in virulence [5], [23]–[26]. The pH-responsive Rim101 transcription factor is involved in the regulation of iron homeostasis, directly binding to the promoters of genes encoding high affinity iron uptake proteins: iron transporters, iron permeases and siderophore transporters [20],[23],[27].
Cryptococcus neoformans is an opportunistic human fungal pathogen. Unlike the distantly related pathogens Candida albicans or Aspergillus fumigatus, C. neoformans grows within a very narrow range of pH values in the host. It grows well at the human physiological pH of the blood and cerebrospinal fluid (pH 7.4) as well as in acidic environments such as the phagolysosome of the macrophage (pH 5) [28],[29]. However, unlike C. albicans and A. fumigatus, C. neoformans demonstrates a severe growth defect above pH 8. Despite this increased sensitivity to alkaline pH, there is still evidence that C. neoformans responds to the slightly alkaline pH of the infected host blood by inducing virulence-associated phenotypes. Capsule production, a major virulence determinate, is optimal at human physiological pH [28], [30]–[32].
On a molecular level, C. neoformans capsule synthesis is transcriptionally regulated by elements of the cAMP/PKA pathway. Strains with mutations in core cAMP signaling elements (such as the Gpa1 Gα protein, adenylyl cyclase Cac1, or the PKA catalytic subunit Pka1) display defective expression of capsule, and these mutant strains are attenuated for virulence in animal models of cryptococcosis. When the regulatory subunit of PKA is mutated in the pkr1Δ strain, the cells have constitutive activation of PKA signaling and display high production of capsule, even in non-inducing conditions [33]–[38]. Other inducing conditions for capsule include iron deprivation, nutrient limitation, and the presence of serum. The mechanisms by which these environmental signals are sensed and subsequently transduced to specific intracellular signaling pathways are not yet known. To explore the interaction of the inducing environmental signals, signal transduction pathways, and downstream effectors controlling capsule synthesis, we have begun to characterize specific transcription factors that are predicted to be targeted by the cAMP pathway, and which also directly control capsule gene expression. By examining the biological function and regulation of the C. neoformans Rim101 homolog transcription factor, we have determined a new pathway for C. neoformans regulation of capsule production. We have also defined a novel interaction between the highly conserved Rim and PKA signaling pathways.
To identify potential transcriptional regulators of C. neoformans capsule that are also directly phosphorylated by PKA, we used a bioinformatic survey of the annotated C. neoformans genome. We first searched the available annotation (GO terms, gene names, homology designators) for proteins likely to be involved in transcriptional regulation: transcription factors, DNA binding motifs, zinc finger domains, and other transcriptional regulators. Given the incomplete annotation of the genome, we accepted that many transcriptional regulators might be initially misidentified or excluded by this approach.
Among this subset of proteins, we searched the predicted protein sequence for consensus sequences for PKA phosphorylation (R/K R/K X S/T), to identify potential direct targets of the Protein Kinase A enzyme. One of these proteins is a homologue of Rim101/PacC, a conserved fungal C2H2 transcription factor (GenBank ID CNH00970). The C. neoformans Rim101 protein contains a RRASSL motif at aa730, and has highest homology to Aspergillus nidulans PacC in the C2H2 domain. Analysis of the protein for conserved domains revealed that the only significant Pfam-A match is the zinc finger C2H2 domain (aa133–155).
To characterize the biological role of this C. neoformans transcriptional regulator, we disrupted the C. neoformans RIM101 gene by homologous recombination. Southern blot analysis confirmed that the rim101::nat mutant allele precisely replaced the native gene in the rim101Δ deletion strain (TOC2) (Table 1) without additional ectopic integration events (data not shown). In addition to this mutant strain in which the entire RIM101 open reading frame was deleted, we created an independent rim101Δ mutant with a partial gene deletion. All phenotypes between these strains were identical, and the TOC2 strain was therefore chosen as the representative rim101Δ mutant strain. To ensure that any phenotypes observed in the rim101Δ mutant strain were due to disruption of the RIM101 gene, we created a rim101Δ+RIM101 complemented strain (TOC4) by integrating a wild-type copy of the RIM101 gene into the genome of the rim101Δ mutant strain.
The rim101Δ mutant strain grew at a similar rate as wild-type on rich media (YPD) and minimal media (YNB) at 30°C, 37°C and 39°C. We found no defect in melanin production on Niger-seed medium. We also examined the mutant strain for resistance to hydrogen peroxide and paraquat by disc diffusion assays and established that the zone of inhibition for the rim101Δ mutant strain was similar to that of wild-type, showing no additional sensitivity to reactive oxygen species.
The rim101Δ mutant strain has a major defect in polysaccharide capsule, an important virulence factor in C. neoformans. We incubated wild-type and rim101Δ mutant strains in a capsule inducing medium, Dulbecco's modified Eagle's medium (DMEM) containing 25 mM NaHCO3, for 72 hrs at 37°C and 5% CO2. Incubation in this media usually leads to large polysaccharide capsules surrounding each cell which can be quantitatively measured by analyzing the percent packed cell volume (Cryptocrit analysis) [31]. The rim101Δ mutant strain exhibits markedly reduced capsule around the cell (3.6% packed cell volume) compared to the wild-type strain (6.2% packed cell volume). The rim101Δ mutant capsule-defective phenotype was not noted in previous reports of the C. neoformans Rim101 protein [39]; therefore, we confirmed our observation in several ways. We examined several independent rim101Δ mutants, including partial and complete gene deletions, which all displayed similar capsule defects in the inducing conditions. These differences in capsule were microscopically visualized by the exclusion of India ink (Figure 1A). Reintroduction of the wild-type allele fully complemented the capsule phenotype (6.8% packed cell volume).
To confirm the role of Rim101 and the Rim pathway in C. neoformans capsule production, we searched the C. neoformans genome for conserved elements of the Rim pathway, and we identified the RIM20 gene. Rim20 is a scaffold protein required for Rim101 cleavage/activation in other fungal species [6], [40]–[45]. When we mutated this gene, we observed a similar capsule defect in the rim20Δ mutant compared to the rim101Δ mutant strain (Figure 1A).
C. neoformans capsule is secreted out of the cell and subsequently bound to the cell wall [46]. To determine whether rim101Δ mutant strains produce and secrete capsular polysaccharide, we used a previously described gel electrophoresis technique to quantify this polymer [47]. C. neoformans cells were incubated in capsule-inducing DMEM for 1 week, after which polysaccharide that was shed into the medium was analyzed for relative abundance and size by reactivity against an anti-GXM antibody (mAb18B7). We noted a similar amount of secreted polysaccharide in the rim101Δ mutant as wild-type (Figure 1B) suggesting that there is no significant GXM synthesis defect in the rim101Δ strain. We repeated this assay using low iron medium [31] as the capsule inducing condition instead of Dulbecco's modified Eagle's medium, and we again observed a similar amount of secreted capsule in wild-type and rim101Δ mutant strains (Figure 1B). As previously described, we used a cap59 mutant strain that is unable to secrete capsule as a negative control [48] and detected no capsule by this assay in this strain. We also demonstrated the quantitative nature of this assay by analyzing the pka1Δ mutant strain, in which there is a previously documented decrease in capsule production compared to wild-type [36] (Figure 1B). Therefore, the hypocapsular phenotype of the rim101Δ mutant strain in the presence of intact polysaccharide production suggests a defect in capsule attachment. This result is similar to the phenotype of the C. neoformans ags1Δ mutant strain which can synthesize and secrete capsule but cannot bind it [49],[50].
In fungi as diverse as Aspergillus, Saccharomyces, Candida, and Ustilago species, the Rim101/PacC proteins control multiple pH-related phenotypes, including regulating iron homeostasis; maintaining membrane and cell wall-associated proteins; and secreting proteases, secondary metabolites, and phosphatases [2],[5],[14],[51]. Many of these factors are important in the virulence of pathogenic fungi.
We tested the C. neoformans rim101Δ mutant strain to determine if this protein retains conserved physiological roles with Rim101/PacC proteins in other fungal species. On alkaline media, the rim101Δ mutant strain exhibits a severe growth defect compared to the wild-type and the reconstituted strains at alkaline pH above 7.6 (p<0.01) (Figure 2A). Similar to other rim101-defective fungal strains, the C. neoformans rim101Δ mutant also displayed sensitivity to media containing 200 mM LiCl or 1.5M NaCl (Figure 2B). To confirm that this growth defect was due to specific sensitivity to ionic stress as opposed to general sensitivity to osmotic stress, we tested the cells for growth on media containing 2.5M sorbitol and detected no difference in growth between the rim101Δ mutant strain and wild-type. We also determined no defects in response to cell wall stress for the rim101Δ mutant strain during growth on calcofluor white, Congo red, or 0.05% SDS. These data suggests that, unlike in other fungal species, there are no major defects in cell wall integrity in the C. neoformans rim101Δ mutant strain.
In other fungal species, the Rim101 protein is activated by cleavage and subsequently localized to the nucleus, as expected for a transcription factor [52]. To test whether C. neoformans Rim101 is cleaved and localized to the nucleus, we created a Gfp-Rim101 fusion protein. We fused the green fluorescent protein gene to the N-terminus of the RIM101 gene, expressing the new transgene under control of a constitutive histone promoter. Introduction of this plasmid (pTO2) by biolistic transformation into the rim101Δ mutant strains fully complemented the rim101Δ mutant capsule phenotype, indicating that the Gfp-Rim101 fusion protein is functional (Figure 1C).
Unlike A. nidulans, in which localization is dependent on activation, C. neoformans Rim101 is nuclear under all conditions tested. Using epifluorescent microscopy, we observed a nuclear pattern of localization for C. neoformans Gfp-Rim101 after 24 hours growth in various conditions, including YNB buffered at pH 8, Dulbecco's modified Eagle's medium, YNB (pH 5.4), and YPD (Figure 3). In contrast, the Gfp-Rim101 protein in the pka1Δ and rim20Δ mutant backgrounds localized to both the nucleus and the cytoplasm under the same growth conditions, suggesting that PKA activity and Rim20-mediated cleavage are both important for complete nuclear localization (Figure 3). In addition, overexpression of the Gfp-tagged Rim101 protein was not able to suppress the pka1Δ or the rim20Δ mutant capsule phenotype (Figure 1C).
To further explore the potential association of PKA and Rim101, we mutated the single Rim101 consensus sequence for PKA phosphorylation by changing serine 773 to an alanine, creating the Rim101-S773A mutant protein encoded in plasmid pTO3. When examining the localization of the Gfp-Rim101-S773A protein, we observed both nuclear and cytoplasmic fluorescence, similar to the localization of wild-type Gfp-Rim101 expressed in the pka1Δ and rim20Δ mutant backgrounds. Introduction of the Gfp-Rim101-S773A protein into the rim20Δ background also resulted in both nuclear and cytoplasmic localization of Rim10. Additionally, the Rim101-S773A mutant protein in the rim101Δ mutant background did not complement the capsule phenotype (Figure 1D). This observation, coupled with our documentation that some Gfp-Rim101-S773A protein is localizing to the nucleus, indicates that serine 773 is necessary for full function of the Rim101 protein. When we transformed the Rim101-S773A mutant protein into the pkr1Δ mutant strain, in which PKA signaling and capsule production are constitutively active, multiple independent transformants had markedly attenuated capsule, even though the wild-type RIM101 gene was still present in these strains (Figure 1D). In contrast, introduction of the plasmid containing the wild-type Rim101 fused to Gfp into the pkr1Δ strain had no effect on capsule. This suggests that the Rim101-S773A mutant protein is acting in a dominant negative manner on C. neoformans capsule.
To examine the interaction of PKA and the C. neoformans Rim pathway, we used western blotting techniques to compare Rim101 protein processing in multiple strain backgrounds. Using an anti-Gfp monoclonal antibody for detection, we identified bands corresponding to the Gfp-Rim101 fusion protein from cell lysates of cultures incubated in YPD to mid-log phase (Figure 4A). The protein we detected when both Pka1 and Rim20 were wild-type had a molecular weight of approximately 120kD (Figure 4A). In contrast, expression of the identical Gfp-Rim101 fusion protein in the pka1 or rim20Δ mutant backgrounds resulted in a protein band of approximately 140kD, which is the predicted size of the full length fusion protein. The Gfp-Rim101-S773A protein also migrated with a reduced electrophoretic mobility in the rim101Δ or rim20Δ backgrounds, resulting in an approximately 140kD band. The 120kD processed form of Gfp-Rim101 was dependent on both PKA and Rim20. Treatment with lambda phosphatases did not alter the mobility of any of these bands (data not shown). We also observed multiple smaller bands in the pka1Δ and rim20Δ mutant backgrounds. These may represent degradation products, suggesting that both Pka1 and Rim20 are necessary to prevent aberrant proteosomal involvement [53]. In addition, the strains with a 140kD Gfp-Rim101 protein were the same strains that had both cytoplasmic and nuclear patterns of Gfp fluorescence. Only the 120kD processed form had predominantly nuclear localization (Figure 3).
A. nidulans PacC undergoes two successive cleavage events that regulate its function as an alkaline-responsive transcription factor [42],[43],[53],[54]. To examine whether C. neoformans Rim101 also undergoes a second cleavage event under activating conditions, we incubated the rim101Δ +Gfp-RIM101 strain to mid-log phase in either YPD or capsule inducing media. When incubated in DMEM, we detected an additional band at approximately 70kD, corresponding to a potentially cleaved N-terminal fragment of the Rim101 protein (Figure 4B). This band was not present when PKA, Rim20, or the PKA phosphorylation consensus sequence were mutated, or under non-inducing conditions, demonstrating that PKA phosphorylation and Rim20 are necessary for this further cleavage of Rim101 under inducing conditions.
To define the downstream targets of the Rim101 transcription factor, we performed comparative transcriptional profiling between the rim101Δ mutant strain and wild-type using whole genome microarrays. We confirmed these results for several genes using quantitative real-time PCR (data not shown). These results indicate that Rim101 controls the transcription of many genes involved in several categories of cellular function (Table 2).
Under capsule-inducing conditions, we were able to document differential expression for a limited number of genes that may be involved in capsule biosynthesis. We observed a 2.9-fold greater expression of UGD1 in the wild-type compared to the rim101Δ mutant strain. UGD1 encodes a UDP-glucose dehydrogenase that is necessary for UPD-glucuronic acid synthesis and thus capsule biosynthesis [55],[56]. A mannosyltransferase (Cmt1) was 2.97-fold greater expressed in the rim101Δ mutant strain than the wild-type. Mannosyltransferases such as Cmt1 have been implicated in the biosynthesis of GXM [57]. In addition, we observed a 3.2-fold decrease in expression of a phosphomannomutase gene (PMM) in the rim101Δ mutant strain. PMM is involved in the biosynthesis of GDP-mannose, another nucleotide sugar essential for capsule production, and is transcriptionally regulated by PKA [24], [58]–[61]. The data also show a small number of other nucleotide sugar-related genes that are differentially expressed and may be involved in capsule production. The fact that many highly inducible capsule genes are not transcriptionally regulated by Rim101 is consistent with our observation that the capsule defect is due to adherence, not production.
Transcriptional profiling also suggested that Rim101 controls the expression of several genes involved in iron or metal homeostasis, including the iron transporter gene CFT1, siderophore importer gene SIT1, and copper transporter gene CTR4 [24], [60]–[62]. In addition, we documented Rim101-dependent expression of homologues of S. cerevisiae iron permeases (FRP1) and reductases (FET3), which are known to be regulated by PKA and by the Cryptococcus transcription factor Cir1 [60],[61],[63],[64]. To demonstrate that decreased expression of these genes was biologically relevant, we incubated the rim101Δ mutant strain in low iron medium, and we observed a distinct growth defect compared to wild-type or the reconstituted RIM101 strain (Figure 5A). C. neoformans strains typically induce capsule in response to growth in this medium. Although the rim101Δ mutant strain grew slowly in low iron media, it eventually reached saturation phase. However, even when grown to saturation, the rim101Δ mutant strain did not exhibit capsule in low iron media (data not shown). Further analysis of these iron homeostasis genes revealed that the promoters of all of these iron regulating genes contain the potential Rim101 consensus binding sequence GCCAAG or the diverged sequence CCAAGAA, recognized by the S. cerevisiae, C. albicans and A. nidulans Rim101 orthologs [7],[27],[65]. These results indicate that C. neoformans Rim101 retains conserved roles in regulating iron homeostasis and import.
The ability to produce capsule is important for the pathogenicity of C. neoformans, and other capsule-deficient strains are severely attenuated for virulence in animal models of cryptococcosis [33], [35], [66]–[68]. In addition, the ability to obtain iron from the host and to grow in low iron conditions is important for microbial survival in the host [60],[63],[69]. We therefore hypothesized that the hypocapsular rim101Δ mutant would be avirulent in animal models of cryptococcosis. However, a recent manuscript in which the investigators tested virulence properties in a large collection of C. neoformans mutants suggested that the rim101Δ mutant strain might be more virulent than wild-type [39]. We therefore tested the role of Rim101 in C. neoformans pathogenicity. Female A/Jcr mice (10 per strain) were inoculated intranasally with 5×105 CFU of the wildtype, rim101Δ mutant, or rim101Δ+RIM101 complemented strains. Mice were monitored for survival and sacrificed at predetermined clinical endpoints predicting mortality (Figure 6A). Infection with either the wild-type or the rim101Δ+RIM101 complemented strain resulted in complete mortality 18 and 19 days after infection respectively; there was no statistically significant difference between the survival of these two groups (p = 0.13). Mice infected with the rim101Δ mutant strain succumbed to the infection 16 days post-infection; this represents a statistically significant decrease in survival compared to animals infected with the wild-type strain (p<0.002). We repeated the infection and sacrificed mice on day 2, day 9 and day 14 to determine fungal burden in the lungs, spleen, and brain. In all organs, we found no statistically significant difference in rates of dissemination among the 3 inoculated strains.
The subtle but reproducible increased virulence of the rim101Δ mutant cells in the inhalation model of C. neoformans infection may be due to enhanced survival in the acidic environment of the alveolar macrophage. We specifically tested intracellular survival of the wild-type, rim101Δ mutant, and rim101Δ+RIM101 complemented strains. As described previously, we co-cultured C. neoformans strains with IFNγ- and LPS-activated J774.1 macrophage-like cells [70]. There was no significant difference in the phagocytosis index of these strains by the macrophages, signifying that the altered capsule in the rim101Δ mutant did not affect fungal cell uptake into macrophages [71]. In contrast, the rim101Δ mutant cells demonstrated increased intracellular survival (p<0.004) within macrophages when normalized against the wild-type (Figure 6B).
Microbial pathogens use varied adaptive mechanisms to survive the harsh conditions of the infected host. Cryptococcus neoformans creates a polysaccharide capsule in response to host conditions such as low iron and high CO2 concentrations [1],[24]. The C. neoformans genome contains a number of genes involved in the biosynthesis of this capsule, and many of these genes are highly transcriptionally regulated, at least partially in response to the PKA pathway [38]. This led us to screen through the genome for transcription factors that are potentially regulated by PKA, and we previously found that the Nrg1 protein regulates capsule. Deletion of the NRG1 gene resulted in a partial capsule reduction, and mutation of the putative PKA phosphorylation consensus sequence prevented full capsule induction. However, not all of the transcriptionally regulated capsule genes appeared to be targets for Nrg1, and many of the nrg1Δ mutant phenotypes were not as severe as mutations in the more upstream components of the cAMP pathway [37]. Therefore, we hypothesized that several transcriptional regulators would control capsule gene induction in response to the PKA pathway. Using a combination of bioinformatic and phenotypic screening, we identified the C. neoformans Rim101 protein as another potential novel PKA-dependent transcriptional regulator of capsule genes.
We hypothesized that the C. neoformans Rim101 protein may be a target of direct PKA phosphorylation due to the presence of a consensus sequence for PKA phosphorylation at amino acid positions 730–736. In contrast, the previously described C. albicans and S. cerevisiae Rim101 proteins do not contain potential PKA phosphorylation consensus sequences. However, there are multiple ways in which PKA can regulate downstream targets, including indirect activation of upstream regulatory proteins as well as by occupying the chromatin of the target genes [72]. Our bioinformatic approach, therefore, does not identify all of the targets of PKA, but does allow us to potentially identify direct targets of PKA phosphorylation.
To determine the relationship between Rim101 and PKA, we used complementary genetic, biochemical, and protein localization experiments. Our results suggest that PKA and Rim20 are necessary for maintenance of Rim101 nuclear localization by altering the cleavage of this transcription factor. Rim20 has been previously implicated in the first cleavage of Rim101, by binding to PEST domains, which are also present in C. neoformans Rim101 [43],[45]. In contrast to the predominantly nuclear localization of Rim101 in wild-type cells, we observed both nuclear and cytoplasmic localization of this protein in the pka1 and rim20 mutant strain backgrounds. We also observed both nuclear and cytoplasmic localization of the Rim101-S773A mutant protein with a putative PKA phosphorylation consensus sequence mutation. In addition, the GFP-tagged Rim101 protein in all of these strains had decreased electrophoretic mobility when compared to the rim101Δ+Gfp-RIM101 strain. The larger band is not due to hyperphosphorylation as this mobility shift was not reversed by treatment with phosphatase. Together, these data indicate that both the cAMP/PKA pathway and the Rim pathway are involved in C. neoformans Rim101 processing and cellular localization.
In Aspergillus and Candida, PacC/Rim101 is activated by two cleavage events, first mediated Rim20 and Rim13 and second by the proteosome [43],[45],[53]. We demonstrate that C. neoformans Rim101 activation may also occur in response to two protein cleavage events, as the Rim101 protein is further cleaved from the 120kD form to a 70kD form in capsule inducing conditions. This further cleavage was not observed when PKA1 or RIM20 were disrupted, suggesting that the initial cleavage to 120 kD is necessary for further processing and activation of Rim101. The multiple smaller bands/laddering observed when PKA1 or RIM20 are disrupted may indicate altered proteosome-mediated processing events, suggesting that both Pka1 and Rim20 are necessary to cause appropriate proteosomal involvement and maintain the balance between processing and degradation. This is consistent with data from A. nidulans, where PacC is first converted by PalB and PalA under alkaline conditions to a 53kD intermediate which exposes the second processing site to the proteosome [53]. Hervas-Aguilar et al. also demonstrated that phosphorylation can accumulate on the 72 and 53kD PacC intermediates during alkaline conditions and affect processing. This is consistent with our results that PKA is involved in regulating processing of Rim101 in C. neoformans, although there is large divergence in the C-terminal and in the potential signaling motifs between these orthologous proteins in these distantly related species. Interestingly, capsule-inducing conditions are not alkaline and thus are not a traditional activating condition for Rim101 proteins. Therefore, CnRim101 may have acquired novel activating conditions in order to respond to the specific host conditions experienced by cryptococcal cells in vivo.
When we examined the targets of Rim101 transcriptional activation, we found that many Rim101 downstream targets and responses from other pathogenic fungi, such as C. albicans, are conserved in C. neoformans. We demonstrated that CnRim101 is important for growth under alkaline conditions in vitro. Using comparative transcriptional profiling, we determined that ENA1, a known downstream target of Rim101 in other fungal species, showed decreased expression in the rim101Δ mutant strain (Table 2). The promoter of the ENA1 gene also had a conserved predicted Rim101 binding sequence, suggesting that it might be a direct target of Rim101 in C. neoformans, unlike in S. cerevisiae, where Rim101 regulates ENA1 through Nrg1 [37]. Idnurm et al. showed that Ena1 is required for C. neoformans survival under alkaline conditions, and that appropriate response to alkaline conditions is necessary for virulence of C. neoformans [73]. Therefore, decreased expression of ENA1 in the rim101Δ mutant strain may explain the defect in alkaline growth of the rim101Δ mutant.
Extracellular pH is involved in regulating iron uptake genes through the Rim101 pathway in C. albicans and S. cerevisiae [3], [5]–[7],[15],[23],[25]. The relationship between iron homeostasis and Rim101 is also conserved in C. neoformans. In order to determine the mechanism for the rim101Δ mutant strain sensitivity to low iron, we compared the transcriptional profile between wild-type and the rim101Δ mutant strain after incubation in capsule-inducing conditions. Our microarray analysis concluded that a number of iron homeostasis genes are differentially regulated between the rim101Δ mutant and the wild-type and we confirmed these alterations in gene expression using quantitative real-time PCR. When we examined the putative promoter regions of the candidate genes, we discovered potential Rim101 consensus binding sequences in CFT1, FET3, and SIT1 among others, suggesting these genes are direct targets of Rim101. Similarly, in C. albicans, Rim101 binds directly to the promoter region of the ferric reductase genes FRE1 and FRP1 to cause increased transcription under iron-limited environments [23].
In C. neoformans, iron uptake is regulated by two pathways: PKA and Cir1. Transcriptional profiling showed that many iron genes, such as the iron permease Cft1 and reductase Cfo1 are differentially regulated by PKA [24],[26],[60],[61]. We have demonstrated that Rim101 is regulated by PKA, thus providing a mechanism for PKA regulation of these iron genes. However, in our transcriptional profiling, we did not demonstrate any difference in expression of Cir1 in the rim101Δ mutant strain, further suggesting that there are two pathways that regulate iron homeostasis. In C. albicans, two signaling pathways regulate iron homeostasis in response to different forms of iron limitation. In C. albicans, the ferric reductase gene FRP1 is differentially regulated by Rim101 and by CBF transcription factors in response to different forms of iron limitation [23]. It is possible that C. neoformans has a similar set of transcription factors to regulate the expression of these iron homeostasis genes under different iron-limiting environments, and that the cell uses both Cir1 and Rim101 to regulate the expression of Cft1 under different environmental stimuli and iron source limitations.
Despite the decreased surface capsule observed in the rim101Δ mutant cells when stained with India ink, this strain was still able to secrete glucuronoxylomannan (GXM) at a similar size and concentration as wild type when the cells were grown in capsule inducing conditions. This data does not preclude other differences in structure and modifications to the GXM in the mutant strain. In accordance with the amount of secreted polysaccharide from the rim101Δ mutant strain, our transcriptional profiling revealed that few capsule biosynthesis genes are transcriptionally regulated by Rim101. In the rim101Δ mutant strain we observed decreased expression of UDP-glucose dehydrogenase Ugd1, mannosyltransferase Cmt1, and phosphomannomutase [55]–[58]. Unlike the iron uptake genes, these capsule biosynthesis genes do not have conserved Rim101 binding sites in the promoter regions, suggesting that these are not direct targets of Rim101. Therefore, our data indicates that CnRim101 is required for the transcriptional activation of some genes involved in capsule biosynthesis; however, the most important effects of Rim101 on capsule are likely due to changes in polysaccharide binding to the cell surface. We hypothesize that Rim101 regulates capsule by altering the expression of genes responsible for anchoring capsule to the cell wall, rather than acting as a direct regulator of these capsule biosynthesis genes.
Unexpectedly for an acapsular strain, the rim101Δ mutant displayed no attenuation in virulence in the mouse inhalation model of cryptococcosis. This confirms prior broad screening experiments of C. neoformans mutants to identify genes required for survival within mice [39]. In these studies, the rim101Δ strain was slightly more virulent than wild-type, as we demonstrated here. Follow-up experiments determining fungal load in the brain, lung, and spleen showed no defects in dissemination. When Rim101 is mutated in Candida, the resulting strains are avirulent as Rim101 regulates processes necessary for fungal virulence [3],[14],[19]. In a fungal pathogen of plants, Fusarium oxysporum, a rim101Δ mutant strain is more virulent than wild-type due to the derepression of acid response genes conferring a survival advantage in the acidic host environment of the tomato [2]. Similarly, our data indicates that the C. neoformans rim101Δ mutant grows better than wild-type within the acidic phagolysosome of the activated macrophage [29],[74]. Perhaps the derepression of acid responsive genes in the rim101Δ mutant could explain the increased growth within the acidic phagolysosome and thus within the lungs of the infected host. Another explanation for the retained virulence of the rim101Δ mutant strain is that the capsular polysaccharide may be shed into the surrounding tissues. This capsular material has well defined immunosuppressant effects. Capsular polysaccharide has even recently been used as an experimental therapy for autoimmune diseases such as rheumatoid arthritis [75]. Therefore, the retained virulence may be attributed to the profound immunomodulatory effects of strains that produce and secrete large amounts of capsule. Also, not all capsule-defective C. neoformans strains are hypovirulent in model systems. The acapsular ags1Δ mutant is fully virulent in the nematode model of cryptococcosis, although sensitive to temperature and thus avirulent in the mouse [49],[50]. The virulence of these strains suggests that capsule may be playing an important role in suppressing the immune system, even when not bound to the cell as an anti-phagocytic mechanism.
It is also possible that the hypocapsular rim101Δ mutant may present an altered cell surface for immune recognition, exposing different antigens resulting in a substantively different immune response than for an encapsulated WT strain. In this model, the increased virulence might result from alterations of the exposed C. neoformans surface antigens leading to over-stimulation of the immune system, such as seen in the response to β-glucan in the C. albicans cell wall [76],[77]. In our microarray data we observed increased expression of MP88 and MP98, two immuno-dominant mannoproteins, in the rim101Δ mutant strain, further supporting a model of an altered antigen surface on the fungus as a result of absent Rim101 activity [78],[79]. MP88 has also been documented as having increased expression in a pka1Δ mutant strain, which may be due to decreased Rim101 activity [60]. A more detailed evaluation of the nature of this cellular infiltration into the infected lungs will help define the varied immune response to different C. neoformans strains.
In summary, we have demonstrated that the C. neoformans Rim101 transcription factor retains conserved functions with orthologous proteins from other fungal species, such as regulation of pH response, cell wall formation, and iron homeostasis. However, the phenotypic output resulting from a C. neoformans Rim101 mutation supports the hypothesis that this conserved protein has been co-opted for unique, species-specific function. In contrast to other fungal species such as Candida or Aspergillus that have adapted to the neutral/alkaline pH of the host lungs and use Rim101 as an inducing signal for virulence, C. neoformans may be better adapted for acidic microenvironments in the host, such as the macrophage phagolysosome. Moreover, our experiments demonstrating PKA regulation of CnRim101 further suggests that conserved signaling elements can be regulated in novel ways to allow adaptation of microorganisms to specific niches in the environment of the infected host.
Cryptococcus neoformans strains used in this study are listed in Table 1. All C. neoformans strains were created in the H99 strain background unless otherwise stated. Strains were maintained on YPD (1% yeast extract, 2% peptone, 2% glucose), a standard yeast medium. Selective media contained nourseothricin (100 mg/L Werner BioAgents, Jena-Cospeda, Germany) or neomycin (G418) (200 mg/L Clontech, Takara-Bio Inc.). Capsule inducing medium (Dulbecco's modified Eagle's media with 25 mM NaHCO3) was prepared as previously described [31]. YNB media (0.67% yeast nitrogen base without amino acids, 2% glucose) was prepared as previously described [33]. Alkaline pH media were created by buffering YNB with 25 mM NaMOPS and adjusting to target pH with NaOH. Resistance to hydrogen peroxide was tested by disc diffusion as described previously [28]. Niger seed agar was prepared from 70 g Niger seed extract (Niger seed pulverized and boiled for 15 min and filtered through cheesecloth) and 4% Bacto agar as previously described [33].
Standard techniques for Southern hybridization were performed as described [80]. C. neoformans genomic DNA for Southern blot analysis was prepared using CTAB phenol-chloroform extraction as described [81].
The wild-type RIM101 gene (NCBI GeneID CNH0097) was mutated using biolistic transformation and homologous recombination with a rim101::nat mutant allele in which the entire RIM101 coding region was precisely replaced with the nourseothrycin resistance gene (nat) [82],[83]. The rim101::nat mutant allele was created using PCR overlap extension as described [84]. Several rim101Δ mutants from independent transformation events displayed identical phenotypes in vitro; therefore one strain (TOC2) was chosen as the rim101 strain for the presented experiments. Putative deletion strains were confirmed by PCR and Southern blot analysis. A second rim101Δ strain was made by creating an identical rim101Δ disruption construct first created by Liu et. al [39] and biolistic transforming it into the H99 strain background. This strain has a partial deletion of the RIM101 gene. Putative mutant strains were confirmed by PCR and Southern blot analysis, and phenotypically compared to the full TOC2 deletion strain.
To reconstitute the wild-type allele, the RIM101 locus was amplified from the H99 wild-type strain using primers (5′ CTGTATCCTTCACTTGAGGC 3′) and (5′ AGCTGTGCGTATCCAATAAT 3′). The neomycin resistance allele was also amplified separately using the M13 forward and reverse primers and both alleles were transformed using biolistic transformation into strain TOC2 to make the reconstituted strain TOC4 as described previously [85]. The reconstituted strain was tested by PCR for presence of the wild-type allele, and examined for reversion of the mutant phenotypes.
The wild-type RIM20 allele (NCBI GeneID CNG00250) was similarly mutated with a rim20::nat mutant allele created by PCR overlap extension. Several rim20Δ mutants from independent transformation events displayed identical phenotypes in vitro; therefore one strain (TOC14) was chosen as the rim20 strain for the presented experiments. This strain was confirmed using PCR and Southern analysis.
We created a green fluorescent protein (Gfp)-Rim101 fusion protein to examine the subcellular localization of Rim101. A histone H3 promoter-GFP fusion [86] was cloned into the neomycin-resistance containing plasmid pJAF, utilizing BamHI and EcoRI to make the resulting plasmid, pCN50. The coding region and the terminator sequence of RIM101 was amplified using primers modified with BamHI sites (5′-AGTTAGGATCCATGGCTTACCCAATTCTCCC-3′ and 5′-ACTGATGGATCCGAGGAAAGCGTCAAGGATATG-3′). The RIM101 gene was then cloned into the pCN50 plasmid at the BamHI site to create the plasmid pTO2 in which the Gfp-Rim101 fusion protein is constitutively expressed under the His3 promoter. pTO2 was then biolistically transformed into C. neoformans strain TOC2, JKH7, CDC7, and TOC17 as previously described to create strains TOC10, TOC12, TOC13 and TOC21 respectively. pTO2 was mutated into pTO3 using PCR-mediated site-directed mutagenesis to change serine 773 to alanine, using primers 5′-GAGAGTGATGCCGCACGTCGATACTGTCCTG-3′ and 5′- AGTTAAGATCTATGGCTTACCCAATTCTCCC-3′. pTO3 was then biolistically transformed into TOC2, CDC7 and TOC17 to create strains TOC18, TOC20 and TOC22 respectively.
Bright field, differential interference microscopy (DIC) and fluorescent images were captured with a Zeiss Axio Imager.A1 fluorescent microscope equipped with an AxioCam mrM digital camera. Confocal images were captured using a Zeiss LSM inverted confocal microscope with the Argon/2 488 laser at×100 magnification. To visualize capsule, cells were grown in inducing conditions (described above), then stained with India ink on glass slides. Images were collected at ×63 magnification. To visualize GFP, cells were washed three times in PBS, and images were collected at ×63 magnification, using 488 nm wavelength for fluorescence. The strains exhibited significant artifactual fluorescence signal when fixed and DAPI stained, therefore all fluorescence images were taken without fixing, precluding DAPI staining.
Estimation of shed capsule polysaccharide size and amount was performed using a technique described by Yoneda and Doering [47]. Conditioned media was made by growing strains in Dulbecco's modified Eagle's medium or low iron medium for 1 week at 30°C with shaking. The cells were incubated at 70°C for 15 minutes to denature enzymes, then pelleted for 3 minutes at 1500 rpm. The resulting supernatant was sterile filtered using a 0.2u filter. 15 uL of conditioned media was mixed with 6x loading dye and run on an agarose gel at 25V for 15 h. The polysaccharides were then transferred to a nylon membrane using Southern blotting techniques. The membrane was air-dried and blocked using Tris-Buffered Saline-Tween-20 (TBS-t) with 5% milk. To detect the polysaccharide, the membrane was incubated with monoclonal antibody 18b7 (1/1000 dilution) [87], washed with TBS-t, then incubated with an anti-mouse peroxidase-conjugated secondary antibody (1/25,000 dilution, Jackson Labs) and detected using SuperSignal West Fempto Maximum Sensitivity Substrate (ThermoScientific).
Protein extracts were obtained using a method previously described [88]. Briefly, cells were incubated to an optical density at 600 nm of 1 in YPD and capsule-inducing conditions. Twenty-milliliter samples of growing cells were pelleted and resuspended in 0.5 mL of lysis buffer containing 2x protease inhibitors (Complete, Mini, EDTA-free; Roche) and 2x phosphatase inhibitors (PhosStop; Roche). To lyse the cells, the supernatant was removed and the cells were lysed by bead beating (0.5 mL of 3uM glass beads in a Mini-BeadBeater-16 (BioSpec), 4 cycles for 30 s each). Following lysis, the samples were immunoprecipitated using 1.6µg anti-GFP antibody (Roche) for 1 hour, then rotated with 80µL protein G Sepharose (Thermo Scientific) for 1 hour. After washing, the samples were eluted by the addition of 40µL 5x Laemmli sample buffer and boiling. Western blots were performed as described previously, using NuPAGE Tris-Acetate gels or NuPAGE Bis-Tris gels to separate the samples. To detect the GFP-labeled proteins, the blots were incubated with anti-GFP primary antibody (1/5,000 dilution) and an anti-mouse peroxidase-conjugated secondary antibody (1/25,000 dilution, Jackson Labs). As a control for non-specific immunoprecipitation, samples were tested in a mock IP (in which no antibody was added) and probed with the anti-GFP antibody, and no bands were observed in this control experiment.
Strains were incubated to mid-logarithmic phase in YPD then washed three times with sterile water before incubation in capsule-inducing media for 3 hours. Cells were washed three times before centrifugation and freezing on dry ice and lyophilizing. RNA was prepared from lyophilized samples using the RNeasy kit (Qiagen). cDNA for real time-PCR was generated using RETROscript (Ambion) using oligo-dT primer. Quantitative real-time PCR was performed as previously described, using the constitutive GPD1 gene to normalize the samples [37].
The microarray used in this study was developed by the Cryptococcus Community Microarray Consortium with financial support from individual researchers and the Burroughs Wellcome Fund (http://genome.wustl.edu/services/microarray/cryptococcus_neoformans). RNA labeling and hybridization were performed by the Duke University Microarray Core Facility according to their established protocols for custom spotted arrays [37],[89]. Data was analyzed using JMP genomics (SAS institute, Cary NC) and initial background subtraction was performed. We used ANOVA normalization and FDR analysis to calculate differences between treatment effects for pairs of inducing conditions. Genes were considered for further evaluation if they showed log2-transformed fold changes with a p-value <0.02.
The virulence of the C. neoformans strains was assessed using a murine inhalation model of cryptococosis as described previously [70]. 10 female A/Jcr mice were inoculated intranasally with 5×105 C. neoformans cells of wild-type (H99), rim101Δmutant (TOC2), or rim101+RIM101 complemented strain (TOC4). Mice were monitored daily for signs of infection and were sacrificed at predetermined clinical endpoints predicting mortality.
The statistical significance between the survival curves of all animals infected with each strain was evaluated using the log-rank test (JMP software, SAS institute, Cary NC). Cell counts were analyzed using Student's t-test. All studies were performed in compliance with the institutional guidelines for animal experimentation.
Survival within alveolar macrophage-like J744.1 cells was tested by aliquoting 50 ul of 1×105 macrophage cells/mL into wells in a 96-well plate. The cells were activated by adding LPS and INFγ and incubating overnight. 50 ul of 2×106cells/mL of each cryptococcal strain were added to the wells and co-incubated for 1 hour after which the excess cells were removed and fresh media was added. The engulfed cells were incubated for 24 hours then disrupted with 0.5% SDS for 5 minutes to lyse the macrophages. The media was removed, and the well was washed 2× with 100 uL PBS. The washes were combined and diluted at 1∶100 before plating. Plates were incubated for 2 days before colony counts [70].
|
10.1371/journal.pntd.0006129 | Expanding molecular diagnostics of helminthiasis: Piloting use of the GPLN platform for surveillance of soil transmitted helminthiasis and schistosomiasis in Ghana | The efforts to control and eradicate polio as a global health burden have been successful to the point where currently only three countries now report endemic polio, and the number of cases of polio continues to decrease. The success of the polio programme has been dependant on a well-developed network of laboratories termed the global polio laboratory network (GPLN). Here we explore collaborative opportunities with the GPLN to target two of the 18 diseases listed as a neglected tropical diseases (NTD) namely soil transmitted helminthiasis (STH) and Schistosomiasis (SCH). These were chosen based on prevalence and the use of faecal materials to identify both polio, STH and SCH. Our study screened 448 faecal samples from the Ghana GPLN using three triplex TaqMan assays to identify Ascaris lumbricoides, Necator americanus, Ancylostoma spp, Trichuris trchiura, Strongyloides stercoralis and Schistosoma spp. Our results found a combined helminth prevalence of 22%. The most common helminth infection was A. lumbricoides with a prevalence of 15% followed by N. americanus (5%), Ancylostoma spp. (2.5%), Schistosoma spp. (1.6%) and S. stercoralis (1%). These results show that it is possible to identify alternative pathogens to polio in the samples collected by the GPLN platform and to introduce new diagnostic assays to their laboratories. The diagnostic methods employed were also able to identify S. stercoralis positive samples, which are difficult to identify using parasitological methods such as Kato-Katz. This study raises the possibility of collaboration with the GPLN for the surveillance of a wider range of diseases which would both benefit the efforts to control the NTDs and also increase the scope of the GPLN as a diagnostic platform.
| The successful campaign being waged against polio has eliminated the disease from most countries where it was once endemic. With this success, it is anticipated that the disease will be eradicated in the coming years with only 37 cases being reported in 2016. Although the efforts to control polio are successful there are a number of low-profile, but no less serious disease, that are still highly prevalent throughout the world. These diseases have been termed the neglected tropical diseases (NTD) and this study aims to test the suitability of the Global Polio Laboratory Network (GPLN) as a platform to screen for two of the NTDs, soil transmitted helminthiasis (STH) and schistosomiasis (SCH). To test the suitability of the samples collected by the GPLN and the suitability of the laboratories themselves 448 samples from the Ghanaian GPLN laboratory were screened with multiplex TaqMan assays for the following six helminth types: Ascaris lumbricoides, Necator americanus, Ancylostoma spp, Trichuris trchiura, Strongyloides stercoralis and Schistosoma spp. Using this method this study was able to identify a prevalence of 22% for the combined helminth infection. The most common infection was A. lumbricoides with a prevalence of 15% followed by N. americanus (5%), Ancylostoma spp. (2.5%), Schistosoma spp. (1.6%) and S. stercoralis (1%). The success of this study indicates that this may be a cost-effective method to passively screen a country for STH and SCH and its success in identifying S. stercoralis infections makes it especially useful as this parasite is hard to identify using traditional surveillance techniques.
| In 1988 the WHO set out to eradicate polio after the successful development of effective polio vaccines and since then the eradication campaign has reduced the number of countries reporting endemic polio from 125 to three in 2016. Control of polio has been a co-ordinated effort involving two main arms; the delivery of vaccination alongside establishing an effective laboratory network for monitoring and surveillance. This surveillance arm is comprised of 145 labs spread throughout the world which taken as a whole forms the Global Polio Laboratory Network (GPLN). The network receives samples from local health clinics where individuals have presented with clinical signs of the disease, typically acute flaccid paralysis (AFP), with the need to confirm or exclude an aetiology of polio. Thus, investigation of AFP initially involves collection of a faecal sample(s) which is then transferred from the regional clinic thence to the central laboratory to undergo a culture screen for polio virus and if found positive is then followed up with a real-time PCR analysis with diagnostic primers able to identify and discriminate if the sample is wild type, vaccine strain or a vaccine-derived virus. Across Africa there are 16 GPLN labs and these have received, in total, an average of 22,017 samples per year in the past 5 years with Ghana contributing, on average, 350 samples per year. These 16 laboratories are divided into three regional reference laboratories (RRLs) and 13 intratypic differentiation laboratories (ITD). The ITD laboratories are responsible for the isolation of poliovirus, molecular characterization of isolates and referral of critical samples to a sequencing laboratory [1]. Currently faecal samples collected by the GPLN are only screened for polio and non-polio enteroviruses; here we explore the potential of the GPLN to screen for other pathogens of public health importance allowing for co-investigation.
Across the world, but especially in Africa, the Neglected Tropical Diseases (NTDs) are an umbrella group of diseases that afflict the poor and retain a cycle of poverty. In total, approximately a billion people from the poorest communities across the globe are infected with at least one NTD [2]. There are currently 18 diseases listed as NTDs [3] with seven of these diseases caused by parasitic helminths. The intestinal nematodes, often referred to as soil transmitted helminths (STH) contribute the greatest number of infections and highest number of DALYs lost for any NTD [4] closely followed by schistosomiasis (SCH), a waterborne trematode infection [5]. Recently the importance of control of these diseases has been recognised by policy makers and steps have been taken to develop cost effective strategies in managing them [6]. Although there are WHO guidelines for classic parasitological surveillance, there is no equivalent for a molecular diagnostic platform of these infections. A key block in doing so is the cost of setting up a standard surveillance platform, de novo, it would therefore be sensible to expand and strengthen existing surveillance structures. The GPLN is a good example and is maintained with substantial annual investments [7]. Thus being able to augment or ‘piggyback’ appropriate NTD surveillance onto the GPLN could have the necessary ‘kick-starting’ effect to provide better access to diagnostic tools needed for control and elimination of STH and SCH [8–12].
Addressing the need for a molecular diagnostics platform for NTDs, in this investigation we explore and develop synergies and necessary steps with the GPLN, taking advantage of its accumulated experience and resources, to include pilot screening for STH (A. lumbricoides, N. americanus, Ancylostoma spp, T. trchiura, S. stercoralis) and SCH (S. mansoni and S. haematobium). The Ghanaian National Polio laboratory based at the Noguchi Memorial Institute for Medical Research was selected to carry out this assessment determining the suitability of the GPLN faecal collections with multiplex TaqMan diagnostic assays.
Ethics applications were approved by LSTM (Research protocol 16-007) and Noguchi Scientific and Technical Committee (Study number 065/16-17) followed by the Institutional Review Board. To obtain approval, initial patient collection forms were amended to later facilitate expanded diagnostic testing with the results made available to the national NTD programme. All participants were anonymised for the final study.
Prior to the work being carried at the Ghanaian GPLN a workshop was carried out to train the staff in the methods used to extract DNA from faecal samples and the subsequent optimisation and running of the qPCR TaqMan assays. The workshop included both practical and theoretical training, this gave the staff of the GPLN a good background knowledge and practical experience in the methods they would use [13, 14].
The samples used in this study were faecal samples sent to the Ghanaian GPLN laboratory from individuals presenting with acute flaccid paralysis. The samples were sent via courier from local health clinics and were kept at 4°C until it reached the GPLN laboratory at which point they were stored at -20°C. The age of the patients that supplied the sample as well as the district from which it originated from were available to this study.
Faecal samples were removed from the -20°C freezer and allowed to defrost at room temperature, once defrosted ~0.1g of faeces was removed and placed into a 2mL screw cap sample tube that was preloaded with 0.9g of 1.4mm ceramic beads. To this 250μL of a 2% PVPP/PBS suspension was added and the sample vortexed for 5-10 seconds. The faecal suspension was then frozen at -20°C overnight. The following day the samples underwent bead-beating at 3000rpm for 30 seconds using the MagnaLyser system. DNA extraction was carried out using the QIAamp DNA Mini kit per the manufacturer’s instructions with the following two modifications: i) an aliquot of phocine herpes virus-1 was added to the AL buffer to act as an internal positive control for the subsequent TaqMan assays, ii) the DNA was eluted in 200μL of nuclease free water. As well as introducing Phocine Herpes Virus (PhHV) into each sample to act as an internal positive control a DNA extraction negative control was introduced after every 47th sample, in total 448 samples were processed [15].
The six helminth types were screened using previously described primers and probes [8–10, 12, 16, 17] and these were used in three triplex reactions, each targeting two helminth types and the internal positive control. The first of these targeted S. stercoralis and N. americanus; the second targeted Ancylostoma spp. and a generic Schistosoma spp. (S. mansoni, S. haematobium, S. intercalatum); the third triplex reaction targeted A. lumbricoides and T. trichiura (Table 1). The primer concentrations were determined individually through primer limiting assays and then tested in the final triplex concentrations using mono, double and triple target DNA assays to ensure there was no internal competition within a reaction. The final volume for each triplex reaction was 20μL, consisting of 12.5μL of iQ supermix, 2μL of DNA template and a final helminth primer concentration of 200nM except for Ancylostoma spp. which ran at 300nM; the concentration of all probes and the PhHV primers was 100nM. All assays were processed using the same ABI 7500 qPCR thermocycler. Each qPCR run consisted of the following cycle, an initial holding step at 95°C for three minutes followed by 50 cycles of 95°C for 15s, 60°C for 30s, 72°C for 30s and a final extension step at 72°C for two minutes.
The 10 regions of Ghana were used to categories the origins of the samples collected by the Ghanaian GPLN. This was then used to determine how evenly across Ghana the origins of the samples were distributed. The region that contributed the most samples was Brong Ahafo, where 23% of the samples originated from following this was the Western region, supplying 15% of the samples. The Ashanti, Central, Greater Accra, Northern, Upper East and Volta regions all contributed a similar amount of between 7-10%. The regions that contributed the least number of samples were the Eastern and Upper West Regions. The distribution of the participants that supplied the samples can be shown to have come from across the country with most regions contributing a similar number of samples.
Due to the anonymity of the samples the sex of the participants was unknown and could not be included as a risk factor, however their age was recorded. It was possible to observe the age range of samples as this would affect the suitability of samples for STH screening. Across the 10 regions the average age ranged from 5 to 6 years and an analysis using ANOVA resulted in a P = 0.16, indicating there was no significant difference in participants age across the six regions of Ghana. Breaking down the ages of participants into pre-school age (PSAC, 0-4 yrs), school age (SAC, 5-16yrs) and adults (17+) the following percentages were found for each group: 60%, 30% and 10% respectively.
The qPCR assay was successful in identifying positive samples for A. lumbricoides, N. americanus, A. duodenale, Schistosoma and S. stercoralis. A total of 102 out of 448 samples were found to be positive for one or more helminth types tested, giving an overall prevalence of 22.7% with 92 of these being single helminth infections and 10 being double infections (Table 2)
The proportion of samples positive for the different helminth types is shown in Fig 1, A. lumbricoides was found to be the most prevalent helminth, being found in 16% of all samples. The two-hookworm species followed with N. americanus found in 6% of samples and Ancylostoma spp. found in 3%. The prevalence of Schistosoma spp. and S. stercoralis was 2% and 1% respectively whilst no samples were found to be positive for T. trichiura.
The distribution of helminth species across the different regions of Ghana varied with Brong Ahafo having the highest proportion of positive samples and the central region having the lowest proportion of positives. The distribution of helminth species across these regions was also not even with the Upper West, Northern and Eastern regions only being positive for two of the helminth types (Fig 2). Other regions contained samples positive for multiple species of helminth.
The purpose of this study is to demonstrate the suitability of adapting a GPLN laboratory for the detection of STH and SCH. It is not within the scope of this study to infer anything from the epidemiological data as the sample size is too small to be representative of the different administrative regions of Ghana. Similarly the samples collected by the GPLN will not be representative of the communities they originate from as they are from individuals that have presented with specific clinical symptoms, notably acute flaccid paralysis.
There were a total of 102 helminth positive samples detected out of a total of 448 samples screened, of which 92 were single infections comprising A. lumbricoides (59), N. americanus (15) and Ancylostoma spp. (10) respectively. Schistosoma spp. (5) was the next most common helminth although surprisingly no cases of Schistosoma spp. were detected in samples from the Volta. S. stercoralis is perhaps the least understood of the intestinal helminths [19] and is difficult to detect with traditional techniques, despite this our study was able to identify five cases of S. stercoralis, three single infections and two co-infections with A. lumbricoides. The total number of samples positive for co-infections was 10 of which half were a co-infection of A. lumbricoides and N. americanus.
The results show that the average age of participants falls between five to six years which means they fall within the SAC age group which is the usual target group for STH and SCH prevalence surveys. The sample contribution from each region varied from 23% to 3% however seven out of the 10 regions contributed a similar percentage of samples. Surprisingly no SCH positives were found in samples from the Volta. The reason for the lack of SCH positives from the Volta region is not yet clear and could be due to insufficient samples from this region, although this is unlikely as regions contributing fewer samples were still found to have SCH positives. An alternative explanation for the lack of Schistosoma s.l. positives is that the method described in this paper is more suited to the detection of S. mansoni than it is for S. haematobium, whose eggs are typically passed in stool samples, whereas S. haematobium predominantly pass their eggs in urine [20].
The scientific community has long acknowledged the likely high burden of disease and morbidity that is caused by S. stercoralis [21]. Current estimates of global infection range from 30 to 100 million [22, 23] although these estimates are based on imperfect sampling techniques and a more recent study has proposed a higher prevalence of 370 million people infected world-wide [24]. This wide range in prevalence estimates highlights the variable reliability of different screening methods. The most common method of screening for helminth infections, Kato-Katz, is poorly suited to the detection of S. stercoralis [25]. The success of the methodology used in this paper to detect S. stercoralis alongside the other STH species and Schistosoma spp. demonstrate the versatility of using qPCR to detect a wider range of helminth infections than more traditional methods. Currently there is no data regarding the distribution of S. stercoralis in Ghana however by screening the samples from the GPLN we were able to identify five samples positive for S. stercoralis.
The current design of the GPLN is not yet suited for its samples to be used to infer the distribution of STH and SCH as the samples are too few in number and are from a specific group within the population, those presenting with clinical signs of polio. The introduces confounding factors and does not provide a representative cross section of the communities at risk, notably adults and school age children were fewer in number than pre-school age children. To become an adequate surveillance platform the range of clinical symptoms for sample collection would need to be widened to include those presenting with STH and SCH symptoms. This would no doubt increase the number of samples being sent in for analysis and subsequently improve the surveillance capabilities of the system. However, this would no doubt incur a greater cost, a possible solution would be to incorporate other pathogens for screening to attract extra funding to cover these costs.
In conclusion, the findings of this study show that it is possible to identify STH and SCH positives in the faecal samples collected by the GPLN and that new diagnostic techniques can be introduced to compliment the work currently being carried out. The current narrow clinical symptoms required to qualify a sample to be sent to the GPLN limits their epidemiological use, a change in sample submission policy would be required to improve their epidemiological relevance. Despite this the study demonstrates a potential way forward in the monitoring and control of NTDs that could be included in the legacy plan of the GPLN.
|
10.1371/journal.pcbi.1003776 | The Fitness Landscape of HIV-1 Gag: Advanced Modeling Approaches and Validation of Model Predictions by In Vitro Testing | Viral immune evasion by sequence variation is a major hindrance to HIV-1 vaccine design. To address this challenge, our group has developed a computational model, rooted in physics, that aims to predict the fitness landscape of HIV-1 proteins in order to design vaccine immunogens that lead to impaired viral fitness, thus blocking viable escape routes. Here, we advance the computational models to address previous limitations, and directly test model predictions against in vitro fitness measurements of HIV-1 strains containing multiple Gag mutations. We incorporated regularization into the model fitting procedure to address finite sampling. Further, we developed a model that accounts for the specific identity of mutant amino acids (Potts model), generalizing our previous approach (Ising model) that is unable to distinguish between different mutant amino acids. Gag mutation combinations (17 pairs, 1 triple and 25 single mutations within these) predicted to be either harmful to HIV-1 viability or fitness-neutral were introduced into HIV-1 NL4-3 by site-directed mutagenesis and replication capacities of these mutants were assayed in vitro. The predicted and measured fitness of the corresponding mutants for the original Ising model (r = −0.74, p = 3.6×10−6) are strongly correlated, and this was further strengthened in the regularized Ising model (r = −0.83, p = 3.7×10−12). Performance of the Potts model (r = −0.73, p = 9.7×10−9) was similar to that of the Ising model, indicating that the binary approximation is sufficient for capturing fitness effects of common mutants at sites of low amino acid diversity. However, we show that the Potts model is expected to improve predictive power for more variable proteins. Overall, our results support the ability of the computational models to robustly predict the relative fitness of mutant viral strains, and indicate the potential value of this approach for understanding viral immune evasion, and harnessing this knowledge for immunogen design.
| At least 70 million people have been infected with HIV since the beginning of the epidemic and an effective vaccine remains elusive. The high mutation rate and diversity of HIV strains enables the virus to effectively evade host immune responses, presenting a significant challenge for HIV vaccine design. We have developed an approach to translate clinical databases of HIV sequences into mathematical models quantifying the capacity of the virus to replicate as a function of mutations within its genome. We have previously shown how such “fitness landscapes” can be used to guide the design of vaccines to attack vulnerable regions from which it is difficult for the virus to escape by mutation. Here, using new modeling approaches, we have improved on our previous models of HIV fitness landscape by accounting for undersampling of HIV sequences and the specific identity of mutant amino acids. We experimentally tested the accuracy of the improved models to predict the fitness of HIV with multiple mutations in the Gag protein. The experimental data are in strong agreement with model predictions, supporting the value of these models as a novel approach for determining mutational vulnerabilities of HIV-1, which, in turn, can inform vaccine design.
| The ideal way to combat the spread of HIV-1 is with an effective prophylactic or therapeutic vaccine [1], [2]. One of the greatest challenges hindering the achievement of this goal is the incredible sequence diversity and mutability of HIV-1 [3], which can limit the effectiveness of the immune response [2], [4].
CD8+ T cells are instrumental in reducing viral load in HIV-1 acute infection [5] and in maintaining the viral set point during chronic HIV/SIV infection [6], [7]. However, HIV-1 is able to escape the CD8+ T cell response through mutations in or adjacent to HIV-1 epitopes that are presented by HLA class I molecules on the surface of the infected cells [7]. One proposed strategy for realizing a potent prophylactic or therapeutic vaccine is to target CD8+ T cell responses to conserved regions of HIV-1, aiming to reduce incidences of immune escape or, if escape occurs, to reduce viral fitness and lower the viral set point, thereby slowing disease course and reducing transmission at the population level [8], [9]. While escape mutations at highly conserved sites often damage the viability of virus [10], this approach is confounded by the development of compensatory mutations which restore or partially restore viral fitness [9]. Thus, to maximize the effectiveness of a vaccine-induced immune response one must look beyond conservation of single residues to identify regions where mutations are not only highly deleterious, but where further mutations elsewhere in the proteome are unlikely to restore lost fitness, but rather, lead to additional fitness costs due to deleterious synergistic effects.
Our group has developed computational models to identify such vulnerable regions of the HIV-1 proteome and to predict the fitness landscape of HIV-1 proteins, providing tools for designing vaccine immunogens that may limit both HIV-1 evasion of CD8+ T cell responses and the development of compensatory mutations [11], [12]. In an early qualitative study we identified groups of amino acids in HIV-1 Gag coupled by structural and functional constraints that cause these residues to co-evolve with each other, but evolve nearly independently of the other residues in the protein [12]. In analogy with past studies on the economic markets and enzymes [13]–[15], we termed these groups of residues “sectors”. This analysis and human clinical data revealed one sector in Gag, which we termed sector 3, where multiple mutations were more likely to be deleterious. This group of residues is naturally targeted more by elite controllers [12]. It is expected to be particularly vulnerable to CD8+ T cell responses that target multiple residues in it since multiple mutations within this sector are likely to significantly diminish viral fitness, thereby restricting available escape and compensatory paths [12].
This approach, however, does not allow us to determine precisely which residues should be targeted, as it does not quantify the relative replicative viability of viral strains bearing specific mutations. Nor does it identify viable escape routes that remain upon targeting residues in the vulnerable regions, or inform how best to block them. To begin to address these issues, we developed a computational model, rooted in statistical physics, which aims to predict the viral fitness landscape (viral fitness as a function of amino acid sequence) from sequence data alone and applied it to HIV-1 Gag [11]. Similar methods have previously been employed to study other complex biological systems, from describing the activity patterns of neuronal networks [16]–[19] to the prediction of contact residues in protein families [17], [20], [21].
The idea underlying our approach is to first characterize the distribution of sequences in the population, which we expect to be correlated with fitness (see below). Due to the small number of available sequences compared to the size of the sequence space, direct estimation of the probability distribution characterizing the available sequences is precluded. Thus, we instead aim to infer the least biased probability distribution of sequences that fits the observed frequency of mutations at each site, and all correlations between pairs of mutations (the one- and two-point mutational probabilities). Mathematically, “least biased” implies the distribution that has maximum entropy in the information-theoretic sense [22]. The maximum entropy distribution that fits the one- and two-point mutational probabilities has a form reminiscent of that describing equilibrium configurations of an Ising model in statistical mechanics. We generated such models using multiple sequence alignments (MSA) for the four subunit proteins of Gag in HIV-1 clade B [11] (described in Supporting Information Text S1, Section 1). This model assigns to each viral strain an “energy” (E), which is inversely related to the probability of observing this sequence.
We expect more prevalent sequences to be more fit, consistent with expectations from simple models of evolution [23] though the precise correspondence between fitness and prevalence may have a more complicated dependence on factors such as the shape of the fitness landscape, as predicted by quasispecies theory [24]. Furthermore, this expectation could be confounded by immune responses in the patients from whom the virus samples were collected, and phylogeny. Recent analyses suggest (described more fully in the discussion) that in spite of these effects, at least for Gag proteins, the rank order of prevalence and in vitro replicative fitness should be similar [25]. Strains with high E values are predicted to be less fit than strains with low E values. Predictions of the model seemed to be in good agreement with experimental data on in vitro replicative fitness, as well as clinical observations on the frequency and impact of viral escape mutations [11].
Our aim in the current work is twofold. First, we present new advances in the inference and modeling of viral fitness landscapes that address previous theoretical and computational limitations. Second, we describe new in vitro fitness measurements for viruses containing multiple Gag mutations, performed to further test fitness predictions using the improved computational methods. To give a broad test of the predictive power of the fitness models, we have performed comparisons for HIV-1 strains containing multiple mutations predicted to harm HIV-1 viability as well as combinations predicted to be relatively fitness neutral. We find that fitness measurements of these mutant strains are in good agreement with model predictions.
Our key hypothesis in formulating models of HIV fitness is that the prevalence of viruses with a given sequence, that is, how often the sequence is observed, is related to its fitness. Simply, fitter viruses should be more frequent in the population than those that are unfit. This hypothesis can be proven for some idealized evolutionary models [23], but cannot be made exact for the complicated nonequilibrium host-pathogen riposte between humans and HIV. However, our theoretical work, backed by extensive computational studies, suggests that the rank order of fitness and prevalence of strains should be strongly monotonically correlated, provided we compare sequences that are phylogenetically close [25]. Thus, if we construct a model to predict the likelihood of observing different viral strains with given sequences, it can predict the relative fitness of the strains. We achieved this goal by constructing a maximum entropy model for the probability of observing sequences in the MSA [26]. The simplest model in this class is an Ising model, a simple model of interacting binary variables from statistical physics which has been widely applied to study collective behavior in complex systems. The parameters of this Ising model are obtained by imposing the constraint that it reproduce the pattern of correlated mutations (relative to the consensus sequence) observed in a multiple sequence alignment (MSA) of HIV-1 amino acid sequences extracted from infected hosts. Specifically, the parameters were chosen such that the frequency of mutations at each single residue and the frequency of simultaneous mutations at each pair of residues were the same in both the Ising model and the MSA. Importantly, the model also reproduced higher order mutational correlations accurately, even though these mutational frequencies were not directly fitted [11].
As described in our previous publication [11], in the Ising model amino acid sequences in the MSA are compressed into binary strings by assigning a 0 to each position where the amino acid matches the consensus sequence (“wild-type”), and a 1 to each position with a mismatch (“mutant”). While this binary approximation greatly simplified our modeling approach, the reduction in complexity has several drawbacks. Firstly, there is a loss of residue-specific resolution. The fitness predictions of our model are insensitive to the precise identity of mutant amino acids, and thus the model cannot resolve fitness differences between proteins containing different mutant amino acid residues in a particular position. Secondly, for relatively conserved proteins such as HIV-1 Gag, where the number of viable amino acids at each position is rather limited, this binary simplification represents a reasonable approximation. However, it is less justified for highly mutable proteins where the wild-type residue in each position is not the overwhelmingly most probable amino acid, as is the case for the HIV-1 Env protein.
In our original approach, we fit the Ising model parameters to precisely reproduce the observed one and two-residue mutational correlations within the MSA. However, simultaneous mutations at certain pairs of residues were never observed. This led to another deficiency in our original modeling approach in that pairs of mutations not observed in the MSA were predicted to be completely unviable (E = ∞). While it is possible that such mutant viral strains have exactly zero replicative fitness, it is more likely that they are highly unfit strains (possessing non-zero replicative fitness) that simply arise too seldom to be observed within our finite-sized MSA.
In this work, we present three significant advances of our original model to predict viral fitness, which also the aforementioned limitations. First, we incorporate Bayesian regularization into our fitting procedure to eliminate the prediction of zero replicative fitnesses for mutations not present within our MSA. Second, we implement a new algorithm for inferring an Ising model from sequence data, which dramatically accelerates the computation of model parameters. Third, we relax the binary approximation to infer viral fitness landscapes that explicitly retain the amino acid identities at each position. We achieve this by describing the viral fitness landscape using a multistate generalization of the Ising model known as the Potts model, another established and well-studied model in statistical physics [27]. We also implement Bayesian regularization into the fitting of the Potts model parameters.
Inference of the parameters of the Ising models, commonly referred to as the inverse Ising problem, is a canonical inverse problem lacking an analytical solution that may be tackled in many ways [16], [17], [19]–[21], [28], [29]. We improve upon our previous techniques described in [11] by incorporating regularization and implementing new inference algorithms, which greatly decrease the computational burden and accelerate model fitting.
To control the effects of undersampling and to improve the predictive power of the inferred fitness models, we incorporate Bayesian regularization into our inference algorithm [18], [19], [30], [31] in the form of a Gaussian prior distribution for the model parameters describing pairwise couplings between residues (see Text S1, Sections 1.3 and 2.5). Regularization of this form is also known as Tikhonov regularization or ridge regression [32]. With this addition, the probability of observing any sequence, including those containing pairs of mutations not observed in the MSA, is nonzero. We have also computed a correction to the energy of each sequence to account for the possible bias that strains near fitness peaks are more likely to be observed than would be expected from their intrinsic fitness when sampled from a finite distribution (see Text S1, Section 3.2).
In an algorithmic advance over our previous fitting procedure, we fit the parameters of our regularized Ising model using the selective cluster expansion algorithm of Cocco and Monasson [18], [19] which identifies clusters of strongly interacting sites and iteratively builds a solution for the whole system by solving the inverse Ising problem for each cluster. With this approach, we cut the CPU time necessary to infer the parameters of the Ising model from roughly 12 years [11] to 5 hours for p24, an improvement by four orders of magnitude. Roughly, we expect algorithm run-time to scale as O(Nn exp(n)), where N is the system size (number of amino acids) and n is the size of a typical “neighborhood” of strongly interacting sites. For a review and applications of this method see [18], [30]. Complete details of our modeling approach and numerical fitting procedures are provided in the Text S1, Section 1.
An ideal model of viral fitness would be able to capture the full (unknown) distribution of correlated mutations throughout the sequence, and thus reproduce the prevalence of every viral strain. Sequences in the MSA represent a sample of the possible strains of the virus, providing information about the distribution of point mutations, pairs of simultaneous mutations, triplets of simultaneous mutations, and all higher orders. However, since the number of available sequences in the MSA is very small compared to the size of the accessible sequence space, and because mutations at most sites are rare, higher order mutations will be severely undersampled. Thus, following our previous approach we appeal to the maximum entropy principle to seek the simplest possible model capable of reproducing the single site and pair amino acid frequencies [11], [22], for which the problem of undersampling is less severe. From this analysis, the Potts model is the least structured model capable of reproducing the one and two-position frequencies of amino acids observed within the MSA [21].
To introduce the Potts model, we represent the sequence of a particular m-residue protein as a vector, , where the elements Ak can take on the q = 21 integer values [1, 2, …, 21] denoting an arbitrary encoding of the 20 natural amino acids, plus a gap [21]. In the Potts model the probability of observing a particular sequence is given by(1)
In analogy with the statistical physics literature, we refer to E as a dimensionless “energy,” the function as the Hamiltonian, and the normalizing factor Z as the partition function [27]. The model is parameterized by a set of m q-dimensional vectors, , and a set of m(m−1)/2 q-by-q matrices, . The hi vectors give the contribution of the identity of each amino acid in each position to the overall sequence energy, and the Jij matrices give the contribution to the energy of pairwise interactions between amino acids in different positions.
To fit the Potts model, we implemented a generalization of the semi-analytical extension of the iterative gradient descent implemented by Mora and Bialek [11], [17]. This approach implements a multi-dimensional Newton search to iteratively adjust the model parameters until the predictions of the model for the one and two-position frequencies of amino acids reproduce those observed within the MSA. In an advance over the original incarnation of this algorithm, we have derived closed form expressions for the gradients required by the Newton search, thereby obviating the need for their numerical estimation by finite differences (which would result in a more computationally expensive and less numerically stable secant search procedure). Our approach is semi-analytical in the sense that while we have analytical expressions for the Newton search gradients, we use a Monte Carlo procedure to numerically estimate the one and two-position amino acid frequencies predicted by the model at each stage of parameter refinement. We are currently developing a Potts generalization of the cluster expansion algorithm [19] to accelerate fitting. We incorporate Bayesian regularization into our fitting procedure in a precisely analogous manner to that described above for the regularized Ising model by introducing a Gaussian prior distribution over the Jij parameters. Inference of the Potts model parameters for p24 required approximately 1.4 years of CPU time using a generalization of the gradient descent approach described in Ref. [11]. Fitting the model parameters by gradient descent is expected to scale as O((Np)2), where N is the number of amino acids in the protein, and p is the characteristic number of mutant residues observed at each position. Full details of the fitting and regularization procedures are provided in Text S1, Section 2. The code implementing the inverse Potts inference algorithm is also provided in Supporting Information Code S1.
To test the accuracy of these models in predicting the fitness landscape of HIV-1 Gag, we performed in vitro experiments to measure the fitness of various Gag mutants. Previously we had measured the in vitro replication capacities of 19 Gag p24 mutants, 16 of which contained single mutations in Gag p24, and compared these with fitness predictions of our original Ising model [11]. Here, we extend this work to measure the replication capacities of HIV-1 strains containing various combinations of mutations, predicted to be either harmful to HIV-1 viability or fitness-neutral, in Gag p24 and p17 and we compare measurements not only to the original Ising model described in Ref. [11], but also to regularized versions of Ising and Potts models that we have developed here. Specifically we considered 17 mutations pairs, one triple, and 25 single mutations within these combinations, as listed in Table 1. These mutations were introduced into the widely used laboratory-adapted HIV-1 clade B reference strain NL4-3.
The tested mutants can be divided into 4 categories, viz. (i) Gag p24 pairs with high E values located within a group of co-evolving amino acids termed sector 3 (cf. Ref [12]), (ii) HLA-associated Gag p24 pairs with high E values, (iii) Gag p24 pairs/triple with low E values, and (iv) Gag p17 pairs (Table 1). These mutation combinations were chosen according to E values predicted by the published Ising model [11], where E>90 or E = ∞ were considered high E values and E<15 were considered low E values. Note that, due to the couplings between mutations at different sites, parameterized by the Jij in equation 1, the E values depend not only on the specific mutations introduced but also on the sequence background. The E values for mutations reported here are computed with the HIV-1 NL4-3 sequence background, which differs from the p17 and p24 MSA consensus sequences by 8 mutations (R15K, K28Q, R30K, K76R, V82I, T84V, E93D, S125N) and 2 mutations (N252H, A340G), respectively. The p24 region of Gag was focused on since this is the most conserved region of the protein. First, we selected six mutation pairs, predicted to be unfavorable in combination, in sector 3 of Gag p24 since we previously found this to be an immunologically vulnerable group of co-evolving residues in which multiple mutations are not well-tolerated [12]. Since it is desirable to identify low fitness/non-viable combinations of escape mutations for vaccine immunogen design aimed at reducing viral fitness or blocking viable escape pathways, we aimed to identify pairs of likely escape mutations with high E values. Virus mutations that are statistically associated with the expression of specific host HLA class I alleles, which also restrict the same epitopes in which the mutations are found, are likely to be CD8+ T cell-driven escape mutations [33]. We therefore tested five high E pairs of mutations located at HLA-associated Gag p24 codons (HLA-associated variants defined in [34], [35]) in or next to optimal CD8+ T cell epitopes (A-list epitopes from the Los Alamos HIV sequence database [36]) that were restricted by the same HLA. For comparison with high E mutation pairs, mutation combinations with low predicted E values were included in testing, comprising known favorable compensatory pairs in Gag p24 where 219Q compensates for the 242N escape mutant [37] and 147L compensates for the 146P escape mutant [10], as well as one pair in sector 3 of Gag p24 and a Gag p24 triple mutant. Additionally, for broader testing, two mutation pairs in Gag p17 were selected. We note that the most commonly observed mutant amino acid at each codon was tested.
We introduced these mutation combinations into the HIV-1 NL4-3 plasmid by site-directed mutagenesis and their presence was confirmed by sequencing, as described previously [38]. Generation of mutant viruses from mutated plasmids and the measurement of their replication capacities were performed as previously [11], [38]. Briefly, mutated plasmids were electroporated into an HIV-1-inducible green fluorescent protein reporter T cell line, harvested at ≈30% infection of cells, and the replication capacities of the resulting mutant viruses were assayed by flow cytometry using the same cell line. Replication capacities were calculated as the exponential slope of increase in percentage infected cells from days 3–6 following infection at a MOI of 0.003, normalized to the growth of wild-type NL4-3 (RC = 1). Three independent measurements were taken and averaged. Mutant viruses were re-sequenced to confirm the presence of introduced mutations.
The values of E predicted by our original and new modeling approaches for the 43 HIV-1 NL4-3 Gag mutants tested here are shown in Table 2. Absolute comparison of the E values between the models are not meaningful, but the relative E values of mutants are generally in excellent concordance between models (Pearson's correlation, r≥0.85 and p≤5.3×10−11, two-tailed test).
The in vitro fitness measurements for all mutants, grouped according to categories, are shown in Figure 1. We initially compared our model predictions and fitness measurements for each category of mutant pairs to evaluate whether mutant combinations with high and low predicted E values corresponded to substantial fitness cost or little/no fitness cost, respectively.
Briefly, all Gag p24 sector 3 mutation pairs with high E values were not viable in our assay system, and were assigned a replication capacity of zero (Figure 1A). Similarly, with the exception of 315G331R, the five high E HLA-associated mutation pairs showed substantial reduction in replication capacity, to between 0–56% of wild-type levels (Figure 1B). Non-viable mutants (RC = 0) were those for which the generation of virus stocks from plasmids encoding these mutation pairs failed, or, in two instances – mutants 186I295E and 186I331R – were not viable unless further mutations developed, confirming unfavorability of the mutation combination. Briefly, concentrated virus stocks for mutants 186I295E and 186I331R were harvested at >22 days post-electroporation compared with the median harvesting time of 6 days post-electroporation for all mutants (at which time the 186I295E and 186I331R mutants had infected ≈1% cells). Sequencing of these viruses revealed the presence of additional mutations and/or reversion of introduced mutations. For mutant 186I295E, amino acid mixtures were detected at codons 63 (Q/R), 177 (D/E) and 186 (I/V), and for mutant 186I331R, mixtures were detected at codons 168 (I/V) and 331 (K/R), as well as reversion of 186I to 186T. On repeating virus generation for these mutants, additional mutations similarly developed – mixtures were observed at codons 214 (R/K) and 271 (N/S) for mutant 186I295E, and 232 (R/M) and 260 (D/E) for mutant 186I331R. With the exception of 186I295E and 186I331R, sequencing confirmed that all mutant viruses had only the specific mutations introduced. The spontaneous mutations 186V, 271S and 232M were not observed in the MSA and the new mutation combinations did not have lowered E values in any of the models, with the exception of the incomplete 186I331R260D combination (complete observed combination 186I, 331R, 232R/M, 260D/E) which displayed a slightly lower energy than 186I331R in the regularized Ising model only (11.5 vs. 13.7) (data not shown). Nevertheless, these observations confirm that 186I295E and 186I331R are unfit mutation combinations requiring compensatory paths to restore viability. Taken together, the data on high E p24 mutants confirm mutation combinations predicted to be unfit, and also identify combinations of HLA-associated mutations in/next to optimal CD8+ T cell epitopes (mutations likely to result in CD8+ T cell escape [33]) that carry substantial fitness costs.
Those p24 mutation combinations, including known compensatory pairs, that were predicted to have low E values displayed replication capacities similar to that of wild-type NL4-3, indicating that these combinations had little or no cost to HIV-1 replication capacity in accordance with predictions (Figure 1C). Similarly, all p17 mutants tested had replication capacities close to that of the wild-type NL4-3 virus, consistent with the predicted E values of all mutants except 86F92M (Figure 1D).
Overall, for only two (86F92M and 315G331R) of the 17 mutant pairs the fitness measurement did not correspond to the E value prediction of high or low fitness cost. It should however be noted that the disparity between E values and measured replication capacities for these mutant pairs is somewhat mitigated in the regularized models. The E values for the regularized Ising model for these mutants (which were assigned an E value of infinity by the original Ising model) are lower than those of other mutants previously assigned infinite energies, and the same is true for mutant 86F92M in the regularized Potts model.
Next, we assessed the relationship between fitness measurements and E values predicted by our original Ising, regularized Ising and regularized Potts models using Pearson's correlation tests. There is a strong correlation between the metric of fitness (values of E, Table 1) predicted by the original unregularized Ising model and our experimental measurements (Pearson's correlation, r = −0.74 and p = 3.6×10−6, two-tailed) (Figure 2A), however this correlation out of necessity excludes mutants with E values equal to infinity (n = 13). The regularized Ising model allows for inclusion of these data points resulting in a stronger correlation between predictions and fitness measurements (Pearson's correlation, r = −0.83 and p = 3.7×10−12, two-tailed) (Figure 2B), which is slightly improved by focusing on Gag p24 mutants only (Pearson's correlation, r = −0.85 and p = 1.4×10−11, two-tailed). There is also a strong agreement between the residue-specific Potts model energies and replication capacity (Pearson's correlation, r = −0.73 and p = 9.7×10−9, two-tailed) (Figure 2C).
In practice, one may be concerned with a more coarse-grained measure of viral fitness: will a virus with a given sequence be able to replicate with similar efficiency to the wild-type, or will it be significantly impaired? To explore this point, we grouped the experimentally tested mutants into two categories, “fit” (RC≥0.5) and unfit (RC<0.5), and tested the ability of the fitness landscape models to predict which class each sequence would belong to based on their E values. This was accomplished by fitting a linear classifier to the data using logistic regression (Text S1, Section 3.1). The regularized Ising model E classifier is highly accurate (91% accuracy at optimal threshold, AUROC = 0.93) – we observed a strong, significant difference in replication capacities between the mutants classified as unfit and those classified as fit (Mann-Whitney U = 32, ) (Figure 3A). Specifically, four mutants (86F92M, 190I, 190I302R and 243P) were not classified correctly. However, 190I302R, which was classified as unfit (E = 8.6), exhibited a fitness close to that of the 0.5 cutoff (RC = 0.56) and 243P, which displayed low fitness (RC = 0.36), had a predicted E value (E = 7.4) bordering on the classifier E value. The Potts model classifier also performs well (81% accuracy at optimal threshold, AUROC = 0.80), but provides a slightly weaker difference between the fit and unfit classes (Mann-Whitney U = 70, ) (Figure 3B). Here, seven mutants were not classified correctly, including the same four not classified correctly by the regularized Ising model as well as mutants 174G, 181R, 269E and 315G331R. Similar to mutant 243P, mutants 174G, 181R and 269E were unfit (RC = 0) but had a predicted E values ranging from 7.3 to 7.7, fairly close to that of the classifier E value.
In this study, we have substantially advanced our modeling approaches and tested the predictive power of these models by in vitro fitness measurements of HIV encoding various mutation combinations in the Gag protein. The in vitro functional data are overall in strong agreement with the viral fitness landscape models and support the capacity of these models to robustly predict both continuous and “coarse-grained” measures of HIV-1 in vitro replicative fitness. Performance of the regularized Potts and regularized Ising models here is similar, which is not unexpected as Gag in general is not highly mutable and the mutants tested here were the most common ones, making the binary approximation a fairly good assumption. Indeed, in instances where the binary approximation is valid, we might encounter poorer performance from the Potts model relative to the Ising due to a diminished ratio of samples (i.e., sequences in the MSA) to parameters (i.e., h and J values) making robust numerical fitting of the former more challenging than the latter. It is nevertheless encouraging that we are capable of fitting a significantly more complicated Potts model that retains residue-specific resolution without compromising the fidelity of our predictions. Improved inverse Potts inference methods which better meet these numerical challenges may also improve performance of the Potts model with respect to the Ising model results.
Simple theoretical analysis suggests that models which differentiate between different mutant amino acids at the same site, like the Potts model employed here, will be necessary to make fitness predictions for highly mutable proteins such as Env and Nef, or to predict the fitness of sequences containing sites with mutations to less frequently observed amino acids. Using a simple toy model, we show in Text S1, Section 3.4 that the binary approximation (Ising model) has several potential deficiencies compared to a Potts model. In particular, the Ising model generically overestimates the fitness of mutant sequences, particularly for sequences containing uncommon mutations. Also, in the Ising case the inferred interaction between mutations at different sites is dominated by the interaction between the most common mutants, while the Potts model is able to accurately capture interactions between rare mutants. Future work will involve testing Ising and Potts model predictions for more highly variable proteins and for mutations to uncommon amino acids.
While this study confirms the usefulness of this method for predicting HIV-1 replicative fitness, at least for closely related sequences, caution will be necessary in applying this method to predict the relative fitness of multiple strains separated by a large number of mutations. In the measure of prevalence used to infer the Ising and Potts model fitness landscapes, factors such as phylogeny are implicitly included. Analysis conducted in [25] suggests that phylogenetic effects influence the value of the inferred fields hi, and that a correction should be included for predictions of energy or fitness. This form of a correction is sensible, as phylogenetic effects should make mutations at individual residues less frequent, leading to larger inferred fields. For closely related strains such as those studied here experimentally, any systematic inaccuracies in the energy due to phylogeny should be similar in magnitude, and thus differences in energy should predict relative replicative fitness fairly accurately. This would not necessarily be true, however, for sequences separated by many mutations. Further theoretical developments may be needed to separate out the contributions of phylogeny and intrinsic fitness from the Ising and Potts model landscapes presented here [25] to predict the relative fitness of strains that differ by many mutations.
In addition to phylogeny, other factors such as host-pathogen interactions and pure stochastic fluctuations affect the observed distribution of sequences, and could complicate fitness predictions. In another work [25] we have investigated these issues by carrying out stochastic simulations that aim to mimic the way the samples were collected and host-pathogen dynamics. In this paper, for the p17 protein, we found that the fitness and prevalence were not the same. However, the rank order of fitness and prevalence were the same as long as the strains being compared were not very far apart in sequence space. This is largely because of the diversity of immune responses due to diverse HLA types in the human population. Additionally, the number of virus particles in single infected individuals from whom the virus sequences were extracted is large, as is the number of patients from whom the virus samples were taken. We find that the one and two-point mutational probabilities in the sequence databases have converged [11]; i.e., these correlations do not change upon removal of some sequences, suppressing the effects fluctuations on the inferred model.
We also note that some caution should be taken in comparing E values for sequences belonging to different proteins. The fitness predictions of the Ising and Potts models are unchanged by a constant shift in energy for all sequences, thus comparisons of absolute energy values are not physically meaningful. Differences in energy between two sequences in the same protein, however, can be unambiguously interpreted as the fitness ratio of those sequences. This is the approach we have taken when examining E values from sequences with mutations in p17 and p24 together: rather than comparing the absolute energies, we compare the differences in energy between the mutant and the NL4-3 reference sequence in each protein, which reflect the fitness of the mutant relative to the NL4-3 reference sequence. Finally, translation from differences in energy to differences in fitness might depend on the specific protein that is being considered. While comparisons of energy differences and relative fitnesses of p17 and p24 mutants performed here exhibit no obvious incongruences, further study is needed to confirm the generality of fitness predictions across proteins.
In the case of two mutant pairs (186I295E and 186I331R) that were predicted by the models to have very low fitness, partial reversions and/or additional mutations spontaneously arose in culture that restored virus viability. However, with the exception of one of the spontaneous mutations (260D) observed in combination with 186I331R that modestly decreased the predicted energy (increased fitness) in the regularized Ising model but not the Potts or original Ising models, the models do not predict lower energies (increased fitness) for these mutant pairs in combination with the additional mutations arising in vitro. Further, three of the spontaneous mutations – 186V, 271S and 232M – were not observed in the MSA and therefore could not be assessed by the Potts model. As a possible interpretation of these findings, we suggest that it may be the case that these mutation patterns observed in vitro are not typically observed in vivo, perhaps since these are infrequently explored mutational routes. As a corollary, this could indicate an inherent limitation of computational models derived from clinical sequence data to identify all possible escape-compensatory pathways, and the importance of in vitro and in vivo experiments to validate and complement model predictions. A mitigating factor, of course, is that mutational pathways observed in vitro but not in vivo may be of less direct clinical relevance.
In future work, the model predictions will be further validated in animal models by testing the viable escape pathways predicted to emerge following immunization with immunogens containing vulnerable HIV-1 regions only. The validated fitness landscape could then be used to design vaccine immunogens containing epitopes from the vulnerable regions that could be presented by people with diverse HLAs and that target residues particularly harmful to HIV-1 when mutated simultaneously, thereby substantially diminishing viral fitness and/or blocking viable mutational escape [11], [12]. Such immunogens potentially represent good therapeutic vaccine candidates to overcome the challenge of HIV-1 evasion of CD8+ T cell responses. However, further work will also be required to optimize design of such immunogens to ensure that epitopes included are processed effectively and that they are sufficiently immunogenic, as well as to test their immunogenicity, optimal delivery methods and protection efficacy in animal models. Furthermore, fitness landscapes of HIV-1 proteins may also be more widely applied to identify effective antibody targets, and help design potent combinations of neutralizing antibodies for passive immunization as well as small molecule inhibitors for therapy.
|
10.1371/journal.pntd.0005377 | Comparative analysis of gut microbiota of mosquito communities in central Illinois | The composition and structure of microbial communities that inhabit the mosquito midguts are poorly understood despite their well-documented potential to impede pathogen transmission.
We used MiSeq sequencing of the 16S rRNA gene to characterize the bacterial communities of field-collected populations of 12 mosquito species. After quality filtering and rarefaction, the remaining sequences were assigned to 181 operational taxonomic units (OTUs). Approximately 58% of these OTUs occurred in at least two mosquito species but only three OTUs: Gluconobacter (OTU 1), Propionibacterium (OTU 9), and Staphylococcus (OTU 31) occurred in all 12 mosquito species. Individuals of different mosquito species shared similar gut microbiota and it was common for individuals of the same species from the same study site and collection date to harbor different gut microbiota. On average, the microbiota of Aedes albopictus was the least diverse and significantly less even compared to Anopheles crucians, An. quadrimaculatus, Ae. triseriatus, Ae. vexans, Ae. japonicus, Culex restuans, and Culiseta inornata. The microbial community of Cx. pipiens and Ae. albopictus differed significantly from all other mosquitoes species and was primarily driven by the dominance of Wolbachia.
These findings expand the range of mosquito species whose gut microbiota has been characterized and sets the foundation for further studies to determine the influence of these microbiota on vector susceptibility to pathogens.
| The microbial communities that reside in mosquito midguts can impact transmission of mosquito-borne pathogens. We used high throughput next generation sequencing to characterize the midgut microbial communities of 12 mosquito species collected in urban residential areas in Champaign County, Illinois. A total of 181 OTUs from 11 phyla and 66 families were identified. Although several bacterial taxa were shared between two or more mosquito species, there was remarkable individual differences in gut microbiota and it was common for individuals of different mosquito species to harbor similar gut microbiota. The microbiota of Ae. albopictus was the least diverse and significantly less evenly distributed compared to 7 of 11 mosquito species. The microbial community of Cx. pipiens and Ae. albopictus differed significantly from other mosquito species and was primarily dominated by Wolbachia. These findings improve current knowledge on the composition and structure of mosquito gut microbiota and provide the framework for understanding their contribution to individual variation in vector competence and potential application in disease control.
| Mosquitoes transmit a wide range of pathogens that cause diseases in humans and other animals. The majority of mosquito-borne pathogens were previously confined to small geographic regions in the tropics but have recently emerged as a worldwide threat to human and animal health. Recent examples of mosquito-borne diseases that have caused major epidemics outside their native geographic range include West Nile virus [1], dengue virus [2], Chikungunya virus [3] and Zika virus [4, 5].
The transmission cycle of mosquito-borne pathogens involve interactions between at least three species: the pathogen, the vector, and the vertebrate host. When the mosquito takes a blood meal from an infected vertebrate host, the pathogen invades the midgut tissue where it undergoes further development and/or replication and then disseminates to secondary tissues such as nerve tissue, fat body, and finally the salivary glands [6]. At this point, the mosquito is considered infectious and is capable of transmitting the pathogen during a subsequent blood meal. However, the mosquito midgut is known to possess factors that may impede successful transmission of the pathogen [7–10]. These factors include the mosquito innate immune system and the digestive enzymes [6, 8, 11].
It is also well established that the mosquito midgut is colonized by a community of bacteria that can affect vector susceptibility to pathogens e.g. [12, 13]. For example, certain bacterial isolates from natural mosquito populations have been shown to reduce mosquito susceptibility to Plasmodium and dengue infection [12, 14, 15]. These effects are exerted through activation of the mosquito immune system [16], generation of reactive oxygen species by certain microbes [15], and formation of a physical barrier to infection [17]. Likewise, modification of midgut microbiota of Anopheles gambiae and Aedes aegypti through antibiotic treatment has been shown to enhance susceptibility to Plasmodium [16] and dengue infection [18], respectively. Other studies have shown that some midgut bacterial isolates can be genetically modified to express molecules that impair pathogen development within the mosquito [19, 20]. Collectively, these findings suggest that the composition of mosquito midgut microbiota likely contributes to within- and between-species variation in vector competence that is typical of many (if not all) mosquito-borne disease systems. Moreover, these studies demonstrate the potential for exploiting microbial functions for symbiotic control of mosquito-borne diseases [21].
Over the last few decades numerous studies have used culture-dependent and culture-independent approaches to characterize the microbial communities in the midguts of mosquito populations. These studies have revealed that the composition and diversity of gut microbiota can vary dramatically within [22] and between mosquito species [23] and are influenced by host diet [24], developmental stage [24], larval environment [25], and pathogen infection [26, 27]. As such, additional studies comparing the microbial communities of different mosquito species can further improve our understanding of mosquito microbiota and propel identification of specific microbes that may be harnessed for disease control.
In this study, we characterized the microbiota of 12 mosquito species collected from Champaign County, Illinois. The aim of this study was to determine how gut microbial diversity, composition and structure differs between mosquito species. Overall, we observed some remarkable similarities in gut microbiota between individuals of different mosquito species that were dominated by one or two bacterial taxa. These bacterial communities tended to vary markedly between individuals. We also found significant differences in bacterial community structure between some mosquito species. These findings advance current knowledge on the microbial communities that reside in mosquito midguts and provide the foundation for investigating their role in mosquito biology and potential application in mosquito-borne disease control.
Mosquito samples for this study were collected once per week (July 2, 2015 to October 15, 2015) outside 19 urban residential houses in Champaign County, Illinois with permission from property owners (Fig 1). The sites were located within a 10 km radius of each other. The collections were done using standard CDC miniature light traps that were baited with dry ice as an attractant. The traps were tied to a tree outside the respective houses and operated between 1800 hours and 0900 hours. Mosquitoes from each trap were transported live in cool boxes, identified morphologically to species [28], and stored at -80°C until further processing.
Individual female mosquitoes were surface sterilized as previously described [23] and dissected in 50 μl of Dulbecco’s phosphate buffered saline (DPBS) solution (Thermo Fisher Scientific, Waltham, MA). Total DNA was isolated from each midgut using QIAamp DNA mini kit (Qiagen, Valencia, CA). A portion of DNA from Culex mosquitoes was used for species identification using real-time polymerase chain reaction [29]. In total, 264 midguts from 12 mosquito species were processed (Table 1). The V3-V5 region of the 16S rRNA gene was amplified and sequenced using Illumina MiSeq Bulk v3 platform at the W. M. Keck Center for Comparative and Functional Genomics at the University of Illinois at Urbana-Champaign as previously described [23]. The following primer set was used: forward 5ʹ -CCTACGGGAGGCAGCAG-3`and reverse 5`-CCGTCAATTCMTTTRAGT-3ʹ.
In brief, all DNA samples were measured on a Qubit (Life Technologies) using High Sensitivity DNA Kit and diluted to 2 ng/μl. A master mix containing 0.5 μl -10X FastStart Reaction Buffer without MgCl2, 0.9 μl -25 mM MgCl2, 0.25 μl -DMSO, 0.1 μl -10 mM PCR grade Nucleotide Mix, 0.05 μl -5 U/μl FastStart High Fidelity Enzyme Blend, 0.25 μl -20X Access Array Loading Reagent, and 0.95 μl -water was prepared using the Roche High Fidelity Fast Start Kit and 20X Access Array loading reagent and aliquoted into 48 well PCR plates along with 1 μl DNA sample and 1 μl Fluidigm Illumina linkers (V3-V5-F357: ACACTGACGACATGGTTCTACA and V3-V5-R926:TACGGTAGCAGAGACTTGGTCT) with unique barcode. In a separate plate, primer pairs were prepared and aliquoted. 20X primer solutions were prepared by adding 2 μl of each forward and reverse primer, 5 μl of 20X Access Array Loading Reagent and water to a final volume of 100 μl.
Four μl of sample was loaded in the sample inlets and 4 μl of primer loaded in primer inlets of a previously primed Fluidigm 48.48 Access Array IFC. The IFC was placed in an AX controller (Fluidigm Corp.) for microfluidic loading of all primer/sample combinations. Following the loading stage, the IFC plate was loaded on the Fluidigm Biomark HD PCR machine and samples were amplified using the following Access Array cycling program without imaging: 50°C for 2 minutes (1 cycle), 70°C for 20 minutes (1 cycle), 95°C for 10 minutes (1 cycle), followed by 10 cycles at 95°C for 15 seconds, 60°C for 30 seconds, and 72°C for 1 minute, 2 cycles at 95°C for 15 seconds, 80°C for 30 seconds, 60°C for 30 seconds, and 72°C for 1 minute, 8 cycles at 95°C for 15 seconds, 60°C for 30 seconds, and 72° for 1 minute, 2 cycles at 95°C for 15 seconds, 80°C for 30 seconds, 60°C for 30 seconds, and 72°C for 1 minute, 8 cycles at 95°C for 15 seconds, 60°C for 30 seconds, and 72°C for 1 minute, and 5 cycles at 95°C for 15 seconds, 80°C for 30 seconds, 60°C for 30 seconds, and 72°C for 1 minute. The PCR product was transferred to a new 96 well plate, quantified on a Qubit fluorimeter (Thermo-Fisher) and stored at -20°C. All samples were run on a Fragment Analyzer (Advanced Analytics, Ames, IA) and amplicon regions and expected sizes confirmed. Samples were then pooled in equal amounts according to product concentration. The pooled products were size selected on a 2% agarose E-gel (Life Technologies) and extracted from the isolated gel slice with QIAquick gel extraction kit (QIAGEN). Cleaned size selected products were run on an Agilent Bioanalyzer to confirm appropriate profile and determination of average size. The final library pool was spiked with 10% non-indexed PhiX control library (Illumina) and sequenced using Illumina MiSeq V3 Bulk system. The libraries were sequenced from both ends of the molecules to a total read length of 300nt from each end. Cluster density was 964k/mm2 with 85.9% of clusters passing filter.
IM-TORNADO 2.0.3.2 platform was used to process the de-multiplexed fasq-formatted files obtained from the sequencing facility. This platform is designed to process non-overlapping reads for analysis as a whole unit without sacrificing one of the reads in the pair and improves accuracy in read analysis compared to single-end read analysis [30]. The 5ʹ PCR primer for forward (R1) and reverse (R2) reads were trimmed using Trimmomatic program [31] with the parameter HEADCROP:17 for R1 read and HEADCROP: 18 for R2 read. The quality filtering process was performed using Trimmomatic program following previously described procedures with slight modifications [30]. Briefly, the sequences were subjected to a hard cutoff of PHRED score Q3 for 5 ʹ and 3ʹ ends of the reads (parameters LEADING: 3 and TRAILING: 3), trimming of the 3’ end with a moving average score of Q15, with a window size of four bases (parameter SLIDINGWINDOW: 4:15), and any reads with less than 150 base pairs removed with parameter R1_TRIM = 150 and R2_TRIM = 150. Reads with ambiguous base calls were discarded. To retain both reads while avoiding misinterpretation of the data, matching R1 and R2 reads were joined using an ambiguous nucleotide character “N” between R1 and R2 [30]. In a single run, IM-TORNADO generates outputs for R1 data only, R2 data only, and paired end data. Only output files related to paired end data were used for taxonomic assignment and downstream analysis. Reads were de-replicated building clusters of reads with 100% similarity and annotated with cluster size. Singletons and reads shorter than the cutoff length were discarded to ensure the use of high quality reads when assigning OTU representation. Reads were sorted by cluster size and processed in USEARCH using the UPARSE algorithm to find the OTU representatives using de novo OTU picking strategy. Chimeric reads are also removed during this step resulting in a set of OTU representatives of very high sequence quality [32]. Operational taxonomic units (OTUs) were assigned at 97% sequence similarity using the Ribosomal Database Project (RDP) version 10 as the reference set with a threshold of 80% bootstrap confidence [33].
Quantitative TaqMan real-time PCR (qPCR) was used to confirm the wsp gene of Wolbachia in mosquito midgut samples using the following primer set: forward: 5’-GSTTTTGCTKRTCAAGYAARAG-3’ and reverse: 5’-GYGCTGTAAAGAACKTTGWDY-3' respectively. Taqman probe sequence was 5’ FAM-TGTTAGTTATGATGTAACTCCRGAA-IABFQ 3’. The primers and probe were synthesized by Integrated DNA Technology, Inc. (IDT, Coralville, IA). Twenty microliter qPCR contained 1× SensiFAST Probe Hi Rox mastermix (BioLine, Taunton, MA), 0.5 μM of each primer, 0.25 μM Taqman probe and 2 μL of the mosquito midgut DNA isolate. The qPCR was run with 1 cycle of heat activation at 95°C for 15 minutes followed by 45 cycles of denaturation at 94°C for 1 minute, annealing at 50°C for 1 minute and elongation at 72°C for 1 minute.
Minigene was constructed using wsp gene segment flanked by the PCR primers and was synthesized by IDT (Coralville, IA). The gene sequences utilized for the minigene were downloaded from GenBank and the accession number was CP001391 for Wolbachia spp wRi. The minigene was used as a positive control for qPCR of Wolbachia wsp gene and as templates for building a standard curve to estimate the quantity of wsp gene in mosquito midgut samples. The copy number of minigene (2063 bp) containing wsp gene segment was calculated based on the DNA concentration determined by NanoDrop 1000 spectrophotometer (Thermo Scientific) and on the assumption that the average weight of a DNA base pair (bp) is 650 Daltons. The formula for copy number calculation is: copy numbers = ((minigene amounts in ng) × (6.022 × 1023)) / (2063 × 650 × 109). The concentration of minigene solution was adjusted to be 5 × 109 copies/μl and 10-fold serially diluted in nuclease free water (BioLine, Taunton, MA). Two microliter of the serially diluted minigene solution was utilized for qPCR. A standard curve was generated using the relationship between the cycle numbers at threshold (Ct values) and the minigene copy numbers in serially diluted minigene solution.
Unless otherwise stated statistical analysis were conducted using R 3.2.3 statistical software (https://cran.r-project.org/bin/windows/base/old/3.2.3/). OTUs accounting for < 0.005% of the total number of sequences were discarded before downstream analysis to reduce the problem of spurious OTUs [34]. The number of sequences varied markedly among individual mosquito midguts (mean ± SE = 6834.72 ± 460.75 per mosquito midgut; minimum = 0, maximum = 39,268). We rarefied the read depth to 1,036 reads per sample to standardize the sampling effort. Sixty nine samples that did not meet this criterion (i.e. had < 1,036 sequences) were excluded from subsequent analysis (Table 1). Alpha diversity metrics including Shannon diversity index, observed species, chao1, and evenness were generated in QIIME [35] and analysis of variance with Tukey adjustments was used to test whether there were any significant differences in these indices among mosquito species. Analysis of similarities (ANOSIM) using the “vegan” package in R was used to test whether microbial communities from samples of each mosquito species were more similar than those of different mosquito species [36]. The computed Bray-Curtis similarity matrix values were used for principal coordinate analysis (PCoA) to determine microbial community differences across mosquito species (“vegan” package in R). Hierarchical clusters based on Bray-Curtis dissimilarity measure were performed in PAST software to highlight the differences in mosquito samples based on the composition and abundance of their gut microbiota [37]. Similarity percentage (SIMPER) analysis was used to identify OTUs that were primarily responsible for observed differences between mosquito species (PAST version 3.14 software [37]).
MiSeq sequencing of the V3-V5 region of 16S rRNA gene amplicons from 264 mosquito samples generated a total of 1,804,366 sequences (Mean ± SE = 6834.72 ± 460.75 per mosquito midgut sample). After quality filtering and rarefying the reads to an even sampling depth of 1,036 sequences, a total of 202,020 sequences from 195 mosquito samples were retained. These sequences were clustered into 181 bacterial OTUs belonging to 11 phyla, 66 families and 111 genera. Only 16 of the 181 OTUs had an overall abundance equal to or greater than 1%. The majority of sequences were from Proteobacteria (81.1%) comprising of Alphaproteobacteria (47.4%), Gammaproteobacteria (29.2%), Betaproteobacteria (3.2%), Epsilonproteobacteria (1.1%), and Deltaproteobacteria (0.3%). Other observed phyla included, Actinobacteria (8.8%), Firmicutes (5.7%), Bacteroidetes (1.8%), Acidobacteria (0.8%), Cyanobacteria (0.6%), Tenericutes (0.5%), Spirochaetes (0.4%), Planctomycetes (0.3%), Parcubacteria (0.03%) and Fusobacteria (0.005%).
The most abundant OTUs were associated with the families Acetobacteraceae (25.7%), Enterobacteriaceae (20.6%), Rickettsiaceae (20.0%), Propionibacteriaceae (8.4%), and Orbaceae (4.2%) (S1 Fig). Acetobacteraceae occurred in high abundance among some individuals of all mosquito species except An. crucians. However, they were found in fewer individuals of Ae. albopictus, An. punctipennis, An. quadrimaculatus, Cx. pipiens and Cx. restuans compared to the remaining mosquito species. Enterobacteriaceae was more common among Ae. triseriatus, Ae. trivittatus, and Ae. vexans and also occurred in high abundance in the guts of a few individuals of the remaining mosquito species. Rickettsiaceae was more abundant and widespread in Ae. albopictus and Cx. pipiens and was also present in high abundance in a few samples of An. crucians, An. punctipennis, and An. quadrimaculatus. Propionibacteriaceae were mostly associated with An. crucians and An. punctipennis and occurred in high abundance in a few individuals of Ae. triseriatus, Ae. vexans, An. quadrimaculatus, Cx. restuans, and Cs. inornata. Orbaceae occurred in high abundance in a few individuals of An. crucians, An. punctipennis, An. quadrimaculatus, Cs. inornata, Ps. ferox, Ae. japonicus and Ae. triseriatus. Overall, only 1–3 major families of bacteria tended to dominate the guts of the 12 mosquito species (S1 Fig). It was also common for some individuals of a given mosquito species from the same study site and collection date to harbor different gut microbiota.
The top 9 OTUs accounted for 69.2% of the total sequences and their relative abundance varied markedly between mosquito species (Fig 2). OTU 1 (Gluconobacter) accounted for 23.1% of the total sequences and was more abundant in all Aedes mosquito species (except Ae. albopictus) as well as Cs. inornata and Ps. ferox. This OTU also occurred in high abundance in a few samples of Cx. pipiens, Cx. restuans, An. punctipennis, and An. quadrimaculatus. OTU 2 (Wolbachia) was more prevalent and abundant in the guts of Ae. albopictus and Cx. pipiens and also occurred in three Ae. japonicus samples and one sample each of An. crucians, An. punctipennis and An. quadrimaculatus. OTU 9 (Propionibacterium) was mostly associated with An. crucians and An. punctipennis but it also occurred in higher abundance in a few samples of other mosquito species. OTU 8 (Morganella) was mostly associated with Ae. triseriatus, Ae. trivittatus, and Ae. vexans and OTU 5 (Providencia) was mostly associated with Ae. vexans. OTU 182 (Gluconobacter) was mostly associated with Ae. japonicus but was also present in high abundance in the guts of some individuals of other mosquito species. OTU 6 (Orbus), OTU 86 (Pantoea), and OTU 12 (Tatumella) occurred in high abundance in one or a few individuals of different mosquitoes (Fig 2). Some individuals of a given mosquito species also tended to differ in their microbial composition despite being collected from the same study sites and collection dates. The majority of mosquito samples were dominated by 1–2 OTUs.
Overall, 57.5% of bacterial OTUs were shared between at least two mosquito species (Fig 3). However, only three bacterial OTUs occurred in all 12 mosquito species. These were OTU 1 (Gluconobacter), OTU 9 (Propionibacterium), and OTU 31 (Staphylococcus).
Shannon diversity indices revealed that on average, the gut microbiota of Aedes albopictus was the least diverse and significantly less even compared to gut microbiota of An. crucians, An. quadrimaculatus, Ae. triseriatus, Ae. vexans, Ae. japonicus, Cx. restuans, and Cs. inornata (Shannon: F = 6.4, df = 11, 179, P < 0.001; Evenness: F = 6.4, df = 11, 179, P < 0.001; Table 2). The gut microbiota of An. crucians was also significantly more diverse and more evenly distributed compared to that of Ae. trivittatus, Cx. pipiens, and Ps. ferox (Table 2). We also calculated Chao1 estimator based on OTUs abundance to determine the expected richness in each sample (Table 2). We were able to detect more than 93% ± 1.3% (mean ± SE) of the expected number of OTUs suggesting that most OTUs were recovered. On average, our results revealed that a mosquito midgut contains 5–10 bacterial OTUs (Table 2). The observed and predicted (Chao1) number of OTUs were significantly lower in Ae. albopictus compared to Ae. vexans (Observed OTUs: F = 3.2, 11, 179, P = 0.0005; Chao 1: F = 2.6, df = 11, 179, P = 0.005; Table 2). Significantly more bacterial OTUs were also observed in An. crucians and Ae. triseriatus guts compared to Ae. albopictus guts.
The ANOSIM analysis based on Bray-Curtis distances revealed a significant difference in microbial communities among the 12 mosquito species (ANOSIM, R = 0.59, P = 0.001). To better visualize the results, a principal coordinates analysis (PCoA) plot was generated based on Bray-Curtis distances (Fig 4). Ordination based on this metric demonstrated a clear separation of Ae. albopictus and Cx. pipiens samples from the other mosquito species indicating that the microbial communities of the two mosquito species differed from those of the other mosquito species (Fig 4). Cluster analysis based on Bray-Curtis distances confirmed that the majority of Ae. albopictus and Cx. pipiens samples tended to cluster together and that it was common for individuals of different mosquito species from different sites and collection dates to harbor similar gut microbiota (S2 Fig).
The SIMPER analysis was used to identify the bacterial OTUs primarily responsible for the observed separation of gut communities between mosquito species, using the relative abundances of bacterial OTUs (S1 Table; S3 Fig). Twelve OTUs accounted for 69.8% of observed differences between mosquito species with OTU 1 (19%), OTU 2 (18%) and OTU 9 (8%) accounting for the largest variation (S1 Table). OTU 1 (Gluconobacter), was found in all mosquito species but was more abundant in Ae. japonicus, Ps. ferox, Ae. trivittatus, Ae. triseriatus, and Cs. inornata (S3 Fig). OTU 2 (Wolbachia) was mainly associated with Ae. albopictus and Cx. pipiens and OTU 9 (Propionibacterium) was mainly associated Cx. restuans, Ae. triseriatus and the three Anopheles species (An. crucians, An. quadrimaculatus, An. punctipennis, S3 Fig).
Real-time qPCR results confirmed the presence of Wolbachia in all three Ae. japonicus samples, 25 of 27 Ae. albopictus samples, 12 of 15 Cx. pipiens samples, and the 1 An. punctipennis sample (S4 Fig). None of the other mosquito species had Wolbachia. Wolbachia wsp gene copy numbers ranged from 0 to 10151 and were relatively higher in Ae. albopictus compared to the other mosquito species (S4 Fig).
In this study we characterized and compared the midgut bacterial communities of 12 mosquito species encompassing four mosquito genera, many of them important vectors of medical, veterinary and wildlife significance. Overall, we found a low diversity of gut microbiota that was characterized by large individual variability and the dominance of one or two bacterial OTUs. Analysis of microbial composition revealed that the bacterial community in mosquito midguts was dominated by a few phyla with only three phyla (Proteobacteria (81.1%), Actinobacteria (8.8%) and Firmicutes (5.7%) accounting for 97% of the total sequences. These bacterial phyla are commonly reported in the guts of mosquitoes and other insects [22, 24, 25, 38, 39]. The Phylum Proteobacteria is highly diverse and contains a wide variety of species that are adapted to a wide range of environments; thus it is no surprise that its dominance in mosquito midguts is well established [22, 24, 25, 40, 41].
Individual variability in gut microbiota was not only restricted to mosquito samples collected from different sites and different dates but was also common among individual mosquitoes collected at the same sites and collection dates. Similar individual variability in gut microbiota and the dominance of a few bacterial taxa in mosquito guts has been reported before [22]. These variations may result from individual variations in external and internal factors such as the gut physiological conditions, larval and adult diet, infection with parasites and pathogens, host aging [24, 26, 27, 38, 42], and host genetic background [43]. Our experimental design cannot decipher the contribution of these factors to the observed pattern of gut microbiota since adult mosquito samples were collected using the CDC light traps and we had no prior knowledge of the factors these mosquitoes were exposed to before collection. Individual variation in gut microbiota may be epidemiologically relevant since some bacterial species are known to enhance [13, 44, 45] or reduce mosquito susceptibility to Plasmodium parasites and dengue viruses [14, 46, 47]. Thus it is possible that differences in gut microbiota observed in this study may be one of the primary factors contributing to individual variation in vector competence that is commonly observed in nature. Future studies targeting the role of specific members of this bacterial community on vector competence and other aspects of mosquito biology may provide important insights into their epidemiological significance.
Ae. albopictus and Cx. pipiens harbored distinct bacterial communities that was primarily dominated by OTU 2 (Wolbachia). We also found Wolbachia sequences in three samples of Ae. japonicus and one sample of An. crucians, An. quadrimaculatus, and An. punctipennis. Real-time qPCR results confirmed the widespread occurrence of Wolbachia in Ae. albopictus and Cx. pipiens samples as well as its presence in the 1 and 3 An. punctipennis and Ae. japonicus samples that had Wolbachia sequences, respectively. We processed only intact mosquitoes and sterilized their surfaces before dissecting their midguts to minimize the potential for contamination. This process is expected to remove bacteria from the body surface but it is still possible these mosquitoes were contaminated with Wolbachia from damaged Ae. albopictus and Cx. pipiens samples either in the traps or during sorting and sample identification. However, the dominance of Wolbachia sequences in one of An. punctipennis samples and three Ae. japonicus samples is unlikely due to cross contamination and may imply that a few individuals of Ae. japonicus and An. punctipennis may harbor Wolbachia endosymbionts. Wolbachia are a genus of maternally-inherited bacterial endosymbionts that are estimated to occur in approximately 65% of insect species [48]. This bacterium acts as a reproductive parasite in arthropods; it induces male killing, feminization, and cytoplasmic incompatibility which facilitate its spread throughout the arthropod population [49]. Both Ae. albopictus and Cx. pipiens are known to harbor Wolbachia endosymbionts [23, 38, 50–52] and our study suggest the need for detailed investigations of Wolbachia infection to ascertain that its absence in other mosquito species as reported in the past is not due to lack of adequate sampling effort. The mechanism underlying the high Wolbachia infection and low diversity of midgut bacteria in Ae. albopictus is unclear but could be due to methodological bias where the rarefaction depth of 1,036 employed in this study may not have been sufficient to detect low abundance OTUs or due to Wolbachia interacting negatively with other bacterial species. Additional studies are needed to develop a better understanding of how Wolbachia interacts with other microbiota. Wolbachia has been shown to inhibit transmission of mosquito-borne pathogens [53–55] and is currently under investigation for potential application in biological control of mosquitoes and associated pathogens [56–58]. Unfortunately, Wolbachia can also enhance transmission of other pathogens such as malaria and West Nile Virus [44, 45, 59]. These effects are dependent on Wolbachia strain and the mosquito-borne pathogen under investigation as it is possible for some Wolbachia strains to inhibit transmission of some pathogens while enhancing transmission of others [60, 61]. These findings reinforce the need to understand the potential impact of Wolbachia on different mosquito-borne pathogens before large scale application of Wolbachia-based disease control strategies.
SIMPER analyses indicated that OTU 1 (Gluconobacter), OTU 2 (Wolbachia), and OTU 9 (Propionibacterium) contributed most to the average dissimilarity between mosquito species. OTU 1 (Gluconobacter) was found in all mosquito species but was strongly associated with Ae. japonicus, Ae. triseriatus, Ae. vexans, Ae. trivittatus, Cs. inornata, and Ps. ferox. Gluconobacter are acetic acid bacteria that are adapted to various sugar- and ethanol-rich environments [62]. These bacteria have been found in association with insects that rely on sugar-based diets including mosquitoes [63, 64]. As an example, the genus Asaia (a member of Acetobacteraceae), are frequently found in the nectar of flowers e.g. [65–67] and have been shown to establish symbiotic associations with mosquitoes [63, 64, 68, 69]. Propionibacterium was mostly associated with Anopheles mosquitoes and Cx. restuans. Propionibacterium is a common bacteria of human skin and other animals [70–72] and has also been isolated in mosquitoes [73]. These bacteria may have been acquired from vertebrate hosts during a blood meal [73]. Another notable OTU accounting for observed differences was OTU 5 (Providencia) which was strongly associated with Ae. vexans. This bacterium is a common gastrointestinal pathogen of humans and animals and also occurs in human and animal wastes [74]. It may have been acquired through contact with blood meal hosts or during larval development. Further studies are needed to investigate the potential role of these bacteria on mosquito biology including susceptibility to pathogens.
In general, there were small differences in bacterial diversity and evenness between most species of mosquitoes. However, the bacterial communities of Ae. albopictus were significantly less diverse and less evenly distributed compared to those of An. crucians, An. quadrimaculatus, Ae. japonicus, Ae. triseriatus, Ae. vexans, Cx. restuans, or Cs. inornata. Similar bacterial diversity and evenness between mosquito species across the four mosquito genera suggest that the mosquito midgut likely plays an active role in regulating the colonization and assembly of bacterial communities. Lower microbial diversity in Ae. albopictus relative to the seven mosquito species may be due to inability of some bacterial taxa to proliferate in the guts of Ae. albopictus either due to species differences in gut physiological conditions [75] and/or modulation of microbial communities by the mosquito innate immune system [12]. The physical presence of some bacterial taxa or other microbes (e.g. fungi) also may render the mosquito midgut uninhabitable to other bacterial taxa due to interspecific competition for resources and/or production of toxins and inhibitory factors. Differences in food sources also may partly account for the observed differences because although all mosquito species tend to feed on microbes as larvae and blood and nectar as adults, different mosquito species portray marked variations in their preferred larval habitats and sugar and blood meal hosts which may pre-expose them to different microorganisms. In addition, sugar feeding and blood feeding can reduce the diversity of gut bacteria in mosquitoes [24]. Although we purposefully selected individuals that were not engorged with blood for microbiome analysis, we could not establish whether our mosquito samples had prior access to a blood meal or a sugar meal. It is possible that the majority of Ae. albopictus that were analyzed in this study had acquired a blood meal and/or a sugar meal leading to major reductions in bacterial diversity.
In summary, our study has characterized the midgut bacterial communities of 12 of the most common mosquito species in the United States, expanding current knowledge on mosquito species whose gut microbes have been studied. We found significant differences in gut microbial composition between some mosquito species and documented marked variation in gut microbiota between individuals of the same mosquito species. The 12 mosquito species included the known vectors of arboviruses of global public health significance such as dengue, chikungunya, Zika, West Nile virus, and La Crosse virus encephalitis. Given the well-documented ability of midgut microbiota to influence vector susceptibility to pathogens [12, 14–16, 25, 46], our results provide critical knowledge that can inspire further studies to determine which of the identified microbial communities could be exploited for disease control.
|
10.1371/journal.pgen.1005286 | A Common Cancer Risk-Associated Allele in the hTERT Locus Encodes a Dominant Negative Inhibitor of Telomerase | The TERT-CLPTM1L region of chromosome 5p15.33 is a multi-cancer susceptibility locus that encodes the reverse transcriptase subunit, hTERT, of the telomerase enzyme. Numerous cancer-associated single-nucleotide polymorphisms (SNPs), including rs10069690, have been identified within the hTERT gene. The minor allele (A) at rs10069690 creates an additional splice donor site in intron 4 of hTERT, and is associated with an elevated risk of multiple cancers including breast and ovarian carcinomas. We previously demonstrated that the presence of this allele resulted in co-production of full length (FL)-hTERT and an alternatively spliced, INS1b, transcript. INS1b does not encode the reverse transcriptase domain required for telomerase enzyme activity, but we show here that INS1b protein retains its ability to bind to the telomerase RNA subunit, hTR. We also show that INS1b expression results in decreased telomerase activity, telomere shortening, and an increased telomere-specific DNA damage response (DDR). We employed antisense oligonucleotides to manipulate endogenous transcript expression in favor of INS1b, which resulted in a decrease in telomerase activity. These data provide the first detailed mechanistic insights into a cancer risk-associated SNP in the hTERT locus, which causes cell type-specific expression of INS1b transcript from the presence of an additional alternative splice site created in intron 4 by the risk allele. We predict that INS1b expression levels cause subtle inadequacies in telomerase-mediated telomere maintenance, resulting in an increased risk of genetic instability and therefore of tumorigenesis.
| Multiple cancer-associated single nucleotide polymorphisms (SNPs) associated with risk of a wide variety of cancers have been identified in the TERT-CLPTM1L region of 5p15.33, identifying this as a multi-cancer susceptibility locus. hTERT encodes the catalytic subunit of the enzyme telomerase, which is responsible for telomere length maintenance in the germline and in most immortalised cancer cells. To date, very little is known regarding the mechanisms by which specific hTERT SNPs predispose to cancer. In this study, we carried out detailed functional analyses on the intron 4 SNP rs10069690, which is associated with a small, but highly significant risk for many types of cancer. We show that the risk-associated minor allele of this SNP results in an hTERT mRNA splice variant, encoding a catalytically inactive protein which acts as a dominant negative inhibitor of telomerase activity and therefore decreases total telomerase activity. We propose that individuals who carry the rs10069690 minor allele have less telomerase activity in some cell types due to cell type-specific alternative splicing, which may result in slightly shorter telomeres, and hence an increased risk of genetic instability and tumorigenesis.
| Telomeres are nucleoprotein structures, which protect the ends of linear chromosomes from being recognized as DNA double-strand breaks [1]. Telomeres shorten with each round of cell division due to the end-replication problem. Normal human somatic cells replicate until their telomeres diminish to a critical threshold, at which point they enter permanent cell cycle arrest and are constrained to a senescent state [2]. Bypass of senescence due to loss of function of the p53 and pRB tumor suppressor pathways results in further telomere shortening which eventually becomes catastrophic, causing end-to-end fusions, genetic instability, and the potential for tumorigenesis.
Telomere shortening may be counteracted by telomerase, a ribonucleoprotein enzyme complex that synthesizes the repetitive telomeric DNA sequence (5'-TTAGGG-3') [3]. The subunits of telomerase include a reverse transcriptase protein, TERT, and an RNA molecule, hTR, which contains a template region. Telomerase activity is detectable during human development from the blastocyst stage to 16–18 weeks gestation in specific tissue types, but is undetectable in most tissues by two months post-natal [4]. In healthy adults, telomerase activity is restricted to germline cells (in the testes and ovaries) [4], peripheral blood mononuclear cells [5,6] and stem cells [7], presumably to support the proliferative requirements of these cell types. In germline cells, there is sufficient telomerase activity to prevent telomere shortening, but in somatic cells the level of telomerase activity is limited, and is only sufficient to slow down the telomere attrition that accompanies normal DNA replication. In contrast, in the great majority of cancers and immortalized cell lines telomere length is maintained; in 85% of cancers this is due to upregulated levels of telomerase and in the remainder this is due to a non-telomerase mechanism [8,9].
One of the ways in which telomerase activity levels appear to be regulated is via alternative splicing of the TERT pre-mRNA [10]. The TERT gene contains 16 exons, and the TERT pre-mRNA can be spliced to yield more than 20 variant mRNAs [11–13]. During human development, loss of telomerase activity in somatic cells is associated with a change in the TERT splicing pattern such that the transcriptional output of the TERT gene consists entirely of splice variants that do not encode active TERT protein [14,15]. Control of alternative splicing is incompletely understood, but there is evidence that this may involve RNA:RNA pairing within the TERT pre-mRNA [16], and that it may be regulated by the cellular microenvironment [17]. Some of the splice variants encode proteins which can act as dominant negative inhibitors of telomerase activity [18,19]. This may be due in part to their ability to become incorporated into the telomerase enzyme complex, because biochemical studies have demonstrated that telomerase exists as a dimer and that its activity is dependent on both of the hTERT active sites being functional [20].
Mutations in hTERT, and in other genes that are required for normal telomere function, can cause short telomere syndromes, which are characterized by proliferative failure in various tissues, especially the bone marrow, lungs, and liver [21,22]. Patients suffering from these syndromes have a substantially increased risk of cancer [22,23], presumably due to excessive telomere shortening and an increased risk of genetic instability. In contrast to the high risk associated with these relatively rare mutations, genome-wide association studies (GWAS) have uncovered numerous single nucleotide polymorphisms (SNPs) in, or close to the hTERT gene which are relatively common in the population and are associated with a small, but highly significant increase in cancer susceptibility ([13,24–27]; reviewed in [28]). Associations have also been found between hTERT SNPs and telomere length [13,24,29].
Despite the statistical robustness of GWAS, the associations remain observational. This is partly because most GWAS have only reported associations with the tagging SNPs in the TERT-CLPTM1L region on the commercial genotyping chips, but fine mapping is required to identify the putative causal SNPs. Mechanistic analysis is required to identify causal rather than correlated variants, and to accurately determine cancer risk. A functional understanding of causative variants may ultimately influence clinical decision-making.
Through our involvement with the Collaborative Oncologic Gene-Environment Study (COGS), we were able to carry out fine mapping of the hTERT locus. This study revealed that the minor alleles at rs10069690 and at rs2242652 were candidate causal variants for risk of estrogen receptor-negative breast cancer, breast cancer in BRCA1 mutation carriers, serous low malignant potential ovarian cancer and serous invasive ovarian cancer with odds ratios ranging from 1.15–1.40 [13,24] and for risk of prostate cancer [13,24]. This rs10069690 SNP is also associated with risk of a wide variety of cancers, including colon cancer, acute lymphoblastic leukemia (ALL), and chronic lymphocytic leukemia (CLL) [25,30–32]. However, no fine mapping has been performed for the latter associations to determine whether it is likely to be directly responsible or simply correlated with a better candidate causal SNP. Consequently, there is a growing need to delineate how this allele confers cancer risk.
We previously generated hTERT minigenes, which included intron 4 with the major (G) or minor, cancer-risk associated (A) allele at rs10069690 [13]. Transfection of these minigenes into the telomerase-positive breast cancer cell line MCF7, and RT-PCR using primers spanning intron 4, revealed an additional band in the presence of the A allele [13]. When this band was excised and sequenced it was identified as a novel hTERT splice variant, INS1b, which contains an intron 4 insert of 480 base pairs. Sequence analysis showed that this transcript was the product of an alternative splicing event at an additional splice donor site created by the G to A polymorphism at rs10069690 (Fig 1A).
In the present study, we investigated the mechanism by which the SNP confers elevated cancer risk. We found that the presence of this polymorphism resulted in the production of both full length and an alternatively spliced hTERT transcript, which were expressed in a cell type-specific manner. The alternative transcript produced a severely truncated protein (INS1b), which retained its hTR binding capability but lacked the reverse transcriptase domain. INS1b directly caused decreased telomerase activity, telomere shortening, and an elevated telomere-specific DNA damage response. Our data demonstrate that INS1b binds to hTR, sequestering it in an inactive form, thereby acting in a dominant negative manner to limit the amount of active telomerase available to extend the telomeres. The combination of the fine mapping data and a mechanism for cancer risk make it highly likely that this is a causal, rather than a correlated cancer risk SNP.
Intron 4 minigene constructs with the Major (G) or Minor (A) alleles of rs10069690 were transfected into three telomerase-positive cell lines (HT1080, MCF7 and A2780) and a fibroblast cell strain (Fre-71s-1). The INS1b transcript is 480 nucleotides larger than the canonically spliced transcript, and includes a premature stop codon in the retained intron 4 sequence (Fig 1B). The truncated protein is predicted to retain the N-terminus, TEN domain and hTR binding domains of hTERT, and to lack the reverse transcriptase domain (Fig 1B), and consequently it is expected to be catalytically inactive. Regardless of which allele was present, all cell lines produced a band that was 38 bp larger than the canonical TERT transcript, corresponding to the INS1 splice variant which results from retention of the first 38 bp of intron 4 in the mRNA [11,33] (Fig 1C). In all four cell types, transfection of the A allele resulted in an RT-PCR product 480 bp larger than the canonical transcript encoding FL-hTERT, corresponding to the INS1b splice variant which we have previously shown by sequencing to result from retention of the first 480 bp of intron 4 [13] (Fig 1C). Other bands were sequenced and found to represent hybrid DNA species (Fig 1C).
We then genotyped a panel of cell lines to identify cell lines that were homozygous (AA) and heterozygous (GA) for the minor allele at rs10069690 (S1 Table). In order to determine the balance of FL-hTERT and INS1b splicing that occurs endogenously, a subset of the cell lines which were either homozygous or heterozygous for the minor allele at rs10069690 (S1 Table) were analyzed by RT-PCR. FL-hTERT was detected in all cell lines, while the levels of INS1 and INS1b transcript varied considerably (Fig 1D). MDA-MB-157 (AA) and B80-T8 (GA) cell lines displayed the highest levels of INS1b transcript, while HEK293T (AA) and BT549 (AA) had lower expression of INS1b. As the canonical splice donor site is retained in the presence of the SNP, FL-hTERT, INS1 and INS1b mRNA were all produced (Fig 1C and 1D). INS1b transcript was undetectable in CAOV3 (AA), MDA-MB-415 (AA), 27/87 (GA) and F80-T1 (GA).
The B80-T8 breast epithelial and F80-T1 fibroblast cell lines are genetically matched telomerase-positive cell lines derived from the same heterozygous individual [34,35], so it is of particular interest that the INS1b transcript is abundant in the breast cell line, whilst undetectable in the fibroblast cell line (Fig 1D). This indicates the potential for cell- or tissue-specific differences in INS1b generation, and suggests that complex regulatory mechanisms maintain the balance between INS1b and FL-hTERT.
Data from the 1000 Genomes Project demonstrated that rs10069690 is in linkage disequilibrium with another intron 4 SNP rs2242652 (r2 = 0.68) and an intron 3 SNP rs7725218 (r2 = 0.54). To investigate whether rs7725218 or rs2242652 influence hTERT splicing together or separately, we generated an hTERT minigene containing both intron 3 and intron 4, and incorporated various combinations of the major and minor alleles at rs7725218, rs2242652 and rs10069690 (S1A Fig). Minigenes were transfected into the MCF7 breast cancer cell line, and the 27/87 ovarian cancer cell line. RT-PCR was performed using primers spanning introns 3 and 4. No additional alternative splice variants were identified with any of these SNPs, when present in combination or individually (S1B and S1C Fig).
The predicted hTERT INS1b protein lacks the reverse transcriptase domain, rendering it catalytically inactive. Therefore, we hypothesized that the relative amounts of FL-hTERT and INS1b in a cell directly determine telomerase activity levels. In order to investigate the function of INS1b, we disrupted the existing cellular balance of FL-hTERT and INS1b by exogenously expressing two different INS1b constructs that we expected would confer variable expression levels of the resulting proteins: a construct (pIRES-hTERT-INS1b) containing the complete cDNA of the alternatively spliced hTERT which consists of the FL-hTERT cDNA with the first 480 bp of intron 4 inserted, and another construct (pIRES-INS1b) which contains the hTERT-INS1b cDNA truncated at the stop codon in intron 4 (Fig 2A).
Variable levels of INS1b protein expression were consistently achieved with the pIRES-hTERT-INS1b and pIRES-INS1b constructs (Fig 2B), with greater expression being observed with the pIRES-INS1b construct. This was evident in multiple cell lines including HT1080 and MCF7, and following both transient and stable transfection (Fig 2B and 2C). Different expression levels were likely due to the relative sizes of the two constructs. The band detected by Western blot corresponding to the predicted size of INS1b (≈73 kD) was confirmed to be an hTERT variant by mass spectrometry analysis (S2 Fig).
To determine whether expression of INS1b had a direct effect on cell proliferation, growth curves were plotted for mass cultures of HT1080 cells stably expressing pIRES-hTERT-INS1b and pIRES-INS1b (Fig 2D and 2E). Protein expression was confirmed by Western blot analysis of immunopurified hTERT variants (Fig 2C). We observed robust expression of INS1b compared to endogenous FL-hTERT at early time points following stable transfection with pIRES-INS1b and pIRES-hTERT-INS1b, but not in the empty vector control. Interestingly, both the INS1b-expressing cultures spontaneously repressed the high INS1b to FL-hTERT protein ratios at later time points (Fig 2C), indicating a substantial selection pressure for FL-hTERT over INS1b. A small, but significant, reduction in growth rate was observed in the cell culture expressing pIRES-hTERT-INS1b compared to the empty vector control (Fig 2E). Overall, however, both INS1b expressing cell cultures maintained replicative capacity across 150 population doublings (pds) (Fig 2D).
These results were supported by the outcomes of transfecting the MCF7 cell line with the intron 4 minigenes containing either the major or minor alleles at rs10069690 (S3A Fig). A small, but significant, reduction in growth rate was observed with expression of the minigene containing the minor allele (A) (S3B and S3C Fig), consistent with the demonstrated production of INS1b transcript in this cell line (S3D Fig). However, both transfected MCF7 lines were also able to maintain replicative capacity over 70 pds. Notably, we were unable to obtain cultures stably expressing INS1b from the pIRES-INS1b construct, which supports the conclusion that expression of high levels of INS1b provides a selective disadvantage to the cells.
To determine whether altering the ratio of INS1b to FL-hTERT affects telomere length, we carried out terminal restriction fragment (TRF) length analysis on HT1080 cells overexpressing pIRES-INS1b or pIRES-hTERT-INS1b. Telomere length was measured approximately every 20 pds for a total of 160 pds (Fig 3A). Telomere length decreased slightly over this time course in the cultures expressing empty vector control, presumably due to gradual telomere length drift as has been observed previously in the HT1080 cell line [36]. In contrast, striking telomere shortening was observed during the first approximately 80 pds in both INS1b-expressing HT1080 cell cultures.
Telomere length was either restored (pIRES-INS1b) or stabilized (pIRES-hTERT-INS1b) at the later time points (Fig 3A). The mechanism underlying telomere length rescue appears to differ between the two cultures. In the pIRES-INS1b culture, the TRF pattern indicates that the small residual population of cells with long telomeres which was present at early pds eventually overgrew the cell population with short telomeres and dominated the culture at later pds. In the pIRES-hTERT-INS1b culture, telomeres underwent rapid shortening initially, but there was no further shortening after 70 pds (on the contrary, there was a small, gradual increase in length). However, even at the latest time points, telomere length did not fully recover and telomeres were substantially shorter than those of the empty vector or early pd cultures. Telomere length rescue coincided with the suppression of INS1b and restoration of FL-hTERT protein levels (Fig 2C).
To determine whether telomere shortening induced by INS1b overexpression resulted in an elevated telomere-specific DNA damage response (DDR), we quantified metaphase telomere dysfunction-induced foci (TIFs) on cytocentrifuged chromosomes from HT1080 empty vector, pIRES-INS1b and pIRES-hTERT-INS1b cell cultures at early (pd 33) and late (pd 190) timepoints (Fig 3B and 3C). An increase in TIFs was found to accompany telomere shortening in cells overexpressing INS1b. The number of TIFs was inversely correlated with telomere length.
To confirm the effects of the minor allele (A) on telomere length, TRF analysis was carried out on MCF7 cell cultures expressing the minigene constructs pIRES-hTERT Intron 4 Major (G) and pIRES-hTERT Intron 4 Minor (A) (S3E Fig). We utilized the minigene constructs to analyze telomere length changes in a more physiologically relevant system in which splice site utilization determines the balance between full-length and INS1b levels. The intron 4 Major (G) minigene generated FL-hTERT but no INS1b transcript (S3D Fig) and caused substantial telomere lengthening (S3E Fig; compare "Parental" lane to PD1). However, no telomere lengthening was observed with expression of the intron 4 minigene containing the minor (A) allele despite this construct producing both full-length and INS1b transcripts (S3D and S3E Fig). Telomere shortening was not observed, indicating that the balance of FL-hTERT and INS1b produced by this minigene is sufficient for telomere maintenance, but insufficient to support telomere lengthening.
To investigate the mechanism by which expression of INS1b causes a decrease in telomere length, we examined the effect INS1b expression has on overall telomerase activity levels. We conducted direct telomerase activity assays on HT1080 cell cultures expressing pIRES-neo empty vector, pIRES-INS1b and pIRES-hTERT-INS1b at three different pds (Fig 4A). Telomerase activity was significantly depleted in both INS1b-expressing cell cultures at early and middle time points, compared to their empty vector counterparts (Fig 4B). At the PD185 time point, telomerase activity in the pIRES-INS1b culture had recovered to a level that was not significantly different from the empty vector, which correlated with full telomere length recovery. Telomerase activity remained low in the pIRES-hTERT-INS1b culture where telomere lengths initially decreased and then stabilized or recovered slightly. Consequently, both telomerase activity and telomere length recovery correlated positively with FL-hTERT protein levels and negatively with INS1b protein levels, directly demonstrating an inhibitory role for INS1b in telomerase-mediated telomere maintenance.
INS1b is predicted to retain the hTERT N-terminal RNA binding domain, but lacks the C-terminal reverse transcriptase domain. To directly determine whether INS1b is able to bind to hTR, FL-hTERT protein and INS1b were immunopurified following their transient co-expression with hTR in the HEK293T telomerase-positive cell line and the GM847 telomerase-negative cell line. GM847 cells do not express endogenous hTERT. Dot blot analysis to detect immunoprecipitated hTR demonstrated that INS1b was able to bind to hTR in both HEK293T and GM847 cells (Fig 4C and 4E). The observation that INS1b binds hTR in GM847 cells in the absence of FL-TERT indicates that binding occurs directly, rather than indirectly by dimerization with FL-TERT protein. Quantification of hTR, FL-hTERT and INS1b expression levels by qRT-PCR demonstrate relative expression of these telomerase components (Fig 4D and 4F).
It has previously been reported that hTERT possesses telomerase-independent functions, including mitochondrial polymerase activity, involvement in apoptosis, as well as a potential role in the Wnt signaling pathway [37–40]. A recent report, however, failed to substantiate a consistent role for hTERT in promoting the expression of Wnt target genes in human cell lines [41]. To investigate whether INS1b displayed non canonical functions in Wnt signaling, we transiently overexpressed the minigene constructs pIRES-hTERT Intron 4 Major (G) and pIRES-hTERT Intron 4 Minor (A), as well as pIRES-INS1b in MCF7 cells and compared expression of Wnt pathway genes to the pIRES-neo empty vector control using Wnt signaling PCR arrays. No activation of Wnt signaling genes was detected with FL-hTERT, low level INS1b expression from the minigene construct, or with the high levels of INS1b overexpression achieved from the cDNA construct (S4 Fig). These data do not support a role for hTERT or INS1b in the transcriptional modulation of Wnt target genes in these cells. Overall, our data demonstrate that INS1b functions as a dominant negative inhibitor of telomerase activity, most likely by sequestering hTR in an inactive enzyme configuration.
Morpholino antisense oligonucleotides can be used to modify pre-mRNA splicing in the nucleus by targeting splice regulatory regions in a highly sequence specific manner [17,42]. To observe the effect of disrupting the balance of hTERT and INS1b endogenously, we designed a morpholino which bound and blocked the primary splice donor site at the exon 4/intron 4 boundary (Fig 5A). We delivered the morpholino (Ex4/Int4) into HEK293T cells, which are homozygous for the minor allele (A) at rs10069690, at a range of concentrations and performed RT-PCR 48 hours post-delivery. With increasing concentration, the morpholino inhibited FL-hTERT transcript production and enhanced INS1 and INS1b transcript levels (Fig 5B).
We conducted a direct telomerase activity assay using equivalent cell numbers to compare the Ex4/Int4 targeted morpholino to a standard non-targeting antisense oligonucleotide (ctrl), and observed a significant reduction in overall telomerase activity (Fig 5C and 5D). This directly demonstrates that endogenous telomerase activity can be inhibited by increasing the proportion of INS1b transcript to full-length transcript (Fig 5E), and indicates that manipulation of hTERT splicing may be a potential anti-telomerase therapeutic strategy, as proposed previously [43].
Candidate gene and genome-wide association studies have identified several SNPs in the TERT-CLPTM1L locus that are associated with cancer risk, and fine mapping has identified rs10069690 as a candidate causal SNP for breast and ovarian cancer risk. Currently, there are few direct functional studies that explain the biological mechanisms of such risk alleles, although a recent study has concluded that a substantial proportion of SNPs may modulate splicing in a tissue-specific way [44]. Here we demonstrate that the minor (A) allele of rs10069690 creates an additional, alternative splice donor site which results in the production of both FL-hTERT transcript and a variant transcript (INS1b) which has a premature stop codon and therefore encodes a severely truncated hTERT. This truncated INS1b protein lacks the reverse transcriptase domain, but maintains the hTR-binding domain, making it catalytically inactive and a dominant-negative inhibitor of telomerase activity. Other correlated SNPs (rs7725218 in intron 3 and rs2242652 in intron 4) were not found to alter hTERT splicing. Consistent with the association between rs10069690 and breast cancer risk, we found that the presence of the minor allele is associated with the presence of the INS1b transcript in telomerase-positive breast epithelial cells, but not in telomerase-positive fibroblasts from the same individual, suggesting that the balance between the two transcripts is subject to cell type-specific regulatory mechanisms.
We used overexpression studies to demonstrate that INS1b protein is produced, retains its hTR-binding capability, and inhibits telomerase activity determined by a quantitative, direct telomerase activity assay. Long-term overexpression of INS1b from two different cDNA expression constructs resulted in telomere shortening, which was accompanied by a telomere-specific DDR. This is most likely due to a dominant negative effect of the INS1b protein, because telomerase exists as a dimer [20] and its catalytic activity requires both hTERT active sites to be functional. It has previously been shown that disruption of the catalytic pocket in one of the two subunits exerts a dominant negative effect [20]. Therefore, even low levels of INS1b are likely to cause substantial effects on telomerase activity by dimerizing with active FL-hTERT molecules and sequestering hTR in catalytically inactive telomerase enzyme complexes. Recent studies have reported associations between minor alleles of rs2736100 [45,46] and rs7726159 [47] and longer telomeres in leukocytes, however population studies did not identify an association between the minor allele at rs10069690 and short mean telomere length [13]. It should be noted that the telomere length measurements were done in surrogate cells (i.e., peripheral blood cells rather than breast or ovarian epithelial cells), and that any effects of physiological levels of INS1b on telomere length in normal cells are likely to be more subtle than the effect of overexpressing this variant transcript in cancer cells.
Multiple hTERT splice variants have been identified at variable, and often abundant, levels in telomerase-positive cancer cell lines [48]. As with INS1b, the vast majority involve truncations of the C-terminus and generate catalytically inactive products [12,49,50]. The abundant hTERT β-deletion splice variant, which results from a 183 nucleotide deletion between exons 6 and 9, was found to be translated into a truncated protein that competed for binding to hTR and inhibited endogenous telomerase activity [19]. Nevertheless, cancer cell lines are able to proliferate rapidly in the presence of abundant inactive splice variants. The mean telomere length of most tumors is shorter than that of the surrounding normal tissue, suggesting that a relatively short, optimal telomere length is selected for and maintained in tumor cells, possibly to decrease the replicative burden presented by longer telomeres. It is possible that expression of alternative splice variants is selected for in telomerase-positive tumors as one means of preventing excessive telomere lengthening from occurring. Conversely, we observed that telomerase-positive cell lines prevented excessive telomere shortening as a result of long-term INS1b overexpression by down-regulating expression of INS1b from the cDNA constructs, via mechanisms that we did not examine, consistent with the notion that there is selection for an optimal telomere length in cancer cells.
Minigene constructs that included intron 4 with each allele at rs10069690 were used to study effects of the additional splice site in a more physiological manner than the cDNA overexpression constructs allowed [51]. Over-expression of FL-hTERT caused telomere lengthening in HT1080 cells [36,52], and this occurred as expected with the minigene containing the major allele, which expresses FL-hTERT but not INS1b. In contrast, we identified a telomere-lengthening defect conferred by the presence of the risk-associated minor (A) allele which co-expresses both FL-hTERT and INS1b. Unlike the INS1b cDNA constructs which encode only INS1b, the minigene did not lead to a telomere shortening phenotype, further supporting the conclusion that the relative levels of INS1b and FL-hTERT affect telomerase activity and telomere length.
By characterizing endogenous INS1b in cell lines homozygous and heterozygous for the minor allele at rs10069690, we identified highly variable levels of INS1b relative to FL-hTERT across different cell lines. Most strikingly, we observed high levels of endogenous INS1b expression in the breast epithelial cell line B80-T8, whereas there was no detectable INS1b expression in the genetically matched fibroblast cell line F80-T1. This demonstrates that there are clear differences in regulation of hTERT splicing between different cell types, including cells isolated from different tissues of the same individual. Indeed, it has previously been shown that differential alternative splicing of hTERT transcripts occurs non-randomly in different tissue types during human development [15]. Differences in splicing may arise from alterations in the chromatin environment encompassing rs10069690, which may render differences in binding of splicing regulatory factors and favor the use of a particular splice site. This is supported by our previous study, in which we used site-specific formaldehyde-assisted isolation of regulatory elements (FAIRE) analysis of an approximately 1 kb region spanning rs10069690, and identified open chromatin signatures in breast stromal and myoepithelial/stem cell samples, whereas closed chromatin signatures were observed in progenitor and differentiated luminal epithelial cell fractions [13].
Finally, to demonstrate the effect of endogenous INS1b we introduced morpholino antisense oligonucleotides designed to sterically block FL-hTERT splicing which resulted in the preferential use of the alternative rs10069690 splice donor site, shifting the balance of transcripts in a dose-dependent manner in favor of INS1b. This resulted in decreased telomerase activity, demonstrating the inhibitory effects of increasing proportions of INS1b transcript in the absence of any exogenous expression constructs. Modulation of hTERT splicing is therefore a potential therapeutic strategy for limiting (or, by inhibiting alternative splicing, increasing) telomerase activity [10,17].
In summary, we provide the first evidence that a well-documented cancer risk-associated polymorphism in the TERT-CLPTM1L locus encodes an alternative splice variant of hTERT, which results in a catalytically inactive telomerase enzyme complex (Fig 6). Because hTERT and hTR levels are both limiting, this results in decreased telomerase activity and limits telomere lengthening, consistent with a dominant negative effect. The balance between INS1b and FL-hTERT levels is critical, and appears to be regulated in a cell-specific manner. We suggest a model whereby expression of INS1b is of functional relevance during development in tissue-specific stem cell populations in which telomerase activity is limiting. Subtle defects in telomerase-mediated telomere extension may result in shorter telomere lengths, and will cause cells to reach proliferative exhaustion or replicative senescence, earlier than in individuals without the risk allele. The increased risk of genetic instability ensuing from shorter telomeres is predicted to increase the risk of tumorigenesis. Cancer risk may be further increased in the context of BRCA1 mutations, which suppress DNA repair mechanisms [53]. This is consistent with the minor allele at rs10069690 conferring additional breast cancer risk in BRCA1 mutation carriers [13]. Therefore, the minor allele at rs10069690 creates an avenue for regulation of telomerase activity where it can increase the risk of cancer-predisposing short telomeres in normal adult tissues and contribute to optimal telomere maintenance in tumor cells.
Summary of experimental design (Fig 7).
pIRES-hTERT [18], pIRES-hTERT Intron 4 Major (G), pIRES-hTERT Intron 4 Minor (A), pIRES-hTERT Intron 4-rs2242652 (A) and pIRES-hTERT Intron 4-(A)(A) (minor alleles at both sites) were developed from pIRES-neo (Clontech) as described previously [13]. A construct containing the first 480 bp of intron 4 of hTERT in a pUC57 vector (GenScript) was subcloned into pIRES-hTERT to generate pIRES-hTERT-INS1b. To generate a plasmid (designated pIRES-INS1b) containing the translated sequence of INS1b, this region was PCR amplified from pIRES-hTERT Intron 4 Major (G) and cloned into pIRES-neo.
Site directed mutagenesis (Agilent Technologies) was used to introduce the minor allele at rs7725218 into pUC57 intron 3 (GenScript). Each version of intron 3 and fragments of pIRES-hTERT Intron 4 Major (G) and pIRES-hTERT Intron 4(A)(A) were then PCR amplified and ligated using the In-Fusion HD Cloning Kit (Clontech). The minigenes generated were designated pIRES-hTERT Intron 3(G)/Intron4(G)(G) (major alleles at all intron 3 and 4 sites), pIRES-hTERT Intron 3(G)/Intron 4(A)(A) (minor alleles at rs2242752 and rs10069690), pIRES-hTERT Intron 3 (A)/Intron 4(G)(G) (minor allele at rs7725218) and pIRES-hTERT Intron 3(A)/Intron 4(A)(A) (minor allele at all sites). All constructs were validated by sequence analysis prior to transfection.
Cell lines were purchased from American Type Culture Collection (ATCC), with the exception of the B80-T8 and F80-T1 telomerase-positive cell lines [35] and the Fre-71s-1 cell strain [54] which were established by L. Huschtscha. Culture media and supplements were purchased from Invitrogen Life Technologies. MCF7, HEK293T, HT1080, MDA-MB-157, CAOV-3, Fre-71s-1 and F80-T1 cells were cultured in Dulbecco’s modified Eagle’s medium (DMEM) with 10% v/v Fetal Bovine Serum (FBS). MDA-MB-415 cells were cultured in DMEM with 15% FCS and 10 μg/mL insulin. A2780, BT549 and 27/87 cells were cultured in RPMI 1640 medium with 10% FBS. B80-T8 cells were cultured in a 1:1 ratio of RPMI 1640 and MCDB-170. All cells were maintained at 37°C and 5% CO2. Cell lines were authenticated by 16-locus short tandem repeat profiling and were confirmed to be free of Mycoplasma species by CellBank Australia (Children’s Medical Research Institute, Westmead, New South Wales, Australia).
pIRES-hTERT Intron 4 Major (G) and pIRES-hTERT Intron 4 Minor (A) were transiently transfected into HT1080, MCF7, A2780 and Fre-71s-1 cells using siPORT NeoFX Transfection Agent (Life Technologies) and harvested after 24 hours for RT-PCR analysis.
For all other overexpression experiments, plasmid constructs were transfected into HEK293T, HT1080, MCF7 and GM847 cells using Fugene 6 Transfection Reagent (Promega) and harvested after 24–72 hours. Overexpression was confirmed using quantitative reverse transcription PCR (qRT-PCR) analysis.
pIRES constructs were stably overexpressed in HT1080 and MCF7 cells, followed by 48 hours recovery and treatment with G418 at 700 μg/mL for HT1080 cells and 800 μg/mL for MCF7 cells. The surviving cells were grown to 95% confluency before initiating growth curve analysis. At this point the cell population was designated population doubling (pd) 1. To measure growth rates of each cell line, cells were counted and seeded at known numbers every 2 to 4 days while being maintained at exponential growth phase.
Genomic DNA was genotyped using a custom Fluidigm 96 multiplex and SNP Type chemistry. Assays were designed by Fluidigm and performed according to the manufacturer’s protocol using a Fluidigm Biomark HD. To ensure quality control, genotyping cluster plots were inspected visually and samples which failed >20% assays were removed from analysis.
The morpholino antisense oligonucleotide “Ex4/Int4 Morph” (5'-TTAAACCAAAGCACAGCCACCCTCT-3') and Standard Control Oligonucleotide (5'CCTCTTACCTCAGTTACAATTTATA-3') were synthesized by Gene Tools. Concentrations of 2–10 μM of Ex4/Int4 Morph were delivered to HEK293T cells with 6 μL/mL Endo-Porter delivery reagent (Gene Tools) and cells were harvested after 48 hours.
Total RNA was extracted using the RNeasy Mini Kit (Qiagen) and DNase I digested (Life Technologies). cDNA was synthesized from 1–5 μg of RNA using the SuperScript III First-Strand Synthesis System (Life Technologies). hTERT and INS1b transcripts were amplified using primers spanning from the exon 3/4 junction to exon 6 (Intron 4 F1: 5'-GGAATCAGACAGCACTTGAAGAGGGT-3'; Intron 4 R3: 5'-TTTGATGATGCTGGCGATGACCTC-3') with Abgene Recombinant Taq DNA polymerase and Buffer (ThermoScientific), 2 mM MgCl2, 2 mM of each dNTP (Roche) and 10% dimethyl sulfoxide (DMSO) (Sigma) for 30 cycles. Intron 3 was amplified using primers spanning Exon 3 to Exon 4 (Intron 3 F1: 5'- AGATCCTGGCCAAGTTCCTGC-3'; Intron 3 R1: 5'-CGACGTAGTCCATGTTCACAATCG-3') and intron 4 amplified using primers spanning Exon 4 to Exon 6 (Intron 4 F4: 5'- ATTGTGAACATGGACTACGTCGTGGG-3'; Intron 4 R3: 5'- TTTGATGATGCTGGCGATGACCTC-3'). For quantitative transcript measurements, GAPDH transcripts were amplified (GAPDH F: 5'- ACCCACTCCTCCACCTTTG-3'; GAPDH R: 5'-CTCTTGTGCTCTTGCTGGG-3') under the same conditions, without DMSO, for 21 cycles. The products were resolved on a 1.3% agarose gel with 500 ng/mL ethidium bromide alongside NEB 100 bp DNA ladder and visualized by the Alpha Innotech FluorChem 5500 system. Quantification was performed by densitometry using ImageQuant software. To quantify overexpression after transient transfection with hTERT, INS1b and hTR constructs, qRT-PCR was performed using SYBR Green (Roche) with LightCycler 96 (Roche) according to the manufacturer’s instructions. One primer set was used to measure both FL-hTERT and INS1b overexpression (MS TERT F: 5'-AAGTTCCTGCACTGGCTGATGAGT-3', MS TERT R: 5'-CACCCTCTTCAAGTGCTGTCTGAT-3') and a another to quantify hTR (hTR F: 5'-CTAACCCTAACTGAGAAGGGCGTA-3', hTR R: 5'- GGCGAACGGGCCAGCAGCTGACATT-3').
Whole cell lysates or telomerase immunoprecipitations were separated by SDS-PAGE and Western Blotting performed using the NuPAGE system (Invitrogen) as per manufacturer’s instructions. Primary antibodies used include HTCS2 sheep anti-TERT [55] at 0.5 μg/mL for INS1b and overexpressed FL-hTERT detection, Rockland rabbit anti-telomerase catalytic subunit diluted 1:500 in PBS-T (8 mM Na2HPO4, 150 mM NaCl, 2 mM KH2PO4, 3 mM KCl, 0.1% Tween 20) for endogenous hTERT detection [56] and Sigma rabbit anti-tubulin diluted 1:500. The secondary antibodies used were rabbit anti-sheep immunoglobulin horseradish peroxidase (Dako) diluted 1:2000 and goat anti-rabbit immunoglobulin horseradish peroxidase (Dako) diluted 1:5000.
FL-hTERT and INS1b were immunopurified using an hTERT antibody (HTCS2), as described previously [57].
Dot blotting of immunopurified FL-hTERT and INS1b samples with probe against hTR was performed as described previously [58].
The direct telomerase activity assay was performed on immunopurified telomerase normalized by hTR amount or cell number as described previously [57].
Terminal restriction fragment (TRF) analysis to determine telomere length was performed as described previously [36].
Meta-TIF analysis was performed as described previously [59] with the exception that after cytocentrifugation the cells were permeabilized in pre-extract buffer (20 mM HEPES, 20 mM NaCl, 5 M MgCl2,, 300 mM Sucrose, 0.5% v/v NP40) with gentle shaking, washed once in PBS-T and washed twice in PBS before fixing in PBS with 4% v/v formaldehyde.
MCF7 cells were harvested 48 hours after transfection with pIRES-neo, pIRES-hTERT Intron 4 Major (G), pIRES-hTERT Intron 4 Minor (A) and pIRES-INS1b, RNA was extracted using the RNeasy Mini Kit (Qiagen) and cDNA was synthesized using the RT2 First Strand Kit (Qiagen). The cDNA samples were then loaded on to qPCR human Wnt signaling PCR arrays (PAHS-043YF-2) and relative transcript levels of each gene measured by qPCR on LightCycler 96 (Roche) according to the manufacturer’s instructions. Results were analyzed on the RT2 Profiler PCR Array Data Analysis pathway software (SABiosciences).
hTERT and INS1b immunopurified samples were prepared and separated on a 4–12% NuPAGE Novex Bis-Tris Mini Gel (Life Technologies). The gel was stained with Coomassie Staining Solution (0.0025% w/v coomassie brilliant blue G, 45% methanol and 10% glacial acetic acid) overnight at room temperature and washed with destaining solution (45% methanol and 5% glacial acetic acid) for 4–6 hours. The gel was rehydrated with two 15 min washes in water and bands corresponding to the sizes of hTERT (125 kD) and INS1b (73 kD) proteins excised. Gel pieces were rinsed with three 10 min washes in water. They were then destained at 37°C in wash solution (50% v/v acetonitrile, 12.5 mM NH4HCO3 pH 7.8) and dehydrated using GeneVac EZ-2 plus. To digest the proteins at lysine residues, 15 ng/μl trypsin (sequencing grade; Promega) in 25 mM NH4HCO3 pH 7.8 was added directly to the gel pieces and, after they became translucent, 25 mM NH4HCO3 pH 7.8 was added to cover the gel pieces. Samples were incubated at 37°C overnight with shaking. They were then briefly spun down, 0.5% v/v trifluoroacetic acid (TFA) was added to cover gel pieces and samples were incubated in a water bath sonicator for 15 min. Samples were loaded onto activated Eppendorf GELoader tips with filter plugs (3M Empore C18 filter membrane), washed in 0.5% v/v TFA and eluted in 70% v/v acetronitrile and 0.5% v/v TFA solution. Samples were spotted onto an Opti-TOF 384 Well Insert with equal volumes of matrix (70% v/v acetonitrile, 0.5% TFA). The insert was analysed by MS/MS on a MALDI-TOF/TOF AB SCiex 5800 system and protein identification performed by the Mascot program.
|
10.1371/journal.pgen.1005397 | Evidence for a Common Origin of Blacksmiths and Cultivators in the Ethiopian Ari within the Last 4500 Years: Lessons for Clustering-Based Inference | The Ari peoples of Ethiopia are comprised of different occupational groups that can be distinguished genetically, with Ari Cultivators and the socially marginalised Ari Blacksmiths recently shown to have a similar level of genetic differentiation between them (FST ≈ 0.023 − 0.04) as that observed among multiple ethnic groups sampled throughout Ethiopia. Anthropologists have proposed two competing theories to explain the origins of the Ari Blacksmiths as (i) remnants of a population that inhabited Ethiopia prior to the arrival of agriculturists (e.g. Cultivators), or (ii) relatively recently related to the Cultivators but presently marginalized in the community due to their trade. Two recent studies by different groups analysed genome-wide DNA from samples of Ari Blacksmiths and Cultivators and suggested that genetic patterns between the two groups were more consistent with model (i) and subsequent assimilation of the indigenous peoples into the expanding agriculturalist community. We analysed the same samples using approaches designed to attenuate signals of genetic differentiation that are attributable to allelic drift within a population. By doing so, we provide evidence that the genetic differences between Ari Blacksmiths and Cultivators can be entirely explained by bottleneck effects consistent with hypothesis (ii). This finding serves as both a cautionary tale about interpreting results from unsupervised clustering algorithms, and suggests that social constructions are contributing directly to genetic differentiation over a relatively short time period among previously genetically similar groups.
| While it is widely recognized that DNA patterns vary across world-wide human populations, the primary features that drive these differences are less well understood. As an example, the Ari peoples of Ethiopia are presently socially divided according to occupation, with Ari Blacksmiths marginalised relative to Ari Cultivators. Two competing theories proposed by anthropologists to explain the existence of these occupational groupings suggest very different histories: (i) the Cultivators reflect migrants who moved into the region occupied by ancestors of the Blacksmiths perhaps many thousands of years ago, versus (ii) the Blacksmiths and Cultivators comprised the same ancestral group before the former was marginalised due solely to their trade. Recent genetic studies showed that Blacksmiths and Cultivators are distinguishable by their DNA, and suggested that overall DNA patterns among the two groups were consistent with (i). However, we demonstrate here that interpreting the results of currently popular algorithms that compare DNA is not always straight-forward. Instead we use a variety of analyses to show that (ii) seems a more likely explanation, perhaps illustrating how social marginalisation can lead to groups becoming genetically distinguishable over a relatively short time period.
| Different ethnic groups in present-day Ethiopia show a substantial amount of cultural [1] and genetic [2] diversity. Some of this diversity falls along societal divisions, e.g. across distinct groups that are segregated through social barriers to interaction and co-operation [1]. Marginalised groups are largely comprised of craft workers (artisans) and hunters [3]. For example, the Ari Cultivators, who are farmers, are said to have limited interaction with the Ari Blacksmiths, who specialize in iron and wood-work and live on the periphery of settlements [4]. Blacksmithing communities are widely regarded as the most marginalised of artisan groups, not just within the Ari but throughout southern Ethiopia [3].
Two alternative hypotheses proposed by anthropologists to explain the origin of marginalised groups in Ethiopia, such as are present in the Ari community, imply very different ancestral histories [1, 3]:
Remnants model (RN)—Under the Remnants model, originally proposed by Biasutti (1905) [5, 1], the Ari Blacksmiths are designated as an early, possibly hunter-gatherer group which existed in Ethiopia prior to the arrival of farmers. The arrival of the Cultivators displaced the remnant group, resulting in the Blacksmiths becoming segregated from society. Marginalisation model (MA)—Under the internal specialisation or Marginalisation model [6, 7, 3], the Ari Blacksmiths and Cultivators share the same ancient history. The adoption of an artisan trade by the Blacksmiths led to their marginalisation within the existing society.
Studying patterns of DNA variation among Ari occupational groups can help shed light on which of these theories is more likely. Under the MA model, which is currently favoured among anthropologists to explain the existence of caste-like occupational groups in southwest Ethiopia [1], observed genetic differences between the two groups should be explained largely by a bottleneck effect in the Blacksmiths consistent with their current isolation, even if the two groups only became isolated from each other very recently. In contrast, under the RN model the two groups descend from two anciently related groups that split perhaps many thousands of years ago, though possibly with subsequent admixture between them. There are alternative theories to the MA and RN hypotheses, including one suggesting the Blacksmiths—along with other artisan groups—migrated to southern Ethiopia after it was occupied by Cultivators, either due to demand for their craft skills or possibly while accompanying invading groups [8, 9]. Here we assume such migrations would result in a genetic relationship between Blacksmiths and Cultivators similar to that expected under the RN model, i.e. such that the two groups split from one another substantially further in the past than under the MA model.
We note that these models are not mutually exclusive [1], as even under a RN model there may have been substantial recent bottleneck effects in the Blacksmiths, as might be expected given their present-day marginalisation. Nonetheless, even after accounting for any bottleneck effects in the Blacksmiths, the RN model implies likely additional genetic differentiation between the two groups due to their ancient relatedness, as we demonstrate using simulations. For example, assuming the remants group consisted of hunter-gatherers [5], the Cultivators might look more genetically similar to other agricultural groups within Ethiopia than the Blacksmiths do.
The most comprehensive genome-wide study of Ethiopians to date [2] analysed 235 individuals from 10 Ethiopian groups, including Ari Blacksmiths and Ari Cultivators. They found that the genetic differentiation between the two Ari occupational groups (FST = 0.04) was at a similar level to that observed between multiple ethnic groups sampled across Ethiopia (FST range 0.02–0.06). The authors used ADMIXTURE [10] to assign individuals’ genetic variation data into clusters based on shared allele frequency patterns, using an “unsupervised” approach which allows each individual’s genetic data to be assigned to multiple clusters. They noted that the Ari Blacksmiths were assigned almost entirely to a single cluster and that a smaller proportion of this cluster was found at varying levels in all other Ethiopian groups including the Cultivators. The authors suggested the ADMIXTURE results were consistent with the RN model of Ari Blacksmith origins, with subsequent assimilation of their indigenous ancestors into the expanding farming community (including the ancestors of present-day Ari Cultivators) [2]. More recently, other researchers [11] applied the same unsupervised model of ADMIXTURE to these data and additional world-wide samples. They similarly suggested that Cultivators were likely the result of admixture between an ancestral group best represented by the Blacksmiths and another ancestral group that diverged from the Blacksmith-like group > 31kya [11]. Note that the original Remnants model proposed by anthropologists, which indeed is not mentioned in [11], does not on its own imply one-way migration from the ancestors of the Blacksmiths into those of the Cultivators, but we will assume that is the case here to match the observations of these two papers. I.e. our RN model, as simulated below, assumes this asymmetrical migration took place, making the groups more similar genetically than they would otherwise be.
Clustering algorithms such as ADMIXTURE [10] and the closely related approaches STRUCTURE [12, 13] and FRAPPE [14] have been applied in a similar manner in many previous studies to explore the genetic ancestries of world-wide [15] and geographically localized [2] populations. For example, STRUCTURE has been used to suggest the presence of distinct (perhaps anciently-related) ancestral groups that have intermixed to form present-day populations in Africa [16]. However, similar to using principal-components analysis (PCA) [17, 18], it can be difficult to assess whether clustering patterns among groups are due to recent admixture between distinct historical populations or to ancestry shared prior to the populations diverging [15], making interpretation challenging.
We used an alternative approach to study 237 samples from 10 Ethiopian and 2 neighbouring (Somalia, South Sudan) populations from [2], which we will refer to as the “Pagani” samples. We also incorporated 850 additional samples from 10 other groups from the 1000 Genomes Project (hencefore “1KGP”; http://www.1000genomes.org/) and 28 individuals from one group (MKK) from HapMap Phase3 [19], giving 23 total labeled populations (Fig 1a, S1 Table). We jointly phased all samples with the program SHAPEIT [20] using 659,857 SNPs. We then used CHROMOPAINTER [21] to explore patterns of haplotype sharing among individuals, which has been shown to be both more powerful than techniques that ignore haplotype information [21] and less susceptible to biases arising from SNP ascertainment schemes [22, 23] such as those leading to the chip data used here. Specifically, CHROMOPAINTER uses a Hidden-Markov-Model (HMM) approach [24, 21] to “paint” each haplotype of a sampled “recipient” individual, identifying—at each location of each recipient’s two haploid genomes—the best matching DNA segment from a set of sampled “donor” individuals. I.e. it infers the donor haplotype with which the recipient shares most recent ancestry relative to all other donor haplotypes at the given genomic locus. Using this approach, for each recipient individual we infer their proportion of genome-wide DNA that shares most recent common ancestry with each donor haplotype, identifying the donors (and groups of donors) that appear to be most related genetically to the recipient individual. By comparing results when using different donor sets, we can distinguish whether genetic differences between groups are more likely attributable to ancient or recent isolation, as described below.
We first clustered the 1115 individuals into 17 groups using CHROMOPAINTER and fineSTRUCTURE [21] (Fig 1a, S1–S4 Figs and S1 Table), removing 56 individuals with ancestry signals inconsistent with that of the majority of individuals with the same population label (S1 and S2 Figs) or that failed other quality control metrics, including 7 Ari Blacksmiths and 1 Ari Cultivator (see Methods). In total, our dataset analysed 10 Ari Blacksmiths (ARIb) and 23 Ari Cultivators (ARIc).
To distinguish between the MA and RN hypotheses, we performed three distinct CHROMOPAINTER analyses that differ in which of the 17 groups are used as donors:
all-donors—recipient groups copy from (i.e. are painted using) all other sampled groups (i.e. MKK and all Pagani and 1KGP groups, including their own) as donors non-Ari-donors—recipient groups copy from all other sampled groups except the ARIb and ARIc as donors non-Pagani-donors—recipient groups copy from 1KGP and MKK groups only as donors
Under each of (A)-(C), we infer a “painting profile” for each individual and world-wide group by measuring the amount of DNA that they copy from each donor group. To compare the ARIb and ARIc under each of (A)-(C), we use a distance-based measure (“total-variation-distance (TVD)”; [25]) that calculates the difference TVDXY in the average “painting profiles” between any two groups (or two individuals) X and Y (see Methods). To account for independent drift effects along independent regions of the genome, we also constructed an alternative measure FXY that scales TVDXY by differences among chromosomes within each of X and Y (see Methods).
Like the unsupervised ADMIXTURE analyses of [2] and [11], our analysis (A) allows any sampled individual to copy from any other individual regardless of group label. In contrast, analyses (B) and (C) compose the Blacksmiths (“ARIb”) and Cultivators (“ARIc”) as genetic mixtures of other non-Ari sampled groups only, which is more similar to a “supervised” ADMIXTURE analysis that pre-defines some clusters using surrogate groups. The important distinction is that ARIb and ARIc are allowed to copy from individuals with their own label only under (A). In a scenario where the Blacksmiths and Cultivators shared identical ancestry prior to recent isolation of the Blacksmiths (i.e. MA hypothesis) and have received no DNA from outside groups since, the inferred ancestry patterns of the two groups are expected to be similar under analyses (B) and (C) even if they are very different under analysis (A) [25]. I.e. analyses (B) and (C) would substantially attenuate the signal of genetic differentiation between the two Ari groups under analysis (A) if that signal is attributable solely to strong bottleneck effects in either of the groups after their split. In contrast, under the RN hypothesis the two Ari groups are expected to look genetically different under analyses (B) and (C) in addition to analysis (A), so long as one of the Ari groups is more recently related to at least one other sampled group, regardless of any bottleneck effects in either group since their split. The key difference between analyses (B) and (C) is that the latter allows the comparison of genetic differences between the ARIb and ARIc to those between other geographically near groups in Ethiopia and surrounding areas, under a scenario where each such group uses identical donors and importantly is not allowed to copy from individuals within their own label. Meanwhile, analysis (B) might have more power to distinguish between the two Ari groups, since it uses more geographically near groups as donors compared to analysis (C).
We illustrate expected genetic patterns under analyses (A)-(C) by performing several simulations designed to capture key features of the Marginalisation and Remnants hypotheses, incorporating one-way migration in the latter to be consistent with previous interpretations of ADMIXTURE results [2, 11]. These include the following four different “full” simulations that simulate 13 world-wide populations with FST values matching that of several of our sampled populations (Fig 2a, S2, S4–S7 Tables):
“MA”—The simulated “Ari” groups split 20 generations ago, followed immediately by a strong bottleneck in the simulated “ARIb”. “RN”—The “Ari” groups split 1700 generations ago, after which migrants from “ARIb” form ≈ 50% of the simulated “ARIc” population over a period 200–300 generations ago. “RN+BN”—The “Ari” groups split 1700 generations ago with subseqent migration from “ARIb” into “ARIc” as in “RN”, followed by a strong bottleneck in the “ARIb” starting 20 generations ago. “RN+BN+80%”—The “Ari” groups split 1700 generations ago with subseqent migration from “ARIb” into “ARIc” as in “RN” but forming ≈ 80% of the “ARIc” population, followed by a strong bottleneck in the “ARIb” starting 20 generations ago.
While it is difficult to discern appropriate parameters for these simulations given the uncertainty surrounding the history of groups in this region, we followed values proposed by [11] as a guide for our Remnants (“RN”) simulations. In particular the authors suggested that the Cultivators likely resulted from a mixture between a group represented by the Blacksmiths and another group that diverged from the Blacksmiths-like group at least 31kya [11]. For the Marginalisation (“MA”) simulations, the aim was to determine whether a very recent split time between the two groups, which we chose as 20 generations, followed by a strong bottleneck in the simulated Blacksmiths can explain observations similar to those we see in our data. We also performed an additional 24 “simplified” simulations that considered only 7 populations in order to explore how different split times between the “ARIb” and “ARIc”, rates of migration from “ARIb” into “ARIc”, and strength of “ARIb” bottleneck affect our power to distinguish the two groups using our CHROMOPAINTER analyses (Fig 2b, S8 Table) under a hypothetical Remnants setting. Throughout we compare the results from our simulations to those from the real data.
Our FST (Fig 1b, S3 Table), ADMIXTURE (Fig 1c), fineSTRUCTURE (S1 Fig) and CHROMOPAINTER analysis (A) (Fig 3a, S7a and S11a Figs) results support previous findings [2, 11] that the ARIb appear genetically distinct from the ARIc.
Below we first show how these results and those from [2] and [11] can be consistent with an MA hypothesis, i.e. where the Blacksmiths and Cultivators have a relatively recent split time with the Blacksmiths experiencing a subsequent strong bottleneck. We then outline several lines of evidence using CHROMOPAINTER analyses (A)-(C) that support the MA over the RN hypothesis as the more plausible explanation of observed DNA patterns among the Ari given these sampled data. In particular the MA hypothesis would predict the following genetic patterns, any one of which is not necessarily expected to be true under the RN hypothesis and thus jointly provide substantial support for the alternative MA hypothesis:
Any differences in inferred ancestry between the Blacksmiths and Cultivators can be explained by bottleneck effects in the Blacksmiths. The Blacksmiths and Cultivators are similarly related genetically to other groups, both within and outside of Ethiopia. After accounting for drift effects likely attributable to a bottleneck in the Blacksmiths, genetic differences between Blacksmiths and Cultivators are similar to differences among Cultivators. The Blacksmiths and Cultivators have similar signals of recent admixture from other sources, including sources likely from both inside and outside of Ethiopia. DNA segments inherited from distinct admixing sources are genetically similar among Blacksmiths and Cultivators. Furthermore, segments from these differenct sources within Blacksmiths show the same strength of bottleneck effects, consistent with the split between the Blacksmiths and Cultivators occurring more recently than the recent admixture.
Under a hypothetical RN setting, we again note that in order to have any power to distinguish the ARIc and ARIb genetically, our analyses must include at least one sampled group whose ancestors split more recently from those of the ARIc than those of the ARIb and ARIc split from each other. Our dataset contains several Ethiopian groups (Afar, Amhara, Anuak, Tigray, Wolayta) assigned as agriculturalists in [27], whose ancestors plausibly could have split more recently from those of the Cultivators than those of the Blacksmiths and Cultivators split from each other, under a hypothetical RN setting. Also, two groups (ORO,MKK) have FST values with ARIc that are lower than those between ARIb and ARIc (Fig 1b, S3 Table), suggesting either or both could represent such a sampled group(s).
We further note that one-way migration from the ancestors of the Blacksmiths into those of the Cultivators, as suggested in [2, 11], will mitigate any genetic differences between the two Ari groups today even if the RN model were true. We assess our power to distinguish the two Ari using simulations under an RN setting, in particular determining the amount of one-way migration that would be necessary to explain observed genetic patterns.
While there are an infinite number of historical scenarios that could be consistent with observed genetic patterns in present-day Blacksmiths and Cultivators, some of which may reflect the RN model, our primary aim is to assess whether the MA hypothesis alone can fit the observations of [2] and [11] as well as the further analyses we perform here. In such a case, we propose the MA model is a more parsimonious explanation given the current marginalised status of Blacksmiths.
If the Blacksmiths experienced a strong bottleneck relative to the Cultivators, then the genetic diversity among Blacksmiths should be lower than that among Cultivators. Consistent with this, the inferred proportion of Identity-by-descent (IBD) sharing among the ARIc is lower than that among the ARIb (PLINK v1.07 [28] PI_HAT = 0.08 compared to 0.18; S13 Table). Indeed, the inferred proportion of IBD sharing was higher for the ARIb than all other groups in our study, with the next highest the Japanese (JPT; PI_HAT = 0.15; S13 Table).
As separate evidence using a different approach, we painted each ARIb separately with CHROMOPAINTER using only other ARIb as donors, and analogously painted ARIc using only other ARIc as donors, after first matching the two groups for sample size (see Methods). In contrast to the IBD approach implemented in PLINK, our haplotype-based approach should be robust to any potential biases arising from ascertainment of chip data [22]. Supporting this, the inferred average size of shared haplotype segments in the ARIb, which we propose as a measure of relative homogeneity under this approach, is no longer the highest out of all 17 groups and instead is lower than that of GBR and FIN (S14 Table, S14 Fig). This pattern is more consistent with the presumed recent bottlenecks in these latter two populations [25, 29] following the major out-of-Africa bottleneck event [30, 31]. Nevertheless under this second analysis, the median length of matching haplotype segments among ARIb is ≈2 times higher than in ARIc (S14 Table), again consistent with bottleneck effects in the ARIb.
In our “MA” full simulations consistent with the Marginalisation hypothesis, we simulated a split time between the Blacksmiths and Cultivators of only 20 generations ago, and then chose a strength of bottleneck in the simulated Blacksmiths that gave a value of FST = 0.025 between the two groups (S4 Table), which is very similar to that of our observed data (FST = 0.023; S3 Table). Under this set-up, we note that the patterns seen in ADMIXTURE results (S6a Fig) and CHROMOPAINTER analysis (A) (S12 Fig, top) are very similar to that observed in the real data. This suggests that a bottleneck event in the Blacksmiths, with a very recent split time between the two Ari groups as expected under the MA model, can explain genetic differences observed between them under these approaches.
The differences we observe between ARIb and ARIc under CHROMOPAINTER analysis (A), ADMIXTURE and FST are no longer present under CHROMOPAINTER analyses (B) and (C) (Figs 1b–1c and 3, S7 and S11 Figs and S10–S12 Tables). A key difference is that for each Ari group we disallow “self-copying” from individuals with the same label under analyses (B) and (C), which should reduce the magnitude of any differences seen in the other approaches that are attributable to bottleneck effects in the Blacksmiths.
Under our approach, we measured differences in inferred ancestry using TVD (Fig 3, S15 Fig, S15 Table). We also used an alternative score FXY (S17 Fig, S16 Table) that is proportional to the TVD score between individuals/groups X and Y but scales this value by ancestry differences across chromosomes within each individual/group to incorporate independent drift effects along the genome (see Methods). Relative to comparisons between other Pagani groups, each of these measures dropped substantially in analyses (B)-(C) compared to analysis (A) when comparing the two Ari groups (S15 and S16 Tables). We also clustered all Ari individuals into two groups based on their inferred ancestry using a novel statistical Markov-Chain-Monte-Carlo (MCMC) algorithm (see Methods). This algorithm correctly classified all Ari individuals by occupational label under analysis (A) but randomly assigned them to the two clusters under analyses (B) and (C) (S19 Table) despite separating other Pagani groups under analyses (A)-(C) (S20–S22 Figs). Furthermore, clustering and TVDXY, FXY patterns closely follow those when applying the same methods to the simulated “Ari” individuals in the “MA” “full” simulations and noticeably differs from the three “RN” “full” simulation scenarios we considered (S15–S18 Figs and S19 Table).
Informatively, even though all Pagani groups are painted using identical donors under analysis (C), the TVDXY and FXY scores between the two Ari groups under analysis (C) are smaller than those between any two other Pagani groups. In particular they are smaller than TVDXY and FXY between groups “AFA” and “ORO” (S15 and S16 Tables), who have the smallest FST among all pairwise comparisons of Pagani groups (Fig 1b, S3 Table) and show similar patterns in our ADMIXTURE results (Fig 1c, S5 Fig). This suggests the Ari groups are more similar to each other, in terms of how their ancestry relates to the non-Pagani donors, than any other groups sampled within Ethiopia used in this study. Furthermore in analysis (B), in contrast to what you might expect under the RN hypothesis as originally formulated [5], the ARIc are not more closely related to groups currently classified as farmers [27] than the ARIb (Fig 3, S8 Fig, S11, S15 and S16 Tables).
Our 24 “simplified” simulations under the “RN” model (Fig 2b) illustrate scenarios where our CHROMOPAINTER analysis has power to tell apart the two groups under hypothetical Remnants scenarios. As we note below (see GLOBETROTTER results), the two Ari groups have similar sources of recent admixture, likely between a West Eurasian source and an African source as inferred by our analyses and other researchers [32], as well as an additional likely African source inferred by our analyses here. Given these similar recent admixture signals, we likely would only have power to distinguish between the two groups under analyses (B) and (C) if there is at least one sampled group whose ancestors split with one of the two Ari more recently than the ancestors of the two Ari groups split from each other. Such a hypothetical setting, which supports the RN model, seems plausible given FST(ARIc,ORO) = 0.015 and FST(ARIc,MKK) = 0.20 are both lower than FST(ARIc,ARIb) = 0.023 (S3 Table), e.g. the ARIc plausibly may have split more recently from the ORO and/or MKK than from the ARIb. Therefore, while it is impossible to evaluate all historical parameters that may lead to diversity patterns observed today, for these “simplified” simulations we fixed the split time between our simulated “ARIc” (i.e. Pop5 in Fig 2b) and “ORO” (Pop4) groups to 700 generations ago, which gave an FST similar to that observed in the real data (FST(Pop5,Pop4) = 0.011 − 0.014, S8 Table) while accounting for levels of inferred recent West Eurasian admixture in the two groups (see Methods). We then altered the split time between the simulated “ARIc” and “ARIb” (Pop5b) from {750, 800, 900, 1000, 1100, 1200, 1300, 1700} generations ago, choosing a strength of bottleneck in Pop5b for each split time that gave similar FST values between the two real Ari groups (FST(Pop5,Pop5b) = 0.019 − 0.027, S8 Table). We also tried three separate rates of migration from Pop5b into Pop5, the direction of migration suggested by ADMIXTURE results as interpreted in [2] and [11], such that ≈ {50%, 75%, 90%} of Pop5 was comprised of migrants from Pop5b over the period 200 to 300 generations ago.
For these “simplified” simulations, we performed an analysis mimicking CHROMOPAINTER analysis (B) in the real data, though note that we used only five surrogate groups to infer ancestry, which could decrease power. Using techniques described in the next section, in these “simplified” simulations we were able to distinguish Pop5 and Pop5b when the split time was ≥ 1300 generations, even when the proportion of admixture from Pop5b into Pop5 was 90%, and when the split time was ≥ 1100 generations when the admixture proportion was 50–75% (S19 Fig, S18 Table). For split times of 1000 generations or less, i.e. such that the split of Pop5 and Pop5b was at most 300 generations older than the split of Pop5 and Pop4, we could not always distinguish the ancestry of Pop5 and Pop5b under our analysis (B). We note that [11] suggest the split occurred > 31kya (i.e. > 1100 generations ago assuming 28 years per generation), which is older than these split times for which our model has no power. Taking these simulation results at face value, our model’s power to distinguish the two Ari groups requires that the split time between the two Ari groups be ≥ 400 generations older than the split time between the ARIc and another sampled group (e.g. ORO) so long as any one-way admixture from the ancestors of the Blacksmiths to those of the Cultivators was ≤ 75%.
In addition to having similar genetic profiles under analysis (B)-(C), a recent split between Blacksmiths and Cultivators followed by a bottleneck in the Blacksmiths (i.e. MA hypothesis) predicts that the genetic diversity of the ARIb might fall somewhere along the spectrum of genetic diversity in the ARIc, assuming drift is relatively low in the Cultivators following this split. In particular, after accounting for bottleneck effects in the ARIb, the differences in inferred ancestry between the ARIb and ARIc should not be substantially greater than differences in inferred ancestry among the ARIc. For example, under the MA hypothesis the following should be true for two Ari individuals X and Y:
Due to the bottleneck, on average FXY should be smaller if X, Y are both ARIb relative to if X, Y are both ARIc. In analyses (B) and (C), FXY where X is ARIb and Y is ARIc should be similar to FXY when X, Y are both ARIc.
Point (1) depends primarily on the magnitude of the bottleneck in ARIb relative to ARIc, while point (2) primarily depends on when the ARIb and ARIc split and any subsequent admixture between them. For each of analyses (A)-(C), in Fig 4 we show the distribution of FXY for all pairings of individuals X, Y such that (i) X, Y are both ARIb, (ii) X, Y are both ARIc, and (iii) X is ARIb and Y is ARIc. Our real data results show the trends expected under point (1) for all three analysis, and for point (2) under analyses (B)-(C). To assess how well point (2) fits the observed data, we calculated the proportion P(ARIc) of ARIc pairs X, Y with FXY greater than the mean FXY across all pairings where X is ARIb and Y is ARIc (Fig 4; see Methods). Under analysis (C), P(ARIc) ≈0.2 is higher than the maximum analogous proportions comparing any two other Pagani groups (S17 Table). Comparing to the results of our “full” simulations, the observed data proportions are similar to their analogues under the “MA” simulations but consistently larger than the “RN”, “RN+BN” and “RN+BN+80%” simulations (Fig 5).
We can also calculate P(ARIb), the proportion of ARIb pairs X, Y with FXY greater than the mean FXY across all pairings where X is ARIb and Y is ARIc (Fig 4). Note that this is < 0.025 under analyses (B) and (C). In general we expect P(ARIb) to be less than P(ARIc) due to the bottleneck in the ARIb. However, even in the presence of a bottleneck in our Remnants simulations, P(Pop5b) is often greater than P(Pop5), in both the “full” (Fig 5) and the “simplified” simulations (S19 Fig, S18 Table), recalling that simulated Pop5b and Pop5 are meant to reflect the ARIb and ARIc, respectively. This suggests that, in contrast to the MA model, under the RN model it is unclear whether the variation among ARIb in inferred genetic relatedness to outside groups should be less than that among ARIc. For example, for some “simplified” simulations where our model has no power to distinguish between the two simulated “Ari” groups, i.e. when migration from Pop5b into Pop5 is ≥ 75% and the split time between Pop5b and Pop5 is ≤ 300 generations older than that between Pop5 and another sampled group (Pop4), nonetheless give P(Pop5b) > P(Pop5). This in turn suggests that historical parameters behind these simulations are less consistent with the real data observation of P(ARIb) < P(ARIc).
We also applied GLOBETROTTER [33] separately to each Ari group for analyses (A)-(C), in order to infer recent admixture events in each group. In brief, within each Ari group GLOBETROTTER explores linkage disequilibrium patterns in order to identify and date any putative DNA admixture event(s) from (unknown) ancestral source groups that have occurred in the past ≈4500 years, using other sampled groups as surrogates for the admixing sources (see Methods). Under each of analyses (A)-(C), GLOBETROTTER found significant evidence (p-value < 0.01) for at least one admixture event in each of the ARIb and ARIc.
In each of analyses (B) and (C), we infer a simple admixture event at a single time between two sources in both Ari groups, with similar inferred dates, admixture proportions, and sources of ancestry (Table 1 and S20 Table, S23, S25 and S26 Figs) between the two groups. Any small discrepancies in inference between the two Ari groups are likely attributable to differences in sample size, with for example inferred values often as consistent between ARIb and ARIc than between all ARIc and a subset of 10 randomly-chosen ARIc (S21 Table). The inferred admixture event corroborates previous inferences of an admixture event ≈3K years ago involving a West Eurasian source [32, 2, 11] and suggests the same such signals in each Ari group. We refer to this admixing source henceforth as originating from “West Eurasia”, noting that our lack of a comprehensive set of world-wide samples, e.g. with no samples from the Near East, prevents interpretation of the precise source of this admixture. The fact the dates under analysis (B) are significantly more recent than those under analysis (C) likely reflects the different surrogates used and/or different inferred sources and proportions. In particular analysis (C) perhaps picks up signals of the original admixture between “West Eurasia” (from a source best represented by CEU out of our sampled groups) and a more “African”-like source (best represented by MKK), which matches results from previous analyses using similar surrogates [2, 11]. In contrast, the date in analysis (B) could reflect admixture between more geographically local groups at a more recent date, i.e. between an already admixed group (best represented by AFA in each Ari group) and another likely African group (best represented by ANU in each Ari group). As the inferred dates in (B) are relatively old and separated by only ≈30–40 generations from the analysis (C) results (Table 1), and/or as there may have been continuous admixture over this timeframe, GLOBETROTTER may not have the power to separate these events/dates reliably with these sample sizes. Indeed there is suggestive evidence of two or more distinct dates of admixture in both Ari groups under analysis (B) (S25 Fig, S22 Table), though the wide confidence intervals in our date estimates when assuming two dates reflects the difficulty in reliably characterizing this signal. If multiple dates or continuous admixture is indeed the case, our inferred dates under analysis (B) might be biased towards more recent intermixing.
GLOBETROTTER results under analysis (A) are more difficult to compare between the two Ari groups, as they do not use the same set of surrogates here as is the case in analyses (B) and (C). Nonetheless, signals in each group are similar and suggest a complex signal where both Ari groups are admixed with a third group, with this admixture dated to a similar time period as that inferred under analysis (B). For example, the ARIc show mixing around 400BCE-330CE between three distinct sources most similar to the ARIb, ORO and ANU, respectively (S20 Table). For the ARIb, GLOBETROTTER under analysis (A) infers mixing between three groups in some analysis (S21 Table) but only two groups in others (S20 Table). This is likely attributable to decreased power in the ARIb due to their smaller sample size relative to the ARIc, as well as the strong bottleneck in the ARIb, which can be thought of as further reducing the effective number of individuals relative to ARIc. To simplify our analysis (A) results, we also applied GLOBETROTTER to each Ari group using only four surrogate groups: ANU, ORO, TSI and the other Ari group (“A-sim” results in S20 Table). This analysis concluded three-way intermixing in both groups, with confidence intervals of inferred dates overlapping (ARIb: 402BCE-690CE; ARIc: 542BCE-270CE) and with at least one inferred source in each group best represented by the other Ari surrogate.
The complexity of the inferred admixture under GLOBETROTTER analysis (A) makes it difficult to interpret reliably [33]. For example, interpreting the three inferred source groups is challenging, as both Ari groups and many other Ethiopia groups (such as ORO) are thought to have substantial admixture from a West Eurasian source (S10 Table, [32]) and hence are subject to the same interpretation difficulties discussed above for analysis (B). Furthermore, as in analysis (B) there is suggestive evidence of multiple dates of admixture in each Ari group (S23 and S24 Figs, S22 Table), though again GLOBETROTTER does not conclude multiple dates of intermixing, perhaps due to the relatively small number of samples in this analysis.
Nonetheless, the GLOBETROTTER results under analysis (A) support three distinct sources intermixing (e.g. see S23 Fig), either at roughly the same time in the past or perhaps with some of the sources intermixing more recently than others. As there was no clear evidence of three separate groups intermixing under analysis (B) in either Ari group, the additional third source captured in analysis (A) is likely more related to the two Ari than any other sampled group. Determining the contribution from this group is difficult. For example, for the strongest inferred events (i.e. “First Event” in S20 Table), the total inferred contribution from the ARIb into the ARIc is ≈12–13% across analyses, while the total inferred contribution from the ARIc into the ARIb is much larger at ≈68–72%. However, these very different proportions are still consistent with the same group contributing DNA to each. In particular, GLOBETROTTER and our related linear modeling methods (see Methods; [25]) tend to down-weight heavily bottlenecked groups (like the ARIb) as surrogates for any putative admixture events that occurred further in the past than the bottleneck (e.g. see simulation results in S12 and S13 Figs). This is not unexpected or necessarily undesirable, as present-day descendants that are heavily bottlenecked from the original admixing source will look less genetically similar to that source. However, as a consequence, if a group equally related to each Ari group contributed DNA to each at the same proportion prior to a bottleneck in the Blacksmiths, GLOBETROTTER’s inferred ARIb contribution to the ARIc would likely be down-weighted relative to the inferred ARIc contribution to the ARIb. We demonstrate this phenomenon using simulations under a MA hypothesis (S27–S30 Figs and S23 and S24 Tables; see Methods).
Therefore we cannot determine whether this additional admixing source inferred under analysis (A) supports an MA model suggesting the same source contributed DNA to the recent shared common ancestor of the two Ari groups, or whether it supports an RN model where the ARIb and ARIc are anciently related and have each intermixed with one another since their initial split. Nonetheless, the inferred date of intermixing is recent (< 3kya) and thus consistent with the Blacksmiths and Cultivators being anciently or relatively recently related. Furthermore, we again note that whether assuming one or two distinct dates of admixture, the inference under each of analyses (A)-(C) is similar between the two Ari groups (Table 1, S20–S22 Tables) and thus consistent with them having recent common shared ancestry.
To further assess whether the Ari share similar genetic origins, we performed an analysis independent of CHROMOPAINTER analyses (A)-(C), based on separating segments inherited from African and “West Eurasian” ancestral source groups. To identify segments from different sources, within each haploid genome of each Ari individual, we fixed the YRI and CEU as surrogates for the two admixing source groups. We then used CHROMOPAINTER to identify all segments containing ≥ 100 contiguous SNPs that we could confidently assign to one of the two surrogates based on new simulations mimicking the presumed recent admixture history of these groups (S31 and S32 Figs; see Methods). Due to a lack of proper surrogate for an ancestral “Ari”-like source outside of the two sampled Ari groups, we did not attempt to characterize all three source groups identified in GLOBETROTTER analysis (A), but instead focused on segments of likely non-African versus African origin. We took each pair of Ari individuals and first extracted all segments within the haploids of each individual that were assigned to one of the two surrogates (i.e. YRI or CEU). Then, separately for each surrogate, we found the proportion of allele matches between haploids from different individuals at all SNPs that overlapped within segments assigned to that surrogate. In this manner, we inferred the genetic similarity between each pair of Ari individuals separately for segments inherited from each source. We used two different methods, called “E-M” and “NNLS” (see Methods), for matching segments to YRI and CEU; each method gave similar results.
Analogous to our comparison of inferred ancestry results for CHROMOPAINTER analyses (B) and (C), for both CEU and YRI segments, the distribution of similarity scores between ARIb and ARIc individuals falls on the distribution of similarity scores among ARIc individuals (Table 2, S33 Fig). Overall patterns match those expected if the Blacksmiths’ ancestors experienced a strong bottleneck effect after splitting from those of the Cultivators under the MA model, as explained above. In contrast, if the RN model were true, genetic differences between the ARIb and ARIc in at least the non-“West Eurasian” segments (i.e. for which YRI acts as a surrogate) are expected to be larger than differences among the ARIc.
While it is difficult with these data to ascertain precisely when the Blacksmiths and Cultivators split (which might be possible with sequencing data in these groups; see Discussion), we can infer whether the split occurred before or after the admixture involving “West Eurasia” if we assume (i) the bottleneck in the Blacksmiths occurred immediately after the two groups split and (ii) the “West Eurasian” and non-“West Eurasian” intermixing ancestral source groups are the same between the two Ari (consistent with results here; Table 2, S33 Fig). We illustrate this in Fig 6a. If the admixture is older than the isolation, then both the introgressed and non-introgressed segments in the Blacksmiths have been subjected to the same amount of drift effects from the bottleneck. In contrast, if the admixture event occurred more recently than the isolation and subsequent bottleneck in the Blacksmiths, then under assumption (i) the introgressed segments have been affected less by drift than the non-introgressed segments. Therefore we can compare the levels of genetic similarity among Blacksmiths within introgressed and non-introgressed segments to infer whether the split occurred before the admixture or vice versa. We assume here the introgressing source is the “West Eurasia” source, though a similar argument follows if the introgressing DNA comes from the non-“West Eurasian” source.
However, in addition to separate drift effects, the levels of genetic diversity within introgressed versus non-introgressed segments can also differ due to varying amounts of diversity in the two source groups at the time of admixture. Therefore, to calibrate differences in the relative amount of genetic homogeneity between the two sources, we infer the relative levels of similarity among ARIc within introgressed and non-introgressed segments. Incorporating assumption (ii), if the introgression is older than the split, then the ratio of genetic similarity in the ARIb among introgressed versus non-introgressed segments should be the same as the analogous ratio in the ARIc. In contrast, if the introgression occurred more recently than the split and bottleneck in the Blacksmiths, the ratio of genetic similarity among ARIb in introgressed versus non-introgressed segments should be less than the analogous ratio in the ARIc (see Fig 6a).
Using the same segments inferred using CEU and YRI as surrogates for each of the two admixing sources, Fig 6b and S34 Fig give the ratios of inferred similarity in introgressed segments versus non-introgressed segments for every pairing of Ari individuals within each of the ARIb and ARIc. Under each of the “E-M” and “NNLS” methods, there is no noticeable difference between the ratios of the two groups. These results are consistent with the split and subsequent bottleneck in the Blacksmiths occurring more recently than the “West Eurasia” introgression event, or at least not substantially before the introgression, and therefore sometime within the last 2.5–4.5K years or so. We note that this observation assumes we have enough power to detect different strengths of bottleneck effects if the split were substantially older than the introgression. Encouragingly, empirical quantiles for similarity scores do not overlap between the ARIb and ARIc under each method for YRI-matched segments and sometimes for CEU-matched segments (Table 2). Assuming the MA hypothesis is true, this demonstrates the approach has some power to separate DNA segments subjected to different strengths of bottleneck effect, plausibly arguing against the split at least being substantially older than the introgression event, even if we cannot be more precise using this technique.
Overall our analyses here suggest evidence for strong bottleneck effects in the Blacksmiths (S13 and S14 Tables), and that these effects appear to be driving differences between the two Ari groups observed today using FST, unsupervised ADMIXTURE, and our CHROMOPAINTER analysis (A). For example, FST(ARIc,ORO) is lower than FST(ARIc,ARIb) (Fig 1b, S3 Table), and TVDXY and FXY under analysis (A) suggest smaller genetic differences between the ARIc and other sampled groups including the ORO than that between ARIc and ARIb (Fig 3, S15 and S16 Tables). Nonetheless our analyses (B) and (C), designed to attenuate bottleneck effects in the Blacksmiths, show discernible differences between the inferred ancestry of ARIc and all other groups, including ORO, but no clear difference between the inferred ancestries of ARIc and ARIb (e.g. Fig 3, S15, S16 and S17 Tables). However, as we demonstrate via simulations, distinguishing between the MA and RN models is challenging if one assumes there was substantial one-way migration from the ancestors of Blacksmiths to those of the Cultivators under an RN model, as suggested from previous interpretations of unsupervised clustering algorithms [2, 11].
These difficulties notwithstanding, we believe the MA hypothesis is a more parsimonious explanation given the Blacksmiths’ currently marginalised status. I.e. such marginalisation can plausibly lead to a substantial bottleneck effect in the Blacksmiths, which in turn is consistent with all of our results. In contrast, harmonizing the RN model with the data analysed here requires an additional assumption beyond this bottleneck effect, namely (1) that we have not sampled a group whose ancestors split more recently from either Ari group than the Ari groups’ ancestors split from each other, or (2) that there were substantial levels of intermixing between the ancestors of ARIb and ARIc since the two groups initially were isolated from one another. Assumption (1) is perhaps less likely given our analyses included other Ethiopian groups described as agriculturalists [27] and groups that are more genetically similar to the ARIc than the ARIc are to the ARIb using the measures noted above. Assumption (2) is plausible under a RN model, given the two groups currently reside together. Indeed we detected likely very recent intermixing (perhaps occuring only a generation ago) between the two Ari groups in a “Blacksmiths” individual that we excluded from our analyses (S3 Fig), though this was the only case of such very recent intermixing observed in these data. Presuming assumption (1) is false, any older intermixing between the two Ari groups’ ancestors would have to be substantial enough to decrease our power to tell the two groups apart today. For example, our analysis (C) results suggest that the two Ari groups are more similar to one another when compared to outside groups than any other pairwise combination of Pagani groups (Fig 3, S11 Fig, S12, S15, S16 and S17 Tables), which is difficult to reconcile with the two Ari groups being anciently related without large amounts of subsequent intermixing.
If the RN model were true, simulations that replicate patterns in our observed data (Fig 5, S19 Fig, S18 Table) suggest our model should have power to distinguish the ancestries of the two Ari groups so long as one other sampled group, which we argue could be the ORO or MKK, split ≥ 400 generations more recently from the Cultivators than the two Ari groups split from each other, even if the Cultivators were comprised of 75% migrants from the Blacksmiths over the period 200 to 300 generations ago. We note again that one-way intermixing from Blacksmiths to Cultivators was proposed based on genetic evidence [2, 11] rather than anthropological findings, and that the overall inferred contribution of the ARIb to the ARIc’s ancestry profile is < 20% in all of our analysis (A) results. In contrast, our analysis (A) GLOBETROTTER results infer the ARIc contribution to the ARIb’s ancestry profile to be > 65%, which might argue for substantial asymmetric migration from the ancestors of the Cultivators to that of the Blacksmiths. However, we note that this need not be the case. In particular the ARIb has its lowest FST with the ARIc out of all other sampled groups (S3 Table), so it is not surprising that GLOBETROTTER infers the ARIb to share the majority of its ancestry with the ARIc relative to the other groups. Overall we argue that there is no evidence in these data that clearly support the RN hypothesis over the MA, with or without moderate levels of intermixing between the two groups, including the difficult-to-interpret GLOBETROTTER analysis (A) results. We note that currently the MA hypothesis is favored among anthropologists for explaining the existence of caste-like occupational groupings in southwest Ethiopia [1], and we show here that this hypothesis is consistent with available genetic evidence.
As further confirmation of the common recent genetic origins of the Ari, we also used the alternative approach of D-statistics (see [34]) to discern whether the ARIc and ARIb form a clade relative to a clade containing any pairing of sampled African groups with little to no inferred recent West Eurasian admixture (see S25 Table). Among six such pairings, we found no D-statistics with a corresponding ∣Z∣-statistic greater than 3, suggesting we could not reject an Ari clade and confirming the Ari groups appear more genetically related to one another than to these other African groups (S25 Table).
An artefact leading to our observations of substantial bottleneck effects in the ARIb could arise if at least some of the sampled ARIb individuals were more closely related (i.e. at a family level) to one another relative to the ARIc, perhaps due to sampling artefacts. However, the ARIb and ARIc from [2] each contained individuals with reported birthplaces spanning a similar number of different locations within the region, suggesting that it is unlikely that any such sampling artefacts are playing a major role. A similar artefact might occur if phase information was captured more accurately for the ARIb than the ARIc via the phasing program SHAPEIT [20]. I.e. the ARIbs’ inferred haplotypes may have fewer “switch errors”, which in turn could lead to them appearing relatively more genetically homogeneous. In fact, better phasing for the ARIb might be expected if they are less genetically diverse than the ARIc, consistent with a bottleneck in the Blacksmiths and the MA hypothesis. However, we note that the average sizes of contiguous DNA segments painted by a single donor haplotype as inferred by CHROMOPAINTER were very similar when forming the ARIc or the ARIb using the non-Ari groups as donors (S9 Table), suggesting higher levels of phasing errors or other genotyping inconsistencies in the ARIc relative to the ARIb are not playing a major role. Furthermore our IBD sharing analysis ignoring phase information gave a similar conclusion of greater homogeneity among ARIb relative to ARIc (S13 Table).
CHROMPAINTER analyses (B) and (C) suggest that the ARIb and ARIc are roughly equally related to all other sampled non-Ari groups. There is some ability to tell the two groups’ inferred ancestries apart under these analyses (e.g. S8 Fig), though we note that these differences are small relative to those between all other sampled groups (Fig 3, S15 and S16 Tables). Strong bottleneck effects in the Blacksmiths can result in their appearing genetically distinct from the Cultivators even under analyses (B)-(C), plausibly over a short time period depending on the strength of the bottleneck, which we try to account for by considering variation in inferred ancestry patterns among individuals’ chromosomes within each Ari group. Increasing the number of sampled individuals from each group (Ari or otherwise) could further increase the power to distinguish Ari groups under these approaches to shed further light on the MA versus RN hypotheses. Increasing the number of outside groups used to describe the Ari ancestry might increase power as well, though likely only if incorporating additional geographically near groups, given that other world-wide groups are not featured prominently in analysis (B). In particular our GLOBETROTTER results under analysis (A) suggest that in addition to admixture from “West Eurasia”, there is admixture in both Ari groups from a source best represented by the Ari out of all of our sampled groups. Further dense sampling of Ethiopia might enable a better genetic description of this group, helping to confirm whether it is the same admixing source for the ARIb and ARIc and whether there were multiple episodes of admixture from varying sources over different time periods. In addition, as GLOBETROTTER is more likely to pick up recent signals over older ones, increased sample sizes might enable detection of any potential older intermixing between the ARIb and ARIc under a hypothetical RN setting.
Using more dense genetic data, e.g. from sequencing, might also increase power in a similar manner. Acquiring sequenced individuals from each Ari group would have the additional benefit of allowing inference of the split time between the two groups using pairwise sequentially Markovian coalescent (i.e. PSMC and MSMC) techniques [35, 36, 37]. For example, a recent study applying these approaches to individuals sampled from Ethiopian groups included in this paper suggested one such group, the Gumuz (GUM in our study), split from each of four other Ethiopian groups (Amhara, Ethiopian Somali, Oromo, Wolayta) ≈20–40K years ago [38]. While that study did not include data from Blacksmiths or Cultivators, given that genetic differences are substantially larger between GUM and each of {AFA,ORO,SOM} relative to differences between ARIb and ARIc in our analysis (C) (S15 and S16 Tables), it is plausible that 40kya provides a very conservative upper bound for the split time of Blacksmiths and Cultivators. Our attempts to refine this upper bound do not use the rich information from sequencing but are consistent with the bottleneck in the Blacksmiths occurring more recently than the “West Eurasia” admixture event, i.e. within the last ≈4,500 years, although this analysis may be influenced somewhat by a lack of power as discussed above. Evidence for the origins of blacksmithing in Ethiopia remain incomplete, but iron and bronze objects were first discovered on sites from the pre-Aksumite period, suggesting the existence of such practices in the mid to first Millennium BC [39, 40]. Therefore our results are consistent with the start of genetic isolation between Blacksmiths and Cultivators corresponding roughly to a time period near the introduction of blacksmithing in the region.
Our findings serve as a cautionary tale for over-interpreting clustering, e.g. ADMIXTURE plots or results from other unsupervised learning techniques applied to genetic data. In particular the ADMIXTURE plots appear similar in each of the “MA” and “RN” simulation scenarios in this case (S6 Fig), though the two hypotheses reflect very different ancestral histories. Previous studies have shown that individuals from a single genetically isolated group can be grouped into a distinct homogeneous cluster by these algorithms, for example the Kalash in an application of STRUCTURE to world-wide populations [41]. We believe a similar effect is causing the Blacksmiths to all be assigned to a single cluster here, although in this case one that is shared by nearby populations. In general this suggests that if such a homogeneous cluster is observed, one should check whether the individuals in the cluster appear to be more genetically homogeneous than the other sampled individuals, particularly when clustering individuals from isolated or geographically localised groups. If so, further investigations such as those performed here are warranted.
Importantly, a comparison of approaches here (analogous to supervised ADMIXTURE; [42]) allows us to distinguish genetic structure attributable to bottleneck effects within a population from that attributable to shared ancestry with outside groups. In particular, after accounting for “self-copying” or high levels of genetic similarity within the ARIb (analysis (A)), we demonstrate that the ARIb and ARIc look genetically similar in terms of shared ancestry with other sampled groups (analyses (B)-(C)). A more parsimonious explanation for this observation favours the Marginalisation model over the Remnants hypothesis, and helps towards resolving a long-standing controversy on the origins of different Ari caste-like occupational groups [1]. Furthermore, this provides evidence that a societal practice, namely the marginalisation of artisan communities, can drive strong genetic differences (FST = 0.02 − 0.04) between groups without involving any outside introgression and possibly occurring within the last 4,500 years.
It is straight-forward to apply these models to samples from other geographic regions, and may be particularly helpful in similar cases where different groups might be subjected to strong isolation effects driving genetic differences due to societal divisions, such as in India [43]. Such careful analyses can help to resolve major questions about whether genetic diversity is primarily driven by ancient demography or by more recent factors such as admixture, social exclusion and drift.
Our dataset consisted of 237 individuals from 12 different populations from Ethiopia, Somalia and South Sudan (“Pagani”, [2]), provided by the authors, 850 individuals from 10 populations from the 1000 Genomes Project [44] (“1KGP”; www.1000genomes.org), taken from the file “ALL_1000G_phase1integrated_v3_impute_macGT1.tgz” from https://mathgen.stats.ox.ac.uk/impute/data_download_1000G_phase1_integrated.html, and 28 individuals from 1 population (MKK) from HapMap Phase3 [19]. These datasets had 659,857 SNPs in common. Our aim was to incorporate data from several world-wide groups in our analyses of the Pagani resource, while still maintaining a large number of densely genotyped SNPs to ensure increased power using our haplotype-based approach. As noted in the Discussion, we do not expect that including individuals from populations not closely related to Ethiopian groups would alter power to test our hypothesis (e.g. given the results of analysis (B)). As noted in [2], all Pagani samples were ascertained such that their self-reported ethnicity matched that reported for the donor’s parents, paternal grandfather and maternal grandmother.
We removed 33 individuals who had an identity-by-descent (IBD) score as inferred by PLINK v1.07 [28] (PI_HAT) ≥ 0.2 with any other individual. Based on this IBD analysis we removed 6 Ari Blacksmiths (the same ones removed in [2] for the same reason), 1 Ari Cultivator (including one of the two removed in [2]), 1 Sudanese (including one of the three removed in [2]), 4 British individuals (GBR), 9 Chinese individuals (CHS), 2 Masaii individuals (MKK) and 10 Luhya (LWK) individuals.
Clustering analysis using fineSTRUCTURE [21] (details below) removed a further 23 individuals whose inferred ancestry looked different from other members assigned to their cluster group (S1 and S2 Figs). In particular, in addition to one Blacksmith with clear Cultivator ancestry (S3 Fig), we removed 2 Ethiopian Somalis, 2 Amhara (including the single Amhara individual removed in [2]), 6 Wolayta, 1 Oromo, 1 Somali, 6 Gumuz (including the single Gumuz individual removed in [2]) and 4 Sudanese individuals (including those removed in [2]). Therefore along with the IBD analysis, 56 individuals were removed in total. As an example of our removal procedure based on this visual inspection, the 6 Wolayta and 6 Gumuz individuals we removed are highlighted with green rectangles in S1 Fig. Note that these 6 Wolayta individuals were clustered together using fineSTRUCTURE, and split quite early (i.e. near the top) of the inferred fineSTRUCTURE tree, suggesting they are not very closely related to the other Oromo and Wolayta individuals (i.e. the ones assigned to the final “ORO” group and hence labeled as “ORO” in S1 Fig). Visual inspection of the heatmap (S1 Fig) showed that these 6 Wolayta individuals were inferred to share a relatively large proportion of ancestry to a set of 6 individuals labeled as Gumuz, perhaps indicating recent admixture between Wolayta and Gumuz individuals. Thus any inferred shared ancestry with these 12 Wolayta/Gumuz individuals could reflect sharing with either the ancestors of the Gumuz and/or the ancestors of the Wolayta, making any such inference difficult to interpret. Therefore we removed these 6 Wolayta and 6 Gumuz individuals from subsequent analyses. Similar decisions were made for the other exclusions based on these fineSTRUCTURE and CHROMOPAINTER results (e.g. S1 and S2 Figs).
Here we explain our exclusions of 7 labeled “Blacksmith” individuals, and how these exclusions relate to those in the Pagani paper [2]. We started with 18 individuals labeled as “Blacksmith” in the dataset provided to us by the authors of [2]. We retained one Blacksmith individual that appeared from our fineSTRUCTURE analysis to be misclassified as an “Ari” and is instead assigned to our “AFA” group; we note that this individual was removed from [2] for a similar reason and was not included among the 17 Blacksmiths reported in that paper. Therefore ignoring this misclassified individual, the 17 Blacksmiths labeled as “ARIb” in our S1–S3 Figs are the same ones reported in [2]. We then removed 6 “ARIb” based on IBD sharing; these are the same 6 Blacksmiths excluded by [2] for the same reason. Finally, in addition we removed one other “ARIb” that appeared to have a high proportion of Cultivator ancestry (see S3 Fig). Thus in total we used 10 “ARIb” individuals in our final analysis, which are the only ones analysed throughout the remainder of this paper, compared to 11 Blacksmiths in the final analysis of [2].
The final 17 clustered groupings, comprising 1059 individuals, are depicted on Fig 1a, with the sample sizes and description of each population label given in S1 Table.
All samples were phased jointly using SHAPEIT [20] incorporating the build 37 genetic map combined across populations available at https://mathgen.stats.ox.ac.uk/impute/data_download_1000G_phase1_integrated.html, using an effective population size (“—effective-size”) of 15000 and otherwise default parameters. Phasing was initially performed across 1176 individuals and 659,881 SNPs. Of these, 24 SNPs monomorphic across individuals were removed. The ASW (61 individuals) from the 1000 Genomes Project dataset were excluded from further analysis because they are known to be recently admixed with Africans and Europeans [45], leaving 1115 individuals and 659,857 SNPs prior to quality control measures mentioned described in the previous section that removed additional individuals.
We ran ADMIXTURE [10] using the 1059 sampled individuals kept after sample exclusions (see below), using several different numbers of clusters K. In this analysis, SNPs were thinned such that no two SNPs within 250kb had squared correlation coefficient (i.e. r2) greater than 0.1. This left 95,648 SNPs for ADMIXTURE analysis. In order to better visualise the Ari groups, ADMIXTURE results for K = 8 are shown for at most 50 individuals for each of the 17 groups in Fig 1c. ADMIXTURE results for K = 7 − 11 for all 1059 individuals are shown in S5 Fig.
We ran CHROMOPAINTER to infer “painting profiles” of each individual for the fineSTRUCTURE analysis and each of analyses (A)-(C). In each case, we initially estimated the mutation/emission (“-M”) and switch rate (“-n”) parameters using 10 steps of Expectation-Maximisation (E-M) algorithm (i.e. “-i 10 -in -iM”), starting with default values and running on a subset of individuals for a subset of chromosomes. We then averaged inferred values of each parameter across these chromosomes, weighting the average by number of SNPs, and then across individuals. We then fixed these values (i.e. using “-M” and “-n”) and ran on all chromosomes and all individuals. We otherwise used all default values, except that for the fineSTRUCTURE analysis we set the size of regions (“-k”) in CHROMOPAINTER to 50 in order to infer the “c” parameter in fineSTRUCTURE.
For the initial analysis of all 1115 individuals for use in fineSTRUCTURE, to estimate the emission and switch rates we used at most 20 individuals from each of the 23 labeled populations (or all individuals for populations with fewer than 20) and chromosomes {4, 10, 15, 22}, giving values of 0.00122 and 419.9 for the emission and switch rates, respectively. For the remaining analyses using the 17 fineSTRUCTURE-inferred groups, we used all individuals and chromosomes {1, 4, 15, 22} to estimate the emission and switch rates across all individuals. Under analysis (A) this gave values of 0.00064 and 390.1, under analysis (B) values of 0.0069 and 403.5, and under analysis (C) values of 0.00119 and 457.2 for the emission and switch rates respectively.
We note that individuals are not allowed to copy from themselves, so that e.g. under analysis (A), each of the 10 Ari Blacksmith individuals is allowed to copy from the other 9 Ari Blacksmith individuals and all individuals from each of the other 16 groups, including all 23 Ari Cultivator individuals. Similarly, under analysis (A) each of the 23 Ari Cultivator individuals is allowed to copy from only 22 Ari Cultivator individuals and all individuals from each of the other 16 groups, including all 10 Ari Blacksmith individuals. This slight asymmetry of donor panels potentially makes comparing the Ari Blacksmiths’ and Ari Cultivators’ copying vectors problematic under analysis (A), though we expect it to have only a small effect. In particular, we have found in practice that removing a single donor individual out of a group of ≈10 donor individuals generally results in a slight increase in copying from the remaining donor individuals, i.e. so that the overall copying from the entire group is not much changed.
We used fineSTRUCTURE [21] to cluster individuals into genetically homogeneous groups. To do so, we first used CHROMOPAINTER as described above to summarize each of the N = 1115 individuals’ ancestries as the total number of haplotype segments they copied from each of the other N − 1 individuals, so that we did not use any group label information when clustering. We set the starting value as 1 cluster and then ran fineSTRUCTURE for 1,000,000 “burn-in” iterations of MCMC, followed by another 1,000,000 iterations where we sampled inferred clusterings every 10,000 iterations, otherwise using default values. This inferred C = 154 final clusters. We next used fineSTRUCTURE to perform 100,000 additional hill-climbing steps to improve the posterior probability and then merge clusters in a greedy step-wise fashion. In particular, starting from the hill-climbing solution, at each step of the tree-building procedure fineSTRUCTURE considers the merging of all ( C 2 ) pairwise combinations of clusters, selects the pairwise merging that minimises the decrease in posterior probability over all such combinations, and continues this process until only C = 2 clusters remain, building a “tree” of relatedness.
Based on the fineSTRUCTURE tree, we classified the 878 individuals from MKK and 1KGP into ten groups. These ten groups differed from the 11 original population labels in two ways: (i) the two groups from China (CHB,CHS) were merged into a single group, and (ii) 23 individuals from Britain (GBR) were separated into their own group (perhaps representing substructure within Britain) and the remaining GBR individuals were merged with the Utah (CEU) samples.
As the MKK and 1KGP individuals comprised a large proportion of the overall sample set yet were not of direct interest in this analysis, we performed a second fineSTRUCTURE run that attempted to further refine clustering in only the Pagani samples. To do so, we treated our ten non-Pagani groups as “super individuals” (“-F”) in this second fineSTRUCTURE run. This means that each of the ten non-Pagani groups (as well as each Pagani individual) was represented as only a single “individual” containing the average number of haplotype segments they copied from each Pagani individual and each of the ten non-Pagani groups. We clustered this new set containing 237 + 10 = 247 “individuals” using fineSTRUCTURE, as before setting the starting value as 1 cluster and running for 1,000,000 “burn-in” iterations, followed by another 1,000,000 iterations where we sampled inferred clusterings every 10,000 iterations and otherwise using default values. This analysis inferred C = 87 clusters (including the ten non-Pagani groups). We considered two independent runs of fineSTRUCTURE using “super individuals” and the final clustering results were very consistent across the two (S4 Fig).
We next performed the re-classification procedure first described in [25]. Briefly for each individual i and each MCMC sample m, this procedure identifies the individuals clustered with i in sample m and calculates the proportion of these individuals contained in each of the c ∈ [1, …, C] final fineSTRUCTURE clusters (e.g. initially C = 87 here, with the 10 1KGP+MKK clusters remaining fixed for this procedure). For each cluster c ∈ [1, …, C] we then average these proportions across all MCMC samples, and then (potentially) re-classify individual i to the c containing the maximum such average proportion across all clusters C. Taking these new re-classifications of all individuals as the new “final fineSTRUCTURE cluster”, we repeat this procedure for 50 iterations. This gave our final classification of C = 87 clusters, though we note that cluster assignments were very similar to those prior to this re-classification procedure. As before we then used fineSTRUCTURE to merge clusters in a greedy step-wise fashion and build the fineSTRUCTURE tree.
These final 87 clusters and corresponding tree are shown in S1 Fig. Labels on the axes refer to the code (i.e. “Pop ID” in S1 Table) we assigned each group based on the population labels among individuals in the given cluster. Using this tree, we removed 23 individuals and grouped the remaining 1059 individuals into 17 genetically homogeneous groups for all subsequent analyses, with these 17 groups detailed in S1 Table and denoted by distinct colors on the axes of S1 Fig. Specifically, we first moved down the tree until reaching the level immediately prior to the Blacksmiths splitting into two distinct groups. The clusters resulting from this level of the fineSTRUCTURE tree are shown in S2 Fig. We then removed 23 individuals whose inferred ancestry visually looked different from the other members assigned to their group; these individuals and all other removed individuals are highlighted with translucent vertical grey bars in S1 Fig and with grey vertical dashed lines in S2 Fig. Including inds removed for having high IBD (see above), from left to right in S2 Fig we removed the Ari individual admixed between Cultivators and Blacksmiths (also shown in S3 Fig), 6 further high IBD Ari Blacksmiths, 1 Ari Cultivator (high IBD), 2 Ethiopian Somali individuals (assigned to the “AFA” group), 2 Amhara individuals (assigned to the “AFA” group), 5 Wolayta (assigned to the “ORO” group), 1 Oromo (assigned to the “ORO” group), another Wolayta (assigned to the “ORO” group), 1 Somali individual (assigned to the “SOM” group), 6 Gumuz (assigned to the “GUM” group) and 5 Sudanese (assigned to the “ANU” group). We removed 31 individuals in total from the Pagani dataset. While not all of our ten non-Pagani “super-individuals” were split into distinct groups at the fineSTRUCTURE tree level depicted in S2 Fig, we nonetheless kept all ten separated for subsequent analyses, giving 17 total groups.
In our additional CHROMOPAINTER analysis that compared the sizes of haplotype segments across groups to assess the relative genetic diversity within each group, we painted each of the 17 world-wide groups using only individuals from their own group as donors. We infered the switch rate separately for each group using 50 steps of E-M algorithm (i.e. “-i 10 -in”) and using the default mutation/emission rate (which was 0.00771). As the expected lengths of segments copied intact from a single donor in the painting can be affected by the number of donor individuals, we used 10 randomly sampled individuals from each group, matching the sample size of our smallest group (“ARIb”). We used all 659,857 SNPs, allowing individuals to copy only from the 9 other individuals of their own group. Our median inferred values across individuals for average haplotype segment size (in cM) and switch rate, plus the 95% empirical quantiles across the 10 individuals, are provided for each group in S14 Table. For each individual, we calculate average segment size by dividing the total proportion of genome-wide DNA copied from all donors by the total expected number of haplotype segments copied from all donors. We note that the often “noisy” process of painting in CHROMOPAINTER [33] suggests exact sizes of haplotype segments should be interpreted with caution and not e.g. related directly to split times as in other approaches [46], though comparing relative sizes across groups is still meaningful.
Our inferred “painting profiles” from CHROMOPAINTER suffer some limitations. For example a priori groups with more individuals will be copied more often when running CHROMOPAINTER, potentially leading to a biased interpretation of results. To cope with this, we use additional linear modeling described in this section to “clean” the raw CHROMOPAINTER inference as in [33, 25].
Following notation in [33], let f j ≡ { f 1 j , . . . , f K j } be the observed “painting profile” inferred by CHROMOPAINTER for recipient group j, with ∑ k = 1 K f k j = 1 . 0 and f k j the proportion of genome-wide DNA that group j paints (or copies) from donor group k ∈ [1, …, K] using CHROMOPAINTER. We use CHROMOPAINTER to calculate analogous painting profiles for each group k ≠ j ∈ [1, …, K] as described above. To measure the relative amount of drift (or “self-copying”) in group j, we introduce a K-vector fj* with fjj*=fjj and all other entries 0. We “clean” the painting of group j using the following linear model:
f j = [ ∑ k ≠ j K β k j f k ] + β SELF j f j * + ϵ . (1)
Here ϵ is a vector of errors, and we seek the estimates β^1j,…,β^Kj,β^SELFj to replace β1j,…,βKj,βSELFj, respectively, that minimize ϵ using least-squares. (Note that fjj*=0 in analyses (B) and (C) for some groups, such as the Ari, so that βSELFj=0 in these cases.) We use the non-negative-least-squares “nnls” package in R to estimate the βkjs under the constraints that all β^kj≥0, β^SELFj≥0, and (β^SELFj+∑k≠jKβ^kj)=1.0. To avoid over-fitting, in practice any β^ij≤0.001 is set to 0, and we then re-scale so that (β^SELFj+∑k≠jKβ^kj)=1.0. We refer to {β^1j,…,β^kj} as our inferred “proportions of ancestry” for group j.
In Fig 3 (top row) and S11 Fig, our inferred βs are shown for each of analyses (A)-(C). To measure uncertainty in the β ^s, we take an approach analogous to [34] and calculate standard errors using a weighted Block Jackknife [47] approach that removes each chromosome one-at-a-time (from each group k ∈ [1, …, K]) and re-calculates the β ^s, weighting each jackknife sample by that chromosome’s number of SNPs. This contrasts from the approach used to measure uncertainty in [25], who instead used bootstrap re-samples of individuals’ chromosomes. In contrast to that approach, the jackknife technique we use here accounts for independent drift effects across chromosomes, as we are particularly interested here in mitigating any differences among groups attributable to drift. We report our β ^s, plus and minus two standard errors, for all 17 groups in S10–S12 Tables for each of analyses (A)-(C).
To compare differences in copying vectors, we use total variation distance (TVD) as in [25]. In particular let f k X be the genome-wide proportion of DNA that recipient X copies from donor group k ∈ [1, …, K] as inferred by CHROMOPAINTER. Then to compare the copying vectors of two recipients X and Y, we calculate TVDXY, with:
T V D X Y = 0 . 5 ∑ k = 1 K | f k X - f k Y | . (2)
Note that f k X might be the copying vector of a single recipient individual, or the average copying vector across all individuals from a particular recipient group (e.g. the average copying vector across all ARIb individuals). For example, in S15 Table and Fig 3 (bottom row) we report TVDXY for fkX,fkY the average copying vector across all individuals from the given group. In contrast, for S15 and S16 Figs we report TVDXY for fkX,fkY the copying vectors of single individuals.
Assessing formally whether the observed value of TVDXY is significantly different from 0, i.e. assessing the evidence (or p-value) for rejecting the null hypothesis that X and Y are ancestrally related in the same way to other groups, is not straight-forward when trying to negate founder and/or population-specific drift effects. For example, if a small number of individuals leave a relatively large population and form a “new” group, the mean inferred such ancestries between the new and old groups may be significantly different (i.e. TVDXY >> 0) due to the decreased variance in the smaller group. But this significant difference is not important if one is only interested in genetic differences attributable to ancient relatedness. One way to account for this is to consider the differences in ancestry between the two groups scaled by the differences within each group among independent genetic regions (i.e. regions separated by historical recombination events). Each such region will relate ancestrally to other groups differently due to independent drift effects, i.e. because drift affects any two unlinked segments of the genome independently, and hence can provide a means to measure variation attributable to drift effects. Here we use each chromosome as separate regions.
Let fikX be the proportion of DNA that recipient X copies from donor group k ∈ [1, …, K] across chromosome i as inferred by CHROMOPAINTER. Then we calculate T V D ˜ X as:
T V D ˜ X = 0 . 5 ∑ i = 1 22 L i L [ ∑ k = 1 K | f i k X - f k X | ] , (3)
with Li the number of SNPs in chromosome i ∈ [1, …, 22] and ∑ i = 1 22 L i = L. We define a new statistic, FXY, to measure the difference between the inferred ancestries of two groups X and Y relative to the difference between chromosomes within each of X and Y:
F X Y = T V D X Y / [ 0 . 5 ( T V D ˜ X + T V D ˜ Y ) ] . (4)
(This formulation has parallels to FST statistics [48].)
Again fikX can refer to the copying vector for chromosome i in a single individual (i.e. so that X refers to an individual) or alternatively can be the average copying vector across chromosome i for all individuals from a particular recipient group (i.e. so that X refers to a group). For example, in S17 and S18 Figs we report FXY for fikX,fikY the copying vectors of single individuals. In contrast, in S16 Table, we report FXY with f i kX , fi kY the average copying vector across chromosome i for all individuals from the given group.
However, this latter version of FXY also has undesirable properties, again attributable to differential drift effects (and/or different sample sizes) between groups X and Y. In particular assuming ancestry component k for each chromosome of an individual from population X is a random draw from some distribution with mean f k X, taking the mean across nX individuals from X to calculate f i kX should move f i kX towards the mean f kX. Now assume another population Y also draws its ancestry components k from some distribution with mean f kX, but the difference is that Y is highly drifted relative to X. In this case, for any given chromosome i the nY individuals from Y are much more similar (relative to individuals from X) in their ancestry component k, so that taking the mean across the nY individuals to calculate fi kY will be some value not necessarily as close to f kX. Therefore in this scenario TVD˜Y>TVD˜X, as we observed in our “MA” simulations. Sample size differences, i.e. having nY << nX, can also lead to the same effect.
One potential approach to assess evidence of whether the difference in inferred ancestry between X and Y is “significantly different”, analogous to those used by other groups [34], is to divide TVDXY or FXY by its standard error calculated using the weighted Block Jackknife [47], weighting by removing e.g. whole chromosomes. However, unlike in [34] it is unclear what the distribution of such a statistic should be, given that TVDXY ≥ 0 and FXY ≥ 0 and so cannot follow e.g. a normal distribution with mean 0. Assessing significance of ancestry differences under these approaches in a manner to mitigate founder and group-specific drift effects remains an area of important future work.
Despite these difficulties, when comparing two groups X and Y we attempt to assess the evidence of a bottleneck (i.e. founder or drift effect) in one of the groups but otherwise very similar ancestry. To do so, we provide an empirical measure of the “significance” of differences in inferred ancestry among sampled individuals within one group (i.e. the non-bottlenecked one) relative to that among individuals between X and Y. In particular, letting X represent the non-bottlenecked group and 1Z an indicator that Z is true, we calculate:
P(X)=1(nX2)∑j=1nX∑k=1nX[ 1j≠k1[ Fjk≥F^XY ] ], (5)
where
F ^ X Y ≡ 1 n X n Y ∑ j = 1 n X ∑ k = 1 n Y F j k . (6)
Values of (Eq 5) range from 0–1 and are provided in Figs 4 and 5 of the main text, and S17, S18 and S24 Tables and S19 and S30 Figs of the SOM. Under a bottleneck in group Y but otherwise identical ancestry patterns between X and Y, or in general if the ancestries of X and Y are not significantly different, we expect (Eq 5) to be relatively large (i.e. closer to 0.5). In contrast, if the ancestries of X and Y are significantly different from each other, or possibly if the ancestries are identical except for a strong bottleneck in X and not Y, we expect (Eq 5) to be at or near 0. We demonstrate the power of this approach to assess evidence of different ancestry between X and Y in our simplified simulation results under a Remnants (“RN”) setting (S19 Fig, S18 Table), results of which are described in “Results”.
We developed a new algorithm to cluster individuals into genetically homogeneous groups according to their CHROMOPAINTER inferred paintings. We did this clustering separately for analyses (A)-(C).
Similar to the algorithm implemented in fineSTRUCTURE [21], but without the restriction that each individual must be painted in CHROMOPAINTER using all other sampled individuals as donors, we cluster based on the inferred genome-wide length of DNA copied from each donor. (We note clustering in fineSTRUCTURE [21] is done based on the inferred counts of DNA segments rather than lengths of DNA copied, though in practice clustering based on the inferred lengths often provides similar results.) In particular, let l j ≡ { l 1 j , . . . , l K j } be the observed CHROMOPAINTER painting for recipient individual j, with l k j the centimorgan (cM) length of genome-wide DNA that individual j paints (or copies) from donor group k ∈ [1, …, K]. Here ∑ k = 1 K l k j equal to the total genome length of DNA in cM (or double that for diploid individuals), and note that fkj ≡ l k j / [ ∑ i = 1 K l i j ].
We want to cluster our N individuals into C sub-groups, where C is fixed (also in contrast to fineSTRUCTURE, which infers C in its current implementation). Let ψj ∈ [1, …, C] be the cluster assignment for individual j, with each cluster equally likely a priori. Let γ ≡ {γ1, …, γC}, with γ c ≡ { γ 1 c , . . . , γ K c } and γ k c the probability of being painted by donor group k if assigned to cluster c. Note ∑ k = 1 K γ kc = 1. We assume:
( l j ∣ γ c , ψ j = c ) ∼ Multinomial ( γ 1 c , . . . , γ K c ) for j = 1 , . . . , N and c = 1 , . . . , C , ( γ c ∣ δ ) ∼ Dirichlet ( δ , . . . , δ ) for c = 1 , . . . , C , Pr ( ψ j = c ) = 1 / C for c = 1 , . . . , C ,
with δ controlling the probability distribution of γc. Intuitively, a larger value of δ makes the parameters of the C clusters more similar to one another, hence discouraging sub-group formation.
We wish to sample the cluster assignments ψj for j ∈ [1, …, N] based on their posterior probabilities conditional on l ≡ {l1, …, lN}. We do so using the following Markov Chain Monte Carlo (MCMC) technique. We start with initial cluster assignments ψ(0) ≡ {ψ1(0), …, ψN(0)} by assigning each of the N individuals randomly to one of the C clusters.
Then for m = 1, …, M:
Sample γ(m) using a Gibbs step, with
( γ c ( m ) ∣ l , ψ ( m - 1 ) , δ ) ∼ Dirichlet ( δ + ∑ j = 1 N l 1 j 1 [ ψ j ( m - 1 ) = c ] , . . . , δ + ∑ j = 1 N l K j 1 [ ψ j ( m - 1 ) = c ] ) ,
for c = 1, …, C, with 1S an indicator for whether S is true. Sample ψ(m) using a Gibbs step, with
Pr ( ψ j ( m ) = c ∣ l j , γ ( m ) ) = Pr ( l j ∣ γ c ( m ) , ψ j = c ) ∑ i = 1 C Pr ( l j ∣ γ i ( m ) , ψ j = i ) ,
for j = 1, …, N. If any of the c = 1, …, C clusters contain 0 individuals, we randomly assign a single distinct individual to each empty cluster. This enables the cluster painting probabilities in step (A) to move away from the prior distribution defined by δ.
For numerical stability, any γ kc values < 1e−7 or > (1 − 1e−7) in step (A) were set to 1e−7 and (1 − 1e−7), respectively, with the values in γc subsequently re-scaled to sum to 1.
For large M, this algorithm is guaranteed to converge to the true posterior distribution of the ψj’s (e.g. [49]). In practice, for all results presented here we use M = 2,000,000, sampling every 10,000th iteration after an initial “burn-in” of 1,000,000 iterations. Also, for all analyses we combined results across ten independent runs of the above procedure.
When clustering the Ari individuals only, we set C = 2 and present our cluster results for analyses (A)-(C) in S19 Table, including analogous results for the simulations, separately for two choices of the prior value δ = {50, 100}. When clustering all 206 Pagani individuals (after sample exclusions), we used δ = 100 and considered several fixed number of clusters C = {2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 50}, providing results for a subset of these values in S20–S22 Figs.
We used GLOBETROTTER to identify, describe and date any putative admixture events occurring within the last ≈4,500 years in the Ari groups, using painting results from analyses (A)-(C) and closely following the application of GLOBETROTTER described in [33]. In short, CHROMOPAINTER identifies the segments of DNA within each Ari individual’s genome that are most closely related ancestrally to each of the S “surrogate” groups. I.e. S = 16 for analysis (A) since your own group is excluded as a surrogate group, while S = 15 for analysis (B) and S = 10 for analysis (C). GLOBETROTTER then measures the decay of linkage disequilibrium (LD) versus genetic distance among the segments copied from a given pair of surrogate groups. Assuming a single “pulse” of admixture between two or more distinct admixing source groups, theoretical considerations predict that this decay will be exponentially distributed with rate equal to the time (in generations ago) that this admixture occurred [13]. GLOBETROTTER jointly fits an exponential distribution to the decay curves for all ( S 2 ) + S pairwise combinations of the S surrogate groups (including where both surrogates in the pair are the same group) and determines the single best fitting rate, hence dating the admixture event. Instead of requiring specific genetic surrogates to represent each admixing source group involved in the admixture, which is necessary for other dating approaches such as ROLLOFF [34], GLOBETROTTER aims to infer each source group as a linear combination of the DNA of sampled groups (i.e. as a linear combination of the S groups). In the case of X distinct pulses of admixture occuring at different times, the decay of LD among segments is expected to be a mixture of X independent exponential distributions. GLOBETROTTER tests for evidence of two distinct admixture events by fitting two exponential distributions jointly to all curves and determining the best fitting rates for each. If two exponential distributions fit the data “significantly” better than a single event (assessed via simulations described in [33]), GLOBETROTTER infers the source groups involved in each of the two events.
We applied GLOBETROTTER to infer any admixture event(s) separately in the ARIb and ARIc. When doing so, we used 10 painting samples from each haploid of each Ari individual. Note that under analysis (A), following [33] we repainted individuals within each Ari group excluding members of their own group as donors to get these painting samples, though we note that (also following [33]) the original painting used elsewhere in this paper (e.g. for Fig 3) was also used here to determine mixing coefficients in GLOBETROTTER analysis (A). Define a “chunk” within one of these painting samples to be a segment of contiguous DNA painted intact from a single donor haploid from one of D donor groups. (For each analysis here, the donor groups were the same as the surrogate groups, so that D = S.) For every pairing of painting samples between and within each Ari individual’s two haploids, we pairwise compared each chunk on one sample to each chunk on the other sample, tabulating the donor group(s) represented at each chunk in the pair, the product of the two chunks’ sizes in cM (with any sizes > 1cM set to 1) and the genetic distance between the two chunks’ midpoints. Then for every pair of donor groups D1 and D2, for each 0.1cM bin g ∈ [0.1, …, 50cM] we sum the products across all chunk pairs separated by g for which one chunk is painted by D1 and one by D2. We repeat this for all pairings D1, D2 of the D donor groups, and re-scale these counts by inferred ancestry coefficients for each of the S surrogate groups (i.e. the β ^ k j values from “inferring mixing coefficients using CHROMOPAINTER output”) as described in [33], subsequently removing any surrogate groups with β ^ k j < 0 . 001. We refer to the plot of these re-scaled counts versus genetic distance g as the “coancestry curves” for surrogate groups S1, S2, of which there are ( S 2 ) + S in total.
We plot these coancestry curves for S1, S2 = {ANU, ARIb, ARIc, ORO, SOM, TSI} for analysis (A) (S23 and S24 Figs), S1, S2 = {ANU, AFA} for analysis (B) (S25 Fig) and S1, S2 = {CEU, MKK} for analysis (C) (S26 Fig). On these plots, we also include our best-fitting exponential distribution assuming a single pulse of admixture (green lines), the rate of which corresponds to our inferred date of admixture (in generations from present). We furthermore include our best-fitting exponential distribution assuming two distinct pulses of admixture (red lines), i.e. the sum of two exponential distributions whose rates correspond to our inferred dates for each event. For these figures and all reported results, we used five iterations of GLOBETROTTER’s alternating source composition and admixture date inference, followed by 100 bootstrap re-samples of individuals’ chromosomes to infer confidence intervals (CIs) around our date estimates. Also, for all results presented here we standardized each coancestry curve by a “NULL” individual designed to eliminate any spurious linkage disequilibrium patterns not attributable to that expected under a genuine admixture event, which simulations suggest is particularly important when the target population has undergone a strong bottleneck (see [33] for details). This gave a p-value < 0.01 testing for evidence of at least one admixture event in each of the ARIb and ARIc under each of analyses (A)-(C), defined as in [33] as the proportion of bootstrap re-samples with an inferred single date = 1 or ≥ 400. We note that under each analysis, each Ari group was inferred to have only a single pulse (i.e. date) of admixture, though we provide results assuming multiple dates of admixture in S22 Table for comparison. For this multiple-date inference, we excluded results if CIs of the two inferred dates overlapped or if the recent data had a point estimate of 1, resulting in the exclusion of analysis (C) results for both Ari groups. In contrast, analyses (A)-(B) suggest visual evidence of multiple admixture dates (i.e. compare red to green lines in S23–S25 Figs), which GLOBETROTTER may not have enough power to detect using the given sample sizes. When there is evidence of only two admixing source groups, we provide the best genetically matching sampled group to each source and the mixing coefficients (for surrogates with coefficients > 10%) that provide a more detailed description of each source as a mixture of sampled groups (Table 1 and S20–S22 Tables). When there is evidence of more than two admixing source groups at a single date, we provide both the best genetically matching sampled groups to each source, the mixing coefficients, and the sampled groups that reflect the greatest difference between inferred ancestries for the two depicted sources (S20–S22 Tables; see [33] for details).
To infer which segments in the Ari were inherited from the “African” and “non-African” (i.e. “West Eurasian”) admixing sources as inferred by GLOBETROTTER, we used CHROMOPAINTER to paint each Ari haploid genome using only the YRI and CEU as donors (using the “-b” switch, which produces the file with suffix .copyprobsperlocus.out). While we could have used TSI or IBS instead of CEU to represent the “non-African” source (which may in fact originate from e.g. the Levant [2]), we were concerned that each of TSI and IBS may have recent admixture from Africa [33] which would diminish accuracy in separating the “African” and “non-African” segments. In contrast, CEU should not have any recent admixture from Africa. Similarly YRI should not have received any recent admixture from outside Sub-Saharan Africa.
We used two different approaches to assign segments to CEU/YRI, which we term “E-M” and “NNLS”.
|
10.1371/journal.pcbi.1004891 | The Role of Genome Accessibility in Transcription Factor Binding in Bacteria | ChIP-seq enables genome-scale identification of regulatory regions that govern gene expression. However, the biological insights generated from ChIP-seq analysis have been limited to predictions of binding sites and cooperative interactions. Furthermore, ChIP-seq data often poorly correlate with in vitro measurements or predicted motifs, highlighting that binding affinity alone is insufficient to explain transcription factor (TF)-binding in vivo. One possibility is that binding sites are not equally accessible across the genome. A more comprehensive biophysical representation of TF-binding is required to improve our ability to understand, predict, and alter gene expression. Here, we show that genome accessibility is a key parameter that impacts TF-binding in bacteria. We developed a thermodynamic model that parameterizes ChIP-seq coverage in terms of genome accessibility and binding affinity. The role of genome accessibility is validated using a large-scale ChIP-seq dataset of the M. tuberculosis regulatory network. We find that accounting for genome accessibility led to a model that explains 63% of the ChIP-seq profile variance, while a model based in motif score alone explains only 35% of the variance. Moreover, our framework enables de novo ChIP-seq peak prediction and is useful for inferring TF-binding peaks in new experimental conditions by reducing the need for additional experiments. We observe that the genome is more accessible in intergenic regions, and that increased accessibility is positively correlated with gene expression and anti-correlated with distance to the origin of replication. Our biophysically motivated model provides a more comprehensive description of TF-binding in vivo from first principles towards a better representation of gene regulation in silico, with promising applications in systems biology.
| A quantitative description of transcription factor (TF) binding in vivo is critical for our understanding of gene regulation. Chromatin Immunoprecipitation followed by sequencing (ChIP-seq) provides a genome-scale map of TF-binding. However, a quantitative characterization of the impact of genome accessibility on TF-binding in bacteria remains elusive. In order to help recruit or block gene expression, TFs must have physical access to regulatory regions. This paper presents a thermodynamics model that describes TF-binding in terms of genome accessibility and binding site affinity. We apply this model in a ChIP-seq dataset for Mycobacterium tuberculosis and observed that genome accessibility is critical to our understanding of TF-binding in vivo. This new model provides practical applications, such as de novo prediction of TF-binding peaks and a framework to measure DNA accessibility from ChIP-seq data. Our model enables us to quantify the relationship of genome accessibility with genomic features and suggest mechanisms that influence genome accessibility in vivo (e.g. distance to oriC). The model proposed in this study gives new perspective for ChIP-seq analysis in bacteria towards an improved description of gene regulation in silico.
| In order to adapt to different environmental challenges, microorganisms need to precisely control the expression of specific sets of genes at defined magnitudes at any given moment [1, 2]. This control is mediated by regulatory proteins such as transcription factors (TF) that are able to recognize and bind specific DNA sequences to recruit or block the gene expression machinery. Recent advances in next-generation sequencing have now enabled us to measure TF-binding in vivo at the genome scale [3–5].
Chromatin Immunoprecipitation followed by sequencing (ChIP-seq) is a popular technology for in vivo measurements of TF binding [6–8], which uses TF-specific antibody selection and high-throughput sequencing to identify the genomic regions that are bound by a query TF. In parallel, technologies for high-throughput characterization of TF-binding in vitro have also emerged [9–13]. Yet, only a fraction of the expected binding sites are bound under physiological conditions [8] and in vivo measurements are poorly correlated with in vitro ones [14, 15].
TF-binding in vivo is often more complex than what can be measured in vitro due to multiple factors [16]. For instance, strength of TF-binding affinity [17, 18], presence of multiple binding sites [19], cooperative interactions [18, 20], and genome accessibility [21, 22] have all been shown to impact TF-binding in vivo. Incorporating these parameters in ChIP-seq analysis can lead to more accurate models of gene regulation across the whole genome [14, 15].
As sequencing costs continue to decrease, challenges in ChIP-seq studies are transitioning from data generation to analysis and modeling [23]. Data analysis methods have moved from purely peak identification to physically-motivated models of ChIP-seq coverage [24]. Early computational methods focused on identifying statistically enriched peaks that correspond to TF-binding regions [5, 25–28]. Recent methods are incorporating mechanistic principles to extract regulatory insights [24, 29–31]. For example, the BRACIL method integrates ChIP-seq coverage, motif score, and thermodynamic modeling through a signal processing representation to predict binding site locations with high-resolution as well as cooperative interactions [24]. The growing abundance of ChIP-seq data creates a greater demand for more comprehensive models [15, 23, 32] and an opportunity to evaluate key parameters of TF-binding in vivo.
Within the cell, transcription factors need to have physical access to the relevant regulatory regions in order to control gene expression. In eukaryotes, genome accessibility is mostly caused by different chromatin states due to epigenetic factors such as histone modification and nucleosome structures [33]. The chromatin state can lead to gene silencing throughout the genome and have been used to estimate genome accessibility. In contrast, bacteria do not organize their genome in nucleosomes, thus genome accessibility is a subtle feature that is hard to be measured. In general, accessibility is not uniform across the genome due to the presence of global factors such as nucleoid associated proteins (NAPs) that alter genomic architecture [15, 21, 22] or local factors such as presence of repressor elements that block recruitment of RNA polymerase [21, 34]. Alteration of global genome structure can lead to changes in gene expression [35, 36]. For example, NAPs are associated with highly expressed genes that are organized into transcription factories [21]. The challenges in measuring and estimating genome accessibility have impeded the incorporation of this feature into bacterial ChIP-seq analysis.
Here, we present a novel biophysically motivated model that incorporate genome accessibility and highlights its importance in assessing TF-binding in bacteria. Extending our previous efforts to mechanistically characterize ChIP-seq coverage information [24], our model treats ChIP-seq binding profiles as a Boltzmann distribution with two parameters: genome accessibility and binding affinity. We applied this model on a large-scale dataset used to map the regulatory network of M. tuberculosis and compared the results to a simplified model that only considers binding affinity. Our results show that genome accessibility can explain variability in ChIP-seq coverage and peaks, and is associated with specific groups of gene function.
Using ChIP-seq data, biophysically motivated models can provide a quantitative framework for determining key parameters of in vivo TF-binding. We represent the ChIPseq profile in region bins of 500 bp and look for the influence of genome accessibility in TF binding. From a thermodynamic perspective, the probability, pij, that a TF j binds to a genome region i depends on the affinity between the TF and the specific sequence it binds, wij, as well as on the degree that this region is accessible, ai. Formally, the probability of binding is defined by the following equation (see Methods for detailed derivation):
log(pij)=ai+wij
(1)
TF-binding is represented in terms of binding affinity alone by constraining ai = 0 for all i. The accessibility parameter is inferred indirectly by performing linear regression on a large-scale dataset of ChIP-seq experiments [15, 32]. The affinity parameter is obtained from the motif score. The parameter ai describes a global trend in the probability of binding to region i by any TF. Here, we refer this as the genome accessibility for better biological interpretation of the results. Fig 1A and 1B illustrates schematically how genome accessibility influences TF-binding. Eq 1 is motivated by the poor correlation between ChIP-seq coverage and motif score (S1 Fig). For example, genomic regions with weak motif scores are observed with strong binding signal and vice versa (Fig 1).
We evaluated the extent to which genome accessibility can explain ChIP-seq data. We model our data according to Eq 1 and use a linear fixed effect model to estimate parameters and predict ChIP-seq profiles. The dataset comprises ChIPseq data for a total of 64 unique TFs obtained under same protocol and growth condition (see Methods). The ChIP-seq profile for a specific TF is defined as the normalized abundance of sequence reads that align to each region. The result suggests that the accessibility parameter is a global trend that provides preferential binding on specific genomic regions. We observe that genome accessibility improves prediction of ChIP-seq profiles when compared to a model that considers only binding site affinity. Quantitatively, the accessibility model explains 63% of the observed variance, while motif score alone explains only 35% (p-value <10−16, Fig 2). We also explored a more complex representation for binding affinity that considers best motif match, number of binding motifs and a combined score for all motif matches. The combined score is defined as the sum of -log(pvalue) for all motif matches. The accessibility values estimated by the more complex model is almost the same as the one estimated by the model that considers only best motif match (correlation above 99.9%; S8 Fig).
Our model can predict functional features that are useful in ChIP-seq analysis. The most common task in ChIP-seq analysis is the identification of TF-binding peaks, i.e. genomic regions that are bound by the TF under query, which shows a peak in ChIP-seq coverage [5, 28]. We classify regions into two groups: peaks or not peaks, according to peak-caller method described in previous work [15]. Each region is ranked with a score that indicates how likely they are to contain a peak. Given a threshold, false positives represent regions classified as peak by peak-calling but labeled as not peaks by the ranking score for de novo peak prediction. Similarly, false negatives represent regions that are classified as not peaks by peak-calling but labeled as peaks by the ranking score for de novo peak prediction. The rank for peak classification is defined according to motif and accessibility score and used to construct the ROC curve. Motif score is defined as the maximum log(p-value) of motif match per region bin and accessibility score is the estimated value for parameter ai from Eq 4. We consider three models for peak classification: motif only, motif plus accessibility, and normalized motif plus normalized accessibility. The first model predicts peaks using only motif score obtained by motif scan; the second model uses the sum of motif score and accessibility value; the last model rescale the values of motif score as well as accessibility in the interval from 0 to 1 and use their sum for peak prediction (see Methods).
The results show that DNA accessibility improves de novo ChIP-seq peak predictions when compared to predictions that consider motif only. As measured by the area under a receiver operating characteristic (ROC) curve, de novo ChIP-seq peak prediction occurs with values 0.69, 0.75, and 0.82 for method that uses motif only, motif score plus accessibility, and normalized motif score plus normalized accessibility, respectively (Fig 3A). The affinity values are sequence specific and by definition do not dependent on experimental conditions while the accessibility parameters may vary depending on experimental condition (S9 Fig). Therefore, given that TF-binding affinity score is previously known, one would only need to measure genome accessibility to predict TF-binding under novel growth conditions or for TFs with known binding motifs. This rationale can significantly reduce the need for additional ChIP-seq experiments.
The ability to predict ChIP-seq peaks de novo depends on the robustness of the genome accessibility metric and the ease to estimate its parameters under novel experimental conditions. The robustness of DNA accessibility values is illustrated by plotting the accuracy of accessibility values as a function of dataset size used for their estimation, i.e. the expected Pearson correlation between the accessibility estimated in a subset of given size versus the accessibility estimated in the entire dataset (S2 Fig). The expected accuracy for the accessibility values is estimated from 100 distinct samples for each subset size. We observe that as low as 10 ChIP-seq experiments is sufficient to estimate the accessibility values with ~90% accuracy (Fig 3B and S2 Fig).
The global trend in genome accessibility is robust to overexpression of a single TF. The ChIP-seq experiments used in this analysis were obtained under the same experimental set, with the exception that the TF under query was overexpressed [15]. We observe that removing any single TF from our dataset does not affect the estimated accessibility value (correlation between estimates are >99%). This indicates that the estimation of genome accessibility is robust to single TF overexpression. Moreover, we observe that just a few ChIP-seq experiments are sufficient to estimate genome accessibility with high correlation to its reference value. Only two ChIP-seq experiments are sufficient to estimate accessibility values with expected 0.7 correlation to the reference (Fig 3B and S2 Fig). We also observed that binding profile of some TFs are better correlated with the estimated accessibility values (S4 Fig). This result may indicate TFs that play a key role on genome structure or good candidates to infer genome accessibility.
Our model can be used to measure the accessibility state of each region in the genome. We sought to determine if genome accessibility is associated with various genomic features. Consistent with previous studies [37], intergenic regions are more accessible than protein coding regions (Fig 4A). Genome accessibility also appears to vary between genes or their regulatory regions based on their Clusters of Orthologous Groups (COG) assignments. In particular, genes or their regulatory regions in COGs for metabolism and transport of amino acids (COG category E) as well as carbohydrates (COG category G) are less accessible, while COGs for translation (COG category J) and transcription (COG category K) are more accessible (p<0.05 after Bonferroni correction; see Fig 4B). The observation of higher genome accessibility in transcription and translation genes is consistent with previous observations that DNA structure plays a critical role in expressing rRNA operons [21, 38, 39]. Finally, we observe that expression levels are positively correlated with genome accessibility (R2 = 0.23, Pearson correlation, Fig 4C). Interestingly, our results show that the expected expression level is the highest at intermediate values of genome accessibility (Fig 4D), which suggest that there may be a non-linear relationship between accessibility and gene expression.
Furthermore, our analysis shows that genome accessibility is biased by genomic position and GC content (Fig 5). Accessibility has a strong negative correlation with GC content (Fig 5A). In addition, accessibility is negatively correlated with distance to the origin of replication, oriC (Fig 5C), while no apparent correlation is observed in comparison to genome position alone (Fig 5B). This suggests two possible mechanisms that may influence genome accessibility: (i) DNA replication makes genomic regions more accessible for TF-binding, or (ii) there is a higher copy number of genomic regions near the oriC, leading to an apparent increase in genome accessibility (Fig 5D). These two mechanisms are not necessarily mutually exclusive and would be interesting to explore in future studies.
In this study, we developed a biophysically motivated formulation for bacterial ChIP-seq analysis that contributes to new biological insights of the role that genome accessibility plays in bacterial gene regulation. The model highlights the importance of binding affinity and genome accessibility for in vivo TF-binding. The model formulates the TF-binding process in thermodynamic terms and derives a linear relationship between accessibility, binding affinity, and probability of binding. This relationship enables us to estimate the model parameters from ChIP-seq data. We optimized our statistical framework with a fixed-effects representation to make parameter estimation more computationally efficient.
Numerous studies have investigated the role of genome accessibility on TF-biding in eukaryotic organisms [30, 40–44]. However, to the best of our knowledge, the work described here is the first attempt for a genome-scale quantitative measurement of DNA accessibility in bacteria. In eukaryotes, reads from DNAse I assays are well-correlated with binding regions [40, 41]. Pique-Regi et al. reported that DNAse I assays can inform genome accessibility for predicting ChIP-seq peaks from ENCODE data using a Bayesian probabilistic model that integrates accessibility with motif information from position weight matrix (PWM), TSS location and evolutionary conservation [29]. Other studies [43, 44] used a threshold on the coverage of DNAse I signal was used to distinguish accessible from silent genome regions and infer TF-TF interaction as well as set of TFs that drive tissue, cell type, and developmentally specific gene expression patterns in Drosophila. Foat et al. developed a thermodynamics model of binding based on equilibrium dissociation constant between bound and unbound states and used a least square regression model to infer binding affinity from ChIP-chip data of Saccharomyces cerevisiae [42]. However, genome accessibility was not considered in the model. Peng et al. developed another thermodynamic model that includes accessibility and binding energy to predict expression dynamics in Drosophila [30]. Accessibility was inferred from DNAse I assays and model parameters were trained based on an objective function that rewards good fit on highly expressed bins.
The method proposed in this paper has several novel features in comparison to those outlined above for eukaryotes. In contrast to eukaryotic genome accessibility models, which are inferred directly from DNAse I assays, our method infer accessibility from binding profiles of multiple ChIP-seq characterized TFs. Our thermodynamics model of TF-binding is derived in terms of binding affinity and genome accessibility by using Lagrange multipliers and free energy of Helmholtz (see Methods). A mixed effects linear regression model is used to make fit efficient and computationally feasible. In addition, the quantitative assessment of DNA accessibility in bacteria provides the possibility of testing hypothesis, novel biological insights, and applications.
The framework described here could be used to assess TF-binding using a reduced set of necessary ChIP-seq measurements. Instead of collecting ChIP-seq data for each TF in every new experimental condition, one would only need to perform a small set of experiments to estimate the state of genome accessibility. Then, in combination with established TF affinity data, one can accurately predict TF-binding genome-wide as demonstrated here. This approach could link both in vitro and in vivo experimental datasets under a unifying framework. Our model provides a step forward in our ability to infer TF-binding at different growth states in silico to capture the dynamic nature of gene regulation in bacteria.
Biophysical processes in vivo as well as experimental protocols should be considered for proper interpretation of accessibility values. Variance in DNA structure, binding competition, or in vitro artifacts in immuno-precipitation affects the measured genome accessibility. NAPs can shape genome structure at a global scale, while specific genome modification factors can affect accessibility within a particular regulon. Multiple transcription factors that bind to the same genomic region may lead to binding competition, causing a decrease in the observed accessibility. Variations in immuno-precipitation protocols and inherent noise in the technique may lead to variation in the estimation of binding specificity and sensitivity. These and other factors may cause genome accessibility to contain bias from ChIPseq experiments and could be helpful in providing better background estimation.
Ultimately, the importance of accessibility in bacteria genome remains to be further explored. In eukaryotic cells, genomic accessibility is critical in fine-tuned gene regulation [45] through controlled activation [46], minimizing biological noise [47], and providing epigenetic regulation [33]. These processes may be similarly important in bacteria physiology. For instance, genomic accessibility could cause stochastic gene expression and influence cell fate [48]. Engineering or altering genome accessibility may lead to new approaches in synthetic gene regulation and advance research in systems and synthetic biology [33].
Our work highlights that new biological insights can be obtained through biophysically-motivated mechanistic models of gene regulation. This approach should inspire more refined models of cellular physiology and adaptation. Here, we showed that thermodynamic principles can improve our understanding of TF-binding and genomic structural states. More refined models that integrate accessibility and binding affinity with other factors such as cooperative interaction and multiplicity will enhance our understanding of gene regulation, which will lead to a more comprehensive representation of whole cell physiology [49].
The probability of TF-binding to a specific region is represented as a Boltzmann distribution that depends on two parameters: accessibility and affinity. The accessibility parameter, ai, is specific to the DNA region and represents how likely a region i is to be bound by any transcription factor. The affinity parameter, wij, represents the specific affinity between a transcription factor j and a region i. Formally, the probability that a TF j binds at region i, pij, is defined as:
pij=eai+wij
(2)
This representation omits negative signs and the temperature parameter because they are not relevant to the approach in this study. In thermodynamic terms, Eq 2 represents a grand-canonical ensemble in which each region bin can exchange particles (i.e. TFs) and energy. The parameter ai represents the chemical potential in region i and the parameter wij represents the energy associated with TF binding (see S1 Text for detailed mathematical derivation).
The probability pij can be measured directly from the ChIP-seq data. In order to make this parameter robust and independent on the sequencing depth, we define pij as
pij=Ci,j∑iCi,j
(3)
where the coverage parameter, Ci,j, represents the number of reads from experiment j that lies in region i. A formal definition for region bins is presented in the next section.
Eq (2) can be transformed to a linear representation. This representation is shown in Eq 1 in the main text and repeated here for clarity:
log(pij)=ai+wij
(4)
Eq 4 permits that we use simple linear regression to estimate the parameters that determine ChIP-seq profiles.
This study is restricted to TFs whose binding sites can be summarized by a position weight matrix (PWM). Motif PWM was obtaining as the output of BRACIL [24]. The PWM provides a first order approximation of the affinity between the TF and the region it binds [10, 50]. We call si,j the affinity score of TF j to region i estimated according to the PWM. This approximation can be placed in Eq 4, and simplify linear regression as following:
log(pij)=ai+tj⋅sij
(5)
The parameter tj is a constant that represents underlying variables specific to each ChIP-seq experiment, such as TF concentration, ChIP-seq coverage as well as quality of immuno-precipitation. The affinity score, si,j, is defined as the -log10(p-value) of motif match with highest score in region i. Motif scan is performed using FIMO [51]. A affinity score of 2 was given to regions without any motif match.
We assume binding affinity to the sequence decreases monotonically with motif p-value. The p-value indicates the probability a score as good as the one observed in motif match occurs by chance according to the reference motif PWM. Thus, the binding affinity is monotonically correlated with–log10(p-value) of a motif match. By expanding it in Taylor series, the term–log10(p-value) becomes a first order approximation for binding affinity that suffices for the purpose of this research.
The genome is binned in regions of 500 bp to create a standardize profile and enable comparison of multiple TF experiments simultaneously. Cases with very low coverage are removed from analysis. In numbers, we classified the M. tuberculosis genome in 8824 region bins and only considered data points in which log(pij) > -10. Our rationale is to set up a threshold that considers data points that are informative for analysis and remove noisy ones that decrease the quality of genome accessibility estimation. 82.5% of the data points are used for analysis after applying the threshold of log(pi) >-10. This choice is supported by a sensitivity analysis that considers a wide range of minimum coverage threshold (S5–S7 Figs). The results are also robust for varying size of region bins (S10 Fig).
We optimized the statistical representation of Eq 5 to make the analysis practical and more efficient. The naïve approach would be to solve Eq 5 by a simple least square minimization. However, the number of data points and parameters needed would exceed 106 and 104, respectively. The least square minimization by QR decomposition (function lm in R) is impractical and we used a linear mixed-effects model (function lmer, R package lfe) instead.
The linear mixed-effects model optimizes regression because the parameter related to regional accessibility can be described as a random effect that shift the intercept of the probability of binding. As most parameters of Eq 5 correspond to the accessibility value of a region bin, the linear mixed-effects representation makes computation much more efficient.
In lmer annotation, our model uses the following formula: `log(p) ~ s·t + (1|region_bin)`, where p, s, and t are general representation of the corresponding parameters in Eq 5 and `(1|region_bin)`represents the random effect caused by accessibility to each region bin. The model that considers binding affinity only is represented as: `log(p) ~ s·t`.
We use three methods for de novo peak prediction: motif only, motif + accessibility and normalized motif + normalized accessibility. Motif only rank regions according to best motif match. Motif + accessibility sums the score of motif match (in terms of -log10(pvalue)) with the accessibility values estimated from fitting Eq 1 in the data. Finally, we define the minimum score to be 0 and maximum score to be 1 and re-scale motif as well as affinity score accordingly. This sum of the re-scaled score is used to rank regions for the method normalized motif + normalized accessibility.
The ChIP-seq data used for this analysis was obtained from a large-scale study that mapped the regulatory network of M. tuberculosis [15, 32]. The TF under query was FLAG-tagged and over-expressed under control of a mycobacterial tetracycline-inducible promoter. The enriched regions were computed according to the log-normal background model described in [15]. The binding motif was obtained as the output of the algorithm BRACIL [24], which uses MEME [51] to perform motif identification. FIMO [51] was then used to scan for binding sites at each region. Only TFs that recognize a binding motif with E-value < 10−5 were selected for this analysis. This resulted in a total of 99 ChIP-seq experiments that comprises 64 TFs.
Gene expression was defined as the median expression from the set of TF overexpression data, as described previously [15, 52].
COG categories were obtained from ftp://ftp.ncbi.nih.gov/pub/COG/COG2014/data and mapped to H37rv loci according to GENBANK annotation.
The code and corresponding documentation are available at https://sourceforge.net/projects/brasolia.
|
10.1371/journal.pntd.0004630 | Chikungunya Fever Cases Identified in the Veterans Health Administration System, 2014 | During December 2013, the first locally transmitted chikungunya virus (CHIKV) infections in the Americas were reported in the Caribbean. Although CHIKV infection is rarely fatal, risk for severe disease increases with age and medical comorbidities. Herein we describe characteristics of Veterans Health Administration (VHA) patients with CHIKV infection and, among those with infections diagnosed in Puerto Rico, investigated risk factors for hospitalization.
We queried VHA’s national electronic medical records to identify patients with CHIKV testing during 2014. Demographics, clinical history, laboratory results, and outcomes were abstracted. We investigated risk factors for hospitalization among patients with laboratory-confirmed CHIKV infection in Puerto Rico.
We identified 180 laboratory-confirmed CHIKV infections; 148 (82.2%) were diagnosed in Puerto Rico, and 32 (17.8%) were diagnosed among returning travelers elsewhere in the United States. In Puerto Rico, where more patients were hospitalized (55.4% versus 20.0%) and died (4.1% versus 0%), risk for hospitalization increased with age (relative risk [RR]/each 10-year increase, 1.19; 95% confidence interval [CI], 1.06–1.32) and, adjusted for age, increased among patients with congestive heart failure (RR, 1.58; 95% CI, 1.25–1.99), chronic kidney disease (RR, 1.52; 95% CI, 1.19–1.94), diabetes mellitus (RR, 1.39; 95% CI, 1.06–1.84), or chronic lung disease (RR, 1.37; 95% CI, 1.03–1.82).
CHIKV infection is an emerging problem among Veterans residing in or visiting areas with CHIKV transmission. Although overall mortality rates are low, clinicians in affected areas should be aware that older patients and patients with comorbidities may be at increased risk for severe disease.
| Infection with mosquito-borne chikungunya virus causes fever and severe diffuse joint pain—an illness known as chikungunya fever, or "that which bends up." Epidemics of chikungunya fever have occurred in Asia, Africa, and Europe. Not until December 2013 were there reports of chikungunya virus infection occurring in the Americas. Since then, it has involved most countries in the Western Hemisphere with >1.1 million cases reported by the end of 2014. Previous data from the Réunion Island outbreak demonstrated that older patients and patients with certain chronic medical conditions may have a higher risk of severe disease. The Veterans Health Administration is the largest health care system in the United States and has facilities in U.S. territories, including Puerto Rico, which has been heavily affected by this epidemic. Among Veterans in Puerto Rico, we investigated risk factors for severe disease and described all chikungunya-associated deaths. Risk for hospitalization increased with age, and for patients of the same age, was increased among those with congestive heart failure, chronic kidney disease, diabetes, or chronic lung disease. Further work is needed to determine whether prevention strategies targeted to those who may be at greatest risk for severe disease could help decrease morbidity and mortality among these populations.
| Chikungunya virus (CHIKV), an alphavirus most commonly transmitted by Aedes species mosquitoes, causes chikungunya fever (CHIK), characterized by acute-onset of fever and what is frequently described as incapacitating polyarthralgia [1, 2]. Since CHIKV was first identified in Tanzania in 1953 [2], epidemics have occurred in South and Southeast Asia, Africa, and Europe [1, 3, 4]. Three distinct genotypes have been described, including East/Central/South African, Asian, and West African [1, 5]. In December 2013, locally transmitted infection in the Western Hemisphere was first reported; the predominant strain is closely related to the Asian genotype [6–8]. CHIKV has rapidly disseminated among this largely immunologically naïve population; the Pan American Health Organization (PAHO) reported >1.1 million suspect cases, involving the majority of Western Hemisphere countries by the end of 2014 [9]. During 2014, over 31,000 cases (14% laboratory-confirmed) were reported in Puerto Rico [10] and over 1,500 cases (17% laboratory-confirmed) were reported in the U.S. Virgin Islands [11]; these figures are thought to underestimate disease burden because they do not include patients who did not present for care nor those who presented, but were not reported or for whom diagnostic testing was not completed [12]. During 2014, 2,811 laboratory-confirmed infections in the United States were reported to the Centers for Disease Control and Prevention (CDC) through the ArboNET surveillance system; the majority were among returning travelers, except for 12 persons in Florida with locally transmitted infection [13].
CHIK is usually self-limited, with the majority of symptoms typically resolving in 7–10 days [1]; however, patients can have prolonged rheumatologic symptoms [14, 15]. CHIKV infection can also be associated with severe illness, involving neurologic, cardiovascular, respiratory, renal, and ocular manifestations [16]. Although overall mortality is low, estimated at 0.3/1,000 population per year on Réunion Island [17], risk for severe disease and death increases with age and is higher among patients with certain comorbidities [17–19].
The Veterans Health Administration (VHA) has health care facilities throughout the United States and U.S. territories. Because 45% of all U.S. Veterans and 62% of Veterans in Puerto Rico are aged ≥65 years [20], and VHA patients have more comorbidities than Veterans who receive care at non-VHA facilities or non-Veterans [21], VHA patients might be at higher risk for severe CHIK than those in the general U.S. population exposed to the virus (i.e. returning travelers and those living in areas with CHIKV transmission). VHA’s Public Health Surveillance and Research Group (PHSR) performs surveillance for emerging infections among VHA patients. After the CHIK epidemic involved U.S. territories, PHSR began performing CHIK epidemiologic surveillance in July 2014 of all patients with laboratory-confirmed CHIKV infection diagnosed at VHA facilities during 2014. Herein, we describe characteristics of these patients, compare clinical findings with patients who tested negative, investigate risk factors for hospitalization, and report phylogenetic analysis of CHIKV strains detected.
We identified patients from all VHA facilities with CHIKV test results for specimens collected during January 1, 2014–December 31, 2014, utilizing VHA’s Corporate Data Warehouse, a national data repository from VHA’s VistA electronic medical record system. Because VHA has no standardized naming convention for laboratory tests and inconsistent application of universal codes that can be utilized for test identification, we queried this data warehouse for any test name containing the term "chik." After distributing CHIKV surveillance reports to VHA facilities beginning in July 2014, PHSR was contacted by VHA facilities that were not listed in the reports but had diagnosed infections. Consequently, we expanded the query to search for "chik" in the comments field to capture additional CHIK diagnostic test results associated with other test names (e.g., “dengue”, “miscellaneous”, and “reference laboratory”). The last query was performed on February 26, 2015.
Apart from VHA’s Public Health Reference Laboratory (PHRL), VHA clinical laboratories do not have CHIK diagnostic testing capability, and thus specimens are sent to non-VHA laboratories for testing. When capacity of non–VHA laboratories in Puerto Rico became overwhelmed by demand [12], not all specimens submitted for CHIKV testing by Veterans Affairs Caribbean Healthcare System (VACHS) in Puerto Rico (1 medical center and 8 outpatient clinics) were processed, including some specimens from patients who subsequently died. To further investigate, we queried the data warehouse for CHIKV testing orders. For patients without results, medical records were reviewed to determine whether tests remained pending, or if cancelled, the reason for cancellation. During December 2014, VHA PHSR provided a list of patients without results to laboratories in Puerto Rico to recover any remaining specimens for testing by PHRL. PHRL performs internally and externally validated CHIKV reverse transcriptase PCR (RT-PCR) (CDC, CHIKV RT-PCR assay protocol, Fort Collins, Colorado) [22] and IgM enzyme-linked immunosorbent assay (ELISA) (Abcam, Inc., Anti-Chikungunya Virus IgM Human ELISA kit, Cambridge, Massachusetts) according to manufacturer’s recommendations. For patients who had specimens collected for CHIKV testing in Puerto Rico or elsewhere in the United States and died before the last query was performed, a physician-epidemiologist reviewed medical records to confirm presenting symptoms were consistent with CHIK (i.e. fever and either oligoarthralgia, polyarthralgia, or myalgia) and the patient resided in or had recently visited an area with CHIKV transmission, and determined whether CHIKV infection might have contributed to death (i.e., the patient had died before recovery from acute illness and death was not clearly from unrelated causes). Although PAHO defines a suspect case as a patient with acute onset of objective fever (>38.5°C or 101.3°F [23, 24]; >38.3°C or 101.0°F [9]; >38.0°C or 100.4°F [9]) and severe arthralgia, not otherwise explained, who resides in or has visited epidemic or endemic areas ≤2 weeks before symptom onset [9, 23, 24], we did not utilize this definition because the majority of our laboratory-confirmed cases did not meet these criteria.
For all patients with available CHIK diagnostic test results, we abstracted demographics, clinical history, including symptoms during the acute illness, laboratory results, and outcomes from medical records. Comorbidities were identified by review of problem lists and provider notes. A laboratory-positive case was defined as a patient with detectable CHIKV RNA by RT-PCR or anti-CHIKV IgM antibody. Patients with positive serology, but without travel to an area with known CHIKV transmission, consistent with a false-positive result, as well as patients with inadequate testing to rule out CHIKV infection (negative RT-PCR >8 days after symptom onset and no serology or negative IgM <4 days after symptom onset and no RT-PCR or convalescent serology) were excluded from statistical analysis [24]. Demographics and clinical characteristics of patients with laboratory-confirmed infection diagnosed in Puerto Rico versus elsewhere in the United States were compared, and clinical findings of CHIKV-positive versus CHIKV-negative patients (all remaining patients with negative CHIKV tests) were compared by using chi-square test for categorical (Fisher’s exact test when n <5 in any cell) and t-test for continuous variables.
To determine whether the association between age and hospitalization was modified by whether CHIK was diagnosed in Puerto Rico versus elsewhere in the United States, analysis was stratified by patient location. Poisson regression with robust error variance [25, 26] was utilized to investigate risk factors for hospitalization among patients with laboratory-confirmed infection in separate age-adjusted models. Potential risk factors investigated included comorbidities, vital signs on presentation, and laboratory abnormalities at presentation and during clinical course. Analyses were performed by using SAS 9.2 (SAS Institute, Inc., Cary, North Carolina).
Finally, we sequenced the CHIKV E1 envelope glycoprotein gene from a convenience sample of patients from Puerto Rico with detectable CHIKV RNA. RNA was extracted from serum by using Qiagen DSP viral RNA mini kits (Qiagen, Germantown, Maryland) according to manufacturer’s instructions. RT-PCR was performed by using Superscript One-Step Platinum Taq HiFi (Thermo Fisher Scientific, Inc., Waltham, Massachusetts) and primers CV1F and CV1R [27]. Resulting PCR products were purified by using Qiagen PCR purification kits and subjected to population-based sequencing by using standard dideoxyterminator techniques. Sequences were assembled, edited, and compared with sequences in GenBank from the Western Hemisphere and previous outbreaks using Geneious software (Biomatters, Inc., San Francisco, California).
This study was reviewed by CDC for human subjects protection and was deemed to be non-research. It was also approved by the Stanford University Institutional Review Board and fulfilled the requirements of regulation OHRP 45 CFR 46.116 (d) for waiver of informed consent. Patient data was anonymized after data abstraction.
We identified 264 patients with CHIKV results during 2014 (Fig 1). Results for 184 patients from VACHS in Puerto Rico were available (22.8% of 806 patients from whom specimens were collected for CHIK diagnostic testing; testing was not completed for the remainder) (Fig 2 and S1 Fig). Results for 80 patients were from other U.S. facilities. Twelve patients were excluded; 1 with an apparent false-positive CHIKV IgM/IgG result and 11 with possible false-negative results because of inadequate testing. The remaining 252 patients, including 180 in Puerto Rico and 72 elsewhere in the United States, were included in the analysis.
Overall, 180 patients had laboratory-confirmed CHIKV infection (Table 1). Specimens for CHIKV testing from Puerto Rico were collected a mean of 4.5 d ± 5.4 d from symptom onset. Specimens from the rest of the United States were collected a mean of 28.7 d ± 40.3 d from symptom onset (p = 0.003). In Puerto Rico, 148 (82.2%) of 180 patients tested were CHIKV-positive; 145 (98.0%) were confirmed by RT-PCR and 3 (2.0%) by serology alone. Outside Puerto Rico, 32 (44.4%) of 72 patients tested were CHIKV-positive; 4 (12.5%) were confirmed by RT-PCR and 28 (87.5%) by serology alone. Fourteen (19%) of 72 patients tested for CHIKV infection outside Puerto Rico had no recorded travel to an area with CHIKV transmission prior to symptom onset, all of whom were CHIKV-negative; none of the patients tested in Puerto Rico had symptom-onset prior to possible exposure to circulating CHIKV. Patients with CHIKV infection in Puerto Rico were older (mean, aged 69 versus 55 years; p < .0001) and had more comorbidities than patients outside Puerto Rico.
Patients with CHIKV infection frequently reported oligoarthralgia or polyarthralgia (88.3%), subjective fever (84.4%), generalized malaise (76.7%), myalgia (69.4%), and rash (44.4%), and reported these symptoms more often than CHIKV-negative patients (Table 2). Among 165 patients with laboratory-confirmed CHIKV infection who had recorded temperatures during their acute illness, only 53 (32.1%) demonstrated objective fever (>38.0°C or >100.4°F) [9], and only 48 (29.1%) had both objective fever and arthralgia. Subjective fever and arthralgia was reported for 140 (77.8%) of 180 patients and subjective fever or any arthralgia for 174 (96.7%).
Among CHIKV-positive patients, 37.0% of 173 had leukopenia (<4,000 white blood cells [WBC]/μL) and 71.4% of 171 had lymphopenia (<1,000 lymphocytes/μL); these findings occurred more frequently compared with CHIKV-negative patients (Table 3). Thrombocytopenia (<150,000 platelets/μL) occurred among 80 (46.5%) of 172 patients with CHIKV infection, with mean nadir platelet count of 104,000/μL. Acute kidney injury (AKI) (≥0.3 mg/dL or 26.5 μmol/L increase in serum creatinine from last level [28]) was experienced by 33 (21.6%) of 153 patients, 4 (12.1%) of whom had stage III AKI. Hepatic transaminitis (aspartate aminotransferase >40 U/L or alanine aminotransferase >45 U/L) was experienced by 52 (40.3%) of 129 patients, 16 (30.8%) of whom had transaminases >3 times the upper limit of normal.
In Puerto Rico, 82 (55.4%) of 148 patients with laboratory-confirmed CHIKV infection were hospitalized, including 10 (6.8%) who required intensive care; elsewhere in the United States, 6 (20.0%) of 30 returning travelers with known hospitalization status were hospitalized (p = .0004). Whereas the hospitalization rate increased with age among patients in Puerto Rico (relative risk [RR]/ each 10-year increase in age, 1.19; 95% confidence interval [CI], 1.06–1.32), it did not increase with age among returning travelers. Because only 6 returning travelers were hospitalized, analysis is only presented for patients in Puerto Rico (Table 4). After adjusting for age, a significantly higher risk for hospitalization associated with having congestive heart failure (CHF) (RR, 1.58; 95% CI, 1.25–1.99), chronic kidney disease (CKD) (RR, 1.52; 95% CI, 1.19–1.94), diabetes mellitus (RR, 1.39; 95% CI, 1.06–1.84), or chronic lung disease (RR, 1.37; 95% CI, 1.03–1.82) remained. Adjusted for age, patients had a greater risk for hospitalization if they were tachycardic (>100 beats/minute; RR, 1.49; 95% CI, 1.12–1.98), had leukocytosis (>11,000 WBC/μL; RR, 1.65; 95% CI, 1.34–2.03), AKI (RR, 1.64; 95% CI, 1.33–2.04), or hepatic transaminitis (RR, 1.38; 95% CI, 1.07–1.80) at presentation.
Two patients with CHIKV infection, confirmed by RT-PCR, developed septic shock without another identified etiology. One patient, with CHIKV infection confirmed by RT-PCR of serum, developed meningitis with a CSF profile consistent with a viral etiology, demonstrating pleocytosis (30 WBC/mm3) with initial predominant monocytosis and elevated protein (88 mg/dL). CSF was not submitted for CHIKV testing and no CSF was recovered for diagnostic testing by VA PHRL. One patient presented with Guillain-Barré syndrome one month after CHIK; recent CHIKV infection was confirmed by positive IgM serology. Two patients with CHIKV infection, confirmed by RT-PCR, presented with diabetic ketoacidosis; one patient with CHIKV infection, confirmed by RT-PCR, presented with pancreatitis; one patient with CHIKV infection, confirmed by RT-PCR, presented with colitis, and one patient with CHIKV infection, confirmed by RT-PCR of serum, experienced monomicrobial nonneutrocytic ascites (recovered peritoneal fluid collected 20 days after symptom onset was CHIKV IgM and RT-PCR negative). Seven patients with CHIKV infection (6 confirmed by RT-PCR and 1 by IgM) experienced pneumonia during their clinical course. Four patients with CHIKV infection, confirmed by RT-PCR, experienced congestive heart failure exacerbations; three patients with CHIKV infection, confirmed by RT-PCR, presented with syncope; and one patient with CHIKV infection, confirmed by RT-PCR, experienced a non ST-elevation myocardial infarction. One patient with CHIKV infection, confirmed by RT-PCR, presented with epididymitis.
Among the 148 patients with laboratory-confirmed CHIKV infection, 6 (4.1%) died. All had viremia demonstrated by RT-PCR (Table 5). In addition to these six patients, there were 15 patients who died after presenting with an illness compatible with CHIK that may have contributed to death; specimens for these patients were submitted for CHIKV testing but were not processed. Although attempts were made to obtain these specimens submitted to non-VHA laboratories, they were not recovered for testing. These 15 patients are thus considered suspect cases (data available upon request).
The mean age of patients with laboratory-confirmed CHIKV infection in Puerto Rico who died was 78 years (range, 66–99 years), compared with 68 years (range, 23–94 years) for those in Puerto Rico who survived (p = 0.12). All 6 patients who died were afebrile on presentation (only one had a recorded temperature >100.4°F or >38.0°C during hospitalization); four were tachycardic on presentation. Of 5 with known medical history, all had multiple comorbidities.
CHIKV E1 envelope glycoprotein gene sequences (GenBank accession numbers: KU724228-KU724266) for 39 patients in Puerto Rico were closely related (<0.5% nucleotide difference) and nearly identical to strains in GenBank from St. Martin and the British Virgin Islands (Fig 3) [6, 7].
Our study is the first to characterize U.S. Veterans with laboratory-confirmed CHIKV infection. The majority received a diagnosis in Puerto Rico, although 32 were returning travelers who received a diagnosis elsewhere in the United States. Although the majority reported subjective fever, only one-third had objective fever, indicating that use of recommended CHIK case definitions, requiring objective fever, might underestimate disease burden among VHA patients [24]. Among returning travelers, approximately one-fifth were hospitalized, whereas in Puerto Rico, where patients were older and had more comorbidities, approximately half of laboratory-confirmed patients were hospitalized, 6.8% required intensive care, and 4.1% died. The fatality rate was higher among VHA patients compared with a preliminary report from Puerto Rico in December 2014 that described only 4 deaths identified by passive surveillance on the island [12]. Although we performed population sequencing of only a portion of the E1 envelope glycoprotein gene, we did not find substantial difference among sequences from Puerto Rico compared with other strains circulating in the Western Hemisphere [6, 7]. As in previous studies [18, 19], we report that age was associated with increased risk for hospitalization. After adjusting for age, CHF (but not coronary heart disease), CKD, diabetes, and chronic lung disease were associated with increased risk for hospitalization. While we cannot be certain of the individual physicians' criteria for hospitalization in many cases, we do know that patients were more likely to be hospitalized if they had unstable vital signs (e.g. tachycardia) or had abnormal laboratory results (e.g. leukocytosis, acute kidney injury, or hepatic transaminitis) at presentation.
Although literature review demonstrates low overall mortality, among persons with CHIKV infection presenting for care, the case fatality rate is not insignificant. During the CHIKV outbreak on Réunion Island (East/Central/South African genotype) [29], among 157 patients with laboratory-confirmed infection who presented to a medical center, 61.8% were hospitalized and 3.2% died [18]. Although patients in that study were >10 years younger and fewer had comorbidities, hospitalization and case fatality rates were similar to VACHS. They reported that diabetes and ischemic heart disease, unadjusted for age, was associated with increased odds of hospitalization [18]; only 11 patients in that study had CKD. Economopoulou et al. reported that among a subset of hospitalized patients with CHIK, cardiac disease, respiratory disease, as well as hypertension were associated with increased risk for severe disease [19].
Because our study included patients with laboratory-confirmed CHIKV infection, we cannot assess the overall burden of CHIKV infection among Veterans, only those who presented to VHA facilities and had appropriate diagnostic testing completed. During 2014, investigation of households surrounding laboratory-positive cases demonstrated that 28% of participants were laboratory-positive for current or recent CHIKV infection, and only 63% of symptomatic persons had sought care [12]. Although our query was robust, any patient with results not entered into the laboratory component of the electronic medical record (e.g., scanned or recorded in a progress note) would not have been captured. Sample size limited analysis of returning travelers, and in Puerto Rico, prevented inclusion of >1 comorbidity in age-adjusted models or assessment of risk factors for intensive care or death among patients with laboratory-confirmed infection. This study identified the lack of testing availability for VACHS, as well as deficiencies in CHIKV diagnostic testing across VHA. We identified 11 patients (4 in Puerto Rico and 7 elsewhere in the United States) with inadequate testing to diagnose CHIKV infection, which may have contributed to underdiagnosis of CHIKV-infection. In Puerto Rico this was because of underuse of serology for patients who presented >8 days after symptom onset. Outside Puerto Rico this was because of underuse of CHIKV RT-PCR or convalescent serology for patients who presented during the first week of symptom onset. Among returning travelers, many of whom presented for care in the U.S. during the convalescent period, when diagnosis is dependent upon serology, some diagnoses could have been missed as CHIKV IgM typically declines after several weeks to months [1]. Only 4 of the returning travelers had CHIKV IgM performed without simultaneous CHIKV IgG, and no patients had a negative CHIKV IgM and positive CHIKV IgG, however, suggesting that few cases may have been missed for this reason. Outside Puerto Rico, only 44% of patients tested for CHIKV had laboratory-confirmed infection; this not only reflects the lower prevalence of CHIKV infection outside Puerto Rico, but also improper testing of patients without symptom-onset after travel to an area with CHIKV transmission. Our surveillance activities determined that CHIKV testing availability for VACHS was lacking and resulted in VHA PHRL offering CHIKV (and dengue virus) testing. Further education of VHA providers regarding CHIKV infection, correct diagnostic testing (RT-PCR versus serology) on the basis of time from symptom onset, and available testing through VHA is needed.
Lessons learned from Puerto Rico are important for areas in the Western Hemisphere with ongoing CHIKV transmission as well as countries, including the United States, with similarly immunologically naïve populations and Aedes aegypti or Aedes albopictus vectors [6]. For clinical management, newly required CHIK public health reporting, and surveillance, having adequate laboratory testing capacity for timely results is helpful. Although testing all symptomatic persons might be infeasible, sufficient capacity to test those with severe (e.g. hospitalized patients) or atypical illness is needed. Clinicians practicing in areas with CHIKV transmission should be aware that CHIKV infection among elderly patients and patients with comorbidities, including CHF, CKD, diabetes, and chronic lung disease may be associated with more severe disease. To determine whether the risk of atypical complications is greater for CHIKV infection compared with other viral infections, a larger cohort of patients presenting with a viral syndrome would need to be studied. Further work to examine risk factors for intensive care and death among a larger sample of patients with laboratory-confirmed infection is needed to provide closer monitoring for those at greatest risk and to investigate the effect of prevention strategies [30] targeted to populations at greatest risk should they acquire CHIKV infection.
|
10.1371/journal.pntd.0006765 | Feasibility and initial outcomes of a multifaceted prevention programme of melioidosis in diabetic patients in Ubon Ratchathani, northeast Thailand | Melioidosis is an infection caused by Burkholderia pseudomallei, a Gram-negative bacillus found in soil and water. Diabetes mellitus is the most important risk factor for melioidosis. The recommendations for disease prevention include avoiding direct contact with soil and water, and drinking only boiled or bottled water.
A prospective intervention study was conducted to evaluate the feasibility and behavioural outcomes of a multifaceted prevention programme for melioidosis. Participants were diabetic adults in Ubon Ratchathani, northeast Thailand. Ten behavioural support groups consisting of 6 to 10 participants per group were conducted. Twelve behaviour change techniques were used: information about health consequences, credible source, adding objects to the environment, reconstructing the physical environment, instruction on how to perform a behaviour, demonstration of the behaviour, commitment, prompts/cues, self-monitoring of behaviour, goal setting, feedback on behaviour, and social support, and their feasibilities evaluated.
There were 70 participants, of median age 59 years and 52 (74%) were female. Participants found the intervention beneficial, interesting and engaging. Participants indicated that they liked to watch videos with information about melioidosis delivered by local doctors and patients who survived melioidosis, and videos showing use of over-the-knee boots by local farmers. Participants felt engaged in the sessions that trialed protective gear and that made calendars with individual photographs and self-pledges as a reminder tool. The proportions of participants reporting that they always wore boots while working in rice fields increased from 30% (10/33) to 77% (28/37, p = 0.04), and that they drank only boiled or bottle water increased from 43% (30/70) to 86% (59/69, p<0.001) at 6 months post intervention.
The programme is highly acceptable to participants, and can support behaviour change. Policy makers should consider implementing the programme in areas where melioidosis is endemic. Making calendars with individual photographs and self-pledges as a reminder tool could be powerful in behaviour change interventions, and further research on this component is needed.
| Melioidosis is a serious infectious disease caused by the Gram-negative environmental bacterium, Burkholderia pseudomallei. Infection in humans occurs following skin inoculation, inhalation or ingestion. The recommendations for melioidosis prevention include using protective gear such as rubber boots when in direct contact with soil and environmental water, and drinking only boiled or bottled water. A multifaceted prevention programme is recommended to achieve the desired behaviour changes. Here, we evaluated the feasibility and behavioural outcomes of a multifaceted prevention programme for melioidosis. Our study participants were diabetic adults in Ubon Ratchathani, northeast Thailand. We found that the multifaceted prevention programme was highly acceptable to participants, and could support behaviour change. A calendar with an individual photograph as a reminder tool engaged participants effectively. Our study also confirmed that commitment and action by the government are essential for the preventive interventions to be successful. We recommend that policy makers should consider implementing the programme in areas where melioidosis is endemic. Since cultures and barriers to adopting the recommended behaviours vary, the intervention strategies would need to be adapted to local contexts.
| Melioidosis is an often fatal infectious disease caused by a Gram-negative bacterium, Burkholderia pseudomallei, which is commonly present in soil and water in tropical regions [1]. A spatial modeling study estimated that there are about 165,000 human melioidosis cases per year worldwide, of which 89,000 (54%) die [2]. The disease is highly endemic and commonly reported in Southeast Asia and northern Australia, where the mortality from melioidosis is about 40% and 14%, respectively [3, 4]. Melioidosis occurs through ingestion, inoculation, or inhalation of the bacterium through direct contact with an environmental soil or surface water [5]. Diabetes mellitus is the most important predisposing factor for melioidosis, and is present in about half of all melioidosis patients [6]. Therefore, persons with diabetes are the main targets for disease prevention interventions. No melioidosis vaccine is currently available [7]. Activities associated with an increased risk of disease acquisition in Thailand, where disease is highly endemic, include working in a rice field, other activities associated with exposure to soil or water, and drinking untreated water [5]. The recommendations for disease prevention include using protective gear such as rubber boots when in direct contact with soil and environmental water, and consuming only boiled or bottled water [5]. However, only a small proportion of people follow such recommendations [8].
Lack of adoption of these preventive behaviours still occurs in Thailand even though the Ministry of Public Health (MoPH) has been recommending wearing rubber boots and drinking boiled water, and provides free rubber boots to prevent leptospirosis since a rise in leptospirosis incidence in 1996 [9, 10]. In a previous focus group study, we identified barriers to adopting recommended preventive behaviours in Thailand [8]. The main barriers were categorized into five domains: (i) knowledge, (ii) beliefs about consequences, (iii) intention and goals, (iv) environmental context and resources, and (v) social influence. People have little knowledge of melioidosis, believe that there is little or no harm in not adopting the recommended preventive behaviours, and are not inclined to use boots while working in muddy rice fields [8, 11]. People perceived rubber boots to be hot and uncomfortable, and they normally followed the behaviour of friends, family and their community, the majority of whom did not wear boots while working in rice fields and did not boil water before drinking [8].
To change behaviour, interventions based on the factors that influence adherence to recommendations are needed [12–14]. In general, providing information and protective gear alone do not change their behaviour [8]. Two related frameworks have been developed to support the investigation of a wide range of possible influences on behaviour: the Theoretical Domains Framework (TDF), and the Behaviour Change Wheel (BCW) [12–14]. The TDF is a useful framework for understanding the barriers and factors influencing specific behaviours [12, 13, 15, 16], while the BCW is a comprehensive framework that links this understanding to designing interventions including the recommended behaviour change techniques (BCTs) [14, 17]. Using these frameworks, we previously selected recommended behaviours, defined barriers to adopting those recommended behaviours, identified intervention options and modes of delivery, and developed a multifaceted prevention programme including a set of BCTs aimed at changing behaviours to prevent melioidosis, based on the local context in Thailand [8].
In this study, our aim was to evaluate the feasibility and behavioural outcomes of this multifaceted prevention programme for melioidosis [8] in diabetic adults in Ubon Ratchathani, northeast Thailand.
We conducted a study of a multifaceted prevention programme for melioidosis between April and December 2015. This was a small group intervention, in which 6 to 10 participants at a time attended a behavioural support group conducted by the study team. Each session lasted about 50 to 60 minutes. The intervention was provided once. Participants were then followed up by phone after one, two, four and five months, and by visiting homes on months three and six after the intervention. Feasibility of the intervention was determined by direct observation during the intervention, and by questionnaires and individual interviews after the intervention and at each follow-up. Components of the intervention were modified after each session based on feedback about feasibility. Two recommended preventive behaviours, wearing protective gear while working in rice fields and boiling water before drinking, were assessed prior to the intervention and at every follow-up by questionnaires and individual interviews.
The study sample was drawn from diabetic patients being followed up at five Tambon Health Promoting Hospitals (THPHs) in Ubon Ratchathani province, northeast Thailand. This comprised Non Noi THPH, Pak Kud Whai THPH, Pak Nam THPH, Ban Kok THPH and Hua Ruea THPH. THPHs are the first level of public health facility in Thailand.
All patients attending for diabetic follow-up who had a physician-confirmed diagnosis of diabetes mellitus, were oriented and could converse normally were invited on the day by the study team to participate. Those who had been diagnosed with melioidosis and had not completed oral eradicative treatment for melioidosis were not eligible to participate because, per standard of care, those patients would be being advised to adopt the preventive behaviours to reduce the risk of melioidosis re-infection [18, 19]. The sample size target was defined by practical and resource considerations as 60 to 100 participants attending 8 to 12 sessions. We ended the study with 70 participants having completed 10 sessions because there was no new feedback to modify the interventions further; saturation was reached. As we aimed to assess intervention and methods feasibility, not effectiveness in terms of either behavioural or clinical outcomes, we did not conduct power calculations.
The interventions included 12 of 13 BCTs recommended in a focus group study evaluating barriers and recommended interventions to prevention melioidosis, conducted in Ubon Ratchathani province, Northeast Thailand in 2012 [8]. The recommended 13 BCTs include information about health consequences (e.g. explaining that not wearing boots while working in rice fields and that drinking untreated water can lead to an often fatal infectious disease called melioidosis), credible source (e.g. a high status professional in the government giving a speech that emphasises the importance of melioidosis prevention), adding objects to the environment (e.g. providing baby powder and long socks to alleviate the problem of discomfort due to heat and humidity when wearing boots), reconstructing the physical environment, instruction on how to perform a behaviour, demonstration of the behaviour, commitment, prompts/cues, self-monitoring of behaviour, goal setting, feedback on behaviour, feedback on outcome(s) of behaviour and social support [8]. The examples of BCTs specific to the two recommended preventive behaviours, wearing protective gear while working in rice fields and boiling water before drinking, had been previously described [8]. The recommended BCT of ‘feedback on outcome(s) of behaviour’ was not used because the study had short study duration and, therefore, could not determine clinical outcome of acquiring melioidosis over the study period. In this study, the objective of the intervention was to increase the frequency of the two recommended preventive behaviours: wearing boots while working in rice fields and dinking boiled or bottled water. Boiling water before drinking, rather than buying bottled water, was the main recommendation among those who were drinking untreated water. Buying bottled water was not primarily recommended because it could be considered expensive and it was not consistent with the national recommendation of boiling water before drinking [9]. Filtering water before drinking was not recommended because filters were rarely maintained properly and B. pseudomallei had been detected in filtered water samples previously [8, 20].
The intervention package included six short videos, three pamphlets, and a calendar with a space for participants’ individual photographs and self-pledge. The materials are publicly available online (https://dx.doi.org/10.6084/m9.figshare.5734155). Each participant also received a pair of long socks and a bottle of baby powder (to reduce itching inside boots) and a 2-litre plastic ice bucket commonly-used to store water to drink while working in rice fields. In each behavioural support group, participants received an introduction by a moderator, watched each short video, and had short group discussions at the end of each video to foster autonomous motivation for the recommended preventive behaviours. Participants then had a protective gear trial session, in which multiple kinds of boots were provided for participants to test them out for wearing (Fig 1A). Next, the study team took a photograph of each individual participant while wearing boots and holding a kettle (Fig 1B) and printed photographs for each participant to use in the next session. Finally, participants attended a session to make their own calendar to act as a reminder tool for the recommended preventive behaviours. We asked participants to attach their individual photograph to the calendar and write their own pledge on the calendar by themselves (Fig 1C). It was recommended that the calendar be hung in participants’ houses (Fig 1D). The moderator also stimulated group discussion during, before and after the sessions. Additionally, we provided social support by giving information to nurses, doctors, participants’ relatives and health volunteers in each participating THPH about the intervention and potential benefits of the intervention. We also ask them to encourage the participant to continue with the recommended behaviours.
Components of the final programme, their related BCTs and intervention functions, and details of each short video are described in Table 1.
Mixed methods were used to evaluate feasibility. Descriptive statistics presented interquartile ranges as 25th and 75th percentiles. Qualitative data from the direct observations during the intervention were analyzed using thematic analysis. McNemar’s exact test was used to compare the percentage of participants reporting that they performed recommended preventive behaviours before and after the intervention. McNemar’s test was used because the evaluation was a repeated measurement of the same subjects over time [21]. Statistical analyses were performed using Stata version 14.0 (StataCorp LP, College Station, TX).
Of the 70 participants, 52 were female (74%) and the median age was 59 years old (interquartile range 52 to 65; range 32 to 77 years old). Forty-nine participants (70%) answered that they were farmers. Of 33 participants who worked in rice fields during the last week prior to the enrollment, six (18%) walked barefoot, 15 (45%) wore sandals, four (12%) wore boots sometimes and 10 (33%) wore boots every time while working in the rice fields. Boiling water from each source every time before drinking was reported in 2 of 4 participants (50%) who drank water from the well, 3 of 22 (14%) who drank borehole water, 4 of 14 (29%) who drank rainwater, and 8 of 29 (28%) who drank tap water. No participants drank water from a pond. Nine of 39 participants who drank bottled water (23%) also drank water from other sources without boiling. Overall, 30 (43%) drank only boiled or bottled water.
During the first four sessions, we received many comments and advice from participants. Participants suggested that the duration of the videos should be shorter, and pointed to information that should be added to the videos or presented by local healthcare workers in the real local setting. Therefore, videos were revised and recut, and the median and maximum duration of the videos were reduced from 3 and 5 minutes to 1:30 and 3 minutes, respectively. The video showing that wearing boots can protect from being cut by golden apple snails was added. Videos showing how to boil water were remade and presented by local healthcare workers in the real local settings rather than presented by the study team in the urban setting.
Based on direct observation, we found that the script of the moderator to stimulate group discussion between each video was too long, and participants took a lot of time to come up with their own pledges. Therefore, the script was shortened and the study team made a list of common and recommended pledges for participants to see and modify for their own pledges. Examples of the pledge included, “I will always boil water before drinking” and “I will always wear boots while working in the rice fields”.
At the end of the first three group sessions, we found that many participants could not remember the name of the disease and the recommended behaviours. Therefore, in the fourth group, a BCT of mental habit formation was employed by asking participants to shout three short two-part phrases repeatedly. The moderator would shout the first part and then asked the participants to shout the latter part together. The three phrases included (1) “to prevent–melioidosis”, (2) “work in rice fields–wear boots” (3) “drink–boiled water”. For each round, each phrase was repeated for three times, and this BCT was conducted for two rounds. Feedbacks at the end of the fourth group session about this BCT were good, and all participants could remember the name of the disease and the two main preventive behaviours. This mental habit formation has been included as one of the main BCTs for the programme since the fourth group (Table 1).
The third session was longer than 60 minutes; after modifications in accordance with feedback, the last seven sessions were shorter than 60 minutes. We received no additional suggestions after the fourth session, and, therefore, no further changes were made from the fifth to the tenth session.
Participants found the intervention beneficial, interesting and engaging. Features of the sessions that participants reported beneficial were the information about disease, and learning that applying baby powder and long socks could make wearing boots comfortable. Most participants had never heard of the disease and the consequences of the disease. Participants indicated that they liked to watch videos about melioidosis delivered by local doctors, relatives of those who died of melioidosis, and patients who survived melioidosis in the local dialect. Participants found that the video showing that farmers who wore over-the-knee boots could easily walk in muddy rice fields and that such boots were durable enough to protect themselves from golden apple snails. Many said that they had never known these things before.
Participants felt engaged in the sessions that trialed protective gear and that made calendars with individual photographs and a self-pledge as a reminder tool. Many participants reported that they saw over-the-knee boots available in the local market, but they had had no chance to try the boots and, therefore, had not known whether or not they would be comfortable and useful. We observed that most participants smiled while having their photos taken, holding their own photos, putting their photos on the calendar and writing their own pledges (Fig 1B and 1C). Based on the interactions and discussions between participants, our judgement was that most participants enjoyed the activities. During the home visits at the third and sixth month follow-ups, we found that 63/70 (90%) and 62/69 (90%) had their calendars hanging in the house, respectively. Many participants reported that they liked their own photos, as they had never had their own photo printed, and the picture of themselves wearing boots and holding a kettle was a good reminder tool for the recommended preventive behaviours. During the home visiting, we also observed that all participants had their boots at home, and many participants informed us that they had never used those boots until they attended our sessions. Most participants said that they would recommend attending the sessions to other diabetic patients.
Sixty-nine participants completed the follow-up at 6 months after the intervention. One participant died of intracerebral hemorrhage 5 months after the intervention.
Proportions of participants reporting that they always wore boots while working in rice fields increased from 30% (10/33) to 74% (32/43) at 1-month post intervention (p<0.001). The proportion was stable at around 75 to 80%, and was at 76% (28/37) six months after the intervention (Table 2).
The proportion of participants reporting that they drank only boiled or bottled water increased from 43% (30/70) to 86% (59/69) at 1-month post intervention (p<0.001). The proportion was stable at around 80 to 85%, and was at 86% (59/69) at 6-month after the intervention.
Our study shows that a multifaceted prevention programme for melioidosis is feasible and acceptable, and can prompt behaviour change in participants. Specifically, the proportion of participants wearing boots while working in rice fields and drinking only boiled or bottle water increased significantly after the intervention. Those increases were sustained for at least six months and are, therefore, likely to lead to lower risk of having melioidosis and other infectious diseases acquired via skin inoculation or ingestion [5]. These positive outcomes could be mainly because the programme was designed systematically based on the identified barriers and enablers, using the TDF and associated BCW [12–14] and taking into account the local context [8].
This may be the first study to show the efficacy of a calendar with an individual photograph as a reminder tool. The individual photograph of the participant wearing boots and holding a kettle could also be categorized as the BCTs “identification of self as role model” and “prompts/cue” [22]. Photography is a very powerful tool to convey a message [23–25]. Because our study could utilize boots and a kettle as part of the reminders, we used individual photographs with those objects. The photograph enables participants to understand the recommended behaviours. The combination of gestures, emotions, attitudes and facial expressions of participants in the photography allows participants to become directly engaged with the intervention. Devising individual photographs into a calendar could enhance the utility of the photographs as participants are likely to look at the calendar frequently, and feel more engaged with their photograph.
The intervention positively affected wearing boots and boiling water before drinking; however, a proportion of participants did not adopt the recommended behaviours. We found that the intervention could not remove all barriers. For example, over-the-knee boots could be used in flooded rice fields without causing difficulty in walking, but were still uncomfortable in hot weather. The BCTs ‘social support’ (including asking nurses, doctors, health volunteers, and families to encourage the person to continue with the recommended behaviours), and the BCT ‘credible source’ (including a high status professional in the government giving a speech emphasizing the importance of melioidosis prevention) in our small study had limited efficacy. This is because, during the follow-up, a number of participants who did not adopt the recommended behaviours did not believe in the ‘information of health consequence’ and noted that if the burden and mortality of melioidosis was real why had they never seen any information or campaign from the government via mass media, particularly on television.
Our study has several strengths. First, we showed that the intervention can lead to adopting recommended preventive behaviours in diabetic patients, who are a key target population for melioidosis prevention in Thailand [6, 7]. Second, the positive effect of the visual tool (a calendar with an individual photograph) to support the behaviour change is an innovation. In this study, the activity of making calendar as a reminder tool implements a number of BCTs; including prompts/cues, credible source, instruction on how to perform a behaviour, demonstration of the behaviour, identification of self as role model, goal setting, commitment and social support (Table 1). Although our study was not designed to evaluate efficacy of each BCT related to this activity, based on our interviews with participants during the home visits, most participants appeared to highly appreciate their own photography on the calendars. Due to strong positive feedback on this component of the intervention, further research should be conducted to evaluate the feasibility and utility of making a calendar with an individual photograph as a reminder tool for other behaviour changes across a range of settings.
The major limitation of this study is that long-term behaviour changes could not be measured and that the follow-ups may be part of the intervention as well as a method of evaluation as they could act as a reminder. Cost-efficacy analysis could not accurately be estimated from this feasibility study but is being evaluated in a subsequent large trial. Also, the programme may not be equally effective for all ages and socioeconomic groupings in the diabetic population in the whole country and beyond. It is possible that some barriers and cultures vary, and that the intervention strategies would need to be adjusted based on local context. Because the reliability of self-report cannot be assumed, we combined it with observation. Our interviews were done together with multiple home visits, during which we observed the boots and kettles that they said they regularly used. It is still possible that some participants may not report accurately, and further evaluating methods such as interviewing relatives and neighbours, and visits to rice fields without notice (but with prior consents from the participants) could be used in the future.
We recommend that health care providers together with policy makers in melioidosis-endemic areas should consider implementating multifaceted interventions for melioidosis prevention. Policy makers, health care providers and researchers should develop a working group to evaluate the feasibility of the interventions, adjust components of the interventions based on their own local context, and gradually implement the interventions. Policy makers should also focus on delivering disease education, particularly through mass media and implementing the multifaceted interventions through healthcare providers. Researchers should also evaluate the efficacy and effectiveness of the interventions which are gradually implemented.
In this study, we evaluated the multifaceted prevention programme of melioidosis and found that the programme is feasible and could lead to adopting recommended preventive behaviours. We strongly suggest that commitment and action by the government are essential for the preventive programmes to occur and be successful. Making calendars with individual photographs and self-pledges as a reminder tool could be powerful in behaviour change interventions, and further research on this component is needed.
|
10.1371/journal.ppat.1004933 | Vibrio cholerae Response Regulator VxrB Controls Colonization and Regulates the Type VI Secretion System | Two-component signal transduction systems (TCS) are used by bacteria to sense and respond to their environment. TCS are typically composed of a sensor histidine kinase (HK) and a response regulator (RR). The Vibrio cholerae genome encodes 52 RR, but the role of these RRs in V. cholerae pathogenesis is largely unknown. To identify RRs that control V. cholerae colonization, in-frame deletions of each RR were generated and the resulting mutants analyzed using an infant mouse intestine colonization assay. We found that 12 of the 52 RR were involved in intestinal colonization. Mutants lacking one previously uncharacterized RR, VCA0566 (renamed VxrB), displayed a significant colonization defect. Further experiments showed that VxrB phosphorylation state on the predicted conserved aspartate contributes to intestine colonization. The VxrB regulon was determined using whole genome expression analysis. It consists of several genes, including those genes that create the type VI secretion system (T6SS). We determined that VxrB is required for T6SS expression using several in vitro assays and bacterial killing assays, and furthermore that the T6SS is required for intestinal colonization. vxrB is encoded in a four gene operon and the other vxr operon members also modulate intestinal colonization. Lastly, though ΔvxrB exhibited a defect in single-strain intestinal colonization, the ΔvxrB strain did not show any in vitro growth defect. Overall, our work revealed that a small set of RRs is required for intestinal colonization and one of these regulators, VxrB affects colonization at least in part through its regulation of T6SS genes.
| Pathogenic bacteria experience varying conditions during infection of human hosts and often use two-component signal transduction systems (TCSs) to monitor their environment. TCS consists of a histidine kinase (HK), which senses environmental signals, and a corresponding response regulator (RR), which mediates a cellular response. The genome of the human pathogen V. cholerae contains a multitude of genes encoding HKs and RRs proteins. In the present study, we systematically analyzed the role of each V. cholerae RR for its role in pathogenesis. We identified a previously uncharacterized RR, VxrB, as a new virulence factor. We demonstrated that VxrB controls expression of the type VI secretion system (T6SS), a virulence nanomachine that directly translocates effectors into bacterial or host cells, thereby facilitating colonization by competing with sister cells and intestinal microbiota. This study represents the first systematic analysis of the role of all RRs in V. cholerae pathogenesis and provides a foundation for understanding the signal transduction pathways controlling V. cholerae pathogenesis.
| Vibrio cholerae causes the diarrheal disease cholera that affects 3 to 5 million people worldwide every year, resulting in 100,000–120,000 deaths annually [1]. V. cholerae produces a number of virulence factors which facilitate colonization of the intestine and subsequent disease. Major virulence factors are cholera toxin (CT), which is responsible for production of profuse watery diarrhea, and a type IV pilus called the toxin-coregulated pilus (TCP), which is required for intestinal colonization [2]. V. cholerae virulence factors are well known to be under extensive transcriptional control. CT and TCP production are controlled by the transcriptional activator ToxT [3, 4]. Expression of toxT, in turn, is controlled by a virulence regulatory cascade involving the membrane-bound transcriptional activators ToxRS and TcpPH. These two regulators activate toxT transcription directly [5–7]. TcpPH expression is activated by the transcriptional activators AphA and AphB [8, 9]. The quorum sensing (QS) regulatory system is also linked to the virulence gene regulatory cascade through HapR, the master QS regulator, which represses aphA expression [10].
Recently, the type VI secretion system (T6SS) has been identified as a new virulence factor in V. cholerae [11, 12]. T6SSs deliver effector proteins into both eukaryotic and bacterial cells in a contact-dependent manner [12, 13]. V. cholerae has one T6SS system with multiple T6SS effectors: VrgG1 and VrgG3 (valine-glycine repeat protein G), which have actin cross-linking activity and peptidoglycan-degrading activity, respectively [14–17]; TseL, which has lipase activity [15, 18]; and VasX, which perturbs the cytoplasmic membrane of target cells [15, 19]. Activity of these effectors is antagonized by corresponding immunity proteins: TsiV3, TsiV1, and TsiV2, respectively, to prevent killing by strains bearing these proteins [15, 16, 20, 21].
The T6SS can be divided into functional sections consisting of the core structural components, the T6SS effector and immunity proteins, and transcriptional regulators. The base of the T6SS apparatus spans the cell envelope, and is a tube within a tube. The inner tube is composed of polymers of the hemolysin coregulated protein (Hcp). The outer tube, also called the contractile sheath, is formed by polymers of VipA and VipB [14, 22]. The Hcp inner tube is capped with a spike complex of trimeric VgrG proteins. The effectors are delivered by contraction of the VipA/VipB sheath, which in turn results in ejection of the inner tube along with VgrG and the effectors towards the target cell [12].
The genes encoding the T6SS components are organized into one large cluster (VCA0105-VCA0124) and two auxiliary clusters (VCA0017-VCA0022 and VC1415-VC1421) [11, 23]. A key positive transcriptional regulator of the V. cholerae T6SS is VasH (VCA0117), which is related to enhancer binding proteins that activate transcription in a σ54 (RpoN) dependent manner [24, 25]. VasH acts on the T6SS auxiliary clusters and vgrG3 of the large cluster, but does not affect expression of the structural genes encoded in the large T6SS gene cluster [24, 26]. Additionally, Hcp production is positively regulated by the master quorum sensing regulator HapR and the global regulator cyclic AMP (cAMP) receptor protein CRP, and negatively regulated by QS regulator LuxO and by global regulator TsrA, a protein homologous to heat-stable nucleoid-structuring (H-NS) [27, 28]. These studies have thus shown that numerous global regulators control T6SS expression, as well as one specific regulator (VasH).
V. cholerae T6SS studies have mainly focused on the V. cholerae O37 serogroup V52 strain because it assembles a T6SS apparatus constitutively [11]. In this strain, the T6SS is required for cytotoxicity towards Dictyostelium discoideum and J774 macrophages, and induces inflammatory diarrhea in the mouse model [29]. In V. cholerae O1 strain C6706, the T6SS is not constitutively produced and conditions that promote T6SS production are unknown. However, production of T6SS can be achieved in other O1 strains by inactivating mutations in genes encoding the LuxO and TsrA negative regulators. In O1 strains, the T6SS translocates T6SS effectors into macrophages, and increases fecal diarrhea and intestinal inflammation in infant rabbits [27]. It was also shown that the V. cholerae O1 C6706 strain T6SS mediates antagonistic interbacterial interactions during intestinal colonization. A strain unable to produce the TsiV3 immunity protein, which provides immunity against the effector VgrG3, exhibited an intestinal colonization defect only when co-infected with strains harboring an intact T6SS locus and VrgG3 [30]. Although T6SS is regulated and expressed differently between V. cholerae strains, production of this system in multiple strains promotes virulence against both eukaryotic and bacterial cells, suggesting the function is largely conserved but the regulation varies.
Pathogenic bacteria experience varying conditions during infection of human hosts and often use two-component signal transduction systems (TCSs) to monitor their environments [31]. TCSs play important roles in the regulation of virulence factors, metabolic adaptation to host environments, and response to numerous environmental stresses including pH, osmolarity, oxygen availability, bile salts, and antimicrobial peptides [32]. TCS rely on a phosphorelay-based signal transduction system. The prototypical TCS consists of a membrane-bound histidine kinase (HK), which senses environmental signals, and a corresponding response regulator (RR), which mediates a cellular response. Response regulators are typically multi-domain proteins harboring a conserved receiver domain (REC) and C-terminal output domain such as DNA-binding, diguanylate cyclase, or methyltransferase [33–35]. Upon environmental stimulation, the HK catalyzes an ATP-dependent autophosphorylation reaction on a conserved histidine residue. The phosphoryl group is transferred from the HK to a conserved aspartate residue on the RR, eliciting a conformation change and subsequent cellular response [32, 34, 35].
The V. cholerae genome reference genome of O1 EL Tor N16961 strain is predicted to encode 43 HK and 49 RR (http://www.ncbi.nlm.nih.gov/Complete_Genomes/RRcensus.html and http://www.p2cs.org). We also included 3 additional RRs (VpsT, VpsR, QstR) which were not annotated in these databases. Thirteen of these 52 putative RRs have been previously characterized and eight have a role in virulence factor production and host colonization: VarA, LuxO, VieA, PhoB, ArcA, FlrC, CarR, and CheY-3 [36–44]. VarA and LuxO repress production of quorum sensing regulator HapR, which represses expression of aphA and, in turn, TCP and CT production [36, 37], VieA regulates ctxAB expression indirectly by affecting production of ToxT through cyclic diguanylate (c-di-GMP) signaling [38, 39]. The RR for phosphate limitation, PhoB, directly controls expression of a key transcriptional regulator, TcpPH, which activates toxT transcription [40]. The RR ArcA controls adaptation to low oxygen environment of the intestine and positively controls the expression of toxT [41]. CarR regulates glycine and diglycine modification of lipid A, confers polymyxin B resistance, and is required for intestinal colonization, although this phenotype is strain dependent [42]. FlrC controls flagellar biosynthesis and CheY-3 is needed for control of chemotactic motility [43, 44]. Both motility and chemotaxis are known colonization factors for V. cholerae [43]. Together, these results show that RRs shown play a role in intestinal colonization have three basic targets: known virulence regulators and concomitant CT and TCP production; lipid A modification enzymes; or motility and chemotaxis. 39/52, however, were not yet analyzed at the time of this study.
To systematically evaluate the role of V. cholerae TCSs in intestinal colonization, we generated in-frame deletion mutants of each RR gene and analyzed the in vivo colonization phenotypes of the resulting mutants. We found 12 RR were required for wild-type intestinal colonization. One RR in particular had a very strong defect, encoded by genomic locus VCA0566. We determined that VCA0566 (now termed Vibrio type six secretion regulator, vxrB) controls expression of several genes including the T6SS genes. We used multiple methods to substantiate that VxrB is required for expression of the T6SS in vitro and in vivo. Lastly, we report that the T6SS contributes to colonization of the V. cholerae O1 strain used in this study.
We have a limited understanding of the V. cholerae TCSs and their role in colonization and adaptation to host environments. To evaluate the importance of the 52 TCS RRs in colonization, we generated in-frame deletion mutants of the 40 RRs. For this analysis, we excluded 12 RR that were either predicted to be involved in chemotaxis (11 CheY, CheV, and CheB proteins) or that we were unable to mutate (VC2368, ArcA) [43, 45]. We then analyzed the ability of 40 RR deletion mutants to colonize the small intestine in an in vivo competition assay where in vivo fitness of a mutant strain is compared to that of wild type strain using the infant mouse infection model (Fig 1A) [46]. While the vast majority of mutants—28—were not different from wild type, we identified 12 RR mutants that had a statistically significant colonization difference as compared to wild type (Fig 1A). We focused on 8 mutants with a statistically significant colonization difference and exhibited at least 1.2-fold difference in CI (Fig 1B). Consistent with previous studies, we identified that ΔVC0719 (phoB), ΔVC1021 (luxO), ΔVC1213 (varA), and ΔVC2135 (flrC) were defective in colonization [36, 37, 40, 44]. The competitive indices (CI) for ΔphoB, ΔluxO, ΔvarA, and ΔflrC were 0.01, 0.02, 0.16, and 0.43, respectively (Fig 1B).
Additionally, we identified a set of genes whose absence slightly but statistically significantly enhanced colonization (at least 1.2 fold higher CI), suggesting that inhibition of their expression and activity may be needed for wild-type colonization. These mutants were ΔVC1050, ΔVC1086, and ΔVC1087, which exhibited subtle and enhanced colonization phenotypes with CIs of 1.43, 1.24, and 1.48, respectively (Fig 1B). VC1050 is classified as an Hnr-type RR, [47] but its function is yet to be determined. VC1086 and VC1087 are part of a predicted eight gene operon encompassing VC1080-VC1087. Both VC1086 and VC1087 have domains that suggest they function in cyclic guanylate (c-di-GMP) regulation. Specifically, VC1086 contains an EAL domain with conserved residues required for enzymatic function, while VC1087 harbors an HD-GYP domain, but this domain lacks the conserved residues required for enzymatic activity.
We also identified one RR that was defective for colonization that had not been previously characterized. This mutant, ΔVCA0566, had a colonization defect with a CI of 0.14 (Fig 1A and 1B). Because this uncharacterized RR was important for colonization, we focused the rest of our studies on this protein.
VCA0566 is the second gene of a predicted five gene operon and had been previously annotated as a RR of the OmpR family. The encoded protein, which we named VxrB for reasons described below, is 245 amino acids in length with an N-terminal REC domain and a C-terminal winged helix-turn-helix DNA-binding domain (Fig 2B). Previously characterized members of the OmpR family in V. cholerae include PhoB, CarR, and ArcA [40–42]. Amino acid sequence alignment of the V. cholerae RRs in the OmpR family and the previously characterized E. coli OmpR [48] was used to identify the aspartate residue that is predicted to be phosphorylated in the REC domain (Fig 2A). Since the phosphorylation state of a RR is likely to determine its activity, we mutated the aspartate residue in the REC domain of VxrB to mimic constitutively active (D78E) and inactive (D78A) versions, as used in other work [48], and replaced the wild-type gene in the chromosome with these altered genes. These mutants were competed against wild type in the infant mouse colonization assay to determine if the phosphorylation state of VxrB is important for colonization. In accordance with our initial colonization screen, ΔvxrB displayed a CI of 0.15 (Fig 2B). Somewhat surprisingly, the CI for vxrB::D78A (inactive form) was 0.53, indicating a modest defect in colonization. This result indicates that the “inactive” form of VxrB does not phenocopy the ΔvxrB mutant, suggesting that VxrB harboring D78A substitution is not fully inactive. The CI for vxrB::D78E (active form) is 1.07, suggesting that constitutive activation of VxrB does not significantly impair V. cholerae (Fig 2B). Collectively, these findings suggest that in vivo phosphorylation of VxrB at D78 is likely to be important for its colonization function, but apparently not absolutely required. It is also likely that VxrB may not function by conventional phosphorylation-dependent signal transduction [49].
The first gene of the vxr loci, VCA0565, is annotated as an HK. The other three genes (VCA0567-69) are predicted to encode proteins of unknown function (Fig 3A). We now termed these genes Vibrio type six secretion regulator (vxr) ABCDE and determined that these genes are co-transcribed using RT-PCR and RNAseq analysis (Fig 3A and S1 Fig). Both the genomic context and organization is conserved in the Vibrio species (S2–S4 Figs) and vxr gene products do not share significant sequence similarity with previously characterized proteins.
To gain a better understanding of the role of the vxr operon in colonization, we investigated whether the cognate HK and other genes in the vxr operon also contributed to mouse colonization. In-frame unmarked deletion mutants of vxrA, vxrB, vxrC, vxrD, and vxrE (Fig 3B) were generated and analyzed in an in vivo competition assay. Each mutant was outcompeted, with CIs of 0.35, 0.16, 0.44, 0.66, and 0.70, respectively (Fig 3B). These findings suggest that while vxrA and vxrB genes are critical for colonization in the infant mouse model, contribution of vxrCDE genes appears to be minor.
To further confirm the phenotype of ΔvxrB colonization defect, a wild type copy of vxrB whose expression was driven from its native promoter was inserted into the Tn7 site (located between VC0487 and VC0488) on the chromosome of ΔvxrB. In vivo competition assay of ΔvxrB-Tn7vxrB had a CI of 0.93, similar to wild type levels, where ΔvxrB had a CI of 0.16 (Fig 3B). Thus, the ΔvxrB colonization defect is restored to wild-type levels by introduction of the wild-type copy of vxrB.
To gain a better understanding of the contribution of VxrB to V. cholerae pathogenesis, we performed high throughput transcriptome sequencing (RNA-seq) analysis to identify the V. cholerae genes controlled by VxrB. We used cells grown under virulence inducing AKI conditions, to mimic the intestinal conditions encountered when we know VxrB is important. 149 genes showed statistically significant differences in gene expression between the wild type and mutant (S2 and S3 Tables). Of these, 80 genes were expressed to greater levels in the ΔvxrB mutant relative to the wild type (S2 Table), while 69 were expressed to lower levels in the ΔvxrB mutant relative to wild type (S3 Table). Of particular interest was the observation that message abundance of most of the T6SS genes in both the large cluster (VCA0105-VCA0124) and the two auxiliary clusters (VCA0017-VCA0022 and VC1415-VC1421) were less in the VxrB mutant relative to wild type (Table 1) (S5 Fig). This finding suggests that VxrB activates expression of the T6SS genes.
To further analyze the role of vxrB in T6SS expression and function, we compared the levels of the major T6SS structural component, Hcp, between wild type and ΔvxrB mutant V. cholerae. Quantitative real-time PCR analysis of hcp revealed that the transcript abundance of hcp was decreased by 3.7-fold under AKI conditions and 4.1- fold under LB conditions in the ΔvxrB mutant relative to wild type (Fig 4A). This finding supports that VxrB regulates expression of hcp and is consistent with the RNA-seq analysis. Additionally the levels of the Hcp protein in ΔvxrB were lower than wild type, in both whole cell samples and culture supernatants (Fig 4B). We also determined that complementation of the vxrB mutation (ΔvxrB-Tn7vxrB) restored Hcp to wild-type levels. Because we found lower amounts of Hcp in the supernatant as well as in whole cells, this finding suggests that VxrB is needed to express and secrete Hcp. As negative controls, we included a Δhcp1Δhcp2 mutant because it is unable to produce the Hcp proteins [11, 50]. As expected, no Hcp production was observed in this mutant. Furthermore, complementation of hcp1 in the Δhcp1Δhcp2 mutant partially restored Hcp levels (Fig 4B). Overall these findings suggest that Hcp production is decreased in ΔvxrB mutant.
Next we analyzed whether VxrB was needed for T6SS function, by examining T6SS-mediated interbacterial killing. Killing assays between the V. cholerae and the target E. coli K-12 strain MC4100 showed that wild-type V. cholerae decreased the numbers of E. coli compared to control experiments. This killing was dependent on the T6SS, as shown by greater numbers of E. coli obtained when incubated with V. cholerae Δhcp1Δhcp2 mutant and ΔvasH mutants, consistent with the findings reported by Ishikawa et al. (Fig 4C) [50]. This phenotype was complemented by introduction of either hcp1 or hcp2 into the Tn7 site on the chromosome. Consistent with our transcriptional and protein analysis presented above, we found that ΔvxrB mutants mediated less E. coli killing. These findings suggest that T6SS regulation by VxrB contributes to interbacterial killing.
Since VxrB regulates T6SS expression and is required to for intestinal colonization, we next asked whether the T6SS itself is required for intestinal colonization. We performed in vivo competition assays of a T6SS null mutant (Δhcp1Δhcp2) against wild type in the infant mouse model. We found that the in-vivo CI for Δhcp1Δhcp2 was 0.17 (Fig 5A). In addition, ΔvgrG3 also had an in-vivo CI of 0.26 suggesting that T6SS components are important for intestinal colonization (Fig 5A). This suggests that structural components of the T6SS are needed to colonize the intestine. Furthermore, this finding also suggests that the colonization defect associated with the ΔvxrB mutant could be caused by diminished T6SS production. To evaluate this possibility, we tested the in vivo competition of Δhcp1Δhcp2 against ΔvxrB and found that these strains competed nearly equally with each other (Fig 5C). Furthermore, in-vivo CI of ΔvxrBΔhcp1Δhcp2 triple mutant against ΔvxrB was 0.07 and ΔvxrBΔhcp1Δhcp2 against wt was 0.10. This finding suggests that the colonization defect by ΔvxrB was not solely due to altered expression of T6SS genes and other factors regulated by VxrB also contribute to colonization. It is also likely that T6SS expression is not completely abolished by the vxrB mutation. Indeed, western analysis (Fig 4B) shows that in vxrB mutant Hcp production is reduced but not completely eliminated. Similarly in vitro killing assay shows that vxrB mutant’s interbacterial killing ability is not identical to that of the strain lacking T6SS.
We next asked whether VxrB plays a role in growth in vitro, by performing an in vitro competition assay. ΔvxrB mutants grew equally well as wild type, suggesting that neither had a competitive advantage over the other in vitro (Fig 5B). This outcome suggests that there may be an in vivo signal produced in the infant mouse that triggers T6SS activity and colonization. We also performed single-strain colonization assays in the infant mouse model with ΔvxrB. There was a 12.7-fold decrease in colonization for ΔvxrB compared to wild type (Fig 5D). This finding suggests that the colonization defect by ΔvxrB was not solely dependent on wild type, and possibly could be caused by competition with the normal flora or ability of the mutant to adapt to the infection microenvironment.
Systematic mutational phenotypic characterization of TCSs has been performed in only a few bacteria, including Vibrio fischeri, E. coli, Bacillus subtilis, Streptococcus pneumoniae, and Enterococcus faecalis [51–55]. In this study, we systematically analyzed the role of all V. cholerae TCS in colonization of the infant mouse small intestine and identified the RRs that play roles in mouse intestinal colonization. Specifically, ΔVC0719 (phoB), ΔVC1021 (luxO), ΔVC1213 (varA), and ΔVC2135 (flrC), and ΔVCA0566 (vxrB) exhibited intestinal colonization defects while ΔVC1050, ΔVC1086, and ΔVC1087 showed enhanced colonization. Many of the RRs had either no statistically significant defect or minor defects in the infant mouse colonization assay. It remains possible, however, that these RRs have a role in colonization in other infection models.
In vivo transcriptome analysis has been performed on different strains of V. cholerae in the infant mouse and rabbit ileal loop infection models. The analysis of the whole genome expression of V. cholerae O1 El Tor C6706 cells accumulating in the ceca of orally infected infant rabbits and the intestines of orally infected infant mice revealed that expression of the genes encoding RRs is altered during in vivo growth conditions as compared to in vitro growth in nutrient broth and that in vivo expression of TCS also differed between the model systems [56]. In the infant rabbit infection model, expression of seven RR (VC1081, VC1082, VC1155, vieA, VC2702 (cbrR), VCA0210, and VCA1105) was increased and 1 RR (carR) was decreased by more than 2-fold significantly in comparison to V. cholerae cells grown in vitro in nutrient broth. In the infant mouse infection model, expression of 17 RR (vpsR, VC1050, VC1081, VC1082, VC1086, VC1087, VC1155, VC1522, flrC, cbrR, ompR, dct-D2, vxrB, uhpA, vpsT, VCA1086, and VCA1105) and 9 RR (qstR, phoB, VC1348, VC1638, vieB, cpxR, ntrC, VCA0532, pgtA) were either decreased and increased significantly by more than 2-fold, respectively, in comparison to V. cholerae cells grown in vitro in nutrient broth [56]. vxrA and vxrB transcript levels were decreased 2 and 3-fold, respectively, in the experiments reported by Mandlik and colleagues, but did not reach statistical significance [56]. This work all used the V. cholerae O1 El Tor strain C6706, and so it is yet unknown whether vxrB expression is similarly regulated in the O1 El Tor A1552 strain used here.
There have been two other studies that analyzed V. cholerae infection phenotypes on a global scale, although they did not specifically target RR. Together, these studies and ours suggests there is a set of genes required for intestinal colonization across multiple models. Fu et al. used random transposon mutants coupled with insertion site sequencing (Tn-seq) in a rabbit model [30]. They identified insertions in two genes—VC1021 (luxO) and VC1155—that showed 8-15-fold reduction in colonization, while strains harboring insertions into RRs VC1348, vieA, vieB, arcA, VCA0256, uhpA, and pgtA had less than a 5-fold reduction in colonization (p<0.001). Another Tn-seq study using the infant rabbit model identified defects associated with luxO and arcA as above, and additionally phoB and varA [57]. Combining the results of these studies with ours identifies luxO, phoB, and varA, as required for in vivo fitness, and others that are variably identified. Because the Tn-seq work used transposon libraries, it is not known whether all RR were eliminated, so it is possible that their studies missed some critical RR. There are hints in their data, however, that the vxr locus is necessary in these other models as well. While Fu et al. did not identify vxrA or vxrB mutants, they did determine that a strain with an insertion into VCA0567 (vxrC) exhibited a 9-fold reduction in colonization (p<0.0001) [30]. Additionally, Kamp et al. found that a strain with a transposon insertion in VCA0565 (vxrA) had a disadvantage in fitness (mean fitness value of 0.6) when the bacteria from rabbit cecum fluid was placed into pond water for 48 hours at 30°C [57]. Collectively, these studies suggest that the Vxr genes play important roles in V. cholerae colonization and environmental dissemination.
Our study revealed that the RR VxrB plays a significant role in colonization and in vitro inter-bacterial competition through its ability to regulate expression of T6SS genes. Neither vxrB nor any of the vxrABCDE operon members show similarity to previously characterized proteins. The vxr loci are conserved among the Vibrio species Vibrio parahaemolyticus, Vibrio vulnificus, Vibrio harveyi, and V. fischeri. BLAST analysis revealed that VxrA protein exhibits 67–80%, VxrB 79–84%, VxrC 56–68%, VxrD 58–74%, and VxrE 68–81% sequence similarity to the same proteins in other Vibrio species (S2–S4 Figs). We also analyzed the predicted structure and function of the VxrCDE proteins using the protein homology/analogy recognition engine (Phyre) [58]. While VxrC and VxrE could not be modeled with high confidence and sufficient coverage, VxrD exhibited structural similarity to outer membrane protein transport proteins (100% confidence, 90% coverage). These analyses suggest that vxr genomic loci are a part of the ancestral Vibrio genome, and therefore likely have an evolutionarily conserved role in Vibrio biology.
Expression and production of T6SS are tightly regulated at the transcriptional and posttranscriptional levels in a variety of bacterial systems [12, 13, 59]. Environmental signals such as iron limitation, thermoregulation, salinity, envelope stress, indole, and growth on surfaces regulate T6SS expression [59]. In V. cholerae A1552, the strain used here, T6SS genes are expressed when cell are grown in high-osmolarity and low temperature conditions [50]. A recent study revealed that the V. cholerae A1552 T6SS genes are part of the competence regulon and their expression is induced when the bacterium grows on chitinous surfaces in a TfoX-, HapR-, and QstR-dependent manner [60]. Our work presented here identified VxrB as a regulator of the T6SS large gene cluster and the two auxiliary clusters. The predicted cognate HK of VxrB, VxrA, does not exhibit similarity to previously characterized sensory domains. The signals that govern expression and activity of VxrAB and how the VxrAB TCS is integrated into the T6SS regulatory network of V. cholerae are yet to be determined. We determined that while the wild-type strain has a competitive advantage in vivo over ΔvxrB, neither strain had a competitive advantage over the other in vitro. Furthermore, single infection studies showed that ΔvxrB had a significant colonization defect compared to wild type, suggesting that VxrB could be involved in competition with normal flora and that ΔvxrB could have a reduced fitness in infection environment. These observations also suggest that there may be an in vivo signal produced in the infant mouse that triggers T6SS activity and colonization. Our studies thus provided significant new insights into the regulation of T6SS in V. cholerae and provided further support that the T6SS is critical for V. cholerae virulence.
All animal procedures used were in strict accordance with the NIH Guide for the Care and Use of Laboratory Animals [61] and were approved by the UC Santa Cruz Institutional Animal Care and Use Committee (Yildf1206).
The bacterial strains and plasmids used in this study are listed in S1 Table. Escherichia coli CC118λpir strains were used for DNA manipulation, and E. coli S17-1λpir strains were used for conjugation with V. cholerae. In-frame deletion mutants of V. cholerae were generated as described earlier [62]. All V. cholerae and E. coli strains were grown aerobically, at 30°C and 37°C, respectively, unless otherwise noted. All cultures were grown in Luria-Bertani (LB) broth (1% Tryptone, 0.5% Yeast Extract, 1% NaCl), pH 7.5, unless otherwise stated. LB agar medium contains 1.5% (wt/vol) granulated agar (BD Difco, Franklin Lakes, NJ). AKI medium contains 0.5% NaCl, 0.3% NaHCO3, 0.4% Yeast Extract, and 1.5% Peptone, as previously described [63]. Antibiotics were used at the following concentrations: ampicillin 100 μg/ml; rifampicin 100 μg/ml; gentamicin 50 μg/ml; streptomycin 50 μg/ml.
An overlapping PCR method was used to generate in-frame deletion constructs of each RR genes using previously published methods [62]. Briefly, a 500–600 bp 5’ flanking sequence of the gene, including several nucleotides of the coding region, was PCR amplified using del-A and del-B primers. del-C and del-D primers were used to amplify the 3’ region of the gene including 500–600 bp of the downstream flanking sequence. The two PCR products were joined using the splicing overlap extension technique [64, 65] and the resulting PCR product, which lacks 80% of amino acids, was digested with two restriction enzymes and ligated to similarly-digested pGP704sacB28 suicide plasmid. Construction of vxrB plasmid harboring point mutations were performed using a similar technique [66] with the following alterations: primers containing the new sequence harboring the point mutations were used in place of the del-B and del-C primers. The deletion constructs were sequenced (UC Berkeley DNA Sequencing Facility, Berkeley, CA) and the clones without any undesired mutations were used. The deletion constructs are listed in S1 Table.
The deletion plasmids were maintained in E. coli CC118λpir. Biparental matings were carried out with the wild type V. cholerae and an E. coli S17λpir strain harboring the deletion plasmid. Selection of deletion mutants were done as described [64] and were verified by PCR. The Tn7 complementation V. cholerae strains were generated by triparental matings with donor E. coli S17λpir carrying pGP704-Tn7 with gene of interest, helper E. coli S17λpir harboring pUX-BF13, and V. cholerae strains. Transconjugants were selected on thiosulfate-citrate-bile salts-sucrose (TCBS) (BD Difco, Franklin Lakes, NJ) agar medium containing gentamicin at 30°C. The Tn7 complementation V. cholerae strains were verified by PCR.
An in vivo competition assay for intestinal colonization was performed as described previously [46]. Briefly, each of the V. cholerae mutant strains (lacZ+) and the fully virulent reference strain (lacZ-otherwise wild-type)) were grown to stationary phase at 30°C with aeration in LB broth. Mutant strains and wild-type were mixed at 1:1 ratios in 1x Phosphate Buffered Saline (PBS). The inoculum was plated on LB agar plates containing 5-bromo-4-chloro-3-indoyl-β-D-galactopyranoside (X-gal) to differentiate wild-type and mutant colonies and to determine the input ratios. Approximately, 106–107 cfu were intragastrically administered to groups of 5–7 anesthetized 5-day old CD-1 mice (Charles River Laboratories, Hollister, CA). After 20 hours of incubation, the small intestine was removed, weighed, homogenized, and plated on appropriate selective and differential media to enumerate mutant and wild-type cells recovered and to obtain the output ratios. In vivo competitive indices were calculated by dividing the small intestine output ratio by the inoculum input ratio of mutant to wild-type strains. For single strain infections, 107 cfu of each strain, including otherwise wild type (lacZ-) strain, were intragastrically administrated to 5-day old CD-1 mice. After 20 hours of incubation, the small intestine was harvested and plated on selective media as previously described above. Statistical analyses for competition infections were performed using Wilcoxon Signed Rank Test. Statistical analyses were performed using Prism 5 software (GraphPad Software, Inc., San Diego, CA) using Wilcoxon Signed Rank Test. P values of <0.05 were determined to be statistically significant.
RNA was isolated as described below. The reverse transcription reaction to generate cDNA was carried out using the SuperScript III Reverse Transcriptase (Invitrogen, Carlsbad, CA) according to the manufacturer’s instructions at 25°C for 5 min, 50°C for 1 h, and 70°C for 15 min using 1 μg of RNA in a 20 μl final volume. The product was used in a PCR using suitable primers, and RNA without RT treatment was used as a negative control.
For qRT-PCR expression analysis, RNA was isolated as described below. cDNA was synthesized using iScript cDNG Synthesis Kit (Bio-Rad, Hercules, CA) from 1 μg of total RNA. Real-time PCR was performed using a Bio-Rad CFX1000 thermal cycler and Bio-RAD CFX96 real-time imager with specific primer pairs (designed within the coding region of the target gene) and SsoAdvanced SYBR green supermix (Bio-Rad, Hercules, CA). Results are from two independent experiments performed in quadruplicate. All samples were normalized to the expression of the housekeeping gene 16S using the Pfaffl method [67]. Relative expression was calculated by normalizing expression at ΔvxrB by that of wt. Statistical analysis was performed using two-tailed student’s t test.
V. cholerae cells were grown aerobically overnight in LB at 37°C, then diluted 1:100 in fresh 10 ml AKI media in borosilicate glass test tubes (diameter, 15mm; height, 150 mm) and incubated at 37°C without shaking for 4 hours. After 4 hours, 10 ml cultures were transferred to 125 ml flasks (for maximal aerated growth on an orbital shaker (250 rpm) for 2 hours. Aliquots (2 ml) of the cultures were collected and centrifuged for 2 min at room temperature. The cell pellets were immediately resuspended in 1 ml of TRIzol (Invitrogen, Carlsbad, CA) and stored at -80°C. Total RNA was isolated according to the manufacturer’s instructions. To remove contaminating DNA, total RNA was incubated with RNase-free DNase I (Ambion, Grand Island, NY), and an RNeasy mini kit (Qiagen, Valencia, CA) was used to clean up RNA after DNase digestion. Five micrograms of total RNA was treated with a MICROBExpress Kit (Ambion, Grand Island, NY) to remove ribosomal RNA, and the efficiency was confirmed by Bioanalyzer analysis (Agilent Technologies, Santa Clara, CA). Three biological replicates were generated for each condition.
Libraries for RNA-seq were prepared using NEBNext Ultra Directional RNA Library Prep Kit for Illumina (New England Biolabs, Ipswich, MA). Twelve indexed samples were sequenced per single lane using the HiSeq2500 Illumina sequencing platform for 50 bp single reads (UC Davis Genome Center, UC Davis, CA) and subsequently analyzed and visualized via the CLC Genomics Workbench version 7.5 (Qiagen, Valencia, CA). Samples were mapped to the V. cholerae genome N16961. Differentially regulated genes were identified as those displaying a fold change with an absolute value of 1.5-fold or greater. Statistical significance was determined by Empirical analysis of Digital Gene Expression (edgeR) test where p<0.05 was deemed significant [68].
V. cholerae strains were grown to an OD600 of 2.0, and the culture (25 ml) was centrifuged at 20,000 g for 10 min to obtain whole cell pellets. The culture supernatant containing secreted proteins were filtered through 0.22 μ membranes (Millipore, Billerica, Massachusetts) and secreted proteins in the culture supernatant were precipitated with 13% trichloroacetic acid (TCA) overnight at 4°C, pelleted by centrifugation at 47,000 g for 30 min at 4°C, wash with ice cold acetone and resuspended in 1x PBS containing Complete protease inhibitor (Roche, Basel, Switzerland). Bovine serum albumin (BSA, 100 μg/ml) was added to the culture supernatant prior to TCA precipitation as a control. Protein pellets from whole cell were suspended in 2% sodium dodecyl sulfate (SDS) and protein concentrations were estimated using a Pierce BCA protein assay kit (Thermo Scientific, Rockford, IL). Equal amounts of total protein (20 μg) were loaded onto a SDS 13% polyacrylamide gel electrophoresis (SDS-PAGE). Western blot analyses were performed as described [69] using anti-Hcp polyclonal antiserum provided by the Sun Wai [28], anti-CRP (Neoclone Inc., Madison, WI), and anti-BSA (Santa Cruz Biotech, Santa Cruz, CA). OneMinute Western Blot Stripping Buffer (GM Biosciences, Frederick, MD) was used to remove the Hcp antibodies and the same blot was used again to probe for CRP or BSA. These experiments were conducted with at least three biological replicates.
Killing assays were performed as described previously [20]. Briefly, bacterial strains were grown overnight on LB plates and resuspended in LB broth containing 340 mM NaCl, as V. cholerae strain A1552 displayed enhanced interbacterial virulence towards E. coli under high osmolarity [50]. V. cholerae and E. coli MC4100 were mixed at a 10:1 ratio and 25 μl was spotted onto LB agar plates containing 340 mM NaCl and incubated at 37°C for 4 hours. Spots were harvested, serially diluted, and plated onto LB plates containing 50 μg/ml of streptomycin to enumerate surviving E. coli prey cells.
The following assay was performed similarly as the intestinal colonization assay except no animal models were used. The V. cholerae mutant strains with wild-type lacZ allele (lacZ+) and reference strain (lacZ-) were grown to stationary phase at 30°C with aeration in LB broth. Mutant strains and wild-type were mixed at 1:1 ratios in 1x PBS. The inoculum was plated on LB agar plates containing X-gal to differentiate colonies formed by the wild-type and mutant strains and to determine the input ratios. The inoculum (50 μl) was spotted on to a LB agar plate and incubated at 37°C. After 20–24 hours of incubation, the 50 μl spots were scraped off the agar plate and resuspended in 1x PBS. The resuspension was serially diluted and plated on appropriate selective and differential media to enumerate mutant and wild type cells recovered and to obtain the output ratios. In vitro competitive indices were calculated by dividing the output ratio by the inoculum input ratio of mutant to wild type strains. Statistical analyses were performed using Wilcoxon Signed Rank Test.
|
10.1371/journal.pntd.0002292 | Evaluation of an Electricity-free, Culture-based Approach for Detecting Typhoidal Salmonella Bacteremia during Enteric Fever in a High Burden, Resource-limited Setting | In many rural areas at risk for enteric fever, there are few data on Salmonella enterica serotypes Typhi (S. Typhi) and Paratyphi (S. Paratyphi) incidence, due to limited laboratory capacity for microbiologic culture. Here, we describe an approach that permits recovery of the causative agents of enteric fever in such settings. This approach involves the use of an electricity-free incubator based upon use of phase-change materials. We compared this against conventional blood culture for detection of typhoidal Salmonella.
Three hundred and four patients with undifferentiated fever attending the outpatient and emergency departments of a public hospital in the Kathmandu Valley of Nepal were recruited. Conventional blood culture was compared against an electricity-free culture approach. Blood from 66 (21.7%) patients tested positive for a Gram-negative bacterium by at least one of the two methods. Sixty-five (21.4%) patients tested blood culture positive for S. Typhi (30; 9.9%) or S. Paratyphi A (35; 11.5%). From the 65 individuals with culture-confirmed enteric fever, 55 (84.6%) were identified by the conventional blood culture and 60 (92.3%) were identified by the experimental method. Median time-to-positivity was 2 days for both procedures. The experimental approach was falsely positive due to probable skin contaminants in 2 of 239 individuals (0.8%). The percentages of positive and negative agreement for diagnosis of enteric fever were 90.9% (95% CI: 80.0%–97.0%) and 96.0% (92.7%–98.1%), respectively. After initial incubation, Salmonella isolates could be readily recovered from blood culture bottles maintained at room temperature for six months.
A simple culture approach based upon a phase-change incubator can be used to isolate agents of enteric fever. This approach could be used as a surveillance tool to assess incidence and drug resistance of the etiologic agents of enteric fever in settings without reliable local access to electricity or local diagnostic microbiology laboratories.
| Every year, 20 million people worldwide suffer from typhoid, a bacterial infection spread by contaminated food and water, and over 200,000 die from the infection. However, few data are available on the prevalence and antimicrobial resistance profiles of the causative agents of typhoid, especially in settings without reliable access to laboratories and electricity. Here, we describe an approach that permits recovery of the causative agents of typhoid that requires no electricity, laboratory infrastructure, or specialized laboratory personnel at the site of patient contact. This approach involves the use of an electricity-free incubator, consisting of an insulated container and reusable packets filled with a chemical that upon warming in hot water or direct sun, maintains 38°C for 24 hours. We used blood culture bottles with a color indicator that signals growth of bacteria and an antibiotic that selects for Gram-negative bacteria. We validated this approach in a clinical study among individuals with fever presenting to a hospital in urban Nepal, demonstrating that the approach performed comparably to conventional blood culture for isolating the bacteria causing typhoid. Such an approach would permit an estimate of the burden of typhoid in areas where such data are lacking, which could inform control strategies.
| Enteric fever is a febrile illness caused by Salmonella enterica serotype Typhi (S. Typhi) or Paratyphi (S. Paratyphi) A, B or C. There are an estimated 21.6 million new infections with S. Typhi and 5.4 million with S. Paratyphi worldwide annually, and over 200,000 deaths [1]. The majority of the disease burden lies in South Asia, where access to accurate diagnostic testing is limited. In rural areas especially, there are very few data on the prevalence of enteric fever and drug resistance among its causative agents. In spite of this, patients presenting to health facilities with fever and no localizing symptoms are frequently presumed to have enteric fever and provided antibiotics directed at this entity. Additionally, with a number of improved vaccines for typhoid in development, there is a need to evaluate the burden of S. Typhi and S. Paratyphi in rural settings to project the need for and potential impact of new vaccines and other control programs. Further, there are increasing reports of S. Typhi and S. Paratyphi with reduced susceptibility or overt resistance to fluoroquinolones and azithromycin in South and Southeast Asia [2]–[4]; these antibiotics are among the most widely used in the treatment of individuals with enteric fever. Consequently, improved surveillance for drug susceptibility will be an important component in directing antimicrobial therapy policies to avert morbidity and mortality from enteric fever.
Serologic tests for typhoid have limited sensitivity and specificity in endemic settings and do not provide information on antimicrobial susceptibility. Blood culture and microbiological identification require skilled personnel, specialized laboratory equipment, and an uninterrupted electricity supply. These are not available in many healthcare facilities in low and middle income countries, particularly in rural areas [5]. This has hampered efforts to assess the burden of enteric fever and antimicrobial resistance in many resource-limited settings.
Here, we describe a simple approach for using a phase change-based incubator to isolate Gram-negative bacteria from the blood of patients with undifferentiated fever, noting that this procedure can be performed without electricity, sophisticated equipment, or specialized laboratory personnel. We compared this with conventional culture for recovery of typhoidal Salmonella. This approach would allow bacteriologic-based assessments of disease burden in difficult environments and permit recovery of organisms for assessing antimicrobial resistance patterns. Such data could assist with targeted roll out and assessment of typhoid vaccine and control programs in these areas.
Approval for this study was obtained from the Institutional Review Board for Human Subjects Research of the Nepal Health Research Council (Kathmandu, Nepal) and the Partners Human Research Committee (Boston, MA, USA). Participants 18 years of age and older were required to provide written informed consent in Nepali for enrollment in this study. For younger patients, parents/guardians provided written informed consent for study purposes after the child verbally assented to have the blood drawn. We followed the Standards for the Reporting of Diagnostic Accuracy Studies (STARD) [6].
This study was performed between July 2012 and October 2012 at Patan Hospital in the Kathmandu Valley of Nepal. All patients presenting to the outpatient or emergency departments with undifferentiated fever (fever and no alternative diagnosis by history and physical exam) during the study period were eligible for enrollment in the study. Patients with presumptive alternative diagnoses (cellulitis, pneumonia, urinary tract infection) were not included. Children under the age of 2 years and pregnant women were excluded; patients who were receiving antibiotics prior to presentation were not excluded.
Demographic and clinical information was obtained from patients who consented to participation in the study. These data included, age, sex, location of residence, clinical symptoms, duration of symptoms, and antimicrobial usage in the week prior to enrollment. An additional 4 ml of blood was collected from study patients as part of the routine venipuncture that was performed for their diagnostic evaluation (complete blood count, biochemical tests, and conventional blood culture).
BacT/ALERT (Biomérieux, Durham, NC, USA) blood culture bottles were inoculated with 4 ml of blood and 250 µg of vancomycin hydrochloride to suppress Gram-positive bacteria. Bottles were placed in a Portatherm electricity-free incubator, which maintains a constant temperature using phase-change materials. Fifty sealed packets containing 1-tetradecanol (a material that melts at 38°C) were prepared by submerging them in a hot water bath (Figure 1a). The reusable packets of 1-tetradecanol packets (Figure 1b) maintain a temperature of 38°C while changing phase from liquid back to solid (Figure 2). The packets were placed into an insulated container (vaccine storage boxes, which are low-cost and widely available in developing countries). The blood culture bottles were then placed together with melted phase-change packets into the containers and closed them (Figure 1c).
All blood culture bottles were inspected daily for discoloration of a CO2 indicator on the base of the bottle, indicating bacterial growth (Figure 1d). Bottles that demonstrated bacterial growth were removed, and the incubator was reloaded with freshly melted packets; this process was repeated daily for up to seven days. After seven days, subcultures were performed on samples from all negative bottles in both study arms onto Blood, MacConkey and Chocolate agars. Subculture and identification was also performed from all bottles that indicated bacterial growth.
Conventional blood cultures were performed by inoculating 4 ml of blood into 30–50 ml of media containing tryptone soya broth and 0.05% sodium polyanetholesulfonate. Per hospital standard practice, BACTEC PEDS PLUS bottles (Becton Dickinson, Sparks, MD, USA) were used for pediatric patients. Care was taken to ensure that an equal quantity of blood was inoculated into the conventional system as into the experimental one. Bottles were incubated at 37°C in a standard electric microbiological incubator. Bottles were treated as before. Isolates originating from both methods were identified using standard biochemical tests and serotype-specific antisera (Murex Biotech, Dartford, UK). Both the conventional and experimental culture procedures were overseen by experienced medical microbiologists. Laboratory personnel examining blood culture bottles from each method were blind to the results of the alternative method.
To evaluate how long after incubation S. Typhi and S. Paratyphi could be isolated from the blood culture bottles, we inoculated bottles with 1 colony forming unit/ml and 5 ml of whole blood, incubated them for a week, and then kept them at room temperature, performing subculture on a weekly basis.
The primary outcome was the proportion of cultures positive for S. Typhi or S. Paratyphi A in the conventional and experimental culture methods as a proportion of the total patients enrolled with undifferentiated fever. Because false positive results for isolation and identification of typhoid using biochemical testing and antisera would be unlikely by either culture system, the proportion that were positive by each system, using the total number of positive samples by either system as the denominator, was also calculated. Conventional blood culture with a sensitivity of around 40–60% is an imperfect reference standard; therefore the percentage positive and negative agreement rather than sensitivity and specificity are reported, consistent with expert recommendations and guidance of the United States Food and Drug Administration [7]. Point estimates together with exact 95% binomial confidence intervals for percent agreement are reported. Furthermore, the proportion of blood cultures that were positive among individuals receiving antimicrobials within the past week was compared with the proportion among those who had not received antimicrobials by Fisher's Exact Test. Time to positivity of culture was compared by the Wilcoxon Signed Rank Test.
Three hundred and thirty-seven patients with undifferentiated fever were approached for participation in this study, of whom 308 (91.4%) consented to participate in the study (Figure 3). An inadequate quantity of blood was drawn to perform both tests in 4 patients; the final enrolled population was 304 patients. The mean age of study participants was 16 years (IQR: 9–25 years) and 41.1% were female. The median duration of fever prior to enrollment was 5 days (IQR: 4–6 days).
Other pathogens were identified in blood from 5 of 304 (1.6%) of patients by the conventional blood culture system; we did not identify an alternative pathogen using the experimental blood culture. The alternative pathogens included Streptococcus pneumoniae (1 patient), other Streptococci (2 patients), Staphylococcus aureus (1 patient) and Acinetobacter spp (1 patient). Twenty-three patients (7.6%) had blood cultures positive for organisms believed to be commensal or skin contaminants; these were seen in 21 (6.9% of total patients) of the conventional blood cultures and 2 (0.7%) of the experimental blood cultures. The majority of commensal organisms (16 of 23; 69.6%) were coagulase-negative Staphylococci.
Accounting for the two “false-positive” cases of skin flora in the experimental procedure that would not have been distinguished from typhoid (without subculture), the overall percent agreement between the experimental blood culture and the conventional blood culture for diagnosis of enteric fever was 94.4% (95% CI: 91.2%–96.7%). The percent positive agreement was 90.9% (95% CI: 80.0%–97.0%) and the percent negative agreement was 95.2% (95% CI: 91.7%–97.5%).
The median time to positivity for typhoid isolates was 2 days for both the conventional blood culture (IQR: 1–3 days; range: 1–4 days) and the experimental blood culture (IQR: 2-2 days; range: 1–5 days). The positive culture result was available at least one day earlier by the conventional culture procedure in 14 patients, by at least one day earlier by the experimental culture in 9 patients, and on the same day in 27 patients (p = 0.38).
Eighty-two patients (27.3%) had received an antimicrobial within the previous week. The median duration of antimicrobial use was 3 days (IQR: 2–5 days). The percentage of patients with positive cultures was higher among individuals receiving antibiotics (30.5%) compared with those not receiving antibiotics (18.0%) (p = 0.027). Antibiotic exposure did not impact percent agreement between the conventional and experimental systems.
After incubation for a week, S. Typhi and S. Paratyphi A remained viable at room temperature, as demonstrated by subculture, for at least six months without supplementation of additional media.
Global estimates of the burden of typhoid are derived primarily from studies in dense urban areas, where culture microbiology is available. In rural areas, where a large proportion of the population resides, there are very few data on the incidence of typhoid due to lack of laboratory capacity. Because typhoid cannot be reliably distinguished from other febrile illnesses—such as viral infections, leptospirosis, and rickettsial infections [8]–[11], syndrome-based surveillance is inadequate. Last year, more than 500,000 cases of typhoid were reported in the public sector alone in Nepal, for an incidence of 1.9 episodes per 100 person-years [12]. This is an order of magnitude higher than the incidence estimated through surveys and other forms of surveillance in high-burden, urban settings [13], [14]. However, the majority of the diagnoses reported in Nepal were made empirically in locations without diagnostic laboratory capacity and may not be accurate. Despite its limited sensitivity, blood culture remains the best available method for establishing a diagnosis of typhoid and the only currently used means for assessing antimicrobial resistance, but it is not available in many high burden settings due to lack of reliable electricity, laboratory infrastructure and trained personnel [5], [15]. As a result, there are few data on the burden of S. Typhi and S. Paratyphi in areas without these resources, especially rural areas and even more limited data on resistance to commonly used antimicrobials in these settings.
Here we describe a simple approach to recovering Gram-negative organisms from blood that is not dependent upon electricity or sophisticated laboratory infrastructure at the site of medical care and can be performed by health personnel with minimal training. This procedure only requires drawing blood, placing the blood into a blood culture bottle containing vancomycin and capable of a colorimetric change, and then incubating the bottle in an insulated container into which reusable, heated packets containing 1-tetradecanol have been placed. We found that this method had comparable yield to conventional blood culture in recovering S. Typhi and S. Paratyphi A. By collecting positive bottles and performing identification and susceptibility testing at reference laboratories, this approach could enable ongoing surveillance for enteric fever prevalence among febrile patients and for antimicrobial sensitivity profiles in rural settings and other settings lacking adequate resources. Of note, as part of this project, we demonstrated that typhoidal Salmonella remain viable in the bottles for at least six months at room temperature after sample collection. Consequently, bottles could be collected from rural sites on a periodic basis for identification and susceptibility testing.
Multiple studies in South and Southeast Asia have demonstrated that S. Typhi and S. Paratyphi account for 90–100% of Gram-negative bacteria isolated from blood among patients presenting to hospitals [8], [16]–[24]. This may be particularly true in children, in whom other etiologies of Gram-negative bacteremia are less common in many resource-limited areas. In our study, 65 of 66 (98.5%) Gram-negative bacteremia diagnoses were attributable to S. Typhi or S. Paratyphi A. However, these studies were performed in an urban setting, and there are few data from rural health centers. By determining surveillance data on the prevalence of enteric fever and drug resistance in rural settings among patients with acute febrile illnesses, practitioners in those settings may be better able to make decisions about antimicrobial use for patients presenting with undifferentiated fever. The limited sensitivity of this culture approach does complicate estimation of prevalence of enteric fever; however, a number of statistical methods have been developed to estimate disease prevalence in the context of imperfect test accuracy [25]. The excellent specificity of culture, in contrast to serologic approaches, makes such estimates relatively straightforward.
Because most Gram-positive organisms isolated in blood cultures in many resource-limited areas are skin contaminants, we added an antibiotic to suppress Gram-positive bacteria, focusing our evaluation on Gram-negative organisms. In the conventional blood culture arm, in which this antibiotic was not added, Gram-positive pathogens only accounted for 6.7% of all pathogenic bacteria isolated. The majority (84%) of Gram-positive bacteria isolated were thought to be commensal skin contaminants. Of note, the antibiotic may be eliminated if surveillance for Gram-positive organisms is of interest in the particular setting being evaluated.
Across Nepal, electricity interruptions or scheduled rationing are common [26]. Even urban areas, such as Kathmandu, face 5 hours of scheduled outages per day during the wet season and up to 18 hours a day of scheduled outages per day during the dry season, with the outages projected to continue to increase for the foreseeable future [26]. Health centers are typically not exempted from these outages. In rural areas, outages can be unpredictable and long term, sometimes lasting for months. Throughout the developing world, maintaining a stable and reliable electricity supply is a common challenge for rural health centers, and this poses a challenge to supporting even basic microbiology laboratories. The phase change-based incubation method we describe has already been used for performing interferon gamma release assays for tuberculosis in resource-limited settings, and could have an array of uses in such areas [27].
In comparing diagnostic procedures with an imperfect reference standard, there has been debate about the appropriate characteristics to report and statistical tests to use for comparisons. In the case of typhoid, multiple studies have demonstrated that the sensitivity of blood culture is only 40–60% [28]–[32]. In cases of an imperfect reference standard, many expert bodies, including the United States Food and Drug Administration, advise against reporting sensitivity, specificity and predictive values. Instead, as we did here, guidelines recommend reporting the percent positive and negative agreement [7]. We chose to also report the proportion of cases identified by each diagnostic approach among all culture-confirmed cases identified by either diagnostic. We believe this approach is valid and useful because Salmonella isolated from either culture system are unlikely to be false positives in the absence of laboratory contamination. We used blood cultures results as our sole diagnostic criteria rather than clinical or serologic criteria because the latter are not yet well established in this setting, and the primary purpose of this study was to evaluate an alternative to conventional cultures for resource-limited settings. While the experimental system identified 10 cases of typhoid not identified by the conventional system (and the conventional system identified 5 cases not identified by the experimental system), these differences were not statistically significant and are consistent with concordance rates seen for paired conventional blood cultures [33].
This study and procedure have several important limitations. First, this procedure focused on recovery of the etiologic agents of enteric fever; other important bacterial pathogens, such as Brucella spp. and fastidious organisms, may be more difficult to isolate with this method [34], [35]. However, many widely used diagnostic tests focus on a single pathogen, including serologic tests for typhoid, brucellosis, Q fever, and leptospirosis; smear or rapid diagnostic tests for malaria or tuberculosis; and PCR techniques for influenza and dengue. Future work could examine the recovery of other bacterial pathogens.
We used only 4 ml of blood for culture, due to local practices shaped by patient concerns about phlebotomy of larger blood volumes, particularly among children. Reller and colleagues utilized a median blood volume of 1.96 ml when investigating pediatric patients for typhoid in Karachi, Pakistan and found all isolates were recovered from <5 ml of blood [19]. Because S. Typhi and S. Paratyphi are often present in low quantities in the blood [36], the culture of larger blood volumes (up to 15 ml) may increase the yield [37], though some studies have failed to demonstrate this [32]. A number of features beyond blood volume and low organisms load lessen the likelihood of recovering viable bacteria in blood, including late presentation and prior use of antibiotics. The presence of sodium polyanetholesulfonate in both the conventional and experimental blood cultures is thought to improve culture yield [38]. As we work to design a reusable bottle that can be produced on site, ox bile broth may serve the same purpose, by lysing blood cells that inhibit bacterial growth [37], [39]–[41]. Ox bile broth has the added feature of inhibiting the growth of many other bacteria including skin flora. The use of vancomycin was designed to suppress Gram-positive organisms, which more often than not represent contaminants (23 of 27 Gram positive isolates in this study were probable contaminants from the skin). However, 4 of the 304 patients screened had Gram-positive bacteremia with pathogenic organisms, which would be missed by the experimental approach. These data demonstrate that the experimental system is not ideal for a setting with more sophisticated laboratory capabilities. Simple chromogenic approaches to distinguish Gram positive and negative isolates in a closed culture system would add value to this procedure while maintaining simplicity for use by non-laboratory personnel.
The one-time cost of the incubator and reusable phase change packets is approximately $50, and the per-use costs of the blood culture bottles and additive were approximately $2.30. This per-use cost is substantially lower than some newer generation serologic tests for typhoid [42]. However, these costs are still high for Nepal, where annual per capita health expenditure is $30. We are currently working to design a lower cost bottle with a colorimetric growth indicator. Additionally, by performing the test on a random sample of patients presenting with fever, as often done in surveillance studies, costs may be minimized. When used for surveillance, subculture and identification costs will accrue for positive bottles; the use of vancomycin averts the costs of identification of skin contaminants. Finally, the technology for the phase change incubator, developed by one of the authors (ABS), is not patented and the phase-change packets could be locally produced.
Reliable, rapid, point-of-care diagnostics for enteric fever and drug resistance are desperately needed in resource-limited settings. While we await such developments, methods for evaluating the prevalence and drug resistance among S. Typhi and S. Paratyphi in regions for which we have no data would be a significant advance. The approach we described may begin to address this. Further studies to evaluate the challenges of implementing this approach in routine clinical environments in rural settings are needed.
|
10.1371/journal.pgen.1006340 | The Strength of Selection against Neanderthal Introgression | Hybridization between humans and Neanderthals has resulted in a low level of Neanderthal ancestry scattered across the genomes of many modern-day humans. After hybridization, on average, selection appears to have removed Neanderthal alleles from the human population. Quantifying the strength and causes of this selection against Neanderthal ancestry is key to understanding our relationship to Neanderthals and, more broadly, how populations remain distinct after secondary contact. Here, we develop a novel method for estimating the genome-wide average strength of selection and the density of selected sites using estimates of Neanderthal allele frequency along the genomes of modern-day humans. We confirm that East Asians had somewhat higher initial levels of Neanderthal ancestry than Europeans even after accounting for selection. We find that the bulk of purifying selection against Neanderthal ancestry is best understood as acting on many weakly deleterious alleles. We propose that the majority of these alleles were effectively neutral—and segregating at high frequency—in Neanderthals, but became selected against after entering human populations of much larger effective size. While individually of small effect, these alleles potentially imposed a heavy genetic load on the early-generation human–Neanderthal hybrids. This work suggests that differences in effective population size may play a far more important role in shaping levels of introgression than previously thought.
| A small percentage of Neanderthal DNA is present in the genomes of many contemporary human populations due to hybridization tens of thousands of years ago. Much of this Neanderthal DNA appears to be deleterious in humans, and natural selection is acting to remove it. One hypothesis is that the underlying alleles were not deleterious in Neanderthals, but rather represent genetic incompatibilities that became deleterious only once they were introduced to the human population. If so, reproductive barriers must have evolved rapidly between Neanderthals and humans after their split. Here, we show that observed patterns of Neanderthal ancestry in modern humans can be explained simply as a consequence of the difference in effective population size between Neanderthals and humans. Specifically, we find that on average, selection against individual Neanderthal alleles is very weak. This is consistent with the idea that Neanderthals over time accumulated many weakly deleterious alleles that in their small population were effectively neutral. However, after introgressing into larger human populations, those alleles became exposed to purifying selection. Thus, rather than being the result of hybrid incompatibilities, differences between human and Neanderthal effective population sizes appear to have played a key role in shaping our present-day shared ancestry.
| The recent sequencing of ancient genomic DNA has greatly expanded our knowledge of the relationship to our closest evolutionary cousins, the Neanderthals [1–5]. Neanderthals, along with Denisovans, were a sister group to modern humans, having likely split from modern humans around 550,000–765,000 years ago [5]. Genome-wide evidence suggests that modern humans interbred with Neanderthals after humans spread out of Africa, such that nowadays 1.5–2.1% of the autosomal genome of non-African modern human populations derive from Neanderthals [2]. This admixture is estimated to date to 47,000–65,000 years ago [6, 7], with potentially a second pulse into the ancestors of populations now present in East Asia [2, 8–11].
While some introgressed archaic alleles appear to have been adaptive in anatomically modern human (AMH) populations [12–14], on average selection has been suggested to act against Neanderthal DNA from modern humans. This can be seen from the non-uniform distribution of Neanderthal alleles along the human genome [9, 13]. In particular, regions of high gene density or low recombination rate have low Neanderthal ancestry, which is consistent with selection removing Neanderthal ancestry more efficiently from these regions [13]. In addition, the X chromosome has lower levels of Neanderthal ancestry and Neanderthal ancestry is absent from the Y chromosome and mitochondria [2, 4, 5, 9, 13, 15, 16]. The genome-wide fraction of Neanderthal introgression in Europeans has recently been shown to have decreased over the past forty thousand years, and, consistent with the action of selection, this decrease is stronger near genes [17]. Finally, a pattern of lower levels of Denisovan ancestry near genes and on the X chromosome in modern humans have also recently been reported [18, 19].
It is less clear why the bulk of Neanderthal alleles would be selected against. Were early-generation hybrids between humans and Neanderthals selected against due to intrinsic genetic incompatibilities? Or was this selection mostly ecological or cultural in nature? If reproductive barriers had already begun to evolve between Neanderthals and AMH, then these two hominids may have been on their way to becoming separate species before they met again [13, 20, 21]. Or, as we propose here, did differences in effective population size and resulting genetic load between humans and Neanderthals shape levels of Neanderthal admixture along the genome?
We set out to estimate the average strength of selection against Neanderthal alleles in AMH. Due to the relatively short divergence time of Neanderthals and AMH, we still share much of our genetic variation with Neanderthals. However, we can recognize alleles of Neanderthal ancestry in humans by aggregating information along the genome using statistical methods [9, 13]. Here, we develop theory to predict the frequency of Neanderthal-derived alleles as a function of the strength of purifying selection at linked exonic sites, recombination rate, initial introgression proportion, and split time. We fit these predictions to recently published estimates of the frequency of Neanderthal ancestry in modern humans [13]. Our results enhance our understanding of how selection shaped the genomic contribution of Neanderthal to our genomes, and shed light on the nature of Neanderthal–human hybridization.
In practice, we do not know the location of the deleterious Neanderthal alleles along the genome, nor could we hope to identify them all as some of their effects may be weak (but perhaps important in aggregate). Therefore, we average over the uncertainty in the locations of these alleles (Fig 1). We assume that each exonic base independently harbors a deleterious Neanderthal allele with probability μ. Building on a long-standing theory on genetic barriers to gene flow [22–27], at each neutral site ℓ in the genome, we can express the present-day expected frequency of Neanderthal alleles in our admixture model in terms of the initial frequency p0, as well as a function gℓ of the recombination rates r between ℓ and the neighboring exonic sites under selection, and the parameters s, t, and μ (see Eq 5, S2 Text). That is, at locus ℓ, a fraction pℓ,t = p0 gℓ(r, s, t, μ) of modern humans are expected to carry the Neanderthal allele. The function gℓ() decreases with tighter linkage to potentially deleterious sites, larger selection coefficient (s), longer time since admixture (t), and higher density of deleterious exonic sites (μ). If a neutral Neanderthal allele is initially completely unassociated with deleterious alleles, pℓ,t would on average be equal to p0. Our model explicitly accounts only for deleterious alleles that are physically linked to a neutral allele. However, in practice, neutral Neanderthal alleles will initially be associated (i.e. in linkage disequilibrium) not only with some linked, but also with potentially many unlinked deleterious alleles. This is because F1 hybrids inherited half of their genome from Neanderthal parents, which leads to a statistical association even among unlinked Neanderthal-derived alleles. Therefore, p0 should be thought of as an effective initial admixture proportion in the sense that it implicitly absorbs the effect of these physically unlinked, but statistically associated deleterious Neanderthal alleles. Technically this is because the effect of unlinked loci (assuming multiplicative fitness) can be factored into a constant multiplier of gℓ(), and so can be accomodated into the model by rescaling p0 (see pages 35 and 36 of [23]). In practice, this means that our estimates of p0 will almost certainly be underestimating the actual proportion of Neanderthal admixture. We will return to this point in the Discussion. We emphasize that, independently of the effect of unlinked deleterious mutations, there may still be more than one linked deleterious mutation associated with any given focal neutral site on average. To assess this possibility, in S2 Text we compare models that explicitly account for one versus multiple linked deleterious mutations.
To estimate the parameters of our model (p0, s, and μ), we minimised the residual sum of squared deviations (RSS) between observed frequencies of Neanderthal alleles [13] and those predicted by our model (see Eq 6 and S2 Text). We assess the uncertainty in our estimates by bootstrapping large contiguous genomic blocks and re-estimating our parameters. We then provide block-wise bootstrap confidence intervals (CI) based on these (Methods and S2 Text). In Figs 2 and 3, we show the RSS surfaces for the parameters p0, s, and μ for autosomal variation in Neanderthal ancestry in the EUR and ASN populations.
For autosomal chromosomes, our best estimates for the average strength of selection against deleterious Neanderthal alleles are low in both EUR and ASN (Fig 2), but statistically different from zero (sEUR = 4.1 × 10−4; 95% CI [3.4 × 10−4, 5.2 × 10−4], sASN = 3.5 × 10−4; 95% CI [2.6 × 10−4, 5.4 × 10−4]). We obtain similar estimates if we assume that the Neanderthal ancestry in humans has reached its equilibrium frequency or if we account for the effect of multiple selected sites (see S2 Text). However, and as expected, the estimated selection coefficients are somewhat lower for those models (S2 Text, Table A in S2 Text). Our estimates of the probability of any given exonic site being under selection are similar and low for both samples (μEUR = 8.1 × 10−5; 95% CI [4.1 × 10−5, 1.2 × 10−4], μASN = 6.9 × 10−5; 95% CI [4.1 × 10−5, 1.6 × 10−4]). These estimates correspond to less than 1 in 10,000 exonic base pairs harboring a deleterious Neanderthal allele, on average. As a result, our estimates of the average selection coefficient against an exonic base pair (the compound parameter (μs) are very low, on the order of 10−8 in both samples (Table 1).
Consistent with previous findings [10, 11], we infer a higher initial frequency of Neanderthal alleles in the East Asian sample compared to the European sample (p0,EUR = 3.38 × 10−2; 95% CI [3.22 × 10−2, 3.52 × 10−2], p0,ASN = 3.60 × 10−2; 95% CI [3.45 × 10−2, 3.86 × 10−2]), but the 95% bootstrap CI overlap (Fig 3). This occurs because our estimates of the initial frequency of Neanderthal alleles (p0) are mildly confounded with estimates of the strength of selection per exonic base (μs). That is, somewhat similar values of the expected present-day Neanderthal allele frequency can be inferred by simultaneously reducing p0 and μs (Fig 4). This explains why the marginal confidence intervals for p0 overlap for ASN and EUR. However, if μs, the fitness cost of Neanderthal introgression per exonic base pair, is the same for ASN and EUR (i.e. if we take a vertical slice in Fig 4), the values of p0 for the two samples do not overlap.
To verify the fit of our model, we plot the average observed frequency of Neanderthal alleles, binned by gene density per map unit, and compare it to the allele frequency predicted by our model based on the estimated parameter values (Fig 5). There is good agreement between the two, suggesting that our model provides a good description of the relationship between functional density, recombination rates, and levels of Neanderthal introgression. At the scale of 1 cM, the Pearson correlation between observed and predicted levels of autosomal Neanderthal introgression is 0.897 for EUR and 0.710 for ASN (see Table C in S2 Text for a range of other scales).
Our estimated coefficients of selection (s) against deleterious Neanderthal alleles are very low, on the order of the reciprocal of the effective population size of humans. This raises the intriguing possibility that our results are detecting differences in the efficacy of selection between AMH and Neanderthals. Levels of genetic diversity within Neanderthals are consistent with a very low long-term effective population size compared to AMH, i.e. a higher rate of genetic drift [5]. This suggests that weakly deleterious exonic alleles may have been effectively neutral and drifted up in frequency in Neanderthals [28–30], only to be slowly selected against after introgressing into modern human populations of larger effective size. To test this hypothesis, we simulated a simple model of a population split between AMH and Neanderthals, using a range of plausible Neanderthal population sizes after the split. In these simulations, the selection coefficients of mutations at exonic sites are drawn from an empirically supported distribution of fitness effects [31]. We track the frequency of deleterious alleles at exonic sites in both AMH and Neanderthals, and compare these frequencies at the time of secondary contact (admixture). We show a subset of our simulation results in Fig 6. Due to a lower effective population size, the simulated Neanderthal population shows an excess of fixed deleterious alleles compared to the larger human population (Fig 6A). This supports the assumption we made in our inference procedure that the deleterious introgressing alleles had been fixed in Neanderthals prior to admixture. Moreover, our estimates of s fall in a region of parameter space for which simulations suggest that Neanderthals have a strong excess of population-specific fixed deleterious alleles, compared to humans (Fig 6B). Over the relevant range of selection coefficients, the fraction of simulated exonic sites that harbor these Neanderthal-specific weakly deleterious alleles is on the order of 10−5, which is in approximate agreement with our estimates of μ. Therefore, a model in which the bulk of Neanderthal alleles, which are now deleterious in modern humans, simply drifted up in frequency due to the smaller effective population size of Neanderthals seems quite plausible. This conclusion has also been independently reached by a recent study via a simulation-based approach [32].
We finally turn to the X chromosome, where observed levels of Neanderthal ancestry are strongly reduced compared to autosomes [9, 13]. This reduction could be consistent with the X chromosome playing an important role in the evolution of hybrid incompatibilities at the early stages of speciation [13]. However, a range of other phenomena could explain the observed difference between the X and autosomes, including sex-biased hybridization among populations, the absence of recombination in males, as well as differences in the selective regimes [33–35]. We modified our model to reflect the transmission rules of the X chromosome and the absence of recombination in males. We give the X chromosome its own initial level of introgression (p0,X), different from the autosomes, which allows us to detect a sex bias in the direction of matings between AHM and Neanderthals. Although our formulae can easily incorporate sex-specific selection coefficients, we keep a single selection coefficient (sX) to reduce the number of parameters. Therefore, sX reflects the average reduction in relative fitness of deleterious Neanderthal alleles across heterozygous females and hemizygous males.
We fit the parameters p0,X, μX, and sX using our modified model to [13]’s observed levels of admixture on the X chromosome (Table 1; S12 and S13 Figs). Given the smaller amount of data, the inference is more challenging as the parameters are more strongly confounded (for an example of μX and sX, see S12 and S13 Figs). We therefore focus on the compound parameter μXsX, i.e. the average selection coefficient against an exonic base pair on the X. In Fig 4, we plot a sample of a thousand bootstrap estimates of μXsX for the X, along with analogous estimates of μs for autosomal chromosomes. For the X chromosome, there is also strong confounding between p0,X and μXsX, to a much greater extent than on the autosomes (note the larger spread of the X point clouds). Due to this confounding, our marginal confidence intervals for μXsX and p0,X overlap with their autosomal counterparts (Table 1). However, the plot of p0 and μs bootstrap estimates clearly shows that the X chromosome and autosomes differ in their parameters.
For reasons we do not fully understand, the range of parameter estimates for the X chromosome with strong bootstrap support is much larger for the ASN than for the EUR samples (Fig 4). For the ASN samples, the confidence intervals for μXsX include zero, suggesting there is no strong evidence for selection against introgression on the X. This is consistent with the results of [13], who found only a weakly significant correlation between the frequency of Neanderthal alleles and gene density on the X chromosome. However, as the ASN confidence intervals for μXsX are large and also overlap with the autosomal estimates, it is difficult to say if selection was stronger or weaker on the X chromosome compared to the autosomes. For the EUR samples, however, the confidence intervals for μXsX do not include zero, which suggests significant evidence for selection against introgression on the X, potentially stronger than that on the autosomes. Note that the selection coefficients on the X (sX, Table 1) are still on the order of one over the effective population size of modern humans, as was the case for the autosomes. Therefore, differences in effective population size between Neanderthals and modern humans, and hence in the efficacy of selection, might well explain observed patterns of introgression on the X as well as on the autosomes. If the exonic density of selection against Neanderthal introgression was indeed stronger on the X, one plausible explanation is the fact that weakly deleterious alleles that are partially recessive would be hidden from selection on the autosomes but revealed on the X in males [33–35].
Our results are potentially consistent with the notion that the present-day admixture proportion on the X chromosome was influenced not only by stronger purifying selection, but also by a lower initial admixture proportion p0,X (Fig 4). Lower p0,X is consistent with a bias towards matings between Neanderthal males and human females, as compared to the opposite. Based on our point estimates, and if we attribute the difference between the initial admixture frequency between the X and the autosomes (p0,X and p0,A) exclusively to sex-biased hybridization, our result would imply that matings between Neanderthal males and human females were about three times more common than the opposite pairing (S2 Text). However, as mentioned above, there is a high level of uncertainty about our X chromosome point estimates. Therefore, we view this finding as very provisional.
There is growing evidence that selection has on average acted against autosomal Neanderthal alleles in anatomically modern humans (AMH). Our approach represents one of the first attempts to estimate the strength of genome-wide selection against introgression between populations. The method we use is inspired by previous efforts to infer the strength of background selection and selective sweeps from their footprint on linked neutral variation on a genomic scale [36–39]. We have also developed an approach to estimate selection against on-going maladaptive gene flow using diversity within and among populations [40] that will be useful in extending these findings to a range of taxa. Building on these approaches, more refined models of selection against Neanderthal introgression could be developed. These could extend our results by estimating a distribution of selective effects against Neanderthal alleles, or by estimating parameters separately for various categories of sequence, such as non-coding DNA, functional genes, and other types of polymorphism(e.g. structural variation) [41].
Here, we have shown that observed patterns of Neanderthal ancestry in modern human populations are consistent with genome-wide purifying selection against many weakly deleterious alleles. For simplicity, we allowed selection to act only on exonic sites. It is therefore likely that the effects of nearby functional non-coding regions are subsumed in our estimates of the density (μ) and average strength (s) of purifying selection. Therefore, our findings of weak selection are conservative in the sense that the true strength of selection per base pair may be even weaker. We argue that the bulk of selection against Neanderthal ancestry in humans may be best understood as being due to the accumulation of alleles that were effectively neutral in the Neanderthal population, which was of relatively small effective size. However, these alleles started to be purged, by weak purifying selection, after introgressing into the human population, due to its larger effective population size.
Thus, we have shown that it is not necessary to hypothesize many loci harboring intrinsic hybrid incompatibilities, or alleles involved in ecological differences, to explain the bulk of observed patterns of Neanderthal ancestry in AMH. Indeed, given a rather short divergence time between Neanderthals and AMH, it is a priori unlikely that strong hybrid incompatibilities had evolved at a large number of loci before the populations interbred. It often takes millions of years for hybrid incompatibilities to evolve in mammals [42, 43], although there are exceptions to this [44], and theoretical results suggest that such incompatibilities are expected to accumulate only slowly at first [45, 46]. While this is a subjective question, our results suggest that genomic data—although clearly showing a signal of selection against introgression—do not strongly support the view that Neanderthals and humans should be viewed as incipient species. Sankararaman et al. [13] found that genes expressed in the human testes showed a significant reduction in Neanderthal introgression, and interpreted this as being potentially consistent with a role of reproductive genes in speciation. However, this pattern could also be explained if testes genes were more likely to harbor weakly deleterious alleles, which could have accumulated in Neanderthals. These two hypotheses could be addressed by relating within-species estimates of the distribution of selective effects with estimates of selection against introgression at these testes genes.
This is not to say that alleles of larger effect, in particular those underlying ecological or behavioral differences, did not exist, but rather that they are not needed to explain the observed relationship between gene density and Neanderthal ancestry. Alleles of large negative effect would have quickly been removed from admixed populations, and would likely have led to extended genomic regions showing a deficit of Neanderthal ancestry as described by [9, 13, 47]. Since our method allows us to model the expected amount of Neanderthal ancestry along the genome accounting for selection, it could serve as a better null model for finding regions that are unusually devoid of Neanderthal ancestry.
We have ignored the possibility of adaptive introgressions from Neanderthals into humans. While a number of fascinating putatively adaptive introgressions have come to light [14], and more will doubtlessly be identified, they will likely make up a tiny fraction of all Neanderthal haplotypes. We therefore think that they can be safely ignored when assessing the long-term deleterious consequences of introgression.
As our results imply, selection against deleterious Neanderthal alleles was very weak on average, such that, after tens of thousands of years since their introduction, these alleles will have only decreased in frequency by 56% on average. Thus, roughly seven thousand loci (≈ μ × 82 million exonic sites) still segregate for deleterious alleles introduced into Eurasian populations via interbreeding with Neanderthals. However, given that the initial frequency of the admixture was very low, we predict that a typical EUR or ASN individual today only carries roughly a hundred of these weak-effect alleles, which may have some impact on genetic load within these populations.
Although selection against each deleterious Neanderthal allele is weak, the early-generation human–Neanderthal hybrids might have suffered a substantial genetic load due to the sheer number of such alleles. The cumulative contribution to fitness of many weakly deleterious alleles strongly depends on the form of fitness interaction among them, but we can still make some educated guesses (the caveats of which we discuss below). If, for instance, the interaction was multiplicative, then an average F1 individual would have experienced a reduction in fitness of 1 − (1 − 4 × 10−4)7000 ≈ 94% compared to modern humans, who lack all but roughly one hundred of these deleterious alleles. This would obviously imply a substantial reduction in fitness, which might even have been increased by a small number of deleterious mutations of larger effect that we have failed to capture. This potentially substantial genetic load has strong implications for the interpretation of our estimate of the effective initial admixture proportion (p0), and, more broadly, for our understanding of those early hybrids and the Neanderthal population. We now discuss these topics in turn.
Strictly, under our model, the estimate of p0 reflects the initial admixture proportion in the absence of unlinked selected alleles. However, the large number of deleterious unlinked alleles present in the first generations after admixture violates that assumption, as each of these unlinked alleles also reduces the fitness of hybrids [23]. These unlinked deleterious alleles should cause a potentially rapid initial loss of Neanderthal ancesty following the hybridization. Harris and Nielsen [32] have recently independently conducted simulations of the dynamics of deleterious alleles during the initial period following Neanderthal admixture. They have shown that the frequency of Neanderthal-derived alleles indeed decreases rapidly in the initial generations due to the aggregate effects of many weakly deleterious loci. The reduction in neutral Neanderthal ancestry due to unlinked sites under selection is felt equally along the genome and as such, our estimate of p0 is an effective admixture proportion that incorporates the genome-wide effect of unlinked deleterious mutations, but not the localized effect of linked deleterious mutations (as formalized by Bengtsson [23]). In practice, segregation and recombination during meiosis in the early generations after admixture will have led to a rapid dissipation of the initial associations (statistical linkage disequilibrium) among any focal neutral site and unlinked deleterious alleles. Therefore, our estimates of p0 can actually be interpreted as the admixture proportion to which the frequency of Neanderthal alleles settled down to after the first few generations of segregation off of unlinked deleterious alleles. As a consequence, the true initial admixture proportion may have been much higher than our current estimates of p0. However, any attempt to correct for this potential bias in our estimates of p0 is likely very sensitive to assumptions about the form of selection, as we discuss below. Conversely, our estimates of the strength and density of deleterious sites (s and μ) do not strongly change when we include multiple deleterious sites or consider large windows surrounding each focal neutral site (up to 10 cM) in our inference procedure (see S2 Text for details). This is likely because much of the information about s and μ comes from the localized dip in Neanderthal ancestry close to genes, and thus these estimates are not strongly affected by the inclusion of other weakly linked deleterious alleles (the effects of which are more uniform, and mostly affect p0).
If the predicted drop in hybrid fitness is due to the accumulation of many weakly deleterious alleles in Neanderthals, as supported by our simulations, it also suggests that Neanderthals may have had a very substantial genetic load (more than 94% reduction in fitness) compared to AMH (see also [28, 29, 32]). It is tempting to conclude that this high load strongly contributed to the low population densities, and the extinction (or at least absorption), of Neanderthals when faced with competition from modern humans. However, this ignores a number of factors. First, selection against this genetic load may well have been soft, i.e. fitness is measured relative to the most fit individual in the local population, and epistasis among these many alleles may not have been multiplicative [48–50]. Therefore, Neanderthals, and potentially early-generation hybrids, may have been shielded from the predicted selective cost of their load. Second, Neanderthals may have evolved a range of compensatory adaptations to cope with this large deleterious load. Finally, Neanderthals may have had a suite of evolved adaptations and cultural practices that offered a range of fitness advantages over AMH at the cold Northern latitudes that they had long inhabited [51, 52]. These factors also mean that our estimates of the total genetic load of Neanderthals, and indeed the fitness of the early hybrids, are at best provisional. The increasing number of sequenced ancient Neanderthal and human genomes from close to the time of contact [7, 17, 53] will doubtlessly shed more light on these parameters. However, some of these questions may be fundamentally difficult to address from genomic data alone.
Whether or not the many weakly deleterious alleles in Neanderthals were a cause, or a consequence, of the low Neanderthal effective population size, they have had a profound effect on patterning levels of Neanderthal introgression in our genomes. More generally, our results suggest that differences in effective population size and nearly neutral dynamics may be an important determinant of levels of introgression across species and along the genome. Species coming into secondary contact often have different demographic histories (e.g. as is the case of Drosophila yakuba and D. santomea [54, 55] or in Xiphophorus sister species [56]) and so the dynamics we have described may be common.
We have here considered the case of introgression from a small population (Neanderthals) into a larger population (humans), where selection acts genome-wide against deleterious alleles introgressing. However, from the perspective of a small population with segregating or fixed deleterious alleles, introgression from a population lacking these alleles can be favoured [57]. This could be the case if the source population had a large effective size, and hence lacked a comparable load of deleterious alleles. Therefore, due to this effect, our results may also imply that Neanderthal populations would have received a substantial amount of adaptive introgression from modern humans.
Here we describe the model for the frequency of a Neanderthal-derived allele at a neutral locus linked to a single deleterious allele. In S1 Text we extend this model to deleterious alleles at multiple linked loci. Let S1 and N1 be the introgressed (Neanderthal) alleles at the selected and linked neutral autosomal locus, respectively, and S2 and N2 the corresponding resident (human) alleles. The recombination rate between the two loci is r. We assume that allele S1 is deleterious in humans, such that the viability of a heterozygote human is w(S1S2) = 1 − s, while the viability of an S2S2 homozygote is w(S2S2) = 1. We ignore homozygous carriers of allele S1, because they are expected to be very rare, and omitting them does not affect our results substantially (S1 Text). We assume that, prior to admixture, the human population was fixed for alleles S2 and N2, whereas Neanderthals were fixed for alleles S1 and N1. After a single pulse of admixture, the frequency of the introgressing haplotype N1S1 rises instantaneously from 0 to p0 in the human population. We discuss the consequences of multiple pulses in S1 Text.
In S1 and S2 Texts we study the more generic case where both S1 and S2 are segregating in the Neanderthal population prior to admixture. Fitting this full model to data (S2 Text), we found that it resulted in estimates which implied that the deleterious allele S1 is on average fixed in Neanderthals. This was further supported by our individual-based simulations (S18 Fig), which show that in a vast majority of realisations, the deleterious allele was either at very low or very high frequency in the Neanderthals immediately prior to introgression due to the high levels of genetic drift in Neanderthals. Therefore, we focus only on the simpler model where allele S1 is fixed in Neanderthals, as described above.
The present-day expected frequency of allele N1 in modern humans can be written as
p t = p 0 f ( r , s , t ) , (1)
where f(r, s, t) is a function of the recombination rate r between the neutral and the selected site, the selection coefficient s, and the time t in generations since admixture (S1 Text).
Based on the derivations in S1 Text, we find that, for autosomes, f is given by
f a ( r , s , t ) = [ ( 1 - s ) ( 1 - r ) ] t [ 1 - r - ( 1 - s ) ( 1 - r ) ] + r 1 - ( 1 - s ) ( 1 - r ) . (2)
We also have developed results for a neutral locus linked to a single deleterious locus in the non-pseudo-autosomal (non-PAR) region of the X chromosome (S1 Text). As above, we also assume that the deleterious allele is fixed in Neanderthals. The non-PAR region does not recombine in males and we assume that the recombination rate in females between the two loci is r. In S1 Text we develop a full model allowing for sex-specific fitnesses. For simplicity, here we assume that heterozygous females and hemizygous males carrying the deleterious Neanderthal allele have relative fitness 1 − s. Following our results in S1 Text we obtain
f X ( r , s , t ) = s ( 1 - 2 3 r ) t + 1 ( 1 - s ) t + 2 3 r 1 - ( 1 - 2 3 r ) ( 1 - s ) , (3)
where the factors 2/3 and (1 − 2/3) reflect the fact that, on average, an X-linked allele spends these proportions of time in females and males, respectively. We also fitted models with different selection coefficients in heterozygous females and hemizygote males, but found that there was little information to separate these effects.
Our results relate to a long-standing theory on genetic barriers to gene flow [22–27], a central insight of which is that selection can act as a barrier to neutral gene flow. This effect can be modelled as a reduction of the neutral migration rate by the so-called gene flow factor [23], which is a function of the strength of selection and the genetic distance between neutral and selected loci. In a single-pulse admixture model at equilibrium, f is equivalent to the gene flow factor (S1 Text).
Lastly, we introduce a parameter μ to denote the probability that any given exonic base is affected by purifying selection. If μ and s are small, we found that considering only the nearest-neighboring selected exonic site is sufficient to describe the effect of linked selected sites in our case (but see Results and Discussion for the effect of unlinked sites under selection). That is, for small μ, selected sites will be so far apart from the focal neutral site ℓ that the effect of the nearest selected exonic site will dominate over the effects of all the other ones. In S1 Text we provide predictions for the present-day frequency of N1 under a model that accounts for multiple linked selected sites, both for autosomes and the X chromosome. We further assume that an exon of length l bases will contain the selected allele with probability ≈ μl (for μl ≪ 1), and that the selected site is located in the middle of that exon. Lastly, the effects of selection at linked sites will be small if their genetic distance from the neutral site is large compared to the strength of selection (s). In practice, we may therefore limit the computation of Eq (1) to exons within a window of a fixed genetic size around the neutral site. We chose windows of size 1 cM around the focal neutral site ℓ, but also explored larger windows of size 10 cM to show that our results are not strongly affected by this choice. Taken together, these assumptions greatly simplify our computations and allow us to calculate the expected present-day frequency of the Neanderthal allele at each SNP along the genome.
Specifically, consider a genomic window of size 1 cM centered around the focal neutral site ℓ, and denote the total number of exons in this window by I ℓ. Let the length of the ith nearest exon to the focal locus ℓ be li base pairs. The probability that the ith exon contains the nearest selected site is then μ l i ∏ j = 1 i - 1 ( 1 - μ l j ), where the product term is the probability that the selected site is not in any of the i − 1 exons closer to ℓ than exon i. Conditional on the ith exon containing the selected site, the frequency pt of N1 at locus ℓ and time t is computed according to Eq (1), with r replaced by ri, the recombination rate between ℓ and the center of exon i. Then, we can write the expected frequency of the neutral Neanderthal allele at site ℓ surrounded by I ℓ exons as
E [ p t , ℓ ] = p 0 g ℓ ( r , s , t , μ ) , (4)
where
g ℓ ( r , s , t , μ ) = ∑ i = 1 I ℓ μ l i ∏ j = 1 i - 1 ( 1 - μ l j ) f ( r i , s , t ) + ∏ j = 1 I ℓ 1 - μ l j . (5)
The last product term accounts for the case where none of the I ℓ exons contains a deleterious allele. Eq (5) can be applied to both autosomes and X chromosomes, with f as given in Eqs (2) and (3), respectively.
We downloaded recently published estimates of Neanderthal alleles in modern-day humans [13], as well as physical and genetic positions of polymorphic sites (SNPs) from the Reich lab website. We use estimates from Sankararaman et al. [13] of the average marginal probability that a human individual carries a Neanderthal allele as our Neanderthal allele frequency, pn. Although pn is also an estimate, we generally refer to it as the observed frequency, in contrast to our predicted/expected frequency pt. Sankararaman et al. [13] performed extensive simulations to demonstrate that these calls were relatively unbiased. We performed separate analyses using estimates of pn for samples originating from Europe (EUR) and East Asia (ASN) (Table 1, [13]).
Although composed of samples from multiple populations, for simplicity we refer to EUR and ASN as two samples or populations. We downloaded a list of exons from the UCSC Genome browser. We matched positions from the GRCh37/hg19 assembly to files containing estimates of pn to calculate distances to exons. We estimated recombination rates from a genetic map by Kong et al. [58].
Our inference method relies on minimizing the residual sum of squared differences (RSS) between E[pℓ,t] and pℓ,n over all nl autosomal (or X-linked) SNPs for which [13] provided estimates. Specifically, we minimize
RSS = ∑ ℓ = 1 n l ( p ℓ , n - E [ p ℓ , t ] ) 2 = ∑ ℓ = 1 n l p ℓ , n - p 0 g ℓ ( r , s , t , μ ) 2 , (6)
where gℓ(r, s, t, μ) is calculated according to Eq (5). For each population, we first performed a coarse search over a wide parameter space followed by a finer grid search in regions that had the smallest RSS. For each fine grid, we calculated the RSS for a total of 676 (26 × 26) different combinations of s and μ. We did not perform a grid search for p0. Rather, for each combination of s and μ, we analytically determined the value of p0 that minimizes the RSS as
p 0 , min , s i , μ i = ∑ ℓ = 1 n l p ℓ , n g ℓ ∑ ℓ = 1 n l g ℓ 2 , (7)
where gℓ is given in Eq (5) and we sum over all nl considered autosomal (X-linked) SNPs. For details, we refer to S2 Text.
We obtained confidence intervals by calculating 2.5 and 97.5 percentiles from 1000 bootstrapped genomes. We created these chromosome by chromosome as follows. For a given chromosome, for each non-overlapping segment of length 5 cM, and for each of 676 parameter combinations, we first calculated the denominator and the numerator of Eq (7) using the number of SNPs in the segments instead of nl. We then resampled these segments (with replacement) to create a bootstrap chromosome of the same length as the original chromosome. Once all appropriate bootstrap chromosomes were created (chromosomes 1–22 in the autosomal case, or the X chromosome otherwise), we obtained for each bootstrap sample the combination of p0, μ, and s that minimises the RSS according to Eqs (6) and (7).
In S2 Text we extend our inference approach to incorporate the influence of multiple selected loci on levels of introgression (in various size windows up to 10 cM in size). We also explored using a more stringent set of Neanderthal calls and using a variance-weighted sum of squares approach. All of these approaches resulted in similar estimates of s and μ, suggesting that our findings are reasonably robust to our choices.
To test whether selection against alleles introgressed from Neanderthals can be explained by the differences in ancient demography, we simulated the frequency trajectories of deleterious alleles in the Neanderthal and human populations, between the time of the Neanderthal–human split and the time of admixture (S3 Text). We assume that the separation time was 20,000 generations (∼600k years). For the distribution of selection coefficients we use those of [31]. This distribution was estimated under the assumption of no dominance [31], and we follow this assumption in our simulations. For the simulations summarized in Fig 6 we assumed an effective population size of 1000 for Neanderthals and 10,000 for humans. Our simulations are described more fully in S3 Text, where we also show versions of Fig 6 for a range of effective population sizes for Neanderthals. The timing of the out-of-Africa bottleneck in humans relative to admixture with Neanderthals is unclear. Therefore, we also explored the effect of a population bottleneck in humans (before admixture) on the accumulation of deleterious alleles (see S3 Text). We allowed the duration of this bottleneck to vary from 10 to 1000 generations. These simulations show that our findings in Fig 6 are robust to the precise details of the demography of the human populations. We acknowledge that our understanding of the human populations that initially encountered Neanderthals is scant, and they may have been small in size. However, importantly the populations that represent the ancestors of modern-day Eurasians do not appear to have had the sustained history of small effect population sizes over hundreds of thousands of years that characterize Neanderthals. Therefore, our simulations likely capture the important broad dynamics of differences in effective population size on deleterious allele load.
For each simulation run, we recorded the frequency of the deleterious allele in Neanderthals and humans immediately prior to admixture. Our simulations show that the majority of deleterious alleles that are still segregating at the end of the simulation are fixed differences (Fig 6). This matches the assumption of our approach, and agrees with the estimates we obtained. Our simulations include both ancestral variation and new mutations, but the majority of the segregating alleles at the end of the simulations represent differentially sorted ancestral polymorphisms.
Harris and Nielsen [32] independently conducted a simulation study of the accumulation of deleterious alleles in Neanderthals, and the fate of these after introgression into modern humans. Their results about the accumulation of weakly deleterious additive alleles in Neanderthals are consistent with ours. In addition, these authors also investigated the introgression dynamics with linked recessive deleterious alleles. They found that, under some circumstances, recessive deleterious alleles may actually favor introgression as a consequence of pseudo-overdominance. However, the majority of weakly selected alleles are expected to act in a close-to-additive manner, as empirical results suggest an inverse relationship between fitness effect and dominance coefficient [59, 60]. Therefore, our assumptions of additivity are appropriate for the majority of deleterious loci.
|
10.1371/journal.ppat.1002023 | Structure-Function Analysis of the Anopheles gambiae
LRIM1/APL1C Complex and its Interaction with Complement C3-Like Protein
TEP1 | Malaria threatens half the world's population and exacts a devastating human
toll. The principal malaria vector in Africa, the mosquito Anopheles
gambiae, encodes 24 members of a recently identified family of
leucine-rich repeat proteins named LRIMs. Two members of this family, LRIM1 and
APL1C, are crucial components of the mosquito complement-like pathway that is
important for immune defense against Plasmodium parasites.
LRIM1 and APL1C circulate in the hemolymph exclusively as a disulfide-bonded
complex that specifically interacts with the mature form of the complement
C3-like protein, TEP1. We have investigated the specificity of LRIM1/APL1C
complex formation and which regions of these proteins are required for
interactions with TEP1. To address these questions, we have generated a set of
LRIM1 and APL1C alleles altering key conserved structural elements and assayed
them in cell culture for complex formation and interaction with TEP1. Our data
indicate that heterocomplex formation is an intrinsic ability of LRIM1 and APL1C
and identify key homologous cysteine residues forming the intermolecular
disulfide bond. We also demonstrate that the coiled-coil domain is the binding
site for TEP1 but also contributes to the specificity of LRIM1/APL1C complex
formation. In addition, we show that the LRIM1/APL1C complex interacts with the
mature forms of three other TEP proteins, one of which, TEP3, we have
characterized as a Plasmodium antagonist. We conclude that
LRIM1 and APL1C contain three distinct modules: a C-terminal coiled-coil domain
that can carry different TEP protein cargoes, potentially with distinct
functions, a central cysteine-rich region that controls complex formation and an
N-terminal leucine-rich repeat with a putative role in pathogen recognition.
| The malaria-transmitting mosquito, Anopheles gambiae, uses a
complement-like pathway to defend against Plasmodium parasites.
The complement C3-like protein, TEP1, binds to the surface of invading
parasites, triggering their destruction and clearance. LRIM1 and APL1C, two
leucine-rich repeat proteins, form a disulfide-bonded complex which stabilizes
mature TEP1 and promotes its binding to parasites. Here, we investigate the
structural and biochemical features of the LRIM1/APL1C complex and its
interaction with TEP1. We identify key amino acid residues responsible for
covalently linking LRIM1 and APL1C and the region of the complex where TEP1
binds. Importantly, we demonstrate that the LRIM1/APL1C complex can interact
with the mature form of three other TEPs, including TEP3, which we characterize
as a novel Plasmodium antagonist. Our results suggest that the
LRIM1/APL1C complex has a modular architecture in which distinct functions map
to different regions. Our study provides important insights into how the
A. gambiae complement pathway helps mosquitoes fight
against the malaria parasite.
| The innate immune system is the primary, and in some organisms, such as insects, the
sole means of defense against infection. The main mosquito defense against invading
Plasmodium is orchestrated by a collection of hemolymph
proteins that closely resembles the vertebrate complement cascade [1]. The majority
of Plasmodium ookinetes traversing the mosquito midgut epithelium
and coming into contact with the hemolymph are attacked and cleared by lysis or by
encasement in a melanin capsule (melanization). Both of these reactions are
triggered by binding on the parasite surface of the thioester-containing protein
TEP1, a homolog of the complement factor C3 [2]. The few parasites that escape
this reaction develop into oocysts and, protected by the oocyst wall, amplify their
numbers and differentiate into sporozoites, the vertebrate infective form of
Plasmodium.
How parasites are recognized by the mosquito immune system and how complement
activation is biochemically regulated remain unanswered, but recent work has
revealed that these reactions involve complex networks of basally expressed
proteins, including LRIM1 and APL1C, two putative pathogen recognition receptors of
the leucine-rich repeat (LRR) immune protein (LRIM) family [2], [3], [4], [5], [6], [7], [8], [9],[10]. LRIM1 and APL1C circulate in the
mosquito hemolymph as a disulfide-bonded high-molecular weight complex and are major
antagonists of mosquito infections with the rodent parasite P.
berghei
[6]. Silencing
the genes that encode LRIM1, APL1C and TEP1 transforms a refractory A.
gambiae strain into a susceptible strain [2], . Importantly, this
triumvirate of proteins contribute to resistance against Plasmodium
of the non-vector mosquito A. quadriannulatus A; their silencing
renders these mosquitoes permissive vectors [11]. The LRIM1/APL1C complex
interacts with proteolytically processed (mature) TEP1 in the mosquito hemolymph
[5], [6]. This
interaction stabilizes this mature and reactive form of TEP1, promoting its binding
to the parasite surface and preventing its reaction with self.
LRIM1 and APL1C share several conserved structural features including a signal
peptide, an LRR domain, a pattern of cysteine residues and a C-terminal coiled-coil
domain [6],
[12]. LRR
domains are common in immune receptors and are flexible in their binding properties,
e.g. Toll-like receptors [13] and the variable lymphocyte receptors of jawless
vertebrates [14],
while coiled-coil domains often mediate protein-protein interactions. The
three-dimensional structure of the LRIM1/APL1C heterodimer has been recently
determined, revealing the presence of a single disulfide bond between the two
proteins formed by conserved cysteine residues and providing a structural framework
for elucidation of the function of this innate immune complex [15]. We designed a
structure-function biochemical study to further our understanding of the
interactions between LRIM1 and APL1C, and to investigate the role of their
constituent domains in interactions with TEP1 and other immune proteins. Using a
panel of engineered LRIM1 and APL1C alleles we
reveal that the cysteine-rich region between the LRR and coiled-coil domains is
crucial for LRIM1/APL1C complex formation and corroborate the identity of the
cysteines involved in the formation of the disulfide bridge that is however not
required for the interaction between the LRIM1/APL1C complex and TEP1. We also show
that the coiled-coil domain is largely dispensable for complex formation, but is
essential for interactions with TEP1 as well as with three other TEP family members,
one of which we show to be a potent antagonist of P. berghei
infection. Our work reveals a modular nature of the LRIM1/APL1C complex, which is
common in LRR-containing innate receptors, and demonstrates that by carrying
different cargoes this putative receptor may serve distinct immune functions.
The LRIM1 and APL1C proteins circulate in the mosquito hemolymph as a
disulfide-linked complex. To determine whether LRIM1 and APL1C have the
intrinsic ability to form complexes or if they require a cofactor, we expressed
LRIM1 and APL1C alone or together in Sf9
cells, an insect cell line derived from the lepidopteran, Spodoptera
frugiperda. Given that the LRIM family is mosquito specific, a
non-mosquito derived cell line should lack potential interacting partners. For
these experiments and those described below we used expression constructs
containing LRIM1 and APL1C transgenes that
incorporate Strep and His epitope tags on the N- and C-termini, respectively
[6].
Following transfection, conditioned medium (CM) was collected from the cells and
analyzed by western blot of non-reducing (NR) sodium dodecyl sulfate
polyacrylamide gels (SDS-PAGE) using antibodies against LRIM1 and APL1C. CM
collected from cells transfected with LRIM1 or
APL1C show that each protein individually has the ability
to form a high-molecular weight homomeric complex (Figure 1). Co-transfection of
LRIM1 and APL1C together preferentially
yields a complex intermediate in size compared to their single transfections. As
expected due to the presence of the tags, the LRIM1/APL1C heterocomplex is
slightly larger than the untagged native complex present in the mosquito
hemolymph. Given the observed sizes of the monomeric tagged LRIM1 and APL1C (63
and 97 kDa, respectively) the size of the complexes produced by single (169 and
235 kDa) and double transfections (197 kDa) best fit with homotrimers and
heterotrimers. However, the stoichiometry of the LRIM1/APL1C complex was
recently determined to be a heterodimer using techniques that are not influenced
by protein shape [15]. The same study also showed that both LRIM1 and APL1C
could form homodimers. Therefore, the LRIM1/APL1C complex appears to migrate
aberrantly slowly on SDS-PAGE gels as is commonly reported for proteins
containing coiled-coils [16], [17], [18].
Given that the LRIM1/APL1C complex is held together by disulfide bonds, cysteine
residues are implicated as a key feature in complex formation [6]. In
addition, the coiled-coil domains of LRIM1 and APL1C are in direct apposition to
each other in the crystal structure [15] suggesting these domains
may instruct the correct assembly of LRIM1/APL1C complexes. To assay the
contribution of the conserved cysteine residues and coiled-coil domain in
complex formation, we engineered alleles of these features based on the
LRIM1 and APL1C expression constructs
described above (Figure 2).
Cysteine to serine missense mutations were generated for each of the 5 conserved
cysteine residues of LRIM1 located between the LRR and coiled-coil domains
(C273S, C305S, C317S, C318S, C352S). A single cysteine missense mutation of
APL1C was generated (C562S) that targets the cysteine residue homologous to
LRIM1 C352. The residue number of this cysteine in APL1C matches the A.
gambiae PEST genome reference strain but differs from what was
reported by Baxter and coworkers [15] which is likely due to a
difference in the polymorphic region of PANGGL repeats [8], [12]. In addition, our
previous unpublished observations indicated that LRIM4, another member of the
Long subfamily of LRIM proteins, forms a disulfide-bonded homodimer. LRIM4 lacks
a cysteine residue homologous to LRIM1 C352 or APL1C C562 [12]. Instead LRIM4 contains
a cysteine residue (C535) at its extreme C-terminus, following the coiled-coil
domain (Figure 2). We
generated a missense mutation of this residue in LRIM4 (C535S) to determine
whether it is responsible for homodimer formation. As LRIM1 and APL1C have
bipartite coiled-coil domains with greater than 80% confidence to have
the ability to form multimers (Figure S1) [19], we generated two alleles
that remove each region separately (ΔCCa and ΔCCb) and two alleles that
remove the coiled-coil domain altogether, one retaining LRIM1 C352 and APL1C
C562 (ΔCC1) and one deleting these residues (ΔCC2) (Figure 2). Additionally, we created
alternative expression constructs for LRIM1, LRIM1C352S,
APL1CC562S and APL1CΔCC1 containing a C-terminal
Herpes Simplex Virus (HSV) epitope tag, as some assays required proteins lacking
a His tag.
To determine which conserved cysteine residues of LRIM1 contribute to homo- and
heterodimer formation, we collected samples of CM from transfected Sf9 cells and
analyzed them by NR western blot. Of the 5 alleles, only
LRIM1C352S produced a protein that was
secreted into the CM (Figure
3A). Unlike wild-type LRIM1, which is secreted as both a monomer and
homomeric complex, only monomeric LRIM1C352S was present in the CM.
We performed immunolocalization experiments using both LRIM1 and Strep-tag
antibodies and found that all cysteine mutant proteins are produced (Figure S2).
Therefore, disruption of C273, C305, C317 and C318 prevents LRIM1 secretion. We
next co-expressed the LRIM1 cysteine alleles with wild-type
APL1C to determine if the presence of the wild-type partner
would facilitate secretion or complex formation. Even when co-expressed with
wild-type APL1C, LRIM1C273S, LRIM1C305S,
LRIM1C317S and LRIM1C318S were still absent from the
CM. Similar to when it was expressed on its own, LRIM1C352S was
present in the CM exclusively as a monomer. Given the importance of C352 in the
ability of LRIM1 to form complexes, we tested whether a missense allele of the
homologous cysteine residue (C562) of APL1C (see Figure 2) would behave similarly. We
expressed APL1CC562S alone or together with
wild-type LRIM1 and found that in both cases the protein was
present in the CM only as a monomer (Figure 3B). These data reveal a crucial role
for the terminal conserved cysteine adjacent to the start of the coiled-coil
domain of LRIM1 and APL1C in their ability to form homo- and heterodimers. To
test the flexibility of the location of the cysteine involved in LRIM dimer
formation, we expressed LRIM4C535S in Sf9 cells. When CM was analyzed
under reducing conditions, wild-type LRIM4 and LRIM4C535S both
migrate at approximately 62 kDa, consistent with the predicted size of a monomer
(Figure 3C). When
analyzed under NR conditions, LRIM4C535S remains a monomer whereas
wild-type LRIM4 migrates at the predicted size of a homodimer. These results
indicate that LRIM4 C535 is functionally equivalent to LRIM1 C352 and APL1C
C562.
Even though LRIM1C352S and APL1CC562S do not form a
disulfide-linked complex, we wanted to determine whether these proteins can
interact non-covalently. We co-expressed His- and HSV-tagged versions of
LRIM1C352S and APL1CC562S in Sf9 cells. After
confirming secretion of the proteins into the CM of the transfected cells (Figure 3D), we performed His
pull-down experiments. Western analysis showed that the His-tagged
LRIM1C352S and APL1CC562S were efficiently captured
from the CM. Probing the captured material using an antibody against the HSV tag
revealed homo- and heteromeric interactions between LRIM1C352S and
APL1CC562S (Figure
3D). The relative strength of the observed interactions directly
parallels the abilities of the wild-type proteins to homo- and heterodimerize
(Figure 1); LRIM1 and
APL1C interact the strongest and APL1C dimerizes more efficiently than
LRIM1.
To analyze how the coiled-coil domain of LRIM1 contributes to secretion and
complex formation, we transfected the set of coiled-coil alleles into Sf9 cells.
All of these produce proteins that are secreted into the CM in similar abundance
and migrate at their expected relative sizes when analyzed under reducing
conditions (Figure 4A). NR
western blot analysis of the CM revealed that LRIM1ΔCCa,
LRIM1ΔCCb and LRIM1ΔCC1 were present as both
monomers and homodimers (Figure
4B). In contrast, LRIM1ΔCC2, missing C352, was
exclusively monomeric. Next we co-expressed the LRIM1 coiled-coil alleles with
wild-type APL1C and analyzed the CM under NR conditions. The
LRIM1ΔCCa, LRIM1ΔCCb and
LRIM1ΔCC1 proteins were each present in an additional higher
molecular weight complex compared to when they were expressed alone, indicating
that these proteins can form heterodimers with wild-type APL1C (Figure 4B). Again, the
LRIM1ΔCC2 protein was only present in the CM as a monomer,
demonstrating that it is unable to form complexes with itself or with APL1C. We
performed a similar series of experiments with His-tagged coiled-coil alleles of
APL1C and these behaved like their LRIM1
counterparts. All the APL1C coiled-coil alleles produced
proteins that were secreted into the CM and migrated at their expected sizes
when analyzed under reducing conditions (Figure 4C). When analyzed under NR conditions
we found that APL1CΔCCa, APL1CΔCCb and
APL1CΔCC1 formed homodimers when expressed alone and
heterodimers when co-expressed with wild-type HSV-tagged LRIM1 (Figure 4D). The
APL1CΔCC2 protein, missing C562, only produced a monomer both
when expressed alone or when co-expressed with wild-type LRIM1 (Figure 4D).
Despite lacking a coiled-coil domain, LRIM1ΔCC1 and
APL1CΔCC1 can form homodimers when expressed alone and
heterodimers when co-expressed with a wild-type partner. This suggests that the
coiled-coil domain is not absolutely required for complex formation. To test
this using the natural protein pair, we co-expressed His-tagged
LRIM1ΔCC1 and HSV-tagged APL1CΔCC1 to
determine if a heterodimer could form between partners both lacking a
coiled-coil domain. LRIM1ΔCC1 and APL1CΔCC1 were
expressed alone or together and analyzed by western blot for complex formation.
As shown in Figure 4B and
4D, when expressed alone LRIM1ΔCC1 and
APL1CΔCC1 form homodimers (Figure 4E). Co-expression of
LRIM1ΔCC1 and APL1CΔCC1 produces a new complex
containing both His and HSV tags that is intermediate in size to the homodimers
(Figure 4E). Therefore,
LRIM1ΔCC1 and APL1CΔCC1 can form homo- and
heterodimers demonstrating that the coiled-coil domain is largely dispensable
for LRIM1/APL1C complex formation.
To identify novel proteins that may function with the LRIM1/APL1C complex in
parasite killing, we performed a large-scale capture experiment from the
mosquito hemocyte-like cell line that was used previously to reveal an
interaction between LRIM1/APL1C and the mature form of TEP1 [6].
His-tagged LRIM1 and APL1C were co-expressed and captured from the CM. Proteins
co-captured with the LRIM1/APL1C complex were separated on a NR SDS-PAGE gel,
visualized by staining with colloidal Coomassie and identified by
mass-spectrometry (MS) (Figure
5A). The LRIM1 and APL1C homo- and heterodimers identified migrate at
a slightly greater molecular weight in this assay because protein samples were
separated on a fixed percentage gel (compare Figure 5A to Figure 1). In addition to LRIM1 monomer and
the N- and C-terminal fragments of mature TEP1, we also identified the N- and
C-terminal fragments of TEP3, the C-terminal fragment of TEP4 and the N-terminal
fragment of TEP9 (Table
1). Therefore, the LRIM1/APL1C complex can interact with the mature forms
of 4 different TEP proteins in the CM of mosquito cells. Band 4 at approximately
90 kDa was identified as LRIM1. Since this is too small to be a homodimer, it is
probably a proteolytic fragment of a high molecular weight LRIM1 complex (band 2
or 3). A positive MS identification could not be made for band 5 that, based on
molecular weight, is likely to be a monomer of APL1C. Finally, band 8 was
identified as an RNA poly-A binding protein; however, given that this is an
intracellular protein, it is likely to be a contaminant.
Finding that the LRIM1/APL1C complex can interact with more than one TEP family
member suggests that these TEP proteins may function in mosquito immune
reactions against Plasmodium parasites. Interestingly, two of
the TEPs identified, TEP3 and TEP4, were previously shown to play an important
role in bacterial defense [3], [20]. We analyzed the role of TEP3 and TEP4 on P.
berghei infection intensity and melanization. After
TEP3 silencing, mosquitoes showed a highly significant
increase in developing oocysts 7 days post infection (Figure 5B). This increase in oocysts was not
as great as upon TEP1 silencing. In contrast, silencing
TEP4 had no effect on oocyst numbers, which is consistent
with a previous report [3]. Given the important role of LRIM1, APL1C and TEP1 in
parasite melanization, we next assayed whether TEP3 or TEP4 also function in
this process. To test this, we silenced TEP3 and
TEP4 together with CTL4, a potent
inhibitor of the melanization cascade [21]. Silencing
CTL4 results in a striking increase in melanized ookinetes
and a decrease in live oocysts (Figure 5C). When TEP1 is silenced together with
CTL4, melanization is completely blocked and there is a
dramatic increase in live oocysts. Silencing of TEP3 and
CTL4 together results in an interesting intermediate
phenotype whereby there is a significant increase in oocysts but melanization is
not significantly reduced (Figure
5C). TEP4 silencing together with
CTL4 has no significant effect on oocysts numbers or
parasite melanization.
Given that the LRIM1/APL1C complex can interact with the processed form of 4
different TEP proteins, we wanted to examine if binding is mediated by the TEP-N
or TEP-C fragments independently or whether both are required. We generated
HSV-tagged expression constructs for full-length TEP1 and TEP3 and their N- and
C- terminal fragments. These TEPs were chosen because both TEP1 and TEP3 are
P. berghei antagonists. CM containing HSV-tagged TEP
protein fragments was mixed with CM containing His-tagged LRIM1/APL1C. Following
incubation, the LRIM1/APL1C complex was captured by the His tag and samples were
assayed for the presence of TEP fragments by western blot using an antibody
against HSV. For both TEP1 and TEP3, we observed the strongest interaction
between the LRIM1/APL1C complex and CM containing both the TEP-N and TEP-C
fragments (Figure 6).
Interactions between the individual fragments and full-length TEP1 and TEP3 were
considerably weaker despite their similar abundance in the starting CM. These
data show that the LRIM1/APL1C complex interacts strongly with TEPs only when
they are processed and both N- and C- terminal fragments are present.
To analyze the interaction of the LRIM1/APL1C complex with TEP proteins, we
performed binding assays between different LRIM1 and APL1C alleles and TEP1-N
and TEP1-C. These assays aimed to reveal whether TEP1 can independently interact
with both LRIM1 and APL1C and whether interaction requires disulfide-bonded
complexes or an intact coiled-coil domain. We chose the Sf9 binding assay
described above because this system lacks endogenous LRIMs and because in this
system LRIM1 and APL1C can interact non-covalently (see Figure 3D). Following separate transfections,
CM containing the LRIM1 and/or APL1C variants was mixed with CM containing
TEP1-N and TEP1-C. Recombinant LRIM1 and APL1C proteins were captured using
their His tag and analyzed for TEP1 binding by western blot using an HSV
antibody. Strikingly, TEP1 is only captured by CM containing the LRIM1/APL1C
heterodimer (Figure 7A). It
was not present in samples containing only LRIM1 or APL1C monomers and
homodimers.
Next we tested the ΔCCa, ΔCCb and ΔCC1 coiled-coil alleles of LRIM1
and APL1C expressed with a wild-type partner as well as co-expressed ΔCC1
and ΔCC2 alleles. Finally, given that co-expressed LRIM1C352S and
APL1CC562S can interact non-covalently we tested whether they can
cooperate to bind TEP1. We found that complexes between ΔCCa alleles with a
wild-type partner captured TEP1-N and TEP1-C, but all of the other combinations
of coiled-coil alleles lacked TEP1 binding despite a similar amount of captured
His-tagged proteins (Figure
7A). We also observed strong binding between TEP1-N and TEP1-C when
we used CM containing both LRIM1C352S and APL1CC562S.
Taken together our results, summarized in Figure 7B, demonstrate that TEP1 binds to the
CCb region of the coiled-coil domain of the LRIM1/APL1C heterodimer and that
this binding requires the presence of both LRIM1 and APL1C but not necessarily
in a covalent complex. The coiled-coil domains of LRIM1 and APL1C are
intertwined within the complex and adopt a helix-loop-helix (HLH) fold [15], which
would provide ample space for protein-protein interaction. Our data reveal that
the TEP1 binding site is situated within this extensive coiled-coil region of
the complex and makes important contacts with the coiled-coil CCb domain.
To test the specificity of the LRIM1/APL1C interaction with TEP1, we investigated
whether the LRIM4 homodimer can also interact with TEP1. We performed binding
assays using the wild-type LRIM4 construct expressed in a mosquito hemocyte-like
cell line described above. His-tagged LRIM4 was captured from the CM and the
samples were analyzed by western blot for TEP1. No interaction was observed
between LRIM4 and TEP1 (Figure S3A). As controls we expressed both
His-tagged LRIM1 and APL1C, which can interact with their endogenous partner
produced by these cells and capture mature TEP1. This demonstrates that the
LRIM1/APL1C interaction with TEP1 is specific and not common to other LRIM
dimers. Furthermore, His-tagged LRIM4 did not interact with endogenous LRIM1 or
APL1C produced by these cells (data not shown). As LRIM4 does not interact with
TEP1, LRIM1 or APL1C, we hypothesized that LRIM4 is not involved in P.
berghei defense. Indeed, upon LRIM4 knockdown
there was no effect on P. berghei infection intensity or
prevalence (Figure S3B).
The mosquito complement pathway plays a pivotal role in infections with
Plasmodium parasites. In this study we biochemically dissect
the structural features of the LRIM1/APL1C complex that contribute to its formation
and interaction with the complement C3-like protein TEP1. LRIM1 and APL1C circulate
in the adult A. gambiae hemolymph exclusively as a heterodimer
[6].
Similarly, LRIM1 and APL1C primarily form a heterodimer when over-expressed in Sf9
or cultured mosquito cell lines. However, they can also form monomers and homodimers
suggesting that these alternative forms are either highly unstable in the hemolymph
or retained intracellularly due to stricter quality control mechanisms.
Numerous disulfide-bonded heterocomplexes, similar to LRIM1/APL1C, have been
discovered with key roles in immunity, hemostasis and complement activation. Notable
examples include the IgG heavy and light chain peptides, platelet glycoproteins
Ibα and Ibβ [22] and importantly, the extensive repertoires of secreted
variable lymphocyte receptor (VLR) antibodies in jawless vertebrates [23], [24]. Our study
reveals that C352 of LRIM1 and the homologous residue, C562, of APL1C play a crucial
role in covalently linking these proteins through a disulfide bond, which is
consistent with direct apposition of these residues in the LRIM1/APL1C crystal
structure [15].
Importantly, we show that this disulfide linkage is not necessary for the
interaction between LRIM1 and APL1C or between the complex and the processed form of
TEP1. This observation raises an important question about the role of the disulfide
bond in the function of the LRIM1/APL1C complex. It is possible that the disulfide
bond is necessary for the release of TEP1 from the complex during an immune
response. For example, it might act as a molecular hinge and allow for a productive
conformational change required for TEP1 release. Alternatively, it may increase the
stability of the complex in the mosquito hemolymph and/or prevent the two proteins
from acting independently, e.g. as homodimers.
The dispensability of the disulfide bridge in the formation of the LRIM1/APL1C
complex also suggests that LRIM family members lacking a free cysteine residue in
their C-terminal region may still be capable of forming homodimers or heterodimers.
In addition, by including LRIM4 in our analysis, we demonstrate that the location of
the bridging cysteine residue is flexible: despite its position on the opposite side
of the coiled-coil domain, C535 of LRIM4 appears capable of forming an
intermolecular disulfide bond. It remains to be seen whether the LRIM4 homodimer
adopts a similar or different structure to the LRIM1/APL1C heterodimer [15]. Although
LRIM4 (also referred to as LRRD5) is highly
upregulated in the A. gambiae midgut after P.
falciparum infection [3], little is known about its functional role in the innate
immune response. We demonstrate here that silencing LRIM4 has no
effect on P. berghei infection and that the protein does not
interact with TEP1, LRIM1 or APL1C.
We show that mutations in LRIM1 of each of the remaining cysteine residues located in
the region between the LRR and the coiled-coil domains (C273, C305, C317 and C318)
yield proteins that are trapped within the cell and unable to be secreted into the
CM even when expressed with a wild-type APL1C. Thus mutation of these cysteines is
likely to grossly affect protein folding. These cysteines form two intramolecular
disulfide bonds [15] and their location in all members of the LRIM protein
family is at the end of the LRR domain [12]. Intramolecular disulfide
bonds between adjacent cysteines in this region may generate a family-specific
C-terminal cap similar to those identified in other LRR proteins [25]. Given the
importance of these cysteines to the correct folding of LRIM1, it is interesting to
note that the double cysteine motif in Transmembrane (TM) LRIMs is replaced by a
tyrosine-cysteine pair [12]. Members of this LRIM subfamily are predicted to only
form a single intramolecular bond and leaving a cysteine free to potentially form an
intermolecular bridge.
Expression of various mutant or deletion LRIM1 or APL1C alleles together with their
wild-type partners, respectively, implicate the cysteine-rich region as the key
determinant in LRIM1/APL1C complex formation and reveal that the coiled-coil domain
of these proteins is largely dispensable for heterodimer complex formation. However,
we show that the most carboxy-terminal coiled-coil region (CCb) may play a role in
the specificity of LRIM1 and APL1C interaction. When co-expressed with their
wild-type partner, proteins lacking CCb form homo- and heterodimers with equal
efficiency whereas those possessing CCb favor heterodimer formation. Importantly,
the CCb region of both LRIM1 and APL1C is critical for binding to mature (processed)
TEP1 altogether raising the intriguing possibility that TEPs may be involved in the
specificity of complex formation.
We have revealed that the combined coiled-coil domain of the LRIM1/APL1C heterodimer
is the binding site of mature TEP1 and that homodimers of LRIM1 and APL1C do not
bind TEP1. In addition to revealing an interaction with mature TEP1, which was
previously shown [5],[6], our MS analysis of proteins interacting with the
LRIM1/APL1C complex revealed mature forms of 3 other TEP family members. It is not
known whether LRIM1/APL1C interacts with each TEP individually and competitively
under different circumstances. However, the reported 1∶1 stoichiometry of TEP1
to LRIM1/APL1C heterodimer [15], makes it likely that different TEPs form independent
complexes with LRIM1/APL1C.
We demonstrate that one of the TEPs we found to interact with the LRIM1/APL1C
complex, TEP3, is also an antagonist of P. berghei infections. As
the increase in oocysts upon TEP3 silencing is not as dramatic as
with LRIM1, APL1C and TEP1, we
speculate that TEP3 is either redundant or has a more indirect role in P.
berghei killing. Another possible explanation is that TEP3 participates
in a Plasmodium defense pathway that is distinct from or
complementary to that of TEP1. This is consistent with the requirement of both TEP3
and LRIM1 for phagocytosis of gram-negative but not gram-positive bacteria whereas
TEP1 is important for both [20]. Importantly, unlike TEP1, TEP3 has an inactive thioester
motif [26],
and although TEPs lacking an active thioester are reported to play a role in immune
reactions such as phagocytosis [27] their function may
be regulatory rather than structural. It remains to be determined whether TEP3
functions against the human malaria parasite, P. falciparum.
The LRIM1/APL1C complex also interacts with the mature form of TEP4 that plays an
important role in bacterial defense and phagocytosis [3], [4], [20]. TEP4 has been previously shown
to be upregulated by P. berghei infections [28], but we show here that it has no
effect against P. berghei. Therefore, we hypothesize that
LRIM1/APL1C and TEP4 cooperate in a defense mechanism against bacteria and that the
TEP4 upregulation is due to opportunistic infections with gut bacteria that occur
during Plasmodium traversal of the mosquito midgut epithelium. Both
TEP3 and TEP4 are strongly upregulated by bacteria [29].
Our finding that the LRIM1/APL1C complex can interact with multiple members of the
TEP protein family opens new avenues for investigating how mosquitoes may generate
pathogen-specific immune responses. For example, LRIM1, APL1C and TEP1 have a
prominent role in mosquito defense against P. berghei, but of these
proteins only TEP1 has been shown to play a role in controlling infections with
human malaria parasites, P. falciparum
[3], [30], [31]. Just as the
LRIM1/APL1C complex can interact with multiple TEP proteins, it is possible that
TEP1 may also interact with multiple LRIM family members. APL1A and LRIM17 are
attractive candidates given that they have been shown to be antagonists of
P. falciparum
[3], [31]. APL1A is a
particularly intriguing candidate since it is 61% identical to APL1C in the
coiled-coil region that contributes to the TEP1 binding site. The same region of
APL1B is 78% identical to APL1C, and although APL1B has been shown to be
dispensable for defense against P. berghei and P.
falciparum
[31], it may
interact with TEP1 in a pathogen-specific manner. Future research is important to
determine whether TEP1 exists in different complexes in the mosquito hemolymph and
if such complexes contribute to pathogen-specific responses.
In this paper we provide the biochemical framework for understanding the role of the
LRIM1/APL1C complex in regulating mosquito immunity to Plasmodium.
Taken together, our data reveal that the LRIM1/APL1C complex is organized into three
distinct modules as summarized in Figure S4. The central region containing a
pattern of conserved cysteine residues is largely responsible for LRIM1/APL1C
complex formation, while the coiled-coil may also contribute to the specificity of
the interaction. The combined C-terminal coiled-coil region functions to carry
different TEP cargoes. We show that at least four different TEP family members with
distinct and overlapping roles in mosquito innate defense bind to the LRIM1/APL1C
complex. Finally, we hypothesize that the LRR domains of LRIM1 and APL1C function in
activation of the complex, possibly through recognition of pathogen surfaces
directly or via an interaction with other immune receptors. What triggers the
release of mature TEP1 from the LRIM1/APL1C complex is important to understanding
how the mosquito complement pathway targets and eliminates malaria parasites.
This study was carried out in strict accordance with the United Kingdom Animals
(Scientific Procedures) Act 1986. The protocols for maintenance of mosquitoes by
blood feeding and for infection of mosquitoes with P. berghei
by blood feeding on parasite-infected mice were approved and carried out under
the UK Home Office License PLL70/6347 awarded in January 2008. The procedures
are of mild to moderate severity and the numbers of animals used are minimized
by incorporation of the most economical protocols. Opportunities for reduction,
refinement and replacement of animal experiments are constantly monitored and
new protocols are implemented following approval by the Imperial College Ethical
Review Committee.
Alleles of LRIM1, APL1C and
LRIM4 were generated by PCR using pIEx10
clones as templates and primers listed in the Table S1.
Products were LIC cloned (Merck Chemicals) into a dual Strep- and His-tag vector
(pIEx10) or InFusion cloned (Clontech) into an HSV-tag
(pIEx1SPmyc) vector. The pIEx1SPmyc vector
is a variant of the pIEx1 (Merck Chemicals) that retains its
C-terminal HSV tag but was modified to contain an N-terminal IgM signal peptide
and Myc epitope tag. Cysteine and ΔCCa alleles of LRIM1 and
APL1C were generated by splice-overlap extension (SOE) PCR
[32]
using outer “His” and inner allele-specific primers. The
LRIM4C535S missense mutation was created
using the QuikChange II Site-Directed Mutagenesis protocol (Stratagene). All
constructs were verified by DNA sequencing.
Protein samples were separated on 8% or 4-15% Criterion SDS-PAGE
gels (Bio-Rad). NR samples were prepared in Lane Marker sample buffer (Pierce).
Reduced samples were made by supplementing NR samples with a
tris(2-carboxyethyl)phosphine (TCEP) solution (Pierce) to a final concentration
of 25 mM and heating at 95°C for 5 min. Transfer to PVDF and western
conditions were previously described [6] except for rabbit
α-GFP (1∶1000 diluted in PBS + 0.05% Tween 20 and
3% milk), mouse α-Strep-tag (1∶200) and goat α-HSV-tag
(1∶1000) diluted in PBS + 0.05% Tween 20 and 3%
BSA.
Analysis of LRIM1, APL1C and
LRIM4 alleles was performed using the CM of transfected Sf9
cells adapted to growth in serum-free (SFM) culture medium (Sf-900 II,
Invitrogen). Cells were transfected using Escort IV reagent (Sigma) and CM was
collected 3-4 days post transfection, cleared of debris by centrifugation or by
passage through a 0.45 µm filter and supplemented with NR sample buffer.
Unless noted, a 2∶1 µg ratio of APL1C to LRIM1 was used in all
co-transfection experiments to achieve comparable expression levels. Hemolymph
was collected as described previously [6]. For MS analysis,
mosquito Sua4.0 cells were transfected using Effectene (Qiagen) at
80–90% confluence in non-vented 175 mm2 culture flasks.
DNA complexes were made by diluting 7.5 µg of
pIEx10-LRIM1 and 17.6 µg
pIEx10-APL1C with 1.3 mL EC buffer and then adding 40
µL of enhancer followed by 125 µL of Effectene reagent. Complexes
were added to cells dropwise and incubated for 12 h in Schneider's
Drosophila (S2) medium containing 10%
heat-inactivated FCS. Cells were washed with and then placed in 30 mL of
serum-free S2 medium for conditioning. CM was collected after 5 days, passed
through a 0.45 µm syringe filter into a conical tube and supplemented with
0.05% triton X-100. A slurry of 650 µL (packed volume) of Ni-NTA
agarose (Qiagen) in PBS was added to the CM and mixed for 2 h at room
temperature. Beads were washed once in the tube with buffer (50 mM
NaH2PO4, 300 mM NaCl and 20 mM imidazole pH 8.0) and
then transferred to a 10 mL disposable column for further washes. Bound proteins
were eluted in 10 mL of wash buffer containing 250 mM imidazole and then
concentrated to approximately 100 µL using a 10 kDa cutoff Amicon Ultra
filter (Millipore). Samples were made by the addition of NR buffer, separated on
an 8% SDS-PAGE gel, stained with Imperial stain (Pierce) and imaged
before MS identification of individual bands (performed at EMBL).
TEP1 binding assays of LRIM1 and APL1C alleles
were performed using the CM of transfected Sf9 cells. Cells were transfected
independently with pIEx1SPmyc-TEP and
pIEx10-APL1C/pIEx10-LRIM1
vectors. 200 µL of each CM was mixed and supplemented with 0.1%
triton X-100 and 5 µL of a 1∶1 slurry of Talon resin (Stratagene) in
PBS. After a 3 h incubation at room temperature the beads were washed in PBS
containing 0.1% triton X-100 and extracted with 35 µL of 2x NR
loading buffer. Binding between LRIM1C352S and APL1CC562S
was determined by performing His pull-down assays using CM from Sf9 cells
co-transfected with pIEx10-LRIM1C352S and
pIEx1SPmyc-APL1CC562S plasmids,
respectively. TEP1 binding assays for wild-type LRIM4 were performed using the
CM of transfected mosquito Sua4.0 cells as described previously [6].
The N'gousso strain of A. gambiae was maintained as
described previously [33], [34]. Mosquitoes were cultured and infected with
P. berghei CONGFP strain [35] as described
previously [11]. Single and double knockdown experiments and parasite
counts in dissected midguts were performed as described previously [6]. Primers
used for synthesis of double stranded RNA are listed in Table
S1.
LRIM1, AGAP006348; APL1C, AGAP007033; LRIM4, AGAP007039; TEP1, AGAP010815; TEP3,
AGAP010816; TEP4, AGAP010812; TEP9, AGAP010830; polyA-binding protein,
AGAP011092; S7, AGAP010592; CTL4, AGAP005335.
|
10.1371/journal.pntd.0001329 | Epidemiology of Concomitant Infection Due to Loa loa and Mansonella perstans in Gabon | The filarial parasites Loa loa and Mansonnella perstans are endemic in the central and western African forest block. Loa loa is pathogenic and represents a major obstacle to the control of co-endemic filariae because its treatment can cause fatal complications such as encephalitis.
4392 individuals aged over 15 years were studied both by direct examination and a concentration technique. The overall prevalence rates were 22.4% for Loa loa microfilaremia, 10.2% for M. perstans microfilaremia, and 3.2% for mixed infection. The prevalence of both filariae was higher in the forest ecosystem than in savannah and lakeland (p<0.0001). The intensity of microfilariae (mf) was also higher in the forest ecosystem for both parasites. The prevalence and intensity of microfilaria were both influenced by age and gender. Correlations were found between the prevalence and intensity of Loa loa microfilariae (r = 0.215 p = 0.036), and between the prevalence of Loa loa and the prevalence of individuals with microfilaria >8000 mf/ml (r = 0.624; p<0.0001) and microfilariae >30 000 mf/ml (r = 0.319, p = 0.002). In contrast, the prevalence of pruritis and Calabar swellings correlated negatively with the prevalence of Loa loa microfilaria (r = −0.219, p = 0.032; r = −0.220; p = 0.031, respectively). Pruritis, Calabar swellings and eye worm were not associated with L. loa mf intensity (r = −0.144, p = 0.162; r–0.061, p = 0.558; and r = 0.051, p = 0.624, respectively), or with the prevalence or intensity of M. perstans microfilariae.
This map of the distribution of filariae in Gabon should prove helpful for control programs. Our findings confirm the spatial uniformity of the relationship between parasitological indices. Clinical manifestations point to a relationship between filariae and allergy.
| Loa loa and Mansonella perstans are blood filarial parasites, endemic in the central and western African forest block, and transmitted by chrysops and culicoides flies, respectively. Loa loa is pathogenic and represents a major obstacle to the control of co-endemic filariae. Treatment of individuals with >8000 Loa loa microfilariae/ml can result in severe adverse reactions. M. perstans is prevalent in the tropics, with undefined clinical symptoms. We screened 4392 individuals for these infections in 212 Gabonese villages. The overall prevalence rates were 22.4% for Loa loa microfilariae, 10.2% for M. perstans, and 3.2% for mixed infection. These rates varied across the different ecosystems: forest, savannah, Lakeland, river (Ogouée), and equator. A correlation was found between the prevalence and intensity of microfilariae, while a negative relationship was found between clinical symptoms (pruritis, Calabar swelling) and the prevalence of Loa loa microfilaremia. This study confirms the spatial uniformity of the relationship between parasitological indices, and provides a map and baseline data for implementation of mass chemotherapy for these infections.
| Loa loa and Mansonella perstans are endemic filarial parasites in the central and western African rainforest. Loa loa infects 2 to 3 million people [1]. M. perstans is considered non pathogenic [2]–[3], although some clinical manifestations have been associated with M. perstans microfilaria [4], [5], [6] including ocular disorders [7], [8]. Interest in loiasis has grown during the last 30 years, for several reasons. First, in endemic areas loiasis is the second reason for medical visits, after malaria [1], [9]. Second, this infection mainly affects active young individuals, who contribute to agricultural productivity [10], and their health is often aggravated by co-infection by other parasites. Two-thirds of infected individuals are amicrofilaremic, despite subconjunctival migration of adults worms, suggesting immunological elimination of microfilariae [1], [11]. Severe adverse events can occur during treatment with diethylcarbamazine (DEC) and ivermectin in individuals with high-level microfilaremia, requiring close treatment monitoring and hindering mass administration of antifilarial drugs aimed at controlling other filariae in areas where Loa loa is co-endemic. This is not the case with M. perstans [12].
Many epidemiological studies of loiasis and Mansonellosis have been carried out throughout the western and central African forest block. These studies mainly focused on the distribution of loiasis and on the possible relationship between the prevalence and intensity of microfilaremia, in order to estimate the risk of adverse events during mass chemotherapy.
The prevalence of L. loa microfilaremia varies from country to country [13], as well as within a given country and even a given geographic area [14]. The highest prevalence is observed in forest areas and the lowest in savannah areas of both Gabon [15], [16] and Cameroon [17], [18], for example. Differences within a given geographic zone are directly linked to the bioecological specificity of a microzone [19]. These observations were recently used to create a predictive geographical model of loiasis endemicity based on satellite, vector habitat, prevalence, vegetation, temperature, relief, pluviometry and topography data [20]. However, when compared to field data, this model showed certain limitations [21].
A linear relationship between the prevalence and intensity of loiasis has been established. A high prevalence is indicative of intense L. loa infection and therefore a high risk of adverse events [22], [23]. The 20% threshold prevalence of microfilaremia at the community level corresponds to about 5% of high microfilaremia loads (>8000 mf/ml) and 2% of very high microfilaremia loads (>30000 mf/ml), the latter being the cut-off point above which there is a risk of severe adverse events during ivermectin treatment [24]. Owing to the difficulties of drawing regional maps based on microscopic analysis, a rapid method for evaluating the prevalence and intensity of Loa loa infection at the community level has been developed (RAPLOA: Rapid Assessment of Prevalence of Loa loa) [25]. RAPLOA is based on interviews assisted by photographs of adult worms in the eye, to detect subconjunctival migration of adult worms (which lasts 1 to 7 days), as reported by interviewees. A 40% prevalence of a history of eye worm corresponds to a 20% threshold prevalence of microfilaremia at the community level, 5% of high microfilaremia loads (>8000 mf/ml) and 2% of very high microfilaremia loads (>30000 mf/ml) [25]. Another clinical manifestation, Calabar swellings, was used to evaluate the risk of adverse events. This sign has shown to correlate with the prevalence of highly microfilaremic individuals [25].
The use of eye worm and Calabar edema to assess the risk of fatal side effects in patients with loiasis suggests a relationship between clinical symptoms and parasitological indices.
Most of these latter studies were performed in Cameroon, Nigeria, Republic of Congo and Democratic Republic of Congo [17]–[25], only a few concerning Gabon.
In Gabon, epidemiological surveys have identified five filarial species (L. loa, M. perstans, O. volvulus, M. streptocerca, and M. rodhaini), and yielded a preliminary map [12]–[16], [26]–[28]. L. loa is the predominant species and co-exists with M. perstans. The prevalence of microfilaremia varies across provinces and even within a given province, being higher in mountain forest than in savannah.
The aim of the present study was to obtain a fuller picture of the distribution of blood-borne filariae in Gabon, using both the wet blood film and concentration techniques, and to detect a linear relationship between the prevalence and intensity of loiasis and between clinical symptoms and parasitological indices. We therefore conducted a large survey, including all the country's ecological niches and recording the main clinical manifestations of Loa loa infection.
We surveyed rural Gabonese populations. The country is 800 km long and 20 to 300 km wide, consists of 80% rain forest, and is bordered to the west by the Atlantic Ocean. The forest zone extends from west to east, from the coastal basin with the grassland forest to the interior and north-eastern forest plates band, through a wide mountainous forest band from 60 to 100 km parallel to the coast. The south and southeast contain isolated areas of savannah and steppe. A coastal and continental marine ecosystem named lakeland is located around the mouth of River Ogooué (Figure 1) [29]. The population is about 1.5 million and there are 2048 villages located in 9 provinces. Rural populations are located along roads and rivers, and few villages have more than 300 inhabitants.
This survey was conducted during nine-month field missions between June 2005 and September 2008. For this survey, a stratified random sampling method was used, based on the 9 provinces. Twenty to 30 villages per province were randomly selected. The required sample size was calculated on the basis of an estimated prevalence of 5 to 10% (using n = ε2 [p (1−p)]/e2; with ε = 1.96 (alpha risk = 5%), e (precision) = 2% and p = expected prevalence; with n varying from 188 to 864). Within each village, individuals over 16 years of age having lived for at least one year in their village and who accepted blood sampling where included in the study. A free medical examination was offered and basic medicines were provided to all participants and non participants, if appropriate. All the villages were georeferenced (Figure 1).
The rationale of the study was explained and a one-page questionnaire was administered to all participants. We collected demographic data (age, sex and occupation), geographic data (name of the village, length of residence, department and province) and the medical history (eye worm, Calabar swellings, chronic arthralgia, pruritus, etc.) (Figure S1).
The study protocol was approved by the Ministry of Health. The Health Director, the governor of each province and the chiefs of each village received written information. Individual written consent was required before blood sampling. The results of the study were transmitted to the Ministry of Health.
Field laboratory facilities were set up in regional hospitals. Blood samples were collected, usually in the villages' healthcare centers, on a daily basis, into two 7-ml Vacutainer tubes containing EDTA (VWR International, France). The tubes were stored in the dark at +4°C before transportation to the field laboratory.
Due to the variability of microfilarial load, the analysis started systematically by direct examination of a wet blood film, followed by a concentration technique. Two experienced technicians read the slides separately, and the results were controlled by a parasitologist. Briefly, microfilariae were counted directly in a 10-µl wet blood film between microscope slide and coverslip, using an optical microscope equipped with a 10× objective. Parasitemia was expressed in microfilariae per milliliter (mf/ml) of blood. A modified Knott's concentration technique [30] was applied routinely to each sample, as follows: 1 ml blood was diluted with 9 ml PBS in a conical tube and 200 µl of saponin (2% w/v) was added to lyze red cells. The tubes were centrifuged (10 min, 500 g) and the supernatants discarded. The entire pellet was then examined under the microscope (10× objective) and microfilariae were counted. Parasite species were identified by their size and motility, and by the absence or presence of a sheath.
Loa loa prevalence rates were estimated nationwide. As mentioned above, the 20% threshold prevalence of microfilaremia is the cut-off above which serious adverse events are likely to occur, and corresponds to 5% of high microfilaremia loads (>8000 mf/ml) and 2% of very high microfilaremia loads (>30000 mf/ml). Thus, prevalence rates were calculated in each province, village and ecosystem as prevalence rates for microfilaremia loads >8000/ml and >30000/ml. The intensity of infection was estimated as described elsewhere [25]. The difference between the results of the two laboratory tests was calculated. The Chi2 test and Fisher's exact test were used as appropriate. Minitab 16 software was used to calculate Spearman's correlation coefficient for the association between parasitological and clinical parameters, and the Mann Whitney U test was used to compare mf intensity among groups. Univariate crude conditional maximum likelihood estimates of odds ratios (OR) and exact 95% confidence intervals (CI) were determined for each potential risk factor, using STATA software version 9.0 (Stata Corporation, College Station, USA). Multivariate logistic regression models stratified by the ecosystem were constructed from risk factors with a significance of ≤0.10 in univariate analysis, using a backwards stepwise elimination procedure. P values below 0.05 were considered statistically significant.
In total, 4392 individuals from 15 to 85 years old were enrolled in 212 villages, representing 10.7% of all villages in the country. The distance between villages ranged from 5 to 30 km. The sex ratio (M/F) was 0.88 (47.4% men and 52.6% women). About 58% of individuals were more than 45 years old and 63.9% had spent more than 10 years in their village. Farmers represented 69.8% of the population and hunters 10.2%. Around 80% of individuals were surveyed in the forest area, 10% in the savannah and the lakeland. The reported proportions of eye worm, Calabar swellings and pruritis were 29.3%, 11.2% and 22.4% (Table 1).
The wet blood smear identified 790 Loa loa and 116 M. Perstans microfilaremic subjects while the concentration technique detected 984 L. loa and 447 M. perstans microfilaremic subjects (difference of 19.7% for L. loa and 74% for M. perstans) (Table 2).
Most of these individuals who were positive only after concentration had microfilaremia below 100/ml, for both species (Table 3).
The overall prevalence rates of L. loa and M. perstans microfilaria were respectively 22.4% (95%CI: 21.2–23.7) (up to 57% in some villages), and 10.2% (95%CI: 9.3–11.1) (up to 67% in some villages), while 3.2% of subjects were coinfected (95%CI: 2.7–3.8) (Table 4, Table S1). The highest prevalence was found in the North Equator region (Figure 2A) for Loa loa (>10–20%) and along the Ogooué river for M. perstans (Figure 2B).
In the administrative regions, Estuaire province had the highest prevalence of L. loa (33.4%), M. perstans (22.9%) and co-infection (9.5%), while Ogooue maritime province had the lowest prevalence rates (respectively 12.1%, 1.4% and 0.5%) (Table 4).
In the ecological regions, the L. loa prevalence rate (Table 5, Figure 3A) was significantly higher (p<0.0001) in the forest (24.1%) than in the lakeland (17%) and savannah (14.8%). No difference (p = 0.4) was observed between lakeland and savannah. Moreover, within the forest ecosystem, the prevalence was significantly higher in grassland (28.9%) than in the mountain (20.5%), interior (24.3%) and north eastern (20.6%) forest regions (p<0.0002). In the same way, the M. perstans prevalence rate (Table 5) was significantly higher (p<0.0001) in the forest region (11.3%) than in lakeland (4.2%) and savannah (7.4%), and no difference (p = 0.053) was observed between lakeland and savannah. Within the forest ecosystem, the prevalence in the north-eastern forest (5.2%) was significantly lower (p<0.0001) than in the grassland (14.6%), mountain (14.9%) and interior forest (11.9%) (Table 5). Finally, most villages with high L. loa prevalence rates were located in the forest area (Table S1, Figure 3B).
In univariate analysis, males had a significantly higher risk of Loa loa infection than females (OR: 2.38, 95%CI: 2.05–2.75, p<00001), and the prevalence of Loa loa parasitemia increased linearly with age (p<0.00001) (Table 6). The prevalence of Loa loa microfilaremia was higher in hunters than in farmers and other occupational groups (p<0.04), and higher in individuals with eye worm (p<0.001) and those without Calabar swellings (p<0.014) (Table 6). Only gender was a risk factor for M. perstans microfilaremia, males having a significantly higher prevalence than females (OR: 1.89, 95%CI: 1.54–2.31, p<0.0001) (Table 7). In multivariate analysis, only age and sex remained significantly associated with Loa loa parasitemia, throughout the country and within the forest ecosystem (Table 8 and 9).
For clinical symptoms, only eye worm and Calabar swellings remained significantly associated with Loa loa parasitemia, both throughout the country and within the forest ecosystem (Table 8 and 9).
Microfilaremia in Loa loa-positive individuals ranged from 1 to 500 000 mf/ml (arithmetic mean: 5441 mf/ml; median: 900 mf/ml), while M. perstans microfilaremia ranged from 1 to 12 000 mf/ml (mean: 189 mf/ml; median: 18 mf/ml) overall. Mean L. loa microfilaremia was significantly higher in the forest ecosystem than in the savannah (median values: 3469 vs 1357; p = 0.048)) and similar to that in the lakeland (3469 vs 3140; p = 0.18). There was no difference between lakeland and savannah (3140 vs 1357; p = 0.8) (Table 10).
Likewise, mean M. perstans microfilaremia was significantly higher in the forest ecosystem than in the savannah (44 vs 4; p = 0.010) and lakeland (44 vs 0; p = 0.014) (Table 10).
The intensity of Loa loa microfilaremia did not vary with age countrywide (r = 0.249, p = 0.634), while it correlated with age in males (r = 0.915 p = 0.011) but not in females (r = 0.684 p = 0.134) (Figure 4). At the district level, the intensity of Loa loa microfilaremia did not vary significantly with age and sex.
The intensity of Loa loa microfilaremia (Figure 5A) correlated with the prevalence of microfilaremia nationwide (r = 0.215 p = 0.036) but not at the regional level (r = 0.163, p = 0.675). The intensity of microfilaremia also correlated with the prevalence of microfilaremia >8000 mf/ml (Figure 5B) and >30 000mf/ml (Figure 5C) (respectively r = 0.624, p = 0.0001 and r = 0.319, p = 0.002).
Furthermore, in the subpopulation of individuals with microfilaremia >8000 mf/ml, this relationship was observed in lakeland (r = 0.839, p = 0.001), savannah (r = 0.625, p = 0.027) and forest (r = 0.575, p = 0.0001), while in individuals with microfilaremia >30 000 mf/ml this relationship was only observed in the forest (r = 0.328, p = 0.005).
The prevalence of pruritis correlated negatively with the prevalence of Loa loa microfilaremia (r = −0.219; p = 0.032) (Figure 6A) but not with the intensity of Loa loa microfilaria (r = −0.144; p = 0.162) or with very intense microfilaremia (>30 000: r = −0.117; p = 0.255). Similarly, microfilaria >8000 mf/ml correlated negatively with the prevalence of prurits (r = −0.22; p = 0.027). Pruritis was associated with Calabar swellings (r = 0.578; p<0.001) and eye worm (r = 0.425; p<0.001). The prevalence of Calabar swellings (Figure 6B) correlated negatively with the prevalence of L. loa microfilaria (r = −0. 220; p = 0.031) but did not correlate with the intensity of microfilaria (r = −0. 061; p = 0.558), microfilaria >8000 (r = −0.185; p = 0.071) or microfilaria >30 000 (r = 0.093; p = 0.370); in contrast, it correlated positively with the prevalence of pruritis (r = 0.578; p<0.001) and eye worm (r = 0.335; p<0.001). The prevalence of eye worm (Figure 6C) did not correlate with the prevalence of microfilaremia (r = −0.05; p = 0.624) or with microfilaremia intensity (r = −0.137, p = 0.182), microfilaria >8000 (r = −0.139; p = 0.178) or microfilaria >30 000 (r = −0.106; p = 0.302), while it correlated positively with pruritis (r = 0.425; p<0.001) and Calabar swellings (r = 0.335; p<0.001). Interestingly, there was no relationship between these three symptoms and the prevalence of M. perstans microfilaria (r = −0.146; p = 0.155 for pruritis; r = −0.090. p = 0.385 for Calabar swellings; and r = −0.164; p = 0.110 for eye worm) or the intensity of M. perstans microfilaria (pruritis: r = 0.004; p = 0.971; Calabar swelling: r = −0.169; p = 0.100; eye worm: r = 0.182; p = 0.075).
We conducted a large-scale survey of two blood-borne filarial parasites, using direct examination and a concentration technique, in rural populations of 212 villages in Gabon, in order to map their distribution throughout the country, to characterize the modalities of population exposure and to explore the relationship between prevalence and intensity, and between clinical symptoms and parasitological indices.
The overall prevalence rates were 22.4% for Loa loa microfilariae, 10.2% for M. perstans, and 3.2% for mixed infection. These rates varied across the different ecosystems, the Ogooue River, and the equator. A correlation was found between the prevalence and the intensity of microfilariae, and between clinical symptoms (eye worm, Calabar swelling) and the prevalence of Loa loa microfilaremia.
As direct microscopic detection of microfilaria in wet blood films is not very sensitive, we combined two techniques for this survey, namely direct examination of 10 µl of blood (wet film) and prior concentration of 1 ml of blood. If we had used direct examination only, 19.7% Loa loa mf carriers and 74% of M. perstans carriers would have been missed. Most of these subjects had fewer than 100 mf/ml. Such underestimation may have implications for estimates of the risk of transmission and even for control programmes. Better sensitivity after sample concentration has been reported [14], [30], although this method is more tedious for large-scale surveys. Previous surveys used direct examination with larger volumes (30–50 µl [22], 50 µl [10] or 75 µl [19]).
The prevalence of Loa loa microfilaremia was 22.4% overall (up to 57% in some villages) while that of M. perstans was 10% (up to 67% in some villages). Gabon is thus a highly endemic country and a zone at high risk of fatal treatment complications. These prevalence rates are similar to those reported in southern Cameroon (up to 38% in the district of Elig-Mfomo) [13] and Equatorial Guinea (27%). This contrasts with Central African Republic (CAR) and Chad, where prevalence is lower (11% and 8.4% respectively). In DRC-Congo, Republic of Congo and Nigeria the prevalence rates range from 1.2% to 97% [13]. It should be noted that these prevalence rates are for specific regions of these countries, whereas our survey covered the whole of Gabon. The prevalence of Loa loa remains high in Gabon [15], [28].
Loa loa was highly prevalent in the north Equator (>20%), compared to the south (10–20%). Most areas crossed by the Ogooue River from the south-east (its source) to the north-west (towards the Atlantic Ocean) had an M. perstans prevalence of more than 10%, while other areas had a prevalence below 10%.
Among the three major ecosystems, forest had a higher prevalence of both parasites than savannah and lakeland. Differences were also seen among the different types of forest, as previously observed in Cameroon [19]. Geographic factors have been implicated in the prevalence of diseases like arteriosclerosis [31]. Sunlight might have a protective effect on some diseases [32], as ultraviolet B radiation stimulates the synthesis of vitamin D, which plays a role in immunity [33]. Geographic factors may influence filarial distribution by affecting the host immune system or the vector. The environment created by Ogooue River may affect the distribution and transmission of M. perstans. Although no soil studies around Ogooue River are available, studies in other areas have shown that low-pH soil, low organic soil content, salty soil, and wet soil promote Culicoides fly breeding [34], [35] while temperature may affect vector competence [36].
The prevalence of Loa loa microfilaremia was influenced by age in both sexes. In some parts of the country the prevalence continued to increase up to 70 years of age, while in others the prevalence appeared to plateau after 60 years. Males tended to be more microfilaremic than females, possibly because men are more exposed to chrysops bites due to their outdoor occupations (hunting, etc.), which become more intense with age, hence the correlation between age and microfilaremia. Genetic factors may also have a role [37]. Furthermore, the negative correlation of the intensity of microfilaremia with age in males may due to concomitant immunity against new incoming infection [38] or natural death of existing microfilariae [39].
In some areas of Cameroon where the general prevalence of microfilaremia exceeds 20%, approximately 5% of individuals have 8000 mf/ml and 2% have more than 30 000 mf/ml [24]. Similarly, in an area with a prevalence of 30%, 9% of carriers had >30 000 mf/ml, while in an area with a prevalence of 40%, approximately 16% of carriers had >8000 mf/ml and 5–6% had >30 000 mf/ml. Therefore, areas with a prevalence of more than 20% are considered to be at a high risk of treatment complications. Such studies have only been conducted in Cameroon [24], [25]. In this study, we observed a positive relationship between the prevalence and intensity of microfilaria. This suggests that the relationship between these two parasitological indices is spatially stable.
Clinical symptoms have also been used to predict the risk of side effects during mass chemotherapy. As previously described, eye worm and Calabar swelling have been found to correlate strongly with prevalence [25]. Photographs of ocular passage of the eye worm were used in previous studies [25]. Whether the lack of photographs in the present study influenced the accuracy of the patients' answers is not known. Yet, in our opinion, the use of photographs would yield a higher prevalence of amicrofilaremic subjects. Another striking observation is the negative correlation of pruritis and Calabar swelling with the prevalence of Loa loa but not M. perstans. Pruritis is a clinical sign of an allergic reaction. The negative relationship suggests that Loa loa filaria may induce desensitization. In Gabon, skin test reactivity against common allergens is low [40], while treatment of helminth infections increases skin test reactivity to mite antigens [41]. Similar observations have been made with M. perstans in Ugandan women [42]. A previous study in Gabon showed a high level of polyclonal IgE and Loa loa-specific IgG4 in permanent residents [27].
Further investigations are needed to elucidate the relation between filaremia and allergy in Gabon.
In conclusion, we provide a map of Loa loa and M. perstans microfilaremia in Gabon, and describe important relationships between parasitological indices and clinical manifestations. A clear and spatially uniform relationship was found between the prevalence and intensity of parasitemia. These data should be of use for planning mass chemotherapy.
|
10.1371/journal.pcbi.1005746 | A deep convolutional neural network for classification of red blood cells in sickle cell anemia | Sickle cell disease (SCD) is a hematological disorder leading to blood vessel occlusion accompanied by painful episodes and even death. Red blood cells (RBCs) of SCD patients have diverse shapes that reveal important biomechanical and bio-rheological characteristics, e.g. their density, fragility, adhesive properties, etc. Hence, having an objective and effective way of RBC shape quantification and classification will lead to better insights and eventual better prognosis of the disease. To this end, we have developed an automated, high-throughput, ex-vivo RBC shape classification framework that consists of three stages. First, we present an automatic hierarchical RBC extraction method to detect the RBC region (ROI) from the background, and then separate touching RBCs in the ROI images by applying an improved random walk method based on automatic seed generation. Second, we apply a mask-based RBC patch-size normalization method to normalize the variant size of segmented single RBC patches into uniform size. Third, we employ deep convolutional neural networks (CNNs) to realize RBC classification; the alternating convolution and pooling operations can deal with non-linear and complex patterns. Furthermore, we investigate the specific shape factor quantification for the classified RBC image data in order to develop a general multiscale shape analysis. We perform several experiments on raw microscopy image datasets from 8 SCD patients (over 7,000 single RBC images) through a 5-fold cross validation method both for oxygenated and deoxygenated RBCs. We demonstrate that the proposed framework can successfully classify sickle shape RBCs in an automated manner with high accuracy, and we also provide the corresponding shape factor analysis, which can be used synergistically with the CNN analysis for more robust predictions. Moreover, the trained deep CNN exhibits good performance even for a deoxygenated dataset and distinguishes the subtle differences in texture alteration inside the oxygenated and deoxygenated RBCs.
| There are many hematological disorders in the human circulation involving significant alteration of the shape and size of red blood cells (RBCs), e.g. sickle cell disease (SCD), spherocytosis, diabetes, HIV, etc. These morphological alterations reflect subtle multiscale processes taking place at the protein level and affecting the cell shape, its size, and rigidity. In SCD, in particular, there are multiple shape types in addition to the sickle shape, directly related to the sickle hemoglobin polymerization inside the RBC, which is induced by hypoxic conditions, e.g., in the post-capillary regions, in the spleen, etc. Moreover, the induced stiffness of RBCs depends on the de-oxygenation level encountered in hypoxic environments. Here, we develop a new computational framework based on deep convolutional networks in order to classify efficiently the heterogeneous shapes encountered in the sickle blood, and we complement our method with an independent shape factor analysis. This dual approach provides robust predictions and can be potentially used to assess the severity of SCD. The method is general and can be adapted to other hematological disorders as well as to screen diseased cells from healthy ones for different diseases.
| Sickle cell disease (SCD), also known as sickle cell anemia, is a type of inherited RBC disorder associated with abnormal hemoglobin S (HbS) [1]. When HbS molecules polymerize inside RBCs, due to lack of oxygen, they affect greatly the shape, elasticity, and adhesion properties of RBCs. Moreover, the RBCs become stiff and more fragile, with vastly heterogeneous shapes in the cell population [2], which makes this problem an ideal candidate for the examination of morphological heterogeneity. Unlike the normal RBCs, which are flexible and move easily even through very small blood vessels, sickle RBCs promote vaso-occlusion phenomena. Hence, SCD patients are afflicted with the risk of life-threatening complications, stroke and organ damage over time, resulting in a reduced life expectancy. According to a recent study [3], as of 2013 about 3.2 million people have SCD while an additional 43 million have sickle-cell trait, resulting in 176,000 deaths in 2013, up from 113,000 deaths in 1990, mostly of African origin. The prime hallmark of SCD is that is surprisingly variable in its clinical severity. Available methods for treating SCD are mainly supportive and mostly aim at symptom control, but lack the active monitoring of the health status as well as the prediction of disease development in different clinical stages [4]. Recent developments in advanced medical imaging technology and computerized image processing methods could provide an effective tool in monitoring the status of SCD patients. Indeed, Darrow et al. [5] recently demonstrated a positive correlation between cell volume and protrusion number using soft X-ray tomography. Van beers et al. [6] have also shown highly specific and sensitive sickle and normal erythrocyte classification based on sickle imaging flow cytometry assay, a methodology that could be useful in assessing drug efficacy in SCD.
Therefore, implementing an automated, high-throughput cell classification method could become an enabling technology to improve the future clinical diagnosis, prediction of treatment outcome, and especially therapy planning. However, there are several major technical challenges for automatic cell classification: 1) RBCs may touch or overlap each other or appear as clusters in the image, which makes it difficult to detect the hidden edge of cells. 2) The RBC region and the background may have low contrast in the intensity. 3) The boundaries of RBCs may be blurry due to the influence of imaging procedure. 4) Very complex and heterogeneous shapes of RBCs are present in SCD. 5) Artifacts may be present, for instance, dirt on the imaging light path, various halos and shading. 6) Finally, because RBCs lack a nucleus, methods utilizing the nuclei location as an apparent marker for cell counting and detection are not applicable.
The objective of the current work is to develop an automated algorithm for sickle RBC classification test, which may prove a powerful complementary clinical test for a) assessing patient’s disease severity via longitudinal tracking and patient-specific RBC mapping, and b) intervention strategies via personalized medicine treatment monitoring. Next, we present a brief overview of the state-of-the-art techniques involved in cell segmentation and classification.
Cell detection methods are prevalent, see e.g. [7–10], and some open source software (e.g., CellProfiler [11], CellTrack [12], Fiji [13] and CellSegm [14], etc.) for 2D and 3D cell detection and counting has emerged recently. However, in SCD we need cell classification, which is quite difficult due to the heterogeneous shapes of RBCs and the existence of touching and overlapped RBCs in the raw microscopy image, and existing software cannot be directly used to obtain RBC boundaries and cannot distinguish among the many different types of RBCs. Presently, there are two kinds of cell classification approaches, i.e., manual and automatic. In the manual approach one inspects the blood samples using the microscope to count the number of cells and examines the outliers in each frame. This, apparently, is subjective, labor intensive and time consuming for batch data processing. Coulter Counters and Laser Flow Cytometers enable cell sorting automatically by detecting the current and light refraction changes during cell pass through the channel. However, there are some shortcomings, such as the high cost and low processing speed (106 cells/hour), and in particular, these instruments are not suitable for the classification of heterogeneous cells. Thus, some cellular data analysis tools have been recently developed targeting this problem. For example, ACCENSE [15] adopted two clustering methods (k-means and DBSCAN) to facilitate the cellular classification automatically, however, the clustering performance relies on the properly initiation of parameters by hand; moreover, the performance of cell classification degrades for the clusters with different size and different density. More recently, RSIP Vision (http://www.rsipvision.com) has developed a commercial software package, allowing the recognition and count of RBCs by using a classifier to classify the hand-crafted morphological features; however, the main drawback of this method is that it requires domain-specific expertise on the feature extraction,f and is also a time-consuming procedure. In addition, the accuracy of the method has not yet been demonstrated for cell classification. Both of the aforementioned methods use machine learning tools but not deep learning algorithms. Likewise, some other similar studies on the HEp-2 cell classification based on the traditional machine learning methods have emerged recently, such as in [16] where multi-variant linear descriptors were adopted to extract the features and applied the SVM method to realize HEp-2 cell classification with an accuracy of 66.6%. Other methods include superpixels-based sparse coding method approach [17], k-nearest clustering method for red blood cell and white blood cell classification [18], etc.
Due to ineffectiveness of the aforementioned methods and given the recent advances of deep learning technique, Gao et al. [19] performed HEp-2 cell classification based on deep CNNs. Also, in order to improve the diversity of single HEp-2 cell data samples, Li et al. [20] carried out classification experiments based on deep CNNs by using four different patients’ datasets under different lighting conditions. However, for the currently available automated machine learning methods, which could be used for cell classification, the following are still drawbacks: 1) the classification studies are mostly directly based on already prepared single HEp-2 cellular data, hence, ignoring the initial key procedure of single cell extraction from the raw image data; 2) the adopted conventional machine learning methods are time consuming for the hand-crafted feature extraction and need specific human expertise; moreover, they need an accurate cell segmentation; 3) the classification accuracy is limited by the selected features and the performance of selected classifier. For our application, since RBCs of SCD exhibit special characteristics in terms of heterogeneous shapes and variant sizes, there is still no efficient tool that can be used to facilitate the automated inspection and recognition of various kinds of RBC patterns which are present in SCD blood.
The main focus of our paper is to develop an automated, high-throughput sickle cell classification method based on the deep Convolutional Neural Networks (dCNNs), taking advantage of the hierarchical feature learning goodness of dCNNs. The rest of this paper is organized as follows: In Section 3 we present our methodology, and in Section 4 we present the experimental results, a comparative analysis, and a discussion. Finally, in Section 5 we present the conclusion. S1–S4 Appendix contain some more details of the collection of raw data, the shape factor analysis, the CNN architecture, and deoxygenation method of sickle RBCs.
On the basis of the raw RBC microscopy image data from SCD patients following cell density fractionation [21] as shown in S1 Appendix, our automatic, high-throughput RBC classification assay consists of four main steps for the RBC-dCNN training: 1) Hierarchical RBC patch extraction, 2) Size-invariant RBC patch normalization, 3) RBC pattern classification based on deep CNN, and 4) Automated RBC shape factor calculation. A detailed overall training flowchart is shown in Fig 1. Each step of the algorithm is described below.
In the traditional learning-based cell image segmentation or classification method, the two most common techniques to obtain the training patches are the exhaustive pixel-wise sliding window with the same size method [22] and the ground truth bounding box method, e.g. Li et al. [20]. However, the major drawback of the pixel-wise block splitting method for the application of RBC classification is that it generates a large number of unwanted and redundant patches for the background and artifacts (e.g., dirt or debris in the light path) to feed for training and testing of the neural network. This redundancy and artifacts significantly hinder the efficiency of the method to take into account the high resolution of the microscopy data and large background area. The ground truth bounding box method was based on manual labeling of cells present in raw images, a process which is labor intensive and needs specific domain knowledge.
In addition, due to the fact that sickle cells are always heterogeneous in shape and at times touch or overlap, it can be difficult to obtain all single RBC patches by using the sliding or bounding box window with a fixed pixel size. Therefore, in our study, a hierarchical RBC patch extraction method was developed to overcome the above problems. The complete flowchart of the proposed method is shown in Fig 2.
Firstly, the raw microscopy images were divided into overlapped patches by using the sliding window technique, with the block size N * N. Then, the entropy containing in each image block was estimated by Eq (1) below:
E = - ∑ i = 1 L P i log P i , (1)
where L is the maximum grayscale level, Pi refers to the probability of occurrence for each intensity level that is encountered in the image block, and it can be derived from the ith histogram count f(i, j) divided by the amount of pixels in each subblock image (the size of the block is N), as shown in Eq (2) below:
P i = f ( i , j ) N 2 . (2)
We have employed the information entropy to measure the uncertainty in RBC regions and the background region; the high entropy regions were extracted as the ROI (region of interest), i.e. the RBC regions in the raw microscopy images. The detailed ROI extraction procedure is shown in Fig 3.
First, raw microscopy images (Fig 3A in high resolution were split into overlapped blocks. Next, the information entropy was calculated for all sub-blocks (including the edges and noises blocks). The blocks with high entropy are shown in white color in Fig 3B, where the entropy threshold (5.0) was obtained from our validation experiments on different datasets. The corresponding ROI mask image was generated by filling the holes and removing the artifacts with area smaller than a common RBC prior area (6*10*10) for the result of Fig 3B. The result is shown in Fig 3C with each color representing a single ROI region. Fig 3D shows the “cleaned” RBC ROI region result corresponding to the ROI mask image. The entropy estimation method can effectively extract the complete RBC regions from the raw images, especially for those RBC boundaries in a low intensity contrast. Moreover, it can also detect the RBC region correctly from various datasets regradless of their brightness differences. Thus, it can effectively overcome the shortcomings of the previous commonly used methods (e.g., Ostu, watershed and Sobel, etc.). To obtain the RBC patch images for the deep CNNs, the high-level ROI boundary is detected and by searching the minimum coordination of pixel (x0, y0) and maximum pixel coordination (x′, y′) from the boundary pixels, the ROI patches are illustrated as shown in Fig 3E.
It should be noted, however, that for the particular situation of overlapped and touching RBCs that may be present in the raw microscopy image, we may obtain some extracted ROI regions containing multiple cells; see the yellow smaller sized box in Fig 3F, where 8 ROI patches contain two or more RBCs, and the pink smaller sized box that includes all segmented single RBC patches. The subimages in the two boxes were obtained by calculating the corresponding bounding boxes of the ROI. Overlapping RBCs were removed from the input of deep CNNs in our work. Therefore, we only focused on the “touching” RBC separation problem by applying the random walk method [23] in conjunction with the distance transform [24] to generate the RBC boundary. This method can obtain the RBC seed points identification automatically. The specific separation procedure is shown in Fig 4.
Because of the RBC heterogenity in size, shape and orientation, the generated single RBC patches from section B were of different sizes (see Fig 3E). In addition, due to varying brightness and intensity contrast conditions during the procedure of raw RBC microscopy data collection, the background of RBC patch images appeared to differ among datasets. Currently, commonly used image scaling methods for the image size normalization are prone to reducing the RBC patch image fidelity (e.g., intensity contrast, noise and distortion), which will accordingly affect the RBC classification accuracy of the CNN. Therefore, to overcome the above issues, a size-invariant RBC patch normalization method based on statistic intensity linear mapping was employed. The algorithmic workflow is shown in Fig 5.
In our work, we adopted a deep CNN architecture with 10 layers, including 3 convolutional layers (C1, C3 and C5), 3 pooling/subsampling layers (P2, P4 and P6), dropout layers (D7 and D9, where p = 0.5) and a fully connected layer (F8). As a result of the computational efficiency, the grayscale RBC image patches were initially resized to 78 * 78. Next, these were then fed into the neural network. A ReLU non-linear activation function was then applied. Following the F7 layer, a logistic regression method combining the softmax function (see Eq (4)) with a cross-entropy loss function (see Eq (5)) was implemented to obtain the final learning probability and predicted labels. The softmax function can “squash” the obtained score vector Q = {qi|i = 1, 2, …, N} to a N-dimension probability vector δ(qi), so as to aid RBC classification efficiency.
Q ′ = δ ( q i ) = e q i / ∑ j = 1 N e q j , (4) D ( Q ′ , Q ) = - ∑ i = 1 N Q i log ( Q i ′ ) , (5)
According to different shape division level for the original RBC patches, two kinds of RBC labeling principles were employed in the experiment: coarse labeling(output = 5) and refined labeling (output = 8). Thus, the output layer had two different dimensions (5 or 8 categories). More details about the deep CNN architecture applied in this paper are shown in Fig 7 (see also S3 Appendix for the specific illustration of the layers of deep CNN).
As mentioned previously in the text, RBCs from SCD patients vary significantly in morphology/shape [25]. In the previous section, deep CNNs was applied to train and learn the diverse RBC patterns from RBC microscopy imaging data (see Table 1). Hence, by utilizing this deep CNNs we can classify sickle RBC in different types according to training. In addition to RBC type classification we perform shape factor analysis for each RBC type to further quantify specific RBC shape parameters derived from the contour analysis of the individual RBCs. Three kinds of shape factors were calculated in this work.The shape factors’ formulas and pseudo-code for the specific implementation of the automatic RBC shape factors quantification method are given in S2 Appendix.
On the basis of the above automated image-based shape factor analysis scheme, we can perform a comprehensive shape analysis for the classified RBCs or unclassified RBCs according to specific practical applications and requirements.
In this section, we conduct several experiments to evaluate the performance of the deep CNN used in the special RBC classification cases and present a comparative analysis of the results. In our experiments in order to validate the robustness of our methodology in dealing with different imaging data, we consider 434 raw microscopy images of 8 different SCD patients collected from two different hospitals. The number of images for each patient in different fractions is shown in Fig 8; all the images in different fractions (F1,F2,F3,F4 and UF) are of the same size (1920*1080 pixels) in TIFF format with 4 color channels. Based on the obtained raw images, 7206 single RBC image patches were extracted by using the proposed method in Section 3.1.Subsequently, all RBC patch images were normalized to the same size (78*78) by using the method described in section 3.2. Namely, all the RBC patch images were assigned to 8 different categories (discocytes, echinocytes, elongated, granular, oval, reticulocytes, sickle and stomatocyte) manually with the corresponding quantity of each RBC category presented in Table 1 as described in [26]. Conventionally, our definition of echinocytes is equivalent to echinocyte type II and III mentioned in [27]. Echinocyte type I is actually the “granular shape” we mention in this manuscript; moreover, wherever the state of oxygenation is not mentioned it implies “Oxy” state. We note that oval shape refers to the shape of the red cells and is not related to Southeast Asian ovalocytosis. This convention is consistent in our training of the dCNN model. A comparison study on the deep CNNs training model for two datasets with different number of patients’ data was conducted. The input data was enhanced with geometric transformations—a method also known as data augmentation technique. This technique adds value to base data by adding information derived from rotation, shifting or mirroring, illumination adjustment, etc., and introduces only a slight distortion to the images but without introducing extra labeling costs. A larger dataset can help evaluate and improve the robustness of RBC classification CNN model as well as restraining the common over-fitting problem. Thus, in our work, five types of data augmentation were performed on the normalized single RBC patch: rotate 90°, 180°, 270° and horizontal and vertical reflection.
In order to test the performance of the deep convolutional neural network model, we conducted systematic convergence studies with respect to the number of iterations and the learning rate; here we show some representative results. For the case of 4 patients (Exp_I), we evaluated the training error and loss in the configuration of different learning rates (0.01 and 0.03), batch size = 20, image size is 78*78 and weight decay is 0.01. In Fig 9, we observe that both the train and the loss errors decay with the increasing number of epochs, and the higher learning rate can accelerate the decay speed, see the corresponding plots of the loss and train error results for two comparative experiments (T1 and T2) with different learning rate settings. Moreover, another significant observation in Fig 9 is that both the train error and loss results start fluctuating after 15 iterations for T2 and 25 iterations for T1. In particular, the fluctuations in the loss increase as the number of iterations increases, but the train error has a relatively smaller fluctuation. In order to better understand the fluctuation problem (so-called “over-training” or “overfitting”), we optimized the batch size and use the “dropout” scheme proposed in [28] to overcome this problem. As described before, the dropout layer is implemented after the convolution layer (p = 0.5). Finally, when the number of iterations reaches 60, our RBC-dCNN model achieved optimal prediction performance. We plotted the two normalized confusion matrices with respect to different number of maximum iteration times (30 and 60) in Fig 10.
In Fig 10, we observed that the Discocytes and Granular classes have relative low prediction accuracy among the 8 classes of RBC before the convergence of loss and training error. However, when the maximum number of iterations was 60, there was a significant improvement in the accuracy of different class prediction due to further decay of the loss and training errors. Table 2 gives detailed performance analysis of the running time, train error, test error and loss with respect to different maximum iteration times based on Exp_I dataset.
Despite a learning model being trained to fit the statistics, the model cannot be assumed to have a successful predictive capability. This is due to the regularization which increases the performance, while the performance on test is optimal within a range of values of the regularization parameter. Thus, accurate evaluation of predictive performance is a key step for validating the precision and recall of a deep neural network classification model.
K-fold cross-validation is an effective way to measure the predictive performance for the deep CNNs model [29]. The K-fold cross-validation procedure is shown in Fig 11. First, the total RBC population was divided into k non-overlapped subsets with equal number of RBCs (here k was chosen to be 5). Then, for every fold or experiment, one of the 5 subsets was chosen as the validation set (green color data block) and the other k − 1 subsets were combined to form the training set (orange color data blocks). Finally, the average validation scores obtained from the five folds were calculated as the final prediction score. Every class of RBC images is divided into 5 equal subsets, the quantity of training data and validation data can in each class can be expressed as Eq (6).
{ Sum ( V ij ) = C ( i ) / 5 , Sum ( T ij ) = 1 - C ( i ) / 5 , i ∈ [ 1 , n ] , j ∈ [ 1 , 5 ] (6)
Where, C(i) is the number of RBC in the ith class, Vij describes the i-th class and j-th validation sub-dataset, and Tij is the corresponding training dataset.n can take the values of 5 or 8 for our studies. Finally, 5 folds can be generated by collecting the same subset from different classes alternately. For example, the j-th fold can be represented by Eq (7).
fold ( j ) = { V 1 j , V 2 j , … , V n j } ∪ { T 1 j , T 2 j , … , T n j } (7)
The main advantage in using k-fold cross validation is that each image is limited to one use during the validation process. This can effectively avoid the inaccurate and unstable phenomenon while artificially forcing multiple common samples into both training and testing.
Hence, in order to evaluate the general performance of our RBC-dCNN model, we performed 5-fold cross validation for the new datasets Exp_II (7 patients), in which we created a supplement for the number of echinocyte, granular, sickle and reticulocyte categories. To evaluate the performance of our deep CNN model for the SCD RBC classification and determine the importance of different types of RBCs present in SCD blood, we perform the experiments according to the following principles:
Precision
=
T
P
/
(
T
P
+
F
P
)
(8)
Sensitivity
=
T
P
/
(
T
P
+
F
N
)
(9)
Specificity
=
T
N
/
(
F
P
+
T
N
)
(10)
F
-
score
=
2
*
T
P
/
(
2
*
T
P
+
F
P
+
F
N
)
(11)
Accuracy
=
(
T
P
+
T
N
)
/
(
T
P
+
F
P
+
F
N
+
T
N
)
(12)
Here, TP, TN, FP and FN are, respectively, the true positive, true negative, false positive and false negative number of RBCs being classified for each class. The above five metrics can help us measure the dCNN’s performance from different perspective; e.g., the precision, or positive predictive value (PPV) can be viewed as a measure of a classifiers exactness. A low precision can also indicate a large number of False Positives. The sensitivity—also called recall or true positive rate(TPR)– measures the proportion of positives that are correctly identified; it can be viewed as a measure of a classifiers completeness. A low recall indicates many False Negatives; the specificity (SPC)—also known as true negative rate(TNR)– measures the proportion of negatives that are correctly identified. F1-score considers both precision and recall; it gets the best accuracy when it reaches 1, worst corresponds to 0. The ROC-AUC curve is a plot for TPR and NPR (Negative Positive Rate), which is explained in the experiments below.
In the following, the experimental results based on 5-fold cross validation for the two kinds of labeling datasets are presented respectively.
Evaluation of coarse-labeled RBC dataset (5 categories): In this experiment, all RBC patch images were coarsely labeled into 5 categories: 1) Dic+Ovl, 2) Ech, 3) El+Sk, 4) Grl, 5) Ret. In accordance with the cross validation scheme in Fig 11, the divided 5-fold cross validation datasets for 5 types of RBC and their corresponding evaluation results are given in Table 3.
As seen from Table 3, the mean accuracy for training of 5 types of RBC classification under different folds is 91.01%, and the mean evaluation accuracy is 89.28%. Here, in order to better visualize the discriminative capability of the training deep CNNs model for RBC classification and investigate the sensitivity of the deep CNN model to various RBC categories, the Receiver Operating Characteristic (ROC) curve was used to plot the true positive rate (TPR) against false positive rate (FPR) for different classes of the 5-fold test. The top-left corner of a ROC plot is the “ideal point” while the diagonal dashed line indicates random chance or luck probability. Therefore, the closer the curve followed the left-top border of the ROC space, the more accurate the test can be considered. We also computed the AUC (Area Under the Curve) for each ROC curve to evaluate the prediction performance of our RBC-dCNN model. Fig 12 shows the corresponding ROC-AUC results for RBC classification with 5 target categories. In the ROC-AUC plot, the average ROC curve was calculated and shown in blue color, and the corresponding averaged AUC for each fold is at least 0.97. Regarding the prediction performance of the five RBC classes, Granular and Echinocytes received a relative low AUC value, and the other two classes (“Discocytes+Oval”and “Elongated+Sickle”) obtained a high AUC value.
In addition, Fig 13A shows the corresponding confusion matrix, which can guide humans to observe the confusing classes in red circles; for instance, Ret and Ech are a pair of confusing classes, and the diagonal represents the correctly predicted number of each observation. The calculated sensitivity (right column) and precision (bottom) for each class in yellow color are consistent with the bars in the statistic Fig 13C. Except for these measures, three other measures are also computed for the performance analysis of the experiment, however, the difference in accuracy among the 5 type of RBCs is small because it refers to the true predictions (TP and TN) among the total validation dataset. However, high accuracy is not enough to demonstrate the goodness of the classifier, nor it can tell the sensitivity, precision, specificity and F-score. Therefore, it is necessary to explore these measures for a more in-depth analysis in the experiment. In Fig 13C, the Ret RBCs have a low recall (sensitivity), and the Ech get the lowest precision among the five classes. F-score can be applied to harmonize the above two evaluation metrics; the comparison results of F-score, precision and recall of 5 classes are shown in Fig 13B. Throughout all the evaluation measurements, we can obviously observe that the deep CNN model get a high accuracy and precision in predicting the different types of RBCs,in particular for “Dic+Ovl”, “El+Sk” and “Ret” types.
Evaluation of refined-labeled RBC dataset (8 categories): To evaluate the robustness of the deep CNN model in the application of more rich types of RBC classification, a refined labeling dataset “Exp_II” was generated, which included 8 types of RBC: Dic, Ech, El, Grl, Ovl, Ret, Sk and Sto. Similarly, 5-fold cross validation was carried out and the classification result is shown in Table 4. The mean evaluation accuracy for the 8 types of RBC classification was 87.50%.
The corresponding mean ROC-AUC result for the refined labeling test is shown in Fig 14. The average AUC value for 8 types of RBC is 0.94 as opposed to an average AUC value of 0.97 for the coarse labeling RBC classification. The RBC Categories (El and Ovl) got a relative low classification performance with an AUC value of 0.92. In addition, in Fig 15A, we see a more detailed confusion matrix for classification of 8 RBC categories. This shows the most confused classes (red circles) for each type of RBC; Fig 15b and 15c give a performance comparison among the 8 categories. Dic reached the best values for each metric and exhibited a sensitivity of 94.4% with high class-specific precision on testing sets of 1434 RBC images. Ovl type achieved the lowest precision and recall due to the misclassification with Dic and El types.
From the prediction result example in Fig 16 we can observe that some Ovl type RBCs (e.g. the RBC in red frame of Fig 16) are misclassified as Dic and the El type RBCs are prone to be classified to Ovl type and Sk type, e.g., the RBCs in blue and green frames in Fig 16.
Based on the proposed deep RBC-CNN model, we perform an independent RBC classification test on 8 raw microscopy images in the highest density RBC fraction i.e. fraction 4 (typically associated with severe SCD). Statistical quantification results for the number of different types of RBC are shown in Fig 17. Notice the significant heterogeneity of cell types even at the highest density fraction.
In addition to the above two experiments (EXP_I, EXP_II) on coarse-labeled and refine-labeled sickle RBC classification, in order to test the RBC-dCNN model for oxygenated and deoxygenated RBCs in SCD, we also performed a patient-specific experiment on the classification of a new experimental dataset that includes the previous coarse-labeled five catergories under normoxic conditions (Oxy) and a new catergory: “El+Sk under deoxygenation (DeOxy)”, see appendix for details on the experimental methodology. The specific experimental dataset (EXP_III) is shown in Table 5 (row 6) and it includes 81 El+Sk (DeOxy) RBCs, which after data augmentation correspond to an equivalent sample of 486 DeOxy RBCs.
In order to appreciate the differences in RBCs under Oxy and DeOxy conditions, we present in Fig 18 images of RBCs before and after deoxygenation. Even under Oxy, these particular RBCs have crenated shape because they are irreversibly sickled. However, upon deoxygenation we see that there is further polymerization of sickle hemoglobin (HbS) inside the RBCs, manifested by the roughening of the contours of RBCs as well as the alteration of the texture inside the RBCs. While the change in the overall shape of these deoxygenated RBCs is relatively small compared to their Oxy state, the differences are subtle and hence they present a new challenge for our dCNN.
Having this new mixed Oxy-DeOxy dataset (EXP_III) and the particular RBC inner pattern alteration characteristics, we carried out the dCNN model training and testing using the previous similar 5-fold cross validation schema, which involves four folds for training and one fold for testing. We have a total of 988 RBCs for training, which we arrange in 50 batches of 20 images each except the last one that has only 8 RBCs; see Fig 19A. So each batch contains 20 different RBCs, which may be in any of the six categories that the dCNN model should learn. The RBCs are randomly shuffled before input to dCNN. The hierarchical features can be extracted by dCNN layer-by-layer. For instance, the learned feature maps in the hidden 5th-layer for batch 1 is shown in Fig 19B. We observe that the convolutional operation can extract and highlight image features based on the raw image data field directly and hierarchically, such as detecting the image key points, edges, curves, etc. This is further illustrated in Fig 20, which presents a sequence of feature maps for different layers (layers 5, 6, 8 ad 10) corresponding to four different classes of RBCs in Oxy and DeOxy states. As we move to high layer numbers, we pick up more features from low level to high level, hence bridging the gap between high level representation and low level features. Within each layer different filters can be learned from the data to help extract different features. The images shown in Fig 20 correspond to arbitrary selection of filters for each layer. The original raw images are shown on the first column of Fig 20. Hence, the learned hierarchical convolutional features corresponding to variant learning filters play an important role in classifying RBCs in SCD, in particular for the classification of Oxy and irreversibly DeOxy sickle RBCs.
The final prediction result for the classification of deoxygenated RBCs is shown in Fig 21 for the elongated and sickle (DeOxy) category. If there is an obvious intracellular pattern change, then the accuracy of our trained dCNN model can obtain a high recall (93.8%) but a relatively low precision (60.0%). The main reason for this phenomenon can be justified as follows:
Taken together, the above observations imply that both intracellular patterns and RBC contours play a significant role in classification (see Fig 18, second row).
Our proposed RBC classification methodology also relies on the extraction of individual RBCs shape factors that is complementary to RBC classification. Two of the most prevalent shape factors are the Circularity Shape Factor (CSF) and Ellipticity Shape Factor (ESF) (Also see S2 Appendix) [30–32]. We computed the CSF and ESF shape factors for the classified RBCs obtained with the RBC-dCNN methodology (see Fig 22). The graph is a statistical visual representation of the classified RBCs (i.e., Elongated, Oval and Discocytes) within the ellipticity and circularity shape factor mapping. In addition to these two factors, we can implement in the workflow and compute any of the additional 12 shape factors mentioned in Table S-I to quantify SCD patient-specific RBC shape parameters. The results here are consistent with results described by Horiuchi et al. [30].
In summary, we have used patient-specific microscopy images to develop an automated, high-throughput, ex-vivo RBC classification method for the sickle cell disease based on pre-extraction of RBC region and deep CNNs. We employed a hierarchical RBC patch extraction method followed by a shape-invariant RBC patch normalization technique for the input of our deep nets, which can exclude unnecessary background patches and save time during both the training and the learning procedures. Moreover, our experiments for two kinds of labeling datasets (5 and 8 classes) based on different partition levels demonstrate the great capability and robustness of our RBC-dCNNs model on the classification of various RBC categories with characteristics of complex patterns and heterogeneous shapes without the need for hand-crafted feature pre-extraction. While most of the dCNN training was done based on oxygenated SCD RBCs, we also conducted the classification of deoxygenated RBCs, and demonstrate that our model can detect the deoxygenated RBCs with high accuracy capturing the subtle intracellular texture alterations. Furthermore, the explicit shape analysis at the end of the procedure can offer a robust morphological quantitative tool expanding the proposed framework to high-throughput, ex-vivo RBC classification.
Our program is written in Python language and C language, and it currently runs on CPUs, but it can also be updated to run on GPUs. It is mainly based on Python open-source libraries Theano, Numpy, SciPy and matplotlib, etc. The program takes only a few seconds on a standard desktop to test over a thousand RBCs using the trained deep neural network model.
In SCD, the shape of sickle RBCs is directly related to the polymerization process inside the RBC, which, in turn, depends on the de-oxygenation rate and hence the specific human organ where a sickle cell crisis may occur, consistent with clinical observations. The ability to perform high-throughput morphological classification utilizing deep CNNs of individual RBCs or other cell types, (e.g., white blood cells) opens up complementary avenues in medical diagnostics for highly heterogeneous cell populations such as in hematological diseases and stored blood used for transfusion.
The framework presented here is powerful but many aspects can be further improved in future work. For example, new work should aim to: (1) develop an accurate segmentation method for the overlapped RBCs in the microscopy image; (2) increase the dataset scale on the number of rare categories, e.g. sickle, granular, stomatocytes, etc.; and (3) build a golden standard library containing diverse SCD RBC categories. Given the success of dCNN in classifying deoxygenated RBCs, having been trained mostly with oxygenated RBCs, we believe that with the proper training of dCNN, the overall methodology for classification we propose could be effective in other hematological disorders, e.g., diabetes mellitus, elliptocytosis, spherocytosis, as well as in classifying other cells, e.g., cancer cells, and even detecting the activation state of platelets.
|
10.1371/journal.pcbi.1005737 | A single Markov-type kinetic model accounting for the macroscopic currents of all human voltage-gated sodium channel isoforms | Modelling ionic channels represents a fundamental step towards developing biologically detailed neuron models. Until recently, the voltage-gated ion channels have been mainly modelled according to the formalism introduced by the seminal works of Hodgkin and Huxley (HH). However, following the continuing achievements in the biophysical and molecular comprehension of these pore-forming transmembrane proteins, the HH formalism turned out to carry limitations and inconsistencies in reproducing the ion-channels electrophysiological behaviour. At the same time, Markov-type kinetic models have been increasingly proven to successfully replicate both the electrophysiological and biophysical features of different ion channels. However, in order to model even the finest non-conducting molecular conformational change, they are often equipped with a considerable number of states and related transitions, which make them computationally heavy and less suitable for implementation in conductance-based neurons and large networks of those. In this purely modelling study we develop a Markov-type kinetic model for all human voltage-gated sodium channels (VGSCs). The model framework is detailed, unifying (i.e., it accounts for all ion-channel isoforms) and computationally efficient (i.e. with a minimal set of states and transitions). The electrophysiological data to be modelled are gathered from previously published studies on whole-cell patch-clamp experiments in mammalian cell lines heterologously expressing the human VGSC subtypes (from NaV1.1 to NaV1.9). By adopting a minimum sequence of states, and using the same state diagram for all the distinct isoforms, the model ensures the lightest computational load when used in neuron models and neural networks of increasing complexity. The transitions between the states are described by original ordinary differential equations, which represent the rate of the state transitions as a function of voltage (i.e., membrane potential). The kinetic model, developed in the NEURON simulation environment, appears to be the simplest and most parsimonious way for a detailed phenomenological description of the human VGSCs electrophysiological behaviour.
| A unifying novel kinetic model of human voltage-gated sodium channels is proposed, which is able to reproduce in detail the macroscopic currents of all the ion-channel isomers, from NaV1.1 to NaV1.9. Its topology consists of six states (two closed, two open, two inactivated) and twelve transitions, and it is particularly well suited to be implemented in biologically inspired multi-compartmental neural cells and neural network models. It represents the most parsimonious kinetic model able to account for the most recently described electrophysiological features, and it has been developed by taking into account the experimental data gathered by published work reporting on each different isomer heterologously expressed in mammalian cell lines. Equipped with original differential equations, the model reproduces in detail the ion-channel macroscopic electrophysiological features with the minimal computational load.
| In computational neuroscience, modelling of ionic channel behaviour represents a fundamental step to develop biophysically detailed neuron models. As key players in the mechanisms underlying excitability, impulse conduction and signal transduction, both the voltage-gated and ligand-gated ion channels are essential components of the electrophysiological behaviour of each neuronal cell and, consequently, of the neural networks these cells make up [1–2].
Until recently the phenomenological behaviours of the voltage-gated ionic channels have been mainly modelled according to the formalism introduced by the seminal and forward looking work of Hodgkin and Huxley [3]. By exploiting their substantially fair approximation to the macroscopic currents of the voltage-gated ionic channels, the models derived by the Hodgkin-Huxley (HH) equations have been instantiated, even recently, in a multiplicity of realistic cellular and network models [4–7].
The overall simplicity and the relative light computational load of the HH formalism especially make them particularly well suited in modelling biologically detailed neural networks.
However, following the continuing achievements in the biophysical and molecular comprehension of these pore-forming transmembrane proteins, the HH formalism turned out to carry limitations and inconsistencies in reproducing in detail the ion-channels electrophysiological behaviour [8–12].
More detailed insights into the single channel kinetics provided by patch-clamp techniques [12–13] and into their molecular structure by means of x-ray crystallography [14] have greatly advanced our comprehension of ion channels to a degree difficult even to conceive when Hodgkin and Huxley developed their impressive and seminal research.
The more information about ion-channel gating has been achieved, the clearer is the need for models with explicit representation of single ion-channel states. In the HH formalism the gating parameters do not represent specific kinetic states of ion channels, and the HH model is sometimes not sufficient to capture various aspects of the channel behavior [15–16].
For the aforementioned reasons, Markov-type kinetic models have been developed to accurately represent an ionic channel as a collection of states and a set of transitions between them.
In recent years, many kinetic models have been developed, specifically focused on single isoforms of different channels or on specific details of them, which are derived from both functional and structural studies (e.g., [17–19]). The developed models, biophysically detailed and aimed at representing the molecular conformational changes of the transitions between states, usually carry a considerable number of states and related transitions. They are well suited to describe the detailed microscopic behaviour of single ion channels, but their computational load makes them much less suitable for the implementation in multi-compartmental conductance-based biologically detailed neurons and neural networks models.
As a result, although a variety of Markov-type kinetic models have been used to analyze the functional biophysical properties of single ion channels, yet very little of this information is used to develop conductance-based models of neural structures [20].
Thus, nowadays the two mutually exclusive options in modelling ionic channels rely on the HH formalism, which is global and computationally light, or the Markov-type kinetic models, which are specific, detailed and computationally heavy.
Among the ionic channels, voltage-gated sodium channels (VGSC) are probably the most studied and modelled voltage-gated ionic channels. They are directly involved in the cellular excitation and in the onset of the spike. In humans, nine subtypes or isomers (NaV1.1 to NaV1.9) of VGSC exist, each of them with peculiar kinetics and tissue distribution [1–2]. They exhibit a diversified and complex, membrane potential-dependent gating behavior [8], and even slight modifications of their gating kinetics by genetic mutations give rise to a number of severe human diseases in peripheral nerves, skeletal muscles, the heart and the central nervous system [2, 21].
This purely modelling study is aimed at developing a Markov-type kinetic model for VGSCs, which is detailed (accounting for different features of the VGSCs macroscopic current), unifying (accounting for all ion-channel isoforms) and computationally efficient (with a minimal set of states and transitions).
By exploiting experimental data gathered from previously published electrophysiological studies from different laboratories investigating single isomers of the VGSC, we derived a simplified common kinetic model for VGSCs, suitable to be adopted in biologically detailed simulation of neural structures.
The obtained results show that it is possible to develop a unifying kinetic model for VGSCs macroscopic currents by adopting a new simplified state diagram and novel equations describing the voltage dependence of the state transitions.
As we were mainly interested in modelling the macroscopic electrophysiologic behaviour of single isoforms of the sodium channel, we searched the literature for experimental data from single human VGSC (NaV1.1 to NaV1.9) α-subunits, heterologously expressed in mammalian cell line (usually Human Embrionic Kidney 293 cells) [22–29]. Studies on NaV1.8 and NaV1.9, whose transfection in non-nervous cell line is practically challenging, were performed in homozygous NaV1.8-cre mice Dorsal Root Ganglia neurons lacking endogenous NaV1.8 [30], and, respectively, in ND7/23 cells (a hybrid from mouse neuroblastoma and rat neurons) [31].
In some of the considered studies one or two β-subunits were also co-transfected: β1-subunit was coexpressed with NaV1.7 [29], and β3-subunit with NaV1.3 [25]; β1- and β2-subunits were both coexpressed in NaV1.1 [22] and in NaV1.2 experiments [23]. The electrophysiological experiments were conducted by means of the whole-cell patch-clamp method, usually at room temperature and our model corrected for the experimental temperature (by means of the temperature coefficient, Q10).
Modelled graphics of every VGSC with reference to the experimental ones are displayed in a Supporting Information file (S1 Appendix. Modelled graphics with reference to the original ones).
For each VGSC isomers, the following electrophysiological data were gathered and in turn reproduced by modelling:
When available, we also compared our simulations with the following electrophysiological data:
The ionic channel current is governed by Ohm's law, wherein conductance is determined by the fraction of channels in the open states, xO (0 to 1):
INa(t)=g¯Na⋅xO(t)⋅(V(t)−VNa),
(5)
where g¯Na is the sodium maximal conductance and VNa is the reversal potential of that ion. Transitions between the O (open), I (inactivated), and C (closed) states are described by conventional Markovian model equations [32] written for the fractions of channel, xO, xI, xC, to be in these states:
dxOdt=ACOxC−(AOC+AOI)xO
(6)
dxIdt=AOIxO−AICxI
(7)
xO+xC+xI=1
(8)
where AXY is the transition rate between the state X and the state Y. The topology of the single, general model and the transitions between its states have been searched for through progressive appoximations using heuristic optimization.
All simulated experiments were performed by means of NEURON version 7.4 simulation environment [33]. The kinetic equations were written and solved directly using KINETIC methods of NMODL language of NEURON, which is a derivative of the MODL description language of the SCoP package [34].
All virtual experiments were performed on an one-compartmental cylindrical 'soma' 50 μm long with a diameter of 63.66 μm, so that the membrane area was set to 10'000 μm2. The membrane capacitance was set at 1 μF/cm2 [35]. The maximal conductance density for each VGCS isomer inserted into the soma was arbitrarily set to 0.1 S/cm2, and the resulting ionic current density was measured in mA/cm2. The capacitive currents were subtracted from the total current in all the simulations. The time for single integration step (dt) was set to 0.025 ms.
At every step, the rate constants of each transition were multiplied by the temperature coefficient, Q10, calculated as follows:
Q10=3(T°−20°10°)
(9)
Original NEURON source code was developed to simulate the protocols needed to yield the electrophysiological features of the channels. The simulated voltage-clamp protocols are depicted in Fig 1.
The source code along with the virtual experimental procedures is provided and available as a ModelDB [36] entry (http://modeldb.yale.edu/230137).
All simulations were performed on an iMac desktop computer running a MacOS version 10.12.5 (™ and © 1983–2017, Apple Inc, Cupertino, CA, USA).
The developed code automatically supplied the appropriate graphics, which replicated the macroscopic currents and the electrophysiological relationships found in the experimental studies (see S1 Appendix. Modelled graphics with reference to the original ones). A both empirical and quantitative curve fitting method was then adopted to reconcile experimental and modelled data. Firstly, the curves and relationships obtained by the simulations were compared by visual inspection to the experimental ones. Then, the modelled curves were fitted for the Eqs (2) to (4), as appropriate, by using a nonlinear least-squares minimization method included in NEURON (Multiple Run Fitter subroutine), which in turn derives from the PRAXIS (principal axis) method described by Brent [37]. Finally, the parameters of the Eqs (2) to (4) of the modelled curves were compared to the experimental ones (Table 1). The agreement of the modelled data with the experimental ones was considered acceptable when the former were within two standard deviations of the latter.
In order to merely test the suitability of our model to be implemented in cell models, and to compare the features of the spikes it provides to those carried out by other channel models (built according to both HH or Markov-type formalisms), we inserted the developed channel model in three previously published cell models [11, 38–39] and performed a series of voltage-clamp and current stimulation simulations.
The previously published cell models were downloaded from the ModelDB [36] repository and are accessible, respectively, with the accession numbers: 3805, 98005, 180370. For model specifications see the corresponding papers [11, 38–39].
The first implementation sample, where our model is compared to a sodium channel model developed according to the HH formalism, is reported in the Result section. The remaining two samples are provided in a Supporting Information file (S2 Appendix. Examples of model channel implementation). The second sample compares the performances of our model with a more complex Markov-type model, the third sample shows how our model behaves in a morphologically detailed neuron model.
Table 1 shows the values of the main electrophysiological features reproduced by the model, alongside the available corresponding experimental values for comparison, in all VGSC subtypes. The displayed values are the parameters of the fitting of experimental and simulated curves with the Eqs (4) to (6). It can be noted that the most of the modelled data are within one standard deviation of the experimental values.
Graphics from a single isomer, namely the Nav1.5 VGSC, comparing experimental and modelled data, are shown in Fig 2. The complete set of graphics for each VGSC isomer, showing the curves obtained during the simulations, with reference to the corresponding experimental data in previously published electrophysiological studies, is provided as a Supporting Information file (S1 Appendix. Modelled graphics with reference to the original ones).
The most parsimonious state diagram able to account for the phenomenological behaviour of all the VGSC isomers was found to be a six-state one (Fig 3). It is arranged in two closed, two open and two inactivated states.
The second inactivated state (I2) was considered as a deeper inactivated state than I1, only connected to I1. This topology was found to be the simplest to consistently account for both the slow and fast inactivations, as well as for the two time constants of the recovery from inactivation.
The second open state (O2), only linked to C2, was added to reproduce in detail a second slower constant of decay from activation, usually detectable on the current-voltage curves. An alternative solution, which considered O2 as only linked to O1 (by analogy with I1 and I2), was discharged because it provided not realistic tail currents of deactivation.
Two closed states (C1 and C2) were found to be sufficient to faithfully reproduce the activation kinetics and the tail currents after a brisk repolarization.
All transitions between two consecutive states were considered reversible (with one exception, see below), and the paired forward and backward transitions were computed by equations carrying numerical values (coefficients) of the same order of magnitude. The only exception was the O1 to I1 transition, where the backward transition (I1 to O1) was described by an infinitesimal value, so that the O1 to I1 transition could be considered irreversible.
The dynamics of the different states of the channels are described by the following set of coupled ordinary differential equations:
dC1dt=I1C1*I1+C2C1*C2−(C1C2+C1I1)*C1
(10)
dC2dt=C1C2*C1+O1C2*O1+O2C2*O2−(C2C1+C2O1+C2O2)*C2
(11)
dO1dt=C2O1*C2+I1O1*I1−(O1C2+O1I1)*O1
(12)
dO2dt=C2O2*C2−O2C2*O2
(13)
dI1dt=I2I1*I2+C1I1*C1+O1I1*O1−(I1C1+I1I2+I1O1)*I1
(14)
dI2dt=I1I2*I1−I2I1*I2
(15)
Moreover, the states obey the law of mass conservation:
O1+O2+I1+I2+C1+C2=1
(16)
Since the studies by Hodgkin and Huxley [3], the voltage dependence of the rate transitions has been mathematically modelled (Fig 4A) as an exponential equation (black line), or as a sigmoid (blue line), or as a combined linear and exponential equation (red line).
In other cases [32, 40] a sigmoid curve with minimum and maximum asymptote has been adopted (Fig 4B inset), described by equations as below,
Aω=τminω+τmaxω⋅[1+exp(V−V1/2ωkω)]−1
(17)
where ω is the transition between two states, τminω and τmaxω are the two asymptotes, V1/2ω the hemiactivation voltage, and kω the slope which describes the voltage sensitivity of the transition rate.
By progressive optimizations, we found the most suitable general equation to be adopted in all the transitions was a sigmoid one. For most of the transitions, the minimum asymptote was conveniently set to zero, while in a few cases, notably for the O1 to I1 transition, it needed a non-zero value. Furthermore, to accommodate in detail the time course of the current-voltage curves, it was found appropriate to slightly modify the sigmoid, adding a bending at the beginning of the rising slope of the curve (Fig 4B).
This way, the modified sigmoid could be mathematically described as the combination of two sigmoids with opposite slope: the first one placed towards the hyperpolarized side and carrying a positive slope factor, the second one toward the depolarized side and carrying a negative slope factor.
As a result, the general equation adopted to describe this double sigmoid was set as follows:
Aω=Bhypω⋅[1+e(V−Vhypωkhypω)]−1+Bdepω⋅[1+e(V−Vdepωkdepω)]−1
(18)
where Bhypω, Vhypω and khypω are the magnitude, the hemiactivation and the slope factor, respectively, of the voltage dependence of the transition rate ω in the hyperpolarized region, and Bdepω, Vdepω and kdepω the corresponding values in the depolarized region.
With this formalism, the slope factor (k) in the hyperpolarized region assumes a positive value and a negative value in the depolarized one. In addition, when the transition rate is better described by a simple sigmoid, which happens in most of the transitions, one of the two terms of the general equation can be conveniently dropped.
The complete set of parameters values for the simulation of all the sodium channel isomers is provided in Table 2.
As a general rule, according to previously proposed Markov-type channel models [e.g., 27, 41], the backward and forward transitions between two consecutive states are described by equations with opposite slopes. This arrangement is usually adopted to account for the wide differences in state occupancy of the channel at different voltage values.
Yet, in modelling the voltage-intensity curves and relations, we obtained more realistic results by adopting for the activation sequence (C1 to C2 to O1 and reverse) equations with identical slopes of the main sigmoid. Moreover, in paired forward and backward transitions an identical hemivoltage point was found to consistently fit the experimental data, as well as a shift of the transitions between C1 and C2 towards more depolarized values than the transitions between C2 and O1. As an example, Fig 5 shows the voltage-intensity curves of NaV1.2 (Fig 5A) and NaV1.9 (Fig 5B). This is an instance of the most divergent electrophysiological behaviour in two channel isomers, yet the model is able to reproduce in detail the real data in both the cases. The transition rates dependences from voltage in C1 to C2 transition (green), C2 to C1 (yellow), C2 to O1 (red), O1 to C2 (purple), and O1 to I1 (black) are depicted in Fig 5C and 5D. The forward and backward transitions between two consecutive states have identical, not opposite, slope values and identical hemiactivation points. The transitions between C1 and C2 are shifted to more depolarized values compared to the transition between C2 and O1.
In addition, the backward transitions (C2 to C1, and O1 to C2) are described by a double sigmoid, whereas the corresponding forward transitions by simple sigmoids. Here the double sigmoid of the backward transitions accounts for the short time constants of deactivation recorded as tail currents (right end of the curves in Fig 5A and 5B). The higher values, indeed, of the backward transition rates at more hyperpolarized voltages drive the channel into more closed states with short latency during brisk repolarizations.
Fig 6 shows the plot of the simulated results obtained during recovery from inactivation (repriming) in NaV1.2.
A relatively long depolarizing conditioning pulse sets all the channels up into inactivated states (Fig 1C). A following repolarizing step, variable in duration, enables the transition from inactivated to closed states (recovery from inactivation) in a proportion of channels, which increases with the duration of the repolarizing step. The subsequent depolarizing test pulse probes the proportion of the channels having recovered from the inactivation. The progressive recovery with increasing duration of the repolarization is shown in Fig 6A. In the graphic of Fig 6B, the relative amplitude of the transient current following the test pulse is drawn against the logarithm of the duration of the repolarizing step. The fast and slow time constants of recovery depend on the interplay between the two inactivated states I1 and I2 during the first depolarizing step and the following repolarizing phase. As can be seen in Fig 6C, where the fractions of I1 (red line) and I2 (blue line) states are plotted against time, a step depolarization (inset) suddenly drives almost all channels into I1 state. As the depolarization lasts (100 ms in this simulation) an increasing fraction of channels slowly moves to the I2 state. When repolarizing, the exit from the inactivated states follows a fast (I1, red) and a slow (I2, blue) course, which together account for the two time constants of recovery. By fine-tuning the parameters of I1 to I2 transition, and those of I2 to I1 transition, the relative fractions of inactivated states and the slow time constant of recovery, respectively, can be adjusted to consistently reproduce the experimental data.
This subsection is only intended as a not exhaustive proof of concept of the feasibility and suitability of the proposed channel model to be implemented in different types of computational models. As such, no in-depth exploration of the implementations here presented has been performed.
In this study, we developed a single unifying deterministic Markov-type kinetic model which consistently accounts for the macroscopic electrophysiological behaviour of all human VGSC isomers, NaV1.1 to NaV1.9.
To date, the developed general model appears to be the simplest and most parsimonious one, able to account in detail for a number of electrophysiological features derived from experimental data, and suitable to be implemented in biologically detailed, conductance-based neurons and neural networks models.
The kinetic model here proposed bridges the gap between what is available from more recent electrophysiological studies and what is needed to construct reliable VGSCs models for use when simulating neural activity [20, 44]. In addition, its implementation in different types of neuron models has been proven straightforward and suitable.
Moreover, though the model is targeted to human VGSCs, its versatility makes it easily of use to simulate VGSCs from other species, provided that experimental data are available to constrain its parameters.
In modelling studies the level of approximation to the physical reality can be made strictly dependent on the temporal and spatial scale and on the complexity of the investigated issue. Thus, when dealing with the macroscopic scale of multiple interacting neural networks, simplified neurons are usually adopted [45–46], without even considering the types and amount of ionic channels which in real neurons are responsible of the electrical behaviour of the cell. On the other hand, in ion-channel modelling, the microscopic biophysical detail can be moved forward to account for the minimal displacements of the sensory domain of the pore-forming protein (e.g., [47]) in response to a transmembrane voltage change. Yet, the biophysical detail also brings a considerable computational load which limits the use of these detailed ion-channels models in simulations of neurons and large networks of those. In this study, in accordance with a common practice within computational neuroscience, a third intermediate way has been adopted, focused on the mesoscopic scale of the electrophysiological behavior of the VGSC. This modelling study, indeed, exclusively deals with the macroscopic ionic currents of the VGSC, which are of interest when building biophysically detailed neurons and neural network models, rather than their discrete and stochastic counterparts at the molecular level.
In their pioneering and seminal works, Hodgkin and Huxley combined the voltage-clamp techniques and quantitative modelling to provide a deterministic and continuous description of the macroscopic ionic currents [3]. They were able to clarify the nonlinear behavior of ions permeation through the cellular membrane in response to membrane depolarizations, and to disclose the relationship between these ions fluxes and the axonal spike. As a consequence, a wealth of data about the electrophysiological behaviour of the excitable membranes were provided with a fairly accurate mathematical description: from the form, amplitude and velocity of a propagated action potential to the subthreshold depolarizations, to the refractory period after a spike, to the inward sodium and outward potassium movements associated to an impulse.
However, thanks to the patch-clamp techniques [13], and the discovery of the molecular structure of the pore-forming proteins [14], it became clear that excitable membranes are studded with discrete ion channels undergoing random fluctuations between open and closed stable states [8]. In this respect, the HH formalism appeared as a simple macroscopic and deterministic description of a phenomenon that ultimately arises from the microscopic and stochastic behaviour of the system [9].
Markov-type models have been proposed as efficient kinetics scheme, suitable to capture the essential properties of a number of neural structures, like voltage-gated channels, transmitter-gated channels and second messenger-activated channels [40].
A Markov model represents an ion channel as a collection of states and a set of transition probabilities between them, and rely on the assumptions that: a) the configuration of a channel protein can be operationally grouped into a set of distinct states separated by large energy barriers, and b) the probability of state transitions is dependent only on the present state occupied [1, 40].
The various states represent a sequence of protein conformations that underlies the gating of the channel. The time evolution of the probability of state Si is described by the Master equation [48]:
dP(Si,t)dt=∑j=1nP(Sj,t)P(Sj→Si)−∑i=1nP(Si,t)P(Si→Sj)
(25)
where P(Si, t) is the probability of being in a state Si at the time t, and P(Si→Sj) is the transition probability from state Si to state Sj.
In the limit of large numbers of identical channels, the quantities given in the master equation can be reinterpreted. The probability of being in a state Si become the fraction of channels in state Si, noted si, and the transition probabilities from state Si to state Sj becomes the rate constants, rij, of the reactions
rijSi⇄Sjrji.
(26)
In this case, the master equation can be rewritten as:
dsidt=∑j=1nsjrji−∑i=1nsirij
(27)
which is a conventional kinetic equation for the various states of the system [49].
Stochastic Markov models, as in Eq (25), are adequate to describe the stochastic behaviour of ion channels as recorded using single-channel recording techniques [50]. In other cases, where a large number of ion channels are involved, as in the whole-cell patch-clamp recordings here considered, the macroscopic currents are continuous and more adequately described by conventional kinetic equations, as in Eq (27).
In the former case, derived from single-channel recordings, a Markov model is usually designated starting from ion-channel molecular representation, with each state of the model corresponding to a different configuration of the molecule. This approach gives a better understanding of the biophysical structure and functioning of the channel, and, by taking into account also the smaller gating currents, it details even the minimal, non-conducting molecular displacement.
Stochastic Markov models derived from single-channel recordings in ligand-gated ion channels have proven to be able to solve the inverse problem, that is the direct fitting of the models with raw data, with provision of estimates for rate constants and estimation of the errors for those estimates [51–52].
However, it is also possible to take a signal-processing approach to the design of Markov models, [20, 49, 53]: the required model is the minimal model that represents with sufficient accuracy the response of the channel to the stimulation protocols. This approach leads to more economical models, more suitable for numerical simulations of large collections of channels and of neurons, and was followed in the present paper.
In particular, this phenomenological approach is more reminiscent of the empirical and classical fitting of ionic macroscopic currents developed by Hodgkin and Huxley [3]. It could be argued that such approach is more similar to the fitting of a curve and hardly suitable to reveal the finest details of the biophysical features of the channel.
Yet, this is specifically in agreement with the main goal of the present study, which was to develop a model without the structural HH limitations and able to include the most recent experimental data on the macroscopic currents of VGSCs.
The phenomenological approach is intended to develop the smallest number of states and transitions necessary to replicate the electrophysiological VGSCs behaviour. Consequently, there is no exact correspondence of the model states with the physical states of the channel. In other words, in our phenomenological model the states do not represent single physical events, but each of them should be considered as an aggregate of molecular configurations suitable to be treated as a functional entity. For example, while a series of four (or more) closed states are commonly hypothesized (usually in adherence with the tetrameric structure of the proteic channel) before an open state can develop following a depolarizing step, our model collapses them all in only two. Two closed states, indeed, are necessary and sufficient to deterministically reproduce: a) the tail currents after a brisk repolaritation, b) the kinetics of the activation sequence, c) the kinetics of fast inactivation.
The aim to develop the computationally lightest model allows us to make one transition practically irreversible (from O1 to I1), and releases the phenomenological model from the principle of microscopic reversibility, like other kinetic schemes based on macroscopic currents (see [54], for a recent example). Microscopic reversibility, indeed, only holds when the states are elementary processes (collisions, molecules, elementary reactions, etc). On the other hand, most, but not all, ion channels obey the law of microscopic reversibility [55], and the law is only true at genuine equilibrium, which could not be the case when some sort of external energy supply is involved (in this case, an ionic gradient) [55].
Kuo and Bean [41] proposed a Markov-type model of VGSC incorporating the results of their study on the kinetics of the recovery from inactivation of sodium channel in rat hippocampal CA1 neurons. Since then, the model or its variants have been adopted as a more detailed alternative compared to the HH models [11, 56–59]. Yet, being provided with 12 states and 32 transitions, its computational load makes it quite heavy to be implemented in multi-compartmental neuron models and networks of those.
Mimicking the open probability of the HH model, an 8-state Markov-type kinetic model of voltage-gated ion channels has also been proposed [59]. More recently, aimed at modelling the slow inactivation of VGSC, a new version of the model has been proposed [27] for the isomer NaV1.5, derived from the 8-state model by Milescu et al [59]. To account for the slow inactivation, 4 inactivated states were added, and all the transitions between states (similarly to the present model) were made voltage-dependent. The resulting model (Figure 3B in [27]) fits very smoothly the complex electrophysiological behaviour of NaV1.5, yet it is equipped with 12 states and 34 transitions.
Compared to the model by Zhang [27], the one here proposed is able to simulate in detail the phenomenological behavior of NaV1.5 as well as of each other VGSC isomers with a significantly lower number of states and transitions (6 and 12, respectively).
In recent years, the increasing availability to the scientific community of powerful computational systems [44, 60] able to process, even in parallel, huge amount of data in relatively short time, has promoted the development of simulations of complex neural structures [4–7], equipped with large amounts of neural cells and synaptic connections. These simulations tried to be realistic and biologically inspired as much as possible, according to a bottom-up plan, and they succeeded, indeed, in replicating a number of electrophysiological experimental features of the cells.
In such bottom-up approaches, however, the value of detailed kinetics of ionic channels, as building blocks of cells electrophysiology, cannot be ignored. Taking examples from the VGSC here described, the presence of complex kinetics with fast and slow inactivations must be considered, as they have direct effects on the recovery from inactivation and, consequently, on the refractory time of the cells which, in turn, affects the ability of the cell to fire repetitive action potentials.
In addition, it should be considered that biologically inspired models have to consistently deal with the diversity and variety of the different channel isoforms. A differential topographic clustering of distinct ionic channels isomers, indeed, has been described also in subcellular compartments [61]. In real neurons, indeed, also subtle differences in ion-channels kinetics do have relevance. This is mainly showed with the clearest evidence by the effects of genetic mutations affecting ion-channels genes, considered pathogenic in a number of channelopathies. As regard to the VGSC, indeed, even slight differences in kinetics sustained by the mutation can give rise to severe diseases [21–22, 30].
The present work is a purely modelling study, and the proposed model relies on previously published experimental data. On one hand, this limits the genuine interpretation the modeller can actively draw from the raw experimental data. On the other hand, it is probably not affordable for a single laboratory to conduct research on the electrophysiological behavior of all the VGSC isomers, and data have necessarily to be collected from different studies.
Lack of interaction with experimenters also acts in the opposite way, as some suggestion of not canonical experimental paradigm (e.g. the adoption of different levels of repolarization during inactivation protocols), arisen during modelling studies, cannot be performed.
Moreover, different laboratories often perform experiments with not exactly similar protocols, and in some cases not all the useful experimental data are available. One example is the slow component of the recovery from inactivation (Table 1), where the second slower time constant of recovery can be only discernible by electrophysiological protocols carrying a long conditioning pulse, followed by longer repolarizing intervals, before the test pulse. A further example is the two time constants of fast inactivation or the ultra-slow inactivation, which have been quantitatively reported on in only few studies.
A further limitation is that some less investigated VGSC electrophysiological features, like the resurgent currents, have not been accounted for in this study.
An ensuing follow-up of the present study will be to evaluate the suitability of the proposed model as a general kinetic model for all voltage-gated ionic channels with similar molecular structure (four-metameric pore-forming proteins with six transmembrane domains in each metamere), calcium and potassium ion channels in primis.
|
10.1371/journal.pgen.1003390 | Genetic and Genomic Architecture of the Evolution of Resistance to Antifungal Drug Combinations | The evolution of drug resistance in fungal pathogens compromises the efficacy of the limited number of antifungal drugs. Drug combinations have emerged as a powerful strategy to enhance antifungal efficacy and abrogate drug resistance, but the impact on the evolution of drug resistance remains largely unexplored. Targeting the molecular chaperone Hsp90 or its downstream effector, the protein phosphatase calcineurin, abrogates resistance to the most widely deployed antifungals, the azoles, which inhibit ergosterol biosynthesis. Here, we evolved experimental populations of the model yeast Saccharomyces cerevisiae and the leading human fungal pathogen Candida albicans with azole and an inhibitor of Hsp90, geldanamycin, or calcineurin, FK506. To recapitulate a clinical context where Hsp90 or calcineurin inhibitors could be utilized in combination with azoles to render resistant pathogens responsive to treatment, the evolution experiment was initiated with strains that are resistant to azoles in a manner that depends on Hsp90 and calcineurin. Of the 290 lineages initiated, most went extinct, yet 14 evolved resistance to the drug combination. Drug target mutations that conferred resistance to geldanamycin or FK506 were identified and validated in five evolved lineages. Whole-genome sequencing identified mutations in a gene encoding a transcriptional activator of drug efflux pumps, PDR1, and a gene encoding a transcriptional repressor of ergosterol biosynthesis genes, MOT3, that transformed azole resistance of two lineages from dependent on calcineurin to independent of this regulator. Resistance also arose by mutation that truncated the catalytic subunit of calcineurin, and by mutation in LCB1, encoding a sphingolipid biosynthetic enzyme. Genome analysis revealed extensive aneuploidy in four of the C. albicans lineages. Thus, we identify molecular determinants of the transition of azole resistance from calcineurin dependence to independence and establish multiple mechanisms by which resistance to drug combinations evolves, providing a foundation for predicting and preventing the evolution of drug resistance.
| Fungal infections are a leading cause of mortality worldwide and are difficult to treat due to the limited number of antifungal drugs, whose effectiveness is compromised by the emergence of drug resistance. A powerful strategy to combat drug resistance is combination therapy. Inhibiting the molecular chaperone Hsp90 or its downstream effector calcineurin cripples fungal stress responses and abrogates drug resistance. Here we provide the first analysis of the genetic and genomic changes that underpin the evolution of resistance to antifungal drug combinations in the leading human fungal pathogen, Candida albicans, and model yeast, Saccharomyces cerevisiae. We evolved experimental populations with combinations of inhibitors of Hsp90 or calcineurin and the most widely used antifungal in the clinic, the azoles, which inhibit ergosterol biosynthesis. We harnessed whole-genome sequencing to identify diverse resistance mutations among the 14 lineages that evolved resistance to the drug combination. These included mutations in genes encoding the drug targets, a transcriptional regulator of multidrug transporters, a transcriptional repressor of ergosterol biosynthesis enzymes, and a regulator of sphingolipid biosynthesis. We also identified extensive aneuploidies in several C. albicans lineages. Our study reveals multiple mechanisms by which resistance to drug combination can evolve, suggesting new strategies to combat drug resistance.
| The evolution of drug resistance is a ubiquitous phenomenon that has a profound impact on human health. With the widespread deployment of antimicrobial agents in both clinical and environmental settings, the rate at which resistance evolves in pathogen populations far outpaces the rate at which new drugs are developed [1], [2]. Drug resistance threatens the utility of the limited arsenal of antimicrobial agents. The economic costs are staggering and exceed $33 billion in the United States alone to cover treatment of drug-resistant infections in patients, eradication of resistant pathogens in agriculture, and crop losses to resistant pests [3]. The evolution of resistance to antifungal drugs is of particular concern given the increasing incidence of life-threatening invasive fungal infections, and the limited number of antifungal drugs with distinct targets [4]. Unlike for antibacterials, fungal-specific drug targets are limited, in part due to the close evolutionary relationships of these eukaryotic pathogens with their human hosts, rendering most treatments toxic to the host or ineffective in combating infections [5]. Even with current treatment options, mortality rates due to invasive fungal infections often exceed 50%, and fungal pathogens kill as many people as tuberculosis or malaria [6], [7]. Thus, there is a pressing need to develop new strategies to enhance the efficacy of antifungal drugs and to minimize the emergence of drug resistance.
A powerful strategy to extend the life of current antimicrobial agents is drug combination therapy [8]. Combination therapy has the potential to minimize the evolution of drug resistance by more effectively eradicating pathogen populations and by requiring multiple mutations to confer drug resistance [9]. Great success has been achieved with combination therapy in the treatment of HIV [10]–[12], and it is currently the recommended strategy for treatment of tuberculosis and malaria [13], [14]. Combination therapies have been less well explored in the clinic for fungal pathogens. However, targeting cellular regulators of fungal stress responses has emerged as a promising strategy to enhance the efficacy of antifungal drugs and to abrogate drug resistance [5], [15]. Two key cellular regulators that are critical for orchestrating cellular responses to drug-induced stress are Hsp90 and calcineurin. The molecular chaperone Hsp90 regulates the stability and function of diverse client proteins [16], [17], and controls stress responses required for drug resistance by stabilizing the protein phosphatase calcineurin [16], [18]–[21]. Compromise of Hsp90 or calcineurin function transforms antifungals from fungistatic to fungicidal and enhances the efficacy of antifungals in mammalian models of systemic and biofilm fungal infections [15], [22]–[24], suggesting that combination therapy with azoles and inhibitors of Hsp90 or calcineurin may provide a powerful strategy to treat life-threatening fungal infections.
Targeting fungal stress response regulators holds particular therapeutic promise for enhancing the efficacy of the azoles, which are the class of antifungal drug that has been used most widely in the clinic for decades. Azoles block the production of ergosterol, the major sterol of fungal cell membranes, by inhibition of lanosterol demethylase, Erg11, resulting in a depletion of ergosterol and the accumulation of the toxic sterol intermediate, 14-α-methyl-3,6-diol, produced by Erg3 [25]. The azoles are generally fungistatic, causing inhibition of growth rather than cell death, and thus impose strong selection for resistance on the surviving fungal population [26]; as a consequence, resistance is frequently encountered in the clinic [27]. Azole resistance mechanisms fall into two broad classes: those that block the effect of the drug on the fungal cell and those that allow the cell to tolerate the drug by minimizing its toxicity [5]. The former class of resistance mechanisms includes upregulation of drug efflux pumps [28], or mutation of the azole target that prevents azole binding [29]. The latter class includes loss-of-function mutations in ERG3, which encodes a Δ-5,6-desaturase in the ergosterol biosynthesis pathway; Erg3 loss-of-function blocks the accumulation of a toxic sterol intermediate, conferring azole resistance that is contingent on cellular stress responses [16], [30]. Azole resistance acquired by loss of function of Erg3 or by many other mutations is exquisitely dependent on Hsp90 and calcineurin [16]; inhibition of these stress response regulators enhances azole sensitivity of diverse clinical isolates, and compromises azole resistance of isolates that evolved resistance in a human host [16], [18], [23], [31]. Inhibition of Hsp90 or calcineurin with molecules that are well tolerated in humans can impair the evolution of azole resistance [16], [20], though the potential for evolution of resistance to the drug combinations remains unknown.
Azole resistance mechanisms have been studied most extensively in the opportunistic fungal pathogen Candida albicans and the model yeast Saccharomyces cerevisiae. C. albicans is the leading cause of death due to fungal infection [32], and the fourth leading cause of hospital-acquired infectious disease [7], [32]. It is a natural member of the mucosal microbiota of healthy humans, but can cause life-threatening illness in immunocompromised individuals, such as transplant recipients and those infected with HIV [7], [33], [34]. Drug resistance can readily evolve in C. albicans in the laboratory and the clinic, and molecular studies have revealed a diversity of resistance mechanisms [35]. Molecular studies with C. albicans are hindered by its obligate diploid state, lack of meiotic cycle, unusual codon usage, and inability to maintain plasmids [36], thus complementary experiments are often performed with its genetically tractable relative, S. cerevisiae, with which it often shares drug resistance phenotypes and underlying molecular mechanisms [37]. For both species, inhibition of Hsp90 or calcineurin reduces azole resistance acquired by diverse mutations [16], [18], [22], [38]. With short generation times and relatively small genomes, these organisms provide tractable and complementary systems to explore the dynamics and mechanisms underpinning the evolution of resistance to drug combinations.
Here, we provide the first analysis of the genetic and genomic architecture of the evolution of resistance to drug combinations in fungi. To recapitulate a clinical context where Hsp90 or calcineurin inhibitors could be used in combination with azoles to render azole-resistant fungal pathogens responsive to treatment, we initiated an evolution experiment with strains that are resistant to azoles in a manner that depends on Hsp90 and calcineurin. We evolved populations of S. cerevisiae and C. albicans that were resistant to azoles due to loss of function of Erg3 with a combination of an azole and an inhibitor of Hsp90, geldanamycin, or calcineurin, FK506, to identify the mechanisms by which resistance evolves to the drug combinations. Of 290 lineages initiated, most went extinct, yet 14 evolved resistance. We identified mechanisms of resistance in the evolved lineages using a hypothesis-driven approach based on cross-resistance profiling and a complementary unbiased approach using whole genome sequencing. Resistance mutations in the drug target of FK506 or geldanamycin were identified and validated in five lineages. Non-synonymous substitutions conferring resistance were identified in a transcriptional activator of drug efflux pumps, Pdr1, and in a regulator of sphingolipid biosynthesis, Lcb1. Resistance also arose by premature stop codons in the catalytic subunit of calcineurin and in a repressor of ergosterol biosynthesis genes, Mot3. Several of the mutations conferred resistance to geldanamycin or FK506, while other mutations transformed azole resistance from dependent on calcineurin to independent of this stress response regulator. Genome analysis also identified extensive aneuploidy in four of the C. albicans lineages. Thus, we illuminate the molecular basis for the transition of azole resistance from calcineurin dependence to independence, and establish numerous mechanisms by which resistance to drug combinations can evolve, providing a foundation for predicting and preventing the evolution of drug resistance.
Inhibition of Hsp90 or calcineurin has emerged as promising strategy to enhance the efficacy of azoles against resistant fungal pathogens, motivating our study to monitor the evolution of resistance to the drug combinations in azole-resistant populations. To do so, we used an experimental evolution approach starting with C. albicans and S. cerevisiae strains that harbour erg3 loss-of-function mutations or deletions, rendering them resistant to azoles in a manner that depends on the stress response regulators Hsp90 and calcineurin [5]. Propagation of these strains in the presence of azole and the Hsp90 inhibitor geldanamycin or azole and the calcineurin inhibitor FK506 at concentrations that exert selection pressure for resistance to the drug combination could lead to the evolution of resistance to geldanamycin or FK506, or the evolution of an azole resistance mechanism that is independent of Hsp90 or calcineurin among extant lineages (Figure 1A). Lineages were propagated by serial transfer for between 33 and 100 generations until robust growth in the presence of the drug combination was observed in extant lineages (Figure 1B). The effective population size per lineage was ∼4.6×106, given that cultures reached saturation (∼107 cells/ml) between transfers. Of the 290 lineages initiated, the majority went extinct. Fourteen lineages evolved resistance to the combination of azole and inhibitor of Hsp90 or calcineurin (Figure 1C); seven of these lineages are C. albicans and seven are S. cerevisiae (Table 1). Six C. albicans lineages evolved resistance to azole and FK506 (Ca-F lineages), and only one evolved resistance to azole and geldanamycin (Ca-G lineage). Four S. cerevisiae lineages evolved resistance to azole and geldanamycin (Sc-G lineages) and three evolved resistance to azole and FK506 (Sc-F lineages).
Resistance levels to the drug combinations of all fourteen evolved lineages were evaluated by performing minimum inhibitory concentration (MIC) assays in the presence of the inhibitors with which they were evolved, azole and FK506 (Figure 2A and 2B) or azole and geldanamycin (Figure 2C–2E). Because the azole resistance phenotypes of the starting strains were abrogated by geldanamycin or FK506, resistance of the evolved lineages was monitored with a fixed concentration of azole and a gradient of concentrations of geldanamycin or FK506. Resistance was monitored for a population of cells from each archived lineage, and for four clones isolated from the evolved population. In all cases, the clones reflected the resistant phenotype of the population (data not shown), suggestive of strong selective sweeps as mutations were rapidly fixed in the population. For each population, a clone was archived and further analyses were performed on that strain. The lineages evolved distinct levels of resistance to the drug combinations (Figure 2), indicating that they acquired different mutations conferring resistance.
To gain insight into mechanisms of resistance to the drug combinations, we assessed cross-resistance profiles. Cross-resistance assays were performed in the presence of a fixed concentration of an azole and a gradient of concentrations of the structurally dissimilar counterpart to the Hsp90 or calcineurin inhibitor with which the population was evolved (native inhibitor), as well as with an azole and an inhibitor of the other stress response regulator not targeted in the evolution experiment (naïve inhibitor; i.e. Hsp90 inhibitor if the population was evolved with a calcineurin inhibitor). Cross-resistance profiles can be used to predict candidate resistance mechanisms based on an understanding of how these inhibitors bind to and inhibit their targets (Figure 3).
Lineages evolved with azole and FK506 were assayed for resistance to azole and geldanamycin (a naïve inhibitor) as well as to azole and cyclosporin A (a structurally dissimilar calcineurin inhibitor) [39], [40] (Figure 3A). FK506 inhibits calcineurin by forming a complex with the immunophilin Fpr1, and it is the drug-immunophilin complex that binds to and inhibits calcineurin [41]. The structurally unrelated calcineurin inhibitor cyclosporin A binds to a distinct immunophilin, Cpr1, to form a complex that binds to calcineurin and inhibits its function [40]. Geldanamycin inhibits Hsp90 by binding directly to the unconventional Bergerat nucleotide-binding pocket of Hsp90 [42], [43]. The level of resistance to these specific drug combinations suggests several candidate mechanisms of resistance (Figure 3B). For example, resistance to the combination of azole and FK506 but not to azole and other inhibitors tested suggests an FK506-specific mechanism of resistance such as mutation of FPR1. If resistance was also observed to the combination of azole and cyclosporin A, this would suggest that calcineurin has been altered in a way that prevents the binding of both immunophilin-drug complexes, or that a calcineurin-independent mechanism of azole resistance has evolved. If resistance was also observed to the combination of an azole and the naïve inhibitor geldanamycin, this would suggest that resistance emerged by a mechanism that is independent of the stress response regulators Hsp90 and calcineurin; candidate mechanisms include those that block the effect of the azoles on their target, such as up-regulation of the drug efflux pump Pdr5 in S. cerevisiae [28], or alteration of the azole target Erg11 that prevents azole binding [29].
Lineages evolved with azole and geldanamycin were assayed for resistance to azole and radicicol, a structurally unrelated Hsp90 inhibitor. Like geldanamycin, radicicol binds to the unusual nucleotide-binding pocket of Hsp90, inhibiting its chaperone function [42] (Figure 3C). These lineages were also assayed for cross-resistance to azole and FK506, a naïve inhibitor to these strains. Resistance to azole and geldanamycin alone suggests that a mutation in HSP90 occurred that prevents the binding of geldanamycin (Figure 3D). This cross-resistance profile is also consistent with a specific increase in geldanamycin metabolism or efflux. Cross-resistance to azole and FK506 suggests that an azole resistance mechanism evolved that is independent of the stress response regulators Hsp90 and calcineurin.
Variation in the patterns of cross-resistance to the distinct drug combinations was observed among the evolved strains (Figure 4, Figure 5, Figure 6, Figure 7), implicating a multitude of distinct resistance mechanisms. Even within a cross-resistance category variation was observed in the level of resistance to the drug combinations between strains, indicating that different mutations were responsible for resistance. This is consistent with the variation in levels of resistance with the native drug combination with which the population was originally evolved (Figure 2).
Two S. cerevisiae lineages evolved with azole and geldanamycin, Sc-G-12 and Sc-G-14, displayed different levels of resistance to azole and geldanamycin relative to the ancestral strain (Figure 2C and 2D). Both Sc-G-12 and Sc-G-14 showed increased cross-resistance to azole and radicicol, although Sc-G-14 was able to grow with higher concentrations of azole and geldanamycin as well as azole and radicicol than Sc-G-12 (Figure 4A). Neither lineage showed any cross-resistance to azole and FK506. This suggests distinct mutations in HSP90 might confer resistance to the drug combinations in these lineages. In S. cerevisiae, Hsp90 is encoded by two genes, HSC82, which is expressed at constitutively high levels, and HSP82, which is induced by high temperature [44]. Sequencing of HSC82 and HSP82 in Sc-G-14 identified a non-synonymous point mutation that maps to the N-terminal domain of HSC82, T1350A. This leads to the amino acid substitution I117N, a residue located in the groove lining the nucleotide-binding pocket of Hsp90 to which geldanamycin and radicicol bind. This residue is highly conserved and thought to be 90–100% buried [43]. The impact of HSC82I117N on resistance to azole and geldanamycin was confirmed by performing an allele swap, where HSC82 was deleted from the ancestral strain and the evolved allele was introduced on a plasmid, and reciprocally, HSC82 was deleted from the evolved strain and the ancestral allele was introduced. Expression of the HSC82I117N allele in the ancestral strain conferred a level of resistance to the combination of azole and geldanamycin equivalent to the evolved Sc-G-14 lineage (Figure 4B). Reciprocally, expression of only the ancestral HSC82 allele in the evolved strain abrogated resistance to the drug combination (Figure 4B). This confirms that HSC82I117N confers resistance to the combination of azole and geldanamycin in the Sc-G-14 lineage, perhaps by blocking geldanamycin-mediated inhibition of Hsp90 function.
Sequencing of HSC82 and HSP82 in Sc-G-12 identified a 4 bp insertion in HSC82 that results in a frameshift mutation and a premature stop codon in the middle of the coding sequence (HSC82K385*). This mutation is expected to render HSC82K385* non-functional [45]. Surprisingly, deletion of HSC82 in the parental strain confers a slight increase in resistance to azole and geldanamycin that phenocopies the resistance of Sc-G-12 (Figure 4C), suggesting that HSC82K385* is indeed non-functional and confers resistance to the combination of azole and geldanamycin in Sc-G-12.
The C. albicans lineage Ca-G-10 exhibited increased resistance to azole and geldanamycin with no cross-resistance to azole and FK506 or azole and radicicol (Figure 4D). It was cross-resistant to azole and 17-AAG (Figure S1), a derivative of geldanamycin, suggesting a mode of resistance specific to ansamycin benzoquinone Hsp90 inhibitors. Sequencing identified a heterozygous, non-synonymous mutation in HSP90, G271T. This mutation causes a D91Y amino acid substitution at a residue in the Hsp90 nucleotide-binding pocket that is thought to be 60–90% buried. This residue is conserved in human Hsp90 although not in S. cerevisiae, where the native amino acid is glutamic acid [43]. To assess the impact of HSP90D91Y on resistance to the combination of azole and geldanamycin we performed an allele swap, replacing one allele of HSP90 in the ancestral strain with the HSP90D91Y allele from the evolved strain, and replacing the HSP90D91Y allele in the evolved strain with the ancestral HSP90 allele. Replacing HSP90D91Y in Ca-G-10 with the ancestral HSP90 allele abrogated resistance in two independent transformants (Figure 4E). Reciprocally, replacing a native allele of HSP90 in the ancestral strain with the HSP90D91Y allele conferred resistance that phenocopied that of Ca-G-10. This indicates that HSP90D91Y confers resistance to azole and geldanamycin and is responsible for resistance of Ca-G-10. Thus, distinct mutations in Hsp90 can block the impact of geldanamycin on azole resistance in both C. albicans and S. cerevisiae, providing a mechanism for resistance to this drug combination.
Sc-F-2 and Sc-F-3 were evolved with azole and FK506, and demonstrate no cross-resistance to azole and cyclosporin A or azole and geldanamycin (Figure 5A and 5B), suggesting a mutation in FPR1 may confer resistance to azole and FK506 in these lineages. Sequencing identified a non-synonymous mutation in Sc-F-3 FPR1, G322T. This mutation leads to a V108F amino acid substitution that was responsible for the azole and FK506 resistance, as determined by an allele swap where the FPR1V108F allele was expressed from a plasmid in the ancestral strain in which the native FPR1 allele had been deleted, and reciprocally, the ancestral FPR1 allele was expressed in Sc-F-3 in which the FPR1V108F allele had been deleted. Expression of the FPR1V108F allele in the ancestral strain conferred resistance to azole and FK506, while replacing FPR1V108F with the ancestral allele in Sc-F-3 abrogated resistance (Figure 5C). This mutation likely reduces but does not completely block binding of FK506 to Fpr1 given that complete deletion of FPR1 confers an even greater level of resistance to azole and FK506. Consistent with this mutation conferring resistance to FK506 rather than altering the dependence of the azole resistance phenotype on calcineurin, deletion of the regulatory subunit of calcineurin required for its activation, CNB1, abrogated resistance of Sc-F-3 (Figure 5E).
Sequencing FPR1 in Sc-F-2 revealed a tandem duplication of nine amino acids that maps to the middle of the coding sequence, FPRdupG53-D61. Expressing FPRdupG53-D61 in the background of the ancestral strain conferred increased resistance to azole and FK506 (Figure 5D). That resistance was not as strong as in Sc-F-2 is likely due to the difference in expression levels of the native gene and the plasmid borne allele, which is driven by the GPD1 promoter. It is unlikely that there are other mutations affecting resistance in Sc-F-2 given that the resistance phenotypes of the ancestral strain and Sc-F-2 with the plasmid borne FPRdupG53-D61 allele as the sole source of Fpr1 were identical. Further confirming the importance of FPRdupG53-D61 for resistance to azole and FK506, replacing FPRdupG53-D61 of Sc-F-2 with the ancestral FPR1 abrogated resistance (Figure 5D). As with the FPR1 mutation identified in Sc-F-3, the FPRdupG53-D61 mutation in Sc-F-2 likely reduces but does not block binding of FK506 to Fpr1 as deletion of FPR1 confers an even greater level of resistance to azole and FK506 (Figure 5D). As with Sc-F-3, deletion of the regulatory subunit of calcineurin required for its activation, CNB1, abrogated resistance of Sc-F-2, consistent with this duplication in FPR1 conferring resistance to FK506 rather than altering the dependence of the azole resistance phenotype on calcineurin (Figure 5F).
Hypothesis driven approaches did not uncover any candidate resistance mutations for several evolved lineages. We therefore turned to whole genome sequencing to provide an unbiased approach to identify mutations that accompany the evolution of resistance to the drug combinations on a genomic scale. For example, S. cerevisiae Sc-F-1 was evolved with azole and FK506 and demonstrated robust resistance to the combination of azole and FK506 as well as azole and cyclosporin A (Figure 6A). This resistance profile suggested a possible mechanism of resistance involving alteration of calcineurin that prevents the binding of both protein-drug immunophilin complexes, or the emergence of a calcineurin-independent azole resistance mechanism. Calcineurin is encoded by the redundant catalytic subunits CNA1 and CNA2 and the regulatory subunit CNB1 in S. cerevisiae [39], [46]. Sequencing of CNA1, CNA2 and CNB1 did not reveal any mutations. Intriguingly, abrogating calcineurin function by deletion of CNB1 did not reduce resistance to azole and FK506 in Sc-F-1, indicating a calcineurin-independent mechanism of resistance had evolved (Figure 6B). Whole genome sequencing at high coverage (Table S1) identified two non-synonymous mutations (Table 2), as well as 58 mutations that were synonymous or in non-coding regions (Table S2 and Table S3); the best candidate for a mutation for affecting resistance was a mutation in MOT3, a transcriptional repressor of ergosterol biosynthesis genes [47]. The non-synonymous substitution in MOT3 resulted in a premature stop codon near the middle of the coding sequence, MOT3G265*, suggesting that this might be a loss-of-function allele. Deletion of MOT3 in the background of the ancestral strain or in Sc-F-1 phenocopied the level of resistance of Sc-F-1, which is consistent with MOT3G265* being a loss-of-function allele that confers resistance in Sc-F-1 (Figure 6C).
S. cerevisiae lineage Sc-G-13 was evolved with azole and geldanamycin and demonstrates only a small increase in resistance to this combination, with no cross-resistance to either azole and FK506 or azole and radicicol (Figure 6D). This resistance profile is consistent with a mutation in HSC82 or HSP82 that partially reduces binding of geldanamycin, however, no mutations were identified upon sequencing HSC82 and HSP82. Genome sequencing of Sc-G-13 identified five non-synonymous mutations, as well as 130 that were synonymous or in non-coding regions (Table 2 and Table S3); the best candidate for a mutation affecting resistance was a C2593G mutation in PDR1, which encodes a transcription factor that regulates the expression of numerous multidrug transporters such as PDR5. Gain-of-function mutations in PDR1 are a well-established mechanism of azole resistance that is independent of Hsp90 and calcineurin [16], [30], [48]. The mild resistance phenotype of Sc-G-13 suggested that the PDR1P865R allele in Sc-G-13 confers only a slight increase in drug efflux pump expression. Cross-resistance to azole and FK506 was not observed, likely because FK506 inhibits Pdr5-mediated efflux [49]. To evaluate the importance of the PDR1P865R allele in resistance to azole and geldanamycin we deleted PDR1 from the ancestral strain and the evolved Sc-G-13 lineage and introduced the ancestral PDR1 allele on a plasmid driven by the GPD1 promoter. Replacing the PDR1P865R allele of Sc-G-13 with the ancestral PDR1 allele reduced resistance of Sc-G-13 (Figure 6E). Resistance remained slightly increased relative to the ancestral strain, likely due to higher expression of PDR1 from the GPD1 promoter relative to the native promoter; consistent with this possibility, simply replacing the ancestral PDR1 allele in the ancestor with the same allele on the plasmid conferred a small increase in resistance (Figure 6E). Since there was no difference in resistance phenotype between the ancestral and evolved strains when the plasmid provided the only allele of PDR1, there are likely no other mutations conferring resistance in Sc-G-13.
For the six C. albicans lineages evolved with fluconazole and FK506 (Ca-F-4, Ca-F-5, Ca-F-6, Ca-F-7, Ca-F-8, and Ca-F-9), candidate resistance mutations were not identified by hypotheses-based cross-resistance profiles. These lineages shared the same cross-resistance profile of resistance to high concentrations of FK506 and increased resistance to cyclosporin A in the presence of azole (Figure 7A). This profile suggested that either a mutation in calcineurin preventing binding of both drug-immunophilin complexes occurred or a calcineurin-independent mechanism of resistance to azoles evolved. We sequenced the genome of all six lineages of this resistance class.
Genome analysis revealed aneuploidies in four of these evolved lineages. For Ca-F-4, we identified extensive aneuploidies in the absence of any non-synonymous mutations (Figure 8). This lineage exhibited increased copy number of chromosomes 4, 6 and 7 as well as an increase in copy number of the right arm of chromosome 5. Since approximately half the genome of Ca-F-4 had elevated copy number, resistance might be conferred by a combination of mechanisms including overexpression of the many relevant genes that were amplified including the gene encoding the drug transporter Mdr1, genes encoding ergosterol biosynthetic enzymes, the gene encoding the calcineurin regulatory subunit CNB1, or those encoding regulators of many other cellular pathways. We also identified increased copy number of chromosome 4 in three of the lineages, Ca-F-5, Ca-F-6 and Ca-F-7, as observed in Ca-F-4 (Figure 8). Ca-F-5 also had an increased copy number of chromosome 7. The remaining two lineages, Ca-F-8 and Ca-F-9, had no copy number variation other than variation in chromosome R, which was observed in all of the C. albicans lineages sequenced. Chromosome R contains the genes coding for rDNA, and extensive variation in size of the rDNA array has been observed in experimental populations of C. albicans [50], likely as a consequence of the highly repetitive nature of the genomic context. Two non-synonymous mutations were identified in C. albicans lineage Ca-F-9 (Table 3), and 7 mutations that were synonymous or in non-coding regions (Table S4). The best candidate for a resistance mutation is the C1201A mutation in CNA1, the gene encoding the catalytic subunit of calcineurin; this mutation leads to a premature stop codon, S401*. Truncation of C. albicans Cna1 at position 499 removes the autoinhibitory domain, resulting in a constitutively activated form of calcineurin [51]. Consistent with this mutation conferring resistance to the combination of azole and FK506 or azole and cyclosporin A, deletion of the evolved CNA1 allele in Ca-F-9 abrogates resistance to the combination of azole and calcineurin inhibitor (Figure 7B). Deletion of one allele of CNA1 in the ancestral strain has no effect on sensitivity to the drug combination. Thus, hyperactivation of calcineurin provides a mechanism by which resistance to azoles and calcineurin inhibitors can evolve.
Five non-synonymous mutations were identified in the C. albicans lineage Ca-F-8 (Table 3), and 16 mutations that were synonymous or in non-coding regions (Table S4). The best candidate for a resistance mutation is the A1169T mutation identified in orf19.6438 resulting in non-synonymous substitution, L390F. orf19.6438 remains uncharacterized in C. albicans but is an ortholog of S. cerevisiae LCB1, which encodes a component of serine palmitoyltransferase that is responsible for the first committed step in sphingolipid biosynthesis, along with Lcb2 [52]. Sphingolipids are a necessary component of the fungal cell membrane and have known interactions with ergosterol [53], while inhibitors of sphingolipid biosynthesis can enhance the efficacy of azoles [54]. To test the model that LCB1L390F confers resistance to the combination of azole and calcineurin inhibitor we used the serine palmitoyltransferase inhibitor myriocin, which inhibits Lcb1 and Lcb2 [55]. Inhibition of Lcb1 with myriocin abrogated resistance to azole and FK506 of the evolved lineage Ca-F-8 but did not affect resistance of Ca-F-9 (Figure 7C), suggesting that LCB1L390F confers resistance to the drug combination. Notably, myriocin caused an increase in resistance of the ancestral strain to azole and FK506 suggesting that resistance phenotypes are exquisitely sensitive to the balance of sphingolipid biosynthesis.
Our study provides the first experimental analysis of the evolution of resistance to drug combinations in fungi, illuminating the molecular basis of a transition of drug resistance from dependence on a key stress response regulator to independence, and a diversity of resistance mechanisms that can evolve in response to selection. This work addresses some of the most fundamental questions about the nature of adaptation. One key question is how many mutations underlie adaptive evolution. For all of the lineages for which we functionally tested the importance of mutations identified, we found that a single mutation was responsible for adaptation, in contrast to other experimental evolution studies with S. cerevisiae where multiple adaptive mutations were implicated [56], [57]. The small number of adaptive mutations identified in our study may reflect the short duration of the evolution experiment and the strength of the selection. Despite the limited number of adaptive mutations, we identified a larger number of total mutations in many lineages than reported in other studies [57]. The elevated number of mutations may be specific to the intense drug selection pressure, as bacterial mutation rates can increase in the presence of antibiotic selection [58], and antifungals have been associated with the rapid appearance of aneuploidies and genomic instability [59]. Another central question is how many genetic routes are there to adaptation. Among only 14 evolved lineages, we identified a diversity of adaptive mechanisms including target-based resistance to Hsp90 or calcineurin inhibitors and distinct mutations that render azole resistance independent of cellular stress response regulators, suggesting a complex adaptive landscape with multiple genotypes leading to high fitness adaptive peaks. Exploring the impact of the adaptive mutations on fitness in different environments, including in the absence of drug, will be key to understanding fitness costs of drug resistance, evolutionary trade-offs, and the limits of adaptation.
By starting the evolution experiment with strains that are resistant to azoles in a manner that depends on Hsp90 and calcineurin, we provide relevance for a clinical context where Hsp90 and calcineurin inhibitors could be deployed in combination with azoles to render azole-resistant isolates responsive to treatment. There is some precedent for the evolution of resistance to these drug combinations, as clinical isolates recovered from an HIV-infected patient over the course of two years evolved increased resistance to the combination of azole and inhibitors of Hsp90 or calcineurin [16]. While this patient was not treated with Hsp90 or calcineurin inhibitors, fever may have provided the selection for Hsp90 independence given that febrile temperatures cause global problems in protein folding that can overwhelm Hsp90 function and reduce azole resistance in a manner that phenocopies Hsp90 inhibition [16]. In our experimental evolution study, most of the 290 lineages initiated went extinct, while the 14 lineages that evolved resistance to the combination of azole and inhibitor of Hsp90 or calcineurin acquired a diversity of resistance mechanisms. These resistance mechanisms included mutations that rendered erg3-mediated azole resistance independent of the stress response regulator calcineurin, mutations that blocked the effects of the Hsp90 or calcineurin inhibitor, and large-scale aneuploidies. This experimental evolution approach provides a powerful system to predict the mechanisms by which resistance to drug combinations may evolve in the clinic. Consistent with the relevance of our findings, the increased resistance to azole and inhibitor of Hsp90 or calcineurin in isolates that evolved in an HIV-infected patient was accompanied by mutations causing overexpression of multidrug transporters [16], [60], as expected for the PDR1 mutation identified in one of our lineages.
One of the most prevalent mechanisms of resistance identified in our evolved populations was mutation in the target of the drug used in combination with azole during the evolution experiment. For Hsp90 inhibitors, it has been predicted that the probability of target-based resistance would be relatively low given that the amino acid residues in the nucleotide-binding site of Hsp90 family members are highly conserved from bacteria to mammals [61], suggesting that mutations that confer resistance would likely inactivate this essential molecular chaperone. This has helped fuel research on Hsp90 as a target for development of anti-cancer drugs, where inhibiting Hsp90 can impair the function of a multitude of oncoproteins [62]–[64]. Despite the constraints, there is precedent for point mutations in Hsp90 conferring resistance to Hsp90 inhibitors. One study engineered S. cerevisiae strains to be hypersensitive to drugs and expressed yeast or human Hsp90 as the sole source of the chaperone; introduction of a single point mutation (A107N for yeast, A121N for human Hsp90α, and A116N for human Hsp90β) conferred resistance to Hsp90 inhibitors [65]. Further, the fungus that produces radicicol, Humicola fuscoatra, harbours an Hsp90 with reduced binding affinity to radicicol but not geldanamycin [66]. Three of our evolved lineages acquired substitutions in Hsp90 that rendered erg3-mediated azole resistance more recalcitrant to the effects of Hsp90 inhibitors (Figure 4). For one S. cerevisiae lineage (Sc-G-14) and one C. albicans lineage (Ca-G-10), the mutations were in the nucleotide-binding domain, consistent with impairing drug binding. For S. cerevisiae lineage Sc-G-12, the mutation led to a premature stop codon (K385*); consistent with this HSC82 mutation causing a loss of function, deletion of HSC82 in the parental strain phenocopied resistance of Sc-G-12. Reducing dosage of a drug target often confers hypersensitivity to the drug rather than resistance [67]; this may suggest compensatory upregulation of the other S. cerevisiae gene encoding Hsp90, HSP82, which could confer elevated resistance. Target-based resistance to Hsp90 inhibitors has yet to emerge in Hsp90 inhibitor clinical trials, suggesting that these mutations may be associated with a fitness cost.
Mutations in the drug target also emerged as a mechanism that renders erg3-mediated azole resistance recalcitrant to the effects of calcineurin inhibitors in our evolved lineages. Two S. cerevisiae lineages acquired mutations in FPR1, which encodes the immunophilin that FK506 must bind to in order to form the protein-drug complex that inhibits calcineurin function [41]. A V108F substitution was identified in Sc-F-3 and a nine amino acid duplication near the protein midpoint was identified in Sc-F-2 (dupG53-D61). These alterations likely reduce but do not block FK506 binding, given that deletion of FPR1 conferred a higher level of FK506 resistance (Figure 5). There is precedent for overexpression or disruption of FPR1 conferring resistance to FK506 in S. cerevisiae [68], as well as for a W430C amino acid substitution in one of the two redundant calcineurin catalytic subunits Cna2 [69]. One C. albicans lineage, Ca-F-9, acquired a mutation in the catalytic subunit of calcineurin, CNA1C1201A, which results in a S401* premature stop codon that confers resistance to azole and both FK506 and cyclosporin A (Figure 7), likely due to hyperactivation of calcineurin [51]. Despite the emergence of target-based resistance to calcineurin inhibitors in vitro, there may be significant constraints that minimize the emergence of resistance in the human host. FK506 (tacrolimus) and cyclosporin A are front line immunosuppressants broadly used in the clinic to inhibit calcineurin function, thereby blocking T-cell activation in response to antigen presentation and suppressing immune responses that contribute to transplant rejection [39], [70]. Invasive fungal infections occur in ∼40% of transplant recipients including those that receive a calcineurin inhibitor as an immunosuppressant [71], however, this immunosuppressive therapy does not select for resistance to calcineurin inhibitors in C. albicans or Cryptococcus neoformans recovered from these patients [72], [73]. That resistance has not been observed in the host suggests that the resistant mutants may have reduced fitness relative to their sensitive counterparts or that other selective constraints alter the evolutionary dynamics.
Several of our evolved lineages took a distinct evolutionary trajectory, and evolved azole resistance mechanisms that are independent of the cellular stress response regulators. S. cerevisiae lineage Sc-F-1 evolved cross-resistance to azole and FK506 as well as azole and cyclosporin A (Figure 6). The azole resistance phenotype was independent of calcineurin but dependent on Hsp90 (Figure 6), suggesting a resistance mechanism that is contingent upon distinct Hsp90 downstream effectors, such as Mkc1 [74]. We identified an adaptive mutation in MOT3 (Table 2), a transcriptional repressor of ergosterol biosynthesis genes [47], which resulted in a premature stop codon, G265* and likely a loss-of-function allele (Figure 6C). Loss of function of Mot3 would lead to overexpression of ergosterol biosynthesis genes, which could minimize the impact of azoles on their target or could lead to a change in sterol balance that reduces the dependence of azole resistance on calcineurin. Changes in membrane composition may also explain the resistance of C. albicans lineage Ca-F-8 to azoles and calcineurin inhibitors, which was attributed to a mutation in the ortholog of S. cerevisiae LCB1 (Figure 7), encoding a regulator of sphingolipid biosynthesis. Notably, Mot3 is also a prion protein, which can convert between structurally and functionally distinct states, at least one of which is transmissible [75]; changes on Mot3 conformation and activity can modulate phenotypic variation in S. cerevisiae, and thus may influence the evolution of drug resistance phenotypes. S. cerevisiae lineage Sc-G-13 evolved a small increase in resistance to azole and geldanamycin associated with a mutation in PDR1, which encodes a transcription factor that regulates the expression of drug transporters such as PDR5 (Figure 6). Gain-of-function mutations in PDR1 are known to confer azole resistance that is independent of Hsp90 and calcineurin [16], [30], [48]. Cross-resistance to azole and FK506 may not have been observed because FK506 inhibits Pdr5-mediated efflux [49]. The weak resistance phenotype could reflect a small increase in transporter expression, or a fitness cost of the PDR1 mutation in an erg3 mutant background [76].
Several of the C. albicans lineages that evolved resistance to azole and calcineurin inhibitors demonstrated a complex genomic landscape of aneuploidies. The emergence of azole resistance in C. albicans has been associated with general aneuploidies as well as the formation of a specific isochromosome composed of two left arms of chromosome 5 (i5L) [77]. The isochromosome confers azole resistance due to increased dosage of two genes located on the left arm of chromosome 5: ERG11, which encodes the target of the azoles; and TAC1, which encodes a transcriptional regulator of multidrug efflux pumps [78]. Our lineages were resistant to azoles at the outset of the experiment, suggesting that the aneuploidies emerged in response to stress or were selected as a mechanism of resistance to the drug combination. Ca-F-4, Ca-F-5, Ca-F-6, and Ca-F-7 all had numerous aneuploidies relative to the parental strain (Figure 8). One aneuploidy that was common to all four lineages was increased copy number of chromosome 4, suggesting that an important resistance determinant might reside on this chromosome. While one might predict that such aneuploidies would be associated with a fitness cost, it is notable that a previous analysis of isolates carrying the i5L isochromosome demonstrated improved fitness in the presence and absence of azoles, relative to their drug-sensitive counterpart [59]. In contrast, many azole resistance mutations are associated with a fitness cost [79], though this cost can be mitigated with further evolution [80]. The prevalence of aneuploidies in the C. albicans lineages underscores the remarkable genomic plasticity of this pathogen [81], and the diversity of genomic alterations that can accompany adaptation.
The landscape of genetic and genomic changes observed in our evolved lineages illuminate possible mechanisms by which resistance to drug combinations might evolve in the human host and suggest candidate targets to minimize the emergence of resistance. Despite optimizing our selection conditions to favour the evolution of resistance to the drug combination, the majority of lineages went extinct (Figure 1). Consistent with constraints that minimize the evolution of resistance to these drug combinations, treatment of organ transplant patients with calcineurin inhibitors has not yielded resistance to these drugs in fungal pathogens recovered from these patients despite the extensive use of these drugs in patient populations [72], [73]. While Hsp90 inhibitors remain at the clinical trial stage for cancer and other diseases [62], [63], [82], [83], resistance has yet to emerge in these patient populations. Although there are a multitude of mechanisms that can confer resistance to the drug combinations, they may not be favoured due to fitness costs in the complex host environments.
The mechanisms by which resistance to the drug combinations evolved in our lineages suggest novel targets that could be exploited to block the evolution of drug resistance. Drug interactions have tremendous potential to influence the evolution of drug resistance [84]. Elegant studies with antibacterials emphasize that the impact of these interactions are often more complex than anticipated [8], [85]–[87]. While synergistic interactions that yield inhibitory effects larger than expected from individual drugs can maximize the rate at which infection is cleared, antagonistic interactions that yield inhibitory effects smaller than expected can suppress the evolution of multi-drug resistance. Ultimately, a systems biology approach incorporating experimental evolution, genetics and genomics, and clinical samples will be crucial for the development of effective strategies to enhance the efficacy of antimicrobial agents and minimize the evolution of drug resistance.
All Saccharomyces cerevisiae and Candida albicans strains were archived in 25% glycerol and maintained at −80°C. Strains were typically grown and maintained in rich medium (YPD: 1% yeast extract, 2% bactopeptone, 2% glucose, with 2% agar for solid medium only), or in synthetic defined medium (SD, 0.67% yeast nitrogen base, 2% glucose, with 2% agar for solid medium only), supplemented with amino acids, as indicated. Strains were transformed using standard protocols. Strains used in this study are listed in Table S5. Strains were constructed as described in Text S1.
Plasmids were constructed using standard recombinant DNA techniques. Plasmids used in this study are listed in Table S6 and oligonucleotides used in this study are listed in Table S7. Plasmids were constructed as described in Text S1. All plasmids were sequenced to confirm the absence of spurious non-synonymous mutations.
Evolution experiments were initiated with three ancestral strains of erg3-mediated azole resistant strains: two haploid S. cerevisiae strains (erg3Δ and erg3W148*) and one C. albicans strain (erg3Δ/erg3Δ; see Table S5). A founder colony was established for each ancestral strain and grown overnight in liquid, rich medium (YPD) without drug. From here, culture was transferred to a plate containing YPD with combinatorial drug concentrations of azole (fluconazole or miconazole) and geldanamycin, or azole (fluconazole or miconazole) and FK506 (i.e. treatments; see Table 1). Geldanamycin and FK506 were selected based on their specificity of target inhibition and their capacity to abrogate erg3-mediated azole resistance [16]; fluconazole and miconazole were selected as clinically relevant azoles of the triazole and imidazole class, respectively [4], [37]. Treatments were selected for the evolution experiment based on growth phenotype in the dose response matrices (Figure S2), such that strong directional selection for resistance would be applied. Concentrations were also varied to favour the emergence of distinct mechanisms of resistance. Lineages were then propagated in replicate in either 96-well plates (Sarstedt; 48 lineages initiated in this format) or 24-well plates (Becton Dickinson Labware; 242 lineages initiated in this format). The plates were formatted as described in Figure 1B. For propagation in 96-well plates, 1 µl of culture was transferred from the overnight culture to a final volume of 100 µl. Lineages were grown in a Tecan GENios plate reader and incubator at 30°C with constant agitation for three days. Subsequently, 1 µl of culture was transferred to a new plate containing YPD and treatment. Transfers occurred every 3 days to allow slow growing lineages to reach carrying capacity. This process was repeated until robust growth was present in some treatment wells. The experimental design for lineages propagated in 24-well plates was the same with the following adjustments: different drug combinations were selected for treatments; 10 µl of culture was transferred to 990 µl of YPD with treatment; plates were maintained at 30°C with constant agitation in a shaking incubator and transfers occurred every two days. With this dilution factor of 1/100, ∼6.6 generations occurs between transfers. The effective population size per lineage of ∼4.6×106 was estimated as described [88], given that cultures reached saturation of ∼107 cells/ml between transfers. Lineages that demonstrated reproducible resistance to the drug combination in which they were propagated were archived. Lineages unable to grow in the presence of the drug combination, either from when the cultures were initiated or over the course of the evolution experiment, were considered extinct. A summary of treatment concentrations, number of transfers and type of plate evolved in can been found in Table 1.
Resistance to drug combinations was assayed in 96-well microtiter plates, as previously described [16], [21]. Minimum inhibitory concentration (MIC) assays were set up to a final volume of 0.2 ml/well. MICs were performed in the absence of fluconazole (Sequoia Research Products) or with a constant concentration of fluconazole or miconazole (Sigma–Aldrich Co.), as indicated in the figures. All gradients were two-fold dilutions per step, with the final well containing no drug. The starting concentration of geldanamycin (Invivogen) gradients was 50 µM for S. cerevisiae strains and 5 µM for C. albicans strains. The starting concentration of FK506 (A.G. Scientific) gradients was 6 µM for S. cerevisiae strains and 100 µM for C. albicans strains. The starting concentration of radicicol (A.G. Scientific) gradients was 25 µM for both S. cerevisiae and C. albicans strains. The starting concentration of cyclosporin A (Calbiochem) gradients was 50 µM for both S. cerevisiae and C. albicans strains. The cell densities of overnight cultures were determined and diluted to an inoculation concentration of ∼103 cells/well. Plates were incubated at 30°C in the dark for the period of time specified in the figure legend. Cultures were resuspended and absorbance at 600 nm was determined using a spectrophotometer (Molecular Devices) and corrected for background of the corresponding medium. OD measurements were standardized to either drug-free or azole-only control wells, as indicated. Data was plotted quantitatively with colour using Java Treeview 1.1.3 (http://jtreeview.sourceforge.net/). Resistance phenotypes were assessed on multiple occasions and in duplicate on each occasion with concordant results, validating that the phenotypes are reproducible and stable.
Dose response matrices, or checkerboard assays, were performed to a final volume 0.2 ml/well in 96-well microtiter plates, as previously described [74]. Two-fold dilutions of fluconazole were titrated along the X-axis from a starting concentration of 256 µg/ml, with the final row containing no fluconazole. Along the Y-axis, either geldanamycin or FK506 was titrated in two-fold dilutions with the final column containing no geldanamycin or FK506. The starting concentration of geldanamycin was 5 µM for checkerboards with either S. cerevisiae or C. albicans strains. The starting concentration of FK506 was 4 µM for checkerboards with S. cerevisiae and 40 µM for checkerboards with C. albicans strains. Concentrations were selected to cover a range that spanned from no effect on growth to near complete inhibition of growth. Plates were inoculated and growth assessed as was performed for MIC assays.
Fluconazole was dissolved in sterile ddH2O. The Hsp90 inhibitors geldanamycin and radicicol and the calcineurin inhibitors FK506 and cyclosporin A were dissolved in DMSO. Myriocin (Sigma) was dissolved in methanol.
C. albicans cell pellets were digested with R-Zymolase for 1 hour (Zymo Research, D2002), prior to genomic DNA extraction with phenol-chloroform (EMD Millipore, EMD6810), and sodium acetate precipitation. Whole genome libraries were prepared using Nextera XT kits (Illumina, FC-131-1096) according to manufacturer's protocol. Libraries were sequenced on the Illumina HiSeq2000 platform using paired reads (101 bp) and version 3 reagents and chemistry.
The yeast genomes were sequenced in a multiplexed format, where an oligonucleotide index barcode was embedded within adapter sequences that were ligated to genomic DNA fragments [89]. Only one mismatch per barcode was permitted to prevent contamination across samples. Next, the sequence reads were filtered for low quality base calls trimming all bases from 5′ and 3′ read ends with Phred scores < Q30. Trimming sequence reads for low quality base calls drastically lowered false positive SNV calls.
De-multiplexed and trimmed reads from the S. cerevisiae strains were aligned to the S288c 2010 genome, a high fidelity sequence from an individual yeast colony (from F. Dietrich's lab at Duke University; it is the SGD reference genome as of February 2011) [90]. Reads from the C. albicans strains were aligned to the SC5314 genome from CGD [91]. While C. albicans is an obligate diploid, the current build of the genome, assembly 21, is a haploid genome, and is more accurate than the original diploid genome, assembly 19 [92], [93]. The diploid assembly was not used because it features 412 supercontigs with non-obvious heterozygosity, whereas the haploid assembly has been curated and organized into 8 chromosomes [93].
Sequence reads were aligned with Bowtie2, which was chosen over other commonly used short-read aligners such as Illumina's Eland, Maq, SOAP and BWA because it has been reported to be one of the fastest accurate aligners [94]–[98]. Additionally, it was chosen because it is updated frequently, supports variable read lengths within a single input file, is multi-threaded with a minimal memory and temporary file footprint and supports the standard Sequence Alignment/Map (SAM) file format [94], [98]–[100]. Alignments and all subsequent sequence data were visualized using the Savant Genome Browser [101]. The average coverage of each genome was calculated and was sufficient for confident variant detection (Table S1).
Aligned sequence reads for S. cerevisiae were subsequently processed using the UnifiedGenotyper package of the Genome Analysis Toolkit (GATK), which features a comprehensive framework for discovering SNVs and calculating coverage across genomic data [102], [103]. Variants detected in the S. cerevisiae parental strains were subtracted from complete variant lists, yielding a set of novel variants that emerged during strain growth in the presence of drug. Since C. albicans is diploid, we processed the reads with a more accurate approach using the probabilistic framework JointSNVMix, which uses paired parental and evolved strain sequence data to determine significant novel variants [104]. After identifying candidate SNVs, the threshold for homozygous SNV calls for both haploid (S. cerevisiae) and diploid (C. albicans) systems was set to 85% alternate (non-reference) basecalls at a specific position. In a diploid system, 35% was the threshold set to identify heterozygous SNVs. All variant positions required a minimum coverage of 15× to be considered as a candidate SNV. The total number of high-confidence novel mutations agrees with mutation rates observed previously for S. cerevisiae (Table S3) [105]. To further verify that the sequence data are of high quality, we compared two distinct sequence runs from two different sequence library preparations of the same parent C. albicans strain CaLC660. The total number of diploid single nucleotide variants that exist between the parent strain and the reference genome (SC5314) is 3748, therefore there is 99.99% concordance between both sequence replicates (Table S8).
The software package CNV-seq was used to identify chromosomal regions that varied in copy number between parental strains and evolved lineages [106]. This analysis found no significant CNVs in the S. cerevisiae strains, but numerous large variants were observed in C. albicans.
Sequence data is publicly available on the NCBI Short Read Archive with accession SRA065341.
|
10.1371/journal.pgen.1003608 | Prediction of Complex Human Traits Using the Genomic Best Linear Unbiased Predictor | Despite important advances from Genome Wide Association Studies (GWAS), for most complex human traits and diseases, a sizable proportion of genetic variance remains unexplained and prediction accuracy (PA) is usually low. Evidence suggests that PA can be improved using Whole-Genome Regression (WGR) models where phenotypes are regressed on hundreds of thousands of variants simultaneously. The Genomic Best Linear Unbiased Prediction (G-BLUP, a ridge-regression type method) is a commonly used WGR method and has shown good predictive performance when applied to plant and animal breeding populations. However, breeding and human populations differ greatly in a number of factors that can affect the predictive performance of G-BLUP. Using theory, simulations, and real data analysis, we study the performance of G-BLUP when applied to data from related and unrelated human subjects. Under perfect linkage disequilibrium (LD) between markers and QTL, the prediction R-squared (R2) of G-BLUP reaches trait-heritability, asymptotically. However, under imperfect LD between markers and QTL, prediction R2 based on G-BLUP has a much lower upper bound. We show that the minimum decrease in prediction accuracy caused by imperfect LD between markers and QTL is given by (1−b)2, where b is the regression of marker-derived genomic relationships on those realized at causal loci. For pairs of related individuals, due to within-family disequilibrium, the patterns of realized genomic similarity are similar across the genome; therefore b is close to one inducing small decrease in R2. However, with distantly related individuals b reaches very low values imposing a very low upper bound on prediction R2. Our simulations suggest that for the analysis of data from unrelated individuals, the asymptotic upper bound on R2 may be of the order of 20% of the trait heritability. We show how PA can be enhanced with use of variable selection or differential shrinkage of estimates of marker effects.
| Despite great advances in genotyping technologies, the ability to predict complex traits and diseases remains limited. Increasing evidence suggests that many of these traits may be affected by a large number of small-effect genes that are difficult to detect in single-variant association studies. Whole-Genome Regression (WGR) methods can be used to confront this challenge and have exhibited good predictive power when applied to animal and plant breeding populations. WGR is receiving increased attention in the field of human genetics. However, human and breeding populations differ greatly in factors that can affect the performance of WGRs. Using theory, simulation and real data analysis, we study the predictive performance of the Genomic Best Linear Unbiased Predictor (G-BLUP), one of the most commonly used WGR methods. We derive upper bounds for the prediction accuracy of G-BLUP under perfect and imperfect LD between markers and genotypes at causal loci and validate such upper bounds using simulation and real data analysis. Imperfect LD between markers and causal loci can impose a very low upper bound on the prediction accuracy of G-BLUP, especially when data involve unrelated individuals. In this context, we propose and evaluate avenues for improving the predictive performance of G-BLUP.
| Many important human traits and diseases are moderately to highly heritable. This, together with advances in genotyping and sequencing technologies, brought the promise of genomic medicine [1]. In the last decade genome-wide association studies (GWAS) have uncovered an unprecedented number of variants significantly associated with important complex human traits and diseases [2]. However in most cases, the combined effects of variants found to be significantly associated with various traits and diseases explain such a small proportion of inter individual differences in genetic risk that the usefulness of genomic information in clinical practice remains limited. In part, this reflects lack of power of standard GWAS to detect phenotype-marker associations for small effect variants [3], [4]. A number of studies have shown that prediction accuracy can be increased by including in the model variants that may not show significant association at the marginal level (e.g., [5]. A few authors [6]–[8] went further and suggested that the analysis and prediction of complex traits may be improved with the use of Whole-Genome Regression methods (WGR; [9]) where phenotypes are regressed on hundreds of thousands of markers concurrently. For instance, using G-BLUP (Genomic Best Linear Unbiased Predictor, one of the most commonly used WGR methods) Yang et al. [7] found that roughly 50% of the genetic variance of human height can be explained by regression on common SNPs. Similar results were confirmed for other complex traits [10].
The ability of a model to predict yet-to-be observed phenotypes (hereinafter referred to as PA, for prediction accuracy) constitutes one of its most important properties from the perspective of its potential use for preventive and personalized medicine. The study by Makowsky et al. [8] assessed PA of G-BLUP and, using family data, reported a cross-validation R2 of 0.25. However, the R2 ranged from 0.36 for individuals having 3 or more close relatives in the training data set to 0.11 for individuals with no close relatives in the training data set. The result confirms previous findings from the field of animal breeding [11] suggesting important influences of close familial relationships on the PA of G-BLUP methods. This raises an important question: what levels of PA could be expected when G-BLUP is used to predict complex human traits and diseases using data from unrelated individuals?
In this article, using theory, simulation and real data analysis we study the factors that affect the extent of missing heritability and the prediction accuracy of G-BLUP for the analysis of human data. The article is organized as follows. The methods section begins with an overview of G-BLUP. We describe the assumptions that define the model and derive analytical expressions that relate genomic relationships to prediction accuracy in two scenarios: (a) when the genotypes used for analysis are those at causal loci (hereinafter referred to as analysis under perfect LD between markers and QTL) and (b) under imperfect linkage disequilibrium (LD) between the markers used to compute genomic relationships and the genotypes at causal loci. The derivation of the R2 formula under perfect LD between markers and QTL follows from standard properties of the multivariate normal density and similar results have been presented before [12], [13]. However, under imperfect LD the model does not hold (because of misspecification of the covariance function) and the standard formulas cannot be used. Based on a few assumptions we derive a closed-form upper bound on prediction R2 for the case of imperfect LD. Predictions from the formulas derived in the methods section are validated in simulated and real data analyses using data from related (Framingham Heart Study [14]) and nominally unrelated (a sub-study of GENEVA [15]) individuals. In the Discussion section the analytical and empirical findings of our research are discussed and put into context and various implications of our results are considered.
Standard quantitative genetic models describe phenotypes as the sum of a genetic value () plus an error term ; that is . For ease of presentation it is assumed throughout this article that phenotypes and genetic values have null mean (i.e., these are expressed as deviations from the sample mean). Genetic values could be a complex function of the genotypes (zi) of individual i at q causal loci, that depends on the genetic architecture of the trait (i.e., the number and exact set of causal loci and the types of interactions among alleles within and between loci, and the distribution of effects). In practice, the genetic architecture of the trait analyzed is unknown and empirical models are built using regressions on marker genotypes. In WGR models [9], phenotypes are regressed on potentially hundreds of thousands of marker covariates concurrently using a regression function which could be parametric or not. The empirical model becomes: , where denotes the number of copies of one of the alleles observed on the ith individual at the lth marker and is a model residual that captures the effects of non-genetic factors () as well as errors which may emerge either because of model misspecification (e.g., unaccounted interactions) or because of imperfect LD between markers and genotypes at causal loci. In most applications, is structured using a parametric linear regression of the form where represents the additive effect of the allele coded as one at the lth marker. Often the number of markers (p) vastly exceeds the number of data points (n) and implementing these large-p with small-n (p>>n) regressions requires shrinkage estimation or use of some form of variable selection. Owing to developments in the field of penalized and Bayesian regressions, there is a multiplicity of methods that can be used to implement these p>>n regressions [16]. Most of the applications in plant and animal breeding and most of the studies involving human data have used G-BLUP and we therefore focus on this method.
Genomic BLUP can be motivated in many different ways: as a Ridge Regression (RR, [17]) on marker genotypes, as a Bayesian Gaussian Regression on markers or as a random effects model. A detailed description of this model is given in Supplementary Methods. Here we briefly describe G-BLUP adopting the random effects perspective where phenotypes are viewed as the sum of a random effect representing genomic signal () and a model residual (),(1)both of which are assumed to follow multivariate normal (MVN) distributions. The vector of genomic values is assumed to follow a MVN distribution with mean equal to zero and variance-covariance matrix proportional to , a marker-derived matrix of realized genomic relationships between pairs of individuals (). Model residuals, are regarded as independent of u and assumed to follow IID normal distributions, centered at zero and with variance . Therefore,(2)where I denotes an identity matrix of dimension n. Importantly, the ability of the model described by expressions (1) and (2) to separate signal (u) from noise () depends completely on how well G describes realized genetic relationships at unobserved causal loci.
In empirical analyses, genomic relationships are usually computed using crossproduct terms between genotypes. In such cases, estimates derived from G-BLUP methods are equivalent to those that can be derived by regressing phenotypes on marker genotypes using a linear model, , with marker effects treated as IID draws from a normal distribution, . See Supplementary Methods for further details about the equivalence of G-BLUP and some linear regressions on marker covariates.
The predictive ability of a model is commonly assessed using the variance of prediction errors (or prediction error variance), , where represents a prediction, for instance, . The proportional reduction in phenotypic variance accounted for by predictions (referred to as R2 in this article) can be quantified usingwhere, represents the phenotypic variance of individual n+1. Below we look at two scenarios: (i) prediction accuracy when markers and QTL are in perfect LD and (ii) prediction accuracy when markers and QTL are in imperfect LD.
To obtain further insight on the impacts of imperfect LD between markers and QTL on the proportion of missing heritability and on PA, a simulation study and real data analysis were performed using data sets from related and from unrelated individuals.
Results from the simulation study are reported first; this is followed by the results of the real data analysis.
Table 1 shows the distribution of allele frequencies (computed among the 5,800 individuals used for analysis in each of the data sets) by set of markers and data set. The distribution of allele frequencies observed in the FHS and GEN was very similar, with a correlation of MAF between the two data sets of 0.997. The distribution of minor allele frequencies of subsets of randomly chosen markers (either those designated as non-causal loci or those designated as causal loci in the RAND scenario) was very similar, with more than 65% of the markers having a MAF greater than 0.15. On the other hand, as a consequence of the sampling scheme used, in the Low-MAF scenario, the distribution of allele frequencies at causal loci had an over representation of low MAF loci. We also computed the squared correlation of genotypes of adjacent markers at various lags (in this case defined as the number of markers in the interval in the map), from lag 1 to lag 100 in FHS and GEN. For the set of SNP used in this study (400 K) the average inter marker distance was 7.2 kb. Plots of the patterns of association between genotypes at adjacent markers are given in the Supplementary Data (see figures S1 and S2). Although, for some pair of markers, the squared correlations in FHS and GEN were different; however, the overall patterns (e.g., the average squared-correlation at lag 1, 2,…, 100, or percentiles of the squared correlations at various lags) were identical in both data sets.
The eigenvalue decomposition of the marker-derived genomic relationship matrices revealed that the cumulative variance explained by the 1st 5 eigenvalues were 0.35, 0.51, 0.64, 0.78 and 0.90% in FHM and 0.35, 0.51, 0.61, 0.69, and 0.77% in GEN, respectively. Ordinary least squares regression of adjusted height on the 1st PC explained a proportion of the variance (in the training sample) equal to 4% in FHM and to 2% in GEN. Therefore, although both data sets exhibit some extent of population stratification, the proportion of variance of genotypes explained by high order principal components was low.
Estimates of and of prediction R2, averaged across 30 MC replicates are displayed in Table 2. Results by MC replicate are provided in Tables S1, S2, S3, S4, S5 of the Supplementary Data.
Table 5 gives estimated posterior means of for the pedigree-model (P-BLUP, applied to FHS only), and of for G-BLUP and wG-BLUP fitted to the full and combined data sets. The estimate of for the P-BLUP model in FHS was 0.857; this value is within the range, slightly higher, of what is generally considered the heritability of human stature (i.e., 0.8). The estimate of in FHS with G-BLUP was slightly smaller (0.837). Both results are in agreement with previous reports for this trait and data set (e.g., Makowsky et al. [8]) as well as with the simulation study presented in this article, with one small difference: in the simulation study the estimate of from marker based G-BLUP was slightly higher than that of P-BLUP, while in the real data analysis the opposite happened. One possible explanation is that in the real data analysis P-BLUP captured some non-additive genetic effects and/or some components of permanent environment that are not captured by G-BLUP. Finally, in FHS, the estimated derived using wG-BLUP was similar, albeit slightly lower, than with G-BLUP (0.814). In short, regardless of the method (P-BLUP, G-BLUP or wG-BLUP) no missing heritability is observed in the analysis of family data.
On the other hand, the analysis of data from unrelated individuals (GEN) exhibited a great extent of missing heritability (roughly 53% for G-BLUP, computed as 100×[1−0.374/0.80]) both for G-BLUP and even greater for wG-BLUP. These results are also in agreement with previous reports for the trait (e.g., [7]) and with the trend observed in the simulation study in scenario Low-MAF. However, the extent of missing heritability was higher than what was observed in the simulation, perhaps suggesting that the levels of imperfect LD between genotypes at markers and those at causal loci affecting human height are even more extreme that those present in the simulation.
In recent years GWAS have uncovered unprecedented numbers of variants associated with many important complex human traits and diseases. However in most cases the joint effects of variants detected so far explain only a small proportion of the genetic variance of those traits, a problem referred to as the missing heritability of complex traits [3], [4]. G-BLUP was first used in human genetic studies partly to address this issue in the study of Yang and co-authors in 2010 [7]. This study showed that inclusion of all available marker information in a joint analysis resulted in a marked increase of the proportion of variance explained, recovering part of the missing heritability. However, the ability of a model to predict yet-to-be observed phenotypes can be markedly different from the proportion of variance accounted for in a training data set, as measured by estimates of . Previous studies [8] suggest that the ability of G-BLUP to predict unobserved phenotypes of individuals that are distantly related to the training samples is very limited.
The purpose of this study was to shed light on some of the factors that affect the predictive performance of G-BLUP and to identify avenues by which this methodology can be improved. In particular we focused on studying how imperfect LD between markers and QTL affects the extent of shrinkage in prediction R2, relative to the prediction R2 obtained with the same sample and data structure if genotypes at causal loci were known. Several authors have studied the factors affecting prediction accuracy of G-BLUP. For instance, Goddard [28] and Daetwyler et al. [29], [30] derived formulas linking prediction R2 to features of the trait (e.g, h2) of the sample (e.g., size of the training data set) and of the genome (e.g., span of LD and how this affects the number of independently segregating segments). The studies of Goddard [28] and Daetwyler [30] have a much broader scope than ours. However, because of the broader scope, their results reside on stronger assumptions and important factors affecting prediction accuracy are not accounted for. One of these assumptions is that genomes can be described as a set of independently segregating segments. This abstraction is conceptually appealing; however the abstraction is difficult to validate and the quantification of the number of independently segregating segments is controversial (various methods leading to very different values of this parameter exist, e.g., [24], [28]). In the approach presented in this study this assumption is not required.
One limitation of Goddard's approach is that it does not account for the effects of recent familial relationships (the derivations are solely based on population LD). Our approach captures, via the regression of marker-derived genomic relationships on those realized at causal loci, both effects of LD between markers and QTL as well as cosegregation between markers and QTL that occurs because of recent family relationships. On the other hand, Daetwyler's approach [30] assumes that the model accounts for all the genetic variance. We have shown that this assumption, which is not present in Goddard [28], is clearly violated in analyses involving unrelated individuals and is not part of the derivations presented in our work.
In Goddard's approach [28] the factors affecting prediction accuracy are decomposed into two components: (a) one related to the accuracy of estimates of effects and (b) one that quantifies the effects of imperfect LD between markers and QTL on prediction accuracy. In our approach, all the factors affecting the accuracy of estimates of the effects of the alleles at causal loci are captured by the R2 under perfect LD; and we make almost no statements about this quantity, other than the ones that follow from the properties of the multivariate normal distribution. Instead, we focus on quantifying the effects of LD on R2 that occurs through misspecification of TRN-TST relationships. Importantly, our simulation results show that the proposed upper bound formulas account for 80–90% of the observed shrinkage in R2.
In summary, our approach is complementary to that of Goddard [28] and Daetwyler [30]; we focused on a much narrower problem and by virtue of that were able to arrive at closed-form formulas that account for a sizable proportion of the shrinkage in R2 due to imperfect LD without making strong assumptions.
The ability of G-BLUP to separate true signal (g) from noise () depends entirely on how well marker derived genomic relationships () describe genetic relationships realized at unobserved causal loci (). Genomic relationships at subsets of loci in the genome (e.g., markers, causal loci) can be viewed as the result of a random process with expected value given by the pedigree relationships () and variation due to Mendelian sampling. Because of the random nature of this process, genomic relationships vary across regions of the genome and therefore, the patterns of genomic similarity at markers and at causal loci may be different.
If the variance of the realized genomic relationships (across regions of the genome) is small relative to their expected value, the patterns of realized genomic relationships at markers will provide a good description of the patterns of realized genetic relationships at unobserved causal loci. Hill and Weir (2011) [20] have characterized various moments of the distribution of genomic relationships and concluded that the coefficient of variability decreases as the expected value, , increases. Therefore, for pairs of unrelated individuals, a large coefficient of variation of genomic relationships across regions of the genome is expected. The analyses reported here support this; indeed, the regression of realized genomic relationships computed at different subsets of markers is close to one (0.98, see Table 4) for closely related individuals and very small (of the order of 0.10, see Table 4) for pairs of nominally unrelated individuals. Therefore, two contrasting situations are encountered: some of the elements of the marker derived genomic relationship matrix represent very well the true covariance function (i.e., the patterns of realized genetic relationships at observed causal loci) but others (all the off-diagonal elements corresponding to distant relatives and to pairs of unrelated individuals) show patterns of realized genomic relationships that do not describe well the patterns of realized genetic relationships at causal loci. This has direct and different impacts on estimation of variance parameters and on PA, because variance parameters and PA are driven, in part, by different forces. To illustrate with an extreme scenario, suppose that G, the matrix of realized genomic relationships at causal loci, is diagonal (i.e., all off-diagonal terms of G equal zero). In this case, it would still be possible to estimate variance parameters and genomic heritability (simply based on the fact that the diagonal elements of G are not constant). Yet, the prediction accuracy for phenotypes in the TST data set will be null because all the off-diagonals of G are equal to zero.
In this study we have chosen to center and to standardize markers using estimates of allele frequency derived from the sample. As stated, centering does not have an effect on predictions or on estimates of variance parameters [25], provided that the model contains an intercept. On the other hand, standardization can have an effect. When markers are standardized to unit variance, the relative contributions of markers to the genomic relationship matrix are the same. This is good practice if it enhances the ability of marker derived genomic relationships to describe the patterns of genetic similarity realized at causal loci. If the distribution of allele frequency at causal loci has a higher representation in the low minor allele frequency spectrum than the one observed at the markers, or if the size of effects is inversely related to minor allele frequency, then standardization may reduce the extent of missing heritability and may improve prediction accuracy.
The results of the simulation study indicate that when markers and QTL are in perfect LD, no missing heritability is observed, as expected. This holds regardless of whether the training sample comprises data from related or unrelated individuals. When markers and QTL are in imperfect LD two contrasting situations were encountered: (a) with family data no missing heritability was observed, and (b) with unrelated individuals, we either observed a small extent of missing heritability (when markers and QTL were sampled from the same distribution of loci, the RAND scenario) or a greater extent of missing heritability (this happened when the distribution of allele frequency at markers and causal loci was different, the Low-MAF scenario). The estimates of variance components and of genomic heritability for human height reported here are consistent with previous results for this trait. In other words, no missing heritability was observed in the analysis of family data [8] and a great extent of missing heritability (roughly 50%) was observed with unrelated individuals [7].
Predictions based on G-BLUP are weighted averages of phenotypes in the TRN data set (see, eq. 4). The weights are heavily determined by the realized TST-TRN genomic relationships (i.e., the off-diagonal entries of G). Therefore the PA that can be derived from G-BLUP is highly dependent on the magnitude of these coefficients and on the extent to which marker derived genomic relationships represent the underlying patterns of genetic similarity realized at causal loci. Using standard properties of the multivariate normal distribution one can derive closed-form expressions for prediction error variances and for prediction R2 (see Supplementary Methods). These expressions are valid if the model holds. This requires, among other things, that the markers used to compute genomic relationships are in perfect LD with genotypes at causal loci. Under such conditions, prediction R2 has an upper bound given by an index that is the product of the heritability of the trait times a weighted sum of squares of the realized genomic relationships between the individuals used for TRN and those in the TST data set (see eq. 5). The expected value of realized genomic relationships is given by the pedigree derived additive relationships. For distantly related individuals the expected value of genomic relationships is small and, consequently, data from unrelated individuals are expected to contribute little to prediction accuracy. Nevertheless, if the model holds, PA is anticipated to increase monotonically with the size of the TRN data set (each additional phenotype in the TRN data set brings additional information) and, asymptotically prediction R2 converges to the heritability of the trait. However, this does not occur when markers and QTL are in imperfect LD. Indeed, under imperfect LD, prediction R2 can have an upper bound that is much lower than the heritability of the trait. Assuming a linear relationship between the realized genomic relationships at markers and at causal loci, an upper bound to prediction R2 under imperfect LD between markers and QTL () was derived (see expression 7). This upper bound is given by the product of two terms : (a) the R2 that can be obtained (using the same TRN sample) if markers and QTL were in perfect LD and (b) a coefficient that depends on the coefficient of linear regression between TRN-TST realized genomic relationships at markers and those at causal loci . This result was derived assuming that realized genomic relationships at causal loci in the TRN data set are known, and therefore, represents an upper bound on prediction under imperfect LD.
The regression coefficient drives the size of the reduction factor on prediction R2. When the TRN and TST data set are related due to close familial relationships, the regression of genomic relationships at markers on those at causal loci is moderately high (e.g., of the order of 0.8–0.9 for pairs of related individuals, or of the order of 0.35 when we consider a mixture of both related and unrelated individuals as in the FHS, see Table 3). Using a value of 0.35 (average for the FHS) the minimum expected reduction factor in prediction R2 due to imperfect LD, , is of the order of 40–50%. On the other hand, when TRN and TST data sets are composed of nominally unrelated individuals, the regression is much smaller (of the order of 0.1). A large reduction factor in prediction R2 is therefore predicted (of the order of 80% computed as 100×[1–2×0.1+0.12]). Importantly, the minimum shrinkage in R2 predicted by our formula matched very closely the observed shrinkage due to imperfect LD estimated in the simulation (roughly, the minimum shrinkage factor was 80–90% of the observed shrinkage in R2, see Table 3).
The maximum R2 that can be attained under perfect LD (assuming infinitely large samples and that the model holds) is h2, the heritability of the trait. Imperfect LD between markers and QTL induces shrinkage in R2; in case of data sets of nominally unrelated individuals similar to GEN a minimum shrinkage in R2 of 80% is anticipated; therefore, the expected asymptotic upper bound for R2 is 20% of h2, or 16% in the case of height. This estimate applies to data sets of similar characteristics that the GEN data set. Prediction problems involving individuals that are less (more) distantly related than the average individual in GEN are expected to have a lower (higher) upper bound on R2. Similarly, our estimates reflect the specifics of the SNP chip used and how genomic relationships were computed.
In finite samples, as pointed out in previous studies [28]–[31], estimation errors in marker effects will reduce the perfect LD R2 to values smaller than h2. Some proposed formulas for the expected value of R2 under perfect LD take the forms [31], or [30], where m is the number of independent causal loci and N is the number of records in the training data set. These formulas could be used to obtain a reference for the expected R2 under perfect LD. However, the derivation of these formulas assumes that genotypes at causal loci are fully orthogonal. We applied these formulas using m = 5,000, and , the setting of our simulation if we assume that causal loci are in linkage equilibrium, and obtained R2 values of 0.47 and 0.37 using the formulas suggested in [30] and [31], respectively. These values are lower than those obtained in our simulation for GEN, where R2 under perfect LD ranged between 0.52–0.54.
GENEVA and the FHS contain samples drawn from relatively homogeneous populations. On the other hand, when allele frequencies vary across subpopulations, so does the relative contribution of each locus affecting the trait to genetic variance in each of the subpopulations. This raises the question of what estimates of allele frequencies should one use when analyzing data involving different subpopulations. In the present study this was not an issue because the correlation of estimates of allele frequencies derived from GEN and FHM was virtually 1 (0.99). However, when this is not the case, if genomic relationships are scaled with estimates of allele frequencies derived from the entire sample, then marker derived genomic relationships will provide a poorer description of the realized genetic relationships in each of the sub-populations. This may result in a lower estimate of and a much higher R2-shrinkage factor.
Both FHS and GEN, especially the former, show some degree of population stratification, as judged by the inspection of the loadings of the 1st two eigenvectors derived from G. However the cumulative proportion of variance explained by the first two eigenvectors was relatively small. In the presence of stratification, there may be reasons to remove between cluster variability, and to obtain within cluster estimates of variance components and of prediction accuracy. Following the approach used by Janss and coauthors [32] one could derive genomic relationships that do not include the contribution to genetic similarity of the 1st k principal components of G. The use of such genomic relationships would yield a within cluster estimate of . These estimates can be plugged into the equations presented here to derive an upper bound on prediction R2 that does not account for genetic similarity attributable to substructure.
The effectiveness of G-BLUP depends critically on the extent to which marker derived genomic relationships reflect the patterns of realized genetic relationships at causal loci. The size of the coefficient of variation of realized genomic relationships across regions of the genome depends on the number of independently segregating segments among the pair of individuals whose realized genomic relationship we wish to assess. For pairs of unrelated individuals this is largely controlled by the span of LD in the population. For pairs of related individuals this is largely controlled by within family disequilibrium. In animal and plant breeding populations G-BLUP has exhibited very good predictive performance because the two conditions needed for G-BLUP to perform well are generally met: LD span over long regions and data include highly related individuals. Under these conditions variable selection is difficult to perform and may not be needed because the patterns of genetic similarity realized at markers and at causal loci are similar.
However, the analysis of human data from unrelated individuals represents the exact opposite situation. Here LD spans over shorter regions [33] and within family disequilibrium cannot be exploited. Under these conditions the use of markers that are in imperfect LD with QTL results in very low prediction accuracy of G-BLUP. Variable selection constitutes a natural way of increasing the extent of LD between markers and QTL. However, for complex traits, stringent variable selection can induce poor coverage of regions with small, but not negligible, contribution to variance. Therefore, we are faced with the need for finding an appropriate balance: as variable selection becomes more stringent, LD between markers and QTL increases, but the some proportion of the variance contributed by QTL of small effects may be lost. The appropriate balance will likely depend on the genetic architecture of the trait but also, importantly, on features of the sample. With family data, the benefits of variable selection are relatively small. However with unrelated individuals, variable selection, including large numbers of markers (e.g., 5 K top SNPs), or perhaps better some form of smooth differential weighting of the contribution of individual markers to genomic relationships, seems to be an effective way of improving prediction accuracy. This could be done either combining information from a prior study, as implemented in this article, or using methods that perform variable selection and differential reduction of estimates of effects simultaneously. The literature on WGR offers several penalized and Bayesian methods that can achieve this goal. The application of these methods to plant and animal breeding data has not shown marked improved gains in PA relative to G-BLUP. However, for the reasons discussed in this paper, we anticipate that the situation may be different when these methods are applied to the analysis and prediction of complex traits using data from unrelated individuals.
In conclusion, we have provided an analytical framework to quantify the maximum prediction R2 that can be attained using G-BLUP and have compared the properties of G-BLUP in samples of related and unrelated individuals. The analytical expressions derived are consistent with our simulation and empirical results and suggest that the analysis of nominally unrelated individuals presents a number of challenges that standard G-BLUP does not address. These can be partly met by incorporating prior knowledge of the relative importance of SNPs for a given trait. Further research will be required to optimize the modeling of such prior knowledge towards improved trait prediction.
|
10.1371/journal.ppat.1000583 | Population Genetic Analysis Infers Migration Pathways of Phytophthora ramorum in US Nurseries | Recently introduced, exotic plant pathogens may exhibit low genetic diversity and be limited to clonal reproduction. However, rapidly mutating molecular markers such as microsatellites can reveal genetic variation within these populations and be used to model putative migration patterns. Phytophthora ramorum is the exotic pathogen, discovered in the late 1990s, that is responsible for sudden oak death in California forests and ramorum blight of common ornamentals. The nursery trade has moved this pathogen from source populations on the West Coast to locations across the United States, thus risking introduction to other native forests. We examined the genetic diversity of P. ramorum in United States nurseries by microsatellite genotyping 279 isolates collected from 19 states between 2004 and 2007. Of the three known P. ramorum clonal lineages, the most common and genetically diverse lineage in the sample was NA1. Two eastward migration pathways were revealed in the clustering of NA1 isolates into two groups, one containing isolates from Connecticut, Oregon, and Washington and the other isolates from California and the remaining states. This finding is consistent with trace forward analyses conducted by the US Department of Agriculture's Animal and Plant Health Inspection Service. At the same time, genetic diversities in several states equaled those observed in California, Oregon, and Washington and two-thirds of multilocus genotypes exhibited limited geographic distributions, indicating that mutation was common during or subsequent to migration. Together, these data suggest that migration, rapid mutation, and genetic drift all play a role in structuring the genetic diversity of P. ramorum in US nurseries. This work demonstrates that fast-evolving genetic markers can be used to examine the evolutionary processes acting on recently introduced pathogens and to infer their putative migration patterns, thus showing promise for the application of forensics to plant pathogens.
| Sudden oak death, caused by the fungus-like pathogen Phytophthora ramorum, has caused devastating levels of mortality of live oak and tanoak trees in coastal California forests and in urban and suburban landscapes in the San Francisco Bay Area. This pathogen also causes non-lethal disease on popular ornamental plants, including rhododendrons, viburnums, and camellias. P. ramorum was discovered in California in the late 1990s and is exotic to the United States. Recently, presence of the disease in wholesale nurseries in California, Oregon, and Washington has led to shipments of diseased plants across the US, thus risking the introduction of the pathogen to other vulnerable forests. We examined the genetic diversity of this pathogen in US nurseries in order to better understand its evolution in nurseries and movement between states. We found that California populations were genetically different enough from Oregon and Washington populations that infestations of the pathogen found in nurseries in other states could be distinguished as having originated from California or the Northwest. Our inferences were consistent with trace forward investigations by regulatory agencies.
| Plant pathogens that have been introduced to a new environment may be characterized by low genetic diversity due to a genetic bottleneck experienced during the process of introduction and establishment, given that only one or a few genotypes are usually introduced [1]–[5]. Genetic diversity may also be lower on the margins of an epidemic or in founder compared to older populations [6]–[9]. In some cases the absence of a mating type may limit the pathogen to clonal reproduction and contribute to its reduced genetic diversity, yet clonality does not necessarily prevent continued evolution. Phytophthora infestans, causal agent of potato and tomato late blight, is a well known example of a plant pathogen able to adapt while reproducing clonally, as observed by changing virulence on host cultivars [10]. Stepwise evolution of new pathotypes in a single clonal lineage has also been observed for stripe rust of wheat, Puccinia striiformis f.sp. tritici, in Australia and New Zealand [4]. The increasing development and availability of polymorphic neutral genetic markers have allowed for detailed exploration of the genetic variation contained within clonal lineages [11]–[13].
Genetic markers are also beginning to be used for forensic purposes in human pathogens. Microbial forensics is “the detection of reliably measured molecular variations between related microbial strains and their use to infer the origin, relationships, or transmission route of a particular isolate” [14]. This approach has been taken to examine high-profile HIV outbreaks and transmission events [15],[16] and characterize anthrax strains associated with bioterror attacks [17],[18]. Forensics requires a sound scientific foundation, including knowledge of the genetic diversity within and among populations of the organism of interest and the evolutionary forces and genetic mechanisms that shape this diversity [19],[20]. The population genetic base required for forensic work remains weak for many plant pathogens that pose economic or environmental threats [19].
Phytophthora ramorum, the causal agent of sudden oak death, was recently introduced to North America and is responsible for the rapid decline of forest populations of tanoak (Lithocarpus densiflorus) and coast live oak (Quercus agrifolia) in northern California coastal forests and parts of coastal southern Oregon [21],[22]. P. ramorum is also a foliar and twig pathogen on common ornamentals, such as Rhododendron, Viburnum, Pieris, and Camellia. Thus, P. ramorum has been found in nurseries in North America and Europe, and nursery shipments have been implicated in the movement of the pathogen. There is serious concern about the inadvertent transfer of P. ramorum to other susceptible ecosystems, such as the Appalachians [23]. P. ramorum has had significant economic and societal impacts [22],[24],[25].
P. ramorum is a diploid oomycete, located in the kingdom Stramenopila along with diatoms, golden-brown algae, and brown algae [26],[27]. Fast-evolving microsatellites in P. ramorum have confirmed the clonal reproduction of this pathogen and have proved valuable for examining its population structure [12],[13],[28],[29]. Three distinct clonal lineages of P. ramorum have been found in nurseries [28],[30]. These lineages appear to have been evolutionarily isolated for at least 100,000 years [31], which together with their initial geographic distributions suggests that there were three introductions of this pathogen to North America and Europe [32]. The lineages have been given the names NA1, NA2, and EU1 by consensus agreement within the P. ramorum research community [33]. The NA1 lineage has been the most frequently isolated lineage from US nurseries and is the cause of oak and tanoak mortality in US forests [13],[28]. The EU1 lineage was initially confined to European nurseries, but is now also found in European parks and North American nurseries [34]–[36]. The third lineage, NA2, has only been documented in North American nurseries [28],[36]. P. ramorum is self-sterile; sexual reproduction requires contact between two different mating types. All tested NA1 and NA2 isolates have been mating type A2 and EU1 isolates mating type A1 with the exception of rare finds of A2 in Belgium [37]. Sexual reproduction has not yet been observed in nurseries where both mating types have been found [34].
Most of the P. ramorum-positive nurseries have been in California, Oregon, and Washington, where annual inspection and sampling is required for nurseries that ship interstate and contain host or associated host plants on the P. ramorum host lists per the Federal Interim Rule of 2007 (7 CFR 301.92). West Coast nurseries that ship non-host nursery stock interstate are also required to be inspected annually. When found, infected plants are quarantined and destroyed under the authority of the Plant Protection Act of 2000. P. ramorum has also been found in states that received shipments from infected West Coast nurseries. For example, shipments of millions of potentially infected plants were made from a large California nursery to over 1,200 nurseries in 39 states in 2004 [25]. When a nursery has been confirmed as infested with P. ramorum and it has been determined that the nursery shipped potentially infected P. ramorum host or associated host plants, the nursery is required to provide to the US Department of Agriculture's Animal and Plant Health Inspection Service (USDA APHIS) a list of all host and associated host plants that were shipped from the nursery during the preceding 12 months. A trace forward protocol (http://www.aphis.usda.gov/plant_health/plant_pest_info/pram/) is implemented to determine whether the receiving nurseries or landscapes have become infested. Similarly, a trace back protocol is implemented at the infested shipping nursery to investigate the potential source of P. ramorum.
Previous studies examining neutral genetic variation in nursery populations of P. ramorum using mitochondrial DNA sequence, AFLP, or microsatellites have focused on the broad diversity of a worldwide sample of P. ramorum isolates [28],[30],[38] and specifically on Oregon [13], California [12], or West Coast [29] populations using isolates collected through 2005. These studies have shown genetic similarity between 2004 nursery isolates and early California forest infestations [12] and migration among West Coast populations in the first half of this decade [29]. The Oregon forest population is an apparent exception to the frequent migration between California, Oregon, and Washington, as it is genetically differentiated from both California forest and Oregon nursery populations [13]. Thus far, microsatellites have been the most informative markers for examining population structure and migration.
Here we report on the population genetic analysis of P. ramorum in US nurseries using 279 isolates collected from infected nurseries from across the US between 2004 and 2007. There is interest in the P. ramorum community in using genetic markers to link new detections of P. ramorum in both nursery and wildland settings to possible sources; therefore, we typed microsatellite loci known to show variation within and between the P. ramorum clonal lineages to examine their utility in confirming or contributing to trace forward and trace back investigations and, more generally, the potential for forensic analysis of P. ramorum. We specifically address four major questions regarding nursery populations of P. ramorum: 1) Do nursery populations show genetic diversity and population structure or are they dominated by a single dominant or founding genotype? 2) Are West Coast infestations more genetically diverse than those in other states, as might be expected if infestations are older and effectively larger in Oregon, Washington, and California? 3) Have the populations of the West Coast states changed between 2004 and 2007 in a way that would indicate that eradication measures have or have not been effective? 4) Can we use these genetic markers to infer the major migration pathways and potential sources of recent migrants?
All 279 isolates produced multilocus genotypes that could be unambiguously assigned to one of the three known P. ramorum lineages and no recombinant multilocus genotypes were observed that would be indicative of sexual reproduction between lineages. Thirty-four EU1 isolates and 17 NA2 isolates were identified in the sample (Table 1). EU1 isolates were found in California (CA), Oregon (OR), and Washington (WA) and produced two genotypes (Figure 1), which differed by two repeats at locus 64. OR and WA isolates were all identical, while all but one of the CA isolates were the second genotype. All of the NA2 isolates were from WA and produced identical genotypes except for one isolate from 2004 (Table 1, Figure 1), which differed by one repeat at both alleles of locus PrMS43a.
The NA1 lineage was the most common and genetically variable lineage in US nurseries, found in all sampled states. We found 53 different multilocus genotypes among the 228 NA1 isolates, including three genotypes with null alleles at PrMS43b (Figure 1, Tables S1 and S2). Unique to the NA1 lineage was apparent uniform homozygosity at loci PrMS39b, PrMS43a, and PrMS43b. These loci also exhibited high numbers of alleles among NA1 isolates relative to the other genotyped loci (Table S1). Loss of heterozygosity was observed for two isolates at locus PrMS45 and one isolate at locus 64.
Sample sizes from many states were very small, e.g. one isolate from one infested nursery in the state (Tables 1 and S3). For sample sizes up to about 15 isolates, there was a positive linear relationship between sample size and number of resulting multilocus genotypes, such that for the NA1 clonal lineage every five additional isolates produced around 3 additional multilocus genotypes (Figure S1). The relationship between sample size and genotypes changed at higher sample sizes and the number of multilocus genotypes was instead correlated with the number of infected nurseries in the state.
The lineages are separated by large genetic distances (Figure 1) and reproduction appears to be completely clonal [12],[13],[28], therefore the three lineages were considered separately. Furthermore, the paucity of EU1 and NA2 isolates and genotypes precluded the need for extensive analysis of these lineages and hence our analyses focused on NA1 isolates. We examined the genotypic diversity, genotypic evenness, and genotypic and allelic richness of NA1 samples by state and year for those with sample sizes of five or more isolates (Table 1). Importantly, given the variation in sample sizes among states, we used rarefaction to estimate genotypic and allelic richness for a standardized sample size of five isolates. Interestingly, genotypic richness in the Connecticut (CT), Georgia (GA), Texas (TX), and Virginia (VA) samples were at levels seen in the West Coast states. Evenness is expected to be influenced by differences in sampling intensity, but tended to decrease over the sampled years in CA and WA. Private alleles were found in OR, WA, TX, and VA. A larger number of states produced multilocus genotypes that were not observed elsewhere (Table S2).
Minimum spanning networks revealed qualitative differences among states and years for the West Coast (Figure 2). By 2007, samples from all three states produced relatively compact networks, indicating that these populations had been limited to a small number of mostly closely related genotypes. The change over time was most evident in the Washington networks, in which there were long chains of genotypes prior to 2007. Private genotypes were generally on the margins of the networks and sometimes were only distantly related to the other genotypes, suggesting that they were either the result of rare mutation events or were immigrants from locations with intermediate genotypes.
We also examined the minimum spanning networks for other states represented by five or more isolates to compare them to the West Coast states (Figure 3). These networks generally showed populations of closely related genotypes. The most common multilocus haplotype in each of the four networks corresponded to one of the two most common multilocus genotypes in the overall sample (either MG 1 or 2 in Table S2) and may thus be the founding genotype. The outlying haplotypes in the networks were often private genotypes.
We tested for significant genetic variation among West Coast states and years using analysis of molecular variance. We found significant variation among years within states, but more variation among states and within states and years (Table 2). Examination of CA, OR, and WA individually showed that variation among years accounted for 0% (P = 0.27), 3.0% (P = 0.13), and 4.9% (P<0.0001) of the total variation, respectively. When data were clone corrected there was significant variation among states but not among years within states (Table 2).
Structure 2.2 and BAPS 5.2 were used to cluster NA1 isolates, without regard to state or year of isolation, into underlying groups. The Structure analysis produced the highest likelihood for two groups (posterior probability that K = 2 was 1.00). AMOVA confirmed significant variation between these groups, which accounted for 33% of the variation. Ten isolates could not be assigned to one or the other group with a probability greater than 0.75 and 31 isolates were not assigned with a probability greater than 0.95 (Figure 4). The optimal partitioning of isolates by BAPS produced 18 clusters (Figure 5). However, these 18 clusters formed two overall groups that largely coincided with the two Structure groups (Figure 5). AMOVA on the BAPS groups indicated that the two overall groups were responsible for 27% of the variation and the clusters within the larger groups explained another 27% of the variation. K-means clustering of individuals based on either allele frequency or AMOVA also produced the best result for two groups based on Calinski and Harabasz's pseudo-F [39]. Differences in group assignment between Structure, BAPS, and k-means clustering were limited to twelve isolates, all of which produced low posterior probabilities for group assignment in Structure. Many states were represented by mostly one group or the other, but there were also mixed populations (Figures 4 and 6). Both groups were represented in WA in all years, OR in 2004, CA in 2006, and GA with high probability. Structure outputs the overall allele frequencies and frequencies within each resulting group, which showed that particular loci and alleles were highly influential in determining group assignments (Table 3). For example, allele 246 of locus PrMS39b had an overall frequency of 0.303, but a frequency of 0.936 in group 2.The influential alleles differed by only one repeat from each other, suggesting that these groups may not be robust to repeated and reverse mutation.
The relative rates of immigration to mutation among West Coast states and from these states eastward were estimated using a coalescent-based approach, as implemented in the program Migrate. We used a migration model in which the three West Coast source populations could both send and receive migrants, but the combined population representing all other states could only receive immigrants. This migration model is consistent with nursery industry shipment patterns. The ratio of immigration rate to mutation rate (m/µ) tended to be higher for the non-West Coast sample, but with a large amount of uncertainty in the estimates (Figure 7). Many of the estimates were not significantly greater than 1.0, indicating that mutation and drift were often more important than migration in generating population genetic variation.
When a nursery that ships P. ramorum host and associated host plants out of state is confirmed to be infested with P. ramorum, the USDA APHIS trace forward protocol is implemented by the receiving state(s). Shipping records are obtained for all host and associated host plants that were shipped in the preceding 12 months. These shipping records are used to conduct inspections to determine whether receiving nurseries or landscapes have become infested. Trace forward shipments from P. ramorum infested nurseries in CA, OR and WA to non-West Coast states resulted in the detection of P. ramorum in 12 states from 2004 to 2007 (Figure 6). In 2004, all but three of the confirmed trace forward detections originated in CA. The remaining three were from OR to CT (2 detections) and MD (1 detection). Additional states received shipments from P. ramorum infested nurseries; however, the movement of any infected plants was not determined or confirmed.
Our analysis of the genotypic diversity of P. ramorum isolates from US nurseries revealed two genetic groups in the NA1 lineage. The composition of these groups suggests that many of the isolates collected in non-West Coast states were associated with California genotypes whereas the Connecticut infestation more closely resembled Oregon and Washington genotypes. This is in agreement with trace forward analyses by USDA APHIS, which established major shipments of P. ramorum-positive plants from California to nurseries across the country and smaller shipments from an Oregon nursery to Connecticut in 2004. The 2004 California shipments also sent P. ramorum-positive plants to Oregon and Washington, perhaps explaining representation from both genetic groups in these states' samples. Migration between all three West Coast states was also inferred by Prospero et al. [29] based on genotyping of California forest isolates and Oregon and Washington nursery isolates collected from 2003 to 2005. Yet, the clustering of isolates into two groups appeared to be highly influenced by three loci that show rapid evolution in NA1. This suggests that over time isolates could mutate between groups and thus grouping based on these markers may not be robust in the long term. The states with representatives from both groups tended to be those with higher numbers of multilocus genotypes and higher genotypic diversities, which could be explained by either more migration to these states or larger populations with more opportunities for mutation. For example, the networks of Washington isolates included chains of genotypes differing by single mutational steps yet assigned to different groups, suggesting that these mixed populations could be the result of large and diverse infestations.
P. ramorum isolates from nineteen states were examined and only five states were found that did not contain the most common genotype in the overall sample (NA1 multilocus genotype 1). Two of these, Connecticut and North Carolina, produced isolates with the second and third most common NA1 genotype, respectively. This suggests that only a few genotypes may be responsible for initiating P. ramorum infestations across the US. This is again consistent with USDA APHIS analysis, which indicated that shipments of infected plant material occurred only a few times. P. ramorum is also present in nurseries in British Columbia, Canada [36],[40] and there has been movement of the pathogen between BC and West Coast states every year since 2003 based on USDA APHIS trace data.
The most genetically variable populations were on the West Coast, as expected based on the large number of infected nurseries that have been found in these states (Table S3), yet we also found relatively diverse samples when we had five or more isolates from other states. The observed variation is likely related to the number of infested nurseries sampled and perhaps also to how long the infestations went undiagnosed, information that we do not have. Georgia and Texas had 14 and 11 confirmed positive nurseries, respectively, which could help explain the observed levels of variation, but Connecticut had only three and Virginia two positive nurseries (Table S3). More extensive sampling within nurseries would be required to elucidate the population structure in infested nurseries as our results suggest that we did not achieve saturation in sampling the diversity of nursery populations. In general, rapid detection and eradication should result in small effective population sizes and low genetic diversities. As the genotyping of nursery isolates becomes increasingly routine, more samples per nursery are being retained for genotyping. In fact, sampling appeared to be nearing saturation in 2007 for California and Washington nurseries.
Providing an interesting contrast to the single genotype shared among many states, we identified 36 NA1 multilocus genotypes that were unique to a state. Destruction of infected plants should ensure that populations in individual nurseries do not have the opportunity to grow large and small populations are subject to genetic drift. The observed genetic diversity and number of private genotypes suggests that there is also rapid mutation following the founding of a new nursery population and little to no gene flow following initial introduction. Interestingly, in California, several recently established P. ramorum forest populations (<5 yrs old) were observed to be as diverse as older forest populations (>10 yrs) and the genetic distance among new populations was greater than that observed among older populations [12], suggesting that a similar process of rapid mutation, genetic drift, and limited gene flow may characterize newly founded populations in both forest and nursery environments.
From 2004 to 2007 NA1 populations in West Coast nurseries appeared to become increasingly dominated by a few closely related genotypes and in 2007 all three states produced compact minimum spanning networks. This pattern is particularly striking for Washington, from which we had the largest numbers of isolates and observed high genotypic and allelic richnesses, and suggests that in 2007 there were fewer nodes of infection or earlier detection and eradication of infections. In fact, West Coast states had many fewer P. ramorum-positive nurseries in 2007 than in previous years (Table S3).
Prospero et al. [13] examined P. ramorum isolates from Oregon nurseries collected in 2003 and 2004, finding four NA1 genotypes in 2003 and six in 2004. Although each year was dominated by two closely related genotypes, there were no genotypes in common between years, which suggested that the 2003 nursery infestations were eradicated and the 2004 infestations were new introductions. In our sample of Oregon nursery NA1 isolates from 2004 through 2007, we did not find significant genetic variation across years. Of 13 multilocus genotypes found in Oregon, 4 of these were found in more than one year and 3 additional genotypes differed by one repeat from a genotype found in multiple years. Thus, some genotypes may have persisted in Oregon nurseries. However, the most common Oregon multilocus genotype (NA1 MG 2) was found in 2004, 2005, and 2006 but not 2007.
California nursery populations were dominated by a single genotype (NA1 MG 1), comprising 20 of the 36 isolates from the state. Mascheretti et al. [12] found the same dominant genotype in their nursery sample, which was also a common genotype in the California forests. This genotype has been observed in nurseries since 2004, thus it is either not being eradicated from nurseries or is re-colonizing nurseries from forest populations.
Given the levels of heterozygosity observed at most of the microsatellite loci [13],[28] and in the nuclear genome [41], the consistent homozygosity at loci PrMS39b, PrMS43a, and PrMS43b is unexpected. Loci PrMS45 and 64 were also heterozygous in all but three isolates and had large differences in allele sizes, therefore this limited homozygosity was likely a result of mitotic recombination. Mitotic recombination generally refers to crossing-over during mitosis, which results in the loss of heterozygosity at all loci distal to the chromosomal breakpoint. Loss of heterozygosity may also be the result of mitotic gene conversion, in which case only a small segment of the chromosome is altered. Mitotic recombination is thought to be responsible for frequent observations of loss of heterozygosity in P. infestans allozymes [42] and P. cinnamomi microsatellites [11]. Mitotic gene conversion has been observed in P. sojae [43]. Mitotic recombination or gene conversion may also provide an explanation for the homozygosity at PrMS39b, PrMS43a, and PrMS43b, where it must occur at a very rapid rate as these are also fast-evolving loci. It is also possible that these three loci are hemizygous or heterozygous for a null allele [11] or that intermediate genotypes have simply not been sampled.
Mitotic recombination may purge deleterious mutations from Phytophthora populations in the absence of sexual reproduction and unmask recessive traits or advantageous new mutations [11],[42]. However, the eradication of infections in US nurseries results in small effective population sizes and populations likely to be structured by genetic drift rather than natural selection. The major effect of mitotic recombination on nursery populations may be to increase the genetic distance between isolates as new mutations are made homozygous and passed on to asexual progeny. The genetic diversity among populations that could conceivably be created by this process may benefit the pathogen in the long term if in the future these populations are allowed to grow unchecked, which would allow natural selection to weed out the more fit recombinants from the less fit. Meiotic recombination through sexual reproduction would further benefit these populations by breaking linkages between beneficial and detrimental mutations. Limiting the distribution of the EU1 lineage, which is primarily the A1 mating type, and its proximity to NA1 and NA2 lineages (A2 mating type) will reduce the possibility of sexual reproduction.
The EU1 clonal lineage has now been found in all three west coast states [34], yet detectable genetic diversity in both this lineage and the NA2 lineage remain low. This could be due to the hypervariability of several of the microsatellite loci in NA1 but not EU1 and NA2, the more recent introduction of EU1 and NA2 to North America, and/or smaller population sizes of these lineages in US nurseries compared to the NA1 lineage. The recent finding of a single nucleotide polymorphism in the mitochondrial DNA of the NA1 lineage implies that this lineage may have a larger effective population size than the other two lineages [30].
The rapid mutation rates of these microsatellite loci has proven valuable for population genetic analyses, but poses a challenge for forensic tracing of P. ramorum when mutation rates are as high as appears to be the case for the PrMS43a and PrMS43b loci in the NA1 lineage. For example, an isolate of interest may differ from a suspected source population at one of these loci, thus raising doubts about their connection. Alternatively, convergence through repeat or reverse mutations may also have caused some Washington isolates to cluster with isolates from California and other states, which could falsely imply a direct connection between states where there is none. On the other hand, the relative homogeneity of the EU1 and NA2 lineages in US nurseries may hinder genetic-based tracing of isolates in these lineages. Nevertheless, our results were consistent with trace forward analyses and thus these microsatellites should be informative when used in conjunction with other data. The identification of more microsatellite loci that exhibit variation within the clonal lineages would strengthen these inferences [32],[44],[45].
Continued genotyping of P. ramorum from nurseries will be necessary to track the movement and diversification of the lineages and to identify new dominant genotypes, newly introduced lineages, or recombinant genotypes. As part of our efforts, the clonal lineage of each P. ramorum isolate genotyped, with permission from the provider of the isolate, is posted to a public website along with its county and state of origin at http://oregonstate.edu/~grunwaln/index.htm. Ongoing genotyping will also be valuable in evaluating how effective eradication efforts are in restricting migration, lowering effective population size, and increasing the effect of genetic drift.
Isolates of P. ramorum were obtained from scientists with State Departments of Agriculture, the US Department of Agriculture's Animal and Plant Health Inspection Service, universities and research institutions as new or recurring findings of infected nurseries occurred. Newly infected sites are subject to federal quarantine and could not be systematically sampled. Thus sampling intensity likely varied by state. For example, isolates from non-West Coast states may each represent one infested nursery, whereas recent samples from OR and WA include multiple isolates per nursery. Isolates for which we had detailed host information came from Camellia japonica, C. sasanqua, C. bonsai, Kalmia latifolia, Laurus noblis, Osmanthus heterophyllus, O. fragrans, Pieris japonica, Rhododendron spp., Viburnum tinus, and from soil and water baits. The 2004 shipments from CA to 39 states contained Camellia species. We do not know how many nurseries with recurrent infestations that were sampled over 2 or more years are represented in our dataset.
Upon receipt, isolates were transferred to cleared 20% V8 agar medium (200 ml V8 juice; 2 g CaCO3; 30 mg/L β-sitosterol (EMD Chemicals, Inc., San Diego, CA); 15 g agar; 800 ml deionized water) and stored at 20°C in the dark. All isolates were handled following the standard operating procedures associated with corresponding USDA APHIS permits and an exemption from the Director of the Oregon Department of Agriculture for work with P. ramorum under containment conditions.
Six microsatellite loci were genotyped that had previously shown variation among isolates within the P. ramorum clonal lineages: PrMS39b, PrMS43a, PrMS43b, PrMS45 [13], 18, and 64 [28]. These loci are also differentiated between lineages. Three additional loci that are invariable within lineages, PrMS6, Pr9C3, and PrMS39a, were also genotyped. Genomic DNA was extracted from mycelia grown in cleared 20% V8 broth using the FastDNA SPIN kit (MP Biomedicals LLC, Solon, OH) following the protocol for yeast, algae, and fungi. Loci were amplified using primers and protocols as outlined in [28] and [13]. PrMS6, Pr9C3, PrMS39a and b, and PrMS45 were amplified using a PCR program of 1 cycle of 92°C for 2 min, followed by 30 cycles of 92°C for 30 s, 52°C for 30 s, 65°C for 30 s, and 1 cycle of 65°C for 5 min. Fluorescent multiplex PCR reactions were performed in 10-µL volumes with the following final concentrations: 1× GenScript PCR Buffer (10 mM Tris-HCl; 50 mM KCl; 1.5 mM MgCl2; 0.1% Triton X-100 buffer), 0.2 µM dNTPs, 3–6 µM of primer pairs, 0.5 U GenScript Taq DNA polymerase (Genscript Corporation, Piscataway, NJ), and 0.5 µL (∼50 ng) DNA template. Loci PrMS43a and b were amplified using the following PCR program: 1 cycle of 92°C for 2 min, 35 cycles of 92°C for 30 s, 52°C for 30 s, and 72°C for 1 min, and 1 cycle of 72°C for 45 min. The final concentrations of the reaction mixture for PrMS43 (10 µL volume) were 1× PCR Buffer, 0.4 µM dNTPs, 0.3 µM forward and reverse primers, 1.0 U DNA polymerase, and 0.5 µL DNA template. Loci 18 and 64 were amplified with the PCR program: 1 cycle of 94°C for 2 min, 30 cycles of 94°C for 20 s, 55°C for 20 s, and 72°C for 30 s min, and 1 cycle of 72°C for 10 min. The final concentrations of the 10 µL reaction mixture for 18 and 64 were 1× PCR Buffer, 0.2 µM dNTPs, 0.2 µM forward and reverse primers, 0.5 U DNA polymerase, and 0.5 µL DNA template.
Three isolates were used as positive controls in identification of the three clonal lineages and to ensure consistency among runs: PR-04-001 (aka 2027.1, lineage NA1 from Curry County, Oregon), PR-04-020 (aka 03-74-D12-A, EU1 from an Oregon nursery), and PR-04-015 (aka wsda3765, NA2 from a Washington nursery). PCR products were sized using capillary electrophoresis on an 3100 Avant Genetic Analyzer (Applied Biosystems, Foster City, CA) using the internal size-standard LIZ 500 (Applied Biosystems). Results were analyzed using GeneMapper 3.7 software packages (Applied Biosystems). Genotyping was replicated for a subset of isolates with independent DNA extractions, PCR, and sizing of fragments. Reproducibility of novel allele sizes was confirmed.
Genetic distances among all identified multilocus genotypes were calculated over eight of the nine loci using Wright's modification of Roger's genetic distance [46],[47] as implemented in the program TFPGA [48]. PrMS39a was excluded from the calculation because it was invariable in NA1 isolates and did not amplify in the other two lineages. Null alleles were coded as missing data. A UPGMA dendrogram was inferred from the distance matrix and visualized using MEGA version 4 [49]. Bootstrapping of data was conducted in TFPGA using 1,000 permutations.
For the NA1 lineage and states and years with sample sizes of at least five isolates, we estimated multilocus genotypic diversity using Stoddart and Taylor's index G [50], multilocus genotypic evenness (the distribution of genotypes in a sample) using the index E5 [51],[52], and multilocus genotypic richness and allelic richness (average number of alleles per locus) corrected for sample size using rarefaction as implemented in ADZE [53].
Analysis of molecular variance (AMOVA) [54] was conducted using Arlequin 3.1 [55] to test for significant variation among years in CA, OR, and WA. The analyses used the standard data setting and 10,000 permutations.
In order to examine genetic distances among isolates as measured by mutational differences, rather than mutation plus mitotic recombination, we collapsed the data to the haploid state. Three loci, PrMS39b, PrMS43a, and PrMS43b, were consistently homozygous. Loci 18 and 64 had two distinct size classes of alleles and only the larger of the two was variable among isolates for both loci. Thus when collapsed to haploid, the larger allele was retained. Three additional isolates appeared to exhibit mitotic recombination rather than mutation at otherwise uniformly heterozygous loci. These were two WA 2004 isolates (locus PrMS45) and a SC 2004 isolate (locus 64), multilocus genotypes 49, 53, and 35, respectively (Tables S1 and S2). These isolates were excluded from the haploid data set. PrMS45 was monomorphic across remaining NA1 isolates and PrMS6 and Pr9C3c were invariable within NA1.
To examine the relationships among isolates, minimum spanning networks were constructed using the genetic distance of Bruvo et al. [56], which incorporates microsatellite repeat number. Here, a distance of 0.10 is equivalent to one mutational step (one repeat) but larger distances do not strictly correspond to a given number of mutational steps. Genetic distance matrices were calculated for the three West Coast states for all available years and for the 2004 samples from CT, GA, TX, and VA using the haploid dataset. MINSPNET [57] was used to generate minimum spanning networks from genetic distance matrices. All tied trees were included in the network, which was visualized using the neato program in the Graphviz package [58].
To examine genetic structure in the NA1 sample, the clustering programs Structure 2.2 [59],[60] and BAPS 5.2 [61] were run using the haploid data set. For Structure 2.2 we used the no admixture model, because the NA1 lineage appears to be completely clonal, and assumed that allele frequencies among populations were correlated. However, very similar results were obtained using the admixture model and independent allele frequencies. Lambda was set to 1.0 and 100,000 MCMC replicates were used after a burn-in of 20,000. The number of underlying groups (K) was varied from 1 to 5 and replicated five times. The posterior probability of the most likely K was calculated assuming a uniform prior as described in the Structure 2.2 documentation. Genetic mixture analysis was run at the individual level in BAPS 5.2 for maximum number of populations (K) from 2 to 31, replicated 3 times. A UPGMA dendrogram of the resulting clusters was produced using Nei's distance as implemented by the program. AMOVA was conducted on the resulting Structure and BAPS clusters. Structure and BAPS results were also compared to those obtained from k-means clustering of individuals as implemented in Genodive [62], which does not assume Hardy-Weinberg equilibrium within populations.
Maximum likelihood estimates of the ratio of immigration rate to mutation rate (m/µ) for West Coast states compared to the non-West Coast sample were obtained using the program Migrate version 2.4.3 [63]–[65]. Isolates from all years were divided into four populations: CA, OR, WA, and all other states. All years of collection were combined to obtain larger population sizes for parameter estimates. The data were coded such that the homozygous loci had one missing allele, to account for the possibility of homozygosity by mitotic recombination rather than mutation. The analysis used a migration model in which the three West Coast source populations could both send and receive migrants, but the fourth combined population could only receive immigrants. We used the Brownian motion approximation to the stepwise mutation model and a search strategy of 10 short chains of 500 steps followed by 3 long chains of 10,000 steps at the default sampling increments with 3 heated chains using the adaptive heating scheme. The search strategy was replicated five times for each locus within each run such that the last chains of each replicate were combined for parameter estimation. Runs for which the profile likelihood calculation failed were discarded. A total of four runs were examined to account for possible variation among runs.
|
10.1371/journal.pmed.1002553 | Universal versus conditional day 3 follow-up for children with non-severe unclassified fever at the community level in Ethiopia: A cluster-randomised non-inferiority trial | With declining malaria prevalence and improved use of malaria diagnostic tests, an increasing proportion of children seen by community health workers (CHWs) have unclassified fever. Current community management guidelines by WHO advise that children seen with non-severe unclassified fever (on day 1) should return to CHWs on day 3 for reassessment. We compared the safety of conditional follow-up reassessment only in cases where symptoms do not resolve with universal follow-up on day 3.
We undertook a 2-arm cluster-randomised controlled non-inferiority trial among children aged 2–59 months presenting with fever and without malaria, pneumonia, diarrhoea, or danger signs to 284 CHWs affiliated with 25 health centres (clusters) in Southern Nations, Nationalities, and Peoples’ Region, Ethiopia. The primary outcome was treatment failure (persistent fever, development of danger signs, hospital admission, death, malaria, pneumonia, or diarrhoea) at 1 week (day 8) of follow-up. Non-inferiority was defined as a 4% or smaller difference in the proportion of treatment failures with conditional follow-up compared to universal follow-up. Secondary outcomes included the percentage of children brought for reassessment, antimicrobial prescription, and severe adverse events (hospitalisations and deaths) after 4 weeks (day 29). From December 1, 2015, to November 30, 2016, we enrolled 4,595 children, of whom 3,946 (1,953 universal follow-up arm; 1,993 conditional follow-up arm) adhered to the CHW’s follow-up advice and also completed a day 8 study visit within ±1 days. Overall, 2.7% had treatment failure on day 8: 0.8% (16/1,993) in the conditional follow-up arm and 4.6% (90/1,953) in the universal follow-up arm (risk difference of treatment failure −3.81%, 95% CI −∞, 0.65%), meeting the prespecified criterion for non-inferiority. There were no deaths recorded by day 29. In the universal follow-up arm, 94.6% of caregivers reported returning for reassessment on day 3, in contrast to 7.5% in the conditional follow-up arm (risk ratio 22.0, 95% CI 17.9, 27.2, p < 0.001). Few children sought care from another provider after their initial visit to the CHW: 3.0% (59/1,993) in the conditional follow-up arm and 1.1% (22/1,953) in the universal follow-up arm, on average 3.2 and 3.4 days later, respectively, with no significant difference between arms (risk difference 1.79%, 95% CI −1.23%, 4.82%, p = 0.244). The mean travel time to another provider was 2.2 hours (95% CI 0.01, 5.3) in the conditional follow-up arm and 2.6 hours (95% CI 0.02, 4.5) in the universal follow-up arm (p = 0.82); the mean cost for seeking care after visiting the CHW was 26.5 birr (95% CI 7.8, 45.2) and 22.8 birr (95% CI 15.6, 30.0), respectively (p = 0.69). Though this study was an important step to evaluate the safety of conditional follow-up, the high adherence seen may have resulted from knowledge of the 1-week follow-up visit and may therefore not transfer to routine practice; hence, in an implementation setting it is crucial that CHWs are well trained in counselling skills to advise caregivers on when to come back for follow-up.
Conditional follow-up of children with non-severe unclassified fever in a low malaria endemic setting in Ethiopia was non-inferior to universal follow-up through day 8. Allowing CHWs to advise caregivers to bring children back only in case of continued symptoms might be a more efficient use of resources in similar settings.
www.clinicaltrials.gov, identifier NCT02926625
| As a result of declining malaria prevalence and increased use of malaria diagnostic tests, it is becoming more common for children seen by community health workers (CHWs) to have non-severe unclassified fever.
Caregivers of children seen on day 1 with non-severe unclassified fever are advised to bring the child back to the CHW on day 3 for reassessment, regardless of whether symptoms have resolved or not (universal follow-up), burdening both the family and health system.
This study assessed the safety of CHWs following-up children with non-severe unclassified fever only when symptoms have not resolved (conditional follow-up), hypothesizing that the conditional follow-up would be as safe as universal follow-up.
From December 1, 2015, to November 30, 2016, a total of 4,179 children were enrolled into the study in the Southern Nations, Nationalities, and Peoples’ Region in Ethiopia, of whom 3,946 (1,953 universal follow-up arm; 1,993 conditional follow-up arm) adhered to the CHW’s follow-up advice and also completed a day 8 study visit within ±1 days.
Caregivers’ adherence with follow-up advice given by CHWs was high: 93.5% of caregivers followed the advice that was given in line with the cluster to which the CHW was assigned.
On day 8, a total of 2.7% of the children had treatment failure (persistent fever, development of danger signs, hospital admission, death, malaria, pneumonia, or diarrhoea), with no significant difference between the 2 follow-up arms.
There were no deaths recorded by day 29.
Conditional follow-up of children with non-severe unclassified fever in a low malaria endemic setting was as safe as universal follow-up.
Allowing CHWs to advise caregivers to return only if a child has continued symptoms may be more efficient than their advising all children to return.
| Mortality in children under 5 years is estimated at 43/1,000 live births globally and 82/1,000 live births in sub-Saharan Africa. This corresponds to the death of 5.6 million children under 5 years old globally each year, 2.8 million in sub-Saharan Africa alone [1].
Although there have been substantial improvements over the last 2 decades, high levels of child mortality persist in many countries, including Ethiopia, where the under-5 mortality rate is estimated at 58/1,000 live births [1,2]. A large proportion of deaths globally are caused by infectious diseases such as pneumonia (15.5%), diarrhoea (8.9%), and malaria (5.2%) [3]. In response, many countries in sub-Saharan Africa have introduced integrated community case management (iCCM), where community health workers (CHWs) are trained to assess, classify, and treat uncomplicated cases of pneumonia, diarrhoea, and malaria in children under 5 years, and refer children with danger signs and malnutrition for facility-based care [4]. While the mortality impact of iCCM has been difficult to demonstrate [5], there is clear evidence that it can increase the treatment rate among sick children [6].
As per WHO iCCM guidelines [7], children diagnosed by a CHW with a non-severe illness are given treatment and counselled to return on day 3 to assess treatment compliance and illness resolution. Children with fever but without a diagnosable cause of illness and without danger signs (i.e., unclassified fever) for whom anti-infective treatment can be withheld are also told to return to the CHW on day 3, even if the child has recovered.
However, febrile illness is common in childhood, and is often due to viruses or other self-resolving illnesses [8,9]. In a large proportion of cases, fever resolves rapidly, most often within 96 hours [10]. A number of studies have suggested that it is safe to withhold medical treatment for children with unclassified fever [11,12]. The Integrated Management of Neonatal and Childhood Illness (IMNCI) manual [13], followed by CHWs in Ethiopia (locally referred to as health extension workers [HEWs]), recommends that such children should return for a reassessment only if the illness persists or deteriorates.
There is limited evidence on which of the 2 follow-up recommendations (conditional, as in IMNCI, or universal, as in iCCM) is safer for the child. Further, it is unclear whether and to what extent caregivers of children actually come back promptly to CHWs for their conditional follow-up visit if the child is not improving, or if they come back at all if the child has improved. Bacterial infections can develop quickly, and delaying care-seeking is a major risk factor for death in both pneumonia and malaria [14,15]; hence, children with untreated persistent fever may be at risk if caregivers do not comply with the conditional follow-up advice. A universal follow-up visit for all children may promote detection of those at risk of developing severe illness. However, it could also potentially lead to delayed care-seeking for children who rapidly deteriorate at home if caregivers wait for their booked follow-up visit. In addition, the visit may add extra burden to families and CHWs, and might be unnecessary if fever has resolved [16,17]. On the caregiver side, opportunity costs and other barriers often hinder care-seeking for sick children, even when community-based providers are near and free of charge [18]. It is therefore unclear whether caregivers and CHWs would comply better with the conditional follow-up advice compared to the universal follow-up advice and whether the universal follow-up visit is even necessary.
Hypothesizing that conditional follow-up does not have a higher risk of treatment failure, the objective of this study was to assess whether conditional follow-up was non-inferior to universal follow-up for non-severe febrile illness in children aged 2 to 59 months in whom malaria, pneumonia, diarrhoea, and danger signs were absent.
This was a 2-arm community cluster-randomised non-inferiority trial (the TRAction study) carried out in 3 woredas (districts) in Southern Nations, Nationalities, and Peoples’ Region (SNNPR) in Ethiopia. The woredas were purposively selected based on (1) strength of iCCM programme (i.e., consistency in HEW supervision and supply), (2) HEW use rate among caregivers (more than 50 children assessed for fever each month over a 12-month period), and (3) regular concurrent community mobilisation and supportive supervision activities under other grants (to ensure that demand was kept high during the study period). Clusters, defined by health centre (the referral centre and practical training institution for HEWs, where their services are coordinated), were randomised into either the conditional or universal follow-up arm. All 25 health centres and 144 health posts with 284 HEWs in the 3 selected woredas were included in the study, and all children seeking care from the health posts in these clusters were potential recipients of the interventions, in addition to having access to routine care available from private and public health services. Caregivers of children who met the inclusion criteria (fever with a negative malaria rapid diagnostic test [mRDT], and in whom the HEW did not diagnose pneumonia or diarrhoea or identify other symptoms requiring referral on day 1) were counselled to either (1) return on day 3 (universal follow-up arm) or (2) return if symptoms persisted (conditional follow-up arm). Caregivers in both arms were advised to go to the health centre immediately if danger signs, such as convulsion, lethargy, not drinking/breastfeeding, or vomiting everything, developed. Of 25 included clusters, 13 were randomised to the universal follow-up arm and 12 to the conditional follow-up arm.
The government of Ethiopia has deployed over 42,000 female CHWs, or HEWs [19,20], to provide preventive, promotive, and curative health services at the community level; since 2010, the full iCCM package has been included in the IMNCI guidelines (with the addition of treatment of pneumonia) and has been scaled up in most regions of the country. There are typically 2 HEWs assigned to a health post in a sub-district with a population of 3,000–5,000; they are supported by the Health Development Army, female volunteers who enhance community engagement and encourage use of maternal and newborn health services [21]. While IMNCI recommends conditional follow-up, HEWs and their supervisors report a range of other practices for children with unclassified fever, including universal follow-up advice, immediate referral to health centres, and treatment with antimalarial tablets [22].
Children aged 2–59 months who presented to the HEWs in the study area with fever (≥37.5°C) or a history of fever, a negative mRDT, no pneumonia or diarrhoea according to iCCM criteria, and no danger signs were eligible to participate in the study. Written informed consent was obtained from each caregiver before enrolment in the study.
Cluster randomisation was at the health centre area level; the 25 study health centres had an average of 5 health posts and 7.5 HEWs each. Allocation to the universal versus conditional arm was done via restricted randomisation whereby clusters were balanced on health area estimates of (1) population size, (2) prior 6-month likelihood of mRDT-negative febrile children (number of children mRDT negative/under-5 population), and (3) geographic distance from health post to zonal referral hospital, was performed to minimise the differences between conditional and universal follow-up arms [23]. Sorting of clusters and random selection of schemes were carried out by the study statistician (MP) in Stata 13 (StataCorp, College Station, TX, US).
HEWs collected data at enrolment (day 1) using an Open Data Kit (ODK) Collect (version 1.9.1) data collection form on mobile phones, including date of enrolment, a child identifier code, and clinical indicators such as fever (axillary temperature ≥ 37.5°C or, if a functional thermometer was unavailable, hot to touch reported by HEW or caregiver-reported fever in past 2 days), cough, respiratory rate, diarrhoea, and danger signs. The enrolment data were synchronised with a server accessed by a data manager for scheduling of follow-up visits. Six independent evaluators (IEs), with a bachelor’s degree in a health-related discipline, clinical experience using IMNCI, and a minimum of 2 years’ research experience, and who were able to communicate in Amharic and English, were trained for 2 days in study procedures and in follow-up of enrolled children. Each district was assigned 2 IEs, who were blinded to the cluster allocation of the children they were reassessing.
In the conditional follow-up arm, HEWs counselled caregivers on how to detect danger signs and to seek care immediately at the health centre if danger signs developed, how to reduce fever using paracetamol, and the need to return at any point to the HEW for reassessment if symptoms remained the same, or worsened. In the universal follow-up arm, caregivers were counselled on all of the above, as well as the need to return on day 3 to the HEW for a follow-up assessment, even if the child had recovered. Caregivers in both arms were informed that a follow-up home visit by an IE would take place. Clinical outcomes were assessed by an IE during a home visit after 1 week (on day 8); if the child had not fully recovered, the child was assessed again by the IE after 2 weeks (day 15) and, if still not recovered, at after 4 weeks (day 29). Caregivers of all children were followed up by a phone call to assess vital status (alive/dead) after 4 weeks. Management of illness at any follow-up visit (i.e., return to HEW on any day, including scheduled assessments) followed national IMNCI guidelines.
IEs initially used ODK to collect reassessment data on enrolled children; halfway through the study the data collection software was changed to CommCare (version 2.38.1, Dimagi, Cambridge, MA, US), which allowed for automatic linking of follow-up forms, as well as scheduling of subsequent visits, once the children were registered in the 1-week follow-up form. The replacement system used an automatically generated child identifier code, which reduced the effort of having to manually link the forms, as well as supporting the IEs in tracking the follow-up visits that were due. The data collected during the household follow-up visits included the child identifier code, clinical data, additional antimicrobial treatment, hospitalisation, care-seeking history, and costs, as well as caregiver and household characteristics.
For children who were brought back on day 3 for reassessment in the universal follow-up arm and for any spontaneous revisit in both arms, a full reassessment was done by the HEW. If the child still had unclassified fever and a negative mRDT on reassessment, the child was referred to the nearest health centre, as recommended in the national IMNCI guidelines.
A rigorous monitoring system implemented by the study team was part of the continuous quality assurance. The data manager reviewed forms submitted to the server daily, and checked for duplicates, completeness, and accuracy before storing them in the project database. Discrepancies, overdue follow-up visits, and other issues were resolved by phone calls to the IEs and during weekly supervision meetings with field research staff. Biweekly field supervision visits to all HEWs were carried out, and district HEW supervisors were trained to monitor HEW trial activities during routine weekly group supervisions. While the protocol stated that a minimum of 10% of all enrolled cases and 50% of children with treatment failure should have a quality control reassessment by a research assistant, the actual percentage was significantly higher. Six months into the trial, all HEWs and their district supervisors had a refresher training in study procedures. In addition, the regional ethical clearance committee members did a field supervision visit during implementation of the project in all 3 districts selected for the study (9 health posts; 3 from each district), and provided feedback recommendations to the study team. The final dataset was analysed in Stata 13.
The primary outcome was treatment failure rate on day 8, defined as the proportion of children whose illness was not resolved (child had any of the following: reported fever, danger sign(s), hospital admission, death, malaria, pneumonia, or diarrhoea). Three progressively stricter, and more objective, definitions of treatment failure were added in post hoc analysis to be consistent with a concurrent sister study in the Democratic Republic of the Congo (DRC) [24]: (1) reported fever ≥3 days, danger sign, hospital admission, death, malaria, pneumonia, or diarrhoea; (2) measured axillary temperature ≥ 37.5°C, danger sign, hospital admission, death, malaria, pneumonia, or diarrhoea; and (3) danger sign, hospital admission, death, malaria, pneumonia, or diarrhoea.
Secondary outcomes were percentage of caregivers who brought the child to the HEW for the follow-up visit on day 3 in the universal follow-up arm; percentage of caregivers who spontaneously re-presented to HEW for persistence or worsening of symptoms in the conditional follow-up arm, and the timing of these visits; percentage of children receiving antimicrobial treatment in each arm; and severe adverse events in each arm. Severe adverse events were defined as hospitalisation or death. A data monitoring committee (DMC) convened twice during the study to review enrolment rates, demographic and clinical characteristics of enrolled children, and follow-up rates at 1, 2, and 4 weeks, in order to monitor the overall conduct of the study. The DMC was advisory to a study steering committee (SC), which comprised the implementing study team from Malaria Consortium and lead study investigators, who jointly had responsibility for the design, conduct, and analysis of the trial. The SC was responsible for reviewing the DMC recommendations, to decide whether to continue or terminate the study, and to determine whether amendments to the protocol or changes in study conduct were required.
We hypothesized that treatment failure at day 8 would not be more common with conditional than universal follow-up. We assumed that about 5% of children in the universal follow-up arm and 6% in the conditional follow-up arm would have treatment failure at day 8 (based on rates of 3% and 8% in previous studies [9,25]). Sample size for a non-inferiority trial was calculated in PASS 15 (NCSS, Kaysville, UT, US). Assuming that the proportion of failure was 5% in the universal follow-up arm and 6% in the conditional follow-up arm, and allowing for non-inferiority if the proportion of failure was as high as 9% in the conditional follow-up arm, a sample size of 2,142 per arm was needed to ensure that the upper limit of the 1-sided 95% confidence interval would exclude a difference in treatment failure of more than 4% with power of 80%. Using a design effect of 3 to account for clustering at the health post and health centre levels, the total sample size required was 4,284 children, with 2,142 per arm; this was inflated to 4,900 to account for potential losses to follow-up. The primary analysis was conducted on the per-protocol population (only including children for whom the primary outcome was collected on day 8 ± 1 and whose caregiver reported receiving follow-up advice from the HEW that was aligned with the study arm). In addition, an intention-to-treat analysis was done, whereby all children with a primary outcome defined were included. We also calculated cluster-specific failure rates on the per-protocol population and with the same model specifications as for the primary outcome.
The primary outcome was compared between arms using generalised linear models with a binomial distribution and identity link using a robust variance estimator, treating cluster as a random effect. We applied a conventional statistical non-inferiority test using a CI approach using the exact binomial CI for the difference in overall treatment failure between study arms. Here, we claimed non-inferiority if the upper bound of the 95% CI lay on the negative side of the 4% margin, using a 1-sided test done at the 2.5% significance level. The main analysis was done using the per-protocol population, as is appropriate for non-inferiority and equivalence studies, together with sensitivity analysis in the per-protocol and intention-to-treat populations [26]. All p-values for categorical data were calculated using the Pearson’s chi-squared test, whereas the adjusted Wald test was used for continuous data, accounting for clustering using the svy command in Stata 13.
The trial protocol was approved by the SNNPR State Health Bureau on September 23, 2015 (ref P026-19/4511). In addition, approval was obtained from the district authorities and local leaders in the study-area woredas. US Centers for Disease Control and Prevention investigators participated under a non-engaged determination from their Office for Human Research Protections. The study protocol has been published [23] and was registered (ClinicalTrials.gov; identifier NCT02926625) after the first participant was randomised due to a miscommunication between study investigators.
From December 1, 2015, to November 30, 2016, 4,784 children were eligible for enrolment; consent for enrolment was not obtained for 8 of these. In all, 4,776 children were enrolled (mean 191 per cluster [range 45–762]), but 181 were excluded due to enrolment violations (fever not reported/measured, presence of diarrhoea or pneumonia at enrolment, or outside the eligible age group) (Fig 1). The mean number of children enrolled per HEW was 20.8 (range 1–103) in the universal follow-up arm and 22.1 (range 1–166) in the conditional follow-up arm.
Baseline characteristics were well balanced between arms (Table 1). There were more children with cough in the conditional follow-up arm than in the universal follow-up arm (p < 0.001).
Follow-up was completed in December 2016, 1 month after the last patient was enrolled. Of the 4,595 children enrolled, 416 were lost to follow-up by day 8, resulting in a total of 4,179 children with the primary outcome defined, including 3,946 who met the per-protocol definition (Fig 1). Late follow-up and loss to follow-up were primarily due to difficulties in accessing the villages during the rainy season, as well as the close-down of the mobile data network (and hence inability of HEWs to send data on enrolled cases) during a state of emergency that was instituted by the Ethiopian government in October 2016.
In all, 97.3% (4,064/4,179) of caregivers reported receiving follow-up advice, and about 97% received advice that was in line with the cluster allocation (Table 2). Caregivers’ reported adherence with the advice given by the HEWs was high: 94.6% of caregivers in the universal follow-up arm reported returning to the HEW, in contrast to only 7.5% in the conditional follow-up arm (risk ratio 22.0, 95% CI 17.9, 27.2).
Overall, 106 (2.7%) of the 3,946 enrolled children had treatment failure at day 8: 0.8% (16/1,993) in the conditional follow-up arm and 4.6% (90/1,953) in the universal follow-up arm. The treatment failure rate varied by cluster and ranged from 0% to 12% (intraclass correlation coefficient 0.07), and the total number of enrolments was proportionate to cluster size for all but 3 clusters (Table 3).
The difference in treatment failure between conditional follow-up and universal follow-up was −3.81% (95% CI −∞, 0.65%) and −3.67% (95% CI −∞, 0.57%) in the per-protocol and the intention-to-treat populations, respectively (Table 4). As the difference between arms was less than the prespecified acceptable margin of 4% for the upper 95% CI, conditional follow-up was non-inferior to universal follow-up. Applying the more stringent treatment failure definitions did not change the result and further reduced the difference between the 2 arms (Table 4). The non-inferiority plot of clinical failure in the intention-to-treat and per-protocol populations is displayed in Fig 2.
All 106 children who had treatment failure at the 1-week follow-up visit were visited at 2 weeks, at which time point all had fully recovered. In all, 3,922 (99.4%) of the enrolled children were followed up at 4 weeks for the vital status check. There were no deaths during the 4-week follow-up period.
Out of the 106 children with treatment failure at 1 week, 78 (73.6%) were from 1 cluster in the universal follow-up arm. In this cluster, the cases all occurred during a 2-month period (February and March 2016), and mainly comprised children who were reported still febrile at 1 week by their caregivers. Only 3 of these had a measured temperature of ≥37.5°C. A sensitivity analysis was done excluding this cluster; this analysis showed no difference in treatment failure between arms, with a risk difference of −0.27% to 0.20% (upper 95% CI 0.57–0.77) in the per-protocol population and −0.4% to 0.04% (0.39–0.76) in the intention-to-treat population.
In the per-protocol population, only 114 (5.7%) children in the conditional follow-up arm returned to the HEW after enrolment, 62 (54.4%) of them because the child was still sick on day 3 or had deteriorated. Only 3 did not recover by the time of the 1 week follow-up, and were therefore defined as treatment failures. Of the children who had treatment failure at 1 week, 91.1% (82/90) and 6.3% (1/16) had previously returned to the HEW in the universal and conditional follow-up arms, respectively. In the intention-to-treat cohort, 155 (7.3%) of the children in the conditional follow-up arm returned to the HEW, 66 (42.6%) because the child was still sick or deteriorating. Of the children who had treatment failure at the 1-week follow-up visit, 91.5% (86/94) and 15.8% (3/19) had previously returned to the HEW in the universal and conditional follow-up arms, respectively. Hence, most children who had treatment failure at day 8 in the conditional follow-up arm were not seen again by the HEW after enrolment, and only 3 children across both arms were taken to another provider. Few children subsequently sought care from another provider after having initially been seen by the HEW: 3.0% (59/1,993) in the conditional follow-up arm and 1.1% (22/1,953) in the universal follow-up arm, on average 3.2 and 3.4 days later, respectively, with no significant difference between arms (risk difference 1.79%, 95% CI −1.23, 4.82, p = 0.244). There was no difference between arms in time or money spent: the mean travel time to the other provider was 2.2 hours (95% CI 0.01, 5.3) in the conditional follow-up arm and 2.6 hours (95% CI 0.02, 4.5) in the universal follow-up arm (p = 0.82), and the mean cost for the efforts to seek care after visiting the HEW was 26.5 birr (US$1.12) (95% CI 7.8, 45.2) and 22.8 birr (US$0.96) (95% CI 15.6, 30.0) (p = 0.69), respectively. Of those who sought care from another provider after having been seen by the HEW, 88.9% went to a health centre, 6.2% to a hospital, and 3.7% to a private clinic. Few children received additional treatment; 3.2% (63/1,993) in the conditional follow-up arm and 1.5% (29/1,953) in the universal follow-up arm (p = 0.253). Based on examination of prescription notes/medicine packs or the caregivers’ report, 26 children received antibiotics (28.3%), 10 (10.9%) received antimalarials, and 8 (8.7%) received oral rehydration therapy and/or zinc. Only 1 child in each arm was admitted to hospital.
Conditional follow-up of children with non-severe unclassified fever in a low malaria transmission setting in Ethiopia was found to be non-inferior to universal follow-up through 1 week, with an average 2.7% of children across both arms having treatment failure at day 8. No deaths were recorded. While iCCM guidelines recommend universal follow-up for all children, regardless of symptom resolution, IMNCI guidelines recommend less intense conditional follow-up after 2 or 3 days (depending on malaria endemicity). To our knowledge, this study and a sister study in DRC [24] are the first to provide evidence that conditional follow-up is no less safe or marginally less safe than universal follow-up in children aged 2–59 months seen by CHWs.
Reported fever is one of the most common presenting symptoms of paediatric illnesses; fever incidence is variable, with country-specific reports from Africa showing a mean of 5.9 fever episodes annually per child under age 5 years [27]. Fever in children signifies systemic inflammation, typically in response to a viral, bacterial, parasitic, or, less commonly, non-infectious aetiology [28]. A number of studies have been conducted to establish the specific cause of fever in children who test negative for malaria, with the vast majority of fevers caused by viruses [8,9]. Present guidelines are based on clinical features that are unfortunately poorly predictive of the diseases causing fever; hence, low-cost, accurate, point-of-care diagnostics are needed to determine which children can benefit from antimicrobial treatment [29].
Several studies from sub-Saharan Africa provide convincing evidence that mRDT-negative febrile children can be safely managed without antimalarial treatment [12,30–33]. While the overall treatment failure rate in our study (2.7%) is similar to the low rate observed in a study of malaria-negative febrile children in 2 sites in Tanzania (3%) [12], it was significantly lower than the rate observed in Zambia (9.3%) [25] or in our sister study in DRC (10.1%) [24]. One explanation may be that the relatively lower malaria endemicity in Ethiopia leads to lower numbers of false-negative cases; parasite prevalence in children under 5 years in Ethiopia is 0.6% [34] versus 10%–75% in children aged 2–9 years in Zambia [35] and 22.6% in children under 5 years in DRC [36]. However, the 14% and 40% of children positive for malaria in the 2 study sites in Tanzania [12] suggest that factors other than malaria prevalence may also play a role.
With a steady decline in malaria transmission, the role and management of unclassified fevers will become more important [37,38]. With the epidemiology and burden of paediatric febrile illness shifting, understanding the aetiology of unclassified fevers in each context, and in particular in high-burden countries, is an important next step to improve management of these cases [28].
The communication between health providers and patients about the purpose and result of mRDTs is often poor [39,40]. Furthermore, primary healthcare workers often have low confidence in managing children with fever symptoms but negative tests for malaria. Recent studies show that mRDT-negative patients with cough or difficult breathing complaints in Malawi had 16.8 times higher odds of antibiotic overtreatment than mRDT-positive patients [41]; in Burkina Faso and Uganda both community and health facility workers prescribed antimalarials to mRDT-negative patients if no other fever cause was identified, often due to parental pressure [41,42]. However, when clear case management instructions were provided, as in this study, and non-malaria fever was introduced as a diagnostic term, HEWs felt empowered to withhold medicines, while simultaneously reassuring caregivers that their child was cared for [22]. This finding is supported by the low number of children in our study who were taken to another provider and provided with secondary treatment after being enrolled by the HEW.
The robust randomised controlled trial design is a particular strength of this study. Compliance both among HEWs, who gave the follow-up advice, and among caregivers, who followed the advice, was very high, indicating that these follow-up recommendations can be easily applied to routine case management practice. Yet, in an implementation setting, it is crucial that HEWs are well trained on counselling caregivers on when to come back for follow-up, as the high adherence seen in the controlled study setting may not transfer to routine practice.
A limitation of this study is that it did not collect sufficient clinical data on children at enrolment to be able to generate an understanding of which other symptoms or possible diagnoses were present. However, almost all children were followed up until 4 weeks, and none of them died or were referred between 1 week and four weeks, indicating that no child deteriorated to a severe condition. We were not able to further investigate fever aetiology among the children who had not recovered by 1 week, and we can therefore not speculate about the possible causes for these treatment failures. However, it is unlikely that this will have affected our results, as our failure outcome was purposefully designed to include any cause for failure, and was balanced between arms. Thus, we feel comfortable arguing that if children with potentially severe illness are excluded based on the presence of danger signs, the rest can be safely managed without a requisite visit on day 3.
In conclusion, we recommend that IMNCI guidelines in Ethiopia, which stipulate conditional follow-up of children with unclassified fever, remain unchanged, as our study demonstrated the safety of this approach in comparison to universal follow-up of similar children. Allowing CHWs to advise caregivers to bring children back only in case of continued symptoms might be a more efficient use of resources in these settings.
|
10.1371/journal.pntd.0007205 | A simple score to predict severe leptospirosis | The case-fatality rate of severe leptospirosis can exceed 50%. While prompt supportive care can improve survival, predicting those at risk of developing severe disease is challenging, particularly in settings with limited diagnostic support.
We retrospectively identified all adults with laboratory-confirmed leptospirosis in Far North Queensland, Australia, between January 1998 and May 2016. Clinical, laboratory and radiological findings at presentation were correlated with the patients’ subsequent clinical course. Medical records were available in 402 patients; 50 (12%) had severe disease. The presence of oliguria (urine output ≤500 mL/24 hours, odds ratio (OR): 16.4, 95% confidence interval (CI): 6.9–38.8, p<0.001), abnormal auscultatory findings on respiratory examination (OR 11.2 (95% CI: 4.7–26.5, p<0.001) and hypotension (systolic blood pressure ≤100 mmHg, OR 4.3 (95% CI 1.7–10.7, p = 0.002) at presentation independently predicted severe disease. A three-point score (the SPiRO score) was devised using these three clinical variables, with one point awarded for each. A score could be calculated in 392 (98%) patients; the likelihood of severe disease rose incrementally: 8/287 (3%), 14/70 (20%), 18/26 (69%) and 9/9 (100%) for a score of 0, 1, 2 and 3 respectively (p = 0.0001). A SPiRO score <1 had a negative predictive value for severe disease of 97% (95% CI: 95–99%).
A simple, three-point clinical score can help clinicians rapidly identify patients at risk of developing severe leptospirosis, prompting early transfer to referral centres for advanced supportive care. This inexpensive, bedside assessment requires minimal training and may have significant utility in the resource-limited settings which bear the greatest burden of disease.
| Leptospirosis, a neglected tropical disease with a global distribution, is estimated to kill 60,000 people every year. Predicting those at risk of developing severe disease is challenging, and a simple scoring system to quantify the risk of severe disease has proven elusive. Identifying the high-risk patient is important, as it might expedite the initiation of life-saving supportive care. This review of 402 adult patients with leptospirosis in tropical Australia determined that three clinical variables identified at presentation independently predicted severe disease (a subsequent requirement for Intensive Care Unit admission, intubation, vasopressor support, renal replacement therapy or the development of pulmonary haemorrhage). These three variables (abnormal auscultatory findings on respiratory examination, hypotension and oliguria) were used to generate a simple, three-point clinical score which can be determined rapidly and reliably at the bedside by health care workers with minimal training. This simple score may help the clinical management of patients with leptospirosis, particularly in lower and middle-income countries that bear the greatest burden of disease.
| Leptospirosis is a zoonotic infection with a global distribution [1, 2]. Although most infections are mild and self-limiting, the disease is believed to kill almost 60,000 people every year [1]. Severe disease–manifesting as pulmonary haemorrhage, acute kidney injury (AKI) or multiorgan failure–develops in 5–15% of cases. The case-fatality rate of severe leptospirosis is as low as 6% if there is prompt access to vasopressors, renal replacement therapy (RRT) and mechanical ventilation [3], but it can rise to greater than 50% if the delivery of this supportive care is delayed [4].
However, identifying the patients who are at risk of developing severe disease can be difficult. Different studies have suggested that the presence of a variety of clinical features, laboratory investigations and imaging and electrocardiography findings can help [5–10]. While these approaches may be helpful in well-resourced settings where there is access to advanced laboratory and radiology support, they may have less utility in low and middle-income countries (LMIC), which bear a disproportionate burden of the disease [1].
Leptospirosis is endemic in tropical northern Australia, and the state of Queensland has one of the highest reported incidences in the developed world [11]. Most of the cases in Queensland occur in relatively remote locations where there is limited access to diagnostic support. Accordingly, given the potential for patient deterioration, if there is clinical uncertainty about a patient’s prognosis, they are often transferred–sometimes great distances–to a tertiary centre for continuing care. Not only is this frequently unnecessary, it is inconvenient for patients and their families, and expensive for the health system.
To improve the triage of patients with leptospirosis, and identify patient characteristics that predict severe disease, we reviewed the presentation of adults with confirmed leptospirosis in Far North Queensland and correlated their clinical findings and laboratory and imaging results with their subsequent clinical course. Our aim was to produce a simple score that could be used to quickly identify the patients at greatest risk of deterioration, expediting their referral for intensive care unit (ICU) support. We also hoped that the score could predict which patients could be safely managed without transfer, providing reassurance for local clinicians and reducing costs for the health system. Recognising that leptospirosis has a significant burden in LMIC–and in remote locations in high-income countries–it was also hoped that the score that might be applicable where access to diagnostic support is limited.
This retrospective study was performed at Cairns Hospital, a 531-bed, tertiary referral hospital in tropical, northern Australia that–with 16 smaller community hospitals–provides medical services to a population of approximately 280,000 people across an area of 380,000km2. The local electronic pathology reporting system (AUSLAB) was used to identify all leptospirosis cases in the region between January 1998 and May 2016.
Adult patients (≥16 years of age) were defined as having confirmed leptospirosis if they met one or more of the following criteria: (1) Leptospires isolated from blood culture; (2) Microscopic agglutination test (MAT) single titre of ≥ 1:400; (3) Fourfold rise in MAT antibody titres; (4) Detection of Leptospira in blood by polymerase chain reaction (PCR).
Medical charts were reviewed at the hospital of first presentation and at Cairns Hospital if a patient required inter-hospital transfer. It was recognized that a proportion of the medical records would be unavailable as the health service has a policy of destroying the paper medical record if there have been no new patient encounters for ten years.
Patient characteristics at the time of presentation to medical attention were reviewed. The World Health Organization was undertaking enhanced surveillance of leptospirosis in the region and a case report form was in use for much of the study period. Clinical findings, haematology, biochemistry, urinalysis, chest x-ray and electrocardiogram results were recorded. The following cut-offs–based on the literature, reference ranges and everyday clinical practice–were used to characterise any association severe disease: hypotension (systolic blood pressure ≤100 mmHg), anaemia (haemoglobin ≤100 g/L), severe thrombocytopenia (platelets ≤50 x 109/L), acidosis (bicarbonate ≤22 mmol/L), AKI (creatinine ≥2 mg/dL), jaundice (bilirubin ≥3 mg/dL) and C-reactive protein ≥200 mg/L. The quick Sequential Organ Failure Assessment (qSOFA) and the quick National Early Warning Score (qNEWS) scores were calculated for the patients with sufficient clinical information [12, 13].
Severe disease was defined as the development of pulmonary haemorrhage, ICU admission, or a requirement for RRT, intubation or vasopressor support. Pulmonary haemorrhage was said to be present if there was frank haemoptysis or if blood was present on tracheal aspirate.
The study obtained approval from the Far North Queensland Human Research Ethics Committee (HREC/16/QCH/37 – 1043LR). As per the approval, this retrospective study used anonymized patient data and did not obtain individual patient consent. This study reviewed human patients only; no animals were involved in any aspect of the study.
Data were entered into an electronic database (Microsoft Excel) and analysed using statistical software (Stata 14.0). Groups were analysed using the Kruskal-Wallis and chi-squared tests. Multivariate analysis was performed using backwards linear and logistic regression. For the multivariate analysis, only variables with an area under the receiver operating characteristic (AUROC) curve of >0.7 in univariate analysis were selected.
There were 738 cases of laboratory-confirmed leptospirosis during the study period. Medical charts were available in 429 cases; 402 (94%) were adults. Their median (interquartile range (IQR)) age was 33 (23–45) years; 362 (90%) were male. Nearly all the cases (397/402 (99%)) were acquired locally, 273/397 (69%) occurred during the region’s November-April wet season, and 355/397 (89%) occurred in a region of high-intensity banana and dairy cattle farming situated approximately 100km south of Cairns. In the 384 in whom an occupation was documented, 327 (85%) had the potential for occupational exposure. There were 50 (12%) patients who developed severe disease, including two (0.5%) who died (Fig 1).
In 331/402 (82%) cases, clinicians included leptospirosis in the differential diagnosis at the time of presentation. Leptospires were isolated from blood culture in 275/402 (68%), MAT was diagnostic in 178/402 (44%) and PCR was positive in 151/402 (38%). Serovars could be determined in 353/402 (88%); Australis and Zanoni were the commonest serovars, and the most likely to cause severe disease (Table 1).
The median (IQR) duration of symptoms was 4 (3–6) days in the patients who developed severe disease compared with 3 (2–4) days in those that did not (p = 0.0001). Comorbidities were documented in 395 patients and were more common in the patients who had severe disease (15/49 (31%) than those that did not (29/346 (8%), p<0.0001).
Clinical findings were similar to those in the published literature, although only 9% had a serum bilirubin ≥3mg/dL and conjunctival suffusion was documented in only 109/397 (27%) who were assessed. The symptoms, signs and laboratory tests and their association with severe disease are presented in Tables 2 and 3. In univariate analysis, the symptoms of diarrhoea, dyspnoea and bleeding were most associated with the development of severe disease (Table 2). The presence of renal impairment and thrombocytopenia were the laboratory tests most associated with the development of severe disease (Table 3).
In the 50 patients with severe disease, 45 (90%) required ICU admission, 27 (54%) developed pulmonary haemorrhage, 27 (54%) required vasopressor support, 18 (36%) required RRT and 24 (48%) required mechanical ventilation. APACHE III scores were available for 39/45 (87%) patients admitted to ICU; the median score was 84 (range 27–169).
There were only two deaths in the study. The first was an 80-year-old man with a history of diabetes mellitus and 5 days of symptoms; he developed multiorgan failure and died one day after presentation despite ICU support. The second, a 73-year-old man with a history of cardiovascular disease, chronic lung disease and connective tissue disease, had four days of symptoms; he also had multiorgan failure and died within one day of presentation.
Multivariate analysis identified four independent variables associated with severe disease; three of these variables were clinical signs–abnormal auscultatory findings on respiratory examination (odds ratio (OR) (95% CI): 8.9 (3.5–22.4), p<0.0001), oliguria (OR (95% CI): 8.2 (3.2–21.2) p<0.0001) and hypotension (OR (95% CI): 3.8 (1.5–9.9), p = 0.006) and one was a laboratory variable (creatinine ≥2 mg/dL) (OR (95% CI): 7.0 (2.7–18.1) p<0.0001). The three clinical findings–awarded one point each–were used to generate a three-point SPiRO score (Systolic blood Pressure ≤100 mmHg, Respiratory auscultation abnormalities, Oliguria, Table 4). The risk of severe disease increased incrementally with the SPiRO score. A score of zero had a negative predictive value (NPV) for severe disease of 97.2% (95% CI: 94.6–98.8%). A score greater than one had a positive predictive value (PPV) (95% CI) for severe disease of 77.1% (59.9–89.6), while a score of three had a PPV of 100% (66.4–100) (Table 5 and Fig 2). The predictive ability of the SPiRO score was compared with the qSOFA and the qNEWS scores. In the 379/402 (94%) patients in whom the scores could be calculated, the AUROC of the SPiRO score (0.87 (95% CI 0.81–0.9) was higher than that of the qSOFA (0.76 (95% CI 0.70–0.83) score (p = 0.003). The difference between the AUROC of the SPiRO score and the qNEWS score (0.81 (95% CI 0.74–0.87) failed to reach statistical significance (p = 0.053).
In adults with leptospirosis, a simple three-point clinical score–the SPiRO score–appears to reliably identify patients at risk of severe disease. The score could be used anywhere that leptospirosis is seen, but as a rapid and inexpensive assessment, which can be performed at the bedside by even junior health care workers, it has significant appeal for a disease that has its greatest impact in resource-poor settings. An absence of hypotension, oliguria or abnormal auscultatory findings–a SPiRO score of zero–was particularly helpful in identifying low-risk patients. The score could therefore determine which patients can be safely managed in remote locations, avoiding unnecessary and expensive transfer. The likelihood of severe disease rose incrementally as the score increased, facilitating recognition of the high-risk patient, expediting the initiation of supportive treatment and prompting consideration of transfer to referral centres.
Although many variables have been shown to predict severe leptospirosis, a simple scoring system to quantify the relative risk of severe disease has proven elusive [5]. As in this series, older age has been associated with severe leptospirosis and worse outcomes in multiple countries including India [4], Brazil [14] and Turkey [15]. Other predictors of severe leptospirosis or leptospirosis-attributable mortality have consisted of a combination of clinical features, laboratory findings and interpretation of imaging and electrocardiography. It is notable that pulmonary involvement is associated with worse outcomes in almost every published series, while renal involvement and hypotension also have also been shown to have significant prognostic utility.
In the French West Indies, dyspnoea, oliguria, white blood cell count >12,900/mm3, alveolar infiltrates on chest X-rays and repolarization abnormalities on electrocardiograms were independently associated with death [10]. In Brazil, pulmonary involvement, oliguria, creatinine >3 mg/dL and platelets <70,000/mm3 were independent predictors of mortality, with pulmonary involvement being the strongest prognostic factor [16]. Similarly, in India [17, 18], Indonesia [19], and Greece [20], pulmonary involvement was associated with increased mortality. In Thailand, pulmonary rales, oliguria, hypotension and hyperkalaemia were all independently associated with death [21].
While an elevated serum creatinine, white cell count and thrombocytopenia were also associated with severe disease in our series, it is important to remember that laboratory support may be limited in the rural and remote settings of LMIC where most cases of leptospirosis are seen. Even where there is access to laboratory support in these locations, results are not always available promptly. Similarly, while an abnormal chest X-ray had prognostic utility in our series, there may not be access to radiology support in leptospirosis-endemic areas and even when there is, accurate interpretation of imaging findings requires high quality images and significant medical training. Finally, although electrocardiography is inexpensive and relatively easy to perform, the identification of repolarization abnormalities also requires some expertise.
These issues may also be relevant in high-income settings like Australia. In rural locations where most cases of leptospirosis are diagnosed, it can take up to 24 hours for the processing, transport and analysis of even simple haematology and biochemistry tests such as platelet count or serum creatinine. Other laboratory tests that have been linked to severe leptospirosis, such as the quantification of leptospires in blood, are unlikely to be routinely available in the foreseeable future [6, 22, 23]. Imaging is also not necessarily accessible, and patients are usually reviewed initially by junior staff.
The entirely clinical SPiRO score therefore has significant appeal. It is simple to perform, reproducible, requires little medical training and addresses the renal impairment, pulmonary involvement and hypotension that have repeatedly been shown to be associated with the worst clinical outcomes [10, 16–21]. AKI in leptospirosis occurs due to direct leptospire invasion resulting in tubulointerstitial nephritis [24]. Renal biopsy most frequently reveals a mononuclear cellular infiltration and interstitial oedema, although an immune-complex glomerulonephritis may also be present [25, 26]. While leptospirosis has traditionally been thought to cause non-oliguric AKI [27], oliguria is an early clinical marker of AKI that is less likely to respond to rehydration and more likely to require RRT [28]. Hypotension in leptospirosis is usually due to vasodilatory mediators and proinflammatory cytokines released in response to the infection; this results in reduced renal blood flow, further exacerbating renal injury [29]. Pulmonary involvement–perhaps the most serious manifestation of severe disease–is frequently overlooked [23, 30]. Leptospirosis impairs the fluid handling of alveolar epithelial cells resulting in pulmonary oedema which can trigger acute respiratory distress syndrome [31–33]. Pulmonary haemorrhage–the most feared respiratory manifestation–is thought to occur from a direct effect of leptospiral proteins or toxic cellular components on multiple components of the alveolocapillary membrane [34]. As larger areas of haemorrhage coalesce, symptoms worsen and clinical signs are more likely to be apparent on auscultation [35]. As pulmonary haemorrhage progresses, pulmonary vascular resistance also increases, further contributing to systemic hypotension [29].
Severe disease was common in our series, but the case-fatality rate was very low with both deaths occurring in elderly patients with significant comorbidities. The wide variation in case-fatality rates reported in the literature has been attributed to differing definitions of severe leptospirosis, although our definition was conservative and the patients’ APACHE III scores were high. It is possible that the local case-fatality rate is higher than we have reported as patients with rapidly fatal leptospirosis may have a negative MAT test early in their disease. However, culture and PCR are used widely locally with only 18% of cases were diagnosed by MAT alone and accordingly, the number of unrecognized, fatal cases is probably small. The excellent outcomes are likely to be the result of early disease recognition and access to prompt ICU support. There is an extensive medical retrieval network in Australia which ensures people living in rural and remote services have access to sophisticated healthcare. However, given the country’s great expanse, these services are costly, frequently relying on a combination of road ambulances, helicopters and aeroplanes. While the coordination of retrieval services is centralised and efficient, organising safe and appropriate medical evacuation is time-consuming. A simple scoring system that facilitates recognition of patients with severe leptospirosis could help guide clinicians to identify which patients are most likely to require critical care support and early referral to retrieval services. Conversely, the SPiRO score could help prevent unnecessary medical evacuation, which would be welcome for patients and reassuring for the clinicians involved in their care.
Our study was retrospective and the SPiRO score requires prospective validation to ensure its applicability in other geographical settings. However, in other locations including Brazil [14, 36], Thailand [21], Moldova [37], Greece [20], and Réunion [38], hypotension, oliguria or abnormal respiratory auscultation have been identified previously as independent predictors of severe disease. Indeed, in a much smaller series from French Polynesia the same three clinical parameters were found to be the only independent variables in predicting severe disease [39]. The SPiRO score therefore has potential global utility. While many counties with a high incidence of leptospirosis do not have access to medical retrieval services or advanced ICU support, the score may still identify those who may benefit from closer monitoring and might be expected to improve outcomes.
The clinical findings of leptospirosis have been linked to the infecting serovar, with serogroup Icterohaemorrhagiae particularly associated with severe disease [6, 35]. This serogroup was uncommon in our series, occurring in only two cases, while severe disease was most commonly associated with serovars Australis and Zanoni. Variations in the prevalence of different serogroups and serovars have the potential to limit the generalizability of our findings, however, as previously noted, the clinical phenotype seen in our cohort was remarkably similar to that seen in the published literature [3, 10, 35].
Evidently, the SPiRO score can only be applied to patients with a diagnosis of leptospirosis, a condition whose prompt diagnosis remains challenging. While clinical findings can inform the clinician, they are non-specific and may not differentiate leptospirosis from other tropical infections including rickettsial disease, malaria and dengue. PCR is rarely available where the disease is endemic, and even in well-resourced settings like Australia, results take several days. Point-of-care tests have the greatest potential to facilitate diagnosis in both resource-poor and rich countries, but although their sensitivity and specificity is improving, these tests are not currently in routine use [40, 41]. If reliable point-of care tests can be developed and coupled with a validated simple predictive tool, the early recognition and management of leptospirosis is likely to improve significantly. That being said, clinicians in leptospirosis-endemic areas often recognize the disease–in our series, leptospirosis was in the initial differential diagnosis in over 82% of cases. Furthermore, even when the diagnosis of leptospirosis cannot be confirmed, a patient presenting to a remote clinic with hypotension and evidence of pulmonary and renal disease is likely to require referral for more sophisticated care whatever the aetiology.
The retrospective nature of our study meant that documentation was sometimes incomplete, and investigations were not standardised. However, clinicians working in the area have a high index of suspicion for the disease and a leptospirosis pro forma was in use for most of the study period. As a result, the clinical features on presentation were generally well documented. Data are now being collected prospectively to confirm these preliminary observations. The antibiotic therapy, its timing, route and duration were available in most cases and almost all patients received at least one appropriate agent. However, the enormous variety of antibiotic regimens prescribed precluded meaningful analysis of their relative efficacies.
In conclusion, a simple three-point clinical based scoring tool appears to help clinicians identify people at risk of developing severe leptospirosis. The score requires prospective validation in other geographical locations, but it has the potential to improve the care of people with leptospirosis, particularly in resource-limited settings where the disease has its greatest clinical burden.
|
10.1371/journal.pntd.0001362 | The Cost of Antibiotic Mass Drug Administration for Trachoma Control in a Remote Area of South Sudan | Mass drug administration (MDA) of antibiotics is a key component of the so-called “SAFE” strategy for trachoma control, while MDA of anthelminthics provides the cornerstone for control of a number of other neglected tropical diseases (NTDs). Simultaneous delivery of two or more of these drugs, renowned as “integrated NTD control,” is being promoted to reduce costs and expand intervention coverage. A cost analysis was conducted alongside an MDA campaign in a remote trachoma endemic area, to inform budgeting for NTD control in South Sudan.
A first round of antibiotic MDA was conducted in the highly trachoma endemic county of Mayom, Unity state, from June to August 2010. A core team of seven staff delivered the intervention, including recruitment and training of 44 supervisors and 542 community drug distributors. Using an ingredients approach, financial and economic costs were captured from the provider perspective in a detailed costing database. Overall, 123,760 individuals were treated for trachoma, resulting in an estimated treatment coverage of 94%. The economic cost per person treated was USD 1.53, excluding the cost of the antibiotic azithromycin. Ninety four per cent of the delivery costs were recurrent costs, with personnel and travel/transport costs taking up the largest share.
In a remote setting and for the initial round, MDA of antibiotics was considerably more expensive than USD 0.5 per person treated, an estimate frequently quoted to advocate for integrated NTD control. Drug delivery costs in South Sudan are unlikely to decrease substantially during subsequent MDA rounds, as the major cost drivers were recurrent costs. MDA campaigns for delivery of one or more drugs in South Sudan should thus be budgeted at around USD 1.5 per person treated, at least until further costing data for delivery of other NTD drugs, singly or in combination, are available.
| Trachoma is one of a group of so-called “neglected tropical diseases” (NTDs) for which safe and effective treatments are available. The International Trachoma Initiative oversees donation of the antibiotic azithromycin to endemic countries. Delivery of this drug to communities affected by trachoma is the responsibility of national programmes and their implementing partners, and should be conducted as part of a comprehensive control strategy termed “SAFE,” which includes trichiasis surgery, health education and water/sanitation interventions. There are little data on how much the different components of a trachoma control programme cost and none from South Sudan. To inform budgeting to scale up control of trachoma, and of other NTDs whose control relies on large-scale mass drug administration (MDA), the present study set out to determine the cost per person treated when antibiotics were delivered through a vertical campaign that covered 94% of the target population in a remote trachoma endemic area of South Sudan. The average economic cost per person treated was USD 1.53, which included all inputs not paid for in cash except for the cost of the donated azithromycin and the opportunity cost of community members attending treatment.
| Since 2007, financial support for the control or elimination of some key neglected tropical diseases (NTDs) has increase substantially [1]–[4]. The NTDs largely benefitting from this attention are onchocerciasis, lymphatic filariasis (LF), soil-transmitted helminth infections, schistosomiasis and trachoma. The rationale for focusing on this group of diseases is based on the fact that safe and effective preventive chemotherapy (PCT) options are available for all of them, either free of charge or at low cost, and on the assumption that previously separate distributions of these drugs by stand-alone vertical programmes could be easily combined under one structure or become part of other large-scale distributions of public health commodities such as insecticide-treated mosquito nets, hence substantially reducing delivery costs [5]. A figure of around USD 0.5 per person treated per year for co-administration of PCT has been extensively used for advocacy [6]–[8], while there is limited empirical evidence supporting this estimate [9]–[11]. For programmatic purposes, particularly to budget for scale up of PCT delivery to previously untreated areas, there is thus a need to collect accurate cost data. Furthermore there is a need to develop systems for routine collection of NTD programme cost data to determine whether economies of scale and/or scope exist and to estimate the cost-effectiveness and cost-benefit of stand-alone and ‘integrated’ NTD control programmes [12].
Investigations into the cost and cost-effectiveness of PCT delivery for the above diseases have been undertaken to varying degrees, depending on the availability of data and the principal interest of the researchers undertaking the analyses. Delivery costs for large-scale deworming through schools in Kenya, for example, have been estimated at USD 0.23 per child treated with albendazole or USD 0.95 for combined albendazole and praziquantel treatment (both estimates including the cost of the drug) [13]. Estimates of delivery costs for mass drug administration (MDA) of anthelminthics from other countries have ranged from USD 0.03 to USD 0.54 [10], [13], [14], [15]. Similar cost estimates have been arrived at for MDA campaigns aimed at LF elimination, although the upper bound was significantly higher at USD 2.23 for the per capita cost of Haiti's first MDA round [12]. As frequently encountered with economic analyses, these results are not necessarily readily comparable, as investigators choose to use different costing perspectives and approaches [16]. The results generated may thus not be suitable for budgeting purposes.
For trachoma, a number of cost and cost-effectiveness studies have been conducted [17]–[22] and detailed guidance on how to conduct a cost-effectiveness analysis for this disease has been provided [23]. The present study aimed to build on this evidence by generating critical information on the per-capita cost of implementing a MDA campaign for trachoma. The intervention was supported by the United States Agency for International Development Neglected Tropical Disease (NTD) Control Program and delivered in a remote trachoma endemic area of South Sudan that had not benefited from intervention. Cost data specific to this setting was required to predict the likely cost of repeating this exercise in the same area, scaling up antibiotic distribution to other trachoma endemic areas of the country and to provide a general estimate of how much it might cost to deliver one or more drugs for NTD control through campaigns in this post-conflict setting.
The MDA campaign was conducted on behalf of the Ministry of Health, Government of South Sudan (MoH-GoSS), and in close collaboration with state and county MoH representatives. Implementation followed treatment guidelines provided by ITI and WHO [24], [25]. The costing study only used data on expenditures and non-financial inputs incurred by Malaria Consortium in the implementation of the campaign. Collection of these data did not involve human subjects and therefore did not require ethical approval.
The overall purpose of the MDA campaign was to deliver antibiotics to the highest possible number of individuals eligible for treatment in an area of South Sudan shown to have prevalences of 57.5% and 39.8%, respectively, of trachomatous inflammation-follicular and trachomatous inflammation-intense in children aged 1–9 years. In the same area, the prevalence of trachomatous trichiasis in adults was found to be 19.2% [26].
The International Trachoma Initiative (ITI) recommends at least 90% of the population should be treated with antibiotics to lower the risk of recurrent infection [24]. Children below six months of age should be treated with tetracycline eye ointment for six weeks (by their parents or care taker), while older children and adults should receive a single dose of azithromycin. To minimize the risk of choking on tablets, children age six months to five years are generally provides with a paediatric oral suspension (POS) formulation of azithromycin, while older children and adults receive the drug in tablet form. For both formulations the dose was determined according to height by means of a gradated dose pole, as recommended by ITI and WHO [24], [25].
Between June and August 2010, 417 villages in all ten payams of Mayom County, Unity State, received antibiotics (azithromycin and tetracycline) through a MDA campaign. South Sudan has a four-tier administrative structure comprised of states (1st), counties (2nd), payams (3rd) and bomas (4th). The rainy season in this part of the country starts in about May, which unfortunately meant that the majority of fieldwork had to be conducted during the worst season for implementation. Ideally, implementation would have been conducted during the dry season, but in the present case this was not feasible due to various logistical and financial delays.
A core team led by a MDA coordinator and consisting of six training and treatment supervisors (TATS) conducted the campaign. A Malaria Consortium field officer and a logistician also supported the team at various stages of the MDA, amounting to a total of 30 and 22 days, respectively. In addition, 542 community drug distributors (CDDs) and 44 supervisors were recruited and trained in the process; both of these staff categories were paid an incentive, while the core team and their support staff were on full salaries.
The number of CDDs required to treat each boma was calculated using population data from the 2008 national census. Approximately one CDD was recruited for every 40 households or 240 people (as the average household size was estimated to be six people). Two CDDs were paired up for the duration of the campaign to ensure that at least one from each pair lived in the area to be treated and that at least one was literate. Each CDD had to attend a three-day training session to acquire a basic background on trachoma and its treatment, and be informed of his/her role in the MDA. The CDD's role was to conduct a census of their allocated households, pick up drugs from an agreed location, return to and correctly treat their allocated households, and return equipment, drugs and the census and treatment registers to a Malaria Consortium staff member. Each CDD was provided with a set of equipment at the end of the training: T-shirt, ID card, backpack, pens/pencils/erasers, notebook, clipboard, plastic folder, census and treatment register, and spray paint/stickers to mark houses that had been treated. Each pair of CDDs was also given a trachoma flip chart to facilitate health education sessions, a wooden dose pole to determine the amount of azithromycin to be given to each individual, a cup, and water purification tablets to treat all drinking water before it was given out to facilitate swallowing of azithromycin tablets. Supervisors were provided with the same equipment as CDDs, as well as with a pair of gumboots and a raincoat. They were also loaned a bicycle for the duration of the MDA in their payam (usually around ten days).
Two Land Cruiser HZJ Troop Carriers (Toyota Motor Corporation, Toyota City, Aichi, Japan) owned by Malaria Consortium and two drivers were allocated to the MDA campaign for its entire duration. Each of the vehicles was equipped with a high frequency radio (Codan Ltd., Adelaide, Australia), a handheld satellite phone (Thuraya, Abu Dhabi, United Arab Emirates) and an in-car charger, to allow communication between teams and with the field office at all times. A third vehicle and driver were hired for one month during the campaign, but it lacked above equipment and was not suitable for off-road driving.
The MDA campaign consisted of three distinct implementation phases: a start-up phase amounting to a total of 57 days during which health education and training materials were developed, equipment was purchased and transported to the state capital Bentiu, TATS were recruited and trained, and local authorities were made aware of the upcoming MDA and its purpose. After the start-up phase was complete, MDA delivery was conducted payam-by-payam using approximately the same procedure and timeframe in each and amounting to a total of 88 days, 27 of which were spent on training CDDs. The procedure applied consisted of CDD and supervisor training (2–3 days), a census in each village (2 days), drug pick-up (1 day), MDA delivery (2 days), and drop-off of drugs and treatment registers (1 day). The MDA was conducted using central point or house-to-house delivery, as agreed upon by the CDDs and the community during the census. Once all payams had been treated, a closure phase of 24 days was required during which equipment was transported and stored, inventories were updated, data were entered and analysed, and a report was prepared.
A detailed costing database was developed using an ingredients approach to capture financial and economic costs as incurred by Malaria Consortium. The costing template and methodology described in an earlier costing study on NTD mapping in South Sudan were used for this purpose [27]. Some costs that could not be captured during the fieldwork were subsequently estimated from financial expenditure records. Average exchange rates for the period 1 May to 31 August 2010 were used for currency conversions, as provided by OANDA (http://www.oanda.com/currency/historical-rates). The rates were: 1 United States Dollars (USD) = 2.26 Sudanese Pound (SDG) or 1 USD = 2,197 Ugandan Shillings (UGX).
Both financial and economic costs were estimated from the perspective of the provider [28], in this case Malaria Consortium. The only input not captured using this approach was the time of community members required to attend sensitization and treatment sessions. In addition, we chose not to include the cost of azithromycin, as ITI will donate and deliver this drug to trachoma endemic countries for the foreseeable future. Furthermore, inclusion of the economic cost of this drug, estimated at USD 278 per bottle of 30 tablets and USD 39 per 1,200 mg bottle of POS (ITI, pers. com.) would have skewed the cost-estimate towards the drug, while we were ultimately interested in estimating a comprehensive delivery cost as incurred by the provider. The cost of tetracycline, however, was included, as this drug has to be procured by the provider to ensure that all eligible age groups can be provided with trachoma treatment.
Financial costs captured were the cash expenditures made to enable implementation of the campaign. For capital items, these were estimated for the total number of days they were required to organize and implement the MDA campaign using straight-line depreciation, followed by calculation of an average financial daily cost. Economic costs captured the value of all resources required for the campaign, including opportunity costs of equipment that was used but not paid for and a proportion of the costs of capital items with a value over USD 100 and an expected useful life of more than one year that were used in the campaign [29]. The time of CDDs was not included as an economic cost, because they were paid a per-diem as compensation for their time. Capital items were discounted over their estimated useful life using the recommended discount rate of 3% [30], [31]. Daily economic costs were calculated for all capital items and multiplied by the appropriate number of days in use during the MDA campaign. Based on our experience of working in the harsh environment of South Sudan, we estimated the useful life of vehicles and high frequency radios (fitted to vehicles) to be four years and two years for all other items.
All resources used to conduct the costing study were excluded from the analysis. Training costs, however, which could potentially be incurred only once if knowledge and staff were retained over subsequent MDA rounds, were included as an integral part of the campaign costing. This was felt to be necessary, because the study aimed to estimate the cost of scaling up MDA to the substantial trachoma endemic areas of South Sudan that have not been targeted with interventions to date [32], hence incurring these training costs, but also because high attrition of community volunteers [11] is likely to results in the same or similar training needs for subsequent MDA rounds in the same geographic area.
To be consistent with our previous costing study in South Sudan [27] we applied an overhead of 25% to the financial cost estimates for all budget lines. In our experience this estimate provides an accurate reflection of overheads associated with implementation in this post-conflict setting.
The outcome that the present analysis aimed to estimate was the cost per person treated through an MDA campaign, regardless of the type of antibiotic treatment or its formulation. Given that high coverage with azithromycin and tetracycline is required for effective control, we decided that there was no reason to investigate potential differences in delivery costs between antibiotic types or formulations.
One-way sensitivity analyses were conducted to explore the effects of key assumptions on the cost estimate. The effects of varying the following parameters were explored: i) the discount rate was reduced to 0% or increased to 10%; ii) the assumed lifespan of vehicles was increased to 7.5 years, and iii) the exchange rate was modified from the average of SDG 2.26 per USD provided by OANDA to an average of SDG 2.45 per USD, as used by the KCB bank in Juba over the implementation period.
The total financial cost of the campaign amounted to USD 169,084 including a 25% overhead (Table 1). The major cost drivers were recurrent rather than capital costs, led by personnel (41.3%) and followed by travel/transport (29.1%) and accommodation/sustenance (10.0%). The largest proportion of financial costs was incurred during the actual MDA delivery period (86.6%), while the start up and closure phases only required moderate investment (Table 2). The total economic cost was USD 189,889, including the same overhead as that included in the financial cost (Table 1). Economic costs were slightly higher than financial costs, because this estimate captured non-financial contributions to the implementation of the campaign. The difference between these two estimates was greatest for the categories of accommodation/sustenance and MDA/IEC consumables and other charges (Table 1), largely because the team was able to stay free of charge with other non-governmental organizations during part of the campaign and because the state/county administrations provided the training facilities. In other settings, even within Unity state, these contributions would have to be paid for, which is why they were included as part of the economic costs.
The MDA campaign delivered antibiotic treatments to a total of 123,760 individuals in Mayom county, which translated into 94% coverage of the estimated population. For the first time ever this area benefitted from at least one of the component of the SAFE strategy for trachoma control. The average economic cost per person treated was USD 1.53, excluding the cost of azithromycin. If the economic cost of the donated antibiotic had been included, the estimated economic cost per person treated would have been USD 34.2 (data not shown).
Varying the underlying assumptions had limited effect on the estimated economic cost per person treated (Table 3). Most pronounced was the use of the exchange rate applied by the main bank in Juba, as opposed to the figure provided by OANDA, reducing the cost per person treated to USD 1.48.
The aim of the present analysis was to estimate the cost of delivering one, or possibly more, drug(s) through MDA campaigns in South Sudan, thus informing budgeting for control of trachoma and other key NTDs in the country. The costing study established that a first round of MDA in the remote area of Mayom county incurred an economic cost of USD 1.53 per person treated. This estimate was considerably higher than that of USD 0.5, frequently quoted as the approximate annual per capita cost for simultaneous delivery of multiple drugs to control some key NTDs [6]–[8]. This substantial difference between the two cost estimates can largely be explained by the fact that the latter figure is not based on in-depth costing work for a specific setting.
In the present case, the setting was one of the hardest to reach, and hence most costly implementation areas in Sub-Saharan Africa. Moreover, logistical and financial delays meant that implementation had to be conducted during the rainy season, slowing down fieldwork. In addition, this was the first MDA round for trachoma control in Mayom county, necessitating a start-up period to develop the training and health education materials, develop a baseline list of villages, which did not exist previously, and develop the most practical MDA implementation methodology. The cost estimate may therefore be considered a ‘worst-case scenario’, rather than a figure that is generally applicable for large-scale PCT (co-)administration in other parts of Sub-Saharan Africa, although this remains to be confirmed. Our estimate would certainly be applicable to the delivery of first rounds of antibiotics or other PCT for NTD control in most areas of South Sudan, as the remoteness and operational constraints of Mayom are the rule rather than an exception in this country.
Unfortunately it is hard to judge how our estimate compares to other settings, as we were unable to identify any other estimates of the delivery cost for trachoma MDA that were directly comparable to the approach used here. The seemingly most comparable result was obtained as part of a comparison of different treatment strategies in Mali [21]. Costs were, however, not directly estimated during the study but extrapolated from data provided by the Malian National Programme for Prevention of Blindness for antibiotic distribution at district level. Assuming the drug was free of charge and including an estimate of the societal cost associated with attending the distribution, the study arrived at an estimated cost per person treated of USD 0.25 for mass treatment of all residence [21]. It is unclear whether this estimate was derived using the same ingredients as in the present study. One ingredient that was clearly not included in the analysis from South Sudan was the opportunity cost of the people attending treatment. Two methods – central point or house-to-house distribution – were used in Mayom depending on the communities' decision, but no data were collected on which approach was used where. It was therefore not feasible to include the correct economic cost associated with community members attending distribution, or to establish whether there may have been a difference in cost between the two distribution methods.
Drug delivery costs in South Sudan would likely be slightly lower in areas nearer the capital Juba, as this would reduce the high transport costs. In the present case, supplies and equipment had to be brought from Juba to Unity, a journey taking at least three days by truck. Implementation in areas near Juba may also be conducted over a shorter timeframe, because accessibility of communities is generally better and a higher proportion of the community is literate. Furthermore, the population density near the capital is higher, meaning that it may be feasible to treat more people with similar inputs, for example by utilizing more central point than house-to-house distribution. This could increase the denominator and hence decrease the cost per person treated. Costs may also decrease when the materials and experience developed during the initial campaign are applied in subsequent rounds. Further cost analysis by implementing partners is clearly required to answer these questions and assist in future budgeting.
Given our experience of operating in South Sudan, we nevertheless feel that implementation in areas other than Mayom county and the delivery of subsequent, rather than initial, MDA rounds are unlikely to lead to significant cost savings. Most trachoma endemic areas of South Sudan are remote, with households dispersedly located, and hence comparable to Mayom. Staff turnover including that of CDDs is high [11], resulting in loss of institutional memory and requiring a large amount of costly re-training from year to year. In the present MDA round there was also no obvious opportunity to increase the scale of the intervention with existing resources, as the core MDA team had no spare capacity to allow coverage of additional areas. Scaling up geographical coverage while maintaining quality, safety and at least 90% population coverage would thus be associated with hiring more personnel and using more transport resources, both of which drove the overall implementation cost in the Mayom campaign and would lead to a similar delivery cost per person treated if the same approach was scaled up. For subsequent MDA campaigns in Mayom, as well as for the scale up of trachoma control to all other counties of Unity state, it therefore seems prudent to budget an average cost of USD 1.5 per person treated, unless obvious cost savings are identified during repeated rounds. Accordingly, the estimated cost of treating the population in all nine trachoma endemic counties of Unity State would exceed USD 1 million annually and require a yearly donation of azithromycin valued at USD 16 million.
Further costing work on drug delivery through community-based distribution mechanisms in South Sudan and elsewhere would be useful, particularly with regards to estimating the actual cost and cost-effectiveness of co-administering PCT in areas endemic for more than one NTD [33]. In the present case, the intervention area is also endemic for schistosomiasis (Malaria Consortium, unpublished) and is being targeted by a programme for integrated community-case management (iCCM) to treat malaria, pneumonia and diarrhoea [34], [35]. While drug safety recommendations would currently not allow the co-administration of antibiotics with praziquantel for schistosomiasis control [25], there may be ways in which iCCM and its support/supervision system could be harnessed to contribute to trachoma and/or schistosomiasis control. There was some collaboration between the trachoma and iCCM intervention during this first MDA round, but both interventions had only just started and were largely concerned with achieving their own objectives rather than exploring opportunities for integration. Now that iCCM has been fully established it would be beneficial to estimate what additional resources are required to deliver MDAs for trachoma and schistosomiasis control as part of this structure rather than through a vertical campaign, and whether this is likely to be as effective but cheaper. Similarly, further experience and cost data is urgently required to determine how South Sudan could establish an innovative platform for community-based delivery of a broad range of public health interventions; one of the opportunities offered by this post-conflict setting that has not been taken up [11].
In a remote setting and for the initial round, the delivery of antibiotics for trachoma control through an MDA campaign was three times as expensive as the figure commonly used in international advocacy for simultaneous delivery of a package of drugs to prevent or treat multiple NTD. Costs for delivering one or more drug(s) through MDA campaigns in South Sudan are unlikely to decrease substantially during subsequent rounds, as the major cost drivers were recurrent costs and unlikely to decline dramatically. Until further cost data are generated, MDA campaigns in South Sudan should thus be budgeted at a delivery cost of about USD 1.5 per person targeted.
|
10.1371/journal.pgen.1001351 | Cancer-Associated Splicing Variant of Tumor Suppressor AIMP2/p38: Pathological Implication in Tumorigenesis | Although ARS-interacting multifunctional protein 2 (AIMP2, also named as MSC p38) was first found as a component for a macromolecular tRNA synthetase complex, it was recently discovered to dissociate from the complex and work as a potent tumor suppressor. Upon DNA damage, AIMP2 promotes apoptosis through the protective interaction with p53. However, it was not demonstrated whether AIMP2 was indeed pathologically linked to human cancer. In this work, we found that a splicing variant of AIMP2 lacking exon 2 (AIMP2-DX2) is highly expressed by alternative splicing in human lung cancer cells and patient's tissues. AIMP2-DX2 compromised pro-apoptotic activity of normal AIMP2 through the competitive binding to p53. The cells with higher level of AIMP2-DX2 showed higher propensity to form anchorage-independent colonies and increased resistance to cell death. Mice constitutively expressing this variant showed increased susceptibility to carcinogen-induced lung tumorigenesis. The expression ratio of AIMP2-DX2 to normal AIMP2 was increased according to lung cancer stage and showed a positive correlation with the survival of patients. Thus, this work identified an oncogenic splicing variant of a tumor suppressor, AIMP2/p38, and suggests its potential for anti-cancer target.
| Lung cancer is one of the most common cancers and a leading cause of death resulting from cancer. Despite intensive investigation, effective therapeutic targets and reliable biomarkers are still limited. Here we found that a tumor suppressor, AIMP2 (MSC p38), produces a variant lacking a part of its structure in cancer tissues. We designated it AIMP2-DX2. This smaller version of AIMP2 compromises the normal tumor suppressive activity of AIMP2 and induces tumor formation. We also found that the expression of AIMP2-DX2 was increased according to cancer progression. In addition, the patients with higher expression of AIMP2-DX2 showed lower survival than those with lower levels of this variant. Suppression of AIMP2-DX2 slowed tumor growth, suggesting it as a new therapeutic target. In summary, this work newly identified a tumor-inducing factor, AIMP2-DX2, that can be used as a therapeutic target and biomarker associated with lung cancer.
| Alternative splicing is implicated in the regulation of gene function and diversification [1]–[3]. Although this process can provide another level of flexibility in gene regulation, the disruption in the balance between splicing variants or the generation of aberrant alternative splicing variants can lead to pathological disorder. In this context, the discovery of aberrant splicing variants that are related to human diseases and the understanding of their mode of action would provide important insights into diagnosis and therapy of the related diseases. In this work, we identified a splicing variant of tumor suppressor, AIMP2, that is associated with cancer formation and characterized its working mechanism and pathological implication.
Nine different aminoacyl-tRNA synthetases (ARSs) form a macromolecular complex with three auxiliary factors, AIMP1, 2, and 3. Many of the enzyme components were previously shown to be involved in diverse signaling pathways with their unique mechanisms [4], [5]. The three AIMPs appear to facilitate the assembly of the whole complex through the interactions with each other as well as with their specific target enzymes [6]. These factors also play diverse regulatory roles. AIMP1/p43 is secreted as a cytokine controlling angiogenesis [7], immune response [8], [9], tissue regeneration [10] and as a hormone for glucose homeostasis [11]. It is also implicated in the regulation of the autoimmune phenotype such as lupus [12]. AIMP3/p18 was demonstrated to be a tumor suppressor responding to DNA damage [13], [14] or oncogenic stimuli [15].
AIMP2 plays a pivotal role in the control of cell fate. It shows anti-proliferative activity by enhancing the growth-arresting signal of TGF-β signal [16]. AIMP2 also promotes cell death via the activation of p53 [17] and apoptotic signal of TNF-α [18]. For this reason, mice lacking AIMP2 were neo-natal lethal due to lung failure resulted from overproliferation of lung epithelial cells. In addition, the AIMP2 heterozygous mice with reduced expression level of AIMP2 showed a higher susceptibility to tumorigenesis [19]. Based on these results, AIMP2 is considered as a haploinsufficient tumor suppressor with unique working mechanism. Here we identified a splicing variant of AIMP2 that can compromise the normal function of AIMP2 and induce tumorigenesis.
The gene encoding AIMP2 is located in chromosome 7 and consists of four exons (Figure 1A). Interestingly, the expressions of its splicing variant lacking exon 2 (encoding 69aa) was reported in the EST database for cervical carcinoma (BI259092) and muscle rhabdomyosarcoma (BI115365). We designated this variant AIMP2-DX2. To see whether the generation of this variant has any association with cancer formation, we compared the expression of AIMP2-full length and –DX2 with normal and lung cancer cells by RT-PCR. With the primers, JTV-13 and 11 (Figure 1A top and Figure S1), two PCR products were apparently generated in lung cancer cell lines (Figure 1B top). Isolation and sequencing of the two products determined that the upper and lower bands resulted from the full-length (AIMP2-F) and exon 2-deleted (AIMP2-DX2) transcripts, respectively. RT-PCR with the primer, DX2-S2, which is specific to the junction of exon 1 and 3, and JTV5 (Figure 1A and Figure S1) generated the PCR product of the expected size (27.8kD/756 bp) at higher levels in lung cancer cell lines compared with normal cells (Figure 1B middle).
To see whether the increased expression of AIMP2-DX2 is also observed in human cancer tissues, we isolated the cancer regions from different types of human lung cancer patients and compared the AIMP2-F and –DX2 by RT-PCR. In many of the clinical tissues, the AIMP2-DX2 expression was increased in cancerous regions compared to the normal regions (Figure 1C). To further evaluate the higher expression of AIMP2-DX2, we compared the AIMP2-F and –DX2 expressions in adenocarcinoma of human lung cancer using the poly-A polymerase alpha (POPOLA) gene as a reference [20] using quantitative RT-PCR. The expression ratio of AIMP2-DX2 to PAPOLA, was increased to 0.32 in the cancer region compared to 0.14 in the normal region. In contrast, the ratio of AIMP2-F expression to PAPOLA was only slightly increased in cancer region (Figure 1D).
To confirm the identity of the two forms of AIMP2, we introduced each of pcDNA3.1 plasmid encoding mouse AIMP2-F or –DX2 into AIMP2-deficient mouse embryonic fibroblasts (MEFs) and each transfectant was subjected to Western blotting with AIMP2 antibody, which can recognize both AIMP2-F and –DX2. When proteins were extracted from the transfectants and subjected to immunoblotting with the anti-AIMP2 antibody, AIMP2-F and –DX2-transfected cells showed specific bands at the location of expected molecular weight while no signal was observed from the EV-transfected cells (Figure 1E left). We also introduced siRNAs designed to specifically target either AIMP2-F or –DX2 into A549 cells in order to validate the identity of the two forms. Each siRNA specifically suppressed the expression of AIMP2-F and –DX2 as expected (Figure 1E right), further confirming the identity of the two transcripts.
To see whether the expression of AIMP2-DX2 can be induced by carcinogenic stress, we treated the WI-26 normal lung cells with chemical carcinogen, anti-benzo[a]pyrene-7,8-dihydrodiol-9,10-epoxide (BPDE), and selected the surviving cells. We then compared the expression levels of AIMP2-DX2 between the normal WI-26 and surviving cells by Western blotting. The surviving cells expressed higher levels of AIMP2-DX2 compared to that of the normal cells (Figure 2A), suggesting that AIMP2-DX2 can be induced by carcinogenic stress. To determine whether the increased expression of AIMP2-DX2 involves any mutations that might affect alternative splicing pattern between AIMP2-F and –DX2, we isolated the genomic DNA encoding AIMP2 from the surviving cells. We determined the DNA sequences and identified a C39 deletion mutation and a base substitution of A152 to G in exon 2. Furthermore there were substitutions of C227 to T and A342 to G in intron 3 (Figure S2).
We then employed the plasmid (pGINT) containing a splicing cassette that can monitor the alternative splicing by the expression of GFP in 293 cells [21] (Figure 2B). The 1263 base genomic fragment spanning intron 1 (630 bp from 3′end), exon 2 (207 bp) and intron 2 (426 bp from 5′end) of AIMP2 was inserted into the gene encoding GFP (pGINT-exon 2) (Figure 2C). If the exon 2 of AIMP2 is included in splicing, it will be inserted into the middle of GFP, thereby ablating green fluorescence. If exon 2 is skipped during splicing, the normal GFP will be expressed and the cells will generate green fluorescence. To see whether this system would work as expected, pGINT or pGINT-exon 2 was transfected into 293 cells with pRINT that expresses RFP in the same splicing cassette as a control. In both transfectants, RFP was expressed to similar degree, indicating that splicing process occurred normally. In contrast, the transfectants of pGINT-exon 2 showed significantly reduced green fluorescence compared with the pGINT transfectants (Figure 2D). This indicates that splicing process including exon 2 of AIMP2 mainly takes place.
We then introduced pGINT containing the exon 2 insert with different mutations and determined whether any of these mutations induced exon 2 deletion during splicing. Among the transfectants of four tested mutations, the cells containing A152G substitution expressed GFP to the level similar to pGINT control. In control, the cells with three other mutations showed the reduced GFP levels as pGINT-exon 2 wild type (Figure 2E). The effect of the four mutations on exon 2 splicing was also determined by RT-PCR of pGINT-exon 2 constructs. Among the four tested mutant sequences, the DNA containing A152G substitution induced exon 2 skipping and mainly generated GFP transcript whereas the three other mutants generated the transcript like the wild type DNA (Figure 2F).
To identify splicing factors involved in this process, exonic splicing enhancer (ESE) finder program ver. 3.0 [22], [23] was used for the serine/arginine-rich (SR) protein-binding motif analysis in exon 2. Motif sequences for the SR proteins, SF2/ASF, SC35, SRp40 and SRp55, were used for the ESE finding and the mutation sites located in exon 2, ΔC39 and A152G, were predicted to be located in the ESEs for SF2/ASF (Figure S3A). Since A152G mutation which induced exon 2 skipping was predicted to severely affect the potential ESE site for SF2/ASF, we tested the effect of SF2/ASF on the expression level of AIMP2-DX2. Knockdown of SF2/ASF reduced both AIMP2-F and -DX2 (Figure S3B) and overexpression of SF2/ASF increased the both forms (Figure S3C), implying that AIMP2 pre-mRNA is a functional substrate for SF2/ASF-mediated splicing.
To demonstrate the direct interaction between SF2/ASF and exon 2 RNA, exon 2 RNA was detected after co-precipitation with SF2/ASF. The association of SF2/ASF was observed with WT exon 2, but not with A152G exon 2 RNA (Figure S3D). Direct interaction between SF2/ASF and exon 2 RNA was also tested by incubating SF2/ASF with exon 2 WT or A15G probe. Among the two probes, only the WT exon 2 probe strongly bound to SF2/ASF (Figure S3E). SF2/ASF controls alternative splicing of tumor suppressors and cell cycle regulatory genes and also shows abnormal expression in various cancers including lung carcinoma [24]-[26]. Although SF2/ASF appears to control the expression of both AIMP2-F and AIMP2-DX2, the expression of the DX2 form seems to be more dependent on the levels of SF2/ASF. Thus, enhanced expression of SF2/ASF would increase the relative ratio of AIMP2-DX2 to AIMP2-F.
We investigated how AIMP2-F (full length) and -DX2 would affect DNA damage-induced cell death by monitoring the sub-G1 portion of A549 cells. The sub-G1 portion was increased by adriamycin treatment and it was further augmented by the introduction of AIMP2-F (Figure 3A). However, the exogenous addition of AIMP2-DX2 reduced the adriamycin and AIMP2-F-induced sub-G1 portion of cells (Figure 3A). Since AIMP2-F was shown to mediate the apoptotic response of p53 to DNA damage, we monitored how AIMP2-F and –DX2 would affect the expression of the luciferase reporter gene with the promoter, Growth Arrest and DNA Damage 45 (GADD45), under the control of p53. Luciferase activity was increased according to the added amount of AIMP2-F in the presence of adriamycin (Figure 3B). However, adriamycin-induced luciferase activity was decreased by the addition of AIMP2-DX2 in a dose-dependent manner (Figure 3C).
We also tested the effect of AIMP2-F and –DX2 by fluorescence staining of the cells. AIMP2-F and –DX2 were transfected into the immortalized MEF cells as GFP fusion proteins and the cells were treated with DNA damaging agent, etoposide, for 8 h and the effects of the two proteins on cell death were monitored by cell morphology change using bright field and fluorescence microscopy. The cells expressing GFP-AIMP2-F showed apoptotic morphology with high frequency while other cells still maintained normal shape (Figure 3D). The cells transfected with GFP-AIMP2-DX2 was observed in 16 h after etoposide treatment. Many cells turned to apoptotic morphology as shown by light microscopy whereas most of the GFP-AIMP2-DX2 expressing cells kept normal morphology (Figure 3E). All of these results further confirmed pro-apoptotic activity of AIMP2-F and suggested that AIMP2-DX2 may render resistance to cell death, perhaps by interfering with the normal activity of AIMP2-F.
Since AIMP2-F mediates the apoptotic response to DNA damage via p53, we investigated whether AIMP2-DX2 would influence the pro-apoptotic interaction of AIMP2-F with p53. To see this possibility, we first tested whether AIMP2-DX2 could bind to p53 like AIMP2-F. We transfected the two forms of AIMP2 into A549 cells. The endogenous p53 was immunoprecipitated and the co-precipitation of AIMP2-F or –DX2 with p53 was determined by Western blotting. Both of AIMP2-F and –DX2 were bound to p53 with a similar affinity (Figure 4A). We then prepared radioactive AIMP2-F and DX2 by in vitro translation and each protein was mixed with GST or GST-p53. GST-p53 was precipitated with glutathione-Sepharose and co-precipitation of AIMP2-F and –DX2 was determined by autoradiography. Both proteins were precipitated with GST-p53 but not with GST, further confirming their ability to bind p53 (Figure 4B).
We then investigated whether AIMP2-DX2 would compete with AIMP2-F for the interaction with p53 by in vitro pull-down assay using GST-p53 and radioactively synthesized AIMP2-F or –DX2. The binding of AIMP2-F to p53 was decreased by the addition of AIMP2-DX2 in dose-dependent manner (Figure 4C). Conversely, AIMP2-F decreased the binding of AIMP2-DX2 to p53 (Figure 4D). These results confirmed that the two forms of AIMP2 would compete for the binding to p53. Since AIMP2-F can block the interaction between p53 and MDM2 [17], we examined how AIMP2-DX2 would affect the interaction between p53 and MDM2. We added different amounts of AIMP2-F and –DX2 to the binding mixture of radioactive MDM2 and GST-p53, and checked how they would influence the association of p53 and MDM2. While MDM2 binding to p53 was decreased as more AIMP2-F was added (Figure 4E), the addition of AIMP2-DX2 gave no inhibitory effect on the p53-MDM2 association (Figure 4F). These results suggest that AIMP2-DX2 does not block MDM2 binding to p53. We then examined the binding of AIMP2-F and –DX2 to p53 at endogenous levels. In normal lung WI-26 cells, we observed UV-dependent association of AIMP2-F with p53 (Figure 4G left). However, in lung cancer A549 cells expressing higher level of AIMP2-DX2, endogenous AIMP2-DX2 bound to p53 to the extent similar to AIMP2-F, further supporting the potential competition between the two forms of AIMP2 for the binding to p53 (Figure 4G right).
We next compared the effect of AIMP2-F and –DX2 on the cellular stability of p53 after knocking down AIMP2-F and –DX2. A549 cells were treated with cycloheximide (CHX) to block the de novo cellular protein synthesis, and AIMP2-F or –DX2 was specifically suppressed with their specific siRNAs. Then, the cells were harvested at time intervals and the cellular levels of p53 were determined by Western blot analysis. When AIMP2-F was suppressed, the cellular level of p53 was more rapidly reduced compared with that of the si-control-treated cells. In contrast, the level of p53 was maintained for longer duration in the cells in which AIMP2-DX2 was suppressed with its specific siRNA (Figure 4H). This property means that AIMP2-DX2 would work like a dominant negative mutant against the pro-apoptotic interaction of AIMP2-F with p53. For this reason, cells showed an increased sensitivity to etoposide-induced cell death when endogenous AIMP2-DX2 was suppressed, (Figure S4A), and this increased sensitivity was abrogated in p53-deficient cells (Figure S4B), indicating that the effect of AIMP2-DX2 involves p53.
Although AIMP2-DX2 expression can be induced, the total amount of AIMP2-DX2 appears to be much lower than that of AIMP2-F in most of the cell lines. This observation poses the question whether AIMP2-DX2 could effectively compete with AIMP2-F for the binding to p53. In this regard, it should be noted that AIMP2-F is mainly anchored to the multisynthetase complex in cytosol [6]. In other words, only a portion of AIMP2-F is recruited to p53 upon DNA damage while the majority of AIMP2-F is still bound to the complex [17]. We checked whether AIMP2-DX2 would also bind to the multi-synthetase complex like AIMP2-F by yeast two hybrid assay. KRS (lysyl-tRNA synthetase), one of the components for the multisynthetase complex, was used as the testing pair for AIMP2-F and –DX2 because AIMP2 make a strong binding to KRS [6], [27]. Although AIMP2-DX2 showed the binding ability to p53 with a similar affinity to AIMP2-F, it lost the normal affinity to KRS as determined by yeast two hybrid assay (Figure 5A). This result was further confirmed by co-immunoprecipitation assay. When Myc-AIMP2-F or –DX2 was immunoprecipitated in 293 cells, KRS was co-precipitated with AIMP2-F, but not with –DX2 (Figure 5B). Conversely, when KRS was immunoprecipitated, co-precipitation of AIMP2–DX2 was significantly lower than that of AIMP2-F (Figure 5C).
We also compared the cellular levels of AIMP2-F and –DX2 that exist unbound to the multisynthetase complex. This time, we used AIMP1, another component of the complex [28], instead of KRS, for immunoprecipitation to make sure that we are not looking at only the binary interaction between AIMP2 and KRS. AIMP1 was immunoprecipitated with its specific antibody, and the co-precipitation of AIMP2-F and -DX2 was determined. AIMP2-F, but not AIMP2-DX2, was co-precipitated with AIMP1 (Figure 5D left). We then examined the amounts of AIMP2-F and –DX2 left in the immune-depleted supernatant. The majority of the cellular AIMP2-DX2 was detected in the supernatant that was comparable to the free form of AIMP2-F (Figure 5D right). All of these results suggest that AIMP2-DX2 would exist mainly as free form, not bound to the multisynthetase complex, whereas AIMP2-F is bound to the complex. Thus, AIMP2-DX2 would be able to compete with AIMP2-F that is dissociated from the multisynthetase complex for the binding to p53 and Far-upstream element-binding protein (FBP) although the total amount of AIMP2-DX2 is lower than AIMP2-F (Figure 5E).
Since AIMP2 works as a potent tumor suppressor, AIMP2-DX2 may exert the opposite activity by interfering with the normal activity of AIMP2. To determine this possibility, we isolated WI-26 cells, in which AIMP2-DX2 expression was induced by BPDE to different degrees, and compared their propensity to anchorage-independent colony formation. The same number of the cells was spread on agar plates and the resulting numbers of colonies were counted. The number of the anchorage-independent colonies showed a positive correlation to the expression levels of AIMP2-DX2 (Figure 6A), implicating the potential oncogenic property of AIMP2-DX2. To further confirm these results, we transfected an empty vector, AIMP2-F and AIMP2-DX2 into MEF, and then performed anchorage-independent colony forming assay. Transfection of AIMP2-F and –DX2 resulted in fewer and more colonies compared to that of empty vector (Figure 6B).
To see whether an increased expression of AIMP2-DX2 may render susceptibility to tumorigenesis, we generated transgenic mice in which AIMP2-DX2 is constitutively expressed, as described in Materials and Methods. The increased expression of AIMP2-DX2 was confirmed by Southern blotting with the primers specific to the junction of exon 1 and 3, and by RT-PCR with the primers specific to CMV transgenic vector (Figure S5). In addition, we compared the expression of AIMP2-DX2 by Western blotting and found that AIMP2-DX2 is highly expressed in the transgenic mice (Figure S5). We isolated MEFs from the wild type and AIMP2-DX2 transgenic mice and compared the cellular levels of p53 and its phosphorylation. The p53 level and phosphorylation were significantly reduced in the transgenic mice (Figure 6C). Normal and AIMP2-DX2 MEFs were also treated with adriamycin and their sensitivity to cell death was compared by flow cytometry. AIMP2-DX2 transgenic cells showed lower cell death (Figure 6D), resulting in the enhanced cell number (Figure 6E) compared to the wild type cells. The wild type (n = 14) and AIMP2-DX2 transgenic mice (n = 18) were also compared for their susceptibility to benzo[a]pyrene-induced lung carcinogenesis. Lungs were isolated and the formation of lung cancers were determined as described previously [19] and tumor area was analyzed by Bio-Image J 2.0. The AIMP2-DX2 transgenic mice showed a significant increase in tumor incidence and regions compared to the wild type counterpart (Figure 6F, 6G and 6H). All of these results suggest the oncogenic potential of AIMP2-DX2.
To see whether we can control tumor progression by suppressing the expression of AIMP2-DX2, we established a lung cancer xenograft model with the lung cancer cell line, H460. We introduced siRNA against AIMP2-DX2 and siRNA control via intratumoral injection four times at three day interval from 5 days after cell transplantation. The tumor growth was retarded by the injection of si-DX2 (Figure 7A). We isolated the tumors and compared the expression levels of p53. The tissues isolated from the si-DX2-injected tumors showed higher expression of p53 (Figure 7B), indicating that the si-DX2-injected tumors may have enhanced cell death through the activation of p53. We also determined whether intratumoral injection of si-DX2 actually reduced the expression level of AIMP2-DX2 in tumor region by quantitative real-time PCR. The cellular levels of AIMP2-DX2 were reduced about 50% by the introduction of si-DX2 compared to those in the siRNA control-injected tumors (Figure 7C). For an additional experiment, we introduced luciferase-expressing A549 lung cancer cells into the mice via tail vein. After 2 months of cancer cell inoculation, siRNA control and si-DX2 were delivered to the lungs via intubation four times in three day interval and the mice were incubated another 2 weeks. The lung cancer cell growth was monitored by the luciferase activity released from the growing cancer cells. While the control siRNA-injected mice (n = 5) showed a 2.4 to 4.8 fold increase in luciferase activity, si-DX2-delivered mice showed 1.1 to 2.8 fold increase or even a shrinkage of tumors (Figure 7D). We also isolated the lungs and compared the tumor area by histological analysis. Compared to the siRNA control-delivered tumors, si-DX2-delivered tumors showed smaller cancer regions (Figure 7E). Specific suppression of AIMP2-DX2 transcript was also confirmed by quantitative RT-PCR (Figure S6). Next, we monitored the effect of AIMP2-DX2 suppression on tumor growth using a carcinogen-induced lung cancer model. Lung cancer in mice was induced by the injection of benzo[a]pyrene (B[a]P). Then, we prepared the plasmid DNA encoding shRNA targeted against AIMP2-DX2, and delivered it into lungs via inhalation device eight times in three day interval (Figure S7). We isolated the lungs from the EV- and sh-DX2 treated mice and compared the tumor growth by H&E staining. The DX2-shRNA treated lungs contained smaller tumor area compared to those treated with an empty vector (Figure 7F, 7G and Figure S8). To assure that the plasmid is efficiently delivered to various lung tissues, the same plasmid encoding GFP was prepared and delivered to lung using the same method. The lungs were isolated from the treated mice and the delivery of the plasmid was monitored by the expression of GFP (Figure S8A and S8B). Delivery of sh-DX2 from the treated lungs was confirmed by PCR (Figure S8C). All of these results consistently demonstrated that suppression of AIMP2-DX2 expression can retard tumor growth.
Since AIMP2-DX2 compromises tumor suppressive activity of AIMP2-F via competitive inhibition, the expression ratio of AIMP2-DX2 to AIMP2-F may have a relationship to cancer progression. To explore this possibility, we obtained squamous cell carcinoma and adenocarcinoma tissues from lung cancer patients who were in different stages, and the expression ratios of AIMP2-DX2 to -F were compared. The AIMP2-DX2/F ratios were gradually increased from those of the normal tissues according to cancer stage (Figure 8A).
We also determined the expression of AIMP2-DX2 and AIMP2-F in the normal and cancerous regions isolated from 97 NSCLC patients by quantitative RT-PCR. The patients were divided into two groups based on the expression ratios of AIMP2-DX2 to AIMP2-F. 43 patients showed higher expression ratios of AIMP2-DX2 to AIMP2-F in cancerous regions compared with those in the normal regions while the other 54 patients did not show apparent difference between the cancer and normal tissues. The patient group with tumors that expressed higher ratios of AIMP2-DX2 to AIMP2-F exhibited a poorer overall (Figure 8B) and disease-free survival (Figure 8C) pattern compared to the group in which the patients did not show higher expression of AIMP2-DX2 although the degree of difference between the two groups varied to some extent. When we examined the prognostic values with several factors, the AIMP2-DX2 to AIMP2-F expression ratio showed the most significant correlation with patient survival (Table 1, Table S1, Hazard Ratio for Overall Survival = 6.83, P = 0.045; Hazard Ratio for Disease Free Survival = 2.25, P = 0.046). Altogether, the results further support that increased expression of AIMP2-DX2 may reflect the aggressiveness of cancer cells.
Although alternative splicing can be a way for the diversification of a gene function [1]–[3], the generation of inappropriate splicing product can be pathologic. In particular, the cancer-associated formation of alternative splicing variants has been reported in prostate cancer [29]. Oncogenic alternative splicing variants of Acute Myeloid Leukemia 1 (AML1) and TrkA have been found in ovarian cancer [30] and neuroblastoma [31], respectively. Similarly, the splicing variant of p53 was also shown to modulate the normal activity of p53 [32], [33]. In p63, another p53 family tumor suppressor, a splicing variant expressed dominant negative activity [34]–[36]. Based on these findings, alternative splicing variants with oncogenic properties are expected to be found in other tumor suppressors. Here we report a generation of an oncogenic alternative splicing variant of tumor suppressor AIMP2 in lung cancer cells and patient tissues (Figure 1).
The base substitution of A152 to G was found at the exon 2 region of AIMP2 in carcinogen-induced transformed cells. This mutation induced exon 2 skipping, resulting in the increase of AIMP2-DX2 production (Figure 2E and 2F). These results further support the association of AIMP2-DX2 with tumorigenesis. Generation of AIMP2-DX2 does not seem to seriously affect the cellular level of AIMP2-F transcript. Perhaps this is because the amount of AIMP2-DX2 is minor compared to that of AIMP2-F or because the loss of AIMP2-F resulting from the generation of AIMP2-DX2 is re-filled by a compensatory expression. Further detailed investigation is needed to see how alternative splicing between AIMP2-F and –DX2 is regulated.
AIMP2-DX2 competes with AIMP2-F for the binding to p53 (Figure 4). In our previous work, deletion mapping determined that the C-terminal region (161–312 aa) of AIMP2 binds to the N-terminal region of p53 [17]. Thus, it is not surprising that AIMP2-DX2 retains its binding ability to p53 because the exon 2 encodes the region of 46–113 aa of AIMP2, which is not involved in the interaction with p53. Based on the conformational modeling, the exon 2 domain is expected to protrude outward and may give a steric hindrance for MDM2 to bind p53 (data not shown). Since AIMP2-DX2 polypeptide is smaller than AIMP2-F and may not have the expected protuberance of exon 2, it may not be able to block the MDM2 binding to p53.
AIMP2 also plays critical roles in the apoptotic activity of TNF-α [18] and the anti-proliferative activity of TGF-β [16]. Although in this study we focused on the antagonistic function of AIMP2-DX2 against the pro-apoptotic activity of AIMP2-F via p53 in response to DNA damage, it may also influence the normal activity of AIMP2 in these two other pathways with a similar mode of action, namely, through the competitive binding to the target proteins. Since AIMP2-F augments TGF-β-dependent growth arrest via downregulation of FBP and c-Myc [16], we examined whether AIMP2-DX2 would also work against AIMP2-F in this pathway. While overexpression of AIMP2-F increased p53 level, it reduced the levels of FBP and c-Myc (Figure S9A). Exogenous supplementation of AIMP2-DX2 reversed these effects of AIMP2-F (Figure S9A). We also examined the effect of AIMP2-DX2 on the ubiquitination of AIMP2-F target proteins. While p53 ubiquitination was inhibited by AIMP2-F, it was interfered by the addition of AIMP2-DX2 (Figure S9B). Conversely, ubiquitination of FBP was enhanced by the addition of AIMP2-F (Figure S9C) and this effect was inhibited by exogenous introduction of AIMP2-DX2 (Figure S9C). We also compared the expression level of c-Myc as well as p53 among WI-26 and other lung cancer cell lines such as H460, H1975 and H292 (all of these cell lines contain wild type p53) in which AIMP2-DX2 levels are increased. In these cell lines, c-Myc level was more increased compared to that of the normal WI-26 cells while p53 levels were more decreased (Figure S9D). Interestingly, all of these cells lines showed increased SF2/ASF levels, further supporting the regulatory connection between this splicing factor and expression of AIMP2-DX2. Based on these results, AIMP2-DX2 is expected to compromise not only the pro-apoptotic activity but also anti-proliferative activities of AIMP2-F.
Considering the functional importance of AIMP2 in cell fate determination as well as in the assembly of multisynthetase complex [37], [38], its cellular level and activity may need to be tightly controlled. For this, expression of AIMP2-DX2 can be a part of a normal regulatory mechanism to finely control the function of AIMP2. Perhaps, the generation of AIMP2-DX2 could become out of control by mutations that can disrupt the normal splicing process and lead to cancer formation. The responsible mutations can be located in trans-acting splicing factors or in the promoter or gene encoding AIMP2. Suppression of AIMP2-DX2 in lung cancer reduced tumor growth in lung cancer model (Figure 7), and lung cancer patient analysis demonstrated the correlation of AIMP2-DX2 expression with cancer progression and patient survival (Figure 8). Combined with the molecular working mechanism and cellular effect, AIMP2-DX2 appears to be a novel therapeutic target against lung cancer. It would be interesting to see whether this variant would be also involved in other types of cancer.
Normal lung cell line, WI-26, was purchased from Korea Cell Line Bank (KCLB) and NL-20 was provided by Dr. M.-H. Cho (Seoul National University). Lung carcinoma cell lines A549 and NCI-H460 were obtained from ATCC, and H322 and H157 from KCLB. The siRNAs targeting AIMP2-F and -DX2 were designed as the sequences of AGAGCUUGCAGAG ACAGGUUAGACU and UCAGCGCCCCGUAAUCCUGCACG UG, respectively. AIMP2-DX2 polypeptide was used as the antigen and the selected hybridoma clone (#324) generated the monoclonal antibody recognizing both of AIMP2-F and –DX2. This antibody was used for immunoblotting and immunohistochemistry. Anti-KRS and –AIMP1 antibodies were purchased from Abchem.
The expression of AIMP2-DX2 and –F were analyzed by quantitative real time RT-PCR. Eleven normal lung samples and fourteen patients with lung adenocarcinoma were retrospectively identified from the surgical pathology files of the Department of Pathology at Samsung Medical Center and their archival formalin-fixed paraffin-embedded (FFPE) tissues were obtained. All the samples were collected anonymously according to Institutional Review Board guidelines. All patients had undergone a surgical operation and had received neither chemotherapy nor radiotherapy before surgical resection. For total RNA extraction from FFPE tissues, each tissue section was stained with hematoxylin and cancer regions were microdissected using laser microdissection system (ION LMD, JungWoo International, Korea). Paradise Whole Transcript RT Reagent System (Arcturus, CA, USA) was used for RNA isolation and RT of FFPE samples. Due to the limitation of RNA amount extracted from FFPE tissues, half RNA and cDNA were used for reverse transcription and quantitative RT-PCR, respectively. PCR primers and Taqman probes for this study are provided from Metabion (Germany).
Cells transfected with the plasmid encoding AIMP2-F or –DX2 were cultivated in the absence or presence of adriamycin, fixed with 70% ethanol for 1 hour at 4°C and washed with ice-cold PBS two times. Then, the 1×106 cells were stained with PI (50 µg/ml) containing 0.1% sodium citrate, 0.3% NP40 and 50 µg/ml RNase A for 40 minutes, and subjected to flow cytometry (FACS Caliber, Beckton-Dickinson) for the determination of apoptotic cells by counting sub-G1 cells. For each sample, 20,000 cells were analyzed using Cell Quest Pro software. All of the experiments were repeated three times and averaged.
pCDNA3 encoding AIMP2-F and -DX2 were transfected into 12.5 days mouse embryonic fibroblasts. The cell lines stably expressing each of the transfected plasmid were established by G418 selection. For soft agar colony assays, the cells were diluted into 0.3% agar in DMEM containing 10% FBS and seeded in triplicate onto 0.6% agar containing culture medium. 200 cells were seeded on each well in 12-well plate. The colonies were fed in every 3 to 4 days and evaluated after 5 weeks. To evaluate the correlation between the expression level of AIMP2-DX2 and colony formation, WI-26 cells (Korean Cell Line Bank) were treated with 0.1 µM BPDE once in every three days for 4 weeks and the surviving colonies were observed after 2 weeks from the chemical treatment. The 20 separate colonies were randomly selected to establish the cell lines.
For immunoprecipitation, AIMP2-F or DX2 was ectopically expressed in HEK293 cells by transfection. The cells were lysed with the lysis buffer containing 1% NP-40, 0.5% deoxycholate and protease inhibitor cocktail (Calbiochem). The lysates were incubated with the antibody against p53 (DO-1, Santacruz), and precipitated by protein G (Invitrogen) 4°C overnight and co-precipitation of the two proteins were blotted with anti-Myc or anti-AIMP2 antibody. For in vitro pull down assay, GST-p53 or GST was mixed the protein extracts with glutathione-Sepharose in the PBS buffer containing 1% Triton X-100 and 0.5% N-laurylsarcosine at 4°C for 2 hours. We synthesized MDM2 and AIMP2-F or -DX2 by in vitro translation in the presence of [35S] methionine using TNT Quick coupled Transcription/Translation system (Promega). The synthesized peptide was added to the GST protein mixtures above, incubated at 4°C for 4 hours with rotation in the PBS buffer containing 1% Triton X-100, 0.5% N-laurylsarcosine, 1 mM DTT, 2 mM EDTA and 300 µM phenylmethylsulfonyl fluoride (PMSF), and washed six times with the same buffer containing 0.5% Triton X-100. We then eluted the proteins bound to Sepharose beads with the SDS sample buffer, separated by SDS-PAGE and autoradiographed. For yeast two hybridization assay, the cDNAs encoding human AIMP2-F, –DX2, p53 and KRS were obtained by PCR using specific primers. The PCR products were digested with EcoRI and XhoI, and ligated into the same sites of pEG202 (LexA) and pJG4-5 (B42). The interactions between LexA-AIMP2 fragments and B42-KRS or –p53 were analyzed for their ability to form blue colonies on yeast medium containing X-gal.
To test the p53-dependent transcriptional activity, the target gene GADD45-luciferase vector was transfected into A549 cells. The cells lysates were prepared and reacted by luciferase assay kit following the manufactuerer's protocol (Promega) and the luciferase activity was analyzed using luminometer.
The 1263 bp genomic DNA of AIMP2 covering exon 2 with flanking introns was amplified with Platinum Taq DNA polymerase high fidelity (Invitrogen) and DNA template from BPDE-transformed WI-26 cells (WI-26T) with the primers 5′GAAGAGTCTAACCTGTC TCTGCAAGCTCTTGAG3′ and 5′AACATGCTCTTGGCT CTGCCTTTG3′. The resulting PCR products were cloned into pGEMT-Easy vector (Promega) and sequenced. To verify the effect of mutations in WI-26T cells on AIMP2-DX2 generation, the splicing reporters, pGINT and pRINT, which express EGFP and RFP, respectively as an indication of splicing, were kindly provided by Dr. Garcia-Blanco [21]. The genomic wild type AIMP2 exon 2 regions including cis-acting splicing elements were amplified with Platinum Taq DNA Polymerase High Fidelity (Invitrogen) using primers (5′GCGCGGATCCTCCCAAAGTGCTGGGATTACAGGT3′ and 5′GCGCGTCGACAACATGCTCTTGGCTCTGCCTTTG3′) and WI-26 genomic DNA template, and cloned into pGINT after BamHI and SalI restriction enzyme digestion. The constructed plasmid pGINT-exon 2 which contains wild type AIMP2 exon 2 with the flanking intron regions was used as a template for site-directed mutagenesis to introduce each of the mutations identified from the WI-26T cells. To assess the effect of mutations on the generation of AIMP2-DX2, 0.5 µg of pGINT-EX2 wild type and its mutant derivatives were transfected into HEK293 cells seeded on a 6-well plate using GenePORTER reagent (Gelantis). The same amount of pRINT was transfected as a splicing and transfection control. After 24 hours incubation, the cells were visualized under fluorescent microscope to detect green and red fluorescence. The same set of transfected cells were used for RNA extraction using RNA extraction kit (Qiagen) and cDNA synthesis. Primer pairs, 5′ACGTAAACGGCCACAAGTTC3′ and 5′AAGTCGTGCTGCTTCATGTG 3′, were used for the detection of EGFP or exon 2-included EGFP transcripts.
Wild type sequence of AIMP2 exon 2 from WI-26 and the exon 2 sequence with mutations found from WI-26T were analyzed for exonic splicing enhancer (ESE) of four serine/arginine-rich (SR) proteins. ESE finder program ver. 3.0 was used for the prediction. The highest sequence score for each SR protein was calculated in a random-sequence pool and the threshold values were set as the median of the highest score. The threshold values were as follows: SF2/ASF heptamer motif, 1.956; SF2/ASF (IgM-BRCA1) heptamer motif, 1.867; SC35 octamer motif, 2.383; SRp40 heptamer motif, 2.670; and SRp55 hexamer motif, 2.676.
HEK293T cells were transfected with GFP-tagged SF/ASF as well as pGINT-exon 2 or pGINT-exon 2 with A152G mutation (pGINT-A152G). After 24 h incubation, nuclear fraction was obtained and subjected to immunoprecipitation using anti-GFP antibody. GFP-SF2/ASF-bound mRNAs were purified using Trizol reagent according to the manufacturer's guideline. Reverse transcriptase PCR was conducted using AIMP2 exon 2-specific primers. To verify the effect of A152G mutation on the interaction with SF2/ASF in vitro, RNA probes encompassing full sequence of AIMP2 exon 2 were synthesized by in vitro transcription using [α-32P]UTP and T7 promoter-containing DNA templates obtained from PCR amplification of pGINT-exon 2 WT and A152G. The RNA probes were incubated at 30°C for 30 min with GFP-SF2/ASF which was immunoprecipitated as mentioned above. After washing 3 times with the reaction buffer (20 mM HEPES, pH 7.8, 200 mM KCl, 20 mM NaCl, 10% glycerol, 2 mM DTT, and 2 mM MgCl2) containing 0.05% NP-40, the RNA-protein complex was eluted using formaldehyde sample buffer, separated by electrophoresis in 7.5% urea polyacrylamide gel, and detected by autoradiography.
Murine AIMP2/p38 sequence (BC026972) lacking exon 2 was cloned into pCDNA3.1(+) (Invitrogen) at HindIII and XhoI. The region of CMV promoter to poly A site was linearized and injected into mouse fertilized egg of C57BL/6 strain. The insertion of AIMP2-DX2 was confirmed by Southern blot and PCR analyses using the genomic DNA isolated from MEF cells. AIMP2-DX2 expression was determined by Northern and Western blot analyses (see Figure S3 for details).
A549 cells expressing luciferase reporter (107) were injected into tail veins of five-week-old female BALB/c nu/nu mice. After 5–6 weeks, the dissemination of the cells was monitored by IVIS (Xenogen). Mice were divided into two group (n = 5/group) depending on their releasing photon flux and the mixtures of si-scramble AIMP2-DX2 (50 µg) or si-AIMP2-DX2 (50 µg) with GDM-PEI (glycerol dimethacrylate polyethyleneimine) [39] in 50 µl of 0.9% saline were delivered intratracheally into lung. To monitor photon flux, mice were anesthetized with isoflurane inhalation, and 100 ul of D-luciferin (7.5 mg/ml, Xenogen) were intraperitoneally (i.p.) injected. Bioluminescence imaging with CCD camera (IVIS, Xenogen) was initiated 30 min after injection for 1–60 seconds, depending on the amount of luciferase activity. The data were expressed as photon-flux (photons/s/cm2/steradian).
For xenograft experiment, NCI-H460 lung cancer cells (107) were suspended in 200 µl of 0.9% saline and subcutaneously injected into 6 weeks old female nude mice. The tumor volumes were monitored three times a week. 50 µg of siRNA mixed with GDM-PEI were directly injected into tumors in three directions for four times in the indicated interval from the point that tumor volume reached 250 mm3. The volume was calculated by (length x width x height)/2.
We designed shRNA against AIMP2-DX2 (TCGAGCGGGCCACGTGCAGGACTA TTCAAGAGATAGTCCTGCACGTGGCCCGCTTTT, underlined regions are matched to the DX2 sequence) and cloned into IMT-700 vector system (Imgenex) using SalI and XbaI. The plasmid was mixed with glucosylated polyethyleneimine (G-PEI) at 1.64:1 weight ratio. After 1 week from last chemical injection, the DNA mixture was delivered into lung through intranasal pathway using the humid vacuum chamber in which the DNA mixture was vaporized. The DNA vapor was inhaled for 30 minutes through the nose of the mice that were fixed in the cylinder using Bio-Rad compressor as previously described [40]. From 6 weeks after the last injection of BP, the administration of DNA was conducted twice a week for 4 weeks and the tumor areas were measured.
To determine expression ratios of AIMP2-DX2 to AIMP2-F in different lung cancer stages, cDNAs from the frozen tissues of 23 patients with squamous cell carcinoma and adenocarcinoma were provided from Roswell Cancer Park Institute and analyzed by real-time PCR as described above. The research subject was reviewed and determined to be non-human subject research (NHSR) by the committee for Research Subject Protection of the Institute. Correlation of AIMP2-DX2 expression with patient survival was also investigated. This study included patients (n = 97) with pathological stages I, II or IIIA NSCLC patients who underwent curative surgical resection at the Kyungpook National University Hospital (Daegu, Korea) between January 2001 and December 2008. Patients who underwent chemotherapy or radiotherapy prior to surgery were excluded to avoid the effects on DNA. All the tissues were obtained at the time of surgery and then rapidly frozen in liquid nitrogen and stored at -80°C until the relevant bioassays were conducted. The histologic types of lung cancers, according to the World Health Organization classifications, were as follows: 44 cases (45.4%) of squamous cell carcinomas and 53 cases (54.6%) of adenocarcinomas. The pathologic staging of the tumors, which was determined according to the standard of the American Joint Committee on Cancer (AJCC), was as follows: 59 patients (60.8%) of stage I, 20 patients (21.6%) of stage II and 18 patients (18.6%) of stage IIIA. Written informed consent was obtained from all patients prior to surgery and this study was approved by the Institutional Review Board of the Kyungpook National University Hospital. Overall survival (OS) was measured from the day of surgery until the date of death or to the date of the last follow-up. Disease-free survival (DFS) was calculated from the day of surgery until recurrence or death from any cause. The survival estimates were calculated using the Kaplan-Meier method. The differences in OS or DFS across different genotypes were compared using the log-rank test. Hazard ratios (HRs) and 95% confidence intervals (CIs) were estimated using multivariate Cox proportional hazards models, with adjustment for age (≤63 versus >63 years), gender (male versus female), smoking status (never- versus ever-smoker), and pathologic stage (I versus II-IIIA). All analyses were performed using Statistical Analysis System for Windows, version 9.1 (SAS Institute, Cary, NC, USA).
All human subject researches were approved by the appropriate ethics committees according to the Declaration of Helsinki. The correlation of AIMP2-DX2 expression with patient survival rate was studied under the review of institutional board in Kyungpook National University Hospital. The title was ‘Identification of Cancer Specific Biomarker in Lung Cancer: 74005-263’. Roswell Park Cancer Institute approved the project ‘Evaluation of AIMP2-DX2 as a Cellular marker for Diagnosis of Lung Cancer: NHR002008′ through their ethics committee.
|
10.1371/journal.pcbi.1005821 | Neural field model to reconcile structure with function in primary visual cortex | Voltage-sensitive dye imaging experiments in primary visual cortex (V1) have shown that local, oriented visual stimuli elicit stable orientation-selective activation within the stimulus retinotopic footprint. The cortical activation dynamically extends far beyond the retinotopic footprint, but the peripheral spread stays non-selective—a surprising finding given a number of anatomo-functional studies showing the orientation specificity of long-range connections. Here we use a computational model to investigate this apparent discrepancy by studying the expected population response using known published anatomical constraints. The dynamics of input-driven localized states were simulated in a planar neural field model with multiple sub-populations encoding orientation. The realistic connectivity profile has parameters controlling the clustering of long-range connections and their orientation bias. We found substantial overlap between the anatomically relevant parameter range and a steep decay in orientation selective activation that is consistent with the imaging experiments. In this way our study reconciles the reported orientation bias of long-range connections with the functional expression of orientation selective neural activity. Our results demonstrate this sharp decay is contingent on three factors, that long-range connections are sufficiently diffuse, that the orientation bias of these connections is in an intermediate range (consistent with anatomy) and that excitation is sufficiently balanced by inhibition. Conversely, our modelling results predict that, for reduced inhibition strength, spurious orientation selective activation could be generated through long-range lateral connections. Furthermore, if the orientation bias of lateral connections is very strong, or if inhibition is particularly weak, the network operates close to an instability leading to unbounded cortical activation.
| Optical imaging techniques can reveal the dynamical patterns of cortical activation that encode low-level visual features like position and orientation, which are shaped by both feed-forward projections, recurrent and long-range intra-cortical connections. Anatomical studies have characterized intra-cortical connections, however, it is non-trivial to predict from this data how evoked activity might spread across cortex. Indeed, there remains an apparent conflict between the reported orientation bias of cortical connections, and imaging studies on the propagation of cortical activity. Our study reconciles structure (anatomy) with function (evoked activity) using a dynamic neural field model that predicts the dynamics of cortical activation in a setting both inspired by and parametrically matched to the available anatomical data.
| Horizontal connections link cells separated within the same cortical area, over a distance of a few millimeters (mm) covering several iso-functional columns [1, 2] and spatially distributed into regular clusters [3–5]. In V1, the lattice-like pattern of connectivity is reminiscent of the spatial regularity observed in orientation maps and has therefore been proposed to link neurons with similar preferred orientation [6]. Functional mapping combined with retrograde labeling [7] has shown that pyramidal cell horizontal axons have a bias to preferentially connect iso-orientation loci (in cat [8, 9], tree shew [10] and monkey [11]). However, this result depends on cell type and location since neurons in layer 4 (L4) [12], pinwheel centers ([13] but see [14, 15] since this result may depend critically on the distance to the pinwheel center) and inhibitory cells [16, 17] connect without orientation bias. As a consequence of this cellular heterogeneity, it is not trivial to predict the selectivity of stimulus-driven horizontal activation.
More information was recently gained from population measures of orientation selectivity at mesoscopic scales. Techniques such as optical imaging has led to the description of cartographic organization of many cortical structures [18, 19]. The development of voltage-sensitive dye [20] has further allowed for investigation of dynamic features and computations arising within these maps, see reviews in [21] and [22]. In [23] voltage sensitive dye imaging (VSDI) was used to study the retinotopic activation with localized oriented inputs in cat V1. A characteristic plateau of activity, coinciding with the retinotopic extent of the stimulus, was independent of stimulus orientation. Within the plateau several peaks of activation would appear, with location strongly dependent on orientation. [24] further explored orientation selectivity outside the retinotopic activation using localized stimuli. Over several hundred milliseconds activation gradually propagated outwards, extending several mm beyond the feedforward footprint (FFF). The dynamic and the spatial range of this activation beyond the FFF is presumably generated by long-range excitatory connections in L2/3 of V1. Note that, although less plausible because their spatial and temporal properties are not in the appropriate range (see discussion in [25] and [24]), alternative circuits such as intra-thalamic, thalamo-cortical divergence and feedback loops cannot be entirely dismissed (but see [26]). Interestingly, only a local component of this activation, circumscribed within the FFF, was found to be orientation selective (Fig 1A), at first glance a surprising finding given the numerous studies showing an iso-orientation bias for long-range connections. However, as discussed in [24] and introduced above, this result may be expected from the anatomy known from studies already published at that time. First, because the iso-orientation connection bias is small and has been quantified only for short horizontal distances (mostly <1–1.5 mm, a distance for which a similar small bias is also reported by [24]). Second, because it can be seen in the few existing intracellular labeling studies that the iso-orientation bias tends to decrease with distance [17, 14]. Thus, the results from [24], although in contradiction with a strict “like-to-like connectivity” principle, call for modification depending on cortical distance: from a like-to-like bias in functional connectivity at short range toward no bias at long range. Importantly, two follow-up studies consolidated these findings with an optogenetics functional confirmation ([27], see also [28, 29]) and a timely anatomical clarification using quantitative and statistical analysis of intracellular labeled neurons [15].
The aim of this modelling study is to explore the relationship between lateral connectivity properties (structure) and the way activity spreads across cortex as evoked by localized, oriented stimuli (function). It remains to reconcile the studies of iso-orientation bias of anatomical connectivity in V1 [7, 9, 14] and the properties of cortical activation observed in other experiments [24, 27, 15]. To achieve this, we have developed a dynamical model to investigate stimulus driven activity in 2D cortical space with a discrete representation of orientation. The neural field equation gives a coarse-grained description of average membrane potential on a continuous domain [30, 31]. This framework has been widely used to model visual cortex (e.g. [32–35]) and cortical dynamics in general (see reviews [36] and [37]). This level of description is well-suited for comparison with VSDI [38–40]. The model’s design allows for connectivity properties such as the spatial profile of excitation and inhibition to be investigated. Model parameters were constrained to be consistent with the level of orientation bias reported in [14]. Our study shows that realistic cortical connectivity schemas [14, 15] govern a spatio-temporal dynamics of cortical activation in accordance with the observed functional dynamics [24, 27]. Our approach allowed us to further elaborate on the non-trivial link between anatomical constraints and predictions of population level activation patterns. To probe the structure-function link, we used our model to make experimentally tractable predictions on the specific role of excitatory-inhibitory balance.
Our neural field equation model [30, 31] gives a mesoscopic, continuous description of neural activity on a 2D plane (x, y) parallel with the cortical surface in layer 2/3 of V1. Neural activity is represented by an average membrane potential ui(x, y, t) evolving with time t in four sub-populations, each associated with a different orientation i = {0°, 45°, 90°, 135°} (the ° notation will be dropped in the remainder of the manuscript). The decomposition into sub-populations is a convenient abstraction for modelling purposes; these sub-population outputs ui(x, y, t) will be combined into activity variables that can be compared with experimental data in Conversion of model output to VSD-like signal. The following integro-differential equation describes the dynamics of the ui population:
τ ∂ ∂ t u i ( x , y , t ) = − ∑ j ρ j u j ( x , y , t ) (1) + ∑ j k j I j ( x , y ) ( 1 + β inp J j ( x , y ) ) (2) + S ( u i ( x , y , t ) ) ⋆ ( x , y ) [ g ex w E loc ( x , y ) − g in w I ( x , y ) ] (3) + ( 1 + β rec J i ( x , y ) ) S ( u i ( x , y , t ) ) ⋆ ( x , y ) g ex w E lat ( x , y ) . (4)
The cortical timescale is τ = 10 ms, and the rate of change of ui is proportional to the right hand side of this equation with the following terms
(1) Decay of the population activity to resting potential (2) Stimulus driven input with feedforward footprint I, modulated by the orientation map J (3) Non-selective intra-cortical interactions gated by sigmoidal threshold function S (4) Orientation-selective interactions, modulated by the orientation map J
Term (1) describes a decay back to an resting potential of ui = 0 (the membrane potential has arbitrary units). Within the sub-population the decay term has strength ρi=j = 1 (an arbitrary choice for this first constant). Between sub-populations the decay term has strength ρi≠j = 0.1 (local, linear cross inhibition). For ρ > 0.2 this cross inhibition can generate undesired above-threshold activation in non-stimulated sub-populations, therefore, a value 50% below this was selected. Term (2) describes localised circular inputs with one of four orientations (e.g. Fig 1B). For an input with say orientation 0, I0 is active and I45, I90 and I135 are zero. The input weighting for a sub-population’s associated orientation ki=j = k1 is larger than the weighting for other orientations ki≠j = k2, with k1 = 2k2. Inputs are modulated by the orientation map Ji with strength βinp = 0.25 (Fig 1D and 1I). For βinp > 0.2 the spatial phase of multi-bump patterns of activity will match the orientation preference map as required, increasing it further has little effect. Term (3) describes non-selective lateral connections within the sub-population ui via a convolution ⋆ over (x, y) (spatial integrals, see (6)) with the components of the spatial connectivity profile. This profile is radially symmetric and describes the average connectivity at any point in cortex. It is broken down into local excitatory w E loc and inhibitory wI components, see Fig 1G and Details and equations for connectivity profile. In the convolutions ui is processed through the sigmoidal threshold function S, which transforms membrane potential into a normalized firing rate. Any locations above threshold will influence their neighbors through lateral connections. Weighting constants gex and gin are determined by theoretical constraints, see Details and equations for connectivity profile and (17). Term (4) describes orientation-selective excitatory connections as modulated by the preference map Ji with strength βrec.
An example of the simulated sub-population activity is shown in Fig 1F for an input with orientation 0. The main parameters of interest in this study are RWex, which modifies the spatial width of excitatory peaks (Fig 1G) and βrec (taking values between 0 and 1), which controls the amplitude modulation of long-range excitation by the orientation preference map (Fig 1H). We will also study a parameter C controlling the strength of inhibition (equivalently the balance between excitation and inhibition) as defined by (16) in Details and equations for connectivity profile.
The firing rate (or threshold) function is given by a sigmoid
S ( u ) = 1 1 + e - μ u + θ - 1 1 + e θ , μ , θ > 0 , (5)
where μ = 2.3 is a slope parameter and θ = 5.6 the threshold. Input strengths k1 = 2.8 and k2 = 1.4 are set such that inputs to the stimulated sub-population are above threshold in (5) μk1 > θ, whilst inputs to other sub-populations are below threshold μk2 < θ. The particular form of S is chosen such that ui = 0 (all populations at resting potential) is a always a solution to the model equations and values of μ and θ are chosen such that this is the only stable solution with no inputs [41]. The spatial convolution terms in (3)–(4) are expressed as integrals, e.g. Term (4) can be computed as one integral
wElat(x,y)⋆(x,y)[ S(ui(x,y,t))(1+βrecJi(x,y)) ]=+∫x,ywElat(x−x′,y−y′)[ S(ui(x′,y′,t))(1+βrecJi(x′,y′)) ]dx′dy′, (6)
with dummy variables x′ and y′. In this case the modulation by Ji is introduced whilst still preserving the convolutional structure, allowing for the numerical methods described in [41], exploiting Fast Fourier Transforms to be applied. The radial inputs Ii(r), r = x 2 + y 2, as plotted in Fig 1I(left), are given by
I i ( r ) = { 1 , r < 0 . 7 Λ h ( r , 0 . 7 Λ , 0 . 3 Λ ) r ≥ 0 . 7 Λ . (7)
where h is the radially shifted Gaussian ring given below in (10). The extent of Ii and its decay at the stimulus border were chosen to match cortical point spread functions measured in [24]. In simulations the input amplitude ramps up linearly from 0 to full amplitude in the interval t ∈ [20, 120] ms, see [40].
For each sub-population a finite differencing scheme for the domain [−L, L] × [−L, L] with L = 30 and N = 128 evenly distributed gridpoints in each spatial direction is used, as described in [41]. Model simulations were run on a domain much larger (2L = 60, relative to Λ = 2π) than the localized patterns of activation studied here, justifying the use of periodic boundary conditions. A standard Runge-Kutta time stepper in Matlab was used for model simulations with default tolerances. The source code for the full implementation of the model has been made available in the Supplemental Information (S1 Code).
Earlier theoretical works characterized connectivity constraints that lead to stable localized patterns of activation in 1D [42, 43] and in 2D [44, 45]. A common choice of connectivity function is a so-called Mexican hat (e.g. a difference of Gaussians) featuring a broader footprint for inhibition than for excitation [36, 37]. [41] suggested a role for longer-range excitation serving to stabilize larger patterns of activation and equating the separation between peaks in excitation with Λ (hypercolumn separation, [46]) The formulation presented here is further inspired by the way [14] quantified their results so that we can link model parameters with their quanitification or orientation bias for long-range connections (see next section). The model’s connectivity profile is broken down into local excitatory w E loc (a local Gaussian bump), lateral excitatory w E lat (Gaussian rings centered at Λ and 2Λ) and inhibitory wI (a broad local Gaussian bump) components, which were plotted in Fig 1G. The use of Gaussian rings was inspired by [16] (see their Fig 16), noting the important features that excitatory connections 1) drop in number at a range Λ/2, 2) have a peak at a range Λ and 3) can extend several mm across cortex. The following details give full definitions of these components and their relative scaling with particular attention to the global balance between excitation and inhibition, which will be controlled by a parameter C. When C = 0 excitation and inhibition are balanced (the area under the 2D radial versions of the blue and red curves in Fig 1G would be equal). In this study C is taken to be negative (net inhibition) whilst remaining close to the balanced condition. We note that when C is larger (or even positive) localized patterns of activity are more likely to destabilize and spread across cortex [41], which is investigated in Reduced inhibition leads to orientation selective activation outside stimulus footprint. The length scale Λ is the mean hypercolumn separation.
Radially symmetric functions for the connectivity components are defined in terms of a radial coordinate r = x 2 + y 2. We define a 2D Gaussian function with spatial decay rate σ:
g(r,σ)=12πσ2e(−r22σ2),(8)
where the pre-factor 1 2 π σ 2 normalizes the area. The number of inhibitory connection, based on a diverse class of inhibitory neurons [16], is assumed to have a Gaussian decay with distance from the origin:
w I ( r ) = g ( r , R W in ) , (9)
with RWin = 0.55Λ (a cross-section of this function is plotted red in Fig 1G). This value gives the qualitative feature that there is more excitation than inhibition at ranges Λ and above. The results are not contingent on the exact value chosen, but varying RWin has a similar effect to varying C (inhibition strength, defined below), which is investigated in Reduced inhibition leads to orientation selective activation outside stimulus footprint. Assuming that there are peaks in the number of excitatory connections ever Λ-distance from the origin up to 2Λ, there are rings of excitation at distances {0, Λ, 2Λ}. We assume that the amplitude of the peaks centered at these distances decay within an exponential envelope
χ ( r , ζ ) = e - r ζ ,
where ζ = 0.625Λ. The exact value chosen is not critical, varying ζ by ±20% does not significantly affect the results. We define a radially shifted 2D Gaussian
h(r,r0,σ)=e(−(r−r0)22σ2), (10)
which describes a ring that is maximal at a radius r = r0 and decays away with spatial scale σ (note that 1 2 π σ 2 h ( r , 0 , R W ex ) = g ( r , R W ex )). The local excitatory component (a Gaussian bump centered at 0) is given by
w E loc ( r ) = h ( r , 0 , R W ex ) , (11)
where RWex is a free parameter. For RWex < 0.1 the number of excitatory connections at Λ/2 would drop to 0, which is not realistic. Further RWex should be less than RWin; for our parameter exploration, we therefore consider a smaller range [0.1,0.4] based on the anatomical constraints introduced later. The long-range excitatory component (Gaussian rings centered at Λ and 2Λ) is given by
w E lat ( r ) = χ ( Λ , ζ ) h ( r , Λ , R W ex ) + χ ( 2 Λ , ζ ) h ( r , 2 Λ , R W ex ) . (12)
The overall profile of excitation w E loc + w E lat is plotted in Fig 1G. Noting the following analytic expression for the zero-mode of the Fourier transform of (10)
H ( 0 , r 0 , σ ) = 2 π σ 2 e - r 0 2 2 σ 2 + π σ r 0 2 π ( 1 + erf r 0 2 σ ) , (13)
where erf is the standard error function, we write the normalisation pre-factor for the combined excitatory components w E loc + w E lat B E = 1 / [ 1 2 π R W ex 2 + H ( 0 , Λ , R W ex ) + H ( 0 , 2 Λ , R W ex ) ] . (14)
We now define the normalized combined excitatory profile as
w E = B E ( w E loc + w E lat ) , (15)
which by design has zero-order Fourier mode of 1. The complete connectivity function is
w ( r ) = P [ w E ( r ) + ( C - 1 ) w I ( r ) ] , (16)
where C is a constant controlling the relative strength of excitation and inhibition. When C = 0 excitation and inhibition are balanced and when C is negative there is net inhibition globally. We introduce a constant P that matches the value of W’s largest Fourier mode with the connectivity used in [41]. This final scaling of the overall connectivity allows us to manipulate any of the connectivity parameters, whilst keeping all non-connectivity parameters constant (e.g. input and threshold function parameters). Failing to do this means that the correct operating region of the model would shift each time, say, RWex was modified. Two constants gex and gin in (3)–(4) are given by
g ex = B E P , and g in = P ( C - 1 ) . (17)
Retinotopic space is mapped to the surface of V1 [47] and different locations show preference for a variety of low-level visual features such as spatial frequency, ocular dominance, orientation and direction of motion [48, 49]. Cortical neurons exhibiting similar preference for low-level features are organized in a columnar fashion [50, 51]. Orientation preference varies incrementally along the cortical surface [52], defining a quasi-periodic organization characterized by linear zones and pinwheels, singularities about which all orientation preferences are present along a circular path [18]. The orientation maps have a quasi-periodic organisation with a regular length scale of around 0.5–1 mm (depending on species and cortical areas), which can be measured as the mean distance between iso-orientation domains. This length scale, which we denote Λ, is reflected as a peak in the Fourier spectrum of the preference map represented as polar argument [53, 46]. Orientation preference can be deduced from single-electrode recordings [54] or with optical imaging of intrinsic signals [18, 10].
The characteristics, artificial generation and biological development of orientation preference maps have been investigated in various modelling studies [55, 56, 46, 57]. [56] showed the importance of long range connections in generating the quasi-periodic repetition of key map features in a canonical pattern forming system with spatially extended complex representation of orientation. [46] showed that when normalized by the regular length scale, orientation preference maps show a constant pinwheel density; see also [58, 59]. [57] explored mechanisms for the stable development using a Hebbian learning method for connections in a two-stage (LGN, V1) model and found an important role for adaptation and normalisation.
In this study we generated realistic maps specifically for our planar model with discrete representation of orientation (in four-sub-populations). The maps were generated specifically for the periodic domain used in our model using a spatial Hebbian-like learning rule operating on the converged model output u i , n fin from a series of localized inputs Ii with random orientations and at random locations. After each simulation the Ji were updated via the following rule:
Ji,n+1=Ji,n+HaIi(1−〈 | Ji,n |⋆G 〉[ 0,1 ])ui,nfin,Ji,n+1=〈 Ji,n+1 〉[ −1,1 ].
Ha is the learning rate, G a smoothing kernel (8) with σ = 0.6Λ, and 〈.〉 rectification on the given interval. The smoothing ensures that regions that already have local structure are modified less than regions without local structure. Learning was initiated from a homogenous initial set of Ji = 0 (where the model produces localized multi-bump states in the stimulated area and sub-population). The learning process converged (e.g. pinwheel density stabilizes) after around 1600 steps as the map takes on structure across the whole domain (after this the term 1 − 〈|Ji,n| ⋆ G〉[0, 1] remains close to 0 across the whole domain, although small changes in the map continue). Final maps (at 6400 steps) were high-pass and low-pass filtered as described in [46]. The maps used in this study are included as part of the simulation code (see Supplemental information S1 Code), thus allowing the results to be independently reproduced. Full details of the method will be the subject of a separate study.
Alternatively, one could use maps obtained experimentally directly with our model, but this would not have any significant effect on the results presented. For the purposes of the present study it is sufficient to show that the maps used in our model show characteristic features of realistic maps including linear zones, pinwheels and the regular length scale Λ. Fig 2A shows the component maps Ji, each associated with a different orientation, that in (1)–(4) modulate inputs with strength βinp and long range excitatory connections with strength βrec. In conjunction, the composite maps combine to give the orientation preference map shown in Fig 2B, which has a regular length scale Λ characterized by a sharp peak in the map’s spectral power curve (Fig 2B).
The effective tuning of lateral connections is computed across ranges of the connectivity parameters RWex (width of peaks in excitation) and βrec (orientation bias of long-range lateral connections). These parameters were chosen as they have a strong effect on the anatomical measure of interest over ranges where the model is well defined. Another choice could be ζ (controlling the decay of peaks in excitation), but this would be redundant with ζ having a similar effect as varying RWex. These computations allow for a direct comparison with the anatomical-data-based model analysis presented in [14], which reported the orientation bias of long-range lateral connections in V1 L2/3. Optical imaging was used to find orientation preference maps and combined with intracellular labeling of lateral projections of pyramidal cells to identify target locations relative to the preference map. The orientation bias was quantified by tuning curves of the orientation preference at axon terminals relative to orientation preference in a region local to the originating cell’s body. Tuning curves were quantified by parameter-fits to a von-Mises distribution (circular normal distribution). Our model was developed to be able to provide a direct point of quantitative comparison with these measures. The distribution is defined on a circular domain and parametrized by a tuning coefficient κ ≥ 0 (0 is untuned, larger is more tightly tuned) and a preferred orientation μ ∈ [0, 180):
f ( x ; μ , κ ) = e κ cos ( x - μ ) 2 π I 0 ( κ ) . (18) I0(κ) is the modified Bessel function of order 0. [14] reported values of κ in the range 0.7–1.2 for population-level tracing of lateral connections.
We perform a similar computation for the connectivity function in our model. The computation of the effective orientation tuning of lateral connections, at one specific map location (Fig 3A(top)), is illustrated in Fig 3A–3E. For the orientation preference map one can compute an orientation tuning curve by counting the number of pixels falling within equally sized orientation bins, as shown in Fig 3A (bottom, circular markers). All orientations are represented with equal probability so the profile is untuned and a best-fit von-Mises distribution (solid curve) has κ ≈ 0 (fit determined using a least-squares minimisation with Matlab’s lsqcurvefit for κ and μ in (18)). A spatial weighting for the pixel count can be introduced in order to find the tuning of orientations in some local region. A 2D Gaussian function (Fig 3B(middle)) was used to give the local weighting shown in Fig 3B(top). In most regions of the map this results in a sharp tuning (note large κ) centered at one specific orientation as shown in Fig 3B(bottom). Introducing the radial profile of excitatory connections from the model as a weighting function (Fig 3C(middle), see (15)) one can compute the effective tuning of the lateral connections. With βrec = 0 the weighting function is radially symmetric, nevertheless there is an expected weak tuning of the weighted connections (Fig 3C(bottom)) due to the structure of the orientation preference map; there is a slight bias toward finding similar orientations at a range Λ (annular ring) relative to the origin (star). Increasing βrec introduces an orientation-specific bias in the weighting profile for the long-range connections (Fig 3D–3E(middle)). This results in locations with similar orientations to the origin being targeted specifically by long-range connections (Fig 3D–3E(top)) and a corresponding increase in the tuning strength κ (Fig 3D–3E(bottom)).
As one might expect the tuning strength of connections (κ) increases monotonically with increasing βrec or decreasing RWex, which allows us to define an operating range for the model before running simulations. Fig 3F shows κ (average value from 50 randomly selected map locations) computed at combinations of RWex ∈ [0.1, 0.4] and βrec ∈ [0, 1]. Map locations close to pinwheels (7/50) identified by having a local tuning (computed as in Fig 3B) with κ < 1 were excluded (this exclusion had a very minor effect). The map shows that κ increases with βrec (as expected) and decreases with RWex. Solid white contours show a band of values for RWex and βrec where the tuning of connections in the model is consistent with the anatomical data (κ ∈ [0.7, 1.2]). These contours are later replotted in Orientation selective activation is restricted to stimulus footprint in the anatomical parameter range and Reduced inhibition leads to orientation selective activation outside stimulus footprint for comparison with model simulations. For RWex much beyond 0.4 the anatomical data cannot be matched as βrec must be <1.
The process of converting the model output from individual simulations into a VSD-like signal, and the method for computing the general and orientation selective activation, is illustrated in Fig 4. For a given oriented simulation, the Ii are weighted toward the specific orientation (Fig 1A). The input is further modulated by Ji if βinp > 0 (Fig 1B) and recurrent connections in each sub-population are modulated by Ji if βrec > 0. The sub-population corresponding to the input orientation responds above threshold in a multi-bump pattern with the location of the bumps determined by Ji, whilst the other sub-populations have a sub-threshold response (Fig 4A). We first transform the sub-population variables ui into a VSD-like signal following a similar method to the one proposed in [38]. The sub-population membrane potentials in Fig 4A are converted into a firing rate, processed through weighted excitatory and inhibitory connectivity profiles, summed across the sub-populations and diffused with a Gaussian profile (representing the attenuation and diffusion of the signal in cortical tissue). The contribution from inhibition is assumed to be in the range 15–20% [60, 61]. The resulting optical imaging signal OI0 in Fig 4B(top left) was computed from the four sub-population responses in Fig 4A. For different orientation inputs, the other OI signals can be computed in a similar fashion (Fig 4B, other panels). The general activation Act is computed as an average across the responses to inputs with four different orientations; the response in Fig 4C was computed as an average of the four responses in Fig 4B. The preference Pref and selectivity Sel (Fig 4D) are computed by transforming the four OI signals into polar coordinates from difference maps (OI0 − OI90 and OI45 − OI135) where the angular coordinate (argument) is the preference and the radial coordinate (magnitude) is the selectivity [62]. Thresholds delineating the general activated area and orientation selective area are defined as a fraction of the average activation or selectivity inside the FFF. The radial profiles (average radial decay) of Act and Sel are characterized by the parameters of best-fit Naka-Rushton functions [63] (a widely-used, smooth, monotonically decaying function (29), which has been used to fit, e.g., contrast response data [64]). Further details and equations of all the processing steps are given below.
Following the method described in [38], to account for the known optical diffusion of the VSD signal (light) in V1 L2/3, we convolved the signal with a Gaussian distribution (8) with σOI = 0.075Λ. Note that the results in this study are not contingent on this specific value, there being little effect of either increasing or decreasing σOI by a factor of 2. The unattenuated imaging signal ui for each sub-population is assumed proportional to the post-synaptic membrane potential. Hence we first computed the dynamics of mean pre-synaptic membrane potential for each sub-population ui as given by (1)–(4), which is converted to a firing rate of the pre-synaptic neurons through S. The postsynatpic population response was then computed via a convolution of the presynaptic firing rate with the connection profile. There is an 85% contribution from excitation and a 15% contribution from inhibition, leading to a weighting for inhibition of pI = 0.177. These are summed across the sub-populations i to give the total unattenuated signal (expression in large parentheses). Finally the optical imaging (OI) signal is computed as a convolution of the total unattenuated signal with a Gaussian:
O I ( x , y ) = ( ∑ i S ( u i ( x , y ) ) ⋆ ( x , y ) [ w E loc ( x , y ) - p I w I ( x , y ) + w E lat ( x , y ) ( 1 + β rec J i ) ] ) ⋆ ( x , y ) g ( x , y , σ OI ) . (19)
This equation converts the model’s state variables ui for an input stimulus with specific orientation (e.g. 0° as in Fig 4A) into a VSD-like signal (e.g. OI0 in Fig 4B, top left).
We are interested to explore how cortical activation spreads over time. One potential issue with the temporal dynamics in the model, and (19), is the assumption that activity generated through long range lateral connections (w E lat) propagate instantaneously. Although the model converges to the correct final state, the transient dynamics may not be captured exactly. To solve this, whilst avoiding the introduction of say delay terms in (1)–(4), we assume that there is a slower timescale τlat = 240 ms for the portion of the OI-signal generated through w E lat:
O I ( x , y , t ) = ( ∑ i S ( u i ( x , y , t ) ) ⋆ ( x , y ) [ w E loc ( x , y ) - p I w I ( x , y ) + ( 1 - e - t / τ lat ) w E lat ( x , y ) ( 1 + β rec J i ) ] ) ⋆ ( x , y ) g ( x , y , σ OI ) , (20)
where t is the time after stimulus onset. The introduction of τlat is done at the post-processing stage only and its value was chosen to match data from [24].
For four sequential simulations, each with an input with different orientation, the OI signal can be computed (Fig 4B). The general activation is the average of these signals:
Act ( x , y , t ) = 1 4 ∑ j O I j ( x , y , t ) , j = 0 , 45 , 90 , 135 , (21)
as shown in Fig 4C.
Before computing the preference Pref and selectivity Sel, the VSD signals are normalized by a scale factor that accounts for differences in the maximum value over (x, y) across the four simulations with different orientations. Two difference maps between the normalized VSD signals from simulations with orthogonal inputs are computed:
D 1 ( x , y , t ) = O I 0 ( x , y , t ) − O I 90 ( x , y , t ) , (22) D 2 ( x , y , t ) = O I 45 ( x , y , t ) − O I 135 ( x , y , t ) . (23)
The orientation preference of the activation is given by
Pref ( x , y , t ) = Arctan ( D 1 ( x , y , t ) , D 2 ( x , y , t ) ) , (24)
where Arctan is the four quadrant inverse tangent, and the selectivity strength is given by
Sel ( x , y , t ) = D 1 ( x , y , t ) 2 + D 2 ( x , y , t ) 2 . (25)
In the results section, all plots of Sel(x, y, t) and Act(x, y, t) are scaled by 1.1× their values at the final time point tfinal in the simulation, thus showing the time-evolution relative to the final state. The thresholds contours Tact and Tsel for activation and selectivity (and corresponding areas Aact and Asel) are set as a fraction of the mean activation and selectivity within the FFF at the final time point,
Act ¯ FF = mean r < r FF ( Act ( x , y , t final ) ) , (26)
Sel ¯ FF = mean r < r FF ( Sel ( x , y , t final ) ) , (27)
where rFF is the FFF boundary. The thresholds are given by
T a c t= η Act Act ¯ FF , T s e l= η Sel Sel ¯ FF , (28)
where ηAct = 0.2 and ηSel = 0.5. These thresholds were chosen ad hoc. In order to ensure the results obtained weren’t contingent only on these choices, we further characterized the radial decay rates of the general and orientation selective activation.
We also look at radial profiles of the general and selective activation patterns (radial decay of Act and Sel). Radial profiles Act(r, t) and Sel(r, t) are computed by re-meshing Act(x, y, t) and Sel(x, y, t) on radial coordinate system (r, θ) centered at the stimulus center and averaging in the angular coordinate θ. The decreasing Naka-Rushton function NR used here decays to zero as r increases:
N R ( r ) = R max ( 1 - r n r n + r 50 n ) , (29)
where the exponent n > 0 describes the steepness, the maximal value is Rmax and the half-max r-value is r50 > 0. Best-fit Naka-Rushton functions were determined by minimising, for example the least-squares distance between Act(r, tfinal) and NR(r) varying the parameters n, Rmax and r50 using Matlab’s lsqcurvefit function.
The spatio-temporal dynamics are investigated for a specific case with inputs modulated by the orientation preference map (βinp = 0.25) but with radially symmetric lateral connections (βrec = 0, no orientation bias in lateral connections). The dynamics produced by the model will be illustrated for two values of the width of the peaks of excitatory connections RWex. When RWex is small, excitatory connections cluster on tightly on rings at distances Λ and 2Λ away from the origin. When RWex is larger the clustering at these specific ranges is more diffuse (Fig 1G).
Simulations were run, sequentially for four different orientations, with a radial input centered at a specific location in the orientation preference map (Fig 5A). Fig 5B and 5E (top) show that the general (averaged) activation spreads outward from the center of the stimulated region (note that simulations are computed on a domain around 3 times larger than the window shown here). Only some portion of this general spread of activation is orientation selective as shown in the bottom panels (area inside white contour is significantly smaller than area inside grey contour). When, RWex is small the selective activation extends outside the FFF (Fig 5B(bottom,right)), and when RWex is larger it is confined to the FFF (Fig 5E(bottom,right)). The temporal dynamics of the general and orientation selective areas (Fig 5C and 5F) shows that the rate of the area increase slows down after roughly the first 100 ms. Nevertheless, when RWex is small the spread of activation persists after 550 ms as shown by the blue/green curves still increasing in Fig 5C (eventually converges at ∼ 700 ms). When RWex is larger the orientation selective activation converges after around 200 ms although the general activation continues to increase slowly (Fig 5F). This latter behavior is compatible with the observations made using VSDI in [24]. The radial profile of the general and selective activation shows the decay with distance from the center of the stimulated region (Fig 5D and 5G). When RWex is small, the profile of the general and selective activation exhibits a similar decay profile as marked by similar values of the exponent n in a best-fit Naka-Rushton function for the two profiles (Fig 5D). For a larger RWex the exponent for the orientation selective activation is substantially larger (nSel = 14.47), indicating a steep transition from high to low selectivity, than for the general activation (Fig 5G).
Even with lateral connections not modulated by the orientation preference map (no orientation bias in lateral connections), orientation selective activation can be generated outside the FFF of the stimulus due to convergent excitation generated from regions directly stimulated within the FFF. Indeed, activated locations with the same preference mutually excite each other at a range Λ. These activated regions can further generate overlapping excitation at Λ-equidistant points, either inside the FFF, or potentially outside the FFF. Orientation selective activation generated in this way, outside the FFF, does not necessarily agree with the orientation preference map (Fig 5B, bottom right, region inside the white contour but not the yellow). This occurs if the range of peaks in excitation are highly specific (small RWex) as in Fig 5B–5D. These findings illustrate that with RWex excessively small, the ringed connectivity can lead to undesired behaviour that is inconsistent with activation observed in experiments. We note that later in section Reduced inhibition leads to orientation selective activation outside stimulus footprint we find that spurious activation (not agreeing with the preference map) can be generated by another mechanism (destabilization of activity). When the peaks in excitation are broader (RWex > 0.2), the orientation selective activation decays quickly at the border of the stimulated region and no such miss-tuned activation (as generated by convergent excitation) is observed outside the FFF.
We have seen how our modelling results allow us to distinguish between patterns of activation that are consistent with imaging studies in terms of 1) the area and range of general (grey) and selective (white) activation relative to the FFF (red) in Fig 5B and 5E(bottom); 2) whether activation reflects the correct orientation with respect to the underlying preference map (difference between yellow and white contours in Fig 5B and 5E(bottom,right); and 3) the rate of decay of general and selective activation in Fig 5D and 5G as quantified by n (larger is a steeper decay). The simulation shown in Fig 5E–5G is consistent with [23] and [24] because 1) in E(bottom,right) the grey contour extends much further than the white, 2) in E(bottom,right) the yellow and white contours agree closely, 3) in G the exponent n is larger for the selective activation (blue) than for the general activation (green). Further, this steep decay of selectivity is consistent with there being a plateau of highly selective activation inside the FFF that transitions sharply to low selectivity outside the FFF.
The effect of introducing an orientation bias to the long-range lateral connections (βrec > 0) is shown in Fig 6. For another location in the orientation preference map (Fig 6A) the orientation selective activation is shown for different values of βrec in Fig 6B–6D with the radial profile shown in F–I. The parameter values for panels Fig 6B and 6F are the same as those for Fig 5E–5G (RWex = 0.25 and βrec = 0), the only difference is the location in the orientation preference map. These two examples show a qualitatively similar spatial profile, and in general, the resulting spatio-temporal dynamics are independent of the location in the orientation preference map (five locations were randomly chosen for the results in this paper). Across the five locations there is a large range of nSel (between 6 and 15) when RWex = 0.25 and βrec = 0. However, this variability in nSel across locations is much less (between 5 and 7) when βrec is increased to say 0.5. As βrec is increased, more orientation selective activation is observed increasing from small isolated patches outside the FFF at intermediate values (Fig 6C–6D) to many larger for βrec = 0.9 (Fig 6E). In contrast with the orientation selective activation outside the FFF observed in Fig 5B(bottom), which does not necessarily reflect the local orientation from the preference map, the activation outside the FFF in Fig 6E reflects the preference map (white and yellow contours agree) as it is generated by orientation-biased long-range connections. The increased selective activation outside of the FFF coincides with a reduced slope (smaller exponent n) in the radial decay of selective activation shown in Fig 6F–6I.
Again, the results shown here for βrec in the range [0, 0.5] are consistent with imaging studies [23, 24], as the region of selective activation is predominantly confined to the FFF (limited small selective regions outside the red circle), the preference agrees with the underlying map (yellow and white contours are very similar) and the exponent n is larger for the selective activation.
The measures of the orientation selective component of the lateral spread of activation as studied in individual simulations in Figs 5 and 6 are now quantified over a range of RWex and βrec. The ranges of these parameters consistent with anatomical data is indicated in Figs 7–9 (between the white iso-κ contours reproduced from Fig 3). The agreement of model simulations with qualitative features from imaging data will be assessed in this and the following section. Fig 7A shows a map of normalized area (relative to the FFF area) of the selective activation, where each value is an average from simulations at 5 randomly selected map locations. Within the red contour (darker regions) the selective area is roughly equal to or smaller than the FFF of the stimulated region (e.g. Fig 7E, top). Outside this region the selective region is larger than the FFF (e.g. Fig 7D, top). Fig 7B shows a greyscale map of the proportion of the selective area with the correct orientation with respect to the underlying orientation preference map. Within the white contour (darker regions) the majority of the selective region has the correct orientation. For small RWex or small βrec a significant proportion (more than 15%) of the selective activation is spurious (not in agreement with the preference map). Fig 7C shows a map of the ratio of the slopes (Naka-Rushton exponent) of the selective activation nSel and the general activation nAct. When this ratio nSel/nAct is large there is steeper radial decay of the selective activation (relative to the general activation) and a sharp transition to non-selectivity at the border of the FFF. Note that nAct and nSel are independent of the choice of ηSel and ηAct defined in (28).
In each case (Fig 7A–7C) there is a significant overlap between the regions with orientation selectivity confined to the FFF, with this selective activation having the correct orientation, with a characteristically steep decay of this region, and the anatomical range for the lateral connections in the model. In the green triangular region (overlayed in Fig 7A–7C), all of these characteristics are satisfied. We highlight that the operating range matching anatomy and functional data covers more than 20% of the (conservative) permissible range of RWex (0.1 < RWex < 0.55 = RWin).
The extent of the operating region for the model depends on values of the thresholds used in Fig 7A–7C. A value of 1.05 for normalized selective area is consistent with observations from [24], that the selective area is close to or marginally larger than the FFF area. A value of 85% for the proportion correct orientation was chosen heuristically, for smaller values isolated spurious selectivity occurs outside the FFF, e.g. compare Fig 7E and 7F. The value nSel/nAct = 1.3, also chosen heuristically, gives a sharp drop-off in the selectivity close to the FFF border. Reducing the threshold on the selective area, or the threshold on nSel/nAct, by 10% would give a reduced by still existing operating region for the model, but increasing the proportion correct orientation threshold much beyond 85% rapidly reduces the operating region. Fig 8 illustrates the dependence of the operating region limits on each of the thresholds. The specific choice ηSel = 0.5 could also affect the size of the model’s operating region for the normalised selective area. For example, increasing ηSel could potentially increase the operating region for a fixed value of the threshold on the normalized selective area. However, the constraint on the ratio nSel/nAct (also providing an upper bound on the green region) would be unaffected as it is independent of the choice of ηsel.
Although an anatomical bias towards iso-orientation in the model’s connectivity profile was introduced through βrec, the bias is not necessarily evident at the population functional level (in the patterns of activity observed in simulations). This may originate from the non-linearities in the functional expression that combines excitatory and inhibitory activation. To further explore the role of the relative balance between E/I in this functional expression, we manipulate the strength of inhibition in the model (controlled by C in (16)). Recall that when C = 0 there is global E/I balance and for the reduced inhibition case here we reduce C toward 0 (from its standard value C = −0.4 to C = −0.2). Fig 9 (computed in the same way as Fig 7A) shows that with reduced inhibition, the region of the (RWex, βrec)-plane with orientation selective activation confined to the FFF footprint (inside the red contour) is smaller. In general, maintaining the same values of other parameters, but reducing inhibition leads to a wider spread of orientation selective activation. Compare Fig 9B (bottom) where the stable pattern of activation extends outside the FFF with Fig 7B (bottom) where the activation is confined within the FFF. Furthermore, with strong orientation bias in the model’s lateral connections (large βrec), the spread of activation can destabilize and continue indefinitely, see Fig 9B (top). This activation outside the stimulated region has spurious orientation preference and is due to destabilized activity that spreads far beyond the boundary of the stimulated region. Note that this a distinct mechanism from the convergent excitation that generates activation with incorrect orientation when RWex is too small (see Fig 5B). For the case illustrated in Fig 9B (top), of the four stimulus presentations with I0, I45, I90 and I135, the condition I0 resulted in an continual spread of activation down and to the right of the FFF (blue-green regions), I45 above the FFF (purple regions), whilst I90 and I135 remained constrained to the FFF. These different responses across different orientations arise from a local imbalance in the preference map, which becomes exaggerated when input drives the model close to spatial instability, as is the case with reduced inhibition. With stronger inhibition, small imbalances in the spread of activity can occur across different orientations, but these don’t impact the validity of tuning of the response (e.g. see the slightly larger response to 135° in Fig 4B). We note that the destabilization is not contingent on there being local imbalance in the preference map, in general, moving in parameter space to the top left of Fig 7A, the unbounded activation would occur for all input orientations. In general, with reduced inhibition, there is no more overlap in the (RWex, βrec)-plane between the anatomical operating range of the model (between the white contours) and the range where orientation selective activation is confined to the FFF.
With increased inhibition strength (C = −0.6) the operating region for the model increases slightly shifting to lower RWex and extending to larger βrec values (Fig 9C). This illustrates how inhibition constrains the spread of activation to the FFF, even with larger orientation bias of lateral connections. Fixing RWex = 0.25, the operating region in terms of C and βrec is illustrated in Fig 9D, with the largest extent in βrec occurring at C = −0.4. For large values of C the model responses become predominantly feed-forward (input driven) and the profile of decay for the general and selective activation becomes similar (nSel/nAct tends to 1). As discussed above, without sufficient balance from inhibition activity can spread unbounded across cortex (top left region of Fig 9D).
This study reported on a neural field model of L2/3 of primary visual cortex, investigating the dynamics of stimulus driven, localized patterns of cortical activity. The aim was to understand the relationship between the anatomical properties of lateral connections and the functional propagation of cortical activation. The model featured a coarse-grained description of average membrane potential in 2D, with discrete sub-populations for different orientations. Orientation was encoded using a realistic orientation preference map that modulates inputs from LGN via L4 and/or the lateral connections within L2/3 sub-populations. Anatomical data quantifying the orientation bias of lateral connections was used to constrain model parameters [14].
VSDI experiments in V1 have shown that local, oriented visual stimuli elicit stable orientation-selective activation within the retinotopic footprint [24]. Recently two studies confirmed these findings, one using optogenetic stimulation [27] and another anatomofunctional study [15]. The cortical activation dynamically extends far beyond the retinotopic footprint, but the peripheral spread stays non-selective, which could be surprising given anatomical studies showing the orientation specificity of long-range connections [7,10,8,11,9]. However, this result could actually be expected given that (i) the quantified orientation bias is small (ii) decreases with distance [14] and (iii) holds only for a sub-part the functionally activated hyper-column (L4, pinwheel and interneuron cells do not show this bias [12, 65], but see [14, 15] for the pinwheel dependence).
Based on these anatomical studies, we designed a new connectivity function, flexibly parametrized to investigate clustering of connections, their orientation bias and balance between excitation and inhibition. We adopted the non-orientation specific nature of local excitatory connections [14] and inhibitory connections [16]. Taking motivation from [16], longer-range excitatory connections are proposed here to, although decaying with distance, form in rings at multiples of Λ (hypercolumn separation). This allows for the following important features to be captured: that excitatory connections 1) drop in number at a range Λ/2, 2) have a peak at a range Λ and 3) can extend several mm across cortex. Two parameters were tuned to agree with the available data from [14], the width of peaks in number of excitatory connections and their orientation bias (note we did not include an explicit representation of L4 nor a special connectivity rule at pinwheel locations). Similar results and conclusions would be expected if other parameters had been varied, e.g. those controlling the amplitude of excitatory peaks, or the width of inhibition.
We found a significant overlap between the anatomically relevant parameter range and patterns of cortical activation consistent with imaging experiments. Hence the imaging results can be reconciled with the reported level of orientation bias from anatomical studies. Specifically, [24] found a sharp decay of orientation selective activation at the stimulus retinotopic footprint border, resulting in peripheral activation that was not orientation selective. Our results demonstrate that this sharp decay is contingent on three factors: the diffuse clustering of long-range connections, the intermediate range (consistent with anatomy) of their orientation bias and the sufficient balance between excitation and inhibition. We note that although long-range connections are orientation biased, the recruitment at the target of a local network without orientation bias could result in a general non-orientation specific activation [27]. Furthermore, if the orientation bias of lateral connections is excessively strong, or if inhibition is particularly weak, the network operates close to an instability leading to unbounded cortical activation (a testable prediction, see below).
Our modelling results predict a qualitative change in the spread of orientation selective activation for localized, oriented inputs if global inhibition strength is reduced. Under those conditions, we should observe selective activation outside of the retinotopic footprint, however, the exact orientation preference may not be preserved. Activation with spurious orientation preference is generated via spatial destablization of the localized activation generated by the input. These predictions could be addressed by manipulating the inhibitory cells pharmacologically [66–68] or optogenetically [27]. To confirm our model’s prediction though, one may have to go close to pathological epileptic conditions [69, 70]. Other approaches could be used by simply comparing anesthetized and awake conditions. Recently, [71] have indeed shown this to be a valuable approach for investigating E/I balance in the integrative properties of the cortical populations. Comparison of the dynamics of propagation of orientation-selective activity in awake or anesthetized conditions could hence provide the appropriate non-pathological test to probe our model’s prediction.
Spurious activation can also be generated by another mechanism. Activated regions inside the stimulated region can generate excitation at an equidistant range outside the stimulated region. The specific range is associated with the peak in the radial excitatory profile. This happens in the model when peaks in excitation are highly specific, in a parameter range that was ruled out from the operating regime of the model. This could be seen as an undesired consequence of the model’s design. However, the fact that, under particular circumstances, the preferred orientation of the horizontal propagation may be at odds with the underlying orientation preference map could unravel some new unexpected computational capacities of the horizontal network, which may be present in visual areas beyond V1/area 17. For instance, the ability to link information of position and orientation for non co-circular filters that could be of importance for processing objects with sharp angles. Inline with this hypothesis, [24] showed that the spread of orientation selective activity is not fixed but can increase when increasing spatial summation.
The transition to an unbounded spread across the network, as contingent on a spatial modulation parameter (like βrec controlling orientation bias in our model), has been observed in a one dimensional theoretical study of the neural field equation with purely excitatory connectivity [72]. A more common choice of connectivity function (e.g. difference of Gaussians) features a broader footprint for inhibition than for excitation [36, 37]. In our model excitation extends further than inhibition with additional peaks in excitation away from the origin [44]. The distance between excitation peaks fixes a regular length scale that stabilizes multi-bump patterns. In [41], without an explicit representation of orientation, localized inputs were shown to produce multiple bumps within a stimulated region. It was proposed that the connectivity’s excitatory peak separation could be equated with Λ (hypercolumn separation) and the spatial phase of multi-bump patterns governed by an orientation preference map. In the present work, we have shown that the localized patterns of orientation selective activation observed in [23] and [24] are well described by a superposition of these intrinsic multi-bump states.
Local models of orientation selectivity [32, 33] have been more widely studied than spatial models capturing interactions across columns. Spatial models of map development [55, 73, 46, 74, 57] do not focus on dynamics on sub-second timescales. Indeed, [74] does not consider dynamics or long-range connections within V1; see also commentaries in [75, 76]. The self-organizing map (SOM) model in [57] included, but did not show, the dynamics of stimuli responses. The focus of the study was on the long-timescale dynamics of map development, however, this class of model could be another candidate to investigate functional activation in the future. Integrate-and-fire neuron models of earlier thalamic and cortical processing stages proved successful for capturing the dynamics of orientation selectivity and tuning, but were restricted to a small patch of cortex without considering superficial layers accessible to imaging [77, 78]. [79] explored the role of patchy long-range connections in cortex in a general setting, whilst [80] looked at larger regions of cortex in an integrate-and-fire network, reporting complex spatio-temporal dynamics, but did not model orientation.
The neural field framework used here is an ideal level of description for comparison with VSDI recordings [38] that allowed us to simulate a large spatial domain extending beyond the local region of interest, thus avoiding issues with boundary effects. The four sub-population implementation used here was chosen because it allows for efficient computation of connectivity integrals using convolutions. In turn, this allowed for a comprehensive multi-parameter investigation of the model’s dynamics. Another choice would be to consider a single mixed population, where connection weights depend directly on orientation difference as has been considered in other spatial models [73, 81], including those posed in an elastic net framework [82, 83]. This would have the advantage of allowing direct implementation of the heterogeneity of long-range connections, however, the increased computational burden may not allow such a broad parameter investigation.
[27] used optogenetic stimulation in combination with visual inputs and observed a non-orientation specific linear addition of the two inputs, which could be explored in our modelling framework. The model could also be used to investigate selective recruitment and spatial summation in regions between localized oriented stimuli. Two further properties were described in [24] that can be further investigated. First, as demonstrated with intracellular recordings, the lack of orientation selectivity observed at the mesoscopic level (VSDI) was due to a diversity of microscopic rules: some cells received untuned presynaptic input, but others a tuned presynaptic input with preferred orientation either agreeing with or different from the recorded cell. Such diversity in local cellular rules is to be linked to our observations that, with narrow excitatory peak widths or strong orientation bias (Fig 7B) or reduced inhibition (Fig 8), our model can easily produce spurious selective propagation. Adding diversity in the connectivity rules can thus easily lead to natural diversity in the tuning of the propagation. Second, increasing spatial summation increases the slope of selectivity decay at the stimulus boundary, whilst selective propagation reaches further across cortex; the model could also account for this property. More generally, the model could be used to make predictions to decipher the selective functional connectivity rules that link position and orientation in cortical space. For example, the model could be extended to differentiate inhibitory cell sub-classes as reported in [16]. As such it could generate functional predictions on e.g. the role of long-range basket cell connections that preferentially target cross orientations.
A lumped description of inhibition and excitation was used here with local cross-orientation inhibition. By separating out inhibitory and excitatory populations, one could consider different profiles for cross-orientation interactions including their spatial profile. This may change the model’s behavior and would be important if it is extended to consider inputs with multiple orientations that overlap spatially. Earlier stages of cortical and thalamic processing could be incorporated with differing properties of orientation tuning and bias of connections across V1 layers [84]. The four sub-population implementation used here can be viewed as a coarse discretization from a continuous representation of orientation, which could be considered in future work. For example, a recent paper studied spatio-temporal patterns with continuous orientation, but only on a 1D spatial domain [85]. Theoretical work characterising localized states in 2D space plus orientation would be an important first step. An extended feature space including spatial frequency (SF) could be used to investigate lateral connections in light of recent work showing interesting interactions between orientation and SF maps [86].
Our model of orientation selectivity in V1 is the first of its kind, capturing the spatio-temporal dynamical spread of localized patterns of activation with a representation of orientation. Our study addressed an apparent conflict between the orientation bias of lateral excitatory connections in L2/3 of V1, as characterized in anatomical studies, and imaging studies on the lateral propagation of cortical activity for localized oriented visual stimuli. Simulations with the neural field model illustrated that observed levels of orientation bias in anatomical studies actually predict long-range activation outside of the retinotopic footprint of the stimulus, but with a sharply decaying profile of orientation selectivity, as observed in imaging studies. Without this sharp decay, which might occur with excessive orientation bias or diminished inhibition strength, the network could destabilize leading to unbounded spread of cortical activation.
|
10.1371/journal.ppat.1002272 | IRAK-2 Regulates IL-1-Mediated Pathogenic Th17 Cell Development in Helminthic Infection | Infection with the trematode parasite Schistosoma mansoni results in distinct heterogeneity of disease severity both in humans and in mice. In the experimental mouse model, severe disease is characterized by pronounced hepatic egg-induced granulomatous inflammation mediated by CD4 Th17 cells, whereas mild disease is associated with reduced hepatic inflammation in a Th2-skewed cytokine environment. Even though the host’s genetic background significantly impacts the clinical outcome of schistosomiasis, specific gene(s) that contribute to disease severity remain elusive. We investigated the schistosome infection in wild-derived mice, which possess a more diverse gene pool than classically inbred mouse strains and thus makes them more likely to reveal novel mechanisms of immune regulation. We now show that inbred wild-derived MOLF mice develop severe hepatic inflammation with high levels of IL-17. Congenic mice with a MOLF locus in chromosome 6, designated Why1, revealed high pathology and enabled the identification of Irak2 as the pathogenic gene. Although IRAK-2 is classically associated with TLR signaling, adoptive transfer of CD4 T cells revealed that IRAK-2 mediates pathology in a CD4 T cell specific manner by promoting Th17 cell development through enhancement of IL-1β-induced activation of transcription factors RORγt and BATF. The use of wild-derived mice unravels IRAK-2 as a novel regulator of IL-1-induced pathogenic Th17 cells in schistosomiasis, which likely has wide-ranging implications for other chronic inflammatory and autoimmune diseases.
| Schistosomes are trematode helminths that cause widespread disease in vertebrates and are responsible for over 200 million human infections worldwide. The species Schistosoma mansoni causes a hepatic granulomatous inflammatory and fibrosing reaction against tissue trapped parasite eggs that varies greatly in humans and among mouse strains, implying that the host’s genetic background plays a critical role in determining disease severity. Although exacerbated hepatic inflammation is known to be associated with an increase in CD4 Th17 cells, specific genes conducive to high pathology are unknown. In this study we used genetically diverse inbred wild-derived mice and found that their natural severe immunopathology and high IL-17 levels are regulated by the interleukin-1 (IL-1) receptor-associated kinase-like 2 (IRAK-2). We demonstrate that T cell intrinsic IRAK-2 affects disease severity by enhancing the development of Th17 cells, which results from an increased sensitivity to IL-1β induced activation of the lineage-specific transcription factors RORγt and BATF. Our findings thus identify IRAK-2 as a single regulator of pathogenic Th17 cell development in murine schistosomiasis and reveal a novel mechanism that is likely to operate in other chronic inflammatory and autoimmune diseases.
| The genetic analysis of complex traits has been critical to our understanding of the molecular mechanisms that underlie disease processes. Quantitative trait loci (QTL) are genomic intervals, whose variation is responsible for the majority of genetic diversity in human disease susceptibility and severity. As a model of human genetics, classical inbred mouse strains have been instrumental in identifying QTL. Murine schistosomiasis represents an extensively characterized model of a major human infectious disease that shares similar mechanistic features with many autoimmune and chronic inflammatory diseases [1], [2]. Although several QTL underlying pathology in schistosomiasis have been identified to date, mouse genetic studies have not entirely recapitulated the genetic complexity that is likely to determine the disease course in humans. One reason for this is the relatively limited genetic diversity among classically inbred strains. These mice are derived from a restricted number of founder animals predominantly within the Mus mus domesticus subspecies and therefore do not reach the level of diversity observed in humans [3], [4]. We reasoned that this limited diversity was a major problem that has made it difficult to identify genes that underlie even well defined traits, leaving a compelling need for new models of genetic analysis.
Wild-derived inbred mice diverged from a common ancestor with classical strains more than one million years ago. As a result of this early divergence, many of the wild-derived strains have large genomic regions originating from the subspecies M. m. musculus and M. m. castaneus[3], [4], which provides them with a unique and more genetically diverse gene pool compared with classically inbred strains. The genetic diversity of wild-derived mice resembles that seen in humans, which makes them more suitable for the analysis of complex traits, such as host-pathogen interactions. Furthermore, novel phenotypes identified in wild-derived mice are likely to have increased biological relevance given that they have arisen in an evolutionarily driven context[5]. Wild-derived mice have proved useful as genetic models in identifying novel phenotypic variants in studies exploring host responses to infection with pathogens, such as Salmonella typhimurium[6], as well as identifying several loci that confer resistance to TNFα induced toxic shock[7].
The main pathology in murine schistosomiasis consists of a granulomatous inflammatory and fibrosing reaction in the liver and intestine against tissue trapped parasite eggs, which is a host adaptive immune response mediated by CD4 T cells specific for schistosome egg antigens (SEA)[1], [8], [9]. Following infection with S. mansoni, most humans develop mild, “intestinal schistosomiasis”, however 5-10% develop a severe inflammatory and fibrosing response, which leads to a potentially lethal form of the disease, “hepatosplenic schistosomiasis”[10]. This variation also exists in a mouse model of schistosomiasis, where CBA mice develop pronounced hepatic granulomatous inflammation, while C57BL/6 (BL/6) mice develop significantly smaller granulomas with milder hepatic inflammation[11], [12].
Th17 cells represent a unique lineage of T cells that act as potent proinflammatory mediators and have been shown to play a significant role in a number of inflammatory and autoimmune diseases, such as experimental autoimmune encephalomyelitis (EAE), collagen-induced arthritis, psoriasis and inflammatory bowel disease[13], [14], [15], [16]. Th17 cells are characterized by their expression of the transcription factor RORγt[17] as well as by their requirement of IL-6, TGF-β, IL-23 and IL-21 to differentiate and expand[18], [19], [20], [21], [22], [23]. IL-1β is also of particular importance in Th17 cell differentiation and pathogenesis [24], [25], [26], [27]. In schistosomiasis, IL-23 and IL-1β are necessary for IL-17 production in response to parasite eggs and severe immunopathology correlates with increased production of IL-17[28], [29], [30]. Furthermore, in vivo neutralization of IL-17 significantly reduces the immunopathology[12].
In an attempt to identify novel mechanisms that govern severe disease, we assessed the schistosome infection in wild-derived inbred mice of the MOLF strain. We previously have shown that in MOLF mice, TLR ligation in macrophages in vitro leads to significantly higher transcription of proinflammatory cytokines than in classically inbred BL/6 mice [31]. To examine if their bias towards a proinflammatory response also occurs in an in vivo infection model, we infected MOLF mice with S. mansoni and found that they develop severe liver immunopathology associated with a high Th17 response. Although we anticipated identifying novel factors that regulate T cell function via APC responses, we were surprised to find a T cell-intrinsic regulator of IL-17 production. We now demonstrate that a single gene, Irak2, is capable of controlling severe pathology in murine schistosomiasis. Furthermore, we provide evidence of a novel role for IRAK-2 in IL-1β-mediated pathogenic Th17 cell development. Our study also demonstrates the utility of wild-derived mice as a model to identify novel gene networks and further refine our understanding of immune signaling pathways.
Seven weeks after infection with 85 cercariae of S. mansoni, MOLF mice were debilitated, as defined by hunched posture, reduced activity and scruffy appearance, and exhibited significantly enhanced hepatic egg-induced immunopathology when compared with BL/6 mice, with granulomatous lesions in some instances larger than those seen in the high-pathology control CBA mice (Fig. 1A). Individual granulomas from MOLF mice consisted of significantly larger perioval aggregates of macrophages/histiocytes and lymphocytes admixed with eosinophils and neutrophils, with a greater tendency to infiltrate the surrounding hepatic parenchyma than in BL/6 mice (Fig. 1B). However, the number of schistosome eggs present in the livers did not significantly differ between the mouse groups, indicating that the parasite load did not correlate with the extent of pathology (Fig. 1C). Analysis of antigen specific cytokine production by schistosome egg antigen (SEA)-stimulated draining mesenteric lymph node (MLN) cells (MLNC) from infected mice revealed that MOLF mice produced strikingly higher amounts of IL-17 than BL/6 and even CBA mice (Fig. 1D). There also was a significant, but less pronounced, increase in IFN-γ, IL-6 and TNFα (Fig. 1E–G). However, there were no significant differences in IL-4, IL-5 or IL-10 between BL/6, CBA and MOLF mice (Fig. 1H–J).
Cytokine production in MLNC typically correlates with that produced in the affected liver. In order to confirm this in MOLF mice, we isolated granuloma cells (GC) and analyzed their specific response to SEA. Similar to the MLNC, MOLF GC produced very high amounts of IL-17 compared with both BL/6 and CBA mice (Fig. 1K). There also was higher IFN-γ, IL-6 and TNFα production in MOLF vs. BL/6 mice, but only in the case of TNFα was the difference statistically significant (Fig. 1L–N). There were no significant differences in IL-4, IL-5 or IL-10 (data not shown). Analysis of cytokines involved in the development of Th17 cells revealed that IL-1β, as well as IL-23p19 and IL-12p40, the subunits that make up IL-23, were expressed at much higher levels in the livers of MOLF mice than BL/6 and CBA mice (Fig. 1O–Q), whereas the difference in the IL-12-specific subunit IL-12p35 was less striking (Fig. 1R). These results demonstrate that wild-derived MOLF mice produce exceedingly high levels of Th17-related cytokines, suggesting a potentially novel mechanism of severe immunopathology.
We previously mapped the IL-6 hyper-responsiveness of MOLF macrophages to TLR stimulation to a dominant locus on chromosome 6, designated Why1[32]. Since MOLF mice reacted to schistosome infection with an overwhelmingly proinflammatory response, we postulated that the Why1 locus may also play a role in this phenotype in the context of live infection. To assess the effect of the Why1 locus directly, we used congenic mice (Why1 mice), which contain the MOLF allele of the Why1 locus on a BL/6 background. After a 7-week schistosome infection, Why1 mice appeared debilitated, as defined above, and displayed a significant increase in hepatic granulomatous inflammation compared with BL/6 mice, although not to the same extent as MOLF mice (Fig. 2A), with the differences not attributable to dissimilar egg burdens (Fig. 2B). There was significantly greater IL-17, IFN-γ, IL-6 and TNFα production by SEA-stimulated MLNC in Why1 vs. BL/6 mice (Fig. 2C–F), although IL-4, IL-5 or IL-10 were not significantly different (Fig. 2G–I). Expression of IL-1β and IL-23p19 (Fig. 2J,K), but not of IL-12p40 or IL-12p35 (Fig. 2L,M), was also higher in Why1 vs. BL/6 mice. These findings demonstrate that Why1 mice largely recapitulate the pathology and IL-17 secretion seen in MOLF mice, and identify Why1 as a key locus associated with increased egg-induced immunopathology and Th17 cell development.
The Why1 locus contains >200 possible causal genes that could underlie pathology in a complex trait such as the response to schistosome infection. We therefore sought to reduce the number of possible candidate genes by further defining the phenotype of Why1 mice. Based on previous mapping of the Why1 locus in macrophages[31], [32], we hypothesized that severe disease and increased IL-17 production were induced by innate immune cells. To this effect, in order to avoid bias towards any one particular APC type, we used an in vitro system involving bulk naïve splenic APC together with CD4 T cells isolated from MLN of infected mice. SEA-stimulated Why1 APC-CD4 T cell cocultures produced markedly higher levels of IL-17 than BL/6 controls (Fig. 3A). However, surprisingly, Why1 T cells in combination with BL/6 APC resulted in higher IL-17 production than Why1 APC in combination with BL/6 T cells, which did not significantly differ from the IL-17 produced by the all-BL/6 coculture; moreover, Why1 T cells in combination with either Why1 or BL/6 APC resulted in high IL-17 production (Fig. 3A). These findings suggest that CD4 T cells play a decisive role in dictating the levels of IL-17; however, since the Why1 T cells were isolated from infected Why1 mice, an influence of Why1 APC on impending Th17 cell development could not be excluded. We therefore adoptively transferred naïve splenic CD4 T cells from uninfected Why1 mice to allow antigen specific Th17 cell differentiation to take place in vivo in the absence of Why1 APC. As shown in Fig. 3B, transfer of Why1 T cells caused a sharp increase in granulomatous inflammation in infected BL/6 recipients, whereas a similar transfer of BL/6 T cells had no effect. Furthermore, SEA stimulated bulk MLNC (Fig. 3C), or MLN CD4 T cells (Fig. 3D), from infected BL/6 mice recipients of Why1 T cells produced significantly more IL-17 than those receiving BL/6 T cells. There was also an increase in IFN-γ, although to a lesser extent than IL-17 (Fig. 3E,F). These results indicate that the Why1 locus controls severe immunopathology and enhances Th17 cell development via a CD4 T cell-mediated mechanism.
Why1 CD4 T cells confer enhanced immunopathology and IL-17 production to infected recipient BL/6 mice, however, the cytokine responses of Why1 T cells in the absence of schistosome infection are unknown. To determine if an inherent proinflammatory bias exists, Why1 and BL/6 CD4 T cells were isolated by negative selection from the spleens of naïve uninfected mice and stimulated with anti-CD3/CD28. Both displayed similar basal levels of cytokine mRNA, however, there were significantly higher levels of IL-17 and IFN-γ in stimulated Why1 T cells than in BL/6 T cells (Fig. 4A,B), while levels of IL-4 were higher in BL/6 T cells than in Why1 T cells (Fig. 4C), suggesting that Why1 CD4 T cells are biased towards a Th17/Th1 proinflammatory phenotype.
Using shRNA knockdown technology, we previously demonstrated that Irak2 is the gene responsible for the enhanced proinflammatory response of Why1-derived macrophages following TLR stimulation [31]. This observation supported the notion that Irak2 is primarily involved in innate immune response signaling. To investigate whether there is a direct involvement of IRAK2 in up-regulation of inflammatory cytokines in activated Why1 T cells, we employed shRNA technology to examine the effect of IRAK-2 knockdown on levels of cytokine mRNA. We observed a significant reduction in IL-17 mRNA induced by IRAK-2-specific (but not by control shGFP) hairpin treatment in Why1 T cells (Fig. 4D). Knockdown of IRAK-2 also suppressed IL-17 by BL/6 T cells (Fig. 4D), indicating that its effect is not specific to MOLF mice. Infection with the IRAK-2-specific hairpin led to a specific decrease in IL-17 expression, which strongly suggests that IRAK-2 is involved in transcriptional up-regulation of IL-17. Interestingly, there was no significant effect on IFN-γ (Fig. 4E) or IL-4 (Fig. 4F). Knockdown of IRAK-2 in BL/6 and Why1 T cells was confirmed by mRNA analysis due to the lack of a suitable antibody against IRAK-2 (Fig. 4G).
To confirm these results, we measured the effect of IRAK-2 on IL-17 protein levels. Why1 T cells stimulated with anti-CD3/CD28 produced significantly more IL-17 than their BL/6 counterparts (Fig. 4H). Furthermore, knockdown of IRAK-2 significantly decreased IL-17 production in both Why1 and BL/6 T cells at 48 and 96 hours following stimulation, indicating that this effect was stable over time (Fig. 4I, J). Unstimulated cells produced no detectable cytokine at either time point, and knockdown of IRAK-2 did not significantly affect IFN-γ or IL-4 protein levels (data not shown). IFN-γ, unlike IL-17, is largely dependent on the JAK1/STAT1 activation pathway, whereas IL-17’s promoter has well characterized consensus sites for the NFAT and NF-κB transcription factors [33], [34], [35]. We thus postulate that the specific effect of IRAK-2 on IL-17 is related to its role in activating NF-κB and p38 MAP kinase. These results uncover a novel role for IRAK-2 in directing Th17 cell polarization.
To directly assess the effect of IRAK-2 on the schistosome infection in vivo, we examined the immunopathology and cytokine profile in IRAK-2-deficient (IRAK-2-/-) mice. Since naturally high-pathology IRAK-2-/- mice are currently not available, we took advantage of a model in which severe egg-induced immunopathology with high IL-17 levels can be induced in infected BL/6 mice by immunization with SEA in CFA (SEA/CFA) [12]. This model shares many attributes with naturally occurring high pathology, and has been instrumental in identifying mechanisms controlling the elevated Th17 responses[12], [30], [36], [37]. After 7 weeks of infection, the SEA/CFA-immunized IRAK-2-/- mice appeared healthier and exhibited significantly reduced granulomatous inflammation when compared to similarly treated WT BL/6 mice and IRAK-2+/- littermate controls (Fig. 5A). SEA-stimulated MLN CD4 T cells from infected, SEA/CFA-immunized IRAK-2-/- mice also produced significantly less IL-17 than BL/6 and IRAK-2+/- controls, and barely above the levels of their unimmunized counterparts (Fig. 5B). Interestingly, while IRAK-2 did not significantly influence the levels of IFN-γ or IL-5 (Fig. 5C,E), its absence led to higher IL-4 production (Fig. 5D). These results demonstrate that IRAK-2 plays a key role in the development of severe immunopathology and IL-17 production in schistosomiasis.
IL-1β and IL-23 have been shown to play key roles in pathogenic Th17 cell development from naive precursors [26], [27], [38]. We have also demonstrated these cytokines to be necessary for Th17 cell differentiation in high-pathology CBA mice [28]. IRAK-2 plays a key role in MyD88 dependent TLR signaling, however, it also can function downstream of the IL-1 receptor [39], [40]. Our results suggest that the role of IRAK-2 in Th17 cell development is T cell intrinsic (Figure 3B–F). To address the molecular basis by which wild-derived IRAK-2 leads to enhanced IL-17 production, we stimulated naïve CD4 T cells with IL-1 and IL-23, alone or in combination, and found that IL-1 per se induced IL-17 production only in CD4 T cells from Why1, but not from BL/6 or IRAK-2-/- mice (Figure 3B–F); IL-23 alone was ineffective. Since IL-23 is known to induce IL-17 production in memory Th17 cells [41], [42] these results suggest that neither Why1 nor BL/6 T cells had a pre-existing memory phenotype. Rather, IL-23 synergized with IL-1 for a significantly greater IL-17 production by Why1 than either BL/6 or IRAK-2-/- cells (Fig. 6A). In CD4 T cells additionally stimulated non-specifically with anti-CD3/CD28, IL-1 again elicited significantly more IL-17 production in Why1 than in BL/6 cells, whereas the lowest levels of IL-17 were produced by T cells from IRAK-2-/- mice (Fig. 6B). IL-23 markedly enhanced IL-1-induced IL-17 production by Why1 and BL/6 T cells, but significantly less so in IRAK-2-/- T cells (Fig. 6C). These results demonstrate that IL-1-induced IL-17 production by T cells is highly dependent on IRAK-2; they also indicate that IL-23 by itself is unable to stimulate IL-17 production, but significantly potentiates IL-1 in carrying out this function.
As further evidence of the role of IRAK-2 in boosting Th17 development, we found a significantly higher expression of the Th-17-cell associated cytokine IL-22 in IL-1-stimulated CD4 T cells from Why1 than from BL/6 or IRAK-2-/- mice (Fig. 6D). IL-23R expression, which is a marker of activated and memory Th17 cells [41], [43], was also significantly higher in anti-CD3/CD28 plus IL-1 stimulated CD4 T cells from Why1 than from BL/6 or IRAK-2-/- mice (Fig 6E). Additionally, in Why1 T cells there was significantly higher expression of the Th17 cell lineage-associated transcription factors RORγt (Fig. 6F) and AP-1 B-cell activating transcription factor (BATF) (Fig. 6G) [17], [44]. Time course analysis revealed an earlier and short-lived induction of Rorγt in comparison with BATF, but at all times both expression levels were higher in Why1 than in BL/6 cells, and were profoundly down-regulated in IRAK-2-/- cells. Interferon regulatory factor 4 (IRF4) and the aryl hydrocarbon receptor (AHR) also play important roles in Th17 cell biology[45], [46], [47], however, there were no significant differences in their expression among the various cell populations (not shown). Likewise, no significant differences were observed in the induction of the Th1 and Th2 cell-associated transcription factors T-bet and Gata-3 (not shown).
For effective induction of responsive genes following TLR/IL-1R stimulation, IRAK family kinases are known to activate a series of downstream signaling events, including NF-κB and certain MAPK family members[48], [49]. To assess the effect of IRAK-2 on these molecular mediators, CD4 T cells were stimulated with IL-1 and the activation of two pathways downstream of the IRAK signaling complex, NF-κB and MAP kinases, were compared. Western blot analysis of phosphorylation levels revealed that in Why1 T cells there was enhanced activation of the IκB kinase (IKK) family member p105 in comparison with BL/6 controls, suggesting increased activation of the NF-κB axis via IL-1 receptor signaling. At the same time, changes in Erk phosphorylation were insignificant, indicating that the MAPK pathway is less affected by the pro-inflammatory IRAK-2 isoform (Fig. 6H). Taken together, these results suggest that IL-1-induced Th17 cell polarization via IRAK-2 is associated with increased expression of the transcription factors RORγt and BATF, likely through enhancement of NF-κB activity.
Murine schistosomiasis is a well-established experimental model of a major human infectious disease. Humans as well as mice develop marked differences in disease severity and it is clear that immunopathology is profoundly affected by the host genome. Thus, a greater understanding of its pathogenic mechanisms and underlying genes has widespread implications. Our laboratory has identified several genetic intervals that are associated with severe disease in mice [29], [50], of which some correspond to regions in the human genome that contain the loci Schistosoma mansoni 1 (Sm1) and Sm2 [51], [52], [53]. Despite these efforts, specific genes that control severe disease have not been identified to date. One reason for this is the genetic redundancy of classical inbred mouse strains, which facilitates genetic analysis of “simple” monogenic and fully penetrant traits. However, greater genetic diversity may be required when investigating traits that are conferred by multiple loci that impart a quantitative contribution to the phenotype. Hence, the limited diversity of classically used strains can make it particularly difficult to identify genes that underlie complex traits, such as those involved in the host response to schistosome infection.
Using the more genetically diverse wild-derived mice as a model, we provide evidence of how genetic mapping of complex traits can be dissected with prior knowledge of the loci or genes identified in relatively simple screens. Previously, we positionally cloned a mutation in the promoter of IRAK-2C that limits the expression of the inhibitory isoform of IRAK-2 in MOLF mice. The outcome of this differential expression is a higher ratio of proinflammatory IRAK-2A relative to the inhibitory isoform IRAK-2C, which in turn leads to an enhanced proinflammatory response in MOLF macrophages[31],[54]. Extending these findings to a physiological setting in vivo, we now show that addition of the MOLF Why1 interval, which contains Irak2, markedly increases the levels of IL-17 and the severity of egg-induced hepatic immunopathology in schistosome-infected BL/6 mice. Using a reciprocal approach, we also observed that the deletion of Irak2 leads to a significant defect in IL-17 production and a marked reduction of immunopathology, thus identifying Irak2 as the causal gene for this in vivo phenotype. The effect of Irak2 on immunopathology is striking since susceptibility to S. mansoni infection is likely conferred by many genes, which have been elusive in previous genetic screens measuring immunopathology as a direct phenotypic read-out. IRAK-2 was not identified in our previous genetic screens in vivo [29], [50]. This is not surprising because these analyses were done in classical inbred mice, which have different levels of pathology but similar IRAK-2 alleles, thus precluding the assessment of the wild-derived IRAK-2 allele in T cell activation during infection. The observed effect of the Why1 locus and Irak2 on pathology thus sets an important precedent for how results of a genetic screen in vitro can be used for identification of genes influencing complex traits in vivo.
IRAK family kinases are central to TLR signaling and a critical factor in innate immunity [55]. Recently, IRAK family kinases have been studied in the adaptive immune response with some discrepancy as to their precise role. IRAK-4 has first been suggested to be an essential factor in TCR induced T cell responses[56]. However, these results have not been confirmed as it was later shown that IRAK-4 is dispensable for normal T cell responses and TCR activity[57], [58]. Here we provide evidence that another IRAK family member, IRAK-2, critically affects T cell biology by regulating the ability of IL-1 to promote Th17 function. Thus, stimulation of T cells with either IL-1 alone, or together with anti-CD3/CD28, resulted in a dramatic increase in IL-17 production by Why1 CD4 T cells compared with BL/6, while IL-17 from IRAK-2-/- T cells was minimal. Stimulation of Why1 CD4 T cells with IL-1 also led to increased activation of the IκB kinase p105, which promotes the degradation of IκB and allows NF-κB to translocate to the nucleus [59]. These observations identify IRAK-2 as a key regulator of Th17 cell biology by enhancing IL-1R signaling through NF-κB activation. This effect of IRAK-2 was specific to Th17 and did not affect IFN-γ production, which is in agreement with recent observations linking NF-κB specifically with IL-17, but not IFN-γ [34]. Our data also show that stimulation with IL-1 in the absence of TCR engagement is sufficient to induce IL-17 production in Why1 T cells, suggesting that their high expression of IRAK-2 is responsible for the increased Th17 responses. However, naive BL/6, Why1 or IRAK-2 -/- T cells did not express the IL-23R or respond to stimulation with IL-23, two hallmarks of activated/memory Th17 cells [18], [20], [41], [42], [60], [61], suggesting that neither one was in a state of activation prior to stimulation. Altogether, these findings imply that the role of IRAK family members in T cell responses is not limited to an effect on TCR signaling, but rather that they can also act via the IL-1R-MyD88 complex to direct Th17 cell responses.
Among several candidate transcription factors, RORγt has been demonstrated to play a central role in Th17 cell differentiation, as its absence significantly impairs IL-17 production [45], [62]. We now show that Why1 CD4 T cells significantly up-regulate RORγt expression following stimulation with IL-1, suggesting that IL-1 per se can activate Th17 cells through an IRAK-2 dependent pathway. More recently, BATF was identified as a key transcription factor in Th17 cell differentiation [44], as BATF-deficient mice displayed impaired Th17 cell activity and were resistant to EAE despite normal IL-6 signaling. BATF synergized with RORγt to enhance IL-17 production and sustained RORγt expression in Th17 cells, although the exact nature of their interaction remains to be elucidated [63]. Here we show that BATF expression is significantly enhanced in IL-17-producing Why1 CD4 T cells compared with BL/6 T cells and that this function is IRAK-2 dependent. Interestingly, in our model RORγt expression peaks earlier than BATF (Fig 6F,G), suggesting that RORγt may up-regulate BATF during Th17 cell development. Our findings also suggest that BATF functions downstream of the IL-1 receptor thus explaining why BATF-/- mice have a defect in Th17 cell differentiation [44], [63]. IRF4 is another transcription factor involved in Th17 cell differentiation via IL-21[45], [64]. In our system, stimulation with IL-1 resulted in increased IRF4 expression, but levels were not significantly different among BL/6, Why1 or IRAK-2-/- T cells. Likewise, we found no significant differences in the expression of AhR, which has a demonstrated regulatory role in Th17 development and function [47], [65] (data not shown).
As demonstrated in this study, wild-derived IRAK-2 confers on T cells a powerful, TCR-independent hypersensitivity to stimulation with IL-1, which is further amplified in the presence of IL-23. Mechanistically, this is due to a deletion in the IRAK-2C promoter leading to unopposed activation of the main proinflammatory isoform of IRAK-2A. This contrasts with BL/6 mice, in which the inhibitory isoform IRAK-2C is abundantly expressed and up-regulated in response to inflammatory stimuli[31]. Tissue inflammation induces large amounts of IL-1 and IL-23 and it has been suggested that non-antigen specific Th17 cells responding to these stimuli may aggravate tissue damage [26]. In schistosomiasis, IL-1 and IL-23 are highly expressed in MLN and hepatic lesions of high-pathology CBA, but not low-pathology BL/6 mice. Furthermore, dendritic cells derived from the bone marrows of normal CBA mice produce abundant IL-1 and IL-23 in response to stimulation with live schistosome eggs, whereas those from BL/6 mice do not, clearly linking these cytokines with exacerbated disease [28], [30]. Interestingly, IL-1 and IL-23 are also key cytokines for human Th17 cell differentiation [66], and given that humans contain only one proinflammatory isoform of IRAK-2 similar to wild-derived mice [31], it is possible that IRAK-2 may enhance the sensitivity of Th17 cells in a TCR-independent manner and further aggravate immune-mediated tissue damage in human inflammatory diseases.
In summary, using wild-derived mice as a model, we illustrate the first example of a gene controlling severe pathology in murine schistosomiasis, setting an example of how analysis of simple monogenic traits in vitro can be applied to complex in vivo models of infection or autoimmunity. We used this model to uncover a novel role for IRAK-2 in CD4 T cell signaling via the IL-1 receptor and show that IRAK-2 is a key regulator of IL-1-mediated Th17 cell biology, which may have wide-ranging effects on other Th17 cell-mediated inflammatory diseases.
All the animal experiments were performed in accordance with the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health and with the permission of the American Association for the Assessment and Accreditation of Laboratory Animal Care. The protocol was reviewed and approved by the Tufts Medical Center Institutional Animal Care and Use Committee and the Division of Laboratory Animal Medicine (Permit Number: B2009-88).
C57BL/6J, CBA/J and MOLF/Ei (MOLF) mice, 5-8wk old, were purchased from the Jackson Laboratory. Why1 mice were produced as previously described[31]. These congenic mice are homozygous for the MOLF allele selected by marker D6Mit328 (chromosome 6 at 112.7 Mb) on a BL/6 background. IRAK-2-/- mice were obtained from Dr. Shizuo Akira (Research Institute for Microbial Diseases, Osaka, Japan). Mice were bred and maintained at the Animal Facility at Tufts University School of Medicine in accordance with the American Association for the Assessment and Accreditation of Laboratory Animal Care guidelines. Some mice were infected by i.p. injection with 85 cercariae of S. mansoni (Puerto Rico strain) obtained from infected Biomphalaria galabrata snails provided by Dr. Fred Lewis (Biomedical Research Institute) through National Institutes of Health/National Institute of Allergy and Infectious Diseases Contract N01-AI-55270. For some experiments, IRAK-2-/-, IRAK-2+/- and BL/6 mice were immunized s.q. with 50 µg of SEA/CFA before and after infection, as previously described[12]. SEA was prepared as previously described[67].
Formalin-fixed liver samples from 7 week-infected mice were processed for histopathological analysis of 5-µm sections stained with H & E. The extent of granulomatous inflammation around schistosome eggs was measured by computer-assisted morphometric analysis using Image-Pro Plus software (Media Cybernetics) as previously described[29]. At least 15 granulomas were counted per liver. Granuloma size was expressed in square micrometers ± SEM. The schistosome egg load was assessed by counting the number of eggs present in 1 mm2 fields of liver tissue in sections stained with hematoxylin/eosin as previously described [68].
MLNC suspensions were prepared from individual mice by teasing the lymph nodes in supplemented RPMI 1640 medium containing 10% FCS (Atlanta Biologicals) as previously described[30]. CD4 T cells from MLN or spleens were purified by negative selection using CD4 MACS columns (Miltenyi Biotec) in accordance with manufacturer’s instructions. CD4 T cell purity was >95% by FACS analysis. Liver granuloma cells were isolated as previously described[12].
Bulk MLNC or GC suspensions (5×106 cells/ml), or purified CD4 T cells (1×106 cells/ml) plus normal irradiated syngeneic splenic APC (4×106 cells/ml), were incubated in the presence or absence of 15 µg/ml SEA for 48hrs. IL-17, IFN-γ, IL-6, TNFα, IL-4, IL-5 and IL-10 protein concentrations in the cell culture supernatants were measured by ELISA using standard cytokines, Abs and protocols from R&D Systems.
1×106 purified MLN CD4 T cells from 7 week-infected BL/6 and Why1 mice were cultured ex vivo with 4×106 irradiated naïve splenic APCs from BL/6 or Why1 mice for 48 hours in the presence or absence of 15 µg/ml of SEA. IL-17 levels in cell supernatants were measured by ELISA as described. For the cell transfer experiments, BL/6 recipient mice were sublethally irradiated (500 rad) 3 days prior to infection and subsequently injected i.v. with 8×106 naïve splenic CD4 T cells from BL/6 or Why1 donor mice, purified by negative selection as described above. After 7 weeks of infection, IL-17 production by SEA-stimulated bulk MLNC, and by purified MLN CD4 T cells plus irradiated naïve splenic APC, was measured by ELISA as described.
Total RNA was isolated from individual samples using TriZol reagent (Invitrogen) as per manufacturers instructions. RNA (1–5 µg) was subjected to DNASE I treatment (Roche) and reverse-transcribed using the high capacity cDNA reverse synthesis kit (Applied Biosystems). Real-time quantitative RT-PCR was performed by SYBR green or Taqman analysis using an ABI 7300 instrument. GAPDH levels were used to normalize the data. Taqman real-time probes for IL-17, IFN-γ, IL-4, IL-12p40, IL-12p35, IL-23p19, IL-22, IL-1β, IL-23R and batf were obtained from Applied Biosystems. Primers for SYBR green analysis of rorγt were described previously[17]. Using the average mean cycle threshold (Ct) value for GAPDH and the gene of interest for each sample, the equation 1.8 e (Ct GAPDH - Ct gene of interest) ×104 was used to obtain normalized values [69].
1×106 cells CD4 T cells were stimulated with IL-1β (4 ng/ml, R&D Systems) for 0, 5, 10, 20 and 30 minutes followed by lysis on ice in a cytoplasmic lysis buffer (50 mM Tris, pH 8, 150 mM NaCl, 2 mM EDTA, 1% Triton X-100, 1 mM NaVanadate and 10 mM NaF) supplemented with Halt protease inhibitor cocktail (Thermo Fisher Scientific) for 10 min. Lysates were then centrifuged at 13,000 rpm at 4°C for 10 min. Cleared lysates were resolved on a 4-12% gradient Bis-Tris SDS gel (NuPAGE; Invitrogen) and transferred to a nitrocellulose membrane. Rabbit polyclonal antibodies to phosphorylated ERK and p105 were obtained from Cell Signaling Technology. After incubation with specific Abs, chemiluminescence was detected using ECL substrate (Thermo Fisher Scientific).
To down-regulate the expression of IRAK-2 in mouse T cells, we used infection with lentiviral particles expressing IRAK-2-targeting shRNA. Lentiviral particles were produced by transfecting (Fugene, Roche) 293-T cells with a plasmid encoding IRAK-2-specific shRNA in the pLKO.1 vector (Open Biosystems clone ID TRCN000022502) together with two other plasmids, pSPAX2 and pMD2.G (Addgene), encoding packaging components of the lentivirus. Supernatants from 293-T cells were harvested on days 2 and 3 after transfection and passed through a 0.22 µm filter.
Naïve CD4 T cells were purified from normal Why1 and BL/6 mouse spleens using the Easysep kit (StemCells). T cells were resuspended to a density of 2×106 cells/ml and plated on 6 well plates that were previously seeded with resident i.p. macrophages from normal BL/6 mice (1.5×106 cells/well). Viral supernatant and medium were added at a 1∶1 ratio for 18 hours. Subsequently, the T cells were washed and allowed to recover for 96 hrs in the presence of macrophages to promote survival. Non-adherent T cells were re-plated at a concentration of 1×106 viable cells/ml for stimulation with anti-CD3/CD28. Cells and supernatants were collected after 48 and 96 hrs.
Naïve CD4 T cells were incubated in either 96 well plates (3.5×105 cells/ml) for ELISA detection, or 48 well plates (1×106 cells/ml) for real-time analysis in triplicates, and stimulated with anti-CD3/CD28 coated beads (3×105, Dynal) together with rIL-1β, at indicated concentrations, and rIL-23 (20 ng/ml). For ELISA, cell culture supernatants were collected after 4 days and analyzed for IL-17 as described. For real-time PCR, cells were collected at 0, 2, 24, 48 and 96 hrs in Trizol reagent and assayed as described.
ANOVA and Student’s t tests were used to determine the statistical significance of the differences between groups and were calculated with GraphPad Prism.
Il17a (Mouse Genome Informatics:107364), Ifng (MGI:107656), Il6(MGI:96559), Tnf (MGI:104798), Il4 (MGI:96556), Il5 (MGI:96557), Il10 (MGI:96537), Irak2 (MGI:2429603), Rorc (MGI:104856), Batf (MGI:1859147), Irf4 (MGI:1096873), Ahr (MGI:105043).
|
10.1371/journal.pntd.0001632 | Linking Oviposition Site Choice to Offspring Fitness in Aedes aegypti: Consequences for Targeted Larval Control of Dengue Vectors | Current Aedes aegypti larval control methods are often insufficient for preventing dengue epidemics. To improve control efficiency and cost-effectiveness, some advocate eliminating or treating only highly productive containers. The population-level outcome of this strategy, however, will depend on details of Ae. aegypti oviposition behavior.
We simultaneously monitored female oviposition and juvenile development in 80 experimental containers located across 20 houses in Iquitos, Peru, to test the hypothesis that Ae. aegypti oviposit preferentially in sites with the greatest potential for maximizing offspring fitness. Females consistently laid more eggs in large vs. small containers (β = 9.18, p<0.001), and in unmanaged vs. manually filled containers (β = 5.33, p<0.001). Using microsatellites to track the development of immature Ae. aegypti, we found a negative correlation between oviposition preference and pupation probability (β = −3.37, p<0.001). Body size of emerging adults was also negatively associated with the preferred oviposition site characteristics of large size (females: β = −0.19, p<0.001; males: β = −0.11, p = 0.002) and non-management (females: β = −0.17, p<0.001; males: β = −0.11, p<0.001). Inside a semi-field enclosure, we simulated a container elimination campaign targeting the most productive oviposition sites. Compared to the two post-intervention trials, egg batches were more clumped during the first pre-intervention trial (β = −0.17, P<0.001), but not the second (β = 0.01, p = 0.900). Overall, when preferred containers were unavailable, the probability that any given container received eggs increased (β = 1.36, p<0.001).
Ae. aegypti oviposition site choice can contribute to population regulation by limiting the production and size of adults. Targeted larval control strategies may unintentionally lead to dispersion of eggs among suitable, but previously unoccupied or under-utilized containers. We recommend integrating targeted larval control measures with other strategies that leverage selective oviposition behavior, such as luring ovipositing females to gravid traps or egg sinks.
| Controlling the mosquito Aedes aegypti, the predominant dengue vector, requires understanding the ecological and behavioral factors that influence population abundance. Females of several mosquito species are able to identify high-quality egg-laying sites, resulting in enhanced offspring development and survival, and ultimately promoting population growth. Here, the authors investigated egg-laying decisions of Ae. aegypti. Paradoxically, they found that larval survival and development were poorest in the containers females most often selected for egg deposition. Thus, egg-laying decisions may contribute to crowding of larvae and play a role in regulating mosquito populations. The authors also tested whether removal of the containers producing the most adult mosquitoes, a World Health Organization-recommended dengue prevention strategy, changes the pattern of how females allocate their eggs. Elimination of the most productive containers led to a more even distribution of eggs in one trial, but not another. These results suggest that behavioral adjustments by egg-laying females may lessen the effectiveness of a common mosquito control tactic. The authors advocate incorporating control strategies that take advantage of the natural egg-laying preferences of this vector species, such as luring egg-laying females to traps or places where their eggs will accumulate, but not develop.
| At present, dengue virus transmission can be controlled or prevented only through suppressing mosquito vector populations [1]. Even with the advent of a licensed dengue vaccine, which is anticipated by 2015 [2], vector control will remain a necessary component of any sustainable program to eliminate dengue transmission in endemic areas or prevent virus introduction into new areas [3]. Unfortunately, few contemporary dengue control programs have achieved the high thresholds of vector population suppression (estimated to be >90% at some locations [4], [5]) needed to prevent epidemics [6]. Controlling Aedes aegypti, the primary dengue vector worldwide, is challenging because it is well-adapted to the domestic environment [7], [8]. Adult mosquitoes rest indoors on clothing and underneath furniture, where they are difficult to reach using traditional aerosol or residual insecticides [7], [9]. Furthermore, females deposit their eggs in a wide assortment of man-made containers, ranging from water storage drums to discarded bottles and cans, making exhaustive larval control impractical in most cases [4], [10], [11].
Ae. aegypti productivity tends to be clustered at most field locations, with the majority of the adult population emerging from a small subset of water-holding containers [10]–[12]. Thus, targeting larviciding and container elimination efforts to these most productive containers may substantially improve the efficiency and cost-effectiveness of dengue control [13]. Proponents of targeted larval control predict that elimination of containers producing, for example, 80% of pupae will lead to a sustained linear reduction in the total adult density [10]. This expectation is based, however, upon two key assumptions: (1) all available Ae. aegypti larval development sites are already at carrying capacity and (2) oviposition behavior has little impact on population dynamics [10]. Field evaluations of targeted larval control programs have yielded mixed outcomes. Investigators in Myanmar and the Philippines reported nearly linear reductions (73–77%) in the Ae. aegypti Pupae per Person Index (PPI) after 5 months [12]. In Thailand, however, only a 15% reduction in PPI was observed after implementing a targeted control campaign designed to eliminate 80% of pupal production. In Iquitos, Peru, a 236% increase in PPI was noted after an intervention designed to eliminate 92% of pupal production [12]. Thus, the efficacy of targeted larval control varies substantially between settings and likely depends upon details of Ae. aegypti ecology and population dynamics at the local scale.
Selection of an oviposition site by a female mosquito directly affects offspring survival and growth [14]–[16], and has consequences for population dynamics [17]. Because evolutionary theory predicts that animals should act to maximize their reproductive success, egg-laying females are expected to select the most suitable sites for their offspring based on reliable cues of habitat quality [18]–[20]. Whether and how female Ae. aegypti select oviposition sites, the impact of oviposition decisions on offspring fitness, and how females adjust to changes in oviposition site availability will affect the validity of the two key assumptions underlying targeted larval control. Previously, we demonstrated that free-ranging Ae. aegypti in Iquitos actively select egg-laying sites [21]. In particular, females exhibited a preference for containers holding conspecific larvae and pupae. Container characteristics of secondary importance included large size, abundant organic material, and exposure to sunlight [21].
In the present study, we assessed whether Ae. aegypti oviposition site choice is correlated with offspring performance. We tested the prediction that females will lay more eggs in containers in which more juveniles successfully complete development and grow to large adult size, two important components of mosquito fitness [22]–[24]. We also investigated how individual females partition their egg batch among available containers. We predicted that, prior to targeted container elimination, individual females would cluster their egg batch in a preferred container, but switch to spreading their eggs widely among more remaining, available containers if preferred sites were eliminated. By examining whether Ae. aegypti females adjust their egg-laying strategies in response to environmental change as well as the implications of oviposition site choice for population dynamics, we hope to better understand why targeted larval control measures may not achieve the desired level of population reduction in some settings. Ultimately, we expect our detailed findings on Ae. aegypti behavior to provide insight for the development of improved strategies for vector population suppression.
Households included in our field experiment were selected based on the home owners' willingness to participate. After explanation of study objectives and procedures, verbal consent was obtained from the head of each household. We did not collect information on household residents. Our study was approved by the local Ministry of Health, Dirección Regional de Salud-Loreto. Institutional Review Boards (IRBs) from the University of California, Davis and the United States Naval Medical Research Center (Project #: PJT-NMRCD.032) determined that our study did not meet the definition of human subjects research and IRB approval was, therefore, not required. A waiver of IRB approval was granted by the UC Davis IRB for feeding laboratory-reared mosquitoes on humans.
Our study was conducted in Iquitos (73.2°W, 3.7°S, 120 m above sea level), a city of approximately 380,000 people in northeast Peru. Iquitos is located at the confluence of the Amazon, Nanay, and Itaya Rivers in the Department of Loreto and has been described in detail previously [25]–[27]. Daily air temperature, relative humidity, and rainfall data collected from a National Oceanic and Atmospheric Administration meteorological station located at the airport (∼6 km from the city center) demonstrated that the climate of Iquitos is relatively consistent year round, with rain falling during all months and small fluctuations occurring in temperature and relative humidity [28], [29]. Our experiments took place during August to November 2008. During these months, mean temperature (± SD) was 26.2±1.3°C, mean relative humidity (± SD) was 81.2±5.1%, and mean daily rainfall (± SD) was 6.0±12.0 mm [28].
Both experiments conducted during this study (described below) required genotyping mosquitoes to match them to parents. We established 18 Ae. aegypti family lines in the field laboratory by collecting Ae. aegypti eggs (F0 generation) from 36 households across 18 neighborhoods in Iquitos. Because our goal was to make these families easily distinguishable, each family originated from a different neighborhood (males and females collected >100 m apart to avoid inbreeding) to maximize the number of alleles shared within a family and minimize alleles shared between families. Field-collected eggs were hatched by immersion in hay infusion overnight and larvae reared according to the standardized protocol described by Wong et al. [29]. Throughout the rearing process, mosquitoes were kept separated by collection house and date. Paired matings were set up as detailed by Wong et al. [30] and all F0 mosquitoes were assigned unique identifying numbers.
Females were offered an opportunity to imbibe blood from a human daily, but were not fed sugar (see [30]). F1 eggs were collected daily, labeled by the mother's identifying number, allowed to embryonate in a moist chamber for 48 hrs, dried for storage, and later hatched for experiments. Upon completion of three gonotrophic cycles or death, F0 parents were transferred to 1.5 mL plastic vials filled with 96% ethanol and stored at −20°C for subsequent genotyping.
Data loggers were used to record weather variables once per hour. During the field experiment, Hobo ProV2 data loggers (U23-001) were deployed in 14 of the 20 houses (attached to the side of a container) to monitor ambient temperature and relative humidity (Onset Computer Corporation, Pocasset, MA). In the same houses, Hobo Pendant loggers (UA-002-64) were placed inside containers to monitor water temperature. We did not have enough data loggers to monitor weather at all 20 houses, but based on previous experience we expected that temperatures would be consistent across the city. Within the semi-field enclosure, loggers were used to record air temperature, relative humidity, and water temperature indoors and outdoors once per hour.
All specimens from this study were transported to the University of California, Davis (UCD) for DNA extraction and genetic analysis. DNA from adults used in paired laboratory matings (to establish families) was purified by potassium acetate/ethanol precipitation [37]. DNA from legs of released females was isolated using the same method, with the exception that reagents were used at 50% volume. Due to the large number of experimentally collected mosquitoes (from the field or semi-field enclosure), DNA from these individuals was purified using the automated BioSprint 96 DNA extractor and reagents from the BioSprint 96 Kit (Qiagen, Valencia, CA).
Individuals were genotyped at ten microsatellite loci using fluorescent-labeled forward primers as described in Wong et al. [30]. Polymerase chain reaction (PCR) products were diluted 1∶60 in ddH2O and submitted to the College of Agriculture and Environmental Sciences Genomics Facility at UCD (http://cgf.ucdavis.edu/home/) for fragment analysis on an ABI 3730 XL capillary sequencer (Life Technology Corp., Carlsbad, CA). Resulting chromatograms were analyzed using ABI Peak Scanner™ software (Applera Corp., Norwalk, CT). Exclusion-based parentage analysis was performed using PROBMAX version 1.2 [38] to identify offspring of parental pairs [30].
During the field study, mean air temperature, water temperature, and relative humidity were consistent across houses and between the two trial periods (Table S1). Mean air temperature ranged from 26.6±1.9°C to 28.1±2.7°C. Water temperature was similar to air temperature, but exhibited less fluctuation throughout the day. Mean relative humidity ranged from 76.3±6.1% to 82.8±6.1%.
Within the semi-field enclosure, air temperature, water temperature, and relative humidity were also similar between trials (Table S2). Mean air temperature ranged from 27.3±0.8°C to 29.3±1.0°C indoors and from 26.4±1.2°C to 29.7±1.3°C outdoors. In general, temperatures fluctuated less in water compared to air, and less indoors compared to outdoors. Mean relative humidity ranged from 70.8±4.8% to 83.6±4.9%.
Data on the mean number of eggs deposited per container per week are shown in Figure 4. The optimal model included a random effect due to house and fixed effects due to container size, fill method, larval density, and week (Table 1). In general, more eggs were laid in large vs. small containers, in unmanaged vs. manually filled containers, with increasing larval density, and to a lesser extent, with week. There was no significant effect of trial (likelihood ratio = 0.57, p = 0.45) or container size by fill interaction (likelihood ratio = 1.29, p = 0.256).
A total of 3,263 pupae were collected from all containers located in the 20 households (mean [± SE] = 40.7±7.3 pupae per container; range = 0 to 384). More pupae were collected from large unmanaged containers (n = 1,933 pupae) than any other container treatment (Figure S1). Due to the time-intensive nature of genotyping all mosquitoes in order to identify those introduced from established families (25 F1 larvae introduced per container), standardized pupation probability was calculated for Ae. aegypti from containers in eight houses from the first trial (Figure 5). In these eight households, mean (± SE) larval density just prior to F1 introduction was 2.56 (±0.83) larvae per L in large unmanaged containers, 0.76 (±0.37) larvae per L in large manually filled containers, 0.75 (±0.32) larvae per L in small unmanaged containers, and 0 larvae per L in small manually filled containers. Of the 996 pupae genotyped, we matched 231 individuals to parental pairs from established families (mean [± SE] = 7.2±1.3 matched individuals per container; range = 0 to 23).
Pupation probability was significantly influenced by treatment (container size by fill method interaction) and house. First instar larvae introduced into small unmanaged containers exhibited significantly higher probability of pupation (β = 3.37, p<0.001) compared to individuals in the three other container types. No differences in pupation probability were observed among individuals developing in small manually filled containers compared to large unmanaged (β = −2.37, p = 0.214) or large manually filled containers (β = −1.79, p = 0.479). There was also no difference in pupation probability between individuals from large containers, regardless of fill method (β = 0.58, p = 0.811). Within each container treatment, we found no significant effect of larval density on pupation rates (likelihood ratio = 0.67, p = 0.414).
Mean wing length of female mosquitoes collected from all 20 houses are shown in Figure 6. Wing lengths of males followed a similar pattern (Figure S2). Female wing lengths ranged from 1.85 to 3.23 mm (median = 2.51 mm) and wing lengths of males ranged from 1.55 to 2.56 mm (median = 2.00 mm). The optimal mixed effects models included a random effect due to house and fixed effects due to container size, fill method, and larval density (Table 2). Female wing length decreased significantly among Ae. aegypti developing in large vs. small containers, in unmanaged vs. manually filled containers, and with increasing larval density. Similar patterns were observed for males, with wing length decreasing in large containers, in unmanaged containers, and with increasing larval density. For both sexes, there was no significant effect of trial (females: likelihood ratio = 1.72, p = 0.190; males: likelihood ratio = 1.38, p = 0.24) or container size by fill interaction (females: likelihood ratio = 0.25, p = 0.803; males: likelihood ratio = 0.61, p = 0.435).
The numbers of females released, eggs collected, and offspring genotyped during each trial within the semi-field enclosure are shown in Table 3. The total number of eggs collected decreased steadily during each successive trial. Detailed results regarding on which days and in which containers individual females laid their eggs (those that could be genotyped) are displayed in Figure S3.
Based on genotyped offspring, we calculated the largest proportion of each egg batch that was concentrated within a single container (Figure 7). During the first trial (pre-intervention), six egg batches were each aggregated within a single container (always in a large unmanaged container). Among all subsequent trials (pre- and post-intervention), there was only a single batch in which all eggs were deposited within a single container (trial 2, concentrated in a small manually filled container). In general, egg distribution was more clumped during the first trial compared to the later three trials.
Values for the Shannon equitability indices for each trial are shown in Figure 8. In our model, Shannon indices were affected by trial, but not by gonotrophic cycle number or female (data not shown). Shannon equitability indices were significantly different between the two pre-intervention trials (trial 1 vs. trial 3, β = 0.18, p = 0.014), but not between the two post-intervention trials (trial 2 vs. trial 4, β = 0.02, p = 0.991). Individual females' egg batches were more clumped during trial 1 (pre-intervention) compared to the two post-interventions trials (β = −0.17, p<0.001). There was no difference, however, in Shannon equitability indices for trial 3 (also pre-intervention) compared to the two post-intervention trials (β = 0.01, p = 0.900).
When containers were examined daily for whether or not they received eggs (all eggs included, genotyped or not), the random effect of trial was not significant (intercept variance = 0). The probability that a container received eggs increased when containers were located indoors (β = 1.36, p<0.001) and if containers were large and unmanaged (β = 1.16, p = 0.012). The overall probability that any container received eggs increased during the post-intervention scenario (only small manually filled containers present in enclosure: β = 1.36, p<0.001).
When presented with a choice of four container types varying in size and organic content, wild female Ae. aegypti consistently deposited more eggs in large containers with abundant organic material. This behavior is expected to be adaptive, with females choosing sites based on cues of habitat quality. After monitoring the development of juvenile Ae. aegypti, however, we did not find a positive association between female egg-laying choice and juvenile growth or survival. The container type most preferred by ovipositing females (large unmanaged) produced individuals with low pupation probability and small adult body size. Pupation probability was highest among Ae. aegypti in small unmanaged containers, which received ample food and relatively few eggs, creating an environment consistent with low competitive pressure for food. In large unmanaged containers, we suspect that high food content was offset by high larval density. Large unmanaged containers may have quickly reached carrying capacity, so that F1 pupation rates were no better than in sites receiving little total food (manually filled containers). Prior to F1 introduction, mean larval density was 3.4 times greater in large unmanaged containers (2.56 larvae per L) compared to small unmanaged containers (0.75 larvae per L). To avoid colinearity with container size and fill method, we did not directly assess larval density as a predictor in our models. We instead examined relative larval density within each container treatment, but found no significant effect of larval density on pupation rates. The negative impact of high larval density was evident, however, in our analysis of Ae. aegypti body size. Large unmanaged containers yielded the smallest adult mosquitoes. Furthermore, within each of the four container treatments, body size clearly decreased with increasing larval density. Our result is consistent with previous field studies in Iquitos [26], Puerto Rico [44], and Thailand [5] that demonstrated negative relationships between the density of larvae in aquatic habitats and the size of emerging adults. Wing lengths of females collected during our study (range = 1.85 to 3.23 mm, median = 2.51 mm) were comparable to those reported by Schneider et al. [26] in Iquitos (range = 1.67 to 3.83 mm, median = 2.60 mm).
Mismatches between female oviposition preference and offspring performance have been reported for several insect species (e.g., [45], [46]), including mosquitoes [47]. Sub-optimal oviposition site selection may result from females' inability to predict stochastic events, sense determinants of site quality, or obtain complete knowledge of the environment [47]. Alternatively, apparent mismatches are sometimes attributed to experimental design and/or failure to examine important variables [46]. We attempted to simulate Ae. aegypti container colonization and water-use patterns typical of Iquitos, but our study was limited in some respects. During the re-introduction of larvae into containers to imitate colonization, eggs were hatched synchronously rather than gradually in installments, as is typical for Ae. aegypti [35]. The faster rate of larval introduction may have disproportionately increased levels of density-dependent competition in the most preferred containers (large unmanaged).
Containers occurring naturally in the field are likely to experience different rates of water evaporation and filling. This may result in dramatic fluctuations in larval densities, as well as variable cycles of desiccation and/or overflowing. To make our study design and analysis tractable, we artificially maintained stable water levels in our experimental containers. For species whose larvae develop in small containers and must mature before the habitat desiccates, maternal ability to assess water permanence would be favored [48]. It is possible that female Ae. aegypti evolved to detect cues associated with water permanence, and thus acted to trade off risks between desiccation and food competition for their progeny. Due to our experimental design, we were unable to assess the importance of container desiccation as a selective force in oviposition site choice. Such an investigation would require a detailed study on water dynamics of naturally-occurring (i.e., non-experimental) containers.
Previously, we observed that the majority of Ae. aegypti eggs tend to be aggregated within a small subset of containers. In addition, females were most likely to oviposit in sites that contained, or had recently contained, conspecific larvae and/or pupae [21]. These findings are consistent with other studies demonstrating that semiochemicals produced by conspecifics [49], [50] and conspecific-associated bacteria [51] act as oviposition attractants for Ae. aegypti (reviewed in [52]). During the present study, we did not attempt to isolate or identify these chemical mediators. Instead, our intention was to complement chemical ecologists' studies by investigating the consequences of conspecific attraction for Ae. aegypti offspring fitness and population dynamics. In our study, large aggregations of larvae in preferred containers led to the production of numerous small adults. For mosquitoes, adult body size can have important impacts on the rate of pupation growth and patterns of virus transmission. Small body size has been correlated with reduced life span and decreased fecundity for females and decreased mating success for males (e.g., [24], [53]–[56]). Female body size also exhibits a complex relationship with several components of vectorial capacity. A population dominated by small females, which are less susceptible to oral dengue infection [57] and less persistent in seeking blood meals [58], may serve to attenuate dengue transmission. On the other hand, small females must feed more frequently [59], [60], which could lead to increased rates of human-vector contact and enhance virus transmission.
Our results indicate that Ae. aegypti oviposition site choices that lead to crowding of larvae may play a role in population regulation by limiting the production and size of adults. In this situation, removal of the most productive containers would reduce adult abundance in the short term, but the long term population-level outcome would depend on the availability of alternative suitable oviposition sites in the area. If all water-filled containers are infested to their carrying capacity, targeted larval control is expected to result in a sustained, linear reduction in adult mosquito density [10]. On the other hand, if suitable unoccupied or under-utilized containers are available, targeted larval control could merely shift production to new containers over the next few generations. Results from our companion study indicated that, in Iquitos, containers suitable for Ae. aegypti development are frequently unoccupied (STS, unpublished). We predict that colonization of previously unoccupied sites could release large numbers of larvae from density-dependent food competition, eventually attenuating or undermining the immediate gains of targeted larval control. Results from a Brazilian field study support this idea. Maciel-de-Freitas and Lourenço-de-Oliveira [61] documented that elimination of the most productive container type (water tanks accounting for 72% of pupae) led to increased productivity from almost all other container classes, most notably in metal drums, which shifted from producing 3.5% to 30.7% of all pupae. Accompanied by this shift in productivity was a rebound in the adult densities to pre-intervention levels within 4–5 weeks. Only after eliminating both water tanks and metal drums (which were considered unimportant prior to the intervention) did investigators observe a long term drop in adult densities. The authors speculate that sustained reductions in Ae. aegypti densities were possible because of the similarity between water tanks and metal drums; both are large, typically shaded, perennial water storage containers. Even then, interventions that were designed to eliminate 75.9% of pupal production resulted in a 45.7% reduction in adult densities [61]. We suspect that in Iquitos and other locations where rain falls year round, large numbers of alternative containers and plasticity in Ae. aegypti oviposition behavior will render the long term results of targeted larval control less effective than anticipated.
The degree and speed of population recovery will also depend on whether females' egg distribution strategies are influenced by the characteristics of available containers. Inside the semi-field enclosure, egg distribution patterns were more aggregated for females during the first pre-intervention trial (trial 1), but not the second (trial 3), compared to the two post-interventions trials (trials 2 and 4). During trial 1, females frequently shared preferred oviposition containers, clustering the overall majority of eggs in these two sites. When only least preferred containers were available (post-intervention, trials 2 and 4), females were less likely to concentrate a large portion of their egg batch in any particular site, leading to a more even overall dispersion of eggs among containers. This pattern of spreading eggs evenly among sites, however, was also observed during our second pre-intervention trial (trial 3). Our mixed results suggest that egg distribution strategies are somewhat plastic and context-dependent. Differences between trials 1 and 3 may be the result of behavioral variation among individuals. Even individuals within a population are expected to vary in oviposition site selection strategies [62]. It is thus conceivable that individuals faced with similar environments could vary in their egg distribution strategies as well. Nonetheless, when we examined all eggs laid within the enclosure (genotyped or not), the overall probability that a container received eggs did increase during the post-intervention trials. The possibility that females may spread eggs more widely after elimination of the most productive containers is consistent with evidence from the field [61] and deserves further investigation.
A major shortcoming of this experiment was our inability to genotype offspring from eggs that failed to hatch. Overall, we were able to assign parentage to 74% of all offspring from the semi-field enclosure. This provides an informative, albeit incomplete, picture of oviposition patterns among the released females. If all unhatched eggs could be attributed to a few uninseminated females, we would expect our conclusions to be unbiased. Alternatively, if a proportion of every female's egg batch failed to hatch, this could lead us to underestimate the number of containers used by ovipositing females. We suspect that the true explanation lies somewhere between these two extremes. Another limitation of our study was that we substituted the two large containers with two small containers under the post-intervention scenario, which is unrealistic for a dengue control campaign. We took this step to prevent confounding between the effects of targeting specific containers as opposed to reducing container abundance in general. Targeted larval control campaigns are specifically directed at the small subset of most productive containers, so we would not expect overall container abundance to change dramatically. For this reason, we were more interested in how females responded to the non-availability of large containers rather than a reduction in container numbers. Had we been able to conduct more trials inside the enclosure, we would have examined effects of container removal without substitution, as well as effects of varying Ae. aegypti female density in the household.
Finally, all females in this experiment were confined to the one household within the semi-field enclosure. This design precluded us from testing whether female oviposition choices would be different if they had access to multiple houses and different container types, as occurs naturally in the field. We had originally planned to address that question during a field validation in which we would release females into the field and search for their progeny in the release house as well as neighboring houses. Due to a dengue-4 epidemic in Iquitos during fall 2008 [63], [64], however, we were unable to release mosquitoes to conduct this field validation.
We do not dispute that larval Ae. aegypti control should be practiced or that interventions such as container elimination, larviciding, and biological control are more cost effective when targeted to the most productive containers [12]. We suggest, however, that targeted larval control alone should not be relied upon as the predominant strategy to prevent dengue transmission. Due to the complexity of Ae. aegypti ecology and the low population threshold densities required for dengue transmission [4], [5], a combination of multiple control measures (e.g., container elimination, egg sinks, autodissemination of insect growth regulators, lethal ovitraps, etc.) will likely be necessary to produce an epidemiologically significant change in vector abundance. For example, elimination of the most productive containers could be coupled with deployment of gravid traps or egg sinks [21], [65]. Such a combined strategy may encourage females to lay eggs in traps, either for themselves (gravid traps) or for their offspring (egg sinks), as well as minimize shifts in productivity to under-utilized containers. Regardless of the specific combination of tools used, successful integrated control strategies should be based on sound understanding of Ae. aegypti behavior and population dynamics.
|
10.1371/journal.pcbi.1000467 | Accelerated Immunodeficiency by Anti-CCR5 Treatment in HIV Infection | In 50% of progressing HIV-1 patients, CXCR4-tropic (X4) virus emerges late in infection, often overtaking CCR5-tropic (R5) virus as the dominant viral strain. This “phenotypic switch” is strongly associated with rapidly declining CD4+ T cell counts and AIDS onset, yet its causes remain unknown. Here, we analyze a mathematical model for the mechanism of X4 emergence in late-stage HIV infection and use this analysis to evaluate the utility of a promising new class of antiretroviral drugs—CCR5 inhibitors—in dual R5, X4 infection. The model shows that the R5-to-X4 switch occurs as CD4+ T cell activation levels increase above a threshold and as CD4+ T cell counts decrease below a threshold during late-stage HIV infection. Importantly, the model also shows that highly active antiretroviral therapy (HAART) can inhibit X4 emergence but that monotherapy with CCR5 blockers can accelerate X4 onset and immunodeficiency if X4 infection of memory CD4+ T cells occurs at a high rate. Fortunately, when CXCR4 blockers or HAART are used in conjunction with CCR5 blockers, this risk of accelerated immunodeficiency is eliminated. The results suggest that CCR5 blockers will be more effective when used in combination with CXCR4 blockers and caution against CCR5 blockers in the absence of an effective HAART regimen or during HAART failure.
| HIV has caused over 30 million deaths. The virus is so fatal because it infects and depletes CD4+ T cells, “helper” immune cells critical for orchestrating and stimulating the overall immune response. No one understands why, in about 50% of HIV infections, a more deadly strain emerges late in infection. The new HIV strain, known as X4, differs from its predecessor, known as R5, because X4 only infects CD4+ T cells displaying the receptor CXCR4, while R5 only infects CD4+ T cells displaying the receptor CCR5. Because CXCR4 and CCR5 are found on different CD4+ T cells, X4 depletes a second set of critical immune cells, accelerating immunodeficiency and death. Recently, the FDA began approving drugs that selectively block R5, and some researchers have touted anti-R5 therapy alone as a potentially safer alternative to current anti-HIV drugs. But an open question is whether anti-R5 treatments push HIV toward the more deadly X4 variant earlier. To understand how X4 emerges and how anti-R5 treatments affect X4, we apply a combination of mathematical analysis and simulation. An important medical result of our work is that anti-R5 treatment alone can accelerate X4 emergence and immunodeficiency. Our results suggest that anti-R5 treatment only be used with anti-X4 treatment or anti-HIV drug “cocktails,” which combat R5 and X4 equally.
| Left untreated, human immunodeficiency virus type-1 (HIV) generally targets and severely depletes a patient's CD4+ T cells over a period of up to 15 years, with a median AIDS onset time of 9.8 years [1]–[4]. HIV's infection of a CD4+ T cell begins when HIV's outer envelope protein gp120 binds to a CD4 receptor and subsequently binds to one of two chemokine coreceptors, CCR5 or CXCR4 [5],[6]. Viral-coreceptor binding exposes a second viral envelope protein, gp41, which mediates fusion of the viral and target-cell membranes, allowing HIV to inject its retroviral material into the cell. HIV strains that use CCR5 as a coreceptor are termed R5 viruses, while those that bind CXCR4 are called X4 viruses.
R5 virus is predominant during early infection where X4 virus has rarely been observed, independent of the route of viral transmission [5], [7]–[9]. Importantly, X4 alone is generally unable to infect humans: individuals homozygous for a 32 base-pair deletion in CCR5, CCR5Δ32, are almost entirely immune to HIV [5]. However, in approximately 50% of progressing HIV patients, a ‘phenotypic switch’ occurs wherein X4 virus emerges late in infection, overtaking R5 virus as the dominant viral strain. The R5-to-X4 switch is strongly associated with a poor clinical prognosis for the patient: it occurs with a steep loss in CD4+ T cell counts and accelerated AIDS onset.
The mechanisms causing R5's early dominance and the subsequent R5-to-X4 switch are poorly understood, however multiple lines of evidence suggest that CCR5's higher cell-surface density on activated and recently activated memory CD4+ T cells enable R5 to infect more of this crucial cellular population than X4. CCR5's cell-surface density has been shown to determine the efficiency of R5 infection [10], possibly because multiple CCR5 receptors act in a cooperative, concentration-dependent manner to facilitate infection [11]. R5 virus' level of infection is thus highest among CD62L− effector memory CD4+ T cells [12], where CCR5's cell surface density is highest. CXCR4's cell-surface density is similarly positively correlated with X4's emergence [13], but CXCR4's per-cell density on memory CD4+ T cells is lower than that of CCR5 [14], giving R5 an advantage on these cells. On dually-positive CCR5+, CXCR4+ CD4+ T cells, the coreceptors compete for association with CD4 [15], which should lend R5 an advantage given CCR5's higher per-cell surface density on dually-positive cells [14].
Thus, R5 virus' early advantage may stem from CCR5's greater per-cell surface density on activated and recently activated ‘effector’ memory CD4+ T cells [14],[15]. These ‘effector’ memory CD4+ T cells are the crucial virion-producing populations as evidenced by snapshots taken during SIV infection, which show approximately five times as many virions surrounding infected, activated effector memory CD4+ T cells as around infected, quiescent CD4+ T cells [16]. Moreover, Li et al. show that CD4+ T cells positive for Ki67 (a marker that is displayed after late G1 cell-cycle progression and indicates T cell ‘activation’) produce over 90% of the virions during the chronic phase of SIV infection [17]. This may also explain why X4 has trouble initiating infection when R5 virus is absent: CXCR4's per-cell density on the most crucial memory CD4+ T cell population is simply too low [14].
The perplexing question underlying the R5-to-X4 phenotypic switch is therefore: how does a switch to X4 occur if R5 virus is simply better at infecting memory CD4+ T cells?
Since the R5-to-X4 switch only occurs during late infection, it is reasonable that there exists an early selection pressure in favor of R5 virus, which is mitigated over the course of infection. In support of this hypothesis, Ribeiro and colleagues recently proposed the idea that increasing target-cell activation over the course of dual infection causes X4 to eventually outcompete R5 [18].
A critical prediction of the Ribeiro model is that CCR5 blockers (small-molecule pharmaceuticals that bind CCR5 and thereby obstruct R5 virus' ability to infect a CD4+ T cell) successfully reduce overall viral loads, decrease cellular activation levels, and inhibit X4 emergence. This prediction is critical since a central question is whether CCR5 blockers lend X4 virus an advantage and promote clinically deleterious switches to X4 during dual R5 and X4 infection. However, in vivo trials of the CCR5 inhibitors CMPD 167 and maraviroc showed CCR5 blockers actually increasing X4 viral loads and decreasing R5 viral loads (approximately reciprocally) in dually-infected patients [19],[20].
Given the recent CCR5 clinical trial data, we analytically probed how changing target cell activation levels could produce a switch and whether such models could account for documented increases in X4 viral load after anti-CCR5 treatment. Our model builds upon [18], but in our generalized setup the R5-to-X4 switch can occur even if the fraction of activated naïve CD4+ T cells increases at a slower rate than the fraction of activated memory CD4+ T cells. In this more general setting, we rigorously show how the R5 to X4 switch occurs and find that CCR5 blockers often do accelerate X4's emergence and attendant immunodeficiency. Fortunately, the results also show that when CXCR4 inhibitors or HAART are given along with CCR5 inhibitors, X4 emergence is unlikely to be accelerated and is instead often delayed.
In the following three models, all variables are capitalized and represent concentrations per microliter (1/µl). Specifically, in Model 1, T represents the concentration of uninfected CD4+ T cells, and (without loss of generality) is given an initial value of 1000 CD4+ T cells/µl. In Models 2 and 3, T is split into uninfected naïve (N) and memory (M) subpopulations, each with an initial value of 500 CD4+ T cells/µl. I4 and I5 reflect the concentrations of abortively, latently, and productively infected CD4+ T cells by X4 and R5, respectively, and in Model 3, we analogously define N4, M4, M5 (see below). V4 and V5, each given initial values of 1000 virions/ml, represent X4 and R5 viral load concentrations.
Defining the parameters, λ is the rate of thymus production of CD4+ T cells and has units cells/(µl•day), k4 and k5 are the respective infection rate coefficients for X4 and R5 infection of CD4+ T cells and have the units µl/(virions•day). All remaining parameters have units 1/day. These include: dT, the death rate of uninfected CD4+ T cells in Model 1, set to λ/T0 to allow for steady-state pre-infection, and dn and dm the analogous death rates of uninfected CD4+ T cells in Models 2 and 3, also set so that equilibrium exists pre-infection. Additionally, δ is the death rate of infected CD4+ T cells, p is the rate of viral production by activated infected cells, and c is the viral clearance rate. an and am are required to satisfy Equation 2 (below) and represent the fractions of activated naïve and memory CD4+ T cells as a function of CD4, the total number of uninfected and infected CD4+ T cells per microliter. Thus, in Models 1 and 2, CD4 = T+I4+I5, and, analogously, in Model 3 CD4 = N+M+N4+M4+M5. Since the total concentration of CD4+ T cells changes over time, an and am vary over the course of infection.
Because over 99% of infected cells are defectively infected [21] and because such non-productively infected cells are indistinguishable from uninfected cells, we make the simplifying assumption that an and am also approximate the fractions of infected naïve and memory CD4+ T cells that are activated. Thus, in Models 1 and 2, anI4 and amI5 represent the concentrations of activated X4 and R5 infected cells, respectively. Analogously, in Model 3, anN4, amM4, and amM5 represent the concentrations of activated X4-infected naïve, X4-infected memory, and R5-infected memory CD4+ T cells, respectively. In our models, it is only these activated subpopulations of infected cells that produce virions. We thus multiply the concentration of activated infected cells (e.g., amM5) by p, the rate of viral production (per-day) from a productively-infected (i.e., activated and infected) cell, yielding the respective total concentration of virions produced each day by a given infected cell type.
We first extended the basic model of viral dynamics [1],[2] to two viral strains, to test whether this simplified, one-compartment model can generate a representative R5-to-X4 switch.(Model 1)Here we make X4's viral production dependent on the fraction of activated naïve CD4+ T cells an, but not on am. One reason for this simplification is that R5 out-competes X4 for dually-positive memory CD4+ T cells [22]. Furthermore, the vast majority of CXCR4-positive T cells are in the naïve subset, where CXCR4's cell surface density is also highest [14]. Since Model 1 lumps all CD4+ T cells into a single target-cell compartment, and because across all lymphocytes CXCR4's median per-cell surface density is almost three times as high as that of CCR5 [14], we also assume k4>k5. As above, this does not imply that X4 productively infects more target cells than R5 at the beginning of infection, since very few naïve cells are activated early in infection [23]. Importantly, given the simplifications employed, the purpose of Model 1 is not to represent the actual dynamics of coreceptor tropism in HIV infection, but to rigorously explore an activation-based R5 to X4 switch in the simplest setting.
To account for the fact that in reality naïve and memory CD4+ T cells are disjoint target cell compartments, we subsequently build upon Model I and divide T into N and M.(Model 2)The equations in this system are analogous to those in Model 1 but the uninfected CD4+ T cell population is now split into uninfected naïve (N) and memory (M) subpopulations. Additionally, f is defined to be the fraction of naïve cells activated via the conventional Ag-TCR interaction, which divide and differentiate into CD45RO+ memory cells. The rest of the activated cells are assumed to have been upregulated via cytokines or other Ag-TCR independent processes and thus remain phenotypically naïve (CD45RA+) [24]–[26]. We note that non-Ag mediated activation of naïve CD4+ T cells is not absolutely necessary for our models' primary conclusions of strain coexistence and phenotypic switching at clinically-representative time-points (i.e., 3–6 years post-infection); we include this activation term for the added realism it brings to the model.
In Model 2, X4 is only able to infect naïve CD4+ T cells, a simplification we employ because of the data in [14] showing that the per-cell density of CCR5 is significantly higher than that of CXCR4 on memory CD4+ T cells. Moreover, a recent paper finds that on dually-positive CXCR4+, CCR5+ CD4+ T cells, R5 generally outcompetes X4 [22], arguably because of CCR5's higher surface density [15]. Finally, naïve CD4+ T cells have been found to be preferentially depleted during X4 infection [27].
Because in practice X4 actually infects both naïve and memory CD4+ T cells, in our final model, Model 3, we extend the two-compartment setup of Model 2 to allow X4's infection of memory CD4+ T cells:(Model 3)In this model, kN4 and kM4 are the infection rate coefficients of X4 on naïve (N) and memory (M) CD4+ T cells, respectively, and kM5 is the infection rate coefficient of R5 on memory CD4+ T cells. kN4, kM4, and kM5 all have units µl/(virions•day). N4 and M4 are the concentrations of abortively, latently, and productively infected naïve and memory CD4+ T cells, respectively, by X4 virus, and M5 is the concentration of abortively, latently, and productively infected memory CD4+ T cells by R5 virus. All other parameters, variables, and initial conditions have been defined above. Because CCR5 is far more strongly expressed on memory CD4+ T cells than is CXCR4 [14], we set kM5≫kM4. Conversely, CXCR4 is more highly expressed on naïve CD4+ T cells than it is on memory CD4+ T cells [14], making kN4≫kM4.
In single target-cell compartment susceptible-infectious (SI) models such as Model 1, the ecological principle of competitive exclusion generally applies [31]. Thus, while Model 1 can produce an R5 to X4 switch in a clinically representative timeframe, it necessarily manifests competitive exclusion (Fig 1a). The lack of steady-state coexistence in Model 1 is significantly different from data, which show long-term coexistence of R5 and X4 variants in post-switch individuals [32]. Moreover, X4's emergence late in infection—well into quasi-steady state—is very difficult to achieve in this single compartment framework, because X4 could have been rendered extinct via competitive exclusion prior to the late-stage switch (Weinberger and Perelson, manuscript in preparation).
In order to prevent the species with the higher effective reproductive ratio from dominating exclusively, which contradicts observed results [32], Model 2 splits the target cell population into naïve and memory CD4+ T cells, and, for simplicity, assumes that X4 solely infects naive cells and that R5 only infects memory cells. The dual-target cell compartment nature of Model 2 makes coexistence possible (Weinberger and Perelson, 2009, manuscript in preparation). Thus, while Model 2 can also produce an R5-to-X4 switch at a clinically representative time, it is able to maintain R5 and X4 coexistence post-switch (Figure 1B).
However, CCR5 inhibition cannot produce a transient increase in X4 (Figure 1C). This is because in models that restrict X4 and R5 to infecting distinct target cell populations (e.g., Model 2), X4 does not infect any of the (memory) CCR5+ T cells that are made refractory to R5 infection by CCR5 inhibition. Quantitatively, Eq. (2) stipulates an′(CD4)<0, so when CCR5 is inhibited and memory CD4+ T cell counts rise, an decreases and the rate of viral production from an X4-infected cell (p*an) is lowered. Due to the lack of competition, the number of X4-infected cells does not increase to compensate for the decreased per-cell virion production rate, so X4 viral loads decrease (Figure S1). This result is in contrast to recent studies on dually-infected rhesus macaques and humans, which demonstrate clear increases in X4 virus after R5 virus is selectively suppressed through the use of a CCR5 inhibitor [19],[20]. To produce a temporal X4 increase upon R5 inhibition and to maintain coexistence in contradistinction to Model 1, we need a multi-compartment model where X4 infects both naïve and memory CD4+ T cells.
In Model 3, our final and most biologically detailed model, we include naïve and memory CD4+ T cell compartments. Since CXCR4 is found on a large number of memory CD4+ T cells, we allow for X4's infection of memory as well as naïve CD4+ T cells (Figure 2A). Thus, Model 3 serves as a union of the two previous models: it includes the X4 and R5 strain competition found in Model 1 and it also includes the separate target cell compartments of Model 2, which allowed for X4's persistence prior to a switch and the coexistence of strains afterward. Model 3 produces X4-to-R5 switches at clinically representative times of 1000–2000 days and also maintains coexistence post-switch in two types of parameter regimes, the “non-competitive” and “competitive” regimes, whose distinctions are elaborated upon below (Figure 2B).
Given that X4 and R5 viruses can coexist in the disjoint two-compartment model, it is reasonable to conjecture that coexistence is a feature of this extended model as well. To show this formally, we define Reff4 and Reff5 to be the effective reproductive ratios of X4 and R5 virus, respectively, which are given by:(3)The effective reproductive ratio, Reff, is thus a time-dependent function for the average number of infected cells produced by an average infected cell at a given point, t, in time. Reff generalizes R0, the basic reproductive ratio, which evaluates the average infectivity only at the initial time point.
Solving the necessary and sufficient conditions for an R5-to-X4 switch, d/dt(V4(t*))>d/dt(V5(t*)) and V4(t*) = V5(t*), we see that a switch occurs in Model 3 if and only if:(4)But am>an for all time, so, in particular, at the switch time t* we have an(CD4(t*))/am(CD4(t*))<1. The right-hand side of Equation (4) must therefore be less than 1, meaning that at the switch point N4(t*)+M4(t*)>M5(t*). Thus, at the switch point t* there are more X4-infected CD4+ T cells than R5-infected CD4+ T cells. This implies that X4 had a higher effective reproductive ratio at some earlier point, t**. However, Reff4 (t**)>Reff5 (t**) does not imply that Reff4 (t)>Reff5 (t) for all t>t**: Equation (3) implies that when N decreases faster than M and when the resulting decrease to N/M is less than the increase to an/am, Reff5 increases relative to Reff4 (i.e., Reff4/Reff5 decreases). But the condition for steady-state coexistence of X4 and R5 is Reff4 = Reff5, so by enabling Reff5 to rebound relative to Reff4 post-switch, the dual-compartment nature of Model 3 makes coexistence possible.
We can grasp the switch threshold in (4) more easily by substituting in the particular equations of (1), yielding the following switch condition (where the right-hand side is positive):(5)Importantly, Equations (3) and (5) imply that, with the exception of changes to kM5, modulating parameters to accelerate CD4+ T cell decline hastens an R5 to X4 switch while changing parameters to mitigate CD4+ T cell decline hinders a phenotypic switch. Thus, successful antiretroviral therapy will generally inhibit X4's emergence. However, because R5 and X4 are now in competition, CCR5 inhibitors generate more complicated kinetics.
CCR5 inhibitors decrease kM5, causing R5's viral load to decline, and, as a result, memory CD4+ T cell counts to increase. The question we sought to answer is whether X4 infects sufficiently many of these R5-immune memory CD4+ T cells to counteract the increase in CD4+ T cells from CCR5 inhibition. We hypothesized that X4's ability to infect memory CD4 T cells would depend on kM4, and with a sufficiently large kM4 (the “competitive regime”), X4 would infect a non-negligible fraction of newly R5-immune cells which and an increase in X4 would ensue. The temporal increase in X4 viral loads would thus cause greatly increased X4 infection of naïve CD4+ T cells, which yields accelerated naïve CD4+ T cell depletion. Indeed, numerical simulations (Text S1 contains more information on how these simulations were done) show that successful CCR5 blockage results in accelerated AIDS onset across much of parameter space (Figure 2C). This result does not change when Model 3 is extended to include the loss of virions due to the infection of new target cells (Figure S2). Importantly, the early immunodeficiency after effective CCR5 blockage is due to accelerating X4 emergence and increasing X4 viral loads as the efficacy of CCR5 inhibition increases in the “competitive regime” (Figure 2D).
Conversely, if kM4 is sufficiently small (the “non-competitive regime”), X4 does not infect a sufficient number of dually-positive memory CD4+ T cells upon CCR5 blockage. This causes the uninfected memory CD4+ T cell population to increase during anti-CCR5 therapy, yielding a drop in an/am and hindering a potential switch to X4 as well as immunodeficiency (Figure 2C, small kM4 regime). The latter result is to be expected from Model 2, because a weak kM4 can be approximated by a complete lack of competition. Thus, a single parameter, kM4, controls the efficacy of anti-CCR5 therapy in dually infected HIV patients, highlighting the need for circumspection in prescribing these treatments.
Given that CCR5 inhibitors accelerate R5-to-X4 switching and immunodeficiency across the wide swath of parameter space in which kM4 is relatively large, the question arises as to whether CCR5 inhibitors are similarly deleterious when used in conjunction with CXCR4 inhibitors, which reduce kM4. Simulations show that adding a CXCR4 inhibitor with an efficacy of at least 5% is sufficient to prevent accelerated AIDS onset in the “competitive regime” (Figure 3A). Because X4 emergence is due to an increase in the relative fraction an/am of activated naïve to memory CD4+ T cells, we also simulated whether a generic antiretroviral therapy such as HAART, which increases CD4+ T cell counts and reduces an/am, also prevents the accelerated X4 emergence that CCR5 inhibitors can engender. The results of dual-treatment with HAART and CCR5 inhibitors are analogous to those shown for dual-treatment with CXCR4 and CCR5 inhibitors, proving that a relatively modest additional HAART therapy (with an efficacy above 7%) obviates the risk of CCR5 inhibitors accelerating immunodeficiency in the “competitive regime” (Figure 3B). Finally, generalizing across kM4 and kM5 parameter space shows that when treatment efficacies are sufficiently strong (e.g. 80% efficacies) dual treatment with CXCR4 inhibitors does not accelerate immunodeficiency relative to untreated individuals (Figure 3C). Similarly, dual-treatment with CCR5 inhibitors and HAART does not accelerate immunodeficiency relative to untreated individuals (Figure 3D).
Here we present a mathematical model of dual-strain R5 and X4 HIV in vivo dynamics and show that CCR5 inhibitors can accelerate the emergence of X4 virus and immunodeficiency. Two equivalent R5-to-X4 switch conditions were found: either the ratio of the relative fractions of activated naïve and memory CD4+ T cells (an/am) must surpass a threshold (Eq. 4) or, equivalently, CD4+ T cell counts must drop below a critical value (Eq. 5). The resultant “phenotypic” switch yields a drastic loss in CD4+ T cell counts, due to X4's depletion of R5-immune naïve CD4+ T cells. Of significant clinical importance, our results show that, across much of parameter space, CCR5 inhibitors may force an early switch to X4 virus, greatly accelerating CD4+ T cell depletion and AIDS onset. However, CCR5 inhibitors do not appear to have the deleterious effect of accelerating X4 emergence and immunodeficiency when they are used in conjunction with CXCR4 inhibitors or HAART.
The result that CCR5 blockers alone may promote X4 emergence is supported by data from a study on dually-infected rhesus macaques injected intravenously with the CCR5 inhibitor CMPD 167 [19]. After beginning treatment, two out of three primates manifested a transient increase of several logs in X4 viral load, essentially canceling the decrease in R5 viral load. Moreover, a clerical error in a recent study on the effect of the CCR5 inhibitor maraviroc on R5-only patients resulted in a dually-infected patient mistakenly being included in the trial [33]. That patient saw no change in total viral load as the X4 viral load increased upon CCR5 inhibition [20].
While CCR5 inhibitors alone may accelerate X4 emergence and AIDS onset, there is still good reason to consider their utility as part of a multi-therapy cocktail. Recent clinical data from the MOTIVATE 1&2 trials show that CCR5 inhibitors together with optimized background therapy yield larger increases in CD4 counts and larger reductions in viral loads when compared with optimized background therapy alone [34]. Our model simulations strongly support this result, showing that across much of parameter space, employing CCR5 inhibitors together with HAART lengthens the time to AIDS when compared with the time to AIDS under HAART alone (Figure S3).
But even if CCR5 inhibitors are a helpful component in a diversified anti-HIV therapy, one has to wonder about the greater immunological cost associated with blocking this chemokine receptor. A recent meta-population analysis of West Nile Virus (WNV) prevalence in four US states found that CCR5Δ32 homozygotes are approximately four times more likely to develop symptomatic WNV as are those with the wild-type receptor [35]. Previous murine models have suggested a mechanism by which CCR5 confers protective advantage against symptomatic WNV: CCR5 may promote the transfer of leukocytes to a WNV-infected individual's brain, aiding in immune control of encephalitis [36]. CCR5's potential protective advantage against symptomatic WNV may also help explain why CCR5Δ32/Δ32 is relatively common (5–14%) among European Caucasian cohorts, but near absent in African populations [37], the latter being at a far greater risk of contracting WNV.
Additionally, it is important to consider the prospect that CCR5 inhibition may lead to HIV evolving to bind to an entirely new coreceptor during early infection. A precedent for the evolution of new lentiviral coreceptor tropisms exists: the SIV endemic to red-capped mangabeys (RCMs) can utilize CCR2b rather than CCR5 [38]. This is likely because a large percentage (estimated at over 80%) of RCMs are homozygous for a 24 base-pair deletion in the gene for CCR5, and CCR5Δ24/Δ24 cells cannot be transfected with R5 virus [38]. The ability of SIVrcm to use CCR2b occurs despite almost all other known SIVs utilizing CCR5 exclusively in vivo [5]. Δ24 appears to be an ancient deletion: it has been found in both red-capped mangabeys and sooty mangabeys, species which diverged more than 10,000 years ago [38]. It is therefore possible that in the long-run HIV may evolve entirely new coreceptor usages in response to coreceptor inhibition.
|
10.1371/journal.pbio.1000399 | Cortical Overexpression of Neuronal Calcium Sensor-1 Induces Functional Plasticity in Spinal Cord Following Unilateral Pyramidal Tract Injury in Rat | Following trauma of the adult brain or spinal cord the injured axons of central neurons fail to regenerate or if intact display only limited anatomical plasticity through sprouting. Adult cortical neurons forming the corticospinal tract (CST) normally have low levels of the neuronal calcium sensor-1 (NCS1) protein. In primary cultured adult cortical neurons, the lentivector-induced overexpression of NCS1 induces neurite sprouting associated with increased phospho-Akt levels. When the PI3K/Akt signalling pathway was pharmacologically inhibited the NCS1-induced neurite sprouting was abolished. The overexpression of NCS1 in uninjured corticospinal neurons exhibited axonal sprouting across the midline into the CST-denervated side of the spinal cord following unilateral pyramidotomy. Improved forelimb function was demonstrated behaviourally and electrophysiologically. In injured corticospinal neurons, overexpression of NCS1 induced axonal sprouting and regeneration and also neuroprotection. These findings demonstrate that increasing the levels of intracellular NCS1 in injured and uninjured central neurons enhances their intrinsic anatomical plasticity within the injured adult central nervous system.
| Following trauma to the central nervous system (brain or spinal cord), neurons show very little capacity to re-grow their axons, which can lead to a permanent loss of function in those regions. In this study, we show that this failure of axon re-growth is associated with low intracellular levels of a small molecule called neuronal calcium sensor-1 (NCS1). We modified a non-replicating virus in two ways so as to increase the level of NCS1 in neurons while simultaneously labelling them with a green fluorescent protein, which allowed us to track neuronal growth. Using this virus to increase the level of NCS1 in a particular group of neurons that communicate between the brain and spinal cord, we showed that new axonal growth occurred in the spinal cord with or without injury to the neurons. Electrophysiological assessments demonstrated that these new processes formed functional connections in the spinal cord, and behavioural experiments revealed that this recovery also helped the animals move their limbs more effectively. Furthermore, an increase in NCS1 protected these neurons, such that fewer of them died after the injury. These findings demonstrate that increasing the intracellular levels of NCS1 in neurons can aid in the recovery from central nervous system injury, and can help improve behavioural function.
| Spinal cord injury is a significant clinical problem that produces life long disability, although in a minority of patients some degree of recovery can occur spontaneously without any therapeutic intervention [1],[2]. There are several possible mechanisms that could be responsible for this, one being anatomical plasticity, but such plasticity is very limited [3]–[5].
There is a growing literature suggesting pharmacological interventions can enhance both axonal regeneration [6]–[9] and anatomical plasticity [10]–[14] within the spinal cord, but little is known about the intracellular mechanisms responsible for such plasticity.
Recently, we have found that following injury, the lentiviral overexpression of retinoic acid receptor β2 (RARβ2) induces regeneration in sensory and central axons [15],[16]. Microarray analysis of CNS tissue transduced with overexpressing RARβ2 lentivector was carried out to identify the intracellular molecular pathways involved in such regeneration. In unpublished data, this analysis revealed a significant upregulation of neuronal calcium sensor-1 (NCS1) in the transduced tissue as confirmed immunohistochemically and by real-time PCR.
NCS1 is highly conserved across species and emerges as a key intracellular calcium signalling component in a number of regulatory pathways in neurons [17],[18]. This small molecule has been implicated in neuronal survival [19], short-term synaptic plasticity [20], and enhanced synapse formation and transmission [21]. Recently, it has also been suggested to regulate neurite outgrowth in pond snails [22],[23] and in primary cultured embryonic chick dorsal root ganglia neurons [24].
Here we show using lentiviral vectors that NCS1 overexpressed in primary cultured adult cortical neurons increases neurite sprouting. Following corticospinal tract (CST) denervation by unilateral pyramidotomy, axons of uninjured corticospinal neurons (CSN) overexpressing NCS1 sprout across the midline to form functional connections in the CST-denervated spinal cord. In axotomized CSN, overexpression of NCS1 induces axonal sprouting and regeneration and also neuroprotection. These studies reveal NCS1 as an important intracellular molecule for promoting anatomical plasticity following CNS injuries in the adult.
To transduce adult cortical neurons with NCS1 at high efficiency and to enable visualisation with an extrinsic marker, we constructed a minimal human immunodeficiency virus (HIV) based lentiviral vector expressing NCS1 and GFP under cytomegalovirus (CMV) and spleen focus-forming virus (SFFV) promoters, respectively, termed HIV-GFP-NCS1 (Figure 1A). A control vector termed HIV-GFP was used that expressed only GFP under the CMV promoter (Figure 1B). It has been shown that although CMV is a stronger promoter than SFFV in GFP expression, the percentage of transduced neurons with GFP expression was similar [25]. The level of NCS1 in transduced primary adult rat cortical neurons measured immunocytochemically was more than 5-fold greater than in the control HIV-GFP-transduced neurons (Figure 1C–1I). In control-transduced neurons, few sprouts were observed from cell bodies or neurites (Figures 1C–1E, 1J–1K, and S1). In contrast, neurons transduced with HIV-GFP-NCS1 showed significant increases in the numbers of sprouts from both cell bodies and neurites (Figures 1F–1H, 1J–1K, and S1). The GFP immunostained cortical cells were confirmed as neurons by co-immunopositive staining with the neuronal growth associated marker GAP43 (Figure 1L). Furthermore, although two different promoters to drive GFP were used, the adequacy of GFP immunostaining for neurite distribution in both the NCS1- and GFP-transduced groups was confirmed to be similar to that obtained with phalloidin staining (Figure 1M–1N).
The type of neurite that has undergone sprouting with NCS1 overexpression was further investigated using the specific dendritic immunomarker microtubule-associated protein 2 (MAP2). MAP2 has been previously shown strongly and weakly to immunolabel dendrites and axons, respectively [26],[27]. In control HIV-GFP-transduced neurons, few sprouts were observed in both dendrites and axons (Figure 2A–2F). In contrast, a significant increase in the number of sprouts on both dendrites and axons was observed in HIV-GFP-NCS1-transduced neurons (Figure 2G–2N).
These data indicate that primary cultured adult cortical neurons overexpressed with NCS1 have significantly more neurites and sprouts from dendrites and axons than the control neurons.
NCS1 has been shown to induce neuronal survival via the activation of the PI3K/Akt pathway [19]. We investigated whether this downstream intracellular pathway was also involved in the NCS1-induced neurite sprouting in primary cultures of adult mammalian neurons. The level of phospho-Akt in NCS1-transduced cortical neurons was significantly higher than in the control GFP-transduced group (Figures 3A–3D, 3I, and S2). Blockade of PI3K/Akt pathway with the inhibitor LY294002 caused a significant decrease in levels of phospho-Akt in the NCS1-transduced cortical neurons (Figures 3E–3F, 3I, and S3). This decrease corresponded to a significant 2-fold reduction in number of neurites from cell bodies and a 5-fold reduction in sprouts from neurites (Figure 3A–3K). However, neither phospho-Akt expression nor neurite sprout number was significantly changed in GFP-transduced neurons treated with LY249002 compared to vehicle treatment (Figures 3G–3K and S3).
These data indicate the levels of phospho-Akt are elevated in neurons with overexpressed NCS1 and blockade of Akt production reduces neurite sprouting in these neurons.
In Western blots, the levels of NCS1 were significantly higher in cortical neurons transduced with HIV-GFP-NCS1 compared with the controls both in vitro and in vivo (Figure 4A–4B, 4E–4F). This increase in NCS1 corresponds with a significant increase in phospho-Akt levels (Figure 4C–4D, 4G–4H). In the presence of LY249002, no significant change in NCS1 level occurred in NCS1-transduced cortical neurons compared to the controls (Figure 4A–4B, 4E–4F). This demonstrates that the reduction in phospho-Akt level was a direct result of LY249002 and not via the NCS1 overexpression itself. These data show that the neurite sprouting induced by NCS1 was indeed via the PI3K/Akt pathway.
With the demonstration that NCS1 overexpression induces neurite sprouting in primary cultured adult cortical neurons, it was next determined whether this also occurred in vivo. HIV-GFP-NCS1 or control HIV-GFP lentivector was injected into the forelimb and hindlimb regions of the left sensorimotor cortex of adult rats. High efficiency was achieved of both GFP and NCS1 expression in the CSN at 3 wk with HIV-GFP-NCS1 (Figure 5A–5C). Similar numbers of GFP labelled neurons in layer V of the sensorimotor cortex were observed in both transduced groups as detected by immunohistochemistry (Figure 5D). Within such GFP labelled neurons, the percentage with NCS1 positive immunostaining was significantly higher in NCS1-transduced neurons than in the control GFP-transduced neurons (Figure 5D). Axons from NCS1-transduced CSN were visible in the pyramidal tract with GFP immunostaining (Figure 5E), in the main dorsal component of the CST, and in its collaterals at the cervical cord level (Figure 5F–5I). These data show that GFP allows a precise identification both of co-labelled transduced neurons overexpressing NCS1 and of their axons, thus obviating the need for later labelling with neuronal tracers or by the use of an independent virus expressing LacZ or GFP.
Adult Wistar rats received unilateral intracortical injections of either HIV-GFP-NCS1 (NCS1-transduced) or the HIV-GFP (control) lentivector 3 wk before receiving on the contralateral side a unilateral pyramidal tract lesion which, in turn, causes CST-denervation of the contralateral side of the spinal cord (Figures S4A and S5A–D). The lesion site was defined with the astrocytic marker GFAP and the loss of PKCγ immunostaining caudal to the lesion site (Figures S5C–D and S8). At 6 wk post-CST-injury, GFP-immunostaining was performed to define axon collateral sprouting from the intact CST at the cervical and lumbar levels and particularly into the CST-denervated side of the spinal cord. The number of GFP-labelled axons in the CST was not significantly different between the control and NCS1-transduced rats (Figure 6A). In control rats, GFP-labelled collaterals were present in the CST-innervated side of the cord but few in the CST-denervated side at the cervical (Figures 6B–6F, 6M–6N, and S6) and lumbar (Figure 7A–7E) levels. In NCS1-transduced rats, GFP-labelled collaterals were present in the CST-innervated side, with a significant increase in the peak number of GFP positive fibres at the mediolateral region (Figures 6G–6I, 6N, 7F–H, 7L, and S7). More importantly, a significant increase also occurred in the number of GFP positive fibers sprouting across the midline into the CST-denervated cord. At the cervical level, GFP positive fibers in the range of 1–2 fibers per section for the control group compared to 4–5 fibers for the NCS1-transduced rats. At the lumbar level, GFP positive fibers of no more than 1 fiber per section for the control group compared to 5–6 fibers for the NCS1-transduced rats. This significant difference was maintained for up to 850 µm and 350 µm from the midline at the cervical and lumbar level, respectively (Figures 6J–6M, 7I–7K, and S7). The completeness of the tract lesions was confirmed by PKCγ immunostaining in the spinal cord (Figure S8). These data show that overexpression of NCS1 in CSN at the cortical level can induce distal axon collateral sprouting across the midline into the CST-denervated side of the cervical and lumbar spinal cord.
It has been shown that a pyramidal tract lesion in adult hamster causes CSN to become atrophied after 2 wk post-injury [31]. To investigate whether NCS1 overexpression can prevent adult axotomized CSN from atrophy, adult Wistar rats received unilateral intracortical injections of either HIV-GFP-NCS1 or the control HIV-GFP lentivector 1 wk before an ipsilateral pyramidal tract lesion at the medullary level. To identify CSN, the retrograde Fast Blue tracer was injected directly into the lesion site immediately after sectioning (Figure S4D). Controls were unlesioned rats with Fast blue injected into the pyramidal tract at the medullary level with no intracortical lentiviral injection. After 2 wk post-lesion, the axotomized CSN in control GFP-transduced rats that have low NCS1 levels exhibited significant cell soma shrinkage compared to the large and healthy CSN in unlesioned rats (Figure 12A–12F and 12J–12K). In contrast, the axotomized CSN in NCS1-transduced rats that have high NCS1 levels did not exhibit significant cell soma shrinkage compared to the CSN in unlesioned rats (Figure 12G–12K).
These data suggest NCS1 overexpression in CSN exerts a neuroprotective effect on axotomized CSN.
This present study demonstrates that the intracellular levels of NCS1 in adult cortical neurons can be significantly elevated by transduction with a lentiviral vector. In culture, neurons overexpressing NCS1 develop extensive neurite sprouting which by immunocytochemistry and Western blotting was shown to be via Akt phosphorylation. Similarly, analogous experiments conducted in vivo show that CSN overexpressing NCS1 with intact CST axons can undergo distal collateral sprouting and cross the midline into the CST-denervated side of the spinal cord. This anatomical plasticity is also functional as demonstrated by the behavioural and electrophysiological outcomes in NCS1-transduced adult rats. Furthermore, studies on the axotomized CSN show that NCS1 overexpression not only induces axonal sprouting and regeneration at the lesion site but also exerts a neuroprotective affect on injured CSN.
To date, several therapies for spinal cord injury models have shown both significant axonal regeneration [6]–[9] and anatomical plasticity [10]–[14] within the spinal cord. However, the intracellular mechanisms for these therapies have been little investigated. Only the purine-sensitive ste20-like protein kinase (Mst3b) has been linked to the anatomical plasticity observed with the purine nucleoside inosine [11],[26]. Interestingly, Mst3b has been shown selectively to induce outgrowth only from axons and not dendrites [26]. Conversely, in the present study, NCS1 overexpression induces sprouting from both axons and dendrites in cultured neurons, suggesting the growth induction process of NCS1 is non-selective. However, despite the existence of morphologically and molecularly distinct differences between dendrites and axons, neurons have been shown to have the capacity to generate axons from dendrites [32].
In addition to observing neurite sprouting in vitro, we demonstrated an NCS1 mediated axon collateral sprouting in vivo following unilateral pyramidotomy. The HIV-GFP-NCS1 lentivector injected into the cortex enabled GFP labelling of neurons overexpressing NCS1. The GFP labelling allowed visualisation and quantification of sprouting of the CST axons without the need of applying an independent tracer, and such labelling of axons and their collaterals can be detected as far distal as the lumbar region.
We report for cervical region the number of GFP-positive collateral fibers from uninjured CST axons that have crossed the midline into the CST-denervated side of the spinal cord, as measured per 40 µm section, is 1–2 for the control and 4–5 for the NCS1-transduced rats. It is of interest to consider the possible total number of crossing fibers over the relevant cervical region. The length of the adult rat C5–C8 cervical cord containing the majority of forelimb motoneurons is approximately 11 mm [33],[34]. Extrapolating using these data, the NCS1-transduced rats would have approximately over 800 additional collaterals to account for the functional plasticity demonstrated behaviourally and electrophysiologically in these rats. Furthermore, this recovery was shown to be dependent on the collaterals provided by the intact CST as demonstrated by its loss following the second pyramidotomy as well as the loss of the crossed EMG activity when the intact CST axons were sectioned in the terminal experiment.
We have previously induced axonal regeneration by lentivector overexpression of retinoic acid receptor β2 (RARβ2) [15],[16]. In this study, NCS1 overexpression has been demonstrated in axotomized CSN to induce axonal sprouting and regeneration. Recently, we showed that the retinoic acid receptor beta agonist (CD2019) overcomes inhibition of axonal outgrowth via the PI3/Akt pathway in injured adult rat spinal cord [35]. From our unpublished data, lentivector overexpression of RARβ2 also induces an upregulation of NCS1 as initially detected by microarray analysis and confirmed immunohistochemically and with real-time PCR. Thus this present study suggests that upregulation of NCS1 is a major intracellular target linking RARβ2 to the PI3K/Akt pathway in inducing anatomical plasticity and axonal regeneration. A similar explanation may account for GDNF-induced anatomical plasticity as other studies have shown NCS1 is upregulated by GDNF [14],[19],[36].
The successful regenerative responses of CSN after pyramidotomy with delayed post-injury NCS1 overexpression suggests axonal sprouting and regeneration can occur without the need to prime with overexpression of NCS1. These data provide promising support for NCS1 overexpression as a possible therapeutic treatment for CNS injury in a clinical setting.
This study also reveals a neuroprotective feature of NCS1 overexpression in reducing cell shrinkage due to retrograde effects of axotomy. Others have shown that neurotrophic factors can prevent atrophy or death of axotomized CSN [31],[37]. Recently, Chondroitinase ABC, which is known to remove the inhibitory scarring at the injury site, has been shown to induce neuroprotection of CSN via a possible retrograde effect mediated at the injured mouse spinal cord [38]. However, our study demonstrates that NCS1 overexpression can increase the intrinsic capacity of CSN to overcome the inhibitory environment and even compensate for the apparent lack of trophic support associated with CNS injury.
Our study establishes that NCS1 is an important intracellular component in the regulation of axonal sprouting and regeneration, and neuroprotection in central neurons of an adult mammalian nervous system, as recently shown for the peripheral nervous system in chick embryo studies on dorsal root ganglion (DRG) neurons [24]. Furthermore, the PI3K/Akt pathway mediating these responses in vitro and in vivo is consistent with the experiments on primary cultured adult DRGs and perinatal cortical neurons linking Akt activation with neurite outgrowth [39],[40] and the survival of primary cultured embryonic cortical neurons [19]. The opposite conclusion, that NCS1 contributes to a retardation of neurite growth, may relate to the use of a rat adrenal medullary pheochromocytoma cell line (PC12 cells) and the additional need for NGF to promote differentiation into sympathetic neuron-like cells [41],[42].
In summary, the limited ability of adult CST neurons to undergo functional sprouting may be due to low endogenous levels of NCS1. Thus NCS1 emerges as a potential intracellular target for therapeutic intervention following injury to the central nervous system.
The complete cDNA sequence of the rat NCS1 was generated by PCR from adult rat cortex using the following forward primer; 5′-ATGGGGAAATCCAACAGCAAG-3′; and the reverse primer, 5′-CTATACCAGCCCGTCGTAGAG-3′ then cloned into a pCR2.1-TOPO vector (Invitrogen). The ncs1 gene was inserted under the control of a CMV promoter in a human immunodeficiency virus type 1 (HIV1) vector containing a 5′ central polypurine tract (cPPT) element and a 3′ woodchuck post-transcriptional regulatory element (WPRE) enhancer. To allow for coexpression of NCS1 and the enhanced green fluorescent protein (GFP), the eGFP gene was inserted into the Cla1 site under the control of a SFFV promoter. Viral vector stocks, pseudotyped with the VSV-G envelope glycoprotein, were prepared by triple plasmid transient transfection of HEK293T cells as previously described [43]. The titre of pRRL-CMV-NCS1-SFFV-eGFP (for simplicity termed HIV-GFP-NCS1) was 3.3–4.4×108 TU ml−1 and pRRL-CMV-eGFP (for simplicity termed HIV-GFP) was 4.7–4.8×108 TU ml−1 determined by transient transfection of the HEK 293T cell line and analysed by flow cytometry.
Adult cortical neurons were cultured as previously described [16],[44]. Adult male Wistar rats (220–250 g) were overdosed with sodium pentobarbitone (Lethobarb), transcardially perfused with heparinized saline, and the cortices removed with as little white matter as possible. The cortices were cut into 0.5 mm longitudinal sections using a McIlwain tissue chopper before digestion in 2 mg/ml papain at 30°C for 30 min. Cortical neurons were mechanically dissociated with a glass Pasteur pipette and separated from debris by centrifugation in four 1 ml steps of Optiprep in HibernateA/B27 medium (7.5%, 10%, 12.5%, and 17.5%) at 600 g for 15 min. Fractions containing neurons were collected, washed, and resuspended in NeurobasalA/B27 medium for plating at a density of 3,000 cells per well on poly-D-lysine (10 µg/ml) pre-coated cover slips. The neurons were allowed to settle onto the cover slips for 1 h, and after washing, HIV-GFP-NCS1 or control HIV-GFP lentivector was added to the media at MOI 10. The neurons were left for a further 3 days in vitro (DIV) before immunocytochemical processing.
For phospho-Akt studies, the reversible PI3K/Akt inhibitor LY294002 (hydrochloride, 40 µM dissolved in DMSO, Tocris Biosciences) or DMSO alone at a final concentration of 0.01% was added to the cultures at 1 DIV and 2 DIV. The neurons were harvested at 3 DIV.
After 3 d in culture, neurons were fixed with 4% paraformaldehyde for 20 min and then permeablized with cold methanol for 3 min. The neurons were washed three times for 5 min with 0.01 M phosphate-buffered saline (PBS) before 2 h incubation with chicken anti-GFP (1∶1000, ab13970, Abcam), rabbit anti-NCS1 (1∶500, NL3750, BioMol International), rabbit anti-phospho-Akt (1∶200, Cell Signalling Technology), or rabbit anti-MAP2 (1∶1000, AB5622, Chemicon). The cover slips were washed 3×5 min with PBS and then incubated with donkey anti-rabbit Alexa Fluor 546 and goat anti-chicken Alexa Fluor 488 secondary antibodies (1∶2000, Molecular Probes) for 45 min. After 3×5 min PBS washes, they were mounted with FluorSave reagent containing 0.5 µl DAPI (10 µg/ml) to visualise cell nuclei. The staining with phalloidin-TRITC (1∶100, P5282, Sigma) was carried out 1 h before immunostaining with other antibodies was carried out.
Image analysis and quantification was made with the observer blinded to the group assignment as previously described [16],[45],[46]. Analyses were restricted only to transduced neurons immunoexpressing GFP. For each experimental group, 50–100 GFP-positive neurons were captured randomly using a Zeiss Axioplan 2 fluorescence microscope. The soma of each neuron was outlined to obtain the fluorescent intensity using the Axiovision V4.6 software to determine the neuronal levels of NCS1 and phospho-Akt immunoreactivity. To minimise variability between each image, the capture settings were fixed throughout the whole study. The number of neurite sprouts from the cell bodies and of the longest neurites, of length greater than cell body diameter was determined. To differentiate whether a GFP positive neurite was a dendrite or an axon, the specific dendritic marker microtubule-associated protein 2 (MAP2) was used. Neurites with strong and weak MAP2 immunostaining were identified as dendrites and axons, respectively.
Western blots were carried out as previously described [47],[48]. After 3 d in culture, primary adult cortical neurons transduced with either HIV-GFP-NCS1 or the control HIV-GFP vector, with or without the PI3K/Akt inhibitor LY294002, media were removed and neurons were harvested in 250 µl ice-cold lysis buffer (20 mM HEPES pH 7.4, 100 nM NaCl, 100 mM NaF, 1 mM Na3VO4, 5 mM EDTA, 1% Nonidet P-40 and 1× protease inhibitor cocktail; Roche). To obtain sufficient protein, the same 250 µl lysis buffer was used in three cultured wells and the lysates rotated for 2 h at 4°C. After centrifugation at 13,500 g for 15 min at 4°C, the supernatant was collected and total protein concentration was determined using a bicinchoninic acid protein assay kit (Pierce).
Intracortical injections of either HIV-GFP-NCS1 or HIV-GFP lentivector in adult male Wistar rats (n = 4–5 per group) were carried out as described below. To determine the role of Akt activation, the PI3K/Akt inhibitor LY294002 (100 mM) or vehicle (DMSO) was injected intracerebroventricularly via an externalised catheter on every other day of the third post-injection week. At the end of the third week, rats were sacrificed and the injected region of the cortex was freshly and quickly removed and stored at −80°C until further processed. The protein obtained for Western blotting was extracted as described above.
Fifteen micrograms of total protein were electrophoresed on 12% acrylamide gel before transfer onto Hybond P membranes (Amersham) and incubated overnight at 4°C with rabbit anti-phospho-Akt (Ser 473, 1∶100, #3787S, Cell Signalling Technology), rabbit anti-NCS1 (1∶1000, NL3750, BioMol International), or mouse anti-β III tubulin (1∶1000, G712A, Promega). Visualisation was performed using secondary antibodies, donkey anti-rabbit IRDye-800CW, and goat anti-mouse IRdye-680CW (LI-COR Biosciences). Fluorescent blots were imaged on the Odyssey Infrared Imaging System (LI-COR Biosciences). To allow for visualisation of the total Akt on the same blot as phospho-Akt (both antibodies were raised in the same species), the blot was first stripped with buffer (62.5 mM Tris-HCl pH 6.8, 2% SDS, 100 mM β-mercaptoethanol) before re-blotting with rabbit anti-Akt (1∶100, #9272, Cell Signalling Technology). Western blotting was carried out with 3–5 independent samples.
The surgery was performed aseptically in accordance with UK Home Office regulations as previously described [16]. Briefly, adult male Wistar rats (n = 8–9 per group) were anaesthetized using a combination of ketamine and medetomidine, then fixed in a stereotaxic frame. The skull was exposed and injections were made at a depth of 2 mm dorsoventrally into the sensorimotor cortex region using the injection coordinates as determined from a microstimulation mapping study [28]. These were, with reference to bregma (AP, anterior-posterior; L, lateral); AP: −1.5 mm, L: 2.5 mm; AP: −0.5 mm, L: 3.5 mm; AP: +0.5 mm, L: 3.5 mm; AP: +1.0 mm, L: 1.5 mm; AP: +1.5 mm, L: 2.5 mm; AP: +2.0 mm, L: 3.5 mm. At each site, 1 µl of HIV-GFP-NCS1 or control HIV-GFP lentivector was directly injected at a rate of 0.2 µl/min using a microinfusion pump via a finely pulled glass micropipette and left in situ for a further 1 min. HIV vector pseudotyped with a VSV-G envelope produced strong expression and anterograde labelling [49]. Three weeks after viral injection, a unilateral pyramidal tract lesion at the level of medulla was performed as described previously [50]. A ventral midline incision was made and the occipital bone exposed by blunt dissection. The ventrocaudal part of the bone was partially removed using fine rongeurs, exposing the right medullary pyramid. The dura was opened and the right pyramidal tract was sectioned approximately 2 mm rostral to the decussation with fine iridectomy scissors using the basilar artery as the midline. Sham operated rats received similar surgery without incision of the tract. In another group of experiments, the left intact pyramidal tract was transected in a second operation.
In the delayed lentivector transduction studies, adult Wistar rats (n = 5–6 per group) received intracortical lentiviral injections 2 d after a unilateral pyramidal tract lesion as described above. To study the sprouting effect of delayed lentivector transduction on uninjured and injured CST axons, the lentiviral injections were administered into the sensorimotor cortex corresponding to the unlesioned and lesioned pyramidal tract, respectively. After 4 wk post-surgery, the rats were perfused transcardially with 4% paraformaldehyde and tissue collected for histology.
In the neuroprotection study, adult Wistar rats (n = 4–5 per group) received intracortical lentiviral injections 1 wk before receiving on the ipsilateral side a unilateral pyramidal tract lesion as described above. Using a microinfusion pump, Fast Blue tracer (200 nl, 2% wt/vol PBS, EMS-Chemie GmbH) was administered at a rate of 0.2 µl/min into the lesion site via a finely pulled glass micropipette and left in situ for a further 1 min. Unlesioned rats without intracortical lentiviral injection had Fast Blue tracer (2%, 200 nl) injected into the pyramidal tract at the medullary region. Care was taken to minimise axonal damage by the injection process. After 2 wk post-injection of tracer, the rats were perfused transcardially with 4% paraformaldehyde and tissue collected for histology.
The analysis of atrophy in CSN was carried out as previously described [38]. The cell area of CSN co-labelled with Fast Blue tracer and GFP immunostaining were acquired using the AxioVision V4.6 program by an investigator blinded to the treatment groups. In unlesioned rats, Fast Blue traced CSN from similar coronal levels as for the lentivector transduced rats were analyzed. Size and frequency distributions of CSN were determined for each rat and a mean distribution calculated for each treatment group. At least six transduced sections were analysed and quantified per rat (n = 4–5 per group). A total of over 2,400 neurons were analyzed.
Following unilateral pyramidotomy, functional recovery was assessed behaviourally using the staircase reaching and grid exploration tests at 2 d post-lesion and then weekly for 6 wk as described previously [15],[16],[50]. In the staircase reaching test, the rats were trained to reach and grasp the food pellets from a baited double staircase (Campden Instruments) before CST lesion. This test allows assessment of extension and grasping ability independently for each forelimb. On the testing day, rats were placed in the staircase box for 15 min and the number of food pellets removed or displaced was recorded. In the grid exploration test, the rats were allowed to explore the grid freely (40 cm×60 cm containing 5 cm×5 cm mesh, raised 50 cm high) where at least 50 forelimb and 20 hindlimb steps were recorded, typically made within 3 min. The “free” exploration removes any possible learning effect due to training as no pre-training was required and that the rats never move around the grid in the same pattern. The grid exploration captured on video camera was replayed and analysed for limb misplacement on the grid. Analysis involved counting the number of limb misplacement from the first 30 forelimb and 20 hindlimb placements of each rat, to ensure no bias between animals and groups. At the end of the behavioural assessment at 6 wk, the rats were sacrificed and perfused transcardially with 4% paraformaldehyde and tissue collected for histology.
Control rats (n = 4) received intracortical injections of HIV-GFP while NCS1-transduced rats (n = 4) received HIV-GFP-NCS1 followed by a unilateral pyramidotomy on the right side 3 wk later as described above. All electrophysiological measurements were performed at least 6 wk post-injury under urethane (1.25 g/kg body weight, i.p.) anesthesia. Following tracheotomy, the rat was fixed into a frame by ear bars and spinal clamps such that the forelimbs were fully pendent. A pair of hooked stainless steel wires insulated to the tip was inserted into the tricep brachii of both forelimbs approximately 6 mm apart for EMG recording. The area of the sensorimotor cortex where lentiviral vectors were injected was exposed by craniotomy, covered by mineral oil, and stimulated through a flat ended silver wire electrode (0.5 mm diameter) ensheathed with plastic to its tip to minimize surface spread of the stimulating current. A 2 mm diameter anode was placed on the skull periosteum rostral to the stimulating electrode. The stimuli repeated at 1 Hz consisted of 1 to 4 pulses, 3 ms apart, and 0.1 ms duration from an isolated stimulator (Digitimer DS2A). The final stimulation site was selected after systematic mapping with varying stimulus parameters until a discrete contralateral (left) forelimb movement was observed with a clear EMG response and a threshold below 25 V for the least number of effective pulses. The EMG was amplified (LF, 30 ms TC; HF, 3 KHz) and digitized using a CED 1401 interface with a sampling rate of 10 kHz. The area of the EMG response (Vs) was measured from 20 averaged responses using Spike 2 V5.0 software.
To check that the EMG response from the CST-denervated forelimb was dependent on contralateral CST input to the spinal cord, the right dorsal CST was transected at the cervical C4 level using a chisel formed by flattening a G25 needle.
Data were analyzed using SigmaStat 3.5 software. Reported values are expressed as mean ± SEM. The in vitro experiments, Western blot analysis, and number of GFP positive axons in the dorsal CST were analyzed with Student's t test. The GFP immunopositive axon collaterals and sprouts in the cord and brainstem, behavioural tasks, and electrophysiology were analyzed with two-way ANOVA followed by Tukey's post hoc test. The cell size cumulative frequency distribution of CSN was analyzed with a two-sample Kolmogorov-Smirnov test, performed against a significant threshold of 0.05 to correct for multiple testing.
|
10.1371/journal.pcbi.1002775 | Functional Connectivity and Tuning Curves in Populations of Simultaneously Recorded Neurons | How interactions between neurons relate to tuned neural responses is a longstanding question in systems neuroscience. Here we use statistical modeling and simultaneous multi-electrode recordings to explore the relationship between these interactions and tuning curves in six different brain areas. We find that, in most cases, functional interactions between neurons provide an explanation of spiking that complements and, in some cases, surpasses the influence of canonical tuning curves. Modeling functional interactions improves both encoding and decoding accuracy by accounting for noise correlations and features of the external world that tuning curves fail to capture. In cortex, modeling coupling alone allows spikes to be predicted more accurately than tuning curve models based on external variables. These results suggest that statistical models of functional interactions between even relatively small numbers of neurons may provide a useful framework for examining neural coding.
| The number of simultaneous neurons that electrophysiologists can record is growing rapidly, and a central goal of computational neuroscience is to develop statistical methods that can make sense of this growing data. Here we present a unified statistical analysis of 10 different datasets recorded from several different species and brain areas. We show how functional interactions between neurons may be used to predict spiking in each of these different areas, and find that, in many cases, modeling interactions between a small number of neurons yields better spike predictions than modeling each neuron's relationship to the outside world using tuning curves. Although these statistical results cannot be linked to specific network architectures, since the measured interactions between neurons are purely functional rather than anatomical, they suggest that modeling interactions between neurons will be a useful approach to understanding neural coding as electrophysiologists record from increasing numbers of neurons.
| One of the central tenets of systems neuroscience is that the functional properties of neurons, such as receptive fields and tuning curves, arise from the inputs that each neuron receives from pre-synaptic neurons. Over the past few decades, a number of experimental techniques have been developed to study exactly how interactions between neurons determine receptive field structure, including in vivo intracellular or paired recording [1]–[6] and pharmacological or electrophysiological interventions [7]–[10]. As electrophysiologists record from increasing numbers of neurons simultaneously [11]–[13], statistical approaches that estimate interactions between neurons have the potential to explain the functional properties of neurons as network effects using only passive spike observations [for review see 14].
To understand how interactions between neurons drive neural activity, recent model-based statistical methods attempt to predict the activity of each neuron based on the activity of other simultaneously observed neurons in addition to any external variables, such as the orientation of a visual stimulus or the direction of hand movement [15]–[19]. This type of inferential approach provides estimates of potential interactions between neurons and allows us to assess how much external variables or interactions between neurons may have contributed to the observed spiking. It is important to note that these models provide only an approximation to the true underlying network structure. Since the vast majority of pre-synaptic inputs to any given neuron are unobserved, the interactions that these approaches describe reflect many different factors including common input in addition to direct and indirect synaptic connections [14], [20]. However, due to the fact that neurons are not independent, these models can improve both encoding accuracy (how well neural responses can be predicted) as well as decoding accuracy (how well external variables can be predicted from neural responses).
Statistical models of interactions between neurons have been used to describe many different aspects of multi-electrode data in retina [21], LGN [22], primary visual cortex [23], [24], motor cortices [17], [25], [26], and hippocampus [27], [28]. Here we present a unified analysis of data from six different brain areas with a particular view towards three questions: 1) How are estimated interactions between neurons related to apparent tuning properties?, 2) How do traditional tuning curve models change as interactions between neurons are included in the model? and 3) How does our ability to predict and decode neural activity improve as increasing numbers of simultaneously recorded neurons are observed?
To make our analysis as broad as possible, we collected ten multi-electrode spike train datasets with at least 30 simultaneously recorded neurons. Datasets were obtained from six different brain areas across four different species performing a variety of tasks. By modeling typical tuning curves for neurons in each area as well as interactions between neurons we determine how much these two factors contribute to spike prediction. We find that including information about the activity of other observed neurons improves both spike prediction and decoding accuracy substantially. By capturing noise correlations and unmodeled features of the external world models of interactions between even a relatively small number of recorded neurons can complement and, in some cases, surpass, models of tuning curves alone.
Although neurons are often characterized by how their firing rate relates to external stimuli or movement variables, the functional properties of most neurons are byproducts of the input they receive from other neurons (Fig. 1A). By modeling typical tuning curves as well as coupling between neurons we aim to determine how well each of these factors explains spiking (Fig. 1B). We fit three, time-instantaneous, generalized linear models (GLMs) to recorded spike trains from 10 different datasets and attempt to predict spiking given: 1) external variables, 2) the activity of other observed neurons, or 3) both external variables and the activity of other observed neurons. After fitting these models to spike data the estimated parameters correspond to a typical tuning curve model, a phenomenological model of interactions between neurons, and a full model that allows functional interactions between neurons to provide an alternative explanation for the spiking that is traditionally attributed to tuning to external variables (see Methods).
Tuning curves can be “explained away” if the other observed neurons provide a better explanation for spiking than the external variables (Fig. 1C). For instance, in a toy network where neuron 1 is tuned to external variables and projects to neurons 2 and 3. Neurons 2 and 3 will appear tuned, despite having no direct relationship to the external world (Fig. 1C, middle). By using the activity of neuron 1 to predict the spiking of neurons 2 and 3, the tuning properties can be explained away by the more direct interactions between neurons (Fig. 1C, bottom). Apparent tuning can appear in any number of network configurations, but given a set of simultaneously recorded neurons the models used here aim to explain spiking as directly as possible. In physiological data, it is unlikely that we are recording from synaptically connected pairs of neurons. The estimated couplings that we observe are likely to be strongly influenced by common input from outside of the recording area and do not necessarily reflect local, recurrent effects. However, tuning curves can still be “explained away” if the activity of the other observed neurons allows better spike prediction.
We fit spike count data from multi-electrode recordings in 6 different brain areas using maximum a posteriori (MAP) estimation for each of the three models (see Methods). Data from motor cortices were recorded during reaching movements to measure tuning to hand direction (Fig. 2A, top). Data from visual cortex were recorded during the presentation of drifting gratings (Fig. 2B, top). Data from auditory cortex were recorded during the presentation of pure tones (Fig. 2C, top). Data from primary somatosensory cortex were recorded during reaching (Fig. 2D, top). Data from hippocampus were recorded during free foraging (Fig. 2E, top). Details of the experiments as well as model fitting and validation procedures are included in the methods.
In the full model, most, but not all, neurons showed decreased modulation to external variables (Fig. 2, bottom). That is, spiking that was previously attributed to tuning properties was more directly explained by functional interactions with other neurons. However, the structure of the tuning curve (i.e. the preferred direction, frequency, or place) remained relatively unchanged. The tuning modulation (minimum-to-maximum) decreased 34–82%, with hippocampus showing the smallest decrease and primary auditory cortex the greatest decrease (Fig. 3A). On the other hand, typical tuning preferences are generally well-preserved (Fig. 3B). Preferred direction, frequency, and place are consistent between the tuning curve model and the full model (correlation coefficient R = 0.34–0.86, circular correlation coefficient where appropriate).
To quantify how coupling in the full model relates to tuning properties we measured the overlap between tuning curves for each pair of neurons in each dataset using the angle between the tuning curve parameter vectors (cosine similarity). An overlap of zero corresponds to orthogonal tuning (i.e. cosine tuned neurons with preferred directions of 0 and 90 deg), an overlap of one corresponds to identical tuning, and an overlap of negative one corresponds to exactly opposite tuning (i.e. cosine tuned neurons with preferred directions of 0 and 180 degrees). We find that tuning curve overlap is clearly related to the bulk spike-count correlation across all stimulus/movement conditions (Fig. 3C). However, coupling strength is only indirectly related to tuning curve overlap (Fig. 3D). Two neurons having similar tuning curves will not necessarily have strong coupling in the full model. This suggests that the explaining away of tuning curves by coupling is not a straight-forward byproduct of stimulus correlation and that including other observed neurons in spike prediction provides information that is not present in the tuning curves alone.
The structure of the coupling terms, particularly the number of connections that each neuron makes with the other observed neurons (the “degree”) provides some insight into how tuning curves are explained away. In contrast to theories of scale-free neural connectivity [29] – which predict power-law degree distributions – the estimated functional interactions in these datasets, under the full model, have uni-modal degree distributions (Fig. 4A). Interestingly, across all datasets, it seems that out-degree (how many outputs a neuron drives) is more narrowly distributed than in-degree (how many inputs a neuron receives). The exact structure of the functional connectivity graphs may be affected by electrode spacing and geometry [24]. However, in-degree is correlated with how well coupling can explain tuning (Fig. 4B). In general, neurons whose tuning curves are well explained by coupling receive input from more neurons compared to neurons whose tuning curves are not well explained by coupling.
How these models behave as the number of simultaneously recorded neurons grows is an important consideration for future modeling. Here we fit the coupling alone and the full model, varying the number of neurons used to predict spikes. Under the full model, we find that, in good approximation, the fraction of variance explained by tuning decreases logarithmically as the number of observed neurons increases (Fig. 5). Place fields in hippocampus are explained away slowly, while tuning curves in motor and sensory cortices are explained more rapidly. In general, 10–70% of the variance initially attributed to tuning curves is explained by coupling between neurons in the full model.
A second metric for studying how these methods scale with the number of observed neurons is spike prediction accuracy (see Methods). As the number of neurons included in the model increases we find that spike prediction accuracy scales, to a good approximation, hyperbolically (Fig. 6A). Note that the full model begins providing the same accuracy as the tuning curve model. As more neurons are included in the model, spike prediction accuracy increases and appears to converge towards a maximum. Interestingly, modeling coupling alone shows this same hyperbolic behavior, beginning at zero and converging towards a maximum. Once 10–30 neurons are included in the model, coupling alone provides more accurate spike prediction that traditional tuning curve models in most datasets.
Hippocampal neurons appear to differ from cortical recordings in that spike prediction accuracy increases approximately linearly. Moreover, modeling coupling alone does not provide more accurate spike prediction than the basic place field model. This may be due to the low correlations between HC neurons. Electrode spacing may also be a factor, since, unlike the 400 µm electrode spacing used in almost all of the intra-cortical arrays, HC recordings had 20 µm vertical electrode spacing. However, the coupling model for spontaneous activity in V1 shows the same hyperbolic behavior despite data being recorded using a polytrode with 50 µm electrode spacing. Scaling of spike prediction accuracy in hippocampus appears to be qualitatively different from that in cortex.
In addition to examining how encoding accuracy scales with the number of recorded neurons, we also examined decoding accuracy for several datasets (Fig. 6B). For the V1, M1, and PMd datasets, we infer which of eight different reach targets or stimuli was presented given the observed spiking on a given trial. Here we use Bayesian decoding under either the tuning curve encoding models or the full encoding model described above (see Methods for details). As with spike prediction accuracy, decoding accuracy grows approximately hyperbolically as more neurons are included in the models. Including coupling between neurons in addition to tuning improves decoding by a small but significant amount: 4.8±0.3%, 7.8±0.4%, 10.3±0.4%, and 7.9±0.3% for the two M1 datasets, PMd, and V1, respectively. Many studies have illustrated how dependencies between neurons can reduce decoding accuracy [30]. By simulating from the tuning curve model we can examine how well we could decode external variables if the neurons were conditionally independent. In this case, decoding from such an independent population of neurons would be ∼25% more accurate than decoding the observed data with the tuning curve model.
It is important to consider what factors may be driving these scaling phenomena. Although the coupling terms are regularized during estimation and the spike prediction accuracy is cross-validated, it may be the case that tuning curves are explained away as a result of over-fitting or, alternatively, as a simple side effect of stimulus correlations. To test for this possibility we simulated spike counts from the tuning curve model, where the neurons are conditionally independent given the external variables. That is, although there may be stimulus correlations, spiking can be completely predicted by external variables. Here we find that no matter how many neurons are included in the full model, tuning explains between 90–100% of the variance (Fig. 7A). This suggests that the results for the full model in real data are not driven by over-fitting or stimulus correlation alone.
Additionally, we can quantify how much stimulus correlation contributes to explaining away by shuffling the data to remove noise correlations. Where possible (M1, PMd, and V1) we shuffle the spike counts within each trial condition (target or grating direction) independently for each neuron. This manipulation retains stimulus correlations while destroying any structure unrelated to the stimulus. Here we find that, in the full model, tuning explains between 85–95% of the variance (Fig. 7B). Furthermore, the spike prediction accuracies of the full and coupling models do not exceed the accuracy of the tuning curve model in shuffled data (Fig. 7C). These two controls demonstrate that the observed explaining away is not simply a byproduct of stimulus correlations or of a poor tuning curve model. Explaining away can only occur when the other observed neurons provide a more direct explanation of spiking than the external variables.
Finally, to examine what drives the shape of these spike prediction accuracy curves we simulated a linear-nonlinear-Poisson neuron receiving sparse, correlated input. As input correlation increases spike prediction accuracy converges more quickly to its maximum (Fig. 8A). When the inputs are strongly correlated, neurons added later are only providing redundant information. However, when the inputs are independent, each additional neuron contributes to more accurate spike prediction. If the inputs are sparse and some of them are irrelevant to the prediction, information added by each neuron is simply smaller on average (Fig. 8B). That is, if only 10% of the inputs are non-zero then it takes 10 times as many neurons to reach a given spike prediction accuracy compared to the case where all of the inputs were non-zero. For input correlation and probability of a given input being non-zero, the simulations are well-approximated by a hyperbolic function where is the maximum spike prediction accuracy and the maximum number of neurons.
Linking the strength of common input and sparseness to the spike prediction accuracy curves observed in real data is difficult. Both a weakly correlated, highly connected network and a highly correlated, highly sparse network will have near-linear growth. However, here we find that neurons in cortex (particularly V1 and A1) tend to be more strongly correlated than neurons in hippocampus (Fig. 3C). This may partially explain the rapid growth in spike prediction accuracy for the cortical datasets and, in comparison, the near-linear growth for hippocampal datasets. Especially in cortex, the fact that neural activity traditionally attributed to tuning curves is more directly explained by interactions between neurons appears to be a byproduct of unobserved common input.
Excepting peripheral neurons such as photoreceptors, the relationship between a neuron's spiking and the external world is a result of the input that each neuron receives from other neurons. Many studies have examined how pre-synaptic input determines receptive field structure and tuning properties both experimentally [e.g. 1], [3], [4], [5] and theoretically [31]–[33]. Here we have used multi-electrode recordings and statistical modeling to examine, broadly, how tuning curves might be explained, in a statistical sense, by functional interactions between neurons. We have found that, in a variety of brain areas, modeling coupling between a relatively small number of simultaneously observed neurons in the same brain area allows more accurate encoding and decoding. As the number of observed neurons grows the fraction of spiking variability attributed to tuning appears to decrease logarithmically, while spike prediction accuracy increases hyperbolically. Once interactions between 10–30 neurons are modeled, coupling alone can often provide more accurate spike prediction than traditional tuning models.
The extent to which the activity of simultaneously observed neurons or tuning properties explain spiking likely depends on a number of factors including the timescales on which we model spiking, the stimulus or task parameters, and the external variables being used to describe tuning. The coarse, instantaneous coupling models used here cannot distinguish between the many possible hidden causes of correlated neural activity. Since the models used here reflect pair-wise dependencies on a long timescale of 100 s of milliseconds, it is likely that unobserved behavioral variables and internal processes make strong contributions to the coupling terms. Modeling these effects explicitly may yield a more nuanced view of the relationship between tuning curves and interactions between neurons [23], [34]. Although we model trial-by-trial spike-count data here, both tuning and coupling can also be modeled as history-dependent effects. Since external variables change fairly slowly, whereas interactions between neurons are likely to be relatively fast, adding such temporal information may result in qualitatively different results. More detailed models that include history-dependent coupling on millisecond timescales may be able to further unpack the functional roles of recurrent, local coupling and instantaneous common input [35].
The datasets used here yield surprisingly similar results considering that they were recorded from different brain areas and species with different electrode configurations. However, in addition to these anatomical differences, it is important to note that the datasets were recorded under a variety of experimental circumstances, which may help to explain some of the remaining differences in the results obtained from each dataset. For data from motor cortices and hippocampus, for instance, the external variables are not controlled in the same way that sensory experiments are. Movement variables such as velocity or body orientation differ even when the monkey's reaches to the same target or when the rat is at the same maze location. These external differences may lead to higher apparent trial-to-trial variability. Additionally, data from V1 was recorded while the animals were under anesthesia, which may lead to higher correlations between neurons [36]–[38].
The tuning models used here, despite their wide-spread use, are relatively simplistic. Tuning functions that take into account more external variables are likely to give more accurate spike prediction, and including these variables may change the degree to which tuning properties are explained away as interactions between neurons are added to the model. At the same time, exploring the space of external variables and determining what causes a neuron to fire can be difficult [39]–[41]. Fitting tuning functions to neurons in medial temporal lobe, for instance, might require exploring the space of all possible objects [42]. Fitting high-dimensional tuning functions, in general, can require large amounts of data as well as sophisticated estimation methods [43]. Rather than exploring the space of external variables, exploring the statistical structure of interactions between neurons may be an alternative strategy for understanding tuning properties.
Neurons receive pre-synaptic input from tens of thousands of other neurons, and each of these inputs, presumably, plays a role in determining the tuning properties of a post-synaptic neuron. How is it possible then that models of interactions between <100 neurons are able to explain spiking more directly than traditional tuning curve models without any guarantee that the neurons are even anatomically connected?
Ultimately, explaining away can only occur when neural activity is not independent. Many studies have examined correlated neural activity [38], [44]–[48] as well as its potential functional roles [30], [37], [49], [50]. Here correlations between neurons are essential in allowing tuning properties to be explained away by the functional interactions between small numbers of neurons. However, the fact that coupling terms do not explain away tuning curves in simulated or shuffled data, suggests that our results are not simply a byproduct of stimulus correlation. Rather, the estimated coupling between neurons is likely to reflect a combination of direct and indirect interactions [e.g. 51] as well as additional unobserved common input [14] and internal processes [52]–[54]. Several studies have made progress in attempting to infer unobserved common input related to the external world [34] as well as internal processes [35], [55]. Here we simply note that unobserved common input may allow more accurate spike prediction in models of interacting neurons by creating correlations that cannot be attributed to the observed external variables. Modeling these dependencies improves decoding by a small, but significant, amount and may be useful for improving brain-machine interfaces [56]. Moreover, the correlations induced by unobserved common input appear to allow neural activity traditionally attributed to tuning properties to be more directly explained by interactions between neurons.
It is important to note, however, that the statistical approaches used here are unlikely to capture anatomical information about the underlying circuitry. These methods still only provide a sketch of the underlying circuit that best explains the observed spiking. The hyperbolic scaling of spike prediction accuracy observed here, for instance, may be a general property of correlated prediction problems [57]. found a similar hyperbolic scaling in accuracy using the firing rates of neurons in motor cortex for linear prediction of hand position.
For many years, studies of the relationship between neural interactions and tuning properties have been based on detailed electrophysiology [58], [59], experimental intervention , or simulation [32], [60]. Most of these studies have addressed data collected in sensory cortices or peripheral areas. However, understanding the response properties of neurons in other areas, such as motor and association cortices, in terms of neural circuits has been difficult. Here we used simultaneous neural recordings and a model-based statistical approach to ask how well tuning properties can be explained, in a statistical sense, by functional interactions between neurons. While these models are able to explain a surprisingly large fraction of the variation in neural spike counts in a variety of brain areas with a relatively small number of observed neurons, they only provide a rough picture of how network architecture might give rise to commonly observed tuning properties.
Understanding how interactions between neurons give rise to tuning properties, will ultimately mean understanding the relative contributions of feed-forward, local, and top-down pre-synaptic inputs, as well as how different subtypes of neurons and neurons with different types of tuning interact. One area where statistical approaches have revealed this type of detailed architecture is in the retina. By recording from dense populations of retinal ganglion cells (RGCs), recent work has shown that RGC receptive fields arise directly and clearly from input received from rods and cones [61]. Moreover, functional interactions between retinal ganglion cells appear to have a strong, local structure [21]. Although photoreceptors are the only elements in the retinal circuit that have direct responses to the external world, the receptive fields of RGC responses can be understood as a byproduct of indirect interactions with photoreceptors, mediated by intermediate neurons, such as horizontal, amacrine, and bipolar cells.
In most areas of the brain, beyond the retina, recording from a complete neural circuit is experimentally infeasible and the complete network of neurons is immensely under-sampled. In these cases, it is difficult to determine whether potential interactions between neurons are direct (mono-synaptic) or indirect (poly-synaptic), and the estimated interactions are likely to be strongly influenced by unobserved common input [14]. What is ultimately estimated by the statistical approaches is a phenomenological model of the circuitry that best describes the observed spikes [20]. For this reason it is difficult to draw conclusions about detailed architecture in current multi-electrode datasets. Here we have examined how modeling interactions between small numbers affects neural coding and how model-based estimates of interactions relate to stimulus and noise correlation. As electrophysiologists record from increasing numbers of neurons [13] these approaches have the potential to reveal more detailed information about the structure of these cortical and sub-cortical areas.
We analyzed 10 multi-electrode spike datasets recorded from 6 different brain areas and 4 different species. Recordings from primary (M1) and dorsal pre-motor cortex (PMd) were made while a macaque monkey performed a center-out reaching task. Recordings from primary sensory cortex (S1) were made while a macaque monkey performed a random-target pursuit task. Recordings from primary auditory cortex (A1) were made while a ferret was exposed to random frequency tone stimuli. Data from primary visual cortex (V1) consisted of recordings of 1) evoked activity while an anesthetized monkey viewed randomly oriented moving gratings and 2) spontaneous activity from an anesthetized, paralyzed cat. Finally, recordings from dorsal hippocampus (HC) were made while a Long-Evans rat was freely foraging for food on a square platform.
All animal use procedures were approved by the institutional animal care and use committees at Northwestern University (M1 & S1), University of Chicago (M1 & PMd), Albert Einstein College of Medicine (V1), University of Maryland College Park (A1), University of British Columbia (V1 spont), or Rutgers University (HC) , and conform to the principles outlined in the Guide for the Care and Use of Laboratory Animals (National Institutes of Health publication no. 86-23, revised 1985). Data presented here were previously recorded for use with multiple analyses. Procedures were designed to minimize animal suffering and reduce the number used.
The aim of our analysis was to examine the relationship between typical tuning curves and receptive fields in each of these brain areas and coupling between neurons. To this end we extracted spike count data from the spike-sorted multi-electrode recordings and focused either on evoked responses for the stimulus and directed movement tasks or binned responses for the foraging and spontaneous tasks. Each dataset contained at least 31 and as many as 107 simultaneously recorded, putative single neurons after spike sorting (Table 1).
Spike count data were fit using either external variables, the activity of the other recorded neurons, or both [14], [17], [21], [28]. In each case we used a class of generalized linear model [71] - a linear non-linear Poisson (LNP) model with exponential nonlinearity [15], [16]. The model and estimation methods have been previously described in detail elsewhere [21]. Briefly, LNP models assume that the covariates (tuning to stimulus/movement or activity of other neurons) are linearly combined, then passed through an exponential nonlinearity such that the firing rate is non-negative. The estimated firing rate for each neuron is then a function of the external variables during each trial and the activity of the other neurons :where denotes one of basis functions that describe the shape of the tuning curve, and the parameters and capture tuning, coupling to other neurons, and a baseline firing rate . The basis functions, described below, will depend on the brain area we are trying to model and the stimulus/task. The spike count is then assumed to be drawn from a Poisson distribution with this rate:where represents the spike count for neuron on trial .
Using this same framework, tuning curves alone were modeled bywhile coupling between neurons was modeled by
Note that, for the coupling model, the spike count for the neuron whose firing rate we are estimating was always excluded. Using this framework we examined the effect of network size on spike prediction accuracy by varying the total number of neurons included in the model and using a random subset of all recorded neurons, again excluding neuron .
For each of these three models – the full model, tuning curve model, and coupling model – we estimated the parameters and directly from the observed spike count data using maximum likelihood estimation (MLE) or maximum a posteriori (MAP) estimation with an L1-penalty to prevent over-fitting [see 21]. Here we compute ML estimates using iterative reweighted least squares (IRLS) with the Matlab package glmfit and compute MAP estimates using path-wise, cyclical coordinate descent [72] with the R package glmnet.
Where regularization is used we optimized the regularization hyperparameter via the cross-validated (10-fold) log-likelihood, and in all cases we evaluated the “spike prediction accuracy” of the models using the cross-validated log likelihood ratio relative to a homogeneous Poisson process. For a firing rate , the log-likelihood is given byand the log likelihood ratio relative to a homogeneous Poisson process (spike prediction accuracy) is given by
In this case, a spike prediction accuracy of zero corresponds to a model that does no better than predicting the mean spike count. Values were calculated in base-2 and rescaled by time to give units of bits/s [see 21,27]. We find that spike prediction accuracy scales approximately hyperbolically, following where is the number of neurons in the model and and are parameters determining the shape of the curve [see 57].
An important component of these models is the choice of basis functions for the external variables. Here we have attempted to choose common tuning models, appropriate for each dataset. For M1 and PMd neurons, for instance,where denotes the target direction on each center-out trial. This linear component of the model corresponds to the traditional cosine tuning models of motor cortical neurons [73], [74]. We used the same model to capture direction tuning in visual cortex [75].
While activity in M1 has also been shown to covary with speed [76], we elected to use the simpler direction tuning model here, and model speed tuning only in S1 neurons, which show direction tuning [77] as well as clear tuning to hand speed [64]. In this case we use
Place fields of the neurons in hippocampus have been well described [78]–[80]. To model these localized response properties we use a set of radial basis functions that tile the foraging area.
Specifically we use K = 25 isotropic Gaussian radial basis functions equally spaced on a 5×5 grid with means and covariance , = 9 cm.
Finally, for neurons in A1 [81]–[83], we again use radial basis functions. In this case K = 7 Gaussians were equally spaced along the log-frequency of the stimulus with standard deviation = 0.64 octaves.
In most cases (TC dimensionality <4), regularization was only applied to the coefficients modeling coupling between neurons. To avoid convergence problems [84] the models using radial basis functions (A1 and HC) included weak regularization on the tuning curve coefficients (with 20% of the L1-penalty used for the coupling coefficients). The tuning curve parameters do not change substantially for penalties ranging from 1–20%; however, there may be unintended shrinkage in these models, and the decrease in modulation observed for these neurons may be somewhat over-estimated.
It is important to note that the models used here differ from previous approaches in that they are time-instantaneous — we model coupling between neurons at the same time. This does not pose any difficulties during fitting, since we are modeling only the conditional distributions for each neuron . However, simulating from the joint spiking distribution is no longer straight-forward. The usual assumption, , does not hold, but, since the conditionals are know, we can use Gibb's sampling to simulate from this joint distribution if necessary (see section on Decoding below).
To quantify the changes in tuning under the full model we evaluate the tuning modulation, tuning preference, and tuning curve overlap between pairs of neurons. Tuning modulation is simply the peak-to-peak difference in firing rate for the tuning curve component of the model, reported in Hz. Tuning preference is defined differently for each dataset: for M1, V1, PMd, and S1 we use the preferred direction, for A1 we use the preferred frequency, and for HC we use the preferred place along the x-axis. Finally, to measure similarity between the tuning curves for pairs of neurons we evaluate the tuning curve overlap between neurons and , . Accordingly, a tuning curve overlap of 1 suggests that the two neurons have identical tuning (up to a constant baseline), while a tuning curve overlap of 0 suggests that the two neurons have orthogonal tuning.
To quantify network properties we also report the spike count correlation (Pearson's correlation). For two neurons with trial-by-trial spike count observations and the correlation is given by . Although the spike count correlation between pairs of neurons is known to increase with both firing rate [85] and time interval [86], [87], we do not attempt to correct for these effects here. In most cases the bin-size is determined by the task and the traditional periods used to measure tuning curves, such as stimulus duration.
To quantify the relative contributions of the tuning curve and coupling components in the full model we summarize the fit using the fraction of variance explained by tuning. For each neuron we calculate
A value of 1 suggests that the coupling terms provide no additional information, while a value of 0 suggests that any tuning information is explained completely by coupling to other observed neurons. It is important to note that there is considerable heterogeneity in how well tuned neurons are to the external variables. Here we analyze all recorded neurons, even those that might be considered un-tuned.
In contrast to the encoding models above, which aim to predict spikes given a stimulus, we can also examine how coupling affects decoding, which aims to predict a stimulus given a set of spike observations . Here, we use Bayesian decoding [88], [89] based on the same encoding models described above. Assuming the stimuli are equally probable
For the tuning curve models, we assume that the neurons are conditionally independent given the stimulus,
However, for the full model, since we assume that coupling is instantaneous, we cannot make this assumption. In this case we use a variation of Gibb's sampling [90]–[92] to approximate the joint distribution . Briefly, in Gibbs sampling we generate samples from by iterating over all neurons and sampling a spike count for each neuron based on the conditional distributionwhere denotes the iteration. Here we use a method know as ordered over-relaxation [93] to improve mixing. This makes the sampler much more efficient for large networks of neurons with strong coupling. In this case we generate samples from the conditional distribution at each update, sort the samples along with , and if is the -th largest value we take the -th sample. After many iterations the set of samples - ignoring a burn-in period - provides an approximation to the joint likelihood that we can then use, via Bayes rule, to approximate the posterior over possible stimuli.
For each model we initialize the sampler with , initialize each successive sample using , and update the spike counts for each neuron in a random order using the conditional distribution with ordered over-relaxation. We then take 5000 samples after a burn-in period of 500 samples to use as an approximation to the joint density.
In practice, it is non-trivial to estimate the probability for each trial given a set of samples from the distribution . When the number of neurons become large the curse of dimensionality makes histogram estimation impossible. We would need samples where is the number of neurons to construct an accurate histogram. Here we use an approximation based on the chain rule of probability
Although we cannot write down the full joint probability analytically, we can approximate each of the marginal distributions in the chain rule using the set of Gibbs sampleswhere denotes the random variables for all other neurons not yet taking a specific value , and the expectations are taken over the set of Gibbs samples. Importantly, each probability in each expectation can now be evaluated analytically based on the conditional Poisson likelihoods of the full model. This approach allows us to approximate the posterior over stimuli and assess the Bayesian decoding accuracy of the tuning curve model with instantaneous coupling.
To examine how the scaling of spike prediction accuracy relates to the underlying structure of the inputs we simulated spikes from a linear-nonlinear-Poisson neuron receiving correlated inputwhere the baseline firing rate parameter was fixed and denotes a set of correlated Poisson random variables. The connection strengths were drawn from a sparse, binary random vector with entries randomly set to zero with probability . Correlated inputs were each assumed to have mean 1 and were drawn from a multivariate Poisson distribution with the covariance matrix where denotes the specified correlation, denotes the mean, denotes the unit matrix, and the identity matrix. Under this covariance matrix all pairs of neurons have correlation and we set the variance of each neuron equal to the mean.
In general, producing correlated Poisson random variables with specific marginal distributions and covariance structure is difficult. Here we use a simplified family of covariance matrices where all neurons have the same correlation and simulate spike counts following [94]. After simulating and fitting the LNP model, we can examine how input correlations and sparseness affect spike prediction accuracy.
|
10.1371/journal.pmed.1002680 | Associations between sex work laws and sex workers’ health: A systematic review and meta-analysis of quantitative and qualitative studies | Sex workers are at disproportionate risk of violence and sexual and emotional ill health, harms that have been linked to the criminalisation of sex work. We synthesised evidence on the extent to which sex work laws and policing practices affect sex workers’ safety, health, and access to services, and the pathways through which these effects occur.
We searched bibliographic databases between 1 January 1990 and 9 May 2018 for qualitative and quantitative research involving sex workers of all genders and terms relating to legislation, police, and health. We operationalised categories of lawful and unlawful police repression of sex workers or their clients, including criminal and administrative penalties. We included quantitative studies that measured associations between policing and outcomes of violence, health, and access to services, and qualitative studies that explored related pathways. We conducted a meta-analysis to estimate the average effect of experiencing sexual/physical violence, HIV or sexually transmitted infections (STIs), and condomless sex, among individuals exposed to repressive policing compared to those unexposed. Qualitative studies were synthesised iteratively, inductively, and thematically. We reviewed 40 quantitative and 94 qualitative studies. Repressive policing of sex workers was associated with increased risk of sexual/physical violence from clients or other parties (odds ratio [OR] 2.99, 95% CI 1.96–4.57), HIV/STI (OR 1.87, 95% CI 1.60–2.19), and condomless sex (OR 1.42, 95% CI 1.03–1.94). The qualitative synthesis identified diverse forms of police violence and abuses of power, including arbitrary arrest, bribery and extortion, physical and sexual violence, failure to provide access to justice, and forced HIV testing. It showed that in contexts of criminalisation, the threat and enactment of police harassment and arrest of sex workers or their clients displaced sex workers into isolated work locations, disrupting peer support networks and service access, and limiting risk reduction opportunities. It discouraged sex workers from carrying condoms and exacerbated existing inequalities experienced by transgender, migrant, and drug-using sex workers. Evidence from decriminalised settings suggests that sex workers in these settings have greater negotiating power with clients and better access to justice. Quantitative findings were limited by high heterogeneity in the meta-analysis for some outcomes and insufficient data to conduct meta-analyses for others, as well as variable sample size and study quality. Few studies reported whether arrest was related to sex work or another offence, limiting our ability to assess the associations between sex work criminalisation and outcomes relative to other penalties or abuses of police power, and all studies were observational, prohibiting any causal inference. Few studies included trans- and cisgender male sex workers, and little evidence related to emotional health and access to healthcare beyond HIV/STI testing.
Together, the qualitative and quantitative evidence demonstrate the extensive harms associated with criminalisation of sex work, including laws and enforcement targeting the sale and purchase of sex, and activities relating to sex work organisation. There is an urgent need to reform sex-work-related laws and institutional practices so as to reduce harms and barriers to the realisation of health.
| To our knowledge there has been no evidence synthesis of qualitative and quantitative literature examining the impacts of criminalisation on sex workers’ safety and health, or the pathways that realise these effects.
This evidence is critical to informing evidenced-based policy-making, and timely given the growing interest in models of decriminalisation of sex work or criminalising the purchase of sex (the latter recently introduced in Canada, France, Northern Ireland, Republic of Ireland, and Serbia).
We undertook a mixed-methods review comprising meta-analyses and qualitative synthesis to measure the magnitude of associations, and related pathways, between criminalisation and sex workers’ experience of violence, sexual (including HIV and sexually transmitted infections [STIs]) and emotional health, and access to health and social care services.
We searched bibliographic databases for qualitative and quantitative research, categorising lawful and unlawful police repression, including criminal and administrative penalties within different legislative models.
Meta-analyses suggest that on average repressive policing practices of sex workers were associated with increased risk of sexual/physical violence from clients or other partners across 9 studies and 5,204 participants.
Sex workers who had been exposed to repressive policing practices were on average at increased risk of infection with HIV/STI compared to those who had not, across 12,506 participants from 11 studies. Repressive policing of sex workers was associated with increased risk of condomless sex across 9,447 participants from 4 studies.
The qualitative synthesis showed that in contexts of any criminalisation, repressive policing of sex workers, their clients, and/or sex work venues disrupted sex workers’ work environments, support networks, safety and risk reduction strategies, and access to health services and justice. It demonstrated how policing within all criminalisation and regulation frameworks exacerbated existing marginalisation, and how sex workers’ relationships with police, access to justice, and negotiating powers with clients have improved in decriminalised contexts.
The quantitative evidence clearly shows the association between repressive policing within frameworks of full or partial sex work criminalisation—including the criminalisation of clients and the organisation of sex work—and adverse health outcomes.
Qualitative evidence demonstrates how repressive policing of sex workers, their clients, and/or sex work venues deprioritises sex workers’ safety, health, and rights and hinders access to due process of law. The removal of criminal and administrative sanctions for sex work is needed to improve sex workers’ health and access to services and justice.
More research is needed in order to document how criminalisation and decriminalisation interact with other structural factors, policies, and realities (e.g., poverty, housing, drugs, and immigration) in different contexts, to inform appropriate interventions and advocacy alongside legal reform.
| Sex workers can face multiple interdependent health risks [1,2]. Between 32% and 55% of cisgender (cis) women working mostly in street-based sex work report experience of workplace violence in the past year [3]. Across diverse settings, both cis and transgender (trans) women and men in sex work are at increased risk of experiencing violence and homicide [4–6], HIV infection [7–9], chlamydia and gonorrhoea [10,11], and poorer mental health than their non-sex-working counterparts [12]. Yet there is considerable variation within sex-working populations [13,14]. The epidemiological context as well as social and structural factors and power relations reproduce inequalities within sex-working populations [2,3,8,9]. For example, cis women working in street-based sex work are more vulnerable to all these outcomes than those working in off-street settings [15,16]. Many vulnerabilities faced by sex workers are multiplicative, closely linked to poverty, substance use, disability, immigration, sexism, racism, transphobia, and homophobia [17].
Qualitative literature demonstrates how social policies and structural factors shape the health and welfare of sex workers. The ‘risk environment’ concept, developed to understand drug-related harms [18] and adapted to HIV and violence experienced by sex workers [19,20], examines different types (physical, social, economic, and political) and levels of environmental influence (micro and macro), in line with broader efforts to address structural determinants of health [21]. This concept has been used to demonstrate how policing, stigma, and inequalities interplay to shape sex workers’ vulnerability to HIV [22], violence [23], and lack of access to healthcare [24] and justice [25,26], and the potential for sex-worker-led interventions to challenge these harms [27]. Epidemiological evidence documents the associations between macro-structural factors (laws, housing and economic insecurity, migration, education, and stigma) and work environment and community factors (policing, work setting and conditions, autonomy, and access to health and peer-led services) and sex workers’ risk of violence and HIV transmission [2,3]. Criminalisation and repressive public health approaches to sex work (e.g., mandatory registration and HIV/sexually transmitted infection [STI] testing) have been shown to hinder the prevention of HIV, where the focus of interventions and research has been directed [28–30]. Conversely, mathematical modelling has estimated that decriminalisation of sex work could halve the incidence of HIV among sex workers and their clients over a 10-year period [2], and evidence from New Zealand indicates that sex workers in decriminalised settings report improved workplace safety, health and social care access, and emotional health [31,32].
Broadly, there are 5 legislative models used to manage, control, or regulate sex work (Table 1) [33]. Full criminalisation prohibits all organisational aspects of sex work and selling and buying sex. Partial criminalisation is where some aspects of sex work are penalised (e.g., soliciting sex in public for sex workers and/or clients, advertising services, collective working, or involvement of third parties). In 1999, Sweden criminalised the purchase, but not the sale, of sex, and various other countries have followed [34]. This ‘criminalisation of clients’ model typically retains laws against ‘brothel-keeping’, which may in practice also target sex workers working together. Regulatory models make the sale of sex legal in certain settings (e.g., in licensed brothels or managed zones) or under certain conditions (e.g., mandatory registration or HIV/STI testing) but illegal in other settings or for individuals who do not meet registration requirements or eligibility criteria (e.g., migrants, cis men and trans sex workers, or people living with HIV) [35]. Full decriminalisation, implemented in New Zealand in 2003, removes criminal penalties for adult sex work, emphasises enforcing criminal laws prohibiting violence and coercion, and regulates the sex industry through occupational health and safety standards [36]. All models criminalise coerced sex work and the involvement of minors, and almost all models—including decriminalisation in New Zealand—prohibit migrants without permanent residency from working legally or in a regulated environment. In practice the implementation of these models through bylaws and enforcement practices is complex, and varies between and within countries and even locally within cities [37,38].
The debate around sex work policy and legislation is highly polarised. Some argue that all sex work is itself gendered violence and should be repressed—a notion that underpins the criminalisation of sex workers’ clients [39,40]. Others argue that this fails to recognise the diversity of experience and identity in the sex industry and the possibility that financial reimbursement for sex between adults can be consensual [41]. At a time of increasing political interest in legislative reform [42–45], there is a critical need to bring together this evidence to inform policies that protect sex workers’ safety, health, well-being, and broader rights. We conducted a systematic review to synthesise evidence of the extent to which sex work laws and their enforcement affect sex workers’ safety, health and access to services, and the processes and pathways through which these effects occur, including in interaction with other macro-structural, community, and work environment factors.
Following a protocol with pre-specified search terms, we searched MEDLINE, CINAHL, PsychINFO, Web of Science, and Global Health for public health and social science literature on studies that combined 3 search domains: (1) sex work, AND (2) legislation OR policing, AND (3) health (physical or emotional, including violence/safety) OR access to services (including health, risk reduction, and social care/support). The complete search terms and review protocol are attached (S1 Text). Meta-analyses were not pre-specified, since they were subject to identifying sufficiently homogenous studies in relation to outcomes and definition of criminalisation.
Three authors screened the sources for inclusion, discussing any uncertainties within the team; a second person re-reviewed relevant sources when necessary. Quantitative data were extracted and analysed by LP and JE, and qualitative data synthesised by PG and RM. For qualitative and quantitative studies, we defined quality-related criteria adapted from the Critical Appraisal Skills Programme (CASP) [46] that papers had to fulfil in order to qualify for inclusion: methods and ethics processes described, appropriate study population clearly defined, and conclusions supported by study findings. Quantitative studies were further assessed according to appropriateness of study design, data collection methods, and analyses, using assessment approaches adapted from the Newcastle–Ottawa scale and CASP [46,47]. A full copy of the quality assessment process for the quantitative studies is available (S1 Table). For qualitative evidence, confidence in review findings was assessed according to CERQual guidance, taking account of methodological limitations, coherence, adequacy of data, and relevance of included studies (S2 Text) [48]. Methodological limitations were assessed using CASP guidelines for qualitative evidence.
We included studies with sex workers of all genders who currently or have ever exchanged sexual services for money, drugs, or other material goods. We included research on all models of sex work legislation and used the following definition of the criminalisation of sex work: ‘a model of intervention in which the criminal law is used to manage, control, repress, prohibit or otherwise influence the growth, instance or expression of prostitution’ [33]. We also included the use of non-criminal penalties to target sex workers, such as fines and displacement orders, including those that do not formally relate to sex work. Within the broad legislative models (Table 1), sex work legislation and policing was operationalised into 8 different categories of police exposure: (1) police repression on an environment in which sex work takes place (workplace raids, zoning restrictions, and displacement from usual working areas), (2) recent (within last year) arrest or prison, (3) past arrest or prison, (4) confiscation of condoms or needles or syringes, (5) extortion (giving police information, money, or goods to avoid arrest), (6) sexual or physical violence from police (negotiated or forced), (7) fear of police repression, and (8) registration as a sex worker at a municipal health authority. Where clear from included papers, we recorded data on gender using the terms ‘cis’ and ‘trans’ to refer to people who do and do not identify themselves with the gender they were assigned at birth, respectively. Conscious of cultural diversity in gender identities, we use the term ‘transfeminine’ to describe feminine-presenting trans populations that do not necessarily describe themselves as female/women [49]. We did not identify any papers that discussed the experiences of people who identify their gender as trans male/masculine or non-binary.
We included quantitative, qualitative, and mixed-methods studies published in English, Russian, or Spanish, and included data specific to the experiences of sex workers. We included papers that measured quantitative associations between criminalisation or decriminalisation of sex work, or repressive policing practices within these contexts, and the following outcomes: threatened or enacted violence, STIs, HIV, hepatitis B/C, overdose, stress, anxiety, depression, risk practices/management (e.g., working with others, reporting violence, condom use, sharing needles/syringes), and access to health/social care services (HIV/STI/hepatitis prevention, testing, and treatment; contraception; abortion; opioid substitution therapy and other drug/alcohol services; mental health and counselling; primary and secondary care; psychosocial support services; housing; and social security). We also included studies that reported qualitative data on the relationships between experiences of criminalisation or decriminalisation and policing and sex workers’ experiences of violence, safety, health, risk management, and/or accessing health or social care services, from the perspectives of sex workers themselves.
We synthesised estimates that adjusted for confounders to assess overall risk of experience of physical or sexual violence, HIV/STI, and condomless sex, stratified by the categories of repressive police activities described above. Where multiple policing practice exposures were presented in the same study, we selected independent estimates in an overall pooled estimate prioritising recent experience of arrest/prison and the most commonly occurring outcomes. Studies including sex workers of different genders were pooled together. We applied random effects models using the DerSimonian and Laird method for all analyses, allowing for heterogeneity between studies and converting all effect estimates into odds ratios (ORs) [50]. We examined heterogeneity with the I2 statistic. We conducted sub-group analyses to describe differences in experience of violence and condom use by partner type (client versus intimate partner/other) and by type of violence (physical versus sexual or sexual/physical combined). We conducted sensitivity analyses to look at overall associations between policing and our specified outcomes, excluding or pooling studies that did not adjust for confounders or reported only STI outcomes (self-reported and biological) or composite HIV/STI, and altering the priority choice of police exposure (from recent arrest/prison to other). We conducted a narrative synthesis of outcomes that were too heterogeneous to pool, including access to services (both mandatory and voluntary uptake of services), harms related to drug use, and emotional health. Studies that measured associations with registration at the municipal health department were also synthesised separately, since this policy was less comparable with all others that involved direct police action. All analyses were conducted using the metafor package in R version 3.4.1 and RStudio version 1.0.143 [51].
For qualitative studies, data were synthesised inductively, iteratively, and thematically. From the body of eligible papers we first focused on the ‘data-rich’ papers that contributed substantive or moderate data and analyses relevant to our research questions. Among the body of papers that had a limited focus on the topic, we then purposively sampled studies that reported on an under-represented population, setting, legislative model, or health issue of interest in this review [52] until no new themes emerged (thematic saturation). For the data-rich papers, we reviewed and wrote summaries of the results and discussion sections, inductively and iteratively drawing out author- and reviewer-identified themes and sub-themes. We then linked sub-themes and themes to 4 core categories, informed by concepts of structural, symbolic, and everyday violence that argue that mistreatment, stigma, exclusion, and ill health often result from intersecting inequalities that become institutionalised and normalised through policies, practices, and social norms [53]. We paid careful attention to the different levels and forms of environmental influence within risk environments [18]. Finally, we reviewed the less data-rich papers (relative to our research questions) against these emerging categories until they required no further refining. We summarise the core categories narratively with illustrative quotes (Box 1), drawing out findings that help to unpack the quantitative associations and their causal pathways. Within each category, we pay close attention to patterns by legislative model.
From 9,148 papers identified, 134 studies met the inclusion criteria, resulting in 40 papers included in the quantitative synthesis, of which 20 were included in the meta-analysis and 20 in the narrative synthesis. A total of 94 met the inclusion criteria for the qualitative synthesis, of which 46 were included in the thematic analysis, 3 were excluded following quality assessment, and 45 were excluded when thematic saturation had been reached (Fig 1).
We estimate that, collectively, lawful or unlawful repressive policing practices linked to sex work criminalisation (partial or full) are associated with increased risk of infection with HIV or STIs, sexual or physical violence from clients or intimate partners, and condomless sex. The qualitative synthesis clearly shows pathways through which these policing practices and health risks are associated: enacted or feared police enforcement—targeting sex workers, clients, or third parties organising sex work—displaces sex workers into isolated and dangerous work locations and disrupts risk reduction strategies, such as screening and negotiating with clients, carrying condoms, and working with others. Specific policing practices, including confiscation of condoms or needles/syringes, are associated with increased odds of HIV, STIs, and violence by a range of actors. Repressive police practices frequently constitute basic violations of human rights, including unlawful arrest and detention, extortion, physical and sexual violence by law enforcement, lack of recourse to justice, and forced HIV testing—violations inextricably linked to increased unprotected sex, transmission of HIV and STIs, increased violence from all actors, and poorer access to health services [3,29,134]. The qualitative synthesis shows how violence and stigma against sex workers are institutionalised, legitimised, and rendered invisible [26,35] in contexts of any criminalisation and regulation [26,35], as sex workers across settings consistently report being further criminalised, blamed, or ignored when they report crimes against them. This structural, symbolic, and everyday violence fosters climates of impunity and under-reporting, and failure to recognise sex workers as citizens deserving protection, care, and support [26]. Targeting and exclusion of the most marginalised sex workers reinforces and obscures the injustices they face.
Our findings build on previous reviews documenting the extent to which and how social and structural factors influence sex workers’ safety and vulnerability to HIV. They do so by showing how these factors interplay with criminalisation to further marginalise sex workers and deprive them of civil, labour, and social rights [134–137]. Fear of prosecution and moral judgement, due to laws against homosexuality and transgenderism [138] and drug use [135], and, in the case of migrant workers [139], fear of deportation, further reduce willingness to report violence and exploitation to the police. Other evidence has shown how evictions based on landlords’ fears of brothel-keeping charges increase vulnerability to homelessness for sex workers and their families, while arrest and criminal records or simply being identified as a sex worker can lead to sex workers’ children being placed in institutional care [135,140].
Despite including search terms relating to broader health outcomes, the majority of epidemiological literature focused on sexual health outcomes and, in more recent evidence, violence. We found few studies that focused on emotional health, but these show detrimental associations with repressive policing and criminalisation. Qualitative and quantitative studies demonstrate that police enforcement and its threat is a major source of anxiety [103,141], whereas working in indoor, decriminalised environments is associated with improved mental health outcomes [32,142]. A recent critical literature review demonstrates that criminalisation, stigma, poor working conditions, isolation from peer and social networks, and financial insecurity have negative repercussions for sex workers’ mental health [13]. Only 1 quantitative study reported on the associations between policing and violence from intimate or other partners, and further research is needed to understand the mechanisms of this relationship [58]. It is clear that criminalisation and stigma interact to reproduce sex workers’ exposure to physical and sexual violence, and limit possibilities to resist or challenge it, and interventions are urgently needed to address violence against sex workers from all perpetrators. Successful sex-worker-led approaches to improving access to justice and challenging institutional stigma in South India offer important examples of what can be achieved with sustained funding and support [99].
Findings clearly show that criminally enforced regulatory models create major disparities within sex worker communities, possibly enabling access to safer conditions for some but excluding the large majority who remain under a system of criminalisation, including trans women, cis men, people who use drugs, migrant populations, and often sex workers operating in outdoor environments, who are at increased risk of HIV in many settings [81,90,126]. In contexts of mandatory HIV testing following arrest, fear of enforcement can hinder voluntary uptake of HIV testing and interventions [71,80], showing how this punitive approach to public health ultimately reduces access to health services. More recent research from Senegal has shown that while registration was associated with better physical health, the stigma attached to being registered has a detrimental effect on well-being; only a minority of sex workers are registered, and those who test HIV positive are excluded [143]. As the qualitative synthesis demonstrates, in New Zealand, following decriminalisation, sex workers reported being better able to refuse clients and insist on condom use, amid improved relationships with police and managers [36,144,145]. Other research in this setting indicates that decriminalisation has the potential not only to reduce discrimination, denials of justice, denigration, and verbal abuse but also to improve sex workers’ emotional well-being [31]. This concords with existing modelling data that suggest a positive effect of decriminalisation on incidence of HIV [2].
We were unable to examine the effects of different legislative models in the quantitative synthesis due to limited data, particularly for the models of decriminalisation and the criminalisation of the purchase of sex. Evidence included in our qualitative synthesis clearly shows that criminalisation of clients does not facilitate access to services, nor minimise violence. This is supported by the epidemiological evidence from Vancouver that showed that sex workers who were stopped, searched, or arrested were at increased risk of client violence despite the introduction of more severe laws against the purchase of sex introduced in 2014 (alongside fewer sanctions for sex workers working together and modelled on the Swedish law) [57]. In addition, the practice of rushing negotiations due to police presence increased and was associated with increased client-perpetrated violence [92]. Findings from our qualitative synthesis suggest that enforcement strategies that seek to reduce the numbers of sex workers [118] or clients [114] are unlikely to achieve these effects, since the economic needs of sex workers remain unchanged, resulting in sex workers having to work longer hours, accept greater risks, and deprioritise health. There is no reliable evidence from Sweden that the numbers of sex workers have decreased since the law changed in 1999 [34].
There are a number of limitations to this review. Findings from our pooled meta-analyses examining condom use and violence were limited by high heterogeneity, although effect estimates remained consistent across sensitivity analyses, suggesting we can be confident in their robustness. By limiting the search to literature written in English, Russian, and Spanish, we may have missed key studies. There was a lack of comparable quantitative data on outcomes such as access to services, drug-related harms, and emotional ill health, which precluded the use of meta-analysis. Similarly, few qualitative studies explored the emotional health effects of criminalisation and enforcement, and its effects on access to health and broader services received less attention relative to safety and health risks, within the rich body of evidence reviewed. Methodologically, some studies did not provide sufficient detail on sampling and analysis methods, and few included reflexive discussions on the position of the researcher. Although a growing number involve sex workers as researchers or advisors, few included discussion of the challenges and benefits of participatory approaches. We found few eligible studies that included trans female or cis male sex workers, who experience particular inequalities in relation to HIV, access to services, and—as the qualitative synthesis shows—police targeting and violence, limiting our ability to generalise findings to these populations. It is also possible that some studies may not have differentiated between trans women and cis men [146], or between cis and trans participants within samples of female and male sex workers, and few disaggregated experiences or outcomes by gender. This is an important area of future research given the specific vulnerabilities experienced by these populations, in contexts where gender and sexual minorities are criminalised, inadequately protected against hate crimes, and, in the case of trans people, not legally recognised. There is particular need for research with trans women, who experience intense violence, discrimination, and exclusion from education and employment, and whose health needs have been obscured by their conflation with ‘men who have sex with men’ [146].
Our review focuses on the implementation of enforcement practices linked to 5 broad legislative models. While it is clear that sex work laws and enforcement practices are inextricably integrated and it is key to link practice to legal frameworks to inform policy-making and advocacy, our findings reinforce previous evidence [37,38] that shows wide variation in how laws are enforced, which vary with sex work setting [126], visibility of sex work, sex workers’ and managers’ relationships with individual officers [99,101], and political and media attention [110,125], or arbitrarily by city [121]. We report on recent and past history of arrest or prison based on the information available to us, but few studies reported whether the arrest was related to sex work, was related to another offence, or had to do with social, gender, or racial profiling. Assessing the extent to which the enforcement practice was lawful or unlawful is beyond the scope of this review, but in some cases unlawful activities are clearly evidenced (e.g., police violence) while in others they are less visible or evidenced. This limits our ability to assess the specific contribution of sex work penalties to the health and safety of sex workers, relative to the use of other penalties and abuses of police powers against sex workers in contexts of criminalisation. Lack of clarity on the lawfulness of police enforcement practices also reflects the difficulties in measuring stigma and its interaction with criminalisation, and the need for mixed-methods approaches to unpack these complexities in context. We found few data on the interplay between criminalisation, collective organisation, and health outcomes. Evidence from India has shown how tackling social injustice and mistreatment by the police as part of a sex-worker-led HIV prevention intervention has resulted in fewer arrests, more explanation of reasons for arrest, and fairer treatment by the police, as well as decreased violence against sex workers [84,99]. However, most evaluations of community-led health interventions have been limited to HIV prevention and have been implemented in India, Dominican Republic, and Brazil [147,148]. Although there are numerous examples of active sex worker organisations advocating for sex worker rights and evidence-based policy internationally, as well as developing guidelines for rights-based HIV programming with, for, and by sex workers [149], the voices of sex workers continue to be dismissed and silenced in policy debates in many settings as well as in the design and evaluation of public health interventions.
The public health evidence clearly shows the harms associated with all forms of sex work criminalisation, including regulatory systems, which effectively leave the most marginalised, and typically the majority of, sex workers outside of the law. These legislative models deprioritise sex workers’ safety, health, and rights and hinder access to due process of law. The evidence available suggests that decriminalisation can improve relationships between sex workers and the police, increasing ability to report incidences of violence and facilitate access to services [36,95,96]. Considering these findings within a human rights framework, they highlight the urgency of reforming policies and laws shown to increase health harms and act as barriers to the realisation of health, removing laws and enforcement against sex workers and clients, and building in health and safety protections [134]. It is clear that while legislative change is key, it is not enough on its own. Law reform needs to be accompanied by policies and political commitment to reducing structural inequalities, stigma, and exclusion—including introducing anti-discrimination and hate crime laws that protect sex workers and sexual, gender, racial, and ethnic minorities. Mixed-methods, interdisciplinary, and participatory research is needed to document the context-specific ways in which criminalisation or decriminalisation interacts with other structural factors and policies related to stigma, poverty, migration, housing, and sex worker collective organising, to inform locally relevant interventions alongside legal reform. This research must go alongside efforts to examine concerns surrounding decriminalisation of sex work within institutions and communities, which influence policy and practice, and sex workers must be involved in decision-making over any such research and reforms [121,150]. Opponents of decriminalisation of sex work often voice concerns that decriminalisation normalises violence and gender inequalities, but what is clear from our review is that criminalisation does just this by restricting sex workers’ access to justice and reinforcing the marginalisation of already-marginalised women and sexual and gender minorities. The recognition of sex work as an occupation is an important step towards conferring social, labour, and civil rights on all sex workers, and this must be accompanied by concerted efforts to challenge and redress cultures of discrimination and violence against people who sell sex. While such reforms and related institutional shifts are likely to be achieved only in the long term, immediate interventions are needed to support sex workers, including the funding and scale-up of specialist and sex-worker-led services that can address the multiple and linked health and social care needs that sex workers may face.
|
10.1371/journal.ppat.1002998 | Identification of a Novel Splice Variant Form of the Influenza A Virus M2 Ion Channel with an Antigenically Distinct Ectodomain | Segment 7 of influenza A virus produces up to four mRNAs. Unspliced transcripts encode M1, spliced mRNA2 encodes the M2 ion channel, while protein products from spliced mRNAs 3 and 4 have not previously been identified. The M2 protein plays important roles in virus entry and assembly, and is a target for antiviral drugs and vaccination. Surprisingly, M2 is not essential for virus replication in a laboratory setting, although its loss attenuates the virus. To better understand how IAV might replicate without M2, we studied the reversion mechanism of an M2-null virus. Serial passage of a virus lacking the mRNA2 splice donor site identified a single nucleotide pseudoreverting mutation, which restored growth in cell culture and virulence in mice by upregulating mRNA4 synthesis rather than by reinstating mRNA2 production. We show that mRNA4 encodes a novel M2-related protein (designated M42) with an antigenically distinct ectodomain that can functionally replace M2 despite showing clear differences in intracellular localisation, being largely retained in the Golgi compartment. We also show that the expression of two distinct ion channel proteins is not unique to laboratory-adapted viruses but, most notably, was also a feature of the 1983 North American outbreak of H5N2 highly pathogenic avian influenza virus. In identifying a 14th influenza A polypeptide, our data reinforce the unexpectedly high coding capacity of the viral genome and have implications for virus evolution, as well as for understanding the role of M2 in the virus life cycle.
| Influenza A virus is a pathogen capable of infecting a wide range of avian and mammalian hosts, causing seasonal epidemics and pandemics in humans. In recent years, the unexpected coding capacity of the virus has begun to be unravelled, with the identification of three more protein products (PB1-F2, PB1-N40 and PA-X) on top of the 10 viral proteins originally identified 30 years ago. Here, we identify a 14th primary translation product, made from segment 7. Previously established protein products from segment 7 include the matrix (M1) and ion channel (M2) proteins. M2, made from a spliced transcript, has multiple roles in the virus lifecycle including in entry and budding. In a laboratory setting, it is possible to generate M2 deficient viruses, but these are highly attenuated. However, upon serial passage a virus lacking the M2 splice donor site quickly recovered wild type growth properties, without reverting the original mutation. Instead we found a compensatory single nucleotide mutation had upregulated another segment 7 mRNA. This mRNA encoded a novel M2-like protein with a variant extracellular domain, which we called M42. M42 compensated for loss of M2 in tissue culture cells and animals, although it displayed some differences in subcellular localisation. Our study therefore identifies a further novel influenza protein and gives insights into the evolution of the virus.
| Influenza A virus (IAV) is a genetically diverse pathogen of global significance, responsible for seasonal epidemics and sporadic pandemics in humans, as well as outbreaks in domestic animals. Its primary reservoir is wild birds, but it can infect a wide range of vertebrate species. For these reasons, there is the need to develop better therapeutics and vaccines [1]. Current vaccines target the surface glycoproteins haemagglutinin (HA) and neuraminidase (NA), but these proteins are subject to antigenic change, necessitating regular updating of the vaccine to ensure a good antigenic match to the circulating strains. Next generation influenza vaccines seek to induce broader or ‘universal’ protection against conserved epitopes; for example, the ‘stalk’ region of HA or the ectodomain of the matrix 2 ion channel protein (M2) [2], [3].
The IAV genome consists of eight segments of negative sense, single stranded RNA (vRNA), each encapsidated into ribonucleoproteins (RNPs) by the viral RNA dependent RNA polymerase and multiple copies of the viral nucleoprotein (NP). Upon infection, incoming RNPs are imported into the nucleus, where the vRNA is transcribed to give positive sense mRNA, and also cRNA, which acts as a replication intermediate. The approximately 13 kb genome has so far been demonstrated to encode up to 13 proteins [4], [5]. Segments 1, 4, 5 and 6 each encode a single protein: PB2, HA, NP and NA respectively. However, segments 2, 3, 7 and 8 have additional protein coding capacity. Segments 2 and 3, whose primary protein products are the polymerase proteins PB1 and PA respectively, additionally produce PB1-F2, PB1-N40 and PA-X proteins from single mRNA species by leaky ribosomal scanning and translation termination-reinitiation in the case of segment 2 and +1 ribosomal frameshifting for segment 3 [4]–[7].
In segments 7 and 8, protein coding capacity is expanded by differential mRNA splicing. For segment 8, a single spliced species has been described, producing NS2/NEP, while NS1 is produced from the unspliced transcript [8], [9]. Segment 7 mRNA splicing is more complex, as three spliced transcripts have been described (designated mRNAs 2–4) in addition to the unspliced mRNA1 [10]–[12]. Unspliced mRNA1 gives rise to M1 protein. The spliced mRNAs use a common 3′-splice acceptor (SA) site, but use different 5′-splice donor (SD) sites (Fig. 1A). To date, only mRNA2 has been demonstrated to encode a protein: the M2 ion channel [13]. mRNA3 is produced from the most 5′-proximal SD, and is proposed to negatively regulate segment 7 protein expression during early infection [14], a non-essential function for virus growth in tissue culture [15], [16]. More recently, mRNA4 has been shown to be produced by the A/WSN/33 (WSN) strain of virus [12], [15], [17]. It hypothetically encodes an internally deleted form of the M1 protein (“M4”; Fig. 1A) but this protein has not been detected [12].
M2 is a 97 aa integral membrane protein, functional as a homotetramer, with multiple important roles during the virus lifecycle [18], [19]. Each monomer consists of a 24 aa N-terminal ectodomain, a transmembrane α-helix and a ∼50 aa cytoplasmic domain that contains a membrane proximal amphipathic alpha helix [20], [21]. M2 has proton channel activity, which is important for acidification of the interior of the virion upon entry [22]–[24]. In some strains of virus, proton conductance plays an additional role in modulating the pH of the Golgi compartment to prevent premature activation of the HA fusion apparatus [22], [25]. The cytoplasmic tail of M2 also has roles in virus assembly, budding and morphogenesis [26]–[32]. The function of the ectodomain is less well described, although along with the transmembrane domain, it likely plays a role in directing the membrane topology of M2 [33], [34]. It may also be important for incorporation of the protein into virions [35]. Nevertheless, the ectodomain is highly conserved amongst virus strains and this has made it an attractive candidate for a universal influenza vaccine [2], [3].
Surprisingly, it has been possible to generate M2 null viruses, either by introduction of stop codons or by mutating the splice donor site, although these viruses are highly attenuated [15], [36]–[39]. Here, we describe a pseudoreversion mechanism of a virus with a mutated mRNA2 SD site, which reveals a new aspect of IAV biology. After serial passage, we identified a single mutation that upregulated mRNA4 expression without restoring M2 synthesis. Instead, mRNA4 encodes an M2 variant with an alternative ectodomain, designated here M42, which nevertheless functionally complements M2, in vitro and in vivo. Furthermore, we present evidence that certain strains of IAV, most notably those responsible for the 1983 Pennsylvania outbreak of highly pathogenic avian influenza (HPAI), normally express M42. Our data extend the known IAV proteome and have implications for virus evolution and vaccine design.
Previously, we used reverse genetics to create an A/PR/8/34 (PR8) virus with synonymous mutations to the mRNA 2 SD sequence. This virus, (M1 V7-T9, hereafter named V7-T9), did not produce detectable levels of M2 and was highly attenuated in tissue culture [37]. To better understand the role of M2 in the virus life cycle, we studied the mechanism by which V7-T9 could regain fitness upon serial passage.
WT and V7-T9 viruses were subjected to six rounds of serial passage via low multiplicity infections of MDCK cells. At each round, outputs were titred by plaque and HA assay. Before serial passage (“P0”), the input V7-T9 virus replicated to a plaque titre 400-fold lower than the WT and had an HA titre 100-fold lower (Fig. 2A). However, on serial passage it regained fitness rapidly, producing similar plaque and HA titres to WT virus within two passages. As a further test of fitness recovery, the plaque areas of the WT and V7-T9 viruses were measured before and after serial passage. Prior to serial passage, V7-T9 displayed a small plaque phenotype ([37]; Fig. 2B). However, after passage six (P6), its average plaque area had increased over four-fold and was no longer significantly different from that of the WT virus (Fig. 2B).
To test if the regained fitness resulted from restoration of M2 expression, we examined infected cell lysates from the original and serially passaged versions of the WT and V7-T9 viruses by western blotting for M1 and M2. All infected cells showed abundant M1 expression, confirming infection (Figs. 2C). Cells infected with the WT virus isolates also contained a polypeptide recognised by the M2 ectodomain-specific 14C2 monoclonal antibody, but as before [37], cells infected with the original V7-T9 virus did not; a phenotype that remained unchanged in the serially passaged isolate (Fig. 2C, lanes 2–5). However, 14C2 antibody recognition is restricted to an epitope encompassing residues 4 to 16 of M2 [40]–[42]. When a polyclonal antibody raised against the entire M2 protein, G74 [43], was used, the original V7-T9 virus still did not show any reactivity (Fig. 2C). However, the P6 V7-T9 virus produced detectable amounts of a G74-reactive polypeptide of similar electrophoretic mobility to that of M2 (Fig. 2C, compare lanes 4 and 5), suggesting that it now expressed some M2 polypeptide, albeit with different antigenicity to the WT protein.
To further investigate M2 expression by the P6 V7-T9 virus, we examined cells infected with passaged or unpassaged WT and mutant viruses by indirect immunofluorescence for NP (to identify infected cells) and M2, using the two M2-specific sera. All infected cells stained strongly for NP, confirming similar levels of infection (Fig. 3; in red). Consistent with the western blot data, WT virus infected cells also stained strongly with both 14C2 and G74 anti-M2 antibodies, showing the expected predominant staining of apical and lateral membranes [44], while neither isolate of the V7-T9 virus reacted with the 14C2 monoclonal (Fig. 3; in green or separate channel in grey). Also consistent with the western blot data, the unpassaged V7-T9 virus did not stain above background levels with the G74 antiserum, but the P6 isolate showed clear reactivity. However, the staining pattern was markedly different to that shown by WT virus, with prominent perinuclear staining and some staining of lateral membranes (Fig. 3B). Overall, these data suggested that the serially passaged M2 null virus had regained fitness by expressing a variant form of M2 that no longer reacted with the ectodomain-specific antibody.
Following serial passage, segment 7 of both the P6 WT and V7-T9 viruses was sequenced. No changes were detected in the WT virus compared to the reference sequence (GenBank accession EF467824). The P6 V7-T9 virus retained the original mutations that destroyed the mRNA2 SD site, indicating pseudoreversion to recover WT growth properties rather than true reversion. It also contained a single additional change, not seen in the original V7-T9 virus, of a U to A substitution at nucleotide 148 (U148A; mRNA sense, Fig. S1). This change is in the M2 intron and is silent in M1. However, the change would be predicted to improve the mRNA4 SD consensus (Fig. S1), from AG/GUU to AG/GUA [45]. As previously noted, mRNA4 is predicted to encode a 54 aa internally deleted version of M1 [12], from the first AUG on the transcript (Fig. 1B). Notably, the Kozak consensus [46] of AUG1 is not optimal, lacking a G at position 4 (Figs. 1B, S1) and in the context of segment 2, an intermediate strength initiation context AUG1 is known to permit translation initiation at downstream codons by a leaky ribosomal scanning mechanism [4], [6], [7]. Inspection of the segment 7 sequence showed another AUG starting at position 114 in frame 2 (Fig. 1B). The predicted protein product from this AUG would have a variant ectodomain compared to M2, but would be identical from amino acid 10 onwards. The predicted size of the protein product would be 99 amino acids, compared to 97 for M2 (Fig. 1C).
Accordingly, we hypothesized that the U148A change induced pseudoreversion of the V7-T9 virus via upregulation of mRNA4, to produce a variant M2 protein (designated here “M42”) from AUG2 via leaky ribosomal scanning. To test this, we used reverse genetics to first ask whether the U148A change was sufficient to restore WT growth properties to the V7-T9 virus. Initially, a PR8 V7-T9+U148A virus was generated, along with WT, V7-T9 and a virus with only the U148A change. Viruses were rescued by transfecting bidirectional plasmids [47] into 293T cells, amplified by one passage in MDCK cells and plaque titred. WT PR8 grew to approximately 7×108 PFU/ml and formed large plaques, whereas V7-T9 had a small plaque phenotype and was attenuated by approximately 3 log10 (Fig. 4A), consistent with previous observations [37]. Introduction of the single U148A mutation into the background of an otherwise WT virus did not alter virus growth properties. However, when the change was added to the V7-T9 background, the double mutant grew to an average of 5×108 PFU/ml and produced normal-sized plaques (Fig. 4A), confirming that the U148A mutation was necessary and sufficient to restore WT growth properties.
To further test the M42 hypothesis, we introduced two mutations that would be expected to block production of the predicted novel polypeptide: either by removing its AUG codon (U115C), or by destroying the mRNA4 SD site (G145A). Each of these mutations was made on the background of WT segment 7, as well as with the V7-T9, U148A or V7-T9+U148A mutations. On a WT background, a virus with only the U115C mutation grew normally and produced plaques indistinguishable from the WT virus (Fig. 4A). When the U115C mutation was combined with the U148A change, the resulting virus grew slightly less well than WT (an average relative titre of 0.44 [n = 4]) and displayed a small plaque phenotype. Addition of U115C to the V7-T9 mutant also had only a minor effect on growth relative to the parent virus. In contrast, its addition to the V7-T9+U148A background reversed the positive effect of the U148A mutation, resulting in a virus that grew poorly (to less than104 PFU/ml) and produced small plaques. Similarly, the G145A mutation had no effect on virus growth as a single mutation or when combined with the U148A change. However, in 3 independent attempts, it was not possible to rescue a virus with V7-T9, U148A and G145A mutations, suggestive of a lethal phenotype.
These data indicated that pseudoreversion of the M2-null virus required mRNA4 and also AUG2. As an additional genetic test of the M42 hypothesis, we introduced a premature stop codon (K70*) into the distal region of the M2 ORF that would be common to both M2 and M42 polypeptides, but outside the M1 (or hypothetical M4) coding region. As a control K70 was also substituted for tryptophan (K70W), a similar sequence change but known to be compatible with M2 function [29]. It was not possible to rescue a virus with V7-T9+U148A+K70stop, although the V7-T9+U148A+K70W mutant grew comparably to WT and V7+U148A viruses. Together, the genetic data are consistent with the hypothesis that the U148A change restores growth of an M2-deficient virus by upregulating expression of mRNA4, allowing expression of an M2 variant from AUG2 of the transcript.
To provide biochemical evidence for the M42 hypothesis, we next examined segment 7 mRNA splicing by the panel of viruses in 293T cells. The V7-T9+U115C+U148A virus grew to insufficient titres to allow high multiplicity synchronous infections and the V7-T9+G145A+U148A virus could not be rescued, so the U115C+U148A and G145A+U148A viruses were used as proxies to analyse the effects of the U115C and G145A mutations on mRNA expression. Following infection, total RNA was extracted and reverse transcriptase-primer extension reactions were performed using a single primer capable of distinguishing segment 7 mRNAs 1–4 [17]. Separate primers specific for segment 7 vRNA and cellular 5S rRNA were also included as controls for virus infection and RNA recovery respectively. The levels of 5S rRNA were equivalent between samples (Fig. 4B), demonstrating equal loading. vRNA levels were also comparable between infected samples, suggesting that all infections had proceeded successfully. The levels of unspliced mRNA1 were also similar between the viruses. However, large differences in the levels of spliced mRNAs 2 and 4 were apparent. In cells infected with the WT virus, the unspliced transcript predominated, but abundant levels of mRNA2 (for M2) were also present (Fig. 4B, lane 1; quantification in Fig. 4C). In contrast, mRNAs 3 and 4 formed minor species that were only visible on long exposure (primary data not shown, but see Fig. 4C for quantification). As expected, mRNA2 was not detected in viruses containing the V7-T9 mutation (Fig. 4B, lanes 5 and 6). Importantly, and as predicted, the U148A mutation, either alone or on a V7-T9 background, strongly upregulated production of mRNA4 (compare lanes 1, 3 and 6). This effect was blocked when the mRNA4 SD was destroyed with a G145A change (lane 7). Interestingly, the changes in levels of mRNAs 2 and 4 were partly reciprocal. Loss of the mRNA2 SD site in the V7-T9 virus was associated with weak upregulation of mRNA4 (compare lanes 1 and 5), while improvement of the mRNA4 SD by the U148A change in an otherwise WT background led to around a three-fold drop in mRNA2 levels (compare lanes 1 and 3). Addition of the U115C change to the U148A virus caused a further decrease in mRNA2 levels, but left mRNA4 levels unaltered (compare lanes 3 and 4). Overall, these data supported the proposed mechanism of pseudoreversion involving increased production of mRNA4.
Next, we analysed segment 7 protein expression from the mutant viruses by western blotting for M1, and for M2 using 14C2 and G74 antisera. Lysates from the primary reverse genetics transfections in 293T cells were used, because V7-T9+G145A or V7-T9+G145A+U148A viruses could not be obtained. Virus polypeptides in these lysates will therefore come from several sources: from RNA Pol II transcription of the bidirectional plasmid, from viral transcription in cells where active RNPs have been reconstituted by transfection, and from spread of viable virus through the cell culture. To control for purely plasmid-mediated expression, lysates from a transfection where PB2 was omitted were probed. In this sample, levels of M1 and M2 were below the limit of detection, although they were readily visualised when all eight plasmids were transfected (Fig. 4D, compare lanes 1 and 2). This suggested that under these conditions, the major signal came from viral gene expression. When mutant and WT transfections were compared, M1 levels were broadly similar between samples, but there was more variation in M2 levels. As expected, 14C2 reactivity was only detected from viruses with an intact mRNA2 SD (lanes 2–6) and was absent from all of the V7-T9 family of viruses (lanes 7–12). G74 reactivity was also readily detectable in all samples from viruses able to make mRNA2. However, in the absence of mRNA2, it was only detectable in the V7-T9+U148A transfected lysates (lane 10). Significantly, this was dependent on the presence of both elevated mRNA4 levels and segment 7 AUG2, as addition of either or the G145A or U115C mutations ablated its expression (compare lanes 10, 11 and 12).
Next, to prove the existence of the M42 polypeptide, we raised a specific antibody against a peptide corresponding to the predicted novel ectodomain of PR8 M42. To validate the serum, we tested it against transfected M42 and M2, both fused to GFP. M42-GFP, M2-GFP or GFP alone were transfected into 293T cells and the resulting cell lysates were probed with anti-M42 and anti-M2 14C2 or G74. Samples were also probed with anti-GFP and tubulin antisera, to confirm expression of the GFP polypeptides and equal sample loading respectively (Fig. 5A). The anti-M42 serum detected M42-GFP with a high degree of specificity over M2-GFP (compare lanes 1 and 2). Conversely, the 14C2 antibody was specific for M2-GFP, while as expected, anti-G74 detected both M42 and M2-GFP. A preimmune bleed from the rabbits immunized with the M42 peptide did not react with either M42-or M2 GFP. To further probe the immunological cross-reactivity between M2 and M42, we tested a polyclonal antiserum raised against the entire M2 ectodomain, anti-M2e [48]. This reacted strongly with M2-GFP and only weakly with M42-GFP, confirming the novel antigenicity of the M42 ectodomain.
Having generated a specific M42 antibody, we investigated expression of the protein in cells infected with the panel of WT and mutant viruses. Western blot analysis of MDCK cell lysates using anti-tubulin sera demonstrated equivalent loading of all samples while anti-M1 and anti-NP sera showed equal levels of infection except in mock infected cells (Fig. 5B). WT, U115C, G145A and G145A+U148A viruses expressed abundant quantities of a polypeptide that reacted with 14C2 and G74 anti-M2 sera (Fig. 5B, lanes 2–5 and 7). In contrast, the V7-T9 virus did not produce detectable levels of any M2-related polypeptide (lane 8). However, concomitant upregulation of mRNA4 on this background by the addition of the U148A mutation led to abundant synthesis of an anti-M42 reactive polypeptide (lane 9), confirming our hypothesis of a novel M2-related polypeptide. The U148A mutation also led to synthesis of readily detectable amounts of the M42 polypeptide when introduced into an otherwise WT background (lane 5). M42 reactivity was however lost on this background by mutation of AUG2 with the U115C change, or by mutation of the mRNA4 SD using G145A (lanes 6 and 7). Consistent with the mRNA abundance data, overall amounts of M2 were reduced by the U148A mutation, as judged by 14C2 and G74 staining (compare lanes 1 and 5). In addition, double staining the same blot with the mouse 14C2 antibody in red and the rabbit M42 antibody in green allowed the creation of a merged image (Fig. 5B, top panel) that illustrates the similar molecular weights of the M2 and M42 polypeptides, as well as their changing relative abundance in response to mutations to SD sites of mRNAs 2 and 4.
Immunofluorescent staining of P6 V7-T9-infected cells suggested that the two forms of M2 localised differently within infected cells (Fig. 3B). To test if this truly reflected a difference in behaviour of M42 compared to M2, we infected cells with the relevant recombinant viruses and examined M2 and M42 localisation by immunofluorescence. In WT virus infected cells, M2 protein localized to the plasma membrane (visible as staining of lateral membranes in single optical slices through the midline of the cells) as well as internally, often in a perinuclear position (Fig. 6A). Double staining for M42 however, only produced background levels of fluorescent signal, similar to mock infected or V7-T9 infected cells. In contrast, cells infected with the V7-T9+U148A virus did not stain for M2 but stained strongly with anti-M42, with the M42 signal largely present in a perinuclear structure. Confirming that the two proteins did indeed localize differently in infected cells, when cells infected with the U148A virus (which expresses both proteins; Fig. 5B) were examined, the two polypeptides displayed limited colocalisation in the perinuclear region, but were largely in separate regions of the cell (Fig. 6A). Double staining of cells infected with the V7-T9+U184A virus with anti-M42 sera and a variety of markers for the cellular exocytic pathway showed good co-localisation with GM130 (Fig. 6B and data not shown), indicating that M42 was largely resident in the cis-Golgi apparatus. To examine intracellular trafficking of M42 further, we created a plasmid encoding an M42-red fluorescent protein fusion (M42-mCherry) and examined the localization of the chimaeric protein in living cells, in comparison to a simultaneously transfected M2-GFP fusion. Both polypeptides co-localised in discrete cytoplasmic puncta and at the plasma membrane, but while the intensity of GFP and mCherry fluorescence was similar in the cytoplasmic dots, there was clearly less of the M42-mCherry protein on the plasma membrane at steady state (Fig. 6C). Examination of time lapse films showed that the cytoplasmic puncta showed the expected pattern of movement for intracellular vesicles (Videos S1, S2, S3). Overall therefore, we conclude that like M2, M42 enters the exocytic pathway, but its altered ectodomain affects intracellular trafficking of the protein, resulting in a lower proportion resident at the plasma membrane.
The marked difference in intracellular localization between M2 and M42 was surprising, given that the two proteins were apparently interchangeable with respect to virus replication in MDCK cells (Fig. 4A). We therefore tested the effect of modulating M42 expression on virus pathogenicity, using the murine infection model. When either BALB/c or C57BL/6 strain mice were infected with 100 PFU of WT PR8 virus, they lost weight rapidly (Figs. 7A, B), showing average peak weight losses of around 20% and substantial amounts of mortality (Fig. 7C). High titres of virus were also recoverable from the lungs of infected C57BL/6 mice at days 2 and 4 p.i., dropping somewhat at day 7 (Fig. 7D). In contrast, mice infected with the same titer of the M2-null V7-T9 virus showed minimal weight loss, few signs of illness or virus replication and no mortality. However, upregulation of mRNA4 synthesis via U148A on the V7-T9 background substantially increased virus replication and pathogenicity in terms of virus titres andweight loss, although the overall mortality was less than observed with WT virus. Conversely, increasing M42 expression by adding the U148A change on the background of a WT virus still able to express M2 had the opposite effect, decreasing the severity of weight loss and overall mortality, although lung titres were not affected. Removal of the M42 AUG codon with the U115C mutation had little effect in BALB/c mice but led to slightly delayed weight loss and decreased mortality in C57BL/6 mice. Overall therefore, altering the balance between M2 and M42 expression modulated virus pathogenicity, but a virus that only expressed M42 still caused significant disease.
The work described above demonstrated that mRNA4 encodes a biologically significant polypeptide that can compensate for loss of M2 expression. The question therefore arose as to whether this might apply to other strains of IAV. The two requisites for M42 expression are production of mRNA4 and the possession of an AUG codon in the appropriate reading frame. mRNA4 was originally discovered in the WSN strain of virus but was not detected in the A/Udorn/72 (Udorn) strain, a difference proposed to result from a single nucleotide difference in the sequences immediately surrounding the splice site: AG/GUU in WSN versus AG/GCU in Udorn ([12]; see Fig. S1, which shows an alignment of the viruses used or discussed in this work, in addition to the consensus sequences of the major virus subtypes that have infected humans this century). To test this prediction, we compared mRNA4 synthesis in a panel of viruses with either GUA, GUU or GCU at the 5′-end of the mRNA4 intron. In agreement with the quality of match with the consensus SD sequence, mRNA4 was not detectable in the two viruses with a GCU sequence: human H3N2 Udorn and H1N1 A/USSR/77 (Fig. 8A, lanes 6 and 7), while it was most abundant in the PR8 U148A mutant (GUA; lane 2). The prediction was also partially supported when mRNA4 synthesis was examined in viruses with a GUU sequence immediately downstream of the SD site. Intermediate quantities of mRNA 4 were detected in RNA from WSN and Cambridge PR8 virus infected cells (lanes 4 and 5; note lower overall amounts of segment 7 mRNAs with the latter virus). Curiously however, mRNA4 was not seen from reverse genetics PR8 (compare lanes 3–5). This was despite segment 7 of this virus only differing from Cambridge PR8 and WSN at two nucleotide positions in the 5′-240 nucleotides, with neither change located near the mRNA4 SD sequence (Fig. S1 and data not shown). There were also notable differences in the amount of mRNA3 produced by the viruses, with WSN and Udorn making abundant quantities, A/USSR/77 rather less and all three PR8 viruses making very little (Fig. 8A). Thus mRNA4 production is predictable by examination of the SD consensus sequence, with GUA>GUU>>GCU, although other unidentified sequence polymorphisms also play a role.
We therefore used this information to interrogate Genbank for IAV segment 7 sequences likely to express M42. As of October 2011, over three-quarters of the 20,236 viruses for which useful segment 7 sequence was available contain the M42 AUG (data not shown). An imperfect initiation context of the M1/M2 AUG codon (which is likely to be necessary to allow leaky ribosomal scanning) is a very highly conserved feature of IAV (only 3 of 17256 sequences covering the M1 AUG have an optimal G at position +4). However, the majority (∼80%) of viruses are unlikely to produce substantial amounts of mRNA4, as they possess an unfavourable AG/GCU or otherwise non-canonical SD sequence. In 1998, Shih and colleagues identified 8 viruses likely to make appreciable amounts of mRNA4 [12]. Now, with an increased number of sequences available as well as a better understanding of the sequence elements necessary for expression of a third biologically active protein from segment 7, we identified around two dozen viruses likely to express M42 (Table 1), by virtue of containing an AUG codon at positions 114–116 and an mRNA4 SD sequence of AG/GU(U/A/G). These mostly fell into three partially overlapping groups: isolates from the early 20th Century (human isolates from the 1930s and two classic fowl plague highly pathogenic avian influenza (HPAI) viruses), viruses that had been adapted to replicate in mice (the WS family, two H3N2 isolates and PR8) and a set of HPAI isolates, mostly from the USA 1983 outbreak [49]. The latter H5N2 grouping seemed the most likely to express large amounts of M42 because of their AG/GUA mRNA4 SD sequence. In addition, the H5N2 outbreak spread widely, persisted for several years and resulted in the culling of 17,000,000 poultry [50], making it an important group of non laboratory-derived viruses, even if represented on Genbank by relatively few sequenced isolates.
We therefore tested whether the AG/GUA mRNA4 SD consensus of the H5N2 viruses was biologically significant. For biosafety reasons, we used reverse genetics to create a PR8 reassortant (MPenn) with segment 7 from A/chicken/Pennsylvania/10210/1986 (Penn) as well as various mutant derivatives with alterations to the mRNA2 or 4 SD sequences or the M42 AUG codon (Fig. S1), and then analysed their segment 7 mRNA expression profiles. Analysis of viral RNA synthesis showed that, as predicted by the MPenn mRNA4 SD sequence, mRNA4 was the predominant species made from segment 7, accumulating to markedly higher levels than either the unspliced transcript or spliced mRNAs 2 and 3; a reversal of the ratios seen with the ‘prototype’ mRNA 4-expressing virus, WSN, where mRNA4 was the least abundant species (Fig. 8B, compare lanes 2 and 10). Mutations to the mRNA2 and mRNA4 SD sequences had the expected effects. Destruction of the mRNA2 SD sequence by a G52C change reduced mRNA2 accumulation to below detectable levels (lane 3). Removal of the mRNA4 splice site with the G145A change blocked detectable synthesis of mRNA4 with, as before, the side effect of upregulating mRNA2 and mRNA3 production (lane 5). Mutations that weakened the mRNA4 SD consensus (A148G/U or C) dramatically reduced mRNA4 accumulation whilst simultaneously improving synthesis of mRNAs 2 and 3 (lanes 6–9). Also as expected, these changes were specific to segment mRNA, as the levels of segment 7 vRNA and segment 5 mRNA and cRNA were much more consistent between viruses (Fig. 8B).
Next the impact of these changes on virus growth were assessed. The WT MPenn reassortant virus grew well, reaching titres of around 107 PFU/ml (Fig. 8C). Abolition of mRNA2 expression (G52C) had no effect on virus replication; in contrast to the attenuation seen when M2 synthesis was blocked in other virus strains [15], [36]–[39]. Similarly, mutations predicted to block M42 expression by destroying its AUG codon (U115C) or mRNA4 production (G145A) had no effect on virus growth. A similar lack of effect on virus titres were seen from the mutations that attenuated mRNA4 production: A148G, A148U and A148C. However, double mutations targeting both M2 and M42 production were deleterious to virus growth. Viruses lacking the M42 AUG codon or mRNA4 SD sequence could not be rescued (in 3 attempts) in combination with the G52C mRNA 2 SD knockout (Fig. 8C). Moderate downregulation of mRNA4 production by an A148G change in the absence of mRNA 2 synthesis led to a virus that grew to high titres but with a small plaque phenotype, while a more severe downregulation of mRNA4 synthesis with an A148U change resulted in an additional phenotype of very poor growth. Overall, these data indicate that the A/chicken/Pennsylvania/10210/1986 segment 7 expresses two functionally redundant versions of the viral ion channel, either one of which is sufficient to support replication in cultured cells. Thus M42 expression is not peculiar to laboratory adapted viruses but is likely to have been a feature of a major group of HPAI viruses that circulated for four years in North America.
Here, we demonstrate expression of a 14th IAV polypeptide; a variant form of the M2 protein with an alternative ectodomain, encoded by a distinct segment 7 mRNA. This novel protein, M42, can functionally replace M2 and support efficient replication in tissue culture cells and pathogenicity in an animal host, despite showing clear phenotypic differences with respect to its intracellular localization. We have not directly tested M42 for proton conductance but the PR8 MUd virus engineered to express M42 rather than M2 retained amantadine sensitivity (data not shown), providing indirect evidence that the protein retains this function, as expected from its identical transmembrane domain sequence to M2. The ability of M42 to support efficient virus replication despite its inefficient transport to the plasma membrane is interesting in light of current theories regarding the role of M2 in membrane scission [32].
Like the three other IAV “accessory” genes that were discovered long after the virus genome was sequenced (PB1-F2, PB1-N40 and PA-X; [4]–[6]), M42 is clearly non-essential for virus replication, as long as sufficient M2 is expressed. Unlike the additional proteins expressed from the P protein genes, M42 expression is likely to be restricted to a minority of IAV strains under normal conditions, as a result of the suboptimal SD sequence of mRNA4. Examination of the consensus sequences for the major subtypes of IAV that have infected humans in the last century showed that (in consensus, with occasional exceptions) all possess(ed) a weak mRNA4 SD sequence (GCU at the intron boundary) of the type found in Udorn (Fig. S1). However, all these viruses except the current 2009 swine-origin pandemic virus also contain the M42 AUG codon as well as an imperfect Kozak consensus around the M1/M2 AUG codon (Fig. S1), suggesting the potential for M42 expression should mRNA4 expression be present. This perhaps argues that there are environments in which it is advantageous for IAV to shift the balance of segment 7 splicing away from the normal mRNA2/M2 route to increased mRNA4/M42. In this respect it is noteworthy that increased mRNA4 synthesis has been selected for on at least three, probably four, occasions on different virus backgrounds during adaptation to growth in mice (Table 1). Also, given the Golgi-biased localisation of M42, it is tempting to draw a link between the requirement for pH-modulation of the Golgi during intracellular transport of HA molecules with polybasic cleavage sites [25], [51] and the overrerpresentation of HPAI viruses in the list of those likely to express M42. We also speculate that the altered antigenicity of the M42 ectodomain might provide the virus with a route to escape selection pressure imposed by a vaccine directed against the M2e sequence, given that in many viruses, a single nucleotide change would be predicted to alter the balance of splicing towards M42.
The viruses in which we can be reasonably confident of M42 expression represent a very small minority (∼0.2%) of the available sequences. However, there are two further considerations that may render M42 expression more widespread in IAV than our conservative prediction in Table 1. Firstly, we do not yet fully understand what controls segment 7 splicing. A sizeable number of viruses (around 15%; mostly from avian hosts) have an mRNA4 SD sequence of AG/GCA. An A at position+3 clearly promotes more efficient use of the splice site when position +2 is U but it remains to be determined if it is sufficient to override a C at +2. Furthermore, the differences in relative splicing seen between PR8 and WSN viruses make it clear that sequence elements outside of the core consensus splice sites affect their use; these sequences are identical in the two viruses but their splicing patterns are very different. Analysis of a 7+1 PR8:WSN reassortant indicates that the difference is intrinsic to segment 7 (HW, PD, unpublished data) but we have not yet identified the sequence determinants. Secondly, there are many precedents for cell-type dependence of alternative splicing in cellular mRNAs [52] so it may be that in some host species and/or cell types, M42 expression is present in a wider array of IAV strains. Further experiments are required to test these hypotheses.
Animal experiments were carried out in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals under the auspices of an NIH Animal Care and Use Committee-approved animal study protocol. The protocol was approved by the NIAID Animal Care and Use Committee (Permit Number LID-6E). All efforts were made to minimize suffering.
Madin-Darby Canine Kidney, 293T human embryonic kidney and A549 human lung adenocarcinoma cells were cultured according to standard procedures [53]. A reverse genetics clone of influenza PR8 and its M2-null derivative, V7-T9 have been previously described [37], [47]. A clone of A/chicken/Pennsylvania/10210/1986 (GU052748) with the UTRs derived from A/chicken/Pennsylvania/1/1983 (CY015074; where a complete segment sequence was available) was synthesized by Genscript and cloned into pDUAL reverse genetics vector [47]. Further mutants were made using the 8 bidirectional promoter plasmid system described in [47], following oligonucleotide-directed mutagenesis to introduce the desired changes, as indicated in Fig. S1. Primer sequences are available upon request. Non-recombinant Cambridge lineage PR8 virus, A/USSR/77 and A/WSN/33 viruses were obtained from the University of Cambridge Division of Virology's collection of viruses. A/Udorn/72 was a gift from Professor Compans [54]. M2 and M42-GFP tagged expression constructs were produced by cloning the coding sequences of the respective proteins into the KpnI/AgeI sites of pEGFP-N1 (Clontech). An M42-mCherry fusion was made by substituting the EGFP open reading frame with mCherry. Purchased monoclonal antisera were against ß-tubulin (clone YL1/2: AbD-Serotec), GM130 (Clone 610822; BD Transduction Laboratory), GFP (clone JL8, Clontech) and anti-influenza M2 (Clone 14C2, Abcam). Further anti-M2 reagents of a goat polyclonal (G74) raised against the whole protein and a mouse polyclonal raised against the M2 ectodomain (M2e) were the generous gifts of Drs. Alan Hay and Xavier Saelens, respectively. Rabbit polyclonal anti-M1 (A2917) and anti-NP (A2915) have been previously described [55], [56]. Affinity purified anti-M42 specific serum was purchased from Genscript. Rabbits were immunized with a peptide corresponding to the N-terminal 16 amino acids of the protein, MSLQGRTPILRPIRNE (where unique sequences compared to M2 are underlined).
Recombinant viruses were rescued by 8 plasmid transfection into 293T cells followed by amplification in MDCK cells as previously described [37]. In some cases, stocks were further amplified by growth in day 10–12 embryonated eggs, also as described [37]. Tissue culture cells were infected by allowing virus to adsorb for 30–60 min in serum free medium. For synchronous analyses of viral RNA and protein synthesis, infections were carried out at an MOI of 3–10. For analyses of virus growth, infections were initiated at low multiplicity and cells overlaid with serum free medium supplemented with 1 µg/ml trypsin (Worthington Biochemicals) and 0.14% bovine serum albumin. Serial passages were performed by infecting 3×106 MDCK cells at an MOI of 0.01. At 48 h p.i., the medium was clarified and 10 µl (of 5 ml) used to infect fresh MDCK cells. This procedure was repeated a further five times.
Plaque assays were carried out in MDCK cells using an Avicel overlay followed by staining with toluidine blue [37], [57]. Plaque areas were measured from scanned images using an oval selection marquee in the program Image J [58] and calibrated with respect to the area of a 6-well dish. HA assays were performed using 1% chicken red blood cells in 96 well plates according to standard procedures [37].
Infection of C57BL/6J or BALB/c mice (strains 664 and 1026, JAX Mice and Services) was carried out under animal BSL3 conditions at the National Institutes of Health. Groups of five 9–10 week old female mice were infected intranasally with 100 PFU of virus in 50 µl DMEM under oxygenated isoflurane anesthesia. Mice were individually identified and weighed daily; mice losing 25% or more of their initial body weight were euthanised. Three mice on days 2 and 7 and four mice on day 4 postinfection were euthanised and lungs collected for weight-normalized homogenization and MDCK plaque titration.
Total cellular RNA was extracted using Trizol (Sigma) and individual RNA species detected using radiolabelled reverse transcriptase primer extension followed by urea-PAGE and autoradiography as previously described [17], [59]. The primer GAAGGCCCTCCTTTCAGTCC, which targeted nucleotides 885–904 in mRNA sense, was used to detect segment 7 mRNA. Quantitation was performed using Fujifilm imaging plates and a Fujifilm FLA-5000 fluorescent image analyser. Data was analysed using AIDA software (Raytest). SDS-PAGE followed by western blotting was performed according to standard procedures. Blots were developed using infrared fluorescent secondary antibodies and imaged using a LiCor Biosciences Odyssey platform. Cells were stained for immunofluorescence after formaldehyde fixation using primary followed by Alexa-fluor conjugated secondary antibodies (Invitrogen) as previously described [60] and imaged on Zeiss LSM510, Leica SPE or TCS-NT confocal microscopes. Live cell imaging was performed in a temperature-controlled hood and CO2-independent medium as previously described [53].
|
10.1371/journal.pntd.0000214 | Use of Short Tandem Repeat Sequences to Study Mycobacterium leprae in Leprosy Patients in Malawi and India | Inadequate understanding of the transmission of Mycobacterium leprae makes it difficult to predict the impact of leprosy control interventions. Genotypic tests that allow tracking of individual bacterial strains would strengthen epidemiological studies and contribute to our understanding of the disease.
Genotyping assays based on variation in the copy number of short tandem repeat sequences were applied to biopsies collected in population-based epidemiological studies of leprosy in northern Malawi, and from members of multi-case households in Hyderabad, India. In the Malawi series, considerable genotypic variability was observed between patients, and also within patients, when isolates were collected at different times or from different tissues. Less within-patient variability was observed when isolates were collected from similar tissues at the same time. Less genotypic variability was noted amongst the closely related Indian patients than in the Malawi series.
Lineages of M. leprae undergo changes in their pattern of short tandem repeat sequences over time. Genetic divergence is particularly likely between bacilli inhabiting different (e.g., skin and nerve) tissues. Such variability makes short tandem repeat sequences unsuitable as a general tool for population-based strain typing of M. leprae, or for distinguishing relapse from reinfection. Careful use of these markers may provide insights into the development of disease within individuals and for tracking of short transmission chains.
| Molecular typing has provided an important tool for studies of many pathogens. Such methods could be particularly useful in studies of leprosy, given the many outstanding questions about the pathogenesis and epidemiology of this disease. The approach is particularly difficult with leprosy, however, because of the genetic homogeneity of M. leprae and our inability to culture it. This paper describes molecular epidemiological studies carried out on leprosy patients in Malawi and in India, using short tandem repeat sequences (STRS) as markers of M. leprae strains. It reveals evidence for continuous changes in these markers within individual patients over time, and for selection of different STRS-defined strains between different tissues (skin and nerve) in the same patient. Comparisons between patients collected under different circumstances reveal the uses and limitations of the approach—STRS analysis may in some circumstances provide a means to trace short transmission chains, but it does not provide a robust tool for distinguishing between relapse and reinfection. This encourages further work to identify genetic markers with different stability characteristics for incorporation into epidemiological studies of leprosy.
| Implementation of standardised multidrug regimens in the 1980s and 1990s has had a major impact on global leprosy prevalence, through shortening the duration of treatment. While it was reasoned that effective treatment of individual patients would reduce the spread of Mycobacterium leprae within communities, there is little evidence that the elimination programme has had a significant impact on disease incidence [1]. Continued controversies over global trends in the epidemiology of leprosy highlight gaps in our knowledge of the basic mechanisms of infection transmission and pathogenesis of this poorly understood disease [2]. While direct spread by aerosol or contact with infected individuals is thought to be the major route for dissemination of M. leprae, a role for zoonotic or environmental reservoirs cannot be excluded.
Publication of the genome sequence of an M. leprae isolate in 2001 revealed an organism which has undergone an extensive evolutionary process of gene decay [3]. Subsequent genetic studies have revealed that current human isolates of M. leprae from around the world show very little variation [4]. Identification of a limited number of informative single nucleotide polymorphisms (SNPs) allowed the elucidation of a high-level global phylogeny for M. leprae, but the extraordinary degree of sequence conservation has so far precluded application of SNP-based typing in the analysis of local epidemiology and mapping of transmission chains.
In contrast to the extreme conservation uncovered by SNP analysis, several researchers have reported a highly dynamic pattern of variation in copy number of short tandem repeat sequences in the M. leprae genome [5]–[8]. Using these loci, considerable variability was observed amongst panels of isolates within restricted geographical areas [5], [8]–[11]. In most, though not all, cases, the pattern of polymorphisms was conserved during passage in experimentally infected nude mice [12]. Inability to grow M. leprae in axenic culture has prevented quantitative measurement of the frequency in which changes in copy number are generated.
The aim of the present study was to assess whether differences in copy number of short tandem repeats provides information that is informative about the transmission of M. leprae. To evaluate this, we have applied molecular typing to leprosy patient biopsies collected in the course of epidemiological studies carried out in northern Malawi and in Hyderabad, India. This has allowed two different evaluations: firstly to assess the stability of the genotype within individuals; and secondly to assess transmission within leprosy multi-case households.
The Malawi (Karonga) series included 43 biopsies which had been collected for diagnostic purposes from 17 leprosy patients at the same or different times. The patients were identified from records of the Karonga Prevention Study [13]–[15]. The skin biopsies, taken by 6 mm punch, and partial thickness nerve biopsies, taken from enlarged sensory nerves, had been fixed in formol-Zenker and sent to the UK for histopathological analysis. Biopsies were read independently, and assigned classifications (TT = polar tuberculoid; BT = borderline tuberculoid; BB = Borderline; BL = borderline lepromatous; LL = Polar lepromatous; IND = Indeterminate) [16] by S Lucas. All patients gave informed consent for biopsy collection, and ethical permissions for this study were obtained from the Malawi Health Sciences Research Committee and from the Ethics Committee of the London School of Hygiene and Tropical Medicine.
The Indian (Hyderabad) series included 20 biopsies from 20 patients from eight families with more than one patient. All these patients were diagnosed at the outpatients clinic of the Blue Peter Research centre (BPRC), Hyderabad, India. All biopsies had been taken by 6 mm punch for diagnostic purposes, fixed in 10% buffered formal saline, embedded in paraffin and processed for histopathology [5]. Ethical approval was obtained from the local BPRC Ethics Committee and informed consent was obtained from all the subjects involved in the study.
Ten sections of 5 µm thickness were collected in a separate vial for each sample. A separate blade was used to cut each block in order to avoid cross contamination. The sections were deparaffinised prior to extraction. 100 µl of extraction buffer (90 µl of 0.5 M EDTA pH 8.0, 0.5% SDS+10 µl Proteinase K–QIAGEN) was added to each sample, mixed and incubated with constant agitation at 56°C overnight. Each sample was centrifuged at 223×g for 5 min and 100 µl of the supernatant then transferred into a tube containing 0.5 ml PB buffer (QIAGEN PCR purification kit) and thereafter DNA extraction was done according to the manufacturer's instructions [17].
A total of seven repeat loci were examined. These were identified by analysis of the M. leprae genome sequence (of an isolate derived from South India) [18] and from previous publications [5]–[7]. Table 1 provides a list of PCR primers used. All PCR products were analysed by gel purification and sequencing to determine repeat copy numbers as described previously. DNA purified from armadillo-grown M. leprae supplied by Dr P Brennan through the NIH Leprosy Contract (http://www.cvmbs.colostate.edu/mip/leprosy/index.html) was used as a positive control.
PCR was performed on a Hybaid Express thermal cycler in a final volume of 25 µl using the ‘Hot-Start’ Excite Core Kit (BioGene) according to the manufacturer's instructions. After an initial denaturation step (10 min at 95°C), 45 cycles of amplification were performed as follows: denaturation at 95°C for 15s, annealing at 58°C for 40s, and extension at 72°C for 30s. A final extension was performed at 72°C for 2 min. PCR products were initially screened by electrophoresis in 2% (w/v) agarose gels. For sequencing, products were separated on 2% (w/v) low melting point agarose (Invitrogen) and bands were excised with a sterile scalpel blade and purified using a GeneClean DNA isolation kit (Q-BIO gene). Cycle sequencing was performed on a PE 2700 system with ABI Big Dye 3.1 Terminator Ready Reaction Kit (Applied Biosystems) according to the manufacturer's protocol, with subsequent analysis on an ABI 3730 Genetic Analyzer.
Based on inspection of the genome sequence of M. leprae, we selected a panel of seven loci containing short tandem repeat elements ranging from 3 to 25 base pairs. To assess the extent of polymorphism of these loci among the Malawi patients, we amplified and sequenced the corresponding products from all 43 skin and nerve biopsies taken from 17 individual patients (Table 2). We were able to determine repeat copy numbers for almost all of the loci, even from biopsies which had been scored as polar tuberculoid and with no bacilli seen by microscopy. A total of 26 different copy number combinations were observed from the 43 samples.
Polymorphism was more extensive in the case of the shorter repeat elements. The two longer repeats, ML2469/70 (23 bp) and ML2418 (25 bp), were uniformly present in two copies in all of the Malawi samples. This differs from the copy numbers in the sequenced isolate (3 and 5 copies respectively) but is identical to a recent sample of armadillo derived M. leprae DNA provided by the NIH Reference Facility at Colorado State University. For subsequent analysis of the Malawi samples, we used only the five shorter repeat loci.
To assess the utility of genotypic analysis for mapping of transmission, we first assessed whether repeat copy number could vary between samples from a single individual, by analysis of Malawi biopsies taken at a single timepoint from different anatomical sites or tissues, and biopsies taken from the same individual at different times.
Copy numbers at three of the repeat loci were analysed for a panel of skin biopsies taken from the Indian patients from eight multicase families (Table 6). With three exceptions (see C, D and F), members of these families lived together in the same household residence. Identical genotypes were found in all the cases in six of the families. Two individuals from family “G” lived in same residence but had bacilli which differed at a single locus. In family “D”, three individuals shared an identical genotype, one differed at a single locus, and one differed at all three loci. Interestingly, the two individuals with bacilli of different genotypes were the oldest in the entire series, and the one which differed at all three loci was the most distantly related (mother of a daughter in law) and lived in a different residence.
Our findings in the Malawi patients are consistent with previous publications demonstrating extensive polymorphism in the copy number of short tandem repeat sequences of M. leprae [5]–[8]. Comparison with the relative paucity of SNP polymorphisms suggests that M. leprae may have acquired a specific lesion in the mechanisms required for maintenance of fidelity during replication of repeat sequences, possibly as one of the consequences of the overall pattern of gene decay in this organism [18]. Comparison of different repeat loci indicates that changes are most extensive in the case of very short repeat motifs comprising only 2 or 3 base pairs. This is consistent with the reported absence of polymorphism in longer repeat sequences resembling the mycobacterial interspersed repeat units (MIRU) that have proved useful in typing of M. tuberculosis [18].
While we cannot exclude the possibility that changes in repeat copy number have potentially selectable phenotypic consequences, inspection of the predicted changes does not indicate any obvious biological significance. Three of the repeats are located outside of coding regions. The 6bp ML1505 locus introduces a variable number of Pro-Ala repeats within a conserved hypothetical protein; the 12bp ML1182 locus encodes a Glu-Val-Val-Glu repeat in a member of the PPE protein family; the ML0058 21bp repeat, ML2469 23bp repeat and the ML2418 25bp repeat are located in pseudogenes.
Analysis of multiple Malawi biopsies collected at the same time reproduced our previous finding among Indian patients [5], with significantly greater genotypic differences between M. leprae samples collected from skin and nerve than between multiple skin samples from the same patient (7/7 skin-nerve versus 1/7 skin-skin patients; p<0.01). This could indicate that the patients with more than one M. leprae genotype had been infected by more than one isolate, and that the isolates were tissue specific, or that progression of the infection results in expansion of different bacterial populations in different anatomical sites. To explore further the possibility that the genotype may change during disease progression, we analysed serial samples from four patients from Malawi. In no patient were the serial isolates identical. While again we cannot totally exclude the possibility of multiple infections, the variation in copy number observed over time suggests that the dominant genotype can undergo changes within a single individual.
The potential occurrence of genotypic variation within individual patients points to a need for considerable caution in any application of this type of analysis to tracking of transmission between individuals. On the other hand, when we analysed repeat copy numbers at three loci for individuals with a high probability of sharing a transmission link as a result of living in the same household, in Hyderabad, India, we observed a strong concordance in bacterial genotype. This is consistent with our earlier observation in this same population [5].
The homogeneity in genotypes between individuals within households in the Hyderabad series contrasts with the differences observed between individuals in Malawi, and also with the differences observed within individuals over time and between tissues in the Malawi series. We offer three comments on these patterns. First, as only three loci were examined in the Hyderabad series, versus five in the Malawi patients, our ability to detect differences was lower for the Indian than for the Malawian series. This may have increased the apparent homogeneity of the Hyderabad household sets. Second, the Indian series included more individuals towards the lepromatous pole than did the Malawi series. Perhaps the relatively unrestrained growth of M. leprae in lepromatous patients allows selection of dominant clones, and/or these patients were infected with large numbers of genetically identical bacilli within their household environments. Third, the Malawi patients were appreciably older than the Indian patients. Though tuberculoid disease is thought to have a shorter incubation period than lepromatous disease, this age difference, and the fact that leprosy has declined rapidly in Malawi in recent years [13],[14], means that the Malawian patients are likely to have been infected for longer than the Indian patients, which could explain the genetic divergence of the M. leprae populations between individuals. The fact that the two oldest patients in the Hyderabad series had isolates which differed from those of their household contacts is consistent with this.
Taken together, our findings suggest that genotyping of M. leprae on the basis of short tandem repeat copy numbers may provide insights into disease progression within individual patients and, when used with care, may assist in analysis of short and recent transmission chains. Current evidence indicates that it does not provide a robust assay to distinguish recent transmission from relapse, or reinfection from reactivation, in the way that molecular tools have proved useful for study of the epidemiology of tuberculosis. It remains possible that further repeat loci will be identified as having an intermediate stability suitable for wider transmission tracking. |
10.1371/journal.pgen.1000512 | Copy Number Variation in Intron 1 of SOX5 Causes the Pea-comb Phenotype in Chickens | Pea-comb is a dominant mutation in chickens that drastically reduces the size of the comb and wattles. It is an adaptive trait in cold climates as it reduces heat loss and makes the chicken less susceptible to frost lesions. Here we report that Pea-comb is caused by a massive amplification of a duplicated sequence located near evolutionary conserved non-coding sequences in intron 1 of the gene encoding the SOX5 transcription factor. This must be the causative mutation since all other polymorphisms associated with the Pea-comb allele were excluded by genetic analysis. SOX5 controls cell fate and differentiation and is essential for skeletal development, chondrocyte differentiation, and extracellular matrix production. Immunostaining in early embryos demonstrated that Pea-comb is associated with ectopic expression of SOX5 in mesenchymal cells located just beneath the surface ectoderm where the comb and wattles will subsequently develop. The results imply that the duplication expansion interferes with the regulation of SOX5 expression during the differentiation of cells crucial for the development of comb and wattles. The study provides novel insight into the nature of mutations that contribute to phenotypic evolution and is the first description of a spontaneous and fully viable mutation in this developmentally important gene.
| The featherless comb and wattles are defining features of the chicken. Whilst the Pea-comb allele was known to show a dominant inheritance and drastically reduce the size of both comb and wattles, the genetics underlying the mutation remained elusive. Chicken comb is primarily composed of collagen and hyaluronan, which are produced by chondrocytes. These cells are formed through the condensation and differentiation of mesenchyme cells during the chondrogenesis pathway, the early stages of which are regulated by SOX transcription factors. Here we pinpoint a massive amplification of a duplicated sequence in the first intron of SOX5 as causing the Pea-comb phenotype. By studying early embryos, we show that SOX5 is ectopically expressed during a restricted stage of development in the cells which underlie the comb and wattles of Pea-comb animals. We hypothesise that the sequence duplication alters the regulation of SOX5 expression when the differentiation of cells essential for comb and wattle development is taking place. Pea-comb adds to the growing list of phenotypic variation which is explained by regulatory mutations and so demonstrates the evolutionary significance of such events.
| In 1902 Bateson [1] reported the first examples of Mendelian inheritance in animals based on the genetic studies of four traits in chicken, one of these being the Pea-comb phenotype (Figure 1). The Pea-comb allele results in reduced comb and wattle size compared to wild-type individuals. Pea-comb shows incomplete dominance and as such the small comb shape can differ slightly between homo- and heterozygous birds. Homozygotes present three longitudinal rows of papillae, whilst heterozygotes can have a well-developed central blade (still of reduced size compared to wild-type) [2]. The wild-type has a single central blade of tissue and is therefore often denoted single comb. Bateson and Punnet [3] reported the first example of an epistatic interaction between genes when they showed that walnut comb is caused by the combined effect of Pea-comb and Rose-comb. Subsequent studies revealed that Pea-comb, besides its effect on comb and wattles, was also associated with a ridge of thickened skin that runs the length of the keel over the breast bone [4]. The Pea-comb mutation may have occurred early during domestication as the phenotype is widespread among both European and Asian breeds of chickens. Furthermore, it has been speculated that a reproduction in the tomb of Rekhmara at Thebes, Egypt, dated to ∼3,450 years before present depicts a rooster with the characteristic Pea-comb phenotype [5].
Chickens were domesticated from the red junglefowl with some contributions from the grey junglefowl [6], two species adapted to subtropical or tropical environments. Chickens do not sweat, instead they dissipate up to 15 percent of their body heat through the comb and wattles [7], making the Pea-comb phenotype adaptive to cold environments since it reduces heat loss. This phenotype has also been favoured in chickens bred for cock-fighting, as noted by Darwin [8] the smaller ornaments provided smaller targets for injury.
In the present study we show that the classical Pea-comb phenotype in chickens is caused by a large expansion of a duplicated sequence in intron 1 of the gene for the SOX5 transcription factor.
Pea-comb has previously been assigned to chromosome 1 [9],[10]. We refined the localization by linkage analysis using a dense set of genetic markers and a large segregating family. The interval harbouring Pea-comb was defined as 67,831,796–68,456,921 bp on chromosome 1, based on flanking markers showing recombination with Pea-comb (Table 1). This interval contains a single gene, SOX5, a member of the SRY-related HMG box family of transcription factors. SOX5 is located in a one Mb gene desert that is enriched for Evolutionary Conserved Non-coding Sequences (ECNS; Figure 2A). This is a typical feature of developmentally important genes [11],[12]. SOX5 was not an obvious candidate gene for Pea-comb but the comb is composed of extracellular matrix and SOX5 has a well-established role in chondrocyte development and production of extracellular matrix [13]. Mouse SOX5 knockouts die at birth from respiratory distress caused by a cleft secondary palate and narrow thoracic cage [13]. Mouse SOX5/SOX6 double knockouts die in utero with severe skeletal dysplasia, demonstrating that these two genes have critical, redundant roles during development [13],[14].
To further refine the localization of Pea-comb we characterized SOX5 haplotype patterns among three breeds of chicken, a French experimental population, the Russian Orlov and the Chinese Hua-Tung. These breeds all carry Pea-comb and, to the best of our knowledge, there has been no exchange of genetic material between them for 100 generations or more. The Orlov and Hua-Tung are not fixed for Pea-comb, allowing recombination to reduce the size of the shared haplotype associated with the mutation. Initial IBD mapping using 12 samples from the three different populations revealed a completely shared haplotype between 67,961,701 bp and 68,061,854 bp (Table 2). SNP genotyping of all Hua-Tung and Orlov individuals available narrowed the shared haplotype further to a 50 kb region spanning positions 67,985,285 bp and 68,035,337 bp (Figure 2A; Table 2). The upstream break-point (67,985,285 bp) was identified using a single Hua-Tung bird. The break was confirmed in two additional individuals from the same population which were homozygous at the six SNPs diagnostic of the Pea-comb haplotype, but heterozygous at this break-point. Downstream, the haplotype was broken at 68,035,337 bp in three Orlov birds (Table 2).
This critical region is located upstream of the first annotated exon however a comparison with SOX5 from mammalian species indicated that exon 1 is missing from the chicken genome assembly and is expected to be found more than 200 kb upstream of exon 2 (Figure 2A). We confirmed the existence of an upstream exon in chicken by 5′ RACE analysis. The obtained nucleotide sequence (GenBank accession number FJ548639) showed 90% identity to human SOX5 exon 1, but did not give a match in the chicken genome, implying a gap in the current chicken assembly.
Resequencing the 50 kb region associated with Pea-comb from a set of Pea-comb and wild-type birds revealed a limited number of sequence polymorphisms, with fixed differences between genotypes. These potentially causative SNPs were interrogated using a larger set of wild-type birds from the AvianDiv panel [15], however none of the alleles were found to be unique to the Pea-comb haplotype (Table 2). The failure to identify a causative point mutation led to a screen of the Pea-comb region for structural changes using Southern blot analysis. The SOX-85kb_SB probe (Table S1) revealed a dramatic increase in the hybridization signal of a 3.2 kb BamHI fragment in Pea-comb birds (Figure 2C) whilst other probes from the region gave identical restriction fragment patterns for both alleles. The result implied that Pea-comb is associated with a large tandem array of a duplicated sequence containing a BamHI restriction site. PCR and sequence analysis revealed that this DNA fragment is also duplicated on wild-type chromosomes which have two copies (Figure 2B), whereas the Pea-comb allele has a large number of copies.
Quantification of the copy number of the duplicated fragment using both pulsed field gel electrophoresis (PFGE) and real-time PCR analysis confirmed that a massive amplification of a duplicated sequence is associated with the Pea-comb allele. PFGE analysis using the restriction enzyme PshA1, which cuts outside the duplicated region, gave a 97 kb restriction fragment in Pea-comb birds in contrast to a predicted 10 kb fragment based on the reference genome sequence from a wild-type bird (Figure S1). The result indicates that the Pea-comb allele contains about 30 copies of the duplicated sequence. Real-time PCR analysis of Pea-comb birds from three breeds confirmed this finding and revealed a 20- to 40-fold sequence amplification (Figure 2D). The real-time PCR analysis did not indicate two clear groupings corresponding to Pea-comb heterozygotes and homozygotes suggesting that the duplication may show further copy number variation among Pea-comb individuals. Interestingly, 100 years ago Bateson and Punnett [16] reported variable expression of the Pea-comb phenotype which may reflect a copy number variation of the duplicated sequence. Although the duplicated sequence is not evolutionary conserved, it is located close to two highly conserved ECNSs (Figure 2A). The distance between these elements is about 10 kb on wild-type chromosomes in contrast to about 100 kb on Pea-comb chromosomes. The duplication includes a sequence repeated in two copies on wild-type chromosomes and each copy contains two partial LINE fragments (Figure 2B). The expansion of this duplication must be the causative mutation because it was the only polymorphism showing complete association with the phenotype.
A closer examination of the duplicated sequence shows that it is particularly GC-rich and contains a small CpG island (Figure 2A and 2B). The wild-type chromosome contains two copies of this CpG island whereas the Pea-comb chromosome contains about 30. This could be relevant for the mechanism of action of this intronic mutation.
The Pea-comb phenotype is apparent at hatch and must therefore reflect altered gene expression during development. Tissue samples from the comb region were collected from both homozygous Pea-comb and homozygous wild-type birds at embryonic (E) days 6, 7, 8, 9, 12 and 19 for expression analysis. Quantitative RT-PCR analysis only revealed significant differences in SOX5 expression at stage E7 and E8 (which were combined due to the low number of E8 samples). The results for E7+8 revealed significant upregulated SOX5 expression in the comb region in Pea-comb birds (t = −5.0, p = 0.002; Figure S2A). Expression analysis was also conducted using primers specific for each exon of SOX5 (including the previously un-annotated exon 1 described above), however the results did not indicate any difference between genotypes in regards to differential splicing of SOX5 (Figure S2B).
Immunohistochemical staining with a human SOX5 antibody as well as in situ-hybridization with a chicken-specific cRNA probe was carried out to investigate SOX5 expression in both Pea-comb and wild-type embryos during development (Figure 3). Specific immunostaining of nuclei was seen in developing cartilaginous structures including the nasal septum, Meckel's cartilage and optic sclera (Figure 3A and 3D). Scattered and rare SOX5 positive cells were seen in the surface ectoderm (Figure 3B and 3M). All structures with SOX5 staining in wild-type embryos were also positive in Pea-comb embryos including the scattered cells in the ectoderm. However in Pea-comb embryos, striking ectopic SOX5 expression was observed in mesenchymal cells located just beneath the surface ectoderm where the comb and wattles will develop (Figure 3A–3J). Differential expression was confirmed with in situ-hybridization (Figure 3G–3H) and quantitative real-time PCR (see above). The ectopic expression is transient. Whereas few cells with ectopic expression are visible in the comb region by day E6, they are prominent at E9, and almost completely absent at E12 (Figure 3K–3P). Thus, Pea-comb appears to be a spatiotemporal-specific, cis-acting regulatory SOX5 mutation.
A major challenge in current genome biology is to reveal the biological significance of the many Evolutionary Conserved Non-coding Sequences (ECNS). The analysis of the functional significance of ECNS is hindered by a paucity of mutations in such regions which show an association with a phenotype. Here we demonstrate the first spontaneous SOX5 mutation associated with a phenotype, despite the rich abundance of ECNS in the SOX5 region (Figure 2A). SOX5 is under complex regulation and as demonstrated here, mutations affecting its regulation can have very specific effects. It would be surprising if regulatory mutations in this gene do not to some extent contribute to phenotypic diversity present in humans. For instance, the human face shows a bewildering array of diversity. The nearly identical facial appearances of monozygotic twins imply that this diversity is nearly 100% genetically determined, but knowledge concerning the underlying molecular basis of this diversity is restricted to certain craniofacial abnormalities [17]. It is likely that regulatory mutations in developmentally important genes shape this type phenotypic diversity, and SOX5 may very well be one of the genes that contributes.
The comb is a sexual ornament that shows strong sexual dimorphism in chickens and the fact that this sexual dimorphism is maintained in Pea-comb birds shows that the Pea-comb tissue maintains the response to the influence of sex hormones (Figure 1). That the comb is under sexual selection is evidenced by red junglefowl females showing mating preferences for males with large combs and reciprocally, males tend to favour females with larger combs [18],[19]. The size of the comb is proportionally larger in many breeds of domestic chickens compared to their wild ancestors. In our previous study of a large intercross between White Leghorn chicken (with larger combs) and red junglefowl, we identified a number of Quantitative Trait Loci (QTL) affecting the size of the comb [20]. Interestingly, one of the QTL controlling the size of the female comb overlaps the SOX5 locus, which now becomes an obvious candidate gene for this QTL. However, the confidence interval for the QTL is large, as is usually the case in an F2 intercross, and the entire SOX5 region needs to be considered in a search for possible causative mutation(s).
SOX genes are defined by their high-mobility-group (HMG) domains and are divided into eight groups (A to H) based on protein sequence comparison [14]. SOX5 belongs to the D family of SOX genes, along with SOX6 and SOX13. SOX5 has been termed an architectural transcription factor [21], as binding to this protein will cause a sharp bend (80–135 degrees) in the bound DNA and may lead to different regulatory regions of a target gene coming into closer proximity. SOX5 has been reported to have a co-operative role in chondrogenesis; during embryonic cartilage formation SOX5 and SOX6 assist SOX9 to activate specific genes [22], and have a repressive role in oligodendrogenesis during neural development [23]. SOX5 is also expressed in the developing neocortex and cranial neural crest during the early stages of development. SOX5 postmitotically regulates migration, axon projection and postmigratory differentiation of certain neocortical neurons [24] but little is known about SOX5 function in neural crest derivatives [25]. With these different roles, the functional consequence of the transient ectopic SOX5 expression in Pea-comb birds is not clear.
The comb is composed of layers of epidermis, dermis and central connective tissue, of which collagen and hyaluronan are the major components [26]. The ectopic SOX5 expression is first seen in E7 (st28) mesenchyme (Figure 3). Previous studies with grafts of comb-primordia from different ages at various locations imply that cells giving rise to the comb are already determined by E4 (st24) [27],[28] and that the determination resides in the mesenchymal components and not in the ectoderm [27]. These experiments also revealed that the morphology of the comb was under control of the mesenchyme [27],[29]. Heterotopic grafts of single-comb primordia to the neck region without beak mesenchyme, lost the serrated single ridge morphology and expanded laterally following the development, resembling that of complex comb types [29] such as the Pea-comb. Hence, changes in the underlying mesenchyme at the time of the ectopic SOX5 expression will not affect the determination and initial stages of the comb development but rather the development of comb shape. Our results indicate that ectopic SOX5 expression changes the modulating properties of the mesenchyme of the nasofacial region beneath the regions of the developing comb and wattles. The serration of a single comb is associated with loosely coherent clusters or points of proliferating mesenchymal cells [30],[31]. Such clusters were not observed in the developing Pea-comb mesenchyme and this difference may be due to the ectopic SOX5 expression.
Pea-comb is an additional example of a Copy Number Variation (CNV) associated with a phenotype. About 12% of the human genome contains tandem duplications that may show CNV [32] and a number of human diseases have been reported to be associated with CNVs [33],[34]. It is important to distinguish CNVs that are due to duplications of single copy sequences (de novo duplications) and expansions or contractions of already duplicated sequences. We have previously reported three de novo duplications associated with phenotypic traits in domestic animals, Dominant white colour in pigs [35], the Ridge phenotype in Ridgeback dogs [36] and Greying with age in horses [37]. In contrast, Pea-comb and most human diseases associated with CNVs involve expansions or contractions of existing duplications. Pea-comb is however an unusual CNV associated with a phenotype because it involves the amplification of a non-coding region located far from any coding sequence. Pea-comb therefore to some extent resembles the massive amplification of a trinucleotide repeat in intron 1 of Frataxin causing Friedrich ataxia [38]. However, the mechanism of action is probably very different since the expansion of the trinucleotide repeat in Frataxin leads to the formation DNA triplexes and “sticky DNA” causing transcriptional silencing [38].
The duplicated sequence in intron 1 of SOX5 is not evolutionary conserved between birds and mammals. This does not exclude the possibility that it contains regulatory elements which are important for SOX5 in birds, or in birds that develop combs and wattles. However, even if the duplicated sequence per se is not functionally important, the massive amplification of this sequence may disturb the action of regulatory elements in the region. For instance, tandem repeats may recruit DNA methylation which abolishes protein-DNA interaction at regulatory elements [39]. Our observation that the duplicated region is not only particularly GC-rich, but contains a small CpG island which becomes repeated about 30 times on the Pea-comb chromosome, suggests that DNA methylation maybe a plausible mechanism for Pea-comb as this effect may spread to neighbouring regulatory sites.
Genetic studies of phenotypic diversity in domestic animals provide a strong case for the evolutionary significance of regulatory mutations. Other examples of cis-acting regulatory mutations underlying phenotypic traits in domestic animals include (i) a nucleotide substitution in intron 3 of IGF2 with a prominent effect on muscle growth in the pig [40], (ii) regulatory mutations in the gene for microphtalmia-transcription factor (MITF) causing white spotting in dogs [41], (iii) regulatory mutation(s) in BCDO2 causing the yellow skin phenotype in chicken [6], (iv) a 4.6 kb duplication in intron 6 of STX17 causing Greying with age in horses [37], (v) an 11.7 kb intergenic deletion causing intersexuality and lack of horns in goats [42] and (vi) a mutation creating an illegitimate microRNA target site in the sheep myostatin gene promoting muscle growth [43]. Furthermore, the ridge phenotype in dogs [36] and the dominant white colour in pigs [35] are caused by large duplications that most likely lead to dysregulated expression of some fibroblast growth factor genes and the KIT receptor, respectively. Most of these examples concern growth factors, growth factor receptors, or transcription factors that have important roles during development and for which null mutations are lethal or sub-lethal. The significance of regulatory mutations is also supported by the identification of mutations underlying morphological variation in Drosophila [44],[45] and stickleback fish [46]. This wealth of data now demonstrates the prominent role of regulatory mutations, at least for morphological evolution, as predicted by King and Wilson more than 30 years ago based on the limited divergence in protein sequences between human and chimpanzee [47].
DNA samples from a French pedigree consisting of 7 parental, 14 F1 and 244 F2 progeny were used for linkage analysis. The parentals consisted of four heterozygous Pea-comb birds and three homozygous wild-type birds. DNA samples from Pea-comb birds for identical-by-descent mapping came from a French experimental population kept by INRA, from a Chinese Hua-Tung population and from the Russian Orlov breed. DNA samples from various domestic breeds collected by the AvianDiv project [15] were used for real-time PCR analysis and to test whether candidate causal mutations from the Pea-comb region could be excluded since they were present among birds homozygous for the wild-type allele at the Pea-comb locus.
Linkage analysis was conducted using the SNPs compiled in Table S1. SNP genotyping was performed with Pyrosequencing (See ‘Linkage primers’, Table S1 for details). Fine-mapping was carried out on a small number of recombinant individuals that more exactly defined the Pea-comb region. In this case, one kb fragments were amplified and sequenced to detect SNPs (see ‘1 kb fragment analysis’, Table S1 for primers).
IBD mapping was initially performed on a panel of 12 chickens; two Pea-comb and two wild-type birds from the linkage pedigree, four homozygous Pea-comb birds from the French pedigree, two Pea-comb birds from the Chinese Hua-Tung population and two Pea-comb birds from the Russian Orlov population. A collection of one kb regions spanning approximately 67,891,800 bp to 68,181,677 bp on chromosome 1 were sequenced for each animal to identify SNPs between lines (See ‘SNPs used for IBD Mapping’, Table S1, for exact positions). In a similar way, the heterozygosity of chromosome 1, fragment 68,181,600 bp to 68,335,500 bp, was determined by sequencing 16 homozygous Pea-comb birds belonging to the linkage pedigree (Primers SOX+130, SOX+140, SOX+200, SOX+260 in Table S1). This re-sequencing effort revealed potential causative SNP that were differentially segregating between the Pea-comb and non-Pea-comb populations. These polymorphisms were subsequently tested in the non-Pea-comb individuals from the AvianDiv panel and used to define the Pea-comb region by six loci, positions 68,038,060 bp, 68,035,337 bp, 68,019,518 bp, 68,011,661 bp, 67,991,941 bp and 67,985,285 bp respectively. Pyrosequencing was used to assay these six variations in 34 Hua-Tung Pea-comb birds and 27 Orlov Pea-comb birds (See ‘Pyro SNPs used for IBD mapping’, Table S1). Lastly, four of these loci were also genotyped for a variety of birds from the AvianDiv panel to check the frequency of the Pea-comb haplotype among wild-type chromosomes.
The copy number of the SOX5 duplication was evaluated by comparing eight populations with wild-type phenotype (red junglefowl, n = 5; commercial broiler, n = 5; Czech Golden Pencilled, n = 5; Friesian Fowl, n = 5; Finnish Landrace, n = 5; Red Villafranquina, n = 5; Transylvanian Naked Neck, n = 5; White Leghorn, n = 5) to three breeds segregating for Pea-comb (French Pea-comb, n = 3; Hua-Tung, n = 13; Orlov, n = 13). The real-time PCR assay contained TaqMan Gene Expression Master Mix (Applied Biosystems), 900 nM of each primer combined with 250 nM of fluorometric probe and 30 ng of genomic DNA. The SOX5 assay was normalised using an assay designed to ribosomal protein S24 (rps24). Primer and probe concentrations of those reactions were 750 nM and 300 nM, respectively. Each assay was performed in triplicate, averaged and referenced to a wild-type red junglefowl. Details of primer and probe sequences are in Table S2. Fold change was calculated using the equation 2−(Normalized Ct peacomb assay−Normalized Ct rps24 assay) and the range of this value was determined from the combined standard errors of both assays.
Seventy kb on chromosome 1 from 67,969,741 bp to 68,041,242 bp were re-sequenced using a panel of ten birds; two wild-type parental birds from the linkage pedigree, two red junglefowl (RJF) birds, two homozygous Pea-comb from the French pedigree, two Pea-comb Hua-Tung birds and two Pea-comb Russian Orlov birds. Primers pairs were used to generate over-lapping PCR amplicons ranging from approximately 1200 bp to 1400 bp in size. Internal primers were used with each primer pair set. Primers were designed using Primer3 [48]. DNA sequences were analysed and edited in Codoncode Aligner (CodonCode, Dedham, MA). The RJF genomic sequence used to generate the chicken genome sequence was used as a reference for alignment.
The chicken genome reference sequence contained three gaps. Gap 1 spanned 67,981,199 bp–67,983,790 bp; gap 2, 68,002,231 bp–68,003,557 bp and gap 3, 68,006,200 bp–68,006,994 bp. Gaps 1 and 3 were closed using a PCR-based 2-step strategy [49] (Primers Dynal-75_gap and Dynal-105_gap primers in Table S1), whilst gap 2 was covered using long range PCR (Primers LR_gap1, Table S1). Gap 2 was found to be a tandem duplication, part of the duplication linked to the Pea-comb mutation. Therefore sequencing was performed after the amplicon was cleaved with XhoI, and both halves sequenced independently.
Southern blot analysis was performed using a set of six different probes (SOX-55kb_SB to SOX-105kb_SB, Table S1) on a panel consisting of three homozygous Pea-comb birds from the linkage pedigree, three red junglefowls, two commercial broiler samples and two White Leghorn birds. The DNA was digested with BamHI and separated by 0.7% agarose gel electrophoresis.
DNA plugs were prepared from nine chickens, three of each wild-type, Pea-comb heterozygous and Pea-comb homozygous birds. The plug preparation and restriction digest protocol follows that of Giuffra et al. [35], with the following modifications. Whole blood stored in 0.5 M EDTA was used as starting material and resuspended to a concentration of 25×108 cells/ml in PBS after washing. Plugs were solidified at room temperature prior to digestion for 24 hours at 50°C in 0.5 mg/ml proteinase K, 1×NDS (0.5 M EDTA, 0.01 M Tris, 0.34 M N-Laurylsarcosine, pH 8.0) with constant shaking. Enzyme digestions were performed as described [35]. PshA1 (New England BioLabs) was selected for this experiment as this restriction enzyme was predicted to cut at position 67,998,520 bp and 68,005,614 bp, i.e. outside the duplicated region.
PFGE of the PshA1 digested plugs was performed in a 1.0% agarose gel, 0.5% TBE at 14°C, 6 V/cm, switch times ramped from 1–25 seconds for 17 hours and fragment sizes were estimated using the MidRange I PFG Marker (New England BioLabs). Southern blot analysis was performed as before, using the 986 bp product from the SOX-85kb_SB amplicon (Table S1) as probe.
The duplicated region was amplified with long-range PCR primers (SOX-Duplication_LR1_F and R, Table S1). In addition, internal primers were used to check the length of the potential duplication through nested PCR of the initial amplicon (Primers SOX-Duplication_F, R11, 12 and 13, Table S1).
Heads from staged embryos were fixed in 4% paraformaldehyde in phosphate buffered saline (PBS) for one hour at 4°C. Fixed heads were incubated overnight in 30% sucrose in PBS at 4°C, embedded in OCT freezing medium (Tissue-Tek, Sakura), frozen and sectioned in a cryostat. Cross sections and sagittal sections, 10 μm thick, were collected on glass slides (Super Frost Plus, Menzel-Gläser). The sections were rehydrated in PBS for 15 min and then blocked in PBS containing 1% fetal calf serum, 0.1% Triton-X and 0.02% Thimerosal. The SOX5 antibody (Abcam, a_6226041) was diluted 1∶500 in blocking solution and incubated on the slides over night at 4°C. The secondary antibodies (Jackson Immunoresearch Laboratories) were incubated at room temperature for two hours at a 1∶200 dilution in blocking solution. Samples were analysed using a Zeiss Axioplan2 microscope equipped with Axiovision software. Images were formatted, resized, enhanced and arranged for publication using Axiovision and Adobe Photoshop.
A cRNA probe was made using a DIG RNA labeling kit (Roche). The SOX5 probe was made from the chEST752i6 cDNA clone acquired from the BBSRC ChickEST Database [50]. The probe was hybridized to untreated sections over night at 66°C under conditions containing 50% formamide and 5×SSC in a humidified chamber. The DIG labeled nucleotides were detected using an alkaline-phosphatase coupled anti-DIG antibody (Roche) followed by incubation with BCIP/NBT developing solution (Roche) for 1–5 hours at 37°C. Images were captured using a Zeiss Axioplan2 microscope equipped with Axiovision software (3.0.6.1, Carl Zeiss Vision GmbH).
Tissue was collected from homozygous Pea-comb birds and homozygous wild-type birds. The ages of the birds sampled were embryonic (E) days 6, 7, 8, 9, 12 and 19 (with hatching occurring at approximately day 21). Two Pea-comb and two normal individuals were collected from each stage, with the exceptions of E7, where nine samples (four Pea-comb and five wild-type) were used and two E8 samples (one of each type). Tissues were initially stored in RNALater (Ambion), with total RNA extracted from embryonic tissues using the Trizol reagent (Invitrogen). The most central part of the presumptive beak and comb were dissected out. cDNA was made from 1 μg of RNA using GeneAmp (Applied Biosystems). Samples were run in triplicate using IQ SyBr Green Supermix (Biorad) and normalized to β-actin and TATA-box binding protein (TBP); primers are given in Table S1. SOX5 was amplified using primers SOX5_cDNA_1 crossing intron/exon boundaries. Control cDNA reactions containing primers but no RNA were performed in parallel. Samples were run on two separate machines: the ABI 7900HT and the Corbett Rotor-Gene 6000. In addition to these primers, primers for each individual exon (2 to 15) were also used to analyse potential alternate SOX5 splicing in tissue from the comb. These were used on cDNA from two E7 samples (Pea-comb and wild-type) and two E9 samples (Pea-comb and wild-type). Statistical analysis was performed by first correcting Ct values for batch effects caused by using two different machines, then conducting a two-sample t-test on the average of each set of triplicates.
Information on the chicken genome sequence is available at http://www.genome.ucsc.edu.
The sequence data presented in this paper have been submitted to GenBank with the following accession numbers FJ548629-FJ548639
|
10.1371/journal.pntd.0001030 | HTLV-1 Tax Specific CD8+ T Cells Express Low Levels of Tim-3 in
HTLV-1 Infection: Implications for Progression to Neurological
Complications | The T cell immunoglobulin mucin 3 (Tim-3) receptor is highly expressed on
HIV-1-specific T cells, rendering them partially “exhausted” and
unable to contribute to the effective immune mediated control of viral
replication. To elucidate novel mechanisms contributing to the HTLV-1
neurological complex and its classic neurological presentation called HAM/TSP
(HTLV-1 associated myelopathy/tropical spastic paraparesis), we investigated the
expression of the Tim-3 receptor on CD8+ T cells from a cohort
of HTLV-1 seropositive asymptomatic and symptomatic patients. Patients diagnosed
with HAM/TSP down-regulated Tim-3 expression on both CD8+ and
CD4+ T cells compared to asymptomatic patients and HTLV-1
seronegative controls. HTLV-1 Tax-specific, HLA-A*02 restricted
CD8+ T cells among HAM/TSP individuals expressed markedly
lower levels of Tim-3. We observed Tax expressing cells in both
Tim-3+ and Tim-3− fractions. Taken
together, these data indicate that there is a systematic downregulation of Tim-3
levels on T cells in HTLV-1 infection, sustaining a profoundly highly active
population of potentially pathogenic T cells that may allow for the development
of HTLV-1 complications.
| The retrovirus, Human T lymphotropic virus type 1 (HTLV-1) infects 10–20
million people worldwide. The majority of infected individuals are asymptomatic;
however, approximately 3% develop the debilitating neurological disease,
HTLV-1-associated myelopathy/tropical spastic paraparesis (HAM/TSP). There is
also currently no cure, vaccine or effective therapy for HTLV-1 infection. The
precise role of CD8+ killer T cells in the control or contribution of
HTLV-1 disease progression remains unclear. The T-cell immunoglobulin mucin
domain-containing (Tim) proteins are type 1 transmembrane proteins. Three human
Tim proteins (Tim-1, -3, and -4) exist and display markedly diverse expression
patterns and functions. Tim-3 is upregulated on CD8+ T cells
during chronic viral infections leading to a population of poorly functioning T
cells. We investigated the expression of Tim-3 on T cells from patients with
asymptomatic and symptomatic HTLV-1 infection and compared this with HTLV-1
uninfected donors. Patients diagnosed with HAM/TSP down-regulated Tim-3
expression on T cells when compared to asymptomatic patients and uninfected
controls. Our study provides evidence of a novel mechanism for the persistent
inflammation observed in HTLV-1 infected patients with neurological deficits and
significantly advances our understanding of how the Tim-3 pathway functions.
| The vast majority of HTLV-1-infected individuals with low and stable HTLV-1 proviral
load levels are clinically asymptomatic for life [1]. However, 1–3% of
subjects develop progressive neurological complications related to HTLV-1 infection,
classically denominated as HTLV-1 associated myelopathy/tropical spastic paraparesis
(HAM/TSP) [2],
[3], [4]. The infection
can also lead to a debilitating malignancy, known as HTLV-1 associated adult T cell
leukemia (ATL) in approximately 2–5% of infected individuals [4], [5], [6], [7].
The immune response, and in particular the cellular immune response, plays an
important role in the control of HTLV-1 infection [8], [9], [10], [11], [12]. In vitro
studies further demonstrate that CD8+ T cell responses are able to
directly lyse HTLV-1-infected CD4+ T cells [9], [11], [13]. In patients with HAM/TSP,
CD8+ T cells are capable of producing multi-cytokine responses
and are able to release cytotoxic molecules [14], [15]. Recent studies have selected
out patients with HLA-A*02 and HLA-Cw08 genes as being associated with lower
HTLV-1 proviral load and a reduced risk of progression to HAM/TSP [16], [17].
While these data support an important protective role for the CD8+ T
cell immune response with the potential for viral control, other studies suggest
that HTLV-1-specific CD8+ T cells may paradoxically contribute to
the neuromuscular immunopathology through autoimmune mechanisms, leading to the
clinical manifestation of HAM/TSP [18]. Furthermore, patients with HAM/TSP also present with
high numbers of HTLV-1 Tax-specific CD8+ T cells in the cerebrospinal fluid
[15], [19], [20], [21], [22] that are
thought to play a immunopathogenic role, either by release of neurotoxic cytokines,
such as TNF-α and IFN-γ [23], [24], or by direct cytotoxicity. It is evident from these
studies that the precise role of CD8+ T cells in the control or
pathogenesis of HTLV-1 disease progression remain unclear. Further knowledge of the
mechanisms leading to T cell induced immunopathology in HTLV-1 infection will be
important in determining successful immune-based therapies and provide insights for
effective vaccine designs.
During chronic viral infections, virus-specific CD8+ T cells undergo
an altered pattern of differentiation and can become exhausted [25], [26]. CD8+ T cell
exhaustion is a transcriptionally altered state of T cell differentiation distinct
from functional effector or memory CD8+ T cells [27].
CD8+ T cell exhaustion leads to profound T cell dysfunction and
the inability of the T cells to control retroviral replication [28], [29], [30]. Conversely, downregulation
of exhaustion markers could lead to a highly functional population of T cells. T
cell immunoglobulin and mucin domain-containing protein 3 (Tim-3), is upregulated on
CD8+ T cells during chronic viral infections [29], [30], [31], [32], [33], [34], [35], [36].
Programmed death receptor-1 (PD-1) is also known as another immune exhaustion
biomarker expressed in chronic viral infections [28], [37], [38], [39], [40], [41], [42], [43]. High levels of PD-1 and Tim-3 on
virus-specific T cells have been shown to lead to poor proliferative capacity and,
in some cases, ineffective Th1 cytokine production [29], [39], [44]. A sustained downregulation
of these receptors would lead to an exacerbated constitutively active T cell
population. The phenotypic profile of immune exhaustion markers on T cells is
unknown in seropositive HTLV-1 individuals. In this study, we show for the first
time that HTLV-1 associated complications may be related to the highly responsive
inflammatory Tax-specific T cells in HTLV-1-infected individuals. These results
support the idea that HTLV-1 infection induces mechanisms resulting in a limited T
cell exhaustion profile, leading potentially to neuro-immunopathology and disease
complications.
The research involving human participants reported in this study was approved by
the institutional review board of the University of Sao Paulo (IRB #0855/08) Sao
Paulo, Brazil. Informed consent was obtained for all subjects. All clinical
investigation were conducted according to the principles expressed in the
Declaration of Helsinki (http://www.wma.net/en/30publications/10policies/b3/index.html).
Patients were serially recruited in the HTLV-1 Outpatient Clinic at the
University of Sao Paulo, Brazil in two stages with written informed consent
approved by the University of Sao Paulo's Institutional Review Board
(#0855/08). The diagnosis of HAM/TSP based on criteria outlined by the WHO [45] (Table 1). The majority of
the patients were female (63%) with a median age of 48 (IQR: 22–66)
years. We enrolled age and sex matched healthy uninfected volunteers without
clinical and laboratory evidence of HTLV-1-associated disease, from the same
demographics as the infected subjects. All HTLV -1 seropositive subjects tested
negative for Hepatitis B, Hepatitis C, and HIV infections. No other inflammatory
diseases or disorders were present in any of the participants. Blood samples
were processed with Ficoll-Paque PLUS (Amersham Pharmacia Biotech, Uppsala,
Sweden) gradient centrifugation, and peripheral-blood mononuclear cells (PBMC)
were isolated and cyropreserved in fetal bovine serum (FBS) containing
10% DMSO in liquid nitrogen.
Conjugated Pentamers were obtained commercially from Proimmune (Oxford, UK). The
HLA-A*02 restricted HTLV-1 Tax (LLFGYPVYV) and CMV (NLVPMVATV) peptides were
obtained from New England Peptide (Gardner, MA). In some experiments rIL-2
[80 IU/ml] (Roche Diagnostics, Mannheim, Germany) and rIL-15 [50
ng/ml] (R&D Systems, Minneapolis, MN) were used during in
vitro culture studies.
Cryopreserved PBMC were rapidly thawed in warm RPMI 1640 with 10% FBS,
washed in FACS buffer (PBS, with 0.5% bovine serum albumin, 2 mM EDTA
(Sigma-Aldrich, St. Louis, MO)). For staining, 5×105 cells were
incubated with conjugated antibodies against Tim-3 (R&D Systems,
Minneapolis, MN), PD-1 (Biolegend, San Diego, CA), CD4, CD8, CD3 (all from BD
Biosciences, San Jose, CA) for 30 min on ice. In some experiments, PMBC were
then fixed and permeabilized prior to staining with conjugated anti-Tax (clone
Lt-4) antibodies [46] or a control labeled IgG. Fluorescence minus one
(FMO) samples were prepared for each fluorochrome to facilitate gating as well
as conjugated isotype control antibodies. Anti-mouse IgG-coated beads were
stained with each fluorochrome separately and used for software-based
compensation. Analysis was performed using a FACSCanto instrument (BD
Biosciences) and at least 100,000 events were collected and analyzed with FlowJo
software (TreeStar, Ashland, OR).
To define pentamer positive cells: staining was initially performed immediately
after thawing with biotin-labeled HLA-A2 Tax or CMV epitope specific pentamer
fluorotags followed a secondary staining step with fluorophore conjugated
antibodies against CD8 (BD), Tim-3 (R&D Systems), PD-1 (Biolegend) and CD3
(BD), and with labeled streptavidin. Cells were washed twice with PBS containing
1% FBS, then fixed in 2% paraformaldehyde and run on a customized
BD FACSCanto within 12 hours.
HTLV-1 proviral DNA was extracted from PBMC using a commercial kit (Qiagen GmbH,
Hilden Germany) and according to the manufacturer's instructions. The
extracted DNA was used as a template to amplify a fragment of 158 bp from the
viral tax region using previously published primers[47]. The SYBR green
real-time PCR assay was carried out in 25 µl PCR mixture containing
10× Tris (pH 8.3; Invitrogen, Brazil), 1.5 mM MgCl2, 0.2
µM of each primer, 0.2 mM of each dNTPs, SYBR Green (18.75
Units/r×n; Cambrex Bio Science, Rockland, ME) and 1 unit of platinum Taq
polymerase (Invitrogen, Brazil). The amplification was performed in the Bio-Rad
iCycler iQ system using an initial denaturation step at 95°C for 2 minutes,
followed by 50 cycles of 95°C for 30 seconds, 57°C for 30 seconds and
72°C for 30 seconds. The human housekeeping β globin gene primers GH20
and PC04[48]
were used as an internal control calibrator. For each run, standard curves for
the value of HTLV-1 tax were generated from MT-2 cells of log10
dilutions (from 105 to 100 copies). The threshold cycle
for each clinical sample was calculated by defining the point at which the
fluorescence exceeded a threshold limit. Each sample was assayed in duplicate
and the mean of the two values was considered as the copy number of the sample.
The amount of HTLV-1 proviral load was calculated as follows: copy number of
HTLV-1 (tax) per 1,000 cells = (copy number of HTLV-1
tax)/(copy number of β globin/2) ×1,000 cells. The method could detect
1 copy per 103 PBMC.
MAIPS4510 Elispot plates (Millipore, Danvers, MA) were coated with anti-IFN-γ
(10 µg/ml) (Mabtech, Nacka Strand, Sweden) in PBS, 50 µl/well,
either overnight at 4°C or for one hour at room temperature. After three
washes with PBS, PBMC (1×105 cells/well) and the appropriate
antigens were added (Tax peptide and CMV peptide), with a final volume of 200
µl/well. Plates were incubated at 37°C in 5% CO2 for
16–20 hours. After washing with phosphate-buffered saline (PBS) plus
0.1% Tween 20 (PBST), biotinylated anti-IFN-γ 1 µg/ml)
(Mabtech), antibodies were added to the appropriate wells in PBS 0.1%
tween 1% BSA (PBSTB) for 30 minutes at room temperature. Plates were
washed again three times with PBST, and alkaline phosphatase-conjugated
streptavidin (Jackson Immunoresearch, West Grove, PA) was added (50 µl of
1∶1,000 dilution in PBSTB) and incubated for 30 min at room temperature.
Plates were washed in PBSTB, soaked for 1 hour in PBSTB and incubated with blue
substrate (Vector Labs, Burlingame, CA) until spots were clearly visible, then
rinsed with tap water. When plates were dry, spots were counted using an
automated ELISPOT reader.
Statistical analysis was performed by using GraphPad Prism statistical software
(GraphPad Software, San Diego, CA). Non-parametric statistical tests were used.
The Mann-Whitney U was used for comparison tests and the Spearman rank test were
used for correlation analyses.
Peripheral venous blood was drawn from 22 HTLV-1 seropositive patients and 7
HTLV-1 seronegative matched donors, all screened for the presence of
HLA-A*02 alleles, and peripheral blood mononuclear cells (PBMC) were
extracted and cryopreserved.
Tim-3 and PD-1 are two cellular molecules expressed on T cells implicated in
immune exhaustion. We evaluated the expression and co-expression of Tim-3 and
PD-1 on T cells derived from HTLV-1 seropositive (both asymptomatic carriers and
patients with the diagnosis of HAM/TSP) and seronegative controls to determine
whether they were modulated in HTLV-1 infection. We observed a significant
decrease in the frequency of Tim-3+ PD-1−
expressing CD8+ and CD4+ T cells among HTLV-1
seropositive subjects (CD8+: median 8.01%, IQR
5.42–10.50; CD4+: median 4.3%, IQR 3.50–5.99)
compared to HTLV-1 seronegative controls (CD8+ median
15.10%, IQR 10.50–17.60; CD4+: median 6.84%,
IQR 5.74–7.85) (Figure 1A and
B). Patients with HAM/TSP (red circles) had significantly lower
levels of Tim-3+ PD-1− expressing
CD8+ (p = 0.002) and
CD4+ (p = 0.004) T cells compared to
healthy uninfected controls (open circles). In contrast, the frequency of
Tim-3− PD1+ T cells trended to an increase
in subjects with HTLV-1 infection (CD8+: median 18.80%,
IQR 10.42–24.90; CD4+: median 20.70%, IQR
13.6–25.35) compared to healthy uninfected controls (CD8+:
median 9.22%, IQR 8.97–15.50; CD4+: median
13.60%, IQR 12.7–18.6) (Figure 1A and B). Only a few T cells
co-expressed both Tim-3 and PD-1, and no differences were observed between
uninfected subjects and those with HTLV-1 asymptomatic infection or HAM/TSP
patients. Using linear regression analysis we observed no association between
the frequency of Tim-3 or PD-1 expression on CD8+ T cells in
HTLV-1 infected subjects and proviral load. (p = 0.68;
r = 0.1043; or p = 0.89;
r = −0.03202, respectively).
HLA- A*02 positive HTLV-1-infected patients have high amounts of circulating
CD8+ T cells specific for an immunodominant HLA- A*02
-restricted epitope, HTLV-1 Tax 11–19 [20], [49], [50]. In HAM/TSP patients, these
HTLV-1's Tax-specific CD8+ T cells correlate with HTLV-1
proviral load [23]. Among this cohort, we identified 15 HLA-A2 positive
subjects (asymptomatic carriers, n = 9 and HAM/TSP,
n = 6; Table
1), and evaluated the Tim-3 and PD-1 receptor expression on
Tax-specific CD8+ T cells. Eight patients had Tax-specific
CD8+ T cells (median 2.45%, IQR 1.11–5.31) as
determined by specific pentamers. Among these patients we also observed
HLA-A*02 -restricted CMVpp65 CD8+ T cells (median 2.49%, IQR
1.87–11.37). Interestingly, Tim-3 levels were dramatically reduced on
CD8+ Tax 11–19-specific T cells (median 24.77%,
IQR 15.2–39.54) compared to the expression of PD-1 (median 48.06%,
IQR 36.81–65) (Figure 2A,C
and E). We also evaluated Tim-3 expression on HLA- A*02 CMV
specific T cells and found a similar pattern of expression with Tim-3 levels
reduced on CD8+ CMV-specific T cells (median 27.62%, IQR
21.48–43.19) compared to PD-1 (median 47.70%, IQR
40.45–51.16) (Figure 2B,D and
F).
To determine whether there was an association with Tim-3 or PD-1 levels on Tax
11–19-specific CD8+ T cells and their functionality, we
evaluated the production of IFN-γ in response to the HLA-A*02-restricted
Tax 11–19 immuno-dominant epitope and in comparison, the CMVpp65 epitope
by an ELISPOT assay derived from PBMCs derived from 8 HLA- A*02 restricted
infected individuals with Tax 11–19- and CMVpp65 specific
CD8+ T cells (Figure 3). We saw no correlation between IFN-γ secretion and
global PD-1 or Tim-3 expression on either the CD4+ or
CD8+ T cells, irrespective of disease status (data not
shown). The frequency of PD-1 expression on Tax-specific or CMV-specific
CD8+ T cells also did not associate with the amount of
IFN-γ secreted (r = 0.1317;
P = 0.7520 and r = 0.2245;
P = 0.594, respectively) (Figure 3B). However, we observed a
statistically significant inverse correlation between the frequency of Tim-3 on
both Tax-specific as well as CMV-specific CD8+ T cells and the
amount of IFN-γ secreted (r = −0.8982;
P = 0.0046; r = 0.9710;
P = 0.0028; Figure 3A).
Tax expression marks HTLV-1 viral replication in both CD4+ and
CD8+ infected T cells. We aimed to determine whether the
downregulation of Tim-3 we had observed was occurring only among infected cells,
or in bystander cells as well. We therefore co-stained for Tax and Tim-3
expression on T cells from HTLV-1 infected subjects. We also stained for PD-1
expression as a control. The culture of PBMC overnight did not alter Tim-3 or
PD-1 expression levels on the HTLV-1-infected T cells (data not shown). We
observed that Tax was expressed on PBMC from some subjects following 24 hours of
culture and was detected on both Tim-3+ as well as
Tim-3− CD4+ T cells (Figure 4A). Similarly, Tax was present on
both PD-1+ and PD-1− T cells. We further
identified a unique subset of Tax expressing CD4+ T cells that were
Tim-3hi and lacked PD-1 in most of the subjects expressing Tax
(Fig. 4A). No difference
in the pattern of co-expression between HTLV-1 seropositive asymptomatic
patients and those diagnosed with HAM/TSP was observed.
An increase in Tim-3 levels on T cells would potentially lead to a downregulation
of T cell functionality. We therefore tested several gamma-chain associated
cytokine mediators that could potentially modulate Tim-3 expression. We observed
that IL-2, and especially IL-15, led to a prominent increase in the frequency of
Tim-3 levels, specifically on the CD8+ T cell population after only 12
hours in culture (Figure
4B,C). No change in the levels of PD-1 expression were observed on
both CD8+ and CD4+ T cells (Figure 4B,C).
CD8+ T cell dysfunction and/or exhaustion are common features of many
chronic viral infections, including HIV-1 and HCV infections [29], [30], [31], [32], [33], [34], [35], [36]. The mechanisms of T cell
dysfunction are complex, but are in part mediated by a distinct set of inhibitory
receptors [27],
[51]. A
high, and sustained, expression of Tim-3 and PD-1, have emerged as hallmarks of T
cell exhaustion in human viral infections, and blockade of these pathways can
reinvigorate immune responses during persisting viral infections [29], [30], [33], [34], [36]. In this
study, we report that CD8+ and CD4+ T cells in
HTLV-1 infection express lower levels of Tim-3, and this was more pronounced in
patients with HAM/TSP. Phenotypically, we observed that Tax HTLV-1-specific,
HLA-A*02 -restricted CD8+ T cells consistently retain a lower
frequency of Tim-3. We propose that this low expression of Tim-3 on HTLV-1
Tax-specific T cells may lead to a persistent and deleterious effector T cell pool
leading to more inflammation.
The pattern of expression of PD-1 in HTLV-1 infection has recently been shown to be
elevated on T cells in HTLV-1 carriers and also on CMV and EBV specific T cells in
asymptomatic carriers compared to healthy controls [52]. This opposing relationship of
PD-1 and Tim-3 expression on T cells in patients with HTLV-1 infection suggests that
the downregulation of Tim-3 expression potentially leads to more vigorous T cell
activity in the HTLV-1-infected individual, whereas PD-1 may not fully reflect T
cell dysfunction, but rather an activated status of the T cell response to
infection. Indeed the association between the frequency of Tim-3 and PD-1 levels
with IFN-γ secretion in response to either Tax or CMVpp65 epitopes show
remarkably different correlations. In a study by Petrovas and colleagues, it was
apparent that PD-1 expressing T cells are able to secrete cytokines in response to
viral peptides [39]. Our data suggests that PD-1 and Tim-3 on antigen
specific CD8+ T cells are functionally different, and this may
reflect a distinct stage of differentiation. PD-1 appears to mark early T-cell
activation and exhaustion, while Tim-3 represents a more terminal stage of
impairment.
The positive association between the frequency of HTLV-1's Tax-specific
CD8+ T cells and HTLV-1's Tax mRNA load and proviral load
is well documented [8], [53], [54]. Studies evaluating the phenotype of CD8+
T cells in HTLV-1 infection have been largely limited to characterizing the
expression of T cell maturation and differentiation markers (CD28, CD45RO) [14]. Our data
suggest that downregulation of Tim-3, rather than PD-1, marks global and
Tax-specific CD8+ T cells, which are hyperfunctional. This contrasts
with HIV-1 and HCV infections, where the expression of Tim-3 is increased, leading
to a population of CD8+ T cells that are rendered dysfunctional both
in terms of proliferative capacity and cytokine release as well as release of
cytolytic granules [29], [36].
Surface receptors known to regulate T cell function like CD244 and PD-1 have been
shown to be upregulated either directly due to Tax or indirectly due to the cytokine
milieu [52], [55]. We
postulate that either direct HTLV-1 viral components led to a downregulation of
Tim-3, or as yet to be defined cytokine(s), suppress Tim-3 expression. In several
human and murine studies, the manifestation of autoimmune diseases such as multiple
sclerosis, have been attributed as a result of downregulated Tim-3 expression on T
cells [56].
It still remains unclear how HTLV-1 infection sustains low levels of Tim-3 on T cells
in infected patients and whether this is a cause or a consequence of disease
progression. Multilayered mechanisms for this regulation may be occurring in the
context of HTLV-1 infection. One strategy to reduce the T cells response would be
through enhancement of the Tim-3 receptor for engagement with its cognate ligand.
This could serve as a novel strategy to dampen the inflammatory inducing T cells.
From our results, PD-1 engagement may not be as effective since both
PD-1− and PD-1+ cells retain the potential for
CD8+ T cell lytic function.
A novel strategy to reverse or prevent the onset of neurological complications would
be through dampening effector T cell functions. From our results, it appears the
γ-chain cytokines elicited higher levels of Tim-3 on specifically on
CD8+ T cells, and such a strategy could be harnessed to dampen T
cell function in the HTLV-1 infected individual. Further work to understand the
mechanisms for HTLV-1 disease progression and devise strategies to effectively
prevent neurological complications will be needed. Targeted modulation of the Tim-3
pathway provides a viable model for this intervention.
|
10.1371/journal.ppat.1000761 | A Broad Distribution of the Alternative Oxidase in Microsporidian Parasites | Microsporidia are a group of obligate intracellular parasitic eukaryotes that were considered to be amitochondriate until the recent discovery of highly reduced mitochondrial organelles called mitosomes. Analysis of the complete genome of Encephalitozoon cuniculi revealed a highly reduced set of proteins in the organelle, mostly related to the assembly of iron-sulphur clusters. Oxidative phosphorylation and the Krebs cycle proteins were absent, in keeping with the notion that the microsporidia and their mitosomes are anaerobic, as is the case for other mitosome bearing eukaryotes, such as Giardia. Here we provide evidence opening the possibility that mitosomes in a number of microsporidian lineages are not completely anaerobic. Specifically, we have identified and characterized a gene encoding the alternative oxidase (AOX), a typically mitochondrial terminal oxidase in eukaryotes, in the genomes of several distantly related microsporidian species, even though this gene is absent from the complete genome of E. cuniculi. In order to confirm that these genes encode functional proteins, AOX genes from both A. locustae and T. hominis were over-expressed in E. coli and AOX activity measured spectrophotometrically using ubiquinol-1 (UQ-1) as substrate. Both A. locustae and T. hominis AOX proteins reduced UQ-1 in a cyanide and antimycin-resistant manner that was sensitive to ascofuranone, a potent inhibitor of the trypanosomal AOX. The physiological role of AOX microsporidia may be to reoxidise reducing equivalents produced by glycolysis, in a manner comparable to that observed in trypanosomes.
| Microsporidia are obligate intracellular parasites responsible for a number of diseases in commercially important animals (e.g. bees) and of significant medical concern, in particular in immunocompromised humans. Though related to fungi, microsporidia have undergone a rapid phase of adaption to the intracellular environment and have in the process reduced many aspects of their biology. Notably, microsporidia have highly reduced mitochondria (powerhouses of the cell) reflected in reduced energy metabolic pathways. Thus they likely produce ATP only through the process of glycolysis. In some parasites, this glycolytic pathway is dependent on an additional step involving a protein called the “alternative oxidase”. We have shown that this protein is also present in several species of microsporidia. Crucially, this protein is absent from humans and so can potentially be exploited as a drug target. Our experiments show that this protein is likely widespread in microsporidia, and is sensitive to the antibiotic ascofuranone, which is currently being tested as a potential treatment for the agent causing sleeping sickness. Our results suggest that knowledge gleaned from drug trials on sleeping sickness is potentially transferrable to the treatment of some cases of microsporidiosis.
| Microsporidia are a large and diverse group of eukaryotic intracellular parasites that infect a wide variety of animal lineages, including humans [1]. Although once thought to be early branching eukaryotes, they are now widely accepted to be very atypical parasitic fungi [2],[3],[4],[5]. They are highly adapted to the infection process, and many typical eukaryotic features have been simplified, reduced, or lost completely. Microsporidian genomes are reduced and organelles such as the peroxisome, mitochondria and Golgi apparatus are absent or altered from their canonical forms [6],[7],[8].
In particular, microsporidian mitochondria have been severely reduced into biochemically and physically streamlined “mitosomes” [8]. Mitosomes lack their own genome, and there is no evidence from the nuclear genome of any microsporidian for genes encoding any of the respiratory chain complexes or an F1-ATP synthase complex. In the absence of the ability to synthesize ATP through oxidative phosphorylation, microsporidia appear to import ATP directly from their host cell via ATP translocases located in the cell membrane [9],[10], using a transporter which may have been acquired by lateral gene transfer from bacterial energy parasites such as Chlamydia and Rickettsia [11]. Identification of which mitochondrial-derived genes have been retained in the complete genome of Encephalitozoon cuniculi, together with immunolocalization studies in E. cuniculi and Trachipleistophora hominis, suggest that the major functional role for the mitosome is not in energy generation, but instead the assembly of iron-sulphur clusters for export to the cytoplasm [6],[12],[13].
Biochemical and genomic evidence all generally point to glycolysis as the major route of energy generation in most microsporidia [6],[9]. In order for ongoing glycolytic activity to be sustainable, however, some mechanism to reoxidise reducing equivalents produced by this pathway is also required. Of the few proteins associated with the microsporidian mitosomes that are not involved in iron-sulfur cluster assembly, one is glycerol-3-phosphate dehydrogenase. This enzyme is the mitochondrial component of the glycerol-3-phosphate shuttle, a pathway used in some eukaryotes to move reducing equivalents into mitochondria [14]. Both cytosolic and mitochondrial components of this shuttle are encoded in the genomes of several microsporidia that have been well studied [6],[15], and it has been suggested that this could provide a mechanism sustaining glycolysis in the cytosol by reoxidising glycerol-3-phosphate [9]. However, the E. cuniculi mitochondrial glycerol-3-phosphate dehydrogenase does not appear to be located in the mitochondrion any longer [12], and even if a working shuttle was present, there is no obvious mechanism for reoxidation of the co-reduced FAD produced by this shuttle in the genome of E. cuniculi [6]. In the bloodstream form of Trypanosoma brucei parasites, the mitochondrial glycerol-3-phosphate dehydrogenase is coupled to an alternative oxidase (AOX) that together achieve this process [16], and a similar system has been postulated to be present in the apicomplexan parasite Cryptosporidium parvum [17].
AOX is a cyanide-insensitive terminal oxidase that is typically located on the inner surface of the inner mitochondrial membrane. It branches from the main respiratory chain at the level of the ubiquinone pool, results in the net reduction of oxygen to water, and is non-protonmotive [18],[19],[20]. It has been found in some prokaryotic lineages, including alpha-proteobacteria [21], and has a wide but discontinuous distribution across eukaryotes: it is widely distributed in plants, and has also been found in a handful of invertebrate animals [22],[23],[24],[25]. In parasitic protists, the distribution of AOX is also uneven: it is known from the amoebozoan Acanthamoeba castellanii, the heterokont Blastocystis hominis, and the trypanosomes. In the alveolates, it is found in the apicomplexan Cryptosporidium and some other distantly related alveolates including some ciliates, but absent from the more closely related Plasmodium parasites [26],[27],[28]. The broad overall distribution of AOX may be indicative of an early origin in eukaryotes, and is perhaps even derived from the endosymbiotic alpha-proteobacterium that gave rise to mitochondria [27],[29].
In fungi, the protein also has a wide but discontinuous distribution [30], but it is absent from the completely sequenced genome of E. cuniculi and from the recent large-scale genome surveys of Nosema ceranae and Enterocytozoon bieneusi [6],[31],[32]. Interestingly, however, we identified a homologue in the partially sequenced genome of Antonospora locustae, demonstrating the pattern of retention versus loss is also uneven within the microsporidia, despite their otherwise common mode of intracellular parasitism and apparently similar metabolism. The possible presence and function of AOX in microsporidia is of practical interest as well, because the absence of AOX in mammals, including humans, renders it a potential therapeutic target for the treatment of microsporidiosis, as is the case in a number of protisitan parasites [16],[33],[34]. This is of particular importance in microsporidia as current medical treatments are not universally effective. The drugs of choice for microsporidiosis are currently albendazole and fumagillin [35]. Whilst albendazole is used in treating many species, some, such as V. corneum and E. bieneusi are resistant and in these cases fumagillin, which is mildly toxic, has to be used [36].
Here, we characterize the phylogenetic distribution of microsporidian AOX, and examine the functional activity of AOX enzymes from the human parasite T. hominis and the insect parasite A. locustae. Phylogenetically the microsporidian AOX is weakly related to mitochondrial homologues from other eukaryotes, and both A. locustae and T. hominis AOX proteins include an N-terminal leader that was demonstrated by confocal microscopy to target the proteins to mitochondria in yeast, altogether suggesting the enzyme is likely derived from the mitosome and may be localized in the organelle still, though direct co-localization would be required to give a definitive location of function. Enzyme assays with recombinant proteins demonstrated both possess cyanide-resistant oxidase activities sensitive to inhibition by the very specific trypanosome AOX inhibitor ascofuranone [37], suggesting the enzyme functions as a terminal electron receptor.
The complete genome of E. cuniculi lacks any gene resembling AOX, but we identified a full-length homologue of the AOX gene in the partial genome of A. locustae (gmod.mbl.edu/perl/site/antonospora01, Antonospora locustae Genome Project, Marine Biological Laboratory at Woods Hole, funded by NSF award number 0135272). To determine the distribution of this gene, degenerate PCR was used to amplify a short fragment of AOX from other species of microsporidia, Glugea plecoglossi (233 bp), Spraguea lophii (235 bp), and T. hominis (236 bp). To examine the complete sequence of an AOX from a human parasite, the ends of the T. hominis AOX gene were also sequenced using 5′ RACE and splinkerette protocols [38], resulting in a full length gene of 957 bp with a translated protein of 318 amino acids (compared to A. locustae AOX, which had a length of 831 bp).
Hypothetical translations of A. locustae and T. hominis sequences contain all conserved sites consistent with AOX activity. Specifically, both genes encode the six conserved di-iron binding ligands that are essential for AOX activity (Figure 1), which are conserved in all alternative oxidases sequenced to date [17],[20],[39]. In addition both sequences contain 4 highly conserved tyrosine residues, one of which (Tyr at the S. guttatum equivalent position 275) is considered to be critical for the net reduction of oxygen to water and probably plays a key role in enzyme catalysis (Figure 1) [19]. Further confirmation that A. locustae and T. hominis sequences encode AOX proteins is the finding that a putative substrate binding site (residues 242–263) [19] is also conserved in both microsporidia. However, one striking difference between the microsporidian AOX sequences and those AOX sequences found in all other mitochondria and protists is the lack of tryptophan-206, which is most unusual since it is highly conserved and has been proposed to play either a structural or catalytic role [18]. In A. locustae the tryptophan has been replaced by serine whilst in T. hominis it has been replaced by alanine. Similar to other parasite AOXs however, none of the cysteines postulated to play a role in the regulation of AOX activity in plants [40], are present in either A. locustae or T. hominis.
Mitoprot I predicted both microsporidian AOX sequences to encode amino-terminal mitochondrial transit peptides, and the T. hominis AOX protein was also predicted by Predotar and TargetP 1.1 to have a mitochondrial targeting peptide. In order to test the degree of conservation and functionality of potential targeting signals, full-length proteins were expressed in S. cerevisiae cells fused to a green fluorescent reporter protein. Expression in yeast shows that GFP overlays mitotracker fluorescence, indicating successful heterologous targeting for both proteins (Figure 2).
The phylogenetic relationship among alternative oxidases is in general poorly resolved. There are several well-supported clades, including the microsporidia, the ascomycete fungi, and the basidiomycete fungi, but the fungi do not form a single well-supported clade (Figure 3A), similar to results recovered in earlier AOX phylogenies [27]. The strong support uniting AOX from A. locustae and T. hominis does, however, confirm the microsporidian genes share a single common origin. Phylogenetic analysis based on the conserved region of the gene amplified from other microsporidia similarly places S. lophii and the G. plecoglossi in the same monophyletic microsporidian group (Figure 3B), further supporting the common origin of all microsporidian AOX genes. The overall distribution of microsporidian AOX was therefore mapped onto an SSU phylogeny including all major clades of microsporidia as defined by molecular and ecological data [41], which showed that AOX is widely distributed in microsporidia, and perhaps only absent from a single clade of predominantly vertebrate and insect parasites, including E. cuniculi, E. bieneusi and N. cerenae (Figure 3C).
To directly examine the function of A. locustae and T. hominis AOX proteins (especially given the sequence difference reported in Figure 1), recombinant A. locustae and T. hominis AOX proteins were expressed in E. coli and the enzyme structure and activity was measured. Antibodies raised against the plant AOX recognize both A. locustae and T. hominis proteins (Figure 4), and both a monomer and a dimer can be detected in Western blots of non-reducing gels, as is the case within the thermogenic plant Sauromatum guttatum, although in the case of A. locustae the monomer is not very prominent. (Figure 4). In E. coli membrane fractions containing either A. locustae or T. hominis recombinant AOX (rAOX), ubiquinol-1 oxidase activity indicates that the activities of both proteins are as expected for AOX (Table 1). In both cases, 1 µM antimycin A, 2 µM myxothiazol and 1 mM potassium cyanide were included in the assay system to ensure inhibition of the cytochrome bo and bd complexes of E. coli, and the specific activities reported in Table 1 have been corrected for auto-oxidation of ubiquinol-1 in the absence of membranes (see methods). It is important to note that, although A. locustae rAOX was more active than T. hominis rAOX, both proteins were equally sensitive to 10 nM ascofuranone (Table 1), the very specific and potent inhibitor of the trypanosomal alternative oxidase [37]. Furthermore, it is apparent from Table 1 that the specific activities of these microsporidia are considerably higher than those reported for rAOX from C. parvum [17] but comparable to those observed with overexpression studies of T. brucei rAOX in E. coli membranes [42].
The genome of E. cuniculi has served as a model for microsporidian metabolism since it was completed [6], however, it has never been clear how this model organism dealt with the reducing potential built up through ongoing glycolysis, since it lacks a terminal oxidase. Here we show that this model does not reflect microsporidia as a whole, because alternative oxidase has a broad distribution amongst microsporidian parasites. This distribution remains discontinuous, however, because we can say with some confidence that AOX is not present in either the E. cuniculi or N. ceranae genomes, which have been sequenced to near completion [6],[32]. It also appears to be absent from the genome of E. bieneusi, although this genome is not completely sampled [31]. Our negative PCR results from E. aedis and A. (Brachiola) algerae are less conclusive (these have previously been shown to have a high AT content that may prevent the successful amplification of the AOX gene by degenerate PCR [43]), but it suggests the gene may also be absent in several other lineages. Whilst G. plecoglossi, T. hominis and S. lophii are quite closely related and within the Marinosporidia clade, Antonospora locustae falls within the distantly related Aquasporidia clade as defined by molecular and ecological analysis [41] (Figure 3C). As we know that the alternative oxidase is present in at least two major clades, and in many fungi, the most parsimonious explanation for its distribution in microsporidia is that it was present in their last common ancestor, but has been lost in E. cuniculi and probably other lineages during their more recent evolutionary history.
Analysis of the AOX sequences from A. locustae and T. hominis reveals that both possess the iron-and substrate-binding motifs found in other AOXs. In S. guttatum, Tyr-253 has been shown to be involved in substrate binding, and Tyr-275 to be critical for catalytic activity [19],[44], and both of these are also conserved in microsporidia. The absences of Trp-206 in A. locustae and T. hominis AOX sequences is somewhat surprising, as it is conserved across all other known mitochondrial AOX sequences. Since A. locustae and T. hominis AOX sequences are demonstrably functional (Table 1), Trp-206 cannot play a universally critical role in electron transport, but it may have a role in other mitochondrial AOXs as helping to anchor the protein to the leaflet of the inner mitochondrial membrane in a manner seen with other monotopic membrane proteins [19],[20],[45].
The demonstration that A. locustae and T. hominis rAOX have a high quinol oxidase activity that is sensitive to ascofuranone at nanomolar concentrations not only solves a significant puzzle in microsporidian metabolism, but also offers a new avenue of treatment for some microsporidian species and further “in tissue culture” trials can establish the efficiency of the drug across the life cycle of the microsporidian. There is currently considerable interest in this antibiotic, originally isolated from the phytopathogenic fungus Ascochyta visiae, for its potential promise in the treatment of trypanosomiasis and cryptosporidiosis. The fact that it also appears to potently inhibit the microsporidian AOX may give the drug a more widespread use than previously considered. Of course several of the microsporidia that parasitise humans lack the AOX (e.g. E. cuniculi and E. bieneusi), but for other human parasites (e.g. T. hominis) the AOX is clearly a potential target, and may also be in other unexplored lineages (e.g., Vittaforma corneae).
With respect to the potential function of AOX in microsporidia a possible role may be similar to that proposed in the bloodstream form of some trypanosomes. In the bloodstream form of Trypanosoma brucei, where glucose is abundant and there is no conventional respiratory chain [16], ATP synthesis is switched from oxidative phosphorylation to substrate level phosphorylation. Glycolysis is contained within a glycosome, a membrane-bound organelle containing glycolytic enzymes. In this system, reducing equivalents generated by glycolysis in the form of glycerol-3-phosphate are shuttled out of the glycosome and re-oxidised by a glycerol-3-phosphate dehydrogenase (G3PDH) located on the outer surface of the inner membrane. G3PDH itself reduces the mitochondrial ubiquinone pool that in turn is then re-oxidised by the alternative oxidase. In this way, glycerol-3-phosphate within the glycosome is continuously being re-oxidised to supply further substrate for the net oxidation of NADH [16]. Thus in an indirect manner mitochondrial alternative oxidase activity maintains the NADH/NAD balance within the glycosomes. In addition to the alternative oxidase, however, trypanosomes also possess a glycerol kinase that under anaerobic conditions helps to maintain the glycosome NADH/NAD balance by converting glycerol-3-phosphate to glycerol [16].
It is plausible that most microsporidia rely on a similar system and that AOX fulfils the role of the terminal oxidase, as shown in Figure 5. Whether the microsporidian AOX functions in the mitosome or cytosol is not completely certain, but its very presence in the cell and its carrying out the functions we have demonstrated in vitro significantly change our view of microsporidian metabolism and drug sensitivity in either event. Overall, the presence of an N-terminal leader with characteristics of a transit peptide, together with the likely mitochondrial origin of the protein, all suggest a mitosomal location is most plausible. This also fits well with previously unusual observations on the glycerol-3-phosphate shuttle. Localization studies on mitochondrial glycerol-3-phosphate dehydrogenase in E. cuniculi show no evidence that the enzyme is confined to mitochondria or specifically localized there, unlike ferredoxin, frataxin, ISCU and ISCS [12],[13], and in E. bieneusi the gene appears to be absent altogether [31]. This suggests that the glycerol shuttle has been displaced in these microsporidia, which is functionally consistent with the absence of the alternative oxidase protein in both species.
The A. locustae alternative oxidase sequence was retrieved from the GMOD MBL A. locustae database and used to design degenerate primers to amplify a fragment of the alternative oxidase gene from T. hominis, G. plecoglossi and S. lophii (Forward 5′-GAAACWGTWGCWGCWGTNCCNGG-3′, Reverse 5′-ATWGCTTCTTCTTCNAKRTANCCNAC-3′). Degenerate PCR was carried out on DNA from E. cuniculi to exclude the possibility that the AOX gene is present in the genome within the subtelomeric regions that have not been fully assembled [6]. This gave negative results. Negative degenerate PCR results were found for Brachiola algerae and Edhazardia aedis. The full-length gene was amplified from T. hominis DNA and RNA obtained from purified spores from cultures maintained in rabbit kidney cells at Rutgers, State University of New Jersey. The 5′ prime end of the gene was amplified using RLM-RACE using primers designed from within the fragment amplified by degenerate PCR. The first round of PCR yielded a product truncated at the 5′ end. Primers were then designed from within that fragment to obtain the presumed full-length gene. A splinkerette strategy was used to obtain 3′ end of the gene [38]. Amplified PCR products were cloned using the TOPO TA cloning system (Invitrogen) and sequenced using Big Dye 3.2 (ABI). Mitochondrial transit peptides were predicted using Mitoprot I [46], Predotar [47], and TargetP 1.1 [48]. (New sequences are deposited in the GenBank Database under the accession numbers GU221909-GU221911).
DNA fragments corresponding to A. locustae and T. hominis AOX open reading frames were amplified by PCR by using primers that generated in-frame restriction sites. PCR products were cloned upstream of green fluorescent protein (GFP)-S65T under the control of the MET25 promoter [49] for analysis by confocal or fluorescence microscopy. Constructs were then transformed into the diploid yeast strain JK9-3da/a (leu2-3,122/leu2-3,122 ura3-52/ura3-52 rme1/rme1 trp1/trp1 his4/his4 GAL+/GAL+ HMLa/HMLa), and plated on uracil and methionine deficient SD plates (2% (w/v) agar, 2% (w/v) glucose and 0.67% (w/v) yeast nitrogen base supplemented with the relevant amino acids). Positive colonies were grown overnight in SD medium lacking uracil and methionine and stained with MitoTracker (MitoTracker Red CM-H2XRos) according to the manufacturer's protocol (Molecular Probes). Yeast cells were visualized using the Zeiss meta confocal microscope.
Separation of yeast mitochondrial proteins on non-reducing SDS-polyacrylamide gels, transfer to nitrocellulose membranes, and detection of AOX protein using monoclonal antibodies raised against the S. guttatum AOX [50] was performed as described previously [51].
The A. locustae and T. hominis gene sequences were amplified using Phusion High-Fidelity Taq (New England Biolabs) and cloned into the pet14b expression vector. Both constructs were used to transform E. coli strain C41, which is especially suited to the expression of transmembrane proteins. Bacterial membranes were prepared using 2.5 L Luria broth cultures, adapted from Berthold [52] and as described in detail by Crichton et al 2009 [53]. Flasks containing Luria Broth, 0.02% glucose, 0.002% FeSO4 and 50 µgml−1 ampicillin were inoculated with 10 mlL−1 starter culture, and incubated at 37°C for 4 hours. The temperature was reduced to 18°C, and the cultures were incubated for one hour prior to induction with 100 µM IPTG. After induction, the cultures were incubated for 18 hours at 18°C. Cells were then harvested using centrifugation at 11,000×g for 10 minutes. After initial centrifugation, cells were resuspended in 60 mM Tris-HCl (pH 7.5), 5 mM DTT, 300 mM NaCl and 0.1M PMSF and then sonicated for 8 minutes at 14 microns. After sonication, cell debris was removed by centrifugation at 12,000×g for 15 minutes, and clear supernatant was further refined by a 2-hour ultracentrifugation step at 200,000×g. Pellets from final spin were resuspended in 60 mM Tris-HCl (pH 7.5), 5 mM DTT, 300 mM NaCl and used for subsequent gel and assay analysis. Ubiquinol oxidase activity (AOX activity) was measured by recording the absorbance change of ubiquinol-1 at 278 nm (Cary UV/vis -400 Scan spectrophotometer). Reactions were started by the addition of ubiquinol-1 (final concentration 150 µM, ε278 = 15,000 M−1cm−1) after 2 min preincubation at 25°C in the presence of rAlAOX and rThAOX in 50 mM Tris-HCl (pH 7.4). Endogenous ubiquinol activities were inhibited by inclusion of 1 µM antimycin A, 2 µM myxothiazol and 1 mM potassium cyanide in the assay medium.
The A. locustae and T. hominis AOX amino acid sequences were aligned to 47 diverse proteins sequences with representatives from animal, kinetoplastid, fungal, heterokont, plant and proteobacterial lineages. Sequences were aligned using ClustalW [54] and manually edited and masked. The alignment was analysed using Modelgenerator to select an appropriate model for amino acid change [55]. Phylogenetic trees were inferred using MrBayes 3 [56] with a Blosum62 matrix and with 2 runs each of 1000000 generations carried out on the freely available Bioportal (www.bioportal.uio.no). A burn-in of 400 trees was removed from each run and a consensus created from remaining trees. Five hundred bootstrapped data matrices were also analysed by maximum likelihood using PhyML 3.0 [57] with a JTT model of amino acid change and an estimated gamma parameters with four rate categories of amino-acid change. A second alignment restricted to the conserved area amplified by degenerate PCR from S. lophii, G. plecoglossi was also analysed. Trees were inferred and 100 bootstrap datasets analysed from this short alignment using PhyML, using the parameters described above. The SSU rRNA backbone phylogeny was based on available SSU sequences from NCBI, which were aligned using ClustalW, manually edited and masked and analysed using PhyML 3.0 with a JC69 nucleotide substitution model with estimated gamma parameter and 4 categories of rate change.
|
10.1371/journal.ppat.1003308 | Microbes Bind Complement Inhibitor Factor H via a Common Site | To cause infections microbes need to evade host defense systems, one of these being the evolutionarily old and important arm of innate immunity, the alternative pathway of complement. It can attack all kinds of targets and is tightly controlled in plasma and on host cells by plasma complement regulator factor H (FH). FH binds simultaneously to host cell surface structures such as heparin or glycosaminoglycans via domain 20 and to the main complement opsonin C3b via domain 19. Many pathogenic microbes protect themselves from complement by recruiting host FH. We analyzed how and why different microbes bind FH via domains 19–20 (FH19-20). We used a selection of FH19-20 point mutants to reveal the binding sites of several microbial proteins and whole microbes (Haemophilus influenzae, Bordetella pertussis, Pseudomonas aeruginosa, Streptococcus pneumonia, Candida albicans, Borrelia burgdorferi, and Borrelia hermsii). We show that all studied microbes use the same binding region located on one side of domain 20. Binding of FH to the microbial proteins was inhibited with heparin showing that the common microbial binding site overlaps with the heparin site needed for efficient binding of FH to host cells. Surprisingly, the microbial proteins enhanced binding of FH19-20 to C3b and down-regulation of complement activation. We show that this is caused by formation of a tripartite complex between the microbial protein, FH, and C3b. In this study we reveal that seven microbes representing different phyla utilize a common binding site on the domain 20 of FH for complement evasion. Binding via this site not only mimics the glycosaminoglycans of the host cells, but also enhances function of FH on the microbial surfaces via the novel mechanism of tripartite complex formation. This is a unique example of convergent evolution resulting in enhanced immune evasion of important pathogens via utilization of a “superevasion site.”
| Complement is an important arm of innate immunity. Activation of this plasma protein cascade leads to opsonization of targets for phagocytosis, direct lysis of Gram-negative bacteria, and enhancement of the inflammatory and acquired immune responses. No specific signal is needed for activation of the alternative pathway of complement, leading to its activation on all unprotected surfaces. Pathogenic microbes need to evade this pathway, and several species are known to recruit host complement inhibitor factor H (FH) to prevent the activation. FH is important for protection of host cells, too, as defects in FH lead to a severe autoreactive disease, atypical hemolytic uremic syndrome. We have now identified at the molecular level a common mechanism by which seven different microbes, Haemophilus influenzae, Bordetella pertussis, Pseudomonas aeruginosa, Streptococcus pneumoniae, Candida albicans, Borrelia burgdorferi and B. hermsii, recruit FH. All microbes bind FH via a common site on domain 20, which facilitates formation of a tripartite complex between the microbial protein, the main complement opsonin C3b, and FH. We show that, by utilizing the common microbial binding site on FH20, microbes can inhibit complement more efficiently. This detailed knowledge on mechanism of complement evasion can be used in developing novel antimicrobial chemotherapy.
| Complement system (C) is an important part of innate immunity in human plasma, and the alternative pathway of complement (AP) is the first line of defense against invading microbes. AP is spontaneously activated on all unprotected surfaces leading to covalent binding of the main complement opsonin C3b to hydroxyl or amine groups. Surface-attached C3b forms a base for enzymatic convertases, which cleave intact C3-molecules until the activator surface is covered with C3b-molecules. This opsonization leads to opsonophagocytosis, propagation of the cascade resulting in release of chemotactic and anaphylatoxic peptides, and formation of lytic membrane attack complexes. To prevent attack against host structures and over consumption of the components in plasma, complement needs to be tightly regulated.
The main regulator of the AP in plasma is factor H (FH). FH is a 150 kDa glycoprotein and consists of twenty globular complement control protein modules (CCPs), each approximately 60 residues long. The AP control activity of FH is in domains 1–4 (FH1-4) [1], [2]. The so-called cofactor activity of FH is needed for inactivation of the central complement opsonin C3b by the serine-protease factor I. In addition to this, FH regulates AP activation by competing with factor B in binding to C3b and accelerating the decay of AP convertase C3bBb [3], [4]. To regulate complement, FH has to discriminate between host and non-host surfaces, as activation is warranted on microbial surfaces, but obviously not on host surfaces. This “target recognition” site is known to be in the carboxyl-terminal domains 19–20 (FH19-20) [5], [6]. Our structures of domains 19–20 alone [7] and complexed with C3d [8] showed how SCR20 can bind to cellular and glycosaminoglycan containing surfaces while SCR19 binds simultaneously to C3d part of C3b facilitating control of the AP. This dual binding ability facilitates target recognition by the AP.
The necessity of FH and its ability to distinguish between host and non-host surfaces is demonstrated by mutations in the carboxyl-terminus of FH. Even heterozygous mutations in this region can lead to uncontrolled AP activation on host cells causing severe damage to endothelial cells, red cells, and platelets, resulting in a serious systemic disease, atypical hemolytic uremic syndrome [9]. Another important target binding region in FH is within domain 7 and polymorphism in this domain is strongly associated with age-related macular degeneration, the most common cause of blindness in elderly people in industrialized countries [10], [11].
FH is utilized by several pathogenic microbes for protection against complement attack [12]. Binding of FH down regulates opsonization and prevents further amplification of the C cascade followed by formation of cytolytic membrane attack complexes. While prevention of opsonization and subsequent phagocytosis is beneficial for practically all microbes, evasion of membrane attack complex formation is especially important for Gram-negative bacteria and spirochetes. Acquisition of FH is important or even essential for pathogens; increasing numbers of them have been shown to bind FH [12]. There are two main interaction sites on FH for microbial binding (Table S1); one is within domains 6–7, and group A streptococci [13] and Neisseria [14], for example, utilize this site. Binding via domains 6–7 facilitates also utilization of FHL-1, an alternatively spliced transcript derived from FH-gene which contains domains 1–7 of FH and has cofactor-activity like FH [15]. Many microbes have been shown to bind both FH and FHL-1 [16].
The other microbial interaction site on FH is in the carboxyl-terminal domains 19–20. It seems that most microbes utilize both sites: for instance, B. burgdorferi sensu stricto, which causes Lyme disease, binds FH via domain 7 using protein CRASP-1 [17] and via domains 19–20 using outer surface protein E (OspE) and its paralogs [18]. This ability for dual binding facilitates efficient protection against the AP attack. Due to the high homology between the C-terminus of FH and C-termini of FH-related proteins (FHRs), some microbes bind also certain FHRs but the significance of this phenomenon for immune evasion is not clear yet.
We wanted to analyze in detail how and especially why different microbes utilize FH via the carboxyl-terminus. We selected pathogens representing Gram-negative, Gram-positive, and eukaryote microbes known to bind FH, and three microbial proteins, OspE (from B. burgdorferi sensu stricto) [18], FhbA (from B. hermsii) [19], and Tuf (from P. aeruginosa) [20]. We discovered that they all share a common binding site in domain 20 that overlaps but is not identical with the heparin and cellular binding sites. We also showed that FH bound to the microbial binding site forms a tripartite complex with C3b and furthermore, formation of this complex not only facilitates regulation of the AP but also enhances it.
We first characterized at the molecular level how microbes bind FH via domains 19–20. We generated point mutations to 14 surface exposed residues of a recombinant fragment of FH domains 19–20 and used five different microbes isolated from patients: three Gram-negative bacteria P.aeruginosa (Pa) [20], (H. influenzae (Hi) [21], B. pertussis (Bp) [22]), one Gram-positive bacterium (S. pneumoniae (Sp) [23]), and one eukaryotic pathogen (C. albicans (Ca) [24]). We also measured binding of full FH to strains used and noticed they all bind FH, as expected on the basis of previous reports (Figure S1).
Binding of 125I-labeled wild type (wt) FH19-20 was measured in the presence of increasing amounts (up to 7 µM) of the mutant FH19-20 constructs. Concentrations of the mutants that inhibited 50% of the wt FH19-20 binding (IC50) were calculated from binding curves of three experiments done in triplicate (examples are shown in Figure S2) and shown in Figure 1 as a reciprocal value (1/IC50) for clarity (diminished value indicating diminished binding). Three mutations, R1182A, R1203A, and R1206A, caused decreased binding to all five microbes (p<0.05); K1188A had reduced binding to four microbes (Hi, Pa, Sp, Ca); R1210A to three (Hi, Pa, Sp); and the K1186A and R1215Q mutations reduced binding to one microbe (Hi) (Figure 1). Four other mutations (W1183L, T1184R, L1189R, E1198A) in domain 20 and three (D1119G, Q1139A, W1157L) in domain 19 showed no reduction in binding compared to wt.
To further characterize interaction of FH with microbial surfaces, similar binding inhibition assays were used with three non-homologous and structurally unrelated bacterial outer surface proteins: OspE, a 15 kDa protein from a Lyme borreliosis agent B. burgdorferi [18], FhbA, a 20 kDa protein from a relapsing fever spirochete B. hermsii [25], and Tuf, a 43 kDa protein from P. aeruginosa [20]. Binding of 125I-FH19-20 to the recombinant proteins was measured in the presence of increasing concentrations of the 14 mutant proteins and the IC50 values were calculated from the binding curves as above. When compared to wt FH19-20, two mutant proteins, R1182A and R1206A, showed decreased affinity to all the three microbial proteins, five mutants (W1183L, L1189R, E1198A, R1203A, R1215Q) to two microbial proteins and one mutant (R1210A) to one protein (p<0.05) (Figure 2, Panels A–C, shown as a reciprocal value (1/IC50) for clarity). The effect of three mutants (T1184R, K1186A, K1188A) in domain 20 and three (D1119G, Q1139A, W1157L) in domain 19 was comparable to wt FH19-20 (p>0.05). Six of the mutants showed decreased binding to both OspE and FhbA suggesting an overlap of the binding sites. The overlap was confirmed using cross inhibition assays with OspE and FhbA (Figure 2, Panels D and E).
Taken together, the binding inhibition assays revealed that all mutants that affected binding were in the domain 20. Furthermore, we identified one mutant (R1182A) with significantly decreased binding to all the microbes or microbial proteins analyzed and two mutants (R1203A, R1206A) with significantly reduced binding to seven out of eight targets (p<0.05) (Table 1). In addition, the three central residues in microbial binding, R1182A, R1203A, and R1206A, are close to each other in the crystal structure of FH19-20 [7]. They are within 14 Å of each other on domain 20 and three residues (K1188A, R1210A, R1215Q) involved in binding to several microbes are also nearby (Figure 3). Folding of all these mutants was comparable to wt FH19-20 according to a circular dichroism analyses (Figure S3).
One binding site for glycosaminoglycans/heparin is located at FH domain 20 [26]. We next analyzed if microbes could utilize this site by analyzing binding of 125I-FH19-20 to OspE, FhbA, and Tuf in the presence of heparin, a model substance for cell surface glycosaminoglycans. We showed that heparin inhibits binding of FH19-20 efficiently to Tuf and slightly also to OspE and FhbA (Figure 4). The data are consistent with previous data showing that glycosaminoglycans bind to residues R1203, R1206, R1210, and R1215 at the very carboxyl-terminus of FH20 [27]. This suggests that the microbial binding site on FH overlaps to some extent with, but is not identical to, the heparin binding site needed for recruitment of FH to eliminate C3b on host cells.
Down-regulation of the AP by FH on host cells occurs because FH20 binds to glycosaminoglycans/heparin while FH19 binds simultaneously to the C3d part of C3b [8], [28]. Next we tested if microbes could utilize FH similarly, i.e. facilitating a two point binding of FH19-20 to surface-bound C3b, one site binding to the microbial protein and the other to C3b. There are two binding sites on FH19-20 for the central complement opsonin C3b, one in domain 19 and the other in domain 20 [8]. Structural analysis shows that the site on domain 20 overlaps with the microbial site, while the site on domain 19 of FH is clearly distinct from it [8]. In agreement with our model, binding of C3d did not inhibit binding of FH19-20 to the microbial proteins (Figure 4, panels A–C) but, to our surprise, actually enhanced it.
As C3d enhanced binding of FH19-20 to microbial proteins, we analyzed further if microbial proteins could enhance binding of FH19 to its main physiological ligand, C3b. We measured the binding of 125I-FH19-20 to C3b in the presence of OspE, FhbA, and Tuf. OspE and FhbA enhanced binding of FH19-20 to C3b statistically significantly while enhancement with Tuf was smaller and not significant (Figure 5, Panel A). This suggests that a microbial protein, FH19-20, and C3b together form a tripartite complex. We were able to prove this by measuring binding of 125I-OspE to solid phase C3b in the presence of FH19-20 (Figure 5, Panel B). This means that the tripartite complex must form, because OspE alone does not bind C3b [18]. Mutation of four central residues in the C3d/C3b binding site on domain 19 of FH (FH19del-20) [8] significantly reduced the formation of the tripartite complex, indicating that the C3d/C3b binding site on domain 19 is essential for the interaction (Figure 5, Panel B). These experiments show that FH19-20 can bind simultaneously to a microbial protein and C3b, and that binding of microbial proteins to FH19-20 enhances the FH-C3b interaction. To further test formation of the tripartite complexes on microbial surfaces we measured effect of C3d (100 µg/ml) on binding of FH19-20 to the surface of whole microbes (B. burgdorferi, S. pneumoniae, P. aeruginosa, H. influenzae and C. albicans). A small increase in FH19-20 binding was observed with all the used microbes, most clearly with S. pneumoniae and C. albicans (Figure S4). No binding of 125I-C3d to any microbes was seen (data not shown).
By modeling the tripartite complex on a surface using the structure of FH19-20 in complex with C3d [8], C3b (containing the C3d part) [29], and our recent crystal structure of FH19-20 in complex with borrelial OspE protein (Bhattacharjee et al., submitted), a model of a microbial surface protein, we could also show that formation of a tripartite complex is possible without any steric clashes. Furthermore, in this model the thioester site of C3b faces towards the membrane indicating that a surface-bound microbial protein can enhance binding of FH to C3b on the same surface (Figure 5, Panel C).
The results above suggested that, by enhancing the interaction between FH and C3b, microbes might be able to down-regulate complement activation more efficiently. The main regulatory function of FH is to act as a cofactor for serine protease factor I in inactivation of C3b. We therefore measured the cofactor activity of full length FH in factor I mediated cleavage of C3b in the presence of the three microbial proteins, OspE, FhbA, or Tuf (Figure 6, Panels A and B). All tested microbial proteins enhanced significantly the cofactor activity of FH (p<0.05 at ≥20 µg/ml for all of the proteins). The enhancement was due to the carboxyl-terminal part of FH, since it did not clearly occur when FH1-4 was used instead of full length FH (Figure 6, Panel C), i.e. enhancement obviously requires domains 19 and 20 that mediate formation of the tripartite complex.
Escape of the complement system, and especially its alternative pathway amplification cascade, is a prerequisite for microbial virulence since this first line immune mechanism is spontaneously activated on all non-protected surfaces. Microbes are known to protect themselves by binding host complement regulators from plasma or other body fluids: FH for protection against the alternative pathway activation and C4b-binding protein for inhibition of the classical and lectin pathways. Binding of FH has been thought to be simple recruitment of host FH onto the microbial surface since FH acts as a cofactor for factor I in the degradation of the central complement component C3b [30]. This inactivation is essential for microbial survival in nonimmune plasma or blood, since it prevents opsonophagocytosis and microbial lysis by the membrane attack complexes [31]. Microbes recruit host FH by binding it via two separate sites, one within the domains 6–7 and the other in the C-terminal FH19-20 (Table S1), but the reason for using these sites has remained unexplained.
Our new data show, first, that the microbes we studied not only use FH19-20, but in particular the same area on FH domain 20, which we have named the “common microbial binding site” (Figure 3, panel B). Second, our data show that binding via this particular site allows the formation of a tripartite microbial protein∶FH∶C3b complex (Figure 5, panel C). Third, and most importantly, our data show that formation of the tripartite complex enhances FH-mediated inactivation of C3b. This explains why many kinds of microbes have evolved to utilize this common microbial binding site on FH.
We analyzed the interaction site between the carboxyl-terminus of FH and microbes by measuring the effect of mutant FH19-20 proteins on binding of wt FH19-20 to five important human pathogens (Gram-negative and Gram-positive bacteria and a yeast). Next we analyzed FH19-20 binding by three structurally non-related, FH binding proteins, two from spirochetes, OspE from B. burgdorferi sensu stricto [18] and FhbA from B. hermsii [32], and Tuf from P. aeruginosa [21]. To our great surprise all the microbes and microbial proteins studied bound FH via heavily overlapping binding sites on domain 20 (Table 1, Figure 3, panel A). We found three key amino acids (R1182, R1203, R1206) that affected binding to all the studied microbes and three more (K1188A, R1210A, R1215A) that affected binding to at least three out of seven microbes analyzed. We believe that this site, the common microbial binding site, will be found to be used by many other pathogenic microbes too. We did not use full length FH with point mutations in these experiments since microbes have often two binding sites for FH (Table 1) and expression and purification of full-length FH with mutations in both the microbial binding sites might not result in easily interpretable results.
Since the different microbial proteins are non-homologous it is expected that they use slightly different residues within or next to the common microbial binding site on FH20 to form, for example, hydrogen bonds and hydrophobic contacts. An example of this is seen with OspE since mutations of two residues of FH19-20 (W1183 and E1198) that are not used by several other microbes had the most striking effect on OspE binding to FH19-20. Use of variable residues within the same area does not compromise the key finding that the used microbes share a common binding area on FH domain 20 but indicates variability in the structure of the microbial molecules binding to the common shared site on FH. It is obvious that only detailed structural analysis of different microbial FH-binding proteins in complex with FH19-20 will show how important each residue within or next to the common site is for the interaction.
At least three non-homologous microbial proteins and, in addition, four microbial species without known homologues of these proteins utilize the same site on FH20. For some of these microbes, it is not known which surface molecule recruits host FH and it is possible that, at least in some cases, the surface molecules are not proteins but carbohydrates. FH is known to bind to several negatively charged carbohydrates [33] and the common microbial binding site on FH20 overlaps with the site responsible for binding to at least one host carbohydrate, heparin (Figure 4, panels A–C) [27]. It remains to be studied if any microbe binds to the common microbial binding site on FH20 via a carbohydrate, and if carbohydrate binding to FH domain 20 could promote the FH∶C3b interaction through formation of a tripartite complex, similarly to the studied microbial proteins.
Why have different microbes evolved to utilize domain 20, and practically the same particular site on this domain, in recruitment of FH? Our results provide three reasons for this. First, our work shows that FH bound to microbial surface via domain 20 can also bind the C3d part of C3b by domain 19 (Figure 5, panels A and B). This brings FH near to its main target, C3b, and allows complement inhibition. On the basis of the superimposition of three structures, our recently solved structure of a microbial FH-binding protein (OspE) in complex with FH19-20 (Bhattacharjee et al, submitted) and the previously solved structures of FH19-20 in complex with C3d, [8] and of the C3b (containing the C3d [29]), it became clear that FH19-20 can bind simultaneously to a microbial protein and C3b (Figure 5, panel C). Furthermore, in this superimposition the microbial binding site is also directed towards the surface to which C3b is bound to via the thioester site and is therefore readily available for the microbial molecules in general. Second, the site on domain 20 is available under physiological conditions: the previously described physiologically important heparin binding site [27], [34], [35] and the common microbial site overlap to some extent (Figure 4, panels A–C). Ferreira and coworkers [36] have also emphasized that the cofactor site on domains 1–4 must not be disturbed upon binding of FH to a microbe and our common microbial binding site fulfills also this criterion. Third, utilization of this particular common microbial binding site provides more efficient down regulation of complement activation on the microbe. The intact α′-chain of C3b disappears more efficiently when FH and microbial proteins are present (Figure 6, Panels A and B). This is due to the carboxyl-terminus of FH, as practically no enhancement was seen by using FH1-4, which has cofactor activity but does not have the common microbial binding site on domain 20 (Figure 6, Panel C).
The tripartite microbial protein∶FH∶C3b/C3d complexes are clearly formed in fluid phase in vitro (Figure 5). It is, however, uneasy to demonstrate the complex formation on microbial surface since C3b is bound covalently to the molecules on the target surface and the formed tripartite complex is easily broken upon purification of C3b from the cells. To indicate the complex formation on cell surfaces we therefore used an experimental setup where binding of FH19-20 to whole intact microbes was analyzed in the presence of soluble C3d and saw that addition of C3d could, indeed, enhance the binding (Figure S4). Since the extrinsic C3d can enhance binding of FH onto the microbial surface it is highly likely that the microbial proteins could also enhance formation of the tripartite complex on the microbial surface, at least if the density of C3b depositions was high enough. The highest C3b concentration occurs on the target surface areas where alternative pathway activation is vigorously amplified via the feedback loop. It would be most beneficial for the microbe if the tripartite complexes were formed within those areas leading to maximal complement down regulation exactly at the spots where it is needed most. For the tripartite complex formation the microbial FH-binding molecule needs to be – or bend – next to the C3b molecule but most, if not all, of the FH-binding microbial proteins which have been structurally characterize are either long molecules (e.g. streptococcal M protein) or have a flexible tail that allows twisting and tilting (e.g. OspE, Bhattacharjee et al, submitted). Therefore at least some of the microbial FH-binding molecules seem to be able to operate on a broader area of the surface than just the exact spot they are attached to. The area where the tripartite complexes could be formed might in addition be expanded by lateral movement of the lipid tail or membrane anchor of the microbial FH-binding molecules on at least surface of Gram-negative bacilli.
An FH-related protein found in plasma, FHR-1, has a C-terminal domain that differs from domain 20 of FH only by two residues. The differences are located close to the microbial binding site and it remains to be studied if FHR-1 binds similarly to the used microbes, and if possible recruitment of FHR-1 is functionally beneficial or unfavourable for the microbes as FHR-1 does not have any cofactor-activity.
Clearly, formation of the tripartite complex is the reason for the increase in the regulatory function of FH caused by the microbial proteins. As far as we know, this kind of enhancement has neither been suggested, nor studied before. Instead, it has been suggested that microbes mimic host structures and thereby bind FH and other complement regulators [37]. Although microbes, heparin or endothelial cells do bind to overlapping sites on FH, this is not exactly molecular mimicry as the binding sites are not identical. The structures involved are completely different and they appear to differ from organism to organism. We and others have recently shown that host cells recruit FH via domain 20 [27], [35] and it remains to be studied if this leads to elevated FH function due to tripartite complex as in the microbial proteins [8], [28]. If this were the case, microbes utilizing the common microbial binding site on FH domain 20 would have functional, not molecular, mimicry of host cells. So far there is, however, no evidence of this.
The identified common microbial binding site on FH domain 20 represents a surprising type of host-pathogen contact – a single site on a host molecule utilized by several kinds of microbes in immune evasion. Such a common immune evasion site for both bacterial and eukaryotic pathogens has not been reported earlier. We call this kind of conserved site for microbial immune evasion a “superevasion site” and suggest that superevasion sites may occur on other powerful down regulators of host immunity, too. The concept of a microbial superevasion site is valid not only for down regulators of immunity, such as FH, but also for host immune activator molecules such as immunoglobulins. It is probable that, for example, staphylococcal protein A [38], streptococcal protein G [39], and E. coli protein EibD [40] are not the only microbial proteins that bind to a conserved site on IgG leading to prevention of the effector functions of immunoglobulins. This site on the Fc part of IgG is probably an example of a superevasion site on immune activator molecules.
In this study we have identified a conserved microbial binding site on domain 20 of the important complement regulator FH. We have shown that, by binding to the common binding site on FH, microbial proteins enhance the FH∶C3b interaction by enhancing their interaction, thereby increasing down regulation of C3b and leading to efficient evasion of complement attack and presumably to increased survival of the microbes in the host. The identified common microbial binding site on FH is the first example of a “superevasion site” pointing to new avenues not only in research on immune evasion by microbes but also in research aimed at novel vaccines and antimicrobial agents.
The outer surface proteins OspE and OspA from B. burgdorferi sensu stricto strain N40 were cloned, expressed and purified as described [18]. FhbA was cloned and purified from B. hermsii strain MAN [32], and Tuf from a P. aeruginosa blood isolate strain similarly as described earlier [20]. Cloning and purification of wt FH19-20 and the FH19-20 mutants have been described earlier [7], [27], [41]. Circular dichroism spectras of six mutants (R1182A, W1183L, K1188A, E1198A, R1203A, R1206A) were compared to wt to confirm proper folding of the mutants (Figure S3, panel A). The capacity of these mutants to form oligomers was compared to wt FH19-20 using gel filtration on a Superdex 75 10/300 GL column (Figure S3, panel B). FH1-4 was produced as described [42]. C3 and FH were purified from human plasma and C3b generated with trypsin as described [43]. C3d was a kind gift from Prof. D. Isenman, Univ. of Toronto, Canada. Factor I was purchased from Calbiochem/MerckMillipore (Merck, Darmstadt, Germany) and BSA, gelatin and heparin from Sigma-Aldrich (St. Louis, MO, US). The wt FH19-20, FH, OspE, and C3b were labeled with 125I using the IodoGen method (Thermo Scientific Pierce, Rockford, IL, US).
The strains of Pseudomonas aeruginosa, Haemophilus influenzae, Streptococcus pneumoniae, Staphylococcus aureus and Candida albicans we used were isolated from blood cultures of septic patients and were kind gifts of Dr. K. Haapasalo-Tuomainen, HUSLAB, Helsinki Univ. Central Hospital, and Univ. of Helsinki, Finland. Bordetella pertussis was a kind gift of Dr. Quishui He, Pertussis Reference Laboratory, Turku, Finland. The used serum sensitive Haemophilus influenzae strain is isolated from a throat swab of a healthy individual.
To detect binding of FH or FH19-20 to the microbes, the bacteria and yeast were first washed three times with PBS. Approximately 1×108 cells/reaction were incubated with radiolabeled FH or FH19-20 (40,000 cpm/reaction) in the absence or presence of C3d (0–100 µg/ml) in 50% PBS containing 0.1% gelatin (GPBS) at 37°C for 20 min with agitation (1,200 rpm). Cell-associated and free radioactive proteins were separated by centrifugation (10,000× g, 3 min) of the samples through 20% sucrose in GPBS. Radioactivities in the supernatant and pellet fractions were measured with a gammacounter (Wallac, Turku, Finland). The amounts of bound proteins were calculated as percentages of the total radioactivities in the corresponding pellets and supernatants. The experiments were performed three times in triplicate.
Nunc Polysorp BreakApart plates (Thermo Scientific, Rockford, IL, US) were coated with either bacteria (1×106/well in phosphate-buffered saline, PBS, at 37°C for 12 hours) or proteins (5–25 µg/ml in PBS at 4°C for 12 hours). The wells were blocked (0.5% BSA/PBS, 60 min at 22°C, or 0.5% BSA/50% PBS for the experiment shown in the Figure 4, Panel C) and washed with PBS. Serial dilutions of proteins were mixed with 125I-FH19-20 or 125I-OspE (50,000 cpm/well) in a separate 96-well microtitre plate (Greiner Bio One, Frickenhausen, Germany) before transferring into the coated wells. After incubation (37°C, 60 min) and washing with PBS (or 50% PBS for the experiment shown in Figure 4, Panel C), the radioactivity in each well was measured with a gamma-counter (Wallac, Turku, Finland). The inhibition curves were fitted using non-linear regression of a “log(inhibitor) vs. response” model using GraphPad Prism software (version 5.0b, GraphPad Software, CA, US). The mean inhibitory concentrations (IC50-values) were calculated from the fitted curves. All the assays were performed three times using triplicate wells.
To measure cofactor activity 125I-C3b (100,000 cpm/assay) was mixed with factor I (16 µg/ml) in the absence or presence of FH or FH1-4 (8–85 µg/ml) and OspE, FhbA, and Tuf (50 µg/ml). Mixtures were incubated at 37°C for 5 min and, after adding β-mercaptoethanol, the samples were heated (3 min at 93°C) and run on 10% SDS-PAGE gels. The gels were subjected to autoradiography and cofactor activity was evaluated as the intensity of the C3b α′-chain measured with GelEval-programme (FrogDance Software, Dundee, UK).
Values are expressed as means ± SD. All statistical analyses were performed using GraphPad Prism software and statistical differences were calculated with unpaired t-tests.
|
10.1371/journal.pgen.0040012 | Dominant-Negative CK2α Induces Potent Effects on Circadian Rhythmicity | Circadian clocks organize the precise timing of cellular and behavioral events. In Drosophila, circadian clocks consist of negative feedback loops in which the clock component PERIOD (PER) represses its own transcription. PER phosphorylation is a critical step in timing the onset and termination of this feedback. The protein kinase CK2 has been linked to circadian timing, but the importance of this contribution is unclear; it is not certain where and when CK2 acts to regulate circadian rhythms. To determine its temporal and spatial functions, a dominant negative mutant of the catalytic alpha subunit, CK2αTik, was targeted to circadian neurons. Behaviorally, CK2αTik induces severe period lengthening (∼33 h), greater than nearly all known circadian mutant alleles, and abolishes detectable free-running behavioral rhythmicity at high levels of expression. CK2αTik, when targeted to a subset of pacemaker neurons, generates period splitting, resulting in flies exhibiting both long and near 24-h periods. These behavioral effects are evident even when CK2αTik expression is induced only during adulthood, implicating an acute role for CK2α function in circadian rhythms. CK2αTik expression results in reduced PER phosphorylation, delayed nuclear entry, and dampened cycling with elevated trough levels of PER. Heightened trough levels of per transcript accompany increased protein levels, suggesting that CK2αTik disturbs negative feedback of PER on its own transcription. Taken together, these in vivo data implicate a central role of CK2α function in timing PER negative feedback in adult circadian neurons.
| The molecular mechanism that governs organization of physiology and behavior into 24-h rhythms is a conserved transcriptional feedback process that is strikingly similar across distinct phyla. Notably, cyclic phosphorylation of negative feedback regulators is critical to time molecular rhythms. Indeed, mutation of a putative phosphoacceptor site in the human PERIOD2 gene, a key negative regulator, is associated with Advanced Sleep Phase Syndrome. This study reveals a critical role for the protein kinase CK2 for setting the period of behavioral and molecular oscillations in Drosophila. Circadian phenotypes due to CK2 disruption are due to a direct requirement in adult circadian pacemakers. These findings further demonstrate that CK2 modification of the negative feedback regulator PERIOD alters its cyclical phosphorylation, protein abundance, nuclear translocation, and transcriptional repression activity. These studies place CK2 as a central kinase in circadian timing.
| Circadian rhythms that orchestrate daily fluctuations in biochemistry, physiology and behavior are observed across distinct phylogenetic kingdoms. Underscoring the evolutionary importance of these clocks, the molecular processes that drive circadian rhythms are also highly conserved. At the core of the circadian pathway is a transcriptional feedback loop. In Drosophila melanogaster, CLOCK (CLK) and CYCLE (CYC) activate expression of target genes such as period (per) and timeless (tim) [1–3]. PER and TIM ultimately translocate to the nucleus and inhibit CLK/CYC transcription [3–6]. Notably, the overall architecture of this feedback loop, as well as some of the molecular players, are observed in organisms as diverse as fungi, plants, and mammals [3].
In addition to transcriptional influence in circadian rhythms, posttranslational modification, particularly for PER, has been shown to play a critical role in normal and disordered circadian timing [7–10]. The most well studied kinase CK1/DOUBLETIME (DBT) is hypothesized to regulate PER nuclear entry, repression, and degradation [7,11,12]. A second enzyme, glycogen synthase kinase (GSK3β)/SHAGGY (SGG), regulates phosphorylation of TIM protein, levels, and nuclear entry [13]. These rhythmic phosphorylation cycles also necessarily include the activity of a phosphatase, and protein phosphatase 2A (PP2A) has been implicated in the rhythmic dephosphorylation of PER [14], while protein phosphatase 1 (PP1) has been implicated in the dephosphorylation of both TIM and PER [15].
Our laboratory has been investigating the function of the protein kinase CK2 in circadian clock function [16,17]. The CK2 holoenzyme is a heterotetramer consisting of two alpha catalytic and two beta regulatory subunits [18,19]. Mutant CK2α and CK2β alleles result in period lengthening phenotypes (<3 h long), consistent with their proposed clock role [16,20]. The manner in which CK2 is important for setting circadian period remains unclear. CK2 also functions in various developmental processes [18], consistent with the pre-adult lethality observed in CK2α and CK2β mutants [16,21]. This developmental function raises the question that CK2 phenotypes may derive from its activity during maturation rather than in adults. While both CK2 subunits are expressed in pacemaker neurons [16,20], it is uncertain if CK2 functions in these neurons to regulate circadian rhythms. RNAi studies in S2 cells suggest that the role of CK2 phosphorylation is to promote transcriptional repression by PER [22]; however, it is not clear if this is true in vivo.
To better address these questions, we expressed a dominant negative CK2α Timekeeper (Tik) mutant [16] in a spatially and temporally controlled manner and queried the effects on behavior, PER protein levels, phosphorylation, repression, and nuclear entry in core pacemaker neurons of adult D. melanogaster. Taken together, these findings reveal remarkably potent effects of manipulating CK2 activity in adult circadian neurons and uncover a role consistent with the regulation of PER nuclear localization and feedback repression.
Prior studies implicate CK2 in the control of circadian function in Drosophila, Arabidopsis, and Neurospora [16,20,23,24]. Testing of the strongest homozygous mutants alleles is limited by developmental lethality [16,21]. More modest period phenotypes raised questions as to the functional significance of CK2 action in the circadian clocks. To determine the consequences of suppressing CK2 activity, we used the GAL4/UAS system to drive expression of CK2α bearing the dominant Tik mutation (CK2αTik) [25]. The CK2αTik allele contains two missense mutations, one of which introduces a charged residue into the putative hydrophobic binding pocket for the phosphodonor nucleotide [16,19]. In vitro analysis indicates that these mutations eliminate most catalytic activity [26]. The molecular lesion, the loss of biochemical activity and the dominant behavioral phenotype suggest that Tik encodes a dominant negative form of CK2α.
To examine the behavioral consequences of CK2αTik expression, we crossed flies bearing UAS-driven CK2αTik (UASTik) with timGal4–62 driver flies [27] and assayed circadian behavior in the progeny (timGal4/+; UASTikT1/+, “timTik”). The Drosophila circadian network consists of six bilateral groups of cells: large and small ventral lateral neurons (lg- and sm- LNv), dorsal lateral neurons (LNd), and three clusters of dorsal neurons (DN1–3) [28]. The tim promoter induces GAL4 expression in all of these key neuronal clusters that coordinate circadian behavior [29]. To our surprise, these timTik flies display extraordinarily long periods averaging ∼33 h relative to control periods of ∼24 h (Figure 1, compare Figure 1A and Figure 1B; Table 1). Moreover, the influence on period is dose-dependent; by increasing Gal4 dosage in timTik flies with a second circadian driver, cry16Gal4 [30], the period is further lengthened to ∼37 h (Table 1). Confirming the circadian specificity of this result, expression of UASTik only in photoreceptor neurons with the GMRGal4 driver [31] does not result in period lengthening (data not shown). Heterozygous Tik/+ mutant flies display periods 2–3 h longer than wild-type controls with a reduction of ∼50% in CK2 activity [16]. The magnitude of the period effects strongly argues that CK2 activity is more gravely inhibited in timTik flies. The fact that the magnitude of period effects exceeds that of nearly all circadian mutant alleles suggests that CK2 activity is critically important for setting circadian period.
By increasing dosage of the dominant allele with double copies of both the broad circadian timGal4 driver and the UASTik transgene (timGal4; UASTikT1, “timTik2x”), rhythmicity is undetectable in constant darkness (Figure 1C; Table 1). The above results suggest that CK2α function in central pacemaker neurons is essential for wild type behavioral rhythms. Thus, CK2α and DBT appear to be the only core circadian kinases demonstrated to be obligatory for free-running behavioral rhythms [11,32]. Mutations in a catalytic subunit of the cAMP-dependent protein kinase (PKA) also result in behavioral arrhythmicity [33]; however, as this lesion leaves core molecular cycling of the clock intact, it is likely to function in an output capacity.
The neuropeptide Pigment-Dispersing Factor (PDF) mediates transmission of timing information from core LNv pacemaker neurons to downstream neural circuits [34]. The CK2α and β subunits are strongly expressed in the pacemaker LNv [16,20]. To test the hypothesis that CK2α functions in pacemaker neurons, CK2αTik was induced in the LNv using a pdfGal4 driver [34]. Similar to timTik flies, CK2αTik expression in PDF+ neurons (pdfGal4/+; UASTikT1/+, “pdfTik”) also results in dramatically long periods (∼32 h; Figure 1D; Table 1). Again, these effects are dose-dependent, as adding an additional Gal4 driver, cry16Gal4, increased the period length to ∼37 h (Table 1). We previously identified a spontaneous revertant allele, TikR, which deletes a portion of the Tik coding region, largely reverts the dominant circadian phenotype but still lacks catalytic activity, consistent with its characterization as a recessive loss-of-function allele [16]. Supporting this hypothesis, pdfGal4 expression of independent UASTikR lines had no significant effect on circadian rhythms (Table 1). These results confirm the hypothesis that the Tik mutation acts as a dominant-negative to inhibit function of endogenous wild type CK2α.
PDF neurons communicate with and reset the clocks in non-PDF pacemaker neurons to synchronize different clusters in the network [35]. The CK2αTik period effects were blocked in a pdf null [34] background or by coexpressing an inwardly rectifying potassium channel that hyperpolarizes the LNv (UASKir2.1, [36], Table 1), indicating that CK2αTik period effects are transmitted by LNv activity and PDF output. These manipulations alone (pdf01 null mutants or expression of UASKIR with pdfGal4) result in short, weak periods (Figure S1; Table 1) ([34,36]). These data provide functional evidence that CK2α operates in pacemaker LNv to regulate circadian period, consistent with published expression data.
While pdfTik flies show a long period phenotype, we also noted variability in the period measurement and reduction of the strength of the rhythm in these flies (Table 1). It was hypothesized that “wild-type” non-PDF clock neurons were unable to entrain to the long period program in PDF+ cells, and were expressing a secondary rhythm. To see if flies were exhibiting more than one period, we performed periodogram analysis using the Lomb-Scargle method [37,38]. This approach eliminates misidentification of periods that are simply multiples of a true period. This analysis reveals that approximately 45% of pdfTik animals display two significant periods (Figures 1E and 2A). The dominant period is 35.3 (+/− 0.3) h while a secondary peak indicates an average period of 23.2 (+/− 0.1) h (Figures 2B and S2). When UASTik is expressed as a heterozygote with the broader expressing timGal4 driver, reduced rhythm strength and splitting is not detectable (Figure 2A; Table 1), suggesting that hyper-elongating period only in PDF-positive LNv causes uncoupling of clock cell groups. We propose that non-PDF neurons are unable to maintain synchrony with PDF clocks with extreme periods. To our knowledge, this is the first example of complex rhythmicity due to altering period length in a subset of pacemaker neurons.
Given that CK2 acts in multiple pathways throughout the life cycle of the fly, we queried if the CK2αTik phenotype is due to developmental/compensatory effects or whether loss of CK2α function during adulthood would still result in lengthening of period. In order to address this concern, we utilized a temperature sensitive Gal80 inhibitor of Gal4 expressed in all cells under the tubulin promoter (tubGal80ts, [39]). This conditional, temporal and regional gene expression targeting (TARGET) system has been previously used to examine the genetic basis of complex behaviors such as memory in Drosophila [39]. tubGal80ts represses GAL4 at the permissive temperature of 18 °C, but is inactivated and fails to repress GAL4 at the restrictive 29 °C temperature. We generated pdfGal4/tubGal80ts; UASTikT1/+ flies and raised them at the permissive temperature (18 °C) to prevent expression of UASTik so that CK2α function would be largely intact during development. Flies were then tested at either permissive (18 °C) or restrictive (29 °C) temperatures and period was calculated during constant conditions. A cardinal feature of circadian clocks is their temperature compensation, i.e., period is roughly invariant over a broad temperature range [40,41]. Consistent with this idea, the control strain here (tubGal80ts/+; UASTikT1/+) shows little period change between 18 °C and 29 °C (Figure 3A and 3B, top panels). Constitutively inhibiting CK2α in pdfTik flies again demonstrates the severe period lengthening effect at both temperatures (Figure 3A and 3B, middle panels); splitting of rhythms in these flies is observed at levels similar to those described above, but only at 29 °C (unpublished data). Interestingly, when dominant-negative UASTik is selectively activated at 29 °C during testing of adult flies, the extreme long period phenotype (>30 h) is still manifested (Figure 3A and 3B, bottom panels). Slight period lengthening is detectable at 18 °C; however, this effect is much smaller than observed at 29 °C, and is likely due to incomplete Gal80 inhibition of Gal4. Additionally, two split periods are again observed at 29 °C in conditionally inhibited circadian CK2α flies, but not at 18 °C (data not shown). These results indicate that CK2α plays a direct role in adult circadian rhythms, and its loss of function in Tik and UASTik animals is not likely due to some developmental artifact. Consistent with this idea, inspection of LNv structure and PDF labeling in UASTik-expressing brains reveals no gross abnormalities of circadian pacemaker anatomy (unpublished data). To our knowledge, this is one of the few temporal investigations of clock gene function demonstrating an acute role of a circadian gene during adulthood [42,43].
To determine the effects of CK2α loss of function on core molecular clock rhythms, we tested whether expression of UASTik in PDF-positive LNv altered cycling of the core clock protein PER. Levels and cellular distribution of PER protein in smLNv were examined quantitatively on the first day of DD in pdfTik or control Gal4 flies. Although we do observe splitting in these flies, behavior remains largely synchronous on the first day of DD (Figure 4A). Control flies show the typical evening peak of activity at ∼CT12 while the long-period pdfTik flies have a delayed evening activity peak, regardless of whether they exhibit split periods or not (Figure 4A, pdfTikL v. pdfTikS). Measurements of pixel intensity indirectly report the amount of PER protein in smLNv [44]; as seen in Figure 4B and 4C, PER levels are elevated in smLNv of pdfTik during the subjective day relative to controls. Wild type PER levels wane from CT4–8 and begin accumulating again in the subjective evening (CT16–20); in contrast, a prolonged decline in PER throughout the day (CT4–12) is evident in pdfTik flies, and levels only disappear during subjective evening (CT12–20), consistent with a long period phenotype. Peak and trough PER levels are also elevated in pdfTik flies relative to controls (p < 0.001 comparing pdfGal4/+ CT0 to pdfTik CT4 for peak and pdfGal4/+ CT12 to pdfTik CT16 for trough, Figure 4C). PER typically transitions from the cytoplasm to a predominantly nuclear distribution during the middle of the night, and such a pattern is observed in pdfGal4/+ control flies (Figure 4B and 4D). However, the amplitude of the localization rhythm (as quantified by the nuclear:cytoplasmic ratio) is seriously reduced in pdfTik flies (Figure 4D, p < 0.001 at CT0, CT4, CT8, and CT20 pdfGal4/+ v. pdfTik). The timing of nuclear localization is also delayed in pdfTik flies; while PER never becomes predominantly nuclear, the time at which the most PER is localized to the nucleus occurs later from CT4–12 in pdfTik smLNv, rather than CT0–4 for the GAL4 control (Figure 4D and 4E). This finding is supported by analysis of nuclear PER levels in pdfGal4/+ and pdfTik smLNv. Nuclear PER levels accumulate to a similar degree in pdfTik as in the Gal4 control; however, nuclear levels do not rise until later in the subjective day relative to control (Figure 4E). The overall fraction of nuclear PER is lower (Figure 4D), as more of the PER protein in pdfTik neurons remains sequestered in the cytoplasm (Figure 4F). Indeed, the reduced nuclear PER levels in the face of elevated cytoplasmic PER levels at CT0 provide the most compelling evidence that CK2α is important for nuclear PER localization independent of regulating its cytoplasmic abundance. These results are consistent with prior reports that reduction of CK2 activity inhibits nuclear entry [16].
To quantitatively examine the effect of CK2αTik on PER cycling and phosphorylation, we used western blots of whole head extracts on the first day of DD. The far majority of PER in whole heads is expressed in the eye [45]. To examine CK2αTik effects we used the timGAL4 driver that includes strong expression in the eye. In wild-type flies, PER phosphorylation (evident as reduced mobility) peaks in the early subjective morning (CT1, Figure 5Aa). PER levels are subsequently reduced, presumably reflecting phosphorylation-induced degradation. PER begins to appear early in the subjective night (CT13, Figure 5Aa), and levels accumulate during the night as PER becomes progressively more phosphorylated. In Tik/+ and timTik flies, both level and mobility rhythms are delayed, consistent with a lengthened period in these flies (Figure 5Ac, 5Ad, and 5B), while expression of UASTikR was not detectably different than wild type (Figure 5Ab).
We then examined flies with two copies of the timGAL4 and UASTik transgenes (timTik2x). These flies did not exhibit any significant behavioral rhythms (Table 1). Severe reduction of CK2α activity in homozygous timTik2x flies exacerbates PER metabolism during constant conditions, causing constitutive elevations in PER protein and minimizing the amplitude of PER cycling (Figure 5Ae and 5B), consistent with reduced behavioral rhythmicity. These effects are most evident at wild type trough times for PER (p < 0.01, significant effect of genotype at CT9). In addition, PER fails to achieve a hyperphosphorylated state in timTik2x flies (Figure 5Ae). This consequence is most evident at CT1, time of peak phosphorylation in wild type. These findings support the notion that CK2α ensures the proper timing of PER cycling and function. While we cannot exclude the possibility that CK2 indirectly regulates the post-translational modification of PER, the strong effects on PER mobility in CK2 loss-of-function flies argue that PER is an in vivo CK2 substrate, consistent with previous studies [17].
Previous studies have implicated CK2 in promoting PER repression of CLK activation [22]. However, these studies were performed in cultured Drosophila S2 cells which do not harbor functioning circadian clocks. To test the hypothesis that CK2 promotes PER repression in vivo, we examined circadian transcription in UASTik expressing flies. Levels of two CLK-activated transcripts, per and vrille (vri) [46] were analyzed using quantitative real-time reverse-transcriptase polymerase chain reaction (qRT-PCR). We hypothesized that if negative feedback is unaffected in UASTik expressing flies, then elevated PER levels would strongly repress CLK, reducing per and vri transcript levels. If negative feedback is disrupted, then elevated PER levels would fail to appropriately repress per and vri transcription. Expression of dominant-negative UASTik in circadian neurons in timTik flies postpones the decline in per transcript until early subjective night (Figure 6A), consistent with the effect of CK2α loss of function in the heterozygous Tik/+ mutant. Whereas wild type per transcription peaks around CT9–13, per levels do not achieve maximum until CT13–17 in timTik flies, and a similar pattern emerges from analysis of the vri transcript (Figure 6B).
The most informative result becomes apparent in timTik2x flies. Further reductions in CK2α activity in timTik2x result in per and vri transcript levels with a severely reduced amplitude rhythm (Figure 6A and 6B). Importantly, per and vri never reach wild-type trough levels (per: p < 0.001 for y w at CT1 relative to timTik and timTik2x at CT5 and p < 0.01 for vri at the same time points), consistent with the hypothesis that elevated PER protein levels are unable to fully repress CLK target genes in UASTik-expressing flies. Taken together, the magnitude of the observed effects suggests that CK2 not only promotes PER repression activity in vivo, but that it has a sizable impact on transcriptional repression.
The role of posttranslational modification in regulating precise circadian timing is well established [9], and indeed may be principally responsible for molecular cycling [8]. CK2 has been implicated in regulating circadian rhythms, PER modification and metabolism [16,17]. The present study sought to determine if CK2α activity is required in adult core pacemaker neurons for molecular and behavioral rhythmicity. Broad spatial expression of UASTik in pacemaker neurons with the timGal4 driver causes severe lengthening of circadian period to ∼33 h, a degree even greater than that of the heterozygous Tik mutant. Radical reductions in CK2α activity by increasing copy number of the transgenes in timTik2x flies ultimately result in behavioral arrhythmicity, demonstrating that CK2α is an obligatory component of circadian rhythms. Previous work demonstrated that overexpression of wild type CK2 mildly lengthens period [17]; taken together, these data indicate that period is highly sensitive to CK2 activity. Expression of UASTik in PDF+ LNv is also sufficient to lengthen period; indeed, the effect requires LNv activity and output of the PDF neuropeptide. That the period length is not exacerbated by additional clock neuron expression in timTik versus pdfTik flies implies that the phenotype originates largely from the LNv; however, the possibility that CK2α additionally functions in other circadian cells cannot be excluded. Given that splitting is eliminated when the genetic programs of both LNv and downstream circadian neurons are identical with respect to UASTik expression, the data imply that this manipulation affects CK2 function in other clock cells. Indeed, if CK2 is a true component of the core transcriptional pacemaker, it would be expected to regulate feedback in all cells that have a functional molecular clock. As CK2 functions to dictate period in LNv cells during constant conditions, it is possible that CK2 activity may also regulate morning behavior, as these cells drive morning activity, while downstream neurons dictate evening activity [47,48]. However, as CK2 may also function in cells responsible for evening behavior, some balance of CK2 between LNv and non-LNv neurons could favor a strong morning or evening activity phase. The contribution of CK2 activity to morning and evening behavior is currently under investigation.
Further evidence that CK2α activity is important in LNv derives from the finding that inhibition by CK2αTik causes delayed nuclear entry of PER in these core pacemaker cells. An unanticipated consequence of pdfTik expression is splitting of the behavioral rhythm into long (∼35 h) and short (∼23 h) components. All of the above behavioral effects are due to acute CK2α activity as adult-specific inhibition is able to induce the rhythm phenotypes. At the molecular level, elevated levels and diminished phosphorylation of PER protein is associated with reduced CK2α function; this effect on PER protein is further correlated with elevated and delayed transcription of per and vri clock genes.
The severity of the observed behavioral phenotype places CK2 as a critical regulator of circadian rhythms. Of known circadian kinases, only mutants of doubletime (dbt) and PKA are also capable of completely eliminating rhythmicity as is observed in timTik2x flies [32,33]. As the core molecular clock is unperturbed by PKA mutations, this kinase is proposed to function in circadian locomotor output [32,33], leaving DBT and CK2 as the only critical core circadian kinases. Originally, DBT was found to regulate PER stability and electrophoretic mobility; this initial study concluded that DBT-mediated phosphorylation led to PER degradation [11,49]. Subsequent studies suggested that DBT may retard the ability of PER to enter the nucleus and repress transcription [12,22,50]. Many other gene mutations that result in arrhythmicity affect either input or output of the circadian system. As the core molecular feedback loop is disrupted in UASTik-expressing flies, CK2 appears to regulate timing of the core clock and shows phenotypes similar to mutants of other core circadian genes such as per, tim, and Clk. However, the magnitude of the period phenotype in both pdfTik and timTik flies is greater than nearly all circadian mutants. The only other alleles which produce a similar degree of period lengthening include the timUL mutation [51] and a novel dominant-negative kinase dead dbt allele whose expression also results in period lengthening or arrythmicity [52].
We present numerous pieces of evidence to support the hypothesis that CK2α acutely functions in the PDF+ LNv neurons. The long period phenotype observed when CK2αTik is expressed in PDF-positive LNv is associated with splitting of the behavioral rhythm into two components: a predominant, long, ∼35 h period and a weak shorter period of approximately 23 h. The splitting is reflected in the low strength of behavioral rhythms observed in pdfTik flies. Splitting was originally observed in Syrian hamsters maintained under constant light; this finding was the foundation for a two-oscillator model whose coordinated output is manifested as an overt circadian rhythm [53]. It has similarly been shown that non-conventional entrainment conditions can induce multi-period splitting and desynchronization of circadian neurons in mammals [54,55]. Early reports indicated splitting of the Drosophila circadian period in sine oculis mutants that have disrupted optic development, suggesting that dual periods may arise from entrainment through different input pathways [56]. Similarly, both wild type flies under low light and mutants of cryptochrome, the major circadian photoreceptor, exhibit split rhythms under constant light [57,58]. These periods include short (∼22 h) and long (∼25 h) components that alternatively decrease or increase with light intensity, respectively, again implicating variation of the oscillator system input pathway. Ectopic misexpression of the PDF output neuropeptide induced multiple periods during DD (of ∼22 h and ∼25 h) [59]. Nitabach et al. [38] further identified complex rhythmicity by activating LNv neurons; at least two periods of ∼22 or ∼25–26 h lengths are observed (with an occasional 3rd, shorter ∼20–21 h peak). Elevated PDF levels and desynchronization of circadian neurons are detected in these flies [38]. Both of the above cases suggest that split periods arise from misregulation of neuronal output from the core pacemaker neurons. The result presented here is the first demonstration of splitting as a consequence of altering a core clock component.
It is hypothesized that driving the oscillator period to such an extreme only in the LNv uncouples them from non-UASTik expressing, PDF-negative “wild type” circadian neurons (i.e., LNd/DNs) that then contribute the shorter, weaker behavioral rhythm (∼23 h), such as that seen in pdf01 mutants [34]. This notion is further supported by the behavior of timTik flies that express UASTik in all circadian neurons; in this case, when the genetic programs of all clock cells are identical, no such splitting is observed and rhythm strength returns to normal levels. These results begin to examine the limits of entrainment of one oscillator by a coupled oscillator in a circadian pacemaker network.
CK2 has a number of roles in cellular biology [18]. It is required at multiple transitions during the cell cycle including mitosis and functions to regulate caspase-mediated apoptosis and cell survival [18]. Developmentally, CK2 regulates proliferation and cell fate decisions [25,60]. Not surprisingly, it is an essential gene, as homozygous Tik mutants are not viable as adults [16]. We were able to utilize the TARGET system [39] to conditionally induce dominant-negative CK2α in adult flies. Interestingly, when CK2α activity was inhibited in LNv solely during adulthood, the behavioral phenotypes are still manifested. Thus, an acute CK2α loss of function impacts rhythmicity in the adult circadian system, presenting it as a critical and direct regulator of the circadian clock. While it has been shown that such adult-specific rescue of per is able to restore rhythmicity in Drosophila [43], it will be important to investigate the life-stage properties of other circadian genes; for example, Clk is also known to have developmental roles [61].
Thus, CK2α is an acute, direct, and essential component of circadian rhythms; we propose that CK2 regulates the core oscillator by phosphorylating PER to promote nuclear entry and repression. There is abundant evidence for PER as a bona fide CK2α substrate. CK2α can phosphorylate PER in vitro at specific predicted CK2α sites [16,17]; moreover, mutation of these CK2α target residues causes period phenotypes similar to the Tik mutation when expressed in vivo, demonstrating the functional relevance of this modification on PER activity [17]. Finally, we present here the clear defects in PER mobility observed with a deficiency in CK2α activity. Expressing UASTik singly or in double dosage with the timGal4 driver results in increased, hypophosphorylated PER. Indeed, the amplitude of PER cycling appears completely diminished when CK2α is severely inhibited in timTik2x flies. Again, the molecular PER phenotype mirrors the behavioral effect of these manipulations; timTik2x flies are arrhythmic under constant conditions, supporting the idea that CK2α activity is critical for the maintenance of a molecular and behavioral clock.
A further consequence of CK2α loss of function in core pacemaker neurons is a pattern of delayed PER decline, consistent with the long period phenotype observed in these flies. Despite increases in overall and cytoplasmic PER levels, nuclear PER levels are lower relative to wild type during the early subjective day, providing further evidence that nuclear translocation is not strictly driven by protein accumulation [50,62]. The dampened and delayed nuclear entry of PER protein of CK2αTik-expressing smLNv provides support that CK2α normally functions to promote PER nuclear translocation. A second possibility is that the high levels of PER protein saturate the nuclear entry pathway, preventing the majority of PER from localizing to the nucleus in pdfTik flies. Yet, the delay in nuclear accumulation is consistent with the hypothesis that CK2α activity typically functions to permit timely PER nuclear entry.
Previous evidence indicated that knock-down of CK2 levels in cultured Drosophila S2 cells limits the ability of PER to repress a Clk-driven luciferase reporter [22]. It is critical to validate such studies in vivo to determine the true function of the kinase in the circadian system. While one may expect that the increased levels of PER associated with CK2αTik expression (particularly at trough time points) would lead to enhanced clock gene repression, we do not see such an effect. Conversely, CK2α inhibition results in delayed per and vri transcription, and elevated trough transcript levels, confirming that CK2α normally operates to promote repression of clock gene transcription. The features of CK2α function are both in opposition with and complementary to those put forth for the DBT kinase. DBT is thought to retard PER nuclear entry [12] and signal its degradation [11]; in contrast, CK2α appears to promote nuclear entry of PER (and hence repression), but may also influence its turnover.
We have outlined a model of the way in which CK2 promotes repression of circadian transcription developed from existing and currently presented data. We speculate that effects of CK2 on PER nuclear localization may operate through the proposed interval timer described in S2 cells. Based on the interval timer model, PER and TIM heterodimerize in the cytoplasm in a time-insensitive manner [63]; after some lag or upon some signal, they dissociate and enter the nucleus independently [44,63] where PER mediates transcriptional repression. The role of nuclear TIM is yet unclear. Recent work indicates that repression is not achieved merely by physical association of PER with CLK, but perhaps by PER acting as a scaffold to bridge CLK and DBT [64]. It is hypothesized that phosphorylation of CLK by DBT diminishes its transactivating capabilities [64], similar to the model proposed in Neurospora [23]. Nawathean et al. conclude that the ability of PER to repress transcription is primarily a function of its nuclear localization, which is, in turn, dependent on phosphorylation [65]; however, they also acknowledge that phosphorylation may secondarily modulate the intrinsic ability of PER to enact repression. While PER is predominantly cytoplasmic in S2 cells (e.g., [65]), altering its subcellular localization by adding a nuclear localization signal or blocking nuclear export increases transcriptional repression; however, this activity is reduced in mutants that lack a critical DBT binding site shown to be important for PER phosphorylation [64,65]. Thus, the data from S2 cells does not resolve whether activity primarily regulates PER nuclear entry or repression, but provides evidence for both functions. The conflicting S2 cell results, particularly for DBT analysis, support the use of our in vivo approach to determine the role of kinase modification on molecular cycling.
We propose that the effects of CK2 loss of function on feedback repression are due, in part, to the inability of PER to properly translocate to the nucleus. Phosphorylation of the PER:TIM heterodimer in the cytoplasm by CK2 may act as the interval-timer signal [63] to dissociate this complex and/or facilitate nuclear entry. Indeed, with reduced CK2α function, PER nuclear accumulation is delayed, and correlates with delayed repression of circadian transcription. DBT functions to induce degradation of free cytoplasmic PER, as well as hyperphosphorylated nuclear PER. We propose that increased and lingering levels of PER in UASTik-expressing brains is due to an inability of CK2 to signal dissociation of the PER:TIM heterodimer. Persistence of PER in this complex and its failure to independently enter the nucleus protects it from DBT-mediated phosphorylation. The lack of DBT phosphorylation would reduce PER degradation and repressor activity, leading to increased cytoplasmic PER and altered feedback repression. Intrinsic PER repressor function does not appear to be greatly compromised by CK2 loss of function, as liberation of PER after light-induced TIM degradation in timTik2x flies results in robust suppression of per RNA (R. Meissner, J. Lin, unpublished observations). As nuclear entry is a critical step in feedback repression, CK2 function is important for the maintenance of a functional molecular and behavioral clock. Alternatively, CK2 may promote PER-TIM dimerization and subsequent nuclear entry. CK2 may facilitate PER's interaction with CLK and thus enhance repression. Lastly, CK2 could function to augment DBT phosphorylation of PER either by modulating DBT activity or by providing a phosphorylated substrate for recognition by DBT, thus controlling PER abundance.
Conditional spatio-temporal expression of this dominant-negative CK2α mutation is a useful tool for in vivo exploration of the myriad roles of CK2 in all aspects of biology. The UASTik transgene has already been used to dissect features of CK2 function during Drosophila eye development [25]. Here, the use of such a strong CK2α allele permits dissection of the molecular, genetic, and neuroanatomical clock. Ultimately, these studies provide a model in which CK2α activity in the core adult pacemaker is critical for the proper timing of the transcriptional feedback loop. The critical role of CK2α in pacemaker function highlights the importance of a concerted post-translational modification scheme to regulate cycling of core clock components in order to manifest precise circadian rhythms at the molecular and behavioral level. Indeed, mutation of a phosphorylation site in the human per2 gene is responsible for Familial Advanced Sleep Phase Syndrome [10]. As CK2 is highly conserved across kingdoms and has been shown to function in the circadian pathway of other species [23,24], it is a priority to investigate its role in mammalian circadian rhythms.
For generation of UASTik lines, we cut the BglII and XhoI fragment containing Tik from pET-Tik [16], and cloned it into pUAST to create transgenic flies. All transgenic DNA constructs were sequenced (Applied Biosystems, Foster City, CA) and injected in y w embryos at CBRC Transgenic Drosophila Core (Charlestown, MA). Transformant lines with inserts on the second or third chromosome were balanced with y w;Bl/Cy0 or y w;;TM2/TM6B, respectively. Distinct UASTik insertions are denoted with S1 or T1 superscripts for second or third chromosome inserts, respectively. The revertant TikR allele was similarly cloned, and UASTikR inserts are also designated with the superscripts T2 or T3 indicating two discrete third chromosome insertions.
For additional fly lines, the pdfGal4 and timGal4–62 drivers [27,34] as well as pdf01 mutants [34] were obtained from Michael Rosbash, Brandeis University. The UAS-Kir2.1 flies are previously described [36] and were acquired from Grae Davis, UCSF. tubGal80ts flies [39] were ordered from the Bloomington Stock Center, Indiana University. Flies were maintained in standard cornmeal-molasses-agar food at 25 °C unless otherwise noted.
As described [66], male flies aged 2–7 days old were entrained to 2–5 days of 12 h light:12 h dark (LD) and exposed to 7–10 days of constant darkness (DD). Activity patterns were monitored by the Drosophila Activity Monitoring system (TriKinetics) in 30 min bins. Clocklab software (Actimetrics) was used to calculate period estimates using a chi-square periodogram (α = 0.01). Rhythm strength was measured as the power of each record minus the significance (“p-s”). Flies were considered rhythmic if they exhibit a p-s value of >10. Visual inspection of actograms was performed to confirm rhythmicity (or lack thereof) in weakly rhythmic flies. Lomb-Scargle periodograms were constructed using Clocklab to score period splitting as previously described [38] except significance was set at α = 0.01. Period peaks were considered if they crossed the significance line; the percent of flies exhibiting a single peak of ∼24 h versus a single long period or two peaks was calculated.
For developmental analysis (tubGal80ts crosses), flies were mated and progeny raised at 18 °C. Flies were then placed into the behavior apparatus and exposed to 5 days of LD and 7–10 days of DD at either 18 °C or 29 °C such that the LD phase allowed Gal4 induction (at 29 °C) while circadian behavior was calculated during the DD phase.
Adult male flies of the indicated genotypes aged 2–10 days were entrained to at least 3 days of LD and shifted to DD. Brains were dissected on the first day of DD as follows (modified from [44]): following brief anaesthetization, flies were pinned to a Sylgard dish and the proboscis was removed. Cold PBS was placed around the wound until all flies for that time point were processed. PBS was replaced with 4% formaldehyde/PBS to fix for 15 min. The brain was then dissected away from the head capsule under cold PBS, fixed again for 20 min, and stored in PBS at 4 °C overnight. Brains were washed 2 × 5 min in PBSTx (PBS+0.3% Triton-X100), blocked 30 min in PBSTG (PBSTx+10% normal goat sera), and incubated in the following primary antibodies overnight: rabbit anti-PER 1:4,000 and rat anti-PDF 1:1,000 [61,67], gifts from M. Rosbash, Brandeis University. Antibody was removed and brains were washed 4 × 10 min in PBSTx, then incubated with secondary goat anti-rabbit-Alexa488 (Molecular Probes, Invitrogen) and donkey anti-rat- Cy3 (Jackson ImmunoLabs) both at 1:500 dilutions for at least 2 h at room temperature. Final washes of 4 × 10 min in PBSTx were performed, brains were rinsed in PBS, and placed in 80% glycerol overnight at 4 °C before mounting.
Immunofluorescence was imaged with a Nikon C1 confocal microscope and analyzed as reported [44]. Briefly, PDF staining was used to determine cytoplasmic/nuclear compartments, and the average pixel intensity for PER staining was measured with Image J (National Institutes of Health). Background levels in nearby pixels were subtracted from the raw data, and total amounts and the nuclear:cytoplasmic PER signal ratio were calculated for each fly using Excel. A t-test was used to determine the significance between total peak and trough PER values for Gal4 controls versus pdfTik at the indicated CT. T-tests were also employed to probe differences in the nuclear:cytoplasmic ratio amongst the genotypes.
Quantification of PER western blotting was performed as described previously with NIH Image J [16]. Equal loading and transfer were confirmed with Ponceau S staining of membranes. The dilutions for the primary and secondary antibodies were 1:10,000 and 1:2,000 in TBST buffer (ECL protocol; Amersham Biosciences). Five TBST buffer washes lasting 5 min each were performed following the primary and the secondary antibody incubations. Blocking was achieved with TBST+5% milk (Bio-Rad). The incubation times for blocking, primary and the secondary antibody incubations were as follows: 1 h at room temperature, overnight at 4 °C, and 1 h at room temperature, respectively. ECL reagents were used for immunoassay signals. A single-factor ANOVA compared the effect of genotype at trough time points.
Total RNA was isolated from frozen whole heads using TRIzol reagent (Invitrogen) according to the manufacturer's protocol. DNA was removed from RNA extracts using RQ1 DNase from Promega. Real-time PCR reactions were run using the Applied Biosystems 7900HT fast real-time PCR instrument. Data were collected using SDS software version 2.2.1. Data were analyzed using the 2−ΔCt method [68] using RP49 expression values to normalize for differences in RNA amount among samples. Statistical significance was evaluated for trough transcript levels by comparing the effect of genotype at the indicated CT using a single-factor ANOVA.
For PCR reactions, ∼100 ng RNA were used per reaction. Reactions were prepared using the reagents from the Qiagen QuantiTect SYBR Green RT-PCR kit. Total reaction volume was 25 ml and reactions were run in 96-well plates. Primer sets used were ordered from Integrated DNA Technologies. Primer sequences are as follows. per: forward primer is 5′-CAGCAGCAGCCTAATCG-3′, and the reverse primer is 5′-GAGTCGGACACCTTGG-3′. vri: forward primer is 5′-TGTTTTTTGCCGCTTCGGTCA-3′, and the reverse primer is 5′-TTACGACACCAAACGATCGA-3′. RP49: forward primer is 5′-CGACGCTTCAAGGGACAGTATC-3′, and the reverse primer is 5′-TCCGACCAGGTTACAAGAACTCTC-3′. RTPCR cycling parameters were as follows: 30 min at 50 °C, 15 min at 95 °C, and 30 cycles of 15 sec at 94 °C, 30 sec at 55 °C, and 30 sec at 72 °C.
Following are the accession numbers and FlyBase identifiers for the genes and protein products in this study. National Center for Biotechnology Information (NCBI) Entrez (http://www.ncbi.nlm.nih.gov/Entrez/) gene numbers: CK2alpha, GeneID 48448; period, GeneID 31251; vrille, GeneID 33759. FlyBase (http://flybase.org/) identifiers: CK2alpha, FBgn0000258; CK2alphaTimekeeper, FBal0141857; CK2alphaTimekeeperRevertant, FBal0141856; period, FBgn0003068; vrille, FBgn0016076. NCBI Entrez protein database numbers: Drosophila CK2alpha (isoform D), NP_001036624; Drosophila PERIOD, NP_525056. |
10.1371/journal.pntd.0004363 | Efficacy and Safety of Pafuramidine versus Pentamidine Maleate for Treatment of First Stage Sleeping Sickness in a Randomized, Comparator-Controlled, International Phase 3 Clinical Trial | Sleeping sickness (human African trypanosomiasis [HAT]) is a neglected tropical disease with limited treatment options that currently require parenteral administration. In previous studies, orally administered pafuramidine was well tolerated in healthy patients (for up to 21 days) and stage 1 HAT patients (for up to 10 days), and demonstrated efficacy comparable to pentamidine.
This was a Phase 3, multi-center, randomized, open-label, parallel-group, active control study where 273 male and female patients with first stage Trypanosoma brucei gambiense HAT were treated at six sites: one trypanosomiasis reference center in Angola, one hospital in South Sudan, and four hospitals in the Democratic Republic of the Congo between August 2005 and September 2009 to support the registration of pafuramidine for treatment of first stage HAT in collaboration with the United States Food and Drug Administration. Patients were treated with either 100 mg of pafuramidine orally twice a day for 10 days or 4 mg/kg pentamidine intramuscularly once daily for 7 days to assess the efficacy and safety of pafuramidine versus pentamidine. Pregnant and lactating women as well as adolescents were included.
The primary efficacy endpoint was the combined rate of clinical and parasitological cure at 12 months. The primary safety outcome was the frequency and severity of adverse events. The study was registered on the International Clinical Trials Registry Platform at www.clinicaltrials.gov with the number ISRCTN85534673.
The overall cure rate at 12 months was 89% in the pafuramidine group and 95% in the pentamidine group; pafuramidine was non-inferior to pentamidine as the upper bound of the 95% confidence interval did not exceed 15%. The safety profile of pafuramidine was superior to pentamidine; however, 3 patients in the pafuramidine group had glomerulonephritis or nephropathy approximately 8 weeks post-treatment. Two of these events were judged as possibly related to pafuramidine. Despite good tolerability observed in preceding studies, the development program for pafuramidine was discontinued due to delayed post-treatment toxicity.
| Sleeping sickness, or human African trypanosomiasis (HAT), is a neglected tropical disease. Because only 2 treatment options are available to treat persons with stage 1 disease, and both require parenteral administration, oral drugs would be of great benefit to the affected population. In this Phase 3, multi-center, randomized, open-label, parallel-group study, we compared oral pafuramidine with intramuscular pentamidine in persons in sub-Sahara Africa with first stage HAT. At 12 months, the overall cure rates (combined clinical and parasitological cure) were similar: 89% in the pafuramidine group and 95% in the pentamidine group. At 24 months, the cure rates continued to be high: 84% and 89%, respectively. Pafuramidine’s safety profile was superior to the comparator drug, and it was consistent with the overall safety profile seen in previous Phase 2 studies. Upon further analysis, however, a renal safety issue was identified as being possibly related to pafuramidine and further clinical development was halted. Nevertheless, the clinical studies conducted in the pafuramidine development program provide a model for future studies in rural Africa.
| Sleeping sickness (human African trypanosomiasis [HAT]) is a neglected tropical disease with limited treatment options that currently require parenteral administration. Trypanosoma brucei (T.b.) gambiense is found in 24 countries in west and central Africa and currently accounts for over 98% of reported cases [1]. Despite the long history of the disease (first cases reported in 1373/1374), the drugs available to treat it are toxic, difficult to administer, and stage-specific [2].
First stage symptoms entail bouts of fever, headaches, joint pains, and itching, and a person can be infected for months or even years without major signs or symptoms of the disease. When more evident symptoms emerge, the patient is often already in an advanced disease stage where the central nervous system is affected (second stage). The majority of current HAT research is focused on stage 2 of the disease, which requires drugs that can cross the blood-brain barrier. Drugs for stage 2 HAT are either too toxic (melarsoprol) or have too complex a regimen (nifurtimox-eflornithine combination treatment) for use against the first stage of the disease.
Only two drugs are approved for treatment of stage 1 HAT: pentamidine (only for T.b. gambiense) and suramin (only for T.b. rhodesiense), which are both administered parenterally. Suramin, synthesized as a dye in 1916 [3], has been used for the treatment of sleeping sickness since 1922 [4] [5], but it can cause undesirable effects in the urinary tract and allergic reactions [1]. Pentamidine, introduced in 1937, was developed as an analog of synthalin, a hypoglycemic agent with anti-trypanosomal activity [6] [7]. Pentamidine is administered by the intramuscular route and has a reported treatment failure rate after a course of five injections of approximately 7% [8] [9] [10]. Though this efficacy profile is encouraging, treatment with pentamidine has limitations. It requires injection, which hampers its use in rural treatment facilities, and though adverse reactions are usually reversible and its most serious long-term consequence, diabetes, is rare, the treatment is accompanied by a high frequency of adverse events, including hypotension, nephrotoxic effects, leukopenia, and hypo- and hyperglycemia [11] [12].
There is no vaccine for T.b. gambiense HAT and there is a great need for new safe and efficacious drugs that would be easy to use in rural health centers and affordable. In 2000, the promising orally administered pro-drug pafuramidine was chosen for clinical development by the Consortium for Parasitic Drug Development, which was founded in 1999 to foster the development of compounds with antiprotozoal activity [13]. Pafuramidine is the dimethoxime prodrug of furamidine (which has demonstrated excellent efficacy in vitro against T.b. rhodesiense) [14]. Pafuramidine was shown to be effective in vivo in the acute model (first stage disease) in mice [15] [16] and in monkeys (green vervet monkey [17] and rhesus monkey [18]). Preclinical evaluation in vitro as well as animal testing indicated no major safety concerns.
In 2000, pafuramidine was further evaluated in Phase 1 studies in healthy patients after single and multiple dosing (up to 21 days) and was well tolerated [19]. The subsequent Phase 2 studies (conducted from 2001 to 2007) in patients with stage 1 sleeping sickness supported this finding [20]. Efficacy after 5 days of treatment was limited; therefore, to attain efficacy comparable to that of pentamidine, the treatment duration was prolonged to 10 days. The pharmacokinetic properties of pafuramidine (in particular, the lack of proportional conversion of DB289 to DB75 at therapeutic doses) precluded using a higher dose to improve efficacy [19] [20] [21] [22].
This single, confirmatory, pivotal Phase 3 study was developed to support the registration of pafuramidine for treatment of stage 1 HAT under a Special Protocol Assessment in collaboration with the United States (US) Food and Drug Administration (FDA). The primary objective of the study was to demonstrate the non-inferiority of oral pafuramidine versus intramuscular pentamidine for treatment of first stage HAT caused by T.b. gambiense. Since safety and efficacy of a new drug should, if at all possible, be established in a study population representative of the target population, the secondary objective was to include pregnant and lactating women as well as adolescents in the study. Reproductive studies of pafuramidine in animals have not indicated any embryo or fetal toxicity or other effects on reproductive function of adult male and female rats or rabbits. Therefore, it was considered appropriate and was approved by the US FDA to proceed with studies including pregnant and lactating women.
This was a multi-center, multi-country, open-label (sponsor-blinded), parallel-group, comparator-controlled, randomized Phase 3 study to compare the efficacy, safety, and tolerability of pafuramidine and pentamidine in 273 patients with first stage HAT caused by T.b. gambiense. The study was conducted at six African sites where T.b. gambiense sleeping sickness is endemic: two trypanosomiasis reference centers (Angola and the Democratic Republic of the Congo [DRC]) and four hospitals (1 in South Sudan and three in the DRC) from August 2005 (first patient enrolled) to September 2009 (last patient follow-up completed).
The study was registered on the International Clinical Trials Registry Platform at www.clinicaltrials.gov with the number ISRCTN85534673. International Protocol #289-C-006.
There was one protocol amendment that provided detailed microscopy instructions for examining blood and CSF for the presence of trypanosomes and determining WBC count in CSF. The amendment also detailed randomization of pregnant and lactating women in a separate strata, and provided additional clarifications and administrative changes.
Male and female patients were eligible to participate if they were ≥12 years of age, weighed ≥30 kg, had first stage T.b. gambiense infection documented by the presence of trypanosomes in the blood and/or lymph, and had no evidence of second stage disease (no trypanosomes detected in the cerebrospinal fluid [CSF] and ≤5 white blood cells [WBCs]/mm3 in CSF). Patients were also excluded if tested positive for malaria or helminth infections. Patients were not tested for HIV prior to treatment. Patients were treated at two trypanosomiasis reference centers (Angola and the DRC) and four hospitals (1 in South Sudan and three in the DRC). Written informed consent was obtained from each patient. If the patient was a minor or mentally impaired, a legal guardian also signed the consent form and if a patient was illiterate, an impartial witness assisted in the consent process.
Pregnant and lactating female patients as well as adolescents 12 to 15 years could be enrolled. Adolescents underwent additional safety laboratory testing at the 3-month post-treatment visit. Eligible pregnant and lactating female patients could participate with the understanding that additional safety measurements regarding course and outcome of the pregnancy and/or the health of their infant would be performed.
Patients were excluded if they had a possible or confirmed second stage T.b. gambiense infection (ie, presence of parasite in the CSF upon microscopic examination or a WBC count in the CSF of >5 mm3); any active, clinically relevant medical conditions that in the investigator’s opinion might jeopardize patient safety or interfere with study participation; presented with a score of less than 9 on the Glasgow Coma Scale; were previously treated for HAT; or displayed other conditions that would compromise participation.
Screening occurred within 7 weeks prior to dosing with pafuramidine or pentamidine (within 6 weeks prior to the baseline evaluation) using the card agglutination test for trypanosomes [11] [12] and microscopic examination (thin and/or thick smear) of blood and lymph node aspirate for trypanosomes either at the trypanosomiasis treatment centers or by mobile diagnostic units. All diagnostic tests performed by mobile teams were repeated in the treatment centers. Lumbar puncture was performed at the treatment centers in all trypanosome-positive cases detected by any method and the disease stage was determined by microscopic examination of CSF for trypanosomes and by WBC counts. If the result was negative, a blood sample was examined (including hematocrit centrifugation [23] and miniature anion exchange centrifugation technique [m-AECT] [24]). All patients were tested for malaria, and filaria using thick and thin blood smears and for diarrhea. If necessary, malaria treatment was given before enrolment; filariasis therapy was administered after study treatment when necessary. Patients were admitted as in-patients to the clinical site for the full duration of the treatment/observation period (11 days for pafuramidine or 7 days for pentamidine). Other baseline documentation included demographics, medical history, signs and symptoms of HAT, and concomitant disease(s) and medication(s).
Clinical supplies of pafuramidine were provided to the sites in bottles (50 tablets of 100 mg) labeled to indicate study drug, strength, expiration date, protocol number, and other information according to local regulations. Pentamidine was provided locally by the agency (generally the national HAT control programs) responsible for each center in the form of pentamidine isethionate for injection in single-dose vials at 200 mg/vial.
For efficacy assessments, patients underwent microscopic examination of blood (thin and/or thick smear), hematocrit centrifugation of blood [25], microscopic examination of lymph node aspirate, and microscopic examination of blood after m-AECT concentration at the end of treatment and at 3, 6, 12, and 18 months post-treatment [26]. Lumbar puncture was performed for microscopic examination of CSF fluid for WBCs and trypanosomes at baseline and 6, 12, and 18 months post-treatment, and at any other evaluation where relapse was suspected or trypanosomes were demonstrated in blood or lymph nodes. Additional assessments of clinical efficacy were performed at 24 months post-treatment.
During the treatment and post-treatment period, safety evaluations included vital signs, physical examination, adverse event monitoring, laboratory tests, electrocardiogram (ECG) monitoring to the extent possible at each site, and documentation of concomitant medications. Signs or symptoms of HAT were queried and graded. Laboratory tests assessed clinical chemistry (aspartate aminotransferase, alanine aminotransferase, total bilirubin, glucose, and creatinine) and hemoglobin. Electrocardiograms were performed at baseline, 1 hour prior to dosing, and 1 hour after dosing for all patients. An additional ECG was obtained on Day 7 post-treatment for pafuramidine-treated patients.
Clinical response definitions are listed in Table 1. The primary efficacy endpoint was the combined rate of cure and probable cure at the 12-month follow-up in the per-protocol data set. The overall cure rate was defined as the proportion of treated patients with no clinical signs and symptoms of HAT, no evidence of trypanosomes in any body fluid examined, and no treatment with other trypanocidal agents for any reason; in addition, ≤5 WBCs/mm3 in CSF obtained from a lumbar puncture was required.
Secondary efficacy endpoints were cure, clinical cure, probable relapse, relapse, and death rates at the end of treatment and all follow-up visits. Parasitological cure, probable relapse, relapse, and death rates were also assessed at the 12-month test of cure evaluation and at the 24-month evaluation; the clinical cure was considered equivalent to the parasitological cure at the 24-month evaluation.
Study efficacy parameters and timing of post-treatment evaluations were based on WHO recommendations for clinical product development for HAT [27]. Although 18 months post-treatment is recommended to assess clinical cure in HAT control programs due to anticipated increased drop-out rates from follow-up after 6 to 12 months, the 12-month evaluation was chosen as the primary endpoint in this study in order to maintain a robust data set for the primary analysis.
Safety was assessed through the end of treatment evaluation and included adverse events, laboratory results, vital sign measurements, physical examinations, and use of any concomitant medications. The term “adverse event” included any of the following events that developed or increased in severity during the study: 1) any signs or symptoms whether thought to be related or unrelated to HAT; 2) any clinically significant laboratory abnormality; or 3) any abnormality detected during physical examination. Adverse events were graded by the investigator (1 = mild, 2 = moderate, 3 = severe, 4 = intolerable). Adverse events were assessed at every study visit and were classified according to the terms found in the Medical Dictionary for Regulatory Activities (MedDRA).
A serious adverse event was defined as any event that suggested a significant hazard, contraindication, side effect, or precaution, it included any event that: 1) is fatal; 2) is life threatening; 3) is a persistent or significant disability/incapacity; 4) requires or prolongs in-patient hospitalization; 5) is a congenital anomaly/birth defect; or 6) is an important medical event, based upon appropriate medical judgment, that may jeopardize the patient or may require medical or surgical intervention to prevent one of the other outcomes defining serious.
There were no changes to any of the outcomes.
A total of 250 patients, 125 patients per treatment group, were originally expected to be treated in order to include 100 patients per treatment group in the per-protocol. This sample size provided more than 90% power to demonstrate non-inferiority of pafuramidine to pentamidine for the primary endpoint, when the study drugs have equivalent probable cure rates of 95% in the per-protocol analysis. Non-inferiority comparison was conducted with an alpha equal to 0.048 and non-inferiority margin (ie, delta) of 0.15.
The sponsor may have terminated this study prematurely, either in its entirety or at a particular site, for reasonable cause or safety concerns. The sponsor remained blinded and the data were provided to the data safety monitoring board (DSMB) for evaluation. Based on these data, the DSMB made recommendations to the sponsor regarding continuation of the study. The study could have been stopped if: 1) any new untoward safety issues were identified in the pafuramidine treatment group such that pafuramidine was significantly less safe than pentamidine; 2) the re-estimated sample size exceeded 500 patients to achieve 90% power for the primary efficacy endpoint; or 3) efficacy analysis indicated that pafuramidine was significantly more effective than pentamidine (p<0.002).
An interim analysis was to be conducted when half of the enrolled patients reached the 12-month post-treatment endpoint, however, this was not done because the pafuramidine development program was discontinued due to delayed post-treatment toxicity (details are provided in the Harms section).
Patients were randomly assigned by the local investigators to receive either pafuramidine or pentamidine in the order in which they were enrolled. Randomization was carried out in blocks of variable size following a randomization schedule prepared by the sponsor; randomization of pregnant and lactating women was stratified. Each study site was provided with series of individual envelopes each containing a card with the treatment assignment for 1 patient and a control number. After a patient signed the informed consent and inclusion/exclusion criteria were confirmed, the investigator opened the next envelope in the randomization list to obtain treatment assignment for that patient and then transferred the control number to the patient’s case report form.
The study was open-label, since pafuramidine is administered orally and pentamidine is administered intramuscularly. However, the sponsor was blinded to treatment assignments.
The primary efficacy analysis, demonstrating the non-inferiority of pafuramidine to pentamidine for the combined rate of cure and probable cure, was conducted with an alpha equal to 0.048 and non-inferiority margin (ie, delta) of 0.15. The comparison was made with a 1-sided 97.6% confidence interval (CI) for the treatment difference in parasitological cure rate. The normal approximation to the binomial distribution with continuity correction was used to construct the CI. The primary data set for efficacy analysis was the per-protocol data set.
The efficacy analysis was carried out for the per-protocol data set (primary analysis), the intention-to-treat (ITT) data set, and the modified ITT (mITT) data set (supportive analyses). The per-protocol data set was defined as patients who received a minimum of 7 days of pafuramidine or five injections of pentamidine and who attended the test of cure assessment at Month 12 or reached an efficacy endpoint (death, non-response, or relapse) at an earlier time. Patients without lumbar puncture at Month 12 were included and their outcome was assessed based on clinical signs and symptoms and parasitological findings from any body fluid examined.
The mITT data set consisted of all patients who received the minimum amount of randomized study drug (7 or 5 days) and for whom an end-of-treatment assessment and at least one follow-up efficacy assessment were available. Patients who had received at least one dose of study drug were included in the ITT data set and patients who were lost to follow-up or discontinued from the study for any reason were considered treatment failures in the ITT analysis. The last observation carried backward was used to account for missing data at an earlier evaluation (in case of cure at a later evaluation). For the mITT analysis, missing data were addressed according to the last observation carried forward principle.
The secondary efficacy variables were summarized at all time points with point estimates and 1-sided 97.5% CIs for the difference between treatments.
The safety data set consisted of all patients who received at least one dose of study drug and had at least one safety evaluation after dosing. Treatment group differences in the proportion of patients who reported treatment-emergent adverse events for Day 1 through Day 11 were assessed with Fisher’s exact test.
The number and percentage of patients who reported treatment-emergent adverse events were summarized for each treatment group at the system organ class, high-level group terms, and preferred term level. Treatment group differences in the proportion of patients who reported each high-level group term were assessed with Fisher’s exact test.
Any clinically significant physical examination changes from baseline were captured as an adverse event.
First stage HAT patients rarely present at a hospital or a treatment center. Therefore, intense screening activities were necessary. Between July 2005 and March 2007, a total of 234,919 patients were screened to find 839 individuals affected with HAT (Fig 1). The exclusion rate was high (566 of 839 patients, 67.5%); primary reasons were that patients had stage 2 HAT and did not meet inclusion criteria. A total of 273 patients were randomized: 136 patients received pafuramidine and 137 received pentamidine; all 273 completed the study. Most of the patients (91%) were enrolled in the DRC (248 of 273 patients); 5.5% were enrolled in Angola (15 of 273 patients), and 3.7% (10 of 273 patients) were enrolled in South Sudan.
As shown in Fig 1, follow-up attendance at Month 24 was good; only 2 patients in the pafuramidine group and 8 patients in the pentamidine group were lost to follow-up.
As seen in Table 2, baseline demographic characteristics between the two treatment groups were similar. The median age of patients in both treatment groups was approximately 30 years, and the majority of patients in both groups were female (70% and 66%, respectively).
As shown in Fig 1, 133 of 136 patients (97.8%) in the pafuramidine group and 129 of 137 patients (94.2%) in the pentamidine group were included in the efficacy analysis. Three patients in the pafuramidine group and 8 patients in the pentamidine group were excluded because they were lost to follow-up.
As shown in Table 3 at the test of cure evaluation (12 months post-treatment), the combined rate of cure and probable cure was 89% (118 of 133 patients) in the pafuramidine group and 95% (123 of 129 patients) in the pentamidine group in the per-protocol population.
Pafuramidine was non-inferior to pentamidine as the upper bound of the 95% CI did not exceed 15%. This finding was supported by the 24-month follow-up data, where cure rates of 84% for the pafuramidine group and 89% for the pentamidine group were maintained (Table 4). Supportive analysis in the ITT and mITT populations generated similar results.
Table 5 summarizes the secondary efficacy variables for each follow-up visit, including the cumulative number of cures, probable cures, probable relapses, relapses, and deaths for each treatment group. There were no deaths in either treatment group during the active study period, and all patients responded to treatment. Relapses in the pafuramidine treatment group appeared to be evenly distributed over the whole follow-up period, whereas a trend for late relapses was observed in the pentamidine treatment group.
The numbers of adolescents and pregnant women (8 and 10, respectively) were too small to make any definitive conclusions about efficacy in these patients. For lactating women, the overall cure rates of pafuramidine and pentamidine at the test of cure evaluation were the same (23 of 26 patients [88.4%] in each group). However, the low number of lactating women also did not allow for definitive conclusions.
No patients prematurely discontinued due to an adverse event during the study.
As shown in Table 6, the most commonly reported adverse events were injection site pain, pyrexia, hypoglycemia, and hypotension. These events occurred more frequently in the pentamidine group than the pafuramidine group, with the exception of pyrexia, which occurred more frequently in the pafuramidine group. The incidence of patients with at least one adverse event overall for Day 1 through Day 11 was statistically significantly less in the pafuramidine treatment group (82%, 111 of 136) than in the pentamidine group (99%, 135 of 137) (p<0.05). Among high-level group terms, there were statistically significant differences in favor of the pafuramidine treatment group versus the pentamidine group for hepatobiliary investigations (7% vs. 77%, respectively); renal and urinary tract investigations and urinalyses (2% vs. 9%, respectively); glucose metabolism disorders (including diabetes mellitus) (6% vs. 18%, respectively); and decreased nonspecific blood pressure disorders and shock (44% vs. 62%, respectively) (p<0.05 for all).
A statistically significantly greater percentage of pafuramidine patients than pentamidine patients experienced epidermal and dermal conditions (5% vs. 1%, respectively) (p<0.05).
The majority of the adverse events were mild or moderate in severity and typical for patients recovering from first stage HAT.
The ECG results from this study were included in a separately published study on cardiac alterations in HAT [28]. In brief, the mean PQ and QTc intervals did not increase during treatment of first stage disease in either treatment group. The appearance and disappearance of repolarization changes at the end of treatment were comparable between groups.
A total of 43 patients experienced serious adverse events during the study (including the follow-up period): 19 of 136 patients (14.0%) in the pafuramidine group and 24 of 137 patients (17.5%) in the pentamidine group. Of these, only 3 patients had serious adverse events while on treatment: 1 in the pafuramidine group (cellulitis considered probably not related to study drug) and 2 in the pentamidine group (hypersensitivity considered not related to study drug and subcutaneous abscess considered probably related to the study drug). All other serious adverse events occurred during the follow-up period.
Of the 43 patients with serious adverse events, it was initially considered probable that only one was related to the study drug (subcutaneous abscess in the pentamidine group). Re-evaluation of the relatedness of these events to the study drug was subsequently performed when serious renal and hepatic post-treatment toxicity was observed in 3 patients in a supportive Phase 1 study of pafuramidine, which was conducted in South Africa (in December 2007), during the follow-up period of the current Phase 3 study. The Phase 1 study included 175 male and female volunteers taking oral pafuramidine 100 mg BID for 14 days [19]. Retrospectively, the glomerulonephritis reported for the 2 pafuramidine patients in the current Phase 3 study was considered to be possibly related to the study drug.
Thirteen patients (6 in the pafuramidine group and 7 in the pentamidine group) died during the follow-up period. All deaths were considered not related or probably not related to the study drug. Two deaths in the pafuramidine group were considered to be related to HAT; one death was considered to be related to relapse of HAT and another was considered to be associated with treatment of a HAT relapse with melarsoprol.
Safety data for adolescent as well as pregnant and lactating women were similar to the observations in the general population.
Although this study was conducted in rural conditions in Angola, South Sudan, and the DRC with local teams that had limited experience in clinical studies, this study was fully compliant with Good Clinical Practice and regulatory standards. In addition, this was the first Phase 3 study of a new drug intended for treatment of sleeping sickness conducted under a US FDA Investigational New Drug Application that followed contemporary International Conference on Harmonisation guidance.
The results demonstrate the efficacy of pafuramidine in the treatment of first stage HAT, with an overall cure rate that was statistically non-inferior to that observed for pentamidine (89% vs. 95%, respectively) at 12 months post-treatment. The results obtained in the per-protocol set were confirmed in the ITT and mITT analysis populations, which included missing follow-up visits and patients lost to follow-up. The Month 24 results in all populations corroborate the 12-month results and demonstrate the robustness of the primary efficacy analysis.
Compared with patients who received pentamidine, pafuramidine-treated patients (total population including the subpopulations of adolescents and pregnant and lactating women) had lower rates of treatment-related adverse events (93% vs. 40%, respectively) and lower rates of adverse events related to hepatic, renal, and metabolic toxicity. An ECG analysis revealed no cardiotoxicity for either drug [28]. These data are consistent with the good tolerability observed for pafuramidine in the previous Phase 2 studies [20].
This study was also designed to evaluate efficacy and safety of pafuramidine and pentamidine in subpopulations that are particularly vulnerable to the long-term socioeconomic burdens associated with HAT, mainly pregnant and lactating women. However, the number of participating pregnant and lactating women was too low for a thorough separate analysis. The low number may be a result of social pressures and fear of treatment, which could be a detriment to seeking HAT treatment and going to a hospital. Low fertility and amenorrhea, which are often associated with HAT, may also have contributed [29]. From the limited number of relevant patients, there was no evidence for reduced efficacy or additional safety issues relating to pafuramidine compared with those observed in the overall study population. Numeric cure rates were similar and no specific safety issues were identified.
The initial safety profile observed for pafuramidine in this study was consistent with the results of preceding studies in the pafuramidine clinical development program [19] [20]. However, 3 patients in the pafuramidine group exhibited glomerulonephritis or nephropathy approximately 8 weeks post-treatment. On further examination, these events appeared to be similar to events that occurred in the previously mentioned supportive South African Phase 1 safety study. After re-evaluation, 2 of the 3 patients were considered to have events that were possibly related to pafuramidine by the principal investigator.
It should be noted that the analysis of safety data, particularly of the serious adverse events that occurred in the HAT study reported here, did not reveal any apparent negative long-term effects of pafuramidine. The patients who experienced renal toxicity recovered without sequelae, and the additional safety data obtained during the follow-up revealed no differences in abnormal biochemistry values between pafuramidine and pentamidine groups.
Eventually, clinical development of pafuramidine was discontinued in early 2008, since the renal toxicity observed in the additional Phase 1 study was considered to be an unacceptable risk. Preliminary evidence of the involvement of the kidney injury molecule (KIM-1) was only very recently provided through the use of a mouse diversity panel [30].
From the perspective of study design, it is noteworthy that the 12-month endpoint for efficacy effectively predicted the clinical outcomes determined at the 24-month evaluation. Thus, 12 months is a meaningful endpoint for a sleeping sickness study with adequately performed follow-up. The infrastructure and technical expertise developed during the Phase 2 development program for pafuramidine were effectively leveraged to guide the screening, enrolment, and oversight of the larger study population included in this Phase 3 registration study, and eventually led to a unique and comprehensive data set. The successful conduct of the study was evidenced by the retention of 97% (265 of 273 patients) of the randomized patients at the 24-month (end-of-study) evaluation. Finally, the lessons learned from the Phase 2 development program were also helpful in ensuring that the Phase 3 study complied with Good Clinical Practice and regulatory standards required for a registration study [20].
The repeated success of clinical study conduct throughout the pafuramidine development program provides a model for future studies in rural Africa and will undoubtedly contribute to continued improvement of HAT control in sub-Saharan Africa.
|
10.1371/journal.ppat.1000222 | Migratory Dermal Dendritic Cells Act as Rapid Sensors of Protozoan Parasites | Dendritic cells (DC), including those of the skin, act as sentinels for intruding microorganisms. In the epidermis, DC (termed Langerhans cells, LC) are sessile and screen their microenvironment through occasional movements of their dendrites. The spatio-temporal orchestration of antigen encounter by dermal DC (DDC) is not known. Since these cells are thought to be instrumental in the initiation of immune responses during infection, we investigated their behavior directly within their natural microenvironment using intravital two-photon microscopy. Surprisingly, we found that, under homeostatic conditions, DDC were highly motile, continuously crawling through the interstitial space in a Gαi protein-coupled receptor–dependent manner. However, within minutes after intradermal delivery of the protozoan parasite Leishmania major, DDC became immobile and incorporated multiple parasites into cytosolic vacuoles. Parasite uptake occurred through the extension of long, highly dynamic pseudopods capable of tracking and engulfing parasites. This was then followed by rapid dendrite retraction towards the cell body. DDC were proficient at discriminating between parasites and inert particles, and parasite uptake was independent of the presence of neutrophils. Together, our study has visualized the dynamics and microenvironmental context of parasite encounter by an innate immune cell subset during the initiation of the immune response. Our results uncover a unique migratory tissue surveillance program of DDC that ensures the rapid detection of pathogens.
| Cutaneous Leishmaniasis is a difficult-to-treat disease affecting millions of people worldwide. Hence, there is high demand for the development of vaccines against Leishmania parasites, begging for a better understanding of immune responses against this pathogen. Dendritic cells, as part of the innate immune system, are thought to act as gatekeepers against intruding pathogens. However, their behavior in the context of intact tissues is incompletely understood. Here, we have used intravital two-photon microscopy to visualize the behavior of skin resident dendritic cells in real time, both in the steady-state and upon parasite encounter. We have found that migratory dermal dendritic cells are capable of rapidly sensing Leishmania parasites injected into the skin. This occurred through the formation of highly motile cellular processes capable of engulfing parasites, followed by parasite uptake into the cell. Together, our study provides a new vista of the orchestration of host cell–pathogen encounter in the three-dimensional context of intact tissues. Our results serve as the basis for a better understanding of the dynamic regulation of tissue surveillance by dendritic cells.
| The skin is the interface between the environment and internal tissues. Dendritic cells (DC), as part of the body's innate immune defense, are strategically positioned in this organ; the epidermis is the home of Langerhans cells (LC), while the dermis harbors dermal DC (DDC). The main function of DC is believed to be the recognition and processing of foreign antigens, and subsequent presentation to naïve T cells [1]. DC normally reside in an immature state in the skin. Upon antigen encounter in the presence of “danger signals”, such as proinflammatory cytokines, DC undergo maturation enabling their migration to draining lymph nodes (LN) [2]. Accumulating evidence suggests that DDC may be responsible for the transport of pathogens to draining LN [3]–[6]; in certain infections, for example with herpes simplex virus, DDC act as an antigen shuttle, i.e. they transfer antigen to LN-resident CD8+ DC, which subsequently present it to T cells [7]. In other infections, including those with Leishmania parasites, they may present antigen directly to T cells [6].
Using intravital confocal microscopy, LC in the skin were found to be immobile with occasional repetitive dendrite movement, termed dendrite surveillance extension and retraction cycling habitude (dSEARCH) [4],[8]. In contrast to LC, very little is known about the migratory and interactive behavior of DDC. This is of significance, as during certain infections DDC may come into close contact with microorganisms, and it is unclear whether DDC are capable of detecting living pathogens directly or take up antigens from dying infected cells or dead pathogens. Since these initial events of an immune response are likely to determine the magnitude and quality of T cell and B cell immunity, it is important to decipher the events of pathogen encounter directly in situ.
Cutaneous Leishmaniasis is a disease caused by a large group of protozoan parasites belonging to the Genus Leishmania, including L. major. It serves as a paradigmatic skin infection, as promastigote stage parasites are directly deposited into the dermis during sand fly bites [9]. While it is thought that the parasites then infect innate immune cells in the skin, primarily macrophages [10],[11], the precise events occurring at the time of infection are not well defined. After entering cells, the parasites rapidly transform to the amastigote form, a rounded non-flagellated stage, which survives and multiplies within the phagolysosome (parasitophorous vacuole, PV) up until the time of cell rupture. After several weeks a lesion at the site of infection develops that is primarily composed of infected and inflammatory cells [12],[13]. In some cases, these lesions are able to resolve over several months, while in others the lesions are chronic and can be associated with severe disease [9]. Current treatment options are scarce, therefore begging for the development of prophylactic vaccines. A prerequisite for this is a thorough understanding of the immune response against the parasites.
Several studies have investigated the response of cutaneous DC to Leishmania spp. Initial reports suggested that LC are infectable by Leishmania, migrate to LN and activate CD4+ T cells [14]. However, more recently these findings have been questioned as DC harboring parasites in LN do not express the LC marker langerin [6]. Also, mice that lack MHC class II expression in LC but not DDC resolve infection similarly to wildtype animals [15]. While several investigators have suggested that DDC transport Leishmania to the paracortex of draining LN [5],[6], others have questioned the role of cutaneous DC during early infection altogether [16],[17]. At later stages of disease, monocyte-derived DC may differentiate directly within the inflamed skin, and then migrate to draining LN where they induce CD4+ T cell activation [18]. To gain further insight into this controversy, i.e. what is the nature of parasite-DC encounter during early infection, ideally, Leishmania infections should be visualized directly in the natural microenvironment of the skin.
In the present study, we made use of intravital two-photon microscopy (2P-IVM) to address the following questions: 1. What is the steady-state behavior of DDC? 2. How do DDC respond to danger signals? And 3. Do cutaneous DC take up Leishmania parasites in the early phase of infection, and if so, what are the dynamics of this process? Surprisingly, we found that DDC were migratory under homeostatic conditions, which is in stark contrast to their epithelial counterparts. After encountering danger signals, DDC underwent a morphological transition into immobile, dendritic-shaped cells. At this point, the cells were capable of taking up parasites through the elaboration of motile pseudopods. Together, these results shed new light on the dynamics and anatomy of host-pathogen interactions.
In order to visualize the behavior of LC and DDC, we made use of CD11c-YFP mice [19], in which DC express high levels of cytoplasmic YFP. To ascertain that skin DC expressed YFP, we analyzed single cell suspensions prepared from separated epidermis and dermis by flow cytometry (Figure 1). CD45+YFP+ epidermal cells were CD11c+CD11b+F4/80+I-Ab+ (Figure 1), and immunofluorescence staining of tissue sections showed that langerin expressing YFP+ cells displayed the characteristic morphology of LC (data not shown). In the dermis, CD45+YFP+ cells were CD11c+CD11b+F4/80+I-Ab-high, and therefore represented DDC [20]. We also detected a subset of CD45+YFPlow cells within the dermis. However, this signal was due to autofluorescence, rather than specific YFP expression, as a similar population of cells was also found in wildtype animals (Figure S1A). These cells were CD11c−CD11b+F4/80+Moma-2+I-Ab-low thereby resembling dermal macrophages [20]. The fluorescence intensity of these cells was, on average, 50 times dimmer than the YFP signal from DDC. Since, under our 2P imaging conditions, we did not detect any signal in the dermis of wildtype animals (Figure S1B), we concluded that LC and DDC in CD11c-YFP mice can be detected by means of specific YFP expression, while other hematopoietic cell subsets remain undetectable.
The distribution of YFP+ DC populations was determined by 2P-IVM in the ear skin of CD11c-YFP mice. Vertical scans revealed the presence of YFP+ cells between 5–20 µm below the outermost epidermal layer (Figure 2A–2C). These cells exhibited numerous, irregularly shaped dendrites, morphologically consistent with LC. The highest density of LC was found 15 µm underneath the stratum corneum (Figure 2C). Below the epidermis, second harmonic generation (SHG) signals highlighted extracellular matrix (ECM) components [21] forming a dense, mesh-like network (vertical depth of 20–60 µm from the outermost surface). Embedded in the lower part of this network, with the highest density between 20–40 µm below the basement membrane and reaching up to a depth of ∼100 µm, were scattered YFP+ cells, of markedly different morphology to LC, i.e. of rounder shape, with fewer, shorter dendrites (Figure 2A–2C). The overall density of LC was approximately 3 times higher than that of DDC (Figure 2C). Together, these results established that cutaneous skin DC populations could be imaged by means of 2P-IVM, and identifed two morphologically distinct cutaneous DC subsets in the different compartments of the skin.
While epithelial DC populations in the skin and gut have been found to be sessile [19],[22], no information is available on DC behavior in the interstitial space within peripheral organs. Nevertheless, DC in the dermis are suspected to be involved in antigen transport from the skin to draining LN thereby regulating the initial phases of host-pathogen responses. We therefore asked whether DDC scanned their microenvironment in a similar fashion to epidermal LC. To this end, we conducted time-lapse 2P-IVM in ear skin of CD11c-YFP mice. When focusing on the epidermis, we found that LC were sessile (mean velocity <2 µm/min), with their dendrites remaining almost completely immobile (Figure 3A–3C and Video S1). As described previously, we occasionally observed dSEARCH [4],[8] (Figure 3B and Video S2). However, in contrast to LC, we discovered that DDC were actively crawling through the interstitial space of the dermis at a mean velocity of 3.7±0.3 µm/min (mean±SEM) (Figure 3A–3C and Video S3). Migrating cells exhibited a polarized morphology, often displaying lamellipodia at the leading edge and a trailing uropod-like structure (Video S3). Since our experiments were performed in non-inflamed ear tissue, these results suggest that continuous migration is a steady-state feature of interstitial cutaneous DC. It may further indicate that the unexpectedly high motility of DDC serves to screen the dermal extracellular space for intruding microorganisms/environmental noxae.
We next sought to define signals involved in spontaneous migration of DDC. When we treated animals with pertussis toxin (PTX), an inhibitor of Gαi protein-coupled receptors, the capability of DDC to translocate within the dermis significantly decreased (reflected by a reduction of their displacement; Figure 4A and Videos S4 and S5). The migratory velocity of DDC did not change after PTX treatment, because cells moved back and forth in the same place (therefore, following cell-centroids resulted in measurable velocity; Figure 4A and Video S5). We concluded that, while PTX does not interfere with the migratory machinery of DDC per se, DDC utilize chemo-attractant signals, most likely chemokines, for their migration through the interstitial space.
Having defined the cellular activities of skin DC in the steady-state, we determined their behavior in the presence of danger signals implicated in DC activation [23]. CD11c-YFP mice were injected intravenously with LPS (50 µg), which mimics systemic bacterial infection [24]. Two to eight hours after LPS treatment, LC remained sessile within the epidermis, without evidence of lateral or vertical movement (Figure 4B). In contrast, we observed dramatic changes of DDC behavior two to four hours after LPS administration. They exhibited significantly decreased migratory velocity (2.12±0.21 µm/min) and displacement, with more than 50% immobile cells (Figure 4B, Figure S2 and Video S6). Six to eight hours post LPS injection DDC partially regained their mobility (70% motile cells; Figure 4B, Figure S2, and Video S7). Thus, upon encounter of danger signals, DDC change their behavior, which may facilitate recognition/uptake of pathogens.
To further test this hypothesis, we used the protozoan parasite L. major as a model pathogen. During natural infection, promastigote stage Leishmania spp. are directly deposited into the dermis by sand flies. Previous in vitro studies have demonstrated that DC can be infected by Leishmania parasites [13],[25]. We therefore speculated that DDC may recognize and interact with L. major upon introduction into the dermis.
1–2×105 DsRed2-tagged Leishmania (LmjF-DsRed2) promastigotes were injected in a small volume (1.5 µl) of saline solution into the superficial dermis. This allowed us to deposit parasites underneath the epidermis at a vertical depth of 25–60 µm while keeping mechanical tissue disruption as minimal as possible (Figure 5A). Within 20 min of injection, DDC in the vicinity of parasites decreased their migratory speed and changed their shape to a more dendritic cell-like morphology characterized by the emergence of multiple dendritic processes (Figure 5B and 5C). This was paralleled by the appearance of several intracellular vacuoles, each of them containing a single red parasite (Figure 5C), which is consistent with the formation of PVs [26],[27]. Interestingly, these vacuoles were mobile, i.e. appeared to move freely within the cytoplasm of the cells. Two to three hours after infection, the percentage of DDC harboring one or more parasite was ∼70% (Figure 5C). Of note, LC morphology and behavior was unchanged after infection with L. major. Further, LC were not found to take up parasites, at least during the first six hours of infection (data not shown). However, it should be pointed out that parasites were injected intradermally. Consequently, LC access to parasites may have been prevented by anatomical barriers, such as the epidermal basement membrane.
To determine whether parasite uptake by DDC was specific for the Friedlin strain (FV1) of L. major, or could be recapitulated with other L. major strains, we injected the LV39 strain under the same conditions as described above. As shown in Figure 5C, this led to ∼55% of DDC containing parasites. We therefore consider L. major uptake by DDC a general phenomenon.
For most of our experiments we made use of stationary phase L. major promastigotes. Since these cultures may contain stages of different infectivity or even a few dead parasites, confirmatory experiments (n = 3) using highly purified metacyclic [28] LmjF-DsRed2 parasites were conducted. These experiments confirmed uptake of parasites by YFP+ DDC into cytoplasmic vacuoles to the same extent as stationary phase parasites (Figure S3).
Since our LPS experiments had shown that DDC markedly reduce their locomotion after exposure to a danger signal, we next assessed the migratory behavior of L. major-bearing DDC. As shown in Figure 5D, infected DDC significantly reduced their migratory velocities. To determine whether parasite uptake and migratory arrest were related phenomena, we also measured the migratory speed of non-infected DDC. We found that the latter revealed a similar reduction in their migratory velocities and displacement as compared to their infected counterparts. Collectively, these results show that DDC, by default, reduce their migration at sites of inflammation.
We also conducted sham infection experiments using a red fluorescent dye, SNARF-1, in order to exclude that the physical manipulation during intradermal injection by itself caused changes in DDC behavior. As shown in Figure 5E, there was no difference in DDC migration between SNARF-1 injected and uninjected skin attesting to the specificity of the infection experiments.
The exact mode by which Leishmania infects cells in vivo is not known. It is thought that parasite uptake by phagocytes involves non-random promastigote attachment to the cell followed by engulfment [29]. However, only in vitro data on this process are currently available, and the cellular and molecular mechanisms remain poorly understood. Our intravital imaging experiments demonstrated that cytoplasmic DDC processes actively extended towards parasites (Figure 6A and Videos S8 and S9) at an average speed of ∼2.5 µm/min and reaching up to 50 µm in length. We occasionally observed that dendrite extension was preceded by parasite contact with the cell membrane followed by engulfment along the long axis of the parasite (Figure 6A and Videos S8 and S9). After capturing parasites, dendrites often rapidly retracted towards the cell body, paralleled by the formation of an intracellular vacuole. These results establish that L. major parasites in the interstitial space were internalized in a free form by DDC during the early phase of infection.
We next asked whether inhibition of Gαi protein-coupled receptor signaling interfered with parasite uptake by inoculating mice with LmjLV39-DsRed2 parasites two to three hours after systemic PTX treatment. Since after PTX application DDC did not translocate through the dermis, we imaged cells that co-localized with the parasite depots. We observed that the formation of pseudopods and parasite uptake was preserved in these non-displacing DDC (Figure 6B). This indicates that parasite sensing was independent of Gαi protein-coupled receptors. Furthermore, these results show that migration and dendrite formation can be uncoupled at the molecular level.
Phosphoglycans (PG), in particular lipophosphoglycans (LPG), are essential during the infectious cycle of Leishmania. For instance, PGs have been implicated in the adherence of parasites to the gut epithelium in the sand fly, the resistance to complement in the blood stream, and have been considered candidate molecules for the uptake by host cells [13],[30]. PG-deficient parasites persist in vivo for months without causing disease, and are therefore considered potential attenuated anti-Leishmania vaccine candidates [31]. While in vitro studies have shown that macrophages can take up PG-deficient parasites, it is not known whether the target cell range is the same for PG-deficient and wildtype parasites in vivo. To gain further insight into the role of LPG in parasite interactions with DC in vivo, we made use of DsRed2-tagged L. major deficient in the LPG2-encoded Golgi GDP-mannose transporter. These parasites fail to synthesize surface and other secreted PG [32]. As shown in Figure 6B, lpg2KO-DsRed2 parasite uptake was similar to that of LmjLV39-DsRed2 control parasites. Therefore, PGs appear to be dispensable in the initial sensing of parasites by dendrites as well as in the binding of parasites to the cell membrane and subsequent internalization.
DC can, in principle, internalize a large variety of particulate material [33]. Thus, we next determined whether parasite uptake was a specific phenomenon, or whether DDC indiscriminately incorporate particles introduced into the dermis. When inert fluorescent beads were injected intradermally, a minority (∼20%) of DDC revealed intracellular beads at a low number (usually 1 bead/cell) two to four hours after injection (Figure 7A, Table 1, and Videos S10 and S11). When counting the ratio between particles present in the immediate vicinity of DDC (i.e. within half a cell diameter) and intracellular particles, it was obvious that there was a clear preference of L. major uptake (ratio 2.7) as compared to bead uptake (ratio 39.5; Table 1). In addition, we never observed dendrite formation after bead injection.
The bead injection procedure did not result in significant changes in migratory velocity or shape change of the cells (Figure 7B), indicating that the mechanical trauma induced by the inoculation was not sufficient to alter DDC behavior. However, it was conceivable that an inflammatory stimulus may have increased phagocytic activity of DDC. To test this further, we co-injected beads and parasites. Interestingly, there was no increase in bead incorporation, demonstrating that DDC are capable of selectively discriminating between L. major parasites and inert material.
Finally, we determined whether L. major uptake was a primary feature of DDC or was facilitated by other innate immune cells present in early infection. In particular neutrophils have been shown to serve as vectors for Leishmania uptake by macrophages [34]. Depletion of these cells prior to intradermal injection of LmjF-DsRed2 did not change the number of DDC containing parasites as compared to controls (Figure 7C). While these results do not exclude a role of neutrophils in the defense against this pathogen, they suggest that parasite phagocytosis by DDC is independent of these cells.
DC are considered gatekeepers in the defense against intruding pathogens. While DC responses to microbes have been studied in great detail in vitro and in cells isolated ex vivo, very little is known about their interactions in the context of intact tissues in real time. The present study aimed to visualize, in a dynamic manner, the behavior of DDC in normal skin and in response to a defined pathogen. Using 2P-IVM, we found that, under homeostatic conditions, DDC were actively crawling through the dermal interstitial space. Remarkably, upon introduction of the protozoan parasite L. major, DDC transformed into stationary, dendritic-shaped cells that were capable of rapid parasite uptake through flexible dendritic processes. Together, our findings define the microenvironmental context of DDC-pathogen encounter in the earliest phase of cutaneous immune responses.
That DDC migrate in the steady-state was unexpected, as other DC populations, such as DC in the gut epithelium and the epidermis have been found to be immobile, or in the case of the T cell area, very slow moving [4],[8],[19],[22]. It is likely that the specific cellular motility patterns adopted by these individual DC populations serve to optimize their functions in their respective microenvironments. For instance, epidermal LC are in continuous close contacts with surrounding keratinocytes. The paucity of extracellular space may not require, or may not allow, movement of the cells for their immunosurveillance function. Rather, soluble antigens percolating through the extracellular epidermal space or signals from neighboring keratinocytes and/or adjacent LC may be sensed by the communicating dendrite network in this environment. DC in the LN T cell zones are characterized by extensive motions of their dendrites, which may be important for sensing of antigens filtering through the conduit system of this organ, and for establishing contacts with naïve T cells [19]. As compared to the epidermis, DDC are localized within the much more extensive dermal space, which, at the same time, contains considerably lower densities of resident cells, primarily fibroblasts. Thus, while keeping in mind that other tissue resident cells were not visualized in our study, DDC appeared as isolated cells embedded within the network of dermal ECM fibers. They were also found to be morphologically distinct from LC, i.e. they did not exhibit dendrites under non-inflammatory conditions. Therefore, the observation of their continuous crawling indicates a fundamental difference in tissue screening as compared to LC as well as DC in the T cell areas of LN. Since signals from intercellular communication by DDC with other dermal cells may be less abundant than for LC in the epidermis or DC in LN T cell areas, spontaneous DDC migration guarantees access to every corner of this specific microenvironment regardless of the activation state or potential damage of other resident cells during infection. Consequently, this ensures the rapid detection of intruding microbes and the subsequent immediate response to danger signals.
Morphologically, DDC appeared to migrate in an amoeboid fashion, similarly to what has been described for T cells in the extravascular space [35],[36]. Thus, crawling DDC exhibited an anterior-posterior asymmetry reflected by the formation of lamellipodia and uropods. This may suggest that similar molecular cues responsible for interstitial T cell migration, for example surface receptors involved in communication with the environment as well as intracellular molecules, mediate DDC locomotion. We found that blocking of Gαi protein-coupled receptors with PTX significantly reduced the displacement of DDC, implying chemoattractant receptors, such as chemokines or lipid mediators, in this process. This is consistent with recent 2P-IVM studies demonstrating that PTX inhibited the migration of naïve T cells within the LN paracortex, and that CCR7 is, at least partially, involved in this process [37]–[39]. However, the T cell zone of LN contains the fibroblastic reticular cell (FRC) network, which provides the structural backbone of this particular microenvironment. Elegant imaging experiments by Germain's group have shown that the FRC network acts as a scaffold for migrating naïve T cells [40]. A similar cellular structure does not exist in the dermis, raising the question as to how migrating DDC orient themselves within the dermis. It is conceivable that interactions with the extracellular matrix, primarily collagen fibers, are responsible for this process. Indeed, high resolution imaging has shown the intimate contact between DDC and the ECM (Figure 3), and it is likely that chemoattractants are deposited along these structures. Future studies will address the role of specific adhesion receptors, such as integrins or the hyaluronan receptor CD44, as well as specific chemoattractant receptors in these interactions.
The fact that DDC seemed to survey the dermis made us wonder whether they were indeed capable of detecting microorganisms introduced into the dermis. We chose the protozoan parasite L. major as a model pathogen, which is ideal in this context because, firstly, the parasite is directly deposited in the dermis during natural infection by sand flies, and secondly, the parasite is of sufficient size to be detected by 2P microscopy both extra- and intracellularly. Furthermore, the innate and adaptive immune responses against Leishmania spp. have been characterized in great detail in the past, even though controversy exists as to whether DC themselves are infected by the parasite during early infection (reviewed in [11],[13],[25]). While the exact number of parasites transferred during sand fly bites is not known, inoculation of as few as 100 metacyclic parasites is sufficient for establishing an infection [41]. Although we could observe parasite uptake by YFP+ DDC by injecting as few as 2×104 parasites (data not shown), this was technically challenging as only very few parasites and DC could be visualized in situ when using such low numbers. Thus, for the experiments in the present paper, 1–2×105 parasites were used in order to obtain data for proper statistical analysis. It should further be pointed out that the use of small volumes (in the range of 1–2 µl) for intradermal injection was critical, as larger volumes (particularly >5 µl) resulted in the disruption of the local microenvironment. This was evidenced by a disturbance of ECM fibers due to excess fluid (edema) and a migratory decrease/arrest of DDC within the injected ear, even after injection of saline solution without an inflammatory stimulus (data not shown). In contrast, using our injection protocol, we did not observe a migratory or morphological change of DDC imaged ∼50–200 µm away from the injection site under control conditions (Figures 5E and 7B). This result bears consideration not only for imaging studies, but for any situation in which the function of DDC is studied.
In our intradermal infection model, we found that the majority of DDC picked up one or more L. major parasites shortly after inoculation. This was consistent when using two independent L. major strains, supporting the hypothesis that DDC are indeed capable of detecting this protozoan parasite in vivo. Interestingly, after the introduction of parasites, DDC underwent a morphological transition into bona fide DC-shaped cells. Strikingly, parasites appeared to be taken up by long, motile pseudopods (Videos S8 and S9). In vitro infection models of macrophages demonstrated that parasites initially adhered to the cell membrane in a non-random orientation, i.e. preferentially with either the tip or the base of their flagellum [29]. Subsequently, the parasites were engulfed by pseudopods wrapping around the parasites (“coiled phagocytosis” [42]). We found that dendrite extension was sometimes preceded by parasite contact with the cell membrane, while at other times no visible contact was obvious. However, the level of 2P-IVM resolution did not always allow for unequivocal visualization of the parasite flagellum. Therefore, it is conceivable that physical contact is the main trigger of DDC dendrite extension observed in the context of Leishmania infection.
Recently it has been suggested that DDC are composed of two separate subpopulations, i.e. the major langerin− subset and a small langerin+ subset [43]–[45]. Langerin+ DDC are distinct from in-transit LC, and have been shown to be capable of inducing cutaneous hypersensitivity reactions independently from langerin− cells. However, these cells are very rare (2–10% of DDC), and it is unclear whether their functions are different to langerin− cells. Since in our experiments 55–70% of all DDC are infected by L. major, the vast majority will represent langerin− cells. Together with previous studies showing that langerin− DC in draining LN present Leishmania antigens to T cells, we therefore speculate that langerin− DDC are the major players in this scenario. Ablation of langerin+ DDC using genetic approaches will enable definitive answers to this question.
What are the mechanisms of parasite recognition by dendrites? In the intestine, subepithelial DC have been found to extend processes between epithelial cells into the gut lumen, often revealing a spherical shape (“balloon bodies”) [22],[46]. While these processes were capable of capturing bacteria in the gut lumen in a passive manner, this appeared to be a rare event. Importantly, sampling of gut material was non-discriminatory, i.e. DC did not distinguish between inert beads and bacteria [22]. In our study, we found that inert material (beads) alone or co-injected with parasites was largely ignored by DDC. In addition, when we injected fluorescently-tagged Bacillus Calmette-Guérin (BCG), we found that two to four hours after inoculation only ∼30% of DDC contained single internalized BCG, comparable to the results using beads (Figure S4 and data not shown). Furthermore, under these conditions we did not observe the transition of DDC into highly dendritic cells, even when they contained bacteria (Figure S4 and data not shown). Together, these results suggest that L. major induces a specific change in DDC in vivo (i.e. pseudopod formation), and may indicate the involvement of specific surface receptor(s) in this process. Previous studies have shown that Fc receptors and complement receptors are involved in Leishmania uptake by phagocytic cells. However, the molecular cues recognized on parasites are not well understood. Our studies have shown that PGs are not involved in parasite uptake by DDC. The use of parasite strains deficient in a variety of other structural and metabolic genes may, in the future, identify the molecular requirements of parasites to be sensed by dendritic processes.
Another key observation from this study was the rapid transformation of migratory DDC into sessile DDC after exposure to microbial products, such as LPS. In addition, both infected and uninfected DDC became non-migratory at sites of L. major injection, suggesting that the inflammatory environment induces the change in migratory behavior, rather than parasite uptake per se. This conceivably also reflects a switch in functionality of DDC, i.e. from surveillance to sampling/antigen uptake. Thus, by arresting DDC in close proximity to the site of infection, they form a network of sessile cells “primed” for uptake of microbes present at the site. That these states are indeed distinct from each other is further reflected by the fact that PTX treatment interfered with DDC translocation, but not with parasite uptake. These findings raise the question as to the fate of DDC infected early during parasite infection. We have noted that DDC loaded with Leishmania remained relatively sessile over a period of up to ∼6 hours post-inoculation (unpublished observation). When imaging at later time points (∼20 hours post infection), there were numerous YFP+ cells present within the dermis. While these cells showed a similar non-migratory phenotype as cells at earlier timepoints, the number of parasite-containing DDC decreased (unpublished observation). Nevertheless, parasites were still present in the dermis, presumably within other cells (unpublished observation). This may suggest that infected DDC leave the dermis at this stage in order to migrate to draining LN, and that these cells may be replaced by newly immigrating DC or their precursors from the bloodstream. Indeed, previous studies have shown that infected DC arrive in draining LN around 24 hours after infection [5]. Alternatively, parasites within DDC in vivo may simply lose fluorescence over time possibly due to an inability to survive for prolonged periods of time within these cells. Future studies will address potential interactions of infected DDC with the lymphatic vasculature in the dermis, and how these interactions are regulated at the molecular level.
Anti-mouse CD11b, CD11c, CD45.2, F4/80, I-Ab (all from BD Biosciences), Langerin (Dendritics, Lyon, France) and Moma-2 (Abcam, Cambridge, UK) antibodies were used for flow cytometry analysis of epidermal and dermal cell suspensions.
CD11c-YFP mice [19] (kind gift of Dr. Michel Nussenzweig) on a C57BL/6 background (10 generations) were bred in the animal facility of the Wistar Institute and the Centenary Institute. Animal experiments were performed with approval of the Institutional Animal Care and Use Committees at both institutions. To generate fluorescent protein expressing L. major parasites, the gene encoding the red fluorescent protein DsRed2 was PCR amplified from pDsRed2 (Clontech) with primers that added BamHI sites to both ends. The PCR product was cut with BamHI and ligated into BglII site of pIR1SAT yielding pIR1SAT-DsRed2 (strain B4787). After SwaI digestion, it was introduced into L. major strain Friedlin V1 (MHOM/IL/80/Friedlin) by electroporation [47]. DsRed2-expressing LV39 clone 5 (Rho/SU/59/P) and its LPG2-deficient derivative were generated by stable transfection of Swa I-cut pIR1-SAT-LUC-DsRed2 (B5947). This plasmid was obtained by ligating the NruI-SalI DsRed2 fragment from pIR1SAT-DsRed2 (B4787) into SalI+NruI digested pIR1SAT-LUC (B5037). Clonal transfectants were obtained and screened for bright red fluorescence and virulence in mouse infections (data not shown). One clone of each strain was selected for work here (L. major FV1 SSU:IR1SAT-DsRED2(b), LmjF-DsRed2; LV39 SSU:LUC:DSRED2, LmjLV39-DsRed2; LPG2KO SSU:LUC:DSRED2, lpg2KO-DsRed2). Promastigotes were grown in complete M199 as described previously [47]. Red fluorescent protein expressing BCG was generated by transforming BCG Pasteur with plasmid pSMT3:mCherry (a kind gift of Dr Wilbert Bitter, VU University Medical Center, Amsterdam, the Netherlands) as previously described [48]. Hygromycin-resistant colonies were selected on Middlebrook 7H11 medium (Difco Laboratories, Detroit, MI, USA) and expanded in liquid Middlebrook 7H9 medium. Fluorescent colonies were selected by flow cytometry.
Epidermal and dermal cell suspensions were prepared as described previously [49] with some modifications. In brief, ear tissues were incubated in trypsin (0.5%) in HBSS buffer (Invitrogen) for 1 h at 37°C. For CD11c staining, we made use of dispase (5 U/ml) instead of trypsin. After enzyme incubation, epidermis was separated from dermis. To obtain single cell suspensions, epidermal sheets were passed through a wire mesh, and dermal sheets were further digested with collagenase D for 1 hour.
For Gαi protein-coupled receptor inhibition experiments, CD11c-YFP mice were injected intravenously with PTX (30 ng/g bodyweight) in saline solution. For LPS experiments, CD11c-YFP mice were injected intravenously with 50 µg of LPS. 2P-IVM was performed at various time points after the injections. For neutrophil depletion, CD11c-YFP mice were injected i.p. with 250 µg of anti-Gr-1 antibody or rat IgG as control 24 hours before the inoculation of L. major promastigotes. Splenocytes from these mice were examined by flow cytometry to ensure the depletion of neutrophils at the end of imaging (data not shown). In order to reduce autofluorescence from hairs, hair was removed from the ears for all imaging experiments [50]. Control experiments without the use of hair remover revealed identical migratory behavior of DDC (Figure S5).
Mice were anesthetized by intraperitoneal injection of Ketamine/Xylazine (80/10 mg/kg). 1–2×105 stationary phase promastigotes in 1.5 µl of saline solution were injected intradermally using a 33 gauge Hamilton syringe. This procedure was performed under a stereoscopic microscope. For the bead experiments either 2.5×105 FluoSphere microspheres (2 µm, Invitrogen) or 2.5×105 microspheres plus 2.5×105 FVI LmjF parasites were injected. For the BCG experiments, ∼2×105 BCG were injected intradermally. As an additional control, the fluorescent dye SNARF-1 (10 µg/ml) was injected intradermally. Images were typically acquired 50–200 µm from the injection site.
Anesthetized mice were placed onto a custom-built stage to position the ear on a small metal platform for 2P imaging. The ear was immersed in saline/glycerol (70∶30, vol∶vol) and covered with a coverslip. The temperature of the platform was maintained at 36°C, while the body temperature was regulated at 37°C through a heating pad placed underneath the mouse. Two-photon imaging was performed on a Prairie Technology Ultima System or a LaVision Biotec TrimScope equipped with a 40× (NA 0.8) water immersion objective [35]. Both setups included four external non-descanned dual-channel reflection/fluorescence detectors, and a diode pumped, wideband mode-locked Ti:Sapphire femtosecond laser (Coherent Chameleon or Spectra-Physics Mai Tai HP). The ear skin was exposed to polarized laser light at a wavelength of 950–960 nm. Three-dimensional (x,y,z) images of the ear skin were acquired (2–5 µm spacing in z-axis over a total distance of 10–25 µm) every 30 s for a total observation period of 1–2 hours. Images acquired were then transformed into time sequence movies using Volocity software (Improvision). Mean migration velocities, cellular displacement, and confinement ratios (total length of track divided by distance between starting and end point) were manually tracked and calculated for 15′ or 30′30″ as described previously [35]. Measurements were typically performed on 31 or 62 consecutive frames of the video. Cells were defined as immobile if the mean velocity was less than 2 µm/min [19]. To quantify the number of DC with internalized parasites, beads or BCG, images from 3D reconstructions of inoculated skin were examined for the colocalization of red signals (L. major, beads or BCG) and yellow signals (DC).
For comparisons, the Student's t test (normally distributed) or the Mann-Whitney test (not normally distributed) or one-way ANOVA were used. A difference was considered significant if P<0.05.
|
10.1371/journal.pgen.1004748 | MMS Exposure Promotes Increased MtDNA Mutagenesis in the Presence of Replication-Defective Disease-Associated DNA Polymerase γ Variants | Mitochondrial DNA (mtDNA) encodes proteins essential for ATP production. Mutant variants of the mtDNA polymerase cause mutagenesis that contributes to aging, genetic diseases, and sensitivity to environmental agents. We interrogated mtDNA replication in Saccharomyces cerevisiae strains with disease-associated mutations affecting conserved regions of the mtDNA polymerase, Mip1, in the presence of the wild type Mip1. Mutant frequency arising from mtDNA base substitutions that confer erythromycin resistance and deletions between 21-nucleotide direct repeats was determined. Previously, increased mutagenesis was observed in strains encoding mutant variants that were insufficient to maintain mtDNA and that were not expected to reduce polymerase fidelity or exonuclease proofreading. Increased mutagenesis could be explained by mutant variants stalling the replication fork, thereby predisposing the template DNA to irreparable damage that is bypassed with poor fidelity. This hypothesis suggests that the exogenous base-alkylating agent, methyl methanesulfonate (MMS), would further increase mtDNA mutagenesis. Mitochondrial mutagenesis associated with MMS exposure was increased up to 30-fold in mip1 mutants containing disease-associated alterations that affect polymerase activity. Disrupting exonuclease activity of mutant variants was not associated with increased spontaneous mutagenesis compared with exonuclease-proficient alleles, suggesting that most or all of the mtDNA was replicated by wild type Mip1. A novel subset of C to G transversions was responsible for about half of the mutants arising after MMS exposure implicating error-prone bypass of methylated cytosines as the predominant mutational mechanism. Exposure to MMS does not disrupt exonuclease activity that suppresses deletions between 21-nucleotide direct repeats, suggesting the MMS-induce mutagenesis is not explained by inactivated exonuclease activity. Further, trace amounts of CdCl2 inhibit mtDNA replication but suppresses MMS-induced mutagenesis. These results suggest a novel mechanism wherein mutations that lead to hypermutation by DNA base-damaging agents and associate with mitochondrial disease may contribute to previously unexplained phenomena, such as the wide variation of age of disease onset and acquired mitochondrial toxicities.
| Thousands of mitochondrial DNA (mtDNA) per cell are necessary to maintain energy required for cellular survival in humans. Interfering with the mtDNA polymerase can result in mitochondrial diseases and mitochondrial toxicity. Therefore, it is important to explore new genetic and environmental mechanisms that alter the effectiveness and accuracy of mtDNA replication. This genetic study uses the budding yeast to demonstrate that heterozygous strains harboring disease-associated mutations in the mtDNA polymerase gene in the presence of a wild type copy of the mtDNA polymerase are associated with increased mtDNA point mutagenesis in the presence of methane methylsulfonate, a known base damaging agent. Further observations suggest that the inability of disease-associated variants to replicate mtDNA resulted in increased vulnerability to irreparable base damage that was likely to result in mutations when replicated. Also, this study showed that trace amounts of the environmental contaminant cadmium chloride impairs mtDNA replication but eliminates damage-induced mutagenesis in the remaining functional mitochondria. This interplay between disease-associated variant and wild type polymerase offers new insights on possible disease variation and implicates novel environmental consequences for compound heterozygous patients.
| Mitochondrial DNA (mtDNA) maintenance is necessary for the majority of ATP production in eukaryotic cells. The inability to properly replicate mtDNA potentially impacts human health in several ways. The premature aging phenotype of POLG exonuclease deficient mice indicates that increased mtDNA mutagenesis can be detrimental [1]–3. Also, mutations in genes encoding the mitochondrial replisome, including DNA polymerase γ (pol γ, encoded by POLG), contribute to mitochondrial diseases characterized by mtDNA depletion, deletions, or point mutations [4]–[14]. Additionally, environmental changes can modify mitochondrial biology and potentially impact health. Chain-terminating nucleotide analogs used in anti-viral therapy impair mtDNA replication and can result in mitochondrial toxicity [15]. Antioxidants and exercise have been shown in model systems to improve mitochondrial function and suppress the premature aging phenotype associated with increased point mutations and deletions [3], [14]. Therefore, environmental changes can be important for mitochondrial function, and the mechanisms that cause mtDNA mutations warrant further study.
Currently hundreds of POLG mutations have been identified in patients with mitochondrial disease such as Alpers syndrome, progressive external ophthalmoplegia, and ataxia-neuropathy syndrome (mutations listed in http://tools.niehs.nih.gov/polg/) [16]. Pol γ-related mitochondrial diseases display a wide variety of severities. For instance, Alpers syndrome manifests in infants and young children, and these patients rarely live through their first decade of life [17]. Alternatively, patients with progressive external ophthalmoplegia (PEO) and sensory ataxia neuropathy, dysarthria, and ophthalmoparesis (SANDO) often are asymptomatic until>20 years of age [4], [18].
The catalytic subunit of pol γ contains DNA polymerase, 3′-5′ exonuclease, and 5′ dRP lyase activities, with known discrete polymerase and exonuclease domains [19]–[22]. Among the POLG mutations associated with mitochondrial disease, many have been characterized biochemically and shown to disrupt polymerase activity [5]–[10], [13], [14], [23]–[27]. POLG polymerase variants H932Y, R943H, and Y955C alter dNTP-interacting side chains and are associated with less than 1% polymerase activity [26]. Polymerase variants G848S, T851A, R852C, and R853Q also reduce polymerase activity to <1% of wild type activity; in addition, G848S also exhibits a DNA-binding defect [24]. Although mutagenic effects of point mutations that disrupt exonuclease activity have been well established, disease-associated mutations in the human exonuclease domain surprisingly do not disrupt the exonuclease activity in Mip1 [9], [10]. These disease associated exonuclease mutations have not been studied in the human enzyme.
S. cerevisiae has been useful to characterize Pol γ functionality with mutations that alter amino acids within conserved stretches between human POLG and yeast MIP1, most of which are in the polymerase domain [28], [29]. Mitochondrial functionality and mtDNA point mutagenesis have been determined in various mutants using assays that measure frequency of petite colony formation (ie, lacking mitochondrial function either with [rho−] or without mtDNA [rho0]) and erythromycin resistance, respectively [30]. For instance, mutations that alter the catalytic aspartates (eg, Asp171 and Asp 230 in yeast) in the exonuclease domain are associated with 1440-fold and 160-fold increases in point mutagenesis [31]–[33] and deletions between direct repeats in mice, respectively [32], [33]; the corresponding increases in yeast are 2000-fold for point mutagenesis [31] and 90-fold for deletions between direct repeats [12], [33]. These increases in mtDNA mutagenesis establish Asp171 and Asp 230 as critical domains for “proofreading” against misinsertions. Surprisingly, disease-associated mutations in the exonuclease domain are associated with only modest increases in mutant frequency, suggesting that exonuclease activity is functionally sufficient to correct misinsertions [9], [10]. Many of the disease-associated mutations to Mip1 eliminated the ability to replicate mtDNA and were associated with petite colony formation, including human variants R807C, R807P, R853W, N864S, G923D, H932Y, K947R, G1076V, R1096C, S1104C, and V1106I and Alpers-associated mutations G848S, T851A, R853Q, D930N, A957P, P1073A, and R1096H [9]. However, several strains that contain variants, including R853Q and Q308H (R656Q and Q264H in yeast) coexpressed with wild type MIP1 to maintain mtDNA showed significant increases in mutagenesis, and no mechanism has been described for this increase [9].
Environmental agents have also been shown to affect mitochondrial DNA replication both positively and negatively. The presence of antioxidants such as MitoQ and dihydrolipoic acid have been shown to improve mitochondrial function in mutants with disease-associated polymerase domain mutations by salvaging reactive oxygen species [14]. These results suggest an increase of oxidative damage in mtDNA in model systems with defective pol γ, a hypothesis supported by the increased levels of 8-oxo-dG in the mtDNA of a transgenic mouse model that overexpressed the Y955C mutant variant in cardiac tissue [34]. Methyl methanesulfonate (MMS) is an alkylating agent that is associated with increases in mtDNA base damage [35]. Interestingly, in embryonic fibroblasts, 2 mM MMS was associated with persistent mtDNA damage but not with loss of mtDNA or mitochondrial function [36]. In yeast, repair of alkylation damage of mtDNA by MMS involves Apn1 nuclease [37], and Ntg1 [38]. Also, chronic exposure to trace amounts of the known human carcinogen, cadmium chloride, resulted in loss of mitochondrial function [39]. Because cadmium also results in extreme nuclear hypermutability [39], the possibility that cadmium alters mtDNA replication warrants further study.
Base excision repair is active in yeast and human mitochondria and protects cells against alkylation damage [40]. However, lesions on single-stranded DNA are not substrates for base excision repair because there is no complementary strand with which it can reanneal and are therefore highly mutagenic [41]–[43]. The proposed model of asymmetrical mtDNA replication of human mtDNA suggests that single-stranded mtDNA is exposed even in optimal conditions and could be more vulnerable under conditions of decreased replication efficiency [44]. To test whether reduction in mtDNA replication efficiency could leave the cell vulnerable to mutagenic base damage, mtDNA mutagenesis in previously characterized disease-associated mutants were tested in the presence of MMS.
Mutations in mip1 that result in changes in conserved amino acids previously have been shown to cause defective mtDNA replication and, in some cases, increased mtDNA mutagenesis when coexpressed with wild type MIP1 [9]. These experiments were performed in haploid heteroallelic strains with intact chromosomal wild type MIP1 and one of 31 mutant mip1 alleles on a centromeric plasmid with the endogenous promoter. This study interrogated some of the heteroallelic strains and newly created diploid heterozygotes to determine mtDNA point mutagenesis of the gene encoding the 16S ribosomal subunit that confers resistance to erythromycin and the fraction of cells unable to grow on glycerol which requires mitochondrial function.
To test whether the presence of the catalytically defective mutant variant increases the vulnerability of mtDNA to base damage and mutagenic replication by the wild type polymerase, heteroallelic S. cerevisiae strains expressing either Q264H, R656W, R853H, or wild type MIP1 on a centromeric plasmid and chromosomal wild type MIP1 were grown in the presence of sublethal concentrations (3 mM) of MMS (see Table 1 for list of all genotypes). None of these mutant proteins are capable of maintaining mtDNA without the presence of wild type Mip1 [9]. MMS exposure caused a modest 2-fold increase in mutagenesis in the wild type control as compared to no exposure to MMS, whereas a greater increase in mtDNA mutagenesis—17-fold, 11-fold, and 6-fold—was observed in strains expressing Q264H, R656W, and R853H, respectively (Figure 1 and Table 1). Absolute mtDNA mutant frequencies after MMS exposures were associated with 30-fold, 18-fold, and 7-fold increases in strains with Q264H, R656W, and R853H mutant variants, respectively, compared with that of the wild type strain. The control strain with wild type MIP1 on both the centromeric plasmid and the chromosome was associated with only 2.7-fold increase in MMS-induce mutagenesis compared with no MMS exposure.
To avoid the possibility of multicopy expression of the Mip1 variant from the plasmid, heterozygotes were created with mutations that encode Mip1 with defective exonuclease activity [31] or amino acid variants Q264H, R656Q, G651S, or D891A (Table 1). D891A is not associated with mitochondrial disease, but the conserved aspartate in human POLG (Asp1135) is essential for binding catalytic Mg2+ in the active site [19]. Alanine substitution of this equivalent residue in other human DNA polymerases has been shown to disrupt binding of the catalytic Mg2+, eliminating DNA polymerase activity but not binding to DNA binding [45]. With the exception of Q264H, biochemical characterizations have been reported for the remaining mutant variants [24]. In agreement with previous observations [9], disruption of Mip1 exonuclease activity increased mutagenesis; however, there was no additional increase in MMS-induced mutagenesis (Figure 2 and Table 1). Compared with wild type, Q264H, R656Q, and D891A heterozygotes were associated with 7-fold, 8-fold, and 19-fold increases in absolute MMS-induce mutant frequency. In fact, the resulting increases in mutant frequency were approximately equal to or greater than that of the exonuclease defective variant (Figure 2). Strikingly, MMS exposure of the D891A variant resulted in an approximately 3-fold increase in mutant frequency compared with that of the exonuclease deficient variant. These results demonstrate that catalytically inactive or less active polymerases, whether generated through a site-directed mutation or a disease-associated mutation, participate in a mechanism that causes MMS-induced mutations.
Mip1 variants R656Q and G651S are homologous to human disease variant R853Q and G848S, respectively, which both exhibit ≤1% catalytic activity [24]. However, G848S uniquely displayed an approximately 5-fold reduction in DNA binding [24]. Interestingly, Mip1 G651S variant resulted in fewer MMS-induced mutations compared with the other polymerase variants, suggesting that DNA binding may be important for the mechanism. To test this hypothesis, a R656Q/G651S double mutant was created, and MMS-induced mutagenesis was compared with each single mutant. Mutagenesis after exposure to MMS in the double mutant was indistinguishable from the G651S variant and lower than in a single R656Q mutant (Figure 2), suggesting that DNA binding is an important component of the R656Q mutator effect.
To test whether Mip1 mutant variants participate in the bulk of mtDNA replication, mutagenesis was measured in heterozygous diploids that had one wild type MIP1 allele and one allele containing mutations that disrupt exonuclease activity and encode the Q264H and G651S variants in cis. Unlike exonuclease-deficient Mip1 without other mutations, eliminating exonuclease activity did not increase mutagenesis in either the Q264H or G651S variant, with or without MMS exposure (Figure 3 and Table 1). In fact, mutation frequency unexpectedly decreased when the exo− and Q264H were in cis. These results suggest minimal contributions of Q264H and G651S mutant variants to mtDNA replication.
Resistance to erythromycin is conferred by one mutation at any of the following nucleotides: 1950 (G to T or G to A), 1951 (A to T, A to G, or A to C), 1952 (A to T or A to G), 3993 (C to G), or an insertion of G between nucleotide 1949 and 1950 of the 21S rRNA gene (Gen Bank accession number L36885) [31], [46], [47]. To determine if there was a change in the spectrum of mutations associated with MMS exposure, PCR fragments containing nucleotides 1797–1995 and 3895–4107 of the 16S ribosomal subunit gene from erythromycin resistant mutants were sequenced. Only one mutant was taken from each original culture to ensure that each mutation represented a separate event. In the absence of MMS, this and prior studies [46]–[48] demonstrate that A:T→G:C and A:T→T:A were the most frequent mutations (Figure 4 and Table S1). In this and previous studies, C to G mutations were not detected in wild type strains and were detected in only 5% of Δrev1 strains [46]. Interestingly, we found that exposure to MMS was associated with a significant change in mutational spectrum, wherein the most common mutation was C:G→G:C transversions in both the wild type strain (40%) and Q264H heteroallelic strain (45%). G:C→A:T was the only mutation other than C:G mutation detected, but it was only detected in the wild type strain and very low levels (7%). These results suggest that cytosine or guanine is especially sensitive to methylation by MMS, leading mostly to misincorporated cytosines or guanines, similar to previous studies [43].
Increased mutagenesis can arise by disrupting exonuclease activity and/or increasing the frequency of nucleotide misincorporation events. Previous work demonstrated that mitochondrial deletions between direct repeats of 21 nucleotides were rare events that were suppressed by exonuclease activity [12]. To test whether MMS promotes deletion formation, haploid deletion reporter strains that were heteroallelic for wild type, exonuclease-deficient, or Q264H variants of Mip1 were used to measure frequency of deletions between 21 nucleotide direct repeats that flank an ARG8 insertion in the mitochondrial genome. Frequency of deletions between direct repeats was increased (40-fold) in the strain with the exonuclease-deficient Mip1 variant but was not significantly different in the strain with the Q264H variant compared with wild type (Figure 5 and Table 1). MMS had no significant effect on deletion formation in any of the three strains, suggesting that the effect of MMS is specific to point mutations.
Mutations associated with MMS exposure occur in strains with mutants that affect mtDNA replication, suggesting that the mechanism requires suboptimal mtDNA replication. Therefore, it is possible that an environmental agent that reduces mtDNA replication may also be associated with MMS-induced mutagenesis in wild type cells. This was tested by treating wild type and mutant mip1 strains with CdCl2. Exposure to 3 µM CdCl2 was associated with increased petite formation frequency of about 30% with stepwise increases at 4 and 5 µM of about 60% and 80%, respectively (Figure 6) in both wild type and Q264H mutants. Exposure to 5 µM CdCl2 was associated with 3.6-fold reduction in mtDNA among rho+ cells, from 30.6±4.0 copies per cell without CdCl2 exposure to 8.5±0.9 copies per cell with 5 µM CdCl2, suggesting that trace amounts of CdCl2 are associated with mtDNA depletion.
Mitochondrial DNA mutagenesis was assayed in homozygous wild type diploid cells and Q264H heterozygotes to test if MMS-induced mtDNA mutagenesis occurs with CdCl2. Exposure to 4 µM CdCl2 had no effect on mtDNA mutagenesis in either the wild type or Q264H mutant strains (Figure 7). Therefore, the mutagenic effect of CdCl2 is specific to nuclear DNA [39]. Exposure to both MMS and CdCl2 resulted in no increase in mtDNA mutagenesis. However, the mutagenic effect of MMS observed in the heteroallelic Q264H strain was completely suppressed by 3 µM or 4 µM CdCl2. The lack of mtDNA mutagenicity and the suppression of the MMS-induced mtDNA mutagenesis by trace amounts of CdCl2 were recapitulated in heterozygotes expressing the D891A mutant variant (Table 1). Although reduced efficiency of mtDNA replication by a mutant variant is associated with MMS-induced mutagenesis, these results suggest that processes that reduce mtDNA replication suppress MMS mutagenesis.
This study demonstrates a novel mechanism of MMS-induced mtDNA point mutagenesis, mostly C:G→G:C transversions, in heterozygous strains with a disease-associated mutation that disrupts polymerase activity. The frequency of the mutagenesis appeared to be modulated by the activity of the mutant variant in that the possible DNA binding defect observed in the human homologue to G651S reduced MMS-induced mutagenesis. This study showed that alleles with added exonuclease defect were not associated with increased mutagenesis (with or without MMS exposure), suggesting that the mutant variant replicated little or none of the mtDNA that remained and was propagated in the cell. Finally, chronic exposure to trace amounts (3–5 µM) of CdCl2 resulted in the inability to replicate mtDNA, which surprisingly did not increase but instead suppressed MMS-induced mutagenesis. These results are the first to support a mechanism to understand the interplay between polymerases in heterozygous cells and reveal a novel pathway for environmentally-induced mtDNA mutagenesis.
There are few known pathways that increase mtDNA mutations in yeast or humans. Mutations that disrupt the exonuclease activity of the mtDNA polymerase have been shown to cause increased mtDNA mutagenesis. One study used a reversion assay in yeast to identify several mitochondrial mutators including pos5, a gene that encodes an NADPH kinase [49]. Other genes associated with increased mtDNA mutagenesis—such as hap2, fen1, and ntg1—have been identified, but their effect on mtDNA mutagenesis has been modest or requiring long incubation times [49]–[51]. Even base damaging agents such as H2O2 and MMS are associated with modest increases in mtDNA but only in strains without crucial repair pathways [52].
Previously, disease-associated mutations were shown to increase mtDNA mutagenesis, but these mutations also led to the inability to maintain functional mitochondria because of mtDNA depletion [9]. The increase in mutant frequency in some strains was evident in this study in the unexposed controls (Figures 1 and 2). Therefore, it was difficult to ascertain how mutant polymerases that in some cases were suggested to have little or no activity could significantly increase mtDNA mutagenesis. Exposing the heteroallelic strains to sublethal doses of MMS resulted in up to 30-fold increases in mtDNA mutant frequency (Figure 1). Interestingly, repeating the experiment in heterozygous strains recapitulated the MMS-induced increase, albeit at a lower frequency (Figure 2). The heteroallelic strain contains the mutant mip1 and its endogenous promoter on a centromeric plasmid that has been previously shown to have 1–2 copies of the gene per cell [13]. It is possible that MMS-induced mutagenesis is sensitive to differences in the number of Mip1 mutant copies. The fact that the heteroallelic strains are haploid whereas the heterozygotes are diploid could suggest that increased copy number of other replication proteins may alter the MMS-induced mutagenesis phenotype, although there is probably no difference in the protein concentration relative to the genome copy number.
MMS-induced mutagenesis was shown in several disease-associated mutants and the polymerase defective mutant but not the exonuclease-deficient mutant. Although Q264H is an exonuclease domain disease-associated variant, it was previously shown to be detrimental to mtDNA replication. Mutations that result in the Mip1 R656Q and G651S variants are homologous to the mutations in the human pol γ thumb domain that were biochemically characterized to have <1% polymerase activity including a 5-fold reduction in DNA binding affinity in the G651S homologue [24]. Interestingly, G651S is associated with reduced mtDNA mutant frequency and suppression of R656Q when the two mutations are in cis. These results suggest that lower DNA binding affinity impedes the mechanism of MMS-induced mutagenesis. Gly651 is in a stretch of amino acids conserved between humans and yeast, making it likely that Gly651 is also involved in DNA binding. However, future studies will be necessary to show that G651S does not possess other characteristics (eg, decreased stability) that impair MMS-induced mutagenesis.
Biochemical evidence suggesting that some disease-associated mutant variants were impaired for mtDNA replication was further supported by the observation that the mutant variants by themselves could not maintain mtDNA [9]. However, it has been unclear to what extent these polymerases function in the cell. It is well known that one of the catalytic aspartates (Asp891 in Mip1) is necessary for mtDNA replication; therefore, D891A is unable to catalyze the polymerase reaction. Interestingly, the heterozygous strain with D891A was associated with the largest increase in MMS mutagenesis, approximately 3-fold more than the exonuclease-deficient strain. Because the generation of mutations requires DNA replication, this result indicates that the wild type polymerase, which is normally accurate, becomes more likely to incorporate the wrong nucleotide in the D891A strain upon MMS exposure. In the case of the disease-associated mutations, it is possible that the mutant variants contribute to the incorporation of the incorrect nucleotide. However, removal of exonuclease activity in cis with Q264H and G651S showed no increase in mutation frequency regardless of MMS exposure and even showed an unexpected reduction of mutagenesis in the Q264H strain. Considering that the mutant frequency in the Q264H/exo− strain was similar to the G651S mutant frequency, this study cannot discount the possibility that the combination of the two mutations may have similar characteristics to G651S (eg, lower DNA binding affinity). Regardless, these results suggest that the mutant polymerase is not contributing directly to the mutations that drive the mutant frequency. It is possible that the mutant variants replicate mtDNA molecules that may be selected against or are not propagated, possibly because of incomplete replication.
Resistance to erythromycin is associated with a limited mutational spectrum that rarely includes C:G→G:C transversions [46]–[48]. Interestingly, exposure to MMS dramatically changed the mutation spectrum such that 40–45% of the mutations were C:G→G:C transversions. These mutations could either arise from a cytosine incorporated opposite a cytosine or a guanine incorporated opposite a guanine. Previously, MMS exposure of artificially formed or random ssDNA in yeast was associated with increased frequency of all 3 kinds of cytosine substitutions, C→T, C→G and C→A [42], [43]. The mutation spectra and strand bias suggested that N3-methyl cytosine is the prominent mutagenic lesion caused by MMS lesion in the yeast nuclear ssDNA [42], [43]. Therefore, it is possible that C:G→G:C transversions that predominate after MMS exposure result from cytosine incorporation opposite of a methylated cytosine.
These combined results support a model (Figure 8) wherein disease-associated mutant polymerase variants bind to and temporarily stall mtDNA replication, an event that could result in ssDNA intermediates (eg, because of polymerase-helicase uncoupling). Although mtDNA base excision repair would normally repair most damaged bases in dsDNA, there are no known repair systems that act on ssDNA in yeast. In the absence of MMS, the endogenous sources of DNA damage (eg, oxidative damage) impart a small but significant amount of DNA damage, whereas MMS magnifies the damage. Either the mtDNA continues to be stalled and degraded (an event that would not yield a mutant colony) or polymerase switching might occur, allowing the wild type polymerase access to the replication fork. In some cases the polymerase would incorporate the incorrect nucleotide leading to the development of a mutation. This model suggests that the mutant polymerase would stably bind DNA but be unable to replicate mtDNA efficiently.
Recently it was shown that the exonuclease domain suppresses mtDNA deletions between 21-mer direct repeats up to 160-fold [12]. One hypothesis could be that MMS also inhibits exonuclease activity which would lead to an increase in mutagenesis. This hypothesis was unlikely because MMS exhibited a modest effect on the diploid wild type strain. This study showed that MMS did not increase the frequency of mtDNA deletion mutants in wild type or Q264H background suggesting that the mechanism that caused MMS-induced point mutations is different than that of mtDNA deletion formation. Furthermore, MMS does not alter the exonuclease activity involved in suppression of mtDNA deletion formation.
A previous report showed that exposure to trace amounts of CdCl2 was associated not only with suppression of nuclear mismatch repair but also increase in petite colony formation frequency [39]. This study recapitulates this finding and further shows that CdCl2 does not significantly affect mtDNA mutant frequency. These results indicate that either there is no efficient mismatch repair system in yeast mtDNA or the amount of CdCl2 needed to observe a defect in mismatch repair is similar to the amount that is associated with a high frequency of dysfunctional mitochondrial. The only mismatch repair homologue associated with yeast mitochondria is Msh1, and it has been proposed to play a role in base excision repair [48]. This study also showed that mtDNA content was reduced in rho+ cells exposed to CdCl2 suggesting that cadmium negatively affects mtDNA replication or maintenance. We tested whether the effect of inhibiting mtDNA replication could mimic the effect of a disease-associated mutation by increasing MMS-induced mutagenesis. Unexpectedly, 3 and 4 µM CdCl2 did not promote MMS-induced mutagenesis in wild type cells. It is possible that CdCl2 does not stall replication but rather affects another replication-related process, such as replication initiation or termination. Even more unexpectedly, CdCl2 suppressed MMS-induced mutagenesis associated with Q264H. It should be noted that mutant frequency is determined among only rho+ cells so increased petite colony formation frequency from CdCl2 exposure does not explain the suppression of mutant frequency. One possible explanation is that the presence of CdCl2 selects against the maintenance of mtDNA molecules that are damaged or stalled either through some direct inhibition of the enzymes involved or as a result of a cellular response to mtDNA stress. Another possibility is that damaged mtDNA is more sensitive to CdCl2-induced inhibition of mtDNA replication, and the replication of these mtDNA molecules are not completed or maintained during the growth of the colony. Therefore, the various combinations of environmental exposures and genetics may be a useful tool to understand different pathways that occur in response to DNA damage.
This study shows a novel gene-environment interaction which greatly increases mtDNA mutagenesis and supports a polymerase-switching mechanism that has not been described in mtDNA replication. The interpretations of this study are limited in humans because there are key differences in yeast mtDNA maintenance compared with mammalian mtDNA (eg, lower mtDNA copy number and high frequency of recombination). Also, the study assumes that the mutant frequency (ie, mutants per culture) is indicative of the mutation frequency (mutations per mitochondrial division). True mutation frequencies would require information on mtDNA kinetics and mitochondrial dynamics that is currently unavailable. However, with the advent of high-throughput genome sequencing, similar studies in mtDNA mutagenesis will be possible in a model system that more closely mimics human mtDNA. Interestingly, this interaction involves heterozygotes containing mutations which are normally associated with disease [53]. It is interesting to consider that if a similar mechanism occurs in mammalian mtDNA, people who are heterozygous for a disease-associated mutation could be sensitive to environmental exposures that would impair mtDNA replication and promote symptoms involved in mitochondrial toxicity or disease.
S. cerevisiae strains were grown at 30°C in YP (yeast extract 1%, peptone 2%) with 2% glucose or glycerol as carbon sources or synthetic complete media. Escherichia coli strains were grown in standard LB media at 37°C. When appropriate, gentamicin and ampicillin were added to YPD (0.2 mg/ml) and LB (0.1 mg/ml), respectively.
The plasmid, pFL39, which contains MIP1 on a centromeric plasmid was previously described [31]. Site-directed mutagenesis of plasmid-encoded MIP1 was performed using the QuikChange Site-Directed Mutagenesis Kit (Invitrogen) as described previously. The construction of plasmids containing ACT1, COX1, and COX2 fragments, used for mtDNA quantitation was previously described.
All S. cerevisiae strains were derived from E134 (MATα ade5-1 his7-2 lys2-A14 leu2-3,112 ura3-52 trp1-289) [54] and its MATa isogenic strain, YH747. Heteroallelic mip1 strains were made by transforming pFL39-MIP1 or a mutant derivative into E134 by selecting for TRP. Strains that measure deletions between direct repeats were created by transforming TRP1-containing plasmids PFL39 [12] containing wild type MIP1 or mip1 encoding an exonuclease-deficient mutant variant (mip1-exo [Het]; JSY114) into trp1::G418 NPY75 (JSY77) [12]. Chromosomal mip1 mutations were created in haploid E134 with wild type MIP1 using a PCR-based delitto perfetto method as previously described. All strains were checked phenotypically for the absence of the CORE casette used in delitto perfetto (ie, screening for Ura− and gentamicin sensitive cells) and were sequenced to confirm the presence of the mutation and the absence of undesired nearby mutations. The mip1 haploid mutant strains were mated with the isogenic E134 with mating type a.
Resistance to erythromycin is conferred by one of several missense mutations in the 21S rRNA gene in mitochondrial DNA [46], [47], [55]. Yeast strains were replica plated onto YPD plates with or without 1–5 mM MMS or 1–5 µM CdCl2 and grown at 30C for 2 days. In most cases 3 mM MMS was the highest exposure that allowed growth and was used for the experiments. After plating on CdCl2, increased chromosomal mutagenesis was confirmed qualitatively by the presence of colonies on media lacking lysine. From these plates, 20–40 independent colonies from each strain were used to inoculate into 4 ml of synthetic media without tryptophan (heteroallelic strains) or YPD (heterozygous strains), and these cultures were incubated to saturation (for 2 days) at 30°C. The cells were plated on YPEG (1.7% ethanol, 2% glycerol) with 4 g/L erythromycin. A small aliquot of 5–10 cultures were used to titer the number of rho+ cells by plating 10−5 dilutions on YPG. Erythromycin resistant colonies were counted after 6 days of incubation at 30°C. The mutant frequency was the median number of erythromycin colonies per 108 rho+ cells plated. To determine the spectrum of erythromycin-resistant mutations, DNA was extracted from one mutant per culture and used for as a template in a PCR reaction to amplify two regions, approximately 200 nucleotides flanking nucleotides 1950 (using 5′-GAGGTCCCGCATGAATGACG and 5′-CGATCTATCTAATTACAGTAAAGC) and 3993 (using 5′-CTATGTTTGCCACCTCGATGTC and 5′-CAATAGATACACCATGGGTTGATTC). The resulting amplified DNA was the template for the sequence reactions.
To measure deletions between direct repeats, all strains were replica plated onto YPD with or without MMS exposure as described above. Independent cultures were grown from at least 20 colonies at 30°C in YP (yeast extract 1%, peptone 2%) with 2% glucose and adenine for 2 days. Appropriate dilutions of samples from the saturated cultures were plated on synthetic complete media lacking arginine to determine total number of cells with mtDNA. The cultures were plated onto YP with 2% glycerol and deletion mutants were counted after 4 days. The mutant frequency was determined as the median number of mutant colonies per 108 Arg+ cells. In all mutagenesis experiments, 95% confidence levels were determined using the method of the median.
Petite frequencies are the frequency of rho− cells (petites) in the total population. Rho−cells are devoid of mitochondrial functions but are not necessarily devoid of mtDNA (rho0). To determine petite frequency in heteroallelic strains, at least 12 fresh transformants per strain were diluted in water and between 200–1000 colonies were plated onto YPD. For monoallelic strains, rho+ cells derived from several tetrad dissections were single colony purified on YPD and then assayed for petite frequency. Cells were incubated at 30°C for 2 days. Rho+ cells were identified either by the accumulation of red pigment as a result of mutations in the ade biosynthetic pathway or by the inability to grow on YPG. Using either or both method, at least 300 colonies per plate were counted, and petite colonies were identified. Frequencies were determined for each plate, and the median number of the frequencies was calculated. 95% confidence levels were determined by using the method of the median (52). For monoallelic strains determined to be 100% petite, no rho+ cells could be isolated from cells derived from 10 different haploid spores.
Mitochondrial DNA copy number was quantified relative to nuclear DNA copy number using real time PCR. Primers and probes designed to specifically amplify within the mitochondrial-encoded COX1 gene and the nuclear-encoded ACT1 gene. Real time PCR reactions using Taqman Universal PCR Master Mix (Applied Biosystems) were performed at 40 cycles of 95 for 30 sec and 50 degrees for 30 sec. Known concentrations of plasmid molecules containing COX1and ACT1 were quantified as a positive control [9] and real time PCR was performed on 7 different dilutions to determine a logarithmic equation of a curve (R2 values>0.98) that represents numbers of molecules as a function of the critical threshold of every reaction. Every reaction was done in triplicate, and three replicates were tested for each experimental condition. Data represent the average ratio (± SEM) of the number of COX1 molecules to the number of ACT1 molecules.
|
10.1371/journal.pbio.1000416 | Two Distinct Mechanisms for Actin Capping Protein Regulation—Steric and Allosteric Inhibition | The actin capping protein (CP) tightly binds to the barbed end of actin filaments, thus playing a key role in actin-based lamellipodial dynamics. V-1 and CARMIL proteins directly bind to CP and inhibit the filament capping activity of CP. V-1 completely inhibits CP from interacting with the barbed end, whereas CARMIL proteins act on the barbed end-bound CP and facilitate its dissociation from the filament (called uncapping activity). Previous studies have revealed the striking functional differences between the two regulators. However, the molecular mechanisms describing how these proteins inhibit CP remains poorly understood. Here we present the crystal structures of CP complexed with V-1 and with peptides derived from the CP-binding motif of CARMIL proteins (CARMIL, CD2AP, and CKIP-1). V-1 directly interacts with the primary actin binding surface of CP, the C-terminal region of the α-subunit. Unexpectedly, the structures clearly revealed the conformational flexibility of CP, which can be attributed to a twisting movement between the two domains. CARMIL peptides in an extended conformation interact simultaneously with the two CP domains. In contrast to V-1, the peptides do not directly compete with the barbed end for the binding surface on CP. Biochemical assays revealed that the peptides suppress the interaction between CP and V-1, despite the two inhibitors not competing for the same binding site on CP. Furthermore, a computational analysis using the elastic network model indicates that the interaction of the peptides alters the intrinsic fluctuations of CP. Our results demonstrate that V-1 completely sequesters CP from the barbed end by simple steric hindrance. By contrast, CARMIL proteins allosterically inhibit CP, which appears to be a prerequisite for the uncapping activity. Our data suggest that CARMIL proteins down-regulate CP by affecting its conformational dynamics. This conceptually new mechanism of CP inhibition provides a structural basis for the regulation of the barbed end elongation in cells.
| Actin is a ubiquitous eukaryotic protein that polymerizes into bidirectional filaments and plays essential roles in a variety of biological processes, including cell division, muscle contraction, neuronal development, and cell motility. The actin capping protein (CP) tightly binds to the fast-growing end of the filament (the barbed end) to block monomer association and dissociation at this end, thus acting as an important regulator of actin filament dynamics in cells. Using X-ray crystallography, we present the atomic structures of CP in complex with fragments of two inhibitory proteins, V-1 and CARMIL, to compare the modes of action of these two regulators. The structures demonstrate that V-1 directly blocks the actin-binding site of CP, thereby preventing filament capping, whereas CARMIL functions in a very different manner. Detailed comparison of several CP structures revealed that CP has two stable domains that are continuously twisting relative to each other. CARMIL peptides were found to bind across the two domains of CP on a surface distinct from its actin binding sites. We propose that CARMIL peptides attenuate the binding of CP to actin filaments by suppressing the twisting movement required for tight barbed end capping. Our comparative structural studies therefore have revealed substantial insights in the variety of mechanisms by which different actin regulatory factors function.
| The actin capping protein (CP) specifically binds to the barbed end of actin filaments with a high affinity and prevents the addition and loss of the monomers at this dynamic end [1],[2]. CP is a heterodimeric protein composed of α- and β-subunits and the molecule displays a pseudo two-fold symmetry due to the resemblance of the tertiary structures between the two subunits [3]. CP caps the filament with its two independent actin binding sites at the C-terminus of each subunit (“tentacles”). The tentacles are functionally non-equivalent: the α-tentacle is more important than the β-tentacle and is responsible for the initial contact with the barbed end [4]. A recent cryo-electron microscopy (EM) study provided a structural model for the barbed end capping by CP [5]. The model depicted the α-tentacle, with its surrounding residues in the β-subunit, wedged between the two end actin protomers, which represents the primary contact between CP and actin. A mutational analysis revealed that three conserved basic residues in this region, CP (α) Lys256, Arg260, and Arg266 (in the chicken α1 isoform), are critical for the barbed end capping [5]. The β-tentacle was predicted to interact with a hydrophobic cleft on the surface of the terminal protomer to stabilize the capping [5].
A growing body of evidence indicates that CP is a key regulator of actin-based lamellipodial dynamics. In vitro, CP is one of the essential proteins required for the formation of the Arp2/3 complex-nucleated branched-actin arrays, which drive lamellipodial protrusion [6]. CP prevents the production of longer filaments and maintains the cytosolic G-actin pool to promote the Arp2/3 complex-based filament nucleation and branching [7]. In mammalian cells, CP depletion leads to the explosive formation of filopodia, rather than lamellipodia [8]. Thus, the local concentration of CP and its affinity to the barbed end are critical determinants of dendritic actin assembly. The dissociation of CP from the barbed end is a rare event (t1/2∼30 min) in actin polymerization assays using purified proteins. However, recent microscopic observations of cultured cells showed that the fluorescent speckle lifetime of CP bound to actin filament network structures is on the order of seconds [9],[10], suggesting that CP does not stably cap the barbed end in living cells.
At present, several molecules have been identified that affect the barbed end capping activity of CP. These regulators can be categorized in two groups: (1) indirect regulators that bind to actin filaments and protect the barbed end from CP and (2) direct regulators that bind CP and modulate its capping activity. Formin is an indirect regulator because it associates with the barbed end and allows filament elongation even in the presence of CP [11]. Ena/VASP is also assumed to antagonize the capping activity without interacting directly with CP [12]. Polyphosphoinositides, such as PIP2, bind directly to CP and reduce the capping activity in vitro [13],[14].
The V-1 and CARMIL proteins are the only direct CP regulatory proteins that have been reported. V-1, also known as myotrophin, is a 13 kDa ankyrin repeat protein that consists of four ankyrin repeat motifs; two full-repeats are sandwiched between additional incomplete motifs at each terminus [15]. V-1 has been implicated in a variety of cellular events, including catecholamine synthesis [16], cerebellar development [17], cardiac hypertrophy [18], and insulin secretion [19]. Although the precise functional roles of V-1 in these processes have not been clarified, it is possible that V-1 acts as a CP regulator in vivo, because V-1 was found to form a complex with CP in primary-cultured cells and cell lines in murine cerebella [20],[21].
CARMIL is a multi-domain protein that reportedly interacts with myosin I, Arp2/3 complex, and CP [22]. Down-regulation of CARMIL resulted in impaired motility in Dictyostelium and mammalian cells [22],[23]. Although CARMIL is a large protein (∼150 kDa), its CP interaction site has been narrowed down to a small region [23],[24], and a ∼20 amino acid sequence in this region [CP-binding motif; LXHXTXXRPK(6X)P] is shared with other proteins, CD2AP, CIN85, and CKIP-1 [25]. All of these proteins (CARMIL proteins) can interact with CP via this consensus motif [25]. CD2AP and its homologue CIN85 are adaptor proteins involved in various cellular processes, such as T-cell activation, apoptosis, and actin cytoskeleton dynamics [26]. CKIP-1 interacts with casein kinase 2 and recruits the enzyme to the plasma membrane [27].
Previous studies have demonstrated that the V-1 and CARMIL proteins inhibit CP in distinct manners. (1) V-1 bound to CP blocks actin filament capping, whereas the CP/CARMIL protein complex has lower barbed end capping activity (KD∼15 nM) than free CP (∼1 nM) [23],[28],[29]. (2) CARMIL acts on the barbed end-bound CP and facilitates its dissociation from the filament (called uncapping activity), but V-1 lacks this activity [23],[25],[28],[29]. (3) The two actin binding sites in CP, the α- and β-tentacles, are not involved in the CARMIL interaction, whereas V-1 recognizes these sites [23],[28]. (4) The CP binding fragment of CARMIL, including the CP-binding motif, has little secondary structure. In contrast, V-1 is a structured ankyrin repeat protein [15],[23].
Although previous studies have revealed the striking functional differences between the two direct CP regulators, the molecular mechanisms by which these proteins inhibit CP remain poorly understood. In particular, the mechanism by which the CARMIL proteins uncap the filament that is tightly bound by CP has remained enigmatic. In this study, we present the crystal structures of CP complexed with V-1 and with peptides derived from the CP-binding motif of CARMIL proteins. Together with biochemical and computational studies, we have elucidated two distinct mechanisms for CP regulation by V-1 and CARMIL proteins—steric hindrance and allosteric restriction of conformational fluctuations.
In this report, we describe the domain movement of CP. To facilitate the description, we refer to the structural motifs of CP as “N-stalk,” “α-globule,” “β-globule,” “central β-sheet,” “antiparallel H5s,” “α-tentacle,” and “β-tentacle” (Figure 1A; a detailed description of the motifs is provided in Figure S1).
To gain insight into the structural basis for the inhibition of CP by V-1, we solved the crystal structure of CP (chicken α1/β1) in complex with V-1 (human). The CP/V-1 complex was crystallized and the X-ray structure was determined at 2.2 Å resolution (R = 0.186, Rfree = 0.237) by molecular replacement, using the CP structure (PDB: 1IZN) as a search model (Figure 1B and 1C, and Table S1). CP contacts V-1 at two binding sites: (1) the basic residues at the C-terminus of the α-subunit and (2) a hydrophobic pocket adjacent to the basic contact site described above (Figures 2A and S1).
Three conserved basic residues in the CP α-subunit, Lys256, Arg260, and Arg266, were shown to be critical for the barbed end capping [5]. Remarkably, this “basic triad” directly participates in the V-1 interaction (Figure 2B). Arg260, the center of the “basic triad,” forms a bidentate salt bridge with V-1 Asp44. In addition, Lys256 and Arg266 form salt bridges with V-1 Glu78. Furthermore, Lys256 also forms a hydrogen bond with the main chain oxygen of V-1 Asp44. These notable ion pairs involving the “basic triad” clearly indicate that V-1 specifically binds conserved residues important for the interaction with actin, thereby effectively abolishing the barbed end capping. The importance of these ion pairs for complex formation was confirmed by a mutational analysis. We determined the CP/V-1 binding affinity by surface plasmon resonance measurements. Mutations of residues which form the “basic triad,” or their ion-pairing residues in V-1, reduced the affinity more than 25-fold compared with the wild type proteins (KD = 21 nM: binding constants for the mutant proteins are summarized in Table S2). The effects of mutations in the “basic triad” on the V-1 interaction are similar to those on the barbed end capping: reverse-charged mutants have lower affinities for V-1 than alanine mutants, and multiple mutations exhibit more severe defects than single mutations [5].
Another striking feature in the CP/V-1 interface is the hydrophobic contact formed around V-1 Trp8 (Figure 2C). In V-1, Trp8 on the V-1 helix 1 inserts its indole ring into a hydrophobic pocket, which is formed by CP (α) Ala257 and Leu258, immediately adjacent to the “basic triad,” and CP (β) Gly138 and Ile144 in “loop S5–S6” (a loop connecting β-strands 5 and 6 of the β-subunit). This hydrophobic contact is further stabilized by a hydrogen bond between the aromatic nitrogen of the tryptophan and the main chain oxygen of CP (β) Ile144. Mutation of this tryptophan, V-1 W8A, drastically reduced the affinity for CP (KD = 6.4 µM).
As expected, the CP binding-deficient V-1 did not inhibit CP in an actin polymerization assay (Figure S2). The wild-type V-1 allowed actin elongation from spectrin-actin seeds, even in the presence of CP. In contrast, the CP-binding deficient V-1 mutants (V-1 W8A, D44R, or E78R) had little inhibitory effect on CP activity.
We superposed the structure of the CP/V-1 complex onto the previous EM model of CP on the barbed end of an actin filament (Figure 3) [5]. This unambiguously demonstrated the collision of a major part of the V-1 molecule with the filament, mainly with subdomain 3 of the penultimate protomer. Furthermore, V-1 should prevent CP from even an initial contact with the barbed end, as it masks the “α-tentacle” by interacting with the “basic triad” residues (Figure 2B). Collectively, V-1 completely inhibits CP from interaction with the actin filament. The structure also indicates that V-1 lacks uncapping activity, because the V-1 binding site on CP is buried deeply between the two end protomers when CP caps the filaments.
Although the association of V-1 with CP has been reported in vivo [20],[21], it remains unknown whether V-1 is involved in the regulation of cellular actin assembly. We addressed this question by using the rat neuronal PC12D cell line V1-69, which is stably transfected with V-1 cDNA and expresses a 5- to 6-fold higher amount of V-1 than the mock transfectant C-9 [16]. Initially, we measured the ratio of F-actin to G-actin by a sedimentation assay and found that more actin pelleted from extracts of V1-69 cells than mock cells (Figure 4A). This indicates that the overexpression of V-1 leads to enhanced actin polymerization in PC12D cells. We next examined the amount of CP in subcellular fractions. In the V1-69 cells, the proportion of CP in the “high speed supernatant” fraction was significantly larger than that of the mock transfectant. This result was inversely correlated with a decrease in the distribution of the “high speed pellet insoluble in detergent” fraction (Figure 4B: see Materials and Methods for the subcellular fractionation procedure). The overexpression of V-1 did not alter the total amount of CP in the transfectants (unpublished data). These results imply that V-1 enhances actin polymerization by inhibiting the interaction of CP with the cytoskeleton structures. Moreover, we observed that, compared to the mock cells, V1-69 cells exhibited membrane protrusive structures with a thick, neurite-like appearance (Figure 4C). Phalloidin staining revealed that these protrusions were enriched with actin filaments (Figure 4C), implying that CP suppression caused by V-1 overexpression leads to the alteration of cell morphology presumably due to the increase in the level of actin polymerization. Taken together, our results demonstrate the possible involvement of V-1 in the regulation of actin polymerization and cellular morphology in living cells.
With the exception of the mobile “β-tentacle,” CP has been considered to be a rigid heterodimeric protein that is stabilized by many intra- and inter-subunit interactions [3]. However, we found that the overall conformation of V-1-bound CP (CPV-1; Figure 5B) is apparently different from the free form (CPfull; PDB; 1IZN; Figure 5A); e.g., the “antiparallel H5s” is straighter and the “N-stalk” and “β-globule” are further apart. Superposition of the two structures was poor, with a root-mean-square displacement (RMSD) over the Cα atoms of 2.55 Å [residues 9–275 (α) and 3–244 (β); the “β-tentacle” was not included] (Figure 5C). This unexpected finding indicates that CP has conformational flexibility. For further structural comparison, we obtained a new ligand-free CP structure crystallized under different conditions from 1IZN (CPβΔC; at a 1.9 Å resolution) (Figure S3) and found that the structure of CPβΔC is substantially different from both CPfull and CPV-1 (RMSDs of 1.34 Å and 1.87 Å, respectively) (Figure 5C and Table S3). These values are much larger than those expected for the same protein crystallized under different conditions (∼0.8 Å) [30]. Therefore, we conclude that CP conformational changes are not induced solely by the binding of a ligand molecule but show that CP is an intrinsically flexible molecule.
A domain motion analysis revealed that CP comprises two structurally stable domains, and the conformational change can be attributed to a twisting movement between the domains (Figures 5D–G and S4). The larger domain contains roughly two-thirds of the CP residues [residues 1–258 (α): 1–42, 175–192, and 235–277 (β)] and consists of the entire “N-stalk,” “α-globule,” and “β-tentacle” motifs together with parts of the “central β-sheet” and “antiparallel H5s,” whereas the smaller domain [residues 259–286 (α): 43–174 and 193–234 (β)] consists of the remaining portion. We refer to these larger and smaller domains as the CP-L and CP-S domains, respectively. Each domain superimposed well across the three forms (RMSDs of 0.80–1.06 Å for the CP-L domain and 0.80–1.04 Å for the CP-S domain) (Table S3). The boundary of the two domains does not directly correspond to the subunit interface; it resides between the “N-stalk” and “β-globule.” The two domains are linked by flexible regions, such as a short linker [Asp43–Leu47 (β)] between the “N-stalk” and “β-globule” and the helix-breaking residues [Thr253 (α) or Gly234 (β)} in “antiparallel H5s.” These regions may act as hinges to facilitate domain movement.
To explore the structural basis of CP inhibition by CARMIL proteins, we attempted to determine the structures of CP in complex with CARMIL proteins. Since the CP-binding motif of the CARMIL proteins is sufficient for the interaction with CP [25], peptides derived from this motif were used for the crystallographic studies; mouse CARMIL (residues 985–1005; referred to as CA21), human CD2AP (485–507; CD23), and human CKIP-1 (148–70; CK23) (we collectively refer to these synthetic peptides derived from CARMIL proteins as CARMIL peptides) (Figure 6A). In addition, we chose CPβΔC for crystallization, since the “β-tentacle” does not participate in the CARMIL interaction [23]. All of the crystals were grown under conditions similar to those for the ligand-free CPβΔC, and the structures were solved at 1.7–1.9 Å resolutions (R = 0.184–0.213, Rfree = 0.238–0.263) (Table S1).
The three crystal structures are shown in Figure 6B–D. As expected from the sequence similarity, all three peptides bound to essentially the same surface on CP. A superposition of the three structures further highlights the structural similarity, especially in their N-termini (Figure 6E). In contrast, the C-termini showed some diversity, probably due to the lack of consensus residues and the different peptide lengths. The peptides in our structures are largely unfolded, as previously indicated by a circular dichroism analysis [23]. Each elongated peptide binds along a continuous curved groove on the surface of the CP β-subunit. The peptides are bent by 100° at the conserved proline residue in the middle of the CP-binding motif. The consensus motif interacts with CP across the two domains: the N-terminus with the CP-L domain and the C-terminus with the CP-S domain (Figure 6E). The conformations of CP within the CP/CARMIL peptide complexes are similar to each other (RMSDs; 0.71–0.90 Å) and are slightly different from either CPfull or CPβΔC (RMSDs; 0.97–1.26 Å) (Table S3), suggesting that, unlike V-1, the CARMIL peptides do not cause a large conformational change to CP.
The binding between CP and the CARMIL peptides is primarily mediated by electrostatic interactions, which are supported by hydrophobic interactions (Figures 7A and S5). The mutation of a conserved arginine in the middle of the motif (Arg493 in CD23; indicated by an asterisk in Figure 6A) reportedly abolished CP binding for all of the peptides [23],[25],[31]. This central arginine makes multiple interactions with both the CP-L and CP-S domains, by forming a salt bridge with CP (β) Asp44, and hydrogen bonds with CP (β) Ser41 and Tyr64 (Figure 7B). We confirmed the importance of the intermolecular interface residues of CP by biochemical assays using mutant CP proteins (Figure 8 and Table 1). Among the mutant CP proteins, CP (β) D44N exhibited the lowest affinity for the CARMIL peptides.
In addition to their extensive interactions through the CP-binding motif, CD23 and CK23 further associate with the CP “N-stalk” via the C-terminal flanking residues of the motif. In the CP/CD23 complex, CD Phe505 contacts the hydrophobic pocket formed by the CP “N-stalk” residues [CP (β) Ile29, Cys36, and Leu40] and the peptide residues (CD Leu501 and Pro502) (Figure S6A). In the CP/CK23 complex, the C-terminal residue of the peptide, CK Arg169, forms an electrostatic interaction with CP (β) Asp30 (Figure S5B). In contrast to these two peptides, CA21 does not contact CP via the C-terminal flanking region (Figures 9A and S5A).
We tested the importance of the C-terminal flanking regions of the CP-binding motif using a binding assay (Table 2; the constructs used for the measurement are shown in Figure 9B). Surprisingly, GST-CD43, lacking CD Phe505 but containing the entire consensus motif, bound to CP only weakly with a KD of 260 nM, suggesting that the CP-binding motif of CD2AP alone is not sufficient for stable interaction with CP. In contrast, longer constructs with extended C-terminal residues showed higher CP binding affinities than the shorter fragments. GST-CD47, containing CD Phe505, bound to CP with a KD of 18 nM and GST-CD56 bound tightly to CP (KD = 4.7 nM), in good agreement with the previously reported value (KD = 5.6 nM for GST-CD2AP fragment containing residues 474–513 [25]). The C-terminus of CD23 extends into the region between the CP-L and CP-S domains (Figure S6B). Thus, the residues immediately C-terminal to CD23 (i.e., CD Gly508∼) are expected to form additional contacts with the domain boundary residues to stabilize the CP/CD2AP complex. Collectively, the C-terminal flanking region of the consensus motif is required for the stable interaction between CP and CD2AP.
We also examined GST-CARMIL fragments (Table 2 and Figure 9C). Both GST-CA55 and GST-CA63, containing the entire CP-binding motif and 10 or more extra residues at either end, bind only to CP with KDs in the micromolar range. This confirms that the consensus motif alone cannot tightly bind to CP. Moreover, unlike CD2AP, the CARMIL residues immediately C-terminal to the motif do not contribute to the stable CP interaction, consistent with our structure in which CA21 does not contact CP in this region. The stable CP interaction was observed in longer CARMIL fragments. GST-CA76 was found to have modest binding affinity to CP (KD = 80 nM) and GST-CA92 bound strongly to CP (KD = 3.3 nM) and with a comparable KD to GST-CD56.
We next evaluated the CP-binding affinity of CK23 by a competition assay and found that both CD23 and CK23 effectively compete with immobilized GST-CA92 for CP binding, whereas CA21 was a less efficient competitor (Figure S7). Thus, CK23 appears to have CP binding affinity comparable to CD23.
The CP binding affinity of the CARMIL peptides directly correlated with their ability to inhibit the barbed end capping. CD23 and CK23 moderately inhibited barbed end capping by CP, while CA21 was a poor inhibitor (Figure 9D). Furthermore, CD30, a peptide with 7 extra residues at the C-terminus of CD23, showed higher CP inhibition activity than CD23 (Figure 9D). Although weaker than CD23 or CK23, CA21 retained the ability to inhibit CP, since CA21 attenuated the barbed end capping by CPβΔC (Figure 9E), which is a less potent capper compared to CPfull [4]. Intriguingly, all peptides tested effectively inhibited CPβΔC, suggesting that CARMIL peptides do not inhibit CP simply by preventing the “β-tentacle” from filament binding. We next tested the CP inhibitory activity of GST-CARMIL constructs. As expected from their CP binding affinities, GST-CA92 showed the strongest CP inhibitory effect (Figure 9F). GST-CA92 appears to have full CP inhibition activity, because it showed a similar level of inhibition as GST-C-1 (residues 962–1084), which has the same activity as the full length CARMIL (unpublished data [23]).
A superposition of the crystal structures of the CP/CARMIL peptide complexes onto the EM model of the CP/actin filament structure clearly revealed that none of the peptides on CP overlap with the barbed end actin protomers (Figure 10). As described above, all of the peptides used for the crystallization have varying degrees of CP inhibition activity (Figure 9D–F). Furthermore, the C-terminal flanking residues of CD23, which greatly contribute to the CP inhibition, cannot reach the nearest surface of the actin filament. Therefore, unlike V-1, the CARMIL peptides do not inhibit the barbed end capping activity of CP by steric hindrance.
This non-overlapping CP interaction, permitting the CARMIL peptides to interact with the filament-bound CP, is a prerequisite for the uncapping activity. Furthermore, the “α-tentacle” including the “basic triad” on the top surface of CP, the primary actin binding site, is still exposed even when CP is bound with CARMIL proteins. This allows the CP/CARMIL protein complex to make an initial contact with the barbed end, and thus CARMIL proteins cannot sequester CP completely from the barbed end.
The CP binding site of V-1 is located on an opposite face from the CARMIL peptide binding site, implying that CP can simultaneously bind both inhibitors. Conversely, we found that the conformation of CPV-1 is significantly different from that of the CARMIL peptide-bound CP (CPCARMILs) (Table S3), because the binding of V-1 induces a twisting movement of the CP-L and CP-S domains. This raises the possibility that the CARMIL peptides allosterically inhibit CP from binding V-1 by restricting the domain twisting, since the peptides bind to CP across the two domains. We tested this prediction using a surface plasmon resonance assay. We immobilized GST-V-1 on a sensor chip, and then perfused with CP premixed with CARMIL peptides. Surprisingly, CD23 and CK23, which possess substantial affinity for CP, strongly inhibited the CP/V-1 interaction, indicating that the peptides restrict the conformation of CP to the “low affinity to V-1” form (Figure 11A). This inhibition depends on the CP/CARMIL peptide interaction because CA21, which has a lower CP binding affinity than the other peptides, exhibited minimal inhibition (Figure 11A). Furthermore, none of the peptides tested could prevent CP (β) D44N, a mutant CP deficient in CARMIL protein interaction (Table 1 and Figure 8), from the V-1 interaction (Figure 11B). Most notably, in addition to its effect on free CP, the CARMIL peptides can act on CP pre-bound to V-1 and facilitate the dissociation of the complex. When the preformed CP/V-1 complex bound on the sensor chip was perfused with CD23 or CK23, CP dissociated from V-1 quite rapidly, as compared with the buffer control (Figure 11C). Again, we found that CA21 was less effective in facilitating the dissociation (Figure 11C), and that the interaction between CP (β) D44N and V-1 was not affected by CARMIL peptides (Figure 11D). This result suggests that the CARMIL peptides possess the ability to interact with CP in a conformation different from CPCARMILs and to shift the CP conformation toward the CPCARMILs form.
We further confirmed the effect of the CARMIL peptides on CP/V-1 interaction by a pull-down assay. Under equilibrium conditions, the binding of CP to GST-V-1 was inhibited by the addition of the peptides in a concentration-dependent manner (Figure S8). Collectively, we concluded that the CARMIL peptides allosterically inhibit CP binding to V-1.
To further explore the intrinsic flexibility of the CP molecule, we performed a normal mode analysis with an elastic network model (ENM). In this model, a protein is considered as a simple elastic object, and the spatially neighboring residues in the native structure are connected by Hookian springs. Based on this approximation, the intrinsic fluctuations originating from the protein shape are revealed. The normal mode analysis on the ENM has been applied to various sizes of proteins, e.g., lysozyme [32], F1-ATPase [33], and chaperonin GroEL [34]. Referring to the lower frequency modes, the analysis succeeded in reproducing large conformational motions that had been experimentally revealed [35]. We applied this method to the CP/CD23 complex (Figure 12A) and the CP structure extracted from the complex (Figure 12B), and focused on the first lowest modes. The first lowest mode of CP can be described as twisting motions relative to two axes, which run through the α- and β-subunits, respectively (Figure 12A and Table S4; see Materials and Methods for more details). In this mode, the directions of the twisting movements about the two axes are opposite from each other (indicated by black and gray sets of arrows in Figure 12). Among these two axes, the β-subunit axis almost coincides with the axis of the twist movement between the CP-L and CP-S domains that was revealed by the structural comparison (red rods with asterisk in Figure 12). This finding strengthens the notion that CP continually undergoes substantial twisting movements about this axis. Furthermore, we found that the CARMIL peptides alter this intrinsic mode, both in the direction of the rotational axis and the amplitude of the motion (Figure 12B). These effects are observed almost exclusively in the twisting motion about the β-subunit axis, yet not about the α-subunit axis, suggesting that the CARMIL peptide suppresses the twisting movement between the CP-L and CP-S domains.
The crystal structure of the CP/V-1 complex revealed that V-1 mainly interacts with the “α-tentacle,” the primary actin binding surface of CP, thereby sterically hindering CP from barbed end capping (Figures 1–3). The structure supports biochemical data that V-1 has no uncapping activity (Figure 13D). A sequence alignment of V-1 indicates that the residues involved in the V-1 interaction are highly conserved through evolution, despite their relatively minor contributions to the protein fold (Figure S9). Furthermore, the “basic triad” in the CP α-subunit, containing the highly conserved residues critical for actin binding is also recognized by V-1. This suggests that the architecture of the V-1 molecule is well suited for the interaction with CP, i.e., CP inhibition is the key role for V-1 in various cellular processes. This notion is further supported by the finding that, in cultured cells, V-1 is involved in the regulation of actin assembly and cell morphology (Figure 4). We note that CARMIL peptides inhibit CP from binding V-1 (Figures 11 and S8), indicating that the effect of V-1 on CP may be under the control of other proteins which interact with CP or V-1. Future studies will verify the role of V-1 in actin-driven cell motility.
An unexpected finding in this study was the conformational flexibility of the CP molecule. A structural comparison analysis revealed that CP consists of two rigid domains, CP-L and CP-S, and undergoes conformational changes even in the absence of a ligand (Figure 5). This intrinsic twisting motion between the two CP domains was further supported by a normal mode analysis of free CP (Figure 12A). Intriguingly, our analysis also predicts that, in addition to the domain twist related to the rotational axis passing through the β-subunit, there might be an analogous twisting movement about the α-subunit axis. This is plausible because CP has pseudo 2-fold rotational symmetry [3]. Thus, the CP-L domain might be further divided into two rigid subdomains, which also undergo a twisting movement relative to each other.
Our data showed that the CP-binding motif of CARMIL proteins cannot bind tightly to CP, despite the multitude of intermolecular interactions present in the structures (Figures 7, 9, S5, and Table 2). This is attributable to the conformational fluctuation of CP, as the consensus motif interacts with residues at the domain boundary that may act as a hinge in the twisting movement. We demonstrated that the regions C-terminal to the CP-binding motif are responsible for the strong interactions between CP and CARMIL proteins (Table 2). Thus, the consensus motif and the flanking region may reciprocally increase their affinity for CP, which in turn would inhibit CP effectively.
The tight interaction between CP and the barbed end is contributed by the extensive inter-molecular surface residues [5]. Consequently, the intrinsic twisting motion between the two CP domains that can cause changes in the overall structure must affect the capping activity of CP. Therefore, for a stable filament capping, CP accommodates its shape to a favorable conformation for the barbed end interaction. Consequently, we have revised the previous two-step capping model [5] as follows: (i) “Basic triad” residues on the CP “α-tentacle” region interact electrostatically with the barbed end. This initial contact is followed by two independent stabilization steps: (ii) an adaptive conformational change to a “high affinity to the barbed end” form that is a twisting movement between the CP-L and CP-S domains and (iii) the supportive binding of the “β-tentacle” to the filament (Figure 13B). Hence, a factor which disturbs either of the capping steps has an inhibitory effect on the filament capping activity of CP. For example, V-1 sterically hinders CP from the barbed end by blocking step (i).
How do CARMIL proteins inhibit the capping activity of CP in an allosteric manner? We showed that CARMIL peptides allosterically inhibit the interaction of CP with V-1 (Figures 11 and S8). This finding indicates that, regardless of the initial CP state (i.e., free or V-1-bound), the peptides binding across the two CP domains shift the conformational distribution to within a narrow range around CPCARMILs, conformations that are unfavorable for V-1 binding. We propose that CARMIL proteins inhibit CP in a similar manner (Figure 13C); CARMIL proteins limit the conformational distribution of CP to mostly the “low affinity to the barbed end” form, leading to attenuation of the barbed end capping activity [i.e., step (ii) in Figure 13B is inhibited]. Fujiwara et al. indicated that CARMIL does not affect the association of CP to the barbed end but accelerates its dissociation from the filament since the on rate of the CP/CARMIL complex to the barbed end is virtually the same as that of free CP (3.7 µM−1s−1 versus 2.6 µM−1s−1), while the affinity of the complex to the filament is significantly lower than that of free CP (KD = 38 nM versus 0.18 nM)[29]. This is consistent with our hypothesis that the CARMIL proteins inhibit CP only by affecting the twisting motion which provides the capping stability, since our data showed that neither the “α-tentacle” (the capping on rate determinant) nor the “β-tentacle” (the other capping stabilizer) is disturbed by the CARMIL protein. Furthermore, our prediction that the conformation CPCARMILs is substantially different from the “high affinity to the barbed end” form is consistent with the concept that CARMIL binding to free CP must involve some surface or conformation that is not available when CP is bound to a barbed end [23]. This is because the affinity of CARMIL for the barbed end-bound CP has been estimated to be 10- to 100-fold [23] or 200-fold [29] lower than that for free CP.
To better understand the mechanism of CP inhibition by the CARMIL proteins, it would be helpful to know the conformation of CP on the barbed end. As such, we fitted all known crystal structures of CP to the 3D electron density map of the CP/actin filament [5] and found that all of the structures tested fit similarly to the model except for CPV-1, which did not fit as well (Figure S10). The mismatch between the EM envelope and CPV-1 is largely due to the shift of the CP-S domain relative to the CP-L domain, suggesting that the CP in the “high affinity to the barbed end” form may not adopt such an “open” conformation as in CPV-1.
In this study, we cannot provide structural information about CP bound to the full activity CARMIL fragments. During the submission of this manuscript, Robinson and colleagues reported a crystal structure of CP in complex with a CARMIL fragment with an extended C-terminal portion (CBR115; human CARMIL residues 964–1078) [36]. This structure revealed that, in addition to the CP-binding motif, a 15 residue motif serves as a second CP binding site (CARMIL-specific interaction motif, residues 1021–1035; highlighted by orange in Figure S11). The motif binds to the CP “N-stalk” in the CP-L domain, on the side opposite to where the CP-binding motif binds. This result also supports the concept that CARMIL proteins inhibit CP in an allosteric manner (see Text S1 for a detailed discussion about the role of the C-terminal flanking region of the CP-binding motif of the CARMIL proteins for CP inhibition).
Recently, intrinsically unstructured proteins or segments of proteins have been recognized to play critical roles in many cellular processes such as transcriptional regulation and signal transduction [37]. These disordered regions usually fold into ordered secondary or ternary structures upon binding to their targets (termed coupled folding and binding processes). We revealed, however, that the CARMIL peptides are functional in suppressing the conformational flexibility of CP, although they have an extended backbone conformation. Consequently, our results provide new insights into the functional expression of intrinsically unstructured proteins.
An important implication of this study is that conformational restraints placed on CP lead to an attenuated affinity of the protein for the barbed end. This raises the possibility that other CP regulators, such as PIP2, also modulate the capping activity. Moreover, the state of the actin filament would also affect the affinity of CP towards the filament; i.e., a certain actin binding protein that changes and/or restricts the structure of the barbed end to an unfavorable form for CP binding can antagonize the filament capping. We assume that such a mechanism may account for the rapid turnover rate of CP in lamellipodia [9],[10].
In this study, we have described the structural basis for CP inhibition by two regulators, V-1 and CARMIL proteins. Our findings suggest that CP is not a constitutively active inhibitor of barbed end elongation; rather, the capping activity of CP is fine-tuned for the highly orchestrated assembly of the cellular actin machinery, and the conformational flexibility of CP provides the structural basis for the regulation.
Expression vectors for chicken CPfull and CPΔβC were constructed in pETDuet-1 by PCR, using pET-3d/CP [38] as the template. CP was expressed in E. coli Rosetta2 (DE3) and was purified as described [3]. V-1 (human), expressed in E. coli Rosetta2 (DE3) as a GST-fusion protein, was affinity-purified and the tag was removed. Synthetic peptides derived from CARMIL proteins were obtained from Invitrogen. For crystallization, CP was incubated with a 1.2–2.0-fold molar excess of V-1 or CARMIL peptides at 4°C for 2 h, followed by gel filtration to purify the complexes. Expression vectors for the GST-CA constructs were prepared from the mouse cDNA clone as previously described [23]. Vectors for GST-CD fragments were constructed by PCR cloning using a human whole brain cDNA library (Clontech) as the template. Amplified DNA fragments were cloned into pGEX-6P-1. GST-fusion proteins were expressed in E. coli Rosetta2 (DE3) cells and affinity-purified using glutathione sepharose resin. Mutations were introduced using a Quikchange mutagenesis kit (Stratagene). Actin was prepared from rabbit skeletal muscle, as previously described [39], and was further purified by gel filtration chromatography. Pyrene labeled-actin was prepared as described [40]. Spectrin-actin seeds were prepared from rabbit red blood cells, as previously described [41].
Each protein complex, at 8–10 mg/ml in 1 mM DTT and 5 mM Tris-HCl (pH 8.0), was mixed with an equal volume of reservoir solution as follows: 10% PEG4000, 20% isopropanol, 20 mM EDTA, 0.1 M Tris-HCl (pH 8.4) for CP/V-1; 12.5% PEG400, 20 mM BaCl2, 0.1 M MES-NaOH (pH 6.0) for CPβΔC; 18% PEG400, 40 mM BaCl2, 0.1 M MES-NaOH (pH 6.0) for CP/CA21; 10% PEG400, 20 mM BaCl2, 0.1 M MES-NaOH (pH 6.5) for CP/CD23; and 17.5% PEG400, 30 mM BaCl2, 0.1 M MES-NaOH (pH 6.0) for CP/CK23. The crystals were grown at 20°C by the hanging-drop vapor diffusion method and were cryoprotected with their reservoir solutions supplemented with 20% glycerol (for CP/V-1) or with 35% PEG400 (for other crystals) prior to flash-cooling in a cold nitrogen stream. Diffraction data were collected in the BL26B1 beamline at SPring-8 [42] and were processed with HKL2000 [43]. Space groups and cell parameters are listed in Table S1. Initial phases were determined by molecular replacement with Molrep [44], using the CP structure as a search model. Model building and refinement were performed with CNS [45], Refmac [46], and Coot [47]. Each crystal contains one CP or CP/inhibitor complex in the asymmetric unit. Data collection and refinement statistics are summarized in Table S1.
The barbed end elongation assay from spectrin-actin seeds was performed essentially as previously described [4]. Briefly, G-actin was stored in G-buffer (0.2 mM CaCl2, 0.2 mM ATP, 0.5 mM DTT and 10 mM imidazole, pH 7.0). At 90 s prior to polymerization, the Ca2+ was replaced with Mg2+, by the addition of 1/10 volume of 10 mM EGTA and 1 mM MgCl2 to G-actin. Barbed end elongation was initiated by mixing the solutions in the following order: Mg2+ actin (5% pyrene-labeled), CP, V-1 or CARMIL protein, a 1/20 volume of 20× polymerization buffer (1 M KCl, 20 mM MgCl2, 20 mM EGTA, 0.2 M imidazole, pH 7.0) and spectrin-actin seeds. Actin polymerization was measured by monitoring the pyrene-actin fluorescence (excitation 370 nm; emission 410 nm) at 25°C.
The binding affinities of CP for V-1 or CARMIL proteins were evaluated by surface plasmon resonance measurements with Biacore 3000 or Biacore 2000 instruments (GE Healthcare). GST-fusion proteins (GST-V-1, GST-CA, or GST-CD) were immobilized onto a CM5 sensor chip up to 200 RU (response units; 200 pg/mm2) via anti-GST antibodies. CP at various concentrations in running buffer (50 mM KCl, 1 mM MgCl2, 0.005% Tween-20, 10 mM imidazole, pH 7.0) was perfused over the chip at 20°C, at a flow rate of 20 µl/min. Response curves were obtained by subtracting the background signal generated simultaneously on a control flow cell with immobilized GST. To measure the effect of the CARMIL peptides on the facilitation of CP/V-1 dissociation (in Figure 11C and 11D), we used the “co-inject” mode for successive injections of the peptides followed by CP. Kinetic parameters were determined by fitting the sensorgrams to a simple 1∶1 binding model, using the Bia-evaluation software (GE Healthcare). KD values were obtained from the kinetic rate constants. For several mutant proteins possessing fast dissociation rates for the ligand (koff >0.1 s−1), we measured the amount of bound-CP at the steady state over a wide concentration range. KD values were evaluated by plotting these values against the concentrations of CP.
The stable V-1 overexpression transfectant (V1-69) and its mock transfectant (C-9), established in the PC12D subclone of rat pheochromocytoma cells, were cultured as reported previously [16]. The concentrations of F- and G-actin were measured using an assay kit (Cytoskeleton), as described previously [48]. For subcellular fractionation, the cells were homogenized by sonication in homogenization buffer (150 mM NaCl, 2 mM EGTA, 10 mM Tris-HCl, pH 7.4, with protease inhibitors). The extracts were centrifuged at 100,000 g for 60 min, and the supernatant was designated as the “high speed supernatant” fraction. The pellet was incubated for 30 min in the homogenization buffer supplemented with 0.5% Triton X-100 and ultracentrifuged. This supernatant was designated as the “high speed pellet soluble in detergent” fraction, and the “high speed pellet insoluble in detergent” fraction was obtained by further extraction of the pellet in 8.3 M urea. The amount of CP in the fractions was determined by Western blotting with an anti-CP β-subunit antibody [21]. For morphological analysis, cells cultured at a density of 5×104 cells per well on the poly-d-lysine-coated culture slides (BD Biosciences) for 24 h were fixed by 3.7% formaldehyde in PBS and permeabilized with 0.1% Triton X-100 in PBS. Fixed cells were pre-incubated with the Image-iT FX signal enhancer (Invitrogen) and counter-stained with Alexa Fluor 546-conjugated phalloidin (Invitrogen) and Hoechst 33258 (Dojin). The fluorescence images were obtained using Leica microfluorescent system (AF6500; Leica Microsystems).
The intrinsic flexibility of CP was examined by the normal mode analysis with the ENM [49],[50],[51]. In this model, only the Cα atoms are considered, and a harmonic potential with a single parameter, C, is introduced between all Cα atoms within a cut-off distance, Å. The potential energy of a protein is given aswhere is the vector connecting the i-th and j-th Cα atoms and is that in the crystal structure. The Hessian matrix, whose elements are the second derivatives of the potential energy, was derived and diagonalized, and we obtained the eigenvectors and eigenvalues, representing the normal modes.
Since the twisting movements were revealed by comparisons of the crystal structures, we estimated the intrinsic rotations from the lowest frequency mode that corresponds to the largest vibration. As the CP free model structure, we employed the CP structure of the CP/CD23 complex (i.e., the CD2 peptide was removed). The displacements of each Cα atom were derived from the displacement vector, the eigenvector of the lowest frequency mode scaled by the reciprocal of the eigenvalue. We consider that the set of Cα atoms with small displacements represents the rotation axis. The Cα atoms, whose squares of the displacements were smaller than 2 Å2, were collected.
We found that these Cα atoms could be clearly divided into two groups, and each of them was separately distributed in the α-subunit or the β-subunit (Table S4). The coordinates of these Cα atoms in each group were evaluated by the principal component analysis, and the first components defined the rotation axes on the α- and β-subunits. In Figure 12, the axes run on the center of Cα atoms with small displacements. The same analysis was applied to the CP/CD23 complex, with a cut-off displacement of 1 Å2.
The Protein Data Bank accession codes for the crystal structures determined in this study are as follows: CP/V-1 (3AAA), CPβΔC (3AA7), CP/CA21 (3AA0), CP/CD23 (3AA6), and CP/CK23 (3AA1).
|
10.1371/journal.pgen.1003760 | Plasticity Regulators Modulate Specific Root Traits in Discrete Nitrogen Environments | Plant development is remarkably plastic but how precisely can the plant customize its form to specific environments? When the plant adjusts its development to different environments, related traits can change in a coordinated fashion, such that two traits co-vary across many genotypes. Alternatively, traits can vary independently, such that a change in one trait has little predictive value for the change in a second trait. To characterize such “tunability” in developmental plasticity, we carried out a detailed phenotypic characterization of complex root traits among 96 accessions of the model Arabidopsis thaliana in two nitrogen environments. The results revealed a surprising level of independence in the control of traits to environment – a highly tunable form of plasticity. We mapped genetic architecture of plasticity using genome-wide association studies and further used gene expression analysis to narrow down gene candidates in mapped regions. Mutants in genes implicated by association and expression analysis showed precise defects in the predicted traits in the predicted environment, corroborating the independent control of plasticity traits. The overall results suggest that there is a pool of genetic variability in plants that controls traits in specific environments, with opportunity to tune crop plants to a given environment.
| Plants can dramatically alter their development in order to cope with new environmental conditions. Such plasticity is especially evident in the root system since it adopts a particular architecture under one condition, but can change architecture by altering the extent of lateral root branching in a different condition. To explore the extent of root plasticity to the critical nutrient nitrogen we analyzed a natural population of the model plant Arabidopsis in both nitrogen-limiting and nitrogen-rich environments. This revealed that root architecture plasticity appears to be the combined effect of many individual root responses to the environment that are independently modulated. Each aspect, such as lateral root length, number, or density seems to be turned on or off separately, giving the whole system flexibility. We then identified specific genes that control these individual component responses by exploring the genetic variation across the natural population in combination with analyzing which genes respond to nitrogen. Together the results help us gain insights into how the environment shapes plant development. This knowledge can be used to better understand how the growth of our existing crop species might change as the climate varies, and identify new crop varieties that will be robust to such variation.
| Nitrogen is a limiting nutrient in plant growth that is typically taken up from the soil by the root system [1]. However, because the soil environment often varies over space and time, a single genotype needs to adjust its root architecture in response to different soil conditions, an example of developmental plasticity. One imperative in agriculture is to develop crops that can grow efficiently while reducing expensive and environmentally detrimental nitrogen supplements; current high yield crops are typically optimized for a single environment of high nitrogen. To breed crops for different or naturally fluctuating nitrogen environments, mechanisms that mediate traits conditioned on the environment may be important targets of crop improvement.
In plants, root architecture is a complex phenotype that arises from adult meristematic activity in primary and lateral roots and lateral root initiation [2], [3]. These traits collectively determine the root's three-dimensional body plan, where specific shapes can provide advantages in certain environments [4]. For example, deeper primary roots are often associated with plants with a greater tolerance to drought [5], [6]. The dynamic and patchy nature of the soil environment also appears to make the post-embryonic adjustment of the body plan a valuable attribute. For example, a strong association was found between local proliferation of lateral roots and nitrogen uptake in competition assays in grasses [7], [8]. Collectively, these studies show that different attributes of root architecture and the ability of individuals to adjust that architecture can confer advantages in the heterogeneous soil environment.
Here, we systematically characterize the way in which root traits can vary in different environments across accessions in one species, Arabidopsis thaliana. At one extreme, a set of traits may be correlated (or anti-correlated) such that trait 1 and 2 may both consistently increase, decrease or show opposite trends in a new environment when examined in many different accessions [7]. At the other extreme, traits may be independent with respect to each other, such that a change in trait 1 has no predictive value in a change in trait 2 when examining many genotypes [4].
We expect that specific genes mediate the response to extrinsic signals to affect intrinsic development programs [2]. For example, genes that mediate the activation of transient stem cell niches in the pericycle will influence lateral root density [4]. In previous work, plasticity mechanisms active in the pericycle were found to coordinate the control of root initiation and outgrowth by nitrogen. It was found that an increase in expression of the transcription factor Auxin Response Factor 8 in response to high nitrogen treatment decreased lateral root growth and increased lateral root initiation [9]. This was an example of trait coupling that could result in an anti-correlation between traits across many genotypes. Another study showed that C∶N ratio appeared to specifically control lateral root initiation without strongly influencing other root traits – a potential example of trait independence [10]. A few other cases of regulatory genes that control root architecture in response to nitrogen have been identified [11], [12]. However, the overall level of customization of phenotype to environmental variation and the genetic architecture underlying plasticity are not well understood.
To characterize the tunability of root traits in response to different environments, we merge concepts from two different fields. The field of phenotypic integration has documented the level of correlation vs. independence in traits, typically across different genetic variants within a species or a set of closely related species [13]. The field of phenotypic plasticity has documented the ability of single genotypes to show variable phenotypes in different environments [13]–[15]. Here, we examine the correlation vs. independence of the difference in root traits in two environments to ask how finely the plant can manipulate its developmental plasticity. In addition, we also seek to determine the genetic mechanisms that mediate plasticity, as there is growing interest in the genes underlying phenotypic plasticity [13], [16], [17].
By carrying out comparative phenotypic analysis of key features of root architecture, we have been able to both analyze the correlation of individual root traits as well as assess the extent of plasticity within and between root traits that form root architecture. The use of genomic expression profiling in combination with high-density genetic marker association analysis enabled identification of genes implicated in controlling independent root parameters. By combining the phenotypic and genomic analyses we were able to functionally validate a number of new root regulators that mediate the response to nitrogen levels in discrete environments.
We characterized the level of correlation vs. independence in root plasticity by asking how traits vary in two distinct nitrogen environments among 96 well-characterized natural accessions or ecotypes of the model species Arabidopsis thaliana [18] (see Materials and Methods). We define developmental plasticity as the ability of a single genotype to exhibit different phenotypes in different environments. If root trait differences in the two environments are correlated, accessions should exhibit similar suites of changes in root architecture in distinct nitrogen environments. Alternatively, if a high level of trait independence exists, genotypes that are similar in one nitrogen environment could alter a subset of root traits in a new nitrogen environment.
To address this hypothesis, we first clustered accessions based on seven root traits capturing root size and architecture (see Methods) in each of the two environments: high and low nitrogen (Figure 1). There was a dramatic rearrangement of the tree topology in the two environments, as shown by the dispersal of clusters formed in low nitrogen mapped onto the high nitrogen phenotype tree (Figure 1). To observe trait behaviors, we also clustered accessions based on their trait differences in the two nitrogen growth conditions, and mapped average trait differences in each accession onto the tree as bar graphs (Figure 2A,B) or a heatmap (Figure S4). In one example, NFA-8 and Sq-8 have similar architectures on low nitrogen, but exhibit much different phenotypes in the high nitrogen environment, where Sq-8 outgrows lateral roots much more dramatically (Figure 2C). In another clade, Kas-2 is a super-responder, dramatically increasing almost all root traits measured in high nitrogen to the extreme levels observed (Figure 2C). On the other hand, roots of Bil-7 are almost completely unresponsive to nitrogen (Figures 1,2A). Overall, the cluster analysis indicates that sharing a phenotype in one nitrogen environment does not predict similarity in root architecture in a second nitrogen environment, arguing that different trait responses are independent of one another.
We used a Principal Components Analysis (PCA) to investigate the degree of correlation vs. independence in the root traits. In the first PC, which accounted for 64% of variation, almost all traits showed the same magnitude and direction in their contribution (Figure 2D, blue lines). This trend suggested that the greatest variation among the difference of traits on high compared to low nitrogen were correlated changes in traits, meaning overall size differences (Figure 2D). However, the different traits made highly varied contributions to the second and third PCs, as shown by the vectors (blue lines) representing the magnitudes and sign of trait coefficients in each component (Figure 2E). The second and third components represented about 17% and 10% of the variance, respectively, indicating a substantial amount of variation in these two components. Interestingly, the star-shaped configuration of the coefficient vectors indicates that traits are highly orthogonal in the space of the second and third principal components. In other words, traits show a high degree of independence and lack of correlation. The same trends were found in a PCA analysis of trait data from high or low nitrogen conditions or the combined high plus low nitrogen dataset (Figure S5). The substantial variation in PCs 2 and 3 shows that there is a significant component of variation in which traits vary freely among accessions in the transition from one environment to another.
Similarly, a mixed model ANOVA of the trait data showed that almost all traits have accession-by-treatment interactions (see Methods). For example, in the ANOVA model, Kas-2 has a high interaction coefficient in LRtot, in which it changes phenotype dramatically in the two nitrogen conditions (Figure 2C). In a different type of trait interaction, Kas-2 also has a high interaction coefficient in LB/PR (root length between hypocotyl and most distal lateral root), but it is one of the few accessions to show almost no phenotypic difference between nitrogen environments (Table S5). Overall, the analysis shows that individual accessions adjust to differing nitrogen environments with variable increases in overall size, which demonstrates trait correlation as in PC1. However, there is a prominent secondary source of variability in which traits vary independently among the accessions and between environments, as demonstrated in PCs 2 and 3. The result shows that much of the variation observed when growing the 96 accessions in two environments is comprised of overall size effects, but, importantly, another large component of variation reflects a high level of fine tuning of each accession to a particular nitrogen environment.
To identify mechanisms involved in plasticity, we employed a genome-wide association study (GWAS) [19]. We associated known SNPs with root traits from plants grown on low or high nitrogen environments, or the difference in a trait value between the two environments (Table S6). In addition, we calculated the total proportion of trait heritability that the SNPs explain (Tables S6,S7). We used 96 accessions, as previous work suggested this number is sufficient to identify associations with relatively strong effect [19]. In total, we found 53 highly significant SNP hits that could be grouped, based on proximity, into 17 SNP groups (a SNP window included all genes within 10 kb on either side of the SNP and such intervals were joined into “groups” if their windows overlapped). In total, the 17 SNP groups encompassed 106 genes (Table S8). Surprisingly, out of 17 SNP groups, only a third of the groups associated with the same trait in the two nitrogen environments. This could mean that we either lacked power to detect SNPs in one environment, or, that there is genetic variation that specifically influences phenotype in one environment. We sought to test the hypothesis that specific genes mediate distinct traits in one nitrogen environment by testing whether mutants in any of the genes found within intervals showed a phenotype in the associated trait in the predicted environment. We focused on lateral root average length because 7 SNP groups encompassing 53 genes showed high significance and because a number of insertional mutants were available for genes in these windows (Table S8).
We sought to narrow candidates within genomic intervals that were implicated by SNPs by focusing on potential plasticity regulators that showed variable gene expression among accessions or between conditions or both. Thus, we profiled root gene expression of seven accessions that represent diverse root architectures (Col-0, Kas-2, Var2-1, Tamm-27, NFA-8, Sq-8, Ts-5; Figures 1,2A–C) using ATH1 microarrays in response to a 2-hour treatment of nitrate vs. control to identify early growth regulators that respond to new conditions (Methods; Table S9). ANOVA followed by a model simplification assignment (FDR<0.1, see Methods and Tables S10, S11) identified 5,043 genes that varied among accessions but with no response to nitrogen and 279 genes with a range of effects due to nitrogen (Figure 3). Of these 279 genes, 29 genes responded to nitrogen in all accessions with no accession-specific variation in the degree of response or direction of nitrogen-regulation (“nitrogen-only effect”). 123 genes responded to nitrogen across all accessions, with the same direction of response in all accessions but with a variation in the degree of response (“nitrogen, accession effect’). The remaining 127 genes had a nitrogen*accession interaction effect whereby the direction and/or degree of nitrogen regulation varied over the seven accessions. To validate the expression analysis for nitrogen responses, we analyzed the 152 genes that responded across all accessions (29 genes with nitrogen effect only; 123 genes with nitrogen and accession effect; Table S11). These 152 core response genes include two key nitrate response genes (AtNRT2.1 (nitrate transporter 2.1, At1g08090) and NIR1 (nitrite reductase, At2g15620)) and there is an overrepresentation of the GO term ‘response to nitrogen’ (8 genes, P = 1.06E-02). In addition, there is an overrepresentation of a number of metabolic functional terms including GO term ‘cellular metabolic process’ (78 genes, P = 3.88E-04) and GO term ‘small molecule biosynthetic process’ (23 genes, P = 6.69E-03), supporting common nitrogen regulation of cellular and metabolic pathways.
We also defined a more stringent list of regulated genes following the hypothesis that genes controlling the nitrogen response of root traits across accessions should have varied nitrogen-response levels across accessions. To generate such a “stringent set” of candidate genes, we took genes that showed a significant nitrogen*accession effect in ANOVA (127 genes) and those in expression clusters that correlated with average lateral root length in either low or high nitrogen or the difference between the two levels of nitrogen (321 genes).
We then conducted a reverse genetic screen in Col-0 to ask whether GWAS refined by expression analysis could identify genes that mediate specific traits in specific nitrogen environments. As a proxy for examining the phenotypic effects of natural alleles, we evaluated T-DNA mutants in 13 genes that fit two criteria for predicting a specific phenotype: the genes were found within genomic intervals associated with lateral root average length and their transcripts demonstrated a significant ANOVA effect (accession-only, nitrogen or nitrogen*accession effect) among the seven profiled accessions (13/53; Table S8).
Out of the 13 loci, three genes passed our criteria for demonstrating root phenotypes with (1) consistent, quantifiable phenotype in specific root traits for two separate T-DNA mutant alleles and, (2) absent or reduced gene expression in the mutant gene (Figure 4; Methods; Table S12; Figure S10). In addition we carried out crosses of the pairs of allelic mutants and confirmed trans non-complementation, supporting that the mutant alleles were responsible for the phenotypes (Table S12). In support of a model in which genes control traits in specific environments, mutant phenotypes from two loci precisely matched predictions for mediating specific traits in specific environments. One GWAS hit included a block of genes containing JR1 (JASMONATE RESPONSIVE 1), which was associated with lateral root length in low nitrogen and the difference between low and high nitrogen (Figure 4A). In addition, JR1 met stringent expression criteria, belonging to a cluster that correlated to the difference in lateral root length between nitrogen environments (Figure S9). The two mutant alleles tested showed a specific defect in lateral root average length in low nitrogen but not high nitrogen, as we predicted from analysis of GWAS and expression data (Figure 4B–C). In one plausible functional role for JR1 in controlling the length of lateral roots specifically when there are low levels of nitrogen in the environment, the jasmonate pathway has been shown to have a role in lateral root development [20]. A second gene, PhzC, which was significantly regulated across accessions, was also consistent with GWAS predictions, having shorter lateral roots on low nitrogen but not high nitrogen (Figure 4D–F). For a third gene identified from GWAS analysis, UBQ14 (polyubiquitin), an association with lateral root length in low nitrogen but not high nitrogen was tested (Figure 4G). However, phenotypic analysis showed phenotypes in both nitrogen conditions. In addition, phenotypes were observed in total lateral root length (LRtot) and total root length (PR+LRtot), as might be expected with severe defects in lateral root length (Figure 4H–I). Thus, this mutant implicates UBQ14 in trait specificity but not environmental specificity. Nonetheless, for two out of three cases for which we identified a root trait phenotype (JR1 and PhzC) the combination of GWAS with expression profiling identified genes that affected specific traits in specific environments, showing that, with this combination of techniques, we can map genotype to both trait and environment.
Within the narrower set of only three candidate genes that met the dual criteria of belonging to the gene expression “stringent set” and a GWAS “group,” two genes showed mutant phenotypes in precisely the predicted trait and environment. We cannot rule out that expression criteria alone could have identified candidates with mutant phenotypes. However, in a preliminary screen on the same data, we used expression criteria to examine mutants in 13 genes, with some showing pleiotropic phenotypes (data not shown) but none demonstrating specific defects in one or even two traits. Thus, we believe the combination of genome-wide association and gene expression greatly assists in identifying genes involved in specific traits in specific environments with high precision.
Overall, mutations in two out of the three loci that we identified by GWAS affected root systems in low nitrogen environments, where the lateral root system was relatively small, but showed normal root length in high nitrogen environments, where the root system was more extensive. This suggests that the mutant phenotypes are not simply due to general defects in lateral root growth, but rather the gene's specific role in one environment. We point out that we do not know the causal polymorphisms for the phenotypic variation in root traits among natural variants. Even if causal polymorphisms map to the same loci as the mutations we identified, the genetic polymorphisms responsible for the trait variation likely control trait values in a different manner than loss-of-function mutations. However, the mutant analysis provides some corroboration that these loci contribute to controlling plasticity in the traits that we identified. Furthermore, the mutant analysis suggests that different mechanisms may predominate in the control of specific traits in specific environments, perhaps because fewer redundant mechanisms are expressed in one condition. Genotype x environment effects have traditionally been seen as a detriment to crop breeding programs, although there is growing interest in accounting for such effects [17]. Our result suggests that mechanisms that alter traits in specific environments may be quite common. Such genes could be exploited to customize crop phenotypes to a specific environment, such as low nitrogen, without, for example, changing an optimal phenotype in high nitrogen environments.
All seeds were obtained from ABRC (set of 96 ‘Nordborg’ lines, CS22660) [21] or NASC (SALK or SAIL T-DNA lines: SAIL_167_A06, SAIL_658_G04, SAIL_448_B08, SALK_026383(BE), SALK_108492C, SALK_000461C, SALK_026685C, SALK_011676(A), SALK_030620(AI), SALK_020347C, SALK_028332 (BO), SALK_112558, SALK_022578C, SALK_060146, SALK_025883C, SALK_068266(BA), SALK_104906C, SALK_045666, SALK_057714(CF), SALK_123616(BV), SALK_047601, SALK_047837(AS), SALK_059126C, SALK_064966(AP), SALK_075567C, SALK_107827C, SALK_086488C, SALK_121520C, SALK_086554C, SALK_111688C) [22]; Table S8 lists location for each T-DNA line.
Our overall goal in preliminary growth experiments was to find conditions that maximized trait differences between high and low nitrogen environments. To carry out this exploratory phase, Arabidopsis seedlings were grown on a combination of different levels of carbon (0, 3, 10, 15, 30, 60 mM sucrose) and nitrogen (0, 0.03, 0.05, 0.1, 0.5, 1, 5, 10, 20 mM KNO3). For each seedling, primary root length was measured and the number of lateral roots counted; lateral root density was calculated from these two parameters (Table S1, Figure S1). Our previous work [9] showed that high levels of nitrogen induce lateral root primordium development and repress lateral root emergence, resulting in a higher pre-emergent∶emergent lateral root ratio than on low nitrate conditions. As a ratio this is also the case here, although total lateral root numbers on high nitrate are larger (due to the nutrient effect and longer primary roots; overall size effect).
An increasing concentration of nitrate was found to result in increased primary root length, particularly with concentrations of 0.5 mM KNO3 or more (Figure S1A). This inductive effect tended to level off at 5 mM KNO3, with primary root length remaining fairly constant at 10 and 20 mM KNO3. At lower levels of nitrate the primary root was longer with no or low sucrose in the media, but as the nitrate concentration increased this effect was reversed (primary root length was longer on higher sucrose concentrations. This is likely due to a C∶N balance effect [23]. It was also on higher sucrose concentrations that the nitrate inductive effect was more pronounced. A similar C/N effect was found on regulation of lateral root number, and again at more than 0.5 mM KNO3 the N effect was most pronounced, leveling off at 5 mM (Figure S1B). At the highest sucrose concentrations (30 and 60 mM sucrose) a significant increase in lateral root numbers was observed. Lateral root density was found to be relatively constant over all C∶N conditions, suggesting that in general, increases in lateral root number were proportional to primary root length (Figure S1C). However, there was a higher lateral root density for combinations of the highest C∶N levels (30,60 mM sucrose∶5,10,20 mM KNO3), suggesting that at these concentrations there is a developmental effect that leads to larger numbers of lateral roots developing. Thus, in order to understand the genetic basis of this developmental effect we decided to use 30 mM sucrose, 5 mM KNO3 as our ‘high N’ condition; on this combination there also appeared to be strong and near-maximal induction of primary root length and lateral root development (as indicated by the leveling off described above). As a comparative low N condition we decided to use 0.03 mM KNO3 (also at 30 mM sucrose) since root growth and development was significantly different from seedlings grown on 5 mM, and the plants would be N-depleted/starved but still viable and growing (compared to 0 mM KNO3, complete N starvation). To confirm that the root architecture difference that we observed were due to the effect of different nitrate levels rather than potassium levels we grew Col-0 seedlings on either 5 mM KNO3, 5 mM CaNO3, or 2.5 mM KNO3, 2.5 mM CaNO3 and found no major differences between overall root architecture (Table S2, Figure S2). Finally, we have some evidence that the root architecture observed in our chosen conditions correlates with that in field conditions, for example Var2-1 is found in sandy regions and exhibits a highly elongated primary root with very few lateral roots as seen in our experiments (see Figure 1).
For phenotypic analysis, seeds from each of 96 Arabidopsis thaliana accessions [18] or T-DNA lines were grown on vertical agar plates containing custom nitrogen and sucrose-free 1× Murashige and Skoog basal medium (GibcoBRL, Gaithersburg, USA) supplemented with 30 mM sucrose and either low (0.03 mM) or high (5 mM) KNO3 with 0.8% agar (pH 5.7). To confirm the effect of nitrate, KNO3 was replaced with CaNO3 for Col-0. For microarray studies 6,000 seeds (per replicate, in triplicate) of each accession (Col-0, Kas-2, Var2-1, Tamm-27, NFA-8, Sq-8, Ts-5) were sterilized and sown on liquid 1× Murashige and Skoog basal medium containing no nitrogen or sucrose supplemented with 3 mM sucrose and 0.5 mM ammonium succinate for hydroponic growth as previous [9]. Plants were grown for 12 days in 16 hr light (50 mmol photons m−2 s−1 light intensity)/8 hr dark cycles at 22°C in growth chambers. For determining growth conditions and for phenotyping of the 96 accessions, 10 seedlings were measured for one replicate of each condition/accession in New York in a Percival growth cabinet (Percival Scientific Inc., Perry, IA,). For T-DNA allele phenotyping, an average of 10 seedlings were measured for each of three independent replicates of each condition/allele: New York, T-DNA phenotyping Rep 1 in a Percival Scientific Inc; Warwick, T-DNA phenotyping Reps 2,3 in a Sanyo MLR-351, Panasonic Biomedical, Loughborough). To confirm presence of T-DNA insertions and loss-of-expression of candidate genes, roots were harvested for genotyping of isolated DNA and qPCR of isolated RNA (see Table S12). For treatments, KNO3 was added to the media to a final concentration of 5 mM for two hours [9]. Control plants were mock-treated by adding the same concentration of KCl. At the end of the two hour treatment, roots were harvested and flash-frozen in N2(l) for subsequent RNA extraction. To confirm trans non-complementation among alleles for each gene, we crossed the pairs of alleles to each other via reciprocal crossing. As a crossing control, individual alleles were also crossed to Col0. Root phenotypes in the F1s were compared to selfed Col0 plants grown in parallel.
In each KNO3 environment, parameters relating to root architecture were measured using ImageJ: primary root length (i, PR), number of lateral roots (ii, LR#), lengths of all LRs and LR distribution (number of LRs per cm of PR). From this the following were calculated: lateral root density (iii, LRdensity), the proportion of the PR that is the root branching zone (the zone of the parent root that extends from the most rootward emerged LR to the shoot base, LB, terminology following Dubrovsky and Forde (2012) [24]) (iv, LB/PR), total LR length (v, LRtot), total LR plus PR length (vi, PR+LRtot), average LR length (vii, LRlengthave); Table S3, Figure S3. Traits designated with roman numerals were used for GWAS. Shoot area was estimated to calculate shoot area to primary root length. Data was scaled from 0 to 1 using the scaling factor (n - low val)/(high val – low val); Table S4. Clustering of phenotyping values was carried out using hierarchical clustering with an average linkage and Pearson correlation using the clustergram function in MATLAB (The MathWorks, Natick, MA, USA). NA values were considered to have a value of 0. Silhouette widths were plotted in MATLAB using the silhouette function for each hierarchical tree and used to determine where to cut the trees and define clusters. A Perl script was written that produces a line drawing illustrating average seedling PR length, and lengths and distribution of LRs in each cm of the PR. This script can be accessed via URL: http://coruzzilab.bio.nyu.edu/cgi-bin/manpreetkatari/drawplant/drawplant.cgi. A positive hit in the reverse genetic screen was determined by satisfying the following criteria: (1) two separate mutant alleles showed the same phenotype, (2) mutant alleles showed a reduction or complete loss of expression using qPCR, 3) both mutant alleles showed a consistent, quantifiable phenotype in three independent screens including separate trials in New York and Warwick growth facilities.
To calculate heritabilities of the within-environment variables, we used the “lmer” function of the lme4 package [25] in R v.2.15.1 [26] and fit a restricted maximum likelihood (REML)-based analysis of variance (ANOVA) model of the form: Phenotype = Accession+Error, where Accession was treated as a random effect [27]. Heritabilities were calculated as σG/σP, where σG is the genetic variance component (the genetic variance component attributable to variation among accessions) and σP is the total phenotypic variance. To calculate the heritabilities of the response variables, we used the same function in R to fit a REML-based ANOVA model of the form: Phenotype = Accession+Nitrogen Level+Accession-by-Nitrogen Level+Error, where Nitrogen Level (high or low) was treated as a fixed effect, and Accession and Accession-by-Nitrogen Level were treated as random effects. Heritabilities were calculated as σGxE/σP, where σGxE is the variance component of the Accession-by-Nitrogen Level interaction effect [28].
GWAS was carried out using the EMMA package in R as described in Atwell et al (2010) [19]. The kinship matrix was constructed using the full set of ∼214k SNPs and SNPs with a minor allele frequency of 0.1 were mapped (∼178k/214k SNPs); see Figure S7. We opted to avoid what we believe is the overly stringent criteria of the Bonferroni correction and adjusted for multiple testing following Moran (2003) and Storey and Tibshirani (2003) [29], [30]. Thus, we calculated Q-values, in which the distribution of P-values is used to correct for the false positive rate [30]. Q-values were calculated separately for each trait using the “qvalue” package [31] in R, a well-established method for finding significant fold changes in the microarray literature (e.g. [32]). We note that, because Q-values are based on the distribution of the raw P-values and because Q-values are calculated separately for each trait, the raw P-value corresponding to our target threshold of Q = 0.05 (the significance threshold for SNP-trait associations) varies for different traits (see Figure S7). Compared to Atwell et al (2010) [19] we used a more stringent location criteria for selection of genes: for each significant SNP association, a window of 20 kb (rather than 40 kb) centered on the SNP (using the SNP-mapped TAIR8 genome version) was used to select genes predicted to be responsible for the association.
To understand the relationship between minor allele frequency (MAF), additive genetic effect size, and the power to detect an additive genetic effect, we performed a power simulation sensu Yu et al. (2006) [33]. Specifically: (i) the empirical phenotypic values for each accession were treated as random deviates; (ii) based on the empirical phenotypic variation, we calculated a genetic effect equal to 0.1, 0.2, 0.5, 0.7, 0.9, or 1 times the standard deviation of the phenotypic mean; (iii) x accessions out of the total n accessions were randomly assigned to one simulated genotype, and the rest of the individuals were assigned to the other simulated genotype, so that x/n equaled the minor allele frequency of interest; (iv) the genetic effect corresponding to an accession's simulated genotype was added to the empirical phenotypic value for that accession; (v) structured association mapping was performed using the real (non-simulated) kinship matrix; (vi) steps 1 to 4 were repeated 1000 times, and the power to detect the additive genetic effect was the proportion of times that the P value from the mapping analyses (see step v) was below the 0.05 significance threshold. We performed this power simulation for each trait and for MAFs ranging from 0.1–0.5; see Figure S8. The power simulations show similar results to Yu et al. (2006) [33], namely that the power to detect a genetic effect is low at small MAFs and at small genetic effect sizes; the power to detect a genetic effect increases dramatically with an increase in the genetic effect size, such that a genetic effect half as large as the random background variation will usually be statistically significant even at a low MAFs.
RNA from the whole root samples for microarray analysis was extracted with TRIzol (Invitrogen, Carlsbad, CA). Standard Affymetrix protocols were then used for amplifying, labeling and hybridizing 1 µg of RNA samples to the ATH1 GeneChip (Affymetrix, Santa Clara, USA). For qPCR tests, RNA was extracted with the RNAeasy kit (Qiagen) then first DNAase-treated using a Precision DNase kit and double stranded cDNA was synthesized using the nanoscript RT kit (both from Primer Design Ltd, Southampton, UK) according to manufacturer's instructions. qPCR was carried out using the Precision-SY MasterMix kit using primers designed by Primer Design Ltd according to manufacturer's instructions on a Roche 480 LightCycler. The mRNA levels were normalized relative to the UBQ10 housekeeping gene using the geNorm REF gene kit (Primer Design Ltd) and quantified using standard curves generated for each primer pair. Expression of At3g16470, At4g02860 and At4g02890 transcripts were used to confirm loss-of expression in the SALK lines vs. Col-0 (primers designed by Primer Design Ltd). SALK lines were PCR-genotyped with primer designed using T-DNA Primer Design (http://signal.salk.edu/tdnaprimers.2.html); for all primer sequences see Table S12.
Affymetrix GCOS software was used to verify that the arrays had similar hybridization efficiencies and background intensities for all accessions. We carried out an analysis to address the use of the Affymetrix Col-0 chip for other Arabidopsis accessions. Given the rate of SNPs between accessions and Col-0, we first observed that mismatches to any of the 11 probes (25mers) for any given gene were likely to be rare. In addition, we compared only N-deplete with N-replete Affymetrix signal values within each accession directly and only focused on genes that showed a difference between the two. Therefore any genes that cannot be detected because of sequence-associated probe hybridization problems do not confound our analysis. However, to ensure that we account for over/under-estimations of N-regulation significance that might result from stronger hybridization of a sequence in one accession compared to another (due to sequence difference), we developed an algorithm to rank the signal values of each element in each probe set across the experiments (7 accessions, 2 conditions (N-treatment and KCl control), 3 replicates). This was based on the expectation that, while overall signal from the probe sets of a given gene may change, the relative hybridization to each probe set for a given should not. The method identifies elements within probe sets whose expression is indicative of that element not hybridizing to accession-derived sequences due to the presence of SNP(s) using Col-0 as a reference. Significant deviation from this order could indicate sequence divergence altering the binding strength of a sequence to a probe element. We derived a null distribution of signal strength orders for Col-0 and then used this to identify significant probe element outliers in hybridizations from the other (see Table S9 for lists of all element outliers). These probe elements were discarded. Microarray data was subsequently normalized with MAS5 using all but these element values and implemented in the Affymetrix GCOS software (Table S10). On average, 10% of all probe sets were analyzed with the complete set of 11 elements and a further 70% analyzed with 9 or 10 probe elements (see Table S9 for details for each replicate set). The reproducibility of replicates was analyzed using the correlation coefficient and r2 value of replicate pairs in R; r2 values were typically in the range of 0.92 to 0.98, with the lowest being 0.91. Probe-gene mapping was made using the latest annotation file (TAIR10 annotation) (ftp://ftp.arabidopsis.org/home/tair/Microarrays/Affymetrix/affy_ ATH1_array_elements-2010-12-20.txt). The following classes of probes were flagged (Table S10): probes matching non-nuclear Arabidopsis thaliana genes or that had no gene match (flag #1), probes that had an ambiguous match to nuclear genes, i.e. matched more than one gene (flag #2), probes where several probes match a single gene (flag #3), probes whose average expression level was found to be below the detection cutoff (flag #4). To identify flag #4 probes we analyzed genes known to be expressed or absent in the root to calculate an expression signal of 100 as a cutoff for detection.
All genes were fit to the following ANOVA model: Y = μ+αaccession+αtreatment+αaccession* treatment+ε, where Y is the normalized signal of a gene, μ is the mean of the reference accession and treatment (intercept), the α coefficients correspond to the effects of accession, treatment (nitrogen) and the interaction between accession and treatment, and ε represents unexplained variance. Potential location effects were handled by growing plants together in a highly controlled environment and randomizing the placement of accessions and treatments in different shelves and locations of the growth chamber. The replicate trials were conducted in rapid succession in identical conditions, where we have not found significant time-effect variation. Thus, the ANOVA was modeled without block effects, where potential confounding effects were handled by randomization. Genes with a model P value less than a cutoff determined by setting the FDR [34] to 0.1 were analyzed further using model simplification to test these genes for significant N*Accession interaction effects, response to N, and variation across Accession. We did this by removing terms from the model one by one and then comparing the models to see if there was a significant difference in explanatory power between the simplified model and the more complex model using an FDR of 0.1. Gene expression values were averaged for each treatment, log2 converted, row normalized and clustered using hierarchical clustering with an average linkage and Pearson correlation using the clustergram function in MATLAB. Silhouette widths were plotted in MATLAB using the silhouette function for each hierarchical tree and used to determine where to cut the trees and define clusters. Clustering was carried out separately for genes that were determined by ANOVA to have a N,Accession effect, a N*Accession effect, or a N only effect, then the cluster patterns visualised together in MATLAB using the clustergram function to create Figure 3A. Two-tailed t-tests assuming equal variance were used to compare trait values for wild-type and mutant seedling roots, and trait values for Col-0 grown on different levels of sucrose and nitrate. For analysis of overrepresentation of GO terms we used the BioMaps function in VirtualPlant with default settings [35].
Root phenotypes for the seven transcriptionally profiled accessions (Figure S6) were analyzed using a mixed interaction model ANOVA using MATLAB (anovan function) with the following model: ROOT_TRAITn = ENVn+GENn+GENn * ENVn+en where Root Traitn is one of n = 7 root traits measured (PR, LRtot, PR+LRtot, LRlengthave, LR#, LRdensity, and LB/PR), ENV is environment, GEN is genotype, and e is error. Environment was modeled as a fixed effect and genotype was modeled as a random effect. P values were taken for each trait separately for the main and interaction effects. Coefficients generated from the ANOVA were used to determine the specific traits that contributed most to significant interaction effects (Table S5). As in the design for expression analysis, placement of plants was randomized in chambers, and this experiment was conducted at one time point.
Principal Components Analysis was performed in MATLAB using the princomp function with default parameters. Rows were accessions and traits were columns, where dimensionality reduction was performed on traits. The biplot function was used to map accessions in specific treatments (average trait values) in the new trait space and observe the contribution of original traits to each new component. We performed separate analyses on the combined HighN and LowN treatments for each accession, each condition alone, and δ highN-lowN of each accession to changes in nitrogen. We plotted two components on each biplot (1 vs 2; 2 vs 3) to analyze the first three principal components. See Figure S5.
We created a network of expression modules to traits by first clustering responses (expression in low nitrogen – expression in high nitrogen) using Pearson correlation and hierarchical clustering (average linkage, tree cut at R = 0.7). To determine significant clusters, we randomized the data and used the same clustering routine. This routine showed that clusters greater than 50 genes were observed less than 10% of the time by chance. Using that cutoff to define major clusters, we took the mean response of these major clusters in all 7 accessions. We then concatenated mean scaled trait values for δ highN-lowN and generated a correlation matrix, where R>0.7 or <−0.7 resulted in a significant edge. This resulted in a correlation matrix between gene expression clusters and traits that was used to generate the network depicted in Figure S9 using the biograph function in MATLAB.
|
10.1371/journal.pntd.0004225 | Molecular Epidemiology of Amoebiasis: A Cross-Sectional Study among North East Indian Population | Epidemiological studies carried out using culture or microscopy in most of the amoebiasis endemic developing countries, yielded confusing results since none of these could differentiate the pathogenic Entamoeba histolytica from the non-pathogenic Entamoeba dispar and Entamoeba moshkovskii. The Northeastern part of India is a hot spot of infection since the climatic conditions are most conducive for the infection and so far no systemic study has been carried out in this region.
Following a cross-sectional study designed during the period 2011–2014, a total of 1260 fecal samples collected from the Northeast Indian population were subjected to microscopy, fecal culture and a sensitive and specific DNA dot blot screening assay developed in our laboratory targeting the Entamoeba spp. Further species discrimination using PCR assay performed in microscopy, culture and DNA dot blot screening positive samples showed E. histolytica an overall prevalence rate of 11.1%, 8.0% and 13.7% respectively. In addition, infection rates of nonpathogenic E. dispar and E. moshkovskii were 11.8% (95% CI = 10.2, 13.8) and 7.8% (95% CI = 6.4, 9.4) respectively. The spatial distributions of infection were 18.2% (107/588) of Assam, 11.7% (23/197) of Manipur, 10.2% (21/207) of Meghalaya, and 8.2% (22/268) of Tripura states. Association study of the disease with demographic features suggested poor living condition (OR = 3.21; 95% CI = 1.83, 5.63), previous history of infection in family member (OR = 3.18; 95% CI = 2.09, 4.82) and unhygienic toilet facility (OR = 1.79; 95% CI = 1.28, 2.49) as significant risk factors for amoebiasis. Children in age group <15 yr, participants having lower levels of education, and daily laborers exhibited a higher infection rate.
Despite the importance of molecular diagnosis of amoebiasis, molecular epidemiological data based on a large sample size from endemic countries are rarely reported in the literature. Improved and faster method of diagnosis employed here to dissect out the pathogenic from the nonpathogenic species would help the clinicians to prescribe the appropriate anti-amoebic drug.
| Most epidemiologic studies in developing countries carried out for amoebiasis is either based on microscopy alone or culture/ microscopy used as a screening tool, have poor sensitivity and specificity and thus fails to figure out its true magnitude. The purpose of this study was to assess the true prevalence of amoebiasis in selected North Eastern states of India using DNA based screening technique followed by PCR assay for species discrimination. In addition, PCR assay confirmed that only 55.8% of the samples, resembling E. histolytica by microscopy, were true E. histolytica, implying that remaining 44.2% of so-called infections were due to other nonpathogenic Entamoeba spp. We found a higher prevalence of amebiasis (13.7%) using DNA dot blot screening compared to conventional microscopy and culture based screening. Poor living condition, previous history of infection in a family member, unhygienic toilet facility, children in age group <15 yr, participants having lower levels of education and daily laborers were identified as significant risk factors for amoebiasis. Thus, the techniques like DNA dot blot hybridization and PCR based detection adopted in the present study over and above the conventional screening methods can reduce misdiagnosis of the disease appreciably from the population living in this endemic area.
| Amoebiasis, an infection by protozoa E. histolytica is appraised as the third leading parasitic cause of human mortality after malaria and schistosomiasis, causing 40 thousand to 100 thousand deaths annually [1]. The re-classification of E. histolytica into Entamoeba complex comprising pathogenic E. histolytica and nonpathogenic E. dispar and E. moshkovskii has further added to the complexity of amoebiasis diagnosis and epidemiology.
Fecal microscopy, the most commonly used clinical diagnostic used for ages; particularly in resource-limited settings are unable to differentiate these three species except in rare invasive cases where fecal samples frequently found to contain hematophagous trophozoites. It was estimated that on an average only 1% of total E. histolytica infections develop into invasive form and rest remain asymptomatic [2]. Likewise, stool culture based diagnostic methods are time-consuming, laborious and often unrewarding, with a sensitivity of only about 50% [3]. Beside microscopy and stool culture, commercial ELISA based method is among the various other approaches followed for specific identification and detection of E. histolytica in fecal specimens [4–6]. However, few studies while diagnosing the parasite directly from the stool samples have shown poor sensitivity and specificity due to cross contaminations with other parasites. [7,8].
A number of polymerase chain reaction based assays have been developed over the years; mostly targeting unique regions of the SSU rRNA, as its high copy number provides increased sensitivity [9–12]. However, since the technique could not be made cost effective, therefore, till today prevalence rate reported from developing countries is either based on microscopy alone or molecular assay performed on culture/ microscopy screened samples which themselves have low sensitivities [12–18]. Thus, so far as epidemiology of amoebiasis is concerned, there is a paucity of available documented figure describing its true magnitude particularly from developing countries including India. In line with this, very little is known about the molecular epidemiology of amoebiasis in North Eastern population of India. The aim of the present study was to assess the epidemiologic picture of amoebiasis in selected North Eastern states of India during the 3 year period (2011–2014) using a sensitive and systematic protocol developed in our previous study [19].
A comparative cross-sectional study based on a single fecal sample per person was conducted to figure out the true prevalence of amoebiasis from January 2011 to January 2014. The study was carried out in four selected North Eastern states of India (Assam, Manipur, Meghalaya and Tripura) at the levels of community, healthcare facilities and hospitals.
After explaining the importance, purpose and procedure of the study, informed consents were obtained from study participants. For children aged 1 to 10 years consent was systematically sought from the family heads or guardians. Prior to our study, the study protocol was reviewed and approved by the Institutional Ethical Committee (IEC) of Gurucharan College, Silchar, Assam and Assam University, Silchar, Assam, India (IEC/AUS/2013-006).
About 5g of fresh fecal samples were collected in a pre-labelled, clean wide mouth screw-capped container. The samples were collected on the following day within 2–3 h of defecation and delivered to the laboratory and divided into aliquots. One aliquot of each of the fecal samples was used immediately for direct microscopy and inoculated for the establishment of a culture. A part of remaining aliquot was stored at 4°C for formal ether concentration of cysts (for screening by DNA dot blot hybridization), and the third aliquot is stored at -20°C for PCR assay. Samples from distant areas were collected in duplicate. One aliquot was preserved in 10% aqueous formalin for microscopy upon arrival in the laboratory. The other aliquot of the sample was inoculated in culture medium on the spot, and the rest was brought to the laboratory in unpreserved condition by maintaining temperature of approximately 4°C.
Iodine wet mounts of fresh unpreserved fecal samples were examined microscopically for demonstrating cysts and trophozoites of Entamoeba species complex. Briefly, a small fraction of feces was mixed with a small drop of Lugols iodine (diluted 1: 5 with water) on a microscope slide, and observed under microscope after placing a cover slip over the preparation. Irrespective of the microscopic analysis results, all fecal samples were cultured for Entamoeba species under xenic condition using biphasic (solid and liquid) Robinson’s medium within 5–6 h of collection as previously described [20]. The presence of characteristic spherical, oval or round shaped quadrinucleated cyst or trophozoites in fecal sample; and trophozoites emerging out of excysted cysts with ingested starch particles in xenic culture often showing clear pseudopodia were considered as the keys to confirm sample as positive microscopically. The culture, showing excysted cysts into trophozoites was further subcultured in biphasic Robinson’s medium and after 3–5 passages; the culture was expanded to increase the number of cells for isolation of genomic DNA.
Data on selected independent variables were collected by interviewing all the subjects using pre-designed questionnaire which consists of three sections: 1) General socio-demographic data: age, gender, residence, education, marital status, income and occupation, etc. 2) Environmental factors: toilet facility, water supply, animal contact, contacts with animal feces, etc. 3) Clinical information: anti-amoebic treatment taken previously, previous history of infection, symptomatic (stomach cramping, presence of mucus and blood in stool etc.) or asymptomatic at the time of sample collection etc.
DNA dot blot hybridization was performed for screening out the Entamoeba (Entamoeba histolytica and Entamoeba dispar) positive samples. The probe used for the purpose was HMe probe (EcoRI+ Hind III) as previously published [21]. Briefly, crude DNA was obtained from enriched cysts from stool samples directly by five freeze-thaw cycles followed by sonication. After denaturing crude cyst DNA using NaOH to a final concentration of 0.25 N, the DNA was spotted in triplicate on to the GS+ nylon membrane pre-saturated in 0.4 M Tris-Cl, pH 7.5 with the help of mini-fold apparatus. The air-dried and UV cross-linked blots were then ready for hybridization with 4.5 kb rDNA fragment (EcoRI—Hind III) from HMe region of EhR1 (rDNA plasmid in HM1: IMSS strain of E. histolytica).
Genomic DNA was extracted from an aliquot of 200 mg fecal sample using a DNA stool kit (Qiagen, Valencia, CA). Briefly, with the addition of five freezing-thawing cycles, samples were vortexed vigorously for 5–10 minutes in lysis buffer (ASL buffer). The samples were then processed according to the instructions of the manufacturer with slight variations, particularly incubation of the DNA in the spin column in elution buffer was carried out for 3 minutes at room temperature followed by centrifugation and this final elution step was repeated twice using 25 μl elution buffer each time to increase the DNA yield. The DNA was then stored at −20°C until used for PCR amplification.
For isolation of genomic DNA from cultured cell, trophozoites from the positive culture medium were harvested from 6–8 fully-grown culture tubes by chilling, followed by centrifugation at 600g for 5 minutes at 4°C. The cell pellet was then washed twice with 20 ml of PBS and finally stored in 70% ethanol at -20°C. The cells were pelleted through centrifugation at 13000 rpm for 4 min; air-dried to remove all traces of ethanol. The DNA was then isolated from pelleted cells using Genomic DNA mini kit (Real Genomics, Taiwan) following manufacturer’s instructions and finally eluted in 30–50 μl of elution buffer.
Forward and reverse oligonucleotide primers targeting the signature sequence of the infecting parasite were used for PCR assay. Amplification for E. histolytica was achieved using a nested PCR protocol with primer set E-1: 5/ TAA GAT GCA CGA GAG CGA AA 3/ and E-2: 5/ GTA CAA AGG GCA GGG ACG TA 3/ for primary PCR and primer set EH-1: 5/- AAG CAT TGT TTC TAG ATC TGA G-3/) and EH-2 (5/- AAG AGG TCT AAC CGA AAT TAG- 3/) for secondary PCR [22]. A common forward primer sequence EntaF was used for amplifying E. dispar and E. moshkovskii, whereas EdR and EmR were used as species-specific reverse primer for distinguishing the two species. Primer sequences used were as follows: EntaF: 5’-ATG CAC GAG AGC GAA AGC AT-3’ and EdR: 5’-CACCACTTACTATCCCTACC-3’) to detect E. dispar; EntaF: 5´-ATG CAC GAG AGC GAA AGC AT- 3´ EmR: 5´-TGA CCG GAG CCA GAG ACA T-3´to detect E. moshkovskii [23]. Briefly, all the PCR amplifications were performed in a final volume of 20μl with approximately 100ng of template DNA, 1 μM of each primer, 1X PCR buffer with 2.5 mM MgCl2, 1X BSA, 0.2 mM dNTPs, and 1U of Taq DNA Polymerase (Thermo scientific, Wattham, USA) in the thermal cycler (Bio-Rad Laboratories, Hercules, CA). Parasitic infection was confirmed by their expected amplicon sizes of 439 bp, 752 bp, and 580 bp for E. histolytica, E. dispar and E. moshkovskii respectively through gel electrophoresis. Some of the random signature amplicons were sequenced directly using respective primer pair using ABI 3500 Genetic analyzer (Applied Biosystems Inc., CA, USA) and subjected to homology search using nucleotide blast (blastn) program available at National Centre for Biotechnological Information (http://www.ncbi.nlm.nih.gov) database for further confirmation.
Data entry and statistical analysis were performed with the aid of SPSS statistical software version 16.0 (SPSS, Chicago, IL, USA) and MedCalc version 15.4 (MedCalc Software bvba, Belgium). Categorical variables were described using numbers and percentages. Descriptive statistics were used to show any association of disease with the variables like age, sex and others. We used Pearson’s Chi-square test at a level of significance P< 0.05 to test the associations of infection frequencies among groups in univariate statistical model. Frequencies of infection were used as the dependent variables, while the independent variables were environmental, socio-demographic factors and clinical status of participants. Odds ratios (OR) and 95% confidence intervals were computed to measure the strength of association between determinants of parasitic infection and burden of infection.
From a total of 1450 study subjects recruited from the 17 selected sites in the North Eastern states of India, 1260 participated in the cross-sectional survey, owing to an overall compliance of 86.9%. Reasons for non-compliance were absence of communication in the following day of sample collection (n = 13), absence of written informed consent/ questionnaire (n = 27), samples not deposited or because of production of insufficient specimens (n = 57) and few withdrew from the study without specific reason (n = 93). In total, 1260 stool samples were finally collected for the study.
Among this 1260 samples, 588 (46.7%) were from the Assam state, 268 (21.3%) from the Tripura state, 207 (16.4%) from the Meghalaya state and 197 (15.6%) from the Manipur state. Samples were collected over a period of three years from 2011–2014 during different seasons of the year, 373 (29.6%) were collected during pre-monsoon (Feb-May) season while 451 (35.8%) and 436 (34.6%) were collected during monsoon (Jun-Sep) and post-monsoon season (Oct-Jan) respectively (S1 Table).
Analysis of the 1260 fecal samples by microscopic examination of a direct saline (wet) mount, E. histolytica-like cysts or trophozoites were detected in 251 (~19%) samples, whereas, 152 (~12%) samples showed positive result in biphasic xenic culture (Fig 1). Interestingly, when crude DNA extracted from all the fecal samples (1260) were passed through third screening technique, DNA dot blot, 260 (20.6%) showed positive spots. On further analysis of our samples it was observed that amongst the 129 (10.2%) fecal samples that were positive in both microscopy and culture (last row of Table 1) only 118 were positive in DNA dot blot assay while remaining 11 were negative suggesting the false positives associated with conventional assay. Similarly, 122 that were exclusively positive in microscopy and 23 fecal samples that were positive in culture only, dot blot hybridization detected Entamoeba complex in 83 and 20 respectively. Thus the remaining 53 samples that were positive in either of the two conventional screening techniques that is the microscopy and culture, but negative in DNA dot blot assay were repeated again for hybridization, but did not yield a positive signal. While, fecal samples in which no cysts and trophozoites stages of Entamoeba complex was detected by microscopic and/ xenic culture examination, 39 were identified as positive when screened by dot blot assay suggesting the association of false negatives with conventional screening method. The samples positive using any of these three screening methods were then subjected to species specific singleplex PCR assay on genomic DNA isolated from stool samples directly. Genomic DNA from each species was used as positive controls.
PCR amplification targeting signature sequence of small ribosomal RNA gene of E. histolytica, E. dispar and E. moshkovskii produced diagnostic amplicons of 439 bp, 752 bp, and 580 bp respectively. Amongst 122 fecal samples that were positive only in microscopy using wet preparation, 19 (15.6%) were E. histolytica (mono-infection) infected, 21 (17.2%) were E. dispar (mono-infection) infected, 28 (23.0%) were E. moshkovskii (mono-infection) infected, 7 (5.7%) were infected both with E. dispar and E. moshkovskii (mixed-infection) and 36 (29.5%) were mixed infections of E. histolytica with either E. dispar or E. moshkovskii or all the three species, while 11 (9.0%) were negative in all the three PCR assays (Table 1). Similarly, when PCR was performed on 23 fecal samples that were positive by culture only, we failed to amplify the signature location in DNA isolated from 3 (13.0%) samples while among the remaining 20 samples E. histolytica (mono-infection) infections were found in 9 (39.1%), E. dispar (mono-infection) infections were found in 4 (17.4%) and mixed infections of E. histolytica with either E. dispar or E. moshkovskii or both were found in 7 (30.4%) samples. Among the 260 dot blot positive samples, mono-infections of E. histolytica and E. dispar were detected in 111 (42.7%) and 87 (33.5%) respectively, and mixed infections of both in 62 (23.8%) samples. Thus, comparison of molecular technique with classical techniques (microscopy and culture based) revealed that the DNA dot blot hybridization technique followed by validation with PCR is necessary to arrive at the true prevalence of E. histolytica among the study population. The detailed analysis of various screening techniques and their outcome is represented in the Fig 1.
The overall prevalence of any of the three morphologically indistinguishable Entamoeba species (pathogenic and non-pathogenic) was 23.2% (95% CI = 20.9%, 25.6%). Table 2 shows that 13.7% (173/1260; 95% CI = 11.9, 15.7) and 11.8% (149/1260; 95% CI = 10.2, 13.8) of the subjects were infected with E. histolytica and E. dispar, respectively. This 13.7% of the total samples (1260) were positive in the PCR assay either singly for E. histolytica or in combination with other intestinal protozoan parasites. Clinical specimens such as fecal sample often contain PCR inhibitors even after purification steps during genomic DNA isolation. In order to rule out this possibility, 21 culture and/ or microscopically positive, but PCR negative samples were further seeded with control DNA of HM1: IMSS strain of E. histolytica. In all the cases spiking with control DNA yielded a positive amplification suggesting that this 21 microscopy and/ or culture positive samples were actually false positive and thus negative results obtained in PCR assay is not because of PCR inhibitors.
Among the 1260 fecal samples collected from the four North Eastern states viz., Assam, Meghalaya, Manipur and Tripura at the level of community health care units and hospitals, highest E. histolytica prevalence was recorded in Assam 18.2% (95% CI = 15.2, 21.6), followed by 11.7% (95% CI = 7.7, 17.2) in Manipur, 10.2% (95% CI = 6.5, 15.3) in Meghalaya, while 8.2% (95% CI = 5.3, 12.3) in Tripura had the least (Table 3). Prevalence of E. dispar was highest with 14.6% (95% CI = 11.9, 17.8) in Assam followed by 12.3% (95% CI = 8.7, 17.0) in Tripura, 9.2% (95% CI = 5.8, 14.2) in Meghalaya and 5.6% (95% CI = 3.0, 10.0) in Manipur. Prevalence of E. moshkovskii was almost equal in Assam with 11.4% (95% CI = 9.0, 14.3) and the Meghalaya state with 11.1% (95% CI = 7.3, 16.4) (Fig 2).
Univariate analysis of demographics and the prevalences of E. histolytica infection were presented in Table 4. The prevalence showed an age dependency association, with significantly higher infection rates among respondents aged less than 15 years (OR = 3.06; 95% CI = 1.90, 4.94; P< 0.001) and in the age group 15–30 (OR = 2.36; 95% CI = 1.40, 3.97; P< 0.001). It was observed that the prevalence rate decreased from 17.6% to 8.9%, with higher education level of the participants (OR = 2.18; 95% CI = 1.31, 3.63; P = 0.003). Six hundred eighty one (54.0%) of the participants were from rural areas. Infection was higher among respondents from the rural population (OR = 1.62; 95% CI = 1.16, 2.27; P = 0.004) than those from urban population.
Marital status and gender bias was not significantly associated with the prevalence of E. histolytica infection, although female (15.2%) had slightly higher prevalence rate compared to male (11.5%). Among other socio-demographic factors, as in various occupational groups, the school students (OR = 2.31; 95% CI = 1.32, 4.05; P = 0·005), followed by truck drivers (OR = 2.10; 95% CI = 1.14, 3.85) and the merchant group (OR = 1.17; 95% CI = 0.60, 2.29) were at higher risk compared to the public service employee group. In context to the seasonal impact on the prevalence, as expected, we observed a significant higher infection rate during the monsoon season (June–September), approximately 21% (OR = 2.78; 95% CI = 1.84, 4.13; P< 0.001) compared to the pre- or post-monsoon seasons. A month wise variation pattern of prevalence rate was shown in Fig 3. Further univariate analysis of the socio-demographic factors showed that the infection was independent of per day income, marital status, gender and family size.
Table 5 showed the results of regression analysis of various potential factors associated with E. histolytica infection rate. We observed participants having unhygienic toilet facility more likely to be infected with E. histolytica compared to those having hygienic toilet facilities (OR = 1.79; 95% CI = 1.28, 2.49; P = 0.001). In terms of percentage value, participants with a family history of gastrointestinal infection and those have taken anti-amoebic treatment previously were 3.2 (OR = 3.18; 95% CI = 2.09, 4.82; P< 0.001) and 1.9 times (OR = 1.93; 95% CI = 1.40, 2.67; P< 0.001) more likely to be infected. Similarly, subjects having a previous history of infection in life were 1.4 times more likely to be infected compared to those who were not previously infected (OR = 1.40; 95% CI = 1.02, 1.94; P = 0.038).
With respect to behavioral characteristics, participants those directly using river, pond water for daily use were more likely to be infected with E. histolytica (OR = 1.78; 95% CI = 1.22, 2.56; P = 0.003) compared to the group using tap water as a drinking water source. The odds ratio of E. histolytica infection in participants who belong to poor quality of living condition is 3.21 times higher than those who live a better quality of living. Infection was higher among participants with clinical signs like stomach pain and cramping, passage of either watery or mucous with bloody stool etc. (OR = 1.57; 95% CI = 1.14, 2.18; P = 0.005). The data confirmed that individuals who were in close contact with domestic animals, i.e., dogs and cats were around 1.3 times (OR = 1.27; 95% CI = 0.92, 1.75; P = 0.149) more likely to be infected with E. histolytica compared to those who do not keep domestic animals as their pets, however the difference was not statistically significant. Similarly, consumption of raw vegetables was not significantly associated with E. histolytica infection.
High rate of parasitic infections encountered in this part of the sub continent and especially the endemic nature of the disease call for improved method of diagnosis. Development of rapid and accurate identification methods are essential for public health efforts to manage the disease. Various techniques such as microscopy, culture, zymodeme analysis, ELISA and DNA based methods are being followed for specific identification of E. histolytica in fecal specimens. Recent study highlighted the failure of TechLab ELISA kit; in detecting E. histolytica in some of the E. histolytica PCR confirmed samples [7,8]. Microscopy has a sensitivity of only 60%, even under optimal standards while fecal culture is less sensitive than microscopy as a detection method [6,22]. In our study, of the 251 samples that were microscopically positive, 56 were E. histolytica and 84 were mixed infections with E. histolytica. Thus, only 55.8% of the samples, resembling E. histolytica by microscopy, were true E. histolytica as confirmed by PCR assay, implying that remaining 44.2% of so-called infections were due to other two Entamoeba spp. (Fig 4). A study conducted among prisoners and primary-school children in Ethiopia highlighted 91.4% of the microscopy positive samples as E. dispar [24]. In another study reported from Australia, 50% of the microscopy positive fecal samples were found to be positive for nonpathogenic E. moshkovskii in the PCR assay [25]. The negative PCR result in 18 microscopy positive fecal samples is probably because of the presence of other Entamoeba species inhabiting the human gut. However, this needs further confirmation using molecular tools to validate the existence of other commonly found Entamoeba species in humans.
As shown by the results of the present study, the three species of Entamoeba namely E. histolytica/E. dispar/E. moshkovskii are prevalent in North Eastern states of India with an overall prevalence of 23.2%. The prevalence rate of E. histolytica observed in our cross-sectional study conducted at the community level healthcare unit and hospital using a molecular technique was 13.7%. Because of a better sensitivity of the two molecular methods employed here like DNA dot blot and PCR based methods together, helped us to correctly arrive at the true prevalence of E. histolytica in the samples collected from this region. This would not have been possible, employing culture and microscopy methods in isolation. Thus the diagnostic sensitivity can be improved by employing above techniques while carrying out epidemiological study in a region particularly endemic for the parasite. According to a recent review 15–20% of the Indian population is affected by E. histolytica [26]. Studies from different parts of the world indicate variable rate of E. histolytica prevalence in the fecal samples. A prevalence rate of 13.2% for E. histolytica and 9.9% for E. dispar was reported from Orang Asli settlements in Malaysia using real time PCR conducted on microscopy positive samples [27]. A much higher E. histolytica and E. dispar prevalence rate of 69.6% and 22.8%, respectively was reported using PCR assay among children in Gaza, Palestine [28]. However, it is very difficult to compare the true prevalence of amoebiasis because of the lack of uniformity in diagnostic methods. Much of the data reported are either based on microscopy alone or PCR assay performed on microscopy screened samples which itself has poor sensitivity. Moreover, it is now well documented that E. dispar infection is much more prevalent than E. histolytica worldwide [29,30]. In Agboville town near Abidjan, PCR analysis of microscopically positive samples demonstrated the ratio of E. histolytica to E. dispar of 1:46 [31]. Presence of non pathogenic E. moshkovskii has also been reported from countries like Bangladesh, Turkey, India, Iran, Australia, Tanzania and Malaysia and usually they are not associated with disease [4,13,25,32–34].
Studies from different geographical areas of the globe reported that the intensity of intestinal parasitic infections (IPIs) including E. histolytica was significantly higher among children [35–37]. However, our results did not show any significant difference in the prevalence of E. histolytica infection when compared between genders. This supported earlier observations made in different parts of the world [35,38,39]. In contrast, most hospital-based studies reported gender dependent E. histolytica infection [40–43]. The association between infection and occupational status indicated that student/ pre-school and daily laborers, including farmer, driver were the two groups who presented more than a twofold increased risk compared to Gov’t employers. This could be attributed to the fact that former groups frequently consume street foods not maintained in required hygienic conditions. Further, the participant’s level of education also exhibited significant association with E. histolytica infection. Rural background of respondents was also significantly associated with E. histolytica infection. As shown by other previous studies [44,45,46], our study further confirmed a higher risk of E. histolytica infection among the rural population, where prevailing poverty, no exposure to health education program, poor socioeconomic status, low standards of sanitation and hygiene are the associated factors that contributed to the high rate of infection. As expected, we observed significantly higher infection rate among participants with diarrhea or other gastrointestinal symptoms compared to asymptomatic group. This finding is in parallel with the studies conducted in Malaysia, Turkey, and Sweden [47–49]. A recent review suggested that asymptomatic cyst passage, with 90% of human infections either asymptomatic or mildly symptomatic, is considered to be the most common manifestation of E. histolytica. However, the above conclusion was based on studies made by fecal microscopy [6]. The risk of harboring the non-pathogenic species cannot be ruled out. In our study, interestingly 7 individuals mono-infected with E. moshkovskii were found to be symptomatic (S1 Table). In a separate study from India and Malaysia, association of E. moshkovskii infection with dysentery has been reported [13,34]. However, further studies on more samples are necessary to validate the role of E. moshkovskii in gastroenteritis disorders and its virulence.
Logistic regression analysis indicated that the factors responsible for infection can be pointed to poor living conditions, unhygienic toilet facility, not washing hands before taking food due to which infection rates increased by 3.21, 1.79 and 1.68 fold respectively. Similar risk factors have been described for the infection in population from Italy and Yemen [50,51]. Our data also revealed that the likelihood of acquiring infection due to the parasite among participants who have a record of pervious infection history and those had taken anti-amoebic chemotherapy were 1.4 and 1.9 fold, suggesting the possibility of harboring higher drug tolerant strains among the North Eastern population. However, further studies are warranted, particularly focusing on the metronidazole sensitivity of natural and clinical isolates of E. histolytica. Our observation of acquiring infection was three times higher in individuals having a history of infection in the family members. This finding was in line with previous studies carried out among the population of El Salvador, Mexican and Orang Asli Ethnic Groups of Malaysia where person-to-person transmission was indicated as the most important determinant of infection [52–54]. Therefore, it is recommended to screen the stool samples of every family member on a routine basis and any person found infected with the pathogenic species should be treated with the antiamebic drug. As expected, we observed highest prevalence of E. histolytica in the monsoon season followed by the pre- and post-monsoon seasons. This could be attributed to the high rate of fecal–oral contamination during monsoon season.
Our study did not reveal any significant association of E. histolytica infection with the individuals having close contacts with domestic animals. In contrast to this, reports from countries like Nigeria, Yemen and Malaysia reported an increase in the prevalence of Entamoeba complex infection among individuals having close association with domestic animal [55–57]. Recently, E. hartmanni, E. coli and E. dispar were isolated from captive non-human primates housed in the zoological garden of Rome, highlighting the risk of zoonotic transmission of this parasite for animal caretakers and visitors [58]. E. histolytica infection was also found to be prevalent among dogs of younger age group [59]. A report on the molecular detection of E. histolytica/dispar infection among wild rats in Malaysia corroborates further the risk of zoonotic transmission [60]. Therefore, potential risk of zoonotic agents cannot be ruled out and indicates the importance of developing control measures to prevent transmission by zoonotic mode. To understand the actual dynamics of transmission in North Eastern population of India, genotyping of E. histolytica strains from humans and animals is highly recommended.
In conclusion, the present study conducted among four North Eastern states showed the highest prevalence rate of E. histolytica among participants from Assam state. In addition, we have been able to resolve using molecular based techniques, the issue of high rates of microscopically positive samples. The techniques like DNA dot blot hybridization and PCR based detection methods adopted in the present study over and above the conventional screening methods can reduce misdiagnosis of the disease appreciably from the population living in this endemic area. The various logistics associated with the disease that are described here would help the clinicians to better diagnose the patients. Adoption of these diagnostic techniques would help to assess the true epidemiology of this endemic disease prevailing in different parts of India.
|
10.1371/journal.pntd.0002461 | Impact of Neutrophil-Secreted Myeloid Related Proteins 8 and 14 (MRP 8/14) on Leishmaniasis Progression | The myeloid-related proteins (MRPs) 8/14 are small proteins mainly produced by neutrophils, which have been reported to induce NO production in macrophages. On the other hand, Leishmania survives and multiplies within phagocytes by inactivating several of their microbicidal functions. Whereas MRPs are rapidly released during the innate immune response, their role in the regulation of Leishmaniasis is still unknown. In vitro experiments revealed that Leishmania infection alters MRP-induced signaling, leading to inhibition of macrophage functions (NO, TNF-α). In contrast, MRP-primed cells showed normal signaling activation and NO production in response to Leishmania infection. Using a murine air-pouch model, we observed that infection with L. major induced leukocyte recruitment and MRP secretion comparable to LPS-treated mice. Depletion of MRPs significantly reduced these inflammatory events and augmented both parasite load and footpad swelling during the first 8 weeks post-infection, as also observed in MRP KO mice. On the contrary, mouse treatment with recombinant MRPs (rMRPs) had the opposite effect. Collectively, our results suggest that rapid secretion of MRPs by neutrophils at the site of infection may protect uninfected macrophages and favor a more efficient innate inflammatory response against Leishmania infection. In summary, our study reveals the critical role played by MRPs in the regulation of Leishmania infection and how this pathogen can subvert its action.
| Parasites of the Leishmania genus have developed multiple mechanisms to subvert the immune response. Among these mechanisms are the activation of host phosphatases and inactivation of cell signaling pathways, which in turn activate the immune response. On the other hand, it has been observed that the Myeloid Related Proteins (MRPs) 8 and 14 are potent activators of some components of the immune response. In this study, we evaluated the effect of MRPs 8 and 14 on the progression of cutaneous Leishmaniasis. To do so, we used immortalized macrophages and stimulated them with MRPs before or after infection with L. major. We observed that stimulating macrophages with MRPs prior to infection induced NO and TNF-α production, as well as phosphorylation of MAPKs and nuclear translocation of transcription factors NF-κB and AP-1. However, when MRP stimulation was performed after infection, these effects where subverted. Moreover, using a murine model of cutaneous infection, we observed that depletion of MRPs caused increased parasite burden and bigger lesions. On the contrary, injection of recombinant MRPs directly into the lesion, considerably reduced lesion size and parasite burden. Our study suggests that MRPs could have a potential therapeutic use in the control of Leishmania infection.
| Myeloid-related proteins 8 and 14 (MRPs 8/14) also known as S100A8 and S100A9 are small calcium binding cytoplasmic proteins secreted mainly by neutrophils and monocytes [1], [2]. These proteins are formed by two Ca2+ binding domains separated by a hinge region [3]. Although these proteins exist as homodimers, a heterodimer (MRP 8/14) is formed in the presence of calcium. Both proteins are expressed abundantly by neutrophils, being around 30 to 40% of their cytoplasmic proteins [4]. MRP 8 and 14 are not constitutively expressed by macrophages; however, expression of MRP 8 can be achieved in those cells by stimulation with LPS, IFN-γ, IL-1β and TNF-α. Interestingly, murine endothelial cells express both MRP 8 and MRP 14 following LPS stimulation [5]. Murine MRP 8 is chemotactic for neutrophils and monocytes, whereas human MRP 14 and the heterodimer MRP 8/14 are chemotactic for neutrophils, stimulate their adhesion to fibrinogen, and enhance monocyte transmigration across endothelial cells. It is also known that MRP 8 and 14 inhibit bacterial growth possibly by zinc chelation and by preventing bacterial adhesion to mucosal epithelial cells [6].
MRP 8 and 14 have been associated with a number of inflammatory diseases leading to the assumption that these molecules are involved in the body's defense against inflammation. Phagocytes expressing MRP 8 and 14 are found in a variety of inflammatory conditions, including rheumatoid arthritis, chronic bronchitis and inflammatory bowel disease [7], [8]. Moreover, Tessier and collaborators [1] reported that in the murine air-pouch model, stimulation with LPS led to an abundant recruitment of neutrophils and subsequent secretion of MRPs [2].
We have previously reported that MRP 8 and 14 play an important role in the nitric oxide (NO) modulation; a key microbicidal function of macrophages [9]. This increase was linked with augmented expression of inducible nitric oxide synthase (iNOS), at the gene and protein levels, concomitantly with ERK and JNK kinases phosphorylation and the rapid NF-κB nuclear translocation. These findings indicate that MRPs play an important role during inflammation.
Although much is known about MRPs during inflammation and inflammatory diseases; little is known about the potential role of MRPs in Leishmaniasis. Leishmaniasis (caused by parasites of the Leishmania genus), is a disease characterized by three main clinical manifestations; cutaneous Leishmaniasis, muco-cutaneous Leishmaniasis and the lethal if untreated visceral Leishmaniasis. Leishmania parasites of different species are able to abrogate the innate immune response in order to survive inside their host cell [10]. In regard of the role of MRPs in Leishmaniasis, only two reports have documented accumulation of macrophages expressing MRP 8 and 14 at the skin lesions of mice infected with L. major [11], [12]. They also found, that amastigotes isolated from skin lesions presented MRP 8 and 14 adhered onto their surface. However, and despite these observations, the role of these proteins during Leishmania infection has not been investigated.
Herein, we report the first study concerning the role of MRPs during Leishmania infection in a murine experimental model. More precisely, we found that MRP-primed macrophages infected by L. major exhibit antimicrobial activity, whereas unprimed L. major-infected cells were fully inactivated, showing no response to MRP stimulation. Using in vivo approaches, we further demonstrated that L. major's capacity to recruit inflammatory cells was accompanied by MRP secretion at the site of inoculation. The use of anti-MRP antibodies in addition to blocking Leishmania-induced leukocyte recruitment in the air-pouch also increased mice footpad swelling and parasite load. Similarly, MRP deficient mice were found more sensitive to develop footpad swelling. Importantly, use of recombinant MRPs (rMRPs) to treat infected footpads led to significantly reduced footpad swelling and lesion development as well as a reduced parasite load. Altogether, this study provides a clear demonstration that MRPs seem to play a critical role in the control of the progression of Leishmania infection by modulating the innate inflammatory and microbicidal responses.
The research involving animals in this work was carried out according with the regulations of the Canadian Council of Animal Care and approved by the McGill University Animal Care Committee (AUP#4859). BALB/c and C57Bl/6 mice were obtained from Charles River and Jackson Laboratory, respectively. MRP14 KO mice (C57Bl/6 background) were obtained from Dr. Philippe Tessier's laboratory at Laval University, QC. Canada.
Immortalized murine bone marrow derived macrophages B10R cell line were grown at 37°C in 5% CO2 in Dulbecco's Modified Eagle medium (DMEM) supplemented with 10% heat inactivated FBS (Invitrogen, Burlington ON, Canada) and 100 U/ml penicillin 100 µg/ml streptomycin and 2 mM of L-glutamine (Wisent, St. Bruno, QC, Canada). Leishmania promastigotes (L. major A2 and L. major luciferase) were grown and maintained at 25°C in SDM-79 culture medium supplemented with 10% FBS by bi-weekly passage. Macrophages were infected at a parasite-macrophage ratio 20∶1 with stationary phase promastigotes for the times specified in each figure legend. When cells were primed with 5, 10 and 25 µg/ml of MRPs 8/14 heterodimer were used 1 hr before infection and remained throughout the infection time. All reagents if not indicated were obtained from Sigma Aldrich (St-Louis MO, USA).
Cloning expression, and purification of mouse MRP 8 and 14 (S100A8/A9) were previously described [1], [2]. Briefly, mouse S100A8 cDNA was cloned into the pET28a expression vector (Novagen). Murine S100A9 cDNA was obtained by RT-PCR and cloned into the PET28a vector (Dr. Philippe Tessier's laboratory, Laval University, QC. Canada). Recombinant protein expression was induced with 1 mM isopropyl-β-D-thiogalactoside in E. coli HMS174 for 16 hr at 16°C. After incubation, the bacteria were centrifugated and the pellet resuspended in PBS/NaCl (0.5 M)/imidazole (1 mM) and lysed by sonication. The pellet was centrifugated and the supernatant collected. Recombinant His-Tag proteins were purified using a nickel column; S100A8/A9 bound to the column were freed from their His-Tag by incubation with 10 U of biotinylated thrombin for 20 hr at RT. Finally the proteins were passed through a polymyxin B agarose column (Pierce, Rockford, IL USA) to remove endotoxins. The lysate, contamination by endotoxins was <1 pg/µg. The proteins were kept at −80°C until further use.
B10R macrophages were plated in 12-well plates (0.5×106 cells/well, in triplicates). The next day, cells were pre-treated for 24 hr with 5, 10 and 25 µg/ml of MRPs 8/14 and then infected with L. major (20∶1) for another 24 hr (MRP-primed-infected); or pre-infected with L. major for 24 hr and then stimulated with 5, 10 and 25 µg/ml of MRPs for further 24 hr (infected-MRP-stimulated). NO production was assessed by measuring the accumulation of nitrites in the cell culture medium using the colorimetric Griess reaction as previously described [13].
B10R macrophages were plated in 12-well plates (0.5×106 cells/well). Next day, cells were stimulated: MRPs alone, MRP-primed-infected or infected-MRP-stimulated. After the different times of stimulation (indicated in the figure legend) or infection, plates were centrifugated at 2500 rpm and 100 µl of supernatant were collected and added to TNF-sensitive L-929 fibroblasts [14] previously pleated in 96-well plates (3×105 cells/well/100 µl), in the following day 100 µl of B10R culture supernatant were added to the L929 cells making a 2-fold serial dilution. Actinomycin D (final concentration of 2 µg/ml) was added to each well and plates were incubated 18 hr at 37°C. Next day, live cells were stained with crystal violet (0.05% in 0.1% acetic acid solution) for 10 minutes. After, plates were washed to remove excess of stain and 100 µl of 100% methanol were added to each well to elute stain from the cells. Plates were red at 595 nm. Data are expressed as unit of TNF referring to the dilution that induced 50% of L929 cell death.
B10R macrophages (2×106) stimulated with MRPs alone (1 hr), MRP-primed (1 hr)-infected (1 hr) or infected (24 hr)-MRP-stimulated (1 hr) were washed three times with PBS to remove non-internalized parasites, and processed for nuclear extraction as previously described [15], [16]. Briefly, macrophages were collected in 1 ml of cold PBS, centrifuged and pellets were resuspended in 400 µl of ice-cold buffer A (10 mM HEPES, 10 mM KCl, 0.1 mM EDTA, 0.1 mM EGTA, 1 mM DTT and 1 mM of PMSF) and incubated 15 min on ice. 25 µl of IGEPAL 10% were added, and samples vortexed for 30 sec. Nuclear proteins were pelleted by centrifugation and resuspended in 50 µl of cold buffer C (20 mM HEPES, 400 mM NaCl 1 mM EDTA, 1 mM EGTA 1 mM DTT and 1 mM PMSF).
Protein concentrations were determined by Bradford assay (Bio-Rad, Hercules CA, USA). 6 µg of nuclear proteins were incubated for 20 min at room temperature with 1 µl of binding buffer (100 nM Hepes pH 7.9, 8% v/v glycerol, 1% w/v Ficoll, 25 mM KCl, 1 mM DTT, 0.5 mM EDTA, 25 mM NaCl, and 1 µg/µl BSA) and 200 ng/µl of poly (dI-dC), 0.02% bromophenol blue and 1 µl of γ-P32labeled oligonucleotide containing a consensus sequence for AP-1 binding complexes (5′-CGTTTGATGACTCAGCCGGAA-3′) (Santa Cruz Biotechnology Inc, Dallas, TX, USA), NF-κB (5′-AGTTGAGGGGACTTTCCCAGGC-3′) (Santa Cruz Biotechnology Inc) and STAT1 (5′-AAGTACTTTCAGTTTCATATTACTCTA-3′). After incubation, DNA-protein complexes were resolved by electrophoresis in non-denaturing polyacrylamide gel 5% (w/v). Subsequently gels were dried and autoradiographed. Competition assays were conducted by adding a 100-fold molar excess of homologous unlabeled AP-1 oligonucleotide, or the non-specific competitor sequence for SP-1 binding (5′-ATTCGAATCGGGGCGGGGCGAGC-3′).
B10R cells (1×106) stimulated with MRPs alone (30 min), MRP-primed (30 min)- infected (1 hr) or infected (1 hr)-MRP-stimulated (30 min) were washed 3 times with PBS and lysed with cold buffer (50 mM Tris-HCl pH 7.0, 0.1 mM, 0.1 mM EGTA, 0.1% 2-mercaptoethanol, 1% NP-40, 40 µg/ml aproptinin, 20 µg/ml of leupeptin 100 mM PMSF, and 20 mM NaVO4). Proteins were dosed by Bradford (Bio-Rad), and 30–60 µg of proteins were separated by SDS-PAGE, and transferred onto PVDF membranes (GE healthcare, Piskataway NJ, USA). Membranes were blocked in 5% bovine serum albumin (Wisent), washed and incubated ON with anti-phospho or total ERK, phospho or total -JNK, iNOS (Cell signaling, Ipswich, MA, USA) or β-actin. After washing, membranes were incubated 1 hr with α-rabbit or mouse HRP-conjugated antibody, and developed by autoradiography.
To determine parasite survival inside B10R macrophages, cells were plated in 12-well plates (0.5×106cells/well) and the following day they were infected with stationary phase L. major-LUC promastigotes (10∶1 ratio). After 6 hr of infection, the non-phagocytosed parasites were removed by washes with PBS, and samples were collected. For the second group of samples, fresh media was added and cells were incubated for another 18 hr. Adherent macrophages were collected and centrifuged 13,000 rpm×1 min. Pellets were lysed in 25 µl of 1× Cell Culture Lysis Reagent (Promega, Fitchburg, WI, USA). 20 µl of lysate were mixed with 90 µl of Luciferase Assay Reagent (Promega) and luciferase counts were determined using a Mini Lumat LB 9506 luminometer (EG&G).
Air pouches were raised on the dorsum of 6 week-old BALB/c by s.c. injection of 3 ml of sterile air on days 0 and 3. On day 6, 1 ml of LPS (1 µg/ml) or 5×106 parasites of Leishmania major in 1 ml of PBS were injected into the air pouches. At 6 hr, mice were sacrificed and air pouches were washed twice with 2 ml of PBS. Exudates were centrifuged at 1200 rpm for 5 min. Cells were counted with a hematocytometer. Characterization of leukocyte subpopulations migrated into the pouch space was performed by diff-quick staining of cytospins. In some experiments mice were injected i.p. with 4 mg of purified rabbit IgG anti-MRP8/14 16 hr before infection.
To determine the concentration of MRP 8/14 (S100A8/A9) in the air pouch, ELISAs were performed as previously described in [2]. Briefly, Costar high binding 96-well plates (Corning Glass, Tewksbury MA, USA) were coated overnight at 4°C with 100 µl of purified rabbit IgG against MRP8 or MRP14, diluted in 1 µg/ml in 0.1 M of carbonate buffer, pH 9.6. The wells were blocked with PBS/0.1% Tween 20/2% BSA for 30 min at room temperature. Then the samples and the standards (100 µl) were added, and after 45-min period at room temperature, the plates were incubated with rat IgG (100 µl/well) against MRP8 and MRP14 diluted in PBS/0.1% Tween 20/2% BSA for 45 minutes. To reveal the immune complex, 100 µl/well of peroxidase-conjugated goat-anti-rat was added and incubated for 45 minutes. Next 100 µl/well of 3,3′, 5,5′-tetraamethylbenzidine substrate (Research diagnostics, Las Vegas, NV, USA) were added according to the manufacturer's instructions, and ODs were read at 500 nm. The lower limit of quantification was determined as 4 ng/ml for both MRP8 and MRP 14, and 10 ng/ml for the heterodimer. All ELISAs were tested using excess amounts of the other S100 proteins and were shown to be specific under conditions reported in this work.
L. major stationary phase promastigotes (5×106 in 50 µl of PBS) were injected in mice's right hind footpad. Footpad thickness measurement was performed as previously described [17] for 10–12 weeks. For the group of MRP neutralization, anti-MRP 8/14 (4 µg/ml) were injected i.p. 1 day after infection and then at days 3, 6, 9, 12, 15, 18 and 21 after infection. After 8 weeks, mice were sacrificed and parasite load was measured by limiting dilution assay. For the group that received recombinant MRPs (rMPR) as treatment, mice were infected as previously described and then treated 3 times per week with 10 µg of the mix MRP8/14 in 50 µl of PBS, during the last four weeks of infection directly in the infected footpad. Thickness of the lesion was measured every week until the end of the infection and parasite load was measured as described below.
Limiting dilution assay was done as previously described [18], [19] with some modifications. Briefly, after 10–12 weeks of infection mice were sacrificed. The infected footpads were disinfected and inflamed area of the pad was excised, homogenized; extracted parasites were serially diluted in a 96 well plate in duplicate. After 8 days, the number of viable parasites was determined from the highest dilution using an inverted microscopy.
Statistically significant differences were analyzed by ANOVA followed by Tukey test using the Graphpad Prism program (version 5.0). For limiting dilution and TNF, non-parametric Mann-Whitney or Kruskal-Wallis test was used. Values of P≤0.05 were considered statistically significant. All data are presented as mean ± SEM.
We have previously described that MRPs induce NO in murine macrophages [9]. Confirming and extending these data, we observed that increasing concentrations of MRPs 8/14 (5, 10 and 25 µg/ml) lead to NO synthesis by macrophages in a dose-dependent manner (Figures 1A and 1B). Subsequent infection of MRP-primed macrophages with L. major did not affect NO production (Figure 1A). However, when cells were first infected and then stimulated with MRPs, NO production was reduced by around 35% (Figure 1B). To evaluate whether the effect of Leishmania infection was affecting iNOS protein levels, we performed western blotting. As expected, stimulation of macrophages with MRPs, led to an increase of iNOS expression (Figure 1C), and in MRP-primed macrophages followed by Leishmania infection, we observed increased of iNOS expression, which was maximal when 25 µg/ml of MRPs were added. At the same concentration of MRPs, the expression of iNOS was reduced when the cells were infected prior to stimulation (Figure 1C). These results revealed that Leishmania infection alters the capacity of MRPs to induce NO production by reducing iNOS expression.
Tumor necrosis factor α (TNF-α) is a multifunctional cytokine produced primarily by monocytes and macrophages. It has been shown that this cytokine is essential for the control of Leishmania at early stages of infection [20]. Therefore, we were interested in investigating whether MRPs were able to induce TNF-α production in macrophages. We performed a time and dose-dependent experiment, stimulating the cells for 1, 3, 6 and 24 hr with 5, 10 and 25 µg/ml of MRPs using TNF-sensitive L929 fibroblasts [21]. As shown in Figure 2A, the maximum peak of TNF-α production by MRP-stimulated macrophages occurred between 1 and 3 hr with 25 µg/ml of MRPs, decreasing thereafter. The time of 1 hr was chosen to evaluate the profile of TNF-α production during Leishmania infection. First, cells were primed for 1 hr with MRPs and then infected with L. major. Second, macrophages were infected with L. major overnight, followed by washes and stimulation with MRPs. As shown in Figure 2B, and similar to our NO data, MRP-primed macrophages subjected to L. major infection did not show altered capacity to produce TNF-α, however; the ability of L. major-infected cells to produce TNF-α in response to MRP stimulation was clearly reduced.
MRP-priming conferred protection against Leishmania infection, as revealed by iNOS expression, NO and TNF-α production. Thus, we next evaluated whether this MRP-inducible microbicidal response correlated with an enhanced intracellular killing of the parasite [22]. To this end, macrophages were primed with various concentrations of MRPs prior to infection with a L. major strain expressing luciferase, then cells were collected at 6 and 24 hr post-infection. As shown in Figure 3A, at 6 hr post-infection, we observed a higher percentage of infection in the cells that were primed with 10 and 25 µg/ml of MRPs, compared to those that were not primed. However, after 24 hr of infection (Figure 3B) primed macrophages reduced the parasite load by 42% in a dose-dependent manner. Altogether these results could suggest that MRPs provide the cells with the ability to phagocytise and kill the parasites more efficiently that unprimed cells.
As we observed that MRP stimulation increased the expression of iNOS, we further analyzed the signaling pathways involved in iNOS/NO production. We have previously reported that MRP-induced macrophage activation involves the participation of the ERK and JNK MAPKs [9]. Thus, it was critical to determine whether Leishmania could influence phosphorylation of these kinases in order to explain the incapacity of infected cells to respond to MRPs, knowing that Leishmania infection can interfere with signaling under the regulation of these kinases by activating host phosphatases [18], [23]. As expected, phosphorylation of both ERK and JNK (Figure 4) was observed in naive macrophages stimulated with MRPs. Phosphorylation of ERK and JNK was not altered in MRP-primed macrophages infected with L. major (Figures 4A and 4C). On the other hand, MRP-inducible ERK and JNK phosphorylation was strongly inhibited in Leishmania-infected cells (Figures 4B and 4D).
To further characterize the activation of macrophage signaling after MRP stimulation, we investigated the nuclear translocation of transcription factors (TFs) involved in iNOS/NO production (e.g., NF-κB, STAT 1 and AP-1) by performing EMSA. A strong nuclear translocation of NF-κB (Figures 5A and 5B) and AP-1 (Figures 5C and 5D) occurred in response to MRPs stimulation. As shown in Figures 5A and 5C, MRP-primed macrophages showed translocation of NF-κB and AP-1 (Figure 5A). Nonetheless, NF-κB and AP-1 MRP-induced translocation was inhibited when macrophages were first infected with Leishmania (Figures 5B and 5D, respectively). In addition, it was possible to detect the p35 fragment (Figure 5B) that is a product of NF-κB degradation by Leishmania infection [24]. We also monitored the nuclear translocation of STAT; however, we did not observe any alteration of this TF in response to MRPs in either case (data not shown).
Whereas MRPs modulate the microbicidal functions of macrophages in vitro, their role in vivo is still unknown. Therefore, using an air-pouch model we attempted to monitor this innate inflammatory event. Previous reports from our laboratory using this model have demonstrated that inoculation of Leishmania promastigotes led to the recruitment of inflammatory leukocytes at sites of injection within hours and this was accompanied by the secretion of various chemokines [25]. In addition, Tessier and collaborators have previously described that injection of LPS into the air-pouch induced neutrophil accumulation and the subsequent secretion of MRPs, reaching a maximum peak at 6 hr post-stimulation [2]. In this set of experiments, BALB/c mice were infected in the air-pouch with 10×106 parasites for 6 hr. Afterwards, we evaluated the number of cells recruited and the secretion of MRPs. As shown in Figure 6A, Leishmania infection induced leukocyte recruitment comparable to LPS, neutrophils being around 80% of the total recruited leukocytes (Figure S1). In addition, Leishmania infection induced MRP 8/14 secretion by the recruited cells within the pouches (Figure 6B). To further monitor the implication of MRP secretion in the Leishmania-induced inflammatory cell recruitment, we neutralized MRPs using anti-MRP antibodies prior to infection with L. major. As shown in Figures 6A and 6B the use of these antibodies led to a significant reduction in cell recruitment, concomitantly with an almost complete abrogation of MRP secretion. It is important to point out that although we could still observe that the majority of the cells recruited in the mice injected with neutralizing antibodies were neutrophils (Figure S1), the total amount of recruited cells was significantly lower in these mice, compared to mice that received PBS, LPS or L. major (Figure 6A).
In the murine model, cutaneous Leishmaniasis is caused by injection of L. major or L. mexicana directly in the footpad. This model has been widely used to measure progression of infection in resistant and susceptible mice under different circumstances and for further isolation of parasites [26], [27]. To additionally investigate the role of MRPs during Leishmania infection, we monitored to which extend the neutralization of MRPs or the inoculation of recombinant MRPs would influence the progression of the infection in vivo. In a first set of experiments, we infected BALB/c mice and performed tri-weekly inoculation of MRP neutralizing antibodies for a period of 4 weeks. The progression of footpad thickening and development of lesion were followed over 8-weeks period. As shown in Figure 7A, mice that received anti-MRPs antibodies developed a significantly bigger footpad swelling during the first 8 weeks of infection comparatively to the untreated group. Significant differences were also detected between treated and control groups regarding the footpad parasitic load (Figure 7A, bar graph). These data suggest that MRPs secreted in the infectious environment could play an important role in the immunological events controlling Leishmaniasis development during the initial weeks of the infection.
To confirm the contribution of MRPs to the regulation of Leishmania infection, we tested whether recombinant MRP 8/14 (rMRP8/14) injected in infected footpads could lead to reduce Leishmania-related pathologies in mice. As reported in Figure 7B, BALB/c mice which started to receive inoculation of rMRPs at 8 week post-infection over a 4-weeks period, showed a clear and significant reduction of their footpad swelling and parasitic load comparatively to the control group (Figure 7B, bar graph).
To further characterize the role of MRPs in the control of Leishmaniasis we used mice deficient for MRP14 that also fail to express MRP8 in peripheral and tissue leukocytes [28]. L. major infection caused a significantly greater pathology in MRP14 KO mice compared with its genetic background control (Figure 8A), as well as higher parasite load in the footpad after 19 weeks of infection (Figure 8B). This experiment was carried out for a longer period of time compared with the two previous experiments (using anti-MRPs or rMRPs) in order to observe the control of infection, as we used C57Bl/6 background mice. Although the parasite burden calculated by the limiting dilution assay was significantly lower in the C57BL/6 mice compared to the parasite burden from the BALB/c mice, we still observe that the absence of MRPs led to a higher parasite burden, even in resistant mice. This last set of experiments strongly suggests that MRPs play a significant role in the immunological mechanisms involved in the regulation of Leishmania infection. Moreover, these data unveil MRPs as potential therapeutic agents to treat Leishmaniasis.
MRP 8 and 14 also known as S100A8 and S100A9 belong to the S100 protein family, a large group of intracellular proteins associated with many cellular functions including contraction, motility, cell differentiation, calcium regulation among others [3]. In addition, the S100 proteins are also associated with different inflammatory diseases [29]–[32]. Recently, we and others have reported that MRP 8 and 14 can modulate macrophage functions including NO production [9]. Given that MRPs activate the macrophage signaling machinery and knowing that Leishmania parasites exert the opposite effect, we were interested in elucidating the role of MRP 8 and 14 during Leishmania infection both in vitro and in vivo.
Our results clearly showed that MRP-primed Leishmania-infected murine macrophages were able to produce NO with the concomitant expression of iNOS. These events correlated with a more efficient killing of the parasites as demonstrated by the luciferase assay. NO plays a key role in the macrophage microbicidal functions and is essential for the control of Leishmania infection [22]. In addition, we also found that MRP-primed macrophages produced high levels of TNF-α and were able to phosphorylate ERK and JNK kinases. More importantly, we observed that this priming resulted in an increased nuclear translocation of NF-κB and AP-1. This finding correlates with the fact that iNOS contains promoter binding sequences for these two transcription factors along with STAT1α [33].
The induction of MAPK phosphorylation and TFs nuclear translocation was observed very shortly after stimulation; the fact that MRPs are able to induce the NF-κB and the AP-1 pathways suggests that these TFs might act in synergy to enhance the expression of iNOS, resulting in high levels of NO and more efficient Leishmania killing. We have also reported that MRPs are recognized by Toll like receptor 4 (TLR4) [9]. This is in line with the observation that NF-κB is strongly induced by MRPs, on the other hand, efficient induction of AP-1 might be due to the fact that ERK and JNK are up-stream activators of c-Jun and c-Fos which dimerize to form active AP-1 complexes [34].
Additionally, we observed that macrophages that were first infected and then stimulated with MRPs, did not have the capacity to respond in the same way as primed macrophages, since the levels of NO and TNF production as well as the phosphorylation of JNK and ERK and the nuclear translocation of TFs were substantially reduced. This suggests that the parasite is able to abrogate the activation of the macrophage signaling machinery induced by MRPs in order to survive inside the host. One of the main mechanisms adopted by the parasite to subvert the immune response is the rapid activation of host phosphatases [18], [23], [35]. This fact might explain the poor MAPK phosphorylation and TFs nuclear translocation observed in macrophages first infected and then stimulated with MRPs.
Studies made by our group have shown that mouse infection with Leishmania parasites in the air-pouch model leads to neutrophil recruitment [25], Here, we demonstrated that MRPs controlled neutrophil recruitment induced by Leishmania or LPS. However, the exact role of neutrophils during cutaneous Leishmaniasis is still controversial. For instance Lima et al. [36] showed that there is a massive infiltration of neutrophils soon after skin infection with L. major, they investigated in more detail the role of neutrophils in resistant C57BL/6 and susceptible BALB/c mice by depleting neutrophils with specific antibodies. They showed that neutrophil depletion in both susceptible and resistant mice accelerated parasite spreading and caused more severe footpad swelling. These data suggested that neutrophils are of crucial importance in early control of parasite infection. In contrast, a study made by Laskay et al. [37] showed that Leishmania uses neutrophils as an evasion strategy, since the parasite survives inside these cells and use them as “Trojan horses” to get access into the macrophages where it will survive and multiply. Later, the same group showed that Leishmania-infected neutrophils also are uptake by dendritic cells inhibiting early immune response against Leishmania in the tissue [38]. Some other reports have shown that depletion of neutrophils in BALB/c mice inhibited the IL-4 response and promoted partial resistance [39]. Using B10R macrophages we also observed significantly increase of parasite infection when cells were treated with MRPs, however, it did not reflect in the survival of the Leishmania within macrophages (Figure 3).
More recently, Peters et al. [40], [41] showed that depletion of neutrophils reduces the ability of the parasite to establish productive infections. Furthermore, they reported that the neutrophils are the initial host cell for a substantial fraction of parasites and that there is more control of the infection when the neutrophils are not present. Interactions between cellular populations have been pointed out as important to either the control or development of the disease, and one of the most important cell interactions is the one between neutrophils and macrophages. It has been demonstrated by some groups that interactions of apoptotic or necrotic neutrophils with macrophages may interfere with the outcome of the infection. Interaction of dead neutrophils with L. major-infected peritoneal macrophages isolated from BALB/c mice, led to an increase in parasite growth, a mechanism mediated by the TFG-β and PGE2 produced by macrophages; however, macrophages isolated from resistant C57BL/6 mice and co-cultured with the same dead neutrophils presented a good microbicidal activity, mediated by TNF-α, therefore controlling the infection [42]. Concurring with this, Afonso et al. demonstrated that phagocytosis of apoptotic neutrophils by L. amazonensis-infected macrophages led to an increase on the parasite burden; however, phagocytosis of necrotic neutrophils resulted in parasite killing in a NO-independent manner, but dependent on ROIs [43]. Later on, it was observed that elastase produced by neutrophils plays a key role in the control of the infection, since this molecule activates the microbicidal mechanisms of the L. major-infected macrophages in a TLR-4-dependent manner [44].
It has also been studied that, depending on the infecting Leishmania species, or even the specific strain, the interaction between neutrophils and macrophages can lead to resistance or susceptibility to the infection. For instance, Novais F. et al, demonstrated that L. braziliensis elimination depends on the interaction between neutrophils and macrophages, in a TNF-dependent mechanism [45]. The same effect was observed in the control of L. amazonensis, where it was shown that TNF-α, elastase and platelet activation factor, produced by neutrophils, were responsible for parasite killing. On the other hand, this study also showed that NO and ROIs were not involved in the clearance of the parasite, as has been observed with other Leishmania species [46]; we consider that, in both cases, it is also possible that neutrophils secrete MRPs, and this secretion helps to control the infection. Contrary to what was shown by Ribeiro et at., it has been demonstrated that L. major-infected macrophages isolated from C57BL/6 resistant mice induce apoptosis of neutrophils, therefore favoring the propagation and survival of the parasite [47].
Our data are partially in agreement with some of the previous observation, since we clearly demonstrated that there is neutrophil recruitment at the site of infection in the air-pouch model; moreover, we showed that these neutrophils are able to secrete MRPs and that depletion of these MRPs significantly reduced the amount of recruited neutrophils and consequently MRP secretion. In addition, infection with L. major in the footpad of susceptible BALB/c mice and depletion of MRPs resulted in an increased parasite load and footpad swelling. These results strongly suggest that MRPs are important to control Leishmania infection. Strengthening this fact, when we treated mice with rMRPs directly in the footpad, we observed a significant reduction in size of lesion and parasite load. A potential mechanism underlying these events could be that injection of MRPs leads to an enhanced neutrophil recruitment, which in turn, can secrete more MRPs creating a positive feedback loop of constant secretion of MRPs, where it is possible that neutrophils are actually containing the progression of the infection, concomitant with the fact that these proteins present by themselves antimicrobial properties [6]. Additionally, we do not rule out the possibility that direct injection of MRPs also induced monocyte recruitment and as observed in our in vitro results which showed that MRP-primed macrophages are able to produce high levels of NO, being this responsible for the killing and more efficient control of the infection. However, whether the control of the infection in vivo is NO-mediated needs further investigation.
In summary our data showed for the first time that MRP 8 and 14 play an important role in the control of Leishmania infection in vivo and in vitro and support the idea that they could have a potential role as therapeutic drugs.
|
10.1371/journal.pgen.1000005 | Identification of Small Molecule Inhibitors of Pseudomonas aeruginosa Exoenzyme S Using a Yeast Phenotypic Screen | Pseudomonas aeruginosa is an opportunistic human pathogen that is a key factor in the mortality of cystic fibrosis patients, and infection represents an increased threat for human health worldwide. Because resistance of Pseudomonas aeruginosa to antibiotics is increasing, new inhibitors of pharmacologically validated targets of this bacterium are needed. Here we demonstrate that a cell-based yeast phenotypic assay, combined with a large-scale inhibitor screen, identified small molecule inhibitors that can suppress the toxicity caused by heterologous expression of selected Pseudomonas aeruginosa ORFs. We identified the first small molecule inhibitor of Exoenzyme S (ExoS), a toxin involved in Type III secretion. We show that this inhibitor, exosin, modulates ExoS ADP-ribosyltransferase activity in vitro, suggesting the inhibition is direct. Moreover, exosin and two of its analogues display a significant protective effect against Pseudomonas infection in vivo. Furthermore, because the assay was performed in yeast, we were able to demonstrate that several yeast homologues of the known human ExoS targets are likely ADP-ribosylated by the toxin. For example, using an in vitro enzymatic assay, we demonstrate that yeast Ras2p is directly modified by ExoS. Lastly, by surveying a collection of yeast deletion mutants, we identified Bmh1p, a yeast homologue of the human FAS, as an ExoS cofactor, revealing that portions of the bacterial toxin mode of action are conserved from yeast to human. Taken together, our integrated cell-based, chemical-genetic approach demonstrates that such screens can augment traditional drug screening approaches and facilitate the discovery of new compounds against a broad range of human pathogens.
| Microbial resistance to antibiotics is a serious and growing threat to human health. Here, we used a novel approach that combines chemical and genetic perturbation of bakers yeast to find new targets that might be effective in controlling infections caused by the opportunistic human pathogen Pseudomonas aeruginosa. P. aeruginosa is the primary cause of mortality with cystic fibrosis patients and has demonstrated an alarming ability to resist antibiotics. In this study, we identified the first small molecule inhibitors of ExoS, a toxin playing a pivotal role during P. aeruginosa infections. One of these compounds, exosin, likely works by modulating the toxin's enzymatic activity. We further show that this inhibitor protects mammalian cells against P. aeruginosa infection. Finally, we used yeast functional genomics tools to identify several yeast homologues of the known human ExoS targets as possible targets for the toxin. Together, these observations validate our yeast-based approach for uncovering novel antibiotics. These compounds can be used as starting point for new therapeutic treatments, and a similar strategy could be applied to a broad range of human pathogens like viruses or parasites.
| Microbial resistance flourishes in hospitals and community settings, and represents a major threat to human health worldwide [1],[2]. Despite the threat, drug discovery methods have failed to deliver new effective antibiotics [3]. This problem is likely to worsen because major pharmaceutical and biotech companies are withdrawing from antibacterial drug discovery [4]. To address the challenge of developing new antibiotics and managing microbial resistance, alternative strategies are needed to define and inhibit pharmacologically validated targets [5]. Several lines of evidence support the hypothesis that bakers yeast Saccharomyces cerevisiae can contribute during early stages of antimicrobial development. Because many essential molecular mechanisms of cells are conserved, we hypothesized that bacterial virulence proteins could act similarly in both yeast and human cells. Indeed, the study of virulence proteins in S. cerevisiae has proved an effective alternative and proxy for a human model of bacterial infection [6],[7],[8]. In addition, S. cerevisiae is well-suited for screening small molecule inhibitors to inhibit overexpressed proteins [9],[10], and to discover molecules that disrupt protein-protein interactions [11]. Finally, the arsenal of available yeast functional genomics tools provides a powerful means to study the targets and pathways modulated by drugs (reviewed in [12]). Together, these observations support the idea that compound screening in S. cerevisiae is a powerful tool to isolate small molecule inhibitors against potential drug targets of human pathogens.
In antibacterial drug discovery, a particular concern is the emergence of multidrug resistant strains that require several drugs for efficient disease management. This problem is exacerbated in immunocompromised patients [13]. For example, P. aeruginosa affects immunocompromised individuals afflicted with cystic fibrosis and is the primary Gram-negative causative agent of nosocomial infections [14]. P. aeruginosa is resistant to the three major classes of antibiotics, namely β-lactams, aminoglycosides and fluoroquinolones [15]. Notably, P. aeruginosa strains have demonstrated an alarming ability to resist antibiotics, underscoring the need to discover novel molecules with new mechanisms of action [16],[17]. Ironically, there are few innovative antibacterial molecules available or under development and the majority of these target Gram-positive bacteria [18]. Therefore, research on the opportunistic Gram-negative bacterium P. aeruginosa is medically relevant and is a logical choice to explore the utility of the yeast-based approach to discover new small-molecule inhibitors.
A key feature of a number of Gram-negative bacterial infection is the Type III Secretion System (T3SS) [19]. P. aeruginosa manipulate host cells by injecting four effector proteins, exoenzyme S (ExoS), exoenzyme T (ExoT), exoenzyme Y (ExoY) and exoenzyme U (ExoU), through the T3SS. ExoS and ExoT are bifunctional enzymes containing an amino-terminal GTPase-activating protein domain and a carboxy-terminal ADP-ribosylation domain. They inhibit phagocytosis by disrupting actin cytoskeletal rearrangement, focal adhesions and signal transduction cascades [20]. ExoY is an adenylate cyclase that elevates intracellular levels of cyclic AMP and causes actin cytoskeleton reorganization [21]. ExoU is a phospholipase whose expression correlates with acute cytotoxicity in mammalian cells [6],[22]. Therefore, targeting P. aeruginosa virulence factors with small molecule inhibitors would be expected to modulate the pathogen's virulence and provide a starting point for antimicrobial drug development [23].
The pivotal role played by ExoS during P. aeruginosa infection validates this toxin as a promising target to discover small molecules that may interfere with P. aeruginosa pathogenicity and infectivity [20]. ExoS was initially described as a secreted ADP-ribosyltransferase (ADPRT) [24]. The toxin is a well-characterized bi-glutamic acid transferase that requires interaction with a 14-3-3 protein (FAS) for its activity [25],[26]. Vimentin was characterized as the first direct target of ExoS ADPRT activity [27]. Shortly after, Ras and related proteins including Rab3, Rab4, Ral, Rap1A and Rap2 were identified as been modified by ExoS [28],[29]. ADP-ribosylation of Ras at arginine 41 blocks the interaction of Ras with its guanine nucleotide exchange factor (GEF), resulting in the inactivation of the Ras signal transduction pathway in the infected host cell [30]. Recently, ExoS was shown to ADP-ribosylate diverse molecules including cyclophilin A and Ezrin/Radixin/Moesin (ERM) proteins [31],[32]. During infection of HeLa and fibroblast cells, P. aeruginosa translocates ExoS and induces cell death by apoptosis [33],[34].
In this study, we used a novel combined phenotypic and chemical genomics screen in yeast to identify the first small molecule inhibitors of P. aeruginosa ExoS. The compound, exosin, modulates the toxin ADP-ribosyltransferase enzymatic activity in vitro suggesting the inhibition is direct. Furthermore, we observed a protective effect with this compound against P. aeruginosa in a well-established mammalian cell infection assay. Interestingly, although we designed the yeast phenotypic screen to assay heterologous proteins, we also observed that yeast Ras2p is directly modified by ExoS and we biochemicaly characterized this modification. This result reveals that bacterial toxins can target similar proteins in both human and yeast and validates our yeast-based approach for the study of toxin function and for the high-throughput screening for small molecule inhibitors. These initial lead compounds can be used as a starting point for new therapeutic treatments or can help to characterize the cellular functions of bacterial proteins. A similar strategy could also be applied to facilitate the discovery of new compounds against a broad range of human pathogens.
We developed a yeast-based strategy where S. cerevisiae was initially used to identify P. aeruginosa PAO1 virulence factors or essential ORFs that inhibit yeast growth (Figure 1A). These particular genes were selected because they provide an attractive starting point to develop antibacterial drugs. Accordingly, we developed a list of 505 potential drug targets of P. aeruginosa (Table S1) [35],[36],[37]. These bacterial ORFs were individually transferred into the yeast expression vector, pYES-DEST52 where the GAL1 promoter controlled their expression. Transformed yeast growing on 2% glucose served as control (i.e., wild type growth) because in these conditions, the expression of the exogenous bacterial genes is repressed. Expression of these genes was induced by growing the yeast on selective solid medium containing 2% galactose + 2% raffinose (Figure 1B). The experiment was repeated (four times), and involved inoculating yeast cultures at different dilutions and spotting variable volumes of culture on agar plates in an attempt to increase the consistency of this test.
Of the 505 P. aeruginosa ORFs screened, nine strongly or partially impaired the yeast growth when overexpressed (Figure 1C). Five of these are essential genes, including; 1) the ribonuclease III – PA0770, 2) two probable transcription regulators - PA0906 and PA1520, 3) the transcription termination factor Rho – PA5239 and 4) a hypothetical protein – PA2702. In addition, four virulence genes were also detrimental to yeast growth, ExoA – PA1148, ExoS – PA3841, ExoT – PA0044, ExoY – PA2191. Interestingly, each of these four toxins are secreted or translocated by the type II (ExoA) or type III secretion system (ExoS, ExoT, ExoY) and each act within the infected host cell. By comparing the phenotype of yeast harboring the empty vector, we could assess the relative strength of the Pseudomonas gene overexpression effect and classify them into three groups (Figure 1C – right panel). Firstly, ExoA, ExoY, PA1520 and PA2702 strongly inhibited S. cerevisiae growth. Secondly, ExoS, ExoT, PA0906 and transcription termination factor Rho showed an intermediate growth impairment whereas expression of ribonuclease III weakly affected yeast fitness.
To assess the influence of ExoA, ExoY and ExoS enzymatic activities on yeast growth, catalytic mutants were assayed. Residues important for the enzymatic activity of ExoA (E553A), ExoY (K81M) and ExoS (R146W and E379A+E381A) were previously reported [21],[38],[39],[40] and served to guide our mutant construction (Figure 2A).
Compared to the empty vector control, overexpression of active ExoA-wt and ExoY-wt induced a severe growth defect (Figure 2B – top and middle panels) whereas expression of the enzymatically inactive ExoA-ADPRT mutant and ExoY-AC mutant did not. This observation suggests that ExoA and ExoY toxicity is conferred by their ADPRT and AC activities, respectively. Moreover, whereas ExoS-wt expression reduced yeast growth, this dominant negative effect was totally abolished when expression of the ExoS-GAP and ExoS-ADPRT mutants were simultaneously induced, indicating one or both ExoS enzymatic activities are causative for the yeast growth defect (Figure 2B – bottom panel). Because normal growth was observed only for the ExoS ADPRT domain mutant and not for the GAP mutant, this suggests that ExoS ADPRT enzymatic activity is responsible for the yeast toxicity consistent with previous observations [41]. Taken together, these observations attribute the yeast growth inhibition to the ExoA-ADPRT, ExoY-AC and ExoS-ADPRT activities and validate the three toxins as appropriate drug target candidates for further study. Due to its critical role in the initial steps of chronic infections of immuno-compromised patients and in the pathogenesis of acute P. aeruginosa infections, ExoS was selected for interrogation using our yeast-cell based inhibitor screen.
To demonstrate that yeast can serve as a model system to mimic human cells during infection, we asked if these bacterial toxins modulate the biological activity of conserved eukaryotic targets. Following binding of P. aeruginosa to human cells, the bacteria inject ExoS directly into the cytoplasm where it inhibits the activity of several targets by ADP-ribosylation. Therefore, overexpressing yeast homologues of ExoS human targets should restore yeast growth by titrating the toxin's enzymatic activity (Figure 3A). To test our hypothesis, forty-six yeast members of the Ras superfamily and cyclophilins were individually overexpressed in yeast in the presence of ExoS (Table S2).
We first verified that individually, the overexpressed yeast proteins did not impair yeast growth. To accomplish this, cells were cultivated on galactose + raffinose in absence of copper such that only the yeast over-expressed candidates, but not ExoS, were expressed (Figure 3B – top right panel). Yeast genes whose overexpression was toxic were eliminated from our analysis. In parallel, cells were grown on galactose + raffinose in presence of copper to assess the rescuing effect of yeast gene overexpression in the presence of the toxic ExoS (Figure 3B – bottom right panel). Comparing yeast growth to the cell harboring the empty vector and yeast expressing ExoS alone (Figure 3B – bottom left panel), ten yeast genes were found to rescue ExoS toxicity (Table S3). Subsequently, only yeast genes demonstrating a strong growth rescue phenotype (such as RAS2) were analyzed further whereas genes showing weak rescue (such as YPT1) were not studied further (Figure 3B). S. cerevisiae possesses two homologues of the human Ras protein (Ras1p and Ras2p). Interestingly, Ras2p was found among these ten ORFs, i.e. overexpression of Ras2p but not Ras1p rescued ExoS-induced toxicity (Figure 3B).
As previously described, ExoS requires Factor Activating Exoenzyme S (FAS) for its ADPRT activity [26]. FAS is a member of the 14-3-3 protein family which has two yeast homologues, the Brain Modulosignalin Homolog (Bmh) 1 and 2. Accordingly, ExoS toxicity was assessed in the absence of Bmh1p or Bmh2p. As detected by the increase in yeast growth, ExoS-induced toxicity was diminished in cells lacking Bmh1p but not in those lacking Bmh2p (Figure 3C – left panel). In a bmh1Δ yeast background, the toxic effect of ExoS was again restored when introducing BMH1 in the presence of the toxin (Figure 3C – right panel). Together, these data imply that Bmh1p acts as ExoS cofactor in yeast.
To better understand the mechanism of this toxicity, we demonstrated that yeast Ras2p was a direct target of ExoS and that Bmh1p was the ExoS cofactor in yeast, using a biochemical assay. To that end, an ADP-ribosyltransferase enzymatic assay was performed using the radioactive substrate [32P]-NAD+, purified P. aeruginosa ExoS, yeast Ras2p and Bmh1p. Autoradiographic analysis showed that radioactive ADP-ribose was incorporated by Ras2p (Figure 3D). Moreover, in absence of Bmh1p, no ADP-ribosylation was observed. These data reveal that in vitro, Ras2p is directly ADP-ribosylated by ExoS with Bmh1p as a cofactor.
Taken together, these results allow us to conclude the following; (i) in yeast, the growth inhibitory effect observed in the presence of the P. aeruginosa ExoS is mediated by its ADPRT activity, (ii) this growth inhibition is due, at least in part, to the inactivation of the yeast protein Ras2p by ADP-ribosylation, (iii) ExoS ADPRT activity is activated by the yeast cofactor Bmh1p. Most significantly, conservation of several toxin targets from yeast to human, such as Ras2p, Rsr1p, Ypt52p and Cpr6p, suggests that P. aeruginosa ExoS acts in a related manner in both organisms.
The sensitivity and specificity of our yeast-based assay allowed us to use S. cerevisiae to detect potential inhibitors of the three selected P. aeruginosa drug targets. Because ExoS-wt inhibited yeast growth when overexpressed, we reasoned that any molecule that inhibits this enzyme should restore yeast growth (Figure 1A). Because we were unable to find any inhibitors when the bacterial toxin was expressed using the strong promoter GAL1, we exchanged the GAL1 promoter with the copper inducible promoter CUP1 which allows a titrable expression of the toxins. Expression from this promoter decreases the toxin level in yeast and renders the conditions of the yeast screen less stringent. Over 56,000 compounds, primarily synthetic small molecules, were tested against ExoS. Effect of the compounds was compared to the yeast growth in absence of compound (as control for inhibition) and to the cells dividing in absence of toxin (as a control for growth). With this strategy, we uncovered six potential inhibitors, Diosmin, Everninic acid, Flavokawain B, 0469-0796, 4296-1011 and E216-5303 based on their ability to restore yeast growth (Figure 4A).
To determine if the observed yeast growth recovery was due to a direct modulation of the compounds on ExoS ADPRT activity, an in vitro fluorescent ADPRT enzymatic assay was performed. Diosmin, 4296-1011, Everninic acid and E216-5303 modulated ExoS ADPRT activity and their IC50 values were determined as 3, 6, 21 and 23 µM respectively (Figure 4A). Due to their intrinsic fluorescence, Flavokawain B and 0469-0706 effects could not be tested in our enzymatic assay. Because, only exosin protected CHO cells from lysis during P. aeruginosa infection in cell culture (data not shown) it was therefore selected for additional studies. Exosin acts as competitive inhibitor against the NAD+ substrate of ExoS as the Vmax values were largely unaffected, whereas the KM values increased from 9 to 30 µM (Figure 4C). The Ki value (dissociation constant for a competitive inhibitor) was 33.0 ± 3.0 µM for exosin (Figure 3D), which agrees favourably with the IC50 value for this compound (Figure 4B). Thus, the drug-like compound exosin directly modulates ExoS ADPRT activity in vitro via competitive inhibition. Therefore, exosin seems to restore ExoS dependant yeast growth defect by directly inhibiting ExoS ADPRT activity.
To determine if the small molecule inhibitor, exosin, could modulate the viability of CHO cells during P. aeruginosa infection, apoptotic CHO cells and living CHO cells were distinguished using the exclusion dye 7-AAD. Here, CHO cells were exposed to P. aeruginosa with or without the small molecule inhibitor for 2 hours, and the fraction of apoptotic cells was measured by 7-AAD staining and flow cytometry.
The mean fluorescent intensities of 7–AAD were plotted as a histogram (Figure S1). When compared to the mean fluorescent intensity from the red peak (background fluorescent intensity - 7.42; n = 3) and from the blue peak (control for P. aeruginosa infection - 18.2; n = 3), the green (20 µM), orange (40 µM), and light blue (80 µM) peaks gave mean fluorescent intensities of 13.4, 11.3, and 10.3, respectively, indicating that exosin exerted its effect in a dose-dependent manner. Therefore, a higher inhibitor dose reduced the number of cells undergoing apoptosis, reflecting a better protective effect. Similar observations were made in dot plots (Figure 5A). In the presence of 80 µM exosin, a significant increase in the percentage of living cells (79.35%; n = 3) was observed with the serious reduction of dead cells (20.31%; n = 3), compared to the infected CHO cells without inhibitor, 0 µM (49.72% and 50.28%, respectively). However, the protective effect of exosin at a concentration of 80 µM was not observed when CHO cells were infected by the P. aeruginosa PA14 strain, a strain expressing ExoT, ExoY and the phospholipase exoenzyme U (ExoU) but not ExoS, indicating the specificity of the compound exosin against ExoS only (Figure 5A – lower panels).
In the CHO cell infection assay, the protective effect of exosin was monitored during an early stage of infection by detecting the number of dying and dead CHO cells using flow cytometry. Moreover, the effect of the inhibitor at the late stage of infection was assessed by the quantification of lactate dehydrogenase (LDH) released from the population of lysed CHO cells. Four hours after P. aeruginosa PAK infection (Figure 5C) revealed a 6.93% decrease in lysis upon addition of 20 µM exosin, a 13.92% lysis reduction in the presence of 40 µM final of inhibitor and a 12.90% reduction at 80 µM. However, the protective effect of exosin at a concentration of 80 µM was not observed when CHO cells were infected by the P. aeruginosa PA14 strain that translocates ExoU instead of ExoS (Figure 5C). Together, these data strongly support the conclusion that the inhibitor exosin is specific for ExoS and is able to reduce ExoS cytotoxicity against mammalian cells.
P. aeruginosa PAK viability was tested by measuring optical density of cultures in the presence of 20, 40 and 80 µM of inhibitor over a period of 10 hours. Addition of exosin did not affect Pseudomonas growth, further confirming the specificity of exosin for ExoS in CHO cells (Figure S2).
Given the specificity of exosin, we screened 50 structural analogues of this compound in yeast to find molecules with increased potency against ExoS ADPRT activity. Seven analogues with an improved effect were found (Figure 4E – exosin-5138, exosin-5316 and the compounds marked by an asterisk). According to the flow cytometry results, all of these compounds protected CHO cells when infected with P. aeruginosa in cell culture (data not shown). However, only exosin-5138 and exosin-5316 showed a protective effect when monitored with the LDH assay. Therefore, only these two compounds were used for further investigation. Exosin-5340 had no protective effect in yeast or in the CHO cell infection assay and served as a negative control. Importantly, results obtained from the yeast studies revealed the importance of the para position of the nitrobenzyl ring for the inhibitory activity of the compounds (Figure 4G). The three different analogues, exosin-5138, exosin-5316 and exosin-5340, were then selected for quantification of the yeast growth recovery and for IC50 determination. Exosin-5138 showed 36.8% recovery, almost double the protective effect of exosin whereas exosin-5340 conferred no protection (Figure 4F). The last analogue exosin-5316 (with 27.4% recovery) demonstrated a protective effect almost equal to the original compound (20.5% recovery). The fluorescent ADPRT enzymatic assay revealed that the three analogues directly modulate ExoS ADPRT activity in vitro (Figure 4F). The IC50 for each compound was calculated and these values paralleled the effect of the small molecules in yeast, most strongly for exosin-5138 and exosin-5340, and to a lesser extent for exosin-5316.
We extended our studies of these analogs in mammalian cells. For this purpose, protection provided by exosin-5138 and exosin-5316 was assessed in the CHO cell toxicity assay. Using the flow cytometry as described earlier, a strong protective effect of exosin-5138 was observed (Figure 5A – left panel). Dot plots of exosin-5138 showed a large reduction in dead cells at a compound concentration of 80 µM (Figure 5A – right panel). Exosin-5138 showed decreases of 25.20, 44.05 and 60.83% in the number of dead/dying CHO cells in the presence of 20, 40 and 80 µM of inhibitor, respectively (Figure 5B). In the LDH assay, exosin-5138 reduced cell lysis by 9.34, 18.61 and 24.64% in presence of the compound at 20, 40 and 80 µM inhibitor, respectively (Figure 5C) demonstrating an improved efficacy of exosin-5138 against ExoS cytotoxicity versus the original hit.
In addition, as shown by flow cytometry, exosin-5316 exerted a protective effect (Figure 5A). The number of dead/dying CHO cells was detectably lower upon addition of the analogue exosin-5316; however, this reduction was not statistically significant compared to the original hit (p>0.05). In contrast, the LDH assay revealed a 7.94, 12.84 and 15.81% reduction in cell lysis at 20, 40 and 80 µM final concentration respectively (p<0.05). The analogue exosin-5316 showed similar protective effect compared the original hit (p<0.05) (Figure 5C).
The data show a correlation between the protective effect of exosin and its analogues in yeast and for the results obtained in the CHO cell infection assay. Moreover, these observations establish yeast as a powerful assay system to estimate the effect of analogues of an original hit and to prioritize lead compounds before tedious subsequent experiments in a more complicated model of infection are undertaken.
In this report, we used the P. aeruginosa virulence factor, ExoS, to demonstrate the utility of the baker yeast S. cerevisiae as a tool to isolate inhibitors against human pathogens. We succeeded in identifying the first known inhibitor of ExoS, exosin, and demonstrated that this assay can be used to uncover structural analogs with improved potency. Therefore, yeast can substitute for traditional human models of infection, and be used to effectively prioritize compounds.
In our report, S. cerevisiae produced a binary readout that allowed us to test 505 P. aeruginosa genes for their inhibitory effect in yeast. Expression of nine bacterial genes, five essential and four virulence genes, reproducibly prevented S. cerevisiae growth. Among the isolated essential genes, the Rho termination factor from E. coli was demonstrated to induce yeast RNA polymerase II release at all pause sites of the mRNA in vitro [42]. Thus, transcription deregulation could explain the yeast growth arrest in the presence of the transcription termination factor Rho. Members of the ribonuclease III superfamily are RNA-specific endonucleases involved in RNA maturation, RNA degradation and gene silencing [43]. We hypothesize that the observed yeast growth defect induced by expression of the P. aeruginosa ribonuclease III was caused by deregulated RNA degradation. Since no clear biological function is associated with the two probable transcriptional regulators – PA0906 and PA1520, nor for the hypothetical protein – PA2702, we cannot speculate on the mechanism of action of these proteins in yeast.
During infection, P. aeruginosa manipulates host cellular function through the action of the toxins ExoA, ExoS, ExoT and ExoY using the type II and III secretion systems [44]. Since these toxins inactivate key molecules directly within the infected cells and because several basic molecular functions are conserved among eukaryotes, it seemed likely that the toxins could act similarly on targets conserved in both yeast and human. Therefore, inactivation of yeast homologues of the toxin human targets is an attractive and simple scenario to explain the growth inhibition effect conferred by the ExoA, ExoS, ExoT and ExoY expression in yeast. There are many possible reasons to account for the limited number of bacterial genes affecting yeast growth including; (i) a proper expression of the bacterial gene (e.g. high CG content in the Pseudomonas DNA sequence [45]), (ii) the presence of the target and/or appropriate co-factor and (iii) required post-translational modifications.
High conservation of basic molecular and cellular mechanisms between yeast and human cells highlights S. cerevisiae as an ideal model organism to study mammalian diseases and their underlying pathways [46],[47]. Indeed, several reports have shown that this conservation can be used to decipher bacterial toxins mode of action [6],[7],[48]. Here, the toxicity caused by ExoA, ExoY and ExoS overexpression in yeast is mediated by their enzymatic activity (Figure 2B). Interestingly, ExoS ADPRT activity alone is sufficient to induce yeast cell growth arrest. This last observation inspired us to study ExoS yeast toxicity into more detail.
Once translocated in human host cells via P. aeruginosa type III secretion system, ExoS inhibits several cellular targets. Its GTPase activating protein activity reorganizes the actin cytoskeleton through RhoA, Rac1 and Cdc42 inactivation. In contrast, ExoS ADPRT activity inactivates a wide range of proteins such as several members of the Ras family and related proteins, cyclophilin A and the Ezrin/Radixin/Moesin (ERM) proteins. Since ExoS inactivates all its protein targets, overexpression of the yeast homologues of the known human targets should restore yeast growth by titrating ExoS enzymatic activity. The ERM proteins play a role during cell polarity establishment of multi-cellular organisms and no homologues were found in yeast. Therefore, only yeast members on the Ras superfamily and cyclophilin family were induced in our overexpression study. Globally, no yeast homologue of the human RhoA, Rac1 or Cdc42 rescued ExoS-induced toxicity indicating that, together with the results obtained in the ExoS mutagenesis experiments, the ExoS GAP activity does not play a role in ExoS yeast growth inhibition. Four homologues of the human ExoS targets Ras, Rap1b, hRab and CyclophilinA were identified as the yeast Ras2p, Rsr1p, Ypt52p and Cpr6p, respectively. Arf1p, a member of the Sar/Arf family and homologue of the human Arf1 protein, rescued yeast expressing ExoS. Arf1p is involved in the retrograde transport of vesicles from the trans Golgi to the plasma membrane. Biological relevance of Arf1p as a direct target of ExoS is questionable since, during P. aeruginosa infection, ExoS trafficking occurs from the plasma membrane to the perinuclear region [49]. In our system, ExoS is encoded by a plasmid, thus deregulation of the retrograde transport could prevent ExoS reaching its plasma membrane targets (e.g. Ras2) and explain why Arf1p overexpression allowed yeast to divide in the presence of ExoS. No homologues of Cin4p (Sar/Arf family), Cns1p, Cpr7p, Cpr8p exist in human confounding the interpretation of these results.
ExoS overexpression in the absence of Bmh1p, the yeast homologue of the ExoS cofactor FAS, was not toxic to yeast (Figure 3C). In vitro enzymatic assay demonstrated that ExoS ADP-ribosylated Ras2p with Bmh1p as a cofactor (Figure 3B). Interestingly, substitution of the yeast Ras2p by the constitutively active Ras2-G19V mutation did not prevent ExoS toxicity (data not shown). Additionally, previous results showed that a deletion mutant of ExoS (lacking aa 51–72 membrane localization domain), which cannot ADP-ribosylate Rasp in vivo, is nevertheless as cytotoxic as wild type ExoS in CHO cells, indicating that Ras ADP-ribosylation is dispensable for ExoS virulence [50]. These observations suggest that, in yeast, the observed growth inhibition may be due to the cumulative inhibitory effects of ExoS on already known targets and/or due to the additional inactivation of an unknown key protein(s).
ExoS plays a pivotal role in the establishment of P. aeruginosa chronic infections and during acute P. aeruginosa pathogenesis. For that reason, this toxin was selected in our yeast phenotypic system to find inhibitors capable of modulating its enzymatic activity. Here, we report the isolation of exosin, the first inhibitor of ExoS ADPRT activity using a yeast cell-based screen. Six compounds were isolated from a library of 56,000 compounds and all rescued ExoS induced toxicity in yeast (Figure 4A). This relatively small number of hits is likely due to several reasons; both chemical and biological. The compound library, though selected for its diversity, nonetheless samples a limited range of chemical space. Furthermore, of 6,000 compounds that were randomly picked from the library and tested against ExoS and ExoY in two different yeast genetic backgrounds (the wild-type and the pdr1Δ+pdr3Δ strains), no difference in potency was observed, suggesting that the yeast genotype did not substantially influence the number of hits obtained in our screen.
In the enzymatic and the CHO cells infection assays, only exosin modulated ExoS biological activity. The apparent inactivity of the five other small molecules can likely be explained by two arguments. First, the five inhibitors could exert their effect on molecules or pathways modulated by ExoS without directly inhibiting ExoS ADPRT activity. Furthermore, because these five compounds exerted no protective effect in the CHO cell infection assay we predict they act on yeast specific pathways. A second explanation could be that the effect observed in yeast requires metabolism of the compound and is not an effect of the original compound itself.
Our results revealed that compounds derived from natural products were more bioactive on yeast (3 primary hits out of 580 natural products). However, the only hit conferring protection during infection of CHO cells by P. aeruginosa belongs to the class of the drug-like synthetic compounds (3 primary hits out of 53,000 compounds). Thus, there is certainly much more chemical space that can be probed using both natural and synthetic compounds.
Exosin was shown to directly interact with ExoS in vitro as a competitive inhibitor against the NAD+ substrate of ExoS ADPRT activity with comparable IC50 and Ki values indicating that exosin likely binds to the NAD+-pocket within the ADPRT domain of ExoS. This was substantiated by our observation of a similar inhibitory effect of exosin on the ADPRT activity of ExoA (IC50 = 17 µM; data not shown). Unfortunately, no high-resolution structure has been determined for the ADPRT domain of ExoS; however, ExoA is a well-characterized ADPRT enzyme for which there is a crystal structure of the ADPRT catalytic domain in complex with substrates and inhibitors [51],[52],[53]. By analogy with the recent X-ray co-crystal structure of ExoA with PJ34 [51], the benzylmorpholine ring of exosin might be expected to intercalate into the nicotinamide pocket within ExoS. In this scenario, the inhibitor amide should form an H-bond with enzyme. Presently, the site of contact within the ExoS active site for the alkyl group on the exosin nitrobenzyl moiety (shown by an arrow in Figure 4G) is not known; however, a single ring is required for in vivo activity and an ethyl is preferred over a methyl at the alkylation site. In summary, although we lack atomic resolution, it appears likely that the in vivo inhibitory activity of exosin against ExoS toxicity is due to a direct interaction of the inhibitor with the ADPRT domain of the toxin.
Using a yeast cell-based screen, the first known inhibitor of the P. aeruginosa ExoS, called exosin, was isolated and several analogues of the original hit were characterized. This work was facilitated by the partial conservation of the proteins inactivated by ExoS in both human and yeast. Thus, S. cerevisiae is a powerful tool to study bacterial toxins and to identify their corresponding inhibitors. Future studies could extend a similar approach to a broad range of human pathogens such as viruses and bacteria.
S. cerevisiae strain W303-1A harboring the plasmid pYES-DEST52 coding each of the 505 Pseudomonas aeruginosa PAO1 drug targets was grown overnight in SD–Ura to maintain selection of the plasmid. Yeast culture was directly submitted to 3 steps of a ten fold dilution using the liquid handling robot Q-Bot (Genetix). The non-diluted and diluted cultures were then immediately inoculated in duplicate on solid medium containing either glucose-Ura (repressing conditions) or galactose+raffinose-Ura (inducing conditions). Plates were incubated at 30°C and monitored for yeast growth defect after 2 and 3 days. Growth was compare to the fitness of yeast harboring the empty vector and to the yeast harboring the toxic pRS316-TUB2 vector [58].
The LOPAC library (1,280 compounds, Sigma-Aldrich), the SPECTRUM library (2,000 compounds, MicroSource Discovery Inc.) and a ChemDiv library (53,000 compounds) were screened at a final concentration of 50 µM. S. cerevisiae strain 14328-pdr1+pdr3 harboring the plasmid pDH105-exoA, pDH105-exoS or pDH105-exoY was grown overnight in SD -Leu to maintain selection of the plasmid and were diluted to a cell density of 5×103 cells/ml. Addition of 0.9 mM (ExoA and ExoY) or 1.5 mM (ExoA) of CuSO4 induced the expression of the toxin in yeast, the cultures were then aliquoted into wells of 96-well plates and compounds were added. Plates were incubated at 30°C and inspected for yeast growth recovery after 24 and 48 hours. As a control, cells containing the empty vector pDH105 were similarly grown, diluted and inoculated with copper and 0.5% DMSO. The effect of the hits on yeast growth recovery was quantified as a percentage of growth as described elsewhere [10].
Chinese hamster ovary (CHO) cells toxicity assay was performed as previously described with minor modifications [59]. CHO cells were routinely grown in F-12 medium supplemented with 10% fetal bovine serum (FBS) and 2 mM glutamine. Prior to infection, confluent CHO cells were washed and incubated with F-12 containing 1% FBS and 2 mM glutamine. P. aeruginosa was grown overnight in LB, subcultured into fresh LB, and grown to mid-log phase. 2.5×105 CHO cells per well were infected with P. aeruginosa at an initial multiplicity of infection (MOI) of 10 in duplicate. After 2 hours infection, CHO cells were harvested by trypsination and resuspended in Hank's Balanced Salt Solution + 1% bovine serum albumin. Induced cell death was measured by flow cytometry after 7-AminoActinomycin D (10 µg/ml) staining, using a Beckman-Coulter EPICS Elite flow cytometer. Culture supernatants from a second duplicate of CHO cells were collected after 4 hours of infection and centrifuged for 10 min at 3,220×g to sediment bacteria and CHO cells. Lactose dehydrogenase (LDH) in the supernatant was measured with a Roche LDH kit in accordance with the manufacturer's instructions. Percent LDH release was calculated relative to that of the uninfected control, which was set at 0% LDH release, and that of cells lysed in absence of inhibitor, which was set at 100% LDH release.
Recombinant yeast FAS, human Ras, yeast Ras2p, yeast Bmh1p and Bmh2p were cloned into pEGX (GE Healthcare), transformed in E. coli BL21 and purified according to manufacturer's instructions. Recombinant ExoS was purified by gel filtration and ion exchange chromatography as previously described [60].
To monitor incorporation of ADP-ribose into yeast Ras2, 50 nM of purified ExoS was added to a 20 µl reaction mixtures containing 1 µM of Bmh1, 20 µM of yeast Ras2, 2 mM of MgCl2 and 200 mM of sodium acetate (pH 6.0). The reaction was initiated by adding 2 Ci of ExoS radioactive substrate, nicotinamide adenine [adenylate-32P] dinucleotide (32P-NAD) (1000 Ci/mmol, GE Healthcare). Reaction mixes were incubated for 0 to 30 min at 30°C. The reaction was terminated with 2× Laemmli sample buffer, resolved by a 15% SDS-PAGE, and analyzed by autoradiography using a Typhoon Trio Workstation (GE Healthcare). The auto-ADP-ribosylation of ExoS served as a positive control in each reaction and reaction missing FAS served as a negative control.
The NAD+-dependent ADPRT assay of ExoS was performed at 30°C with ExoS at 50 nM in the presence of 1 µM of FAS and 20 µM human Ras in 100 mM NaCl, 2 mM MgCl2, 200 mM sodium acetate, pH 6.0 while the concentration of ε-NAD+ varied between 0 and 75 µM. The reaction was initiated with the addition of 50 nM (final conc.) ExoS in an Ultravette (Brand Scientific) 70 µL cuvette and the transferase reaction was monitored by recording the time-dependent change in fluorescence intensity with a PTI AlphaScan-2 fluorimeter (PTI Inc., New Jersey) with 305 nm and 405 nm excitation and emission, respectively. The data were analyzed by nonlinear curve fitting using the Michaelis-Menten equation (OriginLab v6.1; Northampton, MA) and also by linear regression analysis of both the Hanes-Woolf and the Lineweaver-Burk (LB) plots. The Ki values were determined for various inhibitor compounds (1.4% DMSO, final conc.) using Dixon plots, as well as from secondary plots of the slope of the LB plots versus inhibitor concentration. IC50 values, the concentration of the inhibitor that reduces the activity of the enzyme by 50%, were determined by non-linear regression curve fitting using Origin 6.1 [51]. |
10.1371/journal.pgen.1004655 | The Nuclear Immune Receptor RPS4 Is Required for RRS1SLH1-Dependent Constitutive Defense Activation in Arabidopsis thaliana | Plant nucleotide-binding leucine-rich repeat (NB-LRR) disease resistance (R) proteins recognize specific “avirulent” pathogen effectors and activate immune responses. NB-LRR proteins structurally and functionally resemble mammalian Nod-like receptors (NLRs). How NB-LRR and NLR proteins activate defense is poorly understood. The divergently transcribed Arabidopsis R genes, RPS4 (resistance to Pseudomonas syringae 4) and RRS1 (resistance to Ralstonia solanacearum 1), function together to confer recognition of Pseudomonas AvrRps4 and Ralstonia PopP2. RRS1 is the only known recessive NB-LRR R gene and encodes a WRKY DNA binding domain, prompting suggestions that it acts downstream of RPS4 for transcriptional activation of defense genes. We define here the early RRS1-dependent transcriptional changes upon delivery of PopP2 via Pseudomonas type III secretion. The Arabidopsis slh1 (sensitive to low humidity 1) mutant encodes an RRS1 allele (RRS1SLH1) with a single amino acid (leucine) insertion in the WRKY DNA-binding domain. Its poor growth due to constitutive defense activation is rescued at higher temperature. Transcription profiling data indicate that RRS1SLH1-mediated defense activation overlaps substantially with AvrRps4- and PopP2-regulated responses. To better understand the genetic basis of RPS4/RRS1-dependent immunity, we performed a genetic screen to identify suppressor of slh1 immunity (sushi) mutants. We show that many sushi mutants carry mutations in RPS4, suggesting that RPS4 acts downstream or in a complex with RRS1. Interestingly, several mutations were identified in a domain C-terminal to the RPS4 LRR domain. Using an Agrobacterium-mediated transient assay system, we demonstrate that the P-loop motif of RPS4 but not of RRS1SLH1 is required for RRS1SLH1 function. We also recapitulate the dominant suppression of RRS1SLH1 defense activation by wild type RRS1 and show this suppression requires an intact RRS1 P-loop. These analyses of RRS1SLH1 shed new light on mechanisms by which NB-LRR protein pairs activate defense signaling, or are held inactive in the absence of a pathogen effector.
| How plant NB-LRR resistance proteins and the related mammalian Nod-like receptors (NLRs) activate defense is poorly understood. Plant and animal immune receptors can function in pairs. Two Arabidopsis nuclear immune receptors, RPS4 and RRS1, confer recognition of the unrelated bacterial effectors, AvrRps4 and PopP2, and activate defense. Using delivery of PopP2 into Arabidopsis leaf cells via Pseudomonas type III secretion, we define early transcriptional changes upon RPS4/RRS1-dependent PopP2 recognition. We show an auto-active allele of RRS1, RRS1SLH1, triggers transcriptional reprogramming of defense genes that are also reprogrammed by AvrRps4 or PopP2 in an RPS4/RRS1-dependent manner. To discover genetic requirements for RRS1SLH1 auto-activation, we conducted a suppressor screen. Many suppressor of slh1 immunity (sushi) mutants that are impaired in RRS1SLH1-mediated auto-activation carry loss-of-function mutations in RPS4. This suggests that RPS4 functions as a signaling component together with or downstream of RRS1-activated immunity, in contrast to earlier hypotheses, significantly advancing our understanding of how immune receptors activate defense in plants.
| Plant innate immunity relies on two layers of pathogen detection. Cell surface-localized pattern recognition receptors detect pathogen-associated molecular patterns (PAMPs) of invading microorganisms and activate PAMP-triggered immunity (PTI) [1]. Successful pathogens must circumvent PTI to colonize plants, and many bacterial pathogens use type III secretion (T3S) to deliver effectors that suppress PTI into plant cells [1]. Effectors can be detected directly or indirectly by plant disease resistance (R) proteins, which then activate effector-triggered immunity (ETI) generally together with a hypersensitive response (HR) of the infected tissue [2]. Most intracellular R proteins are modular, with an amino-terminal coiled coil (CC) or Toll/interleukin-1 receptor/R protein (TIR) domain, a nucleotide binding (NB) domain and a leucine-rich repeat (LRR) domain [3]. Some NB-LRR proteins also carry an additional carboxyl-terminal extension, the function of which is unknown [3]. In addition, NB-LRR protein function generally requires an intact P-loop motif (GxxxxGKT/S) in the NB domain, presumably for ATP binding and energy-dependent conformational changes [3], [4]. Plant NB-LRR proteins and mammalian Nod-like receptors (NLRs) exhibit both structural and functional similarities [5].
Signaling following TIR-NB-LRR protein activation requires other key regulators such as Enhanced Disease Susceptibility 1 (EDS1), the EDS1-related proteins PAD4 and SAG101, and biosynthesis of the plant hormone salicylic acid (SA) for full immunity [6]. EDS1 was recently reported to interact with several NB-LRR proteins [7], [8]. Mis-regulation of R protein accumulation, localization or activation can cause constitutive defense responses, which are usually deleterious or lethal. For instance, the dwarf suppressor of npr1-1, constitutive 1 (snc1) mutant carries a point mutation between NB and LRR domains of the TIR-NB-LRR protein SNC1, which results in constitutive defense signaling [9], [10]. Suppression of the stunted snc1 phenotype in mos (modifier of snc1) mutants allowed the identification of several genes required for nuclear defense signaling [11]–[14].
Although most R proteins function to recognize a corresponding avirulent effector (Avr), some NB-LRR proteins appear to act downstream of R protein activation. The tobacco and tomato CC-NB-LRR proteins, “N-required gene 1” (NRG1), and “NB-LRR protein required for HR-associated cell death 1” (NRC1), are required for TIR-NB-LRR protein N-mediated resistance to tobacco mosaic virus and receptor-like protein Cf-4-mediated resistance to tomato leaf mold pathogen, respectively [15], [16]. Arabidopsis CC-NB-LRR Activated Disease Resistance 1 (ADR1) family proteins are required for SA-dependent ETI [17]. The Arabidopsis accession Col-0 downy mildew resistance locus RPP2 comprises two distinct closely linked NB-LRR genes RPP2A and RPP2B, both of which are required for resistance [18]. The rice Pia locus for blast (Magnaporthe) resistance comprises two divergently transcribed CC-NB-LRR genes, RGA4 and RGA5, again both required for resistance [19]. In mammals, the NLR NAIP2 confers specific recognition of PrgJ, whereas NLRs NAIP5 and NAIP6 confer responses to flagellin. However, the NLR NLRC4 is required for defense responses to both PrgJ and flagellin [20], [21]. NLRC4 association with either NAIP2 or NAIP5/6, upon provision of PrgJ or flagellin respectively, is required for defense activation [20], [21].
The T3S effectors AvrRps4 and PopP2 from Pseudomonas syringae and Ralstonia solanacearum respectively, are recognized by paired TIR-NB-LRR proteins RPS4 (resistance to P. syringae 4) and RRS1-R (resistance to R. solanacearum 1), and activate ETI in Arabidopsis [22]–[24]. RRS1-R alleles, found in accessions Ws-2, No-0 and Nd-1, confer recognition of PopP2; the RRS1-S allele of Col-0 does not recognize PopP2, but does recognize AvrRps4 [22]–[24]. Lack of AvrRps4 recognition in accession RLD is due to non-synonymous mutations in RPS4, and RRS1-S in Col-0 is truncated compared to RRS1-R because of an early stop codon [24]–[26]. RPS4 and RRS1-R genetically function together, as plants lacking RPS4, RRS1-R or both show similar enhanced susceptibility to bacterial strains expressing AvrRps4 or PopP2 [25], [26]. RRS1 (also annotated as WRKY52) is an atypical NB-LRR protein that also carries a C-terminal WRKY DNA-binding domain [22].
In this study, we delivered PopP2 using Pseudomonas T3S by fusing it with the N-terminal region of AvrRps4 (AvrRps4N). Pseudomonas-delivered AvrRps4N:PopP2 triggers RPS4- and RRS1-dependent HR and immunity in resistant Arabidopsis genotypes when tagged with a nuclear localization signal (NLS) but not when tagged with a nuclear exclusion signal (NES). We show that the delivery of PopP2, or an inactive PopP2C321A variant, from a Pseudomonas fluorescens strain (Pf0-1) that lacks other effectors [27], results in the induction of ETI-specific genes that overlaps substantially with previously reported AvrRps4-regulated genes [28], [29].
The presence of a single amino acid (Leu) insertion in the WRKY domain of RRS1-R (RRS1SLH1 hereafter) causes the recessive lethal phenotype of the sensitive to low humidity 1 (slh1) mutant in No-0 [30]. RRS1SLH1-induced lethality is associated with enhanced defense gene expression and high SA accumulation. Similarly to other mutants displaying spontaneous cell death, slh1 mutant growth can be restored to wild type phenotype at 28°C [30]–[32]. In contrast to snc1, the slh1 mutant allele is recessive and heterozygotes show no constitutive defense activation [30]. RRS1 is also recessive and an RRS1-R/RRS1-S heterozygote is unable to recognize PopP2 [22], [23].
Here, we used the conditional RRS1SLH1-mediated lethal phenotype to gain insights into RPS4/RRS1 gene pair function. Transcriptional profiling of the slh1 mutant shows that genes induced during RRS1SLH1-mediated defense activation in the absence of Avr overlap with those induced by AvrRps4- or PopP2-triggered immunity. Genetic screening for mutations that suppress slh1-triggered aberrant immunity reveals the critical role of RPS4 in RRS1SLH1-mediated activation of defense signaling. Transient expression of RPS4 and RRS1SLH1 in tobacco results in HR in the absence of AvrRps4 or PopP2, which can be suppressed by co-expression of wild type RRS1-R, consistent with the recessive nature of RRS1SLH1. Our study sheds new light on how paired R proteins work cooperatively and illustrates the similarities between auto-active and Avr-dependent defense signaling.
To compare AvrRps4- or PopP2-triggered HR and immunity, we established the delivery of PopP2 via the Pseudomonas T3S. We engineered pEDV6, a Gateway-compatible version of pEDV3 [33], to carry full-length or N-terminally truncated PopP2 variants (Figures 1A and S1A–B). pEDV6 enables expression of a translational fusion between the N-terminal part of AvrRps4 (137 first amino acids; hereafter, AvrRps4N) and an effector of interest. We used a non-pathogenic Pseudomonas fluorescens Pf0-1 engineered to carry a functional T3S system (hereafter, Pf0-1(T3S)) in HR assays because unlike Pseudomonas syringae pv. tomato (Pto) DC3000, Pf0-1(T3S) does not elicit non-specific tissue collapse. When delivered from Pf0-1(T3S) or Pto DC3000, PopP21–488 (full-length) or PopP2149–488 triggered HR and immunity in Arabidopsis accession Ws-2, whereas the PopP2 variants that were further truncated did not (Figure S1C–D). Interestingly, the N-terminal 148 amino acids of PopP2 that include a nuclear localization signal (NLS) are dispensable in our assay. Based on this finding, we used the PopP2149–488 (hereafter, PopP2) variant for the rest of our experiments.
To verify that Pseudomonas-delivered PopP2 confers genotype-specific avirulence, we investigated the responses of Arabidopsis natural variants to PopP2. When delivered from Pf0-1(T3S), PopP2 and AvrRps4 triggered HR in accessions Nd-0 and Ws-2 whereas Col-0 and RLD showed no symptoms at 24 hours post-infection (hpi) (Figure 1B). Col-0 RRS1-S confers HR-deficient disease resistance to Pst DC3000 delivered AvrRps4 but not to PopP2 [22], [34]. In addition, transgenic expression of Ws-2 RRS1-R in Col-0 confers strong HR in response to Pseudomonas-delivered AvrRps4 [35]. HopA1 was used as an additional control; it triggers HR in Nd-0, Ws-2 and RLD, but not in Col-0, as expected. Next, we tested if Pf0-1(T3S)-delivered PopP2 triggers RPS4- and RRS1-dependent HR in Arabidopsis. Pf0-1(T3S)-delivered PopP2 triggered strong HR in wild type Ws-2 whereas Ws-2 rrs1-1, rps4-21, rrs1-1/rps4-21 or eds1-1 mutants did not show any response (Figure 1C). In contrast, Pf0-1(T3S)-delivered AvrRps4 triggered weak but robust HR even in the absence of RPS4 or RRS1 in Ws-2 (Figure 1C). When delivered from Pto DC3000, AvrRps4 triggered immunity in wild type Ws-2, rrs1-1, rps4-21 or rrs1-1/rps4-21 mutants because AvrRps4 recognition leads to RPS4/RRS1-dependent and -independent immunity (Figure 1D) [26]. To test if Pseudomonas-delivered PopP2 can trigger RPS4/RRS1-dependent immunity in Arabidopsis, we engineered a virulent Pto DC3000 to deliver PopP2. Pto DC3000 (PopP2) showed reduced virulence in wild type Ws-2 but not in rrs1-1, rps4-21 or rrs1-1/rps4-21 mutants compared to Pto DC3000 (pEDV5) indicating that Pseudomonas-delivered PopP2 triggers only RPS4/RRS1-dependent immunity (Figure 1D), consistent with previously reported Ralstonia-delivery assay results [26]. By contrast, HopA1-triggered immunity was not affected in rrs1-1, rps4-21 or rrs1-1/rps4-21 mutants compared with wild type Ws-2 (Figure 1D). All tested Pto DC3000 strains showed unrestricted growth in the eds1-1 mutant compared to other genotypes. Taken together, these data indicate that AvrRps4N-mediated delivery of PopP2 from Pseudomonas can trigger RPS4/RRS1-dependent HR and immunity in Arabidopsis.
We further tested if Pseudomonas-delivered PopP2 recognition requires a specific subcellular localization, as reported for AvrRps4 [8]. We engineered a PopP2149–488 variant lacking the native NLS, to carry a NLS or a nuclear export signal (NES) tag at the C-terminus. The avirulence activity of these PopP2 variants was tested in two resistant transgenic Arabidopsis lines, RLD (RPS4Ler) and Col-0 (RRS1Ws-2). Pf0-1(T3S)-delivered PopP2NES, failed to trigger HR in both transgenic lines and in wild type Ws-2, despite being expressed during plant infection, indicating that nuclear localization of PopP2 is required to trigger RPS4/RRS1-dependent HR (Figure S2A, S2E and S3). The PopP2NES variant induced a response comparable to PopP2C321A, an enzymatically inactive variant that does not trigger RPS4/RRS1-R-dependent immunity [36] in wild type Ws-2 when HR-inducing activity was quantified by ion leakage measurements (Figure S2B). We could also show that PopP2NES, in contrast to PopP2NLS, could not restrict the virulence of bacteria when delivered from Pto DC3000, nor trigger expression of defense genes when delivered from Pf0-1(T3S) (Figures S2C and S2D). As these data suggest that PopP2 triggers HR and immunity in the nucleus, we independently assessed previously reported AvrRps4 variants [8]. Unexpectedly, both AvrRps4NLS and AvrRps4NES variants triggered HR and elevated ion leakage in the Ws-2 accession when delivered from Pf0-1(T3S) (Figure S2B and S2E).
RRS1 is a TIR-NB-LRR protein with a WRKY DNA binding domain, which belongs to Group III of the WRKY superfamily [37]. RRS1SLH1, which carries a leucine insertion near the WRKY motif, shows strongly reduced DNA binding by its WRKY domain [30]. This reduced DNA binding correlates with auto-immunity of the slh1 mutant, suggesting a critical role of RRS1 in transcriptional regulation of defense genes. Delivery of PopP2 from Pseudomonas via T3S, combined with the RPS4/RRS1-R dependence of this PopP2-triggered HR, enables direct assessment of RRS1-R-dependent transcriptional regulation. To identify PopP2-triggered and RPS4/RRS1-dependent early transcriptional responses, genome-wide expression profiling was carried out using EXPRSS, an Illumina sequencing based method [38]. Wild type Ws-2 and rrs1-1 plants were infiltrated with Pf0-1(T3S) delivering PopP2WT or PopP2C321A. The infiltrated leaf tissues were collected at 2, 4, 6 and 8 hpi for total RNA extraction, as onset of HR began at 8 hours after bacterial infiltration in an incompatible interaction (PopP2WT/Ws-2).
For differential expression analysis, PopP2WT-infiltrated Ws-2 samples were compared either to PopP2C321A mutant on Ws-2 or PopP2WT on rrs1-1. Essentially complete overlap was observed between the differentially regulated genes in the two comparisons (Figure 2A), consistent with our results showing that Pf0-1(T3S)-delivered PopP2 triggers RRS1- and acetyltransferase activity-dependent immunity (Figures 1 and S2). In total, 719 genes were differentially expressed in an RRS1- and acetyltransferase activity-dependent manner in at least one of the time points surveyed (Table S1). Gene ontology enrichment analysis using ATCOECIS [39] showed that most of the up-regulated genes are involved in defense, while most of the down-regulated genes are involved in photosynthesis and enriched in chloroplast-localized genes (Table S2). Interestingly, the majority of genes differentially expressed at 4 and 6 hpi were up-regulated, while many down-regulated genes were observed at 8 hpi (Figure 2A). The early (4 and 6 hpi) up-regulated genes, such as SID2, FMO1, NudT7, PBS3 and PAD4, have previously been implicated in plant defense (Table S3). Further analysis of mean expression of genes induced at 4 and 6 hpi (Table S3) showed that there was greater gene induction in Ws-2 infiltrated with PopP2WT (∼20–100 fold) than in Ws-2 infiltrated with PopP2C321A or in rrs1-1 infiltrated with PopP2WT (∼2–10 fold). For simplicity, we interpret genes induced by PopP2C321A as induced by the repertoire of PAMPs in Pf0 (thus, PTI-induced), and by PopP2WT as PTI+ETI-induced. To validate our transcriptional expression profiling results, we performed quantitative RT-PCR (qRT-PCR) verification of EDS5, NudT6, WRKY18 and WRKY40 with the cDNA used for Illumina libraries. Expression of EDS5 and NudT6 but not WRKY18 and WRKY40 was specifically regulated by ETI in our expression profiling data. In qRT-PCR experiments, PopP2 but not PopP2C321A variant delivered from Pf0-1(T3S) induced EDS5 and NudT6 in an RRS1-dependent manner, while expression of WRKY18 and WRKY40 was induced in the absence of ETI (Figure S4).
AvrRps4- and PopP2-dependent transcriptional changes in resistant plants have been investigated previously [28], [29], [40]. We compared these available micro-array and RNA-seq data with our results. To minimize the effects of experimental and technical differences from the AvrRps4/Ws-2 data [28], genes altered in expression at 6 hpi due to mock treatment were subtracted from the comparison; similarly, only the GMI1000/GMI1000ΔPopP2-infected Nd-1 data were used from the Hu et al. [40] study. For comparative analysis the differential expression from PTI, PTI+ETI and ETI responses were combined for data presented in this study (Table S1) and the data from Howard et al. [29]. A summary of these comparisons is presented in Figure S5 and details of genes from comparative datasets are presented in Table S4. Transcriptional changes upon AvrRps4 infection on Col-0 and Ws-2 [28], [29] considerably overlapped with PopP2-regulated genes identified both in our study and the GMI1000/GMI1000ΔPopP2 study [40] (Figure S5). The majority of early PTI+ETI-induced genes detected in our study were also found to be AvrRps4-responsive [28], [29] (Figure S5 and Table S4).
We next tested the expression of four PopP2-responsive genes PBS3, SARD1, FMO1 and PR1 by qRT-PCR in Ws-2, rps4-21 and eds1-1. At 8 hpi, Pf0-1(T3S)-delivered AvrRps4WT, HopA1 or PopP2WT triggered similar levels of induction of the four genes in Ws-2 (Figure 2B). Induction of all four genes was strictly dependent on EDS1 and abolished when non-functional variants of the effectors (AvrRps4KRVY-AAAA, AvrRps4E187A and PopP2C321A) were delivered. PTI+ETI-induction of all four genes in response to PopP2 was reduced to PTI-induced levels in both rps4-21 and in rrs1-1 mutants, confirming RPS4/RRS1-R-dependence of PopP2-induced transcriptional changes. AvrRps4-triggered induction of all four genes was reduced but not abolished in the rps4-21 mutant, likely due to RPS4-independent recognition of AvrRps4 in Ws-2 [26], [41]. These expression profiling data thus reveal the genes specifically regulated at very early stages of PopP2-triggered, RPS4/RRS1-dependent immunity in Arabidopsis. Moreover, these ETI transcriptional changes are very similar after AvrRps4 or PopP2 recognition.
To compare slh1 aberrant defense responses to effector-triggered RPS4/RRS1-mediated immunity, we conducted transcription profiling of the slh1 mutant over a time course after shifting plants from 28°C to 19°C, using Illumina tag sequencing [38]. A total of 1821 genes showed temperature-dependent differential expression in RRS1SLH1 after 24 hours (h) compared to wild type No-0 (Figure 3A). We confirmed the temperature-dependent regulation of 3 genes with differential induction in slh1 by qRT-PCR. PR1, PBS3 and CBP60g transcript accumulation was induced in slh1 plants between 9 and 24 h after the shift from 28°C to 19°C whereas it was unaltered in temperature-shifted No-0 plants (Figure 3B).
We compared the slh1/No-0 temperature-shift transcriptional dataset to the PopP2/RRS1-time course dataset by analyzing the pairwise overlap of genes differentially expressed in both experiments (Figure 4). Each time course response was categorized according to the mode of elicitation as PTI, ETI, temperature shift, auto-immunity, or corresponding combinations (e.g. PTI+ETI). We found that most (∼83%) of the PopP2/RRS1 ETI genes were differentially expressed in slh1 auto-immune and temperature shift responses, while up to 54% of ETI genes were also differentially expressed in the auto-immune response but not by temperature shift (Figure 4, black box). Similarly, more than 55% of auto-immune genes were also differentially expressed in PTI and PTI+ETI (Figure 4, dotted block box). Most ETI genes were also differentially expressed in PTI+ETI (more than 85%) and in PTI (up to 70%). However, less than 10% of the PTI genes were differentially expressed during ETI (Figure 4, blue box). This strongly suggests that many ETI responses involve potentiation of a subset of PTI responses, with few genes solely regulated by effector recognition. The ETI-specific genes that are regulated in PopP2 acetyltransferase activity- and RRS1-dependent manner include nucleotide/ATP-binding protein encoding genes such as NB-LRRs (Table S1).
Similarly, we found that most temperature shift-regulated genes (up to 83%) (Table S5) were also differentially expressed by PTI or PTI+ETI, but only 25% were specifically affected by ETI, and less than 5% of the PTI-responsive genes were differentially expressed by temperature shift (Figure 4, green box). Up to 50% of PTI or PTI+ETI genes were also differentially expressed by temperature shift and auto-immune response, while about 25% of PTI or PTI+ETI genes were differentially expressed by auto-immune response (Figure 4, green box). These results indicate that PTI broadly activates genes responsive to heat, auto immunity and ETI.
These analyses indicate that slh1 auto-immunity overlaps strongly with PopP2- and RPS4/RRS1-R-dependent ETI. Thus, RRS1SLH1-induced transcriptional reprogramming results in similar gene expression changes to those observed in AvrRps4- or PopP2-triggered immunity, indicating that the slh1 lethal phenotype mimics RPS4/RRS1-dependent ETI at the transcriptional level.
Lethality of slh1 at 21°C is correlated with constitutive activation of defense responses including high expression of Pathogenesis Related (PR) genes and SA accumulation [30]. We hypothesized that mutations that affect RRS1SLH1-mediated signaling components or RRS1SLH1 expression would suppress slh1 lethality. To identify genetic components required for RRS1SLH1-dependent immunity, we conducted a suppressor screen. slh1 seeds were incubated with ethyl methanesulfonate (EMS), ∼7,000 M1 plants were grown at 28°C and M2 seeds were harvested. By screening ∼500,000 M2 mutant plants at 21°C, we identified 83 families with a suppressor of slh1 immunity (sushi) mutant phenotype. Among them, 69 and 14 could rescue the slh1 lethal phenotype to a wild type-like and an improved morphology, respectively. We further analyzed the progeny of 7 selected fully rescued sushi mutants for morphological development and defense marker gene expression in the M3 generation (Figure 5). Growth of sushi mutants at 21°C was similar to wild type No-0, whereas slh1 plants did not develop beyond the first true leaf stage (Figure 5A). PR1, PBS3 and FMO1 expression was elevated in slh1 mutants grown constantly at 21°C or 24 h after shift from 28°C to 21°C, but not in fully rescued sushi mutants (Figures 5B and S6).
To exclude any contamination with wild type seeds, we confirmed the presence of the slh1 mutation in 72 of the 83 M3 individual sushi mutants identified using a cleaved amplified polymorphic sequences (CAPS) marker [30]. Next, we carried out Sanger sequencing of RRS1 and RPS4 coding regions in these mutants. As expected from the complete suppression of the slh1 phenotype, we identified 6 sushi intragenic suppressor mutants that carry an early stop codon in RRS1SLH1 and 8 other non-synonymous mutations (Table S6). Surprisingly, non-synonymous mutations were also identified throughout the RPS4 coding region in 34 rescued sushi mutants (Table S6). Most of the altered amino acid residues have not previously been shown to be required for RPS4 function [24]. However, sushi52 and sushi22 harbour non-synonymous mutations at positions R28 and E88 that are important for RPS4TIR+80-triggered HR in tobacco [42], further verifying the crucial role of the RPS4 TIR domain function in RRS1SLH1-mediated defense activation.
It was previously reported that mutations in SID2/ICS1/EDS16 or SID1/EDS5 result in suppression of the RRS1SLH1 mutant phenotype [30]. We sequenced the coding region of these genes in the non-RRS1, non-RPS4 mutants, and found one sushi mutant that carried a mutation in SID2/ICS1/EDS16 (sushi70, Table S6), and no mutants that carry mutations in SID1/EDS5. Similarly to Arabidopsis accession Col-0, wild type No-0 carries two copies of EDS1. Therefore, EDS1 coding sequence was not verified in the sushi lines. The 23 remaining unassigned SUSHI mutations are now subjected to further analysis to identify new signaling components of RRS1SLH1-mediated immunity.
Homo- or hemizygous, but not heterozygous, No-0 plants carrying RRS1SLH1 display a stunted phenotype at 21°C due to elevated immunity [30]. To verify that RPS4 is required for RRS1SLH1 function, we crossed 7 sushi lines carrying mutations in RPS4 (sushi17, 64, 24, 12, 41, 58 and 32) to rrs1-1 and rrs1-1 rps4-21 knockout mutants [26]. The resulting F1 individuals from both crosses were hemizygous RRS1SLH1/rrs1 for RRS1 locus (Figure S7) and either RPS4sushi/RPS4WT or RPS4sushi/rps4 at the RPS4 locus. As expected, the F1 plants derived from a cross between the sushi and rrs1-1 were stunted and showed elevated PR1 expression level (Figure 6A–C). These phenotypes were both completely suppressed in the F1 plants derived from a cross between sushi mutants in RPS4 and rrs1-1 rps4-21 double mutant. This result confirms that RPS4 is required for RRS1SLH1-mediated activation of immunity.
To further verify the functional requirement for RPS4 in RRS1SLH1-mediated immunity, we recapitulated RRS1SLH1-mediated defense activation in Nicotiana tabacum. As shown recently [39], Agrobacterium-mediated transient co-transformation (hereafter, agroinfiltration) of RPS4-HA, RRS1-His-Flag and wild type AvrRps4-GFP or PopP2-GFP induced strong HR within 3 dpi (Figure S8A). The specificity of recognition was further verified by comparing functionally characterized mutant variants of AvrRps4 or PopP2 to wild type. As expected, AvrRps4E187A, AvrRps4KRVY-AAAA and PopP2C321A variants did not induce RPS4/RRS1-dependent HR in tobacco (Figure S8A). We have also verified that AvrRps4 and PopP2 recognition in tobacco activate defense genes orthologous to those that are regulated by RRS1 in Arabidopsis. The transcripts of the defense genes NtWRKY51 and NtNudT7 were highly up regulated when PopP2-GFP was co-expressed with RPS4-HA and RRS1-His-Flag in tobacco (Figure S8B). Agroinfiltration of GFP or PopP2C321A-GFP with RPS4-HA and RRS1-His-Flag induced significantly lower accumulation of defense gene transcripts compared to wild type PopP2 (Figure S8B). Taken together, these results further demonstrate that our transient agroinfiltration assay can also be used to investigate RPS4/RRS1 regulated immunity.
Agroinfiltration of epitope-tagged RRS1SLH1-His-Flag and RPS4WT-HA triggered HR in tobacco leaf cells, whereas RRS1SLH1 co-expressed with GFP or RPS4K242A (P-loop mutant) did not (Figures 6D and 8B). Consistent with our Arabidopsis genetic data (Figure 6B), agroinfiltration of RRS1SLH1 with each RPS4SUSHI variant did not trigger HR in tobacco (Figure 6D). Protein accumulation of the 7 tested RPS4SUSHI variants was comparable to that of RPS4WT, indicating that the lack of HR was not due to low protein expression levels (Figure S9). Moreover, as expected from our genetic analysis, RPS4SUSHI variants did not have a dominant negative effect on RPS4WT function, when both were co-expressed with RRS1SLH1 (Figure S10). We then tested whether SUSHI mutant alleles of RPS4 confer RRS1-dependent recognition of AvrRps4 or PopP2. Agroinfiltration of RRS1WT, RPS4WT and either AvrRps4 or PopP2, triggered RPS4 P-loop-dependent HR in infiltrated tobacco leaf sectors [43] (Figure 6D). Importantly, agroinfiltration of the 7 RPS4SUSHI variants did not confer responsiveness to AvrRps4 or PopP2 (Figure 6D). Taken together, these data show that RPS4 is required for RRS1SLH1-mediated and Avr-triggered/RRS1-dependent defense signaling activation. Recently, we showed the physical interaction of RRS1 and RPS4 [43]. We hypothesized that RPS4SUSHI variants may have lost their ability to interact with RRS1SLH1. However, RPS4SUSHI-HA variants and RPS4WT-HA were co-immunoprecipitated by RRS1SLH1-Flag or RRS1WT-Flag (Figure S9A-B). This result suggests that RPS4-RRS1 interaction is insufficient for signaling activation.
We identified six additional sushi mutants that carry mutations in the TIR domain of RPS4, the structure of which is known [43]. The stunted growth and elevated defense transcript accumulation of slh1 at 21°C were considerably suppressed in sushi52 (R28H), 14 (A38V), 22 (E88K), 71 (L101F), 89 (P105L) and 29 (G120R) (Figure S11). The RPS4 TIR domain structure suggests that side-chains from R28 and A38 are surface exposed, while the side-chains of the other mutated residues are buried (Figure 7A). RPS4TIR expression is sufficient to trigger HR in tobacco after agroinfiltration (Figure 7B) [42]. Therefore we introduced these six SUSHI mutations into an RPS4TIR construct (amino acids 1 to 250) to test their individual effect on RPS4 TIR domain signaling. Strikingly, all six mutations suppressed this response, suggesting that each of these residues is important for RPS4 TIR domain defense activation either through interaction with downstream partners or by maintaining the correct signalling-competent structural conformation, as the protein stability/accumulation was not significantly altered when expressed as GFP fusions in tobacco (Figure S12D). Intriguingly, when SUSHI mutations were tested in the RPS4 full-length context by co-expression in tobacco with RRS1 and the effectors, A38V and L101F did not suppress RRS1SLH1- nor AvrRps4- and PopP2-triggered HR (Figure 7C). This discrepancy was not due to inconsistent level of protein accumulation (Figure S12E) but might illustrate a limitation of the transient expression system in tobacco, or subtle differences between defense activation by RPS4TIR, and by the activated RPS4/RRS1 complex.
As nuclear localization of RPS4 is necessary for AvrRps4-triggered immunity [41], we investigated the role of RPS4 nuclear localization in RRS1SLH1-mediated cell death. Co-expression of RRS1SLH1 with RPS4WT or RPS4NLS induced HR (Figure 8A). However, RPS4NES did not induce HR when co-expressed with RRS1SLH1, indicating the importance of RPS4 nuclear localization for RRS1SLH1 function, consistent with a previous report [41]. Nucleotide binding to the invariant Lys residue of the P-loop motif in the NB domain of R proteins is critical for conformational change and immunity activation [4], [44], [45]. Agroinfiltration of RPS4WT, but not the P-loop mutant RPS4K242A, triggered HR when co-expressed with RRS1SLH1 (Figure 8B). However, RPS4K242A does interact with RRS1SLH1 and RRS1WT (Figure S9C). Therefore, a functional RPS4 P-loop motif is required for activation of RRS1SLH1-induced defense but is not an absolute requirement for RPS4-RRS1 interaction. Surprisingly, introduction of the P-loop mutation (K185A) in the RRS1SLH1 protein sequence did not affect HR-inducing activity when co-expressed with RPS4WT (Figure 8B). Thus, P-loop motif-dependent conformational change may not be required for defense activation by RRS1SLH1, consistent with the functionality of an RRS1 P-loop mutant in AvrRps4 or PopP2 recognition [43].
Structural analysis of RPS4 and RRS1 TIR domains revealed an “SH motif” in regions that mediate heterodimerization between RPS4 (S33 and H34) and RRS1 (S25 and H26) [43]. Moreover, RPS4 or RRS1 variants carrying a mutated SH motif (SH-AA) cannot recognize AvrRps4 or PopP2 in tobacco agroinfiltration [43]. To investigate if TIR-TIR domain heterodimerization is also required for RRS1SLH1 function, SH-AA mutations were introduced in RPS4WT and RRS1SLH1 variants. Agroinfiltration of RRS1SLH1 and RPS4SH-AA, or RRS1SLH1/SH-AA and RPS4WT did not induce HR in tobacco suggesting that TIR-TIR domain heterodimerization between RRS1 and RPS4 is required for RRS1SLH1-dependent defense activation (Figure 8B). However, in the context of the full-length proteins the RRS1SLH1/SH-AA variant could still interact with RPS4WT (Figure S9C).
RRS1SLH1-dependent lethality is recessive [30]. In agreement, agroinfiltration of RRS1WT but not of GFP interfered with HR induced by co-expression of RRS1SLH1 and RPS4WT in tobacco (Figure 8C–D). Interestingly, the RRS1K185A variant did not interfere with RRS1SLH1-induced HR whereas the RRS1SH-AA variant did (Figure 8C), indicating that nucleotide-binding function but not RPS4/RRS1 TIR-TIR domain interaction is required for RRS1-mediated interference with RRS1SLH1-induced HR. These agroinfiltration data are consistent with our transcriptomic and genetic analyses and demonstrate the striking similarity of RRS1SLH1 and Avr-triggered/RRS1-dependent defense activation.
As RRS1SLH1/RPS4-dependent constitutive HR is prevented by co-expression of RRS1WT, we tested if RRS1SLH1 interferes with RRS1WT recognition of AvrRps4 or PopP2. Interestingly, in the presence of both RRS1 variants and RPS4, AvrRps4- or PopP2-triggered HR is still observed suggesting that RRS1SLH1 did not completely abolish RRS1WT function (Figure 8D). However, AvrRps4-triggered HR was attenuated considerably compared to PopP2-triggered HR under the same conditions (Figure 8D).
Although both AvrRps4 and PopP2 are recognized by RPS4 and RRS1, a thorough comparison of immune responses, particularly of early transcriptional changes, has been difficult due to the distinct infection modes of the bacterial pathogens from which AvrRps4 (Pseudomonas syringae) and PopP2 (Ralstonia solanacearum) originate. Root infection of Arabidopsis plants with R. solanacearum causes wilting within 2 weeks, whereas Pseudomonas-delivered AvrRps4 triggers HR in Arabidopsis Ws-2 leaf cells within 24 hours. PopP2 delivery from Pf0-1(T3S) allowed us to compare the transcriptional reprogramming caused by recognition of AvrRps4 or PopP2 at the earliest stages and has resulted in the identification of a set of similarly regulated ETI-specific genes. It is interesting that the NLS is dispensable for the avirulence activity of PopP2 in our assays. It was shown that removal of the N-terminal NLS renders localization of PopP2 and co-expressed RRS1-S/R variants nuclear-cytoplasmic [46]. However, the significance of this PopP2 NLS-dependent relocalization of RRS1 is not known, as there have been no studies showing ETI phenotypes triggered by PopP2 lacking the NLS. As shown in Figure S2, a PopP2 variant lacking an N-terminal NLS shows similar levels of avirulence compared to wild type. Thus, PopP2 NLS-dependent relocalization of RRS1 may not be significant in PopP2-triggered immunity. Alternatively, the portion of RRS1 that is localized in the nucleus with the NLS lacking PopP2 might be sufficient to activate ETI.
It is intriguing to find that AvrRps4NES and AvrRps4NLS are comparable in their ability to elicit HR in Arabidopsis Ws-2 (Figure S2E). AvrRps4NES triggers a slightly lower ion leakage level than AvrRps4NLS (Figure S2C). We conclude that regardless of AvrRps4 contribution to defense activation in the cytoplasm, its major role is in the nucleus via interactions with the RPS4/RRS1 complex.
Pseudomonas T3S delivery of PopP2 provides a useful tool to investigate RPS4/RRS1-dependent transcriptional regulation at an early stage of ETI. In addition, by comparing non-functional variants of AvrRps4 and PopP2 to wild type proteins, we could identify the genes whose transcriptional changes were specific to Avr function. As Pf0-1(T3S) carries a mutated HopA1 gene which is unable to trigger RPS6-dependent immunity in Arabidopsis, the gene expression change in rrs1-1 infiltrated with PopP2WT or in Ws-2 infiltrated with PopP2C321A can be considered as PTI resulting from perception of the Pf0-1 PAMP repertoire. We thus report defense gene expression changes as PTI vs. PTI+ETI (Table S3). Gene ontology enrichment has shown that the majority of early up-regulated genes are involved in plant defense.
Comparative analysis with previously published microarray data shows that many PopP2-triggered early gene expression changes overlap substantially with AvrRps4-triggered transcriptional regulation [28], [29]. It is interesting to note that PopP2-regulated genes also overlap substantially with previously reported PopP2-induced genes at a later stage of infection when delivered from R. solanacearum [40]. Our discovery of early responding genes will allow us to test if they are directly regulated by RPS4/RRS1. It has been recently shown that WRKY18 and WRKY40 positively contribute towards AvrRps4-triggered immunity [47]. Consistent with this, WRKY18 and WRKY40 were highly induced at 3 and 6 hpi by AvrRps4 (Table S4). However, our experimental design enabled us to show that both WRKY18 and WRKY40 are primarily induced due to PTI (Figure S4). PTI+ETI and PTI induction of WRKY40 expression are indistinguishable. There is slightly higher PTI+ETI-induced expression of WRKY18 in response to PopP2WT in Ws-2 at later time points (6 and 8 hpi) compared to PTI elicited by PopP2C321A in Ws-2 or PopP2WT in rrs1-1 (Figure S4), but this could be due to elevated SA levels that we presume are responsible for strong PR1 induction at 8 hpi.
It is interesting to note that AvrRps4-induced regulation of ETI genes only partially requires RPS4. This is consistent with AvrRps4 recognition being conferred by both RPS4/RRS1-dependent and -independent mechanisms. Identification of an R gene(s) that confer RPS4/RRS1-independent immunity will enable comparative analysis of how AvrRps4-induced ETI genes are transcriptionally regulated by different R genes. It was remarkable to observe that AvrRps4, PopP2 and HopA1 induced common genes at early stage of defense activation, suggesting a possible EDS1-dependent conserved gene activation mechanism in ETI.
Several auto-active alleles of NB-LRR genes have been found [9], [10], [30], [48], [49], though unlike the recessive slh1, all others are dominant or semi-dominant. Plants carrying an auto-active R gene typically show temperature-dependent lethality and enhanced resistance to virulent pathogens [30]–[32]. However, in many cases the overlap between elevated disease resistance that is conferred by an auto-active R gene allele and by Avr-triggered immunity is poorly defined. Unlike most other auto-active R gene alleles, RRS1SLH1 carries a single amino acid insertion in the WRKY-DNA binding domain that reduces its DNA-binding affinity [30]. To address the role of RRS1 in transcriptional activation or repression, we tested whether RRS1SLH1-induced transcription changes overlap with AvrRps4- or PopP2-triggered transcription changes. Based on previously reported expression profiling data and the present study, we propose that the genes whose transcripts were differentially regulated by RRS1SLH1, and by AvrRps4 and PopP2 are directly regulated by RRS1 upon Avr detection. As exons 6 and 7 of RRS1SLH1 show reduced binding to a W-box in vitro, RRS1 may act as a transcriptional repressor of plant immunity, or at least as a repressor of RPS4 function, and this repression may be relieved upon Avr perception [30]. However, RRS1 could act both as repressor and activator of defense gene transcription, as has been found for other plant transcription factors [50]. Loss of RRS1-DNA binding may be part of the activation of defense transcription, but paradoxically, rrs1 knockout lines do not show enhanced immunity.
Identification of RPS4 mutant alleles among the SUSHI mutations was unexpected, as we had anticipated that RRS1 might act downstream of RPS4 to regulate defense gene transcription directly. Notably, it would have been difficult to recover recombinants between RRS1SLH1 and an adjacent mutant allele of RPS4, so without a genetic screen, this discovery might not have been made. Based on the genetic requirement of RPS4 for RRS1SLH1-induced defense gene transcription, we now hypothesize that RPS4 is required to form a functional immune receptor complex with RRS1. This hypothesis is further supported by the fact that RPS4 and RRS1 interact with each other, in part but not solely by forming a TIR-TIR domain heterodimer [43]. In addition, the requirement of a functional P-loop motif for RPS4 but not for RRS1 function suggests that RPS4 contributes to defense activation by providing ATP-dependent conversion of a repressive immune receptor complex to an activated state. PopP2 interacts with RRS1 [46], as does AvrRps4 [43]. We hypothesize that RPS4 activates defense upon recognition of perturbations in RRS1 by effectors, and that RRS1SLH1 mimics the results of effector action upon RRS1. Can this be reconciled with the observation that a 35S:RPS4 constitutive defense phenotype partially requires RRS1 [51]? Conceivably, RRS1 might also play a chaperone-like role in facilitating conversion of RPS4 from an inactive to an active form, and RRS1SLH1 has enhanced activity in facilitating this conversion.
The TIR domain of RPS4 induces cell death when transiently overexpressed in tobacco. Several amino acid residues were shown to be required for RPS4 TIR domain auto-activity [42]. Among the 33 single amino acid polymorphisms of RPS4 that we identified in sushi mutants, two residues, R28 and E88, were previously implicated as being required for RPS4 TIR domain-induced auto-activity in tobacco. R28H and E88K mutations are unlikely to alter the overall structure of RPS4 TIR domain, judging from the crystal structure of RPS4/RRS1 TIR domain heterodimer [43]. A study on RPS4 natural variants identified Y950 as an important residue for function as a susceptible RLD allele of RPS4 carries a Y950H mutation, and a Y950H substitution in the functional Ler allele of RPS4 abolishes its AvrRps4-recognition capability [24]. Interestingly, we identified several mutations (S914F, G952E and G997E) in this C-terminal domain (CTD) of RPS4. Although the function of the RPS4 CTD remains unclear, it appears to be important for immune signaling. Conceivably, the sushi-mutated residues found in the TIR domain (R28, E88, P105L and G120R) and in the CTD (S914F, G952E, and G997E) are involved in the interaction with RRS1 or other yet unknown partner(s).
AvrRps4 and PopP2 interact directly with RRS1 [43], [46]. Conceivably, after interaction of AvrRps4 or PopP2 with RRS1, dissociation of the activated RPS4/RRS1 immune complex from target DNA induces RPS4 P-loop-dependent de-repression/activation of defense gene transcription, perhaps via WRKY18 and WRKY40 [47]. There may be multiple WRKY transcription factors that can replace the transcriptional repression function of RRS1, but not its Avr-recognition function. However, the Ws-2 RRS1SLH1 allele may make additional contributions to assembling a defense-activating complex beyond vacating W-boxes.
An intriguing feature of RRS1 is that it is the only known recessive NB-LRR-encoding R gene. Consistent with this observation, the slh1 mutation is also recessive. We were able to recapitulate this feature by transiently co-expressing RRS1 with RPS4 and RRS1SLH1 and suppressing RPS4/RRS1SLH1-triggered HR. This suppression is abolished if the RRS1-R carries a mutation in its P-loop motif. Intriguingly, this result suggests that the RRS1-R P-loop is not required for RPS4-dependent HR activation, but potentiates assembly of an inactive, poised complex. Thus, we suggest that the recessive nature of RRS1 in the Col-0(S)/Nd-0(R) or Col-0(S)/Ws-2(R) cross is the result of the Col-0 allele encoding a protein that can interfere in trans with PopP2 responsiveness and thus acts as a “poison subunit”.
There are nine TIR-NB-LRR gene pairs reported in the Arabidopsis Col-0 genome [26]. It is important to better understand how paired R proteins have evolved and recognize effectors. It is interesting to note that all three TIR-NB-LRR-WRKY encoding genes (At5g45260, At5g45050 and At4g12020) found in Arabidopsis are paired with TIR-NB-LRR genes [26]. This suggests that at least some other paired R proteins may function cooperatively in the nucleus by directly regulating transcriptional processes.
In conclusion, the deployment of a Pseudomonas T3S delivery of PopP2 allowed a detailed comparison of AvrRps4- and PopP2-triggered RPS4- and RRS1-dependent transcriptional regulation. We found that an auto-active allele of the TIR-NB-LRR-WRKY protein RRS1, RRS1SLH1, induces immune responses comparable to Avr-triggered immunity. The suppressor of slh1 immunity screening enabled us to uncover the critical role of RPS4 in RRS1SLH1-mediated defense activation. Furthermore, we defined additional properties of RPS4 and RRS1 that are essential for function, and these results significantly enhance our understanding of NB-LRR protein function in plants.
Arabidopsis plants were grown in short day conditions (10 h light/14 h dark) at 21°C or 28°C. Nicotiana benthamiana and Nicotiana tabacum ‘Petit Gerard’ plants were grown in long day conditions (16 h light/8 h dark) at 24°C. No-0 and slh1 are described in [30]; Ws rrs1-1 and Ws rrs1-1 rps4-21 are described in [26].
To create pEDV6 (gateway destination variant of pEDV3), the nucleotide sequence encoding the HA epitope tag was inserted at SalI site of pEDV3 [33] that resulted in AvrRps4N(1-137aa):HA:ClaI:BamHI (pEDV5). Subsequently, pEDV5 was digested with ClaI and BamHI, treated with T4 DNA-polymerase to generate blunt ends and ligated with EcoRV digested Gateway reading frame cassette B (RFB) (Invitrogen) to create pEDV6. Construction of pBBR1MCS-5:avrRps4 was described previously [35]. The NES- or NLS-tagged avrRps4 variants were kindly provided by Jane Parker laboratory and the cloning procedure was described previously [8]. To generate pEDV6:popP2 variants, full-length or truncated popP2 fragments were amplified from Ralstonia solanacearum genomic DNA by polymerase chain reaction and cloned in the Gateway entry vector, pCR8 (Invitrogen). Introduction of popP2 fragments in pEDV6 was performed according to manufacturer's instructions by using LR recombinase II (Invitrogen). The pBin19:RPS4:HA construct was described previously [52]. To obtain C-terminally GFP tagged AvrRps4 or PopP2 variants, avrRps4 or popP2 coding regions were PCR amplified and cloned at ClaI and BamHI sites of EpiGreenB5:GFP. Construction of 35S:RRS1:His-Flag is described in [43]. Wild type and mutant variants of AvrRps4 and PopP2 were PCR amplified from previously reported plasmid constructs [35], [53]. The resulting PCR fragments were cloned in pCR8 (Invitrogen) and correct sequences were confirmed. These pCR8 constructs were used for LR reaction with the Gateway destination vector pK7FWG2 (35S promoter and C-GFP) to generate C-terminally GFP-tagged AvrRps4 and PopP2 variants. Wild type and SH-AA mutant variants of RPS4-HA and RRS1-His-Flag are described in [43]. Introduction of SLH1 and SUSHI mutations in RRS1 and RPS4, respectively, was achieved by using Quikchange II XL site-directed mutagenesis kit (Agilent). The C-terminally GFP-tagged RPS4 constructs were generated by inserting ClaI/BamHI digested RPS4 in EpiGreenB5-GFP-WT/NES/NLS.
Escherichia coli DH5α was used for maintaining plasmid constructs and bacterial conjugation. For hypersensitive response assay and in planta bacterial growth assay, Pseudomonas fluorescens Pf0-1(T3S) and Pseudomonas syringae pv. tomato (Pto) DC3000 strains were used, respectively. To introduce various constructs carrying avrRps4, popP2 or hopA1 in Pf0-1(T3S) and Pto DC3000, standard triparental mating method was used by using E. coli HB101 (pRK2013) as a helper strain as previously described [33]. For transformation of Agrobacterium tumefaciens strain AGL1, standard electroporation method was used.
For hypersensitive response assay, freshly grown Pf0-1 (T3S) strains on King's B agar plates containing appropriate antibiotics were harvested in 10 mM MgCl2. The final concentration of bacterial suspensions was adjusted to A600 = 0.2. Leaves of five week-old Arabidopsis plants were hand-infiltrated by using 1 mL needless syringes and kept 20–24 h further for symptom development. For ion leakage assays, leaf discs were sampled at 0.5 hpi, floated on water for 30 minutes (with gentle shaking at room temperature) and transferred to fresh water (1 hpi sample). Ion leakage was measured at 24 hpi using a conductivity meter. For in planta bacterial growth assays, Pto DC3000 strains were grown and harvested as for Pf0-1(T3S). Leaves of five week-old Arabidopsis plants were hand-infiltrated with bacterial suspensions (A600 = 0.001) by using 1 mL needless syringes and kept 3–4 days further before sampling. Infected leaf samples were ground in 10 mM MgCl2, serially diluted, spotted on L agar plates containing appropriate antibiotics and kept at 28°C for 2 days before counting colonies to estimate bacterial population in infected leaves.
Agrobacterium tumefaciens AGL1 strains carrying the different constructs were grown in liquid L-medium supplemented with adequate antibiotics for 24 h. Cells were harvested by centrifugation and re-suspended at OD600 0.5 in infiltration medium (10 mM MgCl2, 10 mM MES pH 5.6). For co-expression, bacterial suspensions were mixed in 1∶1 ratio before infiltration with needleless syringes in 5 week-old N. benthamiana or N. tabacum leaves. Tobacco hypersensitive response was generally observed and photographed 2 to 3 days after infiltration.
EXPRSS tag-seq cDNA library construction and data analysis was carried out as described previously [38]. The sequence data presented in this publication have been deposited in NCBI's Gene Expression Omnibus [54] and are accessible through GEO Series accession number GSE48247 and GSE51116. Tag to gene associations were carried out using uniquely mapped reads, with the considerations described previously [38]. Bowtie v0.12.8 [55] was used to map short reads to TAIR10 genome and Novoalign v2.08.03 (http://www.novocraft.com/) was used to align remaining reads to TAIR10 cDNA sequences. Differential gene expression analysis was performed using the R statistical language (v2.11.1) with the Bioconductor package [56], edgeR v1.6.15 [57] with the exact negative binomial test using tagwise dispersion and selected genes with false discovery rate (FDR) <0.01. From RNA-seq data for avrRps4 on Col-0 [29], uniquely mapped read counts to genes were used for reanalysis using edgeR and selected gene with FDR <0.05.
Microarray data files from Pto DC3000 (AvrRps4) infiltration (Array Express E-MEXP-546, [28]) and Interaction of Arabidopsis thaliana and Ralstonia solanacearum (NASCARRAYS-447, [40]) were used. Data analysis was performed using the R statistical language as described previously [38], [58]. Differentially expressed genes were identified using the rank products method with FDR <0.05 [59]. As Pto DC3000 (AvrRps4) data has only one replicate, differential expression analysis was carried out with untreated and MgCl2 infiltrated 3 hpi samples as controls and compared against 3 and 6 hpi of avrRps4 and 6 hpi of MgCl2. For GMI1000/GMI1000ΔPopP2 data, only Nd-1 samples were used.
Total RNAs were extracted from 4 to 5 week-old Arabidopsis plants using the TRI reagent (Invitrogen) according to the manufacturer's instructions. First-strand cDNA was synthesized from 5 µg RNA using SuperScriptII Reverse Transcriptase (Invitrogen) and an oligo(dT) primer, according to the manufacturer's instructions. cDNA was amplified in triplicate by quantitative PCR using SYBR Green JumpStart Taq ReadyMix (Sigma) and the CFX96 Thermal Cycler (Bio-Rad). The relative expression values were determined using the comparative Ct method and Ef1α (At5g60390) as reference. Primers used for quantitative PCR are described in Table S7.
The presence of the slh1 mutation in sushi M3 generation and F1 individuals resulting from the genetic cross with rrs1-1 or rrs1-1 rps4-21 was assessed using the CAPS marker described in [30]. For sequencing of candidate genes on sushi mutants genomic DNA, 10, 6, 4 and 4 couples of primers respectively were used to amplify regions of RRS1, RPS4, EDS16 and EDS5 coding sequence (see Table S7). PCR products were purified on Sepharose and sequences were analyzed using the Vector NTI assembly software (Invitrogen).
Protein samples were prepared from N. benthamiana 48 h after Agrobacterium-mediated transformation. One infiltrated leaf was harvested and ground in liquid nitrogen. Total proteins were extracted in GTEN buffer (10% glycerol, 100 mM Tris-HCl pH 7.5, 1 mM EDTA, 150 mM NaCl) supplemented extemporaneously with 5 mM DTT, 1% (vol/vol) plant protease inhibitor cocktail (Sigma) and 0.2% (vol/vol) Nonidet P-40. Lysates were centrifuged for 15 min at 5,000 g at 4°C and aliquots of filtered supernatants were used as input samples. Immunoprecipitations were conducted on 1.5 mL of filtered extract incubated for 2 h at 4°C under gentle agitation in presence of 20 µL anti-FLAG M2 or EZview anti-HA affinity gel (Sigma). Antibodies-coupled agarose beads were collected and washed three times in GTEN buffer, re-suspended in SDS-loading buffer and denatured 10 min at 96°C. Proteins were separated by SDS-PAGE and analyzed by immunoblotting using anti-FLAG M2-HRP, anti-GFP-HRP or anti-HA-HRP conjugated antibodies (Sigma, Santa Cruz and Roche respectively).
|
10.1371/journal.pbio.0060059 | A Role for Parasites in Stabilising the Fig-Pollinator Mutualism | Mutualisms are interspecific interactions in which both players benefit. Explaining their maintenance is problematic, because cheaters should outcompete cooperative conspecifics, leading to mutualism instability. Monoecious figs (Ficus) are pollinated by host-specific wasps (Agaonidae), whose larvae gall ovules in their “fruits” (syconia). Female pollinating wasps oviposit directly into Ficus ovules from inside the receptive syconium. Across Ficus species, there is a widely documented segregation of pollinator galls in inner ovules and seeds in outer ovules. This pattern suggests that wasps avoid, or are prevented from ovipositing into, outer ovules, and this results in mutualism stability. However, the mechanisms preventing wasps from exploiting outer ovules remain unknown. We report that in Ficus rubiginosa, offspring in outer ovules are vulnerable to attack by parasitic wasps that oviposit from outside the syconium. Parasitism risk decreases towards the centre of the syconium, where inner ovules provide enemy-free space for pollinator offspring. We suggest that the resulting gradient in offspring viability is likely to contribute to selection on pollinators to avoid outer ovules, and by forcing wasps to focus on a subset of ovules, reduces their galling rates. This previously unidentified mechanism may therefore contribute to mutualism persistence independent of additional factors that invoke plant defences against pollinator oviposition, or physiological constraints on pollinators that prevent oviposition in all available ovules.
| Much biodiversity ultimately relies on cooperation between different species, interactions called mutualisms. Benefits to one partner are gained by obtaining resources from the other, presenting a problem: what prevents one partner from exploiting the other at an unsustainable level? Fig trees are pollinated by tiny wasps that only develop successfully themselves by each destroying a single female fig flower that would otherwise become a seed. Wasps tend to occur in long flowers near the fruit's centre, with seeds developing near the outer wall. Female wasps therefore favour long flowers for their offspring, and leave short flowers to develop into seeds. To understand why wasps exploit fig trees sustainably, we need to explain why this preference has evolved. The fig-pollinator mutualism is exploited by small parasitic wasps that attack pollinators from outside the fruit. In three Australasian fig species, we found that pollinator offspring in the outer layer of flowers were more likely to be parasitized than those in the inner layer. Our data thus indicate that long flowers provide enemy-free space for pollinator offspring at the fruit's centre. We suggest that the provision of variable length flowers by fig trees may contribute to mutualism stability by indirectly involving a third party: parasitic wasps, previously regarded as detrimental to both mutualists.
| In a biosphere driven by selection at the level of the individual gene [1], explaining the existence of cooperation, such as mutualism, is a major scientific challenge. Mutualisms are interspecific ecological interactions characterised by reciprocal benefits to both partners [2] that usually involve costly investments by each. What factors thus prevent one partner from imposing unsustainable costs onto the other to enable mutualism stability [3–7]? In some mutualisms, the larger, more sessile partner, manipulates the other by directing benefits to cooperative individuals and costs to cheaters [4–7]. However, a general consensus on mutualism persistence has only recently been formulated, and this clearly shows that a high benefit-to-cost ratio of cooperating is one important factor [8,9].
Fig trees (Ficus) and their host-specific agaonid pollinator wasps are a classic example of an obligate mutualism [10,11]. The wasps pollinate the trees, and the trees provide resources for wasp offspring. In monoecious Ficus, female wasps push their way through a specialised entrance into receptive syconia (colloquially, “figs”), which are enclosed inflorescences. The wasps then pollinate the tree while depositing their eggs individually into ovules. Thus, each egg laid costs the tree one seed, but upon emergence, the female wasp offspring disperse that tree's pollen. Trees need to produce both wasps and seeds for the mutualism to persist, but natural selection should favour wasps that exploit the maximum number of fig ovules in the short term, resulting in a conflict of interest between wasp and tree. However, the mutualism has persisted for at least 60 million years and has radiated into more than 750 species pairs [12]. The mechanisms preventing wasps from overexploiting figs remain unknown, despite intensive study over four decades.
Within receptive syconia, the lengths of floral styles are highly variable [13,14], and ovipositing pollinators (foundresses) favour flowers with shorter styles for their offspring [15–18]. Style and pedicel lengths of flowers are negatively correlated. Short-styled ovules develop into seeds or galls (when a wasp is present) near the syconium inner cavity, while most long-styled ovules develop into seeds near the outer wall [19,20] (Figure 1). These patterns have been shown to reflect the oviposition preferences of foundresses, and are unlikely to be the result of greater elongation of pedicels containing eggs during syconial maturation, because in receptive syconia, pollinators' eggs are mainly present in short-styled inner ovules [16]. These widespread observations have been tied to four, not necessarily mutually exclusive, mechanisms that have been proposed to stabilise the fig-pollinator mutualism: (1) Unbeatable seeds—outer ovules may be defended biochemically or physically against oviposition or larval development [21]. However, no mechanism has yet been identified. (2) Short ovipositors—pollinators' ovipositors may be too short to fully penetrate the long styles of outer ovules [14,19]. However, many pollinator wasp species have ovipositors that are long enough to reach most or all ovules [18,19,22]. (3) Insufficient eggs—because pollinators disperse passively over long distances, too few foundresses may arrive to fill all ovules, and they fill inner ovules first because these are likely to be easier for oviposition [19]. Alternatively, the tree may limit the number of foundresses that enter its syconia [20]. However, in the majority of Ficus species, syconia receive enough foundresses to exploit more ovules than in fact produce wasps, leaving a large proportion of seed production unexplained [23,24]. Consequently, these three hypotheses have failed to explain mutualism stability in monoecious Ficus species, but all suggest that the key to the puzzle lies in explaining why pollinating wasps favour inner ovules for oviposition.
Recently, a fourth hypothesis, based on “optimal foraging” by ovipositing foundresses, has been proposed. Simulations have shown that the fig-pollinator mutualism can be stabilised if ovule profitability is correlated with flower style length, and if some foundresses die before laying all of their eggs [24]. The profitability of an ovule to a foundress depends on the expected offspring fitness divided by the handling time needed to lay an egg. Foundresses are thus expected to prefer short-styled flowers (inner ovules) if handling times to enable successful oviposition are lower and/or inner ovules yield higher offspring fitness than outer ovules. Therefore, the greater the relative profitability of inner ovules, the more that foundresses are likely to be selected to spend their short lives searching for them, even as inner ovules become rare. Indeed, when foundress numbers are experimentally controlled, an increased number of foundresses has been shown to result in a higher proportion of exploited inner ovules within a syconium, rather than in the total number of ovules per se [16]. Thus, the predicted consequence to a foundress in a syconium already full of exploited inner ovules is reduced fitness, with seed production in outer ovules being protected because some foundresses are likely to die before exploiting all nonpreferred outer ovules [24].
Although crucial in determining foundress behaviour, the fitness differential for wasp larvae developing in inner versus outer ovules is largely unknown. However, there is evidence that inner ovules develop into larger galls due to increased space near the syconium cavity (the lumen), resulting possibly in larger, more fecund female offspring [15]. Additionally, the fig-pollinator mutualism is ubiquitously exploited by a suite of nonpollinating wasps [10,11,25], which could also alter the relative values of ovules to foundresses. This is because many nonpollinating fig wasps are parasites (parasitoids or inquilines) that insert their ovipositors into the syconium from the outside and deposit parasite offspring that kill pollinator larvae already present in galls [10,11,25,26]. It is widely perceived that nonpollinating fig wasps compete with pollinators for inner ovules [22,27] and have negative effects on the fig by reducing the production of pollinator offspring [3,11,12,25]. However, the relative positions of parasitized and unparasitized pollinators within the same syconium have never been directly and precisely measured. If parasites are more likely to parasitize pollinator offspring in the outer layers of ovules, this will increase the fitness value of inner ovules to foundress pollinators because their offspring will have increased survivorship in the “enemy-free space” at the centre of a syconium. Thus, parasitic fig wasps could make a contribution to the maintenance of the fig-pollinator mutualism, by being one of the selection pressures that have resulted in foundresses favouring inner ovules.
If inner ovules represent enemy-free space for pollinator larvae, we would predict that externally ovipositing parasitic wasps are more likely to kill pollinator larvae in outer ovules. In the syconia of F. rubiginosa, collected from six sites in Queensland, Australia, seeds, parasites, and pollinators were spatially stratified in the same order. Inner ovules were significantly more likely to contain pollinators, and outer ovules, seeds or parasites. The ovules already exited by male wasps (a combination of pollinators and parasites) were intermediate in length between those still containing pollinators and those still containing parasites (Figure 2). While controlling for variation in parasitism rates between sites (Wald = 112.05, p < 0.001), we found that the risk of parasitism to a pollinator offspring decreased significantly, from 80% nearest the fig wall to 0% toward the centre of the syconium (β ± standard error [s.e.] = −2.21 ± 0.14, Wald = 240.16, p < 0.001; Figure 3). Ovule profitability to foundresses, measured as offspring survivorship, therefore shows a strong increase from the wall to the central cavity of a syconium, even before counting any reduced time required to oviposit in inner ovules. In addition, the overall level of parasitism decreases as syconia get larger (β ± s.e. = −0.002 ± 0.00, Wald = 33.41, p < 0.001), supporting our hypothesis that parasites are limited in their ability to reach ovules farthest from the syconium wall. Moreover, across syconia, the mean length of parasite-occupied ovules positively correlates with the mean length of those occupied by pollinators (F1,45 = 4.85, p = 0.033), independent of the effects of site (F5,45 = 17.22, p < 0.001) and syconium size (F1,45 = 6.66, p = 0.013) (longer ovules are further from the syconium wall). This variation probably reflects variation in gall and pedicel elongation during maturation and further shows that parasites consistently fail to attack inner ovules.
Over 90% of the nonpollinating wasps we found belonged to two genera, Philotrypesis and Sycoscapter [28]. Individuals of these genera develop at the expense of a pollinator offspring, either by consuming the pollinator larva directly (parasitoid) or by killing the larva and consuming the developing seed tissue (inquiline). The occurrence of parasites in the outermost ovules demonstrates the prior presence of pollinators (Figure 3), and thus, the spatial stratification of seeds and pollinators in our dataset cannot be explained by all outer ovules being “unbeatable” for pollinators [21] or by the ovipositors of pollinators being too short to reach outer ovules [14,19]. It should be noted, however, that the spatial patterns of pollinators, parasites, and seeds in our data do not eliminate the possibility that a subset of outer ovules might still be unavailable to foundresses for unknown reasons. Additionally, we ranked all ovules across our dataset for length, and then plotted the frequency of occurrence for three categories: seeds, pollinators, and parasites. Although the frequency of parasitism varies considerably in our dataset (Table 1), there is a clear negative relationship between parasite and pollinator presence, and an increase in the likelihood of seed development, as galls get shorter (Figure 4), suggesting that parasite presence contributes to the overall factors that prevent pollinators exploiting outer ovules to enable the trees to produce seeds.
Egg limitation in the pollinator, Pleistodontes imperialis, is unlikely to contribute to the stability of its mutualism with F. rubiginosa. Two foundresses carry enough eggs (mean eggs per foundress ± s.e. = 231.58 ± 12.53, N = 36) to exploit all ovules in a syconium (mean ovules per syconium ± s.e. = 373.25 ± 86.43, N = 64). The mean number of foundress bodies (mean ± s.e. = 2.58 ± 0.12, N = 203) clearly shows that the amount of wasp eggs that enter an F. rubiginosa syconium exceeds the number of ovules (D. W. Dunn, S. Al-beidh, C. Reuter, S. T. Segar, D. W. Yu, J. Ridley, and J. M. Cook, unpublished data).
The contribution parasitic wasps may make to the overall mechanisms that lead to mutualism stability across Ficus is clearly likely to vary across such a speciose and variable genus [12,22]. A comprehensive taxonomic investigation is clearly beyond the scope of this study. However, within their natural geographic ranges, the larvae of pollinators across monoecious Ficus are likely to be subject to attack by externally ovipositing parasitic wasps (e.g., [11,25]). Moreover, the syconia across Ficus species are highly variable in size. In smaller species, there may thus be physical constraints on the spatial segregation of pollinators and parasites. To test whether our parasite pressure hypothesis is likely to be restricted to the Malvanthera section within Ficus, or if syconium size is likely to constrain spatial segregation of pollinators and parasites, we studied two additional monoecious Ficus species. Ficus obliqua is a close relative of F. rubiginosa [29] but has small syconia (mean diameter at wasp emergence ± s.e. = 6.57 ± 0.34 mm, N = 16). In contrast, F. racemosa has large syconia (mean diameter at wasp emergence ± s.e. = 26.17 ± 0.75 mm, N = 14) and also belongs to a distantly related clade of figs, the vast majority of which are dioecious [30]. It therefore represents a different origin of monoecy to Malvantheran Ficus. In both of these additional species, we found similar spatial stratification of pollinators and parasites as in F. rubiginosa (Figure 5; Text S1), suggesting that a potential contribution of parasitic fig wasps to the overall factors that enable stability in the fig-pollinator mutualism is neither lineage specific, nor limited by small syconium size.
Finally, we found that pollinator body size did not correlate with ovule length (F1,55 = 0.44, p = 0.51), although there was considerable between-site variance (F5,55 = 13.19, p < 0.001). Larger syconia tended to contain larger female pollinators (F1,55 = 6.75, p = 0.012), but the relationship was curvilinear, such that the largest syconia contained smaller pollinators (second-order term: F1,55 = 6.69, p = 0.012). Thus, we found no evidence that foundresses may select inner ovules for benefits associated with producing large offspring, although there may be benefits in entering a syconium of intermediate size.
Our study is the first to show that pollinating fig wasps may gain a fitness benefit by selecting inner ovules for oviposition, because these ovules have reduced vulnerability to parasitism. The provision of ovules with high variance in profitability to foundresses clearly demonstrates that the larger, more sessile partner in the symbiosis [7], the fig tree, controls the resources available to the smaller, more mobile partner. Selection could benefit those trees producing syconia that are partially vulnerable to parasitism, via selection on the toughness and thickness of syconial walls and/or variation of floral style, and hence pedicel, lengths (Figure 1). This variance in floral morphology, and the strong likelihood of the occurrence of externally ovipositing parasitic fig wasps across monoecious Ficus, indicates a wide-ranging potential for parasitic wasps to contribute to stability in the fig-pollinator mutualism. At the smaller scale, this variable floral environment is likely to give a fitness advantage to the first foundresses to enter a receptive syconium by “providing” an abundance of safe inner ovules in which to deposit their offspring. Later foundresses, who will carry pollen of a lower value to the tree because early foundresses will have already distributed the pollen they carried, are thus effectively “penalised” for exploiting outer ovules. Our data thus show that the benefits to foundresses exploiting outer ovules are reduced by the parasitism costs to offspring, and demonstrate how a third party may select for more beneficial behaviour in a symbiont.
The potential role played by parasitic wasps may also help to resolve the evolutionary paradox posed by fig trees having generation times several orders of magnitude longer than those of their pollinators [12,22]. Presumably, a coevolutionary arms race should be resolved in favour of the pollinator, but not if a gradient in ovule profitability is produced in part by exposure to parasitic wasps, which have similar generation times to pollinators. However, the inner ovules used favourably by pollinator wasps provide an untapped resource for parasites, and one would expect strong selection for longer ovipositors in parasites to enable the exploitation of more hosts. We suggest, however, that relatively long ovipositors will have costs to the individual parasitic wasps as well as benefits. For instance, the aerodynamic influences on flight will change with a relatively long ovipositor. Likewise, the time taken to insert the ovipositor when searching for a host is likely to increase with ovipositor length, which may lead to an increased risk of predation by ants [31]. If the costs of a long ovipositor outweigh the benefits, then net selection will not favour the evolution of very long ovipositors in all parasites.
Thus, despite the short-term costs posed by parasitic wasps to the mutualists [10,21,25], parasitic wasps may also contribute to the long-term stability of the mutualism between F. rubiginosa and its pollinator P. imperialis. Moreover, we provide evidence to suggest that parasitic fig wasps have the potential to contribute positively to the overall mechanisms that enable the fig-pollinator mutualism to remain stable in other monoecious Ficus species. Although the larger partner, the fig tree, clearly controls resource availability to its pollinator, our data suggest this may be realised in part by indirectly involving parasitic wasps. Our results therefore provide another example of how a third party can shift a symbiosis towards a more mutually cooperative outcome [4,32,33]. Further studies of diverse fig species should help to confirm both the generality of parasite selection pressure and test for the presence of other mechanisms [17,22] in maintaining the fig-pollinator mutualism.
We measured both the probability of offspring mortality through parasitism, and the body sizes of female offspring, in relation to ovule position within the syconium. We used a total of 64 syconia from six populations of the Australian fig F. rubiginosa (section Malvanthera) ranging across 1,700 km of Eastern Queensland, Australia. Nine to 17 syconia were collected from a single crop from each tree. Each tree originated from a different population. Three trees (Cape Pallarenda, Castle Hill, and Mount Stuart) were from the Townsville region of northern Queensland. The other trees sampled were from Hervey's Range (50 km west of Townsville), Yungaburra (near Cairns, far north Queensland), and Brisbane (southern Queensland). All syconia were early in the male flower phase [34] with no exit holes made by male wasps. This was to ensure that female wasps had yet to emerge from their galls. Immediately after collection, all syconia were placed in 80% ethanol.
In the laboratory, each syconium was sliced into eighths lengthways. Every ovule was then systematically removed from all sections. We measured the total length of every fourth ovule (pedicel + seed or gall, excluding what remained of the style) to the nearest 0.024 mm using an eyepiece graticule attached to a binocular microscope. We did not measure the pedicel length separately for two reasons. (1) Galls or seeds at the extreme outside wall of the syconium do not have pedicels, which would result in a series of zeros in the resulting dataset and subsequent problems with data analysis. (2) In F. rubiginosa, there is no distinct landmark where the pedicel joins the gall or seed for repeatable, accurate measurements to be taken. Moreover, although galls containing wasps have been found to differ significantly in size to seeds in other species of Ficus [15], we are unaware of any significant size differences in galls inhabited by the parasite genera present in this study and galls inhabited by pollinators. The level of spatial stratification between parasites and pollinators is also so pronounced that for this pattern to be an artefact of size differences between parasite and pollinator galls, a large and obvious difference, such as that of galling wasps and pollinators, would have to be apparent.
After dissection, each ovule was assigned to one of four categories: (1) seed—ovules containing seeds; (2) exited—ovules with an exit hole made by a vacated male wasp; (3) parasite—in which the ovule contained a parasitic wasp, and (4) pollinator—ovules containing pollinating wasps (P. imperialis). We did not differentiate between the four “cryptic species” of P. imperialis because genetic data are required to distinguish them [35]. It is not possible to separate galls vacated by males into either pollinators or parasites. Both wasp types therefore contributed to the exited category.
The larval biology of most species of nonpollinating fig wasps has been divided into three major ecological groupings [11,22,28]: (1) large gall-making wasps and their parasites; (2) gall-makers of similar size to the pollinators; and (3) inquilines and parasitoids of similar size to the pollinators. Group 1 wasps are rare in F. rubiginosa but can alter development substantially by causing retention of unpollinated syconia. Their large galls are immediately obvious when a fig is opened, and we excluded the few such syconia found (<5%). The nonpollinating wasps found belonged overwhelmingly to group 3, and over 90% of individuals belonged to two common genera, Sycoscapter and Philotrypesis. The remainder (<10%) were split between another group 3 parasitic wasp (Watshamiella sp.) and a gall-maker from group 2 (Eukobelea sp.). Consequently, about 95% of all nonpollinating wasps in this study were identified as inquilines or parasitoids. Only 131 (2.23%) of the 5,866 ovules measured still contained a male parasite (N = 79) or a male pollinator (N = 52).
For an estimate of the body size of female P. imperialis, we measured the length and width of the head to the nearest 0.024 mm using an eyepiece graticule fitted to a binocular microscope. As a measure of syconium size, we took the mean of the width and length of each syconium (as measured to the nearest 0.05 mm with digital calipers) and used this to calculate the volume of a sphere.
Unless otherwise stated, we transformed all measurements to natural logarithms to normalise the error variances. We compared the mean lengths of each of the four categories of ovules, using a general linear mixed model that included ovule category and syconium volume as predictors. Site was included as a random factor.
To test our hypothesis that parasites can only gain access to pollinators in middle and outer ovule layers, we ran a general linear mixed model that used the mean length of ovules occupied by pollinators to predict the mean length of ovules occupied by parasites. Syconium volume was included as an additional covariate to control for any effects of syconium size on ovule length, and site was again included as a random factor [21].
We used a binary logistic regression to measure the relationship between ovule length and the likelihood of parasitism. Ovules containing seeds or those that had been vacated by a male wasp were excluded from the analysis. For the dependent variable, we included those ovules known to contain either a parasitic wasp (1) or a pollinator (0). Site and syconium volume were included as additional covariates to ovule length.
We estimated the head area of female pollinating wasps (length × width). To test whether pollinators were distributed nonrandomly within syconia according to their size, we used a general linear model with head area as the dependent variable, site as a random factor, and both pollinator-occupied ovule length and fig volume as covariates.
|
10.1371/journal.pntd.0006724 | Low incidence of recurrent Buruli ulcers in treated Australian patients living in an endemic region | We examined recurrent Buruli ulcer cases following treatment and assumed cure in a large cohort of Australian patients living in an endemic area. We report that while the recurrence rate was low (2.81 cases/year/1000 population), it remained similar to the estimated risk of primary infection within the general population of the endemic area (0.85–4.04 cases/year/1,000 population). The majority of recurrent lesions occurred in different regions of the body and were separated by a median time interval of 44 months. Clinical, treatment and epidemiological factors combined with whole genome sequencing of primary and recurrent isolates suggests that in most recurrent cases a re-infection was more likely as opposed to a relapse of the initial infection. Additionally, all cases occurring more than 12 months after commencement of treatment were likely re-infections. Our study provides important prognostic information for patients and their health care providers concerning the nature and risks associated with recurrent cases of Buruli ulcer in Australia.
| Mycobacterium ulcerans (M. ulcerans) causes a necrotising infection of skin and soft-tissue known as Buruli ulcer. Since the regular use of antibiotics for Buruli ulcer treatment in Australian populations was introduced at the turn of the century, treatment success rates have been very high. However there is no information from the Australian setting on the risk of recurrent disease following treatment and assumed cure, despite this being important prognostic information for patients, their families and health-care providers. Furthermore, it is also not known if recurrent disease represents late relapse of the initial treated infection or a subsequent new infection. In our study we have shown for the first time in Australian patients living in an endemic area that the incidence of recurrent Buruli ulcer following treatment and healing is low, and that this risk is similar to the estimated risk of primary infection within the general population of the endemic area. Furthermore, we have used clinical, treatment and epidemiological data supported by genomic information of M. ulcerans organisms to determine that the majority of recurrent lesions appear to result from re-infection. This suggests that for a proportion of treated patients’ acquired protective immunity against the development of recurrent M. ulcerans disease does not develop from their initial infection.
| Mycobacterium ulcerans (M. ulcerans) causes a necrotising infection of skin and soft-tissue known as Buruli ulcer.[1] Since the regular use of antibiotics for Buruli ulcer treatment in Australian populations was introduced at the turn of the century, treatment success rates have been very high.[2–4] Disease cure has assumed to occur if lesions have healed and there have been no recurrent lesions within 12 months of commencing treatment.[1,5] However, disease recurrence is known to occur.[6] At present there is no information from the Australian setting on the risk of recurrent disease following treatment and assumed cure, despite this being important prognostic information for patients, their families and health-care providers. Furthermore, it is also not known if recurrent disease represents a late relapse of the initial treated infection or a subsequent re-infection. Clarifying this issue may shed some light on the effectiveness of current treatments if recurrent lesions represent late disease relapse. On the other hand, if they represent re-infection, this may shed some light on the effectiveness of an individual’s immunity against new infections following eradication of an initial M. ulcerans infection, as well as ongoing transmission risk in the community. For the first time, whole genome sequencing has recently been used to examine this issue in four cases of recurrent M. ulcerans disease in Benin, Africa, and suggested that three of the cases represented disease relapse and one re-infection.[6]
The aim of our study was to determine the risk of recurrent M. ulcerans lesions following treatment and assumed cure in an Australian population and to use whole genome sequencing techniques combined with clinical, treatment and epidemiological data to determine whether recurrent lesions represented late disease relapse or re-infection.
All confirmed M. ulcerans cases managed at Barwon Health, a tertiary referral institution in Victoria, Australia, from 1/1/1998-31/12/2016 were included in the study. A M. ulcerans case was defined as the presence of a lesion clinically suggestive of M. ulcerans plus any of (1) a culture of M. ulcerans from the lesion, (2) a positive PCR from a swab or biopsy of the lesion, or (3) histopathology of an excised lesion showing a necrotic granulomatous ulcer with the presence of acid-fast bacilli (AFB) consistent with acute M. ulcerans infection.
Recurrence was defined as a new M. ulcerans lesion appearing after the original lesion had healed that was culture positive for M. ulcerans and occurred ≥ 12 months after initial treatment. Patients were not actively followed up after 12 months from treatment commencement, therefore diagnosis of recurrence relied upon self-presentation or referral to our health service.
A ‘significant risk’ of relapse following initial treatment was defined as a) those who had surgery without at least 2 weeks of known effective combination antibiotics based on our published risk of relapse of 32% in those who have had surgery alone,[7] and our published treatment success rates in those who have surgery combined with at least 14 days of antibiotics)[8], or b) those who had antibiotics alone but did not complete the recommended 56 days duration of known effective combination antibiotics.[9]
Where available, whole genome sequencing and single nucleotide polymorphism (SNP) analysis was performed to examine genetic relationships between pairs of isolates from the same patient (two patients did not have paired isolates available). A total of 10 isolates, derived from five patients with recurrent disease, were subjected to whole genome sequencing (Table 1). Whole genome sequencing was performed as previously described.[10] Reads were then mapped against the M. ulcerans Agy99 genome [11], including the pMUM001 plasmid [12] and core SNPs across the 10 isolates identified using Samtools. Whole genome SNP analysis was also performed on an additional six previously sequenced M. ulcerans isolates obtained from the same endemic region (Bellarine Peninsula) [13].
Data was collected prospectively using Epi-info 6 (CDC, Atlanta, GA, USA) and analysed using STATA 12 (StataCorp, College Staton, TX, USA).
This study was approved by the Barwon Health Human Research and Ethics Committee. All previously gathered human medical data were analysed in a de-identified fashion.
A total of 426 patients with M. ulcerans were managed at Barwon Health during the study period and included in the analysis. The median age was 57 years (IQR 37–73 years) and 225 (52.8%) were male. Thirty-four (8.0%) patients had diabetes and 35 (8.2%) were immune suppressed. Lesions were classified as World Health Organization (WHO) category one for 79.3%, category two for 10.6% and category three for 10.1% of lesions. The clinical type of lesion was classified as an ulcer for 85.1%, nodule for 6.1%, oedematous for 7.8% and plaque for 0.9%. The median duration of symptoms prior to diagnosis was 42 days (IQR 28–75 days).
Of this cohort, seven (1.6%) patients were diagnosed with a recurrent lesion (Table 1). This was over a combined follow-up time since commencement of treatment until the time of study analysis (12/4/18) of 2813 years, with a median follow-up time of 5.7 years (IQR 3.3–9.4 years). The rate of a recurrent lesion was 2.81 per 1000 person years (95% CI 1.19–5.22 per 1000 person years) (Fig 1). There were no significant differences in the baseline characteristics between those with a recurrence and those without a recurrence. (Table 2)
The recurrent lesions occurred a median 44 months (IQR 16–68 months) after treatment commenced for the initial lesion; 5/7 recurrences occurred at least 3.4 years from the initial lesion. Four (57%) recurrences were on a completely separate limb and side of the body, one was on the same limb but different region of that limb and 2 were on the same limb and in the same region.
Treatment of the initial lesion involved surgery alone for 1 patient, antibiotics alone for 2 patients, and antibiotics combined with surgery for 4 patients (Table 1). According to the initial treatment, 3/7 (43%) patients were assessed as having a ‘significant risk’ of relapse; patient #1 had only 37 days of combined antibiotics alone, patient #4 had excision combined with antibiotic monotherapy with clarithromycin, and patient #6 had excision alone without adjunctive antibiotics and had positive surgical margins.
Whole genome pairwise comparisons of the paired isolates revealed close genetic similarity between pairs (Fig 2). Indeed, based on our SNP analysis the paired isolates from the patients #3 (mu77/mu489) and #2 (mu327/mu432) were genetically identical (Fig 2, Table 3). In contrast, paired isolates from patients #1, #4 and #5, contained SNP differences between each pair (Fig 2, Table 3). To put this genetic variation in context, we also performed SNP analysis on an additional six unrelated human M. ulcerans isolates from the same endemic area. Three of the six isolates (mu74, muJKD8049, mu08009899) were genetically identical to each other following SNP analysis (Fig 2). Three isolates (mu146, mu_UK35 and mu487) from the paired cases were also genetically identical to these isolates demonstrating that even apparently unrelated isolates can share a common genotype. Furthermore, this genotype appeared the most dominant within the Bellarine Peninsula isolates we examined. Within two of the three pairs that contained this ‘common’ genotype (#1 and 5), the primary isolate was more genetically divergent from this ‘common’ genotype compared to the second (reoccurring) isolate. The time interval between recurrent lesions did appear to greatly influence the number of SNP differences between the isolates.
Our study has shown that Buruli ulcer has a low recurrence rate in treated Australian patients with an assumed cure living in an endemic region. This provides important prognostic information for patients and their health providers, and may help alleviate the often substantial fears that patients have of becoming reinfected once their initial lesion has been cured. Although the low risk is reassuring, the fact that it can occur means that patients and clinical staff need to be educated and aware of this possibility, so that any recurrent lesions are assessed and diagnosed early when lesions are small, enabling less complex treatment with better outcomes [5]. It is also important to recognise that recurrent lesions can occur many years later and commonly occur on completely different regions of the body compared to the initial lesion. In our study we did not detect an increased risk of recurrent lesions associated with patient characteristics which included age, gender, WHO category and type of lesion, diabetes, immune suppression and the duration of symptoms prior to diagnosis. Although we did not examine host genetics, previous studies have identified genetic factors associated with increased susceptibility to M. ulcerans that may influence the risk of recurrent disease. [14,15] We would suggest future studies be performed to assess whether host genetics can predict those at risk of recurrences, or whether this is more likely determined by the intensity of re-exposure.
The whole genome sequence analysis revealed a mix of genetic relationships between isolates. Paired isolates from some patients (#2 and #3) were genetically identical, possibly suggesting either late relapse of the initial infection or re-infection from a genetically homogenous source. In the case of patient #3, the extended time between recurrence (46 months), the fact that the patient received highly effective treatment, and the fact that the lesions were identified in different body areas (right forearm and left ankle), suggests that re-infection from a genetically homogenous source was more likely. While it’s hard to estimate the degree of genetic change that would occur during a latency period in vivo, we assume that some mutations would occur with longer periods (particularly 46 months). In contrast, the isolates from patient #2 –also genetically identical–were only separated by 12 months, and occurred on the same body region. In this case, a late relapse of the initial infection would appear more likely.
There were genetic differences between three of the paired isolates (patients #1, #4, and #5) which can be interpreted in two ways. Firstly, it’s possible that they are the result of re-infection from a genetically heterogeneous population. In support of this hypothesis, our previous research examining family clusters of M. ulcerans cases in Australia suggests that exposure risk to M. ulcerans is short-term and may not necessarily be from a genetically homogeneous source [10]. However, given that M. ulcerans is highly clonal in Australia, with only minor genetic variation [13,16], it is expected that some re-infection cases will also be from genetically identical sources. The case of patient #3, discussed above, would be an example here. The second possible explanation is that the bacterium genetically evolves during its latency period in vivo and thus the cases represent late disease relapse despite a small number of SNP differences. In the case of patients #1 and #5 this latter hypothesis cannot be ruled out, but seems unlikely as in both cases the primary (first) isolate had genetically diverged more from the ‘common’ dominant genotype compared to the second isolate. This is further supported in patient #1 by the long duration between lesions (44 months) and in patient #5 by the recurrent lesion being situated on a completely different body area and the initial treatment being highly effective for curing BU. Combined, these findings suggests that re-infection with a different genotype was the most plausible explanation for the #1 and #5 cases.
In comparison with the other known study by Eddyani et al. from Africa [6] that looked at recurrent BU cases post treatment between 1989 and 2010 using whole genome sequencing, their recurrence rate (100/4951 cases; 2.0%) was similar to ours (1.6%). However this study included recurrent lesions occurring from 6 months following treatment meaning their recurrence rate according to our definition (≥ 12 months) would have been lower. With information from clinical, treatment and epidemiological data supported by whole genome sequencing, 80% of our cases were classified as re-infection whereas 75% of their cases were classified as relapse. In the African study, none of the three cases classified as relapse received effective antibiotics against M. ulcerans, putting them at higher risk of relapse [7], and in 2 of the three cases the isolates were genetically identical. The third relapse isolate differed by only 1 SNP and occurred on the same body region within a short time interval (9.5 months). In their single case classified as re-infection, the second lesion was on a separate body area and the isolate had a 20 SNP difference compared to the original one. Thus their interpretations were similar to ours whereby the one case we classified as relapse (#2) had a genetically identical isolate on the same region of the body within a short time interval (12 months), whereas those classified as re-infection had a combination of either being genetically distinct isolates (#1,4,5), on separate body areas (#3,4,5), having had highly effective treatment (#3 and 5) or having a long time interval between cases (#1,3,4). From both studies it is evident that whole genome sequencing can be a useful tool in helping to clarify the likelihood of BU relapse versus re-infection post treatment, as has been the case with tuberculosis [17].
The two recurrent cases who did not have paired isolates available for WGS (#6 and 7) were classified as re-infections based on a combination of separate body areas (#6 and 7), highly effective treatment (#7) and having a long time interval between cases (#6 and 7). Additionally, our data suggesting that all recurrent cases which occurred more than 12 months after treatment commenced were classified as re-infections, and the only one occurring after 12 months was classified as disease relapse, would support our previous clinical definitions that treatment failure occurs when a recurrent lesion appears within 12 months of commencing treatment[5].
If, as suggested by our study, most recurrent cases result from re-infection, then at least for a proportion of treated patients acquired protective immunity against the development of recurrent M. ulcerans disease does not develop following an initial infection. Interestingly, the rate of recurrence (2.81 cases/year/1000 population) was similar to the estimated risk of infection in the general population of the Bellarine Peninsula (0.85–4.04 cases/year/1,000 population)[18], suggesting that there may be no significant risk reduction against future infection for previously treated patients. This is in contrast to a study from Uganda in the 1970s which suggested an 88% protective effect over 4 years against recurrent M. ulcerans disease in those with a prior history of the disease.[19]
A limitation of our study is that we relied on self-presentation or referral to our health service for diagnosis of recurrent lesions more than 12 months after treatment commenced and therefore there is a risk that some recurrent lesions were not captured in our study. However, as we are the only specialised health service in our region managing M. ulcerans it is likely that any recurrent lesions in patients would have be managed at Barwon Health and therefore we feel the risk of missing recurrent lesions would be small. Additionally, as the incidence of M. ulcerans in the Bellarine Peninsula has fallen in recent years,[20] if this reduction relates to reduced environmental pressure for infection we may have underestimated the risk of recurrent lesions that would occur if the pressure had remained constant. It is also recognised that the number of recurrent cases where isolates had WGS performed was small meaning our results need to be interpreted with some caution. Further research involving WGS of more isolates from recurrent cases should be performed to further validate these findings.
There is a low incidence of recurrent Buruli ulcer in treated Australian patients living in endemic regions and the risk is similar to the estimated risk of primary infection within the general population of the endemic area. The majority of recurrent lesions appear to result from re-infection suggesting that for a proportion of treated patients lifelong immunity against M. ulcerans re-infection does not develop.
|
10.1371/journal.ppat.1003423 | Rapid Perturbation in Viremia Levels Drives Increases in Functional Avidity of HIV-specific CD8 T Cells | The factors determining the functional avidity and its relationship with the broad heterogeneity of antiviral T cell responses remain partially understood. We investigated HIV-specific CD8 T cell responses in 85 patients with primary HIV infection (PHI) or chronic (progressive and non-progressive) infection. The functional avidity of HIV-specific CD8 T cells was not different between patients with progressive and non-progressive chronic infection. However, it was significantly lower in PHI patients at the time of diagnosis of acute infection and after control of virus replication following one year of successful antiretroviral therapy. High-avidity HIV-specific CD8 T cells expressed lower levels of CD27 and CD28 and were enriched in cells with an exhausted phenotype, i.e. co-expressing PD-1/2B4/CD160. Of note, a significant increase in the functional avidity of HIV-specific CD8 T cells occurred in early-treated PHI patients experiencing a virus rebound after spontaneous treatment interruption. This increase in functional avidity was associated with the accumulation of PD-1/2B4/CD160 positive cells, loss of polyfunctionality and increased TCR renewal. The increased TCR renewal may provide the mechanistic basis for the generation of high-avidity HIV-specific CD8 T cells. These results provide insights on the relationships between functional avidity, viremia, T-cell exhaustion and TCR renewal of antiviral CD8 T cell responses.
| CD8 T cells directed against virus are complex and functionally heterogeneous. One relevant component of CD8 T cells is their functional avidity which reflects their sensitivity to cognate antigens, i.e. how prone T cells are to respond when they encounter low doses of antigens. In patients with chronic and established HIV infection, we observed that the sensitivity of HIV-specific CD8 T cells was not different between patients with progressive or non-progressive disease. In contrast, the sensitivity of HIV-specific CD8 T cells was significantly lower in patients with early and recent HIV infection. Furthermore, CD8 T cells of high avidity were preferentially associated with a state of functional impairment known as exhaustion. Of interest, some patients treated with antiretroviral therapy during acute infection spontaneously interrupted their treatment and experienced a rebound of virus. In these patients, the avidity of HIV-specific CD8 T cells increased and this increase was associated to stronger cell exhaustion and greater renewal of the population of antiviral CD8 T cells, thus potentially providing the mechanistic basis for the generation of high-avidity CD8 T cells. Overall, our data suggest that rapid perturbation in viremia levels drove increases in the functional avidity of HIV-specific CD8 T cells.
| CD8 T cells play a critical role in antiviral immunity and a large number of studies in both human and murine models indicate that virus-specific CD8 T cells are directly involved in the control of virus replication and disease progression [1], [2], [3], [4], [5], [6], [7].
Functional avidity of T cells, also defined as antigen (Ag) sensitivity, is thought to be a critical component of antiviral immunity. Functional avidity reflects the ability of T cells to respond to a low Ag dose and is determined by the threshold of Ag responsiveness. There is a general consensus that high functional avidity CD8 T-cell responses are of higher efficacy against cancers [8] and acute virus infections [9]. However, their relevance in chronic persistent virus infections and established tumors [10], [11], [12] remains to be determined since conflicting results were obtained in these contexts [13], [14] as well as in HIV infection [15], [16], [17], [18], [19]. HIV-specific CD8 T-cell responses in non-progressive infection were associated with high avidity and superior variants recognition [11], [12], [20], [21], whereas other studies indicated similar avidity between patients with progressive and non-progressive chronic infection [16], [18], [19], [22], [23]. In this regard, we have previously shown that polyfunctional virus-specific CD8 T-cell responses during chronic virus infections were predominantly of low functional avidity [24]. Furthermore, it is also well established that high functional avidity T-cell responses preferentially led to viral escape and T-cell clonal exhaustion [10], [24], [25], [26].
However, the factors determining the level of T-cell functional avidity and its relationship with the phenotypic and functional heterogeneity of T-cell responses are only partially understood [15], [16], [17], [18], [19], [22].
Functional avidity is based on the ability of T cells to respond following stimulation with a cognate Ag and it is well established that responding CD8 T cells are clonally heterogeneous (i.e. oligoclonal) [27], [28], [29], [30]. Therefore, the clonotypic composition of the responding T-cell population (and its TCR diversity) can influence functional avidity [27], [28]. Indeed, we and others reported that HIV-specific CD8 T cells responding to various epitopes harbor a diverse TCR repertoire in chronically-infected patients [31], [32], [33].
HIV-specific CD8 T cells in primary HIV infection (PHI) are temporally associated with the initial control of viremia [1]. Lichterfeld and colleagues suggested that high-avidity HIV-specific CD8 T-cell responses are present during early infection (defined as HIV seroconversion within 6 months) and are then preferentially lost overtime [33].
In the present study, we have performed a comprehensive cross-sectional characterization of HIV-specific CD8 T-cell responses in patients with PHI or chronic (progressive and non-progressive) HIV infection in both steady-state conditions as well as following virus rebound. The primary observations of the present study indicate that a) the functional avidity of HIV-specific CD8 T cells is not different between patients with progressive and non-progressive chronic infection, b) the functional avidity of HIV-specific CD8 T cells is significantly lower in PHI patients as compared to patients with chronic infections, c) increased functional avidity is associated with T-cell exhaustion and lack of expression of markers of co-stimulation, and d) great increase in functional avidity is observed after virus rebound following spontaneous interruption of antiretroviral therapy and is associated with increased TCR renewal.
We recruited 85 HIV-infected patients and performed a cross-sectional analysis of the functional avidity of HIV-specific CD8 T-cell responses. The distinct groups included a) 37 patients with very early stage of acute infection (i.e. prior to seroconversion and incomplete western blot; hereafter referred to as PHI), 39 patients with progressive chronic infection (i.e. typical progressors; hereafter referred to as CP) and 9 patients with non-progressive chronic infection (i.e. LTNP) (Table S1). We first investigated the 115 HIV-specific CD8 T-cell responses obtained in 26 untreated PHI (PHI-B), 19 untreated CP (CP-B) patients and 9 LTNP (Fig. 1A–B). As described in the Methods, blood mononuclear cells were stimulated with decreasing concentrations of the cognate peptides and the peptide dose able to induce half of the maximal response (i.e. effect concentration 50%; EC50) was determined (Fig. 1A).
The results of this analysis indicated that the functional avidity of HIV-specific CD8 T cells was lower in PHI-B as compared to CP-B or LTNP patients (both P<0.0001; Fig. 1A–B) while there was no difference between CP-B and LTNP (Fig. 1A–B). However, there were no significant difference in the magnitudes of HIV-specific CD8 T-cell responses among all groups (Fig. 1C) and no significant association between the functional avidity and the magnitude of HIV-specific CD8 T-cell responses (Fig. S1A). Furthermore, the differences in functional avidity of HIV-specific CD8 T cells between the different cohorts were not influenced by distinct peptides/MHC class I associations, since these differences remained significant when common epitopes (i.e. epitopes recognized by patients from distinct cohorts) were analyzed (Fig. 1D). Of note, the B*2705-KRWIILGLNK (i.e. KK10) epitope has been previously reported as a protective epitope [28], [34], [35] and it was one of the common epitopes recognized by CP-B and LTNP patients. While the functional avidity of B*2705-KK10-specific CD8 T-cell responses in CP-B and LTNP patients was almost identical, it was rather low as compared to the other HIV-specific CD8 T-cell responses from both groups (Fig. 1D).
Taken together, these results indicate a lack of association between the functional avidity of HIV-specific CD8 T cells and virus control consistently with the recent study from Chen and colleagues [22]. Furthermore, the HIV-specific CD8 T-cell responses in acute infection have lower functional avidity than in chronic infection.
It has been previously reported that high functional avidity HIV-specific CD8 T-cell responses are selectively deleted early after acute HIV infection [33]. We addressed this issue by repeating the epitopes mapping in patients with acute infection after one year of antiretroviral therapy (ART) (PHI-T1Y). Forty-five HIV-specific CD8 T-cell responses were identified in PHI-B patients using ICS. Among these 45 responses, 38 (85%) remained detectable after one year of ART whereas 7 (15%) became undetectable. Interestingly, at the time of acute infection, these 7 responses were already of lower magnitude as compared to the 38 responses which remained detectable (P = 0.03; Fig. 2A). Furthermore, the functional avidity of HIV-specific CD8 T-cell responses at the time of acute infection was not different between the 7 lost and the 38 remaining responses (P>0.05; Fig. 2A).
These results indicate that the minor proportion of HIV-specific CD8 T-cell responses selectively lost after acute infection did not have higher functional avidity.
It has been suggested that Ag load may influence the responsiveness of HIV-specific CD8 T cells [36]. To address this issue, we assessed whether the functional avidity of HIV-specific CD8 T-cell responses would change after control of virus replication, i.e. after 1 year of successful ART, in patients with acute or chronic infection. Of note, 46 additional HIV-specific CD8 T-cell responses were considered; 17 responses were identified in the initial 26 PHI patients following re-mapping after 1 year of ART and 29 responses were identified in 11 additional PHI patients only mapped after 1 year of ART. Both magnitude and functional avidity of HIV-specific CD8 T-cell responses generated during ART were similar to those measured at baseline (Fig. 2A). Furthermore, PHI patients were treated either with ART alone or with ART+CyclosporinA (CsA) but CSA treatment had no significant impact on the magnitude or the functional avidity of HIV-specific CD8 T-cell responses (Fig. S2).
The functional avidity of the same HIV-specific CD8 T-cell responses measured longitudinally either prior to ART or after 1 year of ART remained stable in both PHI and CP patients (both P>0.05; Fig. 2B). Furthermore, the lack of significant effect of ART on the functional avidity of HIV-specific CD8 T cells was also confirmed in non-longitudinal, independent, T-cell responses from PHI or CP patients (both P>0.05; Fig. 2C). Therefore, HIV-specific CD8 T-cell responses remained of lower avidity (P = 0.0003) in PHI-T1Y as compared to CP-T1Y patients (Fig. 2D). Consistently with the above-mentioned analyses performed in the untreated groups, differences in functional avidity of HIV-specific CD8 T cells between PHI-T1Y and CP-T1Y were not related to distinct peptide-MHC class I associations since the differences remained significant also when common epitopes were considered (P = 0.0003; Fig. 2E).
All together, these observations indicate that even after control of virus replication HIV-specific CD8 T-cell responses from patients with acute HIV infection remain of lower avidity as compared to patients with chronic infection.
We then assessed the functional profile of HIV-specific CD8 T cells from PHI-B, CP-B and LTNP patients. Although, the magnitudes of HIV-specific CD8 T-cell responses were not significantly different between PHI-B, CP-B and LTNP (Fig. 1C), perforin expression was significantly (P≤0.001) higher in HIV-specific CD8 T cells from PHI-B patients as compared to CP-B or LTNP (Fig. S1B and 3A). As previously shown [37], [38], HIV-specific CD8 T cells from LTNP contained more IL-2, whereas those from CP-B patients were mostly composed of single IFN-γ-producing cells (both P<0.0001; Fig. S1B and 3A).
We then performed a phenotypic characterization of HIV-specific CD8 T-cell responses and monitored CD27 and CD28 expression to assess co-stimulation and PD-1, 2B4 and CD160 expression to assess T-cell activation and exhaustion. For these analyses, only HIV-specific CD8 T cells detectable using cognate peptide-MHC class I multimers (Table S2) were taken into consideration. Regarding T-cell co-stimulation, HIV-specific CD8 T cells from PHI-B expressed a higher proportion of CD27+CD28+ cells than those from CP-B (P = 0.02) or LTNP (P = 0.003) patients (Fig. S1C and 3B). Also, analyses of the expression of co-inhibitory receptors indicated that HIV-specific CD8 T cells from CP-B and LTNP were both composed of significantly higher proportions of PD-1+2B4+CD160+ (P<0.006 and P<0.025, respectively) or PD-1−2B4+CD160+ (P<0.025 and P<0.0001, respectively) as compared to PHI-B (Fig. S1D and 3C). HIV-specific CD8 T cells from PHI-B mostly (about 70%) lacked all three markers or expressed 2B4 alone (all P<0.006; Fig. 3C) and expressed lower frequency and intensity of PD-1 as compared to CP-B (both P<0.002; data not shown).
These data suggest that HIV-specific CD8 T cells from patients with acute and chronic infection are functionally and phenotypically distinct.
We then assessed the association between functional avidity and the expression of co-stimulatory or co-inhibitory receptors. The functional avidity of HIV-specific CD8 T cells was negatively correlated to the proportion of CD27+CD28+ cells (P = 0.01; Fig. 4A) and directly correlated to the proportion of cells co-expressing PD-1/2B4/CD160 (P = 0.005; Fig. 4B).
Furthermore, we performed correlations and rank correlation's matrix to explore the partial associations of variables and to assess the dependency and potential hidden effect of confounding variables in pairs associations. These analyses indicated that the proportions of CD27+CD28+ and of PD-1+2B4+CD160+ HIV-specific CD8 T cells were not significantly dependent on each other. This allowed us to perform a regression model analysis and to postulate that the functional avidity of HIV-specific CD8 T cells may be a linear function of the two aforementioned explained variables (after log10 transformation). Interestingly, this regression analysis indicated that about 28% of the functional avidity of HIV-specific CD8 T cells was explained by a combination of the proportion of CD27+CD28+ and of PD-1+2B4+CD160+ CD8 T cells (P = 0.0013; data not shown). We did not, in contrast, observe any significant correlation between the functional avidity of HIV-specific CD8 T cells and their functional profile.
Overall, these observations indicate that high-avidity HIV-specific CD8 T-cell responses are preferentially composed of cells lacking the expression of co-stimulatory molecules but co-expressing high levels of co-inhibitory receptors. However, the functional avidity can only be partially predicted from the expression of co-stimulatory or co-inhibitory molecules.
We then performed a longitudinal analysis to investigate the effect of changes in viremia levels on HIV-specific CD8 T cells. To address this issue, we longitudinally monitored HIV-specific CD8 T cells in two distinct models: a) in conditions of viremia below the limit of detection, i.e. viremia <50 HIV RNA copies/ml of plasma (in patients successfully treated by ART) and b) in conditions of rapid and major changes in viremia occurring in patients experiencing virus rebound following spontaneous treatment interruption (TI) (Fig. 5A). In particular, we evaluated HIV-specific CD8 T-cell responses in PHI-T1Y and compared them with those after 5 years (PHI-T5Y) of uninterrupted successful ART or after TI (PHI-ATI) (Fig. 5A). Nine out of the 37 patients identified during acute infection spontaneously interrupted ART. These patients were treated since PHI for ≥1 year (mean±SE 131±15 weeks) and all had undetectable viremia (<50 HIV RNA copies/ml) at the time of TI. After TI, all patients experienced a virus rebound with an average plasma viremia of 5.18 log10 HIV RNA copies/ml.
The functional profile of HIV-specific CD8 T-cell responses at the time of TI was different from that of baseline. HIV-specific CD8 T cells in PHI-T1Y were mostly polyfunctional (associated to a large fraction of IL-2-producing cells and little perforin) (P<0.0001; Fig. 5B–C) as compared to the typical effector profile (Fig. 3A) observed in PHI-B. In patients remaining on ART, HIV-specific CD8 T cells became more polyfunctional (i.e. further shifted toward IL-2 production) after 5 years of ART as compared to 1 year of ART (Fig. 5B–C). Conversely, in patients interrupting ART, as shown for patient #1023 who interrupted ART after two years of treatment and experienced a virus rebound of 122'000 HIV RNA copies/ml, the proportion of HIV-specific CD8 T cells co-producing IFN-γ and IL-2 decreased (Fig. 5A–B). Cumulative analyses confirmed the significant (P<0.01) decrease in polyfunctionality of HIV-specific CD8 T-cell responses after TI and the significant (P = 0.03) increase in polyfunctionality of HIV-specific CD8 T-cell responses from patients who remained on ART (Fig. 5C).
Then we determined PD-1 (as well as 2B4 and CD160) expression in a subset of PHI-ATI and PHI-T5Y patients with known HIV-specific CD8 T-cell responses using cognate peptide-MHC class I multimers (Table S2). As shown in the representative flow cytometry profiles from patients #1017 and #1023, PD-1 expression increased in patient #1023 who interrupted ART but not in patient #1017 who remained on ART (Fig. 5D). Along the same line, the proportion of triple PD-1+2B4+CD160+ HIV-specific CD8 T cells also increased in patient #1023 but not in patient #1017 (Fig. 5F). Cumulative analyses confirmed that PD-1 expression as well as the proportion of cells co-expressing PD-1/2B4/CD160 in HIV-specific CD8 T cells were significantly increased in patients who interrupted ART (both P = 0.03; Fig. 5E and 5G). No increase in PD-1 expression or in the co-expression of PD-1/2B4/CD160, however, was observed in the patients who remained on ART (both P>0.05; Fig. 5E and G). Finally, consistently with the differences in the functional profile of HIV-specific CD8 T-cell responses between patients who did or did not interrupt ART (Fig. 5B–C), we observed that the proportion of dual IFN-γ/IL-2-producing HIV-specific CD8 T cells was negatively correlated with the proportion of cells co-expressing PD-1/2B4/CD160 (P = 0.009; data not shown).
These data indicate that major changes of viremia levels in TI patients caused reduction of polyfunctional HIV-specific CD8 T cells and were associated with an increased level of exhaustion.
We then analyzed the effects of virus rebound following TI on the functional avidity of HIV-specific CD8 T cells.
As shown for patient #1023, the functional avidity of B*0701-GPGHKARVL- and A*0301-RLRPGGKKK-specific CD8 T-cell responses significantly increased after TI (ATI) as compared to pre-TI (PTI) (Fig. 6A). Furthermore, an additional HIV-specific CD8 T-cell response against B*0701-IPRRIRQGL, which was below the detection level PTI, was observed ATI (Fig. 6A). Cumulative analyses confirmed the increase in functional avidity of HIV-specific CD8 T cells occurring ATI (P = 0.007; Fig. 6B) and also indicated that new responses generated following virus rebound were of high avidity (P = 0.04; Fig. 6B). Consistently, no significant (P>0.05) differences in functional avidity were observed during the same period when a similar analysis was performed in HIV-specific CD8 T-cell responses from patients who did not interrupt ART (i.e. for an average of 4 years; Fig. 6B). Furthermore, the functional avidity of HIV-specific CD8 T-cell responses ATI was in the same range as compared to those observed in CP patients (data not shown). Of note, consistently with the increase in the co-expression of co-inhibitory molecules occurring in patients experiencing virus rebound (Fig. 5F–G), we observed a positive association (P = 0.02) between the fold increase in functional avidity of HIV-specific CD8 T cells and the fold increase in the proportion of PD-1+2B4+CD160+ HIV-specific CD8 T cells (Fig. 6C).
Of interest, we performed a comprehensive statistical modeling of the changes in functional avidity of HIV-specific CD8 T cells and used mixed-effect linear models [39], [40] to assess the evolution of functional avidity as a function of time and virus rebound.
For this analysis, all longitudinal measures (n = 231) of functional avidity were included. The statistical model revealed that the interaction between avidity and time was not significant in steady-state conditions, i.e. neither in patients on ART (Fig. 6D; red line), nor in patients off ART (Fig. 6D; green line). In both conditions, an increase of 0.013 units per month was determined but did not reach statistical significance (P = 0.05; Fig. 6D), thus indicating that functional avidity does not significantly change under steady-state circumstances. However, we found a significant (P = 0.013) interaction between functional avidity and virus rebound. An immediate increase of functional avidity of HIV-specific CD8 T cells of about 1 order of magnitude (0.95 units) occurred directly after TI (Fig. 6D, grey dashed lines) and was not related to the duration of ART prior to TI.
These observations indicated that the functional avidity of HIV-specific CD8 T cells is stable overtime in steady-state conditions regardless of viremia levels, but does increase after rapid increase in viremia levels associated with virus rebound.
We recently demonstrated that the global CD8 TCR repertoire of virus-specific CD8 T cells was diverse and subjected to continuous renewal [32]. We then evaluated the TCR repertoire in PHI patients experiencing a virus rebound following TI. For this purpose, we measured CDR3 diversity and the percentage of renewal of HIV-specific CD8 T cells and compared those to the changes in functional avidity of HIV-specific CD8 T cells occurring before and after virus rebound.
As shown for patient #1023 (Fig. 5A), TRBV usage and CDR3 size pattern were analyzed for B*0701-GPGHKARVL-specific CD8 T cells at week (W) 18, W96 and W125 (Fig. S3A–B). Using our previously-described model to determine CDR3 renewal [32], we calculated a renewal of 76% between W18 and W96 (i.e. on ART) and a renewal of 82% between W96 and W125 (i.e. after TI). Cumulative analyses confirmed a significantly (P = 0.008) higher CDR3 renewal of HIV-specific CD8 T cells after virus rebound than in steady-state condition, i.e. during treatment (Fig. 7A). Interestingly, the level of CDR3 renewal was directly associated (P = 0.036) with the extent of increase in functional avidity of HIV-specific CD8 T cells (Fig. 7B).
Taken together, these observations suggest that increase in CDR3 renewal may contribute to the increase in functional avidity of HIV-specific CD8 T cells occurring after virus rebound.
T-cell functional avidity reflects the ability of T cells to respond to various concentrations of Ag and may be assessed ex vivo through a quantification of a biological function such as IFN-γ production, cytotoxic activity or proliferation capacity. Several parameters concur to determine the threshold of T-cell responsiveness. These include: a) the affinity of the TCR for the peptide-MHC (pMHC) molecule, i.e. the strength of the interaction between the TCR and pMHC [41], [42], b) the density of pMHC-TCR interactions (reflecting both the amount of Ag and the ability of Ag presenting cells (APC) to present Ags) [43], [44], [45], [46], c) the expression of co-stimulatory and co-inhibitory molecules by T cells and APC [47], and d) the T-cell distribution and composition of signaling molecules [44], [48]. However, the factors determining functional avidity and the relationship between functional avidity and the heterogeneity of T-cell responses are not well understood.
In the present study, we comprehensively investigated the functional avidity of HIV-specific CD8 T-cell responses in a cross-sectional study of different cohorts of HIV-infected patients. The evaluation of the functional avidity of HIV-specific CD8 T cells was based on optimal epitopes, i.e. epitopes not necessarily corresponding to the autologous virus sequences. Since pMHC/TCR affinity is one of the parameters potentially influencing the functional avidity [49], [50], a mismatch between the epitope sequences and the TCR or the MHC may impact the determination of avidity. However, the same strategy was used throughout all cohorts of HIV-infected patients, thus minimizing the potential biases in our observations.
HIV-specific CD8 T cells generated during acute infection were of lower functional avidity as compared to those from patients with chronic progressive or non-progressive infection. These differences were not biased by distinct peptide-HLA associations and remained significant after ART-induced control of virus replication. In addition, a preferential deletion of HIV-specific CD8 T-cell responses of higher avidity was not observed, as previously described in a cohort of early HIV infection [33]. The discrepancy between our and the previous study may be explained by differences in the individual cohorts as well as by the fact that all the 37 patients received ART at the time of diagnosis of PHI in our study, whereas only 5 of the 10 patients in Lichterfeld's study received ART [33]. Our results also indicated that the minor proportion of HIV-specific CD8 T-cell responses lost after acute infection had an initial lower magnitude rather than a higher avidity.
Furthermore, consistently with previous studies [16], [18], [19], [22], [23], [51], there were no significant differences in the functional avidity of HIV-specific CD8 T-cell responses from chronic progressive and non-progressive infection. These observations suggest that T-cell functional avidity does not represent a correlate of virus control, at least in the context of chronic and persistent virus infections. Along the same line, HIV-specific CD8 T-cell responses commonly associated with virus control [52], i.e. HLA-B*27-, B*57- or B*5801-restricted T-cell responses, were consistently found in the lower range of functional avidity (data not shown). These observations do not support previous studies showing a relationship between higher avidity T-cell responses and better virus control [11], [12], [20], [21], [53], [54]. One potential explanation is that in most of these studies, specific T-cell epitopes (e.g. TW10 or KK10) were considered predominantly in individuals with non-progressive infection. Of note, consistently with our study, Chen and colleagues recently demonstrated that KK10-specific CD8 T-cell responses in elite controllers showed better virus control and broader viral recognition but similar functional avidity as compared to progressors [22]. They also confirmed the overall lack of difference in the functional avidity of HIV-specific CD8 T cells between patients with progressive and non-progressive infection [22].
Taken together, these observations suggest an association between higher avidity T-cell responses and chronic HIV infection.
Of note, we also assessed the relationship between T-cell functional avidity and the expression of markers of exhaustion. It is important to underscore that HIV-specific CD8 T cells which had higher avidity in chronic infection expressed also higher levels of exhaustion markers. Therefore, these results further indicate that higher functional avidity does not correlate with better virus control but rather with the status of cells activation/exhaustion.
We also determined the impact of the expression of costimulation (i.e. CD27 and CD28) and exhaustion markers (i.e. PD-1, CD160 and 2B4) on the levels of functional avidity in HIV-specific CD8 T cells using a regression model. The regression model indicated that the expression of the above markers only partially accounts for the establishment of the functional avidity of HIV-specific CD8 T cells, thus indicating that additional factors may contribute to determine the levels of functional avidity.
The lower avidity of HIV-specific CD8 T cells in PHI patients may also be potentially explained by the fact that patients were identified very early in the course of infection and received ART within 24 hours. Therefore, one cannot exclude the possibility that this early control of virus replication blunted the natural evolution and maturation of the immune response, as previously shown for T- and B-cell responses [55], [56], [57]. Consistently, it was also shown in mice that functional avidity of antiviral CD8 T cells continuously increased (avidity maturation) during the first month of infection [58].
In this regard, when patients treated during PHI experienced a virus rebound, the functional avidity of HIV-specific CD8 T-cell responses significantly increased. The mixed-effect linear model we used indicated a punctual increase of about one order of magnitude following virus rebound. However, there was no quantitative correlation between either the peak or the steady-state of the virus rebound and the increase in avidity.
Several mechanisms were proposed to modulate T-cell functional avidity maturation including: 1) the formation of clusters comprising several TCRs and other molecules able to reinforce the immunological synapses [59], [60], [61], 2) the optimization of the signal transduction machinery such as an increase in the amount of and in the basal phosphorylation levels of signaling molecules [62], [63] and 3) a selective expansion of high TCR avidity clones and/or the loss of clones with low TCR avidity [33], [64], [65], [66], [67]. We cannot exclude that the same mechanisms may also contribute to explain the increase in avidity observed following treatment interruption and virus rebound.
Interestingly, we showed that TCR renewal was also significantly higher following virus rebound and associated with an increase in T-cell functional avidity. Therefore, our data indicate a potential role of TCR renewal in the modulation of the levels of functional avidity. However, our results do not distinguish between the recruitment of new clones, selective expansion of pre-existing high-avidity clones or depletion of low-avidity clones since the study was performed at the population level.
Taken together, these results support the following model.
HIV-specific CD8 T cells of lower functional avidity are generated during primary immune responses; then, persistence of detectable viremia drives an increase in functional avidity as supported by the major increase in functional avidity associated with the sudden increment in viremia levels; the increase in viremia levels is also associated with massive TCR renewal which, in turn, causes the generation/selection of T-cell clones with higher functional avidity.
These results provide insights on the relationships between functional avidity, viremia, T-cell exhaustion and TCR renewal of antiviral CD8 T-cell responses.
These studies were approved by the Institutional Review Board of the Centre Hospitalier Universitaire Vaudois and all subjects gave written informed consent.
Seventy-six patients with primary (PHI) or progressive chronic (CP) HIV infection were enrolled. Diagnosis of PHI included the presence of an acute clinical syndrome, a negative HIV antibody test, a positive test for HIV RNA in plasma, and ≤3 positive bands in a Western blot. All PHI patients started ART alone or ART+CsA within 72 h as described [68] and were followed for up to 10 years. Patients with chronic progressive (CP) HIV infection were infected for more than a year, were ART-naïve at the time of inclusion, had ≥400 CD4 T-cells/µl, ≥5000 plasma HIV RNA copies/ml and were directly treated with ART upon diagnosis as described [69], [70]. Four CP patients were investigated both prior to (BSL) and then after 1 year of ART (T1Y). Furthermore, 9 additional HIV-infected patients with non-progressive disease, i.e. LTNP, as defined by documented HIV infection since >10 years, stable CD4 T-cell counts >500 cells/µl, and plasma viremia <500 HIV RNA copies/ml were also included. Clinical and virological characteristics of the different cohorts are detailed in Table S1.
Epitope mapping was performed using a panel of 192 HPLC-purified (>80% purity) previously-described optimal epitopes [71]. Confirmation of specificity was achieved based on the HLA class I genotype of the patients and ICS assays. Peptide-MHC class I multimers (listed in Table S2) were purchased from ProImmune (Oxford, UK) except HLA-B*0801-RAKFKQLL, HLA-B*0702-RPPIFIRRL, HLA-B*0702-TPRVTGGGAM and HLA-B*5701-TSTLQEQIGW (Table S2) which were produced as described [72].
The following antibodies were used in different combinations. CD8-PB, CD8-APCH7, CD3-APCH7, CD45RA-PECy5, PD-1-FITC, IFN-γ-APC, TNF-α-PECy7, and IL-2-PE were purchased from Becton Dickinson (BD, San Diego, CA), CD4-ECD, CD3-ECD, CD28-ECD, CD27-APC from Beckman Coulter (Fullerton, CA, USA), Perforin-FITC (clone B-D48) from Diaclone (Besançon, France), CCR7-FITC from R&D Systems (Minneapolis, MN, USA), 2B4-PECy5.5 and CD160-APC from Biolegend (San Diego, CA, USA).
ELISPOT assays were performed as per the manufacturer's instructions (BD Biosciences). In brief, 2×105 cryo-preserved blood mononuclear cells were stimulated with 1 µg of single peptide or peptide pools in triplicate conditions as described [73]. Media only and staphylococcal enterotoxin B (SEB) were used as negative and positive controls, respectively. Thresholds for assay validation and positivity were determined as described [73]. Results are expressed as the mean number of SFU/106 cells from triplicate assays. Only cell samples with >80% viability after thawing were analyzed, and only assays with <50 spot forming unit (SFU)/106 cells for the negative control and >500 SFU/106 cells after SEB stimulation were considered valid. An ELISpot result was defined as positive if the number of SFUs was ≥55 SFU/106 cells and ≥4-fold the negative control.
Cryo-preserved blood mononuclear cells (1–2×106) were stained for dead cells (4°C for 20′; Aqua LIVE/DEAD, Invitrogen) and then stained with appropriately tittered peptide-MHC class I tetramer complexes at 4°C for 30′ in Ca2+-free media as described [74]. Cells were then washed and directly stained at 4°C for 20′ with the following Abs in various combinations: CD3, CD8, CD28, CD27, PD-1, 2B4, CD160. Finally, cells were fixed (CellFix, BD), acquired on an LSRII SORP (4 lasers) and analyzed using FlowJo 8.8.2 (Tree star Inc, USA). Analysis and presentation of distributions were performed using SPICE version 5.1, downloaded from http://exon.niaid.nih.gov/spice/ [75]. The number of lymphocyte-gated events ranged between 0.6–1×106 in the flow cytometry experiments.
Cryo-preserved blood mononuclear cells (1–2×106) were stimulated for 6 h or overnight in 1 ml of complete media (RPMI (Invitrogen), 10% fetal bovine serum (FBS; Invitrogen), 100 µg/ml penicillin, 100 units/ml streptomycin (BioConcept)) in the presence of Golgiplug (1 µl/ml, BD), anti-CD28 (0.5 µg/ml, BD) and 1 µg/ml of peptide as described [74]. Staphylococcus enterotoxin B (SEB; Sigma) stimulation (100 ng/ml) served as positive control. At the end of the stimulation period, cells were stained for dead cells (4°C for 20′; Aqua LIVE/DEAD, Invitrogen), permeabilized (RT for 20′; Cytofix/Cytoperm, BD) and then stained at RT for 20′ with CD4, CD8, CD3, IFN-γ, IL-2, TNF-α and perforin (clone B-D48). Cells were then fixed (CellFix, BD), acquired on an LSRII SORP and analyzed using FlowJo 8.8.2. Analysis and presentation of distributions were performed using SPICE version 5.1, downloaded from http://exon.niaid.nih.gov/spice/ [75]. The number of lymphocyte-gated events ranged between 0.6–1×106. With regard to the criteria of positivity of the ICS, the background in the unstimulated controls never exceeded 0.03%. An ICS to be considered positive had to have >0.03% of cytokine-positive cells after subtraction of the background (media alone) and to be >5 fold higher that the background.
The analysis of the CDR3 diversity and renewal was performed as described [32]. CDR3 renewal corresponds to the percentage of TCR sequences specific for a given epitope that changed between two time points. Briefly, blood mononuclear cells were stained with cognate multimers and anti-CD3, anti-CD8, anti-CD45RA, and anti-CCR7 mAbs (BD Biosciences). CD45RA+ CCR7+ naïve and Ag-specific (multimer+) CD8 T cells were directly sorted (FACSAria, BD Biosciences) in RLT lysis buffer (Qiagen, Hilden, Germany) containing 20 ng RNA carrier (Roche Diagnostics, Rotkreuz, Switzerland) and RNA extracted (Qiagen). Then, cDNA preparation and amplification were performed by using the SuperSMART PCR cDNA Synthesis Kit according to the manufacturer's instructions (Clontech Laboratories, Saint-Germain-en-Laye, France). Amplified cDNA was subjected to TRBV–TCR-b–chain C region (TRBC) PCR reactions as described [32]. For spectratyping, aliquots of positive samples were mixed with Genescan-500 ROX size standards and run on an ABI 3130 capillary sequencer (Applied Biosystems, Foster City, CA). The CDR3 junction (length) was analyzed using the IMGT system as described [32].
Peptide stimulations were performed as described above. Functional avidity of T-cell responses was assessed by performing limiting peptide dilutions (ranging from 2 µg/ml to 1 pg/ml) in in vitro assays as described [24]. The peptide concentration required to achieve a half-maximal IFN-γ response (EC50) was determined.
Four-digit HLA class I genotyping was performed by direct sequencing methods as described [76]. The data were analyzed and alleles were assigned using Assign-SBT version 3.5 (Conexio Genomics, Applecross, Australia).
Mann-Whitney and Wilcoxon-matched paired tests were performed using GraphPad Prism version 6.00 (San Diego, CA). Analyses of the functional avidity of CD8 T-cell responses were performed on log10-transformed data using non-parametric tests. Associations among variables were performed by Spearman test. Rank correlations matrix and linear regression analysis were performed after log10 transformation of variables using R software. Bonferroni corrections for multiple analyses were applied. Regarding SPICE analyses of the flow-cytometry data, comparison of distributions was performed using a Student's t-test and a partial permutation test as described [75]. Furthermore, mixed-effect linear models were used to assess the evolution of functional avidity as a function of the time and virus rebound, as described [39], [40]. In brief, let Y_ij be the measured avidity for subject i at time j (time_ij) and rebound_ij, the covariate coded as 1 if a patient i is an on-therapy at time j and coded as 0 if not on therapy (off-therapy). We fitted the following mixed effect linear model: Y_ij = (β_0+r_i)+(β_1)time_ij+(β_2)rebound_ij+ε_ij where β_0 is the global mean, β_1 the effect of the time on avidity, β_2 the effect of the virus rebound on avidity, r_i the random effect which represents the individual deviation from the global intercept and ε ij are independent measurement errors with mean zero. The interaction between time_ij and rebound_ij was tested.
|
10.1371/journal.pcbi.1003283 | Evidence for Finely-Regulated Asynchronous Growth of Toxoplasma gondii Cysts Based on Data-Driven Model Selection | Toxoplasma gondii establishes a chronic infection by forming cysts preferentially in the brain. This chronic infection is one of the most common parasitic infections in humans and can be reactivated to develop life-threatening toxoplasmic encephalitis in immunocompromised patients. Host-pathogen interactions during the chronic infection include growth of the cysts and their removal by both natural rupture and elimination by the immune system. Analyzing these interactions is important for understanding the pathogenesis of this common infection. We developed a differential equation framework of cyst growth and employed Akaike Information Criteria (AIC) to determine the growth and removal functions that best describe the distribution of cyst sizes measured from the brains of chronically infected mice. The AIC strongly support models in which T. gondii cysts grow at a constant rate such that the per capita growth rate of the parasite is inversely proportional to the number of parasites within a cyst, suggesting finely-regulated asynchronous replication of the parasites. Our analyses were also able to reject the models where cyst removal rate increases linearly or quadratically in association with increase in cyst size. The modeling and analysis framework may provide a useful tool for understanding the pathogenesis of infections with other cyst producing parasites.
| A large portion of people worldwide are chronically infected with T. gondii. Chronic infection with this parasite is characterized by formation of tissue cysts. Bradyzoites slowly replicate within cysts during the chronic stage of infection leading to a corresponding increase in cyst size. Cysts occasionally rupture and release bradyzoites that invade nearby host cells and convert into tachyzoites which can quickly proliferate. Tissue cysts can also be targeted by immune T cells and phagocytes for removal. We developed a differential equation model to investigate the cumulative effects of unknown growth and removal functions on the cyst-size distribution. We then used the AIC to select models that best fit experimental cyst size distribution data obtained from the brains of chronically infected mice. The results suggest that the within-cyst growth of bradyzoites is finely-regulated asynchronous such that the per capita growth rate is inversely proportional to the number of bradyzoites. While it may be surprising that the bradyzoites replicate in a way analogous to a factory producing a product, there may be factors such as nutrient availability, resource allocation, immune response, and other stress factors that may limit replication in cysts.
| Toxoplasma gondii, an obligate intracellular protozoan parasite, is an important foodborne pathogen that can cause various diseases including lymphadenitis and congenital infection of the fetuses in humans. Infection occurs through ingestion of food or water contaminated with cysts or oocysts. The acute stage of infection is characterized by proliferation of tachyzoites in various nucleated cells. IFN--mediated immune responses limit tachyzoite proliferation [1]–[3] and the parasite establishes a chronic infection by forming cysts containing bradyzoites, primarily in the brain (Figure 1). Chronic infection with T. gondii is one of the most common parasitic infections in humans. It is estimated that 500 million to 2 billion people worldwide are infected with the parasite [4], [5].
During the chronic stage of infection, bradyzoites slowly replicate within the cysts and cyst sizes increase in response. In immunocompromised individuals such as those with AIDS and organ transplants, cysts can rupture resulting in release of bradyzoites, conversion of bradyzoites into tachyzoites, and proliferation of tachyzoites, which can cause life-threatening toxoplasmic encephalitis [6], [7] (Figure 1). Even in immunocompetent host, T. gondii cysts occasionally rupture during the chronic stage of infection [8]. In these cases, tachyzoite growth is controlled by the host's immune response, but the parasite is most likely able to form small numbers of new cysts (Figure 1). Such natural rupture of cysts and the formation of new cysts are thought to result in a wide range of T. gondii cyst sizes observed in the brains of chronically infected mice.
There is currently only limited information on the immune responses to the cyst stage of T. gondii [9], [10]. It was generally considered that T. gondii cysts cannot be recognized by the immune system. However, our recent study revealed that the T cells have the capability to remove tissue cysts from the brains of infected mice [9]. Marked decreases in cyst numbers occur during the T cell-mediated anti-cyst immune responses, suggesting that the immunity-mediated removal of the cysts can prevent formation of new cysts (Figure 1). Therefore, host-pathogen interactions during the chronic stage of T. gondii infection appear to have two distinct processes. One is a natural rupture of tissue cysts that can result in formation of new cysts. The other is the T cell-mediated cyst removal not associated with formation of new cysts. In order to better understand the dynamics of host-pathogen interactions during chronic T. gondii infection, in the present study we developed a set of biologically based models of cyst growth and removal including both natural rupture and immunity-mediated removal of tissue cysts and compared these models with actual data on distribution of sizes of T. gondii cysts obtained from the brains of chronically infected mice.
Previous studies by Hooshyar et al. [11] provided limited snapshots of the cyst size distributions in the brains of infected mice during the period of 2–4 months after infection. Typically, sizes of T. gondii cysts are viewed in terms of diameter. However, volume is biologically a more appropriate measure to indicate the size of cysts since it is expected to be proportional to the number of bradyzoites in a cyst. Hooshyar et al. assumed the shape of a cyst was ellipsoidal and measured the two diameters of the ellipsoid [11]. Based on their data, the mean volumes of cysts at 2, 3, and 4 months after infection were (), (), and (), respectively. The number of cysts examined was 17 for each time point. There is a significant difference in the cyst volume between months 2 and 3 (, ), 2 and 4 (, ), but not 3 and 4 (, not significant). These studies support the assumption that cyst volume reaches a steady state distribution within 4 months after infection.
In order to have a larger data set of cysts in the steady state during the chronic stage of infection, we measured sizes of over 200 cysts of T. gondii in the brains of mice at 6 months after infection. Female Swiss-Webster mice (Taconic, Germantown, NY) were infected intraperitoneally with 10 cysts of the ME49 strain (a type II strain) as previously described in [12]. T. gondii has three predominant clonal genotypes (types I, II, and III) [13]–[15]. Type II constitutes a majority of clinical cases of toxoplasmosis and asymptomatic infections in humans in North America and Europe [13], [15], [16]. Six months later, the brain of each of four mice was triturated in 1 ml of PBS [9]. Mouse care and experimental procedures were performed in accordance with established institutional guidance and approved protocols from the Institutional Animal Care and Use Committee. Four to six aliquots (20 microliters each) of each brain suspension were applied to microscopic examination using a Nikon Eclipse 90i microscope and a photograph was taken on each T. gondii cyst detected at ×400 magnification with a Nikon DS-Ri1 digital camera. Photographs of 50–56 cysts from each brain, a total of 213 cysts from four mice, were recorded (see Figure 2 for a photograph of a typical cyst). We measured the diameter of each cyst from two different angles using NIS Elements BR analysis 3.2 software (Figure 3; see also supplemental data). We calculated the volumes of each cyst using the two measured diameters by assuming an ellipsoidal shape: , where is the larger diameter and is the smaller diameter.
There have been several attempts to understand the biology of Toxoplasma gondii infection through mathematical modeling [17], [18], however, none of these previous efforts have tried to model the growth and distribution of cysts as a function of their volume. Because in this study we are solely interested in the distribution of cyst volumes, we do not explicitly model population of free bradyzoites, tachyzoites, and uninfected target cells and, instead, simply assume new cysts are being formed at some rate . See Figure 4 for a schematic of the within-host system and Table 1 for definitions of the functions used in our model. Biologically represents the rate at which uninfected target cells become infected by free parasites and begin forming intracellular cysts. Following [19], we model the growth of these cysts using a partial differential equation (PDE) structured by both time and cyst volume . Specifically,(1)where is the density of bradyzoite cysts of volume at time , is the cyst growth rate, i.e. the rate at which the bradyzoite population grows within a cyst, and is the cyst removal rate, i.e. the sum of the rate at which encysted cells are either cleared by the immune response or through natural cyst bursting.
Conceptually, the PDE defined in Equation (1) describes how the density of cysts of size at time develops over time. For example, the first term on the left hand side of Equation (1) describes the ‘movement’ of cysts of size along the time variable . Since movement along the time axis is constant, we can think of the cysts as being carried along a conveyer belt along the variable. The second term describes how growth ‘stretches’ or ‘compresses’ the distribution of with cyst growth. For example, if we are considering the density of cysts in a region where is increasing with , then the density of cysts will be stretched out along the variable as larger cysts move more quickly along the axis. In contrast, if is decreasing with then will be compressed along as smaller cysts ‘catch up’ with the larger cysts. Finally, if is constant with respect to , similar to with the time variable, the density of cysts can be envisioned as moving along the axis on conveyer belt. The removal term on the right hand side of Equation (1) describes the rate at which the cyst density is being ‘siphoned off’ via the removal process. If the removal rate decreases/increases with , then larger cysts are removed at a lower/higher rate. If is constant with respect to , then the total density of cysts of a particular age (i.e. ) will decline exponentially with time .
Although these two cyst removal processes differ in that bursting can ultimately leads to the production of new cysts while immune response clearance does not, their effects on the relative distribution of cysts as a function of volume are indistinguishable and, hence, combined in Equation (1). Biologically, both and likely vary with the immune response state of the host. However, since we are focusing on the steady state of the system where the immune response state of the host is constant, we do not explicitly model this dependency. For simplicity, we assume that all new cysts have an initial volume . Based on our definition of as the rate at which new cysts are formed, according to [20] the boundary condition for Equation (1) satisfies the equality,(2)
The general solution of Equation (1) can be obtained using the method of characteristics [21]. First, an inverse function must be determined to find the correspondence between size and time. Depending on a cysts initial volume, , the current volume, , can be determined by some function that depends on the elapsed time since infection. This function, is the solution to , where is growth rate. From equation (2), the equation for is the boundary condition. Then, the general solution is:(3)where is the boundary condition (inflow of all new cysts into the system), is the characteristic curve through the time-size domain that is defined by solving the inverse equation above, is the initial time we wish to model. See Calsina and Saldana for a complete derivation [21].
Although Equation (1) can be explicitly solved as a function of time (e.g. see [21]), here we focus solely on the steady state solution. Letting represent the steady state solution of Equation (1), that is, . Under this condition, Equation (1) simplifies to the following ordinary differential equation(4)where is a combined function of the cyst growth and removal functions:(5)Note that is the derivative of with respect to . Equation (4) has a general solution of(6)where represents the steady state density of newly formed cysts and satisfies the boundary condition defined in equation (2) with .
Because the combined function is a function of both and the first parameters of growth and removal functions, and respectively, cannot be uniquely identified. Instead, they can be estimated only as ratios of one another, i.e. , in this setting.
Our data on cyst volume represents a random sample from the larger cyst population, in order to fit our models to this data we generate a probability density function from our steady state solution. We investigate the steady state solution in Equation (6) under several different forms of growth and removal functions; see the function definitions in Table 2. We divide cyst density by the total cyst population size, to get a probability density function for cyst size. Specifically,(7)where represents the parameters of a given combined function (e.g. or and ). Using this probability density function, it follows that the negative log-likelihood of a particular model and parameter set given a random sample of observed cyst volumes is simply,(8)
For each model in Table 2 we estimated the corresponding model parameters by minimizing based on the observed data using the NMinimize routine in Mathematica 8.1. The minimal value and the total number of independent parameters were used to calculate the AIC value for each model. AIC and parameter estimates are also presented in Table 2.
We measured two diameters on each of 213 cysts detected in the brains of 4 mice at 6 months after infection in order to have a larger size of data on volume of cysts in the steady stage during the chronic stage of infection. Distributions of diameters measured and volume of cysts calculated from the diameters by assuming that cysts are in an ellipsoidal shape are shown in Figure 5. While the probability distribution on the diameter scale (Figures 5 (c) and (d)) is unimodal, the probability distribution on the volume scale (Figures 5 (a) and (b)) does not show modality. This difference is due to nonlinear transform between volume and diameter [22]. See the Methods section for calculation of the volume.
We developed a differential equation model to investigate the cumulative effects of unknown growth and removal functions on the cyst- size distribution. As a means for model selection, the Akaike information criterion (AIC) [23] was used to evaluate and compare different models; see Table 2. Based on information entropy, AIC is an estimate of the relative information lost for a given model. The AIC value of a model is calculated using its negative log-likelihood at the maximum-likelihood estimation (MLE) parameters and the number of parameters. Therefore, AIC provides a trade-off between a model's complexity and its goodness of fit. The AIC of a given model is the difference between the lowest observed AIC value and the AIC value of the model [24].
We explored three different growth functions and eight different removal functions. A schematic illustrations of these functions are shown in Figure 6. More detailed descriptions of the function formalities can be seen in Holling [25]. To determine the cyst growth model that can fit best to the experimental data, we explore three different hypotheses as follows. The first hypothesis is that cysts grow at a constant rate i.e. , such that the cyst volume increases linearly with time (indices 1–8 in Table 2). Because bradyzoite number within a cyst increases with its size, this hypothesis corresponds to a per capita growth rate of bradyzoites that is inversely proportional to the number of bradyzoite, implying that bradyzoite replication is finely regulated and asynchronous within the cyst. The second hypothesis is that the cyst volume increases exponentially with time (indices 9–16 in Table 2). This corresponds to a constant per capita growth rate of within-cyst bradyzoites, implying that bradyzoites replicate independently of each other within the cyst. The third hypothesis is that the cyst volume grows logistically with time i.e. (indices 17–24 in Table 2). This hypothesis implies that bradyzoite replication is regulated within the cyst in a simple density dependent manner in which the per capita growth rate declines linearly with cyst volume. The AIC scores indicate that hypotheses two and three are not supported by the data. Therefore, we focused on various removal functions under hypothesis one. In regard to the cyst removal rate, models with constant (index 1), one-parameter type II (index 4), two-parameter linear (index 6), two-parameter type II (index 7), and two-parameter type III (index 8) functions all fell within 2.5 AIC units of the best model (index 5), which is a model with a one-parameter type III function. We can, however, clearly reject models where cyst removal rate increases linearly (index 2) or quadratically (index 3) in association with increases in cyst volume. Comparison between probability distributions of the experimental data and the models using constant growth function (indices 1–8) in Figure 5.
We have developed a mathematical framework to select the most appropriate mathematical descriptions for the growth and removal processes of T. gondii cysts through parameter fitting of experimental data obtained from the brains of chronically infected mice. Population growth often satisfies a linear or logistic growth function [26]. However, experimental data here supports a constant growth rate model, i.e., . We calculated the volumes of cysts by assuming that cysts are in an ellipsoidal shape. We also performed the same analysis by assuming that cysts are in a spherical shape using the effective diameter (data not shown). In both cases, we reached the same conclusion. We assumed the cyst volume is proportional to the number of bradyzoites within the cyst. Therefore, a constant volume growth rate indicates that the number of parasites within the cyst increases linearly over time and the per capita growth of bradyzoites is inversely proportional to the number of parasites within a cyst. This probably suggests that bradyzoites do not replicate synchronously but each bradyzoite divide independently to produce a single new bradyzoite within a certain time interval. For example, a cyst may start with a given number of bradyzoites and a single new bradyzoite may be formed through replication every few hours. This is a contrast to tachyzoites of T. gondii or merozoites of malaria parasite. The tachyzoites and merozoites are the acute stage form of these parasites and they proliferate quickly after invading into host cells. On the other hand, tissue cysts of T. gondii are formed in the chronic stage of infection and the major purpose of cysts is most likely to persist within host cells, rather than proliferate. Therefore, it appears that tachyzoites and bradyzoites within cysts are under distinct regulatory mechanisms to control their proliferation. While it may be surprising that the bradyzoites replicate in a way analogous to a factory producing a product, there may be factors such as nutrient availability, immune response, and other stress factors that may limit their replication.
Based on the analyses on cyst growth described above, we performed parameter fitting of various removal functions with the constant growth rate. The best model was a one-parameter type III function; however several other removal functions performed similarly well and are indistinguishable from one another. Based on the AIC criteria, performances of the following functions (constant, type II, type III, and type III with two parameters) are indistinguishable for the constant growth rate model. Thus, the current data cannot distinguish between several removal functions. However, our analyses were able to reject models where cyst removal rate increases linearly or quadratically with increases in cyst volume. This result would suggest that removal of cysts is the outcome of a complex of multiple biological mechanisms.
In this study, we considered two removal processes: natural rupture and immune-mediated removal. Natural rupture of cysts may not occur simply based on the volume of cysts. It may also depend on cell-types of cyst-containing host cells and location of cysts in the brain. It has been shown that T. gondii can form cysts in both glial cells and neurons [27]–[29]. Removal of cysts by immune T cells and phagocytes could be independent of the sizes of the cysts contained in the infected host cells. To determine the specific removal function that fits best in experimental data, we would need to collect data on the transient dynamics and conduct corresponding studies. Moreover, the current study on steady state can only determine the ratio between the parameters and . Transient data are also needed to estimate these parameters separately.
Recent studies suggested possible contributions of chronic infection with T. gondii with important diseases such as cryptogenic epilepsy and Alzheimer's disease [30], [31]. Thus, it is crucial to understand the mechanisms of host-pathogens interactions in the brain during the chronic stage of infection with this parasite for defining the pathogenesis of this common infection. The present study provided valuable information that may improve our understanding in this aspect. This study also demonstrated a power of mathematical modeling to provide the information that will be difficult to obtain directly from biological studies. In the present study, we obtained the data at only one time point of the chronic stage of infection. Having the data from multiple time points including the acute stage of infection and larger samples numbers at each time points will assist in understanding of dynamics of cyst growth and removal during the course of infection with T. gondii. These data will also assist in better understanding of the roles of natural rupture of cysts and immune response-mediated removal of cysts in the pathogenesis of cerebral infection with the parasite.
|
10.1371/journal.pntd.0001193 | Interferon-γ and Proliferation Responses to Salmonella enterica Serotype Typhi Proteins in Patients with S. Typhi Bacteremia in Dhaka, Bangladesh | Salmonella enterica serotype Typhi is a human-restricted intracellular pathogen and the cause of typhoid fever. Cellular immune responses are required to control and clear Salmonella infection. Despite this, there are limited data on cellular immune responses in humans infected with wild type S. Typhi.
For this work, we used an automated approach to purify a subset of S. Typhi proteins identified in previous antibody-based immuno-affinity screens and antigens known to be expressed in vivo, including StaF-putative fimbrial protein-STY0202, StbB-fimbrial chaperone-STY0372, CsgF-involved in curli production-STY1177, CsgD- putative regulatory protein-STY1179, OppA-periplasmic oligopeptide binding protein precursor-STY1304, PagC-outer membrane invasion protein-STY1878, and conserved hypothetical protein-STY2195; we also generated and analyzed a crude membrane preparation of S. Typhi (MP). In comparison to samples collected from uninfected Bangladeshi and North American participants, we detected significant interferon-γ responses in PBMCs stimulated with MP, StaF, StbB, CsgF, CsgD, OppA, STY2195, and PagC in patients bacteremic with S. Typhi in Bangladesh. The majority of interferon-γ expressing T cells were CD4 cells, although CD8 responses also occurred. We also assessed cellular proliferation responses in bacteremic patients, and confirmed increased responses in infected individuals to MP, StaF, STY2195, and PagC in convalescent compared to acute phase samples and compared to controls. StaF is a fimbrial protein homologous to E. coli YadK, and contains a Pfam motif thought to be involved in cellular adhesion. PagC is expressed in vivo under the control of the virulence-associated PhoP-regulon required for intra-macrophage survival of Salmonella. STY2195 is a conserved hypothetical protein of unknown function.
This is the first analysis of cellular immune responses to purified S. Typhi antigens in patients with typhoid fever. These results indicate that patients generate significant CD4 and CD8 interferon-γ responses to specific S. Typhi antigens during typhoid fever, and that these responses are elevated at the time of clinical presentation. These observations suggest that an interferon-γ based detection system could be used to diagnose individuals with typhoid fever during the acute stage of illness.
| Salmonella enterica serotype Typhi infection is a significant global public health problem and the cause of typhoid fever. Salmonella are intracellular pathogens, and cellular immune responses are required to control and clear Salmonella infections. Despite this, there are limited data on cellular immune responses during wild type S. Typhi infection in humans. Here we report the assessment of cellular immune responses in humans with S. Typhi bacteremia through a screening approach that permitted us to evaluate interferon-γ and proliferation responses to a number of S. Typhi antigens. We detected significant interferon-γ CD4 and CD8 responses, as well as proliferative responses, to a number of recombinantly purified S. Typhi proteins as well as membrane preparation in infected patients. Antigen-specific interferon-γ responses were present at the time of clinical presentation in patients and absent in healthy controls. These observations could assist in the development of interferon-γ-based diagnostic assays for typhoid fever.
| Salmonella enterica serotype Typhi is a human-restricted intracellular pathogen and the cause of typhoid fever. It is estimated that over 20 million cases of S. Typhi infection occur each year, resulting in approximately 200,000 deaths per year globally [1]. Current typhoid vaccines provide 50–75% protection for 2–5 years [2]. Mediators of protective immunity against typhoid are incompletely understood. S. Typhi is an invasive enteropathogen that, following ingestion, transits through intestinal epithelial cells, is taken up by professional phagocytic cells, survives within macrophages, and systemically circulates [3], [4], [5], [6]. Antibody responses to lipopolysaccharide (LPS), flagellin, Vi capsular polysaccharide, and crude whole cell preparations have been documented, and antibody responses are the basis of the Widal serologic diagnostic assay for typhoid fever [7], [8], [9], [10], [11]. However, with the exception of antibody responses against the S. Typhi capsule (Vi antigen) [12], antibody responses may play a limited role in mediating protective immunity during typhoid fever.
S. Typhi is an intracellular pathogen, and cellular immune responses are required to control and clear S. Typhi infections [10], [13], [14], [15]. Unfortunately, there are limited data on antigen-specific cellular responses during human wild type S. Typhi infection. What is known is largely derived from analyses of cellular responses in mice infected with S. Typhimurium [16], [17], [18]; however, S. Typhimurium does not cause a typhoidal illness in humans, and S. Typhi and S. Typhimurium differ significantly at the genomic level [17], [19], [20]. Direct analysis of cellular responses during S. Typhi infection in humans either pre-dates modern immunologic techniques [21] or involves characterizing immune responses in recipients of live attenuated oral typhoid vaccines [22], [23]. These analyses have shown that CD4+ and CD8+ T cells are critical to the development of protective immunity to Salmonella, and control of Salmonella infection involves prominent expression of interferon-γ by both CD4 and CD8 cells [24], [25], [26]. To date, however, there is less information on the cellular responses in humans during wild type infection, especially to purified S. Typhi antigens.
To address this, we used a modification of an automated approach to purify a subset of S. Typhi proteins for use in immunologic assays [27]. We selected antigens for evaluation based on our previous application of a high throughput immuno-affinity screen, In Vivo Induced Antigen Technology (IVIAT), to S. Typhi [28]. Here we describe purification of a subset of these proteins, and evaluation of interferon-γ and cellular proliferation responses to these antigens in humans with S. Typhi bacteremia in Bangladesh. We also assessed responses to a crude membrane preparation of S. Typhi.
We selected 58 S. Typhi proteins contained within operons identified during our previous application of IVIAT to S. Typhi [28]. IVIAT identifies proteins expressed in vivo during human infection and that generate an antibody response [28]. We obtained pDONR221 Gateway Based entry clones of the S. Typhi CT18 genes corresponding to selected proteins from the NIAID-sponsored Pathogen Functional Genomic Resource Center, J. Craig Venter Institute (JCVI, formerly The Institute for Genomic Research). We used LR clonase II enzyme reactions (Invitrogen, Carlsbad, CA) as per the manufacturer's instructions to move inserts into pDEST17 (Invitrogen, Carlsbad, CA) to generate a fusion containing an amino terminal 6× histidine (HIS) tag. We transformed DH5alpha-T1R competent cells with LR reactions and selected for ampicillin resistance. We confirmed insert presence by restriction digestion and PCR analysis and transformed purified plasmids into E. coli protein expression strain BL21 star (DE3) pLysS (Invitrogen).
We grew transformants harboring recombinant plasmids at 37°C as 1.5 ml cultures in 96-well blocks (Marsh Biomedical Products) to an OD600 of 0.6–0.8. We induced cultures with 1 mM isopropyl β-D-1-thiogalactopyranoside (IPTG) on a 96-well plate shaker (Multitron) (×900 rpm). After 3 hours at 37°C, we harvested cells at 4°C and stored preparations at −80°C for further use. We also induced BL21 star (DE3) pLysS containing pDEST17 but lacking an S. Typhi insert. This construct produced a truncated HIS-tagged protein MSYYHHHHHHLESTSLYKKAERERKMI that we recovered and used as a control protein in immunological assays.
We performed protein purifications in 96-well plates using a BiomekFx (Beckman Coulter) robotic liquid handler as previously described [27]. For this 6xHIS denaturing affinity purification, we thawed cell pellets at room temperature for 15 min, lysed them in the presence of protease inhibitors in 115 µl lysis buffer I (100 mM NaH2PO4, 10 mM Tris, pH 8.0), robotically resuspended product in a 96-well block and agitated at 900 rpm for 10 min (5 min in the clockwise direction and 5 min in the counterclockwise direction). We then added 10 µl of DNase mix (10 mg/ml DNase; Sigma Aldrich in 900 mM MgCl2, 100 mM MnCl2) to the lysate and agitated the preparation at 900 rpm for 10 min. Next, we added 115 µl of lysis buffer II (100 mM NaH2PO4, 10 mM Tris, 6 M guanidine hydrochloride, 10 mM, 2-mercaptoethanol, pH 8.0) to create denaturing conditions. We then allowed these cell lysates to bind to 30 µl of MagneHIS beads (Promega) with shaking at 900 rpm for 20 min (10 min clockwise, 10 min counterclockwise), and separated beads using a magnabot (24-pin magnet; Promega). The robotic liquid handler then washed the MagneHIS beads with bound protein three times with wash buffer (100 mM NaH2PO4, 10 mM Tris, 8 M urea). We prevented bead adherence to the walls during washing by shaking the samples at 900 rpm for 2.5 min clockwise and then 2.5 min counterclockwise. We then washed the beads with bound protein using 100 µl of distilled water, and added 50 µl distilled water to make the final suspensions for analysis. We repeated this extraction cycle six times.
We analyzed proteins in a 96-well format using a capillary-based instrument, the LabChip90 (Caliper Sciences). We automated a system that resuspended 3 µl of protein sample in 7 µl analysis buffer (Caliper Sciences), heated these to 96°C for 5 min., cooled them to room temperature, and briefly centrifuged to collect the sample. We added distilled water (35 µl) to each sample prior to analysis. We primed the analysis chip (Caliper Sciences) according to the manufacturer’s instructions. The automated protein analysis generated three different forms of output: a chromatogram that showed migration time; a virtual gel that mimicked a Coomassie stained gel; and a results table that included the estimated size, quality, and quantity of each peak. The LabChip90 analyzed 96 proteins at a time with analysis time of 40 seconds per sample. We parsed the output results and imported them into the Harvard Institute of Proteomics protein database. We assessed for presence of contaminating E. coli LPS using a HEK-Blue LPS Detection kit (InvivoGen, San Diego, CA).
We prepared S. Typhi membrane preparation as previously described [29], [30]. Briefly, we cultured S. Typhi Ty21a on sheep blood agar plates and harvested in Tris buffer (10 mM Tris, pH 8.0, 5 mM MgCl2). We sonicated the mixture, and centrifuged at 1400× g for 10 minutes and transferred the supernatant to fresh tubes, centrifuging at 14900× g for 30 minutes. We suspended the pellet in 10 ml Tris buffer, and determined the protein content by the BioRad Protein Assay per the manufacturer's instructions.
We performed mass spectrometric analysis of the S. Typhi membrane preparation as previously described using a LTQ-Orbitrap XL (Thermo Fisher Scientific) instrument [19], [31]. We identified peptides using SEQUEST (Thermo Fisher Scientific) through Bioworks Browser, version 3.3.1 SR1. MS/MS data were obtained using 10 ppm mass accuracy on precursor m/z and a 0.5 Da window on fragment ions. Fully enzymatic tryptic searches with up to three missed cleavage sites were allowed. Oxidized methionines were searched as a variable modification and alkylated cysteines were searched as a fixed modification. Salmonella databases for CT18 were downloaded from EMBL-EBI and supplemented with common contaminants. We employed a reverse database strategy [32] using concatenating reversed protein sequences for each database entry in SEQUEST. We filtered peptides for each charge state to a false discovery rate (FDR) of 1%, and then grouped peptides into proteins using Occam’s razor logic. A full listing of proteins identified in mass spectrometric analysis of Salmonella Typhi membrane preparation is available in the supplemental material (Table S1).
Individuals (1–59 years of age) with fever of 3–7 days duration (≥39°C) having clinical symptoms and signs suggestive of typhoid fever and lacking an alternate diagnosis who presented to the Kamalapur field site of the International Centre for Diarrhoeal Disease Research, Bangladesh (ICDDR,B) Dhaka hospital were eligible for enrollment. We collected venous blood (for children <5 years of age, 3 ml of blood; for older individuals, 5 ml of blood) for culture (n = 69). We used the BacT/Alert automated system and identified S. Typhi organisms using standard biochemical methods and by reaction with Salmonella-specific antisera [30], [33]. Following informed consent from patients or guardians in the case of children, we collected an additional 5 ml of blood from bacteremic individuals within 72 hours of the patient presenting for medical care, and a follow-up sample 21–28 days later (n = 16; ages 2–22 years). All patients with 3 days or longer of fever were treated initially with amoxicillin or cefixime at the discretion of the attending physician until scheduled follow-up 48–72 hours later, or sooner as clinically indicated. Individuals with documented S. Typhi bacteremia were continued on amoxicillin if they showed signs of improvement and their blood isolates showed sensitivity to first line treatment; or were switched to parenteral ceftriaxone or oral ciprofloxacin, if their isolates were not sensitive and/or they failed to improve by 72 hours; therapy was continued for up to 14 days, or up to 7 days beyond defervescence, whichever occurred first. All patients recovered. We also collected 5 ml of blood from North American volunteers (n = 3) without a history of international travel who had never received typhoid vaccination and who did not have previous known Salmonella infection, and we collected 5 ml of blood from healthy Bangladeshi volunteers (n = 4) who did not have illness, fever or diarrhea in the preceding three months [34]. Studies were approved by the Institutional Review Boards of the ICDDR,B and Massachusetts General Hospital.
We diluted heparinized blood in phosphate buffered saline (PBS; 10 mM, pH 7.2) and isolated peripheral blood mononuclear cell (PBMC) by gradient centrifugation on Ficoll-Isopaque (Pharmacia, Uppsala, Sweden). We re-suspended isolated PBMCs to a concentration of 1×106 cells/ml in RPMI complete medium RPMI-1640 (Gibco, Gaithersburg, Md) with 10% heat-inactivated fetal bovine serum (Hyclone-Thermo Scientific, Waltham, MA, USA), 100 units/ml penicillin, 100 µg/ml streptomycin, 100 mM pyruvate, and 200 mM L-glutamine (Gibco) [35].
We used PBMCs to measure human interferon-γ expression using an ELISPOT format with MabTech antibodies, according to the manufacturers’ instructions (Mabtech Inc, Cincinati, OH, USA). In brief, we coated 96-well nitrocellulose plates (Multiscreen HTS, Millipore) with 100 µl of 15 µg/ml human monoclonal anti-interferon-γ antibody (1-D1K) overnight at 4°C. Following washing the plates and subsequent blocking with 10% FBS for 2 h at room temperature, we added PBMCs from individual patients or controls at a concentration of 2×105 per well for each experimental condition. We added individual S. Typhi antigens or control protein to wells at a concentration of 140 ng/well of total preparation for each purified antigen (in 200 µl culture, final concentration 0.7 µg/ml). In separate wells, we also added S. Typhi membrane preparation at a final concentration of 10 µg/ml in 200 µl culture, phytohaemagglutinin (PHA; Murex Diagnostics Ltd, Temple Hill, UK) at a final concentration of 2.5 µg/ml in 200 µl culture, and keyhole limpet hemocyanin (KLH). We included additional control wells with media but lacking antigen. Following incubation of plates at 37°C in 5% CO2 for 20 hours, we washed plates, added biotinylated monoclonal anti-interferon-γ antibody (7-B6-1-biotin; 1∶500 dilution), incubated plates at room temperature for an additional 2 hours, washed them, added streptavidin-HRP (1∶500 dilution), and re-incubated for 1 hour at room temperature. We developed plates with aminoethylcarbazol plus H2O2, and counted interferon-γ secreting cells using a stereomicroscope. We subtracted results for wells containing media only and expressed results as the number of spots/106 PBMC in each experimental condition [36].
To characterize the interferon-gamma T cell response further, we resuspended PBMCs at a concentration of 1×106 cells/mL in RPMI medium (Gibco, Carlsbad, CA) and supplemented with 10% fetal calf serum (FCS, Gibco). We cultured PBMCs in U-bottom tissue culture plates (Nunc, Denmark) in the presence of Salmonella membrane preparation (MP; 10 µg/ml), StaF (7 µg/ml), PagC (7 µg/ml), KLH (2.5 µg/ml as a negative control) or PMA (5.0 ng/ml as a positive control; Phorbol 12-myristate 13-acetate) with ionomycin (1.0 µg/ml). Samples containing only unstimulated cells were included to assess in vivo stimulation. We used 1.0 µg/ml of anti-CD28 (clone 28.2; BD Pharmingen) and anti-CD49d (clone 9F10; BD Pharmingen) for co-stimulation. We incubated PBMCs and antigens for 2 hours at 37° C in 5% CO2. After 2 hours, we added 10 µg/mL of brefeldin A (BFA, Sigma) and continued incubating the plates for an additional 4 hours [37]. Following stimulation, we washed cells with PBS and 2% FCS. We then stained cells for 30 min at 4°C with the following surface monoclonal antibodies: anti-CD3-APC, anti-CD4–perCP, and anti-CD8-FITC (Becton Dickinson, San Jose, USA). Following surface staining, we washed the cells and incubated the preparations with FACS Lysing Solution (BD Bioscience) for 10 minutes, and then re-washed and permeabilized the preparations with FACS permeabilizing solution (BD Bioscience) for 10 min at room temperature. We washed the permeabilized cells and stained them for 30 min at 4°C with fluorochrome-conjugated anti-IFN-γ-PE (BD Bioscience). Following staining, we re-washed the cells, and fixed them in formaldehyde before performing flow cytometry using a FACS Calibur (BD, San Jose, CA) [37]. We identified the lymphocyte population on forward versus side scatter plot, then gated CD3+CD4+ and CD3+CD8+ subpopulations, and identified CD4+IFN-γ+ and CD8+IFN-γ+ subpopulations. We subtracted unstimulated responses, and expressed results as interferon-γ+ T cells per 10×6 PBMC.
To evaluate proliferative responses to antigens, we cultured PBMCs (105 cell per well) in DMEM/F12 medium (Gibco, GlutaMAX) supplemented with 1% gentamicin and 5% human AB+ serum in triplicate wells in round-bottomed 96-well plates. We added S. Typhi antigens and controls to wells at the same concentrations used in the interferon-γ ELISPOT assay and with a final culture volume of 200 µl. We incubated plates at 37°C in 5% CO2 for 5 days. After 48 h incubation, we replaced 100 µl of the medium per well with fresh medium. After 5 days of incubation, we added 3H-thymidine (1 µCi) to each well under sterile conditions, incubated plates for an additional 8 hours, harvested cells in Bray’s scintillation fluid (Ultimagold, PerkinElmer, Boston, MA) using a cell harvester (Skatron instruments, Norway), and assessed [3H] thymidine incorporation using a liquid scintillation β-counter (Beckman LS6500 multipurpose scintillation counter, USA) as previously described [22], [38]. We expressed results as counts per minute (cpm), and calculated stimulation indices for each antigen according to the formula: net cpm with antigen /net cpm without antigen (media alone) for each individual on each day (day 5 and day 20) [39].
We used Prism4 (version 4.03, GraphPad Software, Inc.) for data management, analysis and graphical presentation. We used unpaired T tests to compare differences between groups, and paired T tests to evaluate differences between study days within groups.
We estimated that we required at least 20 µg of a specific protein for use in our planned immunological assays. Our six production runs resulted in the production of 20 µg or more for 25 of our selected 58 proteins; nine of these samples had purity by LC90 Caliper analysis of >90%, and 17 had purity greater than >80%. Purity was defined as the quantity of protein matching the molecular size of the desired product. The LPS contamination of all preparations was found to be less than the level of detection of our assay kit (<300 fg/µl). Of these 17 proteins with sufficient quantity and purity, we selected 7 proteins for our initial analysis (Table 1) representing a range of cellular location and function, including a number involved in fimbrial attachment or adhesion such as StaF (putative fimbrial protein encoded by STY0202), StbB (fimbrial chaperone encoded by STY0372), CsgF (involved in curli production encoded by STY1177), and CsgD (a putative regulatory protein encoded by STY1179), as well as OppA (a periplasmic oligopeptide binding protein precursor involved in peptide transport encoded by STY1304), a conserved hypothetical protein encoded by STY2195, and PagC, an outer membrane protein encoded by STY1878 whose expression is regulated by the PhoP regulon involved in intra-macrophage survival [19], [28].
Our mass spectrometric analysis of S. Typhi membrane preparation identified 934 S. Typhi proteins (636 with three or more spectral counts), including many involved in energy metabolism, protein synthesis and fate, cell envelope or peptidoglycan synthesis or maintenance, cellular processes, proteins involved in transport, proteins involved in regulatory functions, and proteins involved in virulence and pathogenesis (Table 2 and 3 and Table S1). We also identified two of our 7 selected proteins (OppA and PagC) in the S. Typhi membrane preparation.
We found that patients with S. Typhi bacteremia had elevated interferon-γ ELISPOT responses at both acute and convalescent stages of infection compared to healthy controls for all seven of the purified S. Typhi proteins, as well as against S. Typhi crude membrane preparation (P<0.05) (Figure 1). In contrast, responses to PHA did not differ significantly between patients and healthy controls, and minimal responses were detected against control protein and KLH in both patients and healthy controls. To assess whether interferon-γ responses were CD4 or CD8-derived, we used intracellular cytokine staining following stimulation with a subset of proteins, and found that the majority of interferon-γ expressing cells were CD4-positive, although a CD8 positive response was also detected (Figure 2).
To further evaluate responses, we selected the three proteins associated with the highest interferon-γ expression levels in convalescent phase samples, as well as membrane preparation, for inclusion in cellular proliferation assays. In comparison to healthy Bangladeshi controls residing within the same S. Typhi endemic area, individuals with documented S. Typhi bacteremia had significantly elevated proliferation indices at the acute stage of illness to StaF and PagC (P<0.01−0.0008), but not to STY2195, or crude membrane preparation, and these acute stage responses further significantly increased within bacteremic individuals by the convalescent period compared to the acute stage responses (P≤0.02−0.001) (Figure 3). We also detected a significantly increased proliferation response to STY2195 and S. Typhi membrane preparation in bacteremic patients at convalescence compared to acute phase samples and compared to control patients (p≤0.01).
Cellular immune responses, including CD4 and CD8-mediated interferon-γ responses, play a critical role in clearing and controlling systemic Salmonella infections [23], [40]. Despite this, there has been limited evaluation of cellular responses in humans to wild-type S. Typhi. No animal model fully replicates host-pathogen interactions and immunologic events that occur during this human-restricted infection. Evaluation in humans has largely focused on characterizing responses in recipients of attenuated vaccine strains of S. Typhi [23], [38], [41], [42], [43], [44]. We report here a screening approach that permitted us to evaluate interferon-γ and proliferation responses to a number of bacterial antigens in S. Typhi-infected humans in Bangladesh. We selected proteins that we had previously identified in immuno-affinity screening assays for humoral responses [28], and we recovered these selected proteins using an automated system and high throughput genomic and proteomic technologies. Although we were able to generate adequate samples for only approximately a third of our selected proteins for evaluation in humans, we feel that high throughput approaches such as the one we describe will assist in accelerating analysis of pathogens that express thousands of antigens. For instance, S. Typhi contains approximately 4,400 open reading frames, and although protein microarrays can be used to screen for humoral responses across the immunoproteome, no comparable system has yet been developed to assess cellular immune responses in a high throughput manner, despite the critical role that cellular immune responses play against intracellular pathogens.
We recognize that high throughput purification techniques may be compromised by issues of contamination, including with LPS when expression occurs in E. coli vectors. However, LPS contamination of all preparations was found to be less than the level of detection of our assay kit (<300 fg/µl), we did not detect cellular immune responses to control protein expressed and purified from E. coli in the same manner as our S. Typhi proteins, and we detected cellular immune responses in patients but not healthy controls to purified S. Typhi proteins. All of these observations suggest that the responses we observed were antigen-specific and not due to contaminating LPS.
In the S. Typhimurium mouse model, CD4 and CD8 cells are critical to the development of protective immunity, and control of Salmonella infection involves prominent expression of interferon-γ by both CD4 and CD8 cells [24], [25], [26]. Overall, only a relatively few defined class I and class II epitopes have been identified in the S. Typhimurium mouse model, including epitopes in FliC and SipC for CD4 cells, and OmpC and GroEL for CD8 cells [24], [40], [45], [46], [47], [48]. A number of Salmonella antigens are also able to induce partially protective immunity when included in subunit-based vaccines in mice, including flagellin, MIG-14 and SseB (Salmonella antigens expressed in vivo), suggesting that immune responses against a number of Salmonella antigens could contribute to protective immunity [49], [50], [51].
In comparison to the murine data, evaluation of cellular responses to S. Typhi in humans have largely involved individuals who have received attenuated S. Typhi vaccine strains such as Ty21a and CVD908 [23], [38], [41], [42], [43], [44]. In concordance with the mouse data, these studies have shown induction of interferon-γ-expressing CD4 and CD8 responses following vaccination [23], [24], [38], [42], [52]. Interestingly, CD8 responses may involve both classical (HLA-A, B and C in humans) and non-classical (HLA–E, F, and G) mediated T cell recognition [43], [52]. Using an ex vivo model, Sztein and colleagues have also recently found that direct infection of antigen-presenting dendritic cells with S. Typhi leads to expression of high levels of TNF-α, IL-6 and IL-8, and low levels of interferon-γ and IL-12 p70, but that dendritic cells can also ingest other infected human cells leading to high level expression of interferon-γ and IL-12 p70, with subsequent induction of a population of CD3+CD8+CD45RA-CD62L- effector/memory T cells in co-cultured lymphocytes [53]. Based on these observations, we used recombinant antigens to assess and characterize interferon-γ and proliferation responses in infected humans in Bangladesh. To establish the feasibility of our approach, we focused our initial efforts on a subset of proteins that we had previously identified as generating humoral immunity and being expressed in vivo during human infection [28]. These included a number involved in fimbrial attachment or adhesion such as StaF, StbB, CsgF, and CsgD, as well as OppA, a conserved hypothetical protein encoded by STY2195, and PagC, an outer membrane protein encoded by STY1878. We previously found that humans infected with S. Typhi develop a serum antibody response to PagC and that this response increases at convalescence [28]. Here we furthered this observation and report detection of a parallel cellular response against PagC during human infection, including both interferon-γ and proliferative responses, and show that responses in convalescence were higher than during acute stage illness. Although the role of PagC during human infection is not fully understood, its expression is controlled by the PhoP-regulon involved in intra-macrophage survival [19], [54].
We also detected significant increases in cellular responses during convalescence against StaF, a fimbrial protein homologous to E. coli YadK that contains a Pfam motif believed to be involved in cellular adhesion [17], STY2195, a conserved hypothetical protein of unknown function, and a crude membrane preparation containing over 900 S. Typhi proteins, including GroEL, OmpC, OppA and PagC.
We found that S. Typhi proteins elicit both CD4+ and CD8+ interferon-γ expressing responses, with CD4 responses being more numerous than CD8 responses. Of interest, we were able to detect antigen-specific interferon-γ responses in patients, including at the time that patients presented for clinical care, but similar responses were not seen on controls. These observations suggest that an antigen-specific interferon-γ-based detection system might be used to diagnose individuals with typhoid fever during the acute stage of illness, similar to the approach used to diagnose infection with Mycobacterium tuberculosis [55], [56], [57]. Currently, all available diagnostic tests for typhoid fever lack either sensitivity and/or specificity, especially in areas of the world endemic for typhoid. For example, microbiological culturing of blood has approximately 30–70% sensitivity, depending on the volume of blood obtained and whether previous antibiotics have been administered, and the Widal assay has at best 85% specificity when analyzing both acute and convalescent phase responses in endemic zones where typhoid exacts its highest burden [58], [59], [60].
In summary, we have used a screening format to preliminarily characterize S. Typhi antigen-specific interferon-γ responses in patients with typhoid fever. This is the first characterization of such responses in humans, and further immunologic analysis will be required to assess the role, if any, that these responses play in controlling or clearing S. Typhi infection. Our study has a number limitations, including analysis of a relatively small number of purified S. Typhi antigens, characterization of a limited number of immunologic parameters, and the absence of the inclusion of febrile control patients confirmed not to be acutely infected with S. Typhi; however, our detection of antigen-specific interferon-γ responses could assist in the development of interferon-γ-based diagnostic assays for typhoid fever, and our overall approach could be used to identify antigens capable of inducing cellular immune responses during infection with other intracellular pathogens.
|
10.1371/journal.pbio.1000173 | Hippocampus Leads Ventral Striatum in Replay of Place-Reward Information | Associating spatial locations with rewards is fundamental to survival in natural environments and requires the integrity of the hippocampus and ventral striatum. In joint multineuron recordings from these areas, hippocampal–striatal ensembles reactivated together during sleep. This process was especially strong in pairs in which the hippocampal cell processed spatial information and ventral striatal firing correlated to reward. Replay was dominated by cell pairs in which the hippocampal “place” cell fired preferentially before the striatal reward-related neuron. Our results suggest a plausible mechanism for consolidating place-reward associations and are consistent with a central tenet of consolidation theory, showing that the hippocampus leads reactivation in a projection area.
| Thinking back to an exciting event often includes the scene in which the event took place. Associations between specific places and emotional events are consolidated in memory, but how this is achieved is currently unknown. Two brain areas involved in learning such associations are the hippocampus and the ventral striatum, which represent spatial and emotional information, respectively. A highly valuable object in an environment will prompt humans and animals to take action, such as approaching the object. Here, we demonstrate that a combination of spatial and emotional aspects of a learning experience is replayed in the hippocampus and the ventral striatum during sleep, which is likely to contribute to the consolidation and strengthening of memory traces. This reactivation is coordinated such that the spatial information in the hippocampus is activated shortly before the emotional information in the ventral striatum. This finding is consistent with a central prediction from Memory Consolidation Theory, namely that the hippocampus initiates and orchestrates replay in connected brain areas. In addition, sleep replay occurs at a time scale about ten times faster than during the actual experience, which makes it a mechanism suitable for strengthening synaptic connections associating place with reward. Our results shed new light on the distributed way the brain processes, links, and retrieves different aspects of memories.
| Successful foraging requires that animals maintain a representation of a multitude of reward properties including the location at which a reward can be found. Forming a place–reward association is thought to depend critically on the communication between the hippocampal formation and the ventral striatum (VS). Cells in the hippocampus proper (HC) [1],[2] and adjacent subiculum [3] show location-specific firing (i.e., “place fields”), and these structures are crucial for spatial and contextual learning [2],[4]–[6]. Neurons in the VS fire in relation to rewards, as they are expected or actually delivered, as well as to cues predictive of reward [7]–[9]. Receiving information from a range of structures such as the HC, amygdala, prefrontal cortex, and midline thalamic nuclei [10]–[13], the VS is thought to utilize information of reward-predicting cues and contexts to guide goal-directed behavior [8],[14],[15]. This process is under strong control of the mesolimbic dopaminergic system, and its disruption has been associated with neuropsychiatric conditions such as drug addiction and obsessive-compulsive disorder [16]–[18]. Although the hippocampal formation projects directly to the VS, and this connection has been implicated in contextual conditioning [19], it is unknown how neural representations of contextual and motivational information are integrated and stored to enable the learning of place-reward associations.
In several brain areas, neuronal patterns evoked during behavior are reactivated during subsequent sleep [20]–[24]. Through modification of synaptic connections, this reactivation has been theorized to constitute an important step in memory consolidation [25]–[28]. Because hippocampal CA1 pyramidal cells exhibiting place fields during active behavior have been demonstrated to reactivate during sleep, it may be reasonably assumed that this replay pertains to spatial and contextual information [20],[21],[29],[30]. In contrast, reactivation in the VS is dominated by reward-related information [31]. Joint reactivation of HC and VS may enable the formation of a memory trace comprising both contextual and motivational components. In this study, we recorded activity from neuronal ensembles in the rat HC and VS simultaneously during wake and sleep episodes to examine whether the HC and VS reactivate coherently and to reveal the temporal dynamics of this process. First, during active behavior, much of the dynamics of hippocampal processing is governed by the theta rhythm, which has been hypothesized to function as a “read-in” or encoding mode for information acquisition and provides a means to temporally align spike sequences by way of theta phase precession [26],[32]–[34]. Therefore, we studied whether neural activity modulation by this rhythm in the awake state is correlated to reactivation during sleep. A second foremost question in this field, not yet addressed in previous multi-area recording studies [22],[24],[35], is whether cross-structural replay depends on the type of behavioral information coded by cell assemblies. To address this question, we investigated whether reactivation is preferentially associated with the expression of place fields and reward-related neural responses. Third, we planned to utilize joint HC-VS recordings to test a central tenet of theories of memory consolidation [25]–[28]. These theories posit that, after a learning experience, long-term episodic and declarative memories become gradually independent of hippocampal storage because this structure would repeatedly retrieve stored associative information over time and thereby orchestrate consolidation of memory traces in the neocortex and other target sites. A key point in these hypotheses is that replay is initiated and orchestrated by the HC, which prompted us to examine whether hippocampal activity leads the VS during reactivation.
Four rats were implanted with a tetrode drive allowing joint HC-VS recordings of spike trains of multiple neurons and local field potentials (LFPs) in each area. Daily recording sessions were composed of an episode of reward searching behavior flanked by two episodes of rest, which rats spent on a “nest” next to the track. The task was to run along a triangular track repeatedly and in one direction. On each lap, one of three reward wells was baited with a drop of one of three corresponding reward types; i.e., sucrose solution, vanilla desert, or chocolate mousse. An example of joint HC and VS ensemble recordings during track running and sleep is shown in Figure 1. First, we assessed reactivation of neuronal patterns using an explained variance (EV) method based on the spike correlations of cell pairs across all simultaneously recorded neurons [23],[36]. The EV reflects the extent to which the variance in the distribution of spike correlations during postbehavioral rest is statistically accounted for by the correlation pattern found during track running, factoring out the correlations present in prebehavioral rest. Joint HC-VS reactivation was examined during rest periods in which the rat was immobile, using only spike correlations between pairs composed of one HC and one VS neuron.
We found coherent, cross-regional reactivation between ensembles of the HC and VS as expressed by an EV of 9.7±3.0%, which was significantly higher than the control measure, the reverse explained variance (REV: 1.4±0.5%, p<0.01, n = 21 sessions; Figure 2A and 2B; Figures S1 and S2; Table S1; Text S1). In the analyses conducted in this research, putative interneurons were excluded from the neuronal population, but it should be noted that including these interneurons yielded similar reactivation values (EV: 9.4±2.7%, REV: 1.9±0.7%; p<0.01). Analysis of the temporal dynamics of reactivation in 20-min blocks of concatenated rest revealed a gradually decaying reactivation which was significant for at least 1 h of postbehavioral rest (Figure 2C). Within periods of rest, reactivation was prominent especially during quiet wakefulness–slow-wave sleep episodes (QW-SWS; n = 13 sessions), but it was not significant for rapid eye movement (REM) sleep (Figure 2D). The lack of pattern recurrence during REM sleep was not attributable to its relatively short duration, undersampling of spikes, or its late occurrence after sleep onset compared to QW-SWS (Figure S3; Table S2; Text S1). Reactivation in the HC [36] and VS [31] occurs markedly during sharp wave-ripple complexes, i.e., short-lasting, high-frequency oscillations in the hippocampal LFP with associated bursts of large-scale neuronal firing that characterize QW-SWS [2],[37]. The same trend was observed for joint reactivation; however, the difference in reactivation values for time windows of 200 ms following ripple onset (“Ripples”) and during 200 ms following the onset of interripple intervals (“Intervals”) did not reach statistical significance, which most likely relates to the high variability across sessions (cf. [31]) (Ripples: EV: 5.9±2.6%, REV: 1.0±0.3%; Intervals: EV: 1.9±0.8%, REV: 0.5±0.2%; EV and between Ripples and Intervals [EV−REV]: n.s.; n = 14 sessions).
The existence of joint HC-VS reactivation raises the question of which physiological and behavioral factors are associated with the strength of this process. We examined three not mutually exclusive factors pertaining to (1) the modulation of the neural activity patterns by theta oscillations, (2) the correlation of neuronal firing patterns with behavioral parameters, and (3) the order in which neurons in different areas were activated. First, we computed the degree to which cells in each pair fired together, and then all of these correlation values per episode were pooled across sessions and animals. We next formed subgroups of cell pairs by partitioning the complete set of correlation values on the basis of the factor under scrutiny. Reactivation values were computed for these subgroups, and statistical significance was assessed by applying a bootstrapping procedure with resampling of pooled correlation values [22],[23].
The two-stage model of memory trace formation posits that theta oscillations are crucial for encoding information in the HC in the awake, active state [26]. The hippocampal theta rhythm may also have a role in governing the temporal organization of activity in target structures to ensure efficient communication [38],[39]. Thus, our first hypothesis holds that HC-VS reactivation will only be strong when information is cross-structurally aligned during encoding by a common temporal framing, the theta oscillation, creating windows of near-synchronous firing.
During track running, robust theta oscillations were observed in the HC and VS. In both areas, cells were observed whose firing patterns were modulated by the local theta rhythm (n = 121 out of 263, 46.0%, in HC, and n = 20 out of 243, 8.2%, in VS (Figure 3A; FigureS4A). Ventral striatal units that were modulated by the local theta rhythm generally showed firing rate modulation also by hippocampal theta oscillations (n = 20, 8.2%). When the peak of the theta oscillation recorded near the hippocampal fissure was taken as synchronizing time point, CA1 cells fired at an average angle of 199.9±6.1° (range: 43.9°–356.4°) and VS cells fired at a slightly later phase (217.1±26.0°, range: 4.5°–336.3°; n.s.). Cell pairs were divided on the basis of modulation by the hippocampal theta rhythm, resulting in four subgroups: Both Cells (n = 140), HC only (n = 1,273), VS only (n = 81), and None Modulated (n = 1,422). Reactivation was observed for all but the VS only groups; its strength was significantly stronger in the Both Cells group compared to the None Modulated and HC only groups (Figure 3A; Table S3).
Our second hypothesis departed from the assumption that spike patterns are not reactivated equally, but are reprocessed especially when they convey behaviorally relevant information. For the HC-VS system, we predicted that cells expressing spatial (HC) and reward-related information (VS) should be preferentially reactivated. Location-specific firing was found for 102 out of 263 (38.8%) hippocampal cells. Place fields were distributed uniformly across the track; there was no indication that place fields occurred more frequently near reward sites or corners of the track. In contrast, a subset of VS neurons fired in close temporal relationship with reward site visits (41 out of 243 cells, 16.9%; Text S1). Reward-related responses were generally increments in firing rate and could be generated at one, two, or all three reward sites. Furthermore, they were often sensitive to either the presence or absence of reward. In line with previous studies on the VS [9],[31],[40], we will apply the term reward-related to all VS units showing significant responses time-locked to reward site visits.
Depending on the expression of place fields and reward-related correlates, cell pairs were grouped in four categories: Double Correlates (n = 192), Place Field only (n = 941), Reward-related Correlate only (n = 287), or No Correlates (n = 1496). The Double Correlates group showed very strong reactivation (EV: 22.9%, REV: 0.1%), whereas reactivation in the other three subgroups in this partition was not significant (Figure 3B; Table S3). Accordingly, the strength of coactivation of a place cell and a reward-related VS cell, expressed in the Pearson correlation coefficient, was positively correlated to the degree of spatial overlap of the firing fields on the track during task performance and postbehavioral sleep, but not during prebehavioral rest (prebehavioral rest: n.s.; track running: R2 = 0.25, p<1×10−12, postbehavioral rest: R2 = 0.03, p<0.02; n = 192).
A long-standing assumption in memory consolidation theory holds that the HC initiates and orchestrates reactivation in its projection areas [25]–[28]. This general process may be realized in several ways (see Text S1). In the HC-VS system, evidence suggests a particular variant of replay in which hippocampal ripples initiate reactivation locally and subsequently trigger a wave of enhanced excitability in the VS [23]. This variant implies that reactivation should be strong when a particular firing order is maintained: during replay, a hippocampal cell should fire predominantly in advance of a VS cell. During behavior, VS firing may also precede HC firing, but this order should not be associated with strong reactivation. The HC→VS order would also be consistent with the unidirectionality of the projection from HC to VS [13]. Despite the finding that sleep reactivation occurs in a “forward” direction, meaning that the order of firing during sleep is similar to the order during the preceding behavior [29],[30],[41], this critical assumption has yet to be confirmed or refuted. Hence, our third hypothesis holds that reactivation is strong when the information flow is organized according to a leading role of the HC.
The firing order of each cell pair was assessed by computing cross-correlograms [42],[43] and determining which order of firing was most prevalent using a “temporal bias” measure [29]. Three subgroups were distinguished, i.e., HC→VS pairs (n = 608), VS→HC pairs (n = 796), and No Clear Order, which included pairs that did not show a preferred firing order (n = 1,512). The HC→VS group reactivated strongly (EV: 15.2%, REV: 0.0%). Reactivation was also observed for the other groups, although the observed strengths were significantly lower than for the HC→VS group. (Figure 3C; Table S3).
Variations in reactivation measured across all of the subgroups partitioned according to each of the three factors could not be attributed to differences in varying numbers of cell pairs, differences in correlation strengths, or differences in spike counts (Text S1). Altogether, these results suggest that all three factors analyzed—modulation of both cells by the hippocampal theta rhythm, maintenance of the HC→VS firing order, and expression of a combination of a place field and a reward correlate—are associated with strong reactivation. However, since a reactivating cell pair may display multiple characteristics at the same time (e.g., behavioral correlates and a particular firing order; see Figure S4), we used a multilinear regression model to test whether the contribution of each cell pair to the session EV value was dependent on firing order, theta modulation, behavioral correlates, or any combination of these characteristics. First, the relative contribution of each cell pair to the session reactivation was estimated by excluding a pair from the simultaneously recorded population and recomputing the reactivation values. The difference between the session EV minus the EV after pair exclusion represents the estimated contribution of that pair to the session EV. Multilinear regression showed that both the expression of a double correlate and the HC→VS firing order were significant factors in explaining the contribution to the session EV (p<0.02 and p<0.002, respectively; theta modulation was not significant, p = 0.6). The combination of firing order and expression of a double correlate predicted the pair's contribution better than either one alone (p<0.0002). This analysis confirms the importance of the HC→VS firing order and expression of combined place and reward information during track running for subsequent reactivation and identified theta modulation as a less significant indicator.
We tested whether reactivation in the subgroup reactivating most strongly, i.e., the Double Correlates, was sparsely distributed as we previously showed for VS ensembles [31]. First, we assessed the contribution of each cell pair to the reactivation as explained above. To find an indication of how many cell pairs can be excluded to abolish reactivation, the pairs were sorted in descending order in terms of their contribution to the reactivation value [EV−REV]. Then the pairs were excluded one by one in a cumulative fashion starting with the highest contributor from the population, and reactivation values were computed each time a next pair was excluded. If the 17 (17/192, 8.9%) most contributing cell pairs were left out of the population, the [EV−REV] dropped below 5.0%. A total of 34 (17.7%) pairs could be removed before the [EV−REV] level decreased to 0.0%. This analysis indicates that, consistent with VS ensembles, reactivating cell pairs were also sparsely distributed in the HC-VS population.
We next explored whether HC-VS cell pairs fire in the same order during reactivation as during behavior and whether replay is accelerated relative to active behavior. For each pair of neurons that showed a place field and a reward-related correlate, we constructed three cross-correlograms, one for each task-episode (n = 192, Figure 4A). We compared the time offsets during active behavior and rest for pairs that showed significant peaks in the cross-correlograms of track running and in at least one of the rest episodes (n = 53, 27.6%). The time offsets of the peaks during track running were positively correlated to those in postbehavioral rest (R2 = 0.09, p<0.05; n = 47), but not to those of prebehavioral rest (Figure 4B; n = 26). Thus, the recurrent firing patterns reflected the preceding experience. In this analysis, spatial overlap between the firing fields of a cell pair turned out not to be a prerequisite for concurrent firing during subsequent sleep, as 29.8% of the cell pairs that showed peaks in the cross-correlograms for task performance and postbehavioral rest exhibited nonoverlapping firing fields on the track. The peak offset in the cross-correlograms of track running ranged from −4.5 to 3.8 s and was significantly correlated to the spatial distance between the firing fields (R2 = 0.27, p<0.001).
To determine whether the order of firing on the track was preserved or reversed in the subsequent rest episode the offset sign (+ or −) of the cross-correlogram peak relative to zero was considered. Peaks during track running and postbehavioral rest were consistently found with the same offset sign (43/47 = 89%, sign test, p<0.0001), which demonstrates that replay took place in a forward direction. In combination with the strong reactivation of cell pairs that exhibited the HC→VS firing order during track running observed in the subgroup-based reactivation analysis (Figure 3C), the preservation of firing order suggests that replay should be dominated by activity patterns in which HC firing largely precedes VS firing, both during track running and postbehavioral rest. Indeed, in the large majority of cell pairs that showed forward reactivation, the hippocampal cell fired preferentially before the striatal cell during both periods (36/43 = 83.7%), indicating that most of the reactivating cell pairs express a HC→VS order during behavior and sleep (sign test, p<0.0001). Thus, not only is the firing order preserved from the behavior to ensuing sleep, but apparently the HC also takes the lead in replay and the VS follows.
An additional analysis on all cell pairs with significant cross-correlogram peaks yielded similar results and confirmed that the preferential firing order during reactivation was not attributable to a lack of VS→HC correlations during track running (see Text S1). Like cross-structural reactivation, reactivation within hippocampal and ventral striatal ensembles also took place in a forward direction (see Text S1) (cf. [29] for HC).
Replay may occur at a different time scale than applicable during behavior [29],[30],[41]. We examined whether joint HC-VS firing patterns were replayed on an accelerated time scale. Peak times in the postbehavioral rest cross-correlogram occurred significantly closer to zero than during track running (track: −525.5±201.9 ms, postbehavioral rest: −53.2±28.5 ms; p<0.01; n = 47), showing an approximately10-fold compression (Figure 4). Replayed patterns appeared compressed and not merely truncated, because the shape of the cross-correlogram peaks with offsets of up to several seconds during behavior were re-expressed during sleep, including their offset from zero (Figure 4A).
Altogether, our results demonstrate coherent reactivation between the HC and a subcortical structure, and identify two major factors governing cross-structural reactivation in the HC-VS system, suggesting a plausible mechanism for consolidation of associative place-reward information. The first factor that significantly correlated to strong HC-VS reactivation bears on the dependence of reactivation on the coding of behaviorally relevant information. Cell pairs that exhibited a double correlate—one place field plus one reward-related correlate—showed the strongest reactivation among all four subgroups. In addition, the contribution to reactivation by individual pairs depended specifically on such a coexpression of behavioral correlates. The near-synchronous reiteration of spatial and motivational information during sleep may serve to integrate these types of information and support the learning of place-reward associations. Such associations are essential to predict and localize desired food and liquids within a known environment and are therefore fundamental to foraging behavior and learned behaviors such as conditioned place preference [5],[19],[44]. Like intra-area ventral striatal reactivation [31], cross-structural replay is dependent on a relatively small subset of cell pairs, indicating that it is a sparsely distributed phenomenon. If replay indeed supports memory consolidation, the formation of associations of a specific place-reward combination is likely to depend on a small minority of cells in the HC-VS circuitry.
The second significant factor in joint HC-VS replay is the preferred firing order of HC and VS cells, consistent across track running and subsequent sleep. The HC→VS firing order during track running was associated with a significantly elevated reactivation as compared to other temporal relationships, and the cell pair contributions to reactivation depended on this specific firing order. This organization of firing order obeys the direction of the anatomical projection [13],[45] and presents necessary, although not sufficient, evidence for a central tenet of consolidation theory, proposing the HC to initiate reactivation in its target structures, as predicted by Marr [25] and subsequent theorists [26]–[28] .
Our data provide several indications supporting that the observed cross-structural reactivation is the consequence of a coordinated process between the HC and the VS rather than of two separately, or coincidentally, reactivating ensembles. First, the temporal relationship between a pair of task-related hippocampal and ventral striatal cells was relatively consistent across task phases, resulting in significant peaks in the cross-correlograms of a substantial number of pairs both during track running and postbehavioral rest. If joint reactivation was just coincidental, the temporal firing relationship between cells in different structures is expected to be random, contrary to what was observed. Second, the timing of the peaks during track running and postbehavioral rest was correlated in the cross-correlogram analysis (Figure 4B), and furthermore, the time scale of sequential activation of firing patterns during postbehavioral rest was compressed compared to track running. Thus, temporal firing relations were consistent across different overall brain states (awake active versus SWS) and on accelerated time scales. Observing such results is very unlikely if the two structures would be reactivating without a systematic temporal relationship. Third, during behavior we identified pairs that fired in the order HC→VS and also in the order VS→HC. During reactivation in the postbehavioral rest, however, we found an overrepresentation of HC→VS pairs. If two ensembles reactivated independently, one would expect the ratio of HC→VS and VS→HC pairs to be similar during behavior and reactivation.
An important finding is that the joint reactivation is compressed by a factor of ten compared to the behavioral time scale of neuronal activation. Thus, at least several seconds of “real-time” joint place-reward information during behavior are brought together in a time frame of hundreds of milliseconds during sleep. This further supports the plausibility of a mechanism for the associative storage of place and reward information by way of synaptic weight changes in the HC-VS system. If a cell from the hippocampal formation, coding place, fires consistently and briefly in advance of a VS cell signaling reward (Figure 4), spike timing–dependent plasticity may be induced in their connection [46],[47].
Cross-correlogram analysis revealed that joint reactivation is not restricted to neuronal pairs that exhibit overlapping firing fields; peaks that were separated by up to about 4.5 s during behavior were found to recur during postbehavioral rest. In a scenario in which a series of place fields is followed by a reward-related correlate, this indicates that value information during SWS is not only paired with locations nearby, but also with more remote stages of the path leading to the reward site. Formation of reward-predicting representations should, by definition, obey the temporal order of predictor-reward events, a requirement that is met by the preferential HC→VS firing order during replay. In principle then, the characteristics of hippocampal-striatal replay are suitable for mediating the “backwards” association between reinforcements and cues and contexts situated progressively earlier in time. This temporally backwards referral is a key feature of conditioning theory and models of reinforcement learning [48]–[50].
Although the causal role of ensemble reactivation in memory consolidation remains to be proven, the temporally ordered cross-structural replay of spatial and motivational information during sleep illuminates a plausible offline mechanism by which information processed in different parts of the brain can be integrated to enable the composition and strengthening of memory traces comprising various attributes of a single learning experience.
All experimental procedures were in accordance with Dutch national guidelines on animal experimentation.
Four male Wistar rats (375–425 g; Harlan) were individually housed under a 12/12-h alternating light-dark cycle with light onset at 8:00 am. All experiments were conducted in the animal's inactive period. During training and recording periods, rats had access to water during a 2-h period following the experimental session, whereas food was available ad libitum. Rats were chronically implanted with a microdrive [51] containing five individually movable tetrodes directed to the dorsal hippocampal CA1 area (4.0 mm posterior and 2.5 mm lateral to bregma) and seven to the VS (1.8 mm anterior and 1.4 mm lateral to bregma) [52]. Reference electrodes were placed in the corpus callosum dorsal to the HC, and near the hippocampal fissure. A skull screw inserted in the caudal part of the parietal skull bone served as ground.
Unit activity, local field potentials, and position data were acquired on a 64-channel Cheetah recording system (Neuralynx). Spike sorting was performed offline using custom cluster-cutting software as described in Text S1.
Recordings of hippocampal CA1 neurons were made from 103 locations between 2.6 mm and 4.8 mm posterior and between 1.2 mm to 2.8 mm lateral to bregma compared to an atlas of the rat brain [52]. Ventral striatal tetrodes were situated between approximately 2.2 and 1.2 mm anterior to bregma and between 1.6 and 3.0 mm laterally. From a total of 140 recording sites, 58% was estimated to be situated in the core region and 42% in the shell region of the VS. Although most sessions were likely to contain recordings from both the core and shell region, six sessions were identified to contain core-only recordings. No gross differences were observed in the number, firing rate, or appearance of behavioral correlates that were estimated to be recorded from the core and the shell region. Moreover, cross-regional reactivation was observed for the core-only sessions, with EV and REV values similar to those observed for other sessions. Therefore, core and shell recordings were pooled.
Pre- and postbehavioral rest episodes included all periods of at least 20 s in which the rat was in the flower pot and remained motionless; i.e., episodes of movement were excluded from analysis. Within these periods of rest, episodes of SWS were characterized by the presence of large irregular activity and the occurrence of sharp wave-ripple complexes in the LFP of the CA1 pyramidal layer [2],[37]. Ripples were detected each time the squared amplitude of the filtered LFP trace (100–300 Hz) crossed a threshold of 3.5 standard deviations (SD) for at least 25 ms. Because incidentally short periods of quiet wakefulness may have been included in SWS episodes, as these two phases share principal LFP characteristics, this state is referred to as quiet wakefulness–slow-wave sleep (QW-SWS). REM sleep periods were indicated by an elevated ratio (>0.4) of spectral density in the theta band (6–10 Hz) to the overall power of the LFP trace recorded near the hippocampal fissure. Their borders were refined upon visual inspection of the trace.
Modulation of a cell's firing pattern to the theta oscillation was determined by first filtering LFP traces recorded from the hippocampal fissure and the VS using a Chebyshev type-1 bandpass filter between 6 and 10 Hz. Binned spikes (10°/bin) were then plotted relative to the theta peaks of two successive theta periods. The spike distribution was considered nonuniform when the Rayleigh score was <1×10−5. The phase angle of the spikes was determined by computing the Hilbert transform of the filtered theta signal. Firing of a unit was considered as being modulated by the theta rhythm when shuffling of the spikes abolished the nonuniformity of the spike firing distribution as assessed with the Rayleigh score.
To characterize spatially selective firing fields, instantaneous firing rates were computed for bins of 50 ms. The spatial position of the rat's head was determined by creating a one-dimensional representation of the track and using a resolution of 2.3 cm. Mutual information was computed between the binned spike trains and the position, and corrected for finite sampling size [56],[57]. A cell was considered to express a place field if its firing rate during track running was at least 0.3 Hz and if it carried at least 0.25 bits/spike of spatial information.
Peri-event time histograms were constructed for the rewarded and nonrewarded condition for each reward site and were synchronized on crossings of offline installed “virtual photobeams” positioned at the points where the rat was just reaching the reward sites. Reward-related responses were assessed within a period of 1 s before and 1 s after arrival at a reward site, using a bin resolution of 250 ms. Spike counts in the eight bins comprising the reward period were each compared to three separate control bins taken from the corner passage opposite to the well under scrutiny within the same lap. A bin of the reward period was only considered significantly different when the Wilcoxon matched-pairs signed rank test indicated significance from each of the three control bins (p<0.01). A reward-related response comprised one or more bins that were significantly different from control bins. Control bins did not show marked deviations from overall baseline firing as checked in peri-event time histograms and plots of the spatial distribution of firing rates. Differences between the responses at different reward sites were assessed with a Kruskal-Wallis test (p<0.05) followed by a MWU (p<0.05), whereas rewarded versus nonrewarded conditions were compared using MWU (p<0.05).
Cross-correlograms were constructed according to Perkel et al. [58] and Eggermont [43]. Spikes were binned into 10- or 50-ms intervals, and the cross-correlation was examined across at least three time windows; viz. [−500, 500] ms, [−2,000, 2,000] ms and [−5,000, 5,000] ms. The firing order of pairs of hippocampal and striatal cells was assessed with the temporal bias method [29] using cross-correlograms with the spikes of the striatal cell serving as reference. The ordinate expressed spike counts per second, which was integrated across intervals of 200 ms before (I) and after (II) zero. The difference between II minus I divided by the sum of the spike counts determined the temporal bias score. If this score was negative, the HC was determined to fire preferentially before the striatal cell. If this score was positive, the preferred firing order was in the opposite direction. The classification No Clear Order was assigned when the scores (I and II) were approximately equal or when the cross-correlogram did not have a clear single maximum.
To estimate the significance of peaks in the cross-correlograms, the mean expected number of joint spike counts μ and the levels of μ±3 SD (corresponding to p = 0.0013) were computed to provide indications for nonrandom excursions of spike counts above or below the expected range [43]. Each cross-correlogram was then subjected five times to a spike-shuffling subtraction procedure [42],[58]. Peaks were accepted as significant only when they exceeded the +3 SD threshold above the mean in each of the five repetitions.
|
10.1371/journal.pgen.1006735 | Parallel reorganization of protein function in the spindle checkpoint pathway through evolutionary paths in the fitness landscape that appear neutral in laboratory experiments | Regulatory networks often increase in complexity during evolution through gene duplication and divergence of component proteins. Two models that explain this increase in complexity are: 1) adaptive changes after gene duplication, such as resolution of adaptive conflicts, and 2) non-adaptive processes such as duplication, degeneration and complementation. Both of these models predict complementary changes in the retained duplicates, but they can be distinguished by direct fitness measurements in organisms with short generation times. Previously, it has been observed that repeated duplication of an essential protein in the spindle checkpoint pathway has occurred multiple times over the eukaryotic tree of life, leading to convergent protein domain organization in its duplicates. Here, we replace the paralog pair in S. cerevisiae with a single-copy protein from a species that did not undergo gene duplication. Surprisingly, using quantitative fitness measurements in laboratory conditions stressful for the spindle-checkpoint pathway, we find no evidence that reorganization of protein function after gene duplication is beneficial. We then reconstruct several evolutionary intermediates from the inferred ancestral network to the extant one, and find that, at the resolution of our assay, there exist stepwise mutational paths from the single protein to the divergent pair of extant proteins with no apparent fitness defects. Parallel evolution has been taken as strong evidence for natural selection, but our results suggest that even in these cases, reorganization of protein function after gene duplication may be explained by neutral processes.
| Parallel evolution of protein domain organization following gene duplication has been demonstrated in the spindle checkpoint pathway leading to the hypothesis that this organization is likely to be adaptive. We test this hypothesis by reconstructing budding yeast strains with a spindle checkpoint pathway containing a protein with ancestral domain organization, and systematically perform stepwise duplication, degeneration and complementation of the duplicated protein. We show that, under laboratory conditions where the spindle checkpoint pathway is necessary for growth, degeneration of the ancestral pathway organization to the extant sub-functionalized proteins is consistent with a neutral model of duplication-degeneration-complementation.
| Gene duplication and sub-/neo-functionalization are processes that both increase genomic complexity and are thought to be the major sources of genetic novelty in organisms [1]. Models seeking to explain the retention of paralog genes include relaxation of selection constraints leading to functional novelty through transcriptional changes, splicing divergence or functional repartitioning of protein domains. Particularly interesting are cases of repeated, parallel evolution of these paralog genes that has been taken to be strong evidence for natural selection (several examples reviewed in [2]). In most model organisms, the spindle checkpoint pathway includes the paralogous Bub1 and Mad3 proteins. Surprisingly, although Bub1 and Mad3 are highly diverged, it was recently reported that this paralog pair has arisen from at least nine independent duplication events throughout the tree of life [3]. Perhaps even more striking is that in each event, the duplication almost always leads to a Bub1 homolog, containing a kinase domain and a Mad3 homolog containing a pseudo-substrate APC inhibitor motif [3]. This reorganization of protein function has therefore been thought to be adaptive, leading to a more complex spindle checkpoint pathway [3,4]. However, theoretical work suggests that a causal link between increased genetic complexity and adaptation may not be as clear as commonly assumed [5,6]. For example, if degeneration and complementation of the ancestral bi-functional protein must always lead to Bub1 and Mad3 division of function (i.e., kinase function separated from the pseudo-substrate inhibitor motif), then the repeated observation of Bub1 and Mad3 protein organization might be due to a rate of degeneration that is higher than the rate of reconstitution.
Whether sub-functionalization of a bi-functional protein is adaptive or neutral cannot be easily distinguished by sequence analysis alone. Both models predict complementary conservation of function. However, precise quantitative fitness measurements in organisms with short generation time, such as yeasts, can be used to address questions about these two models of sequence evolution (adaptive vs neutral). Using comparative approaches that delineate functional regions of proteins, we found that this gene duplication leads to sequence signatures that indicate repartitioning of the ancestral function in the extant paralogs. To test whether this increase in complexity (from a pathway containing a bi-functional gene to a pathway containing two specialized genes) in the spindle checkpoint is adaptive, we employ high-throughput quantitative fitness measurements in laboratory conditions stressful for the spindle checkpoint pathway and we systematically dissect the reorganization in protein architecture of an ‘ancestral’ bi-functional gene. To our surprise, we could not detect any increase in fitness for the stepwise functional reorganization (duplication, degeneration and complementation (DDC)).
Preservation of duplicate genes as explained by the duplication-degeneration-complementation model is a neutral process by which repartitioning of the functions of the ancestral protein occurs within the two extant duplicates [7]. Although sub-functionalization is a neutral process, it can lead to adaptation in the case of adaptive conflicts: mutations that are precluded from occurring in the ‘ancestral’ gene but are adaptive when the functions of the ancestral gene are separated. This resolution of adaptive conflicts can sometimes explain the fitness benefit of mutations that ‘specialize’ multifunctional proteins [8]. Nevertheless, complementation by functional repartitioning or complete functional preservation in one of the two copies is still expected in the cases where mutations leading to neo-functionalization have occurred.
It has previously been shown that Bub1 and Mad3 protein reorganization had occurred several times over the eukaryotic tree of life, leading to similar sequence profiles (Fig 1A, and see [3]). To obtain an amino acid resolution view of the evolution of the paralogous proteins, we performed sequence analysis of the duplication that occurred in the whole-genome duplication of budding yeasts ([9], see Methods). Briefly, a phylogenetic hidden Markov model is used to identify short or large regions of conservation relative to the flanking amino acid sequence. This algorithm is applied to single copy proteins from species that diverged before the duplication to identify short motifs or domains that are under selective constraints. We then map these regions to the duplicates and test whether there is evidence of relaxation in constraints in the clade post-duplication using likelihood ratio tests. This analysis revealed several protein regions, in addition to the KEN boxes and kinase domain identified in previous studies [3], that have repartitioned from the ancestral sequence to the paralogs (Fig 1B).
This more detailed view of the changes in constraints allowed us to propose why some of the changes were correlated (S1A Fig for an example). For example, Mad3 is known to interact with Cdc20 via its N-terminal KEN box motif and its tetracopeptide repeat domain (TPR) [10]. According to 3D structural information of Mad3 (S1B Fig), specific residues in the TPR domain (shown in yellow in S1B Fig) appear to contact the KEN box (shown in red in S1B Fig) and may stabilize the interaction of Mad3 with Cdc20 (S1C Fig) [11]. Bub1, on the other hand, does not possess the N-terminal KEN box but still requires the TPR for proper function [12]. We can detect changes in constraints on these specific residues of the TPR of Bub1, suggesting that the degeneration of residues in the TPR domain or the KEN box will disrupt the same function (binding to Cdc20) and that loss of selection constraints on that function will lead to degeneration of both. Indeed, the same correlated changes in constraints exemplified here are also observed in the mammalian, drosophila, and fission yeast Bub1 and Mad3, which occurred through independent gene duplication events (S1D Fig). The changes in constraints at these specific residues are unlikely to disrupt the other functions of the TPR.
On the other hand, the pattern of evolution on the ABBA motifs in the yeast paralogs is more complicated. The ABBA motif is known to be required for binding to Cdc20 [13], and indeed, both Bub1 and Mad3 retain at least one copy of the ABBA motif (Fig 1B). Nevertheless, we can clearly detect motif turnover in the first ABBA motif in Bub1 (Fig 1B). This suggests two possible explanations: 1) the two ABBA motifs serve the same function and the loss of the first motif was compensated in the Bub1 protein by another motif, 2) the two ABBA motifs may bind Cdc20 for different functional reasons and the Bub1 protein underwent a single change in constraint on this motif. Since we have not identified a newly conserved motif in the Bub1 protein (which could compensate for the loss of the first ABBA motif), we believe that the second possibility is more parsimonious.
Consistent with important functions for the domains identified in the ancestral fungal protein, there were no conserved regions that were lost in both paralogs (Fig 1B). The repartitioning of functional regions suggests that the duplication lead to sub-functionalization. We therefore sought to experimentally verify that sub-functionalization had occurred in the paralog pair. As a proxy for the ‘ancestral gene’, we obtained the gene from Lachancea kluyveri which diverged prior to the whole-genome duplication event (see Discussion). We refer to this gene as the ‘single-copy protein (SCP)’, and transformed it into S. cerevisiae.
We first assessed the localization of the single-copy protein because it was known that Bub1 and Mad3 localize to different subcellular compartments: Bub1 is localized to the kinetochores in a Mps1 dependent manner during specific phases of the cell-cycle [14] and Mad3 is constitutively localized in the nucleus [15]. If Bub1 and Mad3 were products of a sub-functionalization event, we would also expect the L. kluyveri protein to localize to both subcellular compartments. To test this, we tagged the three proteins with GFP and visualized their localization by fluorescence microscopy. To account for possible differences in growth or imaging conditions that may influence the apparent localization of the proteins, we designed an assay where the tagged single-copy protein could be visualized alongside the S. cerevisiae tagged protein in the same field of view using the identical GFP tag. Briefly, we expressed additional cytoplasmic fluorescent proteins of other colors (non-GFP) that allow us to differentiate which cells carried the L. kluyveri protein fusion or the S. cerevisiae fusion and obtained fluorescent micrographs of mixed cultures (see Methods). As expected, we observed a distinct localization pattern for the single-copy protein, which was quantitatively different from either Bub1 or Mad3 (Fig 2A). Upon closer inspection, we found that the localization of the single-copy protein appears to be a mixture of the localization patterns of Bub1 and Mad3. Consistent with this, when ordering by bud size as marker of cell stage, we observed that the L. kluyveri protein localized constitutively to the nucleus, and also showed a punctate localization within the nucleus in the early cell-cycle (Fig 2B).
Because Bub1 and Mad3 are required for the spindle-checkpoint pathway, we tested whether the single-copy protein could rescue defects of cells lacking Bub1 or Mad3 with respect to the functionality of the pathway. To do so, we took advantage of a strain carrying a non-essential chromosome containing the ochre suppressor SUP11 [16]. This strain forms red-pigmented colonies upon loss of this chromosome due to the presence of the ade2-101 allele. As has been reported previously [17], we found that strains without a functional Bub1 gene have a very strong increase in chromosome loss rate (29/65 vs 3/131 sectored colonies, p-value < 0.0001). However, we were unable to detect an increase in chromosome loss rate for strains lacking a functional Mad3 gene (5/177 vs 3/131 sectored colonies). Nevertheless, cells lacking Bub1 and Mad3, but carrying the single-copy gene driven by the Bub1 promoter in this strain completely abolished the Bub1 chromosome segregation fidelity defect (3/207 vs 3/131 sectored colonies, Fig 3A). We next sought to verify if the single-copy protein could rescue the growth defects of impaired cell-cycle functions of cells lacking Bub1 and Mad3. Cells lacking Bub1 or Mad3 are highly sensitive to benomyl, a microtubule destabilizing drug [18], because cells fail to correctly detect mitotic spindle attachments. If the single-copy protein can perform the functions of both Bub1 and Mad3, we expect that the single-copy protein would rescue the fitness defect of cells lacking Bub1 or Mad3. If the phenotype is not fully rescued, it may indicate neo-functionalization and adaptation in the Bub1 or Mad3 protein (or it may be due to an artifact of expressing a heterologous gene). To test this, we performed spot dilution assays and found that, if expressing the single-copy protein, cells lacking Bub1 and Mad3 can grow in the presence of benomyl with similar growth characteristics to wild-type S. cerevisiae cells (Fig 3B). We could not simply test the function of the spindle checkpoint in other species as we observed that yeasts other than S. cerevisiae were highly resistant to benomyl (S2 Fig).
At the limits of the resolution of this plate assay (we estimate that this assay can detect at the minimum only a 5% growth difference), and taken together with the localization data, these results suggest that the single copy protein can rescue Bub1 and Mad3 function in S. cerevisiae. Consistent with the DDC model, the sub-functionalization of the Bub1 and Mad3 proteins may confer no fitness advantage to S. cerevisiae.
It is possible that the spot dilution assay did not have enough resolution to detect the fitness advantage of the reorganization of protein function in the paralogs. To address this, we designed a method to more precisely quantify relative selection coefficients (s, see Methods). Briefly, our assay is a competitive fitness assay that takes advantage of flow-cytometry to provide relative counts of fluorescently labeled cells within a growing population [19] and this assay can be performed in a high-throughput fashion where the combination of alleles of interest are tested in cellular backgrounds expressing different fluorescent proteins as experimental replication. Replicate strains can be rapidly created by the SGA cloning strategy (see Methods and [20], S3 Fig). To determine the resolution limit of this fitness assay, we generated 16 spores from a cross between a strain carrying a wild-type Bub1 allele marked with CaURA3 (the URA3 gene from Candida albicans, which complements the URA3 gene from S. cerevisiae) and a query strain (containing SGA mating type reporters) expressing a green or a red fluorescent protein (8 spores for each cross). These spores were competed to form 64 fitness assays. We reasoned that if any additional single nucleotide polymorphisms (SNPs) between the query strains and parental strain of our mutant arrays had a detectable fitness effect (but masked due to epistasis within their respective backgrounds), or if these SNPs showed non-transitive fitness effects, they would be uncovered within these 16 spores. We compared the relative proportions of cells expressing the green fluorescent protein and red fluorescent protein at the 20th generation to the 40th generation and we calculated the selection coefficient for each competition (see Methods). Because our strains are supposedly genetically identical (except for the possibility of non-shared SNPs), we expect an average selection coefficient of zero and the standard deviation obtained from this test can be used to estimate the resolution of our assay (deviations due to growth conditions or to the non-shared SNPs). We measured the selection coefficients of the wild-type strains and observed a mean selection coefficient of 0.00069, with a standard deviation of 0.0017 (Fig 4A). This indicates that the resolution of our assay is in the order of s = 0.0033 (1.96 times the standard deviation) and we believe this represents the difference in growth rate that we can detect. To account for other possible variations that may occur during the course of the study (changes in media, etc) we therefore chose to report as deleterious/beneficial any differences in fitness where both replicates of a competition exceeded a selection coefficient with an absolute value of 0.005 or greater while remaining consistent with all other competitions.
We next compared strains carrying deletions of Bub1 or Mad3, or single-copy protein rescues of these deletions to wild-type strains and found that cells lacking Bub1 have a strong fitness defect when growing on benomyl while cells lacking Mad3 have a more moderate defect (s<-0.3 and s = -0.01317 respectively, Fig 4A). Remarkably, cells lacking both Bub1 and Mad3, but expressing the single-copy protein at the Bub1 locus, retain wild-type growth rate when challenged with the same benomyl concentration (s = 0.002, Fig 4A). Thus, even at the much higher resolution of these quantitative competition experiments, we find no evidence that the duplication and reorganization of the Bub1 and Mad3 proteins (Fig 1) confers a fitness benefit.
The experiments above show that there is apparently no fitness advantage to the Bub1/Mad3 protein organization relative to the single-copy protein. However, the real evolutionary trajectory almost certainly did not replace the single-copy protein directly with the fully formed Bub1 and Mad3. Instead, a series of evolutionary events (likely separated by millions of years) probably occurred from the single-copy protein to the extant paralogs. It is possible that certain steps along this path, particularly at the beginning after the duplication, were advantageous and drove the reorganization of the paralogs.
Because there is insufficient phylogenetic resolution to infer the exact order and number of mutations, based on previous knowledge of the functional elements found in Bub1 and Mad3 (Fig 1), we created strains with genetic make-up of possible evolutionary intermediates that correspond to stepwise mutational events during protein function reorganization. The evolutionary paths assayed include mutations at multiple loci, and therefore possible paths were created using several rounds of the SGA cloning strategy (see Methods). Briefly, we sought to test single and double mutations of the KEN boxes on the SCP at the BUB1 locus, loss of kinase of the SCP at the MAD3 locus, non-functionalization of the gene, or “evolution” to the extant protein. These mutations represent key evolutionary steps from the ancestral protein to the extant Bub1 and Mad3 (Fig 5). We introduced these mutations into the SCP and cloned them individually into different starting strains (for example, we generated a strain containing the SCP with a mutated KEN box at the BUB1 locus). To combine them, query strains carrying the SGA markers and different fluorophores were crossed to the library in an ordered array and selected such that the final products of several rounds of mating were otherwise genetically identical haploid spores carrying different combinations of marked alleles. We then performed an all-by-all competitive fitness assay, and found that genotypes cluster in three distinct fitness classes, which correspond to cells lacking Mad3 function (Δmad3-like, with an average selection coefficient of -0.015 relative to wild-type), cells lacking Bub1 function (Δbub1-like, with an average selection of <-0.3 relative to wild-type (fitness effects larger than -0.3 are not measurable in this assay and so we consider these genotypes to have fitnesses <-0.3, see Methods), or cells with wild-type phenotype (WT-like, with an average selection coefficient of 0.0026 relative to wild-type) (Fig 4B). Although we have not explored the fitness landscape exhaustively, this sample of genotypes suggests that it is made up of three distinct plateaus.
To assay possible paths through the fitness landscape that lead to sub-functionalization, we focused on genotypes that differed from each other by one ‘evolutionary step’. For example, we consider the initial duplication as one evolutionary step (such that we compared the fitness difference of a strain containing one vs two SCP), as well as individual losses of functional motifs. Although in principle it is possible for a non-functional genotype to revert, we consider these events to be rare and therefore have not considered them for simplicity. The results of our analysis are displayed in Fig 5, as an evolutionary landscape with multiple evolutionary paths that ‘travel’ through possible intermediate genotypes. Interestingly, we were able to find at least one path without a detectable fitness defect consisting of at least three degenerations (Fig 5, path through white nodes). If we only consider paths that do not go through nodes of fitness defects (i.e. we do not allow crossing of fitness valleys), then the analysis suggests that the extant network in S. cerevisiae is an absorbing state (see Discussion).
Although we cannot be certain that evolution has taken any of the paths studied here, that we can find at least one seemingly neutral evolutionary path, at the resolution of our assay, strongly supports the DDC hypothesis that the reorganization of protein domains in the Bub1/Mad3 paralogs can be explained by neutral degenerative mutations (Fig 6 and see Discussion).
This model also predicts that the initial neutral step in the process (the duplication) is reversible, and that the mutations must be relatively common to explain the frequency at which the reorganization occurs during evolution (Fig 1). Consistent with this, we have identified at least one phylogenetic clade where the gene duplication in the ancestor of Saccharomyces reverted to the single-copy functional homolog: Vanderwaltozyma polysporus retains a single-copy protein with all the functional elements of Bub1 and Mad3 even though it diverged after the whole-genome duplication (S4 Fig). It is estimated that the ancestor to the lineages leading to V. polysporus and S. cerevisiae had already non-functionalized ~20% of the duplicates [21] suggesting that neither sub-functionalization of Bub1/Mad3 and non-functionalization were rapid.
Neutral processes, such as described by the DDC model [7], have been shown to be important in increasing genomic complexity (see for example [22] for a study on non-adaptive increase in interactome complexity). At the limit of the resolution of our assay (discussed below), we find no evidence that the Bub1/Mad3 protein reorganization after duplication in budding yeast provides a route for adaptive conflict resolution and therefore, further work is necessary to find a fitness advantage over a single-copy protein in the spindle checkpoint pathway.
Consistent with the DDC model, none of the evolutionary intermediates that we consider functional through the comparative genomics analysis showed a fitness defect when tested under laboratory conditions requiring functional spindle checkpoint and when driven by the BUB1 promoter. Nevertheless, there are several caveats to our experiment. First, the resolution of our assay meant that we could only detect fitness effects in the range of s = 0.005. Because of the population size of budding yeasts in nature [23] and our estimated effective population size during the experiment (see Methods), selection could be efficient even on undetected differences in fitness (given a high enough recombination rate). Therefore, it remains possible that the sub-functionalization to Bub1/Mad3 is truly adaptive. However, even if we assume a very small beneficial effect that was not detected, because we performed our assay in the presence of high concentration of benomyl (which was used to characterize all the components of the spindle checkpoint pathway [24]), we believe that under reasonable growth conditions the adaptive effect of this sub-functionalization would be even smaller. Another possibility is that Bub1 and Mad3 participate in completely orthogonal molecular processes to the spindle checkpoint such that our assayed environment would not be able to detect the real functional differences between the duplicate genes and the ancestral single-copy protein. Yet a third possibility is that there exists one context where the effects of the duplication are under much stronger selection. However, if this context exists, it must be rare due to the finding that the phylogenetic clade leading to V. polysporus reverted to the single-copy gene. Ultimately, it is not possible to know the environmental context, nor the genetic context of the ancestral yeast and we cannot rule out that adaptation by escape of an adaptive conflict drove the sub-functionalization of the ancestral protein. Despite these caveats, we note that the functionally relevant motifs in Bub1 and Mad3 were identified in the context of the spindle checkpoint pathway, and this pathway is activated every cell division to ensure proper chromosomal attachment prior to anaphase. We therefore believe that large effects would have been captured even in this laboratory environment.
Although in our study all functional evolutionary paths lead to the same genotype through the same number of degenerations, the probability that a path is taken is dependent on the rate of mutation, the selection coefficient of the intermediates and the effective population size [25]. We here discuss only the scenario of large population size because small population sizes would allow all genotypes, including any potential neo-functionalization, to be effectively neutral. When population size is large and mutation rate is low, then crossing a fitness valley (as measured by our quantitative fitness assay) requires the population to fix each intermediate genotype (this is the deleterious sequential fixation regime discussed in [25]). In this large population size and low mutation rate regime, paths through deeper fitness valleys are essentially never taken because selection is very effective. In this regime, the relative frequencies of the effectively neutral states are entirely dependent on the mutational rate between these states [7]. Because the mutational rate to degenerate is likely to be higher than the mutational rate to re-create a functional element, the evolutionary outcome of Bub1/Mad3 homologs after sub-functionalization could be nearly deterministic. Therefore, the relative probability of observing Bub1/Mad3 functional homologs after duplication is equal to the rate of sub-functionalization divided by the rate of non-functionalization [7]. We propose that the rate of sub-functionalization can be high due to the very small number and mechanistically simple degenerations: we showed here that sub-functionalization and degeneration of the Bub1/Mad3 protein can occur within only two mutations following the gene duplication (generation of stop codon to remove the kinase function in Mad3, and a shift in start position to remove the first KEN box in Bub1), both of which have been observed frequently over evolution in other genes [26,27]. If this sub-functionalization is truly neutral, we propose that the repeated reorganization of the Bub1/Mad3 homologs may be due to the fact that no other possible outcome of the duplication can be easily observed (it is an absorbing state), such that the observed genotype is surrounded by fitness valleys or reversion to the single-copy protein. Fixation of the re-organization may have occurred through hitchhiking with other beneficial mutations, a scenario that often occurs with large population size and a high mutation rate [28], or due to changes in gene expression (discussed below).
Previous studies have considered the functional implications of evolution of individual enzymes and protein complexes through gene duplication and divergence (e.g. [29,30]). However, many proteins that function within regulatory networks contain complex multi-domain architectures and disordered regions [31]. In the case of Bub1 and Mad3, numerous functional steps occurred after gene duplication, and it is not possible to reconstruct them at the resolution of single amino acid substitution. Nevertheless, we chose several key evolutionary steps during the functional reorganization and assessed the fitness of those intermediates. We believe that our study represents a practical way forward for studies of the evolution of complex eukaryotic proteins. Unlike other previous studies (e.g. [29]), we have not performed ancestral gene resurrection via gene synthesis, but instead we have chosen a gene from another species which we believe is representative of the ancestral allele. Our approach has several advantages. First, it is more likely that the gene is functional in at least one genetic environment. Second, the reconstruction of the ancestral gene might not be accurate in proteins with highly diverged disordered regions. Finally, the gene has evolved for the same period of time as the duplicate genes, providing a direct test of whether adaptation from the ancestral allele is due to the resolution of an adaptive conflict. Similar experimental designs have been used to assay functional differences in whole transcriptional regulatory networks that happen on relatively shorter divergence times [32].
Transcriptional evolution has been shown to be an important aspect of functional divergence after gene duplication [33–35]. An important caveat of our study is that we have not tested the evolution at the promoter, due to the difficulty in finding the functional elements of the promoter region [36]. However, we do have evidence that the Mad3 promoter and the Bub1 promoter are not functionally equivalent as the single-copy protein does not fully rescue the spindle-checkpoint defects when placed at the MAD3 locus (S5 Fig). Those promoter changes, however, are still consistent with a model of degeneration where the Bub1 promoter is similar to the ancestral SCP promoter. On the other hand, it is also possible that after duplication the Bub1 promoter acquires beneficial mutations relative to the ancestral promoter, and these changes drive fixation of the gene duplication. Thus, although there is clear evidence that the promoters of Mad3 and Bub1 have diverged, we did not test the effects of these changes.
Under our tested conditions, our study shows that there is no evidence that the reorganization of protein function was to escape an ‘adaptive conflict’ for the spindle checkpoint pathway in mitotic cells of S. cerevisiae. Our data is consistent with the DDC model and our study suggests that parallel evolution through degenerative processes does not have to be rare or adaptive. Further work will be required to see if this is also the case in other organisms where this reorganization has been observed. This situation is reminiscent to the convergent evolution of holocentric chromosomes across the tree of life [37], and it is still unclear whether it provides an advantage during growth, especially considering the more complex meiotic segregation of chromosomes [38].
Interestingly, although the core spindle checkpoint pathway is conserved in all eukaryotic life, several other differences exist in this pathway [39]. These differences include non-conserved proteins important for spindle checkpoint function (such as p31) or different copy number of paralogous proteins (such as Cdc20 in human). Our study provides experimental techniques to test the step-wise effects of evolutionary changes that have been detected through comparative genomics on multiple loci (such as the ones in the spindle checkpoint). We anticipate that these sensitive quantitative fitness measurements will be useful in the computational modeling of sequence and protein evolution within the context of a complete regulatory network [40,41], as has been performed on other important regulatory networks [42,43].
All strains were derived from either BY4741 or BY4742 using standard yeast genetic techniques or synthetic gene arrays (see next method subsection). The single-copy gene was PCR amplified from purified genomic DNA (Fermentas, #K0512) of L. kluyveri (NRRL Y-12651). All integrations were verified by PCR, and key strains containing the single-copy gene were verified for absence of Bub1 or Mad3 when relevant.
Strain construction for the library of alleles was performed using the same method as described in [44]. SGA query strains were created by transferring the Ste3pr_LEU2 marker from Y8205 into the CAN1 locus of BY4742. Fluorophores with the Ste2pr_LkHIS3 were cloned into the pAN200a plasmid (based on pFA6a [45]) using standard cloning techniques and transformed into the CAN1pr locus using delitto perfetto [46].
Benomyl (10mg/mL DMSO stock) is used at outlined concentrations and added to boiling-hot media until completely dissolved. 5-fluoroorotic acid (5-FOA, 100mg/mL DMSO stock) was used to select against uracil biosynthesis prototrophs [47] and plates were poured at 1g/L 5-FOA final concentration (supplemented with all amino acids, including 72ug/mL uracil). Geneticin (G418) was used at 200ug/mL to select for geneticin resistance.
Cells are grown according to slight modifications to the protocols outlined in [48] as shown as a schematic in S3 Fig. We modified our query strains to have the following cassette integrated at the CAN1 locus: RPL39pr_fluorophore_Ste2pr_LkHIS3_Ste3pr_LEU2. Fluorophores used for our study were yeast-enhanced monomeric green fluorescent protein (ymEGFP) and yeast-mCherry (ymCherry). For the general construction of our strains, a query strain is first crossed with all the desired alleles at a particular locus. Diploid selection is performed by selecting for complementary auxotrophies. Overnight diploid cells from plate patches are then scraped into liquid sporulation media (1% Potassium acetate, 0.005% Zinc acetate) and supplemented with amino acid requirements for diploid strains at 25% of the normal usage and incubated on a roller wheel for three days at room temperature. Usually, about 30% sporulation is observed and 5ul of the mixture is spread or spotted on selection plates that select for MATα and other selection markers such as auxotrophies and drug markers.
We found that modifications to the germination and outgrowth procedure were necessary in our hands to obtain colonies after the SGA procedure. In our hands, addition of lysine to the media greatly enhances the initial outgrowth of spores for all strains that were constructed using our query strains (even the ones that did not express a fluorophore or were lysine prototrophs). Spore outgrowth was normal for standard SGA query strains, indicating that strain specific variation, or heterozygous lys2 deletion strains or homozygous LYP1 affected the outgrowth of our strains (LYS2/Δlys2 LYP1/LYP1 compared with LYS2/LYS2 LYP1/Δlyp1). Therefore, to select for lysine prototrophs, the colonies are replicated to media lacking lysine only after the initial growth. To select for lysine auxotrophy, replica plating was used to isolate colonies that did not grow on media lacking lysine, however alpha-aminoadipate could be used instead [49].
The final strains used in the competitive fitness assay all had the following genotypes: MATα, can1::RPL39pr_fluorophore_Ste2pr_LkHIS3_Ste3pr_LEU2, bub1::allele::CaURA3MX, mad3::allele::KanMX, Δlys2, Δhis3, Δura3, Δleu2.
Protein sequences used for the comparative analyses were from the Yeast Gene Order Browser [50]. These proteins were then aligned with MAFFT [51]. Genomic sequences for BY4741 and BY4742 were obtained from the Saccharomyces Genome Database [52].
To perform our comparative analyses, protein sequences were analyzed using methods described in [9] and by visual inspection. To test for changes in constraints in the duplicate protein, we first predict regions of conservation in the proteins pre-duplication using a phylogenetic hidden Markov model that detects regions that have significantly lower evolutionary rates relative to their flanking region [53]. Having predicted these conserved regions, we can then map them to the duplicate protein and ask whether two rates of evolution (one for the pre-duplication clade, and one post-duplication) better explain the evolution of the selected region as opposed to a single rate. Because the rate of evolution of predicted conserved regions in the ancestral protein is very low, when two rates of evolution better explain the data, it typically implies that the region under purifying selection prior to the whole-genome duplication is now under relaxed constraints after the whole-genome duplication. Partitioning of these losses in selection constraints in the two paralogs is an indication of sub-functionalization. To identify potential new motifs in the post-duplication proteins, the phylogenetic hidden Markov model can be used on the post-duplicate protein and the same analysis for changes in constraints can be performed.
X-ray crystallography files (3ESL: Bub1 [54] and 4AEZ: Mad3 [11]) were obtained from the Protein Data Bank [55] and analyzed using PyMOL [56].
Chromosome loss rate was measured in the classical strain carrying a linear non-essential chromosome in the W303 background [16]. Briefly, strains carrying the ochre allele ade2-101 mutation exhibit a visible red pigment when growing in media with low adenine supplementation. This phenotype is suppressed in strains with a single-copy functional SUP11, which is present in the artificial chromosome. Therefore, chromosome loss rate can be measured by counting the number of red-sectored colonies in agar plates containing low adenine. This assay is performed by plating about 200 colonies, and we report the number of sectored colonies over the number of white colonies after two days of growth.
Quantitative fitness assays were performed using the MACSQuant VYB (Miltenyi Biotec Inc.). Briefly, strains are grown for 48 hours in 5mL of cultures on a rolling wheel. The competitive fitness experiment is started by mixing relatively equal proportion of green cells and red cells in deep 96-well blocks (100ul of a single ymCherry expressing strain and 100ul of a single ymEGFP expressing strain into 600ul distilled water) at a 4-fold dilution. To obtain further dilutions, 20ul of the competition is diluted into 300ul distilled water to form a 16 fold dilution, then 20ul of this dilution is diluted into 300ul of defined media supplemented with amino acids and 100ug/mL ampicillin to form a final 16-fold dilution. The cells are therefore diluted 1024-fold and this operation is performed every 24 hours. Given a conservative estimate of 2*108 yeast cells per mL at saturation, we estimate an effective population size (Ne) of approximately 3.44 * 105. Each screen competes 8 genotypes against the others (64 wells), with an additional 16 wells used as contamination control. We use the diagonal of the competitions as an additional negative control as these were competing genetically identical strains constructed independently with different fluorophores. Competitive fitness assays were performed in synthetic media with 10ug/mL benomyl, which is a condition at which wild-type cells still undergo ten divisions per day but seriously impairs growth of cells lacking spindle checkpoint function.
We defined the relative selection coefficient (s) on the basis of the deterministic continuous time model of logistic growth of an allele against another: dR/dt = sRG, where R and G are the frequencies of red and green cells in the population, t the number of generations, and s the selection coefficient [57]. To simplify the parameter inference, we ignore drift (because the timescale of the experiment is very short compared to Ne), mutational processes that happen during the fitness assay (i.e. we ignore mutations occurring in red or green cells that may alter the lineage trajectories), recombination, and any non-transitive effects. Constraining on G = 1-R, this model describes the logistic equation and has solution [57]:
R(t)=R(0)est(1−R(0))+R(0)est
Because we do not observe the frequencies directly, we estimate the parameters s (and R(0) if needed) by counting each cell sampled by the flow-cytometer as a random variable, taking values of red or green (as it is described for a binomial logistic regression). In total, 50 000 cells are analyzed at the 20th and the 40th generation for each competition experiment and well-defined proportions of green and red cells are gated to remove doublets and relative cell counts are obtained from each competition [58,59]. The likelihood function for this counting process is simply:
L=∏t=1Tnt!rt!(nt−rt)!R(t)rt(1−R(t))nt−rt
Where t are the time points, r the number of red cells counted at that time point, n the total number of cells counted at that time point and R(t) the expected frequency of red cells at time point t. We estimate the initial frequency R(0) and s (due to their relationship with R(t)) by maximizing the likelihood, and the maximum likelihood estimate for the selection coefficient can be obtained by performing a log-linear regression on the ratio of red and green cells in the population:
log(R(t)G(t))=log(R(0)G(0))+st
For more than two time points, the parameters can be estimated by an iterative reweighted least-squares linear regression or with Newton’s method. For two time points, the solution is equivalent to the simple linear regression and we use the same formula as in [59]:
log(r(t2)g(t2))−log(r(t1)g(t1))t2−t1=s
Clearly, if s is positive, then the ratio of red cells to green cells increases at the next generation and the value of the selection coefficient is therefore the selective effect of a beneficial allele as in [60].
The upper limit of detection occurs when we expect fewer than 1 out of 50000 cells of the worst genotype at the 40th generation, which occurs at approximately s = -0.3. In practice, some genotypes can no longer be detected at the 20th generation, and we simply report these as s < -0.3. In some other cases, the number of red or green cells is fewer than 50, which leads to highly inaccurate estimates of the selection coefficient and we also report these as s < -0.3. When the total number of cells counted for both strains in a competition was fewer than 50, we reported s = 0 to mean equally lethal.
All genotypes were made in strains expressing red or green fluorescent proteins, and therefore assayed in two biological replicates. Key genotypes with undetectable fitness effects were further assayed with at least two technical replicates (where the same strains were assayed more than once on different days).
Cells were grown in low-fluorescence media with appropriate auxotrophic requirements to log-phase and imaged using a Leica SP8 confocal microscope. Proteins of interest were tagged with EGFP [61], while the cytoplasm of cells were marked with cytoplasmic mCherry or mTagBFP2 [62] under a constitutive ribosomal promoter. We imaged the green fluorescence first, followed by the cytoplasmic marker to prevent bleaching. This setup enables highly controlled and quantitative analysis of localization because strains with differently tagged proteins can be imaged on the same field of view under identical conditions.
Double-blind quantification of the localization pattern was performed by manually inspecting the green fluorescence localization pattern first and scoring for whole nucleus, kinetochore (punctae), or a mixture of both. The scored cells were then assigned their proper genotype by looking at the red or blue fluorescence channels.
|
10.1371/journal.pbio.1001928 | Bistable Expression of Virulence Genes in Salmonella Leads to the Formation of an Antibiotic-Tolerant Subpopulation | Phenotypic heterogeneity can confer clonal groups of organisms with new functionality. A paradigmatic example is the bistable expression of virulence genes in Salmonella typhimurium, which leads to phenotypically virulent and phenotypically avirulent subpopulations. The two subpopulations have been shown to divide labor during S. typhimurium infections. Here, we show that heterogeneous virulence gene expression in this organism also promotes survival against exposure to antibiotics through a bet-hedging mechanism. Using microfluidic devices in combination with fluorescence time-lapse microscopy and quantitative image analysis, we analyzed the expression of virulence genes at the single cell level and related it to survival when exposed to antibiotics. We found that, across different types of antibiotics and under concentrations that are clinically relevant, the subpopulation of bacterial cells that express virulence genes shows increased survival after exposure to antibiotics. Intriguingly, there is an interplay between the two consequences of phenotypic heterogeneity. The bet-hedging effect that arises through heterogeneity in virulence gene expression can protect clonal populations against avirulent mutants that exploit and subvert the division of labor within these populations. We conclude that bet-hedging and the division of labor can arise through variation in a single trait and interact with each other. This reveals a new degree of functional complexity of phenotypic heterogeneity. In addition, our results suggest a general principle of how pathogens can evade antibiotics: Expression of virulence factors often entails metabolic costs and the resulting growth retardation could generally increase tolerance against antibiotics and thus compromise treatment.
| Scientists have recently realized that nature and nurture are not the only determinants of an individual's traits; some organisms also use random molecular processes to generate phenotypic variation among genetically identical individuals. This raises the question of whether such phenotypic variation could be beneficial and what such possible benefits might be. Working with pathogenic Salmonella bacteria, we discovered that phenotypic variation in one single trait—the expression of virulence genes—provides this pathogen with two critical benefits. First, it leads to the division of labor between different phenotypic variants that allows for effective host colonization, and second, it provides tolerance to antibiotics through a “bet-hedging” mechanism. Our results provide a new perspective on how phenotypic differences between individuals can provide benefits to clonal groups of organisms. At the same time, this study contributes to explaining why some pathogens can evade treatment, and could help to find new and better ways for controlling infectious disease.
| Genetically identical bacterial cells can exhibit remarkable phenotypic differences even when grown in homogeneous environments [1],[2]. These differences can arise from stochastic fluctuations in the expression of individual genes [3]. Although there is evidence that the majority of genes are under selection for tight control of expression [4], some genes are expressed heterogeneously. This raises the question of whether phenotypic heterogeneity can provide benefits and what those benefits might be.
Two possible types of benefits have been proposed. First, heterogeneous gene expression can enable a population to hedge its bets in an unpredictable and fluctuating environment [3],[5]–[7]. In this bet-hedging scenario, one part of the population expresses a phenotype optimized for the current environment, allowing it to survive and reproduce at a high rate. Another part of the population expresses a phenotype less well suited to the current environment, yet it is adapted to a state the environment might change into. Second, phenotypic heterogeneity can promote the division of labor in groups of genetically identical individuals [8]–[10]. This allows a population to perform different functions simultaneously that would be costly or impossible to combine within a single individual. Bet-hedging and division of labor are two fundamentally different adaptive strategies: The benefit of bet-hedging only manifests in fluctuating environments over time; the benefit of division of labor does not require environmental fluctuations to manifest, and the payoff to each subpopulation depends on the interaction with the other subpopulation. Both strategies have been shown independently to play important roles in microbial populations [8]–[13]. Whether heterogeneity in a single trait can promote both functions simultaneously, and how these functions can interact, is an open question. Our aim is to address this question and thereby to gain new insights into the functional complexity of phenotypic heterogeneity.
Here, we present a case where phenotypic heterogeneity in a single trait—virulence gene expression in Salmonella typhimurium—shows characteristics of both strategies, the division of labor and bet-hedging. In S. typhimurium, expression of the type three secretion system 1 (ttss-1) is bistable [14]–[16]. S. typhimurium uses ttss-1 for injecting effector proteins into host cells, promoting penetration of the host tissue. It is, therefore, an important determinant of virulence in this pathogen [17]. It has been shown that the bistable expression of ttss-1 leads to the division of labor among the members of a population. One subpopulation expresses ttss-1 (T1+ cells) and a fraction of those cells invade host tissue and evoke an inflammatory response that is beneficial for the S. typhimurium cells that do not invade [18],[19]. This is thus a special case of “cooperative virulence” where the cooperative behavior is only expressed by a fraction of the population. Recently, it has also been shown that members of the T1+ subpopulation have low cellular growth rates [9],[20]. Slow growth has been associated with tolerance to environmental stresses such as exposure to antibiotics [11],[21]–[25], and the formation of a slow-growing and persistent subpopulation has been interpreted as a typical example for bet-hedging in other organisms [11],[26]. This raised the question of whether the slowly growing T1+ subpopulation is more tolerant to antibiotic exposure than the faster growing T1− subpopulation, so that the formation of these two subpopulations could promote bet-hedging during exposure to antibiotics. Although this question does not imply that exposure to antibiotics was the selective force that might have promoted phenotypic heterogeneity in virulence gene expression, it is interesting to ask whether a bet-hedging benefit under exposure to antibiotics is a potentially very relevant consequence of this heterogeneity.
To investigate whether the two subpopulations—T1+ and T1−—exhibit a difference in susceptibility to antibiotics, we grew clonal populations of S. typhimurium in a microfluidic device that allows single cell observation over extended periods of time under precisely controlled conditions, and quantifying cellular parameters of large numbers of individual cells [27] (Figure 1A shows a temporal montage of two microfluidic channels, and Figure S1 shows a schematic drawing of the microfluidic device). During exponential growth in LB medium, the majority of cells are T1−, which is consistent with previous results [20]. We use filtered medium from late exponential phase cultures grown in LB (“spent LB”) to induce the expression of ttss-1 [20], which, by virtue of its bistable expression, leads to the emergence of two phenotypic subpopulations. The first subpopulation remains T1−, whereas the second induces the expression of ttss-1 (as observed based on a reporter for sicA promoter activity; the sicA promoter controls expression of the sicAsipBCDA operon, encoding key parts of the ttss-1 virulence system). Importantly, and as we will show in more detail below, the T1+ subpopulation pays a cost for the expression of ttss-1, and grows and divides at a slower rate, in line with previously published observations [20]. To test for differential susceptibility of the two subpopulations, we then added 0.05 µg/ml ciprofloxacin. After 3 h of ciprofloxacin exposure, the medium was changed to fresh LB to wash out the antibiotic and to allow growth of surviving cells. Survival of ciprofloxacin treatment showed a positive correlation with single cell GFP intensities (Figure 1B and Figure 1C) and a negative correlation with single cell elongation rates (Figure 1D), and thus with the expression of ttss-1 (Movie S1); T1+ cells, having a growth deficit in the absence of antibiotics, were more likely to survive ciprofloxacin exposure than T1− cells. The bistable expression of ttss-1 can therefore have two functional consequences: In addition to promoting division of labor between two phenotypes, it can also promote persistence of the genotype in the face of fluctuating exposure to antibiotics through a bet-hedging mechanism.
We then carried out two important control experiments. First, we subjected a strain that is genetically avirulent (ΔhilD) to the same experimental conditions as used in Figure 1. HilD is a positive regulator of ttss-1, and deletion of hilD yields a population of fast growing T1− individuals [20]. None of the observed ΔhilD cells resumed division after exposure to 0.05 µg/ml ciprofloxacin (Figure S2), showing that the function of HilD is required to survive antibiotic exposure. Second, we tested whether the tolerance observed in the T1+ subpopulation is due to the acquisition of genetic resistance. We subjected cells that had survived exposure to 0.05 µg/ml ciprofloxacin to treatment with the same antibiotic a second time, without inducing expression of ttss-1 through growth in spent LB. None of the observed cells resumed division after the second antibiotic exposure (Figure S3, Movie S2), indicating that antibiotic tolerance is a phenotypic trait, rather than the result of a resistance mutation.
Next, we tested if our results were specific to antibiotic class and concentration. To test whether the differential killing of the two subpopulations is also observed with an antibiotic from a different class, we treated S. typhimurium cells with 16 µg/ml kanamycin, otherwise using the same experimental conditions as in Figure 1. Kanamycin is an aminoglycoside and has a fundamentally different mechanism of action from ciprofloxacin, a fluoroquinolone. Again, survival of antibiotic treatment was positively correlated with GFP fluorescence and negatively correlated with single cell elongation rates, and thus to expression of ttss-1 (Figure S4, Movie S3). Second, we tested whether the tolerance of T1+ cells plays a role at antibiotic concentrations that are in the range of those measured in patients treated with ciprofloxacin [28] and kanamycin [29], respectively. Using the same experimental setup as above, we subjected S. typhimurium to a ciprofloxacin concentration of 10 µg/ml and a kanamycin concentration of 50 µg/ml. These concentrations are higher than the ones we have used before, and correspond to 200 times the minimal inhibitory concentration (MIC) measured for ciprofloxacin, and 6.25 times the MIC measured for kanamycin (Figure S5). Although more cells died in total when subjected to those higher antibiotic concentrations, survival of T1+ cells was again significantly higher compared to T1− cells for both antibiotics tested (Figure 2 and Movie S4 for ciprofloxacin treatment, and Figure S6 for kanamycin treatment). This partial tolerance of T1+ cells against clinically relevant concentrations of antibiotics could potentially explain the observation of relapsing S. typhimurium infections after treatment [30].
Our results raise the question of whether the growth difference between the T1+ and T1− subpopulations can explain the difference in survival. We tested this in two different ways. First, we grew genetically avirulent ΔhilD cells in chemostats at the two different growth rates observed for the T1− and the T1+ subpopulations, respectively. These growth rates were determined in an experiment where wild-type S. typhimurium cells were grown for an extended period of time in spent LB, and single cell growth rates of T1− and T1+ cells were determined to be 0.96 and 0.26 doublings per hour, respectively (see Materials and Methods for details). The ΔhilD strain allowed us to test the effect of growth rate in a phenotypically uniform population. When treated with 0.05 µg/ml ciprofloxacin, viability counts in the chemostat populations remained largely stable for the slow growth condition (corresponding to the rate at which the T1+ subpopulation grows, 0.26 doublings per hour) over a period of 5 h, whereas viability counts for the fast growth condition (corresponding to the rate at which the T1− subpopulation grows, 0.96 doublings per hour) dropped sharply during the first 3 h and then remained stable at around 1 in 105 cells of the initial population (Figure S7). As a second way to reduce cell growth, we manipulated ΔhilD cells into overexpressing LacZ, a gratuitous protein under the growth conditions used, using an IPTG-inducible promoter on a high copy plasmid [31]. Again, we observed a strong negative correlation between single cell growth rate and survival (Figure S8, Movie S5). Cells lacking the plasmid do not survive when exposed to the same IPTG concentrations (unpublished data). We therefore conclude that the growth rate difference between the T1+ and T1− subpopulations can explain a substantial part of the difference in antibiotic susceptibility, and that the expression of abundant protein upon ttss-1 induction is a plausible reason for this growth deficiency.
The link between virulence gene expression and tolerance against antibiotics that we observe has potential consequences for within-host evolution of virulence. In experimental model systems of cooperative virulence [9],[32]–[34] as well as in a clinical setting [35] it has been shown that genetically avirulent mutants can rise in frequency during infection, leading to improved host condition. If exposure to antibiotics kills phenotypically avirulent cells preferentially, one would expect selection against the emergence of such genetically avirulent mutants. In order to test this, we competed the genetically avirulent S. typhimurium mutant ΔhilD against wild-type S. typhimurium cells in the presence and absence of ciprofloxacin in vitro (Figure 3). In the absence of antibiotic, we saw an increase in prevalence of the genetically avirulent ΔhilD strain, as reported previously [20]. If antibiotic was added to the culture, overall population viability counts dropped (Figure S9) and we observed the opposite effect: The wild-type increased relative to the avirulent ΔhilD mutant (Figure 3). The observation that exposure to antibiotics can lead to selection against genetically avirulent mutants in vitro raises the question of whether antibiotic treatment could contribute to the maintenance of virulence in a clinical context. These findings also suggest that the two established functions of phenotypic heterogeneity—division of labor and bet-hedging—can interact. Variation in virulence expression in clonal populations can translate into differential susceptibility to antibiotics and lead to a bet-hedging benefit, which could in turn protect clonal populations against the invasion of avirulent mutants [9],[32]–[35] that exploit and subvert the division of labor within these populations [8],[9]. This reveals a new level of complexity in the functional consequences of phenotpyic heterogeneity.
In this study, we observed a connection between virulence gene expression and tolerance to antibiotics that could be general: The expression of virulence factors often entails metabolic costs [20],[35],[36], possibly as a side effect of the expression of abundant proteins, and the resulting growth retardation could generally increase tolerance against antibiotics and thus compromise treatment. Under this scenario, pathogens would show tolerance to antibiotics even in situations where treatment was not an important selective factor in their evolutionary past. Understanding the cellular basis of antibiotic tolerance, and the consequences it can have on selection for virulence, is important for using existing treatment options effectively and for developing new strategies for controlling pathogens. In addition, our results suggest a general mechanism that could contribute to the evolutionary stability of cooperative behavior in microorganisms: If individuals that express a costly cooperative trait are also better protected against environmental impacts, then this could lead to a stabilization of the cooperative phenotype.
LB Lennox (Sigma) was used as the growth medium for preculturing for all experiments and as the growth medium in microscopy experiments where indicated. Ampicillin (AppliChem) was added at 100 µg/ml to the growth medium when required. LB buffered with mineral salts [37] was used for growth in chemostats. For the medium used in microscopy experiments, BSA (Sigma) and salmon sperm DNA (Sigma) was added to the medium at 150 µg/ml and 50 µg/ml, respectively, to avoid sticking of the cells to PDMS. Spent medium was obtained by growing the same strain as used in the respective experiments in LB without antibiotics, and by filter sterilizing it when the culture reached an OD 600 nm of 0.8–0.9. Ciprofloxacin (Fluka) was used at a concentration of 0.05 µg/ml for the experiments shown in Figure 1, Figure 3, Figure S2, Figure S3, Figure S7, Figure S8, and Figure S9, and at a concentration of 10 µg/ml for the experiments shown in Figure 2. Kanamycin (Sigma) was used at a concentration of 16 µg/ml for the experiment shown in Figure S4 and at a concentration of 50 µg/ml for the experiment shown in Figure S6. Plates containing 50 µg/ml kanamycin were used to determine strain ratios and colony forming units in the experiment shown in Figure 3 and Figure S9. For the experiment shown in Figure S8, spent LB was supplemented with IPTG (Promega) at the indicated concentrations.
All strains are derivatives of S. typhimurium SL1344 [38] (see Table S1 for a list of all strains used). Bacteria were grown overnight in culture tubes (100 mm×16 mm PP reaction tube, Sarstedt, Nümbrecht, Germany) in 5 ml LB shaking at 220 rpm at 37°C, and then diluted 1∶100 in LB 2–3 h before the experiments to obtain exponentially growing cells in steady state. MICs for ciprofloxacin and kanamycin were determined by the standard method ([39], and Figure S5), and a concentration of 2× MIC (Figure S5 and Text S1) was used for antibiotic treatment in all experiments except for the experiments shown in Figure 2 and Figure S6. For microscopy experiments, the flagella mutant strain X8602 [40] was used to avoid loss of cells from the channels. The plasmid psicA gfp [20] driving expression of gfpmut2 from the sicA promoter was introduced in all strains used for microscopy and flow cytometry, and GFP expression from this plasmid was used to assess induction of ttss-1. For the experiment in Figure S2, a hilD deletion allele from strain M2007 was P22 transduced into the X8602 background to yield strain M3139, and subsequently transformed with the psicA gfp plasmid. For the experiment in Figure 3 and Figure S9, 10 clones (cultures grown from single colonies) of a kanamycin-sensitive wild-type strain (SB300) were competed against 10 clones of a kanamycin-resistant ΔhilD strain (M2007) [20], and 10 clones of a kanamycin-resistant wild-type (resistance cassette inserted at the lpfED locus, showing an identical ttss-1 expression pattern to the kanamycin-sensitive wild type; Figure S9) were competed against 10 clones of a kanamycin-sensitive ΔhilD strain (Z19). Deletion mutants were constructed via lambda red recombination as described in [41] and P22 transduced into the clean SB300 or X8602 background, respectively. For the experiment shown in Figure S8, the plasmid pCA24N-lacZ from the ASKA(–) collection [31] was transformed into M3139.
The microfluidic devices were made using a design adapted from the one published by Wang et al. [27] (Figure S1). Masks for photolithography were ordered at ML&C GmbH, Jena, Germany. Two-step photolithography was used to obtain silicone wafers. PDMS (Sylgard 184 Silicone Elastomer Kit, Dow Corning) was mixed in a ratio of 10∶1, mixed by stirring, poured on the dust-free wafer, degassed in a desiccator until no visible air bubbles were present, and incubated overnight at 80°C for curing. PDMS chips of approximately 1.5 cm×3.5 cm were cut out around the structures on the wafer. Holes for medium supply and outlet were punched using 18G needles (1.2 mm×40 mm) that were modified by breaking off the beveled tip and sharpening the edges of the then straight tip. Chemical activation of surface residues on the PDMS chips and on 24 mm×40 mm glass coverslips (Menzel-Gläser, Braunschweig, Germany) was performed by treating them for 6 min in a UV-Ozone cleaner (Novascan PSD-UV). The PDMS chips were then placed on the glass coverslips, the exposed sides facing each other, and put on a heated plate at 90°C overnight for binding. Before an experiment, chips were rinsed with LB containing BSA and salmon sperm DNA (concentrations as mentioned above, 2 ml/h pump speed) until the growth channels were filled. Cells from an early exponential phase culture were concentrated approximately 100× by centrifugation (12,470× g, 2 min) and loaded into the chip using a pipette. The process of cells entering the channels was observed microscopically, and when sufficient occupation of the channels was observed (after 10–20 min), medium was pumped through. For all experiments, syringe pumps (NE-300, NewEra Pump Systems) with 60 ml syringes (IMI, Montegrotto Terme, Italy) containing the media were used. Tubing (Microbore Tygon S54HL, ID 0.76 mm, OD 2.29 mm, Fisher Scientific) was connected to the syringes using 20G needles (0.9 mm×70 mm), which were directly inserted into the tubing. Smaller tubing (Teflon, ID 0.3 mm, OD 0.76 mm, Fisher Scientific) was then inserted into the bigger tubing and directly connected to the inlet hole in the PDMS chip. Medium change was performed by disconnecting the bigger and smaller tubings and reconnecting to the bigger tubing of a second medium supply. All experiments were run at a pump speed of 2 ml/h.
Microscopy was performed using an Olympus IX81 inverted microscope system with automated stage, shutters, and a laser-based ZDC autofocus system. Several different positions were monitored in parallel on the same device, and phase contrast and fluorescence images (where applicable) of every position were taken every 5 min. Images were acquired using an UPLFLN100xO2PH/1.3 phase contrast oil immersion objective (Olympus) and a cooled CCD camera (Olympus XM10). For image acquisition, the CellM software package (Olympus) was used. Fluorescence images were acquired using a 120W mercury short arc lamp (Xcite 120PC Q) and the U-N41001 GFP filter set (450–490 nm ex/500–550 em/495 dichroic mirror, Chroma). The whole microscope was placed in an incubated box (Life Imaging Services, Reinach, Switzerland) at 37°C during all experiments.
Images were analyzed using the plugin MMJ (available at https://github.com/penamiller/mmJ) for ImageJ [42]. It allows extracting of fluorescence intensities and cell length for the bottom cell of each channel during the course of the whole experiment, and scores division events based on cell length. For analysis of the experiments shown in Figure 1, Figure S2, Figure S3, Figure S4, Figure S8, and for determining growth rates, the standard version of MMJ was used. For the experiments shown in Figure 2 and Figure S6, a modified version of MMJ (MMJAll) was used that allows the extraction of all cells on single frames. Data were then further processed and plots were generated using R [43]. Cells were counted as surviving if they divided at least once after removal of antibiotic.
For competition experiments, strains were mixed in a 1∶1 ratio in fresh LB, according to their optical density in overnight cultures. After 3.5 h of growth, all cultures were diluted 1∶100 in spent LB to extend the time cells spend in an induced state, and 0.05 µg/ml ciprofloxacin was added where indicated. Samples were taken at the indicated times, optical density was measured, and dilutions were spread on LB agar plates. After overnight growth, colonies were counted on the LB agar plates, and replica plated on LB agar plates containing 50 µg/ml kanamycin. Surviving colonies on LB kanamycin plates were counted the next day, and ratios of strains were determined. To control for a possible influence of the placement of the kanamycin resistance marker, the experiment was performed with two different strain combinations, 10 replicates each: Kanamycin-sensitive wild type (SB300) was competed against kanamycin-resistant ΔhilD (M2007) and kanamycin-resistant wild type (resistance cassette inserted at the lpfED locus) were competed against kanamycin-sensitive ΔhilD (Z19). Statistical analysis showed no significant influence of maker placement on the time-dependent relative frequency of the strains (three way ANOVA, treatment×time×marker p = 0.19). Data from both strain combinations were pooled for the plots shown in Figure 3 and Figure S9.
To determine the growth rates of the T1+ and T1− subpopulations, we grew ΔfliCΔfljB psicA gfp cells in the same microfluidic devices as used in Figure 1. After initial growth in LB for 2 h, 45 min, we changed the medium to spent LB (see above), which still contains enough nutrients to sustain growth, and monitored growth and gene expression for 13 h, 45 min. We identified all cells (11 cells in total) in the experiment whose fluorescence levels were higher than 10 standard deviations above the fluctuations in background fluorescence (i.e., fluorescence of areas not containing cells in the vicinity of the respective cells measured), and determined the number of cell divisions during that time. To determine the growth rate of T1− cells, we used 11 cells from channels neighboring channels harboring T1+ cells that do not show a significant increase in fluorescence and determined the number of doublings of those cells during the same time period as for the T1+ cells in the neighboring channel.
Data on cell length and divisions were extracted for a period of 100 min (20 frames)—75 min (15 frames) before and 25 min (5 frames) after addition of the antibiotic—and linear regression was performed on the natural logarithm of cell lengths between divisions, between the start of the period and the first division in the period, and between the last division in the period and the end of the period, respectively, to determine the slope of the length increase during every division. Arithmetic means of the slopes of every individual cell were then calculated and multiplied by 12 to get a number for length doublings per hour.
Chemostat growth was performed using a Sixfors system (Infors HT, Bottmingen, Switzerland) with six parallel reactors. Buffered LB was used as growth medium, as described in Ihssen et al. [37]. We inoculated 400 ml of medium in each reactor with 1 ml early exponential phase cultures that were previously diluted 1∶100 from six overnight cultures of individual clones. The reactors were stirred at 800 rpm, aerated with sterile air, and the temperature was controlled to be at 37°C. Growth as batch cultures was allowed for 3.5 h. Then fresh medium was pumped into the reactors at 104 ml/h (0.26 volume changes per hour, corresponding to a doubling time of 2.66 h), and total volume in the reactors was kept constant at 400 ml. After 16 h, pumping speed was changed for three of the six reactors to be 384 ml/h (0.96 volume changes per hour, corresponding to a doubling time of 0.72 h). After 6 h, 0.05 µg/ml ciprofloxacin was added to the medium supply, and all reactors were spiked with ciprofloxacin to a final concentration of 0.05 µg/ml, to keep the amount of the antibiotic constant in all replicates. However, we cannot rule out the possibility that metabolic differences between the slow and fast growing chemostat populations could lead to differences in the pharmacokinetics of the antibiotic. Samples were taken every 30 min, optical density at 600 nm was determined, samples were diluted up to 1∶107 in a series of 1∶10 dilutions, and 5 µl of each dilution were spotted on LB plates. After the spots dried, plates were incubated at 37°C overnight. Spots with numbers of colonies suitable for counting were then identified, and the number of colony forming units for every time point was calculated and normalized to the total number of cells as determined by the measured OD.
For the experiment in Figure S10, overnight cultures were diluted 1∶20 in LB Lennox, and analyzed in a flow cytometer (LSRII, Becton Dickinson) at an OD 600 nm of 0.9. Bacteria were identified by side scatter, and GFP emission was measured at 530 nm. Data were analyzed using FlowJo software (Tree Star, Inc.).
|
10.1371/journal.pgen.1000062 | How To Perform Meaningful Estimates of Genetic Effects | Although the genotype-phenotype map plays a central role both in Quantitative and Evolutionary Genetics, the formalization of a completely general and satisfactory model of genetic effects, particularly accounting for epistasis, remains a theoretical challenge. Here, we use a two-locus genetic system in simulated populations with epistasis to show the convenience of using a recently developed model, NOIA, to perform estimates of genetic effects and the decomposition of the genetic variance that are orthogonal even under deviations from the Hardy-Weinberg proportions. We develop the theory for how to use this model in interval mapping of quantitative trait loci using Halley-Knott regressions, and we analyze a real data set to illustrate the advantage of using this approach in practice. In this example, we show that departures from the Hardy-Weinberg proportions that are expected by sampling alone substantially alter the orthogonal estimates of genetic effects when other statistical models, like F2 or G2A, are used instead of NOIA. Finally, for the first time from real data, we provide estimates of functional genetic effects as sets of effects of natural allele substitutions in a particular genotype, which enriches the debate on the interpretation of genetic effects as implemented both in functional and in statistical models. We also discuss further implementations leading to a completely general genotype-phenotype map.
| The rediscovery of Mendel's laws of inheritance of genetic factors gave rise to the research field of Genetics at the very beginning of the last century. The idea of traits being determined by the effects of inherited genes is thus the conceptual core of Genetics. After more than one century, however, we still lack a completely general mathematical description of how genes can control traits. Such descriptions are called genotype-phenotype maps, or models of genetic effects, and they become particularly cumbersome in the presence of interaction among genes, also referred to as epistasis. The models of genetic effects are necessary for unraveling the genetic architecture of traits—finding the genes underlying them and obtaining estimates of their individual effects and interactions—and for meaningfully using that information to investigate their evolution and to improve response to selection in traits of economical importance. Here, we illustrate the convenience of using a recently developed model of genetic effects with arbitrary epistasis, NOIA, to inspect the genetic architecture of traits. We implement NOIA for practical use with a regression method and exemplify that theory with a real dataset. Further, we discuss the state of the art of genetic modeling and the future perspectives of this subject.
| There is an increasing interest in Quantitative Genetics and Evolutionary Biology to identify genetic effects, and more particularly gene interactions, on a genome-wide scale and to understand its role in the genetic architecture of complex traits [1],[2]. Genome scans for quantitative trait loci (QTL) have proven to be a successful strategy for identifying genetic effects and interactions. Two of the main issues in the development of QTL mapping methods are which models of genetic effects to use and how to test for effects in regions between marker locations. The second issue is important not only for considering the genome as a virtually continuous space where to map the QTL, but also to efficiently analyze incomplete data sets, which are the norm in practice [3]. Lander and Botstein [4] developed the classic interval mapping (IM) method, in which they showed how to perform a QTL mapping strategy implemented with the most likely genotypes for the genome regions in between marker locations, given the genotypes at the flanking markers. This method has been extended in several ways [5]–[8]. Albeit the computation of those likelihoods is complex and time demanding, Haley and Knott [9], (see also [10]) provided a convenient approximation of them by means of a simple regression method.
Regarding now the first issue mentioned above—the models of genetic effects—the definition of the genetic effects in Haley and Knott's [9] regression (hereafter HKR) comes from a model that has been extensively used in Quantitative Genetics, the F∞ model [11],[12]. However, other models of genetic effects have recently been shown to be more appropriate in QTL mapping. The genetic effects depend not only on the genotypic values but also on the genotype frequencies of the analyzed population (e.g. [13]–[16]). By taking into account these frequencies, it is possible to build orthogonal models that are convenient for several reasons [13]–[19]. First, orthogonal estimates do not change in reduced models, which considerably facilitates model selection for finding the genetic architecture of traits. Second, the estimates of genetic effects obtained by orthogonal models are meaningful in the population under study—they provide the effects of allele substitutions in that population. Third, they directly lead to a proper, orthogonal decomposition of the genetic variance from which to compute important measures, like the heritability of that trait in that population. The statistical properties of HKR could therefore be improved by implementing it with a genetic model that is orthogonal for any possible genotype frequencies in the population under study.
The statistical formulation of the recently developed NOIA (Natural and Orthogonal InterActions) model of genetic effects is orthogonal in situations where previous models are not—for departures from the Hardy-Weinberg proportions (HWP) at any number of loci—and it is therefore more appropriate choice for estimating genetic effects from data in genetic mapping [16]. Furthermore, a novel feature of NOIA is its implementation to transform the genetic effects estimated in the population under study, in two ways. First, they can be transformed into how they would look like in a population with different genotype frequencies at each locus, like an ideal F2 population or into an outbred population of interest. Second, using the functional formulation of NOIA, it is possible also to express the genetic effects as effects of allele substitutions from reference individual genotypes—instead of from population means like in the statistical formulation. In other words, starting from the orthogonal genetic effects of a population or sample under study, which are the ideal ones for performing model selection and have a particular meaning, NOIA enables us to obtain the values of the genetic effects that are associated to other desired meanings and are useful, therefore, to inspect different aspects of the evolution of a population, or selective breeding for increasing or decreasing a trait values.
Our motivation for this communication is to show how to use models of genetic effects to obtain estimates of genetic effects from data that have the desired meaning of any particular scientific purpose. To this end we first inspect how much of a difference it makes to use the classical models for ideal populations, such as ideal F2 populations, to compute genetic effects in a non-ideal situation, under departures from the HWP. We address this issue by generating simulated populations that depart from the HWP in several degrees and analyzing them with NOIA and other models. We quantify the deviances from orthogonal estimates due to using models that assume ideal conditions in the populations under study, thus showing the practical convenience of using the NOIA model for performing real estimates of genetic effects in QTL experiments. Second, we develop an implementation of NOIA with HKR, allowing it for immediate practical use and illustrate its performance using an example with real data. By this example we provide estimates of genetic effects with different meanings and, for the first time, functional estimates of genetic effects—using an individual genotype as reference—from a real data set. We discuss on how this feature opens new possibilities of using real data to analyze important topics in Evolutionary Genetics.
Figure 1 shows the results of estimating, with three different models (NOIA, G2A and F2), the genetic effects of a two-locus and two-allele genetic system (Table 1) in nine simulated populations under linkage equilibrium (LE) with various degrees of departure from the HWP (see Methods). The eight genetic effects plus the population mean in the only model that is orthogonal in all simulated populations—the statistical formulation of the NOIA model—respond to the increasing departures from HWP in three groups. The first and most influenced group contains the three genetic effects involving the additive effect of the locus affected by departures from HWP, αA, αα, and αδ. These genetic effects increase substantially with increasing departures from HWP and are doubled when the homozygote A2A2 is almost completely absent. The second group contains the reference point—the mean of the population, μ—and the single locus effects of locus B (the one at HWP), αB and δB. The estimates in this group decreased with increasing departures from HWP. The third group contains the remaining three genetic effects, δA, δα and δδ, whose estimates are not affected by departures from HWP at locus B. The genetic effects measured by the G2A model show the same qualitative behavior described above for NOIA (i.e. also responds in three distinct groups), but are quantitatively different. The reason for this is that G2A can adapt the measurements to the changes in the allele frequencies of the population, but not to the precise departures of the genotype frequencies from the HWP. The genetic estimates obtained using the F2 model always give the same values independently of the genetic constitution of the population. The F2 thus fails to capture the effects of departures from HWP at all. Thus, unless when the studied population is an ideal F2 (and the deviances from HWP are zero, see Figure 1), the estimate of the population mean from G2A and F2 is biased and the genetic estimates do not reflect the average effects of allele substitutions in the population under study. Those deviations become more severe as the departure from HWP increases (Figure 1).
Figure 2 shows the variance component estimates obtained in the nine simulated populations, which were obtained by computing the variance over the individuals of the sample population of the correspondent genetic effects (additive effect at locus A, additive effect at locus B, etc). For orthogonal models, the sum of the three components of variance gives the total genetic variance—which in this case equals the phenotypic variance, since there is no environmental variance in the simulated populations. Here, this is only observed for the variances computed using NOIA. The other two models are not orthogonal in the populations under study (except in the ideal F2 population, where the three models coincide), and thus there exist covariances between the genetic effects that would need to be accounted for to obtain the true genetic variance of the population [20]. The decomposition of the genetic variance made by the G2A and F2 models is, thus, non-orthogonal. The G2A leads to a greater departure form an orthogonal decomposition of variance than the F2 model by the particular kind of departures from HWP simulated here. Both the G2A and F2 models underestimate the additive variance and therefore also the heritability of the trait in the simulated populations.
For illustrating the advantage of using NOIA for analyzing experimental data, we reanalyze a two-locus (A and B) genetic system with epistasis affecting growth rate in an F2 cross between Red junglefowl and White leghorn layer chickens [21]. The two loci are on different chromosomes, thus avoiding linkage disequilibrium (LD). Locus A departs significantly from the HWP when considered alone, but not when correcting for multiple testing (see Methods). Table 2 shows the genetic effects and the components of variance for this two-locus system using several models of genetic effects—NOIA, G2A, F2 and F∞. As explained in the previous subsection, NOIA is orthogonal under departures from the HWP, whereas the other models are not. The F∞ model deviates severely from the estimates obtained by NOIA. Deviations are expected since the F∞ model is non-orthogonal even in an ideal F2 population with no deviations from the expected frequencies due to sampling errors. The F2 and G2A models, on the other hand, would be orthogonal under ideal circumstances and the observed deviations from orthogonality of those models when analyzing these experimental data are due to sampling (as explained above). Table 2 shows that the estimates obtained using F2 and G2A differ substantially from these of NOIA (up to 18/42% for the G2A and 53/138% for the F2 model, for the genetic effects/variance component estimates). This example with real data, thus, shows that it makes a substantial improvement to use NOIA to compute genetic effects and variance decomposition in QTL mapping experiments over the classical models of genetic effects designed to fit ideal experimental situations.
From the statistical estimates in Table 2, we have computed functional estimates of genetic effects using an analogous expression to (S6), shown in Text S1, derived by Álvarez-Castro and Carlborg [16]. The variances of the statistical estimates can also be transformed to give the variances of the functional estimates using (6), as derived in the Methods section. Choosing “A1A1B1B1” as reference genotype, the estimates of functional genetic effects, and the standard deviations associated to these estimates, are shown in Table 3. Whereas statistical genetic effects describe the average effects of allele substitutions in a population, functional genetic effects describe the genotype-phenotype map as a series of allele substitutions performed in the genotype of a particular—reference—individual genotype [16],[22], in this case the genotype of the Red junglefowl, “A1A1B1B1”.
To illustrate the usefulness of these functional genetic effects for understanding how epistatic effects can contribute to phenotype change, we consider the role of this QTL pair in increasing the growth rate in the Red junglefowl. For simplicity, we assume hereafter that A and B are the only two loci affecting growth rate. From the marginal genetic effects in Table 3, it can be deduced that the White leghorn layer allele at locus A slightly increases the phenotype whereas the White leghorn allele at locus B actually decreases it, when considered in homozygotes. However, the dominance effects are positive and have a higher absolute value than the additive effects. Therefore, if one White leghorn layer allele appeared by mutation in a Red junglefowl population at any of the two loci, A or B, it would be maintained at a certain frequency because of balancing selection—superiority of the heterozygote—but it would neither disappear nor reach fixation. This suggests that one mutation could be present at some frequency in the population when the second one appeared.
For analyzing what would happen if eventually the two mutations were present at the same time in the population, we have to consider also the interaction effects. The double homozygote for White leghorn layer allele increases the phenotype with roughly forty grams (four times aa, in Table 3 as it can be deduced from G = S⋅E, with the reference of R = G1111), relative to the expected value without epistasis, which is a decrease in roughly 20 grams from the Red junglefowl. In total, this makes the phenotype of the White leghorn layer 20 grams higher than the Red junglefowl. However, for inspecting if this results support the White leghorn layer alleles being likely to reach fixation we also need to consider the phenotypes of the heterozygotes. Interactions involving dominance in locus B are all negative, thus favoring the fixation of the White leghorn layer allele, B2. The role of allele A2 is not as obvious, since da is positive. The genotypic value of “A1A2B2B2” is roughly 30 grams higher than the Red junglefowl (computed again from Table 3 and G = S⋅E) and ten grams higher than the pure White leghorn layer. The expected, therefore, would be that the two alleles segregate at locus A. The standard deviations of the estimates are however rather large and thus do not rule out the possibility of fixation of the White leghorn layer allele at locus A.
The statistical formulation of NOIA is orthogonal under random deviations from ideal experimental populations and outbreeding pedigrees [16]. Therefore, NOIA can provide meaningful estimates of genetic effects—as allele substitutions made in the population or sample under study—and a proper decomposition of the genetic variance under those circumstances. In this article, we illustrate the practical implications of these achievements for estimation of genetic effects and QTL analysis in two ways. First, we simulated a two-locus genetic system under departure from the HWP affecting one of the loci underlying the trait under study. This scenario can have a biological origin or be due to sampling alone and it is commonly occurring in experimental data both from natural and experimental populations, such as for the QTL pair we have studied (see below). We therefore deemed it relevant to test the performance of NOIA in practice—by assessing how departures from HWP cause other models to deviate from the orthogonal values. Our results show that departures from HWP substantially affect both the genetic effects and the decomposition of variance. The cause for this is that epistasis makes the genetic effects dependent on the genetic background, which is different under different degrees of departures from HWP. NOIA can capture the proper, orthogonal genetic effects, and thus also their orthogonal variances, in the simulated populations whereas the deviances from these values due to using the other—nonorthogonal—models increases with the departures from HWP.
Second, we used experimental data on epistatic QTL from a previously published study [21] to explore how much of a difference it makes to use NOIA instead of previous statistical models, when departures from HWP are not larger than expected by sampling. Even though the population we studied was rather large (approximately 800 individuals), the random deviations from the HWP in this set of available individuals cause considerable differences in the estimates of genetic effects performed with models that would be orthogonal in totally ideal situations, as compared to the estimates obtained using NOIA. These differences become even more noteworthy for the components of variance estimated using the different models. These values influence consequential quantities, like the heritability of one trait, which may be needed for instance for performing artificial selection at the available sample of individuals. Orthogonal models are also important for finding the genetic architecture of traits—albeit this has not been our focus in this communication. In principle, when testing the effect of a particular locus or set of loci in a QTL analysis, the choice of the model of genetic effects to use does not matter. However, it does matter when it comes to compare which of several putative sets of loci is the most likely genetic architecture underlying the trait, i.e., when performing model selection in QTL analysis. This is so because orthogonal models have the convenient property that the estimates and their variances remain the same when considering reduced models, which facilitates model selection strategies [19].
After model selection and the estimation of genetic effects have been properly carried out using an orthogonal model, the obtained estimates provide the effects of allele substitutions in the sample of individuals used in the study, and the decomposition of variance is also the appropriate one in that particular sample of individuals. The NOIA model provides convenient tools for transforming those estimates into the ones with any other desired meaning, like the orthogonal estimates and the decomposition of variance in a different population [16]. This is useful to compare results from QTL studies performed in different populations, and to use the results obtained with one orthogonal model in one population to study the evolution of the same trait in a different population.
One example of the previous is removing the characteristics of the data that are not supposed to be properties of a target population from the estimates. The departures from HWP of the experimental data we dealt with in this article are in fact supposed to be only due to sampling, instead of being caused by real Hardy-Weinberg disequilibrium in the F2 population. If we were interested in the genetic effects or in the decomposition of variance of the ideal F2 as a target population—in which the departures from HWP are absent—we could use the transformation tool of NOIA to obtain (from the original estimates with the reference of the mean of the sample population) the ones with the reference of the mean of an ideal F2 population. Further, as illustrated in the example with real data, it is possible to transform statistical estimates of genetic effects into functional ones, using a particular reference genotype. Another situation in which these transformations are valuable is, for instance, in a three-locus genetic system with pairwise epistasis. In this case, NOIA would easily permit to consider only the significant genetic effects and to re-compute the genotypic values only from the significant genetic effects (assuming the non-significant third-order interactions to be zero).
Statistical models of genetic effects are necessary for QTL analysis and for performing orthogonal decompositions of the genetic variance in populations. Functional models of genetic effects, on the other hand, are convenient—especially in the presence of epistasis—for studying evolutionary properties of the populations such us adaptation in the presence of drift and speciation (see e.g. [23],[24]). NOIA is the first model framework that successfully unifies functional and statistical modeling of genetic effects [16]. This enables researchers to feed models of functional genetic effects, so far mainly used in simulation studies (see e.g. [2],[24]), with real data obtained using statistical models in QTL mapping experiments. Here, we have actually transformed statistical genetic effects, obtained from real data of an F2 experimental population, into functional genetic effects as allele substitutions performed from a reference individual. Concerning these functional estimates of genetic effects, we have shown in the previous section how they can improve the understanding of the genetic system by inspecting a two-locus model obtained from real data. Notice that when changing the reference of the model, the genetic effects can change their magnitudes and even their signs (see Tables 2 and 3). Therefore, for reaching the kind of conclusions we obtain above for the evolution of a population from an ancestral genotype “A1A1B1B1”, the genetic effects have to be described with a model that uses that particular genotype as reference point. Those are the only ones that are meaningful for analyzing the problem under consideration.
The computation of genetic effects using NOIA in the example with real data required the use of the theory developed in this article, the implementation of the model to handle missing data (1). When performing IM for searching for the positions and estimates of genetic effects in QTL mapping experiments, missing data occurs at two levels. First, the genotype of the QTL located in a marker interval is not known and needs to be estimated from the observed flanking marker genotypes. Second, in most experimental datasets there are missing genotypes for many genetic markers that can be imputed from genotypes at closely linked informative markers. Thus, the implementation of HKR with NOIA enables us to perform IM with a regression method and using a model of genetic effects that is orthogonal regardless of how far the available data is from the HWP.
The HKR has been assessed as a good approximation of IM when dense marker maps are available and missing data are few and random [25],[26], but some disadvantages of this method have also been reported. The residual variance of the HKR has been found to be biased, as first pointed out by Xu [27]. Kao [26] further characterized that bias and found it to be noticeable under LD or strong epistasis. Nevertheless, even in those cases, the estimated genetic effects themselves are not biased [26]. Feenstra et al. [25] have developed a new method, the estimating equation method, which reduces the reported bias of the HKR and is therefore more suitable in the cases when it has proven to be strongly biased. However, the traditional HKR is still popular and convenient mainly due to its dramatic advantage in computational time [25], and this is why in this study we have chosen this method for implementing NOIA for IM.
Models of genetic effects need to be further generalized. Two important cases that need to be accounted for are multiple-alleles and LD, which have been addressed in several recent publications dealing with statistical models of genetic effects. Yang [18] has developed a model to test the importance of LD in QTL data, by designing a component of variance due to LD. This statistical model, like the statistical formulation of NOIA, actually accounts for departures from HWP, although it is restricted to the two-locus case. Wang and Zeng [20] have developed a statistical model with multiple alleles in which they also test the importance of LD, in this case by computing all the covariances between the components of variance, due to LD. It is, however, restricted to HWP. Mao et al. [28] have developed a model to account for LD when computing genetic effects in a two-locus model specially designed for single nucleotide polymorphisms. The desired situation, which we are currently aiming toward is to consider all the different departures from ideal situations gathered under the umbrella of a general formal framework of genetic effects.
We use a simulated numerical example to show how departures from the HWP affect the estimates of genetic effects in several models of genetic effects. We simulate a trait controlled by two biallelic loci, A and B, generating several populations with the second locus affected by departures from the HWP in several degrees. The genotype-phenotype map corresponds to the phenotype mean of the population and all the genetic effects being equal to one in an ideal F2 population (Table 1). We first constructed data for an ideal F2 population of 800 individuals in strict HWP and LE. From this population we subsequently removed 24 A2A2 individuals and added eight A1A1 and 16 A1A2 individuals in a balanced way, without affecting the population size, the frequencies at locus B, the proportion of A1A1 versus A1A2 individuals or LE. Only deviations from the HWP against the A2A2 homozygote were introduced in the data. We repeated this procedure eight times in total and saved each population data, until only eight A2A2 individuals remained. We measured the departures from HWP in these populations by computing the percentage of reduction of A2A2 individuals relative to A1A1, which of course was zero in the ideal F2 population we started from.
We analyzed the simulated data by computing the genetic effects of the system using three models: NOIA, G2A and F2. The F2 model, described in Text S1, is constructed for F2 populations, although it is only orthogonal in ideal F2 populations with the genotypic frequencies being exactly ¼, ½, ¼. The NOIA model is as described in Text S1. The G2A model [19] accounts for any gene frequencies of—and it is orthogonal at—populations under exact HWP. Álvarez-Castro and Carlborg [16] obtained it as a particular case of NOIA by constraining (S5), in Text S1, to HWP:where p is the frequency of allele A1. The genetic effects were computed for each individual genotype using the genetic-effects design matrices and the estimates of genetic effects from each of the three models, which produced different outcomes. The additive, dominance and interaction variances were obtained as the correspondent sums of the variances of each genetic effect (for instance, the sum of the variances of the additive effects of each of the loci gives the additive variance).
We recall the required theory behind the HKR and NOIA in Text S1. Here we extend the NOIA model to IM with HKR. We do this by implementing the genetic-effects design matrix of the statistical formulation of NOIA, SS (S5), in the HKR method, as we do with the F2 model in Text S1. The original genotype frequencies p11, p12 and p22 in the NOIA statistical formulation (S5) are the exact genotype frequencies at the considered loci. In the HKR, the genotype frequencies are not known, but can be estimated as:where N is the number of individuals in the population under study. We implement this model in the general expression of the HKR (S4), in Text S1, and obtain:Let G* be the column-vector of observed phenotypes, G*k, k = 1,…,N, ε the corresponding vector of errors, and Z, which is an N×3-matrix whose rows are the vectors ωk (S4). With this notation, the general expression of regression (S4) is:(1)This has a straightforward extension to several loci with LE. The SS matrix and the E vector can be extended as in Álvarez-Castro and Carlborg [16]. The Z matrix can be extended as the row-wise Kronecker product of the matrices of the single loci, also as in Álvarez-Castro and Carlborg [16], albeit in that article the matrix accounted for only complete marker information, instead of for IM with HKR, or for missing data probabilities. For instance, for a two-locus (A and B) case, the ZAB matrix is an N×9-matrix that is built as:
Carlborg et al. [21] identified 10 genome-wide significant QTL for growth rate in chicken from eight to 46 days of age in an F2 intercross of roughly 800 individuals between one Red junglefowl male and three White leghorn females. A simultaneous two-dimensional genome scan was performed to identify pairs of interacting loci regardless of whether their marginal effects were significant or not. We have studied in more detail one of the detected pairs involving QTL on chromosome 2 (486 cM) and 3 (117 cM), hereafter loci A and B respectively. This pair was selected for a number of reasons. First, these loci interact epistatically, in spite of showing no significant marginal effects in the studied population. Second, since they are located in different chromosomes, there is no physical linkage between them. Third, the genotype frequencies at locus A depart significantly from the HWP (p<0.05) when considered independently, but the departure is not significant after applying multiple testing correction accounting for the rest of the detected QTL. Thus, locus A is an example of the departure of the HWP that is expected in QTL experiments just due to sampling. The level of departure from the HWP for the evaluated pair roughly equals the 30% deviation in Figures 1 and 2.
We have computed the genetic effects of the epistatic pair involving loci A and B, using several models of genetic effects. First we used the F∞ model, which was the one also used by Carlborg et al. [21] as it was the model originally implemented in HKR [9],[29]. Second, the F2 model, which was designed for F2 populations. Third, the G2A model, which can account for departures of the gene frequencies from ½, and finally the statistical formulation of NOIA, which can adapt to the genotype frequencies of the sample used for the estimation of QTL effects. In these analysis we have made use of the theory developed in this article: the implementation of HKR with NOIA. These developments enable us to deal both with missing data and with the estimation of genetic effects of positions inside the marker intervals.
Álvarez-Castro and Carlborg [16] have shown how to transform genetic effects obtained using an orthogonal-statistical model in one population, into statistical genetic effects at any other population or into functional genetic effects from any reference individual. In each of these two cases, the transformation is done as in expression (S6), in Text S1, using the S matrix—the genetic-effect design matrix—of the orthogonal system, G = S1⋅E1, and the inverse of the S matrix in the new system, G = S2⋅E2:(2)Let(3)be the transformation matrix. From (2) and (3), the estimates in E1 can be expressed as functions of the estimates in E2 as:(4)where the letters and their superindexes indicate the vector, or matrix, they are scalars of and the subindexes indicate the position of the scalars inside the vectors or matrices. From (2), the variances of the estimates E2, can be computed from the ones in E1 as:(5)Now for obtaining the vector of variances of the estimates E2, V2, from the vector of variances of the estimates E1, V1, we just rewrite (3) in algebraic notation as:(6)where the open circle stands for the Hadamard product—giving the matrix whose scalars are the product of the scalars at the same position in the original matrices. |
10.1371/journal.pntd.0004689 | Redefining the Australian Anthrax Belt: Modeling the Ecological Niche and Predicting the Geographic Distribution of Bacillus anthracis | The ecology and distribution of B. anthracis in Australia is not well understood, despite the continued occurrence of anthrax outbreaks in the eastern states of the country. Efforts to estimate the spatial extent of the risk of disease have been limited to a qualitative definition of an anthrax belt extending from southeast Queensland through the centre of New South Wales and into northern Victoria. This definition of the anthrax belt does not consider the role of environmental conditions in the distribution of B. anthracis. Here, we used the genetic algorithm for rule-set prediction model system (GARP), historical anthrax outbreaks and environmental data to model the ecological niche of B. anthracis and predict its potential geographic distribution in Australia. Our models reveal the niche of B. anthracis in Australia is characterized by a narrow range of ecological conditions concentrated in two disjunct corridors. The most dominant corridor, used to redefine a new anthrax belt, parallels the Eastern Highlands and runs from north Victoria to central east Queensland through the centre of New South Wales. This study has redefined the anthrax belt in eastern Australia and provides insights about the ecological factors that limit the distribution of B. anthracis at the continental scale for Australia. The geographic distributions identified can help inform anthrax surveillance strategies by public and veterinary health agencies.
| This study explores the spatial ecology of Bacillus anthracis, the causative agent of anthrax disease, in Australia. Globally, anthrax is a neglected zoonotic disease that primarily affect herbivores and incidentally humans and all warm-blooded animals. Here, we used historic anthrax outbreaks for the period 1996–2013 and environmental factors in an ecological niche modelling framework to quantitatively define the ecological niche of B. anthracis using a genetic algorithm. This was projected onto the continental landscape of Australia to predict the geographic distribution of the pathogen. The ecological niche of B. anthracis is characterized by a narrow range of ecological conditions, which are geographically concentrated in two disjunct corridors: a dominant corridor paralleling the Eastern Highlands runs from north Victoria to central east Queensland through the centre of New South Wales, while another corridor was predicted in the southwest of Western Australia. These findings provide an estimate of the potential geographic distribution of B. anthracis, and can help inform anthrax disease surveillance across Australia.
| Anthrax is a zoonotic disease caused by Bacillus anthracis, an aerobic, gram-positive spore-forming bacterium. Bacillus anthracis primarily affects herbivores; though most warmed-blooded mammals may be susceptible [1], including humans. Anthrax is an ancient disease that has caused losses of livestock and wildlife populations prior to and throughout the 20th century and remains enzootic with seasonal variations in many parts of the world [2, 3]. Transmission remains poorly understood, but ingestion of spores is the dominant hypothesis for herbivores [4]. Grazing mammals (e.g. cattle, sheep, zebras) can be infected by ingesting spores present in contaminated soils, while browsers (e.g. deer) may also ingest the pathogen with contaminated foliage [5, 6]. Biting flies may be involved in transmission on some landscapes [7, 8] and inhalation cannot be ruled out [9]; each mechanism requires further study. In each case, transmission is indirect and occurs where a susceptible host interacts with an environment that supports pathogen persistence. These environments can be characterized and mapped to define areas at risk for anthrax [5, 10].
The first recorded livestock anthrax in Australia dates to 1847 at Leppington, New South Wales where the disease slowly spread through cattle and sheep movements along stock routes [11]. In Victoria, anthrax was initially reported in the area around Warrnambool in the southwestern area of the state in 1886. From there, the disease apparently spread throughout the western districts of the state to Melbourne and elsewhere via the transport of infected sheep [12]. Historical records of livestock anthrax from the early 1900s to the 1920s indicate that the disease was more recurrent in New South Wales, where 80 confirmed anthrax outbreaks were recorded during that period [13]; twice as many as reported in Victoria during the same period. The 1930s saw an increase in livestock anthrax in Australia, especially in New South Wales and Victoria. For instance, in New South Wales, a total of 147 outbreaks were officially recorded from 1930 to 1936 [13], and about 200 during the period 1949–1962 with sheep most commonly infected, while an increase of incidence was observed in cattle in Victoria [12]. Additional outbreaks in Victoria in 1968 caused cattle and sheep deaths on 27 farms in the Yarrawonga/Shepparton area [14]. There is little published literature on anthrax in Australia during the period 1970–1990; though there were confirmed reports throughout NSW and Victoria.
Summarizing the available Australian literature, the continent experienced overall reductions in the size and spatial distribution of livestock anthrax outbreaks across the latter half of the 20th century. Similar patterns were documented in the United States [15] and the Ukraine [16]. The majority of anthrax outbreaks in recent decades have taken place across the Australian anthrax belt, which predominantly runs through the center of New South Wales [17]. The geography of the anthrax belt was originally described by Henry [13] and later roughly delineated by Allan [18] and Durrheim et al. [17] to map the extent of the endemic zone in Australia (Fig 1). This description of the anthrax belt is based solely on locations of disease incidents. The belt lies between the tablelands and the western plains in New South Wales, and reaches from northern Victoria at the southern extent, northward through New South Wales to the southern border of Queensland.
Occasionally, outbreaks have occurred outside of the historically defined anthrax belt. For example, an outbreak occurred in 2007–2008 in the Hunter Valley, a valley located ~350 km east of the belt. In that outbreak, 11 dairy farms in the Hunter Valley experienced unusual anthrax outbreaks in the summer of 2007 with clinical cases mainly observed in cattle. The last recorded livestock anthrax cases in the Hunter Valley prior to the 2007–2008 outbreaks occurred in 1939 [17].
An unprecedented livestock epizootic also occurred in the Goulburn Valley in the north of Victoria in the summer of 1997, affecting 83 dairy farms in the Stanhope/Tatura area. This was the largest anthrax epizootic reported in Australia since official reports of livestock anthrax began in 1914 in Victoria [19]. The Goulburn Valley in northern Victoria borders southernmost New South Wales and intersects the southern extension of the anthrax belt into Victoria [17]. Although the number of reported anthrax outbreaks within the anthrax belt has decreased in recent decades, outbreaks continue within and beyond its currently defined boundaries. Therefore, there is a need for ecological investigations of the distribution of B. anthracis and the identification of all potential risk zones in Australia to better inform anthrax surveillance.
Many environmental factors including climate and soil are known to prolong the survival of anthrax spores in the environment. Van Ness [20] postulated that suitable soils with high soil moisture, alkaline pH, and organic nutrients referred to as “incubator areas” may be conducive to the germination, vegetative growth and sporulation of B. anthracis independently of a mammal host. Recent experimental spore germination using a grass-soil model system also supported the possibility that this dynamic state occurs in the soil [21]. However, the study of Dragon et al. [22] demonstrated that in natural conditions, growth of B. anthracis outside a host leads to a rapid loss of virulence, and that vegetative forms cannot compete with other bacteria species in the soil. This latter study supports the “persistent spore theory” according to which, spores persist in the soil for very long periods of time until they come into contact with a susceptible host causing disease [23, 24]. Irrespective of which of these theories is correct, both recognize that soil is the natural reservoir of anthrax spores, which therefore implies that a greater understanding of the ecological conditions that allow spores to “persist” or “incubate” in the soil environment is essential for the prediction of the potential geographic distribution of B. anthracis.
Ecological niche modeling is one approach to estimate the potential geographic distribution of a species and has been applied to map B. anthracis habitat suitability for several landscapes [10, 25–29]. These approaches relate environmental covariates with historic occurrence data on the species (e.g. outbreak locations) using pattern matching genetic algorithms or statistical approaches [30]. Occurrence data are generally obtained from a subset of the landscape accessible by the species [31] and related to larger landscapes described by environmental covariates [5]. Broadly, ecological niche modeling techniques can be divided into presence-absence and presence-only approaches. In the former, the user provides both occurrence locations and locations where the species was not detected. In the latter, the modeling algorithm will exhaustively sub-sample pseudo-absence points from a user-defined amount of the sampling area or background, which has been recommended when spatial information on species’ absence is unavailable [32] or occurrence points derived from idiosyncratic data sources, as is common with historical disease data. Many ecological niche modeling studies have used the presence-only approach to successfully predict the potential geographic limits of organisms in disease ecology [25, 33–35], biogeography [36] and conservation biology [37] across spatial scales. Here we used a presence-only modeling approach to predict the geographic distribution of B. anthracis across Australia.
A geographic information system (GIS) database of historical occurrence of livestock anthrax was constructed using anthrax locations heads-up digitized from Seddon and Albiston [12] and presence data provided by the Department of Economic Development, Jobs, Transport and Resources (DEDJTR) in Victoria, the Department of Primary Industries (DPI) in New South Wales and the Department of Agriculture, Fisheries and Forestry (DAFF) in Queensland, Australia (Fig 1). To ensure that all occurrence data were anthrax related deaths, only confirmed outbreaks (carcasses tested positive for B. anthracis or clinical confirmation), in the states of Victoria, New South Wales, and Queensland were retained for further analyses (Fig 2). An outbreak was defined as any location (infected farm or property) with one or more anthrax cases. Ideally, to predict the geographic distribution of B. anthracis, one would use occurrence data obtained from positive soil samples indicating the presence of the pathogen in the environment. Instead, outbreak locations were used as a proxy for B. anthracis occurrence data because anthrax-related death occurs after a relative short period of time following infection. For this study, we assumed that there were not great distances between infection source and carcasses. Additionally, in this study, B. anthracis infections and deaths occur on the same farms, and outbreak locations were represented by the geographic coordinates of infected farms. For each outbreak, the latitude and longitude were recorded along with additional attributes including date (day, month, and year), and total number of cases per animal species. Table 1 summarizes the spatial resolution and data collection methods for outbreaks for each state.
Duplicate coordinates were removed from the database for ecological niche modeling experiments. We then filtered the database to include one outbreak location per 8x8 km pixel, the resolution of environmental data used for modeling (hereafter referred to as the spatially unique presence points) [26]. The ecological niche modeling algorithm, the genetic algorithm for rule-set prediction (GARP) utilizes a single point per pixel to indicate the presence of B. anthracis. Using more than one point per grid cell for model development is equivalent of using the same data for both the training and testing of a GARP model, which can lead to inflation of accuracy metrics [26].
We used three groups of environmental coverages including bioclimatic (temperature and precipitation), edaphic (vegetation and soil properties), and topographic (altitude) factors known to influence the persistence of B. anthracis in the environment. Bioclimatic variables were downloaded at 30 arc-seconds (approximately 1x1 km spatial resolution) from the WorldClim website (http://worldclim.org) and are described in detail elsewhere [38]. Vegetation indices (8x8 km spatial resolution) were obtained from the Trypanosomiasis and Land Use in Africa (TALA) research group [39]. Soils data were extracted from the harmonized world soil database v1.2 (HWSD) available at the International Institute for Applied System Analysis (IIASA) (http://www.iiasa.ac.at) [40]. The HWSD data were available at 1x1km spatial resolution. The variables used for the ecological niche modeling are presented in Table 2.
Correlated environmental variables were eliminated using a Pearson correlation test to retain the variables presented in Table 2, which were then clipped to the boundary of Australia. Since the environmental data were at different spatial resolutions (1x1 km and 8x8 km), all data layers were resampled to the coarsest cell size (8x8 km) using the GARP Datasets extension in ArcView 3.3 (Environmental Systems Research institute, Redlands, CA).
In this study, we employed the genetic algorithm for rule-set prediction (GARP) and experiments were performed in DesktopGARP version 1.1.3 (DG). GARP is an expert-system, machine-learning algorithm that has been tested and widely used for species’ range prediction [32, 41–43]. Briefly, GARP develops a set of if/then logic string rules to relate observed occurrence data to environmental variables (bioclimatic, edaphic/substrate and topographic) [10]. Predicted presence or absence of a species within an ecological space are defined by one of four types of conditional rules including atomic, logit, and range or negated range rules [43]. Atomic rules use specific values or categories for each environmental variable (e.g. IF temperature = [35°C] AND precipitation = [325 mm] AND pH = [8.5] AND ndvi = [0.5] THEN species = PRESENCE/ABSENCE). Logit rules are fitted logistic regression functions (e.g. IF temperature*0.0078—precipitation*2.5 + pH*0.0039 + ndvi*0.0039 THEN species = PRESENCE). Upper and lower bounds for each environmental variable are specified in range rules (e.g. IF temperature = [14.6–19.5°C] AND precipitation = [348.25–757.51 mm] AND pH = [7.5–8] AND ndvi = [0.01–0.25] THEN species = PRESENCE). Negated range rules define conditions outside of variable ranges (e.g. IF NOT temperature = [15.5–28°C] AND precipitation = [143–1693 mm] AND pH = [6.5–8] AND ndvi = [0.25–0.45] THEN species = ABSENCE). The rules are developed through evolutionary refinement by testing and selecting rules on random draws of presence points from known occurrences data and pseudo-absences localities generated internally from the wider study area. A one-tailed significance χ2 test is then calculated in order to evaluate the quality of a rule at predicting the ecological distribution (presence or absence) of the species [43]. The stochastic process of deriving and evolving rules results in random walks through variable space resulting in multiple models. Each model is a set of 50 presence/absence rules that are projected onto the geographic landscape to estimate the potential geographic distribution of the species as a binary output (absence = 0, presence = 1).
Spatially unique presence points (N = 96 anthrax outbreak locations) were partitioned into training and independent test datasets for model building and evaluation. The geospatial modeling environment (GME, www.spatialecology.com) was used to randomly select 75% of the occurrence data points (n = 72) for models building and 25% of the points (n = 24) for calculating accuracy metrics [44–46]. To evaluate the effects of randomly sub-setting presence points, the selection process was repeated 10 times to develop 10 different GARP experiments. For each experiment, we ran up to 200 models with a maximum of 1,000 iterations and a convergence limit of 0.01. We allowed GARP to internally partition training data into a 75%/25% for model development and rule selection. We used the best subset procedure to select the best 20 models under a 10% hard omission threshold and a 50% commission threshold. Those 10 best subset models from each GARP experiment were then imported in ArcMap and summated using the raster calculator tool in the Spatial Analyst extension. The resulting composite raster layer, with pixel values ranging from 0 to 10, is a surface depicting the potential geographic distribution of B. anthracis in Australia. The higher the pixel values, the greater the potential that the environmental conditions will support pathogen persistence [25]. Model agreements from 0 to 5 were reclassified as not suitable and those greater or equal to 6 were considered most suitable to support B. anthracis persistence [26].
Model accuracy for each GARP experiment was calculated with the 25% independent testing data withheld from model building. Three metrics were used to measure accuracy: the area under curve (AUC) in a receiver operating characteristic (ROC) analysis, omission (a measure of false negatives) and commission (the proportion of the landscape falsely predicted as present) [44, 47]. The AUC was used to evaluate the overall performance of each composite predictive model (10-best subset model). An AUC of 0.5 indicates a random model whereas an AUC of 1 suggests a perfect model [42, 47]. Total omission was calculated as the percent of independent test points predicted absent by the composite predictive model and the average omission as the average omission across each of the 10 best models. Total and average commissions are the percent of pixels predicted as presence by the composite predictive model and the average of this value for the 10 best models, respectively [48].
Overall predicted area and the accuracy metrics were used to rank the 10 GARP composite predictive models. The best composite predictive model with the higher AUC value and lower omission error was retained to describe the potential geographic distribution of B. anthracis for Australia (S1 Fig) and to perform the rule-set analysis.
Rule types from each of the 10 best models in the highest ranked experiment were extracted using a python script (K.M. McNyset, US NOAA) to illustrate the relative number of each rule type [35]. From each rule-set, dominant rules that cumulatively predicted over 90% of the landscape were also identified in order to extract maximum and minimum values of the environmental variables of each presence rule type. The median of minimum and maximum values for each covariate in a given rule were calculated in Microsoft Excel 2010 and plotted as bar graphs to illustrate the ranges of each covariate [49].
Each GARP experiment reached convergence of accuracy (0.01) prior to the maximum 1000 iterations. The accuracy metrics of all ten GARP experiments are ranked and summarized in S1 Table. Metrics indicated all ten experiments were accurate and predicted highly similar geographic distributions. The potential geographic distributions of B. anthracis predicted by the 10 GARP experiments are illustrated in S1 Fig. Experiment number 5 had the highest AUC score and lowest omission errors; its AUC score was 0.966 and significantly different from a line of no information (p<0.01). And the total and average omission errors were 0.00% and 0.83% respectively, meaning that all independent testing data were correctly predicted by each of the 10 models in the best subset. The total and average commission for experiment number 5 were 6.24% and 12.15% of the landscape, respectively (Table 3).
Broadly, experiment number 5 predicted areas that stretch from north Victoria to northeast Queensland and running parallel with the eastern coastal region of Australia (Fig 3). The predicted areas also expand from northwest Victoria into small areas in the south of South Australia. In the southern part of Western Australia, the predicted geographic space of B. anthracis spans an area from the south to the southwest of the state. The interior of the country and the state of Tasmania were not predicted to be suitable for B. anthracis persistence (based on the conservative criteria of 6 or more models in a best subset).
S2 Table summarizes the rule types, number and proportion of each of the 10 best subset models from GARP experiment 5. Range rules represented 97.8% of the rules in rule-set, whereas negated rules accounted for only 1.8%. There were only 2 logit rules kept in experiment 5. There were no atomic rules.
Fig 4 illustrates narrow median range values for the following environmental variables: soil pH, calcium sulfate, organic content, and annual precipitation.
This study aimed to improve our understanding of the landscape ecology of B. anthracis and to predict the geographic distribution of the pathogen across Australia. We revised the geographic extent of the historical anthrax belt [17] that was defined by reported outbreaks and did not explicitly consider ecological conditions. Here we modeled the geographic distribution of B. anthracis based on environmental covariates known to be correlated with pathogen persistence enabling a quantitative redefinition of the anthrax belt. The distribution of B. anthracis has long been associated with environmental factors including soil and climatic parameters [20, 22, 50]. Incorporating these covariates into an ecological niche modeling framework provides a more accurate estimation of the geographic distribution of the pathogen, and therefore risk of anthrax, for Australia.
The predicted areas of B. anthracis are distinctly separated into two anthrax zones: the southeast-northeast and southwest corridors (Fig 3). The southeast-northeast corridor, hereafter referred to as the ‘redefined anthrax belt’, parallel the Eastern Highlands, stretching from north Victoria to central eastern Queensland through New South Wales where it traverses the western region of the Hunter Valley. The redefined anthrax belt extends far beyond Durrheim et al. [17] and captures many of the historical anthrax locations (Figs 5 and S2). In Victoria, the models also predicted the northern area of Goulburn Valley. This prediction includes South Australia along the Spencer Gulf on the southern coast of Australia in an area disjunct from the redefined anthrax belt. In a second disjunct area in Western Australia, the models predict part of the Nullarbor Plain on the Great Australian Bight coast, and the Darling Range in the Perth area.
The rule-set analysis indicated that the predicted ecological niche of B. anthracis is defined by a narrow range of high soil pH, low organic content, calcium sulfate, and annual precipitation (Fig 4). Across the best subset, a single rule per model captured nearly all of the predicted presence (S3 Fig). This is in contrast to the models developed for the United States, where presence rules captured presence with rules delineating eastern or western conditions [15] or presence rules in Kazakhstan dominated by northern and southern rules [49].
Historically, livestock anthrax was widespread in Australia, in particular Victoria and New South Wales. A comparison of past (1914–1963 and 1968–1995; Table 4) and recent epizootics (1996–2013; included in the model building process) confirmed a decrease in the number and spatial extent of anthrax outbreaks in the affected states, New South Wales and Victoria. This decrease is most likely due to the implementation of improved surveillance measures, livestock vaccination, the destruction of infected carcasses by burning, site decontamination and quarantine of affected livestock and properties [13]. A similar pattern of decrease of anthrax incidence was observed in the United States [15] and Ukraine [16]. In the United States, the use of an efficacious vaccine, along with better anthrax disease management strategies also resulted in a decrease in number of reported endemic counties from the 1950s onwards. In Ukraine, mass vaccination campaigns and effective control measures (burning of contaminated carcasses and sites decontamination) resulted in a reduction of anthrax foci from the early 1970s to the post-soviet period (1991 to the present day) [16].
It has been reported that during the mid-19th century, intensification of farming activities in Australia was associated with the use of unsterilized bone meal imported from India as a mineral supplement fed to livestock and as a fertilizer which led to the introduction of the pathogen [12]. The pathogen likely spread within Australia through the movement of diseased livestock along the southeast to northeast coastal corridor, and the contamination of stock routes with B. anthracis spores [12]. Contemporary livestock movement trajectories produced by the AusVet Animal Health Services [51] and East and Foreman [52] agree with historical livestock movements [12]. These movement trajectories perfectly intersect with areas of high model agreement.
The predicted geographic distribution of B. anthracis defines some suitable areas with no historical outbreak records, which may be due to over-prediction of the models. Nevertheless, it is worthwhile to note that over-predicting the geographic distribution of a species does not necessarily infer prediction error. The potentially over-predicted geographical distribution areas may represent an accurate illustration of the spatial extent of B. anthracis [53], despite the lack of presence records that could be used for testing the accuracy of our model in those areas. For example, using GARP, Blackburn et al. [25] successfully predicted suitable distributional areas for B. anthracis in the northwest corner of Montana, US, that had experienced anthrax outbreaks in 2005 although specific localities were unavailable for modeling. In South Australia, it has been reported that twenty three cattle died from anthrax in 1906 on a government farm at Islington [12], and six years later the disease also occurred in a metropolitan piggery at Unley after feeding pigs the carcasses of two horses, that had previously died from anthrax [12]. The source of infection at Islington was not mentioned, and the reported cases at Unley were not associated with direct contact to soil spores. However, these two areas overlap with the high agreement areas of our GARP models.
The models did not predict two outbreak locations that were withheld from model building, one in Western Australia and the other in Queensland. The anthrax cases in Western Australia were recorded in 1994 on three cattle properties in a localized area north of Walpole, where 29 cattle died from unknown sources from January to April 1994 [54]. In 1993, one cow died from anthrax on a grazing property near Rockhampton in Queensland, apparently from ingestion of contaminated feed [55]. In each case, the affected properties were outside of the predicted areas. Since anthrax is primarily a soil-borne disease, we hypothesize that these isolated cases, as well as the early outbreaks in the south coast areas of Victoria outside the predicted geographic distribution areas (S2 Fig), are likely attributable to causes other than ingestion of spores at grazing sites.
Anthrax was first recognized in Victoria in 1879 at Warrnambool followed by other areas in the south west of the state. The disease was later identified in southern and central Victoria following shipment of diseased sheep [12]. Seddon and Albiston [12] thought it is unlikely that the initial outbreak in the southwest of Victoria resulted from the spread of the disease from southern New South Wales, indicating that the introduction of the disease into this area came from other sources, followed by rapid spread over long distances to new areas by movement of stock by rail. The later distribution of the disease into the north of Victoria is considered to be most probably due to stock traveling over the border from NSW [12]. The distribution of anthrax throughout Victoria has changed over time with the majority of outbreaks post-1968 falling within the predicted zone and those prior to 1968 falling outside of this zone (Table 4). We hypothesize that the presence of disease incidents along the south coast of Victoria outside of the predicted geographic distribution prior to 1968 may represent constant reintroductions of the disease into these areas, given their proximity to ports and transport routes combined with possible short term survival and local spread.
This study redefines the anthrax belt of Australia, which is presently defined by the location of anthrax cases, by integrating ecological niche modeling and GIS. This approach provides insights about the ecological factors that limit the distribution of B. anthracis at the continental scale for Australia. The geographic distributions presented here can help inform anthrax surveillance strategies by public and veterinary health agencies.
|
10.1371/journal.ppat.1003807 | Variable Suites of Non-effector Genes Are Co-regulated in the Type III Secretion Virulence Regulon across the Pseudomonas syringae Phylogeny | Pseudomonas syringae is a phylogenetically diverse species of Gram-negative bacterial plant pathogens responsible for crop diseases around the world. The HrpL sigma factor drives expression of the major P. syringae virulence regulon. HrpL controls expression of the genes encoding the structural and functional components of the type III secretion system (T3SS) and the type three secreted effector proteins (T3E) that are collectively essential for virulence. HrpL also regulates expression of an under-explored suite of non-type III effector genes (non-T3E), including toxin production systems and operons not previously associated with virulence. We implemented and refined genome-wide transcriptional analysis methods using cDNA-derived high-throughput sequencing (RNA-seq) data to characterize the HrpL regulon from six isolates of P. syringae spanning the diversity of the species. Our transcriptomes, mapped onto both complete and draft genomes, significantly extend earlier studies. We confirmed HrpL-regulation for a majority of previously defined T3E genes in these six strains. We identified two new T3E families from P. syringae pv. oryzae 1_6, a strain within the relatively underexplored phylogenetic Multi-Locus Sequence Typing (MLST) group IV. The HrpL regulons varied among strains in gene number and content across both their T3E and non-T3E gene suites. Strains within MLST group II consistently express the lowest number of HrpL-regulated genes. We identified events leading to recruitment into, and loss from, the HrpL regulon. These included gene gain and loss, and loss of HrpL regulation caused by group-specific cis element mutations in otherwise conserved genes. Novel non-T3E HrpL-regulated genes include an operon that we show is required for full virulence of P. syringae pv. phaseolicola 1448A on French bean. We highlight the power of integrating genomic, transcriptomic, and phylogenetic information to drive concise functional experimentation and to derive better insight into the evolution of virulence across an evolutionarily diverse pathogen species.
| Pseudomonas syringae are environmentally ubiquitous bacteria of wide phylogenetic distribution, which can cause disease on a broad range of plant species. Pathogenicity requires the master regulator HrpL. HrpL controls the activation of virulence factor genes, including those encoding the type III secretion system which facilitates translocation of bacterial proteins into host cells. Here we overlaid transcriptome profiling of genes onto their phylogenetic distribution by characterizing the HrpL regulon across six diverse strains of P. syringae. We identified novel putative virulence factors, discovered two novel effector families, and functionally characterized an operon most likely involved in secondary metabolism that we show is required for virulence. We demonstrated that the size and composition of the HrpL regulon varies among strains, and explored how genes are recruited into, or lost from, the virulence regulon. Overall, our work widens the understanding of P. syringae pathogenicity and presents an experimental paradigm extensible to other pathogenic bacterial species.
| Many Gram-negative bacteria attach to host cells and translocate effector proteins into them via type III secretion systems (T3SS). Such systems are necessary for pathogenesis, are horizontally transferred across species, and are accompanied by dynamically evolving repertoires of type III effector (T3Es) genes [1], [2]. The T3SS is essential for Pseudomonas syringae pathogens to thrive in plant tissues. P. syringae represents an excellent example of the plasticity of T3E repertoires [3]. Despite a collectively broad host range for the species, individual isolates of P. syringae typically display pathogenic potential on a limited set of plants and either elicit immune responses, or simply fail to thrive on other plant species. Strains can be isolated from diseased plants, as epiphytes from healthy plants [4], and from various environmental sources [5].
The hrp/hrc group I T3SS is essential for P. syringae pathogens to cause disease on plants [1], [6]. The genes that encode the hrp/hrc T3SS and accessory proteins are clustered in a conserved pathogenicity island in P. syringae [7]. The genes for the associated T3Es can be scattered across the genome, often in association with mobile elements indicative of horizontal transmission [8]–[10]. Each strain's T3E repertoire ranges from 15–30 genes sampled from at least 57 different families and these collectively modify host cell biology to suppress immune response and favor bacterial proliferation and dispersion. However, the action of individual T3E proteins can be recognized by plant host disease resistance proteins, and this triggers immune responses sufficient to limit pathogen growth [11]. The conflicting selective pressures to retain a collection of T3E sufficient to suppress host defenses without triggering effector-specific immune responses [11] drives diversity in the suites of T3Es in plant pathogenic P. syringae isolates [3].
Transition from saprophytic to epiphytic or pathogenic lifestyle requires significant transcriptional reprogramming. Expression of genes encoding the P. syringae T3SS structural components and the associated T3E suite is controlled by the ECF-type sigma factor HrpL [12]–[14]. The expression of hrpL is induced in bacteria that encounter the leaf environment [13]. Subsequently, HrpL binds to promoters carrying a “hrp-box” consensus sequence to up-regulate the expression of the corresponding gene(s) [12]–[15].
Previous studies in P. syringae identified proteins that are neither T3Es nor structural components of the T3SS (hereafter, non-T3Es), but are HrpL-regulated [3], [16]–[19]. Non-T3Es coordinately regulated with the T3SS and its substrates were also found in other T3SS-expressing plant pathogens such as Erwinia amylovora [20], Ralstonia solanacearum [21], Xanthomonas campestris pv. vesicatoria [22], [23] and Pectobacterium carotovora [24]. Some HrpL-regulated non-T3E genes affect virulence on host plants in the well-studied strain P. syringae pv. tomato DC3000 (PtoDC3000); these include the corR regulator of coronatine toxin production [18], [25]. Notably, CorR expression is not HrpL-regulated in other strains, such as P. syringae pv. glycinea PG4180 [26].
Multi-Locus Sequence Typing (MLST) separates plant pathogenic P. syringae into at least 5 distinct phylogenetic groups [3], [27]. The fifth group, represented initially by P. syrinage pv. maculicola ES4326, was recently renamed P. cannabina pv. alisalensis ES4326 [28]. Many P. syringae genome sequences are now available, including three closed genomes from isolates representing major pathogen clades [29]–[31], and ∼120 additional draft sequences. Newly sequenced genomes also trace P. syringae disease outbreaks across the globe and over time [3], [32]–[39] attesting to the continued importance of the species. Recently, isolation and sequencing of saprophytic and epiphytic strains provided insight into a subgroup from group II that carries a non-canonical T3SS [40]. To date, transcriptome analyses using high throughput short read cDNA sequencing (RNA-seq) have been applied only to PtoDC3000, providing a well-curated reference gene annotation, but not specifically informing studies of the HrpL regulon [41]–[44].
In this study, we defined the HrpL regulon of six distinct strains of P. syringae with complete or draft genomes using RNA-seq coupled with the GENE-counter software package [45]–[47]. We sought primarily to compare the diversity of non-T3E HrpL-regulated genes between strains and secondarily to determine if there were additional type III effectors not found in our DNA-based analyses [3]. We detect non-T3E genes regulated directly or indirectly by HrpL. Those directly regulated by HrpL are distributed throughout the P. syringae clades in a mosaic pattern. However, most are either absent or not HrpL-regulated in MLST group II. We demonstrate that a novel cluster of non-T3E genes is required for P. syringae pv. phaseolicola 1448A virulence. We also identified two novel T3E families from a previously understudied clade. Our study reveals the mechanisms for gene recruitment into, and loss from, the key virulence regulon in P. syringae, and provides a roadmap for future functional studies.
We defined the HrpL regulons of P. syringae pv. phaseolicola strain 1448A (Pph1448A), P. syringae pv. lachrymans strain 107 (Pla107) representing MLST group III; P. syringae pv. syringae strain B728a (PsyB728a), P. syringae pv. japonica strain MAFF 301072 PT (Pja) representing MLST group II; P. syringae pv. tomato strain DC3000 (PtoDC3000) representing MLST group I and P. syringae pv. oryzae strain 1_6 (Por), belonging to the relatively poorly studied clade, MLST group IV [3], [27]. The native hrpL gene from each isolate was cloned downstream of an arabinose-inducible promoter for controlled, high-level expression in the strain of origin. Isogenic strains carrying either the appropriate hrpL construct, or an empty vector (EV) as negative control, were grown with arabinose to induce the expression of the cloned hrpL gene in a minimal medium [19]. Expression of the native hrpL was repressed by addition of peptone to the media [48]. Figure S1 depicts our experimental pipeline and control validation.
We generated Illumina cDNA libraries from two biological replicates of each strain. Because our goal was to compare transcript abundance more than to improve annotation of transcribed genes, we used a simple cDNA method to minimize the RNA processing steps where transcripts could be lost. Therefore, we did not enrich for 5′ ends or distinguish transcript orientation. Transcript abundance was compared between isogenic HrpL and EV samples using GENE-counter [45]. Similar to other RNA-seq analysis methods like EdgeR or DESeq [49], [50], GENE-counter determines differential expression. While EdgeR and DESeq use the standard negative binomial distribution, GENE-counter relies on the negative binomial p distribution which better accounts for the over-dispersion observed in mRNA-seq data [51]–[53]. We bootstrapped the GENE-counter output for each isolate (Materials and Methods) to control for noise introduced by sample normalization. Between 1.6 and 5.6 million unambiguous reads per sample (mapping to only one location in the reference genome) were used for our analyses (Table 1). The sequencing depth ranged from 9.5 to 16.2 times the genome size, with the exception of the PsyB728a samples, which we sequenced to higher coverage (Table 1). On average 93.5% of the total number of annotated coding genes in a genome were covered by at least one read in at least one sample (Table 1). Bootstrapped-GENE-counter analysis established a median read count for every sample, a median q-value and a B-value, for every gene covered by at least one read in one biological replicate (Table S1). Genes not covered by any unambiguous reads are not represented in our GENE-counter output. The B-value represents the percentage of bootstraps in which a particular gene was called differentially expressed.
We further considered only genes with B-values≥50%. Like all “significance thresholds” the B-value cut-off is somewhat subjective. We selected a B-value of 50% to apply to all genomes because this threshold captured 95% of the known HrpL-regulated genes identified in our control genome, PtoDC3000, with a median q-value greater than 0.05. We identified between 59 to 192 genes differentially expressed across the strains (Table S1). For all strains, the large majority of the differentially expressed genes were up-regulated (between 53 and 180 genes, Table 2). These genes mainly encode T3SS components and known T3Es. Surprisingly, we identified few HrpL-down-regulated genes (Table S2): ranging from none in Pph1448A to 45 in Pla107. Genes called down-regulated in our analysis had relatively low q-values, reflecting low differences in read coverage between HrpL and EV samples. Lan et al. 2006 and Ferreira et al. 2006 identified down-regulated genes in their microarray studies for PtoDC3000. However, almost no overlap was found between the list of down-regulated genes from previous studies and ours, indicating that the down-regulated genes identified are most likely neither biologically, nor statistically robust, and thus unlikely to be biologically relevant. In contrast, there was stronger overlap between our HrpL-induced genes and those shared between these earlier studies (see below). Down-regulated genes were therefore not further analyzed. Finally, we manually inspected and curated all genes with B-values greater than or equal to 50% to define the HrpL regulon for each strain (Table 2, Table S3; Table S4; Materials and Methods).
To evaluate the reproducibility of our method, we compared the read coverage within and between biological samples for all PtoDC3000 genes covered by at least one read in our normalized GENE-counter data set. Biologically replicated samples had highly correlated results (R2 = 0.93 between EV replicates 1 and 2; R2 = 0.96 between conditional expression replicates HrpL1 and HrpL2, Figure 1A, lower panels). Comparing HrpL and EV replicates from two biological replicates, the majority of the data points correlate and cluster around the trend line (Figure 1A, upper panels). The outlier data points in red represent genes defined as differentially up-regulated by GENE-counter and having a B-value≥50%. We plotted the log of the median q-value of each PtoDC3000 gene defined to be differentially up-regulated (before manual curation) and their corresponding B-values ranked from smallest to largest (Figure 1B). As expected, genes with highly significant q-scores also have high B-values. Several genes not previously reported to be HrpL-regulated (marked in red) had more significant q-value scores (3.8E-02) than avrE (marked in blue), a well-characterized conserved HrpL-regulated type III effector [54].
We analyzed the same PtoDC3000 RNA-seq data set using either the complete PtoDC3000 genome sequence [30] or a draft PtoDC3000 genome sequence [36] as references. The draft genome sequence covers 85% of genes at over 90% of their length [36]. Using either the complete or the draft genome as a reference resulted in similar sequencing depths (Table 1). Using the draft genome as a reference, GENE-counter identified 124 HrpL-upregulated genes out of the 133 found using the complete PtoDC3000 genome (Table 2). Most of the genes that were not identified as differentially expressed using the draft genome were missing from the draft genome (data not shown). The high correlation between the Log(median q-value) of genes in the two data sets (Figure 1C) indicates that our method will effectively identify the majority of genes of the HrpL regulon from P. syringae isolates for which only a high quality draft genome is available.
To further validate our pipeline to define HrpL–regulated genes, we compared our manually curated list of 110 PtoDC3000 HrpL-regulated genes (Table 2) to HrpL-regulated genes identified by three previous studies: one promoter probe study using an arabinose-inducible hrpL gene and two custom microarray analyses which compared expression between wild type and hrpL deletion mutant strains [16], [17], [19]. These studies produced largely overlapping, but not identical, lists of putatively HrpL-regulated genes (Table S5). Our PtoDC3000 HrpL-regulated gene set included 57 out of the 66 genes previously identified as HrpL-regulated in at least two of the previous studies (Table S5), even though our induction and analysis methods differed from these studies. 96 of the 110 genes we identified were also found to be HrpL-regulated in at least one of the previous studies [16], [17], [19] or were downstream genes in HrpL-regulated operons (Table 2). Overall, we found 91% of the previously identified HrpL-regulated genes in PtoDC3000. Our analysis also identified 14 novel HrpL-regulated genes (Table 2); six out of eight tested were confirmed to be HrpL-regulated using qRT-PCR (Table 3, see below).
Notably, four of the nine missing genes were not present in our laboratory strain, which has lost part of the PtoDC3000 plasmid A. One gene, shcA (PSPTO_5353) was found differentially expressed in our analysis but had a B-value less than 50%. Further, GENE-counter discards RNA-seq reads that map non-uniquely to more than one location in the genome, and HrpL-regulated duplicated genes account for three missing PtoDC3000 genes: T3E genes hopAM1-1(PSPTO_1022) and hopQ1-2 (PSPTO_4732), and the non-T3E gene plcA2 (PSPTO_B0005) (Table S5). Finally, hopK1 (PSPTO_0044), was covered by RNA-seq reads but the differences in expression in HrpL and EV treatments were not statistically significant (Table S1, S5).
Two previous studies focused on the identification of HrpL-regulated genes in Pph1448A [18], [19] and identified 43 HrpL-regulated genes comparing expression between wild type and hrpL mutants. We identified 35 (∼80%). Four of the missing eight genes were covered by reads but not found significantly differentially expressed, hopAK1 (PSPPH_1424), a gene encoding a MarR transcriptional regulator (PSPPH_1519), avrRps4 (PSPPH_A0087), and hopAS1 (PSPPH_4736). Those four genes had a median read coverage ranging from 100 to 1000, indicating that the absence of differential expression in our analysis is not due to weak or undetectable levels of expression. One, PSPPH_2294 is a pseudogene. PSPPH_1525 encoding a putative effector related to Ralstonia Hpx30 [55], PSPPH_A0009 and A00075 encoding truncated hopW1 are duplicated and had very low to no read coverage (Table S5). Our GENE-counter analysis pipeline results are consistent with previous transcriptional studies, reinforcing the validity of our methods. Additionally, we identified robustly HrpL-induced genes that were not previously identified.
We identified between six and 32 genes previously not known to be HrpL-regulated in each strain with corresponding q-values ranging from E-02 to E-54 (Table 2, Table S3). Some of these are shared across strains. We could not identify a consensus upstream hrp-box in the promoters of several, and suggest that these could be indirectly activated by HrpL. We performed qRT-PCR using samples derived from strains expressing HrpL in the pBAD system and confirmed 19 of 23 tested (Figure S2). Additionally, we confirmed HrpL-dependent expression of 19 genes out of 20 tested, by comparing wild type expression with expression in a hrpL deletion mutant in hrpL-inducing minimal medium (Table 3, Figure S3). We observed a high correlation between RNA-seq data and either qRT-PCR profiling method, especially for genes with a q value>E-03 (Table 3, Figures S2, S3). In sum, we identified the majority of previously identified HrpL-regulated genes in two well-studied strains and we confirmed wild type HrpL regulation for nearly all of the newly identified members of this key virulence regulon.
Most of the known T3E and candidate T3E genes in our tested strains and those previously defined by similarity and/or functional criteria were included in the HrpL regulons we defined in our RNA-seq analyses (Figure S4). Most of strains used in this study had previously been screened for novel type III effector genes by functional translocation assays with the exception of Por and Pja [3], [19]. Therefore, we searched the Por and Pja HrpL regulons for potential novel effector genes based on the criteria of having an identifiable upstream hrp-box sequence and no homology to previously identified T3E families. We chose six Por genes (Porcurated_02784, 04644, 04640, 03530, 02145, and 04371) to investigate as potentially encoding novel T3Es. Pja also carries a gene homologous to Porcurated_04644; but only the Por allele was tested. All six putative T3E were tested for their ability to be translocated via a native T3SS using an established assay [56] (Materials and Methods) from PtoDC3000D28E, an “effector-less” PtoDC3000 strain [57]. Only PtoDC3000D28E carrying Porcurated_02784-Δ79avrRpt2 or Porcurated_04640-Δ79avrRpt2 triggered a Hypersensitive Response (HR) in Col-0 (Figure 2A). We verified that HA-tagged versions of all six T3E candidates were expressed in PtoDC3000D28E indicating that lack of HR in our translocation assay was unlikely due to a lack of protein accumulation (Figure 2B). No HR was observed in the rps2 mutant, indicating that the response was avrRpt2-specific and not the result of toxicity. These two new P. syringae effectors will henceforth be referred to as HopBH1Por and HopBI1Por according to proposed T3E naming guidelines [58].
None of the 19 P. syringae strains for which we previously performed comparative genomic analysis encode either hopBH1 or hopBI1 [3]. However, each can be found in P. syringae strains isolated from various sources ranging from non-symptomatic plants to snow [33], [35], [40], [59]–[62] (Figure S5). Amino acid sequence alignments suggest that HopBH1 is a bi-modular effector exhibiting sequence conservation within its C-terminal domain and sequence diversity toward its N-terminal half (Figure S6). In the non-pathogenic strain Psy642, the putative HopBH1 protein appears to have been disrupted by a frameshift mutation, leading to two putative open reading frames designated as ORF29-30 [40]. Phylogenetic analysis of strains carrying either hopBH1 and/or hopBI1 indicates that both effector genes occur with a mosaic distribution across the P. syringae, but are absent from the phylogenetic group III [3], [27] (Figure S5). Neither HopBH1 nor HopBI1 contain known protein folds, nor do they display sequence or structural homology to proteins of known function.
The composition of the HrpL regulon across strains was surveyed by functional classification based on protein annotation and sequence homology determined by BLASTP (Table S6). As summarized in Figure 3 and Table S7, PtoDC3000 and Por possess the largest and most diverse HrpL regulons among the sampled strains, while the Group II strains Pja and PsyB728a have the smallest. We are confident that the less complex HrpL regulons are not a sampling artifact, because the data collected from Pja has a transcriptome depth similar to the other strains, and the PsyB728a HrpL regulon was sampled at relatively high depth compared to our other transcriptomes. We conclude that HrpL regulons vary in size and composition across the P. syringae phylogeny.
We observed variable HrpL-dependent expression for several highly conserved non-T3E genes present in all six strains (Table S6). We identified polymorphisms in the hrp-box sequences from two of these (Figure 4A). In the first case, new HrpL-regulated genes we identified, PSPTO_2130, PSPPH_1906 and Lac107_00061530, are orthologs that encode a DNA-binding response regulator. HrpL-dependent induction was confirmed by qRT-PCR (Table 3, Figure 4B). Orthologous genes are also present in Pja, PsyB728a, Por (Pjap_00016990, Psyr_1940, and Porcurated_00527, respectively) but were not identified as differentially expressed (Table S1). PSPTO_2130 and all of its orthologs have conserved hrp-box motifs. However, the promoters of the orthologs from Por and all other group II strains contain single nucleotide polymorphisms (in red, Figure 4A) in the consensus hrp-box sequence. Our RNA-seq data suggested that expression of these polymorphic alleles was not HrpL-dependent, a finding confirmed by qRT-PCR performed with both of our HrpL-regulation experimental tests (Figure 4B, Figure S7A).
PSPTO_2130 and its orthologs are part of a putative operon composed of four genes, PSPTO_2128-2131 (Figure S8A). Unusually, the hrp-box sequences were located within the first ORF of the putative operons of PSPTO_2130 and its orthologs. We monitored HrpL-dependent expression using qRT-PCR of all genes from PSPTO_2131 to 2128 from three strains (Figure S8B, C, D). In none of these strains was the first gene of the operon, containing the putative hrp-box, differentially expressed. By contrast, HrpL-dependent expression was observed for genes downstream of the predicted hrp-box, including coding sequences, PSPTO_2130 and PSPPH_1906, in all but the group II reference strain PsyB728a (Figure S8). Deletion mutants in PtoDC300 and Pph1448A of PSPTO_2130 and PSPPH_1906 did not display any growth defect on Arabidopsis accession Col-0 or French bean cultivar Tendergreen (susceptible to PtoDC3000 and Pph1448A, respectively) (data not shown). Thus, the role of PSPTO_2130 and its orthologs in virulence remains unclear.
In the second case, PSPTO_2105 and its orthologs, which encode a putative ApbE-family protein, are highly conserved across P. syringae and are HrpL-regulated in Pph1448A, Pla107, PtoDC3000 and Por but not in the group II strains PsyB728a or Pja (Table S5, S6). qRT-PCR (Figure 4C, Figure S7B) support our RNA-seq data. PSPTO_2105 is required for full virulence of PtoDC3000 on Arabidopsis [18]. We also observed significantly reduced virulence when we tested two independent deletion mutants of the Pph1448A ortholog PSPPH_1855 for growth on the native host, French beans (Figure S9). Every group II strain analyzed has variations in the otherwise well conserved hrp-box sequence in at least two positions (Figure 4A). Collectively, these data demonstrate that promoter erosion within the hrp-box is a mechanism to remove genes from the virulence regulon.
Both PsyB728a and Pja appear to have relatively small HrpL regulons; both belong to the MLST group II. To address whether this was a general feature of group II strains, and to address the distribution of the genes that we identified experimentally across the phylogeny, we extended our investigation of non-T3E HrpL regulon diversity to BLAST homology searches of 44 sequenced Pseudomonas spp. strains [3], [35], [63]. Our non-T3E gene search set included genes likely to be directly HrpL-regulated, derived from either previous studies [19], [64] or this study. From our study, these included genes we experimentally confirmed for HrpL-dependent expression, genes that encoded proteins found not to be translocated, or genes unlikely to encode a translocated product by annotation. We removed T3SS genes and known T3Es (Figure 5). Most of the directly HrpL-regulated non-T3E genes we identified are absent from group II strains, but distributed across strains from groups I and III. Some are present in the previously described group IV and V, as well as the novel MLST groups VII, IX, X (Berge et al., personal communication, Figure S5) for which we had limited sampling. Further, the promoters of group II homologs of Porcurated_02977, 01635 are divergent, and lack canonical hrp-boxes (data not shown). Thus, not only do group II strains possess lower numbers of known T3E genes on average than the other phylogroups, group II strains also possess fewer non-T3E genes in their HrpL regulon suggesting a potential shift in virulence mechanisms of this clade [3].
Both Pph1448A and Pla107 contain avrD, a gene required for synthesis of syringolides, small molecules sufficient for HR on soybean cultivars expressing the Rpg4 disease resistance gene [65]–[67]. avrD is a non-T3E gene, as defined above (Figure 5), and its expression in E. coli is sufficient for production of syringolides [65]. RNA-seq analysis identified a series of orthologous, HrpL-regulated genes directly downstream of avrD in both Pph1448A and Pla107 (Table S3, S6). In Pph1448A, those genes are arranged in two clusters composed of PSPPH_A0112-A0110 and PSPPH_A0109-A0106, which are flanked by transposable elements (Figure 6A). While most of these genes seem to encode hypothetical proteins, PSPPH_A0112, A0109, A0108, A0107 encode putative enzymes: a phosphoglycerate mutase, a sulfotransferase, an amino transferase, and an oxidoreductase respectively. We confirmed the HrpL-dependent expression of PSPPH_A0112, A0110, A0109, and A0107 (Table 3, Figure S2, and S3). This operon is typically found as a presence/absence polymorphism; when present, it is almost always downstream from avrD (Figure 6B). PSPPH_A0111 corresponds to a 99 bp sequence present in Pla107, P. syringae pv. mori (Pmo), P. syringae pv. glycinea R4 (PgyR4), P. syringae pv. tomato T1 (PtoT1), and P. syringae pv. actinidiae (Pan) but not annotated as an ORF, thus it is not represented in the graphical representation of the conserved neighborhood region (Figure 6B). In P. syringae CC1629, this putative operon appears to have been disrupted by insertion of a transposable element. In P. syringae pv aesculi 0893_23 (Pae) this locus is not entirely sequenced. To determine whether avrD is part of an operon with PSPPH_A0112-A0106, we used RT-PCR to confirm that the intragenic regions between avrD/PSPPH_A0112 and between PSPPH_A0110/PSPPH_A0109 were transcribed in wild type Pph1448A but either very weakly or not at all in the ΔhrpL mutant (Figure 6C).
We generated two independent deletion mutants for avrD and PSPPH_A0107 (ΔavrD #1 and 2, ΔPSPPH_A0107 # 1 and 2, respectively) and tested their growth on French bean cv. Tendergreen (Figure 6D). All mutants displayed reduced growth compared to wild type Pph1448A (Pph), indicating that both avrD and PSPPH_A0107 are required for full virulence on cv. Tendergreen. We confirmed that the HrpL-dependent expression of several downstream genes was not disrupted by mutations (Figure S10). However, PSPPH_A0112, A0107 and A0106 were consistently slightly up-regulated in avrD mutants compared to the wild type. The intact remaining hrp-box is closer to PSPPH_A0112-A0106 in the avrD mutants, which could account for increased transcript levels. Additionally, these data could explain why the avrD mutants displayed a reduced growth defect compared to the ΔPSPPH_A0107 mutants (Figure 6D). We speculate that these non-T3E genes are involved in the synthesis of a secondary metabolite(s) required for virulence of Pph1448A.
P. syringae is a broadly distributed and agronomically important pathogen of many plant species. Full virulence for many strains within this species requires expression of genes induced by the sigma factor HrpL, but the HrpL regulon has only been systematically surveyed using microarrays in PtoDC3000 [16], [17] and to a limited extent by promoter probe studies in a few strains [3], [19]. Using RNA-seq, we successfully defined HrpL regulons across six phylogenetically diverse strains. We benchmarked our data set with previous transcriptional studies of two reference genomes [16]–[19] (Table 2) and with qRT-PCR analysis (Table 3, Figure S2, Figure S3). Our approach allowed us to efficiently define the HrpL regulon of multiple strains, even those for which only draft genome sequence is available. We found a plethora of non-T3E genes in these regulons and experimentally verified both newly identified T3Es and non-T3E virulence factors. Additionally, we identified a variety of mechanisms that could drive recruitment into and loss from, the main virulence regulon of P. syringae.
We identified HopBH1Por and HopBI1Por, defining two novel effector families. Both have a mosaic phylogenetic distribution across P. syringae [35], [40], [63] (and an unpublished strain, TLP2, JGI taxon ID: 2507262033). Both are present in CC1513 and CC1629, two other strains belonging to the MLST group IV. They appear to be absent from sequenced MLST group III strains. HopBH1 has a bi-modular structure. The ∼170 amino-acid N-terminus is divergent compared to the relatively well conserved ∼250 amino acid C-terminal domain across HopBH1 alleles (Figure S6). The HopBH1 C-terminal domain is 50% identical to a protein from P. fluorescens SS101 which lacks a putative hrp-box or a T3SS secretion competent N-terminal sequence [68], suggesting that it may have been recruited as an effector by N-terminal assortment [69]. Several putative proteins present in Pantoea, Serratia, Burkholderia species, as well as Myxobacteria, display ∼50% identity with the HopBH1 C-terminal domain. Remarkably, about 150 amino acids of the HopBH1 C-terminal domain also shares 40% identity with part of the ∼1000 amino acid long P. savastanoi pv. savastanoi NCPPB3335 HrpK. Notably, this hrpK gene (PSA3335_2516) is from a rhizobia-like type III secretion and is different from the hrpK(Pto) (PSA3335_1389) of canonical T3SS conserved in plant pathogenic P. syringae [70]. HopBI appears to be confined to Pseudomonas. Neither HopBH1, nor HopBI1 display similarity to known-effectors. Their virulence functions remain to be determined.
Although analysis of type III virulence systems focuses mainly on the characterization and function of T3SS and T3E proteins, several non-T3E genes are co-regulated with the T3SS. They encode hypothetical proteins, transporters, or enzymes likely involved in secondary metabolism (Figure 5). In contrast to T3E genes, for which functional redundancy is predominant and generation of multiple effector mutants is often required to affect virulence [54], [57], [71], [72], several single knockout mutants of non-T3E HrpL-regulated genes in PtoDC3000 and Pph1448A displayed reduced virulence on Arabidopsis and beans [18], [73]. In general, little is known about the non-T3E genes in HrpL regulons, but homology provides reasonable scenarios for several that we identified, and we functionally validated others (below).
Among our collection of diverse HrpL-regulated, non-T3E genes, none are present in the HrpL regulon of all six strains tested, and nearly all are distributed in a mosaic pattern among the genomes of available strains (Figure 5).
PSPTO_0370 and orthologs encode a MATE efflux transporter present in an operon with iaaL which is involved in auxin conjugation to IAA-Lys [74]. Porcurated_02977 encodes a putative indole-3-glycerol phosphate synthase. Both potentially alter auxin signaling and could interfere with the balance between immune response and growth and development [75].
Several other putative transporters were identified as HrpL-regulated. PSPTO_2691 encodes a putative membrane protein TerC; PSPTO_0871 a putative macrolide efflux protein; Porcurated_01635 a putative threonine efflux protein; and PSPTO_0838 a putative major facilitator family transporter. Co-regulation of putative transporters with the T3SS suggests that promotion of nutrient acquisition, export of secondary metabolites, or detoxification of plant-encoded antimicrobials are important features of the virulence regulon.
PSPTO_0834, encoding a putative alcohol dehydrogenase, is the first gene of a putative operon comprising five genes (up to PSPTO_0838). This operon includes genes of unknown function, genes encoding a putative bifunctional deaminase-reductase enzyme and a transporter. The function of this operon remains unknown but at least PSPTO_0834 is required for full virulence of PtoDC3000 on Arabidopsis [18].
The PSPTO_0873-0875 putative operon is widely distributed across Pseudomonas and Erwinia species and also present in Pantoea stewartii pv. stewartii DC283. In Erwinia and P. stewartii, this operon is physically linked to the T3SS and is HrpL-regulated. PSPTO_0873 is a putative amidinotransferase that makes ornithine and homo-arginine from arginine and lysine. Ornithine or homo-arginine may be then incorporated into a tri- or di-peptide natural product generated by the rest of this operon. Most interestingly, hsvC, hsvB, hsvA from Erwinia amylovora, corresponding to PSPTO_0873-0875, are required for full virulence on apple shoots [76].
PSPTO_2105 and orthologs encode a protein similar to ApbE from Salmonella typhimurium involved in thiamine synthesis. ApbE was identified through the analysis of several mutants defective in thiamine biosynthesis, and was implicated in iron-sulfur cluster biosynthesis/repair, as well as FAD binding [77]–[79] suggesting a role during oxidative stress [78]. PSPTO_2105 is required for full virulence of PtoDC3000 on Arabidopsis [18]. We extend this finding by showing that the PSPPH_1855 ortholog of PSPTO_2105 is required for full virulence of Pph1448A on French bean (Figure S9).
PSPTO_2130 and orthologs encode LuxR family DNA-binding response regulators that may be involved in regulation of regulons downstream of HrpL. Our deletion mutants of this gene in PtoDC3000 and Pph1448A, or of the entire PtoDC3000 operon, did not alter growth on Arabidopsis or French bean cv. Tendergreen, respectively (data not shown), undermining the probability of a necessary function during plant colonization in our experimental conditions. However this operon is conserved across Pseudomonas, and PFLU_2937, the ortholog of PSPTO_2129 from P. fluorescence SBW25, was identified as a plant-induced gene [80]. It therefore remains plausible that this operon is involved in plant association.
Porcurated_04644 appears to encode a putative RNA N-methyltransferase, while the hypothetical protein Porcurated_03530 has homology to FliB which, in Salmonella, is responsible for methylation of flagellin [81]. We speculate that both may be involved in modification of conserved molecules known to induce host defense responses [82]–[84].
avrD is widely distributed across bacteria and is involved in the synthesis of syringolides [85]. Syringolides are elicitors of cell death in soybean expressing the Rpg4 disease resistance gene [86], [87]. The putative function of avrD is discussed below.
One of our most striking comparative observations is the relatively small size and diversity of the HrpL regulons of the phylogenetic group II strains PsyB728a and Pja. We observed that most of the non-T3E genes known to be HrpL-regulated in other strains are not present, or lack HrpL-regulation in group II strains, underpinning the conclusion that the limited regulon observed for PsyB728a and Pja can most likely be generalized to all group II strains (Figure 5). They also contain fewer T3Es than the other clades [3]. The group II strains carry genes for phytotoxins not shared by other P. syringae groups. Expression of these phytotoxins is not regulated by HrpL, and could compensate for missing T3E functions, making a smaller T3E repertoire sufficient to suppress plant defenses [3].
Turnover within the HrpL regulon is known to be influenced by gene gain and loss, mediated by association of genes within the regulon with mobile elements and horizontal gene transfer (data not shown, Figure 6 A, B). However, we also observed that all the group II strains analyzed here have polymorphisms in the hrp-box sequence that correlated with the loss of HrpL-dependent regulation of PSPTO_2105 and orthologs (likely encoding AbpE). Several different polymorphisms within the hrp-box were observed, suggesting independent mutational events (Figure 4). Additionally, the group II strain orthologs of PSPTO_2130 (LuxR family), carry nucleotide polymorphisms in the consensus hrp-box, and are not HrpL-regulated (Figure 4). Orthologous genes from Por also display nucleotide variation in this hrp-box, also leading to loss of HrpL-regulation. The substitution patterns of these alterations suggest multiple, independent losses of HrpL-regulation. PSPTO_2130 and its orthologs are part of an operon where the consensus hrp-box is embedded within the first ORF in this operon (Figure S8) and is thus likely to be constrained by the genetic code. Interestingly, PSPTO_2130 and its orthologs have variation in the second half of the hrp box where CCAC is replaced by TCAC. This hrp-box motif, while uncommon, is also found in PSPTO_0370, PORcurated_01251 (hopAO1Por), and Pjap_00002060 (hopC1Pja), each of which we defined as HrpL-regulated.
The promoter erosion we observe could be driven by negative host selection pressure, or weak selection for maintenance of HrpL regulon membership combined with subsequent drift. Similarly, reversion of at least the SNPs could quickly recruit genes back into the HrpL regulon. Because the ORFs have not accumulated stop mutations, these promoter mutations are either relatively recent or there is active maintenance of the ORF sequence, perhaps for expression under different conditions.
Horizontal transfer or other types of recombination could explain how 5′ regions diverge and how these regions and associated genes are recruited in to the HrpL regulon. Porcurated_02977, 01635, and 04371, encode an indole-3-glycerol phosphate synthase, a putative threonine efflux transporter and a hypothetical protein, respectively, that are HrpL-regulated. Similar genes are present in Pja and PsyB728a but are not HrpL-regulated (Figure 5). Putative hrp-boxes can be identified in all three Por genes, but not for the corresponding genes in Pja and PsyB728a. These genes are not syntenic (data not shown). They display high similarity in their coding sequence (data not shown); however their corresponding 5′ upstream regions are highly divergent. This could be the result of horizontal transfer, though there is no obvious footprint of mobile element DNA, or independent recombination events.
Lastly, loss of transcription termination regulation could lead to read-through transcription, and thus provide a mechanism for recruitment of non-T3E genes into the HrpL regulon. This mechanism was first highlighted by the recruitment into the HrpL regulon of the corR gene which was recombined downstream of the hrp-box associated hopAQ1 gene, in PtoDC3000 [25]. We observed that several genes found differentially expressed in our analysis were located downstream of HrpL-regulated T3E genes (Table S3) and could potentially be recruited into the HrpL regulon via loss of transcription termination regulation and subsequent transcriptional read-through.
We identified a cluster of HrpL-regulated genes, PSPPH_A0106-A0112, downstream from avrD that were recruited into a novel HrpL-regulated operon transcribed from the avrD promoter. These genes are flanked in Pla107 and Pph1448A, by transposable elements, suggesting that they could be acquired by horizontal gene transfer (Figure 6). Deletion mutants of either PSPPH_A0107 or avrD resulted in reduced virulence on French bean. The slightly reduced virulence we observed is in conflict with observations that allelic replacement of avrD by the nptII gene did not result in any growth defect in completive index assays [72]. This discrepancy could be explained by transcription from the nptII promoter in the previous work, or by the use of different growth assays, time points, and bean cultivars.
The PSPPH_A0106-A0112 operon is most likely involved in small molecule(s) synthesis promoting bacterial growth on host plants. Component(s) synthesized by the products of this operon and their effect on plants remain to be determined. However, since syringolides can be made from AvrD-expressing E. coli, and since the PSPPH_A0106-A0112 operon is not present in E. coli, we speculate that that the PSPPH_A0106-A0112 operon is not required for syringolide production. When present, AvrD shares no less than 84% amino acid identity across P. syringae strains. Genes encoding an AvrD-like protein with about 30% identity are widely distributed among bacteria, including Bacillus, Streptomyces and Vibrio. In general, these avrD-like genes are not found as singletons, but instead are linked to genes encoding various enzymes not related to any of the PSPPH_A0112-A0106 genes. In Streptomyces coelicolor A3(2), AvrD is part of an mmy operon responsible for synthesis of methylenomycin [88]. The PSPPH_A0110 to PSPPH_A0107 locus and to some extent the PSPPH_A0106 genes have similarity to genes in operons from Xanthomonas, Acidovorax, Pectobacterium and Ralstonia. Only the Ralstonia solanacearum PSI07 megaplasmid, carries both an avrD-like gene and a PSPPH_A0110-A0106 cluster of genes, but they are not contiguous on this plasmid. PSPPH_A0112 is mainly limited to P. syringae, but shares some homology with HMPREF9336_00100 (29% amino acid identity) found in Segniliparus rugosus ATCC BAA-974, an opportunistic pathogen associated with mammalian lung disease [89]. HMPREF9336_00100 and an avrD-like gene are linked in Segniliparus rugosus, being separated by only two genes and encoded on the same strand. We additionally observed that this operon has been disrupted by insertion of a transposable element in P. syringae CC1629, reminiscent of transposon disruptions of T3E genes commonly observed across the P. syringae phylogeny [3].
hrpL is widely distributed, and tightly linked in all hrp/hrc group I T3SS [1] and the non-canonical T3SS found in some P. syringae, as well as the T3SS of P. viridiflava, P. fluorescens, Erwinia, Pantoea stewartii, and Dickeya. It is the key virulence regulator in most if not all of these species. Our work highlights the advantages of integrating next generation transcriptional and genomic data to better understand the role of non-T3E HrpL regulon genes in plant-pathogen interactions. Our approach is readily applied to strains with sequenced genomes and broad phylogenetic sampling [63] to better understand P. syringae virulence mechanisms and their evolution.
For maintenance and transformation, P. syringae were grown in King's B media (KB) at 28°C. E. coli DH5α was grown in Luria-Bertani (LB) media at 37°C. Antibiotics were used at the following concentrations: 50 µg/ml rifampicin, 25 µg/ml kanamycin, 10 µg/ml tetracycline, and 25 µg/ml gentamycin, according to vector selection. Strains used or analyzed in this study are listed with their abbreviation in Table S8.
Native hrpL from the various P. syringae were PCR amplified using LA-Taq (TaKaRa) and oligonucleotides listed in Table S9 containing XbaI and Hind III sites, then cloned into pTOPO-TA (Invitrogen). The pTOPO-TA::hrpL was sequenced, digested with XbaI and HindIII, and cloned into NheI/HindIII-digested pCF340 (Newman and Fuqua, 1999) and designated pBAD::hrpL.
Porcurated_02784, 04644, 03530, 01245, 04371 and their respective upstream region containing the hrp-box were PCR amplified using Pfx (Invitrogen) and primers described in Table S9. Resulting PCR fragments were cloned into pENTR-D-TOPO (Invitrogen) and sequenced. Porcurated_04644 was amplified similarly using primers containing attB1/attB2 sites and cloned into the pDONR 207 vector (Invitrogen). All resulting constructs were sub-cloned into either the gateway-compatible pJC532 vector containing the in-frame Δ79avrRpt2 sequence for translocation assays or the pJC531 vector containing an in-frame HA sequence [3] to check for protein expression. All vectors used in this study were transformed into P. syringae strains using tri-parental mating with an E. coli helper strain containing pRK2013.
Pseudomonas strains containing pBAD::hrpLnative or pBAD::EV were grown overnight at 28°C, in KB media supplemented with tetracycline, then sub-cultured in fresh media at OD600 = 0.2, and grown until OD600 = 0.4–0.5. Bacteria were washed twice with 10 mM MgCl2 and resuspended in minimal medium [48] (MM is 50 mMKPO4 pH 5.7, 7.6 mM (NH4)2SO4, 1.7 mM MgCl2, 1.7 mM NaCl) containing 10 mM mannitol and supplemented with 1% glycerol and 0.1% peptone which suppresses hrpL induction. Bacteria were then inoculated in supplemented minimal media at OD600 = 0.2, and incubated shaking for 30 min at 28°C. Expression of hrpL was induced by addition of 200 mM L-arabinose. Aliquots of cell culture were taken 1, 3, 5 hours post-induction and treated with RNAprotect reagent (Qiagen). RNA isolation was performed by using the RNeasy minikit (Qiagen). Isolated total RNA was treated twice with TURBO DNase (Ambion). Total RNAs derived from each time point were pooled at a 1 to 3 ratio. 10 µg of pooled total RNA was depleted of 16S and 23S ribosomal RNA using RiboMinus (Invitrogen). cDNA were prepared from ∼1 µg of ribosomal depleted RNA, using random hexamer primers and Superscript II reverse transcriptase (Invitrogen). Second strand cDNA was prepared using DNA polymerase I and Ribonuclease H (Invitrogen). Double stranded cDNA was purified using Qiaquick spin columns (Qiagen) and eluted with EB buffer. Double stranded cDNA was sheared using a Covaris Disruptor. Library was prepared according to the manufacturer's protocol (Illumina). Sequencing of the library was performed according the manufacturer's protocol on either Illumina GAII including single-end, 36 cycles or Illumina HiSeq 2000 including single-end, 70 cycles.
We analyzed our RNA reads using the GENE-counter pipeline. For the PtoDC3000, PsyB728A, and Pph1448A datasets, we used the publically available genomes provided by NCBI, along with the transcriptome constructed by NCBI's gene prediction pipeline. For the Por, Pjap and Pla107 dataset, we used in-house assembly for the genome and used JGI's Integrated Microbial Genomes – Expert Review gene prediction pipeline for the transcriptome. All ribosomal RNA genes were excluded from the transcriptome file for all datasets. Transcriptome sequences for each strain were blasted against their corresponding genome and GFF files were constructed from the Blast reports using an in-house script. We processed the RNA reads and aligned the reads using the default parameters of GENE-counter's CASHX read mapping algorithm. Reads mapping to multiple genomic locations were excluded. Annotated genes were included in the analysis only if at least one read in one sample matched that gene which can lead to highly duplicated genes not being considered. The false discovery rate cutoff for determining differential expression was set to 0.05. We made a small modification to GENE-counter's findDGE.pl script that allowed for random seeding during the sample depth normalization process. By repeating the normalization process 300 times we generated B-values to measure and control for normalization effects. The GenBank accession (http://www.ncbi.nlm.nih.gov/) and Gold ID (http://img.jgi.doe.gov/cgi-bin/w/main.cgi) of the genomes used in this study are CP000058-CP000060, Gi04410, CP000075, Gi07003, Gi03478, and AE016853-AE016855. RNA-seq data have been deposited in NCBI Gene Expression Omnibus and will be accessible through GEO Series accession number GSE46930 (http://www.ncbi.nlm.nih.gov/geo/).
First, protein sequences of genes found up-regulated in our analysis with B-values≥50% were used to search each genome used in this study with BlastP to identify genes split up in different contigs/scaffolds. Possible duplication was ruled out by comparing the size of the query to the size of the subject sequence (of complete genomes, principally). Putative sequencing errors leading to stop codons and discontinuous ORFs, led to consecutive queries matching the same subject sequence. Only the entry with the most significant q-value was kept. Secondly, genes encoding open reading frames shorter than 60 amino acids were excluded from our data set. Thirdly, loci of genes not previously found HrpL-dependent were assessed for linkage to genes with a hrp-box. As previously described [41], [90], we observed potential transcriptional read-through artifacts for which directly HrpL-targeted genes led to apparent up-regulation of adjacent genes. Therefore, genes found differentially expressed adjacent to a HrpL-regulated gene, but on the opposite DNA strand were considered to be putative transcriptional read-through and removed from our analysis. Genes encoded on the same strand as the HrpL-regulated gene were kept. Fourth, genes with a hrp-box embedded within their ORF on either sense or anti-sense strands were not included. Adjacent genes encoded on the same strand as the manually predicted hrp-box were included in the defined HrpL regulon, but genes on the opposite strand of the hrp-box were excluded. All genes removed from the HrpL regulons after manual curation are listed Table S4.
For native gene expression, bacteria were grown for 4 hours in KB media from OD600 = 0.2, washed twice with sterile 10 mM MgCl2 and transferred into MM minimum media containing 10 mM mannitol for PtoDC3000, PsyB728a, Pla107, Pja, Por strains or MM minimum media containing 10 mM fructose for Pph1448A strain. Cells were collected after 5 hours of incubation shaking at 28°C and treated with RNAprotect reagent (Qiagen). Total RNA derived from cells grown in MM media or arabinose inducing media (as above) was extracted using the RNeasy minikit (Qiagen), DNase treated twice (Ambion Turbo DNase), and cleaned up with Qiagen RNeasy Mini kit. Reverse transcription was performed using SuperScript II (Invitrogen) with 2 µg total RNA. Diluted cDNA was analyzed with SYBR green (Applied Biosystem) using the Opticon 2 System (BioRad). Primers used are listed in Table S9.
Four week old Col-0 and Col-0 rps2–101c (rps2) plants were hand inoculated with PtoDC300028E [57] carrying Δ79avrRpt2 fusion clones at OD600 = 0.1. Plants were scored for Hypersensitive Response (HR) and pictures were taken 24 h after inoculation.
Knockout constructs were generated using MTN1907, a modified version of pLVC-D which allows for SacB counter-selection [3], [91]. To create Pph1448AΔPSPPH_A0107, Pph1448AΔPSPPH_A0113 mutants, 5′ and 3′ regions flanking the gene of interest were amplified using Pfx (Invitrogen) and combined by overlap extension PCR (Table S9), then cloned into pENTR-D-TOPO and sequenced. To generate the Pph1448AΔPSPPH_1855, PsyB728aΔhrpL and PorΔhrpL mutants, nucleotide sequences corresponding to the fused flanking regions of each gene were synthesized including Gateway recombination sites and cloned in the pUC17 vector (GenScript). All five clones were recombined into MTN1907 and transformed into either Pph1448A, PsyB728a or Por by tri-parental mating. After selection on tetracycline plates, merodiploids resulting from homologous recombination were verified by PCR. Two independent merodiploids carrying either a 3′ or 5′ insertion were grown on KB agar containing 5% sucrose to select for the loss of sacB via a second recombination event. Putative clean-deletion mutants were verified by PCR using flanking primers and gene specific primers.
Before inoculation, Pph1448A and mutants were grown overnight and sub-cultured from OD600 = 0.2 for 4 hours in KB media, then washed twice with 10 mM MgCl2. Two week old French bean cv. Tendergreen improved (Livingston Seed Co.) were dip inoculated with freshly grown bacteria at OD600 = 0.001 bacteria in 10 mM MgCl2 and 0.04% Silwet L-77. Four plants were dip inoculated for each strain. Three days and an half after inoculation leaf discs were cored (12 to 16 replicates, each 4 cores), ground in 10 mM MgCl2, serially diluted and plated on KB/50 µg/ml rifampicin and bacteria counted. Each set of mutants were tested side by side with the wild type strain at least 3 times.
|
10.1371/journal.pcbi.1007000 | Charting pathways to climate change mitigation in a coupled socio-climate model | Geophysical models of climate change are becoming increasingly sophisticated, yet less effort is devoted to modelling the human systems causing climate change and how the two systems are coupled. Here, we develop a simple socio-climate model by coupling an Earth system model to a social dynamics model. We treat social processes endogenously—emerging from rules governing how individuals learn socially and how social norms develop—as well as being influenced by climate change and mitigation costs. Our goal is to gain qualitative insights into scenarios of potential socio-climate dynamics and to illustrate how such models can generate new research questions. We find that the social learning rate is strongly influential, to the point that variation of its value within empirically plausible ranges changes the peak global temperature anomaly by more than 1°C. Conversely, social norms reinforce majority behaviour and therefore may not provide help when we most need it because they suppress the early spread of mitigative behaviour. Finally, exploring the model’s parameter space for mitigation cost and social learning suggests optimal intervention pathways for climate change mitigation. We find that prioritising an increase in social learning as a first step, followed by a reduction in mitigation costs provides the most efficient route to a reduced peak temperature anomaly. We conclude that socio-climate models should be included in the ensemble of models used to project climate change.
| The importance of anthropogenic CO2 emissions on climate change trajectories is widely acknowledged. However, geophysical climate models rarely account for dynamic human behaviour, which determines the emissions trajectory, and is itself affected by the climate system. Here, using a coupled socio-climate model, we show how social processes can strongly alter climate trajectories and we suggest optimal intervention pathways based on the model projections. Steps to increase social learning surrounding climate change should initially be prioritised for maximum impact, making a subsequent reduction in mitigation costs more effective. Policymakers will benefit from a better understanding of how social and climate processes interact, which can be provided by socio-climate models.
| According to many ancient myths, humans did not invent fire-making de novo but rather learned it from personalities like Prometheus and subsequently spread the practice amongst themselves. These stories reveal how ancient myth-makers already grasped the fundamental importance of social learning—the process whereby individuals learn new behaviours, values and opinions from others [1]. Social learning is no less relevant in the era of human-environment challenges [2–4]. The importance of social learning and social processes more generally in climate change mitigation and adaptation is well recognised [5–8]. Increasingly sophisticated geophysical climate models are helping us understand the impacts of anthropogenic greenhouse gas (GHG) emissions [9–11], and the importance of these models is hard to understate. However, climate projections depend strongly on the assumed trajectory of GHG emissions [12]. This trajectory is determined by human behaviour and yet climate models generally do not incorporate dynamic social processes relevant to GHG emissions. Rather, GHG emissions are assumed to follow some specified trajectory. These trajectories are constructed with socio-economic factors in mind, (see Representative Concentration Pathways [12] and Shared Socioeconomic Pathways [13] for instance), but are not coupled to climate dynamics and do not capture human responses to climate change in a mechanistic way.
Just as human behaviour influences climate trends, climate change in turn influences human behaviour concerning GHG emissions, including both climate change mitigation and adaptation [5, 6, 8, 14, 15]. Individuals in places with rising average temperatures are more likely to perceive climate change [15], and social effects are apparent when individuals take steps in response to such shifting perceptions [6, 8]. There is also an important distinction between social learning and social norms—socially accepted and widely practised modes of conduct [16]. Social norms are known to have a strong influence on human behaviour [17] including aspects relating to climate change [7, 16, 18] and therefore play an important role in determining emission trajectories [6]. Multiple studies show a tendency for individuals to conform to emerging norms in support of climate change mitigation [7, 16]. Moreover, it appears that individuals are often not consciously aware of the importance of social norms in their decision-making and instead falsely ascribe their decisions to other factors [18]. However, it is important to note that social norms do not automatically promote socially beneficial outcomes. They can equally well force conformity to a destructive norm such as political extremism [19]. This also happens in the context of climate change behaviour, where it has been found that individuals may also conform to a norm of non-mitigation, by adjusting their habits to match those of less environmentally friendly neighbours [18, 20].
Hence, Earth’s climate and human subsystems are part of a single coupled system where social dynamics play a vital role. Yet, models of Earth’s coupled climate-behaviour system remain essentially undeveloped. One such approach [21] couples a sophisticated climate model [11] to a model for individual behavioural change based on the theory of planned behaviour—a dominant paradigm in psychology [22]. The authors find that the sensitivity of global temperature change to human factors such as response to extreme events, social norms and perceived ability to adopt mitigative strategies is of a similar magnitude to its sensitivity to geophysical factors. They deduce that quantifying behavioural uncertainty and physical uncertainty in climate projections deserve equal attention. The model focuses on how individual psychology and behaviour are influenced by extreme weather events. Social effects are modelled phenomenologically (i.e., exogenously imposed): individuals do not learn behaviour or opinions from one another, and social norms are treated as a fixed effect that does not depend on the population’s current composition of attitudes.
Here, we treat social learning and social norms endogenously, by modelling their dynamics as they emerge from rules governing how individuals interact, learn and behave. Our first objective is to develop qualitative insights into how different aspects of the system—endogenous social processes, temperature trends, and mitigation costs—separately and together determine possible dynamics of the larger socio-climate system. Our second objective is to illustrate potential uses of coupled socio-climate models to chart social and economic policy pathways that mitigate climate change as quickly as possible. To meet these objectives, we sought to develop a model that (1) could capture a range of IPCC climate change scenarios, ranging from 4 degrees of warming by 2100 (RCP 8.5 scenario) to sub 2 degrees of warming (RCP 2.6 scenario), (2) was simple enough to analyse so that we could learn which mechanisms drive the predicted socio-climate dynamics, (3) was based on existing approaches for modelling social dynamics and climate dynamics, and (4) captured the salient features of social and climate systems. Given the model’s simplicity, it is primed for insights as to how social and climate processes interact, though limited in its predictive capacity due to the complexity of the socio-climate system. The development of more complex socio-climate models will be an important research avenue, once the mechanisms of socio-climate dynamics are better understood.
Geophysical models in the climate science literature span a wide range of different complexities depending on the associated research objective. Highly complex models are the state-of-the-art for weather and climate prediction [11, 23, 24], whereas simple models allow us to assess processes and feedbacks, thereby improving our intuition of climate system dynamics [25–29]. Likewise, the behavioural sciences have benefited from a variety of modelling approaches, that address the diverse set of social processes that take place on the individual and societal level [30]. Here, we use minimal models for both social and climate dynamics. Starting simple allows us to build intuition on the effect of socio-climate feedbacks that have yet been considered in the climate change literature. The social model is widespread and, despite its simplicity, captures the salient aspects of social dynamics [2, 30, 31]. Moreover, the simple Earth system model that we use [25] accurately follows the projections of the state-of-the-art CMIP5 models when forced with the IPCC emission scenarios (S1 Fig).
Over the period from 1800 to 2014, the socio-climate model is simulated with a fixed social component, forced with historical anthropogenic carbon emissions. Initial conditions for all climate variables are zero since they represent deviations from pre-industrial values. Social dynamics are initiated in 2014 with an initial proportion of mitigators x0 = 0.05. The ensuing dynamics of ϵ(t) follow an increasing but saturating trend corresponding to the world’s increasing but saturating population size and energy demands. Specifically
ϵ ( t ) = { linear interpolation of historical emissions t ≤ 2014 ϵ 2014 + ( t − 2014 ) ϵ max t − 2014 + s t ≥ 2014 (14)
where ϵmax is the saturating value, and s the half-saturation constant, of ϵ(t). This expression is shown graphically in S7 Fig. The system of (delay) ordinary differential equations is simulated using the NDSolve package in Wolfram Mathematica. Historical CO2 emissions were obtained from the CDIAC data repository [35].
Baseline climate parameters are obtained from the original Earth system model [25] where they were fitted to obtain historical trends of temperature and carbon dynamics. Social parameters are more speculative and so are given wide upper and lower bounds. The relative cost of warming (f(T)) with respect to the net cost of mitigation (β) is chosen in accordance with the argument that the costs of preventative action will be far less than the cost implied otherwise by global warming [36]. For sensitivity analyses we draw parameters from triangular distributions that peak at baseline values and extend to upper and lower bounds (S2 Table). Parameters are kept fixed preceding 2014 to retain historical trends in the simulations.
The model demonstrates how the social learning rate can strongly determine temperature trends. We first consider a null hypothesis where adaptive behaviour is removed from the model by forcing the proportion of mitigators in the population to remain constant. In this case of fixed behaviour, emissions saturate and the temperature anomaly increases indefinitely (S2 Fig). However, once social learning is added and the proportion of mitigators is allowed to evolve dynamically as in our baseline model, the predicted average global temperature anomaly can peak anywhere from 2.2°C, near the Intergovernmental Panel on Climate Change (IPCC) limit [37] (in the case of very rapid social learning) to 3.5°C (in the more realistic case where social learning unfolds on a generational timescale) (Fig 1a–1c). Whether people discuss climate change more or less often can therefore strongly influence temperature trends. Because we model social norms as something that tends to reinforce majority behaviour and attitudes—whatever they might be—one might think that social norms act as a double-edged sword. In fact, they operate more like an unhelpful scimitar, as illustrated by comparing cases of low and high strength of social norms. Because the population starts off from a state of largely non-mitigating behaviour, increasing the strength of social norms suppresses the spread of mitigating behaviour for decades by entrenching non-mitigation as a norm, even when rising temperatures strongly justify an immediate shift (Fig 1d–1f). (This model dynamic echoes not only current climate norms reinforcing non-mitigation [20] but also past social shifts occurring on decadal timescales, such as evolving social norms about when and where smoking is acceptable.) However, when mitigating behaviour eventually does become widespread, a higher strength of social norms does not significantly accelerate its spread. Rather, the two curves for cases of high and low social norm strength simply move in parallel to one another because by this time, the utility function that determines behaviour change is dominated by the large temperature anomaly (Fig 1d). In this parameter regime, social norms generate a perverse asymmetry, in contrast to findings from other socio-climate models that assume social norms can only support climate change mitigation [20].
The model also shows how a reduction in net mitigation cost can significantly accelerate the onset of social change. For instance, a 67% reduction in the mitigation cost increases the percentage of mitigators by 2060 from 10% to 90% (Fig 1g–1i). Therefore, policies that reduce the cost of mitigation (through e.g. subsidies, tax cuts) will benefit from the accelerating effects of social learning and must be timed correctly.
Our baseline model assumes that individuals’ perceived cost of climate change impacts depends on a linear extrapolation of the recent temperature anomaly over the previous ten years (Methods). If individuals instead base their decisions only on the current temperature anomaly, the simulated global temperature anomaly lies well above the 2°C target set by the IPCC, and exhibits wide variation in sensitivity analysis (Fig 2). This contrasts with our baseline model where the population movement towards mitigative strategies ignites earlier, significantly reducing the global temperature anomaly. This predicted dynamic stems from the multi-decadal lag between GHG emissions and the consequent global temperature rise [38].
Our model predicts medium-term GHG emission trajectories (Fig 1b, 1e and 1h) that are qualitatively similar to those often assumed under various future emissions scenarios. This raises the question of how such models can be useful. The socio-climate model enables us to explore how socio-climate dynamics might respond to changes that are under the control of policymakers. For instance, it is possible to compare social and economic policy interventions by considering the effects of simultaneous parameter changes, instead of one at a time. This enables us to chart out the quickest pathways from highest to lowest temperature anomalies. The relative merits of increasing the social learning rate vs. reducing the net cost of mitigation are illustrated with a contour plot where the contours represent peak temperature anomaly as a function of the two parameters (Fig 3). Increasing the social learning rate (e.g. through media coverage and public fora devoted to climate change) is particularly effective when social learning is slow, but has saturating benefits, as indicated by the increasing vertical spacing of contour lines for at higher learning rates. In contrast, reducing the net mitigation cost (e.g. through tax breaks) drives a more linear response in peak temperature anomaly. Crucially, it should be noted that both a reduction of net mitigation cost and an increase in the social learning rate are required to achieve the IPCC target. The arrows in Fig 3 show the ‘path of steepest descent’—the most efficient combination of the two measures. Starting from a situation of high projected temperature anomalies, the model predicts that increasing the social learning rate should first be prioritised, followed by a reduction in net mitigation cost once the benefits of social learning begin to saturate. This approach gets us to the region of parameter space corresponding to the IPCC target faster than alternative trajectories.
A sensitivity analysis reveals the relative influence of each parameter on the peak temperature anomaly (Fig 4). The time horizon of individuals’ temperature projection, social learning rate and costs of mitigation are major factors, all of which may be influenced by appropriate intervention. The importance of social parameter uncertainties in determining climate predictions indicated by our model has also been predicted by other socio-climate models [21]. Interestingly, the system is relatively insensitive to the initial proportion of mitigators, suggesting that the mediation of social processes, as opposed to the current social state, is key to guiding the socio-climate system to a trajectory of reduced emissions. Sensitivity analyses such as these can help investigators determine priorities for data collection: the parameters exhibiting the greatest influence on predictions should be targeted for data collection so we can best reduce model uncertainty.
A striking feature revealed by the sensitivity analysis is the asymmetry in many of the parameter dependencies. Consider the three parameters with highest impact on the peak temperature anomaly (concerning forecast horizon, learning rate and global warming costs). A decrease in these parameters is more detrimental than an increase is beneficial. For example, a forecast horizon 10 years above baseline value results in a 0.6 degree decrease in peak temperature anomaly, whereas a forecast horizon 10 years below baseline value results in a 1 degree increase. This imbalance is a manifestation of the nonlinear interactions between and within each of the social and climate system.
The sensitivity analysis also reveals non-monotonic relationships between the peak temperature anomaly and the parameters. For example, both an increase and a decrease in solar flux results in a higher peak temperature anomaly. Interestingly, this is not the case if the climate subsystem is considered in isolation. For a fixed emissions scenario, a higher (lower) solar flux will always result in a higher (lower) peak temperature anomaly, since the solar flux is proportional to the net downward radiation absorbed by the planet’s surface. The coupling to social dynamics fundamentally alters this relationship. In the socio-climate system, a reduced solar flux results in a slower increase in surface temperature. As a consequence, individuals are less incentivised to mitigate, causing the social system to maintain a regime of non-mitigative behaviour. The accompanying high rate of CO2 emissions quickly overcompensates for the reduced solar flux, yielding a higher peak temperature anomaly. Thus seemingly useful interventions to the physical system can actually end up doing more harm than good when there is strong coupling to a social system, as is the case for global warming.
This study has shown how social processes can influence climate dynamics, according to one possible way of modelling social dynamics and norms. However, other frameworks for modelling human behaviour could yield different predictions. For instance, the socio-climate model of Ref. [21] does not include social learning. Individuals respond directly to changes in the climate, and not through interactions with one another. As a consequence, the rate at which individuals adopt mitigative strategies only varies with the current climate situation, and not with current population consensus. Mitigation efforts can therefore be expected to closely follow the severity of climate change in the model. In our model, social learning manifests as a feedback within the social system, resulting in qualitatively different socio-climate trajectories. Mitigative behaviour is initially suppressed–even as temperatures rise to levels that should incentivise mitigation–due to low numbers of mitigating individuals and therefore little turnover of behaviour in the population. However, social learning creates a positive feedback loop once there is a net positive utility to mitigate, and so as the numbers of mitigators increases, so too does the rate at which non-mitigators switch to being mitigators. This results in a sharp non-linear increase in mitigators, as a combined outcome of both the social and the climate system dynamics. We note that, all else being equal, adding social learning to a model has the effect of slowing down behaviour change in the human population (since a process takes time, by definition), and therefore the mitigation response of human populations.
Conversely, the case of very rapid social learning recovers a ‘best response’ model similar to those assumed in classical economics, where individuals immediately adopt the highest payoff strategy without learning the behaviour from others. Whether or not this assumption can approximate behaviour in real human populations hinges upon how fast social learning occurs—individuals would need to sample others rapidly enough to enable complete population behaviour change within 5 years for this approximation to work in our model, which seems implausible (Fig 1a–1c).
In a different vein, we assumed a homogeneous population with respect to mixing and individual utilities. The model’s social dynamics capture interactions at the individual level, though there are many different scales of social organisation that the model does not consider, from families/neighbourhoods to cities/states and up to interacting countries. Future models could include this more hierarchical social structure. Similarly, these models could include different types of individual with correspondingly different utilities. For instance, the model could include industrial corporations with utilities biased toward shareholder profit, and social institutions (such as laws, taxes, the education system) that reflect the current governmental stance. Social learning may also take on different forms due to diverse individual psychologies and values [39–41]. Such heterogeneities are known to affect the dynamics of a wide variety of systems [42] and can prevent population consensus by permitting development of echo chambers [43]. Our model also makes the simplifying assumption that individuals base their temperature projection on linear extrapolation of past temperatures. This could be generalised to a non-linear extrapolation to reflect an individual’s perception of ‘accelerating’ change. Extending socio-climate models to include these finer details should prove valuable in further investigations.
Climate change is a manifestation of coupled human-environment dynamics and therefore we should start coupling climate models to social models [5, 44]. Our simple coupled socio-climate model shows that the rate at which individuals learn socially strongly influences the peak global temperature anomaly, to the point that variation of this parameter within plausible ranges changes the peak temperature anomaly by more than 1°C. Therefore, it matters whether social processes cause slow or fast uptake of climate change mitigation measures. We found that social norms may not provide help when we most need it, although this finding could be nuanced by adding social heterogeneity. Finally, we illustrated how exploring the parameter space of socio-climate models suggests optimal paths for mitigating climate change. A more sophisticated policy impact assessment model based on a coupled socio-climate approach could therefore be useful to decision-makers facing a mandate to reduce GHG emissions with a fixed budget. In summary, it is essential for climate change research to account for dynamic social processes in order to generate accurate predictions of future climate trends, and the paradigm of coupled socio-climate modelling could help us address this challenge.
|
10.1371/journal.pntd.0001988 | Functional Mapping of Protein Kinase A Reveals Its Importance in Adult Schistosoma mansoni Motor Activity | Cyclic AMP (cAMP)-dependent protein kinase/protein kinase A (PKA) is the major transducer of cAMP signalling in eukaryotic cells. Here, using laser scanning confocal microscopy and ‘smart’ anti-phospho PKA antibodies that exclusively detect activated PKA, we provide a detailed in situ analysis of PKA signalling in intact adult Schistosoma mansoni, a causative agent of debilitating human intestinal schistosomiasis. In both adult male and female worms, activated PKA was consistently found associated with the tegument, oral and ventral suckers, oesophagus and somatic musculature. In addition, the seminal vesicle and gynaecophoric canal muscles of the male displayed activated PKA whereas in female worms activated PKA localized to the ootype wall, the ovary, and the uterus particularly around eggs during expulsion. Exposure of live worms to the PKA activator forskolin (50 µM) resulted in striking PKA activation in the central and peripheral nervous system including at nerve endings at/near the tegument surface. Such neuronal PKA activation was also observed without forskolin treatment, but only in a single batch of worms. In addition, PKA activation within the central and peripheral nervous systems visibly increased within 15 min of worm-pair separation when compared to that observed in closely coupled worm pairs. Finally, exposure of adult worms to forskolin induced hyperkinesias in a time and dose dependent manner with 100 µM forskolin significantly increasing the frequency of gross worm movements to 5.3 times that of control worms (P≤0.001). Collectively these data are consistent with PKA playing a central part in motor activity and neuronal communication, and possibly interplay between these two systems in S. mansoni. This study, the first to localize a protein kinase when exclusively in an activated state in adult S. mansoni, provides valuable insight into the intricacies of functional protein kinase signalling in the context of whole schistosome physiology.
| Schistosome blood flukes are formidable parasites. They can survive for many years as male-female pairs in the blood vessels of their vertebrate hosts where they copulate and produce large numbers of eggs that become lodged in tissues causing schistosomiasis. Over 200 million people are infected with schistosomes, mainly in tropical and sub-tropical countries; in terms of parasitic diseases, the socio-economic impact of human schistosomiasis is second only to malaria. Understanding how cellular mechanisms regulate schistosome form and function is a vital part of global research efforts on schistosomes, not least because identification of novel mechanisms might yield opportunities to develop new drugs against the parasite. Here we use a novel approach to provide a comprehensive atlas that displays the localization of an important protein (protein kinase A) when in an exclusively activated state within schistosomes. We show that this protein is activated within various tissues including those of the musculature and nervous system and that its activation in nerves visibly increases when paired adult male and female schistosomes separate. We also show that PKA plays a vital role in the co-coordination of schistosome muscular activity. Our findings offer valuable insight into this protein at the functional level and provide a much-needed physiological framework for further work on PKA in schistosomes, which has been highlighted previously as a potential drug target.
| Schistosoma mansoni is an important parasitic blood fluke that causes human schistosomiasis, a neglected tropical disease that ranks second only to malaria when considering the number of people infected (∼200 million) and at risk (∼779 million) [1]. The life cycle of this parasite is complex involving snail-intermediate and human-definitive hosts. After the free-living cercariae infect the human host, they transform into parasitic schistosomules which mature in the vasculature via an adolescent stage to separate sex adults; sex organs develop approximately three weeks after infection and copulation between male and female worms begins after approximately four weeks [2]. The intimate association that exists between adult male and female worms in copula is vital to maintaining the full maturation of the female worm [3]–[5], fertilization of eggs, and thus high levels of egg production to facilitate parasite transmission. Not all of the eggs produced by adult female schistosomes escape from the host. The immune response to those eggs that become trapped in tissues such as the gut wall, liver or spleen and the granulomatous reaction evoked by secretory egg antigens gives rise to chronic/advanced schistosomiasis, with an associated disease burden of ∼70 million disability adjusted life years [6], [7]. Praziquantel is the current drug of choice for the treatment of schistosomiasis but after three decades of use in mono-therapy there remains a possibility that resistance to praziquantel will emerge. Recently the genomes of the three most medically-important schistosomes, S. mansoni [8], Schistosoma japonicum [9], and Schistosoma haematobium [10] were published, providing a valuable resource for integrative biological studies on schistosomes [2] and for identifying potential drug targets [11].
Cyclic AMP (cAMP)-dependent protein kinase/protein kinase A (PKA) is one of the best-characterized members of the protein kinase super-family [12], [13]. In eukaryotes, PKA regulates diverse cellular processes including cell cycle progression [14], proliferation/differentiation [15], [16], cytoskeletal dynamics [17], and flagellar beat [18]. In an inactive state, the PKA holoenzyme comprises two identical catalytic (C) subunits bound non-covalently to two identical regulatory (R) subunits. Activation of PKA occurs in the presence of cAMP that is produced by G-protein coupled receptor (GPCR)-mediated activation of adenylyl cyclase. cAMP binds cooperatively to two sites on each R subunit driving a conformational change within the holoenzyme that results in the release of the catalytically active C subunits enabling them to phosphorylate serine/threonine residues in specific cytosolic and nuclear substrate proteins altering their biological functions [19], [20]. Phosphorylation also plays an important part in the activation of PKA. In mammalian cells, the C subunit is phosphorylated at Thr197 in the activation loop by another C subunit or by phosphoinositide-dependent protein kinase 1 (PDK1) [21]–[23]; in addition Ser338 is phosphorylated, which although not required for enzyme activation is important for processing and maturation of PKA [23]. The broad but selective substrate specificity of PKA is achieved by compartmentalization at different sub-cellular regions through interaction with A-kinase-anchoring proteins (AKAPs) [13]. Furthermore, endogenous protein kinase inhibitor (PKI) peptides inhibit the activity of the C subunit independently of cAMP and also serve to traffic free C subunits from the nucleus to the cytoplasm [24].
In 2009, the first definitive evidence of PKA activity in adult worms was published with a full description of a gene encoding a S. mansoni PKA catalytic subunit (Sm-PKA-C) [25]. The putative Sm-PKA-C shared 70% similarity with PKA-C subunits from other organisms including the nematode Caenorhabditis elegans, the fruit fly Drosophila melanogaster, and Homo sapiens, and was most similar to PKA-C of the mollusc Aplysia californica. Furthermore, using both RNA interference (RNAi) and pharmacological approaches, PKA expression and activity were found to be essential for schistosome survival [25], highlighting PKA as a possible anti-schistosome chemotherapeutic target.
In the current paper we provide valuable insights into the precise locations and possible functions of phosphorylated (activated) PKA within intact adult S. mansoni. Our findings highlight particularly a neuromuscular role for PKA in schistosomes and the detailed analysis of PKA activation within worms provides an important physiological framework for future work on schistosome neurobiology and host-parasite interactions.
Laboratory animal use was within a designated facility regulated under the terms of the UK Animals (Scientific Procedures) Act, 1986, complying with all requirements therein; regular independent Home Office inspections occurred. The experiments involving mice in this study were approved by the Natural History Museum Ethical Review Board and work was carried out under Home Office project licence 70/6834.
The Belo Horizonte strain of S. mansoni was used in all experiments. Adult schistosomes were recovered by hepatic portal perfusion of female mice (BKW strain) that were infected approximately 45 days earlier by paddling in water containing 200 cercariae. Worm pairs were collected carefully and were either placed immediately in Dulbecco's modified Eagle's medium (DMEM; Invitrogen, Paisley, UK), or were fixed immediately in ice-cold absolute acetone and stored at 4°C for immunohistochemistry.
Freshly collected adult worm pairs were placed individually in wells of a 12-well tissue culture plate (Nunc, Thermo Fisher Scientific, Loughborough, UK) each containing 1 ml DMEM and were incubated in forskolin (50 µM or 100 µM; Calbiochem, Merck, Nottingham, UK), KT5720 (25 µM or 50 µM; Calbiochem), dimethyl sulphoxide (DMSO) vehicle (0.02% (v/v)), or DMEM alone for 1 h at 38°C. Forskolin was used to activate adenylyl cyclase and produce cAMP to in turn activate PKA; KT5720, a competitive antagonist of the ATP binding site on the PKA catalytic subunit, was employed as a PKA inhibitor. After treatment, each worm pair was homogenized on ice in 25 µl 1× RIPA buffer (20 mM Tris-HCl, 150 mM NaCl, 1 mM EDTA, 1 mM EGTA, 1% (v/v) NP-40) containing 1 µl protease and phosphatase inhibitor cocktail (Pierce; Thermo Fisher Scientific). The resulting homogenate was centrifuged at 13,000 rpm for 10 s at 4°C to remove insoluble material and protein estimations were carried out on the supernatant using the Bradford assay. An appropriate volume of 5× SDS-PAGE sample buffer was added and samples heated to 90°C for 5 min. Once cooled on ice, a further 1 µl of protease inhibitor and phosphatase inhibitor cocktail were added to the extracts and samples stored at −20°C for subsequent electrophoresis. SDS-PAGE was performed using 10% Precise pre-cast gels (Pierce, Thermo Fisher Scientific) and proteins were transferred to nitrocellulose membranes (GE Healthcare, Amersham, UK) using a semi-dry electrotransfer unit (Bio-Rad, Hemel Hempstead, UK). After transfer, membranes were stained with Ponceau S (Sigma, Poole, UK) to confirm homogeneous transfer, and were blocked for 1 h in 5% (w/v) non-fat dried milk in tris-buffered saline containing 0.1% (v/v) Tween-20 (TTBS), and briefly washed in TTBS prior to incubation overnight at 4°C in rabbit anti-phospho-PKA-C (Thr197) polyclonal primary antibodies (Cell Signalling Technology, New England Biolabs, Hitchen, UK; 1∶1000 dilution in 1% (w/v) BSA in TTBS). Next, blots were washed with TTBS and incubated for 2 h at room temperature with horse-radish peroxidase-conjugated secondary antibodies (Cell Signalling Technology; 1∶5000 in 1% BSA (w/v) in TTBS) and exposed to West Pico chemiluminescent substrate (Pierce) for 5 min. Immunoreactive bands were then visualized using a cooled CCD GeneGnome chemiluminescence imaging system (Syngene, Cambridge, UK). Equal loading of proteins was checked by stripping blots for 3 h at room temperature with Restore western blot stripping buffer (Pierce) before briefly washing blots in TTBS and incubating blots with anti-actin antibodies (Sigma, Poole, UK; 1∶3000 in TTBS) followed by secondary antibodies and chemiluminescent imaging. Relative band intensities were quantified using Gene Tools software (Syngene). In addition, to confirm that the anti-phospho-PKA-C (Thr197) primary antibodies only detected the phosphorylated form of PKA-C, western blots were either incubated in primary antibody that had been pre-adsorbed for 30 min to the phosphorylated immunizing peptide or were pre-treated with lambda phosphatase (New England Biolabs; 400 U/ml in TTBS containing 1% BSA and 2 mM MnCl2) for 4 h prior to incubation in primary antibodies; secondary antibody labeling and detection were then performed as described above.
Worms processed for confocal laser scanning microscopy included samples fixed immediately after removal from the host and samples fixed after exposure to forskolin (50 µM) as detailed above. Acetone fixed worms were washed twice with 1 ml phosphate buffer saline (PBS) and were further permeabilized with 0.3% (v/v) Triton-X100 in PBS for 1 h. After a brief wash in PBS, worms were blocked for 2 h with 10% (v/v) goat serum (Invitrogen, Paisley, UK) followed by incubation in anti-phospho-PKA-C (Thr197) antibodies (1∶50 in PBS containing 5% (w/v) BSA) for 72 h. Worms were then washed three times in 1 ml PBS for 20 min each and incubated in Alexa Fluor 488 secondary antibodies (Invitrogen; 1∶500 in BSA) and 200 ng/ml rhodamine phalloidin (Sigma) for 24 h in the dark. After further washing with PBS for 1 h, worms were placed on microscope slides and mounted in Vectashield anti-bleaching medium (Vector Laboratories, Peterborough, UK). All washes and incubations were performed in screw-capped microfuge tubes on a microfuge tube rotator at room temperature. Worms were then visualized using a Leica TCS SP2 AOBS confocal laser-scanning microscope using 20× dry objectives or 40x/63x oil immersion objectives and images collected and analyzed with associated Leica software. Because adult S. mansoni autofluoresced at the same detection wavelength as the secondary antibody, the signal received from the negative controls (i.e. those not incubated in primary antibody) was negated from the positive samples by reducing the power level of the photomultiplier tube (PMT) and then maintaining constant PMT voltage throughout all observations. Worms were also incubated with anti-phospho-PKA-C (Thr197) antibodies that had been pre-absorbed to the phosphorylated immunizing peptide to check for antibody specificity in immunocytochemistry.
Freshly collected worm pairs were placed in DMEM at 28°C for 30 min to equilibrate. They were then observed until the first worm pairs uncoupled naturally, and the individual separated male and female worms (five of each) were immediately collected and fixed in ice-cold acetone. At the same times, five coupled worm pairs were removed from the medium and fixed. This provided for analysis worms that were paired and those that had just separated. In addition, immediately upon separation, individual worms were transferred separately to wells of a 24-well culture plate (Nunc) each containing 0.5 ml DMEM maintained at 28°C. These individual male and female worms were then fixed (as described above) at 15 min, 30 min and 60 min post separation (five of each for each time point) to allow for analysis of PKA signalling during pair separation. Similarly, paired worms (five pairs for each time point) were collected, incubated and fixed after these durations to provide paired-worm controls for each separation time point. Fixed worms were then kept at 4°C until they were processed for immunohistochemistry.
Freshly collected adult worm pairs were placed in individual wells of a 12-well tissue culture plate (Nunc) each containing 1 ml of DMEM at 28°C. After 30 min, worms were treated with 50 µM or 100 µM forskolin, or DMSO (0.02% (v/v)) vehicle. Exposing one sample at a time, adult worms were videoed over 30 min using an Olympus SZ4045 binocular dissecting microscope attached to a JVC TK-1481 composite colour video camera operating with Studio Launcher Plus for Windows software with 1 min long movies captured at 0 min, 5 min, 10 min, 15 min, 20 min, 25 min and 30 min post-treatment; videos were compressed using Panasonic dv codecs. A minimum of five worm pairs per treatment were analyzed. During analysis, cold light sources were employed and light intensity kept constant in order to stabilize light condition. The number of gross random muscular movements/min was then assessed visually for each sample at each time point. A gross random muscular movement was defined as a rapid observable change from the existing body position; an extreme example of such movement can be seen with the whip-like motion observed following forskolin treatment in the supplementary movie (Video S1). Next, movies taken for treatments displaying the maximum phenotypic effects were imported into the publicly-available software ImageJ for Windows (Rasband, W.S., ImageJ, U. S. National Institutes of Health, Bethesda, Maryland, USA, http://rsb.info.nih.gov/ij/, 1997–2009) to further assess the nature of worm movement. Prior to import, videos were decompressed to avi format using Movavi Suite 10 SE for Windows and were converted to 8-bit grayscale; background subtraction was performed using ImageJ and the resulting movie was converted to binary format using the automatic Otsu thresholding algorithm. Binary objects representing the worms were then tracked for ‘thrashing’ (body bend) analysis using the open-source publicly-available custom ImageJ plugin, wrMTrck (http://www.phage.dk/plugins) with a bend threshold of 6°. Graphs depicting the angular movement (degrees) of individual worms (shapes) were generated against time (frame number) along with the total number of body bends above the threshold per treatment.
Analysis of variance (ANOVA) and a post-hoc Fisher multiple comparison test were done to analyze the effects of individual treatments at specific time points using Minitab 15.
Phospho-specific antibodies that bind to PKA-C only when phosphorylated at a site conserved with threonine 197 (Thr197) of human PKA-C within the PKA activation loop were employed in an attempt to detect phosphorylated PKA in adult S. mansoni. Because phosphorylation at this residue is crucial for PKA maturation, optimal conformation and catalytic activity [26], [27], these antibodies are used to determine PKA activation [27]–[29]. Bioinformatic analysis of SmPKA-C [25] comprising 350 amino acids revealed that the amino acid sequence (RVKGRTWTLCGTPEY) surrounding and including Thr197 to which the anti-phospho PKA antibodies bind (www.phosphosite.org) is identical between S. mansoni and human PKA, with Thr195 being the crucial threonine phosphorylation site in S. mansoni PKA-C. Western blotting revealed that the anti-phospho PKA (Thr197) antibodies detected two closely migrating bands with apparent molecular weights of approximately 40 KDa and 42 KDa in adult S. mansoni homogenates (Figure 1A). Treatment of western blots with lambda phosphatase for 4 h prior to exposure to anti-phospho PKA (Thr197) antibodies resulted in a total loss of immunoreactivity of both protein bands; in addition, incubation of these antibodies with the immunizing peptide before exposure to the nitrocellulose resulted in the same effect (Figure 1A). When live adult worms were exposed to the PKA activator forskolin (50 µM or 100 µM) for 1 h the immunoreactivity of both bands increased when compared to controls (Figure 1B). In contrast, exposure to KT5720, a competitive antagonist of the ATP binding site of PKA, decreased the phosphorylation of both bands with 50 µM KT5720 attenuating phosphorylation by approximately 90% (determined by image analysis of two independent bots) (Figure 1B). These results are consistent with the expected effects of antigen competition and dephosphorylation on antibody immunoreactivity and of activation and inhibition on PKA phosphorylation status (e.g. [28], [29]), identifying the anti-phospho-PKA (Thr197) antibodies as suitable for studying PKA activation in S. mansoni. The two bands of phosphorylated PKA-C therefore likely represent the 40.4 KDa SmPKA-C characterized by Swierczewski and Davies [25] and an additional PKA-C or splice variant thereof. The S. mansoni kinome [30] includes up to five predicted PKA-like proteins, and multiple sequence alignment of these proteins using ClustalW2 (http://www.ebi.ac.uk/Tools/msa/clustalw2/) reveals that the residue corresponding to Thr197 of human PKA-C is conserved in all of these proteins (data not shown).
To visualize activated PKA in adult S. mansoni, anti-phospho-PKA (Thr197) antibodies and confocal laser scanning microscopy were used; all images were obtained from whole mounts of intact worms. Schistosomes that were only incubated in secondary antibody (negative control) displayed almost no fluorescence when autofluorescence was negated by reducing the PMT voltage; only the F-actin staining was evident (Figures 2 A–C). In addition, when adult worms were incubated with anti-phospho-PKA (Thr197) antibodies that had been pre-adsorbed to the phosphorylated immunizing peptide no significant fluorescence was observed, with worms appearing similar to those shown in Fig. 2B. In contrast, incubation with anti-phospho-PKA (Thr197) antibodies revealed activated PKA in various regions of the worms (Figures 2, 3). Intense PKA activation was observed in the tegument of both sexes (Figures 2D, 3A) and this was particularly associated with the tubercles (Figures 2G–2J). PKA activation in the tegument was also more prominent in males than females, and towards the central dorso-lateral region of the male where larger tubercles are present (Figures 2D, 2F; cf. Figure 2M which shows the posterior of the worm). High magnification imaging revealed that regions of the tegument displaying PKA activation included the centres of the tubercles surrounded by spines and foci within the canyons between the tubercles (Figures 2G–J); these regions might represent putative sensory structures in the tegument surface. Furthermore, analysis of serial optical z-sections revealed activated PKA in the musculature immediately underlying the tegument (data not shown). In adult males, deeper scanning revealed activated PKA in the muscular wall of the seminal vesicle (Figures 2E, 2K, 2L), and gynaecophoric canal muscles (Figure 2E) with activation in the latter apparently associated with circular contractile rings (Figure 2E); the oesophagus and the highly muscular ventral sucker also possessed activated PKA as revealed by scanning the ventral side of the worm (Figure 2F). Additionally, activated PKA was visible around an area that resembled the collecting duct of the excretory system at the posterior of the male worm (Figure 2M) and in uncharacterized tubular structures located dorso-laterally between the oral and ventral sucker (Figures 2N–2P); these tubular structures extended approximately to the region where the seminal vesicle is sited (Figures 2K, 2L). In adult female schistosomes and additional to the tegument, activated PKA was associated with the oesophagus, ventral sucker and uterus (Figure 3A), ootype (Figure 3D), vitelline follicles and collecting ducts (Figures 3G, 3H); the common vitelline duct did not possess detectable activated PKA. Deeper scanning and cross-sectional analysis of the uterus and ootype regions revealed activated PKA to be associated with the muscular walls of these organs (Figures 3C, 3E). In addition, in some female worms diffuse PKA activation was detected in the ovary (Figure 3F). Finally, a striking ring-like distribution of activated PKA was detected in the female worm uterus surrounding the egg during egg expulsion along the uterus (Figures 3I–3K); such staining was only observed when an egg was present. Although the rhodamine phalloidin staining appeared diffuse in some cases, analysis of individual channels revealed these ring-like structures to be stained by rhodamine phalloidin thus defining their muscular nature.
Normally, activated PKA was not significantly detected in the nervous system of adult male or female worms when paired; however, in about 10% of specimens used (paired and separated) which were all from a single batch of worms collected from a pool of mice from only one infection, striking activation was seen (Figures 4, 5), the general distribution of which was similar between males and females. The basic neuro-anatomy of S. mansoni is similar in both sexes; however, the nervous tissue is particularly evident in males due to their larger size. Selective z-scanning to reveal the nervous system demonstrated that activated PKA was present in the anterior ganglia and their connecting commissures and in the dorsal and ventral nerve cords (Figures 4A, 5A); the typical orthogonal arrangement was clearly visible due to activated PKA within the nervous system. Furthermore, activated PKA was detected in the neuropile of the anterior ganglia and connecting commissure that comprised a widespread plexus of nerve fibres (Figure 5B). Activated PKA also existed within the complex innervation of nerve fibres and plexus of the oral sucker and ventral sucker in both sexes (Figures 4A, 4B, 5A, 5B) that originated from the anterior ganglia (determined by analyzing individual z-sections, data not shown). This plexus extended to nerve endings on the sucker tegument surface possessing distinct foci of PKA activation which we propose might have a sensory function (Figures 4B, 4D, 5B); activated PKA was associated with similar nerve endings and underlying plexus over much the worm body (Figures 4A, 4C, 5B).
High-resolution deep body scanning of the male worm peripheral nervous system revealed that activated PKA was also associated with large lateral peripheral ganglia that were positioned between the main nerve cord and more slender lateral nerve cord which both also displayed activated PKA (Figures 4E, 4F); analysis of optical z-sections revealed that some of these ganglia (Figure 4F) were close to the surface of the gut lining (data not shown). Nerve fibres, cell bodies and the complex nerve plexus also stained positive for activated PKA (Figures 4E, 4F). Analysis of z-sections revealed that this nerve plexus served the musculature of the gynaecophoric canal, sub-tegument and tegument surface where activated PKA was seen associated with the nerve endings amongst the spiny tubercles (Figure 4E). Diffuse PKA activation was also seen in the testicular lobes with more evident activation in the nerve fibres around the testes (Figure 4G). In the female, activated PKA was detected in the nerves associated with the ootype and Mehlis' gland complex (Figure 5C) and the ovary and seminal receptacle complex (Figures 5D, 5E).
Although extensive PKA activation in the nervous system was only observed in a small proportion of S. mansoni recovered, and was not commonly detected, we reasoned that it should be possible to activate neuronal PKA robustly in live S. mansoni with 50 µM forskolin. Incubation of worms in forskolin for 1 h resulted in extensive PKA activation in the nervous system (Figure 6), with control worms (not shown) appearing essentially as in Figures 2F and 3A. This increased PKA activation observed within intact worms in response to forskolin mirrors that observed by western blotting (Figure 1B).
Although the cause of the extensive activation of neuronal PKA observed in a proportion of worms from only one batch of mice was not known, given the distinct neuronal and muscular distribution of PKA we hypothesized that PKA might become activated in the nervous system during worm un-pairing. In agreement with our previous observations, confocal microscopy revealed that PKA was not activated extensively in the nervous system of paired adult worms immediately after perfusion (data not shown). Increased PKA activation was however observed specifically within the nervous system, including the nerve cords 15 min after pair separation in both male and female worms when compared to their paired counterparts and was sustained for 30 min and 60 min post-separation (Figure 7). Partial PKA activation was also observed in the nerve cord of a female worm that remained in copula at 30 min but activation was seen only in areas where the worm had protruded considerably from the male's gynaecophoric canal (Figure 7F); analysis of worms from at least three independent experiments revealed that such neuronal PKA activation was consistently absent when the worms remained closely coupled but was present when they separated. Although it is theoretically possible that separated worms permeabilize more effectively than paired worms, enabling better antibody penetration and labeling, this was not the case in our hands. On occasions when paired worms became separated within rotating tubes during the primary antibody incubation stage no differences in neuronal phospho-PKA labeling could be seen between these paired or separated samples. In addition, PKA activation in the nervous system appeared similar in forskolin treated worms regardless of whether they were paired or separated (data not shown).
Because activated PKA localized to the musculature and nervous system of adult S. mansoni, an experiment was conducted to ascertain the effect of PKA activation on worm movement. Preliminary assays revealed that forskolin treatment induced a phenotype that displayed considerable random contractile movements. Movies of adult worms were therefore captured for visual semi-quantitative analysis. When worms were treated with 50 µM forskolin there was a significant increase in gross muscular movements with time (P≤0.001; Figure 8A). After 15 min, the mean number of gross muscular movements in control worms was 7.2/min whereas the frequency in forskolin treated worms increased 2.6 times to 18.9/min (P≤0.05). This forskolin-mediated effect was even more pronounced after 20 min (P≤0.001) and was sustained until the end of the experiment (Figure 8A) by which time movements had increased to 4.3 times that of control (P≤0.001). Forskolin-treated worms also displayed excessive ‘coiling’ when compared to their non-treated counterparts and this was particularly evident after 20, 25 and 30 min exposure (30 min shown in Figure 8B). The effects of 100 µM forskolin on gross random muscular movements were more pronounced than with 50 µM forskolin, showing a significant increase after only 5 min (P≤0.05) when compared to controls (Figure 8A). After 10, 15, and 20 min, the frequency of movements was significantly greater in the presence of 100 µM forskolin than in 50 µM forskolin (P≤0.01), and after 20 min exposure worms displayed 41.5 movements/min compared to only 7.7 in the control group (Figure 8A; P≤0.001). Despite the increased motility observed in the presence of forskolin, there was no apparent difference in the number of worm pairs that separated during the course of the experiment. The movie (Video S1) provides a visual inspection of the effects of forskolin treatment on worm movement at 20 min exposure. Movies of worms exposed to forskolin for 20 min were then subjected to quantitative thrashing (body bend) analysis using the ImageJ plugin wrMTrck. Both the number of body bends and the extent of angular movement increased considerably following exposure to either 50 µM or 100 µM forskolin when compared to DMSO controls (Figure 8C), with irregular movements observed. Using a threshold of 6° change, the average number of body bends per worm following exposure to these concentrations of forskolin was 4.6 and 6.9 times that of controls, respectively, and was broadly similar to that determined for this time point by semi-quantitative analysis (Figure 8A). Importantly, wrMTrck analysis also revealed the extent of change in angular movement of worms following treatment. Whereas thrashing in excess of ±25° was infrequent in controls, it was considerably more common with forskolin (Figure 8C). Moreover, thrashing in excess of ±50° did not occur in controls, but did on eight occasions with 50 µM forskolin and 24 occasions with 100 µM forskolin. Collectively, this data highlights the nature of the hyperkinetic effect of forskolin and thus PKA activation on S. mansoni worm movement.
By using anti-phospho-specific antibodies and laser-scanning confocal microscopy we have mapped in detail activated PKA, the major transducer of cAMP signalling in eukaryotes [31], to discrete tissues of intact male and female adult S. mansoni. In male worms, activated PKA was found particularly associated with the tegument, gynaecophoric canal muscles, oral and ventral suckers, oesophagus, seminal vesicle wall, other areas of somatic musculature, and anterior tubular structures of unknown function. In females, activated PKA was observed particularly in the tegument, suckers, oesophagus, uterus and ootype wall, and ovary. In addition, in a subset of worms obtained from a pool of mice all from the same infection and in forskolin-treated worms, striking PKA activation was observed throughout much of the central and peripheral nervous systems including at nerve endings at the worm surface. Activation of neuronal PKA appeared to increase during worm separation in vitro, and pharmacological activation of PKA by forskolin induced hyperkinesias in worms in a time and dose-responsive manner. Collectively, these data are consistent with PKA likely playing a vital part in S. mansoni muscular activity and neuronal communication, and interplay between these two systems.
Schistosomes employ a variety of biogenic amines (e.g. 5-hydroxytryptamine (5-HT/serotonin), dopamine and histamine) and neuropeptides in their nervous system [32], [33]. Biogenic amines signal through GPCRs and in some cases activate adenylyl cyclase, elevating intracellular cAMP levels which in turn activate PKA [34], [35]. 5-HT increases the motility of intact schistosomes in vitro [e.g. 36]–[38] and has been localized to the male gynaecophoric canal and oral and ventral suckers [37] with a distribution similar to that seen with activated PKA in the current study. Other GPCRs such as SmGPR-3 [39] and SmD2 [40] which are activated by dopamine and are expressed in the central/peripheral nervous systems and body wall musculature, respectively, might influence S. mansoni movement in a complex fashion given that dopamine suppresses S. mansoni motor activity [39], [41] but induces cAMP production via SmD2 [40]. Furthermore, l-glutamate induces muscle contraction in isolated S. mansoni muscle fibres [42] likely via l-glutamate receptors [43] and kainic acid, an agonist that mimics the effect of glutamate, causes hyperkinesias and coiling in adult worms [44] similar to that observed with forskolin in the present study. Motor activity in S. mansoni thus appears to be under complex regulatory control from neuromodulators and classical neurotransmitters some of which will likely signal to PKA. Our findings should thus help drive forward research aimed at elucidating some of the crucial downstream signalling mechanisms that govern muscular activity which is central to parasite survival and reproduction in the host. It is important to note, however, that the mechanisms by which PKA influences muscle contraction and relaxation in mammals are complex and are not fully understood (see for example, [45]). Elucidating mechanistic control of motility in schistosomes will therefore require significant endeavor.
Extensive PKA activation was evident in the muscular walls of the uterus and ootype. The ootype which comprises regularly arranged circular and longitudinal muscle fibres [46] is the site of egg formation where an egg is produced from a fertilized ovum, with secretions from the vitelline cells and Mehlis' gland. The uterus, which possesses mainly circular fibers [46] leads from the ootype to the genital pore and opens close to the ventral sucker of the female. Swierczewski and Davies [25], [47] demonstrated that the PKA inhibitor H-89, or PKI 14–22 amide, significantly impaired egg output by S. mansoni, and when used at 10 µM or higher, H-89 stopped egg production during the first day of observation. Our findings suggest that active PKA helps regulate S. mansoni muscular activity and thus coordinated activation of PKA in the ootype wall could be vital for successful egg formation. In addition, because activated PKA was observed around eggs within the uterus during egg expulsion and also in ring like structures that circumscribed the egg, PKA activity could be necessary for egg propulsion, a process presumably enabled through peristaltic movement. Thus, the effects of PKA inhibitors on egg output by female worms observed by Swierczewski and Davies [25], [47] may have been, at least in part, due to blockade of egg formation in the ootype or dysregulated peristaltic movement along the uterus. Swierczewski and Davies [25] also reported that H-89 caused dissociation of worm pairs, although the duration required for separation was not reported. Here, in the presence of the PKA activator forskolin for 30 min, there was no difference in the number of worm pairs that separated when compared to controls despite the hyperkinetic effects of forskolin on worm motility.
Striking PKA activation was observed in an extensive network of nerve endings at the surface of the tegument, including those associated with the oral and ventral suckers. These endings were linked to the underlying nerve plexus associated with the peripheral and central nervous systems, which in some worms also displayed extensive PKA activation. The presence of such an evolutionarily-conserved signalling pathway [31] at tegumental nerve endings is intriguing as it suggests that S. mansoni might use these structures to transduce signals from the host and perhaps from other individual worms via PKA signalling. In this context, it is interesting that after worm pair separation in vitro, PKA activation visibly increased specifically in the nervous tissue, including in the central nervous system; in closely coupled worms, active PKA in the nervous system was essentially absent. From the behavioral stand point this is an important finding as it suggests that worm pairing and maintenance of the in copula state may somewhat be governed by sensory neuronal mechanisms mediated by PKA. As recently highlighted [2], integration of cell signalling into research on schistosome sensory biology will help drive forward this important area of research.
The presence of activated PKA in the muscular uterus, oesophagus, suckers, and ring-like structures in the gynaecophoric canal and in the uterus during egg propulsion, the nerves innervating the musculature, and the effects of forskolin on gross worm movements (hyperkinesia) shown here, coupled with un-pairing in the presence of H-89 reported previously [25] support a role for activated PKA in the regulation of S. mansoni motor activity. It is worthy to note that PKA possesses wide ranging regulatory roles in animals. For example, in neurons PKA has also been implicated in processes such as protein degradation, protein trafficking and gene transcription [48]–[50]. Indeed, PKA signalling through the nervous system might well be one mechanism by which schistosomes transmit signals to maintain homeostasis. Certainly, the lethal effect of PKA-C knockdown by either RNAi or PKA inhibition [25] signifies the central importance of this enzyme to worm function.
In S. mansoni PKA has also been implicated in regulating the ciliary motion of miracidia [51] likely through conserved axonemal mechanisms (discussed by Ressurreicao et al. [52]). In addition, PKA inhibition by H-89 has been shown to speed up the rate of transformation of miracidia to mother sporocysts [53], possibly through the effect of attenuated locomotion and thus early loss of epidermal ciliated plates. Expression levels of PKA-C were recently found to differ between different life stages of S. mansoni, and inhibition of PKA by H-89 or PKI 14–22 amide was found to be lethal to cercariae [47] as in adult worms. Our overall knowledge of the signalling mechanisms that regulate schistosome development and function is poor (discussed in [2]) and it is therefore important for future studies to encompass the importance of protein kinases, such as PKA, PKC [54] and MAPKs [55] to the development of definitive host stages particularly in the context of organism function and possible host-parasite interplay. In the case of PKA, consideration should be given to the importance of the regulatory subunits (also highlighted in [25]) and also AKAPs which permit targeting of PKA to certain sub-cellular locations. The current work, which includes a vital atlas of exclusively activated PKA in adult male and female S. mansoni, should prove invaluable for future studies into PKA function during worm development, pairing, and host-parasite interactions, as well as for studies into upstream effector mechanisms and downstream target molecules.
|
10.1371/journal.pcbi.1000381 | Functional Brain Networks Develop from a “Local to Distributed” Organization | The mature human brain is organized into a collection of specialized functional networks that flexibly interact to support various cognitive functions. Studies of development often attempt to identify the organizing principles that guide the maturation of these functional networks. In this report, we combine resting state functional connectivity MRI (rs-fcMRI), graph analysis, community detection, and spring-embedding visualization techniques to analyze four separate networks defined in earlier studies. As we have previously reported, we find, across development, a trend toward ‘segregation’ (a general decrease in correlation strength) between regions close in anatomical space and ‘integration’ (an increased correlation strength) between selected regions distant in space. The generalization of these earlier trends across multiple networks suggests that this is a general developmental principle for changes in functional connectivity that would extend to large-scale graph theoretic analyses of large-scale brain networks. Communities in children are predominantly arranged by anatomical proximity, while communities in adults predominantly reflect functional relationships, as defined from adult fMRI studies. In sum, over development, the organization of multiple functional networks shifts from a local anatomical emphasis in children to a more “distributed” architecture in young adults. We argue that this “local to distributed” developmental characterization has important implications for understanding the development of neural systems underlying cognition. Further, graph metrics (e.g., clustering coefficients and average path lengths) are similar in child and adult graphs, with both showing “small-world”-like properties, while community detection by modularity optimization reveals stable communities within the graphs that are clearly different between young children and young adults. These observations suggest that early school age children and adults both have relatively efficient systems that may solve similar information processing problems in divergent ways.
| The first two decades of life represent a period of extraordinary developmental change in sensory, motor, and cognitive abilities. One of the ultimate goals of developmental cognitive neuroscience is to link the complex behavioral milestones that occur throughout this time period with the equally intricate functional and structural changes of the underlying neural substrate. Achieving this goal would not only give us a deeper understanding of normal development but also a richer insight into the nature of developmental disorders. In this report, we use computational analyses, in combination with a recently developed MRI technique that measures spontaneous brain activity, to help us to understand the principles that guide the maturation of the human brain. We find that brain regions in children communicate with other regions more locally but that over age communication becomes more distributed. Interestingly, the efficiency of communication in children (measured as a ‘small world’ network) is comparable to that of the adult. We argue that these findings have important implications for understanding both the maturation and the function of neural systems in typical and atypical development.
| The mature human brain is both structurally and functionally specialized, such that discrete areas of the cerebral cortex perform distinct types of information processing. These areas are organized into functional networks that flexibly interact to support various cognitive functions. Studies of development often attempt to identify the organizing principles that guide the maturation of these functional networks. [1]–[6].
A major portion of the work investigating the nature of functional human brain development is based on results from functional magnetic resonance imaging (fMRI) studies. By examining the differences in the fMRI activation profile of a particular brain region between children, adolescents, and adults, the developmental trajectory of that region's involvement in a cognitive task can be outlined [3], [5], [7]–[10]. These experiments have been crucial to our current understanding of typical and atypical brain development.
In addition to fMRI activation studies, the relatively new and increasingly utilized method of resting state functional connectivity MRI (rs-fcMRI) allows for a complementary examination of the functional relationships between regions across development. Resting state fcMRI is based on the discovery that spontaneous low-frequency (<∼0.1 Hz) blood oxygen level dependent (BOLD) signal fluctuations in sometimes distant, but functionally-related grey matter regions, show strong correlations at rest [11]. These low frequency BOLD fluctuations appear to relate to spontaneous neural activity [11]–[13]. In effect, rs-fcMRI evaluates regional interactions that occur when a subject is not performing an explicit task (i.e., subjects are “at rest”) [11], [12], [14]–[23]. To date, rs-fcMRI has been used in several domains to examine systems-level organization of motor [11], memory [24],[25], attention [26], and task control systems [21],[22],[27].
In addition, because rs-fcMRI does not require active engagement in a behavioral task, it unburdens experimental design, subject compliance, and training demands. Thus, rs-fcMRI is becoming a frequently used tool for examining changes in network structure in disease [28]–[31], in aging [24],[29], and across development [22], [32]–[35].
In previous work regarding task-level control in adults, we applied rs-fcMRI to a set of regions derived from an fMRI meta-analysis that included studies of control-demanding tasks. This analysis revealed that brain regions exhibiting different combinations of control signals across many tasks are grouped into distinct “fronto-parietal” and “cingulo-opercular” functional networks [21],[36] (see Table 1 and Figure 1). Based on functional activation profiles of these regions characterized in the previous fMRI study, the fronto-parietal network appears to act on a shorter timescale, initiating and adjusting top-down control. In contrast, the cingulo-opercular network operates on a longer timescale providing “set-initiation” and stable “set-maintenance” for the duration of task blocks [37].
Along with these two task control networks [21],[36], a set of cerebellar regions showing error-related activity across tasks [36] formed a separate cerebellar network (Figure 1). In adults, the cerebellar network is functionally connected with both the fronto-parietal and cingulo-opercular networks [21],[22]. These functional connections may represent the pathways involved in task level control that provide feedback information to both control networks [22],[36].
Another functional network, and one of the most prominent sets of regions to be examined with rs-fcMRI, is the “default mode network”. The default mode network (frequently described as being composed of the bilateral posterior cingulate/precuneus, inferior parietal cortex, and ventromedial prefrontal cortex) was first characterized by a consistent decrease in activity during goal-directed tasks compared to baseline [38],[39]. Resting-state fcMRI analyses have repeatedly shown that these regions, along with associated medial temporal regions, are correlated at rest in adults [15],[16],[32],[40]. While the distinct function of the default mode network is often linked to internally directed mental activity [39], this notion continues to be debated [25], [32], [41]–[44].
In two prior developmental studies, we used rs-fcMRI to examine the development of the task control and cerebellar functional networks [22] and, separately, the default mode network [32]. The first study, addressing functional connectivity changes within and between the two task control networks and the cerebellar network [22], showed that the structure of these networks differed between children and adults in several ways (see [22]). In general, many of the specific changes showed trends of decreases in short-range functional connections (i.e., correlations between regions close in space) and increases in long-range functional connections (i.e., correlations between regions more distant in space). We suggested that these global developmental processes support the maturation of a dual-control system and its functional connections with the cerebellar network [22]. These results have now been replicated in a developmental resting connectivity study targeting sub-regions of the anterior cingulate [34].
The development of the default mode network was independently examined in a separate analysis [32]. In children, the default mode network was only sparsely functionally connected. Many regions were relatively isolated with few or no functional connections to other default mode regions. Over age, correlations within the default mode network increased and by adulthood it had matured into a fully integrated system. Interestingly, as opposed to the task-control and cerebellar networks, very few short-range functional connections involving the default mode network regions existed in children. Hence the numerous strong short-range functional connections that decreased with age when investigating the dual control networks were not seen within the default network. In fact, some connections such as the functional connection between the ventromedial prefrontal cortex (vmPFC; −3, 39, −2) and anterior medial prefrontal cortex (amPFC; 1, 54, 21) regions, which are fairly close in space (i.e., short-range at ∼2.7 cm), had a substantial increase in correlation strength over development [32].
The observation that different analyses suggested different developmental features suggests a need for a more nuanced and integrated characterization of the development of functional networks. The goal of this manuscript is to employ several different network analysis tools to provide such a characterization. Visualization techniques such as spring embedding, and quantitative measures, including ‘small world’ metrics and community detection algorithms, will be applied to these networks in an attempt to identify principles for the changes observed across development.
Because of the overlapping and sometimes inconsistent use of terminology between neuroscience and the computational sciences, we will briefly define two terms for the purposes of this paper. The term “networks” will be used in the typical cognitive neuroscience formulation: a group of functionally related brain regions (as described above). The overall collection of regions (encompassing all four “networks”) will be referred to as the “graph.”
Graph theory analyses were applied to 210 subjects, aged 7–31, to investigate the emergence of temporal correlations in spontaneous BOLD activity between regions of the default mode, cerebellar, and two task-control networks. For this initial analysis, average age-group matrices were created using a sliding boxcar grouping of subjects in age-order (i.e., group1: subjects 1–60, group2: subjects 2–61, group3: subjects 3–62, etc.). This generated a series of groups with average ages ranging from 8.48 years to 25.58 years. Each of the groups' average correlation matrices was converted into a graph, with correlations between regions greater than or equal to 0.1 considered as functionally connected.
In a first analysis, we used a visualization algorithm commonly used in graph theoretic analyses known as spring embedding that aids in the qualitative interpretation of graphs (Figure 2 and Video S1) [45]. In spring embedding, the positions of the nodes (i.e., regions) in a graph are based solely on the strength and pattern of functional connections instead of their anatomical locations. In this procedure, each functional connection between a pair of nodes is treated as a spring with a spring constant related to the strength of the specific correlation. The entire system of pair-wise regional functional connections is then iteratively allowed to relax to the lowest global energetic state, i.e., groups of nodes that are strongly interconnected will be placed close together even if anatomically distant.
By creating spring embedded graphs for each of the sliding boxcar groups in age-order, a movie representation can be made that shows the development of the network relationships (from average age 8.48 to 25.48 years) (Video S1). The panels in Figure 2 provide snapshots from child, adolescent, and adult average ages in this movie. In both Figure 2 and Video S1, each node is color-coded in two ways: the outer border represents the general anatomical location (i.e., cerebral lobe) of the node; the inner core color represents the coding by “function” as defined by a large number of fMRI studies.
One of the primary observations from the movie relates to this anatomical-functional distinction. In children, regions appear to be largely arranged by anatomical proximity. This arrangement can be seen in Figure 2 and Video S1 where, in children, regions can be readily grouped by cerebral lobe (outline colors of spheres in Figure 2 and Video S1). Over age, as functional connections mature, the node arrangements change such that anatomically close regions are now largely distributed across the graph layout, in a pattern more aligned with the mature networks' functional properties (core colors of spheres in Figure 2) [21], [36]–[39]. Thus, across development, local clusters of regions “segregate” from one another and “integrate” into more distributed adult functional relationships with more distant regions.
A group of regions in the frontal cortex provides a particularly salient example of segregation. Frontal cortex contains regions that, in adults, are members of each of the task-control networks (e.g., dlPFC, frontal, dACC/msFC) and the default network (e.g., vmPFC, amPFC). As can be seen in Figure 2A (and Video S1), extensive correlations exist between most of these frontal regions in childhood (see blue cloud Figure 2A). Over the developmental window afforded by the current dataset, some of these strong “frontal-frontal” correlations begin to weaken. With increasing age, regions in the frontal cluster segregate into 3 separate functional networks.
Accompanying this segregation is strong integration within the functional networks. The default mode network provides the clearest example. As illustrated in Figure 2B (and in Video S1), correlations between regions of the default mode network are weak (or absent) in children (red cloud, Figure 2B). Just as functional connections between the set of frontal regions are related to their anatomical proximity in children, the regions of the default mode network are each functionally connected to anatomical neighbors, and not to other members of the anatomically dispersed default mode network. Over age, however, the functional connections between default mode network regions mature and the network integrates into a highly correlated system in adults (Figure 2B and Video S1) (also see [32]). We note that these results were not specific to the 60-subject boxcar, and persist with smaller subject boxcars as well (see Video S2).
The qualitative observations noted above can be quantified using community structure detection tools. Using such an approach is particularly important because of the bias inherent in relying on qualitative methods for deciding whether groups of regions that appear to be clustered are indeed clustered, and because of the a priori definitions of each network. As stated by Newman:
Among the many methods used to detect communities in graphs, the modularity optimization algorithm of Newman is one of the most efficient and accurate to date [46]. This method uses modularity, a quantitative measure of the observed versus expected intra-community connections, as a means to guide assignments of nodes into communities. We applied the modularity optimization algorithm to the group connectivity matrices derived from the sliding boxcars described above.
Measures of modularity (Q) were high, and did not show large changes across the age range (Figure 3A and Figure S1 and Figure S2). This result was not dependent on any particular threshold (Figure S1). Although comparable community structure was detected at all ages examined, the components of the communities varied by age. As per our qualitative approach described above, in children, region clusters were largely arranged by cerebral lobe; while in adults, regions were largely clustered by their adult functional properties (Figure 4A). Again, this result was not unique to any particular threshold (Figure 4B and 4C) or size of boxcar (Figure S3). We do note, however, that limited data points (i.e., subjects) are available between the ages of 16 and 19 years (see Materials and Methods) and that our estimate of the specific transitions within this period should be interpreted with care.
As previously reported [22],[34], the segregation of closely apposed regions and the integration of distributed functional networks is associated with a general decrease in correlation strength between regions close in space and an increase in correlation strength between many regions distant in space. This trend is shown in Figure 5 and also Figure S4. Long-range functional connections tend to be weak, but increase over time (warm colors above the diagonal in Figure 5C and 5D and Figure S4C and S4D), integrating distant regions into functional networks. Short-range functional connections tend to be stronger (i.e., higher correlation strength) in children, yet those regions that do change predominantly become weaker over age (cool colors below the diagonal in Figure 5A and 5B and Figure S4A and S4B).
However, there are some interesting nuances to this trend that deserve mention. For instance, not all short-range functional connections decrease in strength over age (Figure 5A and 5B and Figure S4A and S4B). While few, some of the short-range functional connections, typically those in the same network, increase in strength over age (Figure 5A and Figure S4A). Similarly, although many long-range functional connections increase in strength, many others do not statistically change across development (Figure 5C and,5D and Figure S4C and S4D, grey connections).
In a seminal 1998 paper, Watts and Strogatz noted that the topology of many complex systems can be described as “small world”, a type of graph architecture that efficiently permits both local and distributed processing. Graphs with a regular, lattice-like structure have abundant short-range connections, but no long-range connections. Local interactions are thus efficient, but distributed processes involving distant nodes require the traversal of many intermediate connections. Conversely, completely randomly connected graphs are fairly efficient at transferring distant or long-range signals across a network, but they are poor at local, short-range information transfer.
Watts and Strogatz, and others, often describe “small world” properties with two metrics: the average clustering coefficient and average path length of a graph. The clustering coefficient measures how well connected the neighbors of a node are to one another. The average path length measures the average minimum number of steps needed to go between any two nodes. Lattices, optimized for local processes, have high average clustering coefficients but long average path lengths. Conversely, random graphs, which have no preference for short-range connections, have low average clustering coefficients and short average path lengths, making them well suited for communication between distant nodes. One of Watts & Strogatz's key insights was that by randomly rewiring a relatively small number of connections in a lattice graph (i.e., introducing a few long-range connections), a graph could retain its high average clustering coefficient, but dramatically reduce its average path length, thereby enabling efficient short- and long-range processes. It is this hybrid graph topology (i.e., high clustering coefficients and short path lengths) that matches the observed “small world” networks in many complex systems [47].
As previously reported [21],[48],[49], relative to comparable lattice and completely random graphs, the adult graph architecture showed high clustering coefficients and short path lengths, consistent with the ‘small world’ architecture (Figure 3B and 3C). Interestingly for these networks, in children (i.e., as early as age 8), these metrics were quite similar to adults (Figure 3B and 3C), and over age there was very little change in path lengths and clustering coefficients relative to comparable random and lattice graphs. It was originally anticipated that path lengths would decrease over age as long-range anatomical connections were added. Yet even at the youngest ages examined, path length was already quite short, near those of random graphs. Importantly, these results were not dependent on any particular threshold (Figure S5). We note that while the results shown here are largely descriptive, the error bars provided in Figure 3B and 3C constructed from random graphs underscores the difference between random configurations and the observed trends.
The combination of graph theoretic analyses and rs-fcMRI allowed for the examination of the dynamic relationships between multiple networks over development. In the current manuscript, we examined four networks - the cingulo-opercular, fronto-parietal, cerebellar, and default mode networks. As illustrated by qualitative observations in Figure 2 (and Video S1) and modularity analysis in Figure 4, locally organized groups of regions “segregate” over development into multiple distributed adult functional networks, while the functional networks themselves “integrate.” These results support the hypothesis that functional brain development proceeds from a “local” to “distributed” organization. However, despite the “local to distributed” developmental trend, ‘small world’ organizational properties are present in both 7–9 year old child and adult graph architecture.
In the following section, these results are discussed considering two postulates: (1) the temporal pattern of spontaneous activity measured by rs-fcMRI represents a history of repeated co-activation between regions, and (2) the brain attempts to use the most efficient processing pathways available when faced with specific processing demands.
As early as 1875 spontaneous synchronized neural activity has been used to study various aspects of adult brain organization [50]–[53]. However, despite the passing of over 130 years since its initial use, there remains uncertainty as to the role of intrinsic spontaneous brain activity in brain function. In adults, spontaneous correlated activity has been suggested to be important for gating information flow [54], building internal representations [43],[44],[54], and maintaining mature network relationships [43],[44],[54]. Much less work has been done in regards to development, but there are suggestions that spontaneous activity is important for the establishment of early cortical patterns (e.g., ocular dominance columns) [55]–[58] and may over time represent (in a Hebbian sense) a history of repeated co-activation between regions [21],[22],[27],[32],[34],[59],[60]. Within this framework, the changes in the correlation structure of spontaneous activity over development seen in this report may provide insight regarding the arrangement by which brain regions are communicating in children compared to adults.
If we consider the previously mentioned postulates, our results suggest that, typically, the most efficient way for children to respond to processing demands is to utilize more “local” level interactions as compared to adulthood. That is, in childhood there is, relatively greater co-activation of anatomically proximal regions than for adults with similar processing demands. A clear example of this is seen in Brown et al. [3], where identical task performance on lexical processing tests strongly activates a large set of visual regions in children, but strong visual activation is much more restricted in adults. These relationships may be reflected in correlated spontaneous activity measured via rs-fcMRI. The correlations in our youngest children would then represent the anatomical and spontaneous activity-defined initial regional relationships plus 7 years of experience-dependent Hebbian processes tuning these developing connections.
The “local to distributed” organizing principle resonates with recent suggestions that perceptual and cognitive development involve the simultaneous segregation and integration of information processing streams [1],[22],[76],[79],[80]. For instance, the “interactive specialization” hypothesis advanced by Johnson and colleagues, is consistent with these findings [1], [81]–[83]. Johnson points out that cortical regions and pathways have biased information processing properties at birth due to anatomic connectivity, yet they are much less selective than in adults (i.e., they are “broadly tuned”).
Interactive specialization predicts that shortly after birth, large sets of regions and pathways will be partially active during specific task conditions, However, as these pathways interact and compete with each other throughout development, selected regions will come online, be maintained, or become selectively activated or “tuned” as particular pathways dominate for specific task demands. Thus, regional specialization relies on the evolving and continuous interactions with other brain regions over development. If one extends this framework to the network level, the increases, decreases, and maintenance of correlation strengths seen between regions may reflect “specialization” of specific neural pathways to form the functional networks seen in adults.
The “local to distributed” developmental trajectory, discussed above, seems to be driven by an abundance of local, short range connections that generally decrease in strength over age as well as distant, long range connections that generally increase in strength over age. Given the more prevalent short-range connections in children, we expected a more lattice-like structure, with high clustering coefficients and relatively high path lengths. The results, however, clearly indicated that path lengths were near those of equivalent random graphs, and that the child functional networks are already organized as small world networks.
This result can be explained in the context of the re-wiring procedure discussed by Watts and Strogatz [47]. Randomly rewiring a small percentage of local connections in a lattice has a mild linear effect on clustering coefficients, but a highly non-linear effect on path lengths. This is to say, that by rewiring a small fraction of a lattice's connections, substantial drops in path lengths can be seen, with almost no change in the clustering coefficient. In late childhood, as shown in Figure 5 and Figure S2, there are already a significant number of long-range short cuts present. These long-range functional connections are likely responsible for the relatively short path lengths in the child group. We anticipate that if the developmental trajectory of short and long-range functional connections were extended to younger ages, fewer long-range ‘short-cut’ functional connections would be present, and more short-range functional connections would exist. Hence, the path lengths at these younger ages (<7 years old) would likely be longer. Nevertheless, by 8 years old, the networks already display ‘small world’ properties similar to those of adult networks, indicating that efficient graph structures are already in place for both local and distant processing, though they are organized differently than in later development.
While we identified small world properties in both child and adult graphs, the size of the graph is relatively small with only 34 nodes. Therefore, it is possible that with an increased number of nodes the specific results identified here will change, a possibility that will be addressed in further studies.
The regions used in the present analyses were all derived from adult imaging studies. It seems likely that additional regions may be included in one or more of these networks in childhood. In addition, individual differences with regards to the regions and networks chosen likely exist. Future work that includes regions derived from studies using a child population and obtaining the functional connections within subjects from individually defined functional areas may refine the networks and developmental timecourses presented here [84].
Of note, resting-state functional connectivity has been reported to be constrained by anatomical distance (i.e., correlations between regions decrease as a function of distance following an inverse square law) [85]. Thus, if a shift in this general bias occurred with development, then it is feasible that some of the changes seen here could be related to such a shift. With this said, the specificity of the connection changes observed over age, the number of connections that run opposite to the general trends, and the similarity of the distance relationship in connectivity between children and adults when plotting all possible connections (see Figure S6), all suggest that the majority of changes observed here are not related to changes in this bias. In addition, while there are now reports suggesting that changes observed over development with blood oxygen level dependent (BOLD) fMRI are not the product of changes in hemodynamic response mechanisms over age [86],[87], differences in the hemodynamic response function between children and adults could conceivably affect our results [88].
A limitation of rs-fcMRI in general is the restricted frequency distribution that can be examined. rs-fcMRI is used to measure correlations in a very low frequency range, typically below 0.1 Hz. Dynamic changes in correlations in other frequency distributions could exist (for example see [89]). It is also possible that there are undetected developmental changes in power across frequency bands orthogonal to the changes visualized here. The combination of other imaging and psychometric techniques with rs-fcMRI will likely help address these considerations. Characterizing additional networks and how these changes map onto behavior will also help further characterize functional brain development. Specifically, future work that demonstrates a direct relationship between behavior and the developmental trajectory seen here with rs-fcMRI, is presently needed to confirm (or reject) many of the theories presented here and elsewhere. Importantly, consideration of these issues need not be limited to developmental studies, but should be considered whenever investigators compare groups with rs-fcMRI.
Nonetheless, the general results presented here represent a strong set of hypotheses to be tested in broader domains and larger-scale brain graphs. First, that by age 8 years, regional relationships, as defined by rs-fcMRI, are organized as small-world-like networks, which, relative to adults, emphasize local connections. Second, that for the same regions, adult networks show similar network metrics but with regional relationships that have a longer-range, more distributed structure reflecting adult functional histories. In other words, the modular structure of large-scale brain networks will change with age, but even school age children will show relatively efficient processing architecture.
Subjects were recruited from Washington University and the local community. Participants were screened with a questionnaire to ensure that they had no history of neurological/psychiatric diagnoses or drug abuse. Informed consent was obtained from all subjects in accordance with the guidelines and approval of the Washington University Human Studies Committee.
fMRI data were acquired on a Siemens 1.5 Tesla MAGNETOM Vision system (Erlangen, Germany). Structural images were obtained using a sagittal magnetization-prepared rapid gradient echo (MP-RAGE) three-dimensional T1-weighted sequence (TE = 4 ms, TR = 9.7 ms, TI = 300 ms, flip angle = 12 deg, 128 slices with 1.25×1×1 mm voxels). Functional images were obtained using an asymmetric spin echo echo-planar sequence sensitive to blood oxygen level-dependent (BOLD) contrast (volume TR = 2.5 sec, T2* evolution time = 50 ms, α = 90°, in-plane resolution 3.75×3.75 mm). Whole brain coverage was obtained with 16 contiguous interleaved 8 mm axial slices acquired parallel to the plane transecting the anterior and posterior commissure (AC-PC plane). Steady state magnetization was assumed after 4 frames (∼10 s).
Functional images were first processed to reduce artifacts [23],[90]. These steps included: (i) removal of a central spike caused by MR signal offset, (ii) correction of odd vs. even slice intensity differences attributable to interleaved acquisition without gaps, (iii) correction for head movement within and across runs and (iv) within run intensity normalization to a whole brain mode value of 1000. Atlas transformation of the functional data was computed for each individual via the MP-RAGE scan. Each run then was resampled in atlas space (Talairach and Tournoux, 1988) on an isotropic 3 mm grid combining movement correction and atlas transformation in one interpolation [91],[92]. All subsequent operations were performed on the atlas-transformed volumetric timeseries.
For rs-fcMRI analyses as previously described [16],[23], several additional preprocessing steps were used to reduce spurious variance unlikely to reflect neuronal activity (e.g., heart rate and respiration). These steps included: (1) a temporal band-pass filter (0.009 Hz<f<0.08 Hz) and spatial smoothing (6 mm full width at half maximum), (2) regression of six parameters obtained by rigid body head motion correction, (3) regression of the whole brain signal averaged over the whole brain, (4) regression of ventricular signal averaged from ventricular regions of interest (ROIs), and (5) regression of white matter signal averaged from white matter ROIs. Regression of first order derivative terms for the whole brain, ventricular, and white matter signals were also included in the correlation preprocessing. These pre-processing steps likely decrease or remove developmental changes in correlations driven by changes in respiration and heart rate over age.
Resting state (fixation) data from 210 subjects (66 aged 7–9; 53 aged 10–15; 91 aged 19–31) were included in the analyses. For each subject at least 555 seconds (9.25 minutes) of resting state BOLD data were collected. 34 previously published regions comprising 4 functional networks (i.e., cingulo-opercular, fronto-parietal, cerebellar, and default networks; see Table 1 and Figure 1) were used in this analysis [16],[21],[22],[37]. For each region, a resting state timeseries was extracted separately for each individual. For 10 adult subjects, resting data was continuous. For the remaining 200 subjects, resting periods were extracted from between task periods in blocked or mixed blocked/event-related design studies [22]. These concatenated-extracted rest periods were shown to be equivalent to continuous resting data in a recent study describing this method [23]. In addition, several previous findings using this technique [21],[22],[32] have now been replicated using continuous resting blocks [27],[33],[34] and other continuous resting data [89].
To examine the functional connections within and between the large set of regions used in this manuscript we chose to use graph theory. Graph theory is particularly well suited to study large-scale systems organization across development, but requires the data be organized into specific correlation matrices. To do this, for each of the 210 subjects, the resting state BOLD timeseries from each region was correlated with the timeseries from every other region, creating 210 square correlation matrices (34×34). Average group matrices were then created using a sliding boxcar grouping of subjects in age-order (i.e., group1: subjects 1–60, group2: subjects 2–61, group3: subjects 3–62, … group151: subjects 151–210), thus generating a series of groups with average ages ranging from 8.48 years old to 25.48 years old with each group composed of 60 subjects. Average correlation coefficients (r) for each group were generated from the subjects' individual matrices using the Schmidt-Hunter method for meta-analyses of r-values [21],[85],[93]. In cases when the terms “child” or “adult” are used, the matrices or results referred to are the first and last of the sliding boxcar groups respectively, i.e., the child group is the youngest 60 subjects, with an average age of 8.48 years old, and the adult group is the oldest 60 subjects, with an average age of 25.48 years old.
To generate a dynamic representation of the functional connections between regions across development, each of the groups' correlation matrices was converted into a thresholded graph, such that correlations higher than r≥0.1 were considered connections, while correlations lower than the threshold were not connections.
For our initial analyses [21],[22],[32] graphs in child and adult groups were presented in either a pseudo-anatomical fashion or in their actual 3D positions (in Talairach space). Here we add another representation often used in graph theory - spring embedding. In this procedure, a spring constant is added to all of the connections in the network allowing for the pairwise regional connections to relax to their lowest energetic state. The algorithm applied in the present analysis is known as Kamada-Kawai [45] - one of the most commonly used strategies for displaying graph network data. In brief, each functional connection between a pair of nodes is treated as a spring with a spring constant related to the strength of the specific correlation. The nodes are then randomly placed in a plane, which places high strain on the “spring-loaded” connections. The algorithm then iteratively adjusts the positions of each node to reduce the total energy of the system to a minimum. As the pair-wise connections relax to their lowest energetic states the “natural” configuration of the network is revealed. By observing multiple “spring embedded” graphs across the subjects in age-order, approximately representing a 6 month temporal sliding box car (i.e., group1: subjects 1–60, group2: subjects 2–61, etc.), a movie representation can be made that shows the development of the full system (see Video S1). The interpolations, algorithm application, and movie production were performed using MATLAB (The Mathworks, Natick, MA) and SoNIA (Social Network Image Animator) [94].
Communities for our graph were detected with the modularity optimization method of Newman [46]. The modularity, or Q, of a graph is a quantitative measure of the number of edges found within communities versus the number predicted in a random graph with equivalent degree distribution. A positive Q indicates that the number of intra-community edges exceeds those predicted statistically. A wide range of Q may be found for a graph, depending on how nodes are assigned to communities. The set of node assignments that returns the highest Q is the optimal community structure sought by the modularity optimization algorithm, which follows a recursive two-step process. First, a modularity matrix similar to a Laplacian is constructed from the nodes in question, comparing observed versus expected edges. If this matrix has a positive eigenvalue, the eigenvector of the largest eigenvalue is used to split the nodes into two subgraphs, and Q is calculated. Second, nodes are swapped individually between the two subgraphs to see if an increase in Q can be found. Once a maximal Q is found from these swaps, the process is repeated on the subgraphs. At any point in this process, if the matrix has no positive eigenvalues, or if a proposed split does not increase Q, the subgraph is set aside, and defines a community. To detect communities in our networks over a range of ages, we used the sliding boxcar group average correlation matrices described above in “Generation of average group correlation matrices across development.” With weights retained, the modularity optimization algorithm was applied to each matrix along the sliding boxcar. A range of thresholds was explored to define connections for these calculations (see Figure 4 and Figure S1). Any particular threshold did not change the conclusions presented in the main manuscript. A threshold of 0.10 was chosen to display in the main manuscript because it balances two principles: (1) eliminating a multitude of weak correlations, which may obscure more physiologically relevant correlations, and (2) retaining high graph connectedness, so that communities arise from partitioning and not thresholding. Graph connectedness captures the extent of nodes fragmented from the main graph due to increasing thresholds. It is defined for a graph of N nodes as the mean of an NxN matrix, where cell i,j is 1 if a path exists between node i and node j (self-connections are allowed), and is 0 otherwise. A graph in which all nodes can reach each other has 100% graph connectedness, whereas a fragmented network in which some nodes cannot reach the rest has a lower connectedness. The modularity optimization analysis returned a set of community assignments for the nodes, as well as the Q of the graph with those assignments. The group assignments for the nodes were converted to colors and are displayed in Figure 4. The robustness of the community assignments was also tested using a different information theoretic procedure implemented by Meila, [95], which utilizes the measure ‘variation of information (VOI)’ (see Figure S7 and also [96]). All calculations were performed in MATLAB (The Mathworks, Inc., Natick, MA).
To characterize the relationship between connection length and the change in correlation strength over development, we split all 561 possible connections into 4 groups based on vector distance. Since using vector distance as an approximate for connectional distance is much more inconsistent when comparing ROIs across the midline, only intrahemispheric connections or connections to midline structures (i.e., within 5 mm of the midline) were examined. These connections were then sorted by connection length and plotted on a graph where the x-axis corresponds to the child correlation strengths and the y-axis corresponds to the adult correlation strengths (Figure 5 and Figure S2). On both the graphs (Figure 5) and the cortical surfaces (Figure S2), the color of the lines denotes the strength of correlation. Significant differences seen in Figure 5 and Figure S2 were obtained via direct comparison between children (the youngest 60 children out of 210 total subjects; age 7.01–9.67; average age 8.48) and adults (the oldest 60 adults out of 210 total subjects; age 22.47–31.39; average age 25.48). Two-sample two-tailed t-tests (assuming unequal variance; p≤0.05) were performed on all potential connections that passed the above criteria. Fischer z transformation was applied to the correlation coefficients to improve normality for the random effects analysis. To account for multiple comparisons the Benjamini and Hochberg False Discovery Rate [97] was applied. Connections that were significantly different between groups, but r<0.1 in both groups, were not displayed.
The small-world metrics were calculated according to descriptions by Watts and Strogatz [47]. In the main manuscript, calculations were performed on the group average correlation matrices thresholded at 0.10 and converted to binary matrices (for analysis across varying thresholds see Figure S3). For each matrix across age, the average clustering coefficient and average path lengths were compared to those values in lattices with equivalent N (number of nodes) and K (number of connections). To ensure that our matrices also differed from random graphs, 100 random graphs with equivalent degree distributions were also created. From these graphs mean average path lengths and clustering coefficients were calculated. These metrics are presented in Figure 3 and Figure S3. All calculations were performed in MATLAB (The Mathworks, Natick, MA).
|
10.1371/journal.pntd.0006603 | A cluster of cases of severe fever with thrombocytopenia syndrome bunyavirus infection in China, 1996: A retrospective serological study | A cluster of eleven patients, including eight family members and three healthcare workers with fever and thrombocytopenia occurred in Yixing County, Jiangsu Province, China, from October to November 1996. However, the initial investigation failed to identify its etiology. Severe fever with thrombocytopenia syndrome (SFTS) is an emerging infectious disease caused by SFTS bunyavirus (SFTSV), which was first discovered in 2009. The discovery of novel SFTSV resulted in our consideration to test SFTSV on the remaining samples of this cluster in September 2010.
We retrospectively analyzed the epidemiological and clinical data of this cluster. The first case, one 55-year-old man with fulminant hemorrhagic diseases, died on October 14, 1996. His younger brother (the second case) developed similar hemorrhagic diseases after nursing him and then died on November 3. From November 4 to November 15, nine other patients, including six family members and three medical staffs, developed fever and thrombocytopenia after exposure to the second case. The sera of six patients were collected on November 24, 1996. IgM antibodies against SFTSV were detected in all of the six patients’ sera using enzyme-linked immunosorbent assay (ELISA), while IgG antibodies were detected in one patient’s serum using an indirect immunofluorescence assay (IFA). We also found that IgG antibodies against SFTSV were still detected in four surviving patients’ sera 14 years after illness onset.
The mysterious pathogen of the cluster in 1996 was proved to be SFTSV on the basis of its epidemiological data, clinical data and serological results. It suggests that SFTSV has been circulating in China for more than 10 years before being identified in 2009, and SFTSV IgG antibodies can persist for up to 14 years.
| SFTSV was first discovered in 2009. It can be transmitted through tick bites, direct contact with SFTS patients’ blood or bloody secretion, and probable aerosol transmission. SFTS was listed as one of the nine infectious diseases on the WHO priority list in 2017 because of its trend of wider distribution and rising threat imposed on global health. It is worth mentioning that a cluster of eleven patients including eight family members and three healthcare workers developed fever and thrombocytopenia in China, from October to November 1996, but the initial investigation failed to identify its etiology. A retrospective analysis was conducted in September 2010. Based on the epidemiological, clinical, and serological findings, we speculated that SFTSV was the mysterious pathogen of the cluster. Meanwhile, high-titer SFTSV IgG antibodies were detected in four surviving patients’ sera. These SFTS patients preceded two cases in Japan in 2005 reported by Kurihara et al., which were once regarded as “the earliest SFTS cases worldwide”. These results suggest that SFTS should have existed without being diagnosed for a long time, and SFTSV IgG antibodies can persist long for 14 years, and perhaps even a lifetime after infection.
| Severe fever with thrombocytopenia syndrome (SFTS), an emerging haemorrhagic fever, was firstly confirmed among the rural areas in the central and eastern regions of China in 2009[1]. The main clinical features include fever, thrombocytopenia, leukocytopia, lymphadenopathy, and gastrointestinal symptoms. It has an average case-fatality rate of 12% but can be as high as 30%[2]. The causative agent, SFTS bunyavirus (SFTSV), is classified into the Phlebovirus genus, Phenuiviridae family, Bunyavirales order. It was once called as fever, thrombocytopenia and leucopenia syndrome virus (FTLSV)[3], or Huaiyangshan virus (HYSV)[4]. SFTSV is believed to be transmitted through tick bites[1, 5, 6], direct contact with SFTS patients’ blood or bloody secretion[7, 8], and probable aerosol transmission[9]. SFTS cases outside China were first reported in North Korea in 2009[10], South Korea in 2012[11] and Japan in 2013[12]. A closely related virus called Heartland virus was isolated from patients with similar symptoms in the United States [13]. Hence, SFTS was listed as one of the nine infectious diseases on the WHO priority list in 2017 because of its trend of wider distribution and rising threat imposed on global health.
Serological investigation showed that SFTSV infection was widespread in domestic animals (e.g. goats, sheep, cattle, dogs, etc.) and wild animals (e.g. rodent and shrews)[14, 15]. The seroprevalence of SFTSV in healthy people in China varies from 0.23% to 9.17%, depending on the investigated population and geography as well as the test reagent and methods, but only a small proportion of exposed persons develop clinical symptoms[16]. From the published documents, two SFTS cases in Japan in 2005 reported by Kurihara et al. have been regarded as the earliest cases in the world [17]. A recent phylogenetic study on SFTSV in China, South Korea, and Japan demonstrated that SFTSV could be divided into the Chinese clade and the Japanese clade, which may have evolved separately over time, except for the rare occasion of overseas transmission[18].These results suggest that SFTS may have existed without being identified for some time. In this study, we completed a retrospective analysis of a cluster of eleven patients with unexplained fever and thrombocytopenia in China in 1996 to determine whether SFTSV was responsible for this cluster.
The cluster of eleven patients with unexplained fever and thrombocytopenia occurred from October to November in 1996 in a township in Yixing County, which is located in southern Jiangsu Province of China and is characterized by hilly terrain. When the cluster was detected, public health workers were dispatched immediately to record the clinical and epidemiological information of patients and explore its causative agent. Sera of six patients were collected on November 24, 1996. Although the delay between the illness onset and sampling ranged from 9 to 20 days, all efforts were made to explore the causative agent at that time. Common pathogens including Hantavirus, Crimea-Congo Hemorrhagic fever virus, Orientia tsutsugamushi, Spotted fever group rickettsiae, Coxiella burnetii, Rickettsia Prowazeki, Rickettsia Mooseri, Salmonella typhi and other bacteria were excluded by blood culture and antibody tests in December 1996. Then, a small amount of the remaining samples were stored at a temperature of -80°C in the laboratory of Jiangsu Provincial Center for Disease Control and Prevention (JSCDC).
The discovery of novel SFTSV and impact of clinical manifestations of SFTS resulted in our consideration to test SFTSV on the remaining samples of this cluster in September 2010. Meanwhile, a retrospective investigation was conducted through interviewing the patients’ family members, neighbors and medical staffs, cross-checking several written timelines of the cluster, and collecting surviving patients’ sera.
Acute-phase sera of SFTS patients were detected for SFTSV-specific IgM antibodies using an ELISA kit (Xinlianxin, Wuxi, China) according to the manufacturer’s protocol [19]. In the initial screening, an undiluted serum sample was used to determine whether the sample was positive for antibodies against SFTSV. Positive serum samples were further diluted in 2-fold increments starting at 1:2 for titration of antibody titers with the same assay.
SFTSV-specific IgG antibodies were detected in all human sera by IFA as previously described [20].Twenty microliters of diluted (1:2 to 1:1280) serum samples were added to the cell-spotted coverslips with viral antigens and incubated for 45 minutes at 37°C. After washing, 20μL of FITC-conjugated goat anti-human IgG (Abcam, UK) diluted 1:80 with Phosphate Buffered Saline (PBS) containing Evans Blue (1:20,000) was added for further incubation for 30 minutes at 37°C. After washing for three times, the slides were mounted in glycerin and examined under an immunofluorescence microscope.
The study was approved by the Ethics Committee of JSCDC and informed consent was obtained from the participants. All data were analyzed anonymously.
On October 2 1996, a 55-year-old man developed dizziness, fatigue and sore throat, followed by fever and chills on October 4. Then, he developed nausea, vomiting, hematemesis and melena on October 12. Laboratory testing revealed that he had thrombocytopenia, leukopenia, elevated serum alanine and aspartate transaminase levels, proteinuria, and hematuria the next day.
On the morning of October 14, he was admitted to the local township healthcare center. Physical examination showed conjunctival congestion, scleral icterus, ecchymosis on the back of his right hand and right wrist joints, and sporadic hemorrhagic spots on his soft palate. He was administered with dexamethasone and oxygen. He died on that evening.
Retrospective investigation of the patient’s family members revealed that he was a mine safety supervisor, and his hobby was hunting. He caught three hares in the woods near his residence approximately 30 days before illness onset. He had no wife or child. During his illness, his younger brother attended to him day and night.
The second case was patient A’s younger brother. Patient B not only attended to patient A during his illness, but also cleaned his body and dressed him in funeral clothes before cremation. He had sudden onset of fever, chills, and headache on October 25, 11 days after patient A’s death. He was admitted to the same township healthcare center on October 28. On November 1, he was transferred to People's Hospital of Yixing County because of the severe condition. Physical examination on admission revealed supraclavicular lymph node enlargement and hepatomegaly. On November 2, he bled from his mouth and nose, and developed neurological symptoms such as seizures and extensive skin ecchymosis. He died the next morning despite the intensive care including transfusion and hemostatic therapy.
From November 4 to November 15, nine cases, including six family members and three medical staffs, all developed fever and thrombocytopenia. Patient C (patient B’s brother) developed symptoms on November 4 firstly, followed by patient D (patient B’s doctor), patient E (patient B’s daughter) and patient F (patient B’s doctor) on November 7, November 8 and November 10; patient G (patient B’s doctor), patient H (patient B’s elder son) and patient I (patient B’s younger son) presented with fever on November 11; The last two patients, patient J (patient B’s brother-in-law) and patient K (patient B’s nephew) developed symptoms on November 14 and 15, respectively. The mean age of these nine cases was 38.5 years (ranged from 22 to 63 years). Eight patients were male and one patient was female. Compared to the two fatal cases (patient A and B), eight follow-up cases had shorter duration from illness onset to admission with milder symptoms, and finally recovered after supportive treatment. The timeline of key events is shown in Fig 1, and all patients’ demographic and clinical information is shown in Table 1.
Two doctors (patient D&G) worked in People's Hospital of Yixing County and lived in the center of Yixing County. Patient F was an intern doctor in 1996. There was no clinical information about patient F in existing records, because he returned to his college in another city for medical treatment after his illness onset. Prior to the onset of the disease, the three medical staffs had provided medical services for patient B, while six family members had participated in attending to him in People's Hospital of Yixing County. Meanwhile, these nine cases had no contact with patient A before illness onset. Retrospective interviews showed that the three medical staffs had contact with blood or bloody secretion of patient B while rescuing the critically ill patient B on the evening of November 2; patient C and patient J cleaned up his body’s blood after patient B died; Detailed exposure histories of other patients were not remembered clearly by them and their family members.
Sera from patient C, patient D, patient G, patient H, patient I and patient K were collected on November 24, 1996. No serum was available from patient A and patient B. The time span from illness onset to sampling ranged from 9 to 20 days. SFTSV-specific IgM antibodies were detected in all of the six patients’ sera by ELISA, and SFTSV-specific IgG antibodies were detected in the sera of patient D by IFA. However, no SFTSV was isolated by using Vero, Vero-E6 and BHK 21 cell culture and no viral RNAs were detected by real-time reverse transcription PCR from these serum samples.
Sera of the four surviving patients were collected on September 17, 2010, nearly 14 years after illness onset. SFTSV IgG antibody titers were 1:80 in patient C, patient G and patient J, and 1:640 in patient D.
The cluster of eleven patients with unexplained fever and thrombocytopenia in 1996 occurred 14 years before the discovery of SFTSV. These cases were initially diagnosed as a viral haemorrhagic fever caused by Hantavirus or Crimea-Congo Hemorrhagic fever virus, which were known as the most common viruses causing severe hemorrhagic diseases in China, due to the main clinical manifestations including fever, thrombocytopenia and hemorrhages. However, the antibody test and nucleic acid test for these two viruses were negative and further analysis was carried out. Then, differential diagnosis including Orientia tsutsugamushi, Spotted fever group rickettsiae, Coxiella burnetii, Rickettsia Prowazeki, Rickettsia Mooseri, Salmonella typhi and other bacteria were considered, but the test results were all negative. Therefore, we suspected an outbreak of a severe transmissible infection of unknown etiology in People's Hospital of Yixing County and requested notification of all similar cases from the local medical institutions. However, there was no evidence of more cases at that time. Remaining serum samples from six cases were kept stored in a freezer at a temperature of -80°C from November 1996. To test SFTSV on these samples was considered owing to the discovery of novel SFTSV in 2009 and the impact of clinical manifestations of SFTS.
There are five reasons to extrapolate that SFTSV was the mysterious pathogen of the cluster. Firstly, all ten patients with medical records developed typical symptoms of SFTS such as fever and thrombocytopenia. Moreover, six of them had leucopenia, and two of them died with severe hemorrhage compatible with severe SFTS. Secondly, the cluster occurred in Yixing County, which was characterized by hilly terrain. Serological results showed that SFTSV had been circulating widely in Yixing County. The overall SFTSV seroprevalence in urban and rural residents in Yixing County in 2011 was 0.20% [21]. Average SFTSV seroprevalence in animal species in Yixing County in 2012 were: goats(66.8%), cattle(28.2%), dogs(7.4%), pigs(4.7%), chickens(1.2%), geese(1.7%), rodents(4.4%) and hedgehogs(2.7%)[14]. Thirdly, SFTSV infection was reported to occur from April to October annually in Jiangsu Province [22]. The index case had illness onset in October, which was in accordance with the seasonal pattern of SFTSV. Although the index case had no clear history of tick bite, he had high exposure risk for tick due to his occupation of a mine safety supervisor and his hobby of hunting. Fourthly, transmission was closely associated with blood or bloody secretion exposure from the index case or patient B, both of whom died of a fulminant febrile illness with hemorrhage. This is consistent with SFTSV transmission patterns reported in the previous literatures[7, 8, 23]. Last but not least, IgM antibodies against SFTSV were detected in the acute-phase serum samples of six patients by ELISA. Although SFTSV isolation and viral RNA detection are the gold standards for diagnosis, the appearance of anti-SFTSV IgM by ELISA is useful and has become one of the diagnostic criteria for a laboratory-confirmed SFTS case in China since the specificity and sensitivity of ELISA test is similar to those of the microneutralization assay and anti-SFTSV IgM exhibit no cross-reactivity with these antibodies to other closely related viruses such as hantavirus, Rift Valley fever virus, dengue virus, and so on[24–26]. Based on all of these findings, the cluster of eleven patients with unexplained fever and thrombocytopenia in China in 1996 was most likely caused by SFTSV.
Although one recent research by Qing-Bin Lu et al. found that SFTSV specific IgM antibody could be detected at a median of 9 days and remained persistent until 6 months after disease onset[27], it seems to be a theoretical concern more than a practical one (i.e., the chance of a person acquiring SFTSV infection during a given transmission season, maintaining a significant level of virus-specific IgM activity over the ensuing 6 months, and then again being re-exposed to SFTSV during the subsequent transmission season is highly unlikely, because our previous studies indicate that the seroprevalence rate of SFTSV in high risk population is less than 2% in Yixing County and the incidence of SFTSV infection is less than 5 cases/100,000 population in the highest incidence county[14, 28]). Therefore, the appearance of anti-SFTSV IgM is still a possible indicative sign of the clinical disease.
One highlight of our study was that SFTSV IgG antibody titers were detected in surviving patients with high titers 14 years after illness onset, suggesting that SFTSV IgG antibody could last for more than 10 years, perhaps even a lifetime after infection. At present, only one research about the persistence of SFTSV IgG antibodies found that SFTSV IgG antibody could be detected 3 years after infection[27]. It should be noted that the IFA used to detect IgG antibody against SFTSV in our study has good specificity and sensitivity compared to RT-PCR, and no any cross with other arbor-virus including hantavirus, and Japanese encephalitis virus, etc. The other highlight was that the cluster comprising eleven SFTS patients occurred in Yixing County, China, in 1996, which preceded the cases in Japan in 2005 reported by Kurihara et al. that might be mistaken as the earliest SFTS cases worldwide[17]. Our result suggests that SFTS have existed for a long time without being recognized.
Three limitations exist in our study. Firstly, no tissue or serum sample of patient B and patient F, especially of the index patient, was available for retrospective laboratory detection. Secondly, no viral RNAs were detected by real-time reverse transcription PCR from the six patients’ sera. This might be because the time of sera collection was later than the patients’ viremia period, or because the long preservation time and the sera freeze thawing resulting in viral RNA degradation. Thirdly, the cluster occurred such a long time ago that detailed disease-related exposure history could not be clearly remembered and completely recorded. Memory bias may exist in our research.
Our findings suggest that SFTSV has been circulating in China for more than 10 years before being identified and SFTSV IgG antibodies can persist for as long as 14 years.
|
10.1371/journal.pbio.0060097 | Iroquois Complex Genes Induce Co-Expression of rhodopsins in Drosophila | The Drosophila eye is a mosaic that results from the stochastic distribution of two ommatidial subtypes. Pale and yellow ommatidia can be distinguished by the expression of distinct rhodopsins and other pigments in their inner photoreceptors (R7 and R8), which are implicated in color vision. The pale subtype contains ultraviolet (UV)-absorbing Rh3 in R7 and blue-absorbing Rh5 in R8. The yellow subtype contains UV-absorbing Rh4 in R7 and green-absorbing Rh6 in R8. The exclusive expression of one rhodopsin per photoreceptor is a widespread phenomenon, although exceptions exist. The mechanisms leading to the exclusive expression or to co-expression of sensory receptors are currently not known. We describe a new class of ommatidia that co-express rh3 and rh4 in R7, but maintain normal exclusion between rh5 and rh6 in R8. These ommatidia, which are localized in the dorsal eye, result from the expansion of rh3 into the yellow-R7 subtype. Genes from the Iroquois Complex (Iro-C) are necessary and sufficient to induce co-expression in yR7. Iro-C genes allow photoreceptors to break the “one receptor–one neuron” rule, leading to a novel subtype of broad-spectrum UV- and green-sensitive ommatidia.
| Most sensory systems follow the rule “one receptor molecule per receptor cell.” For example, photoreceptors in the fly eye and cones in the human eye each express only one light-sensitive rhodopsin. Rhodopsins are G-coupled protein receptors, a class of ancient signaling molecules that mediate not just vision but also the sense of smell, the inflammatory response, and other physiological processes. However, the mechanisms that regulate mutual exclusion of receptor genes in the visual and olfactory systems are poorly understood. Each ommatidium in the fly eye consists of eight photoreceptors (R1–R8); six of which mediate broad-spectrum motion vision (R1–R6) and two that mediate color vision (R7 and R8). We identified a new class of photoreceptors in the fly retina that violates the one rhodopsin–one receptor rule. This subset of ommatidia, located in the dorsal third of the eye, co-expresses two ultraviolet-sensitive rhodospins (rh3 and rh4) in R7, while maintaining discrimination between green and blue opsins in R8. We took advantage of the genetic tools offered by the fruit fly to show that this co-expression depends on the Iroquois Complex (Iro-C) genes that are both necessary and sufficient to allow the two ultraviolet-sensitive rhosopsins to be expressed in the same R7 cell. These results shed new light on the mechanisms regulating co-expression of rhodopsins in the eye, and may well have implications for regulating co-expression in olfactory receptors and other G-protein coupled systems.
| The primary role of sensory organs is to probe the environment and to transmit precisely this information to the brain for processing. The visual and olfactory systems are composed of sensory epithelia with thousands of sensory receptor cells, each specifically expressing a single sensory receptor gene out of a much larger repertoire [1–6]. This “one receptor–one neuron” rule allows specific detection of sensory information at the periphery. Together, the architecture of the visual or olfactory organs, the correct specification of the sensory neurons, and the expression of specific sensory receptor molecules are crucial for the acquisition of sensory information. Sensory organs have thus adapted for optimal detection of specific stimuli and often exhibit spatial regionalization within the sensory organ itself. This regionalization also extends into topographic maps in the brain (retinotopy of the visual system, chemotopy in the olfactory system) [7].
The Drosophila compound eye is composed of approximately 750 simple eyes called ommatidia. Each ommatidium contains eight photoreceptor cells named R1–R8. The light-gathering structures (rhabdomeres) of outer photoreceptors (R1–R6) form an asymmetric trapezoid whose center is occupied by the rhabdomeres of the inner photoreceptors, where the R7 rhabdomere sits on top of that of R8 [8]. The last step in photoreceptor differentiation is the selective expression of one of the photosensitive pigments, the rhodopsins. The expression of a given rhodopsin, along with additional filtering or sensitizing pigments, dictates the color sensitivity of a photoreceptor. Five rhodopsins are expressed in the compound eye. They respect the general rule of “one receptor–one neuron”— R1–R6 cells express Rh1 [2]. The rhodopsins are similar in function to the vertebrate rods in that they are sensitive to a broad range of wavelengths. They are involved in motion detection. Inner photoreceptors (R7 and R8) mediate color vision [9,10], and are thus comparable to vertebrate cones [11,12]. These photoreceptors express the remaining four rhodopsins, which have a restricted spectrum of absorption ranging from ultraviolet (UV) in R7 to blue or green in R8 [1,3,4,13–15].
Although the eye appears to be composed of morphologically identical ommatidia, the main part of the retina consists of a mosaic of two stochastically distributed subtypes of ommatidia: pale type (p) contains a UV-absorbing Rh3 in R7 and a blue-absorbing Rh5 in R8; yellow type (y) contains a different UV-absorbing Rh4 in R7 and green-absorbing Rh6 in R8 [1,14,16]. A filtering pigment, “yellow” sharpens the sensitivity of yR7 and filters out the blue light reaching the green-sensitive underlying yR8 [17,18]. y ommatidia represent ∼70% of ommatidia in flies ranging from Musca to Drosophila. These ommatidia can now be defined more accurately based on their Rh content. The Drosophila homolog of the vertebrate dioxin receptor spineless (ss) is responsible for the specification of the retinal mosaic [19]. ss expression in ∼70% of R7 cells in pupae commits them to the yR7 fate and to express rh4. The cells that do not express ss become pR7, express rh3, and instruct pR8 to express rh5. By default, the remaining yR8 express rh6 [1,20]. Thus, 30% of ommatidia (p) appear to be more involved in the discrimination of shorter wavelengths, whereas the remaining 70% (y) should be more appropriate for the discrimination of longer wavelengths.
The p and y ommatidia appear to be randomly distributed. The Drosophila eye, like in many insects, has also developed a particularly striking example of sensory system specialization in the dorsal rim area (DRA). DRA ommatidia develop in the dorsal-most row of the eye and have distinct morphological characteristics that enable them to be used to detect the electric vector (e-vector) of light polarization [21,22]. Because polarized light comes from UV-rich sunlight scattered by the atmosphere, this row of ommatidia is limited to the dorsal edge of the eye and must therefore be specified by positional cues [22,23].
Regionalization of tissues often starts very early during organogenesis and often involves conserved molecular mechanisms that are important for patterning tissues as different as Drosophila sensory systems or vertebrate limb buds. In the Drosophila eye imaginal disc, dorso-ventral compartmentalization involves the differential expression of genes of the Iroquois Complex (Iro-C). Iro-C genes encode conserved homeodomain transcription factors from the TALE class [24]—araucan (ara), caupolican (caup), and mirror (mirr)—and their genomic organization as a cluster of three genes is conserved from flies to mammals [25,26]. In Drosophila, ara and caup have almost identical patterns of expression [27], whereas mirr is more divergent. Among other functions, these three genes have been implicated in very early stages of eye-antennal disc development as “dorsal selectors” that are required for the correct specification of dorsal head structures and for the formation of the dorsal compartment of the eye [28–30]. During larval development, the Iro-C genes are expressed in dorsal nondifferentiated cells of the eye imaginal disc and are then down-regulated once neurogenesis has begun. This expression distinguishes different cell fates on either side of the dorso-ventral boundary and is necessary to establish the organizer center at the equator (reviewed in [26]). Although expression of Iro-C genes fades away after the morphogenetic furrow, their expression reappears in the adult. Iro-C genes are necessary to specify the DRA: ommatidia near the edge of the disc are exposed to wingless signaling and become DRA ommatidia only when they are located dorsally [22,23].
Here we describe a new function for Iro-C genes in photoreceptor development: they define a subtype of ommatidia that is restricted to the dorsal region of the eye in which the “one receptor–one neuron” rule is broken. These ommatidia are positioned in the dorsal part of the retina and co-express the two genes encoding UV-absorbing Rhs—rh3 and rh4—in R7 cells. This co-expression results from the induction of rh3 in yR7 cells while pR7 are normal. Therefore, the mutual exclusion pathway that prevents co-expression of sensory receptors appears to be disabled by the activity of the Iro-C genes, allowing the expression of two sensory receptors in a single cell.
It is widely accepted that individual Drosophila photoreceptors express a single rhodopsin gene: rh1 in R1–R6, rh3 or rh4 in R7 [15,31] (Figure 1A), and rh5 or rh6 in R8 [1,3]. However, careful examination of antibody stainings on whole-mounted retinas revealed a surprising exception to this rule: a fraction of R7 cells co-expresses both rh3 and rh4 (Figure 1A and 1B) in a region that starts near the dorsal edge of the eye, outside the DRA, and extends toward the equator, spanning approximately one-third of the eye at its maximum point (Figure 1A). This phenomenon is also clearly observed in cross-sections of the eye (Figure 1C). In the ventral region of the eye, Rh3 and Rh4 proteins are present at a high level in R7 cells and are never found in the same cell (Figure 1C, “V”). In contrast, all R7 cells located in the dorsal eye contain Rh3, either alone or in combination with Rh4 (p and y subtypes, respectively, see below) (Figure 1C, “D”). In R7 cells co-expressing rh3 and rh4, the level of Rh3 protein is lower than in non–co-expressing cells (Figure 1A and 1B). Together, these data suggest that a subset of dorsal ommatidia induce rh3 expression in rh4-expressing yR7 cells (Figure 1B). Rh3 and Rh4 colocalization was observed using different combinations of primary antibodies, indicating that this is not an artifact of a particular pair of antibodies (unpublished data), and co-expression is present in all wild-type backgrounds tested to date (yw and all other Gal4 and upstream activating sequence (UAS) lines used in this study), suggesting that this is a conserved feature of the Drosophila eye.
In our previous studies, we had detected expression of an rh3 promoter fusion to a green fluorescent protein (GFP) reporter [32] in most ommatidia located in the dorsal eye. To distinguish whether the mutual exclusion or co-expression of rhodopsins in one cell results from transcriptional or post-transcriptional regulation, we performed double in situ hybridization to visualize rh3 and rh4 mRNA. In the ventral and central regions of the eye, rh3 and rh4 mRNA are present at high levels in a mutually exclusive manner (Figure 1D, “V”). However, in the dorsal eye, all R7 cells contain rh3 mRNA, either alone or in combination with rh4 mRNA (Figure 1D, “D”). Moreover, staining of rh3-lacZ reporter constructs consistently reveals expanded, weak rh3 transcription in all ommatidia in the dorsal eye, whereas restricted expression to p ommatidia is observed in the remaining part of the retina (unpublished data and [33]). Together, these data indicate that there is localized transcriptional control of rh3 and rh4 that allows their co-expression in the dorsal retina.
We quantified the frequency of R7 cells co-expressing UV-opsins. In line with previous observations, antibody stainings on dissociated ommatidia identified the three previously described subtypes of ommatidia [1,3]: DRA ommatidia that contain Rh3 in both R7 and R8 (Figure 2A), p ommatidia that contain Rh3 and Rh5 (Figure 2B), and y ommatidia that contain Rh4 and Rh6 (Figure 2C) [16]. In addition, a small proportion (5.7%; 6/106) of all ommatidia (dorsal or ventral) express Rh3 in R7 (without Rh4) associated with Rh6 in R8. These likely correspond to the previously described rare Rh3/Rh6 “odd coupled” ommatidia where the signal from pR7 fails to induce rh5 in R8 (unpublished data and [19,20,22]). However, we also identified a fourth subtype of R7 cells that contain both Rh3 and Rh4 in R7 cells (Figure 2D). These represent ∼10% of all ommatidia and are always coupled with Rh6-expressing R8 cells (Figure 2D). Stainings with anti-Rh4, anti-Rh5, and anti-Rh6 antibodies never revealed expression of Rh4 and Rh5 in the same ommatidium (0/200). yR7 cells contain a pigment that gave rise to their name (“yellow”) that is visible under the confocal microscope after neutralization of the cornea [17]. To further confirm that these co-expressing cells are yR7, we imaged the eyes of flies by confocal microscopy to visualize the “yellow” pigment as well as red fluorescent protein (RFP) controlled by the rh3 promoter (rh3>RFP). As expected, “yellow” and rh3>RFP do not overlap in the ventral eye, because “yellow” marks yR7 cells and rh3>RFP labels pR7 cells. However, in the dorsal eye, “yellow” overlaps with rh3>RFP (Figure 2E).
We have thus identified a class of dorsal ommatidia that express both rh3 and rh4 in R7, and rh6 in R8. These ommatidia represent a subset of y ommatidia that also express rh3 in addition to the endogenous rh4. Ommatidia containing Rh3/Rh5 make up ∼30% of all ommatidia as evaluated by quantification of dissociated ommatidia (there is no Rh4/Rh5 coupling, and Rh5-positive ommatidia represent ∼30% [178/636] of all ommatidia). The remaining ∼70% of ommatidia express rh6 (458/636). Stainings with anti-Rh3 and anti-Rh4 antibodies revealed that ∼30% of R7 express only rh3 (85/273), ∼60% express only rh4 (158/273), and ∼10% co-express rh3 and rh4 (30/273). Thus, the ∼70% rh6-expressing ommatidia can be divided into two subtypes: ∼60% of all ommatidia express rh4/rh6 and ∼10% express (rh3 + rh4)/rh6, representing y ommatidia in the dorsal region of the eye.
Iro-C genes control dorsal identity during early eye development; therefore, we analyzed their expression in an effort to identify the determinants of this “dorsal” identity [28,30]. As mentioned earlier, these genes are expressed transiently during early larval stages of eye disc development (Figure 3A). ara and caup (but not mirr) are re-expressed in the adult in the dorsal retina. To perform a more detailed analysis of the Iro-C gene expression pattern, we used reporter lines (Iro-C-nuZ or Iro-C-Gal4), that are insertions in the Iro-C complex and are believed to reflect the expression of both ara and caup [21,27]. At 24 h after puparium formation (APF), the Iro-C-Gal4 reporter is highly expressed in all photoreceptors in the dorsal eye (Figure 3B). The level of expression gradually decreases toward the equator due to fewer and fewer cells per cluster that express the reporter. Ultimately, only R7 cells, identified with the R7-specific marker Prospero (Pros), express the reporter (Figure 3C). In the adult, the expression of the reporter persists in outer photoreceptors, as well as in R7 and R8 as previously shown [22]. This expression pattern correlates with the distribution of y ommatidia that co-express rh3 and rh4 in R7 and express rh6 in R8 (Figure 3E and 3F) (see below for discussion). Thus, in the adult retina, the Iro-C genes ara and caup are specifically expressed in the region of the eye where there is co-expression of rhodopsins.
The similarity between the expression profile of the Iro-C genes ara and caup in the region of the eye where rh3 and rh4 are co-expressed suggested that these transcription factors regulate this newly defined subset of ommatidia. To test this hypothesis, we induced clones of cells that were mutant for Iro-C by using a deficiency that covers ara and caup and deletes most of the regulatory sequences of mirr [27,34]. Ventral clones are easily recovered but, as expected, they do not have a visible phenotype. While small dorsal clones do not produce a strong morphological phenotype, large clones often lead to the formation of ectopic eye tissue near the dorsal head cuticle, presumably because they create a new organizer between Iro-C + and Iro-C – cells (unpublished data)[28,30]. In the few dorsal mutant clones recovered, R7 cells co-express rh3 and rh4 in the surrounding heterozygous tissue, whereas in mutant tissue, R7 cells contain only Rh3 or Rh4 (Figure 3G). Thus, similar to the adult ventral eye where Iro-C is not expressed, dorsal Iro-C mutant R7 cells exclusively contain either Rh3 or Rh4. Therefore, Iro-C expression in the dorsal eye appears to be required for rhodopsin co-expression in R7 cells of dorsal y ommatidia.
To study whether the ara and caup genes are sufficient to induce rhodopsin co-expression, we performed a series of mis-expression experiments. We observed essentially the same phenotype when over-expressing ara and/or caup, with the only difference being that the over-expression of both genes produces a more severe morphological phenotype than the expression of either one of them alone. We only show experiments using UAS-caup, but the same set of data is presented for ara in Figure S1. Because mirr is not expressed at this stage, we did not investigate its mis-expression phenotype. To mis-express ara and caup genes, we used the long glass multiple reporter-Gal4 (lGMR-Gal4) driver whose expression is restricted to all photoreceptors. lGMR expression starts during larval stages, after photoreceptors are specified at the morphogenetic furrow and is maintained throughout photoreceptor development and adulthood [19,35]. Over-expression of caup or ara at 25 °C leads to strong morphological defects in the eye, likely due to the prolonged expression of Iro-C genes when they are normally down-regulated during photoreceptor development. However, lowering Gal4 activity by raising flies at 18 °C induces robust lGMR>caup–dependent co-expression of rh3 and rh4 specifically in all yR7 cells (Figure 4A), whether ventral or dorsal. Importantly, caup-induced expansion of rh3 in yR7 cells does not repress rh4 expression. lGMR>caup over-expression does not induce ectopic expression of rh3 in outer (R1–R6) or in R8 photoreceptors, and co-expression of rh5 and rh6 is not observed. However, lGMR>caup does increase to various degrees the proportion of rh6-expressing R8 cells with a corresponding decrease in rh5-expressing cells (Figure 4B). This expansion of Rh6 in R8 cells produces mis-coupling between R7 and R8 cells, resulting in an increase in ommatidia containing Rh3 in R7 and Rh6 in R8. Our interpretation is that, because lGMR>Iro-C produces morphological defects in the eye, the communication between R7 and R8 might be disrupted. In the absence of a signal from R7 to R8, most R8 cells express the default rh6 (as in sevenless mutant eyes) [1,20].
We have previously shown that the decision between p and y fates is made during early pupation, when ss is activated in yR7 precursors, after lGMR-Gal4 starts to be expressed and long before rhodopsins are expressed [19]. To test whether Iro-C genes can cell-autonomously induce rhodopsin co-expression after the p versus y decision is made, ara and caup genes were expressed using a promoter that is expressed at late stages in development. PanR7-Gal4 is a combination of rh3 and rh4 promoters that is expressed in every R7 cell and in DRA R8 cells [19], starting at late pupal stages when rhodopsin expression starts [36]. Over-expression of caup using this late driver induces co-expression of rh3 and rh4 in the majority of R7 cells (Figure 4C), which are likely yR7 cells. To test whether a very late signal can induce co-expression in yR7, we expressed Iro-C genes using a rh4-Gal4 driver, which is only expressed in yR7 cells. This should allow R7 to be normally specified as yR7 and turn on rh4, which would then supply the Iro-C signal. Again, mis-expression of caup using this driver induces expression of rh3 in most rh4-expressing cells (Figure 4E). The phenotype is stronger in the central or more-dorsal areas than in ventral regions where this driver is not able to transform all yR7 cells, because it might lack the strength of the PanR7 driver. Together, these results suggest that there is an endogenous sub-threshold level of Iro-C in the dorsal eye close to the equator that is not sufficient to induce co-expression of rh3 and rh4 in a wild-type situation. The PanR7- and rh4-Gal4 drivers must only add limited amount of ara or caup, or provide it late, such that not all yR7 cells co-express. Neither PanR7- nor rh4-Gal4 drivers induce phenotypes in R8 cells (31.7% and 33% of rh5 expression, respectively) (Figure 4D and 4F), suggesting that, as expected, the early decision between p and y fates is not affected. In addition, the expression of caup only in R8 with rh5- and rh6-Gal4 drivers does not produce a visible phenotype (Figure S1G). Therefore, the presence of the Caup or Ara transcription factors in yR7 cells, even very late in development, instructs them to co-express rh3 and rh4.
The Drosophila retina presents a stochastic distribution of ommatidial subclasses. As described before, in ∼30% of ommatidia, R7 express rh3 and R8 express rh5, whereas in the remaining ∼70%, R7 express rh4 and R8 express rh6 [1,16,37]. These numbers are correct if we consider only R8 rhodopsin expression. However, the data presented here indicate that, if one considers the distribution of R7 rhodopsins, the y ommatidial subtype should be divided into two subpopulations: the “classical y” subtype that expresses only rh4 in R7 and rh6 in R8; and the “dorsal y” subtype that expresses both rh3 and rh4 in R7 and rh6 in R8. Thus, Drosophila contain four (or even five, if we consider the “odd coupled” ommatidia) rather than the three classes that were previously described: DRA, p, y, and “dorsal y” ommatidia.
The expression of ara and caup allows co-expression of rhodopsins in R7 cells by inducing the expression of rh3 in rh4-expressing cells. Although Iro-C gene products could activate the rh3 promoter directly, it should be noted that they are expressed in all photoreceptors in the dorsal eye, but rh3 is only induced in rh4-expressing R7. In addition, over-expression of Iro-C in all photoreceptors using lGMR-Gal4 only induces rh3 in R7, and not in other photoreceptors (Figure 3A).
During development, photoreceptors are subdivided first into two different subtypes, inner (R7 and R8) and outer (R1–R6) by the expression of the two transcription factors encoded by the spalt (sal) complex in inner photoreceptors (Figure 5A). After photoreceptors acquire a generic “inner” fate, prospero is expressed in R7 and directs it away from the R8 fate and toward an R7 fate. Similarly, senseless plays a parallel role in R8 cells to prevent R7 differentiation [38]. At this stage, R7 and R8 are specified as photoreceptors, but they are not patterned in terms of rhodopsin expression. The dorsal-most row of ommatidia is then specified as DRA by the expression of homothorax (hth), which differentiates them from the rest of the retina to become polarized light detectors (Figure 5A and 5B). The main part of the retina is then patterned into the y and p subtypes by the expression of ss during pupation in a subset of R7 cells (Figure 4A). R7 cells that do not express ss become pR7 cells by default [19]. rh3 is activated by orthodenticle (otd), which is present in all photoreceptors, and thus its activity must be actively repressed in yR7 cells. ara and caup might be the signal in yR7 cells that allows the expression of the default state rh3 and breaks the mutual exclusion pathway between rhodopsins (Figure 4A and 4B). R8 cells that are located below dorsal yR7 cells cannot co-express rhodopsins because a bi-stable loop between warts and melted does not allow an ambiguous choice between the rh5 and rh6 fates after the decision is made [39].
The co-expression of rhodopsins in R7 is restricted to the dorsal eye, which faces the sky. The biological significance of these particular ommatidia in Drosophila is not known. The Drosophila “dorsal y” ommatidia that contain both UV-Rh3 and UV-Rh4 in R7 and green-absorbing Rh6 in R8 provide a unique configuration to measure the ratio between UV and long wavelengths: They contain two UV opsins in R7, providing broad UV sensitivity that is expanded toward shorter wavelengths by Rh3, along with a blue-filtering pigment that prevents short wavelengths to penetrate the R8 layer containing the green-absorbing Rh6 [8]. These ommatidia might be used to discriminate between the “solar” and “antisolar” halves of the sky, necessary to navigate in the correct direction [40].
Although the exclusion of sensory receptors is a general rule, co-expression to achieve a novel sensitivity might be used in special cases when the expression of a single receptor is not sufficient to confer high enough sensitivity. Although the mouse retina is dominated by rods, it also contains cone cells. The majority of these cone cells co-express both S (blue) and M (green) opsins [41]. Presumably, mice live in a dark environment and are mostly color blind; the co-expression might be useful for optimal utilization of cones. The eye of butterflies also displays co-expression of two rhodopsins in several of their photoreceptors, perhaps to expand the spectrum of sensitivity of photoreceptors in species that do not have a rhodopsin with broad absorption spectrum such as Rh1, which is unique to Diptera [42–44].
Vertebrate olfactory neurons also express only one olfactory receptor gene per olfactory receptor neuron, and a direct feedback from the expressed receptor molecule has been proposed to ensure that this rule is stringently applied [45–47]. However, it cannot be excluded that two olfactory receptor genes are co-expressed, because their large number prevents comprehensive expression studies. Indeed, in Drosophila, a striking example of co-expression of two chemosensory receptors that mediate sensitivity to CO2 was recently described for the olfactory system [48]. The expression of each receptor is not sufficient to confer olfactory CO2-chemosensation on its own, but their combined expression does. Therefore, the addition of multiple receptors might not only increase the receptive spectrum of cells, but might also confer sensitivity to new stimuli. In the CO2 sensitivity case, the co-expression is crucial for the fly to detect a repellent smell that indicates danger. Thus, precise regulation of receptor co-expression must be achieved.
The spatial specialization induced by Iro-C genes in the fly retina is not the only example where regionalized specification occurs within sensory systems. For example, in the “love spot” of the housefly Musca, the antero-dorsal region of the male eye has presumably lost color vision, because R7 cells are transformed into motion detecting outer photoreceptors that express Rh1 [49]. The human eye also has geographic specialization: the center of the eye (fovea) contains exclusively cones that are involved both in acute and color vision in bright light. The periphery of the eye is mostly composed of rods and is involved in dim light vision (reviewed in [50]). The mouse olfactory system also exhibits specialization where the main olfactory epithelium that is responsible for detection of general odorants is separated from the vomeronasal organ that is involved in pheromone detection [51]. Drosophila also has two olfactory organs, the antenna and the maxillary palps, which express different sets of olfactory receptors and are likely involved in the detection of different types of odors [52].
Iro-C genes may not only be responsible for relieving the “one receptor–one neuron” constraint in the Drosophila eye, but may also allow receptor co-expression elsewhere. For instance, members of the orthologous family, the Irx genes, are expressed in mouse photoreceptors where opsin co-expression is observed [53–55]. Although the terminal differentiation of bipolar cells is affected in mice with mutant Irx5 [54], it will be of interest to study cone opsin expression in this and other Irx mutants to test whether these genes are also involved in the co-expression of opsins.
Mouse olfactory neurons do not express Irx5 or Irx6 [53], and they do not express more than one olfactory receptor gene [56]. In contrast, recent comprehensive studies in the Drosophila antenna and maxillary palp have identified a subgroup of olfactory receptor neurons that co-express two divergent receptors [57,58]. Interestingly, cells that co-express different olfactory receptor genes are the only neurons that express Iro-C genes in the maxillary palp (EOM, AC, and CD; unpublished observations). Unfortunately, the loss of Iro-C function in this tissue leads to re-specification of these neurons toward other non-neuronal fates (EOM, AC, and CD; unpublished observations), preventing us from further testing the involvement of Iro-C genes in the lack of exclusion.
Genes directly controlled by Iro-C transcription factors are still elusive. Binding sites for Mirr that presumably mediate repression of fringe in the dorsal eye disc were recently described [59]. The identification of target genes of the Iro/Irx family might shed some light on the regulation of the pathway that maintains mutual exclusion of sensory receptors.
Flies were raised on standard corn meal–molasses–agar medium and grown at room temperature (24 ± 1 °C) unless otherwise noted. y1w67 flies were used as control for Rhodopsin expression. As the red color of adult eyes interferes with fluorescent immunostainings, the eyes were rendered white by using an RNAi construct against the white gene [60] when a white marker gene was introduced in the genetic background by P-element transgenes. lGMR-Gal4 was produced by a pentamerized Glass binding site [22], UAS-ara and caup were gifts from J. Modolell. irorF209-PZ and Df(3L)iroDFM3 were obtained from the Bloomington Stock Center. Iro-C-Gal4 was created by replacing the P element in irorF209-PZ with one containing Gal4. rh3-, rh4, and PanR7-Gal4 drivers were described in [19]. To visualize “yellow” and rh3 expression with a reporter, we used flies containing rh3-LexA and lexAop-RFP. Clones were generated using the standard FLP/FRT technique.
Antibodies and dilutions used were as follows: mouse anti-Rh3 and anti-Rh4 (1:100) and mouse anti-Rh5 (1:50) (gift from S Britt, University of Colorado); rabbit anti-Rh4 (1:400) (gift from C. Zuker, University of California San Diego); rabbit anti-Rh6 (1:5000); rabbit anti-βGal (1:5000) (Cappel); mouse anti-βGal (1:500) (Promega); mouse anti-pros (1:50) (Developmental Studies Hybridoma Bank); rat anti-ElaV (1:10) (DSHB); and rabbit anti-GFP (1:800) (Biogenesis). Chicken anti-Rh3 was from [61]. All secondary antibodies were Alexa-conjugated (1:800) (Molecular Probes). Throughout the paper, Rh3 and Rh4 were stained using antibodies generated in mouse and rabbit, respectively, because they are significantly better that the other two.
Antibody stainings for larval and pupal retinas were essentially the same except for the collection of tissue. The protocols merge after the fixation step. Cerebral complexes of late third instar larvae were dissected in phosphate buffered saline (PBS) (1×) and fixed in PBS + 4% paraformaldehyde for 20 min at room temperature (RT). Pupal cases were collected at 24 h after puparium formation at 25 °C and the head was dissected in ice cold PBS (1x). Several eye-brain complexes were extracted by gentle pipetting and collected in PBS (1×) on ice. After 20 min fixation using PBS (1×) + 4% formaldehyde at RT, the samples were washed four times with PBS + 0.1% Triton-X-100 (PBT). The first antibody was added overnight at 4 °C. After four washes with PBT, the secondary antibody was added for at least 2 h at RT. After another four washes in PBT, each retina was separated from the brain by using two tungsten needles and then mounted flat in Vectashield (Vector Laboratories).
10-μm horizontal eye sections were produced using a cryostat (Zeiss) and deposited on Superfrost PLUS slides (Fisher). The slides were then fixed 15 min in PBS (1×) + 4% formaldehyde. After four washes with PBT, the first antibody was added overnight at 4 °C. After four washes with PBT, the secondary antibody was added for at least 2 h at RT. After four washes with PBT, the slides were mounted in Aquamount.
Adult retinas were dissected out and after a rinse with PBS (1×), they were fixed for 15 min with 4% formaldehyde at RT. After three washes in PBT, the retinas were incubated with the primary antibodies diluted in BNT (PBS, 0.1% BSA, 0.1% Tween-20, 250 mM NaCl) overnight at 4 °C. After two rinses and a 30 min wash with PBT, the retinas were incubated with secondary antibodies for 2–4 hours at RT. Two quick rinses with PBT were followed by an overnight wash at 4 °C. Retinas were cleaned of any remaining cuticle and mounted in Vectashield.
The retina of 3–5 dissected eyes were removed from the cornea and dissociated on a slide using dissection needles in a drop of PBS. After the samples dried at RT, they were fixed with 4% formaldehyde and staining was carried out as for frozen sections.
Adult retinas were dissected as described for antibody stainings. Dissected retinas were mounted on Superfrost PLUS slides (Fisher) and dried for 2 h at 65 °C. After fixation for 15 min with 4% paraformaldehyde, the slides were washed in PBS, treated with Proteinase K for 5 min at 37 °C and refixed for 10 min. Following a short PBS wash, the slides were treated with 0.2 M HCl for 10 min, washed in PBS, and acetylated with 0.1 M triethanolamine. Retinas were hybridized overnight at 65 °C with 100 μl hybridization buffer (50% formamide, 5× SSC, 5× Denhardt's, 250 μg/ml yeast tRNA, 500 μg/ml herring sperm DNA, 50 μg/ml heparin, 2.5 mM EDTA, 0.1% Tween-20, 0.25% CHAPS) containing a digoxygenin-labeled rh3 probe and a fluorescein-labeled rh4 probe. After a series of washes in 5× SSC; 50% formamide, 2× SSC; 2× SSC; 0.2× SSC, and 0.1× SSC (5, 30, 20, 20, and 20 min, respectively), the rh3 probe was detected using HNPP/Fast Red (Roche) and the rh4 probe was detected using the TSA Biotin System (Perkin Elmer) and streptavidin-Alexa488 according to the manufacturers suggestions.
Anesthetized flies were fixed to a Petri dish using nail polish. Then, flies were submerged in water and visualized using a 20× water immersion lens. To visualize “yellow”, FITC settings were used [31].
|
10.1371/journal.pcbi.1006686 | Exploring chromatin hierarchical organization via Markov State Modelling | We propose a new computational method for exploring chromatin structural organization based on Markov State Modelling of Hi-C data represented as an interaction network between genomic loci. A Markov process describes the random walk of a traveling probe in the corresponding energy landscape, mimicking the motion of a biomolecule involved in chromatin function. By studying the metastability of the associated Markov State Model upon annealing, the hierarchical structure of individual chromosomes is observed, and corresponding set of structural partitions is identified at each level of hierarchy. Then, the notion of effective interaction between partitions is derived, delineating the overall topology and architecture of chromosomes. Mapping epigenetic data on the graphs of intra-chromosomal effective interactions helps in understanding how chromosome organization facilitates its function. A sketch of whole-genome interactions obtained from the analysis of 539 partitions from all 23 chromosomes, complemented by distributions of gene expression regulators and epigenetic factors, sheds light on the structure-function relationships in chromatin, delineating chromosomal territories, as well as structural partitions analogous to topologically associating domains and active / passive epigenomic compartments. In addition to the overall genome architecture shown by effective interactions, the affinity between partitions of different chromosomes was analyzed as an indicator of the degree of association between partitions in functionally relevant genomic interactions. The overall static picture of whole-genome interactions obtained with the method presented in this work provides a foundation for chromatin structural reconstruction, for the modelling of chromatin dynamics, and for exploring the regulation of genome function. The algorithms used in this study are implemented in a freely available Python package ChromaWalker (https://bitbucket.org/ZhenWahTan/chromawalker).
| A new era in chromatin research started with the availability of Hi-C data and new experimental techniques driving improvements in data resolution enable us to achieve a deeper understanding of the chromatin structure and function, while calling, at the same time, for the development of more advanced analytical methods. Though instrumental in the analysis of Hi-C data, both model-driven polymer models and data-driven statistical approaches are always based on several assumptions and require tweaking parameters. We interpret the Hi-C frequencies of chromatin interactions in terms of pairwise contact energies, obtaining corresponding energy landscape that represents the structure and interactions in chromatin. The ruggedness of this landscape is explored by the random walk of a travelling probe, which is formalized in the framework of a Markov State Model. The multilevel energy landscape induces metastability in the Markov process, revealing the hierarchy of chromatin structural organization. Structural partitions determined by the basins in the energy landscape are, thus, naturally obtained at different levels of hierarchy without any preliminary assumptions. Effective interactions between partitions are evaluated, providing a blueprint of the whole-genome organization and functional interactions, which is further substantiated by mapping of information on gene expression regulators and different epigenetic factors. The notion of affinity between partitions complements the picture by reflecting the degrees of association between partitions, calling for the modelling of chromatin dynamics and exploring its functional modulation.
| The packing of two meters of DNA in the few-micrometer nucleus results in a structure that performs multiple roles, from forming a structural scaffold of chromatin to facilitating active expression and silencing of genetic material [1, 2]. The beginning of interest in the biophysical characterization of chromatin dates to about 50 years ago, spanning from experimental measurements of DNA persistence length [3–5] and thermal stability [4] to pulling individual DNA-protein (DNP) fibrils by convection flows in solution [6], exploring fibril morphology and stability under different media conditions [7], and exposure to ionizing radiation [5].
Before the chromosome conformation capture (3C) [8] era, the classical view of chromatin organization included several successive levels of packing with archetypal structural patterns, ranging from the compaction of nucleosome-bound 10nm fibers [9] with a roughly 200 base-pair periodicity, to the transient 30nm solenoid (hard to detect in vivo) presumably working in the regulation of gene expression [10, 11], then to the 30-100kbp loops/domains that are apparently instrumental in shaping large-scale chromatin organization and gene expression [1, 12–19]. With the development of the chromosome conformation capture (3C) protocol [8], it has become possible to study chromatin interactions between distant genomic loci. In less than a decade, the original 3C protocol evolved from the analysis of selected pairs of genomic loci to the detection of pairwise interactions between loci and the rest of the genome using chromosome conformation capture on-chip (one-to-all, 4C, [20]), carbon copy (many-to-many, 5C, [21]), and high-throughput 3C (all-to-all, Hi-C, [22]). Finally, improvement of the signal-noise ratio was achieved by performing DNA proximity ligation before nuclear lysis, implemented in in-situ Hi-C [23].
Computational approaches for the analysis of chromatin interaction data developed in recent years can be classified as model-driven or data-driven [24]. Generally, the goal of model-driven studies is to validate physical polymer simulations using Hi-C data. Among them are models of chromatin as a crumpled (fractal) globule [22, 25–27], scenarios of loop formation [28, 29], analyses of the role of epigenetic factors in driving the chromatin organization [30–34], to name a few. In data-driven studies, on the other hand, experimental Hi-C interaction maps are used for extracting information on statistically significant chromatin interactions, for defining topologically associating domains (TADs) and A/B compartments [35, 36], and for the 3D reconstruction of chromatin. Several algorithms have been introduced to study the hierarchical organization of chromatin and its correlation with the distribution of various epigenetic features [37–40], including graph-based approaches for exploring sparse networks of Hi-C interaction peaks, as well as ChIA-PET and HiChIP interaction pairs [41–43]. A recent work by Pancaldi et al. defined chromatin assortativity as a metric for the analysis of correlation between distributions of epigenetic marks and chromatin structure [44]. To date, many methods developed for domain detection [23, 45–47] essentially adopt an image segmentation approach aimed at identifying domain regions as a function of short-range interactions along the chromosome, and domain boundary positions are often highly sensitive to the choice of heuristic tuning parameters [48]. Recent network-based methods incorporate effects of long-range interactions in characterizing structural organization [37, 39, 49] and observe spatial couplings at multiple scales associated with the regulation of gene expression [50]. Spatial reconstructions of chromatin using Hi-C interaction data yield consensus structures [51, 52] or ensembles of possible chromosomal conformations [53, 54], providing an overall picture of chromatin organization [55].
In this work, we propose a new approach for extracting robust genomic partitions from Hi-C data, seeking to capture the footprints of chromatin structure and organization by considering the entire interaction landscape of this complex system. Specifically, our objectives here are to identify and study structural features of chromatin from Hi-C interaction data and to find a connection between these features and data on epigenetic regulation. Introducing a Markov State Model (MSM) with minimal assumptions and parameters on the chromatin interaction network, we aim to identify structural partitions and interactions between them. By analogy with a biomolecule moving and interacting in condensed chromatin, the MSM allows one to explore chromatin structure using a “probe” randomly walking in the contact energy landscape derived from Hi-C data. Given the multiscale nature of the data-derived contact energy landscape and the metastability of the corresponding MSM, we can identify regions of dense intra- and inter-chromosomal interactions, linkers between these regions, as well as the overall topology of individual chromosomes and the complex structures that chromosomes form by interacting with each other. We found that multiple levels of hierarchy exist in the structure of each chromosome with a layer-by-layer splitting of partitions into subunits with distinct structural and epigenetic features, and presumably, distinct roles. The notion of effective interaction between partitions is introduced and shown to be instrumental in uncovering the hierarchical organization, as well as functional dynamics and epigenetic modulation, of individual chromosomes. Looking at the whole-genome picture, the matrix of effective interactions delineates how chromosomal partitions form a major cluster—with several chromosomes linked by significant inter-chromosomal interactions—as a structural scaffold for genome architecture. The notion of affinity between partitions complements the picture of effective interactions by evaluating the degree of association between partitions, which may contribute to the formation of topologically associated domains, transcription factories and other functional elements, thereby organizing the regulation of genome expression.
In this paper, we propose a novel computational method for exploiting Hi-C data in the study of chromatin organization. Since Hi-C reads represent interactions between pairs of loci, it is natural to consider Hi-C data as an undirected network of contacts between genomic loci, which, as a highly complex system at a resolution of 50kbp, contains tens of thousands of nodes at the whole-genome level. In the following, we first provide the motivations for adopting a Markov State Model approach for the analysis of Hi-C data, then introduce a toy model of a chromosome that serves to elucidate the most important notions associated with the method. Finally, a specific description of the major steps in the proposed Markov State Model approach is presented for the case of a single chromosome (human chromosome 17), followed by a genome-wide analysis.
A common strategy to study complex network data structures is to combine them with a discrete state Markov process, commonly called a Markov State Model (MSM), with the goal of characterizing hidden network properties [56–58]. MSMs enable one to systematically explore network structure via random walks, where traveling probes form virtual trajectories through the whole network by connecting pairs of nodes. It has been shown that studying the spectral and metastability properties of the network-associated MSM allows one to obtain a reduced description of the underlying complex data.
In order to illustrate how MSMs can be used for studying Hi-C data, we introduce here a toy model of a chromosome. Let us consider a linear system characterized by a discrete set of loci S = {1,…,N}, with N = 500. We assume that the number of loci N determines the maximal resolution of this relatively large system. Each locus of the system is associated with an energy Ei, which is linked to the intrinsic stability of the locus i at the given resolution. For the sake of argument, we assume the intrinsic stabilities Ei to follow a hierarchically shaped energy profile (Fig 1A). The energy profile considered here contains 18 wells separated by barriers ranging from 0.5 to 2 energy units. On the first level of hierarchy, there are two basins separated by a barrier of 2 energy units (black diamonds in Fig 1A), each divided into three sub-basins (indicated as red circles in Fig 1A), which in turn are split into three basins on the third level of hierarchy (black circles in Fig 1A).
A traveling probe in such an energy profile is assumed to make two types of moves: sliding between adjacent loci and hopping between non-adjacent ones. We do not make any assumption about the three-dimensional structure of the system and assume a power law contact probability between non-adjacent loci, namely (d0/dij)α, where d0 and dij are the distance between adjacent and any non-adjacent loci respectively, which is equivalent to the genomic distance between loci i and j and such that dij = d0 for adjacent loci. Thus, for each pair of loci i and j, we define the contact energy landscape Eij = (Ei + Ej)/2 − αln d0/dij with α = 1.5. Assumptions on the power law dependence and the value of exponent α are made on the basis of empirical observations on Hi-C data and polymer models of chromosomes [22]. The contact energy landscape is represented in Fig 1B.
To construct the MSM describing the motion of a probe, we define the corresponding Markov generator L for transitions between loci i → j by the Laplacian Lij=e−β(Ej−Ei)/2eβln(d0/dij)α and Lii = −∑j≠iLij, with transition matrix pij = Lij/∑j≠iLij,pii = 0(∑jpij = 1), flux πij = Lijμj, where steady state probabilities are given by μi=e−βEi/∑je−βEj, and β is an inverse temperature parameter. A network of nodes (loci) and edges (contacts) is obtained from the matrix of fluxes πij, which represents the symmetric probability of contact between a pair of loci. With the set of rules given by the above Markov generator, a probe will tend to explore regions of the network in the neighborhood of the loci that are more stable, i.e., within an energy well, and will rarely connect loci in different energy wells. This property relates to the “metastability” of the corresponding Markov process. Specifically, in a metastable MSM only a few nodes function as “hubs” of the network, which means that the probe tends to spend most of the time in the neighborhood of these hub nodes, instead of anywhere else. In other words, a probe departing from a generic node in the network is likely to hit the closest hub node in the set hub-nodes M. Additionally, the probability for a probe departing from a hub-node in M to return to itself is larger than that for the probe to reach another hub-node in M. As a result, nodes in the neighborhood of hubs tend to cluster together in a modular manner. This is a condition that allows one to find a reduced size MSM that approximates the original Markov process associated with the initial network. One can quantify how well the probe motion satisfies this condition by defining a metastability index ρM. A metastability index is the ratio of two probabilities (see Eq 6 in Methods for a precise definition): the probability Pout for a probe to connect two different hubs in the set M (as small as possible) over the probability Pin for a walker to hit any hub in the set M irrespective of the starting point (as large as possible) [59]. In a metastable MSM the metastability index is expected to be a small number (ρM=Pout/Pin<1) characteristic of the hub set M.
To understand how metastability works, it is instructive to consider the large changes in kinetic properties of the MSM upon increase of the inverse temperature parameter β (annealing condition). These changes are clearly illustrated by the mean first passage time MFPT τij, which is the average time (number of steps) a probe takes to connect the pair of states i and j (where states represent loci of the toy chromosome, see Eq 4 in Methods). Fig 1C shows the MFPT matrices in the case of low β = 1 and high β = 10, respectively. A clear separation of time scales emerges upon increasing 𝛽, as reflected in the partitioning of the MFPT matrices. The nested squares emerging in the MFPT matrix (Fig 1C) at high β identify pairs of states/loci (i,j) with comparable values of the MFPT τij, which is a result of the hierarchically shaped energy profile. As β is increased, the emerging separation of time scales in the MFPT matrix is the result of the dominant barrier that separates a given pair of loci in the energy landscape (see Fig 1A): each of the separated regions contains one or more hubs that cause probes to stay within its vicinity. As a result, the effect of high β on the MFPTs of the MSM elucidates how dominant interactions in the system can be captured using just a subset of loci, the hub set M.
To quantitatively identify the hub set, an optimization procedure is performed in order to find the sets M that minimize the metastability index ρM (see details in Materials and Methods) as a function of increasing β. Fig 1D shows the optimized profile of the index ρM as a function of the hub set sizes, and at different values of β. All the profiles of ρM clearly show three minima corresponding to the hub sets M(2),M(6), and M(18) (of sizes 2, 6 and 18, respectively), which correctly identify locations of the energy wells in the hierarchically shaped energy landscape in Fig 1A. The hub sets obtained by optimizing the index ρM are suitable as cores of partitions, which characterize the coarse-grained state space of an approximated MSM. Fig 1E (top) depicts the toy network associated with the contact energy landscape shown in Fig 1B. Nodes are colored according to the partitions constructed around nodes in the hub set M(6). A reduced network corresponding to the hub set M(6) is also shown in Fig 1E (bottom). The nodes in this network are defined as soft partitions of the initial set of loci S, whereas the links characterize the “effective interactions” between nodes with values Fab = ∑i∈Sqa(i)πib (see Eq 11 in Materials and Methods). The quantity qa(i) is a committor probability, which is the probability for a probe departing from a locus i to hit the locus a∈M before any other locus in the hub set M (see Eq 8 in Materials and Methods).
Using the intuition acquired with the help of this toy model, we describe in the following section how a MSM can be constructed from the Hi-C dataset of a single chromosome and how metastability analysis can be performed in order to infer chromosomal architecture and effective interactions between partitions.
We now consider the random walk through the interaction network of a single chromosome, using the example of Hi-C data on human chromosome 17 in the human B lymphoblastoid cell line GM12878 at 50kbp resolution [23] and describing it via a Markov process. To do that, we start from the number of times fij a pair of genomic loci i and j is found in a contact. After applying a Gaussian smoothing filter on the raw data (see Hi-C data preprocessing in Materials and Methods), a pairwise contact energy Eij = −lnfij is defined for each pair of genomic loci. With this interpretation, the larger the contact frequency the more stable (lower contact energy) pair of genomic loci is involved. The representation of this two-dimensional contact energy landscape is shown in Fig 2A.
A probe moving in such a landscape is expected to spend most of the time in pairs characterized with low contact energy and rarely connecting across high contact energy pairs. In the toy model presented in the previous section, a pairwise contact energy landscape (Fig 1B) was constructed from the one-dimensional energy landscape (Fig 1A). Here, we use a reverse logic and consider the one-dimensional projection (Fig 2B) of the two-dimensional contact energy landscape (Fig 2A). To do that we define the contact energy of a genomic locus i as Ei = −lnfi, where fi = ∑jfij is the total number of times a genomic locus i is found in any contact, hence loci involved in more contacts are more stable as they exhibit lower contact energy. Fig 2B shows the 1D projection of the pairwise contact energy landscape (for both raw and Gaussian-smoothed data), which presents multiple features—minima, maxima, and barriers—characterizing the architecture of the chromosome.
Here, we briefly describe the metastability analysis applied to chromosome 17 (steps 1–4) and consider whole-genome interactions (step 5) using a coarse-grained approximation.
Step 1. In order to explore chromosomal architecture, the MSM describing the motion of a probe in the contact energy landscape Eij is implemented by introducing the Maxwell-Boltzmann probability πij(β)=e−βEij/Z(β), where Z(β)=∑(i,j)e−βEij is the partition function, β is the inverse temperature parameter, and πij is the symmetric flux of probes between pairs of loci. Using a 50kbp resolution for the Hi-C dataset, a total of N = 1625 genomic loci comprise the state space S of the MSM for chromosome 17. The transition matrix associated with the MSM is defined as pij = πij/μi, where μi(β) = ∑jπij(β) is the Boltzmann weighted probability (steady state probability distribution) of observing locus i involved in any contact. The effect of annealing (increasing the inverse temperature β) on the kinetics of a random walker is clearly reflected in the MFPT matrices (Fig 2C), obtained at low and high β, respectively. While at low β (β = 1) MFPTs show no partitioning, a separation of time scales becomes evident at high β (β = 9). Indeed, the 1D projection of the pairwise pseudo-energy landscape (Fig 2B) shows that, apart from the centromere that naturally separates the two chromosome arms, the highest barrier in the 1D projection is about 1.5 in β−1 units (see Fig 2B). Therefore, partitioning of MFPTs scales is observed only for significantly higher values of β.
Step 2. Optimization of the metastability index ρM (see details in Materials and Methods) over the hub set M of different sizes was performed as a function of the inverse temperature parameter β, revealing the levels of structural hierarchy of chromosome 17. The ρM profiles upon increasing β (Fig 2D) converge towards five minima, which correspond to the hub sets M(2),M(5),M(8),M(12), and M(27), of sizes 2, 5, 8, 12, and 27, respectively. The M(2) hub set is not considered as it trivially identifies the chromosome arms separated by the centromere. It should be noted that the locations of the obtained hub sets correspond to the locations of the multiple wells present in the projected contact energy landscape, as shown in Fig 2B.
Step 3. Given the hub sets obtained at different levels of structural hierarchy, one can identify chromosomal partitions, namely regions of the chromosome compacted around corresponding hubs and, at the same time, separated from one another. Soft partitions are defined around corresponding hubs using the committor probability qa(i) [59], which in this case is interpreted as the probability for a locus i to belong to the partition defined by the hub a∈M. To identify physical partitions of the chromosome in relation to other chromosomes, a coarse-grained description is adopted here by considering hard partitions. In this case, a step function θA(i) characterizes whether a locus i belongs to a partition A, specifically θA(i) = 1 if i ∈ A, θA(i) = 0 otherwise, and ∑AθA(i) = 1 for any locus i (see Eq 9 in Materials and Methods). Fig 2E illustrates the partitioning of the network for human chromosome 17 that is obtained from the hub set M(12).
Step 4. To complete the description of chromosome structure, one needs also to characterize the strength of interactions between the partitions obtained at different levels of hierarchy. As in the example illustrated in the toy model, we consider the effective interaction between two soft partitions located around the hub loci a and b of a chromosome as the mean contact energy acting between them, which corresponds to the weighted flux connecting loci a and b via the committor probability qa(i), namely Fab = ∑i∈cqa(i)πib (see Eq 11 in Methods).
Step 5. In the context of whole-genome interactions, a coarse-grained description is adopted (see Step 3) for estimating the mean contact energy between pairs of partitions in the 23 chromosomes: FAB = ∑i∈gθA(i)∑j∈gπijθB(j), where θA(i) and θB(i) are step functions and πij is the flux of probes between corresponding loci (see Materials and Methods for details).
Fig 3A–3C show the partitioning of chromosome 17 at three levels of hierarchy with corresponding effective interaction matrices (Fig 3D–3F), and the band representation of partitions at all three levels (M(5),M(12),M(27); Fig 3G). The major partition boundaries that emerge at the first level of hierarchy persist through the other levels (Fig 3G; similar for all chromosomes, see S1 Fig) and show a qualitative agreement with the borders of euchromatic and heterochromatic bands obtained from Giemsa staining (Fig 3G and S1 Fig). Unfortunately, as Giemsa staining is a very basic and crude cytological method for identifying densely-packed (heterochromatic, dark stain) and low-density (euchromatic, light stain) genomic regions, it is not possible to perform an accurate quantitative analysis on staining bands [61, 62].
At the lowest level of hierarchy (Fig 3A), we observed the bulk topology of the chromosome where two chromosomal arms are brought together via strong interactions between partitions 1, 4, and 5. Most partitions at the lowest levels of hierarchy are found to contain both euchromatic and heterochromatic bands, and they have highly distinct structural and/or functional characteristics. At the second level of hierarchy (Fig 3B), partitions 1.2, 4.1, 4.2, 5.1, and 5.4 form several non-adjacent contacts, working as hubs responsible for most of the network structure. The third level of hierarchy (Fig 3C) yields further details of chromosomal architecture: the p-arm is loosely connected and is weakly centered on 1.2.2 and 1.2.4, while the q-arm is densely connected by multiple hubs (4.1.1, 4.2.2, 4.3.1, 5.1.2, and 5.4.1). At this level many partitions are homogeneous, either eu- or heterochromatic, interacting more strongly with partitions with similar packing densities, resembling the phenomenology of the so-called A/B (active/inactive) chromatin compartments [22]. For instance, partition 1.2 is split into mostly euchromatic (1.2.2, 1.2.3) and heterochromatic (1.2.1, 1.2.4) partitions, while partition 4.1 is split into predominantly euchromatic (4.1.1, 4.1.2, 4.1.3) and heterochromatic (4.1.4, 4.1.5) ones. Partition 4.1.1 is the largest among these, forming significant interactions with the p-arm through partition 1.2.2. Another noticeable interaction between chromosomal arms occurs via the partition 4.3.1, which links heterochromatic partitions 5.1.1–2 and 1.2.4. Interestingly, the mostly euchromatic partition 1.2.2 is responsible for many non-adjacent contacts with the q-arm, whereas heterochromatic 1.2.4 forms non-adjacent contacts only with 4.3.1 and 3.1.1. With these observations, one may conclude that heterochromatic partition 1.2.4 acts as a structural foundation that link the mostly euchromatic partitions 1.2.2, 1.2.3, 2.1.1, 2.1.2, and 3.1.1.
To investigate how the hierarchical organization of chromosomes facilitates their function, we first analyzed the average density of various epigenetic factors in partitions (Fig 4 and S3 Fig), using chromosome 17 as an illustration for this analysis and operating at the third level of structural hierarchy. Fig 4A, in which node sizes depict partition sizes, shows that heterochromatic partitions 5.1.1 and 5.1.2 apparently form a structural foundation of chromosome 17 architecture, linking the p- and q-arms through the large mixed partition 4.3.1 and the heterochromatic partition 1.2.4. Next, we consider two transcription factors commonly associated with chromatin structure studies, namely CTCF (transcriptional repressor, Fig 4B) and RAD21 (cohesin, S3H Fig). The CTCF graph (Fig 4B) shows that the heterochromatic partitions (1.2.4, 5.1.1, 5.1.3, and 5.2.2) and the pericentromeric partition 3.1.1 have the lowest CTCF levels, while the highest CTCF levels were found on 4.2.2, 4.2.1, and 5.4.3. The euchromatic or mostly euchromatic partitions 1.1.1, 1.2.2, 2.1.2, 2.2.1, 4.2.3, 5.4.1, and 5.4.2 show average levels of CTCF in the overall eight-fold variation in the density of this transcription factor across partitions. Among the hub partitions, namely those that form extensive non-adjacent contacts, only 4.2.2 shows high CTCF levels. Unlike CTCF, RAD21 (a component of cohesin) exhibits only a two-fold variation in densities across partitions at this level of hierarchy. The correlation between CTCF and cohesin binding sites has been noted previously [63, 64], and indeed the distribution of RAD21 (S3H Fig) chiefly follows the same general trends as that of CTCF. The strongest among the few exceptions are the increased density of RAD21 in 4.2.3 and decreased density in 5.4.3.
Turning to histone modifications, we note that H3K9ac (Fig 4C) and H3K9me3 (S3A Fig) are associated with activation and silencing of transcription in corresponding promoter regions and, therefore, are expected to show opposite density trends. Indeed, densely packed heterochromatic partitions (1.2.4, 5.1.1, 5.1.3) and pericentromeric 3.1.1 show very low levels of the activating H3K9ac histone modification, while the silencing H3K9me3 modification shows increased density in these partitions (highest in the case of 3.1.1). At the same time, euchromatic and mostly euchromatic partitions 1.1.1, 1.2.2, 2.1.2, 4.1.2, 4.2.2, 5.4.2, and 5.4.3 show an increased density of both epigenetic factors, with some slight variations. The opposing trends are observed in heterochromatic partitions for the activating H3K27ac (decreased density, S3B Fig) and inhibiting H3K27me3 (increased density, S3C Fig) modifications, with the most pronounced effects being on 1.2.1, 4.1.4, 4.1.5, 5.1.1, and 5.1.3. Distributions of the H3K4me1 (S3D Fig) and H3K4me3 (S3E Fig) modifications—both activators—show higher densities in most euchromatic partitions, and in few heterochromatic ones—1.2.1, 4.2.1, and 5.1.2. Interestingly, the heterochromatic partitions 1.2.1, 4.2.1, and 5.1.2 are enriched in all activating histone modifications considered here (H3K4me1, H3K4me3, H3K9ac, H3K27ac), and, at the same time, are depleted in the inhibiting modifications H3K9me3, H3K27me3. These trends suggest that the above partitions may contain facultative heterochromatin that switches between active and repressed states.
Overall, the DNA accessibility graph, indicating the DNase-Seq signal (Fig 4D), shows that most of the euchromatic partitions (1.1.1, 1.2.2, 2.1.2, 4.2.2, 5.4.2, and 5.4.3) are rather open and accessible for contacts or interactions. Increased accessibility observed for partitions 4.2.1 and 1.2.1 is consistent with the conclusion that these partitions may contain facultative heterochromatin, which was inferred from the distribution of activating and inhibiting histone modifications. The partition 5.1.2, on the contrary, is less accessible, suggesting that it contributes mostly to the structure formation. Finally, the distributions of RNA polymerases II and III (S3F and S3G Fig) complement the picture of the potential functional involvement of different partitions in chromosome 17. RNA polymerase II (POL2), crucial component of mRNA synthesis, is distributed quite evenly in both euchromatic and heterochromatic partitions (except the high level in 2.2.1). The synthesis of tRNA, 5S rRNA, and small RNAs through the action of RNA Polymerase III (POL3) is distributed in a more specific way across different partitions. The POL3 signal is high in euchromatic partitions 1.2.2, 4.1.2, 4.2.3, in mixed 4.2.1 and 5.3.1, as well as in some heterochromatic ones (4.1.4, 4.1.5, 5.1.2, 5.1.3, and 5.2.2).
Peculiarities in distributions of epigenetic factors, DNA accessibility, and RNA polymerases revealed in the analysis of individual chromosomes should be further considered in the framework of whole-genome organization, exploring the interplay between intra- and inter-chromosomal interactions in the regulation of gene expression. To this end, we moved from single-chromosome analysis to studying the whole-genome effective interaction matrix. Given that chromosomes are spatially segregated into chromosomal territories (CTs), one can approximate the whole-genome organization by merging single-chromosome partitioning schemes at appropriate levels. Using a selected representative level from each chromosome (see Materials and Methods: Chromosome partitioning), we formed a whole-genome description with 539 partitions, with an average partition size of about 5Mbp (S2 Table).
The matrix of effective interactions between chromosomal partitions (Fig 5) provides a general view of the overall physical interactions in chromatin. It shows that chromosome 1 and small chromosomes (14–20 and 22) massively interact with others, while chromosomes 4, 5, 9, 21, and X appear to be relatively isolated from the rest of the genome (Figs 5 and 6). Several partitions form consistently stronger intra- and inter-chromosomal interactions with other partitions. We classified interaction strengths into 5 layers with equally-spaced threshold values: the scaffold layer is the strongest, followed by layers 1, 2, etc. (see also Materials and Methods for the definition of the interaction strength at different layers).
Fig 6A shows the major cluster in the whole-genome partition set: partitions from different chromosomes form tight sub-clusters highlighted by color and marked by chromosome labels. All displayed partitions are linked by the two strongest layers of interactions (scaffold interactions are represented by black edges, and layer 1 by grey edges).
It is easy to see that most of the intra-chromosomal contacts and some inter-chromosomal interactions are established on the scaffold layer, giving rise to a structural foundation for genome-wide architecture (Fig 6). Specifically, chromosomes 1, 14, 16, 17, 19, 20, and 22 are densely interconnected, while other chromosomes in the major cluster are linked to them via only a few interactions. Partitions 1–2.1.2, 14–3.4, and 22–6, for example, act as contact hubs between these massively interacting chromosomes and others. On the other hand, partitions such as 3–2.2.2, 8–6.2.1, 10–3.4.1, connect less-strongly interacting chromosomes to the strongly interacting ones (see S3 Table for interaction strength layers for these interactions between chromosomes). Notably, chromosomes 1 and 2, the two largest ones (about 250Mbp each), behave differently in the context of the whole-genome interactions. While chromosome 1 serves as a hub in the interactions between the highly- and less-interacting chromosomes, chromosome 2 does not show many interactions with other chromosomes (Fig 6A). Turning to functional regulation, most of the partitions involved in significant inter-chromosomal interactions exhibit higher densities of several epigenetic factors, such as CTCF (Fig 6B), H3K9ac (S5A Fig), and DNase accessibility (S5B Fig). These partitions may participate in the formation of active epigenetic compartments facilitated by the structural role of CTCF [65]. Active processing of genomic information taking place in these structures is regulated by the activating histone modifications (H3K9ac) and transcriptional repressors (CTCF). The opposite trend is observed for partition 14–3.4, which is coupled with a higher density of the silencing H3K9me3 histone modification. Therefore, partition 14–3.4 and its interactions with partitions in other chromosomes, for instance 10–3.4.1, with low activating factor densities may indicate the formation of dense structural heterochromatin and/or silencing facilitated by Polycomb bodies [1, 66].
To evaluate how the distribution of epigenetic signals may be associated with interaction between partitions, we calculated correlations between effective interaction strengths and the expected enrichment of factor densities across partition pairs that are mostly euchromatic (EC) or heterochromatic (HC). The enrichment of factor densities is estimated here as the product of factor densities per partition (S12 Fig). To obtain the strongest signals, we limited our consideration to interactions between partitions that are dominated by either eu- or heterochromatin (see legend for S12 Fig for the definition of EC and HC partitions): EC-EC pairs (S12A Fig), HC-HC pairs (S12B Fig), and EC-HC pairs (S12C Fig). Despite the relatively weak correlations, general trends appear to be quite clear, with the strongest ones seen between euchromatic (EC-EC) partitions (S12A Fig). Transcription factors CTCF and RAD21 are always positively correlated, as well as POL2 in EC-EC (S12A Fig) and EC-HC (S12C Fig) pairs, whereas POL3 shows no correlation. The positive correlation for CTCF and RAD21 with effective interaction strength agrees with current literature on the role of CTCF and cohesin in mediating chromatin structure through looping interactions [65, 67, 68]. Stronger interactions between EC partitions appear to be linked to higher transcriptional activity, as suggested by the positive correlation with active histone modifications and POL2. Absence of correlation for HC-HC pairs in the case of POL2 can be related to the fact that transcriptional activity is suppressed in heterochromatin. Potential active involvement of interacting euchromatic partitions in the formation of transcription factories is corroborated by the most pronounced correlation observed for DNA accessibility in pairs of euchromatic partitions (S12A Fig). Activating histone modifications, except for H3K4me3, show positive correlations in all types of interacting partition pairs. Interestingly, silencing histone modifications appear also to be weakly correlated with effective interactions between partitions.
The original partitioning analysis was performed on the GM12878_primary (B lymphoblastoid) Hi-C dataset by Rao et al. [23] (GEO accession GSE63525). We also applied our model to four other datasets: GM12878_replicate (a biological replicate of GM12878_primary dataset), IMR90 (lung fibroblast), HUVEC (umbilical vein endothelial cells), and HMEC (mammary epithelial cells). Our goal in this analysis was two-fold: (i) to benchmark robustness and reproducibility of the method using the replicate dataset; (ii) to examine the sensitivity of the method in detecting alterations in chromatin organization in different cell lines, associated with corresponding genome functional states and gene expression levels. S13 Fig shows side-by-side comparisons of the partitioning schemes for GM12878_primary and the other datasets, and S6 Table shows some indicative statistics comparing the results from each case. First, we observed a high consistency between the biological replicates of GM12878: S13A Fig shows that the partitioning was highly consistent between the two sets of Hi-C data, with partition boundaries being identical in most cases, resulting in the high Rescaled Mutual Information (RMI) of 0.70 (see Chromosome partitioning in Materials and Methods for definition of RMI). The composition of the major cluster was also largely identical. Comparing the results from other cell lines, we observed significant differences: IMR90, HUVEC and HMEC cells each had significantly shifted partition boundaries compared to GM12878_primary, leading to lower RMI values of 0.39 to 0.48. The major-cluster structures in these cell lines are also significantly different (see S16 Fig for IMR90 and HUVEC), especially that of HMEC, where no strong inter-chromosomal interactions were observed between partitions, and the chromosomes remained isolated in the whole-genome network. Notably, in both IMR90 and HUVEC, a large partition on chromosome 9 forms extensive inter-chromosomal interactions: the overlapping region (chr9:1268000000–1412500000) contains two genes (OLFM1 and MVB12B) with the RNA-expression profiles different from that of GM12878. The MVB12B (a component of endocytic protein system [69]) gene is activated in both IMR90 (lung fibroblast) and HUVEC (umbilical vein endothelial cells) cell lines, and OLFM1 (lung cancer marker [70]) in IMR90, while both genes are silenced in GM12878. These preliminary observations call for future in-depth investigation of the structural basis, functional mechanisms, and specifics of epigenetic regulation behind the observed differences between cell types.
While effective interactions between partitions characterize the overall architecture of genome organization, it may not fully discriminate functionally relevant interactions between chromosomes and their parts. Indeed, most partitions are presumably in constant motion within the nucleus, and as Hi-C experiments are typically conducted on unsynchronized cell populations, effective interactions capture the average contact probability arising from both random diffusion and specific transient interactions. Therefore, in addition to effective interactions, the affinity between partitions was also calculated, which reflects how the observed interaction frequency differs from the expected frequency (from random diffusion), because of possible associations between partitions. Defined as the ratio between observed and expected contact probabilities between pairs of partitions (see Eq 15 in Methods), the affinity is indicative of the degree of association between partitions, and high affinity values may serve as a manifestation of biologically-relevant contacts. Fig 7 contains the whole-genome matrix of pairwise affinities (blue: high affinity, white: low affinity) between corresponding partitions. Like the observations in the whole-genome effective interaction matrix (Fig 5), the largest chromosomes 1 and 2 exhibit different behavior, with chromosome 1 containing partitions with high affinity to those in several other chromosomes (especially with chromosomes 14–22) and chromosome 2 generally showing low affinity to partitions in other chromosomes. Smaller chromosomes 14–22 form more, presumably functional, contacts with each other, compared to other chromosomes. At the same time, the number of partition pairs with high affinity is much lower than number of pairs with significant effective interactions (compare Fig 5 and Fig 7). In total, we observed 687 high-affinity pairs (S4 Table), which are seemingly crucial for whole-genome structural organization and function.
Interestingly, several large partition pairs (>2Mb) with high effective interactions and affinity are located in the telomeric regions of corresponding chromosomes (yellow cells in S5A Table), having moderately high densities of epigenetic/transcription factors and increased DNA accessibility (S5A Table). Two other groups of partitions with high affinities are characterized by smaller partition sizes and highly elevated concentrations of various transcription factors and epigenetic modifications (S5B Table): (i) pericentromeric partitions (red cells in Table) show high concentrations of activating (H3K4me3) and silencing (H3K27me3 and H3K9me3) histone modifications and high levels of POL3 and RAD21; (ii) telomeric partitions (yellow cells in Table) show strongly increased concentrations of all activating histone modifications, POL2, and CTCF, as well as high DNA accessibility. This separation between types of activating histone modifications, transcription factors, and DNA accessibility in centromeric and telomeric regions signals a specificity of functional interactions between partitions with high affinities to each other. Examples of partitions involved in high-affinity interactions and characterized by the over-representation of different epigenetic factors and modifications are collected in Fig 8 and S6 Fig, where high-affinity clusters of partitions enriched in these epigenetic marks are plotted.
A comparison of the inter-chromosomal interactions in the major cluster of effective interactions (Fig 6) with interactions in affinity clusters (Fig 8 and S6 Fig) highlights several relatively-small partitions, e.g. 9–5.6.3, 9–5.6.5, and 11–5.1.3, that act as junctures between different chromosomes. These partitions yield increased density of CTCF along with other juncture-partitions (3–2.2.2, 6–1.4.3, 8–6.2.1 to name a few), pointing to the potential importance of these partitions in whole-genome structural organization. This inference is further supported by multiple interactions detected for partitions 8–6.2.1, 9–5.6.5, and 11–5.1.3 in the CTCF affinity graph (S6D Fig). Additionally, H3K9ac (Fig 8A) and H3K27ac (Fig 8C) affinity graphs hint at the functional importance of some of these partitions: the central part of the H3K9ac graph is formed by partitions 9–5.6.3, 9–5.6.5, and 11–5.1.3, while 9–5.6.3 and 11–5.1.3 are also present in the H3K27ac graph.
Focusing on individual epigenetic factors, the activating H3K4me3 mark links more partitions than the activating H3K4me1 histone modification. The silencing H3K9me3 histone modification functionally links many centromeric partitions, whereas the activating H3K9ac modification works in both mostly euchromatic and mixed euchromatic/weakly-heterochromatic non-centromeric regions. Similarly, the activating H3K27ac modification affects mostly non-centromeric partitions, unlike the very active silencing H3K27me3, for example, in partitions 1–4.11.1, 2–3.1.1, and 10–2.1.1 (Fig 8). These partitions are also characterized by the high levels of POL3 (S6B Fig) and RAD21 (cohesin, S6C Fig), whereas the insulator CTCF links several euchromatic partitions across different chromosomes (S6D Fig).
It is evident that centromeric partitions 1–4.11.1, 2–3.1.1, and 10–2.1.1 are enriched with almost all regulatory factors (see Fig 8 and S6 Fig), yielding high affinities to other partitions and pointing to important functional interactions and intense regulation taking place in these partitions. Interestingly, while activating histone marks (Fig 8A, 8C, 8E and 8F) are dominant in several euchromatic partitions, these marks are also present in partitions containing large sections of heterochromatin and centromeres, which are commonly associated with dense packing and transcriptional repression. Similarly, silencing histone marks (Fig 8B and 8D) are dominant not only in heterochromatic and centromeric partitions, but also in some partitions that are mostly euchromatic. Furthermore, dominating regions for the transcription factors CTCF and cohesin (S6C and S6D Fig) appear to have significant overlap with activating and silencing histone marks, respectively. These overlaps show the complexity of functional interactions in chromatin, even at the coarse-grained level of partitions: opposing factors are found acting in the same regions, allowing for switching between transcriptional states in response to other biochemical cues.
We proposed here a computational framework for exploring chromatin organization based on Markov State Modelling of chromatin interactions. Given the multilevel hierarchical packing of chromatin, we introduced a reduced description of the complex network of chromatin interactions and its organization via interactions between structural units at different levels of hierarchy. By interpreting Hi-C data as a pairwise contact energy landscape, a Markov State Model approach was used to explore the chromatin interaction network through the random walk of a probe. While steady-state distributions obtained from the Markov process of randomly-moving molecules can serve as a measure of the chromatin accessibility for epigenetic factors [71], taken alone they describe neither the genome architecture, nor structural and functional interactions between genome partitions and regulatory factors. In this work, analysis of the Markov State Model under thermal annealing shows the key role played by the ruggedness of the contact energy landscape in shaping chromosome structural organization. Specifically, metastability analysis of the Markov State Model associated with the chromatin interaction network allowed us to identify levels of structural hierarchy and to observe structural units—partitions of different scales. These structural partitions serve as a coarse-grained description of chromosomes, which form the basis for introducing the notion of intra- and inter-chromosomal network of effective interactions. The analysis of effective interaction networks across levels of hierarchy in individual chromosomes shows that chromosomes adopt highly varied topologies. While the lower levels reveal an overall architecture of the folded chromosome, the higher levels can provide structural details in relation to functional organization and regulation of gene expression.
Biological insight on the structural organization of chromosomes can be obtained with our method by considering peculiarities in the distributions of transcription and epigenetic factors in eu- and heterochromatic partitions in relation to interactions between them. Partitions at the highest levels of hierarchy may be seen as analogous to TADs, or to the so-called A/B (active/inactive) epigenomic compartments [72]. In the future, with the development of common standards and benchmarks by the community, it would be important to compare results and insight obtained from various genomic segmentation approaches. In this work, however, we based our analysis on the obtained sets of partitions, showing how studying distributions of activating and silencing histone modifications in these partitions can help to understand the role of structural organization of chromosomes in the regulation of gene expression.
Shifting our focus from structural analysis of individual chromosomes and the functional involvement of partitions to whole-genome architecture, we considered the set of 539 partitions obtained at high levels of hierarchy in corresponding chromosomes with effective interactions between them, which were obtained by adopting a fast “mean field” approximation. In the context of the “partition space”, an analysis of genome-wide effective interactions provides a blueprint of inter-chromosomal contacts, showing that despite the strong crowding of partitions in chromosome territories, most chromosomes are significantly connected with each other, giving rise to a bulky cluster in the core of the effective interaction network. The strongest interactions were observed between chromosomes 14–22, which are characterized by small chromosome sizes. Heterochromatic partitions are apparently mostly involved in the formation of chromosomal territories, interacting within the corresponding chromosomes and providing structural integrity, and showing only low levels of activating factors. On the other hand, most of the inter-chromosomal juncture partitions, while relatively small in size, are enriched with CTCF, H3K9ac, and DNase-Seq, which may lead one to conclude that these partitions are involved in the formation of inter-chromosomal active epigenomic compartments [72]. Correlations of effective interactions between partitions with distributions of epigenetic factors in these partitions show that: (i) most active regulation apparently takes place in pairs of interacting euchromatic partitions; (ii) DNA accessibility, CTCF and activating histone modifications H3K4me1, H3K9ac, and H3K27ac are major potential contributors in the regulation of genome function. Additional biological insight was obtained by determining partitions that may form transient function-related contacts, thereby triggering alternate chromatin states. To this end, the affinity measure was introduced here to evaluate the level of association between partitions, so as to identify partitions with functionally-related interactions. Irrespective of the effective interaction strength, high affinity between partitions point to the mutual functional involvement of corresponding partitions. Since different factors are likely to play dominant roles in different genomic regions, our affinity analysis is complementary to the concept of chromatin assortativity introduced by Pancaldi et al. [44], which may identify epigenetic factors associated with multiple high-affinity communities across the whole genome.
There are different challenges in extending the original analyses of the Hi-C data to exploring the structure-function relationships in the genome. Several previous studies using a hierarchical clustering approach for the analysis of Hi-C data are based on the a priori assumption of the existence of structural hierarchy in chromatin [38–40]. While the work by Boulos et al. is free from such an assumption [37], it employs a tunable scale parameter in establishing the hierarchy. Our approach is based on an energy landscape representation of the chromatin interaction network, which is formalized and explored via a Markov State Model. Metastability analysis of the Markov State Model allows one to detect natural levels of hierarchy in chromatin structure. The novelty of our approach does not free it, however, from certain limitations. For example, the “mean-field” approximation for coarse-graining of whole-genome interactions, rooted in the computational challenge in calculating committor probabilities on very large networks, is currently a necessary step for processing massive whole-genome datasets. On the other hand, because of its rapid computational time, this approximation allows us to explore higher resolution Hi-C datasets and to obtain partitions of smaller sizes and, ultimately, uncovering higher levels of hierarchy. Further, as the rugged morphology of the energy landscape is key to determining chromosomal partitions, the effect of noise associated with the Hi-C data is of critical importance. For example, all contact frequencies and derived quantities are affected by the characteristic noise, since most Hi-C experiments are conducted on unsynchronized cell populations. While a possible solution would be to study single-cell Hi-C [73] data, which shows great promise in capturing differences between transient states in chromatin organization, the current protocol yields too few interaction pairs for a meaningful analysis of the interaction network. Although specialized variants of the Hi-C protocol, such as capture Hi-C (cHi-C) [74], do not provide a full view of the physical organization of chromatin, they can nonetheless be useful in targeting specific subsets of genomic loci, such as promoters. The biological implications of our analysis, particularly on the relationship between factor enrichment and effective interaction strength and affinity, may also be strengthened by incorporating additional experimental approaches, such as ChIA-PET [75] and HiChIP [76], which identify interacting genomic elements that are concurrently associated with specific binding proteins. Genome architecture mapping (GAM), a newly devised experimental protocol that determines the frequency at which genomic loci lie on the same spatial plane by sequencing fragments isolated in cryosections of the nucleus [77], can be a great source of constraints for future 3D whole-genome reconstruction.
To conclude, there is no doubt that scientific interest in chromatin structure will continue to drive the development of a variety of specialized experiments and computational approaches in the field of 3D genomics. The method presented here, aimed at detecting and characterizing the hierarchical organization of chromatin, is a step towards unravelling causal relationships in chromatin structure and dynamics of function-related transient molecular phenotypes. The great potential of new experimental data combined with constant methodological improvement are critical in the quest for a more detailed understanding of chromatin architecture, 3D reconstruction, dynamics, and epigenetic regulation.
No human or animal subjects and/or tissue were used in the work.
Ethics rules of the Bioinformatics Institute, A*STAR were followed during the work on the project and preparation of the paper.
In the following, a Markov jump process is introduced to describe a random walk in the chromatin interaction network, where a probe connects pairs of interacting genomic loci, which represent the states of the Markov State Model (MSM). We first focus on a single chromosome c and denote the corresponding matrix element of Hi-C counts fij for a pair of loci (i,j). We define a pairwise interaction pseudo-energy Eij = −lnfij, which characterizes a strength of interaction between a pair of loci (i,j): the higher the observed counts the more stable the corresponding interactions (lower pseudo-energy) are. Therefore, a Maxwell-Boltzmann probability distribution of counts is defined as
πij(β)=1Z(β)exp(−βEij)
(1)
where Z(β) = ∑(i,j)∈c exp(−BEij) is the partition function and β is the thermal parameter (inverse of a temperature). Eq 1 evaluates the joint interaction probability of a pair of loci (i,j), which can be modulated by the thermal parameter β. For low values of β (high temperature), the interaction energies tend to contribute equally in the exponential, whereas for high values β (annealing, low temperature) only the highly interacting loci contribute significantly in the exponential. The transition probability associated with the Markov jump process is defined by the following conditional probability
pij=πijμi
(2)
where
μi=∑j∈cπij
(3)
is the probability for the genomic locus i to form any interaction in chromosome c. Additionally, μi is the steady state distribution of the transition matrix pij. The transition probability matrix in Eq 2 uniquely identifies a discrete time Markov jump process that governs the trajectories of a random walker across the state space. A random walker is interpreted as a probe particle traveling between genomic loci, for instance a protein such as a transcription factor. Accordingly, the steady state distribution μi can be interpreted as a distribution of probes in a locus i, whereas the distribution πijc in Eq 1 describes an undirected flux of probes connecting the loci i and j.
The kinetic distance between pairs of loci is the mean first passage time (MFPT), the mean number of discrete steps τij between two different genomic loci (i,j) in chromosome c, which is obtained by solving the system of equations [78]
τij=pij+∑k≠i,jpik(1+τkj)
(4)
If the departure and arrival states coincide, the MFPT is called mean recurrence time MRT τi, which gives the mean time for a walker to return to its initial state i. The MRT is obtained from the MFPTs in Eq 4 via the formula
τi=pii+∑k≠ipik(1+τki)
(5)
S7 Fig shows the MFPT matrices (with the MRTs in diagonal) for chromosomes 1, 17, and 20 at β = 1 (left) compared to annealing condition at high β (right). A separation of time scales emerges upon increasing the β parameter, which is reflected in the partitioning of the MFPT matrices. The squares depicted in the annealed MFPT matrices identify sets of pairs (i,j) with a similar τij, which are the results of the partitioning in the network of interactions. S8 Fig shows the MFPT matrices for all chromosomes at corresponding high β. For each chromosome, the value of β was chosen as large as possible to observe the fine partitioning structure in the chromosome, while at the same time avoiding singular values in the calculations of the MFPTs and MRTs. The values of β used are listed in S1 Table.
The optimization of the metastability index upon annealing conditions allows one to obtain an optimal hub set M. It has been shown that in a MSM with a large state space, an optimal hub set and its corresponding partitions can be used for reducing the state space and obtaining a smaller MSM [59]. In the context of our model of chromosome interactions, we introduce a scheme for coarse-graining chromatin structure and quantifying effective interactions between the obtained partitions. In the previous section, we have described the interaction network of a chromosome c in terms of a MSM with transition matrix pij (Eq 2), representing the probability for a probe to reach locus j from locus i, with the steady-state distribution of probes μi in locus i (Eq 3), and with the undirected flux of probes πij between loci i and j defined in Eq 1. Considering the optimal hub set M with its associated committor probability qa(i) in the chromosome c, the effective flux of probes between any two soft-partitions a,b is given by
Fab=∑i∈cqa(i)πib
(11)
where πib is the undirected flux of probes between the loci i and b (Eq 1) in chromosome c. In other words, Eq 11 measures the portion of flux between any locus i and hub b passing through hub a (see S2 Fig for illustration of the notion of effective interactions). Eq 11 is an exact calculation on a single chromosome.
We are also interested in evaluating the effective fluxes between hub loci in different chromosomes. However, Eq 11 cannot be simply extended to the whole genome as the committor is by construction qa(i) = 0 for any locus i ∉ c. As computational limitations do not allow us to calculate the exact committor for the entire genome, a mean field formulation of Eq 11 was used to estimate the effective flux between any two partitions A,B in the genome. To this end, the effective flux between partitions, irrespective of the chromosomes to which they belong is calculated as
FAB=∑i∈gθA(i)∑j∈gπijθB(j),
(12)
where the summations are carried over the entire genome g, πij is the flux of probes between any pair of i and j in the genome (Eq 1 with fij the Hi-C matrix of paired-end reads of counts is now extended to the entire genome), and θA(i) the hard-partitioning committor defined in Eq 8. The rationale of Eq 12 is to efficiently, though indirectly, estimate the flux between any two partitions A, B in terms of all the intermediate pairwise fluxes πij. Within the logic of a MSM, the effective fluxes in Eq 12 serve as a measure of chromatin effective interactions.
Chromosome partitions are obtained from the optimal hub set as a result of the metastability analysis upon annealing conditions. They offer a coarse-grained description of the genome as the interactions between partitions are characterized via effective interaction strengths (Eq 12). Given a genome-wide set of partitions obtained above, a putative reduced model of the major partition interactions can be constructed by directly coarse-graining the matrix of counts fij for the entire genome. The observed joint probability of interaction between two partitions A and B is
P(A∩B)=∑i∈A∑j∈Bfij∑(X,Y),X≠Y∑i∈X∑j∈Yfij,A≠B
(13)
where the summation in the denominator is carried out on the pairs (X,Y) of distinct partitions to ensure proper normalization. Because of the law of total probability, the probability for a partition A to be involved in any interaction other than itself is
P(A)=∑Y≠AP(A∩Y)
(14)
which by construction adds up to one over all possible partitions A. In general, in the case of independent partitions, namely with no association between them, the relation P(A ∩ B) = P(A)P(B) would hold for the interaction probability. Therefore, to provide a measure of the degree of association between partitions, we define the following affinity as
CAB=P(A∩B)P(A)P(B)
(15)
which is a positively defined quantity. This quantity is also known as the observed to expected ratio o/e where P(A ∩ B) and P(A)P(B) are the observed and expected probabilities, respectively. In the case of CAB > 1 where the observed probability exceeds the expected, this is interpreted as a degree of association between partitions, either a contact or functional relationship. On the contrary, if CAB ≤ 1 observed and expected probabilities either coincide or the expected probability exceeds the observed one. These situations are interpreted as either no association (between partitions CAB = 1) or dissociation (partitions repel each other for CAB < 1). Thus, high values of affinity indicate a high degree of association between partitions, suggesting the presence of active binding and/or co-localization mechanisms. Intra-chromosomal pairs show very high affinities, typically with CAB > 10, while inter-chromosomal pairs have affinities CAB < 4.
In this work, we analyzed 50kbp in-situ Hi-C interaction maps obtained by Rao et al. [23] for human B lymphocyte cells (GM12878, two replicates) at both single-chromosome and whole-genome levels (GEO accession GSE63525). Three other datasets listed under the same GEO accession were also analyzed: IMR90 (lung fibroblast), HUVEC (umbilical vein endothelial cells), and HMEC (mammary epithelial cells). Epigenomic data tracks for GM12878 were obtained from the ENCODE Consortium web portal, with signal tracks for transcription factor ChIP-Seq from ENCODE/Stanford/Yale/USC/Harvard, histone ChIP-Seq from ENCODE/Broad Institute, DNase-Seq from ENCODE/OpenChrom (Duke).
Z-scored fractions of epigenetic factors were calculated in order to investigate their distributions within partitions. In the single-chromosome case, for a given signal track density xf(A) of factor f in a partition A of chromosome c, the Z-scored density of factor f is:
Zf(A)=xf(A)−μfσf
(16)
where μf and σf are the weighted mean and standard deviation of densities of factor f across partitions in the chromosome c. For the Z-score calculations on the whole-genome, the weighted mean and standard deviation across all 539 partitions were used.
For the network representation of the effective interactions, the force-directed layout in Cytoscape was used [79] with the force constants parametrized as logFAB, where FAB is the effective interaction between partitions (Eqs 11 and 12). The node sizes are proportional to the partition size or Z-scored epigenetic factor density, respectively. Only partitions of size larger than 2Mbp are shown. Edge width scales with logFAB and only interactions above a certain threshold are shown. For intra-chromosomal networks, width of edges is defined according to fixed thresholds of the interaction strength at each level of hierarchy.
In the whole-genome network of effective interactions, given the large number of partition pairs with a wide spread of effective interaction strengths, we classify interaction strengths into discrete levels and ignore weaker interactions. Histograms of the distribution of effective interaction strengths are plotted in S4 Fig, with intra-chromosomal (red), inter-chromosomal (green), and all (blue) interactions shown on the same axis. Layers of successively weaker interactions provide finer details to the interaction network structure (see S4 Fig): (i) Scaffold-Layer interactions are the strongest 2000 interactions, or the top 1.35% of all interactions; (ii) Layer 1 interactions comprise the top 1.35% to 1.5% of all interactions, compared with the Scaffold Layer; (iii) Layer 2 interactions represent the top 1.5% to 1.7% of interactions; (iv) Layer 3 interactions represent the top 1.7% to 2.0% of interactions. In our analysis for the GM12878_primary network, we considered only the scaffold and Layer 1 interactions (the top 1.5% of all interactions) to be significant.
Before performing the MSM analysis for single chromosomes, a Gaussian Filter (GF) was employed to reduce the effects of sampling noise and systematic errors in Hi-C data: the matrices of raw interaction counts were convolved with a Gaussian kernel. With the interaction matrix at 50kbp resolution, a width parameter in the Gaussian kernel σ = 200kbp was used, truncated at 4σ. S11 Fig shows a comparison of the raw Hi-C matrices with those after the GF preprocessing on chromosomes 1, 17, and 20. Performing partitioning analysis with and without GF preprocessing showed that both approaches yielded similar hub sets and partitions, and also when σ is varied within reasonable bounds, but the optimization on filtered datasets converged more rapidly. The Gaussian kernel width σ was chosen to balance between retaining structural information and computation speed: while increasing σ improved convergence rate, doing so smears out structural information in the high-resolution interaction matrices. Computation of effective interactions between partitions (Eq 12) is not affected directly by GF as the raw interaction matrices are used for obtaining the πij values.
The algorithms used in this study are implemented in a freely available Python package ChromaWalker (https://bitbucket.org/ZhenWahTan/chromawalker), built on the standard SciPy stack of libraries (NumPy, SciPy, Matplotlib, and Pandas), using a serial implementation on CPU. The run time for a full genome at 50kbp resolution, on a 3.4GHz Intel Core i7 CPU with 8GB RAM, is approximately 1 week.
|
10.1371/journal.pgen.1008076 | Heterogeneous pathway activation and drug response modelled in colorectal-tumor-derived 3D cultures | Organoid cultures derived from colorectal cancer (CRC) samples are increasingly used as preclinical models for studying tumor biology and the effects of targeted therapies under conditions capturing in vitro the genetic make-up of heterogeneous and even individual neoplasms. While 3D cultures are initiated from surgical specimens comprising multiple cell populations, the impact of tumor heterogeneity on drug effects in organoid cultures has not been addressed systematically. Here we have used a cohort of well-characterized CRC organoids to study the influence of tumor heterogeneity on the activity of the KRAS/MAPK-signaling pathway and the consequences of treatment by inhibitors targeting EGFR and downstream effectors. MAPK signaling, analyzed by targeted proteomics, shows unexpected heterogeneity irrespective of RAS mutations and is associated with variable responses to EGFR inhibition. In addition, we obtained evidence for intratumoral heterogeneity in drug response among parallel “sibling” 3D cultures established from a single KRAS-mutant CRC. Our results imply that separate testing of drug effects in multiple subpopulations may help to elucidate molecular correlates of tumor heterogeneity and to improve therapy response prediction in patients.
| Commonly occurring genetic alterations and patient-specific genetic features are increasingly used to predict the possible action of targeted cancer therapies. Although several lines of evidence have suggested that preclinical and clinical responses concur, the heterogeneity of tumors remains a severe obstacle in routinely translating preclinical data to patient treatments. Here we present a rapid work flow that integrates drug testing of three-dimensional patient tumor-derived (organoid) cultures and assessment of their genetic make-up as well as that of their donor tumors by amplicon sequencing and targeted proteomics. While the organoid cultures largely recapitulated the genomic profiles of donor tumors, the overall treatment responses and inhibitor effects on the intracellular signaling system were quite variable. Notably, organoid cultures obtained by synchronous multi-regional sampling of the same colorectal tumor showed an up to 30-fold difference in drug response. A combinatorial drug treatment improved the response. These data were confirmed in matched mouse xenograft models from the same tumor. Our findings may help to refine preclinical testing of individual tumors by modelling heterogeneity in cultures, to better understand therapeutic failure in clinical settings and to find ways to overcome treatment resistance.
| Colorectal cancer (CRC) is the third most common cancer worldwide. The major molecular alterations driving cancer of the colonic epithelium involve dysfunction of the WNT/APC and KRAS/MAPK pathways, DNA repair and methylation, and chromosomal instability [1, 2]. Concerted analysis of colorectal cancer transcriptomes has identified consensus molecular subtypes (CMS) that comprise genes controlling cell-type-specific functions, signaling pathways, and, in part, prognosis-relevant characteristics [3, 4]. CMS1 features genes involved in hyper-mutation and hyper-methylation. CMS2 represents the properties of epithelial tumor characteristics mainly driven by WNT/MYC signaling. CMS3 characterizes KRAS-mutated CRC and reflects genes controlling metabolic pathways. CMS4 exhibits expression of stromal components and activation of TGF-beta and VEGFR pathways. Like in other cancer entities, the administration of targeted drugs to colorectal cancer patients is based on the molecular profile of their tumors. Mutations in RAS gene family members or BRAF turn out to be negative predictors for anti-receptor tyrosine kinase therapies [5], while at least a subset of RAS wild-type tumors shows a therapeutic response [6].
Three-dimensional cell culture systems provide accurate and physiologically relevant models for studying the biology of diseases, and they support clinical research as well as drug development [7]. Recently, several groups have described patient-derived colorectal cancer organoids as a discovery platform for therapeutics and for validating the predicted impact of molecular features on therapy responses. Since the tissue architecture, tumor cell-specific genomic alterations, and consensus molecular signatures are essentially maintained in organoid cultures, these models are an excellent source for studying tumor biology in general under conditions reflecting clinically manifested heterogeneities of mutational patterns and epigenetic alterations [8–12]. For example, we identified the hedgehog pathway as a critical driver of colon cancer stem cell survival and tumorigenesis [13]. In scenarios approximating clinical behavior, drug treatments of CRC organoid cultures can even be used to predict personalized therapeutic options for individual patients [10]. A recent report has indeed documented the close relationship between drug effects observed in patient-tumor-specific organoid cultures and clinical responses in donor patients enrolled in clinical trials [14]. CRC organoid cultures also recapitulated the clinically well-known resistance of KRAS-mutant cells toward anti-receptor kinase therapies and showed limited effects of combinatorial targeted therapies against KRAS pathway effectors [15].
It is well known that the molecular diagnosis of primary and metastatic tumors, typically based on single biopsies representing snapshots of ongoing tumor evolution, may be compromised by intratumor heterogeneity (ITH) [16, 17]. Multi-region sampling of solid tumors revealed substantial ITH, as indicated by diverse patterns of cancer gene mutations [18] and copy number alterations [19] in different tumor areas. Individual tumors may harbor up to 10 or more preexisting subpopulations. Tumor evolution is assumed to be based on multiple initial alterations [20], and spatially separated subpopulations of cancer cells are generated during tumor progression [21]. Therefore, ITH is very likely to significantly influence initiation, establishment, fitness, and molecular characteristics of patient-tumor-derived organoid cultures.
Here, we used a previously characterized set of CRC organoids [11], supplemented by novel models, to investigate the impact of intertumor and intratumor heterogeneity on drug response. We chose to analyze the mutation status of the organoids by targeted amplicon sequencing, because interrogating a limited number of driver genes is often used as the standard diagnostic procedure [22, 23]. In addition, we analyzed signaling pathways in a subset of organoid models by targeted proteomics [24]. We correlated mutation patterns and pathway activity with the response to selected compounds targeting receptor tyrosine kinases, the RAS/MAPK, and PIK3CA pathway. Finally, we modelled the consequences of ITH on tumor cell growth and drug response in independently generated “sibling” cultures from the same donor tumor in vitro and in vivo.
Currently, our collection of patient-tumor-derived cell culture models comprises 91 organoids originating from 158 primary or metastatic tumor specimens (S1 Table). Due to the limitations of clinical material, not all tumor biopsies obtained gave rise to proliferation-competent organoid cultures. The time intervals required for the first transfer of primary to secondary cultures varied substantially but was not related to the UICC stage of donor tumors (Fig 1A, S1 Fig). Sufficient cell material for subsequent assays was available in 70% (51/71) of cultures 3 to 4 weeks post-explantation, while the remaining cultures required 5 to 12 weeks. Successfully established cultures maintained typical CRC marker expression and an adenoma-like architecture that retains higher-order organization and apical-basal polarity of the colonic epithelium, as reported earlier [11]. In view of the reported genomic heterogeneity of CRCs, the cultures were expected to be invariably polyclonal during establishment and early passaging.
We evaluated 50 cancer genes by ultra-deep targeted amplicon sequencing [25] in 49 organoids and 29 matched donor tumors (Fig 1B, S2 Table). Donor tumors and derived models both harbored common driver mutations in EGFR/RAS/RAF/MEK, PIK3CA/AKT, and TGF-beta signaling pathways, confirming previous exome sequencing results [11]. There were few exceptions, most likely due to population dynamics in the original tumor tissues [11, 18, 19]. Donor tumor tissue OT288 harbored an ABL1 mutation at a frequency of 6.3%, not detected in the corresponding organoid culture. The SMAD4 and PIK3CA mutations manifested in OT159 and OT161 cultures were not detected in the donor materials. A low-frequency APC mutation in tumor OT281, represented by only 3% of sequencing reads, was retrieved as a homozygous mutation in the corresponding organoid culture. As the received samples represented a macro-dissected admixture of tumor cells and adjacent healthy cells of different cell types, increased mutation frequencies in vitro were most likely due to enrichment of tumor cells by the used culture system and medium.
Notably, we observed a trend for earlier passaging in cultures that had accumulated mutations in RAS, PI3K/AKT, or SMAD4, as indicated by robust establishment and short culture intervals (Fig 1C).
For anti-EGFR treatment testing of 38 KRAS-mutant and wild-type organoid cultures, we focused on the small molecules gefitinib, afatinib, and sapitinib (Fig 2A–2C, S3 and S4 Tables) rather than on the monoclonal antibody cetuximab because its effects in cell culture do not correlate well with clinical action. Earlier reports described IC50 values for cetuximab in the millimolar range, either directly measured or extrapolated [9, 26, 27], while biologically relevant concentrations were reported to be in the nanomolar range [28]. In addition to preventing the binding of ligand to its receptor, cetuximab exerts antibody-derived cellular cytotoxicity, not recapitulated well under cell culture conditions [29, 30].
We considered potentially relevant inhibitory effects of the compounds at a range of concentrations achievable in patient plasma—i.e., at or close to the css steady-state concentration (Fig 2, grey areas) that reduced viability of >50% of cells under the testing conditions. The IC50 values for gefitinib ranged from 0.12 μM to >60 μM. Thirty-one of 38 cultures were resistant to gefitinib, including 13 quadruple-negative cultures (KRASwt, NRASwt, BRAFwt, and PIK3CAwt) and 16 cultures harboring mutations in at least one of these critical genes. Four quadruple-negative, two PIK3CA-mutant, and one exceptional BRAFV600E-mutant cultures responded to gefitinib treatment. The resistance to receptor tyrosine kinase inhibitors corroborated previous preclinical and clinical findings [4], although the range of growth responses indicated substantial heterogeneity of cultures regardless of critical driver mutations. The IC50 values for afatinib and sapitinib ranged from 0.025 μM to 11.7 μM and from 0.033 μM to >60 μM, respectively. In total, 34 cultures did not respond to afatinib, and 25 did not respond to sapitinib.
In view of the heterogeneous drug response, we added another layer of molecular information to the models by contrasting protein profiles of nine organoid cultures, matched tumor tissues, and morphologically normal tissues adjacent to the tumors (S2 Fig, S5 Table). We applied the novel DigiWest bead-based western blot method, designed for in-depth protein profiling of cellular signal transduction [24]. Overall, protein and phosphoprotein abundance varied substantially among cultures and tissues. Next, we correlated the drug response with the activation status of the MAPK pathway (Fig 3). Surprisingly, gefitinib-resistant cultures OT227 (KRASG13D), OT209-M (KRASG12V), OT161 (KRASG12S), and OT151 (NRASQ61K) exhibited very diverse pathway activation, ranging from throughout high expression of p-ERK1, p-ERK2, and pRSK1, for example, to low expression and phosphorylation, respectively. Expression of DUSP6, an antagonist of MAPK signaling involved in feedback regulation, was not consistently associated with diminished phosphorylation of signaling kinases. We observed a similarly diverse pattern of pathway activation in gefitinib-responsive RASwt organoids OT299 and OT276 and the resistant cultures OT347-M01 and OT278.
In addition to blocking receptor tyrosine kinase activity, we treated organoid cultures with the MEK inhibitor selumetinib, the multi-kinase inhibitor regorafenib, and a novel mTORC1/2 inhibitor BI-860585 [31] (Fig 2D–2F, S3 and S4 Tables). We found resistance to selumetinib in 16 out of 17 quadruple-negative organoids and in 13 out of 16 RAS-mutated ones. Three PIK3CA-mutated, RASwt organoid cultures were equally resistant, while the BRAFV600E cultures OT261 and OT212 were sensitive. This finding would suggest that in the absence of damaging RAS and PIK3CA mutations, signaling downstream of BRAF was efficiently blocked. The two KRASG12D cultures OT302 and OT288 were exceptionally sensitive to MEK inhibition; however, the susceptibility towards the inhibitors was not exclusive for this kind of KRAS mutation.
Treatment with the multi-kinase inhibitor regorafenib that targets CRAF and VEGF receptors resulted in responses irrespective of the mutational status of RAS and PIK3CA, with the distribution of sensitive versus resistant cultures being approximately equal for wild-type and mutant cultures (Fig 2E). The range of IC50 values in most cultures was relatively narrow (3.05–23.9 μM), except for cultures OT114 and OT152 (>60 μM) and OT234 (0.58 μM). Since regorafenib specifically targets the tumor vasculature, any conclusions regarding the proliferation of organoids are very limited [14].
Nineteen out of 38 organoid cultures responded upon treatment with BI-860585, and 19 were resistant (Fig 2F). All PIK3CA/PTEN-mutant, KRASwt organoids responded to the mTORC1/2 inhibitor, three PIK3CA/KRAS mutant cultures were resistant, and two of them were sensitive. Analysis of protein extracts by the bead-based western blotting method revealed that neither the level of p85 (alpha), p110 beta or PTEN expression nor the phosphorylation status of PTEN and AKT correlated with the response to mTORC1/2 inhibition (Fig 3).
Next we analyzed the effects of the RTK inhibitors gefitinib, afatinib and sapitinib by profiling 3 selected organoids OT151, OT276 and OT347-M01 employing DigiWest protein expression profiling after drug exposure for 72h. To avoid substantial cell loss and accumulation of cellular debris during drug exposure, we chose drug concentrations primarily based on the clinically achievable plasma concentrations (css, S3 Table). There were two exceptions, in which the IC50 was achieved already below the relevant plasma concentration in patients. Accordingly, we treated OT276 with 0.35 μM gefitinib and OT347-M01 with 0.20 μM sapitinib. The detailed results of DigiWest analysis are shown in S3 Fig and S6 Table. Hierarchical clustering of the analyzed data showed that patient derived organoids formed three separate branches in the dendrogram. These were split further into branches distinguishing gefitinib from afatinib and sapitinib treatment indicating a dominant but distinctive effect of the tested drugs on each individual organoid culture.
For example, the treated NRAS mutant organoid OT151 showed an increase in phosphorylation of central proteins of the MAPK signaling cascade, including ERK, RSK, and S6 ribosomal protein (S3 Fig, S6 Table). While this expression- and activity pattern could easily be reconciled with the resistance to anti-receptor tyrosine kinase inhibition, there was no clear correlation between treatment resistance and drug-modulated signaling patterns in OT276 and OT347-M01 cells. OT276 cells were sensitive to afatinib and sapitinib treatment although p-MEK was increased. Sapitinib sensitivity in OT347-M01 cells was associated with low MAPK activity. We observed a distinct modulation of WNT signaling indicated by GSK3β phosphorylation and a high increase in active beta-catenin in OT347-M01 cells. MTOR activity appeared to be high in OT276 but low in OT347-M01. Overall, most alterations observed 72h after drug exposure are very likely due to the rewiring of the entire signaling system involving adaptive feed-back and cross-talk mechanisms.
To investigate the impact of ITH on treatment effects in organoid cultures, we analyzed mutational patterns and drug responses of five organoid “sibling” cultures established in parallel from separate regions of an individual primary colon carcinoma (CC0514-R1, -R2, -R3, -R4, and -R5). Tissue architecture, Ki67 expression, and expression patterns of markers routinely used to characterize tumors of the colonic epithelium recapitulate those of the donor tumor (Fig 4). Using the 50-gene cancer panel, we detected common KRASG12D, PIK3CAH1047R, and TP53C242F mutations in the tumor tissue-of-origin and the separate sibling cultures, as well as an additional homozygous SMAD4R361H mutation in cultures CC0514-R1 and CC0514-R2 (Fig 1B, S2 Table). This mutation affects the MH2 domain of the protein and abolishes R-SMAD/SMAD4 heterodimerization [32]. All sibling cultures represented the molecular subtype CMS2 (S7 Table), according to the classification by Guinney and colleagues [3].
First, we assayed drug responses in the sibling cultures using inhibitors targeting EGFR, MEK, ERK, p110α, and mTORC1/2 as well as sorafenib and regorafenib (Fig 5A–5C, S4 Table). The drug responses of sibling cultures were not uniform. Organoid culture R1 was unique by being unaffected by inhibitors targeting EGFR, MEK, ERK, PIK3CA, and mTORC1/2. In contrast, culture R4 was sensitive toward EGFR, PI3Kα, and mTORC1/2 inhibition. The remaining cultures, R2, R3, and R5 shared resistance with R1 to EGFR inhibition, but responded heterogeneously to the other compounds.
As the underlying mutations did not sufficiently explain the observed differences in drug response, we expanded molecular analysis by exome and RNA sequencing. Exome sequencing confirmed the mutations found via panel sequencing. An additional shared heterozygous mutation was found for beta-catenin (CTNNB1R582W). The original tumor tissues and sibling cultures exhibited a remarkable genetic heterogeneity (Fig 6). Unbiased analysis of their somatic mutation landscapes highlighted specific mutations in every single region or derived culture (Fig 6A, S8 Table). Phylogenetic trees of the tumor mutations suggested a grouping of tumor regions and organoids CC0514-R1/R2 versus R3, R4, and R5. Moreover, unbiased principal component analysis (PCA) based on the complete transcriptomes of the cultures showed a high variance between cultures R1/R2 and R3, R4, and R5 (Fig 6B). Overall, we identified 646 differentially expressed genes when contrasting SMAD4R361H and SMAD4wt cultures (R1 and R2 vs. R3, R4, and R5; adj. p < 0.05, logFC >1, S9 Table). We visualized the mRNA transcriptome of sibling cultures in heatmaps, focusing on ERK/MAPK-, PI3K-, and mTOR-signaling pathways (Fig 6C). Although retrieving the same sample dichotomy between the R1/R2 and R3–5 groups, we observed substantial heterogeneity in mRNA expression of target genes encoding components of the respective pathways.
To model the therapy response in a heterogeneous tumor cell population in vivo, we first labeled organoid cultures CC0514-R1 and CC0514-R4 by transduction with phosphoglycerate kinase (PGK) promoter-driven expression vectors encoding the fluorescence markers GFP and mCherry (mCh), respectively. We then prepared single-cell suspensions of R1-GFP and R4-mCh cells embedded in Matrigel, separately or as a 1:1 mixture of R1-GFP and R4-mCh cells, injected them subcutaneously into nude mice, and calculated the tumor volumes over time (S10 Table). The R1-GFP cells rapidly formed tumors with a mean volume of 1.36 (±0.16) cm3 after 18 days, while the R4-mCh tumor xenografts achieved less than one-tenth of these volumes (mean 0.11 cm3 ±0.03) after the same period. Even at >80 days after inoculation, R4-mCh tumors were smaller (mean volume 0.76 cm3 ±0.66) than R1-GFP tumors that had formed much earlier. The mixed suspensions of R1-GFP and R4-mCh cells produced tumors after a short lag phase. Tumor volumes comparable to the ones produced only by R1-GFP cells were reached approximately 10 days later (Fig 7A, S10 Table). Notably, the architecture of mixed-culture xenografts indicated strict clonal outgrowth in vivo, as the R1-GFP and R4-mCh cells always formed separate organoid structures (Fig 7B).
For targeting the EGFR/RAS/MEK-signaling system in vivo, we chose the clinically approved EGFR antibody cetuximab and the MEK inhibitor trametinib (S11 Table). The duration of treatment with individual and combined inhibitors was planned for 36 days, following an initial period of 10 days post-injection to achieve engraftment. However, in cases of rapid tumor establishment and progression, the experiment had to be terminated earlier. After measuring tumor volumes, we sacrificed the mice, dissociated the tumor cells, and determined the ratio of R1-GFP and R4-mCh cells by FACS (S12 Table).
In xenografts derived from mixed cultures, R1-GFP cells formed the dominant population, indicated by its greater potential for expansion (Fig 7C, S10 Table). Compared to inoculation of single-cell suspensions, co-injection of R1-GFP and R4-mCh cells, left untreated or treated with vehicle, retarded the onset of tumor growth slightly, and progression to 1.0 cm3 tumors was apparent after an additional week (Fig 7D). Treatment of the KRAS-mutant mixed cultures with cetuximab or trametinib alone reduced tumor take rate (determined on day 38 post-injection) and substantially prolonged the period required for achieving an equal 1.0 cm3 tumor mass in two out of three tumors (Fig 7E and 7F). At the end of this experiment, the R1-GFP population outnumbered the R4-mCh population in controls and inhibitor-treated tumors. Reduction of tumor takes and volumes was more pronounced following cetuximab/trametinib co-treatment. Moreover, the R4-mCh population outnumbered the R1-GFP population, indicating that the latter population was vulnerable to co-treatment, although a minor fraction of R1-GFP cells survived combination treatment (Fig 7G, S12 Table). The initial response to cetuximab treatment of R1-GFP tumors may be due to a surprisingly low activation level of MAPK signaling; however, antibody-dependent toxicity cannot be excluded (Fig 7H, S5 Table). In contrast, the R4-mCh population exhibited enhanced levels of p-ERK1/2 in spite of its low growth potential in vivo. The growth kinetics of R4-mCh organoids in vivo suggest that the duration of inhibitor exposure until day 45 post-injection into mice was too short for this slowly expanding population, although in vitro testing showed susceptibility to anti-EGFR treatment (Fig 5). The presence of GFP-positive cells beyond day 45 suggested that after initial debulking, this population escaped from treatment by anti-receptor tyrosine kinase antagonists, in-line with the results from in vitro testing. The slowly expanding R4-mCh cells may even have provided a niche for surviving R1-GFP cells.
The R1/R4 xenografts responded partially to administration of everolimus, regorafenib, and pictilisib, as well as to combinations of these three drugs with cetuximab (S11 Table). In five out of six experiments, the proportion of residual R1-GFP cells relative to R4-mCh cells after treatment indicated partial response of the rapidly proliferating R1-GFP population and the relative resistance of the R4-mCh population able to increase in number at the expense of R1-GFP cells.
Here we show heterogeneous pathway activity and inhibitor responses in CRC organoid cultures profiled by cancer gene panel sequencing and DigiWest-based signaling-pathway analysis. Eight KRAS G12V/S/D cultures were resistant to EGFR inhibition by gefitinib, as were four KRASG13D and three NRASQ61K cultures, one KRASA146T and one BRAFV600E culture. Four quadruple-negative and two RAS wild-type/PIK3CA-mutant cultures responded to gefitinib. Basically, these results are in line with diagnostic and clinical experience regarding the administration of therapeutic antibodies targeting EGFR [4] and a recent report on colorectal cancer organoids [15]. In our cohort, 12 quadruple-negative cultures were resistant to gefitinib treatment. The magnitude of drug resistance observed in quadruple-negative cultures is reminiscent of clinical experience in tumor treatment. For example, the organoid culture OT330 and its donor tumor harbored an ERBB2 amplification [11]. High ERBB2 expression detected in vitro is known to cause resistance to EGFR-targeting compounds [33, 34]. Moreover, OT330 shared increased levels of MET mRNA with the KRASwt cultures OT216 and OT281. High MET expression in the absence of activating KRAS mutations may mediate primary resistance and an escape mechanism from anti-EGFR receptor treatment [35].
Since extracellular signal-regulated kinases (ERKs) control the biological outcomes of EGFR/RAS signaling [36], we analyzed their activation status in organoid cultures. Interestingly, the presence of KRAS/NRAS mutations did not robustly coincide with MAPK pathway activation determined by the enhanced steady-state phosphorylation of effector kinases BRAF, CRAF, MEK1/2, ERK1/2, or RSK1 p90. Consequently, the wild-type status of KRAS or NRAS proteins did not predict gefitinib response. Similarly, monitoring the activation status of the MAPK pathway downstream of RAS did not substitute as a diagnostic approach for predicting treatment response. For example, KRASG12S organoid OT161 exhibiting low pathway activation, reminiscent of a “functionally wild-type” status, was gefitinib-resistant. Conversely, KRAS/NRAS/BRAFwt culture OT299 exhibited a “functionally mutated,” highly active pathway status in spite of its susceptibility to gefitinib and other EGFR inhibitors. The polyclonal nature of organoids analyzed in early passage most likely reflects the intrinsic molecular heterogeneity of the donor tumor and may even preclude the detection of homogeneously strong MAPK activation. Eventually, subcloning of organoid cultures may allow enrichment of cells exhibiting uniform pathway activation in the presence of the corresponding upstream mutations. However, the value of cloned organoids for serving as avatars for the donor tumor will be diminished, since ITH is no longer maintained. The uncoupling of mutational KRAS status and pathway activation is not unprecedented, as shown in pancreatic adenocarcinoma and colorectal cancer [37, 38]. BRAFV600E mutant colorectal cancers have been shown to express differential activation of the KRAS/AKT pathway and cell cycle proteins, indicating a diverse biology for a subtype of colorectal cancers even driven by the same driver mutation [39]. Here we show that MAPK signaling activity is diverse and not directly affected by the presence of a KRAS mutation. Hence, the true biology of these tumors may only be understood after conducting further comprehensive mechanistic studies. MAPK pathway activation is known to be modulated by dual-specificity phosphatase 6 (DUSP6) and Sprouty (SPRY) isoforms, which constitute transcriptional feedback loops and exert post-translational functions that constrain signaling-kinase activities [40–42]. However, the protein levels of DUSP6 and SPRY1, 2, and 3 were not correlated with MAPK activation in our organoid cultures.
Drug susceptibility testing was done in polyclonal organoids in low passage early after establishment of cultures. It is very likely that tumor subpopulations present in newly established models exhibit different degrees of fitness and pathway activity, a diverse potential to expand, and non-homogenous inhibitor responses. In general, the big-bang hypothesis [20] predicts that during tumor evolution, many subpopulations exhibiting diverse mutation patterns, phenotypes, and treatment responses can emerge. Variable clonal population dynamics can substantially modulate treatment response and tolerance [43]. We investigated this scenario through multi-region sampling from a single colorectal tumor and established sibling organoids. Molecular analysis by sequencing crucial cancer genes, exomes, RNA, and targeted proteomics identified uniform KRAS, TP53, and PIK3CA driver gene mutations in all parallel cultures, but also provided evidence for substantial ITH. Not surprisingly, the sibling cultures showed profound differences in treatment response. Organoid culture R4 responded to 7 out of 10 inhibitors in vitro, while R1, harboring an additional SMAD4 loss-of-function mutation, was resistant to all small molecules tested, except for regorafenib and sorafenib. The response pattern was unequal in R2 organoids despite sharing the SMAD4 mutation and overall RNA expression pattern. The proliferation potential of R1 and R4 organoids in vivo differed substantially in spite of common KRASG12D, PIK3CAH1047R, and TP53C242F mutations. During the cetuximab/trametinib treatment cycle in xenografts, R4 cells survived and later formed the majority of the tumor post-therapy. The role of high ERK activation in R4 cells is not clear. High ERK activation does not necessarily drive proliferation, but it can be growth-inhibitory [44, 45].
The homozygous SMAD4R361H mutation was not detectable in the bulk tumor CC0514. We assume that acquisition of the mutation was not due to a de novo event in culture, but rather was present in a minor subpopulation of the original tumor. The loss of SMAD4 function, known to trigger tumor progression and metastasis [46–48], may explain the high proliferative potential of R1 organoids. In contrast, R4 organoids express presumably growth-inhibitory levels of phosphorylated ERK and resemble a cancer stem cell or progenitor phenotype, as described in Blaj et al. [38]. Interestingly, SMAD4 proteins can be degraded via ERK1/2 signaling [49], suggesting that SMAD4 does not function normally, even in poorly proliferative R4 organoids.
In summary, our findings imply that the administration of therapeutics to bulk organoid cultures may fall short of correctly displaying the functional diversity of existing subpopulations, due to heterogeneities of their genomes and transcriptomes. The great challenge of interpreting and reconciling multi-omics and response data in apparently closely “related" tumor cells was recently demonstrated by Roerink and colleagues [50]. The authors investigated organoid cultures established from single cell clones of different tumor areas of colorectal cancer patients. Via genomic sequencing, methylome analysis and RNA sequencing, they reconstructed evolutionary trees with high concordance between (epi)genomics and gene expression data. Yet, drug response in different organoids from the same tumor displayed substantial differences in IC50 values of up to 1,000-fold for chemotherapeutic agents and targeted inhibitors. The molecular events that lead to differential drug sensitivity of closely related organoids from the same tumors are currently unknown.
In future diagnostics settings, multisampling of malignant tissues, drug sensitivity testing in culture and the depth of subsequent molecular analysis will have to be carefully balanced under consideration of sample availability, time and even funding constraints. The molecular analysis of organoid cultures in this study was limited by the focus on frequent oncogenes and tumor suppressor genes as well as by the availability of validated antibodies used in DigiWest assays. Notwithstanding such limitations, analyzing the dynamics of sibling cultures may help to overcome the limitations of predicting precision targeting of mutations retrieved by genomic analysis alone [51] and to better inform therapeutic decision-making.
EPO strictly follows the EU guideline European Convention for the Protection of Vertebrate Animals Used for Experimental and Other Scientific Purposes (EST 123) and the German Animal Welfare Act (revised version Art. 3 G v. 28.7.2014 I 1308). Furthermore, we handle our animals according to the Regulation on the Protection of Animals Used for Experimental or for Other Scientific Purposes (Tierschutz-Versuchstierverordnung- TierSchVersV: revised version Art. 6 V v. 12.12.2013 I 4145). Compliance with the above rules and regulations is monitored by the Landesamt für Gesundheit und Soziales (LAGeSo), which is the responsible regulatory authority monitoring animal husbandry.
Organoid cultures were generated and propagated as previously described [11, 52]. Overall, 86 patient-derived three dimensional (PD3D) cell cultures were generated, including five samples that were isolated from mouse xenografts (PDX) and one sample that underwent transient engraftment and was then reintroduced into an in vitro organoid culture (OT151-PDX-PD3D). Additionally, for patient CC0514 we generated five organoid cultures from five separate regions of the primary tumor. The full cohort is described in S1 Table.
Two μm de-paraffinized FFPE tissue sections of donor tumors or organoid cultures grown for five days were stained using the primary antibodies anti-CK7 (clone OV-TL12/30, Dako, Germany), anti-CK20 (clone KS20.8, Dako), anti-CDX2 (clone CDX2-88, BioGenex, U.S.A.), and anti-KI67 (clone MIB-1, Dako) for 32 min at 37°C, using the ultraView DAB detection kit (Ventana, U.S.A.) on the BenchMark XT instrument (Ventana), as recommended by the manufacturer. Counterstaining was performed with Hematoxylin II Counterstain and Bluing Reagent (Ventana) for 4 min. Microscopy was performed with a Zeiss Axiovert 400 microscope (Zeiss, Germany).
For immunofluorescence imaging, organoid cultures were fixed in 4% paraformaldehyde for 30 min at room temperature and permeabilized with 0.1% Triton X-100 for 30 min. Samples were blocked in phosphate-buffered saline (PBS) with 10% bovine serum albumin (BSA) and incubated with primary antibodies overnight at 4°C. Antibodies used were Anti-Ezrin (clone EP886Y, Abcam, diluted 1:200) and EPCAM (VU1D9, Cell Signaling Technology, diluted 1:500). Samples were stained overnight with a conjugated secondary antibody at 4°C. F-actin was stained with Alexa Fluor 647 Phalloidin (#A22287, Thermo Fisher, diluted 1:20) for 30 min at room temperature. Nuclei were counterstained with DAPI (Sigma Aldrich). Cells were then transferred to microscope slides for examination using a Zeiss LSM 700 laser scanning microscope.
Genomic DNA from organoid cultures and fresh tumor tissues was prepared using the AllPrep DNA/RNA Kit (QIAGEN, Germany) according to the manufacturer’s protocols, and it was quantified using a Qubit 2.0 Fluorometer and the appropriate assay kits (Life Technologies, Germany). Genomic DNA from FFPE material from at least three consecutive 10 μm sections was isolated using the QIAGEN QIAamp DNA FFPE Tissue Kit to obtain sufficient amounts for targeted panel sequencing. The yields of FFPE-derived DNA were quantified with the TaqMan RNase P detection assay (Life Technologies).
Targeted sequencing was performed on an IonTorrent PGM bench-top sequencer (Life Technologies). Ten nanograms of genomic DNA were used to prepare Ion AmpliSeq Cancer Hotspot Panel v2 (Life Technologies) amplicon libraries in conjunction with the Ion AmpliSeq Library Kit 2.0 (Life Technologies). For multiplexing purposes, unique Ion Xpress barcode adapters were assigned to every sample. For amplification, 17 or 20 PCR cycles were applied for cell-culture-based and FFPE-tissue-based samples, respectively. Amplicon libraries were quantified with the Ion Library Quantitation Kit (Life Technologies), diluted to 100 pM, and pooled in an equimolar concentration. Clonal amplification of single PCR templates was carried out on an Ion OneTouch 1 cycler in conjunction with the Ion OneTouch 200 Template Kit v2 DL. Sequencing of four to five pooled libraries was performed using Ion 316v2 sequencing chips. Base calling, read mapping, and coverage analysis were performed with the default settings of Torrent Suite software version 4.0.2. Targeted panel sequencing yielded over 44.0×106 mapped reads, and the oversampling rates averaged to 3.6×103-fold. The mean uniformity of coverage (i.e., the distribution of reads across all 207 pooled PCR products per sample) was above 98%. Primary data analysis and variant calling were performed with the built-in Variant Caller tool (version v4.0-r76860), set to the “somatic, high stringency” option and down-sampling to 2,000 reads. Variant calls were visually inspected using the Integrated Genomics Viewer (IGV) [53] and annotated in accordance with HGVS recommendations [54]. To assess the effect of the retrieved mutations, the prediction tools SIFT [55], PolyPhen2 [56], and MutationTaster2 [57] were used. Visualization of mutations was carried out using the R Script ComplexHeatmap [58].
Additional Sanger sequencing was performed for sibling cultures CC0514-R1, CC0514-R2, and CC0514-R5 as early as six weeks post-culture initiation (S5 Fig), confirming the panel-sequencing results found for regions R1 and R2 (both SMAD4R361H) and R5 (SMAD4wt).
We also compared the mutation information gathered via panel sequencing with whole genome and whole exome data generated for the recently published study by Schuette et al. [11].For an overlap of 40 organoid cultures for both methods and overlapping genes, 78% (103/132) of the mutations were covered by the targeted sequencing approach, while 22% (29/133) of the mutations found with exome sequencing were not covered by the panel’s PCR amplicons (S13 Table). These included mutations in APC (86%, 25/29), PIK3CA (2/29), TP53, and PTEN (1/29 each). Microsatellite status was analyzed as previously described [11, 59].
To perform systematic and parallel testing of drug responses in vitro, the organoid culture system was adapted to a 384-well microtiter plate format, as shown earlier [60]. In short, the assay system is based on a luminescence readout of cell viability, measured via ATP consumption. The population doubling time was determined by time-course based measurements using CellTiter-Glo luminescent cell viability assessment. Treatment duration covered two doubling times of the individual cultures, and small molecules were tested at concentrations ranging from 3.0 nM up to 60.0 μM (assay cutoff), as previously shown [11]. Treatment duration of organoid sibling cultures derived from patient CC0514 was uniformly 72 h (S4 Table). The half-maximal inhibitory concentration of a compound determined in vitro (IC50) is a measure of the potency of the compound in blocking cell growth and survival. To assess the potential clinical relevance of the drug response assays, we compared the relative IC50 values determined in vitro for 43 organoid cultures (S4 Table) with steady-state plasma concentrations (css) of the therapeutic compounds achievable in patients (S3 Table). We used reference data obtained from publications or the investigator’s brochures (IB) available at clinical-trial centers. For compounds that have not yet entered late-phase clinical studies, we deduced css values from in vivo mouse studies (BI-860585) or early (phase I) clinical studies (alpelisib, ravoxertinib). In addition, we determined the maximum inhibition (efficacy, Emax (%)) per drug per organoid culture model.
RNA from all organoid sibling cultures of patient CC0514 was extracted and quantified as described above. Whole transcriptome sequencing was performed at the Genomics and Proteomics Core Facility (GPCF) of the German Cancer Research Center (DKFZ). An average of 2 μg total RNA per sample was used to generate barcode-labeled libraries using the Illumina TruSeq RNA sample preparation kit (Illumina, San Diego, CA, U.S.A.). Sequencing of 125 bp paired-end reads was performed on an Illumina HiSeq2000 sequencer with equal distribution of pooled libraries over two sequencing lanes. Mapping to the GRCh37 genome was performed with a STAR aligner [61], allowing maximally two mismatches/alignment gaps and 0.3% total mismatches per alignment. On average, RNAseq yielded 5.1×107 mapped reads (95% unique) covering 5.9 GB coding bases. Reads covering exonic regions per gene ID were counted using HTseq [62]. Downstream analysis was performed using AnnotationDBi, DESeq2 [63], limma [64–66], ggplot2 [67], and GSEA [68].
In total, 11 WES samples were processed, including a blood sample, five separate primary tumor regions from CC0514, and their respective organoid sibling culture derivatives. Whole exome sequencing was performed by the Genomics and Proteomics Core Facility (GPCF) at the German Cancer Research Center (DKFZ). The Agilent SureSelectXT Human All Exon V5 kit was used to generate 125 bp paired-end libraries subsequently sequenced on an Illumina HiSeq2000. The WES pipeline was organized as follows: First, poor-quality reads were filtered out using Trimmomatic [69], and the remaining reads were mapped to the GRCh37 (hg19) human reference genome using the BWA aligner [70]. Afterwards, the Genome Analysis Toolkit [71] was employed for duplicate removal, indel realignment, and base-quality recalibration. On average, 9.5×107 reads were uniquely mapped to the reference genome. We called germline, somatic, and loss of heterozygosity (LOH) mutations using the blood sample as reference with Varscan2 [72]. The minimum coverage for calling variant reads was set to 8 and minimum variant frequency to 0.09. We set tumor purity to 0.5 and minimum tumor frequency to 0.09. Rare SNP and indel mutations were selected as exonic, non-synonymous mutations with an ExAC [73] MAF score below 0.001. Evolutionary trees were built on all somatic mutations in exonic regions for cultures R1–R5. A neighbor-joining algorithm [74], implemented in the “APE” R package [75], was used on the Euclidean distance matrix generated from the binary mutation matrix. We used an artificial null-mutated control as root of the tree. In case of very low tumor cellularity, exome sequencing identified fewer somatic mutations (e.g., in tumor CC0514-R4). Following high-coverage panel sequencing, low-frequency mutations were detected after manually inspecting the sequencing reads (see mutation frequencies for CC0514-R4 tissue in S2 Table).
Tissue samples and organoid cultures were subjected to multiplex protein profiling analysis of up to 150 (phospho-)proteins. In addition, three organoid cultures were subjected to drug treatment prior to protein extraction. For this, 2.4 x 105 cells (= 4 x 104 cells/well in 6 wells total) per condition were plated, followed by a growth period of 72h and subsequent treatment with gefitinib, afatinib, sapitinib and vehicle control (0.03% DMSO) for the duration of 72h. Drug concentrations were chosen based on clinically achievable plasma concentrations (css, S3 Table). In order to monitor the pathway activity status and to gather a sufficient amount of cells for subsequent DigiWest analysis after drug treatment, IC50 values were selected for the gefitinib treatment of OT276 (IC50 = 0.35 μM versus css = 1.04 μM) and sapitinib treatment of OT347-M01 (IC50 = 0.20 μM versus css = 1.52 μM). After treatment, collected organoids were washed with ice cold PBS, sedimented, treated with Cell recovery solution (Corning) on ice for 30 min, washed twice with ice cold PBS, pelleted and snap frozen at -80°C until cell lysis.
The NuPAGE SDS-PAGE gel system (Life Technologies) was used to separate cellular lysates and blotting. Proteins (20 μg per sample) were fractionated by electrophoresis through 4–12% Bis-Tris gels according to the manufacturer’s instructions. Blotting onto PVDF membranes (Millipore) was performed under standard conditions. For high-content western analysis, the DigiWest procedure was performed as previously described [24]. Briefly, proteins immobilized on the blotting membrane were biotinylated (NHS-PEG12-Biotin, Thermo Scientific), and individual sample lanes were cut into a comb-like structure (strip height = 0.5 mm each, strip width = 6 mm) using an electronic cutting tool (Silhouette SD). The resulting 96 strips corresponded to 96 molecular weight fractions immobilized on individual membrane strips; they covered a range from 15 kDa to 250 kDa.
For protein elution, individual strips were placed in separate wells of a 96-well plate and incubated for 2 hours in 10 μl elution buffer (8 M urea, 1% Triton-X100 in 100 mM Tris-HCl, pH 9.5) for solubilization. After adding 90 μl of dilution buffer (5% BSA in PBS, 0.02% sodium azide, 0.05% Tween-20), 96 different Neutravidin-coated Luminex bead sets (60,000 beads/well) were added to the individual wells, and biotinylated proteins were captured on the bead surface. After overnight incubation, the Luminex beads were pooled, washed, and stored in a storage buffer (1% BSA, 0.05% Tween-20, 0.05% sodium azide in PBS) at 4°C. For antibody incubation, an aliquot of the bead pool (approximately 0.3% of the available bead pool) was transferred into an assay plate, and 30 μl of diluted western blot antibody in assay buffer (Blocking Reagent [Roche Applied Science], 0.05% Tween 20, 0.02% sodium azide, and 0.2% milk powder) was added per well. The list of antibodies is shown in S5 and S6 Tables. One hundred fifty-four antibody incubations were performed overnight at 4°C for each protein sample. For visualization, the beads were washed twice with 100 μl of PBST before species-specific PE-labeled secondary antibody (Jackson Dianova) was added in 30 μl of assay buffer for 1 hour. After two washes with 100 μl of PBST, analyte signals were generated in a FlexMAP 3D instrument (Luminex).
Data analysis: Data generated by the Luminex instrument were analyzed using a proprietary analysis tool that visualizes the fluorescent signals as bar graphs and identifies specific protein peaks detected by antibodies. Each graph was composed of the 96 values derived from the corresponding molecular mass fractions obtained after antibody incubation. The software tool identified specific peaks, and a relative molecular mass was assigned to each of the 96 fractions. After background correction, specific signal intensities were calculated as the integral of the identified peak.
Cell cultures CC0514-R1 and CC0514-R4 were transduced with lentiviruses carrying a phosphoglycerate kinase (PGK) promotor controlled eGFP (pLenti PGK GFP Puro, w509-5, Addgene) or mCherry. The pLenti PGK mCherry Puro vector construct was created via PCR cloning of mCherry cDNA from plasmid vector 7TCF (Addgene, Plasmid #24307) and modified with BamHI and BsrgI restriction sites for directed ligation into plasmid vector pLenti PGFK GFP Puro. Plasmids were tested for correct insert orientation via Sanger sequencing. Transduced cells were selected with puromycin. Following organoid cell culture expansion, cells were harvested at passage 08, digested, and filtered, and single cells were counted. Transplantation into nude mice was achieved by subcutaneous injection of single populations or 1:1 mixtures of 1.0×106 cells each at EPO GmbH (Experimental Pharmacology and Oncology, Berlin-Buch, Germany). Mice (BOMTac: NMRI-Foxn1nu; Genotype NMRI NU-F Sp/Sp) were obtained commercially from Taconic (Lille Skensved, DK).
Organoid models were injected at 1:1 dilution with Matrigel (Corning, NL) and were allowed to expand for 10 days, reaching PDX sizes of 0.1 cm3, before treatment with selected compounds for 36 days. The compounds were administered either orally or intraperitoneally. Following the first day of treatment, the tumor volume and body weight were recorded twice weekly. Animal welfare was controlled regularly. Animals in poor health were sacrificed independent of treatment duration. Tumor volume was calculated from the measurement of the length and width of subcutaneous tumors (TV = (length+width)2/2). For further information on the handling of PDX models, please refer to [76]. Mice were sacrificed after the PDX tumors reached a size of approximately 1.5 cm3. Retrieved PDX tumors were subjected to whole-mount (immunofluorescence) imaging using a Zeiss Axiovert 400 microscope (Zeiss, Germany). PDX tumor tissues were digested to produce single-cell suspensions and analyzed by FACS at the German Rheumatism Research Centre (DRFZ, Berlin, Germany) to record the distribution of GFP- and mCherry-positive cells.
|
10.1371/journal.pcbi.1006644 | A computational knowledge-base elucidates the response of Staphylococcus aureus to different media types | S. aureus is classified as a serious threat pathogen and is a priority that guides the discovery and development of new antibiotics. Despite growing knowledge of S. aureus metabolic capabilities, our understanding of its systems-level responses to different media types remains incomplete. Here, we develop a manually reconstructed genome-scale model (GEM-PRO) of metabolism with 3D protein structures for S. aureus USA300 str. JE2 containing 854 genes, 1,440 reactions, 1,327 metabolites and 673 3-dimensional protein structures. Computations were in 85% agreement with gene essentiality data from random barcode transposon site sequencing (RB-TnSeq) and 68% agreement with experimental physiological data. Comparisons of computational predictions with experimental observations highlight: 1) cases of non-essential biomass precursors; 2) metabolic genes subject to transcriptional regulation involved in Staphyloxanthin biosynthesis; 3) the essentiality of purine and amino acid biosynthesis in synthetic physiological media; and 4) a switch to aerobic fermentation upon exposure to extracellular glucose elucidated as a result of integrating time-course of quantitative exo-metabolomics data. An up-to-date GEM-PRO thus serves as a knowledge-based platform to elucidate S. aureus’ metabolic response to its environment.
| Environmental perturbations (e.g., antibiotic stress, nutrient starvation, oxidative stress) induce systems-level perturbations of bacterial cells that vary depending on the growth environment. The generation of omics data is aimed at capturing a complete view of the organism’s response under different conditions. Genome-scale models (GEMs) of metabolism represent a knowledge-based platform for the contextualization and integration of multi-omic measurements and can serve to offer valuable insights of system-level responses. This work provides the most up to date reconstruction effort integrating recent advances in the knowledge of S. aureus molecular biology with previous annotations resulting in the first quantitatively and qualitatively validated S. aureus GEM. GEM guided predictions obtained from model analysis provided insights into the effects of medium composition on metabolic flux distribution and gene essentiality. The model can also serve as a platform to guide network reconstructions for other Staphylococci as well as direct hypothesis generation following the integration of omics data sets, including transcriptomics, proteomics, metabolomics, and multi-strain genomic data.
| Methicillin-resistant Staphylococcus aureus (MRSA) USA300 strains have emerged as the predominant cause of community-associated infections in the United States, Canada, and Europe [1]. Today in the United States more deaths are attributed to MRSA infections than to HIV/AIDS [2,3]. USA300 was first isolated in September, 2000, and has been implicated in wide-ranging and epidemiologically unassociated outbreaks of skin and soft tissue infections in healthy individuals [4]. In 2006, the CDC reported that 64% of MRSA isolated from infected patients were of the USA300 strain type, an increase of 11.3% since 2002 [5], indicating a rapid spread throughout the country. Today, vancomycin resistance amongst S. aureus strains is on the rise, further complicating antibiotic treatment [6]. USA300 is capable of producing rapidly-progressing, fatal conditions in humans that cause a wide variety of diseases, ranging from superficial skin and soft tissue infections to life-threatening septicaemia, endocarditis, and toxic shock syndrome. Many efforts are geared towards designing new antibiotic regimens to combat MRSA. However, these endeavors are impaired by the lack of replicability in antibiotic potency and bioactivity across different media [7]. Little is known about the systems-level effects of the nutritional environment on S. aureus growth and metabolism.
While multi-omics data-sets allow for the interrogation of complex interactions occurring on a cellular level, the results can often be hard to interpret. Thus, there is a need for a common knowledge base that enables the integration of disparate data types. Genome-scale models (GEMs) of metabolism have been successfully utilized as a common platform for omics data contextualization and integration [8,9]. GEMs represent mathematically structured knowledge bases of metabolism that contain all of the molecular mechanisms known to occur in an organism. They are built through iterative curation efforts and are constantly updated to reflect the current state of knowledge pertaining to the organism [10]. The S. aureus GEM has undergone several such iterations over the past 15 years [11–14]. The more recent iterations relied more heavily on semi-automated workflows whereby annotations were pooled from online databases. Unfortunately, online databases rely on a combination of manual curation and sequence homology gene function assignment which is often not organism specific. In general, the more manual curation that goes into a GEM, the more reliable and organism-specific the GEM derived predictions are [15]. The rise of antibiotic resistance amongst S. aureus strains has created strong momentum in the field of molecular biology and many novel S. aureus-specific mechanisms have been discovered over the past decade. However, many online databases as well as the current S. aureus GEM [11] are still lagging behind and do not reflect newly uncovered metabolic capabilities.
In this work, we developed an S. aureus str. JE2 (strain LAC cured of its plasmids) GEM integrated with protein structures and used a combination of experimental data and computational methods to analyze systems-level metabolic characteristics under different growth conditions [16]. We geared our efforts towards incorporating the newly discovered molecular mechanisms and metabolic pathways of S. aureus into an updated GEM and brought the most recent S. aureus GEM through one reconstruction iteration [15]. This iteration is guided by literature findings, experimentally derived gene essentiality data, analysis of protein structures, and microarray growth phenotypes. Such efforts are valuable in that the final S. aureus G`EM is up to date with online databases, constitutes a blend of the curation efforts of several groups, and quantitatively and qualitatively recapitulates flux and growth phenotypes. We built condition-specific GEMs by integrating time-course quantitative exo-metabolomic datasets and used flux sampling and predicted gene essentiality to compare the metabolic flux state across growth conditions.
We followed an established workflow for the reconstruction of genome-scale metabolic networks [15] to curate and update the most recent genome-scale model (GEM) of S. aureus [11] with new content. The basic steps outlined in a reconstruction workflow include: Step 1: building a draft reconstruction from a genome annotation; Step 2: refining the reconstruction using literature evidence; Step 3: converting the reconstruction into a computable format; and Step 4: evaluating and validating the network against experimentally observed phenotypes [15]. We conducted detailed and extensive manual curation that brought about major modifications to the S. aureus metabolic network across 56 metabolic sub-modules (Table S1 in S2 Appendix). Our efforts were guided by a combination of literature review and network evaluation and proceeded in an iterative fashion. iYS854 contains 854 unique ORF assignments, 1,202 metabolic processes (excluding biomass and exchange reactions), 1,084 metabolic species, and 681 3D protein structures (Fig 1A, S1 Data). We also designed an updated condition-specific biomass objective function “BIOMASS_iYS_wild_type” and a general biomass objective function “BIOMASS_iYS_reduced” (Fig 1B). We enriched the objects included in the reconstruction (genes, proteins, reactions, and metabolites) with layers of metadata and cross-references (Fig 1C).
We proceeded to validate the GEM against experimental observations (step 4). In this step of the reconstruction, analyzing the discrepancies between model predictions and experimental outcomes can highlight model errors and areas of knowledge gaps. Ultimately, the systems-level view can give a deeper understanding of the organism’s metabolism and guide the generation of testable hypotheses.
Here, we demonstrate a case of integrating an omics dataset with iYS854 to analyze the effect of the addition of D-glucose to the extracellular medium on the intracellular metabolic flux state. Once a model is reconstructed and validated, a condition-specific GEM can be built by constraining the model further using values obtained from experimental measurements. A condition-specific GEM differs from the baseline GEM in that it has a reduced solution space and simulates a flux state that is more representative of the cell’s metabolic state under the tested culture conditions.
This study presents the most recent and up-to-date genome-scale metabolic reconstruction for the gram-positive pathogen S. aureus. We validated iYS854 both quantitatively and qualitatively against a variety of data sets and observed a significant improvement with respect to the starting model in both carbon catabolic capabilities as well as gene essentiality prediction. We then integrated time course quantitative exo metabolomics with iYS854 to analyze the effects of exogenous glucose on the intracellular flux distribution.
Inconsistencies of model-driven predictions with in vitro observations highlighted gaps in knowledge as well as non-essential biomass components (including cell wall components, haem, and menaquinone). Interestingly, cell wall deficient strains are involved in persistence in vivo [80], while menD mutants exhibit the small colony variant (SCV) phenotype, a phenotype known to be associated with increased persistence and resistance to antibiotics in vivo. Taken together, these results hint towards an altered biomass composition as a result of exposure to environmental stresses such as antimicrobials.
Gene essentiality predictions on synthetic physiological media and chemically defined media revealed the essentiality of purine, pyrimidine, and amino acid biosynthesis for growth under nutrient limited conditions. Nucleotide biosynthesis, which was predicted to be essential in RPMI and SNM3, has been shown to be essential for growth in blood for a variety of bacteria including S. aureus, E. coli, Salmonella, and B. anthracis [72,81]. Additionally, iYS854 predicts amino acid biosynthesis to be essential in SNM3 (a synthetic nasal medium), and our results were confirmed experimentally by Krismer et. al [53].Together, these findings point towards putative antibiotic targets for the treatment of bloodstream and nasal infections.
Elevated concentrations of blood glucose is common across diabetic patients and S. aureus is the most frequently isolated and virulent pathogen from diabetic foot infections [82]. Condition-specific models revealed drastic systems-level differences in flux distributions across strains when they were exposed to D-glucose. For example, despite the availability of amino acids in the medium, S. aureus was predicted to utilize both extracellular glucose and amino acids towards protein production in order to satisfy its biomass requirements. This finding is supported by the observation that genes involved in amino acid biosynthesis are highly expressed when D-glucose is added to the medium [55] and hints towards a kinetic constraint favoring the uptake and utilization of extracellular glucose over that of extracellular amino acids towards amino acid biosynthesis. Indeed, S. aureus strains express four glucose transporters suggesting that together, they can induce high levels of glycolytic flux [83].
The addition of glucose to the medium induced significant metabolic rewiring, with production of ATP switching from the Krebs cycle to overflow pathways as evidenced by the large acetate secretion rate. Significantly higher glycolytic fluxes were predicted in the presence of glucose. Aerobic fermentation also occurs in E. coli when the glucose consumption rate is large, and the cell cannot reoxidize reduced equivalents at a sufficient rate [84]. Importantly, glycolytic activity exhibited by S. aureus strains has been shown to induce hypoxia inducible factor 1α signalling and promote the proinflammatory response to infection [85]. The absolute consumption of oxygen was predicted and experimentally shown to be higher in the presence of glucose (as confirmed by experimental evidence [39]). We also predicted an elevated flux through the PPP in the presence of glucose, which was confirmed by an experimentally observed higher NADP/NADPH ratio [55]. In agreement with our predictions, the inactivation of the TCA cycle was found to cause an increase in the carbon flow across the PPP in S. aureus [86]. Here we show that the generation of NADPH is mediated by both the PPP and the TCA cycle and that the increased flux through the PPP compensates for the decreased flux in the TCA cycle.
Our results demonstrate that the updated S. aureus GEM, iYS854, accurately captures experimentally measured differences in central metabolism in the presence and absence of glucose and that the importance of metabolic modules changes drastically under different in silico physiological growth media. This study is a first step towards understanding the systems-level metabolic response of S. aureus to differing media compositions from a constraint-based modeling perspective.
A draft core S. aureus GEM was built by taking the common reactions between the E. coli core GEM [87] and the starting S. aureus GEM [11] and adopting the E.coli core biomass objective function (BIOMASS_Ecoli_core_w_GAM) [88]. We curated the network after reviewing literature using the COBRApy toolbox (Tables S1, S2 and S3 in S2 Appendix) [89]. Modifications included reaction, gene and metabolites addition/removal, and annotations of reactions and genes with confidence scores, references and metadata. We assigned confidence scores as per the standards set by Thiele et. al [90] and novel instance IDs as per BiGG standards [91]. We downloaded the genomic sequence for S. aureus str. JE2 from NCBI (accession number CP020619.1). Genes were updated with names as assigned in the literature (when available) or as generated during automatic genome annotation. We added the E.C. numbers obtained from the genome annotation as metadata to the modelled genes. We then downloaded the S. aureus Swiss-prot knowledge base (which contains manually reviewed proteins and protein metadata specific to S. aureus) and cross-referenced the modelled genes with Swiss-prot IDs using bi-directional best BLAST (PID > 80%, e-value< 10−3) [92,93].
One of the crucial steps involved in a reconstruction is the evaluation of the network flux carrying capability. In addition to ensuring the successful production of biomass precursors, we examined some general properties of the flux distribution. The starting model could not simulate flux through the full TCA cycle and erroneously simulated the dissipation of 13 energy carriers when all exchanges were closed including ATP, CTP, GTP, UTP, ITP, NADH, NADPH, FADH2, FMNH2, MQH7, acetyl-coa, L-glutamate and intracellular proton. Such an aberration is commonly found to be caused by a set of reactions constituting together erroneous stoichiometrically balanced energy generating cycles (ECGs) [23]. To search for energy generating cycles we followed the workflow established by Fritzemeier et. al [94]. Briefly, we blocked all extracellular exchanges by constraining the upper and lower bounds to 0 and iteratively added 14 energy dissipation reactions (S4 Table in S2 Appendix). An energy dissipation reaction is a reaction that consumes high energy metabolites. We simulated maximal flux through one dissipation reaction at a time using flux balance analysis (FBA) [95]. Energy generating cycles existed when the maximal flux through any energy dissipation reaction was larger than 0. We found that EGCs were caused by: 1) sets of reactions carrying out the same function but with inverted reversibility, 2) the inclusion of reactions that are not known to occur in S. aureus nor have any genetic basis for their inclusion (such as 2-oxoglutarate synthase and fumarate reductase allowing reductive TCA), and 3) reversible reactions that could generate energy carrying moieties when the flux was running in the reverse direction. As a result of removing and adjusting the network accordingly, iYS103 successfully simulated flux through the TCA cycle and could not freely charge any of the 13 high energy carriers (Fig S1 in S1 Appendix). The final core network contains 103 unique ORF assignments, 70 metabolic processes and 58 metabolic species and can successfully simulate the utilization of the Krebs cycle.
Reactions were annotated with COG subsystems following the same classification scheme as previous GEM reconstructions [70,91]. The subsystems consisted of: 1) amino acid metabolism, 2) carbohydrate metabolism, 3) cell wall and membrane metabolism, 4) cofactor and prosthetic group metabolism, 5) energy production and conversion, 6) transport, 7) nucleotide metabolism, 8) lipid metabolism and 9) inorganic ion transport and metabolism. We also added as a note to each metabolic reaction the metabolic sub-module that it is described to participate in throughout the literature. We annotated metabolic reactions with 65 metabolic sub-modules. To visualize the amount of novel content added to each metabolic subsystem, we compared the updated metabolic gene content with the metabolic gene content across the 4 previous metabolic reconstructions. Genes were then classified in sub-modules according to the metabolic reactions they participate in. For each sub-module, a fraction representing the ratio of novel genes to the total number of genes it contains was computed. A gene was considered “novel” when it was not accounted for in the previous reconstruction (Fig 2A).
The structural systems biology (ssbio) pipeline was run to map crystallized 3-dimensional structures of proteins deposited in the Protein Data Bank (PDB) to the genes included in the genome scale reconstruction [48,49]. A blast cutoff was chosen at 70%. Genes that could not be mapped through this method to a crystal structure were mapped to their nearest homolog with an existing structure (S4 Table in S2 Appendix). Homology models were built from this template and subsequently modified to match the amino acid sequence of the USA300 query protein (S5 Table in S2 Appendix).
We adapted the weight fractions for the 5 polymer categories and the pool of solutes from Heinemann et. al [12]. The authors computed a biomass composition by averaging experimentally derived weight fractions across several S. aureus strains grown in different media conditions. We proceeded to compute the relative ratios of the DNA precursors using the S. aureus genomic sequence and the RNA and protein weight fractions using transcriptomics data derived for S. aureus str. plasmid cured LAC (JE2) grown on a chemically defined medium with galactose as the main gluconeogenic nutrient source [96]. Computations were performed via BOFdat, a python package for biomass objective function derivation [51]. We included amino acids in their tRNA bound form because two of the twenty amino acids are only synthesized while complexed with tRNA [97]. The relative quantities for the cell wall precursors and lipids were adapted again from Heinemann et. al. However, the updated metabolic network includes the biosynthesis of downstream precursors for some of the cell wall precursors. For example, we replaced the peptidoglycan monomer with a wall teichoic acid bound peptidoglycan dimer, and lipoteichoic teichoic acids with charged lipoteichoic acids. We adjusted the relative coefficients according to the replaced precursor’s molecular weight. The pool of solutes was adapted from [52] and updated with metals and trace molecules (chosen based on literature evidence [98] and protein cofactor utilization obtained from the metadata associated with the 3-D protein structures; S7 Table in S2 Appendix).
We modelled growth on a defined medium by setting the lower bound to the reactions exchanging metabolites that are present in the medium to -10 mmol/gDW/hr. A negative value signifies exchange from the medium to the cell. The lower bound to all other exchanges was set to 0 mmol/gDW/hr. The simulated media types are available in S8 Table in S2 Appendix. When growth could not be achieved, we searched for minimal medium supplementations needed to support growth. For that purpose, we changed the objective of the optimization problem to the minimization of the number of additional open exchange reactions and constrained flux through the biomass objective function to a minimal value of 1 hr−1 (S9 Table in S2 Appendix). We set the lower bound to all exchange reactions to -10, and the solver configuration tolerance feasibility to 10−9 using COBRApy.
Aerobic environments were simulated by setting the lower bound for oxygen exchange to -20 mmol/gDW/hr. Oxygen exchange was blocked to simulate an anaerobic growth environment. The utilization of nitrate as an alternative electron acceptor was simulated by setting the lower bound for nitrate exchange to -20 and the lower bound for oxygen exchange to 0.
Model benchmarking on carbon sources was performed using Biolog plates PM1 and PM2 (BIOLOG Inc. Hayward, CA). The recommended protocol was followed as described by (Zuniga et al., 2016), with the following modifications. S. aureus USA300 was grown to mid-log phase in modified TSB media, pelleted via centrifugation at 4,000 x g for five minutes, washed and resuspended in fresh media to a final OD = 0.1. Aliquots of 100 uL were inoculated into Biolog plates and examined in the plate reader at time zero, then each hour from 1–12 h, and finally at 24 h. Plates were housed in a plate reader under sterile conditions. The plates for both the PM1 and PM2 plate (carbon sources) were run at 490 nm to examine dye absorbance alterations and 750 nm to assess optical density. M9 minimal medium supplemented with niacin and thiamin was used as the minimal medium to simulate for the utilization capability of 68 carbon sources. The simulation results were then compared against experimental observations (S10 Table in S2 Appendix).
The predicted mutant growth phenotypes were obtained by simulating a gene knockout using the cobra.flux_analysis.single_gene_deletion command. The mutants were cultured on tryptic soy broth (TSB, a rich and complex medium for which the composition is unknown) and the observed gene essentiality for this condition was reported. To mimic TSB, we simulated growth by allowing inward flux of all the extracellular nutrients included in the reconstruction. We set the objective function to BIOMASS_iYS_reduced (which excludes the pool of measured intracellular solutes detected by NMR for growth of S. aureus on CDM+glucose). A gene was deemed to be essential when its knockout resulted in a maximal growth of less than 0.0001 hr^-1 or when the solution status was not optimal (S11 Table in S2 Appendix).
To interrogate the capability of iYS854 to recapitulate the mannitol fermentation capability across mutants, we first confirmed that the model could simulate growth on mannitol in an oxygen depleted environment by allowing uptake of mannitol, M9 minimal medium components, thiamin and niacin. Extracellular oxygen exchange was blocked to mimic the anaerobic environment. We subsequently assessed gene essentiality by using the cobra.flux_analysis.single_gene_deletion command and compared the results against experimental observation. In order to assess the GEM’s capability to predict pigment formation, we set the production of staphyloxanthin as the objective of the maximization problem. Growth on rich medium was then simulated by allowing inward flux across all exchanges. Again, we determined gene essentiality to assess the effect of gene knockouts on the production of staphyloxanthin (S12 Table in S2 Appendix). Gene essentiality on all other media types was determined by setting the lower bound to exchange reactions to -10 when they imported a metabolite that was present in the medium (S13 Table in S2 Appendix).
Single colonies of S. aureus str. LAC were inoculated into 5 mL Roswell Memorial Park Institute (RPMI) 1640 supplemented with 10%LB (RPMI+10%LB) and incubated overnight at 37oC with rolling. Overnight cultures were diluted into tubes containing 18 mL fresh media to a starting OD600 0.01 and incubated at 37oC with stirring until cultures OD600 0.4. Precultures were diluted back into 6 new tubes, containing 20 mL fresh media to OD600 0.01 and growth was monitored until cultures reached OD600 0.5. The 6 tubes were mixed in baffled flask and OD600 was taken. Preweighed 0.2 μm filter was placed into a clean glass filter holder above. 40 mL culture was passed through filter and unit was washed with 15 mL ddH2O. A final OD600 reading was taken from remaining culture. Filter disc was transferred to a clean petri dish placed in incubator at 80oC ON. The next day, filter discs were acclimated to room temperature for 45 min and reweighed. Dry cell weights were taken as the average of three weight measurements. We obtained an average dry cell weight of 9.6 mg at an OD600 of 0.58 (S14 Table in S2 Appendix).
Hasley et. al reported absolute concentration measurements for extracellular ammonium, acetate, glucose and all 18 amino acids and complemented these measurements with corresponding time-course OD readings. We calculated the growth rate and uptake rates in both conditions as specified below:
SUR=μ×m,
(5)
μ=slope(loggDW,t)fort∈(0,2,4,6,8)
(6)
m=slope([X],gDW)
(7)
Where [X] is the set of concentration measurements across t in mmol/L, gdW is the gram dry weight in g/L, t is the time in hours, μ is the growth rate, SUR is the substrate uptake rate in mmol/t/gDW.
For each condition, we verified that the uptake and secretion rates were mass balanced and that the overall flux of elements going towards biomass production (i.e. the total influx minus the total outflux) is larger than the total flux of elements needed to support biomass production at the experimentally measured growth rate.
Where Ne,i is the base ratio for element e in the metabolic structure for nutrient i, SURi is the substrate uptake rate for nutrient i, bk is the relative coefficient for the biomass precursor k, Pe,k is the base ratio of element e in the metabolic structure of the biomass precursor k.
We proceeded to build two condition-specific GEMs by constraining the reactions exchanging the extracellular nutrients to +/- 10% of the measured corresponding calculated uptake and secretion rates (S15-S17 Tables in S2 Appendix). We subsequently ran flux balance analysis (FBA) to simulate maximal biomass production.
We calculated a theoretical growth associated maintenance by changing the objective for both condition-specific GEMs to ATP production (i.e., the objective coefficient for the ATP maintenance reaction was set to 1). The maximal flux through the ATP maintenance reaction obtained was 1.93 mmol/gDW/h for the CDM-constrained GEM and was 12.64 mmol/gDW for the CDMG-constrained GEM. As a result, we obtained a growth associated maintenance of 9.59 mmol/gDW/h (which is the slope of the line obtained from plotting maximal flux through ATPM against growth rate) and a non-growth associated maintenance of −5.87 mmol/gDW. Since non-growth associated maintenance should be positive, we hypothesize that data-sets {vATPM,μ} covering a larger range of conditions (e.g. anaerobic conditions as well as alternate carbon/nitrogen sources) coupled with measurements of CO2 secretion rates is needed to lead to more accurate values. We instead used the GAM and NGAM values experimentally obtained for E. coli and iteratively computed maximal growth for decreasing percentages of the initial value. For each condition, the GAM and NGAM value is chosen when the simulated maximal growth corresponds to the observed experimental growth rate (S1 Appendix).
We sampled the steady state flux space a total of 10,000 times using the cobra.flux_analysis.sample() command from the cobrapy package. To obtain a proxy for the predicted relative intracellular ATP concentration we calculated the ratio of the sum of all metabolic fluxes producing ATP across both condition:
rATPCDMCDM+glucose=median(∑i=1naATP*vi)s1median(∑j=1maATP*vj)s2∀i∈R1,∀j∈R2,∀s1∈S1,∀s2∈S2
(9)
where R1 are the set of reactions yielding ATP and S1 is the set of 10,000 sampled fluxes in the CDM-specific GEM, R2 is the set of reactions yielding ATP and S2 is the set of 10,000 sampled fluxes in the CDMG-specific GEM, aATP is the stoichiometric coefficient for ATP in reaction i, and vi is the calculated flux in mmol/gDW/hr through reaction i, and the median is taken across 10,000 samples.
Similarly, we computed a proxy for the relative oxygen level by computing the relative flux for the oxygen exchange:
rO2CDMCDM+glucose=median(vO2)s1median(vO2)s2∀s1∈S1,∀s2∈S2
(10)
Where vO2,i is the flux through EX_o2_e in mmol/gDW/hr and the median is taken across 10,000 sampled fluxes.
Flux balance analysis was run in both conditions and the fluxes were sampled 10,000 times. All reaction fluxes were normalized by dividing by the growth rate to account for growth differences across the two media types. The flux distribution for each metabolic process was compared across both conditions using the Kolmogorov-Smirnov test, a non-parametric test which compares two continuous probability distributions. For this purpose we used the command scipy.stats.ks_2samp from the scipy package. The distribution across two reactions was deemed to be significantly different when the Kolmogorov-Smirnoff statistic was larger than 0.99 with an adjusted p-value < 0.01. We proceeded to plot the reactions highlighted in this process in Fig 5.
|
10.1371/journal.pgen.1006634 | Epilepsy-associated gene Nedd4-2 mediates neuronal activity and seizure susceptibility through AMPA receptors | The neural precursor cell expressed developmentally down-regulated gene 4–2, Nedd4-2, is an epilepsy-associated gene with at least three missense mutations identified in epileptic patients. Nedd4-2 encodes a ubiquitin E3 ligase that has high affinity toward binding and ubiquitinating membrane proteins. It is currently unknown how Nedd4-2 mediates neuronal circuit activity and how its dysfunction leads to seizures or epilepsies. In this study, we provide evidence to show that Nedd4-2 mediates neuronal activity and seizure susceptibility through ubiquitination of GluA1 subunit of the α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid receptor, (AMPAR). Using a mouse model, termed Nedd4-2andi, in which one of the major forms of Nedd4-2 in the brain is selectively deficient, we found that the spontaneous neuronal activity in Nedd4-2andi cortical neuron cultures, measured by a multiunit extracellular electrophysiology system, was basally elevated, less responsive to AMPAR activation, and much more sensitive to AMPAR blockade when compared with wild-type cultures. When performing kainic acid-induced seizures in vivo, we showed that elevated seizure susceptibility in Nedd4-2andi mice was normalized when GluA1 is genetically reduced. Furthermore, when studying epilepsy-associated missense mutations of Nedd4-2, we found that all three mutations disrupt the ubiquitination of GluA1 and fail to reduce surface GluA1 and spontaneous neuronal activity when compared with wild-type Nedd4-2. Collectively, our data suggest that impaired GluA1 ubiquitination contributes to Nedd4-2-dependent neuronal hyperactivity and seizures. Our findings provide critical information to the future development of therapeutic strategies for patients who carry mutations of Nedd4-2.
| Many patients with neurological disorders suffer from an imbalance in neuronal and circuit excitability and present with seizure or epilepsy as the common comorbidity. Human genetic studies have identified many epilepsy-associated genes, but the pathways by which those genes are connected to brain circuit excitability are largely unknown. Our study focused on one of the epilepsy-associated genes, Nedd4-2, and aimed to dissect the molecular mechanism underlying Nedd4-2-associated epilepsy. Nedd4-2 encodes a ubiquitin E3 ligase. Several neuronal ion channels have been identified as its substrates, including the GluA1 subunit of AMPAR. Our results first demonstrate up-regulation of spontaneous neuronal activity and seizure susceptibility when Nedd4-2 is reduced in a mouse model. These deficits can be corrected when GluA1/AMPAR is pharmacologically or genetically inhibited. In addition, we found that three epilepsy-associated missense mutations of Nedd4-2 inhibit the ubiquitination of GluA1 and fail to reduce GluA1 surface expression or spontaneous neuronal activity when compared to wild-type Nedd4-2. These findings suggest the reduction of GluA1 ubiquitination as a crucial deficit underlying insufficient function of Nedd4-2 and provide critical information to the development of therapies for patients who carry mutations of Nedd4-2.
| A hyperactive brain circuit is a common abnormality observed in patients with various neurological and psychiatric disorders, including epilepsies (1). Evidence from human genetic studies implicates genes encoding ion channels or their regulators in the etiology of those pathophysiological conditions [1, 2]. Characterizing those genes and their function in regulation of brain circuit activity is likely to reveal novel therapeutic targets for these diseases. One of those genes is the neural precursor cell expressed developmentally downregulated gene 4-like (Nedd4-2) [3]. Nedd4-2, is an epilepsy-associated gene containing at least three missense mutations identified through genomic mutation screening in patients with epilepsy [3–6]. Nedd4-2 encodes a ubiquitin E3 ligase that belongs to the Nedd4 family of ubiquitin E3 ligases [7] but is the only member encoded by an epilepsy-associated gene [3]. Because of an N-terminal lipid-binding domain, Nedd4-2 has high affinity toward binding and ubiquitinating membrane proteins [8]. Several neuronal membrane substrates of Nedd4-2 have been identified, such as voltage-gated sodium channel Nav1.6 [9], voltage-gated potassium channels Kv7/KCNQ [10–12], neurotrophin receptor TrkA [13, 14] and the GluA1 subunit of the α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid receptor (AMPAR) [15]. Our previous work has demonstrated elevated seizure susceptibility in mice when Nedd4-2 is knocked down [16]. However, the mechanisms by which the dysfunction of Nedd4-2 contributes to epileptogenesis are unclear. Presumably wild-type (WT) Nedd4-2 mediates or represses circuit activity by ubiquitinating one or more of its substrates while the epilepsy-associated mutants fail to do so and lead to seizures and/or epilepsies. To test this possibility, we needed to identify the relevant substrate of Nedd4-2 in regulation of neuronal excitability and characterize the effect of epilepsy-associated mutations on substrate recognition.
The AMPAR is a major subtype of ionotropic glutamate receptors and is the most commonly found receptor in the mammalian nervous system [17, 18]. AMPARs are assembled as homo-or hetero-tetramers and are comprised of combinations of GluA1–GluA4 subunits [19]. Each subunit has a non-conserved C-terminal, an intracellular domain that harbors regulatory elements subject to various post-translational modifications such as ubiquitination. All four AMPAR subunits can be ubiquitinated, but only GluA1 ubiquitination has been specifically described upon different activity stimulations [20, 21]. Studies have shown that GluA1 ubiquitination contributes to its internalization [22, 23]. This internalization, part of AMPAR trafficking mechanisms [24], is critical for synaptic depression as well as homeostatic regulation of synaptic strength [25–28]. Because GluA1-GluA2 is the predominant AMPAR heteromer [29], and GluA1 is required for successful trafficking and targeting of GluA2 [30], we hypothesized that Nedd4-2 is required for limiting GluA1 surface expression and functionality of AMPAR. Because GluA1 levels affect neuronal activity [31], and dysregulation of AMPARs has been shown to be linked to epilepsy [32, 33], Nedd4-2 may play a role in affecting neuronal activity, seizures, and/or epilepsy through fine-tuning of AMPARs.
In our current study, we provide in vitro and in vivo evidence to demonstrate GluA1- and AMPAR-dependent elevation of neuronal activity and seizure susceptibility induced by functional insufficiency of Nedd4-2. To our knowledge, our findings provide the first mechanism underlying Nedd4-2-associated circuit hyperactivity and seizures and open up a new avenue for the development of therapeutic strategies to potentially treat epileptic patients who carry Nedd4-2 mutations.
It is unknown how dysregulation of Nedd4-2 is involved in seizures or epilepsies. To answer this question, we employed a mouse model, Nedd4-2andi, in which the long-form (isoform 1) of Nedd4-2 is selectively deleted due to a spontaneous mutation in exon-2 (Fig 1A). Because Nedd4-2 knockout mice are not viable [34], Nedd4-2andi serves as an ideal, in vivo knockdown model to study Nedd4-2. To assess the question of whether Nedd4-2 mediates neuronal activity, we employed a multielectrode array (MEA) recording system (S1 Fig) to record extracellular spontaneous spikes of electrical activity in primary cortical neuron cultures prepared from WT or Nedd4-2andi mice. As shown in Fig 1B, the frequency of spontaneous spikes was significantly elevated in Nedd4-2andi cultures in comparison to WT cultures. The average spontaneous spike amplitude did not differ between WT and Nedd4-2andi cultures (Fig 1C). These data indicate that spontaneous neuronal activity is basally elevated in Nedd4-2andi cortical cultures.
To determine whether elevated spontaneous neuronal activity in Nedd4-2andi cultures is accompanied by altered synaptic transmission, we performed whole-cell patch-clamp recording to obtain miniature excitatory post-synaptic current (mEPSC) from WT or Nedd4-2andi cortical neuron cultures. As shown in Fig 1D, Nedd4-2andi neurons exhibit elevation of both mEPSC amplitude and frequency when compared to WT neurons. These data suggest that Nedd4-2 likely mediates both pre- and post-synaptic properties, and the elevation of spontaneous neuronal activity observed in Nedd4-2andi cortical cultures (Fig 1C) is likely contributed by multiple factors.
We previously identified the GluA1 subunit of AMPAR as a substrate of Nedd4-2 [15]. We therefore asked whether AMPAR mediates spontaneous neuronal activity and whether it is responsible for the hyperactivity observed in Nedd4-2andi cultures. An AMPAR agonist, AMPA (1 μM), was applied to determine how spontaneous neuronal activity was affected when AMPAR was activated. MEA recordings from WT cultures before and after AMPA treatment for 15 min indicated elevated spontaneous spike frequency (Fig 2A1 and 2A3) suggesting that spontaneous neuronal activity can be modulated by AMPARs. The same treatment produced significant, but smaller, effects on Nedd4-2andi cultures (Fig 2A2 and 2A3; Significant interaction between treatment and genotype was detected, p<0.05.). The average of spontaneous spike amplitude (Fig 2B) and electrode burst activity did not differ after AMPA treatment for either genotype (S2A Fig). These data suggest that Nedd4-2 contributes to the elevation of spontaneous neuronal activity, particularly spontaneous spike frequency, when the AMPAR is activated.
We then asked whether Nedd4-2andi cultures respond differently when AMPARs are pharmacologically inhibited. We employed a specific AMPAR antagonist, NBQX (2 μM), and again recorded the spontaneous neuronal activity before and after a 15-min treatment. As shown in Fig 3A, NBQX slightly, but not significantly, reduced spontaneous spike frequency in WT cultures, while NBQX significantly reduced spike frequency in Nedd4-2andi cultures (Significant interaction between treatment and genotype was detected, p<0.05.). Again, average spontaneous spike amplitude did not differ between either treatments or genotypes (Fig 3B). Interestingly, we observed elevated burst activity without changes in interburst interval (IBI) after NBQX treatment (S2B Fig), which was also observed in another study [35]. Although we suspect the lack of changes in IBI is potentially because NBQX was applied acutely, we did not pursue further experiments with prolonged treatments since our data suggest that NBQX-induced burst activity is independent of Nedd4-2 (S2B1 and S2B2 Fig).
We previously showed elevated synchrony of spontaneous neuronal activity in Nedd4-2andi cultures [16]. Elevation of synchronized activity indicates potentially elevated network activity. To determine whether AMPAR is involved in this phenomenon, we analyzed the synchrony index in WT and Nedd4-2andi cultures treated with either AMPA or NBQX. As shown, although AMPA treatment elicits some effect toward elevation of synchrony, no difference was observed between genotypes. NBQX, on the other hand, produces no effect on synchrony in either WT or Nedd4-2andi cultures (S3 Fig). These results suggest that, although prolonged stimulation of AMPAR might further elevate the synchrony of neuronal activity, the effect is unlikely to be Nedd4-2-dependent. Furthermore, AMPAR is also unlikely to be responsible for basally elevated synchrony when Nedd4-2 is compromised [16] since NBQX exerts no effect. Therefore, whether and how Nedd4-2 mediates synchrony of spontaneous neuronal activity or other network activity, such as network spikes and bursts, through other substrates would be an important future direction.
In summary, we showed that Nedd4-2andi cultures were less sensitive to AMPAR activation but very sensitive to AMPAR blockade with regard to spontaneous spike frequency. These results suggest that altered GluA1/AMPAR signaling in Nedd4-2andi mice contributes to basally elevate spontaneous neuronal activity.
We have previously demonstrated that Nedd4-2andi mice exhibit greater seizure susceptibility induced by systematic administration of kainic acid, a potent agonist for kainate-class ionotropic glutamate receptors that is widely used to induce seizures in animal models [16, 36]. Greater sensitivity to AMPAR blockade in reducing spontaneous neuronal activity, as seen in Fig 3, led to our hypothesis that elevated GluA1 level contributes to elevated seizure susceptibility in Nedd4-2andi mice. To test this hypothesis, WT or Nedd4-2andi mice were crossed with GluA1 knockout mice to obtain the following four genotypes: 1) Nedd4-2wt/wt GluA1+/+;, 2) Nedd4-2wt/wt GluA1+/-, 3) Nedd4-2andi/andi GluA1+/+, and 4) Nedd4-2andi/andi GluA1+/-. As shown in Fig 4A, GluA1 levels in Nedd4-2wt/wt GluA1+/- and Nedd4-2andi/andi GluA1+/- mice were reduced by 31% and 46%, respectively, when compared to their control littermates (Nedd4-2wt/wt GluA1+/+ and Nedd4-2andi/andi GluA1+/+, respectively). Most importantly, the GluA1 level in Nedd4-2andi/andi GluA1+/- mice was similar to Nedd4-2wt/wt GluA1+/+ mice (Fig 4A). We then determined seizure susceptibility in these mice by intraperitoneal injections of kainic acid. Four-week-old mice were injected with kainic acid (15, 30, or 60 mg/kg) as done in our previous study [16]. Behavioral seizures were monitored and scored during a 1-hr observation period. As shown, Nedd4-2andi mice (Nedd4-2andi/andi GluA1+/+; Fig 4C, left panel) had enhanced seizure response in comparison to WT mice (Nedd4-2wt/wt GluA1+/+; Fig 4B, left panel). Reducing GluA1 levels in Nedd4-2andi mice (Nedd4-2andi/andi GluA1+/-; Fig 4C, right panel) significantly reduced seizure response in comparison to control Nedd4-2andi mice (Nedd4-2andi/andi GluA1+/+; Fig 4C, left panel). Furthermore, reducing GluA1 level in Nedd4-2andi mice produced a seizure response similar to WT mice (Nedd4-2wt/wt GluA1+/+; Fig 4B, left panel). A slight, but not significant, reduction of seizure response was also observed when GluA1 level was reduced in WT mice (Nedd4-2wt/wt GluA1+/-; Fig 4B, right panel). In conclusion, our results indicate that genetically reducing GluA1 level is able to correct elevated seizure susceptibility caused by insufficient function of Nedd4-2 in Nedd4-2andi mice.
Our data suggest that impaired GluA1 ubiquitination may be responsible for Nedd4-2-mediated seizure and/or epilepsies. To test this hypothesis, we sought to characterize the functional consequence of GluA1 ubiquitination by Nedd4-2. We first attempted to map the Nedd4-2-ubiquitinated residues of GluA1 as ubiquitination at different residues may affect the function of GluA1 differently [20, 21]. We employed human embryonic kidney (HEK) cells because they do not express a detectable level of GluA1 or Nedd4-2 endogenously (S4 Fig). There are four lysine residues (K813, K819, K822, and K868) located on the carboxyl-terminal, intracellular domain of GluA1 (Fig 5A) [21–23]. As reported previously, mutating all four residues completely abolishes GluA1 ubiquitination; this rules out other lysine residues as targets [21]. Accordingly, WT Nedd4-2 was then co-transfected with WT GluA1 or mutant GluA1s in which each lysine was replaced by arginine (R) at each individual site (K813R, K819R, K822R and K868R) or all four lysine residues together were mutated to R (4KR). As shown in Fig 5B, the GluA1 with either K868R or all four lysine residues mutated to arginine (4KR) showed significantly reduced ubiquitination when co-expressed with Nedd4-2. A trend toward reduced ubiquitination is also observed when GluA1 carries K822R, suggesting a potential alternative residue for Nedd4-2-mediated ubiquitination. In summary, these results suggest that K868 is the most critical residue ubiquitinated by Nedd4-2.
GluA1 ubiquitination at K868 has been shown to affect its surface expression [22, 23]. To determine whether Nedd4-2 mediates surface expression of GluA1, we labeled surface proteins with biotin in WT or Nedd4-2andi cortical neuron cultures followed by purification of biotinylated proteins with streptavidin beads. As shown in Fig 5C, Nedd4-2andi cultures indeed exhibited elevated surface GluA1 when compared with WT cultures. N-cadherin serves as a control and did not differ between WT or Nedd4-2andi cultures. These results confirm the role of Nedd4-2 in limiting GluA1 surface expression. Because elevated surface GluA1 level has been linked to enhanced seizure susceptibility [37, 38], our findings further support our hypothesis that altered GluA1 ubiquitination contributes to Nedd4-2-associated seizures and/or epilepsies.
There are three epilepsy-associated missense mutations of Nedd4-2 (S233L, E271A, and H515P) identified in patients with epilepsies [4, 5]. Based on our findings, we aimed to test the hypothesis that one or more of these mutations could disrupt GluA1 ubiquitination. To avoid potential inference from other neuronal E3 ligases for GluA1 [39, 40], we applied reconstitutive systems to determine GluA1 ubiquitination using either HEK cells or in vitro ubiquitination. When using HEK cells co-transfected with GluA1 and Nedd4-2, we found that GluA1 is less ubiquitinated when co-expressed with any of the Nedd4-2 mutants in comparison to WT Nedd4-2 (Fig 6A and S5 Fig). When in vitro ubiquitination was performed using recombinant full-length GluA1 with WT or any of the expressed Nedd4-2 mutants (partially purified from HEK cells), we also found that in comparison to WT Nedd4-2, all three mutant Nedd4-2s exhibited reduced ability to ubiquitinate GluA1 in vitro (Fig 6B).
Previously we showed that Nedd4-2-mediated-GluA1 ubiquitination leads to degradation of GluA1 [15]. To determine whether mutant Nedd4-2s fail to degrade GluA1, HEK cells were co-transfected with GluA1 and WT or mutant Nedd4-2. Using cycloheximide (100 μg/ml) to inhibit protein translation and follow protein degradation, significant GluA1 down-regulation is only observed when co-expressed with WT Nedd4-2 but not with any of the mutant Nedd4-2s (Fig 6C). Slightly enhanced levels of GluA1 or Nedd4-2 after cycloheximide treatment were observed in some groups after normalization with the internal control Tubulin. This is due to a lower turnover rate of GluA1 or Nedd4-2 than that of Tubulin when cells were transfected with mutant Nedd4-2s. To strengthen the idea that GluA1 degradation is altered when co-expressed with mutant Nedd4-2s, HEK cells co-transfected with GluA1 and WT or mutant Nedd4-2 were treated with proteasome inhibitor MG132 (10 μM) (S6 Fig). GluA1 significantly accumulates when co-expressed with WT Nedd4-2, but not with any of the mutant Nedd4-2s, after MG132 treatment. Altogether, our data suggest that all three missense mutations disrupt Nedd4-2-mediated GluA1 ubiquitination and degradation.
When expressing WT or mutant Nedd4-2s in HEK cells, it was observed that the mutant Nedd4-2s exhibited increased basal levels and reduced degradation when compared with WT Nedd4-2 (Fig 6C3 and S6 Fig), suggesting enhanced stability. Because Nedd4-2 can self-ubiquitinate, the reduced down-regulation of both GluA1 and Nedd4-2 suggests that these missense mutations most likely affect the general ubiquitination process mediated by Nedd4-2 [41, 42]. Furthermore, all three mutations are located on or near one of the three protein-protein interaction domains (WW domain) in Nedd4-2 (S7 Fig), suggesting potentially altered interaction with its interacting proteins. To test this possibility, we studied the adaptor protein 14-3-3, which directly interacts with Nedd4-2 and has been shown to mediate Nedd4-2’s substrate recognition [43–45]. We first aimed to determine whether 14-3-3 mediates Nedd4-2-mediated GluA1 ubiquitination. In vitro ubiquitination using recombinant GluA1 and Nedd4-2 yielded some GluA1 ubiquitination (Fig 7A1, lane 1). Remarkably, in the presence of recombinant 14-3-3ε, one of the 14-3-3 isoforms known to interact with Nedd4-2 [46], GluA1 ubiquitination was significantly enhanced (Fig 7A1, lane 2). To validate the role of 14-3-3, a peptide-based general 14-3-3 inhibitor, R18 trifluoroacetate (R18; 0.025 mg/ml), which is known to disrupt the interaction between 14-3-3 and its binding partners, was used [47–49]. As predicted, R18 reduced GluA1 ubiquitination to a level similar to Nedd4-2 alone (Fig 7A1, lane 4). For controls, the same reactions in the absence of either Nedd4-2 or ubiquitin showed nearly undetectable GluA1 ubiquitination (Fig 7A1, lanes 5–8). These data suggest that, while Nedd4-2 is capable of ubiquitinating GluA1 in the absence of 14-3-3, 14-3-3 significantly facilitates this ubiquitination.
We then asked whether reduced GluA1 ubiquitination by epilepsy-associated missense mutations of Nedd4-2 occurred through altered interactions with 14-3-3. WT or mutant Nedd4-2 was transfected into HEK cells. Co-immunoprecipitation showed that all three mutants have reduced interactions with endogenous 14-3-3 in HEK cells (Fig 7B). Similar results were obtained when using recombinant 14-3-3ε to immunoprecipitate WT or mutant Nedd4-2 expressed in HEK cells (S8 Fig). To determine whether the Nedd4-2 mutants fail to respond to 14-3-3 when ubiquitinating GluA1, in vitro ubiquitination using recombinant GluA1 and WT or mutant Nedd4-2 partially purified from HEK cells was performed. While WT Nedd4-2 strongly ubiquitinated GluA1 and responded to additional 14-3-3ε with further GluA1 ubiquitination, all of the mutant Nedd4-2s failed to do so (Fig 7C). Because the level of 14-3-3 does not seem to be regulated by Nedd4-2 (S9 Fig), our results suggest that the epilepsy-associated missense mutations of Nedd4-2 disrupt GluA1 ubiquitination, at least partially through reduced interaction with 14-3-3.
Because the epilepsy-associated missense mutations of Nedd4-2 disrupt GluA1 ubiquitination and degradation, we hypothesize that these mutations fail to mediate surface GluA1 and spontaneous neuronal activity. To this end, we performed surface protein biotinylation to obtain and measure surface GluA1 from WT cortical neuron cultures lentivirally transduced with WT or mutant Nedd4-2 for 5 days. Surprisingly, no significant effect was observed (S10 Fig). We suspect that the level of Nedd4-2 during early development might reach a threshold in WT cultures, and therefore expression of additional Nedd4-2 fails to elicit significant effects. Furthermore, because mutant Nedd4-2s exhibit significantly reduced affinity toward interacting with 14-3-3 and therefore GluA1 (Figs 6 and 7), they might not be dominant-negative, at least in the context of GluA1 ubiquitination. The endogenous Nedd4-2 in WT cultures potentially dominates even in the presence of mutant Nedd4-2s. Therefore, to determine the effects of mutant Nedd4-2s without the interference from endogenous Nedd4-2, we repeated this experiment in Nedd4-2andi cortical neuron cultures. Using Nedd4-2andi cultures possesses the advantage of studying the behavior of mutant Nedd4-2s while minimizing the concern of overexpression. As shown in Fig 8A, WT Nedd4-2 significantly reduced total and surface GluA1 in Nedd4-2andi cultures while all three mutant Nedd4-2s showed no effect. When expressing WT Nedd4-2 or any of the mutant Nedd4-2s in Nedd4-2andi cortical neuron cultures for 5 days, we found that the cultures transduced with WT Nedd4-2 showed significantly lower spontaneous spike frequency when compared with untransduced cultures or cultures transduced with any of the mutant Nedd4-2s (Fig 8B1 and 8B2). The average spontaneous spike amplitude did not differ between transduced and untransduced cultures. Altogether, our data showed that the three epilepsy-associated missense mutations of Nedd4-2 disrupt the ability to regulate surface GluA1 and spontaneous neuronal activity.
In this study, we present evidence to show that neuronal hyperactivity in vitro and increased seizure susceptibility in vivo associated with Nedd4-2 dysfunction are modulated by altered GluA1 and AMPAR signaling. These findings are further supported by the data showing that three epilepsy-associated missense mutations of Nedd4-2 partially, but significantly, disrupted GluA1 ubiquitination through reduced interaction with the adaptor protein 14-3-3. All mutant Nedd4-2s retain partial function toward ubiquitinating GluA1, reflecting the fact that the mutations were located on or near protein-protein interaction domains but not the lipid-binding or catalytic HECT (Homologous to the E6-AP Carboxyl Terminus) domain. Nevertheless, this is the first report that demonstrates a mechanism to explain Nedd4-2-dependent epilepsy in patients. Although the mutations of Nedd4-2 increase its stability, which is different from the in vivo knockdown mouse model we used (Nedd4-2andi mice), we showed that the reduction of Nedd4-2 and the mutations each reduce the ability of Nedd4-2 to ubiquitinate GluA1. Furthermore, because our data showed that genetic reduction of GluA1 normalized the seizure response in Nedd4-2andi mice, it suggests that inhibition of AMPARs might be a suitable treatment plan for Nedd4-2-associated epilepsy. One of the antagonists of AMPAR, Perampanel, has been approved to clinically reduce partial-onset seizures with or without secondary generalized seizures in epileptic patients [50–52]. Such medication might therefore be specifically useful for epilepsy patients who carry mutations in Nedd4-2. A future study on Perampanel will be very important to determine whether and to what the extent Nedd4-2-associated seizures and/or epilepsy can be ameliorated.
We used Nedd4-2andi mice, in which the long form of Nedd4-2 is disrupted, to study Nedd4-2. Because Nedd4-2 knockout mice exhibit perinatal lethality [34], Nedd4-2andi mice serve as an ideal model to study Nedd4-2 in vivo. However, another question is thus raised regarding the differential contribution of long versus short form of Nedd4-2 to the regulation of spontaneous neuronal activity and brain circuit excitability. The short form of Nedd4-2, which lacks an N-terminal C2 domain (S7 Fig), has also been identified in humans [53, 54]. Indirect evidence has suggested that the C2 domain mediates membrane-targeting of Nedd4-2 [8]. The C2-containing (long) and C2-lacking (short) isoforms therefore target different intracellular locations and substrate pools [8]. Currently, it is unclear whether the short form of Nedd4-2 exhibits similar affinity toward binding to and ubiquitinating GluA1. If it does, the question arises as to whether epilepsy-associated mutations affect the function of short form Nedd4-2 in a similar manner as to the long form of Nedd4-2. If it does not, the second question is whether the short form serves as a dominant-negative Nedd4-2 to sequester interacting or signaling molecules to affect the functions of the long form of Nedd4-2. Because single-nucleotide polymorphisms (SNPs) in human Nedd4-2 lead to differential expression of these isoforms, examining the functional differences of these isoforms may increase our understanding of neuronal plasticity and associated seizure susceptibility in different populations [53, 54].
In the adult brain, the AMPAR has been shown to mediate the majority of excitatory synaptic transmission with GluA1 being one of the major subunits [55]. Activity-mediated changes in the numbers and properties of GluA1/AMPAR are essential for excitatory synapse development and synaptic plasticity. Ubiquitination of GluA1 has been linked to AMPAR surface expression and trafficking, which subsequently may affect many synaptic plasticity mechanisms, such as homeostatic synaptic scaling and synaptic depression [21–23, 56, 57]. We previously demonstrated that Nedd4-2 mediates GluA1 ubiquitination upon chronic neuronal activity stimulation, suggesting a potential role of Nedd4-2 in homeostatic synaptic downscaling [15]. Whether Nedd4-2 participates in other synaptic plasticity mechanisms is unknown. One speculation would be that because Nedd4-2 functions to limit the amount of surface GluA1 as seen in Fig 5C, neuronal activity that mediates depression or elimination of excitatory synapses might induce Nedd4-2-mediated GluA1 ubiquitination. We recently found that the expression of Nedd4-2 is modulated by another ubiquitin E3 ligase murine double minute-2 (Mdm2) and its downstream effector tumor suppressor p53 [15]. Mdm2 is known to be crucial for activity-dependent synapse elimination [58], which is crucial for brain circuit development and maturation. Activation of Mdm2-p53 signaling and Nedd4-2 expression might therefore contribute to elimination of excitatory synapses. Activation of Mdm2-p53 signaling and Nedd4-2 expression might therefore contribute to elimination of excitatory synapses. Further studies are required to delineate the broader effects of Nedd4-2.
In addition to GluA1, other neuronal substrates of Nedd4-2 potentially involved in neuronal activity regulation are voltage-gated sodium channels Nav1.6 and voltage-gated potassium channels Kv7.3/KCNQ3 [9–12]. These two substrates are both crucial to modulating action potential firing and intrinsic excitability. Although our data showed that GluA1 mediates Nedd4-2-associated neuronal hyperactivity and seizures in mice, it does not rule out the potential contributions of Nav1.6 and Kv7.3/KCNQ3 in Nedd4-2-associated brain circuit excitability. Our data also suggest that presynaptic defects are potentially involved in the neuronal deficits associated with Nedd4-2 (Fig 1D). Multiple substrates of Nedd4 family members are known to mediate presynaptic vesicle release and activity, including α-synuclein [59–61] and tyrosine kinase A receptors [13, 14, 62]. Altered ubiquitination of these substrates when Nedd4-2’s function is compromised could contribute to aberrant synaptic transmission. The ubiquitination status, expression level, and subcellular distribution of Nedd4-2’s other substrates are pending further investigation to obtain the full picture of synaptic abnormality and excitability caused by pathogenic functions of mutant Nedd4-2s. As we described previously, future studies are expected to elucidate broader effects of Nedd4-2 and provide better understanding of this important, yet underdeveloped, molecule in the central nervous system.
All experiments using animal data followed the guidelines of Animal Care and Use provided by the Illinois Institutional Animal Care and Use Committee (IACUC) and the guidelines of the Euthanasia of Animals provided by the American Veterinary Medical Association (AVMA) to minimize animal suffering and the number of animals used. This study was performed under an approved IACUC animal protocol of University of Illinois at Urbana-Champaign (#14139 to N.-P. Tsai.)
The Nedd4-2andi mice, GluA1 knockout mice and WT control mice were obtained from The Jackson Laboratory. Primary cortical neuron cultures were made from p0-p1 mice as described previously [58] and maintained in NeuralQ basal medium (Sigma) supplemented with 1X B27 supplement (Invitrogen), 1X GlutaMax (final concentration at 2 mM; Invitrogen), and Cytosine β-D-arabinofuranoside (AraC, final concentration at 2 μM; Sigma). The medium was changed 50% on DIV 2 and every 3 days thereafter.
Dimethyl sulfoxide (DMSO) was from Fisher Scientific. AMPA was from Cayman Chemical and NBQX was from Alomone Labs. Recombinant GluA1 and 14-3-3ε were from Origene. Recombinant Nedd4-2 was from Abnova. R18 was from Sigma-Aldrich. Cycloheximide, poly-D-lysine and Protein A/G beads were from Santa Cruz Biotechnology. The antibodies used in this study were purchased from Santa Cruz Biotechnology (anti-α-Tubulin), Cell Signaling (anti-Nedd4-2, anti-pan-14-3-3, anti-N-cadherin and anti-Ubiquitin), Millipore (anti-GluA1), Abcam (anti-MAP2), Thermo Scientific (anti-HA) and GenScript Corporation (anti-Gapdh). The epilepsy-associated mutations were generated using site-directed mutagenesis reagent (Agilent) to introduce mutations into pCI-HA-Nedd4-2 [15]. The primers used are as below.
S233L: 5’-GGACGTGTCCTCGGAGTTGGACAATAACATCAGAC-3’,
5’-GTCTGATGTTATTGTCCAACTCCGAGGACACGTCC-3’;
E271A: 5’- GGGCGGGGATGTCCCCGCGCCTTGGGAGACCATTTC-3’,
5’- GAAATGGTCTCCCAAGGCGCGGGGACATCCCCGCCC-3’;
H515P: 5’- CGTTTGAAATTTCCAGTACCTATGCGGTCAAAGACATC-3’,
5’- GATGTCTTTGACCGCATAGGTACTGGAAATTTCAAACG-3’.
After washing cultured cells with PBS three times, 0.1 mg LLC NHS-LC-BIOTIN (from Apexbio Technology) was added to cultures for 30 min. at room temperature. At the end of the reaction, cultures were washed with PBS three times. The cell were harvested and lysed in an IP buffer (50 mM Tris, pH 7.4, 120 mM NaCl, 0.5% Nonidet P-40) followed by purification using Magnetic Streptavidin Beads (from Cell Signaling).
HEK cells were transfected using Lipofectamine 3000 (Invitrogen). Primary neuron cultures were transduced using lentivirus. WT and mutant Nedd4-2s were sub-cloned into Lenti-CMV-GFP-2A-Puro Vector (from Applied Biological Materials). Lentivirus was produced in HEK cells as described previously [58].
For immunoprecipitation (IP), cell lysates were obtained by sonicating pelleted cells in IP buffer. Eighty μg of total protein or protein mixtures after in vitro ubiquitination was incubated 2 hours at 4°C with 0.5 μg primary antibodies. Protein A/G agarose beads were added for another hour followed by washing with IP buffer three times. For western blotting, after SDS-PAGE, the gel was transferred onto a polyvinylidene fluoride membrane. After blocking with 1% Bovine Serum Albumin in TBST buffer (20 mM Tris pH 7.5, 150 mM NaCl, 0.1% Tween-20), the membrane was incubated with primary antibody overnight at 4°C, followed by three 10-min washings with TBST buffer. The membrane was then incubated with an HRP-conjugated secondary antibody (from Santa Cruz Biotechnology) for 1 hour at room temperature, followed by another three 10-min washings. Finally, the membrane was developed with an ECL Chemiluminescent Reagent [15]. All the western blot results were semi-quantitatively normalized to the control groups before statistical analysis.
HA-Ub (Boston Biochem), Ubiquitin Activating Enzyme (UBE1) (Boston Biochem) and UbcH5b/UBE2D2 (Boston Biochem) were obtained. Recombinant WT Nedd4-2 was obtained from Abcam. When HA-tagged WT or mutant Nedd4-2s were produced in HEK cells, 250 μg of total protein lysates were subjected to immunoprecipitation with anti-HA antibody to partially purify HA-tagged Nedd4-2s. Recombinant GluA1 (Origene) was used as substrate for in vitro ubiquitination with recombinant Nedd4-2 (Fig 7A) or Nedd4-2s obtained from transfected HEK cells (Figs 6B and 7C) following a protocol previously described [63]. Recombinant 14-3-3ε was obtained from Origene.
Each MEA plate was coated with poly-D-lysine for 30 minutes and plated with 2x105 cells counted using a hemocytometer. Recordings were done at DIV 13–14 (Figs 1, 2 and 3) or DIV 9 and DIV 14 (Fig 8) in the same culture medium using an Axion Muse 64-channel system in single well MEAs (M64-GL1-30Pt200, Axion Biosystems) inside a 5% CO2, 37°C incubator. Field potentials (voltage) at each electrode relative to the ground electrode were recorded with a sampling rate of 25 kHz. Right before a recording, if an electrode channel displays excessive “noise” (>10 μV) that channel is grounded for the entirety of the recording to avoid interference with other channels [64]. After 30 min of baseline recording, the MEA was treated with the drugs specified and recorded for another 30 min. Due to changes in network activity caused by physical movement of the MEA, only the last 15 min of each recording were used in data analyses. AxIS software (Axion Biosystems) was used for the extraction of spikes (i.e. action potentials) from the raw electrical signal obtained from the Axion Muse system. After filtering, a threshold of ±7 standard deviations was independently set for each channel; activity exceeding this threshold was counted as a spike. Only MEAs with more than 2,000 spikes during the last 15 minutes of recording were included for data analysis [16, 65]. The total spikes obtained from each MEA culture was normalized to the number of electrodes, as described in a previous study [66].
The settings for burst detection in each electrode were a minimum of 5 spikes with a maximum inter-spike interval of 0.1 sec as described previously [16]. The burst duration, number of spikes per burst, and interburst interval were analyzed by AxIS software. Synchrony index was also computed through AxIS software, based on a published algorithm [67] as we conducted previously [16], by taking the cross-correlation between any two spike trains, removing the portions of the cross-correlogram that are contributed by the auto-correlations of each spike train, and reducing the distribution to a single metric.
To ensure consistency when acquiring MEA data, all the experiment procedures, including the animal dissection, cell counting and plating, medium changing, and recordings are conducted by the same individual in each experiment. Throughout culture maturation and before recording, each MEA is visually inspected under the microscope and any MEA with poor growth or a patchy network is excluded. Recordings of each experiment were alternate between genotypes. For all before and after drug treatment comparisons, to minimize the variability between cultures, the recording from each MEA culture after treatment was compared to the baseline recording from that same culture.
Immunocytochemistry was done as previously described [68]. In brief, primary neurons grown on poly-D-lysine coated coverslips were fixed at DIV 14 with ice-cold buffer (4% paraformaldehyde and 5% sucrose in PBS). After washing and permeabilization with an additional incubation with 0.5% Triton X-100 in PBS for 5 min, an incubation with anti-MAP2 antibody was performed for 4 hours. After washing three times with PBS, fluorescence-conjugated secondary antibodies were applied to the cells at room temperature for 1 hour. After additionally washing the cells three times with PBS, the coverslips were mounted and observed on Zeiss LSM 700 Confocal Microscope.
Male mice at age 4-weeks old were intraperitoneally injected with kainic acid, prepared in saline solution (Hannas Pharmaceutical), at doses of 15, 30, or 60 mg/kg. The total injection volume was kept close to 0.15 ml. After injection, mice were closely observed in real time for 1 hour. The intensity of seizure was assessed by Racine’s scoring system [69]. To clearly determine seizure susceptibility, only stage 4 (rearing) and stage 5 (rearing and falling) were considered positive for seizures, as previously performed [16, 70]. Mice showing stage 4 seizure and above are counted as 1 while mice showing stage 3 seizure or under are counted as 0 for analysis.
ANOVA with post-hoc Tukey HSD (Honest Significant Differences) test was used for multiple comparisons between treatments or genotypes. Student’s t-test was used for analyzing spontaneous neuronal activity in Fig 1B and 1C, and mEPSC data in Fig 1D. One-sample t-test was used when experimental groups were normalized to control groups, such as western blotting in Fig 5C. Each “n” indicates an independent culture. Differences are considered significant at the level of p < 0.05.
|
10.1371/journal.pgen.1006925 | Rare coding variants pinpoint genes that control human hematological traits | The identification of rare coding or splice site variants remains the most straightforward strategy to link genes with human phenotypes. Here, we analyzed the association between 137,086 rare (minor allele frequency (MAF) <1%) coding or splice site variants and 15 hematological traits in up to 308,572 participants. We found 56 such rare coding or splice site variants at P<5x10-8, including 31 that are associated with a blood-cell phenotype for the first time. All but one of these 31 new independent variants map to loci previously implicated in hematopoiesis by genome-wide association studies (GWAS). This includes a rare splice acceptor variant (rs146597587, MAF = 0.5%) in interleukin 33 (IL33) associated with reduced eosinophil count (P = 2.4x10-23), and lower risk of asthma (P = 2.6x10-7, odds ratio [95% confidence interval] = 0.56 [0.45–0.70]) and allergic rhinitis (P = 4.2x10-4, odds ratio = 0.55 [0.39–0.76]). The single new locus identified in our study is defined by a rare p.Arg172Gly missense variant (rs145535174, MAF = 0.05%) in plasminogen (PLG) associated with increased platelet count (P = 6.8x10-9), and decreased D-dimer concentration (P = 0.018) and platelet reactivity (P<0.03). Finally, our results indicate that searching for rare coding or splice site variants in very large sample sizes can help prioritize causal genes at many GWAS loci associated with complex human diseases and traits.
| Genome-wide association studies (GWAS) have identified thousand of genetic associations between common DNA sequence variants (e.g. single nucleotide polymorphisms (SNPs)) and complex human diseases or traits. In most cases, these associations highlight non-coding variants, and thus fall short of identifying causal genes. The discovery of rare coding variants within these GWAS loci can greatly simplify causal gene identification. Here, we tested the association between 137,086 rare coding and splice site variants (minor allele frequency (MAF) <1%) and 15 blood-cell traits in 308,572 participants. We found 56 rare coding variants associated with hematological traits, including 31 variants reported for the first time. Thirty of these 31 variants are located within blood-cell trait GWAS loci, thus prioritizing candidate genes. We replicated an association between a rare interleukin 33 (IL33) splice site variation and eosinophil count and asthma risk, and showed that it also associates with hay fever risk. Finally, we found a new rare missense variant in plasminogen (PLG) associated with platelet count, D-dimer concentration, and platelet reactivity. In conclusion, it is possible to use the “rare coding variant association study” strategy to pinpoint causal genes at GWAS loci, but very large sample sizes are required.
| Although genome-wide association studies (GWAS) have identified thousands of associations with the number or distributional characteristics of red blood cells, white blood cells, and platelets [1], the genes responsible for this phenotypic variation remain elusive at most of these loci. For the most part, the difficulty in connecting association signals with genes resides in the fact that most GWAS variants are non-coding and in linkage disequilibrium (LD) with many other variants. This problem is not specific to hematological traits, but rather a general bottleneck in the functional characterization and clinical translation of most GWAS findings for common human diseases. Causal gene identification is important to shed light onto disease pathophysiology, but also to develop new, genetically-guided, therapeutic targets.
The discovery of rare variants that alter amino acid sequence or splicing is a strong indication that the mutated genes are involved in the phenotypes of interest. This approach is the cornerstone of Mendelian genetics. For complex human phenotypes, the discovery of rare coding variants associated with blood lipid levels [2, 3], coronary artery disease [4, 5], or more recently height [6] has helped prioritize causal genes and identify new drug targets (e.g. PCSK9 for atherosclerosis). For blood-cell related phenotypes, the identification of rare missense variants in the chemokine receptor gene CXCR2 [7] or the sphingosine-1 phosphate signalling gene S1PR4 [8] has highlighted the importance that these biological pathways play in neutrophil count variation in humans.
Although extremely useful tools to pinpoint causal genes, the discovery of rare variants of modest effect sizes have previously been limited because they require very large sample sizes. Here, we took advantage of genotype data at 137,086 rare (minor allele frequency (MAF) <1%) variants in 308,572 participants to find new genes implicated in human hematopoiesis. We identified 56 rare coding variants, including 31 new variants, which highlight specific genes that contribute to inter-individual variation of red blood cell (RBC), white blood cells (WBC), and platelet (PLT) traits.
In an effort to identify rare coding (defined in this study as missense or nonsense, excluding synonymous) or splice site variants that could implicate new genes in human hematopoiesis, we performed meta-analyses at 137,086 variants for 15 blood-cell traits in 308,572 participants (S1 Fig). These analyses found 81 missense or splice site variants at P<5x10-8 with a minor allele frequency (MAF) <1% across the tested studies. We excluded 25 variants because they were not present in phase 1 of the Blood-Cell Consortium (BCX1) dataset (N = 2), the direction of effect was discordant across studies (N = 11), or they mapped to the HLA region (N = 15)(S1 Table). As positive controls to validate our approach, 25 rare coding or splice site variants previously associated with hematological traits were genome-wide significant in the current analyses (S2 Table) [1, 8–11]. In addition, we found 31 new rare coding or splice site variants in 29 different genes associated with blood-cell trait variation (Table 1 and S3 and S4 Tables). The two variants in ATR are in strong linkage disequilibrium (LD), as are the two variants in EXOC3L4 (D′>0.8).
All 31 variants are novel discoveries because they were not identified in previous large-scale association studies for blood-cell traits. However, although the variants are new, many of them map to known blood-cell trait loci. Indeed, using conditional analyses in the UK Biobank, we could sub-divide these 31 variants into three categories: 16 variants are in LD with markers previously reported to associate with hematological traits (Pcond >0.002 in Table 1 and S5 and S6 Tables), 14 variants are not in LD with known blood-cell variants, but are located within 500 kb of one of these previously reported variants, and one variant (PLG-rs145535174) defines a new locus associated with PLT count variation in humans.
Plasminogen (PLG) encodes the precursor of plasmin, an important proteolytic enzyme that degrades many plasma proteins and fibrin blood clots. The rare G-allele at this new PLG variant (rs145535174, MAF = 0.05%) is associated with increased PLT count (+20x109 platelets/l, P = 6.8x10-9). It replaces an arginine residue at position 172 into a glycine, an amino acid change that is predicted to be probably damaging by Polyphen and deleterious by SIFT [12, 13]. This genomic position is highly conserved, as evidenced by a GERP score >200 and a CADD score of 23.8, ranking this particular amino acid change in PLG among the top 1% of the most deleterious substitutions in the human genome [14].
Given the role of plasmin in hemostasis, we tested the association between rs145535174 and several hemostatic and coagulation factors in the FHS. Although we only found a maximum of 16 carriers of the rare allele (varying depending on overlapping phenotypes), we could detect nominal associations between rs145535174-G and decreased D-dimer concentration and platelet reactivity (Table 2). Because D-dimer is a by-product of fibrin clot degradation by plasmin, this result is consistent with rs145535174-G being a PLG loss-of-function variant. Since plasminogen/plasmin activity also plays a role in thrombotic diseases, we tested if the G-allele at PLG-rs145535174 increases the risk of myocardial infarction (MI), stroke, or venous thromboembolism (VTE) in the UK Biobank (S7 Table). This variant was not associated with MI or stroke risk (P>0.05). However, there was an association with increased VTE risk in the UK Biobank (P = 0.0087, odds ratio = 4.01), although it was not replicated in the Montreal Heart Institute Biobank and the Women’s Health Initiative (P>0.05, S7 Table). The absence of genetic associations between PLG-rs145535174 and thrombotic events could simply reflect our limited statistical power given the rarity of the G-allele.
When we considered genes with known hematopoietic functions, we found little evidence that the rare coding or splice site variants found here implicated different potentially causal genes than candidate genes from previous blood-cell trait GWAS (S8 Table). In other words, we have, with one exception, no loci with two strong blood-cell trait candidate genes. On chromosome 11, we found an association between mean platelet volume (MPV) and a rare coding variant in SIRT3, a gene implicated in platelet biology [15]. At the same locus, Gieger et al. had nominated PSMD13 as a candidate platelet gene on the basis of reduction in plasmatocyte numbers in Drosophila [16]. We also noted few examples where rare coding or splice site variants implicated the same genes than those prioritized by GWAS SNPs due to expression quantitative trait loci effects (e.g. SLC12A7, FLT3, TMPRSS6)(S8 Table).
Conditional analyses identified 16 rare variants in LD with previously reported blood-cell variants (S5 and S6 Tables). Despite this dependence due to LD, at some of these loci, we may have identified a rare variant with even greater biologic plausibility for the blood-cell trait association than the previously identified LD surrogate. Indeed, whereas all 16 variants from our study are coding or splice site variants, most LD surrogate variants are non-coding variants (S6 Table). To clarify whether the new (Table 1) or previously known [1] blood-cell trait variants are more likely causal variants, we performed the reciprocal conditional analyses in which we conditioned the previously known blood-cell trait variants on genotypes at the newly described 31 rare coding or splice site variants (S9 Table). Comparison of the conditional results (S6 and S9 Tables) highlighted a few loci where the common variants are better causal candidates (e.g. WDR66, CLIP1, HIP1R). There were also a few loci where the new rare coding or splice site variants were statistically equivalent to rare non-coding variants previously reported (e.g. IL33, FLT3, BORA). We did not find examples of rare variants completely explaining the association signals at known common variants, as we would expect given limited LD.
On chromosome 1, we found a rare missense variant in CSF3R (rs148916169, MAF = 0.7%) associated with total WBC and neutrophil counts. Although this variant is correlated with common and low-frequency intronic variants (S6 Table), it is the first missense variant to directly implicate CSF3R in WBC count variation in the general population. CSF3R encodes the receptor for CSF3, a cytokine that controls the proliferation and differentiation of granulocytes, and is also mutated in a severe form of congenital neutropenia (MIM# 617014) [17]. Although rare loss-of-function mutations in CSF3R are associated with reduced neutrophil levels, the rare CSF3R missense variant found in our experiment might represent a gain-of-function since its rare allele is associated with higher WBC and neutrophil counts.
Another compelling example is the identification of a rare splice acceptor variant in IL33 (rs146597587, MAF = 0.5%) associated with reduced eosinophil count (Table 1). This likely functional IL33 variant maps to a locus with three non-coding variants, including a rare intergenic variant, associated with WBC traits (S6 Table) [1]. Interestingly, GWAS have associated IL33 common non-coding variants with risk for asthma [18], allergic rhinitis [19], and endometriosis [20]. In the UK Biobank, we found that the rare C-allele at IL33-rs146597587 associated with reduced eosinophil count is also associated with lower risk of asthma (P = 2.60x10-7, odds ratio = 0.56) and allergic rhinitis (P = 4.21x10-4, odds ratio = 0.55) (Table 3). These associations remained significant after accounting for the known GWAS variants previously identified at this locus (Table 3). Our finding strongly reinforces the clinical importance that eosinophils play in the pathophysiology of these diseases, and highlight the IL33 pathway as a promising therapeutic target. However, in a mediation analysis in the UK Biobank in which we corrected for eosinophil count, we found that the protective effect of IL33-rs146597587 was weaker but remained significant (Pmediation = 3.4x10-5). This suggests that IL33 may influence asthma risk in part by controlling blood eosinophil count, but also through currently unknown additional biological pathways. Finally, we note that the same rare IL33 variant was recently shown to associate with reduced eosinophil count and protection from asthma in the Icelandic population [21]. Because there is no sample overlap, our study represents a strong replication of this original report: in a meta-analysis of both studies, association results now reach genome-wide significance (in 24,030 asthma cases and 421,096 controls, P = 1.4x10-10, odds ratio = 0.54).
We found 14 rare coding or splice site variants that appear to be associated with blood-cell traits independently of other common and/or rare variants previously identified (Table 1 and S6 Table). Although there are a few well-known blood-cell-related genes, such as SLC12A7 and TMPRSS6, most of the highlighted genes do not have assigned functions in hematopoiesis (e.g. MFSD2B, SDPR, C9orf66, SUSD1, EXOC3L4)(Table 1). We found two correlated rare missense variants in ATR (rs28910273, MAF = 0.6%; rs77208665, MAF = 0.7%), which encodes a checkpoint protein that coordinates cellular responses to replication stress and DNA damage. ATR is mutated in Seckel syndrome 1 (MIM #210600), which is characterized by hematological defects in some patients [22]. We also uncovered a rare splice acceptor variant in the apolipoprotein gene APOC3 (rs138326449, MAF = 0.2%) associated with RDW, a RBC trait that is considered a non-specific inflammatory marker and a predictor of cardiovascular diseases [23]. The same APOC3 variant has previously been associated with lower triglyceride levels and coronary artery disease risk [5, 24].
We performed three biological pathway-enrichment analyses using 58 genes that harbour a rare coding or splice site variants associated with RBC, WBC, or PLT traits (Table 1 and S2 Table). In total, we found six, 13, and five biological pathways or terms enriched for genes with variants associated with RBC, WBC, and PLT, respectively, at a false discovery rate <10% (Table 4). Most of these pathways were highly redundant in terms of gene content. For instance, five of the six RBC-enriched pathways are involved in iron homeostasis. For WBC traits, there were many significant inter-connected biological pathways that implicated six genes: CSF3R, CXCR2, S1PR4, IL17RA, AMICA1/JAML, and FLT3 (Table 4). These genes highlight the importance that genetic variation plays on cytokine signalling and chemotaxis to modulate circulating numbers of immune cells in the blood.
Using a large sample of >300,000 participants, we identified 31 new missense or splice site variants associated with blood-cell traits. These include a rare p.Arg172Gly missense of plasminogen associated with higher platelet count, lower D-dimer concentration and lower platelet reactivity and a rare splice acceptor variant of IL33 associated with lower eosinophil count and lower risk of asthma and allergic rhinitis. At other genomic loci, our findings may prioritize the causal gene(s) that can be pursued with molecular techniques to further dissect and define the mechanism of association. Because phenotypic variance explained is determined by the effect size and allele frequency of the variants, the rare coding and splice site variants identified here do not have a large contribution to the heritability of blood-cell traits. For instance, the rare IL33-rs146597587 variant explains only 0.06% of the variation in eosinophil count.
The chromosome 2p23 region spans several genes and contains six common, non-coding variants associated previously with WBC or RBC traits [1]. We identified a new rare coding variant in this region associated with lower MCV located in MFSD2B, a transporter gene of unknown function that is highly and specifically expressed in blood and bone marrow cells, particularly of the erythroid lineage [25]. As another example, at the LY75-CD302 locus, four common non-coding variants and a low-frequency missense variant have been associated with PLT and WBC traits [10]. A second novel rare LY75-CD302 missense variant was associated with higher PLT count. The LY75-CD302 locus represents a naturally occurring read-through transcript between the lymphocyte antigen 75 (LY75) and CD302 genes, and is expressed in leukocyte and hematopoietic stem and progenitor cells [25]. Alternative splicing results in fusion LY75-CD302 gene products that are expressed during dendritic cell maturation [26] and Hodgkin’s lymphoma cell lines [27]. Another example of a potential read-through transcript involves a gene-rich region on chromosome 15, where several common variants were recently associated with WBC traits [1]. At this locus, we discovered a rare coding variant (rs184575290) located near the ST20 termination codon associated with higher monocyte count. This locus represents a naturally occurring read-through transcript that produces a fusion protein between ST20 and the neighboring MTHFS gene. ST20 is highly expressed in myeloid cells, while the fusion product is expressed at lower levels in blood neutrophils and dendritic cells of the skin [25].
We found a rare missense variant within SDPR, a phosphatidylserine-binding protein originally isolated from human platelets [28], that may be involved in modulating activation of the platelet protein kinase C substrate pleckstrin and mediating the downstream effects of platelet granule secretion [29]. Common missense variants of EXOC3L4, which has no known function, have been previously associated with both platelet and liver enzyme traits [30]. In progenitor blood-cell expression data from the BLUEPRINT Project [31], there is a large increase in EXOC3L4 expression in megakaryocyte progenitors. Several rare, non-coding variants at the chromosome 13q21 region have been associated with RBC traits [1]. We identified a rare coding variant of BORA, an activator of the protein kinase Aurora A involved in centrosome maturation and spindle assembly during mitosis and whose expression pattern during hematopoiesis increases in RBC and PLT precursors [31].
Matriptase-2, encoded by TMPRSS6, has recently emerged through both complex trait and Mendelian disorders as an important genetic regulator of RBC and iron metabolism [32]. We report here the first rare coding variant of TMPRSS6 (p.V280L/p.V289L) associated with RBC traits in individuals unselected for hematologic disease. Other TMPRSS6 coding variants have been reported in patients with familial iron-refractory iron-deficiency anemia (IRIDA), a rare autosomal-recessive disorder characterized by hypochromic microcytic anemia and impaired iron balance [33].
Plasminogen plays an important role in fibrinolysis as well as wound healing, cell migration, and tissue re-modeling. Congenital plasminogen deficiency is a rare autosomal recessive disorder [34]. Severely affected individuals (type I plasminogen deficiency) can exhibit defective extracellular fibrin clearance during wound healing, leading to “ligneous conjunctivitis'' or thick, pseudomembranes on conjunctival and other mucosal surfaces. The p.Arg172Gly variant of plasminogen associated with higher PLT count is located in the first kringle domain, which mediates binding of plasminogen to fibrin and cell surfaces [35]. The observed association of the PLG p.Arg172Gly variant with lower D-dimer is consistent with reduced fibrinolysis, which might result from reduced circulating plasminogen, plasmin activity, or substrate binding. Overall, we did not observe an association between PLG p.Arg172Gly and risk of thrombotic disease. Similarly, mutations associated with congenital plasminogen deficiency do not appear to be a strong risk factor for VTE [36–38].
The reasons for the association of the PLG p.Arg172Gly variant with higher PLT count and lower platelet reactivity are not readily apparent. In vitro, plasmin has been reported to proteolytically inactivate thrombopoietin [39]. Plasmin also is capable of activating platelets through protease-activated receptors (PAR)-1 and -4 [40, 41]. Thus, it is possible that the PLG loss-of-function variant may influence thrombopoietin-induced platelet production, or PAR-induced platelet aggregation, respectively. Finally, it is worth noting that moderately reduced platelet count is a feature of a congenital platelet disorder characterized by a gain-of-function fibrinolysis defect due to increased expression and storage of urokinase plasminogen activator (PLAU) during megakaryocyte differentiation [42].
In conclusion, focusing on 137,086 rare coding or splice site variants in 308,572 participants, we discovered 31 new variants associated with blood-cell traits. Thirty of these 31 variants map to loci previously implicated by GWAS for hematological phenotypes. Because we used the ExomeChip or the UK Biobank array, there is an enrichment of GWAS signals among the loci tested, although many of the loci identified here did not harbour blood-cell trait genetic associations at the time the arrays were designed. As discussed, many of these associations prioritize strong candidate genes at these loci. Our study was limited by the content of the genotyping arrays utilized. As we expand our genetic experiments to complete genome sequence in very large cohorts, we anticipate that we will uncover more rare coding variants and, maybe more importantly for gene identification within GWAS loci, additional independent variants in the same candidate genes. Together with a recent report for adult height [6], our findings reinforce the importance that rare variants play in the architecture of complex human phenotypes.
Written, informed consent was obtained for all participants in accordance with the Declaration of Helsinki. This project was also reviewed and approved by the Montreal Heart Institute Ethics Committee and the different recruiting centers (approval number 2014–1707).
Single variant association results considered in this effort were obtained from participants of European ancestry using an additive genetic model. All phenotypes were corrected for confounding factors (see below) and normalized using inverse normal transformation. The final analyses included samples from BCX1 (up to 157,622 participants), AIRWAVE (up to 14,866), and the first release of the UK Biobank (up to 136,084). Genotypes for BCX1 and AIRWAVE were obtained from the Illumina ExomeChip array; the content is available at: http://genome.sph.umich.edu/wiki/Exome_Chip_Design. The ExomeChip includes ~250,000 exonic variants obtained from whole-exome sequencing of ~12,000 participants. The UK Biobank samples were genotyped on the UK Biobank Axiom array; the content of the arrays is available at: http://www.ukbiobank.ac.uk/scientists-3/uk-biobank-axiom-array/. Whereas the Axiom array targets >800,000 variants, we only analyzed exonic variants that overlap with the content of the ExomeChip.
Phenotypic information, and ExomeChip quality-control steps and association results from cohorts involved in the BCX1 Consortium and AIRWAVE have been described elsewhere [10, 11, 43]. Briefly, we excluded variants with low genotyping success rate (<95%, except for WHI that used a cutoff <90%). Samples with call rate <95% after joint or zCALL calling and with outlying heterozygosity rate were also excluded. Other exclusions were deviation from Hardy-Weinberg equilibrium (P<1x10-6) and gender mismatch. We performed principal component analysis (PCA) or multidimensional scaling (MDS) and excluded sample outliers from the resulting plots, using populations from the 1000 Genomes Project to anchor these analyses. Prior to performing meta-analyses of the association results provided by each participating study, we ran the EasyQC protocol [44] to check for population allele frequency deviations and proper trait transformation in each cohort. In terms of hematological phenotypes, we excluded individuals with blood cancer, leukemia, lymphoma, bone marrow transplant, congenital or hereditary anemia, HIV, end-stage kidney disease, dialysis, splenectomy, and cirrhosis, and those with extreme measurements of RBC traits. We also excluded individuals on erythropoietin treatment or chemotherapy. Additionally, we excluded pregnant women and individuals with acute medical illness at the time the complete blood count (CBC) was done. For all traits, within each study, we adjusted for age, age-squared, gender, the first 10 principal components, and, where applicable, other study-specific covariates such as study center using a linear regression model. Within each study, we then applied inverse normal transformation on the residuals and tested the phenotypes for association with the ExomeChip variants using either RVtests (version 20140416) [45] or RAREMETALWORKER.0.4.9 [46].
A description of methods and quality-control procedures for the blood-cell data for the first release of the UK Biobank can also be found elsewhere [1]. Description of the exome-component of the genotyping arrays used for the UK Biobank samples can be found at: http://www.ukbiobank.ac.uk/scientists-3/uk-biobank-axiom-array/. In the UK Biobank, we modelled blood-cell traits using the following covariates: age, sex, menopause status for women, assessment centre where the blood samples were collected, machine counter that processed the blood samples, month, day of the week, time inside the day that the samples were collected, self-reported ethnic background of the individuals, smoking status and smoking frequency, alcohol drinker status, alcohol intake frequency, height and weight. In the first release of the UK Biobank, we tested the association between directly genotyped or imputed variants and normalized hematological traits with BOLT-LMM [47].
We meta-analyzed results from BCX1, AIRWAVE and the UK Biobank using inverse-variance weighting as implemented in METAL [48]. We limited our analyses to variants with a mean minor allele frequency (MAF) <1% in the meta-analyses that are annotated as coding (missense or nonsense) or splice site (acceptor or donor) using ENSEMBL’s Variant Effect Predictor (VEP). Furthermore, the variants had to be present on the Illumina exome array used by the BCX1 studies. In total, we tested 137,086 variants against 15 blood-cell traits. These phenotypes are divided between the main three cell types found in blood: red blood cells (red blood cell count (RBC count), hemoglobin concentration (HGB), hematocrit (HCT), mean corpuscular hemoglobin (MCH), mean corpuscular volume (MCV), mean corpuscular hemoglobin concentration (MCHC), and RBC distribution width (RDW)), white blood cells (total white blood cell count (WBC count), neutrophil count (Neutro), lymphocyte count (Lympho), monocyte count (Mono), basophil count (Baso), and eosinophil count (Eosin)), and platelets (platelet count (PLT count) and mean platelet volume (MPV)). The meta-analysis results are available at: http://www.mhi-humangenetics.org/en/resources. We used a conservative α = 5x10-8 to declare statistical significance. At this statistical threshold, our sample size (N = 308,572) provides >99% power to discover variants with MAF = 0.1% that explain >0.03% of the phenotypic variance.
To test whether the 31 rare variants newly identified in the meta-analyses are associated with blood-cell traits independently of other known genetic loci, we regressed out the effect of the known blood-cell variants previously identified in the first release of the UK Biobank [1] from the normalized blood-cell phenotypes. All these analyses were done per phenotype; that is we fitted 15 different models for each of the 15 blood-cell phenotypes tested. We then re-tested in the UK Biobank (using linear regression implemented in R) the association between the “residual” blood-cell phenotypes and genotypes at the rare variants identified in the meta-analyses. For instance, for hemoglobin, we conditioned the new rare coding or splice site hemoglobin variants on all variants (across the genome) previously reported to associate with hemoglobin levels. We provide the complete list of variants on which we conditioned in S5 Table per blood-cell trait. If the rare variants were not significantly associated with the residual phenotypes, we then ran pairwise conditional analyses to identify which previously known variants at the locus accounted for the association signal identified in this project. To declare statistical independence from previously reported hematological trait association signals, we used α = 0.002 (Bonferroni correction for 31 variants tested).
We sought association of the PLG variant (rs145535174) with platelet reactivity, as well as hemostatic and coagulation factors in the Framingham Heart Study (FHS) [49]. Genotyping for rs145535174 was conducted using the Illumina Human Exome BeadChip v.1.0 (Illumina, Inc., San Diego, CA) [50]. Multiple measurements were assessed for platelet reactivity [51], including maximum percentage platelet aggregation in response to agonists, i.e. ADP and epinephrine; minimal concentration of each agonist to produce a >50% aggregation response (EC50); and lag time in response to collagen stimulation. Hemostatic factors and coagulation factors, including antigens of plasminogen activator inhibitor-1 (PAI-1), tissue plasminogen activator (tPA), D-dimer, clotting factor VII (FVII) and von Willebrand factor (VWF), were measured using enzyme-linked immunosorbent assays [52–54] while fibrinogen levels were assessed using the Clauss method [52]. Association analyses in the FHS were conducted using either RAREMETALWORKER (http://genome.sph.umich.edu/wiki/RAREMETALWORKER) or seqMeta (http://cran.r-project.org/web/packages/seqMeta/index.html). A linear mixed effects model that accounts for familial correlation was used and adjustments were made for age, sex and principal components. The phenotypes were log-transformed or inverse normal transformed, as needed.
We tested the association between the rare G-allele at PLG-rs145535174 and myocardial infarction (MI) and stroke in the UK Biobank. We used the “Health and Medical History” records to identify MI and stroke cases. For MI, we used the search terms “heart attack”, “myocardial infarction”, “acute myocardial infarction”, “subsequent myocardial infarction” and “old myocardial infarction” to retrieve affected individuals. We used the terms “stroke”, “ischaemic stroke” and “cerebral infarction” to define stroke cases. As controls, we excluded UK Biobank participants with MI, stroke or transient ischemic attack, percutaneous coronary intervention, coronary artery bypass graft surgery, peripheral vascular disease, congestive heart failure, and angina. For analysis of venous thromboembolism (VTE), we identified cases in the UK Biobank, the Montreal Heart Institute Biobank, and the Women’s Health Initiative as individuals with pulmonary embolism or deep vein thrombosis. We tested the genetic association by logistic regression in PLINK or R, correcting for age, sex and the first ten principal components when available.
We identified asthma, allergic rhinitis (hay fever), and endometriosis cases using the detailed “Health and Medical History” UK Biobank participant records. All other individuals were assigned as controls. We tested the genetic association by logistic regression in R, correcting for age, sex and the first ten principal components. To determine if the rare IL33-rs146597587 variant is independent from the previously reported common SNPs at the locus, we conditioned on genotypes at these variants (rs343496, rs7032572, rs72699186, rs1342326, rs2381416, rs928413, rs10975519) and re-run the logistic regression model.
We used the default parameters in DAVID [55] to perform biological term and pathway enrichment analyses. For these bioinformatic analyses, we used as reference set all genes with at least one rare coding or splice site variant tested in the meta-analyses. We retrieved genes associated with RBC, WBC, or PLT traits from Table 1 (new independent variants from our study) and S2 Table (known positive controls), and tested their enrichment in biological pathways in comparison with the reference set. Due to the relatively low number of genes that were used as input for this kind of analysis, we lowered the initial, minimum number of genes in a seeding group to 3 (default = 4) to ensure that the clustering algorithm will include as many genes as possible into functional groups. All other parameters were left at their default values.
|
10.1371/journal.pntd.0007262 | Application of long read sequencing to determine expressed antigen diversity in Trypanosoma brucei infections | Antigenic variation is employed by many pathogens to evade the host immune response, and Trypanosoma brucei has evolved a complex system to achieve this phenotype, involving sequential use of variant surface glycoprotein (VSG) genes encoded from a large repertoire of ~2,000 genes. T. brucei express multiple, sometimes closely related, VSGs in a population at any one time, and the ability to resolve and analyse this diversity has been limited. We applied long read sequencing (PacBio) to VSG amplicons generated from blood extracted from batches of mice sacrificed at time points (days 3, 6, 10 and 12) post-infection with T. brucei TREU927. The data showed that long read sequencing is reliable for resolving variant differences between VSGs, and demonstrated that there is significant expressed diversity (449 VSGs detected across 20 mice) and across the timeframe of study there was a clear semi-reproducible pattern of expressed diversity (median of 27 VSGs per sample at day 3 post infection (p.i.), 82 VSGs at day 6 p.i., 187 VSGs at day 10 p.i. and 132 VSGs by day 12 p.i.). There was also consistent detection of one VSG dominating expression across replicates at days 3 and 6, and emergence of a second dominant VSG across replicates by day 12. The innovative application of ecological diversity analysis to VSG reads enabled characterisation of hierarchical VSG expression in the dataset, and resulted in a novel method for analysing such patterns of variation. Additionally, the long read approach allowed detection of mosaic VSG expression from very few reads–the earliest in infection that such events have been detected. Therefore, our results indicate that long read analysis is a reliable tool for resolving diverse gene expression profiles, and provides novel insights into the complexity and nature of VSG expression in trypanosomes, revealing significantly higher diversity than previously shown and the ability to identify mosaic gene formation early during the infection process.
| Antigenic variation is a system whereby pathogens switch identity of a protein that is exposed to the host adaptive immune response as a way of remaining one step ahead and avoiding being detected. African trypanosomes have evolved a spectacularly elaborate system of antigenic variation, with variants being used from a library of ~2,000 genes. Our ability to understand how this rich repository is used has been hampered by the resolution of available technologies to discriminate between what can be closely related gene variants. We have applied a long read sequencing technology, which generates sequence information for the whole length of the antigen gene variants, thereby avoiding having to try and piece together antigen sequences from lots of small fragments, the pitfall of standard sequencing. Applying this technology to material taken at specific time points from batches of mice infected with trypanosomes reveals that the diversity of variants is much higher than previously suspected, and that there is a clear semi-predictable pattern in the gene expression. Additionally, using this technology we have been able to detect the presence of ‘mosaic’ genes, which are created by stitching together fragments from several donor genes in the library, much earlier in infection than has been shown previously. Therefore, we shed new light on the complexity of antigenic variation and show that long read sequencing will be a very useful tool in analysing and understanding the expression patterns of closely related genes, and how pathogens use them to cause persistent infections and disease.
| Antigenic variation is used by many pathogens as a means of staying one step ahead of the host’s adaptive immune response. Trypanosoma brucei has long been a paradigm for the study of antigenic variation, and the protein responsible, the variable surface glycoprotein (VSG) has been the focus of much research [1–3]. Each trypanosome in a population expresses a single species of protein, and an inherent, parasite-driven switching process causes a proportion of the population to replace their active VSG gene with a different VSG gene, resulting in the expression of a protein in those cells with different epitopes exposed to the host immune system (at a rate of up to 10−2 switches per cell/generation [4]). The post-genomic era has revealed T. brucei’s antigenic variation system to be unrivalled in its elaboration, particularly in terms of the scale of the numbers of genes that comprise the VSG family. Sequencing the genome of T. brucei has uncovered a gene family much greater in numbers and complexity than was previously thought. Characterisation to date suggests that at least 2,000 VSG genes are in the genome of each trypanosome, providing a spectacularly large repertoire of potential antigens [5, 6], particularly when compared to other pathogens that undergo antigenic variation, such as Plasmodium Falciparum (60 genes in PfEMP1 family [7]), Anaplasma marginale (~10 members in the msp2 & msp3 gene families [8]), and Borrelia burgdorferi (15 members in the vls gene family[9]).
The scale of the gene family size is also reflected in the complexity of switching mechanisms employed to change the identity of the surface antigen. The VSGs are expressed from one of approximately 20 bloodstream expression sites (BES)[10], the active expression occurring in a dedicated sub-nuclear organelle, the expression site body (ESB)[11], with the remainder of BESs being transcriptionally silent. A minor mechanism of VSG switching, accounting for only approximately 10% of events in wild type trypanosomes [12], is to turn off the transcription of the active BES and activate one of the silent BESs (‘transcriptional switching’). However, the majority of switching is through replacing the gene sequence in the active BES via gene duplication, which involves the copying of variable amounts of sequence, ranging from within the gene to the whole telomere [13, 14]. Insights into mechanisms involved in switching suggest that replacing expressed VSG sequence is driven by DNA recombination, and DNA repair/homologous recombination pathways and proteins (e.g. RAD51) have been identified to be involved in the gene duplication process of VSG switching [15] (reviewed in [16]). A further layer of complexity is the construction of novel VSG sequences in the BES from multiple donor VSG sequences, a form of segmental gene conversion termed ‘mosaic’ gene formation [17, 18]. Mosaic gene formation was previously considered to be a rare and minor mechanistic component of overall VSG switching in an infection (e.g. [14]). However, the revelation upon the sequencing of the T. brucei genome that a significant proportion of the VSG repertoire (80–90%) consisted of pseudogenes [19] that cannot be expressed as functional proteins began to alter that perception [5, 20]. It has become clear from subsequent experimental work that mosaic gene formation is an integral component of VSG switching, particularly after the early stages of infection (i.e. beyond the first peak of parasitaemia in mouse infections)[5, 21].
One of the challenges of analysing VSG expression in vivo, and in particular gaining an accurate measurement of the level of expressed diversity given the extent of the VSG repertoire (i.e. to what extent is the repertoire actually used during infection), has been the relatively limited resolution of available techniques–in particular the manual cloning and sequencing of individual VSG cDNAs that has been undertaken in recent studies [5, 21]). While this approach clearly provides accurate data at the level of individual VSG transcripts, the limitations have undoubtedly resulted in a low estimate of the diversity and complexity of VSG expression at the population level, and particularly with respect to minor variant populations. Additionally, although transcriptomics potentially provides the ability to overcome the resolution limitations of manually cloning and sequencing transcripts, the application of RNAseq to VSG expression from in vivo samples has long been deemed challenging, due to the requirement for assembling multiple closely related gene variants from a mixed population using short reads of 100–200 base pairs (e.g. Illumina)–this has similarly been an issue when attempting to resolve, for example, the diversity of the mammalian immunoglobulin gene repertoire underpinning the antibody response (e.g. [22]). However, a recent study subjected in vivo samples to Illumina sequencing (100bp, single-end reads) and demonstrated the utility of transcriptomics in terms of increased resolution [23], and were able to detect minor variants (0.1% of population) and up to 79 variants at a time point, although they were not able to identify significant mosaic gene expression.
Long read sequencing potentially provides the ability to further increase our resolution, particularly as the length of reads commonly reached with such technologies (average read length in Pacbio, for example, is quoted as 10,000–20,000 bp; http://www.pacb.com/smrt-science/smrt-sequencing/read-lengths/) far exceeds the length of the VSG transcript (approximately 1600 bp), meaning that the issue of assembly of closely related VSGs from multiple reads should be bypassed. Here, we present analysis of VSG expression from replicate in vivo T. brucei TREU927 infections in mice at 4 time points over 12 days using almost 500,000 Pacbio Sequencing reads. We demonstrate that long read technologies provide significant advantages for analysing the diversity of VSG expression. Our data suggest that the VSG population comprises significantly more variants even at an early stage of infection (up to 190 variants at day 10 post-infection), that the pattern of VSG expression is surprisingly reproducible (using the novel application of ecological diversity indices), and that mosaic gene expression can be detected much earlier in infection than has been possible previously. Our data also provide insights into the nature of mutations introduced by Pacbio long-read sequencing technology, as the dataset includes significant coverage of one sequence (>140,000 reads).
Using PacBio long read RNA sequencing of 20 blood samples enriched for VSG transcripts from replicate in vivo T. brucei TREU927 infections in mice at 3, 6, 10 and 12 days post infection, we obtained 486,343 reads with an average read length of insert of 1,569 bp (Table 1, Fig 1B). Reads were filtered by length (1400-2000bp) based upon both literature on VSG genes [21, 24] and the read distribution in our dataset (Fig 1B) to remove reads resulting from sequencing artefacts and shorter fragments (i.e. partial reads), and on the basis of similarity to known VSGs (blastn ≥60% alignment against TriTrypDB-v26 [25]–note that the reads include both N-Terminal and C-Terminal domain sequences) (Fig 1C), resulting in a dataset of 296,937 ‘VSG’ reads. Of the reads that were of the appropriate length (1400-2000bp) but did not have ≥60% match to VSGs in the reference database (n = 102,940), 90,810 (88.2%) mapped partially to VSGs, 3,513 (3.4%) mapped to non-VSGs, and 8,617 (8.3%) did not produce any match to the TREU927 reference genome. Within the dataset of 296,937 VSG reads, each read on average represented the consensus sequence from 6.50 passes of the full length fragment by the DNA polymerase (‘full passes per read’; summarised in Table 1; full data in S1 Table), and for each of these reads there was robust identification of a donor gene for the N-Terminal domain (NTD); therefore, for these 296,937 reads we have high confidence that they contain all of the features necessary to be consistent with being full length VSG transcripts. The 296,937 reads represent a total of 449 VSGs (74.77% of VSG a-type and 25.22% VSG b-type [24]) across 20 samples, with the number of reads per VSG following a power-law distribution (Fig 1D), and provide a unique insight into the in vivo VSG transcriptome across time and animal replicate.
ORFs were identified in the 296,937 reads with a conservative minimum nucleotide size of 1200 nucleotides (reported size ranges of VSG NTDs and C-Terminal domains [CTDs] are approximately 300–350 and 100 amino acids, respectively [5, 21, 24]). Surprisingly, only 33,234 reads (11%) resulted in predicted ORFs. Although the percentage of reads with predicted ORF increased with increasing number of full passes, it remained well below 50% even for reads having 10 full passes or more (Fig 2A). Since the distribution of the number of reads with a detectable ORF over all VSGs was similar to total expression level distribution (Table 2), we hypothesize that the lack of identified ORFs was due to random sequencing errors rather than any systematic biases in the data, despite PacBio claiming an accuracy of more than 99% for reads with 15-fold coverage [26]. To investigate this hypothesis in more detail, we focused on the most abundant VSG (Tb08.27P2.380, 1551bp, 141,822 high-confidence reads) and annotated each discrepant base pair of each aligned read as either an insertion, deletion or mismatch with respect to the Tb08.27P2.380 reference genome sequence. All reads had an alignment score greater than 90% over the first 1266bp (the N-Terminal domain) (Fig 2B). The distribution of sequence errors showed a clear bimodal pattern across the N-Terminal domain, with 145 nucleotide positions having a consistent mismatch (131), deletion (10) or insertion (4) across more than 80% of the reads, and 1,112 nucleotide positions having errors in at least one but fewer than 2% of reads (Fig 2B). This suggests that the former represent genuine mutations already present in our inoculum (with respect to the reference genome sequence), whereas the latter represent either random sequencing errors introduced by Pacbio or low level genuine mutations that we cannot currently distinguish from Pacbio error. Previous studies have indicated accumulation of mutations over time in expressed VSGs, and we examined this in our data for reads aligning to Tb08.27P2.380 (for the N-terminal domain) by assessing the error rate for mismatches, insertions and deletions (S1 Fig). While these data indicated statistical support for differences in the data distribution across time points for all 3 mutation classes, due to the skewed nature of the data distribution (most bases have an error rate close to zero) this conclusion must be treated with a degree of caution. The assertion that the errors present in >80% of reads were ‘genuine’ mutations was further supported by these 145 mutations being consistently present in PCR amplicons sequenced by Sanger sequencing. These PCR amplicons had been generated from cDNA extracted from multiple samples (n = 7 for Tb08.27P2.380; representing sequences independently cloned and sequenced from 4 mice on days 3 and 10, S2 Fig). Insertions were the most common Pacbio-introduced error (average per-base error rate of 0.79% across the N-terminal domain sequence), followed by deletions (0.73%) and mismatches (0.33%) (Fig 2C), in agreement with what has been reported before [27]. Consistent with the ORF prediction pattern (Fig 2A), the overall error percentage was lower for reads with higher number of passes, but introduced sequencing errors (i.e. interpreted as mutations not present in the genome of the inoculated trypanosomes) remained present at more than 1000 nucleotide positions even for reads with 10 passes (Fig 2D). The nature of our data, comprising >141,000 reads of the same sequence, therefore provides an unusually robust insight into the nature of Pacbio errors and the caveats that must be placed upon interpretation of such data, as most studies involve much less coverage per single base pair.
Our data demonstrate that we can detect multiple VSGs in each sample, and that we can identify changes in VSG expression and diversity over time. We identified a median of 27 unique VSGs per sample at day 3 post infection (p.i.), which progressed to 82 VSGs at day 6 p.i., peaking at 187 VSGs at day 10 p.i. and reducing to 132 VSGs by day 12 p.i. (Fig 3A). When identified VSGs that mapped to single reads from single samples were removed, this resulted in an identification of 334 VSGs (median of 27, 81, 170 and 126 VSGs per sample at 3, 6, 10, and 12 days p.i., respectively).
Not only were the number of distinct VSGs consistent across samples for the same time point, but the expression pattern (proportion of reads per sample mapping to particular VSGs) was also highly reproducible between samples and over time (Fig 3B), albeit bearing in mind that these analyses are of batches of mice at four different time points rather than longitudinal samples of the same mice. The VSG that is dominant at day 3 (Tb08.27P2.380), presumably introduced as the dominant VSG in the inoculum, remains dominant in all mice at day 6, but is the single VSG with the most reads aligned in only two of five mice at day 10. Interestingly, by day 12, the VSG with the most reads per sample is the same in all five mice (Tb09.v4.0077) and this VSG was also most common at similar timepoints in previous analyses [21]. Additionally, the other eight VSGs that reads map to in mice at days 10 and 12 (Tb927.4.5730, Tb927.10.10, Tb11.v5.0932, Tb927.9.300, Tb09.v4.0088, Tb05.5K5.330, Tb927.9.16490 and Tb927.3.480; Fig 3B) are present in all ten mice suggesting a degree of conservation in the sequential expression of VSGs in independent infections, consistent with previous observations [21, 28, 29]. However, in all mice there were reads that mapped to VSGs distinct to these most favoured 10 VSGs (‘others’ in Fig 3B, which account for 10.36% of all VSG-mapped reads), and in some mice this proportion was particularly high (e.g. mice 3.5, 6.1 and 10.4; Fig 3B). This is particularly evident at day 6, where although the dominant VSG (Tb08.27P2.380) makes up most reads, the majority of reads that do not map to Tb08.27P2.380 map to VSGs other than the other top 9 VSGs in all mice. Additionally, at Day 10 we observe both the greatest number of VSGs and the least domination by any single VSG, but the proportion of ‘others’ either reduces or remains stable. These analyses combine to indicate that while there is a broad predictability in expression, with dominant VSGs at the beginning and end of infections, in between these timepoints there is a degree of stochasticity in the system–although eight VSGs comprise the majority of reads that do not map to either of the two dominant VSGs, the relative proportion of these ‘minority’ VSGs is not consistent, and there are furthermore many VSGs that are expressed at very low levels in all mice.
The analysis described thus far (Fig 3) has not taken into account any sequence similarity between VSGs, but relied on mapping reads to identified VSGs in the reference database. To analyse the population diversity of VSGs within and across samples using a method that is independent of mapping to existing databases (which are likely to be incomplete), we applied information theoretic measures more commonly used to quantify the biodiversity of ecosystems [30]. This approach initially applied a clustering algorithm to a proportion of reads (n = 33,205; comprising reads with predicted ORF) in order to enable identification of the reads that clustered on the basis of sequence similarity, as putatively distinct VSGs (Fig 4A). These data showed significant congruity with those described for the VSG mapping approach described above (Table 2). The top 10 clusters comprised 89.34% of all reads, compared to 89.68% for the VSG mapping approach, and the relative proportion of reads that either map to VSGs or cluster by sequence similarity is very comparable for the 10 most abundant VSGs (Table 2). These data indicate that the clustering algorithm applied was robust in terms of identifying individual VSGs, and therefore indicated a very similar pattern of a dominant early VSG, followed by an intermediate period of significant greater VSG diversity, ending up with a second dominant VSG by day 12.
The sequence similarity data also allowed the analysis of variability between mice using a new measure of population differentiation called normalised beta diversity [30] (Fig 4B). When looking at a single day, beta diversity is the effective number of distinct VSG profiles present on that day, giving information on the differentiation between the animals. This analysis indicates (similar to the VSG mapping data) the greatest beta diversity across individuals is at day 10 (Fig 4B solid line).
Further exploring each time point and variation between mice (Fig 4B dots), we can see that although the mice at day 3 show some distinct VSG profiles (albeit with overexpressed VSGs in individual mice common to all mice, SI Fig 3), at day 6 most mice (except for mouse 6.5) are broadly consistent with respect to which VSGs are present and how common they are. The effective number of VSG profiles increases further on day 10 with maximal divergence between mice at any time point, (Fig 4B, solid line). This value then decreases on day 12 (though mouse 12.2 is distinct), as the mice begin to express similar profiles again. These analyses again indicate that there is stochasticity in the process of VSG expression considered as a progression over 12 days, and there is semi-predictability rather than strict hierarchical progression through VSG expression, as has been described previously [21, 29, 31, 32]
Mosaic genes were considered identified where BLAST hits for a particular read demonstrated non-overlapping homology to more than one distinct VSG in the reference database. This was commonly seen in the C-Terminal domain, where the same N-Terminal domain was in many instances observed with different C-Terminal domains (“3’ donation” in [21]). Using pairwise alignments of all reads that mapped to Tb08.27P2.380, based on the alignment coverage over the gene, donors were filtered based on the region representing the C-Terminal domain (the 3’ region approximating to 30% of the gene shown in Fig 5A). Donors were selected based on at least 80% alignment coverage to the CTD. These data show that the reads aligning to Tb08.27P2.380 consists of three subgroups based on their CTD donors, which are derived from either the reference gene Tb08.27P2.380 (43% of all reads), but also from Tb10.v4.0158 (29%) or Tb927.6.5210 (28%). The proportion of the three donor CTDs varies across time points, with the proportion of reads deriving from the donor Tb08.27P2.380 gene decreasing by days 10 and 12 (reducing from 46.55% at day 3 to day 26.19% at day 12, although the number of reads in total aligning to Tb08.27P2.380 is low by days 10 and 12). The frequent nature of this recombination has been observed previously [21]. We detected N-Terminal domain mosaics (within the constraints of our stringent selection criteria) at a much lower frequency (n = 45 over all 20 mice; three sequences at day 3, five at day 6, 13 at day 10 and 23 and day 12 –S2 Table), and in most cases these are single read examples, and so must be treated with some caution (albeit 12 of the putative mosaic reads have coverage of at least 7 full passes, a coverage level at which our analysis–Fig 2D–suggests should effectively remove sequencing-derived error). However, we have two examples where we have more than one read indicating N-Terminal domain mosaicism, with the additional support for one of these sequences that it is only detected in one mouse–given the complex nature of previously identified mosaic N-Terminal domains [5, 21], it is unlikely that identical mosaics would emerge in separate individual infections. Nevertheless, we do also have one putative mosaic sequence that occurs in two separate mice (balbc_6_0/100673/ccs5 and balbc_12_1/30571/ccs9 in mice 6.1 and 12.1, respectively; S2 Table)–this may either represent a gene currently not annotated in the TREU927 genome or be a true mosaic gene that was present in the initial inoculum and has remained at low levels throughout infection. The N-Terminal domain mosaic examples we have detected are mostly relatively simple mosaic genes (e.g. Fig 5B). Although we cannot formally rule out that at least a proportion of these mosaic genes were present in the original inoculum, the increased frequency over time is consistent with expectations that this process is rarest early in infection but becomes more prevalent as infections progress.
The results illustrate the power of long read sequencing when applied to expressed gene diversity–we identified 449 VSGs across 20 individual samples, covering four time points post-infection (3, 6, 10 and 12 days). The identification of the VSGs was achieved by two approaches; mapping reads to a reference VSG database, and secondly clustering read sequences to identify distinct variants–importantly without reliance upon a reference genome or sequence database. These independent approaches were highly congruent in the number of VSGs and the proportion of reads that were attributed to individual VSGs (Table 2), meaning that the clustering approach may be particularly valuable for analysis of long read data generated from infections with trypanosome strains (or species) where a genome is either not available or is incomplete. When compared with previous approaches, such as manual cloning (801 VSG sequences that comprised 93 distinct VSGs or ‘sets’ across 11 mice across 19 days of sampling each [21]) or short read Illumina sequencing (289 VSGs for 4 mice– 3 mice sampled 9 times over 30 days and one mouse sampled 13 times over 105 days [23]), the Pacbio approach gives significantly higher resolution per sample. It must be acknowledged that in the present study the starting volume of infected blood for each sample was higher (200 μl versus 50–100 μl in [23] and approximately 15 μl in [21]), and additionally the inoculum in the current study was significantly greater and not clonal, meaning the study design may predispose to more expressed variants being detectable. The TREU927 clone used was also highly virulent, giving rise to a high parasitaemia early that was maintained for the 12 days of infection. This is not representative of the classical fluctuating profile of less virulent strains (or clones of this strain, e.g. [33]); however, for the purposes of assessing the utility of Pacbio this was advantageous. A proportion of the identified VSGs (115/449; 25.6%) derive from single reads in single samples, and therefore a degree of caution must be employed with these variants. However, when the singleton VSGs are removed, we can confidently conclude that we have identified 334 VSGs across our datasets–this ranges from a median of 27 VSGs in day 3 samples to 170 in day 10 samples. Therefore, despite these caveats, we can still conclude that the resolution in terms of diversity is significant for the long read approach, and likely to be of great utility for studies incorporating VSG diversity going forward.
Despite the limitations of the study design, where we have analysed batches of mice at four time points rather than longitudinal surveys of individual mice, our data across 20 mice and four time points are very consistent with a highly reproducible pattern of VSG expression over time (Fig 3 & Table 2). There was a remarkable degree of consistency in identity of dominant VSGs across independent infections–particularly as the inoculum used was not a single cell or a cloned inoculum (this is very distinct from, for example, Borrelia, where Pacbio analysis has indicated very little overlap in expressed antigen diversity across replicates from the same starting inoculum [34]). The data demonstrated a consistent emergence of the two sequentially dominant variants at the beginning and end of the infection period (Tb08.27P2.380 and Tb09.v4.0077), although during the period in between the dominant VSGs there was significant diversity in expressed VSGs that was consistent with an inherent degree of stochasticity in the system. This was reinforced by the application of biodiversity analysis (Fig 4), which illustrated the semi-predictable nature of the variant progression across the mice and timepoints. This chimes with previous work that described the semi-predictable expression of VSGs in T. brucei [21, 28, 29], and modelling approaches that have also reflected semi-predictable use of the VSG repertoire [31, 32, 35].
When analysing our data set and comparing with that of Hall et al, 2013, who used the same TREU927 strain, we have significant overlap in detected expressed variants. 90% of our reads correspond with a VSG detected in Hall et al. The dominant early VSG is different (corresponding to ‘Set_23’ in the Hall data, Table 2), although the Tb09.v4.0077 which becomes dominant by day 12 was similarly dominant by ~day 20 in Hall et al; differences are presumably due to the use of either a stabilate with a distinct passage history, or the use of a larger inoculum rather than single trypanosomes (i.e. inoculation of a population from a previous infection presumably expressing the dominant VSG at that particular stage). The dominant VSG in our dataset (Tb08.27P2.380) was annotated as a pseudogene in the reference genome (predicted to be truncated due to insertion of a stop codon). The annotation as a pseudogene is not consistent with our data as a dominant early VSG, as it would suggest mosaic gene formation providing a dominant early gene–indeed, recent reannotation has classified this gene as intact, which would be more in keeping with early expression favouring intact over pseudogene or mosaic VSGs [5, 21]. However, given the 1 × 105 inoculum used in this study, it is also feasible that the transfer of Tb08.27P2.380 as the dominant expressed VSG from the donor mouse infection may have given rise to it being the dominant expressed VSG in the infections analysed.
We have identified mosaic genes (classified as reads demonstrated non-overlapping homology to more than one distinct VSG N-Terminal domain in the reference database) earlier in infection than has previously been identified, although we cannot formally rule out that at least some of these were introduced in the original inoculum. The rate of mosaic gene detection was very low in our study, mostly either single or very few reads, which probably reflects our timeframe being only 12 days post-infection; this also pertains if the trypanosomes expressing mosaic VSGs derived from the inoculum, which was also generated over a short duration (5–7 days) in the donor mouse. However, these data do indicate that the nature of the long read sequencing is highly beneficial in terms of mosaic gene identification; even low frequency expressed genes (within the limitation of the four orders of magnitude of coverage that the read number per sample provides) can be identified with some confidence due to the acquisition of the whole gene sequence–in order to achieve this with short read approaches a reasonable degree of read coverage would be required to identify and confirm putative mosaic genes. This has potential implications for the application of long read sequencing to significantly further our understanding of infection dynamics and the role of mosaic genes as infections progress. This is likely to be important in terms of ability to gain insights into the mechanisms of mosaic gene formation because of consequent increased ability to resolve defects in switching rate (e.g. analysis of DNA recombination gene mutants such as RAD51 that have been implicated in DNA recombination-based VSG switching [15])–at present it is not known if mosaic gene formation involves a mechanistic switch in terms of pathways; the ability to detect low frequency mosaic gene expression should provide the ability to study this. Additionally, detection of low frequency VSGs would enhance the ability overall to more fully analyse the temporal kinetics of VSG switching–providing an avenue for improved quality of inputs into modelling dynamics of VSG expression. The clustering approach developed in this study that does not rely upon a reference database would also make analysing expressed VSG diversity in the animal trypanosomes, T. congolense & T. vivax, feasible—the reference genome (and therefore genomic VSG repertoire) is less well annotated in these species than in T. brucei. One challenge for taking a similar approach in these species is the lack of conserved 3’ UTR sequence in expressed VSGs to enrich transcripts. However, such analyses may be particularly enlightening given the different structure and content of VSG repertoires recently described between the three genomes [36, 37], as well as the strikingly different arrangement of VSG expression sites in T. congolense compared to T. brucei [38].
We detected indels consistently when comparing Pacbio transcripts to the reference gene (Fig 2). While these differences may indeed be real, with our protocol we have somewhat limited resolution for conclusively differentiating indels introduced by the trypanosome from those potentially introduced by PCR. However, PCR is unlikely to be the sole cause of the observed mutations, because in the dominant VSG in our dataset (Tb08.27P2.380), which represents 141,822 reads across all 20 samples–therefore, 20 independent PCR reactions—we observe a consistent set of variations from the reference genome sequence (145 nucleotide positions across the NTD having a consistent mismatch (131), deletion (10) or insertion (4) across more than 80% of reads with respect to the reference sequence) across all reads–these are consistently present across all reads for this variant, including those reads with high fold coverage (i.e. greater than 10 full passes per read) (Fig 2). These data, across technical and biological replicates, lead us to conclude that these differences were present in most likely the genome copy, but also potentially a distinct BES-resident copy of this VSG that has accumulated mutations distinct to the genome basic copy of the gene, and mutations were not introduced by PCR. One possible explanation for this is that there is very likely a significant (and unknown) divergence in passage history between the sequenced reference genome TREU927 trypanosomes and those used in this experiment. This would be consistent with data from many pathogens of the increased mutability of telomeric/subtelomeric gene families [39]. Previous data indicated accumulation of point mutations in expressed VSGs over time within infections [5, 21], and in our data we saw some support for this process, but the skewed nature of the data distribution limits our ability to conclude increased mutations over time as an important aspect of VSG expression (it should be noted that a timeframe of 12 days is relatively short and will have limited our resolution). However, our data indicate that application of long read analysis over longer infection timeframes is likely to be a useful means of characterising the nature and role of this mechanism.
However, the multiple mutations that were present across multiple VSG sequences in our data, did enable detailed analysis of the nature of mutations detected in Pacbio sequencing (Fig 2). Ideally, to enable clear differentiation of PCR bias and artefact, errors introduced by Pacbio, and mutations introduced by the trypanosome, unique molecular identifiers (UMIs) would be added prior to PCR amplification (e.g. [22, 40]). While we did not incorporate this step, we can draw some conclusions from analysis of our data. When data for Tb08.27P2.380, which represents 141,822 reads, is analysed across the range of fold coverage per read, it is clear that most of these mutations are removed as the coverage increases (Fig 2)–although notably even at a high number of passes some introduced mutations remain. This strongly suggests that most of these are errors that are introduced by the Pacbio process, and the proportion we observed across the dataset (insertion 0.79%, deletion 0.73%, mismatches 0.33% per base pair) is consistent with that reported in other studies (e.g. [41]). The mutations also directly influenced the ability to predict open reading frames in our data—ORFs only being detected in 11.22% of VSG reads (33,234 of 296,937). Clearly, with these reads being generated from cDNA one would have expected most if not all to have identifiable ORFs. Therefore, these data indicate some of the limitations when using Pacbio, even with data that comprises multiple passes–the introduction of mutations does provide a layer of complexity to the analysis that must be addressed with care. This is particularly pertinent when trying to analyse multiple closely related genes, as in the case of VSGs. We were able to draw conclusions on the basis of sufficient coverage of a highly expressed dominant gene, combined with the inclusion of multiple biological replicates; without these elements interpretation would have been very difficult without parallel short read sequencing to correct errors introduced by the technology.
A further issue for consideration for the application of long read technologies to the analysis of expressed gene diversity is the number of reads per sample. Our data provided coverage over four orders of magnitude—although significantly greater in resolution than previous manual and laborious methods, this contrasts relatively poorly with the numbers of reads that short read applications deliver (millions). However, it should be noted that with the short read approach many reads will be required to robustly identify full length single variants (in particular to enable differentiation of closely related transcripts, either similar genome-encoded variants or related lineages of mosaic genes [5, 21]), whereas in theory at least a single pacbio read should provide the ability to robustly identify a particular VSG transcript. While the coverage is being improved with the newer platforms (e.g. the Pacbio Sequel potentially delivers a further tenfold increase in data per run), this may limit resolution in terms of detecting minor variants, for example. We did detect significant expressed diversity, and this is partly explained by our use of a relatively large inoculum, which was not cloned, of a virulent isolate that resulted in high and sustained parasitaemia. Therefore, we started with what was probably a relatively diverse population (albeit dominated by expression of Tb08.27P2.380), reflected in the diversity of VSGs detected at day 3 post-infection, which would be significantly lower in the event of a clonal or smaller initial inoculum.
While our data indicate that long read sequencing provides increased resolution in terms of identifying VSG diversity, clearly questions still remain. For example, why the VSG repertoire is so evolved and large? Our data suggest an increased proportion of repertoire is involved, even at early stages, compared to previous studies, which indicates a bigger proportion of the repertoire may be utilised during the lifetime of an infection (which in cattle can be many hundreds of days) than previous data suggests. This is consistent with the data of Mugnier et al [23], where multiple minor variants were observed using an Illumina sequencing approach. However, that study and ours both have limitations, one with relatively few biological replicates (albeit one mouse was followed for ~120 days) and one that only ventured to 12 days post-infection. Therefore, assessing antigen dynamics in the chronic phase of infections with tools that give significant resolution of expressed antigen diversity will be critical to furthering our understanding of the mechanisms of trypanosome antigenic variation. Key to studying this will be analysing the picture in the truly chronic stages of infection (as was done by Mugnier et al in the context of mouse infections), but particularly doing so in relevant hosts (e.g. cattle [42]) where the total population of trypanosomes in the animal will be potentially 1,000 times greater at peak parasitaemia and where infections may last for 100s of days–this will have a profound influence on the usage of the repertoire (our data, for example, was representative of a total population of approximately 1 × 108 parasites per mouse). Additionally, recent studies indicates T. brucei populations inhabit different niches in the mammalian host (e.g. skin and adipose [43, 44]), to the extent that some show evidence of local adaptation with respect to metabolism ([44])–how this population compartmentalisation interacts with antigenic variation and immunity is likely to be important for parasite maintenance and transmission. Therefore, understanding the dynamics in both the chronic stages of infection and in clinically relevant hosts will potentially provide ideas on the selective pressures that maintain such an elaborate system. Additionally, given the significant advantages described above in terms of identifying low frequency variants (including mosaic VSGs), it may be that a combined long and short read approach is likely to be the optimal way of holistically and accurately identifying expressed VSG diversity; the increased read number of short read technologies in combination with the better resolution of long read technologies would provide significant power to examine the complexity of VSG expression in trypanosomes.
Animal experiments were carried out at the University of Glasgow under the auspices of Home Office Project License number 60/3760. Care and maintenance of animals complied with University regulations and the Animals (Scientific Procedures) Act (1986; revised 2013).
All mice were infected with Trypanosoma brucei brucei TREU927, the genome reference strain [19, 45]. A cryostabilate from liquid nitrogen was thawed and inoculated into BALB/c mice in order to amplify a viable in vivo population. Donor mice were euthanased at first peak parasitaemia (approximately 1 × 107 trypanosomes/ml), and blood extracted. Trypanosomes were counted in triplicate under an improved Neubauer haemocytometer, diluted to inocula of 1 × 105 trypanosomes in 200 μl Carter’s Balanced Salt Solution, which were then inoculated via the intraperitoneal route into 20 recipient BALB/c mice. Mice were maintained for 12 days post-infection, and groups of 5 mice were euthanased at 3, 6, 10 and 12 days post-infection. Parasitaemia was monitored daily by venesection of the lateral tail veins using the rapid matching technique [46], and was counted in triplicate under an improved Neubauer haemocytometer on the sampling days.
At each sampling day, RNA was extracted from 200 μl infected blood using the Qiagen RNeasy kit (Qiagen), according to the manufacturer’s instructions. Approximately 1 μg RNA was treated with DNase Turbo (Ambion), according to manufacturer’s instructions, and cDNA was generated as in Hall et al, 2013 [21], including a column purification step on generated cDNA using the PCR Purification kit, according to the manufacturer’s instructions (Qiagen). VSG transcripts were enriched by carrying out PCR with proof reading Herculase II Fusion polymerase (Agilent) on the cDNA template with oligonucleotide primers specific to the T. brucei spliced leader sequence (TbSL) and a reverse primers complementary to a 13 base pair conserved region in VSG 3’ untranslated regions (3UTR); primer sequences and PCR conditions were as previously described [21, 23]. A subset of PCR transcripts was subjected to cloning and sequencing; PCR products were ligated into pGEMT-Easy vectors, transfected into One Shot TOP10 cells, bacteria were grown up and cloned under suitable antibiotic selection (all using the TOPO cloning kit, Invitrogen), and plasmid DNA extracted using a Miniprep kit (Qiagen); these procedures were all carried out according to manufacturer’s instructions. Extracted plasmid DNA of appropriate concentration was sent for sequencing (Eurofins MWG).
1 μg of PCR amplicon template as measured by Nanodrop (ThermoScientific) and Bioanalyser (Agilent) was submitted to the Centre for Genomic Research, University of Liverpool for sequencing using the Pacbio RSII platform (Pacific Biosciences). DNA was purified with 1x cleaned Ampure beads (Agencourt) and the quantity and quality was assessed using Nanodrop and Qubit assay. Fragment Analyser (using a high sensitivity genomic kit) was used to determine the average size of the DNA and the extent of degradation. DNA was treated with Exonuclease V11 at 37°C for 15 minutes. The ends of the DNA were repaired as described by the Pacific Biosciences protocol. Samples were incubated for 20 minutes at 37°C with damage repair mix supplied in the SMRTbell library kit (Pacific Biosciences). This was followed by a 5 minute incubation at 25°C with end repair mix. DNA was cleaned using 0.5x Ampure beads and 70% ethanol washes. DNA was ligated to adapter overnight at 25°C. Ligation was terminated by incubation at 65°C for 10 minutes followed by exonuclease treatment for 1 hour at 37°C. The SMRTbell libraries were purified with 0.5x Ampure beads. The quantity of library and therefore the recovery was determined by Qubit assay and the average fragment size determined by Fragment Analyser. SMRTbell libraries were then annealed to the sequencing primer at values predetermined by the Binding Calculator (Pacific Biosciences) and a complex made with the DNA Polymerase (P4/C2chemistry). The complex was bound to Magbeads and this was used to set up 3 SMRT cells for sequencing. Sequencing was done using 180 minute movie times. Data (raw sequencing files) is available through Gene Expression Omnibus (https://www.ncbi.nlm.nih.gov/geo/ - accession number GSE114843).
|
10.1371/journal.pntd.0006621 | Characterization of the microtranscriptome of macrophages infected with virulent, attenuated and saprophyte strains of Leptospira spp. | Leptospirosis is a bacterial zoonosis, caused by Leptospira spp., that leads to significant morbidity and mortality worldwide. Despite considerable advances, much is yet to be discovered about disease pathogenicity. The influence of epigenetic mechanisms, particularly RNA-mediated post-transcriptional regulation of host immune response has been described following a variety of bacterial infections. The current study examined the microtranscriptome of macrophages J774A.1 following an 8h infection with virulent, attenuated and saprophyte strains of Leptospira. Microarray analysis revealed that 29 miRNAs were misregulated following leptospiral infection compared to control macrophages in a strain and virulence-specific manner. Pathway analysis for targets of these differentially expressed miRNAs suggests that several processes involved in immune response could be regulated by miRNAs. Our data provides the first evidence that host miRNAs are regulated by Leptospira infection in macrophages. A number of the identified miRNA targets participate in key immune response processes. We suggest that post-transcriptional regulation by miRNAs may play a role in host response to infection in leptospirosis.
| Leptospirosis is a zoonotic disease, distributed worldwide, affecting millions of people each year, and leading to sixty thousand deaths per year. These bacteria are found in soil and water and are eliminated by the urine of rodents, their natural reservoir. Through skin contact, bacteria can be acquired, infecting the host. Infection process in leptospirosis is not completely understood and here we add another layer of disease regulation. Recent studies have shown that pathogens can modulate host response. Our current study examined the expression of microRNAs in murine macrophages following an 8h infection with virulent, attenuated and saprophyte strains of Leptospira. This study provides the first evidence that these post-transcriptional regulatory molecules, microRNAs, are modulated in macrophages in a species and virulence-specific manner, following infection with different strains of Leptospira spp. These microRNAs are involved in the regulation of inflammatory and antimicrobial responses in the host and could lead to the identification of biomarkers or therapeutic targets for this disease.
| Leptospirosis is a zoonosis of global importance, particularly in developing countries with tropical climates [1], and is caused by a highly invasive gram-negative spirochete known as Leptospira. This genus is comprised of 12 species and 250 serotypes between pathogenic and non-pathogenic strains [2–3], with Leptospira interrogans being the most common pathogenic species. Mortality rate is around 60,000 deaths per year and the annual number of severe cases can reach 1 million, placing leptospirosis as a major player in morbidity, and number of deaths, by zoonotic causes [4–5].
Rodents are natural reservoirs for these bacteria and they shed the pathogen in their urine, contaminating water and soil in urban and rural environments. Humans can be infected by skin contact, mainly in areas lacking sanitation [6]. During early infection, antibiotics are effective, however most vaccines available for veterinary application provide limited protection against more than 250 pathogenic Leptospira serovars [7–8]. Advances in research have been made through the use of conserved leptospiral proteins, aiming at better vaccine candidates for leptospirosis [9].
Macrophages play a central role in leptospirosis by phagocytizing bacteria in humans and other mammals [10–11]. Cinco et al. [12] suggest that Leptospira inside macrophages are fully capable of replication, and Li et al. [13] showed that Leptospira are capable of escaping host defense responses, however only in human macrophages.
Leptospira interrogans has higher pathogenicity due to components such as lipopolysaccharides, peptideoglycans, lipoproteins, glycoproteins and membrane proteins, which induce a robust inflammatory response [14–17]. Activation of innate immunity by toll-like receptors (TLRs 2/4) in macrophages is essential for host defense [18]. TLR activation and MyD88 recruitment, signaling through several pathways like mitogen-activated protein (MAP) kinases, NF-κB and pro-inflammatory cytokines, lead to B and T cell activation [19–21]. Another rapid response induced by L. interrogans is apoptosis in macrophages and hepatocytes. This pathway is activated by caspase 3 and 6 through a FADD–caspase-8-dependent pathway [22] involving intracellular free calcium ion (Ca2+) [23].
It is known that the pathogen/host interaction can significantly modify gene expression profiles in an infected host, and it was suggested that post-transcriptional regulation might be acting in this interaction [24]. MicroRNAs (miRNAs) are small non-coding RNAs spanning 20–22 nucleotides, and have a major role in posttranscriptional regulation of gene expression. These small RNAs negatively regulate protein synthesis through base pairing with partially complementary sequences in the 3´UTR region of target mRNAs, favoring their degradation or translational repression [25–27]. Each miRNA has the potential to target, thus controlling, hundreds of genes [28]. An extensive body of research indicates that several pathogens (viruses, parasites and bacteria) can affect miRNA expression in host cells [29–31]. Further, miRNAs associated with disease can act as biomarkers or therapeutic targets [32–33].
For this reason, miRNAs and their targets, modulated by the pathogen, are of great importance to comprehend the pathophysiology of leptospirosis. Based on the identification of targets of differentially expressed miRNAs, following infection of murine macrophages with different types of Leptospira, we report several canonical pathways that could be affected by infection. We have established, for the first time, that modulation of miRNAs is present in Leptospira infection, and obtained potential miRNA signatures for different strains, varying in virulence. We suggest that posttranscriptional regulation by miRNAs may play a role in host response to infection in leptospirosis. These findings add to the growing list of infectious diseases that involve miRNAs regulation by the host.
In total, we identified 29 miRNAs that were modulated in macrophages after 8h of infection with different strains of Leptospira spp (fold change ± 1.5; p <0.01). When compared to non-infected control cells, 17 miRNAs were significantly altered (15 upregulated and 2 downregulated) following infection with the virulent strain, 16 miRNAs were modulated (12 upregulated and 4 downregulated) as a response to the attenuated strain, and 9 miRNAs were altered by infection with the saprophyte strain (5 upregulated and 4 downregulated) (Fig 1). The intersection of treatments in the Venn diagram shows that three miRNAs are modulated by all strains. MiRNAs were also modulated in a strain-specific manner, where 7 miRNAs were modulated specifically by the virulent strain, 7 by the attenuated and 5 only by the saprophyte bacteria (Fig 2). Average signals (log2) of samples were hierarchically clustered using Pearson´s correlation and complete-linkage, we observed a clustering of samples based on species and virulence, with the virulent and attenuated strains clustering closer together, followed by the saprophyte strain (Fig 3), and all infected samples clearly separating from non-infected controls. Validation of chosen regulated miRNAs by quantitative realtime PCR (miR-155-5p; miR-7667-3p; miR-203-3p and 222-5p), corroborated the microarray results with respective correlation coefficients between techniques of 0.99, 0.79, 0.92 and 0.99 (Fig 4). The highest fold change was observed for miR-155-5p with an upregulation of >12 fold for L. interrogans and >5 fold for the saprophyte L. biflexa, when compared to non-infected cells. In Fig 5, we depict the significant canonical pathways for miR-155-5p targets, identified in the virulent treatment compared to noninfected control cells.
For prediction of target genes to the differentially expressed miRNAs in all treatments, we used the tool miRNA Target Filter present in the IPA software. We utilized databases from miRecords, Tarbase, TargetScan, and the Ingenuity Knowledge Base, and filtered for targets with high prediction and that have been experimentally observed only. In Table 1, we demonstrate that 10 out of the 29 differentially expressed miRNAs have predicted mRNA targets. To identify the relationships, mechanisms, functions, and pathways relevant to the list of target genes, we employed the feature Core Analysis available in the IPA software package with values of significance–log (BH corrected p-value)>1.3. From our list of predicted genes for each treatment, we report 21 and 5 canonical pathways for virulent and attenuated treatment, respectively (Tables 2 and 3). For the saprophyte treatment, specific miRNAs did not have predicted targets. Further, only the virulent treatment had specific pathways identified.
Five biological processes, Molecular Mechanisms of Cancer, Fcγ Receptormediated Phagocytosis in Macrophages, PI3K/AKT Signaling, PTEN Signaling and Role of Macrophages in Rheumatoid Arthritis are potentially regulated by miRNAs in response to L. interrogans infection, regardless of virulence. These five pathways were therefore classified as common to L. interrogans. In regards to specific processes, 16 pathways were identified only in response to the virulent strain compared to noninfected controls.
Despite its prevalence, morbidity and mortality rates, pathogenicity of leptospirosis is still largely unknown. To better understand the processes involved in the host-pathogen interplay of Leptospira infection, we carried out, for the first time, a global evaluation of microRNA modulation of murine macrophages infected with different strains of Leptospira spp at 8h post-infection in vitro. Here, we compared miRNA profiles to address whether Leptospira affect macrophageal miRNA expression, if this modulation is species specific and if bacterial virulence plays a role in this modulation.
In Fig 3, we show that Leptospira spp, regardless of species or virulence, modulate the miRNAs mmu-miR-155-5p, mmu-miR-155-3p and mmu-miR-221-5p. Notwithstanding, specific miRNAs signatures were also obtained for each strain, and they differed within species based on virulence. Differential host response has been previously suggested to be associated to protein differences between the strains. In fact, Haake et al., [34] reported that attenuated L. interrogans cultures expressed different proteins and LPS profiles when compared to virulent cultures. In addition, Picardeau et al. [35], compared the genomic sequence of saprophyte L. biflexa with that of the pathogenic strain, L. interrogans, identifying differences such as a larger number of genes encoding proteins containing leucine-rich repeat (LRR) domains in the pathogenic strain, shown to be involved in attachment and invasion of host cells in other bacteria. Therefore, it is not surprising that differing bacterial genomic and protein profiles elicit different responses in their host cells. Further, Xu et al. [36] demonstrated significant differences in gene expression, particularly in genes related to antigen processing and presentation, regulation of membrane potential, cell migration, cytoskeleton organization and biogenesis are mostly up-regulated in murine macrophage cells, following infection by different species of Leptospira. Our current results indicate that, beyond the previously reported genomic and transcriptional differences, control of gene expression by means of post-transcriptional modifications may be dependent on species and virulence in leptospiral infection.
Among the miRNAs identified in our study as commonly regulated by the genus Leptospira sp, independent of virulence, is mmu-miR-155-5p. This miRNA is known to be upregulated in inflammatory processes such as rheumatoid arthritis, cancer, cardiovascular disease, as well as in other bacterial infections in macrophages like Listeria, Salmonella, Helicobacter and Mycobacteria [37] reviewed by [31,38]. MiR155 has a considerable number of mRNA targets, and all canonical pathways identified here, in virulent and attenuated treatments, have highly predicted and experimentally observed targets of mmu-miR-155 (S1 and S2 Tables), suggesting an important role for miR-155 as a post-transcriptional regulator in leptospiral induced host response.
Macrophages are fundamental against leptospiral infection. They have different receptors that activate a plethora of responses, such as phagocytosis, cytokine/chemokine production and antigen presentation [39–41]. The cross-linking of IgGs (immunoglobulin-g) with Fc receptors in macrophages initiates crucial cellular events for host immune response. The pathway Fc-gamma receptors in macrophages, play an important role in recognizing IgG-coated pathogen targets during the phagocytosis process in the host [42]. Our target analysis identified previously reported genes in this pathway, significantly identified following virulent and attenuated infection, such as SHIP-1, VAV3 and VAMP3 suggested to be downregulated by mmumiR-155-5p, and PTEN, downregulated by mmu-miR-222-3p, involved in phagosome formation and recycling of cell membrane, respectively. Phagocytosis is vital for internalization of leptospires, and previous work has indicated that one of the resistance mechanism of pathogenic Leptospira in the host, is evasion of the alternative and classical complement system pathways [43]. VAV3, is a potential target to miR-155-5p, and downregulation of this gene can cause inhibition in B- and T-cell development and activation, given its involvement in activating pathways that lead to actin cytoskeletal rearrangements and greater cellular movement [44]. More studies are needed to correlate Leptospira evasion and VAV3 function. Inositol polyphosphate-5-phosphatase D (INPPD5 or SHIP-1) is a well established target for mmu-miR-155-5p [37] and this correlation is associated with cell proliferation. Increased expression of mmu-miR-1555p, with a resulting decrease of INPPD5, promotes transcription of major proinflammatory cytokines in macrophages [45–46]. Restoration of INPPD5 levels is related to an inhibition in PI3K-AKT signaling and anti-inflammatory response in raw264.7 cells and primary bone marrow-derived macrophages (BMDMs), as reported in bowel disease [46]. It is tempting to hypothesize that, at early infection (8h in this study), INPPD5 is downregulated by mmu-miR-155-5p, leading to the production of major pro-inflammatory cytokines, such as 1L-1α and TNF-α, previously observed in macrophages [36], as a first response to infection by Leptospira.
Another common pathway identified in this study, between attenuated and virulent strains, was PI3K/AKT signaling with several target genes for mmu-miR-1555p involved in pro-inflammatory response. This pathway is responsible for B cell development trough activation of several genes. Inhibited PI3K signaling leads to immunodeficiency, autoimmunity activation and leukemia [47–48]. Cheung et al., [49] have shown that IL-10 inhibits miR-155, but this process is dependent on the presence of INPPD5 (a target to miR-155), and also that the activation of PI3K/AKT signaling abolished IL-10-inhibition of miR-155, resulting in an increase of miR-155 [49]. It is plausible to infer that miR-155 can be dependent of PI3K/AKT signaling to promote inflammation in L. interrogans infection, and that mmu-miR-155 could be a master regulator of pro-inflammatory process in leptospiral infection and a potential therapy target or biomarker.
Another goal in the present study was to identify differences in macrophageal response to strains, varying only in virulence. We have identified 16 specific canonical pathways potentially regulated by miRNAs modulated following infection with virulent L. interrogans. Cellular movement and cell-to-cell signaling interactions are functions related with the canonical pathways IL-8 (CXCL8) signaling, Tec kinase signaling, Epithelial adherens junction signaling and Gα12/13 signaling. It is well know that Leptospira causes changes in adherens junctions and endothelial cells increasing vascular permeability, potentially leading to severe illness [50]. The small GTPase RhoA (ras homolog family member A), a protein that regulates actin cytoskeleton and the remodeling of cell junctions, appears in most of the pathways mentioned above, as a direct target of miR-155-5p. In a recent study, Sato & Coburn [50] found a slight elevation of RhoA protein levels in endothelial cells following infection of both virulent and saprophyte Leptospira strains. In our macrophage cells, we observed an increase of miR-155-5p, which could lead to a decrease in RhoA, contrary to what was observed by Sato & Coburn in the endothelial cell line. This could be due to a cell type difference, to a difference in time of infection (8h in our study versus 24h in the aforementioned study), or to a difference in regulation of RhoA between the cell types.
A very common occurrence in patients with leptospirosis is a coagulation disorder causing lung hemorrhage [4]. The mechanism behind this Leptospira-induced hemorrhage is not fully understood. Fernandes et al, [51] provide evidence that lower serum levels of prothrombin and antithrombin III in patients with the disease is related to the observed hemorrhage. Liver-produced prothrombin remains inactive in circulation until being proteolytically cleaved to form thrombin to start clot formation by converting fibrinogen to fibrin. Here, we can suggest that Prothrombin (Coagulation Factor II) is potentially downregulated by miR-155-5p, suggesting a role for epigenetically mediated post-transcriptional control of clot formation in leptospirosis patients. In fact, recent studies support the idea that miR-155-5p can be secreted to act as modulators elsewhere. Wang et al. [52] have just shown that macrophages secrete miR-155-5p that can act as paracrine regulators of inflammation during cardiac injury. Alexander et al. [53] have also demonstrated that miR-155 present in exosomes can pass between immune cells in vivo and promote endotoxin-induced inflammation in mice.
Pathogen invasion into host cells is crucial for pathogenicity. Through phagocytosis, macrophages can kill the invading bacteria early in the process of infection. Both macropinocytosis signaling and clathrin-mediated endocytosis signaling are pathways significantly identified following infection with the virulent strain. In the macropinocytosis process, two genes appear to be down-regulated, Colony Stimulating Factor 1 Receptor (CSF1R), a target of miR-155-5p and SRC (Proto-oOncogene, nNonReceptor Tyrosine Kinase (SRC), targeted by mmu-miR-203-3p. CSF1R is a type III tyrosine kinase receptor, involved in cell proliferation and survival, and when phosphorylated, this receptor activates SRC kinases to initiate the signal cascade required for macropinocytosis process [54–55]. Therefore these miRNAs have potential to control initial signaling cascades of macropinocitosis.
During phagocytosis, one of the most effective weapons used by macrophages to kill invading leptospires is nitric oxide (NO) and reactive oxygen species (ROS), which induce an antimicrobial response [56–57]. On the other hand, an excessive production of O2 by macrophages can affect homeostasis [56]. This burden of intracellular oxygen demand by macrophages to kill pathogens has important collateral effects that can contribute to the inflammatory process through hypoxia in tissues [58] and DNA damage [59].
Further, Luo et al., [56] reported that Erythropoietin signaling, significantly identified in our study as a response to virulent infection, has a vital function in regulation of acute inflammatory conditions in hypoxia. This process of inflammation regulation appears to be inhibited in our acute infection, which could ultimately be associated with the exacerbated inflammation commonly seen in leptospirosis.
Hu and colleagues [59] report that leptospiral infection in macrophages induces cell cycle arrest dependent of p53/p21. Here we identified that the target pathway p53 signaling is regulated, following virulent infection, by modulation of miRNAs. This pathway can be activated by DNA damage, hypoxia, cytokines, metabolic changes, viral infection or oncogenes [60–61], and our study adds bacterial infection by Leptospira sp as another activator. This pathway triggers three important processes in the host cell, cell cycle arrest for DNA repair, apoptosis and cell survival. Here we found that BCL2 (anti-apoptotic gene) is potentially downregulated by mmu-miR-7667-3p following infection with L. interrogans, leading us to suggest that cell survival could be at risk following L. interrogans infection of macrophages.
Lastly, another virulent specific pathway identified in our study, the retinoic acid receptor (RAR) activation, has not been previously described in leptospirosis. This canonical pathway is related to development, differentiation, apoptosis and homeostasis, mostly participating in phosphorylation of several signaling pathways, as reviewed by [62], and also has fundamental importance in acquired and adaptative immune responses, with an important role in clonal expansion, differentiation, survival of pathogen-specific CD8 T cells, and bacterial clearance, as confirmed by a knockout model of RAR in mice [63]. These nuclear receptors act on recruitment of the transcriptional machinery to DNA response elements, regulating other complexes like nuclear factor kappa B (NF-κB complex) [62], SMAD complex [63], and also interact with other signaling pathways like PI3K/Akt and PTEN [62], which are vital to immune response. Here we found genes like Mothers Against Decapentaplegic Homolog (SMAD1/2) as a predicted target to mmu-miR-155-5p, SMAD7/9 targeted by miR- 7667-3p, TGF-β, that indirectly increases activation of SMAD complexes and is a target of mmu-miR-7069-3p and, finally, nuclear factor kappa B (NF-κB complex), a predicted target of mmu-miR-155-5p [64]. We suggest that upregulation of these miRNAs following macrophageal infection by L. interrogans can negatively regulate immune response to virulent leptospires through modulation of RAR activation. Interestingly, it has been demonstrated that vitamin A, a ligand of RAR, has antimicrobial activity against monocytes infected with Mycobacterium tuberculosis, in a mechanism dependent of intracellular cholesterol transporter 2 (NPC2) [65], raising the question as to whether this effect extends to other bacterial infections.
Our data provides the first evidence that host miRNAs are regulated by Leptospira infection in macrophages in a virulence- and species-specific manner in vitro. A large number of the identified miRNA targets participate in key processes involved in the immune response. Characterization of this regulatory network may help to understand the pathogenesis of leptospirosis and to identify miRNAs as biomarkers of infection or as targets for therapy. In conclusion, we suggest that post-transcriptional regulation by miRNAs play a role in the host’s response to leptospirosis infection.
Three types of bacterial samples were utilized in this study, Leptospira interrogans serovar Copenhageni strain FIOCRUZ-L1-130, as a virulent strain; the pathogenic culture-attenuated L. interrogans serovar Copenhageni strain M-20; and Leptospira biflexa serovar Patoc strain FIOCRUZ-Patoc I as a saprophyte strain.
Bacteria were maintained in Fletcher semi solid culture medium, and incubated at 30°C. To restore bacterial virulence in strain L1-130, 1mL of cultured bacteria was inoculated intraperitoneally in hamsters (Mesocricetus auratus) and later recovered from kidneys. Attenuated strain did not undergo intraperitoneal inoculation in hamsters [66]. The inoculum was quantified using the camera of Petroff-Hausser.
Bacterial samples were provided by Laboratory of Preventive Veterinary Medicine of University of São Paulo (USP). Production of these samples were in accordance with Ethics Committee for Animal Use (FOA-FMVA UNESP), under protocol number 2015–00895. Following bacterial arrival, no animal experimentation was performed in the experiments described herein.
Murine monocyte-macrophage cells (Mus musculus monocyte-macrophage cell line J774A.1), provided by the Paul Ehrlich cell bank, Rio de Janeiro, Brazil, was maintained in RPMI1640 media (Sigma, USA) supplemented with 10% heat-inactivated fetal bovine serum (Gibco, USA), 100ug/mL streptomycin (Sigma Chemical Co St. Louis, MO), 0.03% Lglutamine solution (Sigma) and 100 UI/mL of penicillin. Cells were incubated at 37° C, 5% CO2 until formation of a confluent monolayer in 6-well cell culture plates (3cm/well).
Cultured cells were washed three times with sterile phosphate buffer solution (pH 7,2) for removal of antibiotics and non-adherent cells. L. interrogans and L. biflexa were harvested by centrifugation and the pellet was resuspended in RPMI-1640 media (Sigma), and 100:1 bacteria:cell were added to macrophages at confluency (MOI of 100), as previously described [24]. Treatments, performed in three biological replicates, were carried as follows: infection of macrophages with a virulent strain (L. interrogans), infection with attenuated strain (L. interrogans), saprophyte strain (L. biflexa) and non-infected macrophages (control). All treatments were incubated in fresh RPMI medium, without antibiotics, for 8h at 37° C, 5% C02. Rate of infection did not differ between strains (78, 85 e 80% for saprophyte, attenuated and virulent, respectively). Following this period, RNA extraction was immediately performed as described below.
Total RNA was extracted from macrophages with a miRVana miRNA Isolation Kit (Ambion, Austin, TX, USA) according to the manufacturer’s instructions. RNA samples were immediately stored at -80°C. Quantification was performed using NanoDrop (ND-2000 spectrophotometer, Thermo Scientific, Wilmington, DE, USA) and quality of samples was assessed using capillary electrophoresis (Bioanalyzer 2100 Agilent, Santa Clara, CA, USA). All samples used for microarray analysis had a RIN of 10.
MicroRNA profiles were obtained from 250ng/sample of total RNA (RIN 10) using the FlashTag Biotin HSR RNA Labeling Kit, and the Affymetrix miRNA 4.1 Array strip (Affymetrix, Santa Clara, California, EUA), containing 3195 murine specific probes of miRNA, according to the manufacturer’s instructions. A recommended ELOSA quality control assay was run for all samples, and hybridization of samples to the strips was carried at 48°C for 20h. Following this period, strips were processed and scanned using the GeneAtlas System (Affymetrix). Raw intensity values were background corrected, log2 transformed and then quantile normalized by the software Expression Console (Affymetrix) using the Robust Multi-array Average (RMA) algorithm. Data files were deposited at Gene Expression Omnibus (GSE105104). Statistical analysis was performed in the TAC software (Affymetrix) by ANOVA (fold change ± 1.5, p <0.01).
For identification of target genes we employed the miRNA Target Filter Analysis from the Ingenuity Pathway Analysis (IPA) software (Qiagen). For selection, we opted to use conservative filters, allowing only experimentally observed and highly predicted targets to be selected (Supplementary table). For the identification of canonical pathways potentially regulated by the differentially expressed miRNAs, we employed the Benjamini-Hochberg (BH) correction for multiple testing (BH corrected p<0.05).
For validation of miRNA expression in infected macrophages (saprophyte, attenuated and virulent strains) and non-infected control macrophages, we employed the miScript miRNA PCR System (Qiagen-Valencia, CA, USA) for preparation of cDNA and realtime PCR, according to manufacturer´s instructions. Validated inventoried primers employed were purchased from Qiagen. PCR was performed using a Stratagene QPCR System Mx3005P (Agilent Technologies, Santa Clara, CA, USA), following instructions on the miScript miRNA PCR System´s manual. Expression levels were determined using standard curves for all miRNAs at each individual run, and the expression of candidate miRNAs is presented as a ratio to the control miRNA SNORD96A.
Differential expression of each miRNA was determined by ANOVA with two criteria, a fold change of ±1.5 comparing all infected groups to the non-infected control and p-value<0.01. Real time PCR data was analyzed using least-squares analysis of variance and the general linear model procedures of SAS (SAS Institute, Cary, NC, USA; p<0.01). Comparison of means was done using Duncan’s multiple range test, and significance was set at p<0.05.
mmu-miR-155-3p (MIMAT0016993); mmu-miR-155-5p (MIMAT0000165); mmu-miR-221-5p (MIMAT0016070); mmu-miR-1946b (MIMAT0009443); mmu-miR-3473b (MIMAT0020367); mmu-miR-1946a (MIMAT0009412); mmu-miR-203-3p (MIMAT0000236); mmu-miR-222-3p (MIMAT0000670); mmu-miR-7667-3p (MIMAT0029841); mmu-miR-7069-3p (MIMAT0028045); mmu-mir-3473c (MI0018015); mmu-mir-7676-1 (MI0025017); mmu-mir-7676-2 (MI0025018); mmu-miR-1894-5p (MIMAT0007877); mmu-miR-702-5p (MIMAT0022931); mmu-miR-7053-3p (MIMAT0028011); mmu-miR-6987-3p (MIMAT0027877); mmu-miR-6370 (MIMAT0025114); mmu-mir-717 (MI0004704); mmu-miR-7067-5p (MIMAT0028038); mmu-mir-124-3 (MI0000150); mmu-mir-124-1 (MI0000716); mmu-mir-124-2 (MI0000717); mmu-miR-222-5p (MIMAT0017061); mmu-miR-7119-3p (MIMAT0028136); mmu-miR-5100 (MIMAT0020607); mmu-miR-2137 (MIMAT0011213); mmu-miR-5046 (MIMAT0020540); mmu-miR-8100 (MIMAT0031403).
|
10.1371/journal.pgen.1007677 | Formation of phenotypic lineages in Salmonella enterica by a pleiotropic fimbrial switch | The std locus of Salmonella enterica, an operon acquired by horizontal transfer, encodes fimbriae that permit adhesion to epithelial cells in the large intestine. Expression of the std operon is bistable, yielding a major subpopulation of StdOFF cells (99.7%) and a minor subpopulation of StdON cells (0.3%). In addition to fimbrial proteins, the std operon encodes two proteins, StdE and StdF, that have DNA binding capacity and control transcription of loci involved in flagellar synthesis, chemotaxis, virulence, conjugal transfer, biofilm formation, and other cellular functions. As a consequence of StdEF pleiotropic transcriptional control, StdON and StdOFF subpopulations may differ not only in the presence or absence of Std fimbriae but also in additional phenotypic traits. Separation of StdOFF and StdON lineages by cell sorting confirms the occurrence of lineage-specific features. Formation of StdOFF and StdON lineages may thus be viewed as a rudimentary bacterial differentiation program.
| We show that the std fimbrial operon of Salmonella enterica undergoes bistable expression, a trait far from exceptional among loci that encode components of the bacterial envelope. However, an unsuspected trait of the std operon is the presence of two genes that encode pleiotropic regulators of gene expression. Indeed, StdE and StdF are DNA-binding proteins that control transcription of hundreds of genes. As a consequence, StdEF govern multiple phenotypic traits, and the fimbriated and non-fimbriated Salmonella lineages may differ in motility, virulence, conjugal transfer, biofilm formation, and potentially in other phenotypic features. We hypothesize that pleiotropic control of gene expression by StdEF may contribute to adapt the non-fimbriated lineage to acute infection and the fimbriated lineage to chronic infection.
| Phenotypic differences in isogenic bacterial cells grown in the same environment can be a consequence of noisy gene expression [1,2]. In other cases, the occurrence of distinct phenotypes is a programmed event that causes bistability, the split of the bacterial population into two lineages [3–5]. The mechanisms that cause bistability include genetic rearrangement, expansion and contraction of DNA sequence repeats, epigenetic control of gene expression by DNA methylation, and formation of regulatory feedback loops transmissible to daughter cells [5]. This variety, obviously indicative of independent evolution, may suggest that the ability of a given bacterial species to diversify into phenotypic lineages is a product of natural selection. Indeed, game theory shows that lineage formation can have selective value either as a division of labour or as a bet hedging [3,6–8]. In both kinds of strategies, the key biological entity is not the individual cell but the population [9]. Division of labour permits use of resources that are not available to a single cell type. In bet hedging, the fitness of each cell type is higher under different circumstances, and differentiation into distinct lineages preadapts the population to environmental changes.
In bacterial pathogens, phenotypic variation is often associated with virulence [10]. Variation in surface antigens such as flagellin, fimbrial and non-fimbrial adhesins, and the lipopolysaccharide help to evade the immune system [11]. In Salmonella, bistability in surface structures can also prevent cross-immunity between different serotypes [12]. Additional benefits from the formation of bacterial lineages include protection against host defence mechanisms other than the immune system [13], resistance to bacteriophages [14] and antimicrobial substances [15,16], and optimization of metabolic adaptation [17].
Fimbriae are virulence factors that promote attachment of bacterial cells to specific host tissues [18]. In S. enterica serotype Typhimurium, the ability of fimbriae to agglutinate yeast or red blood cells was described in 1966 [19]. Later studies have identified a large number of fimbrial loci including fim [20], csg [21], [22], stf [23,24], saf [25], stb, stc, std, sth, sti, and stj [26].
This study deals with std, one of the Salmonella operons initially identified by genome sequencing [26]. Absence of std in enterobacterial genera other than Salmonella suggests acquisition by horizontal transfer [27]. Std fimbriae bind specific receptors of the cecal mucose in the large intestine, and may play a role in chronic intestinal infection [28,29]. The std operon contains six genes (stdABCDEF) which are co-transcribed from a promoter located upstream of stdA [30]. Expression of the std operon is under transcriptional control by DNA adenine methylation and by HdfR, a poorly known LysR-like transcription factor [31].
This study shows that std expression is bistable, a trait shared with other adhesin-encoding operons [32]. A difference, however, is that std appears to be more than just a fimbrial operon: two products of the std operon, StdE and StdF, have DNA-binding capacity and activate or repress transcription of multiple genes. As a consequence, StdON and StdOFF subpopulations may differ not only in the possession of Std fimbriae but in additional phenotypic traits. Pleiotropic control of gene expression upon formation of StdON and StdOFF lineages may thus be viewed as an example of rudimentary, inconspicuous bacterial cell differentiation.
Single cell analysis of std expression was performed by flow cytometry in a stdA::gfp strain (SV9597). This strain carries a gfp transcriptional fusion downstream of stdA. Insertion of gfp does not cause polar effects on downstream genes of the std operon. A representative experiment presented in Fig 1, panel A reveals the existence of two subpopulations: a major subpopulation that does not show stdA::gfp expression (StdOFF, >99% of cells) and a minor StdON subpopulation that shows stdA::gfp expression (StdON, <1% of cells). Formation of the StdON subpopulation was abolished in a strain that lacks HdfR, a LysR-type transcription factor previously described as an activator of std transcription [31] (Fig 1, panel A). HdfR is thus necessary for formation of the StdON lineage.
Evidence that expression of the std operon occurs only in a subpopulation of S. enterica cells was confirmed by fluorescence microscopy. Wild type S. enterica cells were labelled with rabbit anti-StdA serum and goat anti-rabbit IgG-FITC conjugated antibody. A small fraction of cells harboured Std fimbriae, thereby confirming the existence of StdOFF and StdON subpopulations (Fig 1, panel B).
Evidence that the products of two downstream genes of the std operon, StdE and StdF, controlled the expression of genes located outside the std operon [30] led us to investigate the extent of gene regulation by StdE and StdF. Transcriptomic analysis was performed in a strain that constitutively expressed stdEF (StdEF+, SV8141) and in a strain carrying an in-frame deletion of both genes (StdEF–, SV8142). In strain StdEF+, both stdE and stdF are transcribed from the PLtetO promoter inserted upstream of stdE on the Salmonella chromosome, and the native stdA promoter and all genes upstream of stdE are deleted [30] (S1 Fig). Choice of PLtetO was based on the fact that this promoter renders moderate, constitutive expression [33,34]. Strain StdEF−contained the same insertion of the PLtetO promoter and a complete deletion of the std operon (S1 Fig, panel B). Transcriptomic analysis was performed using a previously described S. enterica ser. Typhimurium SL1344 gene array [35]. Raw data from transcriptomic analysis were deposited at the Gene Expression Omnibus (GEO) database (http://www.ncbi.nlm.nih.gov/geo/), with accession number GSE45488.
A large number of S. enterica genes showed different RNA levels in StdEF+ and StdEF−strains. Table 1 include only loci whose RNA levels differed more than 4-fold between the StdEF+ and StdEF−strains. A detailed gene description of these loci, including each individual fold change, is provided in S1 Table.
Downregulation by StdEF was observed at many loci, suggesting that StdE and StdF are often repressors of gene expression in Salmonella. The list of downregulated loci was heterogeneous, and included genes located in pathogenicity islands SPI-1, SPI-4 and SPI-5, the flhDC master regulatory operon, the flagellar operons flg, flh, fli, fljB and motAB, the che operon, the trg, tcp, tsr and aer genes involved in chemotaxis, the poorly known ygiD gene involved in biofilm formation, and additional loci involved in metabolism or having miscellaneous or unknown functions.
The list of upregulated loci was also heterogeneous, and included the tra operon encoded on the pSLT plasmid, genes involved in metabolism, and loci with miscellaneous or unknown functions. An interesting observation was the presence of hdfR among the StdEF-upregulated genes. Because the HdfR gene product is a transcriptional activator of std expression [31] (Fig 1A), upregulation of hdfR transcription by StdE and StdF may suggest the existence of a positive feedback loop for autogenous regulation of the std operon.
Validation of transcriptomic data was achieved by quantitative real time PCR, monitoring RNA production at loci that had shown differential expression in StdEF+ and StdEF−strains. The list includes genes involved in motility (motA, flgE), conjugation (traA), chemotaxis (trg), virulence (hilA, sipB) and transcriptional regulation (hdfR). All the loci analyzed were found to be under StdEF control (Fig 2), and the gene expression changes detected by RT-PCR correlated well with those obtained by transcriptomic analysis.
The subcellular localization of StdE and StdF was determined using chromosomal 3xFLAG-tagged versions of the StdE and StdF proteins (strains SV9324 and SV9325, respectively). Since the std operon is only expressed in a small fraction of cells in the wild type (Fig 1), this analysis was performed in a Dam− background [31]. Electrophoretic separation of cell fractions (cytoplasm, cytoplasmic membrane and outer membrane) was followed by Western blot analysis of the separated cell samples using a commercial anti-FLAG antibody. DamX, TraT and Lon were used as localization controls [36]. Both StdE and StdF were found in the cytoplasmic fraction (Fig 3, panel A).
Cytoplasmic localization of StdE and StdF, together with the evidence that they control the expression of multiple S. enterica genes, raised the possibility that these proteins might have DNA binding ability. This hypothesis was tested by chromatin immunoprecipitation followed by DNA sequencing (ChIP-seq). The strain used for ChIP-seq carried the stdEF genes under the control of PLtetO (as in the strain used in transcriptomic analysis), and contained a StdF variant labeled at its C-terminus with a 3xFLAG epitope (PLtetO-stdEF-3xFLAG; SV7850). This construct allowed the detection of StdE with a cognate anti-StdE antibody while StdF was detected with an anti-FLAG antibody.
Raw and processed data from ChIP-seq analysis have been deposited at the Gene Expression Omnibus (GEO) database (http://www.ncbi.nlm.nih.gov/geo/), with accession number GSE113562.
ChIP-seq analysis revealed that StdE and StdF bind multiple sites in the S. enterica genome (Fig 3, panel B). The number of DNA sequence reads detected upon immunoprecipitation with the StdE antibody was higher than that found for StdF (Fig 3, panel B), suggesting that StdE and StdF may bind DNA independently, and that StdE may bind more efficiently than StdF. The pie chart shown in Fig 3, panel C, summarizes the number of binding sites detected for StdE (171), for StdF (105), and for both StdE and StdF (60). Peak boundary sequences for StdE (231) and StdF (165) were extracted from the reference genome, and were analyzed with a motif-finding algorithm. Distinct 6-bp motifs for StdE and StdF binding were identified (Fig 3D).
Binding of StdE and/or StdF to specific promoters or upstream regulatory regions permitted a tentative interpretation of data from transcriptomic analysis. For instance, binding of StdEF was detected upstream of the flhDC flagellar operon and also upstream of the conjugal transfer tra operon (see below). StdE and StdF binding sites were also detected within coding regions, perhaps indicating the existence of uncharacterized promoters [37]. On the other hand, intra-ORF binding has been documented previously for other transcriptional regulators [38–40].
Phenotypic validation of the observations provided by gene expression analysis and ChIP-seq was pursued by monitoring motility, epithelial cell invasion, biofilm formation, and conjugal transfer of the virulence plasmid. In certain cases, genetic analysis was also performed to identify regulatory mechanisms and epistatic relationships. Relevant observations were as follows:
Considering that the std operon is expressed in a minor subpopulation of cells (Fig 1), the phenotypic traits detected in the StdEF+ strain can be expected to occur in a small fraction of cells only. To monitor the occurrence of lineage-specific traits, StdON and StdOFF cell lineages were separated by two independent single cell techniques: fluorescence activated cell sorting (FACS) and magnetic activated cell sorting (MACS). A difference between these procedures is that FACS yields live cells while MACS yields fixed cells. The small size of the StdON lineage made these experiments challenging as sorting is unable to yield pure StdON and StdOFF cell lineages. However, the size of the StdON lineage increased >230 fold (from 0.3% to 70%), thus permitting comparison with the StdOFF lineage.
Motility assays on soft agar plates provided evidence that StdON cells show reduced motility as predicted by transcriptomic analysis (Fig 7A and 7B). In turn, the occurrence of lineage-specific transcriptional patterns was confirmed by RT-PCR analysis of sipB, traA, and flhD gene expression upon separation of StdOFF and StdON cells by MACS (Fig 7C). The stdA gene was included as a control to validate cell separation (Fig 7C). The differences in the relative levels of sipB, flhD and traA mRNAs detected between sorted StdOFF and StdON cells were consistent with the observations made upon constitutive expression of StdE and StdF (Fig 2).
The addition of the std operon of S. enterica to the list of bacterial loci that show bistable expression is far from exceptional: subpopulation formation is common among loci that encode envelope structures such as the flagellum [52], the O-antigen of the lipopolysaccharide [14,52] and fimbrial and non-fimbrial adhesins [53–55]. However, an unsuspected ability of the std operon is control of host genes. Constitutive expression of the StdE and StdF proteins, which are encoded by downstream genes in the std operon, brings in major changes in the transcriptome (Table 1). At most loci, StdE and StdF appear to be repressors of transcription (143 loci, Table 1). However, positive control by StdEF is also detected (27 loci, Table 1). Note that Table 1 includes only loci with differences in RNA content of 4-fold or higher; with a threshold of 2-fold differences, the number of downregulated loci would increase to 187, and the list of upregulated loci to 116 (GEO, accession GSE45488).
StdE and StdF are cytoplasmic proteins with DNA binding capacity (Fig 3), and a number of StdE and StdF binding sites are located at or near promoters of genes under StdE and/or StdF control (Figs 4, 5 and 6). The conclusion that StdE and StdF may exert direct transcriptional control is supported by the relatedness of StdE with the transcriptional activators GrlA from E. coli and CaiF from Enterobacter cloacae [30]. In turn, StdF is related to the transcriptional regulator SprB of S. enterica [30]. Phenotypic analysis confirmed that distinct transcriptomic profiles of StdEF+ and StdEF−strains resulted in phenotypic differences: the StdEF+ strain showed reduced motility (Fig 4), reduced invasion of epithelial cells (Fig 5), increased conjugal transfer (Fig 6), and reduced biofilm formation (S3 Fig).
Reduced motility of the fimbriated lineage, a feature reminiscent of the StdEF+ strain, was confirmed upon cell sorting (Fig 7A and 7B). The smaller difference in motility between StdOFF and StdON cells may be explained by formation of StdOFF cells during growth on motility agar. Note that growth can be expected to partially blur differences, and a long incubation time is required due to the small number of cells recovered by sorting. Hence, the real differences between StdON and StdOFF lineages are probably larger than shown under the conditions of our experiments.
The occurrence of a lineage-specific transcriptional pattern in StdON cells was also confirmed by RT-PCR analysis upon separation of fimbriated and non-fimbriated cells by magnetic activated cell sorting (MACS) (Fig 7). Hence, it seems reasonable to conclude that the StdON lineage may display phenotypes similar or identical to those detected upon constitutive expression of StdE and StdF. Note than phenotypes other than motility were not tested in sorted StdON cells because the experiments required previous growth, which would yield a mixture of StdOFF and StdON cells with outmost predominance of the StdOFF lineage.
Adhesion to the cecal epithelium by Std fimbriae may adapt the StdON lineage to colonize the large intestine [28]. The size of the StdON lineage in the intestine remains unknown at this stage. However, a small StdON subpopulation may be sufficient for chronic infection, and production of StdOFF cells can permit Salmonella shedding into the environment as in other types of chronic infection [56]. Formation of StdON and StdOFF lineages may thus be interpreted as a division of labour that adapts the S. enterica StdON subpopulation to acute infection and the StdOFF subpopulation to chronic infection. An alternative interpretation is that lineage formation may increase the chances of host colonization by undertaking two distinct strategies [7]. Whatever the case, the occurrence of additional lineage-specific traits may contribute to adaptation, and the adaptive value of some such traits can be envisaged. For instance, downregulation of flagellar synthesis in StdON cells may contribute to immune evasion and reduce the energetic burden of building a machinery that may be superfluous in cells attached to a surface. A similar argument may explain the adaptive value of downregulation of pathogenicity island 1 (SPI-1) in the StdON lineage as Salmonella does not invade epithelial cells in the large intestine, and the SPI-1 type 3 secretion apparatus is a costly structure [45]. Activation of the tra operon of the virulence plasmid is a more enigmatic trait. However, high rates of pSLT conjugal transfer have been previously detected in the mammalian intestine [57]. The potential adaptive value of inhibition of biofilm formation in the StdON lineage cannot be understood at this stage. Speculation on the existence of additional lineage-specific phenotypic differences would be likewise premature. However, the high number of genes under StdE and StdF control revealed by transcriptome analysis suggests that such differences may exist.
Pleiotropy may be an unusual capacity of a fimbrial locus. However, additional examples of pleiotropic switching have been described in the bacterial world. Phase variation in beta-hemolytic properties in the dental pathogen Streptococcus gordonii is accompanied by changes in adhesive properties and surface antigens [58]. In Pseudomonas aeruginosa, small colony variants show increased biofilm formation and impaired motility and chemotaxis [59]. In a Salmonella strain that causes asymptomatic infection in pigs, phase variation of type I fimbriae is accompanied by changes in adhesion, uptake by phagocytes, and survival within phagocytes [60]. Colony variants of Haemophilus influenzae also differ in multiple surface proteins [61]. While these studies were mostly descriptive, a molecular mechanism that may exert pleiotropic control of gene expression has been described in the phasevarions of Neisseria, Campylobacter, Haemophilus, and other bacterial pathogens [62,63]. In a phasevarion, bistable expression of a DNA methyltransferase gene gives rise to bacterial lineages that differ in the presence or the absence of methylation at multiple genome locations. As a consequence, the lineages may differ in the expression of loci sensitive to the methylation state of their promoters. If transcriptional control by DNA methylation occurs in multiple genes, a tentative analogy with the StdE/StdF pleiotropic switch can be drawn.
In Salmonella, sequential acquisition of pathogenicity islands and other genetic determinants has enabled the pathogen to colonize animals [48,64], and the accommodation of such entities in the host regulatory network has been made possible by complex transcriptional and postranscriptional controls [65]. In the case of std, accommodation in the host regulatory network is known to be exerted by Dam methylation and HdfR [31], and additional controls may exist. However, reciprocal control of host loci by the std operon introduces an interesting twist into the contribution of horizontal transfer to Salmonella evolution. Regulation of host genes has been described in prophages [66,67] and in plasmids [68–70]. Occurrence of an analogous capacity in a small genetic entity devoid of autonomous lifestyle is remarkable, especially if one considers the high number of genes under StdEF control.
Salmonella enterica strains listed in S3 Table belong to serovar Typhimurium and derive from the mouse-virulent strain SL1344 [71]. For simplicity, Salmonella enterica serovar Typhimurium is often abbreviated as S. enterica. E. coli BL21 [F−dcm ompT hsdS (rB−mB–) gal [malB+]K12(λS)] (Stratagene, La Jolla, CA) was used for protein purification. Targeted gene disruption was achieved using plasmids pKD3, pKD4 or pKD13 as templates to generate PCR products for homologous recombination [72]. Antibiotic resistance cassettes introduced during strain construction were excised by recombination with plasmid pCP20 [72]. Addition of a 3xFLAG epitope tag to the stdA coding sequence was carried out using plasmid pSUB11 as template [73]. Primers used in strain construction are shown in S4 Table. Transductional crosses using phage P22 HT 105/1 int201 [74] were used for strain construction operations involving chromosomal markers. The transduction protocol has been previously described [75]. To obtain phage-free isolates, transductants were purified by streaking on green plates. Phage sensitivity was tested by cross-streaking with the clear-plaque mutant P22 H5. Construction of strains SV9324 (Δdam231 stdEF::3xFLAG) and SV9325 (Δdam231 stdF::3xFLAG) was performed by transductional crosses from SV6749 and SV6502 [30] to JH3294 (Δdam231) [76]. Construction of strain SV9287 (PLtetO-stdEF::3xFLAG) was performed by transduction from SV6510 [30].
For construction of the transcriptional gfp fusion of strain SV9597 (stdA::gfp), a fragment containing the promoterless green fluorescent protein (gfp) gene and the chloramphenicol resistance cassette was PCR-amplified from pZEP07 (Hautefort et al., 2003) using oligonucleotides stdASTOPGFP P1 and stdAGFP P2 (S4 Table). The fragment was integrated into the chromosome of S. enterica [72], and integration was verified using oligonucleotides stdA E1 and stdA E2 (S4 Table). The CmR resistance cassette was changed to KmR using oligos Cm-P1-R and FliC-GFP-Km-P4, and later excised with plasmid pCP20 [72].
Construction of SV8141 (PLtetO-stdEF) and SV8142 (PLtetO-ΔstdEF) was performed by transducing the wild type with lysates from SV6503 (PLtetO-stdEF) and SV6634 (PLtetO-ΔstdEF) [30]. In both constructions, the PLtetO promoter is inserted upstream of stdE on the Salmonella chromosome. In both strains, the upstream std genes and the promoter of stdA are deleted. A CmR cassette in reverse orientation linked to the PLtetO promoter provided a selectable marker. For construction of SV7553 and SV7552, the CmR resistance gene was replaced with a KmR cassette amplified from pKD13 [72] with oligos Km PLtetO P1 and Km PLtetO P2 (S4 Table). The resulting PCR product was transformed into strains SV8141 (PLtetO-stdEF) and SV8142 (PLtetO-ΔstdEF) containing pKD46 [72]. KmR colonies were selected on LB + kanamycin. Finally, the KmR cassette introduced during construction was excised by recombination with plasmid pCP20 [72].
Bertani’s lysogeny broth (LB) was used as standard rich medium. Solid media contained agar at 1.5% final concentration. Cultures were grown at 37°C. Aeration of liquid cultures was obtained by shaking at 200 rpm in an Infors Multitron shaker. Antibiotics were used at the final concentrations described elsewhere [77].
To prepare cells for RNA extraction, 3 ml of fresh LB was inoculated with a 1:100 dilution from an overnight bacterial culture, and incubated with shaking. A 2 ml aliquot from a stationary culture (OD600≅2) was centrifuged at 13,000 rpm, 4°C, during 5 min. The pellet was then resuspended in 100 μl of lysozyme (3 mg/ml in water; Sigma Chemical Co.), and cell lysis was facilitated by a freeze-thaw cycle. After lysis, RNA was extracted using 1 ml of TRIsure reagent following manufacture’s instructions (Bioline, Taunton, Massachusetts, USA). Total RNA was resuspended in 150 μl of RNase-free water, and subsequently cleaned by extraction with acidic phenol, followed by a second extraction with chloroform:isoamilic alcohol (24:1). After extraction, RNA was precipitated with ethanol and 3 M sodium acetate, and the dried pellet was resuspended in RNase-free water. The quantity and quality of the RNA was determined using a ND-1000 spectrophotometer (NanoDrop Technologies).
Transcriptomic analyses were performed using the Salmonella enterica serovar Typhimurium SL1344 4X72K array [35]. Hybridation and microarray scanning were performed at the Functional Genomics Core of the Institute for Research in Biomedicine, Baldiri Reixac, Barcelona, Spain (http://www.dnaarrays.org/). Normalization of the expression signals was done with RMA (Irizarry et al., 2003) using Partek Genomics suite 6.5 (6.11.0207). Raw transcriptomic data were deposited at the Gene Expression Omnibus, G.E.O, database (http://www.ncbi.nlm.nih.gov/geo/) under accession number GSE45488. Differential gene expression was assessed using the Limma’s R package [78]. Background correction and normalization of gene expression were done using RMA algorithm [79]. A gene was considered significant for a Benjamini and Hochberg-corrected (BH) p-value of <0.05 in a moderated t-statistic and a log2 fold change > 2. Gene identities were annotated according to the S. enterica ser. Typhimurium strain SL1344 genome sequence (ftp://ftp.sanger.ac.uk/pub/pathogens/Salmonella/STmSL1344.dbs). RNA isolation from ~106 S. enterica cells sorted by MACS was performed using Direct-zol RNA MiniPrep (Zymo Research, Irvine, California, USA), following manufacturer’s instructions.
For quantitative RT-PCR, Salmonella RNA was extracted from stationary phase cultures (OD600≅2) and from sorted cells as described above, and the concentration was determined using a ND-1000 spectrophotometer (NanoDrop Technologies). An aliquot of 1 μg of RNA was used for cDNA synthesis using QuantiTec Reverse Transcription Kit (Quiagen) following manufacturer’s instructions. Quantitative RT-PCR reactions were performed in a Light Cycler 480 II apparatus (Roche). Reactions were carried out in a total volume of 10 μl on a 480-well optical reaction plate (Roche), using Takara SYBR Premix Ex Taq reagent. Each reaction contained 4μl cDNA (1/10 dilution), 5 μl of 2X SYBR mix, 0.2 μl DYE II, and two gene-specific primers at a final concentration of 0.2 mM each. Real-time cycling conditions were as follows: (i) 95°C for 10 min and (ii) 40 cycles at 95°C for 15 s, 60°C for 1 min. A non-RT control (without reverse transcriptase) was included for each primer set. Triplicates were run for each reaction, and the Ct value is averaged from them. Absence of primer dimers was corroborated by running a dissociation curve at the end of each experiment to determine the melting temperature of the amplicon. Melting curve analysis verified that each reaction contained a single PCR product. Gene-specific primers were designed with ProbeFinder software (http://www.universalprobelibrary.com) from Roche Applied Science, and are listed in S4 Table.
For quantification, the efficiency of each primer pair was determined to be between 90%-110%, following the instructions for efficiency determination described in the “Guide to Performing Relative Quantification of Gene Expression Using Real-Time Quantitative PCR” (Applied Biosystems). Relative RNA levels were determined using the ΔΔCt method as described in the same guide. Each ΔΔCt determination was performed at least in three different RNA samples.
A DNA fragment containing stdE was amplified using oligonucleotides NdeIstdE-FOR and EcoRIstdE-REV, and cloned into NdeI- and EcoRI-digested pET28a (Novagen). The recombinant plasmid (pIZ1991) was verified by restriction analysis and DNA sequencing. For 6×His-StdE purification, plasmid pIZ1991 was transformed into E. coli BL-21. BL-21/pIZ1991 was grown in LB broth containing kanamycin, and expression of 6×His-StdE was induced with 1 mM isopropyl β-D-thiogalactopyranoside (IPTG). After 3 h induction, cells were centrifuged and resuspended in 10 ml of lysis buffer (20 mM Tris, 300 mM NaCl, 10 mM imidazole) per g of pelleted cells, and were lysed by sonication with 4 cycles of 30 seconds. The suspension was centrifuged at 10,000 rpm for 20 min at 4°C and the supernatant containing the soluble fraction of 6×His-StdE was transferred to a HisTrap HP nickel affinity chromatography column (GE Healthcare, Wauwatosa, WI, USA). The column was washed 3 times with 4 ml of washing buffer (20 mM NaH2PO4∙H2O, 0,5 mM NaCl, 30 mM imidazole). Protein elution was performed with 3 ml of elution buffer (20 mM NaH2PO4∙H2O, 0.5 mM NaCl, 300 mM imidazole). Elution fractions enriched in 6×His-StdE were selected. Imidazole was removed by dialyzing in cellulose membranes with PBS 1X. Purified 6×His-StdE protein was sent to Biomedal S.L (Sevilla, Spain) for polyclonal antisera production in rabbits. The working dilution was prepared based on manufacturer’s recommendations.
Strain PLtetO-stdEF::3xFLAG was used to perform ChIP-seq experiments. Twenty ml of fresh LB was inoculated with a 1:100 dilution from an overnight bacterial culture, and incubated with shaking at 200 rpm at 37°C. Cells collected at OD600≅2 were cross-linked with 1% formaldehyde at 37°C for 25 min, followed by quenching of the unused formaldehyde with 450 mM glycine for an additional incubation of 5 min. Cross-linked cells were harvested and washed with 10 ml of TBS pH7.6 (2.42 g/l Trizma base, 8 g/l NaCl). The washed cells were resuspended in 1 ml of lysis buffer (10 mM Tris-HCl pH 8, 20% sucrose, 50 mM NaCl, 10 mM EDTA) and after an additional centrifugation step, the cells were resuspended in 0.5 ml of lysis buffer with lysozyme (20 mg/ml; Sigma Chemical Co.). The cells were incubated for 30 min at 37°C and then treated with 4 ml of IP buffer (50 mM HEPES-KOH pH7.5, 150 mM NaCl, 1 mM EDTA, 1% Triton X-100, 0.1% Na deoxycholate, 0.1% SDS and 1mg/ml PMSF). The lysate was then sonicated using a Bioruptor (Diagenode) with 5 cycles of 7 minutes at high setting. Cell debris was removed by centrifugation for 20 min at 4°C and the supernatant was used as a cell extract for immunoprecipitation. The range of fragment sizes resulting from sonication was 100–500 bp, and the average fragment size was 300 bp (S2 Fig).
To immunoprecipitate StdE-DNA and StdF-DNA complexes, 800 μL of chromatin, 20 μL of Ultralink Immobilized protein A/G beads (Pierce) and 2 μL of the corresponding antibody were used. A control sample (mock-IP) with no antibody was included. Four samples were used for each antibody and four samples for the control. Incubation for 90 min was performed at room temperature on a rotating wheel. Beads were transferred to a Spin-X column tube (Costar) and centrifuged at 3,000 rpm for 1 min. Beads were gently re-suspended in 500 μL of IP buffer and incubated on the wheel for additional 3 min. This step was done twice. Beads were washed with 500 μL of IP salt buffer, IP wash buffer and TE pH 8.0 by resuspending and centrifuging the sample. The column was transferred to a fresh tube and the beads were resuspended in 100 μL of elution buffer (50 mM Tris-HCl at pH 7.5, 10 mM EDTA, and 1% SDS) and incubated at 65°C for 20min. After centrifuging at 3,000 rpm for 1 min, the flow-through was treated with 10 μL of 40 mg/ml Pronase (Roche) made up in TBS. The samples were heated at 42°C for 2h and 65°C for 6 h. The reactions were then kept at 4°C overnight. The samples were cleaned using a PCR clean-up Kit (Promega) and resuspended in 50 μL of H2O.
Input and ChIP DNA samples were sent for sequencing at the Functional Genomics Core Facility of the Institute for Research in Biomedicine, Barcelona (Spain). Next generation sequencing was carried out using Illumina’s sequencing technology. Ultra DNA Library Prep Kit (Illumina) was used for library preparation. Libraries were sequenced on Illumina’s Genome Analyzer II system. 50 nucleotides single end reads were obtained following strictly manufacturer’s recommendations. Illumina sequencing data were pre-processed with the standard Illumina pipeline version 1.5. BAM files reported by the sequencing facility were converted to FASTQ format with the BAM2FASTQ tool (https://gsl.hudsonalpha.org/information/software/bam2fastq).The quality of the sequence reads was examined using FASTQC [80] that reported the presence of Illumina adapters. The adapters were trimmed with the FASTX_CLIPPER tool of the FASTX-Toolkit suite (http://hannonlab.cshl.edu/fastx_toolkit/). Reads shorter than 40 nt were discarded. NCBI GCA_000210855.2 genome assembly of S. enterica SL1344 was used as reference genome. Mapping was performed with Bowtie [81] allowing two-mismatches for only unique alignment. Peaks were called using CisGenome version 2.0 [82] using default parameters. The IGV browser [83] was used for data visualization. Genes closest to a ChIP peak were identified using the bedtools suite [84]. Peak boundaries sequences were extracted from the reference genome using the fastaFromBed utility from the BEDTools suite (E) and analyzed with DREME [85] for motif discovery.
Levels of β-galactosidase activity were determined using the CHCl3-sodium dodecyl sulfate permeabilization procedure [86]. β-galactosidase activity data (Miller Units) are the averages and standard deviations from ≥3 independent experiments.
HeLa human epithelial cells (ATCC CCL2) were grown in DMEM containing 10% fetal calf serum and 1mM glutamine (Life Technologies). The day before infection, approximately 105 HeLa cells were seeded, using 24-well plates (Costar, Corning, New York, NY) containing 1 ml of tissue culture medium without antibiotics per well, and grown at 37°C, 5% CO2 to obtain 80% confluency. One hour before infection, the culture medium was removed and replaced by 0.5 ml fresh tissue culture medium without antibiotics. Bacteria were grown overnight at 37°C in LB with shaking, diluted into fresh medium (1:50), and incubated at 37°C without shaking up to O.D.600 0.6–0.8 (overnight). Bacteria were added to reach a multiplicity of infection (MOI) of 50:1 bacteria/HeLa cell. HeLa cells were infected for 30 min, washed 3 times with PBS, incubated in fresh tissue culture medium containing 100 μg/ml gentamicin for 1.5 h, and washed 3 times with PBS. Numbers of viable intracellular bacteria were obtained by lysing infected cells with 1% Triton X-100 (prepared in PBS) and subsequent plating. Invasion rates were determined as the ratio of viable intracellular bacteria vs. viable bacteria added to infect the HeLa cells.
Motility assays were carried out on motility agar plates containing 10 g/l tryptone (Difco), 5 g/l NaCl, and 0.25% bacto-agar [87]. A sterile stick was soaked in saturated bacterial cultures grown in LB, and used to inoculate motility agar plates. Bacterial motility halos were compared after growth at 37°C for 6 h. For motility assays of sorted cultures, the incubation time was 12-18h. A simultaneous viability test was performed by plating on LB, to warrant that the number of cells was the same for each subpopulation.
Biofilm formation was tested in LB medium [50]. Cultures were grown at 22°C for 7–10 days. For better visualization, the biofilm was stained with a 0.1% solution of crystal violet.
Cultures of the donor and the recipient were grown overnight in LB broth. Cells were harvested by centrifugation and washed with LB. Aliquots of both strains, 1ml each, were sucked onto a membrane filter with a 0.45 μm pore size with a donor/recipient ratio of 1:1. The filters were then placed on LB plates and incubated during 4 h at 37°C in a GasPak microaerophilic jar [88]. Conjugation frequencies were calculated per donor cell as previously described [57,88].
Cells from 1.5 ml of an exponential culture (OD600≅0.5) were collected by centrifugation, washed, resuspended in 1 ml TE buffer and fixed by adding the same volume of cold 70% ethanol. Ethanol-fixed cells (100 μl) were stained with polyclonal rabbit anti-StdA serum 1:250 [89]. After extensive washing with PBS + gelatin 0.02%, goat anti-rabbit antibody conjugated to FITC (fluorescein isothiocynate, 1:500) was used. Immunostained cells were placed in 10 μl mounting medium (40% glycerol in 0.02 M phosphate buffered saline, pH 7.5). 20 μl of ethanol-fixed cells were spread onto a poly-L-lysine-coated slide, and dried at room temperature. Slides of stained samples were stored at room temperature in the dark. Images were obtained by using an Olympus IX-70 Delta Vision fluorescence microscope (Olympus, Tokyo, Japan) equipped with a 100X UPLS Apo objective. Pictures were taken using a CoolSNAP HQ/ICX285 camera (Roper Technologies, Sarasota, FL) and analysed using ImageJ software (Wayne Rasband, Research Services Branch, National Institute of Mental Health, MD).
Bacterial cultures were grown at 37°C in LB until stationary phase (OD600≅2). Cells were then diluted in PBS to a final concentration of ~107 cells/ml. Data acquisition and analysis were performed using a Cytomics FC500-MPL cytometer (Beckman Coulter, Brea, CA). Data were collected for 100,000 events per sample and were analysed with CXP and FlowJo 8.7 softwares. Data are shown by dot plots (forward scatter [cell size] vs fluorescence intensity).
Stationary cultures were washed and resuspended in PBS to a final concentration of 5 × 106 cells/ml. Cells were sorted using a MoFlo Astrios EQ cytometer (Beckman Coulter, Brea, CA). Immediately before sorting, 5 × 106 cells were analyzed for GFP expression. Based on this analysis, gates were drawn to separate the ~0.3% of cells expressing GFP (StdON state) from the ~99.7% of non expressing GFP cells (StdOFF state). The whole population (ALL) was also analyzed after FACS.
Five hundred ml from a stationary culture (OD600≅2) of strain SV9600 (stdA::3XFLAG) grown at 37°C with shaking were collected by centrifugation. The pellet was washed with 10 ml of TE buffer and fixed by adding the same volume of cold 70% ethanol. Ethanol-fixed cells were washed with PBS containing 0.05% of Tween (PBS-T). This step was repeated three times. The pellet was resuspended in 5 ml of lysozyme solution (2 mg/ml lysozyme, 25 mM Tris-HCl pH 8.0, 50 mM glucose and 10 mM EDTA) and incubated at room temperature for 10 min. Cells were washed 3 times with PBS-T and resuspended and incubated for 30 min in 10 ml of 2% BSA made up in PBS-T. After collecting cells by centrifugation, anti-flag-PE antibody (Miltenyi Biotec S.L.) was added. After 1 h of incubation at room temperature and extensive PBS-T washing, anti-PE microbeads (Miltenyi Biotec S.L.) were added and incubated overnight at 4°C followed by 3 PBS-T washing. Separation of labelled and unlabelled cells was performed using an autoMACS Pro Separator (Miltenyi Biotec S.L.).
|
10.1371/journal.pntd.0006477 | Causes of acute undifferentiated fever and the utility of biomarkers in Chiangrai, northern Thailand | Tropical infectious diseases like dengue, scrub typhus, murine typhus, leptospirosis, and enteric fever continue to contribute substantially to the febrile disease burden throughout Southeast Asia while malaria is declining. Recently, there has been increasing focus on biomarkers (i.e. C-reactive protein (CRP) and procalcitonin) in delineating bacterial from viral infections.
A prospective observational study was performed to investigate the causes of acute undifferentiated fever (AUF) in adults admitted to Chiangrai Prachanukroh hospital, northern Thailand, which included an evaluation of CRP and procalcitonin as diagnostic tools. In total, 200 patients with AUF were recruited. Scrub typhus was the leading bacterial cause of AUF (45/200, 22.5%) followed by leptospirosis (15/200, 7.5%) and murine typhus (7/200, 3.5%), while dengue was the leading viral cause (23/200, 11.5%). Bloodstream infections contributed to 7/200 (3.5%) of the study cohort. There were 9 deaths during this study (4.5%): 3 cases of scrub typhus, 2 with septicaemia (Talaromyces marneffei and Haemophilus influenzae), and 4 of unknown aetiologies. Rickettsioses, leptospirosis and culture-attributed bacterial infections, received a combination of 3rd generation cephalosporin plus a rickettsia-active drug in 53%, 73% and 67% of cases, respectively. Low CRP and white blood count were significant predictors of a viral infection (mainly dengue) while the presence of an eschar and elevated aspartate aminotransferase and alkaline phosphatase were important predictors of scrub typhus.
Scrub typhus and dengue are the leading causes of AUF in Chiangrai, Thailand. Eschar, white blood count and CRP were beneficial in differentiating between bacterial and viral infections in this study. CRP outperformed procalcitonin although cut-offs for positivity require further assessment. The study provides evidence that accurate, pathogen-specific rapid diagnostic tests coupled with biomarker point-of-care tests such as CRP can inform the correct use of antibiotics and improve antimicrobial stewardship in this setting.
| Fever remains an important reason why people are hospitalised in Southeast Asia. We do not know the most common causes of fever in many regions of the tropics. This knowledge would help doctors decide on the most appropriate treatment in areas where access to diagnostics is difficult. Establishing diagnostic tests for all possible diseases in an area is expensive and often impractical. An alternative is to measure ‘marker’ chemicals in the blood which the body produces in response to infection. These are usually higher in patients with bacterial infections. Differentiating bacterial from viral infections will help reduce inappropriate antibiotic use, which can contribute to the development of antibiotic-resistant bacteria. In this study, we investigated the causes of fever in hospitalised patients in Chiangrai, northern Thailand, and assessed if two chemical markers (CRP and procalcitonin) could distinguish bacterial from viral infections. Scrub typhus, dengue and leptospirosis were the major causes of fever, and these were not always accurately diagnosed and managed. We also found that CRP was better than procalcitonin in differentiating bacterial from viral infections. These results should help improve the management of febrile patients and increase the awareness of these neglected tropical diseases that are potentially deadly if missed.
| Acute undifferentiated fever (AUF) remains the leading cause of hospitalisation among adults and children in urban and rural regions of Southeast Asia. The causes include common diseases such as dengue, scrub typhus, murine typhus, leptospirosis, and enteric fever, which continue to contribute significantly to the febrile disease burden [1–4]. Although malaria may present similarly, its overall incidence and impact on health in this region is declining [5].
In Laos, a prospective multicentre study investigating the causes of non-malarial fever revealed dengue, scrub typhus, Japanese encephalitis and leptospirosis as the major aetiologies in hospitalised adults and children once influenza was excluded [6]. In rural Thailand, dengue, scrub typhus, leptospirosis, murine typhus, and influenza have been identified as the most common causes of AUF among adults and children [4, 7]. Scrub typhus, enteric fever, flavivirus infection, leptospirosis and malaria were the main causes of fever in adults and children in the 1970s in rural Malaysia [8]. In febrile pregnant women on the Thai-Burmese border and in Laos, malaria, rickettsial infections, dengue, leptospirosis, typhoid and pyelonephritis predominate [9, 10]. Adverse neonatal and maternal outcomes were high in this group, particularly in those diagnosed with rickettsial infections [10, 11]. In Cambodian children, dengue, scrub typhus, bacteraemia (Salmonella enterica serovar Typhi was the commonest pathogen) and Japanese encephalitis were the major diagnoses [3].
These “causes-of-fever” studies that address a wide range of infectious diseases in diverse geographies are useful in informing clinicians and epidemiologists alike [12]. However, the majority of currently available fever studies suffer from selection bias, often rely on suboptimal diagnostic tools and non-uniform positivity criteria—limiting estimates of disease incidence and burden [13]. Performing these prospective studies correctly is costly, difficult and challenging–especially if representative geographical coverage is desired [12]. The current literature highlights a panel of AUF that represents the leading causes of fever in Asia and similarities in their clinical presentation and poor access to high-quality, affordable diagnostic tools frequently result in sub-optimal management [14]. Although progress in developing accurate, validated and cost-effective diagnostic tools for non-malarial pathogens, such as disease-specific rapid diagnostic tests (RDTs), have been made (e.g. combining NS1-antigen with IgM detection in dengue), recent modelling approaches suggest that testing for viral infections is unlikely to be cost-effective when considering direct health benefits, whereas RDTs for the detection of prevalent bacterial pathogens could be [15, 16].
Biomarkers such as C-reactive protein (CRP) and procalcitonin have some utility in delineating between bacterial and viral infections and guiding healthcare workers on the appropriate use of antibiotics in patients with respiratory tract infections in high income settings [17]. A retrospective study based on well-characterised samples of adults and children with febrile illnesses from Cambodia, Laos and Thailand demonstrated CRP was highly sensitive and moderately specific for discriminating between bacterial and viral infections [18]. Recently, CRP testing has been shown to reduce antibiotic prescription for acute respiratory illnesses in adults and children in primary healthcare settings in Vietnam [19]. In resource-constrained tropical settings, common treatable infections are being missed and inappropriate use of antibiotics is widespread. This highlights the potential impact of CRP RDTs on the precision of antibiotic use and contribution to the global strategy to combat antimicrobial resistance.
In this prospective study, we investigated the causes of AUF in adults admitted to the provincial hospital in Chiangrai, northern Thailand, and evaluated the use of CRP and procalcitonin tests in guiding appropriate antibiotic use.
The Chiangrai Hospital Ethical Committee, Thai Ministry of Public Health and the Faculty of Tropical Medicine Ethics Committee, Mahidol University, Bangkok, granted ethical approval for this study (MUTM 2006–035). All patients provided written informed consent prior to sample collection, and parents or guardians provided informed consent on behalf of all child participants. Chiangrai Prachanukroh hospital is located in Chiangrai province, the northernmost province in Thailand, and near “the Golden Triangle” where Thailand, Laos and Myanmar converge. The province population of 1.2 million consists mainly of ethnic Thais with 12.5% belonging to hill tribes and other minority ethnic groups.
Between August 2006 and October 2008, we prospectively recruited a total of n = 231 patients age ≥15 years old at Chiangrai Prachanukroh hospital with a fever >37.5°C or a history of fever within the past 21 days, no evidence of a primary focus of infection (e.g. consolidation on chest X-ray, symptoms and signs of a urinary tract infection, cellulitis) and negative for malaria on blood film. Demographic, clinical and laboratory data related to the admission were collected individually on study case-record forms (CRFs) from patient notes and hospital records. Demographic data included age, sex, and occupation. A rural/agricultural occupation was defined as those working as farmers, gardeners, agricultural/plantation workers, or fish and animal farm workers. Clinical data included symptoms, examination findings and vital signs on admission along with details of the current illness, prior antibiotic use, antibiotic treatment during admission, and illness outcome (e.g. fever days, death). Laboratory data included haematology (complete blood count) and biochemistry (renal and liver blood tests) results from admission samples. Chest x-ray findings were also recorded if performed.
An acute study blood sample was collected by study staff on enrolment in addition to the routine tests requested by the treating physician (10ml EDTA whole blood and 10ml clotted blood for serum). Blood and other routine cultures were performed if requested by the local clinician and processed using conventional techniques at the hospital microbiology laboratory. HIV testing was performed as part of routine hospital work using RDTs at the discretion of the treating physician. Follow-up was carried out by study staff 7–14 days after enrolment and involved a clinical review and collection of a convalescent blood sample (10ml clotted blood for serum).
There were 19 patients with incomplete CRFs/datasets and 12 patients with incomplete sample collections. These 31 patients were excluded resulting in a total of 200 study eligible patients. Of these, 171/200 (86%) provided paired samples obtained on admission and follow-up between days 7–14, and 29/200 patients had a confirmatory diagnosis made from admission samples alone. Both admission and follow-up samples were used for the diagnostic assays outlined below. Inflammatory biomarkers were tested on acute samples only. The clotted blood samples were processed for serum, aliquoted, stored locally at -30°C, and batch transported on dry-ice for storage (-80°C) and subsequent analysis at the central laboratory of Mahidol-Oxford Tropical Medicine Research Unit (MORU) in Bangkok. EDTA whole blood samples were transported at ambient temperature on the day of collection to Bangkok for further analysis. Some whole blood samples were processed immediately for culture for leptospirosis and scrub typhus (see below) with the remainder stored as aliquots of whole blood, plasma and buffy coat at -80°C.
In addition, meteorological data comprising average monthly temperatures and total monthly rainfalls were retrospectively collected for the study period from the local Thai Meteorological Department office of Mueang district in Chiangrai province. The data was collected from the district’s weather station near the airport. Chiangrai Prachanukroh Hospital is located within this central district, which is its main catchment area, but the hospital also admits severely ill patients from surrounding districts.
The diagnostic panel included diagnosis of dengue, scrub typhus, murine typhus, leptospirosis and Japanese encephalitis. Dengue diagnosis was performed in paired sera using the following ELISA tests: PanBio Dengue Early NS1 (Alere), PanBio Dengue IgM capture (Alere), PanBio Dengue IgG capture (Alere), and PanBio Japanese Encephalitis/Dengue IgM combo (Alere). An admission titer ≥10 U of NS1 PanBio units and/or ≥4-fold increase of IgM antibodies in the convalescent sample was considered diagnostic of acute primary dengue virus infection. Patients with anti-JEV IgM levels of >40 U were classified as having acute JEV infections only if anti-dengue IgM levels were <40 U using the combination ELISA test. Leptospirosis culture was performed at MORU within 24–48 hours by injecting 100μL of whole blood and 200μL of plasma sediment (the bottom fraction obtained from centrifuging 500μL of heparinized plasma) into 3 mL of Ellinghausen, McCullough, Johnson, and Harris (EMJH) medium, supplemented with 3% rabbit serum and 0.1% agarose. Both culture tubes were incubated aerobically at 25°C–30°C and examined every week for 3 months for evidence of growth. The leptospirosis SD Bioline RDTs were used for detecting anti-leptospira IgM and IgG. Scrub typhus and murine typhus were diagnosed using the indirect immunofluorescence assay (IFA) to detect IgM antibody titers in paired sera (or in admission samples only if convalescent samples unavailable) against Orientia tsutsugamushi antigens (Karp, Kato and Gilliam strains for scrub typhus) and Rickettsia typhi antigens (Wilmington strain for murine typhus), respectively. The new diagnostic IFA cut-off titer of ≥1:3,200 in an admission sample or ≥4-fold rise to ≥1:3,200 in a convalescent-phase sample was used [20]. For scrub typhus, culture and polymerase chain reaction (PCR) assays were also performed as previously described [21]. Briefly, the PCR assays included conventional PCR assay to detect the 56kDa gene and real-time PCR assays to detect the 47kDa htra and groEL genes. To fulfil the PCR criteria for diagnosis, a consensus of two out of three PCR assays was required.
The inflammatory biomarker procalcitonin was measured by the ELISA-based VIDAS PCT kit with a detection range of 0.05-195ng/ml (BioMérieux, France), and CRP serum levels were measured with the NycoCard Reader II (Axis Shield, Norway), with a detection range of 5-150mg/L in serum [22, 23]. Testing was performed on admission samples and two independent operators, blinded to the microbiological diagnoses, performed the procalcitonin and CRP assays in duplicate. Control reagents were provided with each test kit and calibration performed as per manufacturers’ instructions. The following thresholds were evaluated for their usefulness in predicting bacterial causes of fever; for procalcitonin 0.25ng/mL and 0.5ng/mL, and for CRP 20mg/L and 40mg/L plasma levels upon admission, respectively [24–26].
The diagnostic results were considered in relation to each other, and a final diagnosis was attributed to each case by the strength of evidence supporting each diagnosis, as previously described; (I) PCR/antigen/culture positivity > (II) dynamic serology (4-fold rise) > (III) single titer and/or unjustified serological cut-off titer [27].
Blood, urine, sputum and stool culture results from admission were collected from the hospital reporting system if performed. A final conservative diagnosis of culture-attributed infection (CAI) was made on the balance of clinical information, haematological and biochemical results, and results of our diagnostic panel.
Proportions, percentages and averages (median and interquartile range [IQR] or mean and standard deviation [SD]) were calculated controlling for any missing data. Seasonality was assessed by calculating proportions of patients (and 95% confidence intervals) admitted during discrete time-periods and assessing for overlap as well as performing two-sample tests of proportions. Univariate and multivariate logistic regression analysis were performed to determine predictor variables independently associated with the outcomes (e.g. viral/bacterial/unknown aetiologies or specific diagnoses such as scrub typhus or dengue). Categorical data were analysed using Pearson’s Chi-squared test or Fisher’s exact test as appropriate where specified. Comparisons of receiver operating characteristic (ROC) curves evaluated the sensitivity, specificity and likelihood ratios for procalcitonin and CRP in differentiating between bacterial and viral aetiologies. Classification and regression trees were generated for scrub typhus and dengue using Salford Predictive Modeler Software Suite v8.2 (Salford Systems, San Diego, CA, USA). Other analyses were performed using STATA 14 software (College Station, Texas, USA).
Our study cohort of 200 adult patients with AUF admitted to Chiangrai Prachanukroh hospital between August 2006 and October 2008 was predominantly male (114/194, 58.8%), had a median age of 41 (IQR 29–52), and most had a rural/agricultural occupation (64/136, 47.1%). 34/200 patients (17%) received antibiotic therapy prior to admission to the provincial hospital and the median number of days from onset of fever to admission was 4 (IQR 3–7).
77/200 patients (38.5%) had a bacterial aetiology for their fever, 24/200 (12%) a viral aetiology, and 97/200 (48.5%) had an unknown aetiology (the 2 remaining patients were diagnosed with invasive fungal infection, details below). Scrub typhus was the leading bacterial cause of AUF with 45/200 (22.5%), followed by leptospirosis with 15/200 (7.5%) and murine typhus 7/200 (3.5%), while dengue was the leading viral cause with 23/200 (11.5%) and there was a solitary JEV patient (0.5%).
A total of 12/200 (6%) cases had multiple positive tests (11 dual, 1 triple) that required scrutinizing with the criteria described above. Anti-JEV IgM positive cases were superseded by scrub typhus PCR positivity +/- dynamic serology in three cases and dengue NS1 antigen +/- dynamic serology in four cases. One case had weakly positive scrub typhus PCR for a single target (2 out of 3 targets required to fulfil the diagnostic criteria) with negative serology and was superseded by dengue NS1 antigen and IgM positivity. Two leptospirosis RDT positive cases were overruled by scrub typhus PCR-positivity in one case and dynamic murine typhus serology in the other. One case with dynamic rise in anti-dengue IgM but negative NS1 antigen was assigned a diagnosis of scrub typhus on the basis of PCR-positivity and dynamic serology. Finally, one case with positive leptospirosis RDT and anti-dengue IgM dynamic serology with negative NS1 antigen was diagnosed with scrub typhus on the basis of positive PCR assays.
142/200 (71%) patients had blood cultures performed of which, 126 were reported as no growth, 9 had microbiologically non-significant growth (mainly Gram positive organisms e.g. coagulase-negative staphylococci, aerobic spore bearers), and 7 had microbiologically significant growth (3.5%). Blood culture findings included 2 Talaromyces marneffei, 1 Haemophilus influenzae, 1 Staphylococcus aureus, 1 Burkholderia pseudomallei, 1 Escherischia coli, and 1 Enterococcus faecium. The patients with talaromycosis and Haemophilus influenzae bacteraemia tested positive for HIV antibodies using in-house RDTs. In addition, there were 2 significant urine cultures (heavy growth of E.coli), 2 significant sputum cultures (Klebsiella pneumoniae in patients with severe respiratory syndromes), and 1 significant stool culture (Salmonella spp.). In summary, there were 12 additional diagnoses in the culture-attributed infections group (CAI), 10 due to bacteria and 2 due to fungi.
Table 1 summarises the characteristics of patients in the viral, bacterial and unknown aetiology groups. Patients who were younger (OR 0.966, 95%CI 0.937–0.996, p = 0.026), had lower CRP (OR 0.967, 95% CI 0.953–0.981, p = 0.000), lower white blood count (OR 0.713, 95%CI 0.615–0.828, p = 0.000), lower neutrophil count (OR 0.694, 95%CI 0.586–0.822, p = 0.000) or higher haemoglobin (OR 1.259, 95%CI 1.023–1.549, p = 0.029) were significantly more likely to be diagnosed with a viral aetiology on univariate logistic regression analyses. Only low CRP (aOR 0.972, 95%CI 0.957–0.987, p = 0.000) and low white blood count (aOR 0.573, 95%CI 0.331–0.992, p = 0.047) remained as significant predictors for viral infection on multivariate logistic regression analysis. Significant predictor variables for bacterial infection on univariate analyses included the presence of an eschar (OR 11.74., 95%CI 3.849–35.807, p = 0.000) and a higher lymphocyte count (OR 1.366, 95%CI 1.027–1.816, p = 0.032) but only the eschar remained a significant predictor on multivariate analysis (aOR 11.590, 95%CI 3.754–35.784, p = 0.000). The finding of an eschar within the bacterial aetiology group was almost exclusively seen in patients diagnosed with scrub typhus (21/22, 95.5%), the exception being one patient with Staphylococcus aureus bacteraemia (1/22, 4.5%). Significant predictor variables for the unknown aetiology group are shown in Table 1 but are clinically less useful. Details of univariate and multivariate analyses of the predictor variables in Table 1 can be found in S1 Table. When comparing the viral and bacterial aetiology groups directly (excluding unknown group), eschar, CRP, Hb, WBC, neutrophil count and lymphocyte count were significant variables on univariate analyses. A lower CRP (aOR0.969 95%CI 0.951–0.987, p = 0.001) was an important predictor for viral infection while presence of an eschar (completely absent in the viral group) and a higher CRP (aOR1.032 95%CI 1.014–1.052, p = 0.001) remained as significant predictor variables for bacterial infection on multivariate analysis.
For a breakdown of demographics, symptoms and signs, chest x-ray findings and detailed laboratory results for patients diagnosed with scrub typhus, dengue, leptospirosis and murine typhus, please refer to S2 Table. Significant predictors for scrub typhus on multivariate logistic regression analysis included the presence of an eschar (aOR 42.408, 95%CI 4.956–362.905, p = 0.001), a higher lymphocyte count (aOR 2.063, 95%CI 1.146–3.713, p = 0.016), and elevated aspartate aminotransferase (AST, aOR 1.014, 95%CI 1.004–1.023, p = 0.004) and alkaline phosphatase (ALP, aOR 1.004, 1.000–1.008, p = 0.036). For dengue, a lower CRP (aOR 0.956, 95%CI 0.927–0.986, p = 0.005) was the only consistently significant predictor variable on multivariate analysis. Elevated creatinine was significantly associated with leptospirosis on univariate analysis (OR 1.132, 95%CI 1.001–1.279, p = 0.048) but was not significant in multivariate analysis. Details of the analysis can be found in S3 Table. In addition, classification and regression trees (CART) were generated for scrub typhus (S1 Fig, panel A) and dengue (S1 Fig, panel B) which revealed a similar set of significant variables when compared with the multivariate logistic regression analyses above. The presence of an eschar, ALP>289IU/L and AST>88IU/L were used as decision nodes for scrub typhus while CRP≤37mg/L and WBC≤7.9x103/mm3 were used for dengue virus.
The majority of cases occurred during the months of June to November, coinciding with the rainy and early winter seasons. Proportions of patients (95% confidence intervals) admitted from June to November and from December to January were calculated for the study cohort, scrub typhus, dengue, leptospirosis, murine typhus, CAI and unknown groups: total 0.82 (0.77–0.88):0.18 (0.12–0.23) p<0.001, scrub typhus 0.91 (0.83–0.99):0.09 (0.01–0.17) p<0.001, dengue 0.96 (0.87–1.00):0.04 (0.00–0.13) p<0.001, leptospirosis 0.80 (0.60–1.00):0.20 (0.00–0.40) p = 0.001, murine typhus 0.57 (0.20–0.94):0.43 (0.06–0.80) p = 0.595, CAI 0.83 (0.62–1.00):0.17 (0.00–0.38) p = 0.04 and unknown 0.77 (0.68–0.86):0.23 (0.14–0.32) p<0.001. Apart from murine typhus, there were no overlaps of 95% confidence intervals. To illustrate further, scrub typhus and dengue cases were plotted against time along with average monthly temperatures and total monthly rainfall for the district (Fig 1).
A total of 9 deaths were recorded during this study. Three patients had a diagnosis of scrub typhus (3/45, 6.7%), two patients had bloodstream infections (Talaromyces marneffei–previously Penicillium marneffei–and Haemophilus influenzae, both HIV positive cases), while 4 patients had unknown aetiologies (4/97, 4.1%). The 9 patients consisted of 5 males and 4 females, with a median age of 44 (IQR 40–51), 7 worked in agriculture, none had received pre-admission antibiotics and all were treated with antibiotics upon admission with the exception of 1 patient who died soon after presentation. The median number of fever days prior to hospital admission was 4 (IQR 3–6) and the number of days admitted was 3 (IQR 2–4). The majority were febrile on or after admission (8/9, 88.9%) and had neurologic (7/8, 87.5%), respiratory (5/8, 62.5%), gastrointestinal (5/8, 62.5%) or severe disease (5/8, 62.5%). The median (IQR) CRP, PCT, WBC, neutrophil count values for this sub-group were 150 mg/L (149–150), 37.2 ng/mL (2.0–59.7), 10.2x103/mm3 (8.3–14.8) and 9.3x103/mm3 (6.9–12.2), respectively.
169 out of 200 patients (84.5%) received antibiotics during the study (pre-admission and/or during admission). Of the 31 patients who did not receive any antibiotics, 6 had a viral infection (exclusively dengue), 7 had a bacterial infection (5 scrub typhus, 1 leptospirosis, and 1 bacteraemia) and 18 had an unknown aetiology. For monotherapy, ceftriaxone was the most commonly used antibiotic (131/169, 77.5%) followed by doxycycline (118/169, 69.8%) and chloramphenicol (26/169, 15.4%). Use of combination antibiotic therapy was common and particularly applied to patients during their in-patient stay 105/168 (62.5%) compared to those who received antibiotics prior to admission 9/34 (26.5%). Ceftriaxone and doxycycline was the most commonly used combination with 79/169 (46.7%) patients receiving this therapy.
Eighteen of twenty four patients (75%) with a viral diagnosis received antibiotics while patients with a bacterial diagnosis and those with an unknown aetiology received antibiotics in 93.3% (70/75) and 92.9% (79/85) of cases, respectively. Among patients with a diagnosis of scrub or murine typhus, 82.4% (42/51) received anti-rickettsial antibiotics (mainly doxycycline or chloramphenicol), which meant 17.6% of patients (9/51) received antibiotics ineffective against both diseases. In contrast, 93.3% of patients (14/15) with leptospirosis received appropriate treatment (ceftriaxone +/- doxycycline). An overview of antibiotic use is shown below in Table 2.
Nevertheless, the strategy of combining a beta-lactam with doxycycline was often used, and 53%, 73% and 67% of patients with a rickettsiosis, leptospirosis and culture-attributed bacterial infection received an antimicrobial treatment regimen combining a third generation cephalosporin with a rickettsia-active drug, respectively. Ceftriaxone monotherapy was most commonly used for leptospirosis and bacterial causes, while doxycycline monotherapy was commonly used for the rickettsial/dengue subgroups.
CRP on admission was a significant predictor variable for the viral aetiology group (low CRP) when analysing the whole AUF cohort. When the unknown group was excluded, it remained an important predictor for the viral (low CRP) and bacterial (high CRP) groups. 92% and 86% of bacterial cases had CRP levels above the pre-defined cut-offs of >20mg/L and >40mg/L, respectively. For the viral aetiology group, 73% and 86% of cases had CRP levels below these cut-offs, respectively. The >20mg/L and >40mg/L CRP cut-offs correctly identified 87.2% and 86.2% of bacterial and viral cases, respectively. The CRP and PCT results are summarised in Table 3.
The optimal CRP plasma level cut-off to accurately distinguish between bacterial and viral causes for fever in this study was calculated to be >36mg/L [sensitivity 88.9% (95%CI 79.3–95.1) and specificity 86.4% (95%CI 65.1–97.1)], with 88.3% of cases correctly identified. If we compare the choice of CRP cut-offs according to available CRP assays of 20mg/L and 40mg/L, then using the 40mg/L cut-off would provide an improved balance between sensitivity and specificity, with a higher specificity than the lower cut-off of 20mg/L.
On excluding the unknown aetiology group, PCT was good at defining bacterial cases, but poor at selecting for viral aetiologies, which is reflected by the poor specificity values. The higher cut-off at 0.50ng/mL improved specificity from 40.9 to 63.6 when compared to 0.25ng/mL, and was accompanied with a moderate drop in sensitivity and a minor reduction in the proportion of correctly identified cases. If a higher cut-off of 0.7ng/mL for PCT was chosen, sensitivity will fall slightly while the specificity will increase, but with the same number of correctly identified cases [sensitivity 79.2% (68.0–87.8), specificity 68.2% (45.1–86.1), correctly identified cases 76.6%].
Receiver operating characteristic (ROC) curves were generated to visualise the performance of CRP (S2 Fig, panel A) and PCT (S2 Fig, panel B) in differentiating bacterial versus viral infections for specified cut-off values. The areas under the ROC curve were 0.91 (0.85–0.96, 95% CI) and 0.80 (0.72–0.88, 95% CI) for CRP and PCT, respectively.
In this study the cause of non-malarial AUF was determined in 51.5% of enrolled cases. Rickettsial illnesses (scrub typhus and murine typhus) continue to be leading causes of AUF in northern Thailand, and although awareness of these treatable illnesses is increasing at the hospital level–as reflected by the high proportion of cases correctly managed by local physicians—this is not the case at the community level where doxycycline is seldom used. It is notable that by deploying diagnostic tests for as few as five diseases and utilising conventional microbiological culture techniques in the local hospital microbiology laboratory, the causes of more than half of the AUF cases could be identified.
Scrub typhus was the leading cause of AUF followed by dengue, leptospirosis, murine typhus, and bloodstream infections (22.5%, 11.5%, 7.5%, 3.5% and 3.5%, respectively) in this study. The incidence of both scrub typhus and dengue exhibited pronounced seasonality and were more common in the rainy season through to early winter (June to November). Similar to previous studies, the clinical finding of an eschar was strongly associated with the diagnosis of scrub typhus and represents a useful diagnostic clue [7, 28, 29]. However, eschars are not always present in scrub typhus patients, and their formation can be influenced by the degree of past exposure to various Orientia tsutsugamushi strains and the presence of strain-specific immunity [30]. Previous studies on paediatric scrub typhus in northern Thailand reported the presence of an eschar in approximately 70% of children [31, 32], while only 7% of children from Songkhla, southern Thailand and 7% of adults from Udon Thani, north-eastern Thailand with scrub typhus were reported to have an eschar [33, 34]. Whether this represents the spectrum of regular re-exposure to circulating Orientia tsutsugamushi strains in these regions remains to be determined in longitudinal studies.
Five patients presented with eschars but tested negative for scrub typhus. One patient had Staphylococcus aureus bacteraemia while the other four patients were in the unknown aetiology group. Additional testing of samples from one of these four patients revealed one 17kDa qPCR positive blood sample suggestive of Rickettsia spp. As such, alternative causes for febrile patients presenting with an eschar, such as spotted fever group rickettsial infections, should be considered. It is important to note that true eschars are completely painless–a central feature to distinguish them from eschar-like lesions including spider and (manipulated) insect bites which are typically itchy and/or painful [35].
In addition, we have shown that elevated hepatic enzymes (ALP and AST) were important predictors of scrub typhus in patients admitted with AUF on multivariate logistic regression and classification and regression tree (CART) analyses (ALP>289IU/L and AST>88IU/L). Raised hepatic enzymes have previously been described in scrub typhus observational studies in northern Thailand and India although not in cause of fever studies [36–38].
The overall mortality rate in our study cohort was 4.5% (total of 9 deaths) of which a third were attributable to scrub typhus. The scrub typhus mortality rate of 6.7% (3/45) was comparable to previous reports of untreated disease, as summarised in a recent systematic review, but was lower than the previously reported mortality from northern Thailand of 13.1% from 2004–2010, possibly reflecting better awareness and treatment decisions [36, 39].
The majority of patients (84.5%) received empirical antibiotic treatment after admission to the provincial hospital, and 82.4% of patients subsequently diagnosed with scrub or murine typhus received an anti-rickettsial regimen. Doxycycline and chloramphenicol were the two main anti-rickettsial antibiotics used during the study and the majority of scrub typhus patients in our study recovered, despite previous reports of doxycycline and chloramphenicol resistant strains of O. tsutsugamushi in Chiangrai [40]. Of the 3 patients who died with scrub typhus, one did not receive any effective antimicrobial, one had delayed administration of chloramphenicol, and another received doxycycline and chloramphenicol from admission onwards. Azithromycin has been shown to be an effective alternative treatment in scrub typhus patients and appears also to be effective against resistant Chiangrai isolates of O. tsutsugamushi [41–43]. However, azithromycin was not used during this study due to the unavailability of more cost-effective generic formulations at the time. Nevertheless, the fact that 53% and 73% of patients with rickettsioses or leptospirosis, respectively, received a combination of a third generation cephalosporin plus a rickettsia-active antibiotic, and that in the remaining patients ceftriaxone was most commonly used for leptospirosis or bacterial causes, while doxycycline was commonly used for the rickettsial/dengue subgroups, demonstrates a high level of clinical experience and awareness among medical staff in this endemic area (Table 2).
Roxithromycin was used in 1 patient with scrub typhus in combination with doxycycline. There have been limited clinical studies into the effectiveness of roxithromycin in the treatment of scrub typhus [31, 44] and none reported for murine typhus. One case series from Chiangrai reported low efficacy of roxithromycin when compared to doxycycline or chloramphenicol in 20 children with scrub typhus [31]. In vitro susceptibility testing to roxithromycin has not been reported for Orientia tsutsugamushi while Rickettsia typhi appears susceptible [45]. Two patients with scrub typhus received ciprofloxacin, one as the sole anti-rickettsial antibiotic and the other in combination with doxycycline and chloramphenicol. Fluoroquinolones have been shown to be moderately effective in in vitro susceptibility tests and in limited clinical studies against murine typhus [45–47]. However, Orientia tsutsugamushi may be intrinsically resistant to fluoroquinolones which may explain the poor efficacy reported in clinical studies [48–51].
In contrast to antibiotic use in the hospital setting, only 34/200 (17%) of study patients received antibiotics prior to admission. Rickettsial infections make up 25% of patients presenting with AUF, and only 10 of 34 patients received pre-admission antibiotics with anti-rickettsial activity (5% of the study cohort)–of these only 5 patients received effective treatment for scrub typhus (2.5% of the study cohort). Supporting this observation prescription data from primary care units from the central Mueang district of Chiangrai province (2015) revealed low utilisation of anti-rickettsial antibiotics and doxycycline use was absent altogether. This highlights the need for improving the availability of specific antibiotics—particularly doxycycline—in rural endemic areas and for providing effective diagnostics to guide appropriate management of febrile patients, as inappropriate use of antibiotics has led to the development of antibiotic resistance, particularly impacting regions where access to effective antimicrobials is already limited [52].
The paucity of diagnostically useful clinical symptoms and signs in AUF cases should spur the development of affordable and effective rapid diagnostic tests (RDTs). Even at a provincial hospital in Thailand, it is unrealistic for costly and expertise-reliant tests such as IFAs, ELISAs and PCR assays to be performed routinely [53]. Previous studies have shown that robust and high-quality RDTs for common causes of AUF provide the best balance for diagnostic cost-effectiveness [15]. This though requires up-to-date and representative local epidemiological data–ideally based on fever surveillance studies. The provision of effective RDTs to diagnose scrub typhus, dengue and leptospirosis will cover 41% of cases of AUF presenting to the provincial hospital in Chiangrai. The use of algorithms incorporating both clinical findings with accurate RDTs +/- basic laboratory tests to guide early appropriate antibiotic management of patients presenting with AUF will likely improve this further.
In high income countries, biomarkers such as C-reactive protein (CRP) and procalcitonin, have been shown to be safe, cost-effective, and improve correct antibiotic use in the management of respiratory tract infections in primary care settings [17, 25, 54]. In Southeast Asia, it has been demonstrated that CRP can discriminate between bacterial and viral infections in acutely febrile patients and reduce antibiotic use in patients with non-severe respiratory tract infections in the community [18, 19]. Modelling the impact and cost-effectiveness of pathogen-specific RDTs and CRP tests using data from febrile outpatients in Laos revealed that tests for common prevalent bacterial infections (scrub typhus in that setting) and CRP levels were likely to be cost-effective for direct health benefits while tests for viral pathogens (e.g. dengue) were not [15].
This study demonstrated that low CRP and low WBC were significant predictors of a viral infection (mainly dengue, CRP≤37mg/L and WBC≤7.9x103/mm3 on CART analysis). CRP was highly sensitive and very specific for defining bacterial infections (AUROC curve 0.9059), when directly comparing bacterial and viral groups, consistent with data from previous fever studies from Southeast Asia [18]. Currently, two CRP cut-offs are under investigation– 20mg/L and 40mg/L. The results in our study suggest that from a statistical point-of-view, choosing the higher cut-off improves specificity by almost 14%, thus reducing false positivity. However, this needs to be put into clinical context as the reduction of incorrectly treated viral cases from 6/22 (27.3%) to 3/22(13.6%) is offset by “missing” 4/72 (5.6%) of potentially severe bacterial cases–thus, reducing 3 cases with inappropriate antibiotic treatment comes at a cost of not treating 4 cases that would require antibiotics. If the test is employed at a community/primary care level, where monitoring facilities are limited, it could be argued that incorrectly treating an additional 3/22 (4.8%) febrile patients with a viral aetiology may be acceptable if an additional 4/72 (5.6%) patients with a potentially severe bacterial aetiology can be treated appropriately.
When comparing viral and bacterial groups, high procalcitonin was sensitive for the detection of bacterial infections but low levels were poor at selecting viral infections leading to low specificity. In Laos, elevated WBC counts have been shown to be significantly associated with fevers of bacterial aetiologies [6]. Previous fever studies from Southeast Asia have not specifically reported any association between neutrophilia and bacterial infections [3, 4, 6, 8, 55], although multiple reports have associated neutrophilia, lymphopaenia and elevated neutrophil-to-lymphocyte ratios with bacteraemic medical emergencies in high-income settings [56–58]. Our results suggest that simple laboratory tests such as full blood count and CRP could be beneficial in differentiating between bacterial and viral infections in acutely febrile patients at the hospital level, while a CRP-based POCT test (at USD 0.5–2.0 per test) is likely to be cost-effective in community settings in rural Southeast Asia. As Thailand expands its community health care system to fulfil one of five core priorities in partnership with the World Health Organization—this information is relevant to the development and commissioning of diagnostics in the community/district hospitals [59].
Although we were able to assign diagnoses to 51.5% of the febrile patient cohort, a large number of patients with unknown aetiologies demonstrated elevated laboratory markers described above and median CRP levels comparable to the diagnosed bacterial group–suggesting that a large proportion of potentially antibiotic treatable diseases go undiagnosed. The study has important limitations: i) due to budget constraints all cultures were performed in the local microbiology laboratory at the discretion of the treating physician; ii) there was an imbalance in the diagnostic investigations performed due to costs and limited access to high quality tests which may have led to bias (i.e. diagnostics for scrub typhus included PCR, culture and serology, while for leptospirosis only RDTs and culture were performed); and iii) the sample size is relatively small and external validity is limited although some conclusions can be drawn when the results are taken in context of previously published dataset from other fever studies from the region.
In conclusion, this study has highlighted the importance of scrub typhus and dengue in the aetiology of AUF in Chiangrai province, northern Thailand. It has provided more evidence for including anti-rickettsial antibiotics into empirical hospital treatment guidelines and management strategies of AUF in the community. It contributes to the mounting evidence that good quality, accurate, pathogen-specific RDTs are urgently needed, which together with biomarker POCTs such as CRP, may aid healthcare workers in the correct use of antibiotics as part of the wider focus on antimicrobial stewardship. Finally, it emphasises the need for further prospective studies into the causes of AUF in the community along with evaluations of CRP POCTs in improving disease management algorithms, diagnostic accuracy, patient safety and reducing inappropriate antibiotic use in the tropics.
|
10.1371/journal.ppat.1000688 | Lymphangiogenesis and Lymphatic Remodeling Induced by Filarial Parasites: Implications for Pathogenesis | Even in the absence of an adaptive immune system in murine models, lymphatic dilatation and dysfunction occur in filarial infections, although severe irreversible lymphedema and elephantiasis appears to require an intact adaptive immune response in human infections. To address how filarial parasites and their antigens influence the lymphatics directly, human lymphatic endothelial cells were exposed to filarial antigens, live parasites, or infected patient serum. Live filarial parasites or filarial antigens induced both significant LEC proliferation and differentiation into tube-like structures in vitro. Moreover, serum from patently infected (microfilaria positive) patients and those with longstanding chronic lymphatic obstruction induced significantly increased LEC proliferation compared to sera from uninfected individuals. Differentiation of LEC into tube-like networks was found to be associated with significantly increased levels of matrix metalloproteases and inhibition of their TIMP inhibitors (Tissue inhibitors of matrix metalloproteases). Comparison of global gene expression induced by live parasites in LEC to parasite-unexposed LEC demonstrated that filarial parasites altered the expression of those genes involved in cellular organization and development as well as those associated with junction adherence pathways that in turn decreased trans-endothelial transport as assessed by FITC-Dextran. The data suggest that filarial parasites directly induce lymphangiogenesis and lymphatic differentiation and provide insight into the mechanisms underlying the pathology seen in lymphatic filariasis.
| The nematode parasites Brugia malayi and Wuchereria bancrofti are the major organisms responsible for lymphatic filariasis. Lymphatic filariasis is characterized by the dysfunction of the lymphatics that can lead to severe (and often) irreversible lymphedema and elephantiasis. Current advances in distinguishing blood vascular from lymphatic endothelial cells have allowed the direct study of the interaction between live filarial parasites and their lymphatic niche. In the quest towards understanding parasite-lymphatic endothelium interactions, we observed that the filarial antigens have a specific but differential stimulatory capacity towards the lymphatics and cause them to differentiate into tube-like vascular networks in vitro that resemble the formation of collateral lymphatics in vivo. This was a lymphatic-specific phenomenon, as the filarial parasites or antigen did not exhibit similar effects on the human umbilical vein endothelial cells. The differentiation of the lymphatic endothelial monolayers into vascular networks was not dependent on typical markers of lymphangiogenesis but rather involves the matrix metalloproteases and their inhibitors that suggest lymphatic matrix remodeling rather than rendering of the lymphatics hyper-permeable as has been postulated previously.
| Among the clinical manifestations of human infection with the filarial nematodes Wuchereria bancrofti and Brugia malayi, the most debilitating are those associated with lymphatic dysfunction and/or obstruction (e.g. lymphedema, elephantiasis). It is known that the establishment of the adult filarial parasite within the lymphatic vessels of the host triggers a cascade of events that leads to abnormalities in lymphatic integrity and function. In addition, the tissue tropism of these lymphatic-dwelling parasites suggest that the host provides cues to the parasite that provide developmental signals to the parasites [1]. Individuals infected with the lymph-dwelling filariae can develop tissue scarring and fibrosis within and around the lymphatic vessels resulting in permanent and characteristic pathology manifested clinically by irreversible lymphedema or elephantiasis [2],[3]. An extensive body of work supports the notion that some of the severe pathology is immune mediated [4],[5],[6],[7] but that lymphatic dilatation and alterations in lymphatic flow may be parasite-derived and unrelated to the adaptive immune response. Indeed, filaria-infected SCID mice have been shown to exhibit lymphatic dilatation [8],[9], a process that could be reversed by removing or killing the worms [10],[11]. Moreover, lymphoscintigraphy studies showed that patients with subclinical disease have considerable structural abnormalities and aberrant lymph flow [12],[13],[14], and ultrasonography has suggested that the anatomic location of the worm nests changes very little over time but that lymphangiectasia and dilatation are not restricted to the anatomic location of the adult worm nests [15],[16],[17]. In addition, in histologic studies of lymph nodes taken from patients with bancroftian filariasis, there were intact adult filarial worms with little or no associated inflammation [18],[19]. Together, these observations suggest a primary role of the parasites or parasite-encoded proteins in modulating lymphatic integrity and function. It has been speculated that, with chronic infection, it is the death of the adult worms (be it immune mediated or just intrinsic worm lifespan) and the release of parasite antigens (or Wolbachia) that induce proinflammatory responses that render a poorly functioning (and dilated) patent lymphatic completely obstructed [20].
Based on the available animal models, it has been assumed that the filarial parasites promote lymphangiogenesis, a developmental process that involves lymphatic endothelial proliferation followed by differentiation (tube formation). In vitro studies of parasite-endothelial interactions have been limited to those using human umbilical vein endothelial cells (HUVEC) [21] that differ quite extensively from lymphatic endothelial cells (LEC). In the present study, we have examined directly the influence filarial parasites have on LEC function and lymphangiogenesis. We demonstrate that filarial proteins do stimulate the LEC to undergo proliferation and differentiate into tube-like structures that involves matrix metalloproteases and tissue inhibitors of matrix metalloproteases. We further demonstrate that, contrary to data in murine models, the filarial parasites render lymphatics less (rather than hyper-) permeable and suggest possible new mechanisms underlying filarial-induced lymphedema.
Analysis by flow cytometry and quantitative reverse transcriptase real-time PCR (qRT-PCR) demonstrated that the LEC were pure and possessed all of the characteristics of lymphatic endothelium (Figure S1). To evaluate the effect of filarial antigens on LEC proliferation, we stimulated LEC with stage-specific filarial antigens and measured proliferation at 96 h. As shown in Figure 1A, antigen derived from adult Brugia malayi parasites (BmA) or from adult male B. malayi parasites (BmMAg) induced proliferation of LEC with stimulation indices ranging from 8 to 35 and did so optimally at a concentration of 10 µg/ml (optimization data not shown). In contrast, antigens from microfilariae (MfAg) failed to induce proliferation at levels comparable to adult antigens. The extent of proliferation induced by BmA or BmMAg was comparable to that induced by recombinant soluble human vascular endothelial growth factor (VEGF)-A at 25 ng/ml used as a positive inducer of proliferation.
To address the specificity of this observed parasite-induced proliferation, both LEC and HUVEC, as a model representative of blood vascular endothelial cells were stimulated with the optimal concentrations of BmA, BmMAg, and MfAg for 96 h (based on preliminary dose and time-point assays) and the extent of proliferation quantified. As shown in Figure 1B, adult filarial antigens failed to induce proliferation of HUVEC, in marked contrast to the proliferation induced by these antigens in LEC. These data suggest that the filaria-induced EC proliferation was more specific to LEC compared to the HUVECs.
To address further the specificity of this response, non-filarial helminth antigens were tested. Because schistosome egg antigen (SEA) had been shown previously to induce proliferation of the HUVEC [22],[23], we examined the response of crude soluble extracts of Schistosoma mansoni adults (SWAP) and SEA on LEC and HUVEC proliferation. As seen, SEA induced proliferation of HUVEC but not LEC (Figure 1C), while SWAP failed to induce proliferation of either LEC or HUVEC (Figure 1D). Together, our data suggest that adult filarial antigens specifically induce LEC proliferation.
Having observed that filarial antigens induce LEC proliferation, we evaluated the potential of plasma (containing circulating filarial antigens and host angiogenic factors) to stimulate LEC proliferation. Using pooled plasma from filaria-uninfected endemic normal (EN) individuals, microfilaria-positive (MF) individuals with asymptomatic (or subclinical) infection, individuals with chronic lymphatic obstruction (CP), or AB serum from uninfected individuals (Figure 2), we were able to demonstrate that plasma from patients with lymphatic filariasis promoted LEC proliferation in contrast to that seen for uninfected EN plasma or control (AB) serum. As can be seen, the MF and CP plasma exhibited a significant increased proliferation compared with the EN plasma (p = 0.0028 and p = 0.028, respectively) or the AB serum (p = 0.0048 and p = 0.04, respectively). There was no difference in the proliferative capacity between the EN plasma or the AB serum. These data implicate filarial antigens and or soluble angiogenic factors circulating in the serum/plasma as having the capacity to induce LEC proliferation.
Because patent (microfilaraemic) filarial infection has been associated with lymphatic dilatation in both animal models and in human infection, we examined the effect of microfilarial antigen and live microfilariae (Mf) on LEC differentiation. When cultured in the presence of microfilarial antigens, the LEC were found to differentiate and form lymphatic tube-like structures when compared with unstimulated controls. Among the antigen preparations tested, MfAg was the most potent inducer of LEC differentiation, a process that could be observed as early as 48 h; in contrast, adult antigens could induce similar tube-like structures but did so only after 96 h (Figure 3, A–D). More importantly, live parasites (microfilariae, adult males, adult females) were also capable of inducing LEC differentiation. Live microfilariae (Figure 3F), compared with media controls (Figure 3E), induced LEC to form early tube like structures by 48–72 h in culture, whereas the adult male or female parasites induced tube formation only after 96 h in culture and did so in a limited and less complete fashion (Figure 3H and 3J). Increasing the number of live adult parasites did not appear to augment LEC differentiation (data not shown). To evaluate whether tube formation was contact dependent or independent, LEC were cultured with the parasites in transwell inserts. Interestingly, microfilariae co-cultured with LEC in transwell inserts also induced tube formation (Figure 3G), while the adult parasites in transwells did not (Figure 3I and 3K). These data suggest that this contact-independent LEC differentiation is primarily mediated by the excretory-secretory products of the parasites and of microfilariae most prominently.
Formation of lymphatic capillaries in vivo (in animal models) has been shown to be regulated by the expression of Prox-1 and podoplanin, among other molecules [24]. To evaluate whether the capillary tube formation induced by microfilariae involved upregulation of these and other molecules known to be associated with lymphangiogenesis, we examined the expression of CEACAM-1, LYVE-1, podoplanin, PROX-1, VEGF-C, and VEGFR-3 by qRT-PCR (Figure 4). As can be seen, MfAg did not alter expression of any of these molecules, suggesting that microfilariae-induced LEC differentiation is not related to expression of these particular markers.
To assess more globally the molecules involved in LEC differentiation induced by filarial pathogens, we carried out microarray studies comparing unstimulated LEC and MfAg-stimulated LEC. Notably in these analyses (Tables 1 and 2), expression of a number of genes associated with lymphangiogenesis (e.g. CADM1, TBK-1, GJA4, CD36, ANGPTL4, ICAM-2, MMP-1, CASK, TGFBI, VIP, CTGF, OLFML3, SORBS2, GJA1, MARCKS) were found to be altered by MfAg compared with the unstimulated control; many of these (based on pathway analysis) were associated with network functions of cellular development, cellular assembly and organization, hematologic system development and function, and cancer/tumor morphology. Furthermore, the results from these microarrays indicated that the expression of the molecules involved in maintenance of endothelial cell-cell contact (tight junctions and adherence junctions) are altered upon stimulation with filarial antigens (Figure 5). In addition, when qRT-PCR was performed independently on the majority of the genes shown to have filarial-induced alterations in gene expression, there was a strong relationship between the expression levels assessed by microarray and those assessed independently by qRT-PCR (Figure S2).
With the microfilaria-induced increases in gene expression associated with pathways involved in cell-cell contact, tissue remodeling, and angiogenesis, we next assessed the levels of the proteins (gene products) associated with these processes. Following stimulation of LEC with MfAg, it could be shown clearly that TIMP-1 and TIMP-2 levels were significantly inhibited (up to 75%) (Figure 6A and 6B) compared with either BmA (TIMP-1: P = 0.01; TIMP-2: P = 0.02) or BmMAg (TIMP-1: P = 0.03; TIMP-2: P = 0.02) -exposed LEC. There was no significant up- or downregulation of angiopoietin-2 (Ang-2), fibroblast growth factor (FGF), platelet-derived growth factor (PDGF), vascular endothelial growth factors - VEGF-A, VEGF-C, or VEGF-D by the parasite antigens. Other growth factors such as heparin-binding epidermal growth factor (HB-EGF), hepatocyte growth factor (HGF), keratinocyte growth factor (KGF), and thrombopoietin (TPO) were barely (or not) detectable (data not shown).
As MMP are regulated by the TIMPs, we next analyzed the parasite antigen-exposed LEC for the presence of active MMPs (MMP-1, -2, -3, -7, -8, -9, -10, and 13). Of all the MMPs assayed, only MMP-1 and MMP-2 were consistently altered by filarial antigen stimulation, with MfAg preferentially inducing expression of MMP-1. As can be seen, MfAg increased expression of MMP-1 significantly more compared with adult antigens BmA (p = 0.001) and BmMAg (p = 0.001) (Figure 6C). The adult antigens induced higher levels of MMP-2 by the LEC compared with MfAg (p = 0.05) (Figure 6D). These results suggest that MfAg-induced LEC differentiation may be associated with induction of MMP-1 and inhibition of TIMP-1 and TIMP-2.
Lymphedema in filaria-infected individuals is assumed to result from the lymphatic endothelium being rendered hyperpermeable, but the results from the microarrays (Tables 1 and 2 and Figure 5) suggest an increase in the cell-cell junction genes coding for the proteins that would actually limit trans-endothelial transport. Nevertheless, we evaluated the antigen-induced permeability of LEC in vitro. As shown (Figure 7), monolayers of LEC exposed to BmA or MfAg for 16–20h had permeability measurements similar to those of unstimulated controls, whereas known inducers of permeability (e.g., TNF-α, IL-1α) were capable of inducing significant trans-endothelial permeability. These data suggest that the changes induced by filarial antigens do not increase trans-endothelial passage of small molecules. In addition, BmA or MfAg were able to impede the permeability effect of TNF-α and IL-1α.
The pathologic hallmark of lymphatic filariasis is lymphedema and elephantiasis that has been associated with lymphatic dilatation, tortuosity, and obstruction. It is known, however, that individuals carrying adult worms (with or without microfilaremia) may have lymphangiectasia, acute lymphangitis, and lymphedema of the extremities [17]. Further, it has been postulated that with sufficient infection chronicity, individuals who harbor adult Wuchereria bancrofti will develop lymphangiectasia in the vicinity of the worm nests [20]. Although it is not clear what induces the lymphatic dysfunction in filaria-infected individuals, previous studies have shown that filarial antigens do not induce proliferation of HUVEC in vitro and, in fact, inhibited their proliferation [21]. More recent findings, implicate the role of VEGF family members in filarial pathology [25],[26],[27],[28]. The availability of markers (Prox-1, podoplanin, LYVE-1, VEGFR-3, among others) to differentiate LEC from blood vascular endothelial cells has facilitated the study of LEC directly. Although LEC have a lower turnover rate compared with blood vascular endothelial cells, on appropriate stimulation in vivo, LEC are capable of proliferating and migrating to organize into new lymphatic vessels (lymphangiogenesis) [29]. That this process occurs in response to filarial infection/antigens is demonstrated by our findings that filarial antigens specifically induce LEC proliferation and/or differentiation and do so in an antigen-specific manner. Compared to the LECs, under identical serum deprived conditions, VEGF-A induced proliferation of the HUVEC was similar to that of filarial antigens. It is not clear if the observed lower proliferative response of the HUVECs (compared to the LEC) to both the filarial antigens and VEGF-A reflects on their maximum proliferative capacity in vitro. Further, the failure of the filarial antigens to stimulate both the LECs and HUVECs in the presence of serum supplemented media compared to unstimulated controls probably reflects on various growth factors in the sera masking the proliferative capacity of the antigens, which could possibly account for similar observation with HUVECs previously [21]. Though VEGF-C (native or the mutant form of VEGFC-156Ser) specifically activates LEC, it was unable to induce significant proliferation of the LEC at the concentration similar to VEGF-A (25ng/ml). Our data show that, unlike the schistosome antigens (SEA or SWAP) that have no effect on LEC (present study) but do induce the proliferation of HUVECS (Figure 1C, 1D) [22], both adult filarial antigens and serum from patients with lymphatic filariasis induce LEC proliferation. As has been shown by others [25],[26],[27],[28], it is possible that circulating levels of pro-lymphangiogenic factors or circulating parasite antigens (>32,000 Og4C3 units in plasma of infected individuals, data not shown) themselves could be stimulating LECs in vivo. In addition, MfAg appears to induce lymphatic vessel differentiation and remodeling that may be aided by the alteration of the extracellular matrix.
The lymphangiogenic response requires both upregulation of lymphangiogenic factor expression and downregulation of the inhibitors of lymphangiogenesis [30] (typically involved in degradation and synthesis of the extracellular matrix components [31],[32]). Although VEGF family members have been considered the prime mediators of lymphangiogenesis, many other mediators have been identified that can also influence the lymphatic vasculature [33],[34],[35],[36],[37]. Among these HGF, bFGF, PDGF, angiopoeitins, and IGF-1 are known inducers and/or regulators of lymphangiogenesis; however, we were unable to detect any significant increase in the levels of HGF, b-FGF, ANG-2 or PDGF in the culture supernatants of LEC stimulated with MfAg. This however does not preclude the release of VEGF in in vivo conditions by the immune cells such as the macrophages that could aid in the filarial pathology. Thus, the host factors mediating lymphangiogenesis and lymphatic differentiation awaits clarification.
Typically, confluent monolayers of lymphatic endothelium in the presence of growth factors form tubular networks that are similar in their structural arrangement to lymphatic capillaries in vivo [38]. EM analysis of these tubular structures suggested that the flattened EC form a luminal space comprising one to several EC. Typical studies on lymphangiogenesis utilize collagen- or gelatin-coated culture plates or, for 3-dimensional studies, Matrigel™; however, filarial antigen-induced tube formation did not require any pre-existing matrix, suggesting that laminin and collagen type IV (major components of Matrigel™) are not absolutely essential for differentiation of LEC in vitro as has been demonstrated in previous in vitro studies [39],[40]. This is indicative of the capillary EC inherent capacity to produce and secrete factors that can aid in tissue remodeling (e.g. MMP) [41]. The activity of MMP is further controlled by endogenous TIMP that can bind to zymogens or inhibit the activated forms of MMP [42],[43],[44],[45]. TIMP-1 and TIMP-2 are the best-known inhibitors that can suppress capillary endothelial cell function in vitro. The filarial antigen-induced tube formation seen in the present study seemed most dependent on regulation of TIMP-1 and TIMP-2 that were, in turn, associated with increased levels of MMP-1 and MMP-2. Preliminary data suggest that addition of exogenous TIMP-1 and TIMP-2 inhibit the antigen-induced tube formation to a considerable extent (data not shown). It remains to be seen how TIMP and MMP regulate LEC activity upon stimulation with filarial antigens, although their role in development of collateral/accessory lymphatics in cats infected with Brugia pahangi [46] and in humans with lymphatic filariasis has been postulated previously.
A characteristic feature of long-term filarial infection in humans and animals is the fibrosis and cellular hyperplasia in and around the lymphatic walls. Infection with the parasites for long periods results in the fibrosis of the infected lymph nodes, which eventually become non-functional and are bypassed by new lymphatic vessels [46]. Several animal studies [47],[48],[49] have previously shown that EC from infected animals had a decreased number of vesicles (which presumably transport fluid) and an increase in the number of vacuoles (which presumably results from cell damage), suggesting that EC lining the lymphatics inhabited by the filarial worms are affected by the parasites. The injury probably renders these EC less effective in transporting edematous fluid and thereby contributes to the edema and collagen accumulation. Studies on LEC permeability (as an indirect measure of their transport capability) demonstrated that, upon stimulation with filarial antigens, LEC are not rendered hyperpermeable but rather are made less leaky, in large part by the induction of adhesion molecules involved in maintaining cell junctions. A recent study using Dirofilaria immitis antigen on EC also suggests that antigen-treated monolayers do not differ from the unstimulated controls in the trans-endothelial passage of FITC-Dextran [50]. These data suggest that the lymphedema associated with filarial infection is associated with an inability of the parasite-exposed lymphatics to resorb interstitial fluid rather than for fluid to pass through the LEC junctions abnormally. Alternatively, the proinflammatory responses (IL-1, TNF-α) seen in immune effector cells from patients with chronic lymphatic obstruction [51] could themselves render the LEC hyperpermeable. The proliferative effects of the adult antigens and more pronounced tube-formation activity of the microfilarial antigens, perhaps is an indication of the lymphatic dilatation/damage seen in the sub-clinical infections. It is not absolutely clear, but the consensus on the progression and development of lymphedema/elephantiasis lies in the death of the parasites that triggers a massive inflammatory response against the parasite products and its endosymbiont wolbachia. By whatever mechanism, the interstitial fluid remains and leads to fibrosis and tissue injury.
Thus, our studies reveal that filarial parasites are capable of inducing lymphangiogenesis in vitro, a process that is LEC specific and relates to the excretory-secretory components of the filarial parasites and/or the elevated levels of circulating lymphangiogenic factors [26]. The new vessels formed appear to be associated with intrinsic extracellular matrix modeling and are not related to expression of prototypical markers of lymphangiogenesis seen in tumor biology. Our data also suggest that filarial lymphedema may not be due to parasite-induced alterations in trans-endothelial transport (as has been hypothesized) but may involve host-expressed inflammatory components. With the elucidation of the stage-specific excretory-secretory components of B. malayi [52],[53], further studies on the effects of these ES products (and their purified components) on the LECs would enable a clearer understanding of the filarial induced lymphatic dysfunction associated with lymphatic filariasis.
Human dermal microvascular endothelial cells (HDMEC), human lymphatic microvascular endothelial cells (here called LEC), and human umbilical vein endothelial cells (HUVEC) were obtained from Cambrex. LEC and HUVEC (together referred to as EC) were cultured in EGM-2MV (Cambrex), serially subcultured, and maintained through 7–8 passages in T75 tissue culture flasks (Sarstedt; Fisher Scientific). All experiments were performed using cells from passages 3–5. The purity of LEC was verified by their cell surface expression of podoplanin, VEGFR-3, and LYVE-1 by flow cytometry and confocal microscopy.
For flow cytometry analyses, the cells were detached from the culture flasks by trypsinization per manufacturer's instructions (Reagent Pack, subculture reagents; Cambrex). The detached cells (for flow cytometry) or adherent cells in 8-well chambered glass slides (Lab-Tek; NUNC) (for confocal microscopy) were stained individually or in combination with mouse monoclonal anti-human LYVE-1 (MAB2089; R&D Systems), anti-human ICAM-1-FITC (BBA20; R&D Systems), anti-human VEGFR3-APC (FAB3492A; R&D Systems), mouse anti-human CD44-PE (BD Pharmingen), or mouse polyclonal anti-human podoplanin (Cell Sciences). Anti-mouse IgG coupled with Alexa Fluor 594 or Alexa Fluor 488 were from BD Pharmingen. Staining with isotype control Ab were performed in parallel. Flow cytometry data were analyzed by FlowJo (Treestar Inc.). Confocal microscopy was carried out using a Leica-SP2-AOBS unit at the NIAID Biologic Imaging Facility.
QRT-PCR was performed on total RNA samples using commercially available assays (TaqMan™; Applied Biosystems) for podoplanin, LYVE-1, Prox-1, VEGF-C, VEGFR-3, MRC-2, CEACAM, and CD44 on an ABI Prism 7900 sequence detection system (Applied Biosystems). All data are initially expressed as fold change from the endogenous control (18s rRNA). Details of the genes targeted for assessment of quantitative RT-PCR are listed in Table S1.
Soluble extracts of B. malayi adult parasites (BmA), adult male (BmMAg), and microfilariae (MfAg) were prepared as described previously [54], as were soluble extracts of Schistosoma mansoni (SWAP) and egg antigen (SEA) [22]. B. malayi antigens were used both because of their ability to be maintained in animal hosts and because of their extensive homology to the other lymphatic dwelling filariae, W.bancrofti that has as its only host, humans.
Live adult parasites and microfilariae were obtained from the NIH/NIAID-Filariasis Research Reagent Repository Center at the University of Georgia, Athens. The animal procedures were conducted in accordance with ACUC guidelines with the approval from the Institutional Review Boards at the National Institutes of Health and at the University of Georgia. The adult worms were washed thoroughly in RPMI-1640 supplemented with penicillin/streptomycin, 2 mM L-glutamine, and 10 g/l of d-glucose (MF-media). Microfilariae were washed and resuspended in MF-media, layered onto LSM (MP Biomedicals), and centrifuged at 1750 rpm for 20 min. The pellet of live microfilariae was cleared of all contaminating cells by incubation with lysis buffer for 10 min, washed in MF-media, and enumerated.
Primary EC were detached from culture flasks by trypsinization and seeded at a density of 5×103 cells/well in 96-well flat-bottomed plates in EGM-2MV media (Cambrex). The cells were starved overnight in endothelial basal media (EBM-2) before being stimulated with various concentrations of BmA, BmMAg, or MfAg. SWAP and SEA were used as nonspecific parasite stimulus. VEGF-A and VEGF-C were used as specific stimulators of EC. Untreated cells with media alone served as negative controls. EC were cultured at 37°C for 96 h in a 5% CO2 incubator in EBM-2 media with/without any growth factors. The cells were pulsed with 1 µCi of H3-thymidine 16 h prior to harvesting. The cells were harvested at 96 h and the amount of incorporated H3-thymidine measured by Wallac scintillation counter (Perkin Elmer).
Proliferation of LEC in the presence of 0.5% to 1% previously well characterized serum from endemic normals (EN), asymptomatic microfilaraemic (MF), and chronic pathology (CP), or control AB was assayed by incorporation of BrDU (Invitrogen) as per the manufacturer's instructions.
LEC were seeded at a density of 0.5×106 cells in 6-well plates (Costar; Fisher Scientific) in EGM-2MV media. The cells were starved overnight in EBM-2 media and were stimulated with antigen preparations (BmA, BmMAg, and MfAg) or live parasites. The live adult parasites (3–4 adults/well) or live microfilariae (50,000) were co-cultured with LEC directly or in transwells. Tube formation was observed and imaged using a Zeiss confocal microscope under a PlastDIC lens every 24 h.
The culture supernatants from LEC stimulated with filarial antigens or live parasites after 24, 48, 72, and 96 h were screened for Ang-2, FGF-b, HB-EGF, HGF, KGF, PDGF-BB, TIMP-1, TIMP-2, VEGF, VEGF-C, VEGF-D, TPO, or MMP-1, -2, -3, -7, -8, -9, -10, and -13 by multiplex arrays (Searchlight). The intra- and inter-assay coefficients of variances for the multiplexed array were observed to be less than 10% and 25% respectively.
Permeability assays on LEC were carried out as the manufacturer's instructions using the in vitro vascular permeability assay kit (Chemicon Intl.). Briefly, LEC were seeded at a density of 2×105 cells on collagen-coated transwell inserts in a 24-well tissue culture plate. The cells were cultured for 48 h in EGM-2MV medium for confluent monolayers. The cells in the monolayers were starved overnight with EBM-2 basal medium, the media was replaced, and cells were stimulated with antigens for 16–20 h. Permeability was measured based on the amount of FITC-dextran that diffused from the transwell into the lower compartment.
Equal quantities of total RNA from LEC and LEC exposed to MfAg or BmA were converted to aminoallyl-labeled cDNA using the Ovation aminoallyl amplification and labeling system per the manufacturer's instructions (NuGEN Technologies, Inc.). The aminoallyl-labeled cDNA were labeled using Amersham CyDye (Cy3, unstimulated controls; Cy5, stimulated) reactive post-dye protocol (GE Healthcare) and hybridized onto custom microarrays manufactured by the Microarray Research Facility of the NIAID. The human microarray (Hsbb) probe set is based on 70mer oligonucleotides from Qiagen's Human Genome Oligo Set V2.0. A comprehensive gene list is available at http://madb.niaid.nih.gov/cgi-bin/gipo.
Images of the scanned arrays were analyzed, and data were filtered using the criteria of fold-change of ≥2 (upregulation) or ≤2 (downregulation) of the genes compared with unstimulated controls. For categorizing and pathway analysis, ingenuity pathway analysis (IPA, www.ingenuity.com) was used. Selected genes involved in cell-cell junctions were analyzed by qRT-PCR. The details of the primer/probe sets used are listed in Table S1.
The microarray data are available at the NCBI GEOS website http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE15454.
Data analyses were performed using GraphPad PRISM (GraphPad Software, Inc., San Diego, CA). Unless otherwise noted, geometric means (GM) were used for measurements of central tendency. Statistically significant differences between two groups were analyzed using the non-parametric Mann-Whitney U test. Correlations were calculated by Spearman Rank correlation test.
|
10.1371/journal.pgen.1003087 | Phenome-Wide Association Study (PheWAS) for Detection of Pleiotropy within the Population Architecture using Genomics and Epidemiology (PAGE) Network | Using a phenome-wide association study (PheWAS) approach, we comprehensively tested genetic variants for association with phenotypes available for 70,061 study participants in the Population Architecture using Genomics and Epidemiology (PAGE) network. Our aim was to better characterize the genetic architecture of complex traits and identify novel pleiotropic relationships. This PheWAS drew on five population-based studies representing four major racial/ethnic groups (European Americans (EA), African Americans (AA), Hispanics/Mexican-Americans, and Asian/Pacific Islanders) in PAGE, each site with measurements for multiple traits, associated laboratory measures, and intermediate biomarkers. A total of 83 single nucleotide polymorphisms (SNPs) identified by genome-wide association studies (GWAS) were genotyped across two or more PAGE study sites. Comprehensive tests of association, stratified by race/ethnicity, were performed, encompassing 4,706 phenotypes mapped to 105 phenotype-classes, and association results were compared across study sites. A total of 111 PheWAS results had significant associations for two or more PAGE study sites with consistent direction of effect with a significance threshold of p<0.01 for the same racial/ethnic group, SNP, and phenotype-class. Among results identified for SNPs previously associated with phenotypes such as lipid traits, type 2 diabetes, and body mass index, 52 replicated previously published genotype–phenotype associations, 26 represented phenotypes closely related to previously known genotype–phenotype associations, and 33 represented potentially novel genotype–phenotype associations with pleiotropic effects. The majority of the potentially novel results were for single PheWAS phenotype-classes, for example, for CDKN2A/B rs1333049 (previously associated with type 2 diabetes in EA) a PheWAS association was identified for hemoglobin levels in AA. Of note, however, GALNT2 rs2144300 (previously associated with high-density lipoprotein cholesterol levels in EA) had multiple potentially novel PheWAS associations, with hypertension related phenotypes in AA and with serum calcium levels and coronary artery disease phenotypes in EA. PheWAS identifies associations for hypothesis generation and exploration of the genetic architecture of complex traits.
| In phenome-wide association studies (PheWAS) all potential genetic variants in a dataset are systematically tested for association with all available phenotypes and traits that have been measured in study participants. By investigating the relationship between genetic variation and a diversity of phenotypes, there is the potential for uncovering novel relationships between single nucleotide polymorphisms (SNPs), phenotypes, and networks of interrelated phenotypes. PheWAS also can expose pleiotropy, provide novel mechanistic insights, and foster hypothesis generation. This approach is complementary to genome-wide association studies (GWAS) that test the association between hundreds of thousands, to over a million, single nucleotide polymorphisms and a single phenotype or limited phenotypic domain. The Population Architecture using Genomics and Epidemiology (PAGE) network has measures for a wide array of phenotypes and traits, including prevalent and incident status for clinical conditions and risk factors, as well as clinical parameters and intermediate biomarkers. We performed tests of association between a series of genome-wide association study (GWAS)–identified SNPs and a comprehensive range of phenotypes from the PAGE network in a high-throughput manner. We replicated a number of previously reported associations, validating the PheWAS approach. We also identified novel genotype–phenotype associations possibly representing pleiotropic effects.
| Phenomic approaches are complementary to the more prevalent paradigm of genome-wide association studies (GWAS), which have provided some information about the contribution of genetic variation to a wide range of diseases and phenotypes [1]. While a typical GWAS evaluates the association between the variation of hundreds of thousands, to over a million, genotyped single nucleotide polymorphisms (SNPs) and one or a few phenotypes, a common limitation of GWAS is the focus on a pre-defined and limited phenotypic domain. An alternate approach is that of PheWAS, which utilizes all available phenotypic information and all genetic variants in the estimation of associations between genotype and phenotype [1]. By investigating the association between SNPs and a diverse range of phenotypes, a broader picture of the relationship between genetic variation and networks of phenotypes is possible.
A challenge for PheWAS is the availability of large studies with genotypic data that are also linked to a wide array of high quality phenotypic measurements and traits for study. Biorepositories linked to electronic medical records (EMR) have been an initial resource for PheWAS, but these EMR-based studies are often limited to phenotypes and traits commonly collected for clinical use and may represent sets of limited racial/ethnic diversity [2], [3]. While there is no U.S. national, population-based cohort [4], several diverse, population-based studies exist with tens of thousands of samples linked to detailed survey, laboratory, and medical data. These large population-based studies have limitations [5], but collectively [6] they offer an opportunity to perform a PheWAS of unprecedented size and diversity.
To capitalize on the potential for collaborative discovery among some of the large population-based studies of the U.S., the National Human Genome Research Institute (NHGRI) funded the Population Architecture using Genomics and Epidemiology (PAGE) network. PAGE includes eight extensively characterized, large population-based epidemiologic studies where data were collected across multiple racial/ethnic groups, supported by a coordinating center [7], providing an exceptional opportunity to pursue PheWAS with a large number of SNPs, and thousands of phenotypic measurements including a wide range of common diseases, risk factors, intermediate biomarkers and quantitative traits in diverse populations. Herein, we illustrate the feasibility and utility of the PheWAS approach for large population-based studies and demonstrate that PheWAS provides information on, and exposes the complexity of, the relationship between genetic variation and interrelated and independent phenotypes. We have found PheWAS results that replicate previously identified genotype-phenotype associations with the exact phenotype in previous associations or closely related phenotypes, as well as a series of novel genotype-phenotype associations. This data exploration method exposes a more complete picture of the relationship between genetic variation and phenotypic outcome. PheWAS provides the unbiased, high throughput design achieved by GWAS in the genome and phenotype domains simultaneously. This approach changes the paradigm of phenotypic characterization and allows for exploratory research in both genomics and phenomics.
Data from five PAGE study sites were available for this PheWAS: Epidemiologic Architecture for Genes Linked to Environment (EAGLE) using data from the National Health and Nutrition Examination Surveys (NHANES); the Multiethnic Cohort Study (MEC); the Women's Health Initiative (WHI); and two studies of the Causal Variants Across the Life Course (CALiCo) group: the Cardiovascular Health Study (CHS) and Atherosclerosis Risk in Communities (ARIC). Text S1 provides full information on study design, phenotype measurement, and genotyping for each study. These studies collectively include four major racial/ethnic groups: European Americans (EA), African Americans (AA), Hispanics/Mexican Americans (H), and Asian/Pacific Islanders (API). All PAGE study sites included both males and females, except for WHI (which includes only women). Table 1 provides an overview of the sample sizes by PAGE study site as well as the number of SNPs and phenotypes available for this PheWAS. Sample size and the number of phenotypes varied across studies, and the sample size for various phenotypes within each study varied dependent on the number of individuals for which a given phenotype was measured. The number of phenotypes available for this PheWAS ranged within studies from 63 (MEC) to 3,363 (WHI). Study sites also had differing numbers of genotyped SNPs, and Table S1 contains the list of all SNPs available for two or more sites in this study, arranged by previously associated phenotypes. The PAGE network has focused on characterization of well-replicated variants across multiple race/ethnicities, so each study independently genotyped a set of SNPs with previously reported associations with phenotypes such as body mass index, C-reactive protein, and lipid levels.
Tests of association assuming an additive genetic model were performed independently by each PAGE study site for each SNP and each phenotype, stratified by race/ethnicity. The last column of Table 1 presents the total number of comprehensive associations with and without a p-value cutoff of 0.01, showing the proportion of significant results for this many tests of association. The total number of tests of association ranged from >20,000 (MEC) to >1 million (WHI) reflecting the variability in both the number of phenotypes available for study as well as the number of SNPs genotyped by each PAGE study site. As expected, the total number of significant tests of association (p<0.01) represented a fraction of the total number of tests performed.
Results from these tests of association were then compared across study sites to identify overlapping significant associations, as these results most likely represent robust findings. To facilitate determining overlapping significant associations, similar phenotypes that existed across more than one study were binned into 105 distinct phenotype-classes. For some phenotypes, the specific phenotype existed across more than one PAGE study, such as for the phenotype “Hemoglobin”, where hemoglobin measurements were available for ARIC, CHS, EAGLE, and WHI. Other groups of phenotypes binned within phenotype-classes were within similar phenotypic domains but were not represented in exact same form across studies. Table S2 contains a list of the study level phenotypes, the study from which the phenotype is available, and the phenotype-class for each phenotype that overlapped with another study.
The same or similar phenotypes may or may not have been collected by each PAGE study. Thus, the number of studies that were available for comparison of results across studies varied from one phenotype-class to another phenotype-class. Table 2 presents the number of results where at least two of five independent studies had SNP-phenotype associations with p<0.01 for single phenotype-class and single race/ethnicity group, compared to the total number of SNP-phenotype association tests performed. For example, >8,500 tests of association for the same SNP and same phenotype were available from two PAGE study sites whereas only 906 and 58 tests of association were available from four and five PAGE study sites, respectively. There were 3 results where two or more of the groups had a SNP–phenotype association p<0.01 for a single phenotype class across 5 groups represented.
For this PAGE-wide PheWAS, tests of association were considered significant across PAGE study sites where two or more phenotypes in the same phenotype-class in the same racial/ethnic group passed a significance threshold of p<0.01 with a consistent direction of genetic effect. Based on these criteria, a total of 111 PheWAS associations were identified (Table S3). Overall, among the 111 significant PheWAS associations identified, 52 PheWAS results replicated previously published genotype-phenotype associations (Table S4), 26 represented phenotype-classes closely related to previously known genotype-phenotype associations (Table 3), and 33 represented novel genotype-phenotype associations (Table 4).
Almost half of the PAGE PheWAS results (52/111; 48%) replicated previously known genotype-phenotype associations. These replicated results serve as positive controls and demonstrate that the high-throughput PheWAS approach is feasible and valid. As an example, low-density lipoprotein cholesterol (LDL-C) has previously been associated with rs4420638 near APOE/APOC1/C1P1/C2/C4 in European Americans [8], [9]. In the PAGE PheWAS, a significant association between the same SNP and LDL-C phenotypes of the “LDL-C” phenotype-class in European Americans as reported in the literature [8], [9] was observed in two PAGE study sites, with the same direction of effect (β) as well as a third PAGE site with near significant results: ARIC (p = 1.27×10−15, β = −5.75), CHS (p = 7.89×10−12, β = −7.06), and WHI (p = 0.06, β = −4.15). Figure 1 shows the significant PheWAS LDL-C results, as well as other associations considered significant for rs4420638 across PAGE study sites for other phenotype-classes in a similar racial/ethnic group passed a significance threshold of p<0.01 with a consistent direction of genetic effect.
Approximately one-fourth of the PAGE PheWAS results (26/111; 23%) represented SNP-phenotype associations in phenotype-classes closely related to previously known genotype-phenotype associations. For example, rs10757278 near CDKN2A/CDKN2B has been robustly associated with myocardial infarction (MI) [10], [11]. In this PheWAS, rs10757278 was associated with the “Cardiac” phenotype-class, but also with the related phenotype-classes of “Artery Treatment” and “Angina”. Specifically, rs10757278 was associated with phenotypes in the Artery Treatment phenotype-class, such as “percutaneous transluminal coronary angioplasty” (WHI, p = 2.86×10−6, β = −0.17, EA), and “coronary bypass surgery” (CHS, p = 9.60×10−3, β = −0.26, EA). The SNP rs10757278 was also associated with phenotypes in the Angina phenotype-class, such as presence or absence of angina (WHI, p = 6.59×10−3, β = −0.14, EA) and the phenotype “Ever see a doctor because of chest pain?” (ARIC, p = 4.44×10−3, β = −0.31, EA). Replication of association of this SNP with previously known phenotypes were also found with the phenotype-class “Cardiac”, with phenotypes such as “MI (Y/N)” (WHI, p = 1.39×10−4, β = −0.11, EA), and “MI status at baseline (Y/N)” (CHS, p = 6.35×10−3, β = −0.18, EA). Significant PheWAS associations at p<0.01 for rs10757278 are plotted by phenotype in Figure 2, as well as additional results at p<0.05.
Another example of PheWAS associations for phenotype-classes closely related to known genotype-phenotype associations existed for rs599839 near the CELSR2/PSRC1/SORT1 gene cluster. The SNP rs599839 has been associated with serum LDL cholesterol levels [8], [12]–[14], and coronary artery disease [13], [15]. In our PheWAS, associations were found for the “LDL-C” phenotype-class, as well the coronary artery disease related “Angina” and lipid related “HDL-C” phenotype-classes, including specific phenotypes such as “angina, presence or absence of” (WHI, p = 2.10×10−4, β = 0.25, EA), and “HDL-C” (WHI, p = 1.23×10−3, β = −0.04, AA). As expected, a significant association was also identified for the LDL-C level related phenotype “LDL-C (mg/dl)” (ARIC, p = 5.25×10−22, β = 6.42 EA). Significant PheWAS associations at p<0.01 for rs599839 are plotted by phenotype in Figure 3, as well as additional results at p<0.05.
PheWAS results were considered novel, if the significant phenotype-class associations varied substantially from the previously reported GWAS and candidate gene studies. Approximately one-third of the PAGE PheWAS results (33/111; 30%) represented novel genotype-phenotype-class associations. Further research will be required to determine the further validity of these exploratory results.
The most statistically significant of the novel phenotype-class associations identified by this PheWAS include multiple associations involving phenotype-classes for hematologic traits in African Americans (Figure 4). SNPs rs599839 (CELSR2/PSRC1), rs10923931 (NOTCH2), rs2228145 (IL6R), rs2144300 (GALNT2), rs10757278 (CDKN2A,CDKN2B), and rs7901695 (TCF7L2) were each associated with white blood cell count phenotypes among AA (significant p-values ranging 7.96×10−3 to 9.99×10−15). IL6R rs2228145 was also associated with neutrophils and lymphocyte numbers in AA with p-values ranging from 2.44×10−4 to 4.66×10−10. These SNPs were previously associated with LDL-C, total cholesterol levels, and coronary artery disease (rs599839) [8], [12]–[14]; type 2 diabetes (rs10923931) [16]; C-reactive protein (rs2228145) [17]; coronary heart disease, HDL-C and triglycerides (rs2144300) [13]; MI (rs10757278) [11]; and type 2 diabetes (rs7901695) in EA [18]–[20]. It is likely that the majority of the significant findings for three of the SNPs on chromosome 1 [rs599839 (CELSR2/PSRC1), rs10923931 (NOTCH2), rs2228145 (IL6R)] are not truly novel given that these variants are likely in linkage disequilibrium with the white blood cell count-associated Duffy null allele (DARC rs2814778) [21], [22] in African Americans. Of note is GALNT2 rs2144300 (p = 3.32×10−6 in WHI and 7.96×10−3 in CHS), located outside the 90 Mb region known to be associated with white blood cell counts in African Americans [21] and possibly representing a novel genotype-phenotype association for this trait. Also for chromosome 1, novel associations were identified in African Americans at p<0.01 for the phenotype-class “Hemoglobin” and ANGPTL3 rs1748195, previously associated with triglycerides in European-descent populations [13], [19].
Of the remaining hematologic trait associations identified that were not on chromosome 1, rs10757278 near CDKN2A/B on chromosome 9 and TCF7L2 rs7901695 on chromosome 10 were both associated with white blood cell count, neither of which were previously reported in GWAS for this trait [21], [22]. For CDKN2A/B rs1333049, a SNP previously associated with type 2 diabetes, coronary artery disease, and hypertension in European-descent populations [15], [23] p<0.01 associations were identified for the phenotype-class of Hemoglobin. Finally, a novel association in European Americans was noted between FADS1 rs174547, a SNP previously associated with LDL-C [13], [19], and the phenotype-class of “Platelet Count” at p<0.01.
Aside from hematologic traits, the most significant novel association identified in this PheWAS was identified for phenotypes in the phenotype-class “Forced Expiratory Volume in 3 Seconds (FEV3)” and GALNT2 rs2144300 in African Americans (p-values ranging from 8.82×10−3 to 4.90×10−4). GALNT2 rs2144300, previously associated with HDL-C in European Americans and African Americans [13], [24], has not previously been associated with lung function or asthma quantitative traits. Interestingly, GALNT2 rs2144300 was also associated with phenotypes in the “Hypertension” phenotype-class among African Americans in this PheWAS Specifically the phenotypes were “High blood pressure ever diagnosed?” (ARIC, p = 1.61×10−3, β = 0.24) and “Pills for hypertension ever?” (WHI, 8.27×10−3, β = 0.15). Indeed, GALNT2 rs2144300 displayed the most suggestion of pleiotropy among all the SNPs tested in this study. In addition to the associations identified in African Americans, rs2144300 was associated with phenotypes in the phenotype-classes “Serum Calcium” (p-values ranging from 1.47×10−4 to 8.10×10−3) and “Artery Treatment”, specifically the phenotypes “Coronary artery bypass graft (CABG)” (WHI, p = 2.46×10−3, β = 0.24) and “Aortic aneurysm repair” (CHS, 5.49×10−3, β = 0.57) in European Americans. Significant PheWAS associations at p<0.01 for rs2144300 are plotted by phenotype in Figure 5, as well as additional results at p<0.05.
The remaining significant novel PheWAS results have identified potentially pleiotropic effects for SNPs previously associated with lipid traits, type 2 diabetes, inflammation, myocardial infarction, and body mass index. The lipid trait-associated SNPs were associated with the “Menstruation” phenotype-class (specifically age at menarche) in European Americans (CETP rs3764261), the “Dieting” phenotype-class (APOB rs562338 in African Americans and CELSR2/PSRC1/SORT1 rs599839 and rs646776 in European Americans), “Thyroid Goiter” in European Americans (LIPG rs2156552), “Artery Measurements” in European Americans (LDLR rs6511720) and “Artery Treatment” in African Americans (PCSK9 rs11591147), “Plasma Serum Glucose Levels” (APOE/APOC1/APOC4/APOC2/APOC3 rs4420638) in European Americans, and the “Angina” phenotype-class in European Americans (CELSR2/PSRC1/SORT1 rs646776). For the type 2 diabetes-associated SNPs, the PheWAS-identified associations were observed for the phenotype-classes of “Dieting” (IGFBP2 rs4402960) in European Americans, “Artery” and “Ever Smoked” (ADAMTS9 rs4607103) in European Americans, “Hypertension” (NOTCH2 rs10923931) in African Americans, “Heart Rate” (LGR5 rs7961581) in European Americans, and “Menstruation” (specifically age at menarche) in European Americans (FTO rs8050136). Like type 2 diabetes-associated ADAMTS9 rs4607103, BMI-associated NEGR1 rs2815752 was associated with the phenotype-class of “Ever Smoked” in European Americans. The final two PheWAS-identified significant associations involved nutrient based phenotype-classes: MI-associated CDKN2A/B rs2383207 was associated with the phenotype-classes of “Vitamin B12” in European Americans, and inflammation-associated IL6 rs1800795 was associated with the phenotype-class of “Carotene” in African Americans.
The PheWAS results herein present the result of tests of association between a large number of SNPs and an extensive range of phenotypes and traits available within five studies of the PAGE network. For this first PAGE PheWAS analysis we have emphasized associations that replicated across two or more independent PAGE studies for the same phenotype class and same race/ethnicity. Most of the robust findings reported here represent previously known genotype-phenotype relationships, but a tantalizing few also represent potentially novel pleiotropic relationships.
The 33 novel results presented here are intriguing, but it is important to emphasize that these first-pass analyses are considered hypothesis-generating, exploratory, and require additional scrutiny before the findings are further considered for follow-up, unlike the directed a priori hypothesis-testing analyses within PAGE that involve SNPs hypothesized to be associated with specific phenotypes. Further analysis of PheWAS results will be on an individual result basis and will include careful phenotype harmonization for traits and outcomes that cross two or more PAGE studies, as well as considerable investigation of the possible effect of covariates such as age, sex, and environmental exposure(s) on the association between genetic variation and phenotypic outcome.
One of the many challenges for the interpretation of PheWAS results is dissecting the genetic effect observed among correlated phenotypes. In some cases, the relationship is likely attributable to a common biological process with known genetic contribution (e.g., body mass index and waist circumference). In other cases, the networks that exist between intermediary and/or outcome related phenotypes add complexity to interpreting association results. For instance, genetic variation may impact the variation of a single phenotype, but variation in that phenotype could then result in changes in other downstream phenotypes indirectly. Examples of added complexity include obesity leading to impaired immune function [28], and metabolic syndrome where there is a spectrum of risk factors that are all associated with increased risk of cardiovascular disease and type 2 diabetes [25]. As a result, significant associations between a genetic variant and many phenotypes could represent a network or cascade of events. This is a potential interpretation of results found for SNP rs10923931 (NOTCH2) in AA, where type 2 diabetes was the previously reported association for this SNP and the novel result was found for hypertension, and type 2 diabetes and hypertension are often a co-occurrence. Further analysis of individual PheWAS results is necessary to conclusively establish the impact of the relationship between phenotypes on significant SNP-phenotype associations.
With the large number of phenotype-genotype associations calculated, there will be an increase in type 1 error due to multiple testing. A Bonferroni correction could be used within each individual study to choose a cutoff for significance that controls for multiple hypothesis testing. However, this would not take into account the correlations that exist between the phenotypes in these studies that impact the assumption of independence between tests as well as the correlations between the genotypes.
For our first PAGE PheWAS analysis, we chose to seek replication of results across studies and required the same direction of effect as one approach to reduce the false discovery rate. Significant results can still be found by chance across more than one study. Multiple challenges arise when attempting to get a metric of the type 1 error rate across multiple studies. First, as with individual studies, correlations between phenotypes and previous associations for the SNPs are still present. Also, there are varying type 1 error rates depending on the number of studies available for seeking replication. Quantification of how many results were found with a p-value cutoff, and without a p-value cutoff, depending on the number of studies where replication could be sought (2, 3, 4, or 5) provides some information about the number of significant results we found, in Table 2. Table 1 has the total number of results with and without p-value cutoff for individual studies. It is important to note that in cases where replication could be sought in more than two studies, there were cases where the result replicated in 3 or more studies, further increasing our confidence in the result.
A potential limitation of this study is the granularity of phenotypes within our phenotype classes. The phenotypes within some phenotype classes are the same or extremely similar, such as white blood cell count measurements across studies. However, the phenotype class “Artery Treatment” is broad in terms of the types of phenotypes included, such as presence/absence aortic aneurysm repair and presence/absence of angioplasty of the coronary arteries. For some classes, the replicated results encompass more variation in the phenotypes captured, compared to other results. As a result, significant associations between a genetic variant and all phenotypes in a network may be present. PheWAS is an exploratory and hypothesis generating exercise, thus the choice was made to have a broader match for some groups of phenotypes in order to allow for those phenotypes to be part of the exploration of the data. In addition, misclassification of phenotypes when matching is possible, and thus can limit identification of significant associations across studies. Other potential limitations include sample size/power, study heterogeneity, and the SNPs selected for study. As shown in Table 1, there is much variability across independent PAGE studies. While each PAGE study is sizeable, individual tests of association may be underpowered depending on the availability of the genetic variant, phenotype class, and race/ethnicity. Tests of association that failed to reach statistical significance may represent underpowered genotype-phenotype relationships and will require larger epidemiologic or clinic-based samples to identify. In regards to the potential impact of heterogeneity, we have some cases where replication existed in only two or three studies out of those where replication could be sought. In some instances this may be due to power, but this also may reflect the heterogeneity between studies, such as how various phenotypes are measured in individual studies and variation in mean age across the different studies. Finally, SNPs were originally selected for this study to replicate known genotype-phenotype associations and to generalize them to diverse populations. A comprehensive set of genome-wide“agnostic” SNPs may uncover additional pleiotropic or novel genotype-phentoype relationships not tested here.
Despite the the limitations present for this PheWAS, there are multiple strengths within our study. We have had the opportunity to perform a PheWAS of substantial size with an unprecedented diversity of high quality phenotypic measurements and traits, across multiple races/ethnicities. In addition, because of this PheWAS was conducted across multiple independent studies, we were able to identify the most robust genotype-phenotype relationships across studies
This initial PheWAS within PAGE has presented challenges in terms of generating high-throughput tests of association across large epidemiologic studies as well as the synthesis of the resulting data and its eventual interpretation. Even with these limitations, this PheWAS demonstrates the utility of investigating the relationship between genetic variation and an extensive range of phenotypes by validating known genotype-phenotype associations as well as identifying novel genotype-phenotype associations, revealing complex phenotypic relationships and perhaps actual pleiotropy. The utility of this hypothesis-generating approach will continue to improve over time as more samples, variants, and phenotypes/traits across diverse populations are available for study in PAGE and other genomic resources. Larger, richer datasets coupled with methods development promise to more fully reveal the complex nature of genetic variation and its relationship with human diseases and traits.
All studies were approved by Institutional Review Boards at their respective sites (details are given in Text S1). The Population Architecture using Genomics and Epidemiology (PAGE) study includes the following epidemiologic collections: Atherosclerosis Risk in Communities (ARIC), Coronary Artery Risk in Young Adults (CARIDA), Cardiovascular Health Study (CHS), the Multiethnic Cohort (MEC), the National Health and Nutrition Examination Surveys (NHANES), Strong Heart Study (SHS), and Women's Health Initiative (WHI). For this PheWAS, data were available from ARIC, CHS, MEC, NHANES III, NHANES 1999–2002, and WHI (Table 1). The PAGE study design is described in Matise et al [26] and the PAGE PheWAS study design is described in Pendergrass et al [1].
All SNPs considered for genotyping in PAGE were candidate gene or GWAS-identified variants for phenotypes and traits available in the epidemiologic collections accessed by PAGE study sites. Cohorts and surveys were genotyped using either commercially available genotyping arrays (Affymetrix 6.0, Illumina 370CNV BeadChip), and/or custom mid- and low-throughput assays (TaqMan, Sequenom, Illumina GoldenGate or BeadXpress). Quality control was implemented at each PAGE study site independently. Study specific genotyping details are described in Text S1.
In this PheWAS, data were available for SNPs previously associated with HDL-C, LDL-C, and triglycerides [27], body mass index, obesity [28], type 2 diabetes, glucose, insulin [29], and measures of inflammation (C-reactive protein), among other diseases/traits. A total of 83 SNPs overlapped across at least PAGE study sites: ten were specifically selected for body mass index traits replication, three for C-reactive protein, six for coronary/cardiac traits, three for gout/kidney, 41 for lipids, and 20 for type 2 diabetes. Table S1 lists these SNPs, along with references reporting phenotypic associations from the NHGRI GWAS catalog [30] and the open access database of GWAS results of Johnson et al. 2009 [31]. The NHGRI GWAS catalog was most recently accessed in October, 2011. If no references were available from either of those two sources, a PubMed search was performed to retrieve relevant citations.
All tests of association were performed independently by each PAGE study site using the following analysis protocol: Linear or logistic regressions were performed for continuous or categorical dependent variables, respectively, assuming an additive genetic model (0, 1, or 2 copies of the coded allele). For variables with multiple categories, binning was used to create new variables of the form “A versus not A” for each category, and logistic regression was used to model the new binary variable. Linear regressions were repeated following a y to log (y+1) transformation of the response variable with +1 added to all continuous measurements before transformation to prevent variables recorded as zero from being omitted from analysis. All analyses were stratified by race-ethnicity.
Test of association were calculated for the number of SNPs and phenotypes listed in Table 1. The software used to calculate the associations for each study was as follows: ARIC (StatSoftware), CHS (R [32]), MEC (SAS), MEC (SAS v9.2), WHI (R), EAGLE (SAS v9.2 using the Analytic Data Research by Email (ANDRE) portal of the CDC Research Data Center in Hyattsville, MD).
All association results from the tests of association were reported in standardized templates designed by the PAGE coordinating center to facilitate data sharing. All results were then imported into a relational database (MySQL). The database was also used to match previously reported GWAS data with the SNPs analyzed in this study.
The software PheWAS-View was developed for data visualization of the PheWAS results as well as for plotting “Sun Plots” [33]. Synthesis-View [34], [35] was also used to present results within this manuscript. Both software packages are freely available software for academic users: http://ritchielab.psu.edu/ritchielab/software, and can be used with a web interface at: http://visualization.ritchielab.psu.edu/.
A total of 105 phenotype-classes were developed to manually match related phenotypes across studies. To bin related phenotypes into classes the following steps were used as visualized in Figure 6: First, using a MySQL database, the data from EAGLE, MEC, CHS, ARIC, and WHI were independently filtered for any tests of association results at p<0.01, and then lists of the unique phenotypes for each individual PAGE study were generated. The number of phenotypes that passed this significance threshold for each of the four groups was 604 (ARIC), 331 (CHS), 63 (MEC), 324 (EAGLE), 1,342 (WHI). Resulting phenotypes were then manually matched up between ARIC, CHS, MEC, EAGLE and WHI using knowledge about the phenotypes and the known focus of specific PAGE study survey questions (such as bone fracture questions used primarily for collecting information about osteoporosis). For some phenotypes, the specific phenotype existed clearly across more than one PAGE study, such as for the phenotype “Hemoglobin”, where hemoglobin measurements were present for ARIC, CHS, EAGLE, and WHI. Other groups of phenotypes that fell within similar phenotypic domains but were not represented in the same form across studies were also collected into phenotype classes. One example is the phenotypes grouped together for the phenotype class of “Allergy”. EAGLE collected specific quantitative data from allergy skin testing and had survey questions about the presence of allergies in participants. ARIC and MEC did not have skin allergy testing, but did have survey questions about the presence of allergies. Thus these allergy phenotypes were grouped together. Finally, phenotypes from all studies, regardless of significance from genotype-phenotype tests of association, were matched to the already-defined phenotype classes using the criteria described above. A phenotype that matched a phenotype class but was not associated with a SNP at the significance threshold of p<0.01 for a single study would still be included in the phenotype-class list. Using these criteria, a second curator reviewed the resultant phenotypes and phenotype classes for consistency and accuracy. To provide examples of the phenotype-classes, and which subphenotypes were matched with phenotype-classes, we show three phenotype-class examples in Table 5, and Table S2 contains the matched phenotypes across studies within the phenotype-classes for all phenotype-classes used within this study.
It is important to note resources that can be used for further investigation of the phenotypes listed in Table S2, as well as in the results presented in this paper. The following study websites contain additional information about all collected study information, including how those phenotypes were collected:
After creating phenotype-classes, significant PheWAS tests of association for single genotype-phenotype associations across PAGE studies were identified using a database query. Our criteria for considering a PheWAS test of association significant included a threshold of p<0.01 observed in two or more PAGE studies for the same SNP, phenotype class, and race/ethnicity and consistent direction of effect.
A total of 111 PheWAS tests of association met our criteria for significance (Table S3). Significant results were then binned based on class of association: known, related, and novel. In this PheWAS, Known Associations are positive controls and represent previously reported genotype-phenotype associations. Related Associations are SNPs significantly associated in this PheWAS with phenotypes judged to be closely related to phenotypes among Known Associations found here and the literature. Novel Associations are significant PheWAS results where 1) the association does not match a known association and 2) the phenotype for the PheWAS association is not within a similar phenotypic domain as the phenotype of known association.
All participating studies were approved by their respective IRBs, and all study participants signed informed consent forms.
|
10.1371/journal.pbio.1001637 | Effects of Diet on Resource Utilization by a Model Human Gut Microbiota Containing Bacteroides cellulosilyticus WH2, a Symbiont with an Extensive Glycobiome | The human gut microbiota is an important metabolic organ, yet little is known about how its individual species interact, establish dominant positions, and respond to changes in environmental factors such as diet. In this study, gnotobiotic mice were colonized with an artificial microbiota comprising 12 sequenced human gut bacterial species and fed oscillating diets of disparate composition. Rapid, reproducible, and reversible changes in the structure of this assemblage were observed. Time-series microbial RNA-Seq analyses revealed staggered functional responses to diet shifts throughout the assemblage that were heavily focused on carbohydrate and amino acid metabolism. High-resolution shotgun metaproteomics confirmed many of these responses at a protein level. One member, Bacteroides cellulosilyticus WH2, proved exceptionally fit regardless of diet. Its genome encoded more carbohydrate active enzymes than any previously sequenced member of the Bacteroidetes. Transcriptional profiling indicated that B. cellulosilyticus WH2 is an adaptive forager that tailors its versatile carbohydrate utilization strategy to available dietary polysaccharides, with a strong emphasis on plant-derived xylans abundant in dietary staples like cereal grains. Two highly expressed, diet-specific polysaccharide utilization loci (PULs) in B. cellulosilyticus WH2 were identified, one with characteristics of xylan utilization systems. Introduction of a B. cellulosilyticus WH2 library comprising >90,000 isogenic transposon mutants into gnotobiotic mice, along with the other artificial community members, confirmed that these loci represent critical diet-specific fitness determinants. Carbohydrates that trigger dramatic increases in expression of these two loci and many of the organism's 111 other predicted PULs were identified by RNA-Seq during in vitro growth on 31 distinct carbohydrate substrates, allowing us to better interpret in vivo RNA-Seq and proteomics data. These results offer insight into how gut microbes adapt to dietary perturbations at both a community level and from the perspective of a well-adapted symbiont with exceptional saccharolytic capabilities, and illustrate the value of artificial communities.
| Our intestines are populated by an almost unimaginably large number of microbial cells, most of which are bacteria. This species assemblage operates as a microbial metabolic organ, performing myriad tasks that contribute to our well-being, including processing components of our diet. The way this incredible machine assembles itself and operates remains mysterious. One approach to understanding its properties is to create artificial communities composed of a limited number of sequenced human gut bacterial species and to install them in the guts of germ-free mice that are then fed different diets. In this report, we adopt this approach. We describe the genome sequence of a new gut bacterial isolate, Bacteroides cellulosilyticus WH2, which is equipped with an unprecedented number of carbohydrate active enzymes. Deploying four different “omics” technologies, we characterize the response to diet, the relative stability, and the temporal dynamics of a 12-species artificial bacterial assemblage (including B. cellulosilyticus WH2) implanted in germ-free mouse guts. We also combine high-throughput substrate utilization screens and RNA-Seq to generate reference data analogous to a “Rosetta stone” in order to decipher what types of carbohydrates B. cellulosilyticus encounters and uses within the gut, and how it interacts with other organisms that have similar and/or distinct “professions.” This work sets the stage for future ecological and metabolic studies of more complex assemblages that more fully emulate the properties of our native gut communities.
| A growing body of evidence indicates that the tens of trillions of microbial cells that inhabit our gastrointestinal tracts extend our biological capabilities in important ways. Microbial enzymes process many compounds that would otherwise pass through our intestines unaltered [1], and cases of particular nutrient substrates favoring the growth of particular taxa are being reported [2]–[5]. Changes in diet are therefore expected to lead to changes in the composition and function of the microbiota [6]–[10]. However, our understanding of diet–microbiota interactions at a mechanistic level is still in its infancy.
The absence of a complete catalog of the microbial strains and associated genome sequences that comprise a given microbiota complicates efforts to describe how particular dietary substrates influence individual taxa, how taxa cooperate/compete to utilize nutrients, and how these many interactions in aggregate lead to emergent host phenotypes. Gnotobiotic mice colonized with defined consortia of sequenced human gut microbes, on the other hand, provide an in vivo model of the microbiota in which the identity of all taxa and genes comprising the system are known. Within these assemblages, expressed mRNAs and proteins can be attributed to their genome, gene, and species of origin, and findings of interest can be pursued in follow-up in vitro or in vivo experiments. These systems also afford an opportunity to tightly control experimental variables to a degree not possible in human studies and have proven useful in studying microbial invasion, microbe–microbe interactions, and the metabolic roles of key ecological guilds [11]–[15]. Studies aiming to better understand community-level assembly, resilience, and adaptation are therefore likely to benefit from a focus on such defined systems. However, the limited taxonomic and functional representation within artificial communities of modest complexity requires that caution be exercised when extrapolating results to more complex, naturally occurring gut communities (see Prospectus).
Culture-independent surveys of the healthy adult gut microbiota consistently conclude that it is composed primarily of members of two bacterial phyla, the Bacteroidetes and Firmicutes [16]–[21]. The dominance of these two bacterial phyla suggests that their representatives in the human gut are exquisitely adapted to its dynamic conditions, which include a constantly evolving nutrient environment. Members of the genus Bacteroides are known to be adept at utilizing both plant- and host-derived polysaccharides [22]. Comparisons of available Bacteroides genomes with those from other gut species indicate that the former are enriched in genes involved in the acquisition and metabolism of various glycans, including glycoside hydrolases (GHs) and polysaccharide lyases (PLs), as well as linked environmental sensors that control their expression (e.g., hybrid two-component systems, extracytoplasmic function (ECF) sigma factors and anti-sigma factors). Many of these genes are organized into polysaccharide utilization loci (PULs) that are distributed throughout the genome [23],[24]. Recent studies have focused on better understanding the evolution, specificity, and regulation of PULs in the genomes of species like Bacteroides thetaiotaomicron and Bacteroides ovatus [25],[26]. Little is known, however, about the metabolic strategies adopted by multiple competing species in more complex communities, how dietary changes lead to reconfigurations in community structure through changes in individual species, or whether dietary context influences which genes dominant species rely on to remain competitive with other microbes, including those genes that are components of PULs.
Here, we adopt a multifaceted approach to study an artificial community in gnotobiotic mice fed changing diets in order to better understand (i) the process by which such a community reconfigures itself structurally in response to changes in host diet; (ii) how aggregate community function, as judged by the metatranscriptome and metaproteome, is impacted when host diet is altered; (iii) how the metabolic strategies of its individual component microbes change, if at all, when the nutrient milieu is dramatically altered, with an emphasis on one prominent but understudied member of the human gut Bacteroides; and (iv) whether a microbe's metabolic versatility/flexibility correlates with competitive advantage in an assemblage containing related and unrelated species.
Though at least eight complete and 68 draft genomes of Bacteroides spp. are currently available [27], there are numerous examples of distinct clades within this genus where little genomic information exists. To further explore the genome space of one such clade, we obtained a human fecal isolate whose four 16S rRNA gene sequences indicate a close relationship to Bacteroides cellulosilyticus (Figure S1A,B). The genome of this isolate, which we have designated B. cellulosilyticus WH2, was sequenced deeply, yielding a high-quality draft assembly (23 contigs with an N50 value of 798,728 bp; total length of all contigs in the assembly, 7.1 Mb; Table S1). Annotation of its 5,244 predicted protein-coding genes using the carbohydrate active enzyme (CAZy) database [28] revealed an extraordinary complement of 503 CAZymes comprising 373 GHs, 23 PLs, 28 carbohydrate esterases (CEs), and 84 glycosyltransferases (GTs) (see Table S2 for all annotated genes in the B. cellulosilyticus WH2 genome predicted to have relevance to carbohydrate metabolism). One distinguishing feature of gut Bacteroides genomes is the substantial number of CAZymes they encode relative to those of other intestinal bacteria [29]. The B. cellulosilyticus WH2 CAZome is enriched in a number of GH families even when compared with prominent representatives of the gut Bacteroidetes (Figure S2A). When we expanded this comparison to include all 86 Bacteroidetes in the CAZy database, we found that the B. cellulosilyticus WH2 genome had the greatest number of genes for 19 different GH families, as well as genes from two GH families that had not previously been observed within a Bacteroidetes genome (Figure S2B). Altogether, B. cellulosilyticus WH2 has more GH genes at its disposal than any other Bacteroidetes species analyzed to date.
In Bacteroides spp., CAZymes are often located within PULs [30]. At a minimum, a typical PUL harbors a pair of genes with significant homology to the susC and susD genes of the starch utilization system (Sus) in B. thetaiotaomicron [30]–[32]. Other genes encoding enzymes capable of liberating oligo- and monosaccharides from a larger polysaccharide are also frequently present. The susC- and susD-like genes of these loci encode the proteins that comprise the main outer membrane binding and transport apparatus and thus represent key elements of these systems. A search of the B. cellulosilyticus WH2 genome for genes with strong homology to the susC- and susD-like genes in B. thetaiotaomicron VPI-5482 revealed an unprecedented number of susC/D pairs (a total of 118). Studies of other prominent Bacteroides spp. have found that the evolutionary expansion of these genes has played an important role in endowing the Bacteroides with the ability to degrade a wide range of host- and plant-derived polysaccharides [25],[33]. Analysis of deeply sampled adult human gut microbiota datasets indicates that B. cellulosilyticus strains are common, colonizing approximately 77% of 124 adult Europeans characterized in one study [18] and 62% of 139 individuals living in the United States examined in another survey [20]. We hypothesized that the apparent success of B. cellulosilyticus in the gut is derived in part from its substantial arsenal of genes involved in carbohydrate utilization.
To test the fitness of B. cellulosilyticus WH2 in relation to other prominent gut symbionts, and the importance of diet on its fitness, we carried out an experiment in gnotobiotic mice (experiment 1, “E1,” Figure S3). Two groups of 10–12-wk-old male germ-free C57BL/6J animals were moved to individual cages within gnotobiotic isolators (n = 7 animals/group). At day zero, each animal was colonized by oral gavage with an artificial community comprising 12 human gut bacterial species (Figure 1A, Table S3). Each species chosen for inclusion in this microbial assemblage met four criteria: (i) it was a member of one of three bacterial phyla routinely found in the human gut (i.e., Bacteroidetes, Firmicutes, or Actinobacteria), (ii) it was identified as a prominent member of the human gut microbiota in previous culture-independent surveys, (iii) it could be grown in the laboratory, and (iv) its genome had been sequenced to at least a high-quality draft level. Species were also selected for their functional attributes (as judged by their annotated gene content) in an effort to create an artificial community that was somewhat representative of a more complex human microbiota. For example, although more than half of the species in the assemblage were Bacteroidetes predicted to excel at the breakdown of polysaccharides, several were also prominent inhabitants of the human gut that are thought to have limited carbohydrate utilization capabilities (e.g., Firmicutes from Clostridium cluster XIVa). Some attributes for the 12 strains included in the artificial community are provided in Table S4.
For 2 wk, each treatment group was fed a standard low-fat/high-plant polysaccharide (LF/HPP) mouse chow, or a “Western”-like diet where calories are largely derived from fat, starch, and simple sugars (high-fat/high-sugar (HF/HS)) [12]. Over the course of 6 wk, diets were changed twice at 2-wk intervals, such that each group began and ended on the same diet, with an intervening 2-wk period during which the other diet was administered (Figure S3).
Using fecal DNA as a proxy for microbial biomass, the plant polysaccharide-rich LF/HPP diet supported 2- to 3-fold more total bacterial growth (primary productivity) despite its lower caloric density (3.7 kcal/g versus 4.5 kcal/g for the HF/HS diet; Figure S4A). The HF/HS diet contains carbohydrates that are easily metabolized and absorbed in the proximal intestine (sucrose, corn starch, and maltodextrin), with cellulose being the one exception (4% of the diet by weight versus 46.3% for the other carbohydrate sources). Thus, in mice fed the HF/HS diet, diet-derived simple sugars are likely to be rare in the distal gut where the vast majority of gut microbes reside; this may provide an advantage to those bacteria capable of utilizing other carbon sources (e.g., proteins/oligopeptides, host glycans). In mice fed the LF/HPP diet, on the other hand, plant polysaccharides that are indigestible by the host should provide a plentiful source of energy for saccharolytic members of the artificial community.
To evaluate the impact of each initial diet and subsequent diet switch on the structural configuration of the artificial community, we performed shotgun sequencing (community profiling by sequencing; COPRO-Seq) [11] of DNA isolated from fecal samples collected throughout the course of the experiment, as well as cecal contents collected at sacrifice. The relative abundances of the species in each sample (defined by the number of sequencing reads that could be unambiguously assigned to each microbial genome after adjusting for genome uniqueness) were subjected to ordination by principal coordinates analysis (PCoA) (Figure S5A). As expected, diet was found to be the predominant explanatory variable for observed variance (see separation along principal coordinate 1, “PC1,” which accounts for 52% of variance). The overall structure of the artificial community achieved quasi-equilibrium before the midpoint of the first diet phase, as evidenced by the lack of any significant movement along PC1 after day five. A structural reconfiguration also took place over the course of ∼5 d following transition to the second diet phase. Notably, the two treatment groups underwent a near-perfect inversion in their positions along PC1 after the first diet switch; the artificial community in animals switched from a LF/HPP to HF/HS diet took on a structure like that which arose by the end of the first diet phase in animals consuming the HF/HS diet, and vice versa. The second diet switch from phase 2 to 3 resulted in a similar movement along PC1 in the opposite direction, indicating a reversion of the artificial community's configuration to its originally assembled structure in each treatment group. These results, in addition to demonstrating the significant impact of these two diets on the structure of this 12-member artificial human gut community, also suggest that an assemblage of this size is capable of demonstrating resilience in the face of substantial diet perturbations.
The assembly process and observed diet-induced reconfigurations also proved to be highly reproducible as evidenced by COPRO-Seq results from a replication of E1 (experiment 2, “E2”). In this follow-up experiment, fecal samples were collected more frequently than in E1, providing a dataset with improved temporal resolution. Ordination of E2 COPRO-Seq data by PCoA showed that (i) for each treatment group in E2, the artificial community assembles in a manner similar to its counterpart in E1; (ii) structural reconfigurations in response to diet occur with the same timing as in E1; and (iii) the quasi-equilibria achieved during each diet phase are highly similar between experiments for each treatment group (compare Figures 1B and S5A). As in E1, cecal data for each E2 treatment group overlap with their corresponding fecal samples, and DNA yields from E2 fecal samples vary substantially as a function of host diet (Figure S4B).
COPRO-Seq provides precise measurements of the proportional abundance of each member species present in the artificial community. Data collected in both E1 and E2 (Table S5) revealed significant differences between members in terms of the maximum abundance levels they achieved, the rates at which their abundance levels were impacted by diet shifts, and the degree to which each species demonstrated a preference for one diet over another (Figure S4C). Changes in each species' abundance over time replicated well across animals in each treatment group, suggesting the assembly process and diet-induced reconfigurations occur in an orderly, rules-based fashion and with minimal stochasticity in this artificial community. A species' relative abundance immediately after colonization (i.e., 24 h after gavage/day 1) was, in general, a poor predictor of its abundance at the end of the first diet phase (i.e., day 13) (E1 R2 = 0.23; E2 R2 = 0.27), suggesting that early dominance of the founder population was not strongly tied to relative success in the assembly process.
In mice initially fed a HF/HS diet, four Bacteroides spp. (Bacteroides caccae, B. cellulosilyticus WH2, B. thetaiotaomicron, and Bacteroides vulgatus) each achieved a relative abundance of ≥10% by the end of the first diet phase (day 13 postgavage), with B. caccae attaining the highest levels (37.1±4.9% and 34.2±5.5%; group mean ± SD in E1 and E2, respectively). In animals fed the plant polysaccharide-rich LF/HPP chow during the first diet phase, B. cellulosilyticus WH2 was dominant, achieving levels of 37.1±2.0% (E1) and 41.6±3.9% (E2) by day 13. B. thetaiotaomicron and B. vulgatus also attained relative abundances of >10%.
Changes in diet often resulted in rapid, dramatic changes in a species' proportional representation. Because the dynamic range of abundance values observed when comparing multiple species was substantial (lowest, Dorea longicatena (<0.003%); highest, B. caccae (55.0%)), comparing diet responses on a common scale using raw abundance values was challenging. To represent these changes in a way that scaled absolute increases/decreases in relative abundance to the range observed for each strain, we also normalized each species' representation within the artificial community at each time-point to the maximum proportional abundance each microbe achieved across all time-points within each mouse. Plotting the resulting measure of abundance (percentage of maximum achieved; PoMA) over time demonstrates which microbes are strongly responsive to diet (experience significant swings in PoMA value following a diet switch) and which are relatively diet-insensitive (experience only modest or no significant change in PoMA value following a diet switch). Heatmap visualization of E1 PoMA values (Figure S5B) indicated that those microbes with a preference for a particular diet in one animal treatment group also tended to demonstrate the same diet preference in the other. Likewise, diet insensitivity was also consistent across treatment groups; diet-insensitive microbes were insensitive regardless of the order in which diets were introduced.
Of the diet-sensitive taxa, those showing the most striking responses were B. caccae and B. ovatus, which strongly preferred the “Western”-like HF/HS diet and the polysaccharide-rich LF/HPP diet, respectively (Figures 1C and S4C). Among the diet-insensitive taxa, B. thetaiotaomicron showed the most stability in its representation (Figures 1C and S4C), consistent with its reputation as a versatile forager. Paradoxically, B. cellulosilyticus WH2 was both diet-sensitive and highly fit on its less-preferred diet; although this strain clearly achieved higher levels of representation in animals fed the LF/HPP diet, it also maintained strong levels of representation in animals fed the HF/HS diet (Figures 1C and S4C).
When taking into account the abundance data for all 12 artificial community members, proportional representation at the end of the first diet phase (i.e., day 13) was a good predictor of representation at the end of the third diet phase (i.e., day 42) (E1 R2 = 0.77; E2 R2 = 0.84), suggesting that the intervening dietary perturbation had little effect on the ultimate outcomes for most species within this assemblage. However, one very low-abundance strain (D. longicatena) achieved significantly different maximum percentage abundances across the two treatment groups in each experiment, suggesting that steady-state levels of this strain may have been impacted by diet history. In mice initially fed the LF/HPP diet, D. longicatena was found to persist throughout the experiment at low levels on both diet regimens. In mice initially fed the HF/HS diet, D. longicatena dropped below the limit of detection before the end of the first diet phase, was undetectable by the end of the second diet phase, and remained undetectable throughout the rest of the time course. This interesting example raises the possibility that for some species, irreversible hysteresis effects may play a significant role in determining the likelihood that they will persist within a gut over long periods of time.
These diet-induced reconfigurations in the structure of the artificial community led us to examine the degree to which its members were modifying their metabolic strategies. To establish an initial baseline static view of expression data for each microbe on each diet, we developed a custom GeneChip whose probe sets were designed to target 46,851 of the 48,023 known or predicted protein-coding genes within our artificial human gut microbiome (see Materials and Methods). Total RNA was collected from the cecal contents of each animal in E1 at the time of sacrifice and hybridized to this GeneChip. The total number of genes whose expression was detectable on each diet was remarkably similar (14,929 and 14,594 detected in the LF/HPP→HF/HS→LF/HPP and HF/HS→LF/HPP→HF/HS treatment groups, respectively). A total of 11,373 genes (24.3%) were expressed on both diets (Figure S6A), while 2,003 (4.3%) were differentially expressed to a statistically significant degree, including 161 (6.1%) of the 2,640 genes in the microbiome encoding proteins with CAZy-recognized domains. Figure S6B illustrates the fraction of the community-level CAZome and several species-level CAZomes expressed on each diet (see Table S6 for a comprehensive list of all genes, organized by species and fold-change in expression, whose cecal expression was detectable on each diet and all genes whose expression was significantly different when comparing data from each treatment group).
Among taxa demonstrating obvious diet preferences (as judged by relative abundance data), B. caccae and B. cellulosilyticus WH2 provided examples of CAZy-level responses to diet change that were different in several respects. Our observations regarding the carbohydrate utilization capabilities and preferences of B. caccae are summarized in Text S1. However, our ability to evaluate shifts in B. caccae's metabolic strategy in the gut was limited by its very low abundance in animals fed LF/HPP chow (i.e., our mRNA and subsequent protein assays were often not sensitive enough to exhaustively sample B. caccae's metatranscriptome and metaproteome). In contrast, the abundance of B. cellulosilyticus WH2, which favored the LF/HPP diet, remained high enough on both diets to allow for a comprehensive analysis of its expressed genes and proteins. This advantage, along with the exceptional carbohydrate utilization machinery encoded within the genome of this organism, encouraged us to focus on further dissecting the responses of B. cellulosilyticus WH2 to diet changes.
Detailed inspection of the expressed B. cellulosilyticus WH2 CAZome (503 CAZymes in total) provided an initial view of this microbe's sophisticated carbohydrate utilization strategy. A comparison of the top decile of expressed CAZymes on each diet disclosed many shared elements between the two lists, spanning many different CAZy families, with just over half of the 50 most expressed enzymes on the plant polysaccharide-rich LF/HPP chow also occurring in the list of most highly expressed enzymes on the sucrose-, corn starch-, and maltodextrin-rich HF/HS diet (Figure 2A). Twenty-five of the 50 most expressed CAZymes on the LF/HPP diet were significantly up-regulated compared to the HF/HS diet; of these, seven were members of the GH43 family (Figure 2B). The GH43 family consists of enzymes with activities required for the breakdown of plant-derived polysaccharides such as hemicellulose and pectin. Inspection of the enzyme commission (EC) annotations for the most up-regulated GH43 genes shows that they encode xylan 1,4-β-xylosidases (EC 3.2.1.37), arabinan endo-1,5-α-L-arabinosidases (EC 3.2.1.99), and α-L-arabinofuranosidases (EC 3.2.1.55). The GH10 family, which is currently comprised exclusively of endo-xylanases (EC 3.2.1.8, EC 3.2.1.32), was also well represented among this set of 25 genes, with four of the seven putative GH10 genes in the B. cellulosilyticus WH2 genome making the list. Strikingly, of the 45 predicted genes with putative GH43 domains in the B. cellulosilyticus WH2 genome, none were up-regulated on the “Western”-style HF/HS diet.
The most highly expressed B. cellulosilyticus WH2 CAZyme on the plant polysaccharide-rich chow (which was also highly-expressed on the HF/HS chow) was BWH2_1228, a putative α-galactosidase from the GH36 family. These enzymes, which are not expressed by humans in the stomach or intestine, cleave terminal galactose residues from the nonreducing ends of raffinose family oligosaccharides (RFOs, including raffinose, stachyose, and verbascose), galacto(gluco)mannans, galactolipids, and glycoproteins. RFOs, which are well represented in cereal grains consumed by humans, are expected to be abundant in the LF/HPP diet given its ingredients (e.g., soybean meal), but potential substrates in the HF/HS diet are less obvious, possibly implicating a host glycolipid or glycoprotein target.
Surface glycans in the intestinal epithelium of rodents are decorated with terminal fucose residues [34] as well as terminal sialic acid and sulfate [35]. Hydrolysis of the α-2 linkage connecting terminal fucose residues to the galactose-rich extended core is thought to be catalyzed in large part by GH95 and GH29 enzymes [36]. The B. cellulosilyticus WH2 genome is replete with putative GH95 and GH29 genes (total of 12 and 9, respectively), but only a few (BWH2_1350/2142/3154/3818) were expressed in vivo on at least one diet, and their expression was low relative to many other CAZymes (see Table S6). Cleavage of terminal sialic acids present in host mucins by bacteria is usually carried out by GH33 family enzymes. B. cellulosilyticus WH2 has two GH33 genes that are expressed on either one diet (BWH2_3822, HF/HS) or both diets (BWH2_4650), but neither is highly expressed relative to other B. cellulosilyticus WH2 CAZymes. Therefore, utilization of host glycans by B. cellulosilyticus WH2, if it occurs, likely requires partnerships with other members of the artificial community that express GH29/95/33 enzymes (see Table S6 for a list of members that express these enzymes in a diet-independent and/or diet-specific fashion).
Among the 50 most highly expressed B. cellulosilyticus WH2 CAZymes, 12 were significantly up-regulated on the HF/HS diet compared to the LF/HPP diet, with members of family GH13 being most prevalent. While the enzymatic activities and substrate specificities of GH13 family members are varied, most relate to the hydrolysis of substrates comprising chains of glucose subunits, including amylose (one of the two components of starch) and maltodextrin, both prominent ingredients in the HF/HS diet.
GeneChip-based profiling of the E1 cecal communities provided a snapshot of the metatranscriptome on the final day of the final diet phase in each treatment group. The analysis of B. cellulosilyticus WH2 CAZyme expression suggested that this strain achieves a “generalist” lifestyle not by relying on substrates that are present at all times (e.g., host mucins), but rather by modifying its resource utilization strategy to effectively compete with other microbes for diet-derived polysaccharides that are not metabolized by the host.
To develop a more complete understanding of the dynamic changes that occur in gene expression over time and throughout the artificial community following diet perturbations, we performed microbial RNA-Seq analyses using feces obtained at select time-points from mice in the LF/HPP→HF/HS→LF/HPP treatment group of E2 (Figure S3).
We began with a “top-down” analysis in which every RNA-Seq read count from every gene in the artificial microbiome was binned based on the functional annotation of the gene from which it was derived, regardless of its species of origin. In this case, the functional annotation used as the binning variable was the predicted EC number for a gene's encoded protein product. Expecting that some changes might occur rapidly, while others might require days or weeks, we searched for significant differences between the terminal time-points of the first two diet phases (i.e., points at which the model human gut microbiota had been allowed 13 d to acclimate to each diet). The 157 significant changes we identified were subjected to hierarchical clustering by EC number to determine which functional responses occurred with similar kinetics. The results revealed that in contrast to the rapid, diet-induced structural reconfigurations observed in this artificial community, community-level changes in microbial gene expression occurred with highly variable timing that differed from function to function. These changes were dominated by EC numbers associated with enzymatic reactions relevant to carbohydrate and amino acid metabolism (see Table S7 for a summary of all significant changes observed, including aggregate expression values for each functional bin (EC number) at each time-point). Significant responses could be divided into one of three groups: “rapid” responses were those where the representation of EC numbers in the transcriptome increased/decreased dramatically within 1–2 d of a diet switch; “gradual” responses were those where the representation of EC numbers changed notably, but slowly, between the two diet transition points; and “delayed” responses were those where significant change did not occur until the end of a diet phase (Figure 3, Table S7). EC numbers associated with reactions important in carbohydrate metabolism and transport were distributed across all three of these response types for each of the two diets. Nearly all genes encoding proteins with EC numbers related to amino acid metabolism that were significantly up-regulated on HF/HS chow binned into the “rapid” or “gradual” groups, suggesting this diet put immediate pressure on the artificial microbial community to increase its repertoire of expressed amino acid biosynthesis and degradation genes. Genes with assigned EC numbers involved in amino acid metabolism that were significantly up-regulated on the other, polysaccharide-rich, LF/HPP diet were spread more evenly across these three response types (Figure 3).
Careful inspection of our top-down analysis results and a complementary “bottom-up” analysis in which normalization was performed at the level of individual species, rather than at the community level, allowed us to identify other important responses that would have gone undetected were it not for the fact that we were dealing with a defined assemblage of microbes where all of the genes in component members' genomes were known. For example, an assessment of the representation of EC 3.2.1.8 (endo-1,4-β-xylanase) within the metatranscriptome before and after the first diet switch (LF/HPP→HF/HS) initially suggested that this activity was reduced to a statistically significant degree as a result of the first diet perturbation (day 13 versus day 27; Mann–Whitney U test, p = 0.03; Figure S7A). Aggregation by species of all sequencing read counts assignable to mRNAs encoding proteins with this EC number revealed that over 99% of the contributions to this functional bin originated from B. cellulosilyticus WH2 (note the similarity in a comparison of Figure S7A and Figure S7B), implying that the community-level response and the response of this Bacteroides species were virtually one and the same. A tally of all sequencing reads assignable to B. cellulosilyticus WH2 at each time-point disclosed that although this strain maintains high proportional representation in the artificial community throughout each diet oscillation period (range, 10.3–42.5% and 11.6–43.3% for E1 and E2, respectively), its contribution to the metatranscriptome is substantially decreased during the HF/HS diet phase (Figure S7C). This dramatic reduction in the extent to which B. cellulosilyticus WH2 contributes to the metatranscriptome in HF/HS-fed mice “masks” the significant up-regulation of EC 3.2.1.8 that occurs within the B. cellulosilyticus WH2 transcriptome following the first diet shift (day 13 versus day 27; Mann–Whitney U test, p = 0.03; Figure S7D). A further breakdown of endo-1,4-β-xylanase up-regulation in B. cellulosilyticus WH2 when mice are switched to the HF/HS diet reveals that most of this response is driven by two genes, BWH2_4068 and BWH2_4072 (Figure S7E). Our realization that we were unable to correctly infer the direction of one of the most significant diet-induced gene expression changes in the second most abundant strain in the artificial community when inspecting functional responses at the community level provides a strong argument for expanding the use of microbial assemblages comprised exclusively of sequenced species in studies of the gut microbiota. This should allow the contributions of individual species to community activity to be evaluated in a rigorous way that is not possible with microbial communities of unknown or poorly defined gene composition.
In principle, protein measurements can provide a more direct readout of expressed community functions than an RNA-level analysis, and thus a deeper understanding of community members' interactions with one another and with their habitat [37],[38]. For these reasons and others, much work has been dedicated to applying shotgun proteomics techniques to microbial ecosystems in various environments [39],[40]. Though these efforts have provided illustrations of significant methodological advances, they have been limited by the complexity of the metaproteomes studied and by the difficulties this complexity creates when attempting to assign peptide identities uniquely to proteins of specific taxa. Recognizing that a metaproteomics analysis of our artificial community would not be subject to such uncertainty given its fully defined microbiome and thus fully defined theoretical proteome, we subjected cecal samples from two mice from each diet treatment group in E1 (n = 4 total) to high-performance liquid chromatography-tandem mass spectrometry (LC-MS/MS; see Materials and Methods). We had three goals: (i) to evaluate how our ability to assign peptide-spectrum matches (PSMs) to particular proteins within a theoretical metaproteome is affected by the presence of close homologs within the same species and within other, closely related species; (ii) to test the limits of our ability to characterize protein expression across different species given the substantial dynamic range we documented in microbial species abundance; and (iii) to collect semiquantitative peptide/protein data that might validate and enrich our understanding of functional responses identified at the mRNA level, particularly with respect to the niche (profession) of CAZyme-rich B. cellulosilyticus WH2.
Given the evolutionary relatedness of the strains involved, we expected that some fraction of observed PSMs from each sample would be of ambiguous origin due to nonunique peptides shared between species' proteomes. To assess which species might be most affected by this phenomenon when characterizing the metaproteome on different diets, we catalogued each strain's theoretical peptidome using an in silico tryptic digest. This simulated digest took into account both the potential for missed trypic cleavages and the peptide mass range that could be detected using our methods. The results (Figure S8A, Table S8) demonstrated that for an artificial community of modest complexity, the proportion of peptides within each strain's theoretical peptidome that are “unique” (i.e., assignable to a single protein within the theoretical metaproteome) varies substantially from species to species, even among those that are closely related. We found the lone representative of the Actinobacteria in the artificial community, Collinsella aerofaciens, to have the highest proportion of unique peptides (94.2%), while B. caccae had the lowest (63.0%). Interestingly, there was not a strong correlation between the fraction of a species' peptides that were unique and the total number of unique peptides that species contributed to the theoretical peptidome. For example, C. aerofaciens (2,367 predicted protein-coding genes) contributed only 81,894 (1.5%) unique peptides, the lowest of any artificial community member evaluated, despite having a proteome composed of mostly unique peptides. On the other hand, B. cellulosilyticus WH2 (5,244 predicted protein-coding genes) contributed 241,473 (4.5%) unique peptides, the highest of any member despite a high fraction of nonunique peptides (18.4%) within its theoretical peptidome. The evolutionary relatedness of the Bacteroides components of the artificial community appeared to negatively affect our ability to assign their peptides to specific proteins; their six theoretical peptidomes had the six lowest uniqueness levels. However, their greater number of proteins and peptides relative to the Firmicutes and Actinobacteria more than compensated for this deficiency; over 60% of unique peptides within the unique theoretical metaproteome were contributed by the Bacteroides.
We also found that the proportion of PSMs uniquely assignable to a single protein within the metaproteome varied significantly by function, suggesting that some classes of proteins can be traced back to specific microbes more readily than others. For example, when considering all theoretical peptides that could be derived from the proteome of a particular bacterial species, those from proteins with roles in categories with high expected levels of functional conservation (e.g., translation and nucleotide metabolism) were on average deemed unique more often than those from proteins with roles in functions we might expect to be less conserved (e.g., glycan biosynthesis and metabolism) (see Table S8 for a summary of how peptide uniqueness varied across different KEGG categories and pathways, and across different species in the experiment). However, even in KEGG categories and pathways with high expected levels of functional conservation, the vast majority of peptides were found to be unique when a particular species was not closely related to other members of the artificial community.
Next, we determined the average number of proteins that could be experimentally identified in our samples for each microbial species within each treatment group in E1. The results (Figure S8B, Table S9) illustrate two important conclusions. First, although equal concentrations of total protein were evaluated for each sample, slightly less than twice as many total microbial proteins were identified in samples from the LF/HPP-fed mice as those from mice fed the HF/HS diet (4,659 versus 2,777, respectively). While there are a number of possible explanations, both this finding and the higher number of mouse proteins detected in samples from HF/HS-fed animals are consistent with the results of our fecal DNA analysis, which indicated that the HF/HS diet supports lower levels of gut microbial biomass than the LF/HPP diet (Figure S4A,B). Second, a breakdown of all detected microbial proteins by species of origin (Figure S8B) revealed that the degree to which we could inspect protein expression for a given species was dictated largely by its relative abundance and the diet to which it was exposed.
Our ability to detect many of B. cellulosilyticus WH2's expressed transcripts and proteins in samples from both diet treatment groups allowed us to determine how well RNA and protein data for an abundant, active member of the artificial community might correlate. These data also allowed us to evaluate whether or not the types of genes considered might influence the degree of correlation between these two datasets. We first performed a spectral count-based correlation analysis on the diet-induced, log-transformed, average fold-differences in expression for all B. cellulosilyticus WH2 genes that were detectable at both the RNA and protein level for both diets. The results revealed a moderate degree of linear correlation between RNA and protein observations (Figure S8C, black plot; r = 0.53). However, subsequent division of these genes into functionally related subsets, which were each subjected to their own correlation analysis, revealed striking differences in the degree to which RNA-level and protein-level expression changes agreed with one another. For example, diet-induced changes in mRNA expression for genes involved in translation showed virtually no correlation with changes measured at the protein level (Figure S8C, red plot; r = 0.03). Correlations for other categories of B. cellulosilyticus WH2 genes, such as those involved in energy metabolism (Figure S8C, green plot; r = 0.36) and amino acid metabolism (Figure S8C, orange plot; r = 0.48), were also poorer than the correlation for the complete set of detectable genes. In contrast, the correlation for the 110 genes with predicted involvement in carbohydrate metabolism was quite strong (Figure S8C, blue plot; r = 0.69), and was in fact the best correlation identified for any functional category of genes considered. The significant range of correlations observed in different categories of genes suggests that the degree to which RNA-based analyses provide an accurate picture of microbial adaptation to environmental perturbation may be strongly impacted by the functional classification of the genes involved. Additionally, these data further emphasize the need for enhanced dynamic range metaproteome measurements and better bioinformatic methods to assign/bin unique and nonunique peptides so that deeper and more thorough surveys of the microbial protein landscape can be performed and evaluated alongside more robust transcriptional datasets.
Several of the most highly expressed and diet-sensitive B. cellulosilyticus WH2 genes in this study fell within two putative PULs. One PUL (BWH2_4044–55) contains 12 ORFs that include a dual susC/D cassette, three putative xylanases assigned to CAZy families GH8 and GH10, a putative multifunctional acetyl xylan esterase/α-L-fucosidase, and a putative hybrid two-component system regulator (Figure 4A). Gene expression within this PUL was markedly higher in mice consuming the plant polysaccharide-rich LF/HPP diet at both the mRNA and protein level. Our mRNA-level analysis disclosed that BWH2_4047 was the most highly expressed B. cellulosilyticus WH2 susD homolog on this diet. Likewise, BWH2_4046/4, the two susC-like genes within this PUL, were the 2nd and 4th most highly expressed B. cellulosilyticus WH2 susC-like genes in LF/HPP-fed animals, and exhibited expression level reductions of 99.5% and 93% in animals consuming the HF/HS diet. The same LF/HPP diet bias was observed for other genes within this PUL (Figures 2A and 4B) but not for neighboring genes. The same trends were obvious and amplified when we quantified protein expression (Figure 4C). In mice fed LF/HPP chow, only three B. cellulosilyticus WH2 SusC-like proteins had higher protein levels than BWH2_4044/6, and only two SusD-like proteins had higher levels than BWH2_4045/7. Strikingly, we were unable to detect a single peptide from 9 of the 12 proteins in this PUL in samples obtained from mice fed the HF/HS diet, emphasizing the strong diet specificity of this locus.
A second PUL in the B. cellulosilyticus WH2 genome composed of a susC/D-like pair (BWH2_4074/5), a putative hybrid two-component system regulator (BWH2_4076), and a xylanase (GH10) with dual carbohydrate binding module domains (CBM22) (BWH2_4072) (Figure 4A) demonstrated a strong but opposite diet bias, in this case exhibiting significantly higher expression in animals consuming the HF/HS “Western”-like diet. Our mRNA-level analysis showed that this xylanase was the second most highly expressed B. cellulosilyticus WH2 CAZyme in animals consuming this diet (Figure 2A). As with the previously described PUL, shotgun metaproteomics validated the transcriptional analysis (Figure 4B,C); with the exception of the gene encoding the PUL's presumed transcriptional regulator (BWH2_4076), diet specificity was substantial, with protein-level fold changes ranging from 10–33 across the locus (Table S10).
Recent work by Cann and co-workers has done much to advance our understanding of the regulation and metabolic role of xylan utilization system gene clusters in xylanolytic members of the Bacteroidetes, particularly within the genus Prevotella [41]. The “core” gene cluster of the prototypical xylan utilization system they described consists of two tandem repeats of susC/susD homologs (xusA/B/C/D), a downstream hypothetical gene (xusE) and a GH10 endoxylanase (xyn10C). The 12-gene PUL identified in our study (BWH2_4044–55) appears to contain the only instance of this core gene cluster within the B. cellulosilyticus WH2 genome, suggesting that this PUL, induced during consumption of a plant polysaccharide-rich diet, is likely to be the primary xylan utilization system within this organism. A recent study characterizing the carbohydrate utilization capabilities of B. ovatus ATCC 8483 also identified two PULs involved in xylan utilization (BACOVA_04385–94, BACOVA_03417–50) whose gene configurations differ from those described in Prevotella spp. [25]. Interestingly, the five proteins encoded by the smaller xylanase-containing PUL described above (BWH2_4072–6) are homologous to the products of the last five genes in BACOVA_4385–94 (i.e., BACOVA_4390–4). The order of these five genes in these two loci is also identical. The similarities and differences observed when comparing the putative xylan utilization systems encoded within the genomes of different Bacteroidetes illustrate how its members may have evolved differentiated strategies for utilizing hemicelluloses like xylan.
Having established that expression of BWH2_4044–55 and BWH2_4072–6 is strongly dictated by diet, we next sought to determine if these PULs are required by B. cellulosilyticus WH2 for fitness in vivo. A follow-up study was performed in which mice were fed either a LF/HPP or HF/HS diet after being colonized with an artificial community similar to the one used in E1 and E2 (see Materials and Methods). The wild-type B. cellulosilyticus WH2 strain used in our previous experiments was replaced with a transposon mutant library consisting of over 90,000 distinct transposon insertion mutants in 91.5% of all predicted ORFs (average of 13.9 distinct insertion mutants per ORF). The library was constructed using methods similar to those reported by Goodman et al. ([42]; see Materials and Methods) so that the relative proportion of each insertion mutant in both the input (orally gavaged) and output (fecal) populations could be determined using insertion sequencing (INSeq). The INSeq results revealed clear, diet-specific losses of fitness when components of these loci were disrupted (Figure 4D). Additionally, as observed in E1 and E2, expression of each PUL was strongly biased by diet, with genes BWH2_4072–5 demonstrating up-regulation on the HF/HS diet and BWH2_4044–55 on the LF/HPP diet. The extent to which a gene's disruption impacted the fitness of B. cellulosilyticus WH2 on one diet or the other correlated well with whether or not that gene was highly expressed on a given diet. For example, four of the five most highly expressed genes in the BWH2_4044–55 locus were the four genes shown to be most crucial for fitness on the LF/HPP diet. Of these four genes, three were susC or susD homologs (the fourth was the putative endo-1,4-β-xylanase thought to constitute the last element of the xylan utilization system core). Though the fitness cost of disrupting genes within BWH2_4044–55 varied from gene to gene, disruption of any one component of the BWH2_4072–6 PUL had serious consequences for B. cellulosilyticus WH2 in animals fed the HF/HS diet. This could suggest that while disruption of some components of the BWH2_4044–55 locus can be rescued by similar or redundant functions elsewhere in the genome, the same is not true for BWH2_4072–5. Notably, disruption of BWH2_4076, which is predicted to encode a hybrid two-component regulatory system, had negative consequences on either diet tested, indicating that regulation of this locus is crucial even when the PUL is not actively expressed. While many genes outside of these two PULs were also found to be important for the in vivo fitness of B. cellulosilyticus WH2, those within these PULs were among the most essential to diet-specific fitness, suggesting that these loci are central to the metabolic lifestyle of B. cellulosilyticus WH2 in the gut.
The results described in the preceding section indicate that B. cellulosilyticus WH2 prioritizes xylan as a nutrient source in the gut and that it tightly regulates the expression of its xylan utilization machinery. Moreover, the extraordinary number of putative CAZymes and PULs within the B. cellulosilyticus WH2 genome suggests that it is capable of growing on carbohydrates with diverse structures and varying degrees of polymerization. To characterize its carbohydrate utilization capabilities, we defined its growth in minimal medium (MM) supplemented with one of 46 different carbohydrates [25]. Three independent growths, each consisting of two technical replications, yielded a total of six growth curves for each substrate. Of the 46 substrates tested, B. cellulosilyticus WH2 grew on 39 (Table S11); they encompassed numerous pectins (6 of 6), hemicelluloses/β-glucans (8 of 8), starches/fructans/α-glucans (6 of 6), and simple sugars (14 of 15), as well as host-derived glycans (4 of 7) and one cellooligosaccharide (cellobiose). The seven substrates that did not support growth included three esoteric carbohydrates (carrageenan, porphyran, and alginic acid), the simple sugar N-acetylneuraminic acid, two host glyans (keratan sulfate and mucin O-glycans), and fungal cell wall-derived α-mannan. B. cellulosilyticus WH2 clearly grew more robustly on some carbohydrates than others. Excluding simple sugars, fastest growth was achieved on dextran (0.099±0.048 OD600 units/h), laminarin (0.095±0.014), pectic galactan (0.088±0.018), pullulan (0.088±0.026), and amylopectin (0.085±0.003). Although one study has reported that the type strain of B. cellulosilyticus degrades cellulose [43], the WH2 strain failed to demonstrate any growth on MM plus cellulose (specifically, Solka-Floc 200 FCC from International Fiber Corp.) after 5 d. Maximum cell density was achieved with amylopectin (1.17±0.02 OD600 units), dextran (1.12±0.20), cellobiose (1.09±0.08), laminarin (1.08±0.08), and xyloglucan (0.99±0.04). Total B. cellulosilyticus WH2 growth (i.e., maximum cell density achieved) on host-derived glycans was typically very poor, with only two substrates achieving total growth above 0.2 OD600 units (chondroitin sulfate, 0.50±0.04; glycogen, 0.99±0.02). The disparity between total growth on plant polysaccharides versus host-derived glycans, including O-glycans that are prevalent in host mucin, indicates a preference for diet-derived saccharides, consistent with our in vivo mRNA and protein expression data.
We also determined how the growth rate of B. cellulosilyticus WH2 on these substrates compared to the growth rates for other prominent gut Bacteroides spp. After subjecting B. caccae to the same phenotypic characterization as B. cellulosilyticus WH2, we combined our measurements for these two strains with previously published measurements for B. thetaiotaomicron and B. ovatus [25]. The results underscored the competitive growth advantage B. cellulosilyticus WH2 likely enjoys when foraging for polysaccharides in the intestinal lumen. For example, of the eight hemicelluloses and β-glucans tested in our carbohydrate panel, B. cellulosilyticus WH2 grew fastest on six while B. ovatus grew fastest on two (Table S11). B. caccae and B. thetaiotaomicron, on the other hand, failed to grow on any of these substrates. Across all the carbohydrates for which data are available for all four species, B. cellulosilyticus WH2 grew fastest on the greatest number of polysaccharides (11 of 26) and tied with B. caccae for the greatest number of monosaccharides (6 of 15). B. thetaiotaomicron and B. caccae did, however, outperform the other two Bacteroides tested with respect to utilization of host glycans in vitro, demonstrating superior growth rates on four of five substrates tested (Table S11).
B. cellulosilyticus WH2's rapid growth to high densities on xylan, arabinoxylan, and xyloglucan, as well as xylose, arabinose, and galactose, is noteworthy given our prediction that two of its most tightly regulated, highly expressed PULs appear to be involved in the utilization of xylan, arabinoxylan, or some closely related polysaccharide. To identify specific mono- and/or polysaccharides capable of triggering the activation of these two PULs, as well as the 111 other putative PULs within the B. cellulosilyticus WH2 genome, we used RNA-Seq to characterize its transcriptional profile at mid-log phase in MM (Table S12) plus one of 16 simple sugars or one of 15 complex sugars (Table S13) (see Materials and Methods; n = 2–3 cultures/substrate; 5.2–14.0 million raw Illumina HiSeq reads generated for each of the 90 transcriptomes). After mapping each read to the B. cellulosilyticus WH2 reference gene set, counts were normalized using DESeq to allow for direct comparisons across samples and conditions. Hierarchical clustering of the normalized dataset resulted in a well-ordered dendrogram in which samples clustered almost perfectly by the carbohydrate on which B. cellulosilyticus WH2 was grown (Figure 5A). The consistency of this clustering illustrates that (i) technical replicates within each condition exhibit strong correlations with one another, suggesting any differences between cultures in a treatment group (e.g., small differences in density or growth phase) had at best minor effects on aggregate gene expression, and (ii) growth on different carbohydrates results in distinct, substrate-specific gene expression signals capable of driving highly discriminatory differences between treatment groups. The application of rigorous bootstrapping to our hierarchical clustering results also revealed several cases of higher level clusters in which strong confidence was achieved. These dendrogram nodes (illustrated as white circles) indicate sets of growth conditions that yield gene expression patterns more like each other than like the patterns observed for other substrates. Two notable examples were xylan/arabinoxylan (which are structurally related and share the same xylan backbone) and L-fucose/L-rhamnose (which are known to be metabolized via parallel pathways in E. coli [44]).
Importantly, these findings suggested that by considering in vitro profiling data alongside in vivo expression data from the artificial community, it might be possible to identify the particular carbohydrates to which B. cellulosilyticus WH2 is exposed and responding within its gut environment. To explore this concept further, we compared expression of each gene in each condition to its expression on our control treatment, MM plus glucose (MM-Glc). The results revealed a dynamic PUL activation network in which some PULs were activated by a single substrate, some were activated by multiple substrates, and some were transcriptionally silent across all conditions tested. Of the 118 putative susC/D pairs in B. cellulosilyticus WH2 that we have used as markers of PULs, 30 were dramatically activated on one or more of the substrates tested; in these cases, both the susC- and susD-like genes in the cassette were up-regulated an average of >100-fold relative to MM-Glc across all technical replicates (Figure 5B). At least one susC/D activation signature was identified for every one of the 17 oligosaccharides and polysaccharides and for six of the 13 monosaccharides tested (Table S14). The lack of carbohydrate-specific PUL activation events for some monosaccharides (fructose, galactose, glucuronic acid, sucrose, and xylose) was expected, given that these loci are primarily dedicated to polysaccharide acquisition. Further inspection of gene expression outside of PULs disclosed that B. cellulosilyticus WH2 prioritizes use of its non-PUL-associated carbohydrate machinery, such as putative phosphotransferase system (PTS) components and monosaccharide permeases, when grown on these monosaccharides (Table S14).
Several carbohydrates activated the expression of multiple PULs. Growth on water-soluble xylan and wheat arabinoxylan produced significant up-regulation of five susC/D-like pairs (BWH2_0865/6, 0867/8, 4044/5, 4046/7, and 4074/5). No other substrate tested activated as many loci within the genome, again hinting at the importance of xylan and arabinoxylan to this strain's metabolic strategy in vivo. Cecal expression data from E1 showed that 15 of these activated PULs were expressed in vivo on one or both of the diets tested (see circles to the right of the heatmap in Figure 5B). In mice fed the polysaccharide-rich LF/HPP chow, B. cellulosilyticus WH2 up-regulates three susC/D pairs (BWH2_2717/8, 4044/5, 4046/7) whose expression is activated in vitro by arabinan and xylan/arabinoxylan. The three most significantly up-regulated susC/D pairs (BWH2_1736/7, 2514/5, 4074/5) in mice fed the HF/HS diet rich in sugar, corn starch, and maltodextrin are activated in vitro by amylopectin, ribose, and xylan/arabinoxylan, respectively. All three PULs identified as being up-regulated at the RNA level in LF/HPP-fed mice were also found to be up-regulated at the protein level (Figure 5B). Two of the three PULs up-regulated at the mRNA level in HF/HS-fed mice were up-regulated at the protein level as well. The presence of an amylopectin-activated PUL among these two loci is noteworthy, given the significant amount of starch present in the HF/HS diet. The up-regulation of four other PULs in HF/HS-fed animals was only evident in our LC-MS/MS data, reinforcing the notion that protein data both complement and supplement mRNA data when profiling microbes of interest.
Two of the five susC/D pairs activated by xylan/arabinoxylan form the four-gene cassette in the previously discussed PUL comprising BWH2_4044–55 that is activated in mice fed the plant polysaccharide-rich chow (see Figure 4A). Another one of the five is the susC/D pair found in the PUL comprising BWH2_4072–6 that is activated in mice fed the HF/HS “Western”-like chow (see Figure 4A). Thus, we have identified a pair of putative PULs in close proximity to one another on the B. cellulosilyticus WH2 genome that encode CAZymes with similar predicted functions, are subject to near-identical levels of specific activation by the same two polysaccharides (i.e., xylan, arabinoxylan) in vitro, but are discordantly regulated in vivo in a diet-specific manner. The highly expressed nature of these PULs in the diet environment where they are active, their shared emphasis on xylan/arabinoxylan utilization, and their tight regulation indicate that they are likely to be important for the in vivo success of this organism in the two nutrient environments tested. However, the reasons for their discordant regulation are unclear. One possibility is that in addition to being activated by xylan/arabinoxylan and related polysaccharides, these loci are also subject to repression by other substrates present in the lumen of the gut, and this repression is sufficient to block activation. Alternatively, the specific activators of each PUL may be molecular moieties shared by both xylan and arabinoxylan that do not co-occur in the lumenal environment when mice are fed the diets tested in this study.
Elucidating generalizable “rules” for how microbiota operate under different environmental conditions is a substantial challenge. As our appreciation for the importance of the gut microbiota in human health and well-being grows, so too does our need to develop such rules using tractable experimental models of the gut ecosystem that allow us to move back and forth between in vivo and ex vivo analyses, using one to inform the other. Here, we have demonstrated the extent to which high-resolution DNA-, mRNA-, and protein-level analyses can be applied (and integrated) to study an artificial community of sequenced human gut microbes colonizing gnotobiotic mice. Our efforts have focused on characterizing community-level and species-level adaptation to dietary change over time and “leveraging” results obtained from in vitro assessments of individual species' responses to a panel of purified carbohydrates to deduce glycan exposures and consumption strategies in vivo. This experimental paradigm could be applied to any number of questions related to microbe–microbe, environment–microbe, and host–microbe interactions, including, for example, the metabolic fate of particular nutrients of interest (metabolic flux experiments), microbial succession, and biotransformations of xenobiotics.
Studying artificial human gut microbial communities in gnotobiotic mice also allows us to evaluate the technical limitations of current molecular approaches for characterizing native communities. For example, the structure of an artificial community can be evaluated over time at low cost using short read shotgun DNA sequencing data mapped to all microbial genomes within the community (COPRO-Seq). This allows for a much greater depth of sequencing coverage (i.e., more sensitivity) and much less ambiguity in the assignment of reads to particular taxa than traditional 16S rRNA gene-based sequencing. Short read cDNA sequences transcribed from total microbial community RNA can also often be assigned to the exact species and gene from which they were derived, and the same is also often true for peptides derived from particular bacterial proteins. However, substantial dynamic range in species/transcript/protein abundance within any microbiota, defined or otherwise, imposes limits on our ability to characterize the least abundant elements of these systems.
The effort to obtain a more complete understanding of the operations and behaviors of minor components of the microbiota is an area deserving of significant attention, given known examples of low-abundance taxa that play key roles within their larger communities and in host physiology [2],[45]. Developing such an understanding requires methods and assays that are collectively capable of assessing the structure and function of a microbiota at multiple levels of resolution. The need for high sensitivity and specificity in these approaches will become increasingly relevant as we transition towards experiments involving defined communities of even greater complexity, including bacterial culture collections prepared from the fecal microbiota of humans [46]. We anticipate that the study of sequenced culture collections transplanted to gnotobiotic mice will be instrumental in determining the degree to which physiologic or pathologic host phenotypes can be ascribed to the microbiota as well as specific constituent taxa.
The recent development of a low-error 16S ribosomal RNA amplicon sequencing method (LEA-Seq) and the application of this method to the fecal microbiota of 37 healthy adults followed for up to 5 years indicated that individuals in this cohort contained 195±48 bacterial strains representing 101±27 species [47]. Furthermore, stability follows a power-law function, suggesting that once acquired, most gut strains in a person are present for decades. New advances in the culturing of fastidious gut microbes may one day allow us to capture most (or all) of the taxonomic and functional diversity present within an individual's fecal microbiota as a clonally arrayed, sequenced culture collection, providing a perfectly representative and defined experimental model of their gut community. In the meantime, first-generation artificial communities of modest complexity such as the one described here offer a way of studying many questions related to the microbiota. However, the limited complexity and composition of our 12-species artificial community, and the way in which it was assembled in germ-free mice, make it an imperfect model of more complex human microbiota. Native microbial communities, for example, are subject to the influence of variables that are notably absent from this system, such as intraspecies genetic variability and exogenous microbial inputs. There are also taxa (e.g., Proteobacteria, Bifidobacteria) and microbial guilds (e.g., butyrate producers) typical of human gut communities that are absent from our defined assemblage that could be used to augment this system in order to improve our understanding of how their presence/absence influences a microbiota's response to diet and a spectrum of other variables of interest. These future attempts to systematically increase complexity should reveal what trends, patterns, and trajectories observed in artificial assemblages such as the one reported here map or do not map onto natural communities.
Finally, one of the greatest advantages of studying defined assemblages in mice is that they afford us the ability to interrogate the biology of key bacterial species in a focused manner. The artificial community we used in our experiments included B. cellulosilyticus WH2, a species that warrants further study as a model gut symbiont given its exceptional carbohydrate utilization capabilities, its apparent fitness advantage over many other previously characterized gut symbionts, and its genetic tractability. This genetic tractability should facilitate future experiments in which transposon mutant libraries are screened in vivo as one component of a larger artificial community in order to identify this strain's most important fitness determinants under a wide variety of dietary conditions. Identifying the genetic elements that allow B. cellulosilyticus to persist at the relatively high levels observed, regardless of diet, should provide microbiologists and synthetic biologists with new “standard biological parts” that will aid them in developing the next generation of prebiotics, probiotics, and synbiotics.
All experiments involving mice used protocols approved by the Washington University Animal Studies Committee in accordance with guidelines set forth by the American Veterinary Medical Association. Trained veterinarians from the Washington University Division of Comparative Medicine supervised all experiments. The laboratory animal program at Washington University is accredited by the Association for Assessment and Accreditation of Laboratory Animal Care International (AAALAC).
A strain of B. cellulosilyticus designated “WH2” (see Figure S1A,B) was isolated from a human fecal sample during an iteration of the Microbial Diversity Summer Course overseen by A. Salyers (University of Illinois, Urbana-Champaign) at the Marine Biological Laboratory (Woods Hole, MA). The genome of this isolate was sequenced using a combination of long-read and short-read technologies, yielding 51,819 plasmid and fosmid end reads (library insert sizes: 3.9, 4.9, 6.0, 8.0, and 40 kb; ABI 3730 platform), 333,883 unpaired 454 reads (FLX+ and XL+ chemistry), and 10 million unpaired Illumina reads (HiSeq; 42 nt read length). A hybrid assembly was constructed using MIRA v3.4.0 (method, de novo; type, genome; quality grade, accurate) with default settings [48],[49]. Gene calling was performed using the YACOP metatool [50]. Additionally, the four ribosomal RNA (rRNA) operons within the B. cellulosilyticus WH2 genome were sequenced individually to ensure high sequence accuracy in these difficult-to-assemble regions. Further details for the B. cellulosilyticus WH2 assembly are provided in Table S1.
Details regarding the 12 bacterial strains used in this study are provided in Table S4. Cells were grown in supplemented TYG (TYGS; [42]) at 37°C under anaerobic conditions in a Coy anaerobic chamber (atmosphere: 75% N2, 20% H2, 5% CO2). After reaching stationary phase, cells were pelleted by centrifugation and resuspended in TYGS medium supplemented with 20% glycerol. Individual aliquots containing 400–800 µL of each cell suspension were stored at −80°C in 1.8 mL borosilicate glass vials with aluminum crimp tops. The identity of each species was verified prior to its use in experiments by extracting DNA from a frozen aliquot of cells, amplifying the 16S rRNA gene by PCR using primers 8F/27F (AGAGTTTGATCCTGGCTCAG; [51]) and 1391R (GACGGGCGGTGWGTRCA; [52]), sequencing the entire amplicon with an ABI 3730 capillary sequencer (Retrogen, Inc.), and comparing the assembled 16S rRNA gene sequence to the known reference sequence.
Details regarding the construction of each inoculum are provided in Table S3. The inocula used to gavage germ-free mice in each experiment were prepared either directly from frozen stocks (experiment 1, E1) or from a combination of frozen stocks and overnight cultures (experiment 2, E2). The recoverable cell density for each batch of frozen stocks used in inoculum preparation was determined prior to pooling, while the same values for overnight cultures were calculated after pooling. To do so, an aliquot of cells from each overnight culture or set of frozen stocks was used to prepare a 10-fold dilution series in phosphate-buffered saline (PBS), and each dilution series was plated on brain-heart-infusion (BHI; BD Difco) agar supplemented with 10% (v/v) defibrinated horse blood (Colorado Serum Co.). Plates were grown for up to 3 d at 37°C under anaerobic conditions in a Coy chamber, colonies were counted, and the number of colony-forming units per milliliter (CFUs/mL) was calculated. The volume of each cell suspension included in the final inoculum was normalized by its known or estimated viable cell concentration in an effort to ensure that no species received an early advantage during establishment of the artificial community in the germ-free animals. Total CFUs per gavage were estimated at 8.0×107 and 4.2×108 for experiments E1 and E2, respectively.
Experiments were performed using protocols approved by the animal studies committee of the Washington University School of Medicine. For each experiment, two groups of 10–12-wk-old male germ-free C57BL/6J mice were maintained in flexible film gnotobiotic isolators under a strict 12 h light cycle, during which time they received sterilized food and water ad libitum. Animals were fasted for 4 h prior to gavage with 500 µL of a cell suspension inoculum containing the 12 sequenced, human gut-derived bacterial symbionts. After gavage, animals were maintained in separate cages throughout the course of the experiment. Fresh fecal pellets were periodically collected directly into screw-cap sample tubes that were immediately frozen in liquid nitrogen. At the time of sacrifice, the contents of each animal's cecum were divided into thirds and snap-frozen in liquid nitrogen for later use in DNA, RNA, and total protein isolations.
Animals were subjected to dietary oscillations comprising three consecutive phases of 2 wk each (see Figure S3). Prior to inoculation, germ-free mice were maintained on a standard autoclaved chow diet low in fat and rich in plant polysaccharides (LF/HPP, B&K rat and mouse autoclavable chow #73780000, Zeigler Bros, Inc). Three days prior to inoculation, one group of germ-free animals was switched to a sterile “Western”-like chow high in fat and simple sugars (HF/HS, Harlan Teklad TD96132), while the other continued to receive LF/HPP chow. After gavage, each group of animals was maintained on its respective diet for 2 wk, after which each treatment group was switched to the alternative diet. Two weeks later, the mice were switched back to their original starting diet and were retained on this diet until the time of sacrifice.
DNA and RNA were extracted from fecal pellets and cecal contents as previously described [11].
COPRO-Seq measurements of the proportional representation of all species present in each fecal/cecal sample analyzed were performed as previously described [11] using short-read (36 nt) data collected from an Illumina sequencer (data were generated using a combination of the Genome Analyzer I, Genome Analyzer II, and Genome Analyzer IIx platforms). After demultiplexing each barcoded pool, reads were trimmed to 25 bp and aligned to the reference genomes. An abundance threshold cutoff of 0.003% was set for determining an artificial community members' presence/absence, based on the proportion of reads from each experiment that were found to spuriously align to distractor reference genomes of bacterial species not included in this study. Normalized counts for each bacterial species in each sample were used to calculate a simple intrasample percentage. In order to make changes in abundance over time more easily comparable between species with significantly different relative abundances, these percentages were also in some cases normalized by the maximum abundance (%) observed for a given species across all time-points from a given animal. This transformation resulted in a value referred to as the percentage of maximum achieved (“PoMA”) that was used to evaluate which species were most/least responsive to dietary interventions.
COPRO-Seq proportional abundance data were subjected to ordination using scripts found in QIIME v1.5.0-dev [53]. Data from both E1 and E2 were combined to generate a single tab-delimited table conforming to QIIME's early (pre-v1.4.0-dev) OTU table format. This pseudo-OTU table was subsequently converted into a BIOM-formatted table object that was used as the input for beta_diversity.py to calculate the pairwise distances between all samples using a Hellinger metric. PCoA calculations were performed using principal_coordinates.py. These coordinates and sample metadata were passed to make_3d_plots.py to generate PCoA plots. Plots shown are visualized using v2.21 of the KiNG software package [54].
The ability of B. cellulosilyticus WH2 and B. caccae ATCC 43185 to grow on a panel of 47 simple and complex carbohydrates was evaluated using a phenotypic array whose composition has been previously described [25]. Growth measurements were collected in duplicate (two wells per substrate) over the course of 3 d at 37°C under anaerobic conditions. A total of three independent experiments were performed for each species tested (n = 6 growth profiles/substrate/species). Total growth (Atot) was calculated from each growth curve as the difference between the maximum and minimum optical densities (OD600) observed (i.e., Amax−Amin). Growth rates were calculated as total growth divided by time (Atot/(tmax−tmin)), where tmax and tmin correspond to the time-points at which Amax and Amin, respectively, were collected. Consolidated statistics from all six replicates for each of the 47 conditions tested for each species are provided in Table S11.
Whole genome transposon mutagenesis of B. cellulosilyticus WH2 was performed using protocols originally developed for B. thetaiotaomicron [42],[46], with some modifications. Initial attempts to transform B. cellulosilyticus WH2 with the pSAM_Bt construct reported by Goodman et al. yielded very low numbers of antibiotic-resistant clones, which we attributed to poor recognition of one or more promoters in the mutagenesis plasmid. Replacement of the promoter driving expression of the transposon's erythromycin resistance gene (ermG) with the promoter for the gene encoding EF-Tu in B. cellulosilyticus WH2 (BWH2_3183) dramatically improved the number of resistant clones recovered after transformation. The resulting library consisted of 93,458 distinct isogenic mutants, with each mutant strain containing a single randomly inserted modified mariner transposon. Of all predicted ORFs, 91.5% had insertions covering the first 80% of each gene (mean, 13.9 distinct insertion mutants per ORF).
At 11 wk of age, male germ-free C57BL/6J mice (individually caged) were fed either a diet low in fat and rich in plant polysaccharides (LF/HPP) or high in fat and simple sugars (HF/HS). After a week on their experimental diet, animals received a single gavage containing the B. cellulosilyticus WH2 transposon library and 14 other species of bacteria (i.e., this artificial community consisted of the 12 species listed in Figure 1A, plus B. thetaiotaomicron 7330, E. rectale ATCC 33656, and Clostridium symbiosum ATCC 14940). After 16 d, fecal pellets were collected, and total fecal DNA was extracted.
500 ng of each fecal DNA extraction was diluted in 15 µL of TE buffer and digested with MmeI (4 U, New England Biolabs) in a 20 µL reaction supplemented with 10 pmoles of 12 bp DNA containing an MmeI restriction site (to improve the efficiency of restriction enzyme digestion) [42]. The reaction was incubated for 1 h at 37°C and then terminated (80°C for 20 min). MmeI-digested DNA was subsequently purified using 125 µL of AMPure beads (after washing the beads once with 100 µL of sizing solution (1.2 M NaCl and 8.4% PEG 8000)). The digested DNA was added to the beads and the solution incubated at room temperature for 5 min. Beads were pelleted with a magnetic particle collector (MPC), washed twice (each time using a mixture composed of 20 µL TE buffer (pH 7.0) and 100 µL sizing solution, with bead recovery via MPC after each wash), followed by two ethanol washes (180 µL 70% ethanol/wash) and air-drying for 10 min. Samples were resuspended in 18 µL TE buffer (pH 7.0), and DNA was removed after pelleting beads with the MPC. Ligation of adapters was performed in a 20 µL reaction that contained 16 µL of purified DNA, 1 µL of T4 Ligase (2000 U/µL; NEB), 2 µL 10× ligase buffer, and 10 pmol of barcoded adapter (incubation for 1 h at 16°C). Ligations were subsequently diluted with TE buffer (pH 7.0) to a final volume of 50 µL, mixed with 60 µL of AMPure beads, and incubated at room temperature for 5 min. Beads with bound DNA were pelleted using the MPC and washed twice with 70% ethanol as above. After allowing the ethanol to evaporate for 10 min, 35 µL of nuclease-free water was added, and the mixture was incubated at room temperature for 2 min before collecting the beads with the MPC. Enrichment PCR was performed in a final volume of 50 µL using 32 µL of the cleaned up sample DNA, 10 µL 10× Pfx amplification buffer (Invitrogen), 2 µL 10 mM dNTPs, 0.5 µL 50 mM MgSO4, 2 µL of 5 µM amplification primers (forward primer: 5′CAAGCAGAAGACGGCATACG3′, reverse primer: 5′AATGATACGGCGACCACCGAACACTCTTTCCCTACACGA3′), and 1.5 µL Pfx polymerase (2.5 U/µL; Invitrogen) (cycling conditions: denaturation at 94°C for 15 s; annealing at 65°C for 1 min; extension at 68°C for 30 s; total of 22 cycles). The 134 bp PCR product from each reaction was purified (4% MetaPhor gel; MinElute Gel Extraction Kit (Qiagen)) in a final volume of 20 µL and was quantified (Qubit, dsDNA HS Assay Kit; Invitrogen). Reaction products were then combined in equimolar amounts into a pool that was subsequently adjusted to 10 nM and sequenced (Illumina HiSeq 2000 instrument).
All short read Illumina data used for COPRO-Seq and RNA-Seq analyses, GeneChip data, and genome sequencing/assembly data are available through GEO SuperSeries GSE48537 and NCBI BioProject ID PRJNA183545. The draft genome assembly for B. cellulosilyticus WH2 has been deposited at DDBJ/EMBL/GenBank under accession number ATFI00000000. Raw MS data are available from the Dryad Digital Repository: http://dx.doi.org/10.5061/dryad.7fj1k.
|
10.1371/journal.pntd.0002242 | Functional Transcriptomics of Wild-Caught Lutzomyia intermedia Salivary Glands: Identification of a Protective Salivary Protein against Leishmania braziliensis Infection | Leishmania parasites are transmitted in the presence of sand fly saliva. Together with the parasite, the sand fly injects salivary components that change the environment at the feeding site. Mice immunized with Phlebotomus papatasi salivary gland (SG) homogenate are protected against Leishmania major infection, while immunity to Lutzomyia intermedia SG homogenate exacerbated experimental Leishmania braziliensis infection. In humans, antibodies to Lu. intermedia saliva are associated with risk of acquiring L. braziliensis infection. Despite these important findings, there is no information regarding the repertoire of Lu. intermedia salivary proteins.
A cDNA library from the Salivary Glands (SGs) of wild-caught Lu. intermedia was constructed, sequenced, and complemented by a proteomic approach based on 1D SDS PAGE and mass/mass spectrometry to validate the transcripts present in this cDNA library. We identified the most abundant transcripts and proteins reported in other sand fly species as well as novel proteins such as neurotoxin-like proteins, peptides with ML domain, and three small peptides found so far only in this sand fly species. DNA plasmids coding for ten selected transcripts were constructed and used to immunize BALB/c mice to study their immunogenicity. Plasmid Linb-11—coding for a 4.5-kDa protein—induced a cellular immune response and conferred protection against L. braziliensis infection. This protection correlated with a decreased parasite load and an increased frequency of IFN-γ-producing cells.
We identified the most abundant and novel proteins present in the SGs of Lu. intermedia, a vector of cutaneous leishmaniasis in the Americas. We also show for the first time that immunity to a single salivary protein from Lu. intermedia can protect against cutaneous leishmaniasis caused by L. braziliensis.
| Sand fly saliva contains potent, biologically active proteins that allow the insect to stop host responses to acquire a blood meal. After repeated exposures, a number of these salivary proteins also induce a response in the host such as antibody production and/or cellular-mediated immunity. In animal models, these immune responses affect Leishmania infection. On one hand, immunity to Phlebotomus papatasi saliva protected animals against cutaneous leishmaniasis, while on the other hand, immunity to Lutzomyia intermedia saliva did not protect but exacerbated this disease. These differences are probably due to the types of proteins present in the saliva of these different sand fly species. The present work focused on isolation and identification of the secreted proteins present in the salivary glands of Lu. intermedia, an important vector of L. braziliensis, the agent of mucocutaneous leishmaniasis. Saliva from this sand fly contains a number of proteins not present in P. papatasi saliva and, with some exceptions; proteins that are homologous between the two species are very divergent. Furthermore, we identified one protein that, after vaccination, induced a cellular immune response able to protect mice against Leishmania braziliensis infection. This is the first evidence that a single salivary protein from Lu. intermedia can protect mice against this cutaneous leishmaniasis.
| Protozoan parasites of the genus Leishmania cause a broad spectrum of diseases, collectively known as leishmaniasis, that occur predominantly in tropical and subtropical regions. The sand fly vector delivers the Leishmania parasite while acquiring a blood meal, and during this process, the sand fly injects saliva into the host's skin. Salivary proteins have pharmacologic activities that assist in acquisition of a blood meal [1] and, in parallel, these proteins also modulate the function of cells of the immune system [2], [3], [4], [5]. Mice are protected when immunized with bites from Phlebotomus papatasi [6] or with plasmid DNA encoding salivary proteins from P. papatasi [7] or from Lutzomyia longipalpis [8] suggesting that salivary molecules can be envisaged as components of a vaccine against leishmaniasis [9].
Because the composition of salivary molecules varies among distinct sand fly species, it is important to investigate whether the concept of vector-based vaccines can be extended to other Leishmania species such as L. braziliensis. Of note, American Cutaneous Leishmaniasis, caused by L. braziliensis, is distinguished from other leishmaniases by its chronicity, latency and tendency to metastasize in the human host leading to muco-cutaneous leishmaniasis [10]. Surprisingly, immunization with Lutzomyia intermedia SGH did not protect mice against L. braziliensis infection [11]. An association between the presence of antibodies to Lu. intermedia salivary proteins and active disease was reported, suggesting that a humoral response to Lu. intermedia SGH may favor L. braziliensis infection [11].
Although the salivary gland (SG) transcriptomes of various sand fly species, including Lu. longipalpis [12], have been well documented, information regarding the repertoire of Lu. intermedia salivary molecules is lacking. The outcome of Leishmania infection in mice immunized with Lu. intermedia SGH (disease) [11] compared to P. papatasi SGH (protection) [13] is distinct. We then hypothesized that such discrepancies could be due to difference in the repertoire of salivary proteins or the difference in the sequences of their salivary proteins. We took the opportunity to characterize the transcriptome from the salivary glands (SGs) of Lu. intermedia, the main vector of L. braziliensis in Brazil. We also examined the immunogenic properties of a group of salivary proteins and identified one component that inhibited the development of cutaneous leishmaniasis caused by L. braziliensis in mice.
Adult Lu. intermedia sand flies were captured in Corte de Pedra, Bahia. Sand flies were morphologically identified according to the identification key proposed by Young and Duncan. SGs were dissected and stored in groups of 20 pairs in 20 µl NaCl (150 mM)-Hepes buffer (10 mM; pH7.4) at −70°C. Immediately before use, SGs were disrupted by ultrasonication in 1.5-ml conical tubes. Tubes were centrifuged at 10,000×g for two minutes, and the resultant supernatant—SGH—was used for the studies. The level of lipopolysaccharide (LPS) contamination of SGH preparations was determined using a commercially available LAL chromogenic kit (QCL-1000; Lonza Biologics, Portsmouth, NH, USA); LPS concentration was <0.1 ng/ml.
Lu. intermedia SG mRNA was isolated from 50 SG pairs using the Micro-FastTrack mRNA isolation kit (Invitrogen, San Diego, CA, USA). The PCR-based cDNA library was made following the instructions for the SMART cDNA library construction kit (BD-Clontech, Mountain View, CA, USA) with some modifications [14]. The obtained cDNA libraries (large, medium, and small sizes) were plated by infecting log phase XL1-blue cells (Clontech, Palo Alto, CA, USA), and the number of recombinants was determined by PCR using vector primers flanking the inserted cDNA and visualized on a 1.1% agarose gel with ethidium bromide (1.5 µg/ml).
Lu. intermedia SG cDNA libraries were plated to approximately 200 plaques per plate (150-mm petri dish). The plaques were randomly picked and transferred to a 96-well polypropylene plate (Novagen, Madison, WI, USA) containing 75 µl of water per well. Four microliters of the phage sample were used as a template for a PCR reaction to amplify random cDNAs. The primers used for this reaction were sequences from the triplEX2 vector. PT2F1 (5′-AAG TAC TCT AGC AAT TGT GAG C-3′) is positioned upstream of the cDNA of interest (5′- end), and PT2R1 (5′-CTC TTC GCT ATT ACG CCA GCT G-3′) is positioned downstream of the cDNA of interest (3′ end). Platinum Taq polymerase (Invitrogen) was used for these reactions. Amplification conditions were 1 hold of 75°C for 3 minutes, 1 hold of 94°C for 2 minutes, and 30 cycles of 94°C for one minute, 49°C for one minute, and 72°C for one minute 20 seconds. Amplified products were visualized on a 1.1% agarose gel with ethidium bromide. PCR products were cleaned using the PCR multiscreen filtration system (Millipore, Billerica, MA, USA). Three microliters of the cleaned PCR product were used as a template for a cycle-sequencing reaction using the DTCS labeling kit from Beckman Coulter (Fullerton, CA, USA). The primer used for sequencing, PT2F3 (5′-TCT CGG GAA GCG CGC CAT TGT-3′) is upstream of the inserted cDNA and downstream of the primer PT2F1. Sequencing reaction was performed on a 9700 Thermacycler (Perkin-Elmer, Foster City, CA, USA). Conditions were 75°C for two minutes, 94°C for two minutes, and 30 cycles of 96°C for 20 seconds, 50°C for 10 seconds, and 60°C for four minutes. After cycle sequencing the samples, a cleaning step was done using Excel Pure 96-well UF PCR purification plates (EdgeBiosystems, Gaithersburg, MD, USA). Fluorescently labeled extension products were purified following Applied Biosystems BigDye XTerminator purification protocol and then processed on an ABI 3730xL DNA analyzer (Applied Biosystems, Inc., Foster City, CA).
Bioinformatics analysis was performed as previously described and raw sequence files were analyzed using a customized program [15]. DNA sequences with Phred quality scores lower than 20, including primer and vector sequences, were discarded. Sequences were then grouped into clusters using a customized program based on identity (95% identity) and aligned into contiguous sequences (contigs) using the CAP3 program [16]. Contigs were then analyzed by blastx, blastn, or rpsblast programs and compared to the non-redundant (NR) protein database of the National Center for Biotechnology Information (NCBI), the gene ontology (GO) FASTA subset, and the conserved domains database (CDD) of NCBI, which contains KOG, protein families (Pfam), and simple modular architecture research tool (SMART) databases. The three potential translations of each dataset were submitted to the SignalP server to detect signal peptides. All the analyzed sequences were combined in an Excel spreadsheet and manually verified and annotated. Sequences were aligned using ClustalW (version 1.4) [17]. For Phylogenetic analysis, statistical neighbour-joining (NJ) bootstrap tests of the phylogenies were done with the Mega package [18].
Lu. intermedia SGH (equivalent to 60 SG pairs) were run on NuPAGE (4–12%), 1 mm thick (Invitrogen) according to manufacturer's instructions. Proteins were visualized by staining with SimplyBlue (Invitrogen). The gel was sliced into 30 individual sections that were de-stained and digested overnight with trypsin at 37°C. Identification of gel-separated proteins was performed on reduced and alkylated trypsin digested samples prepared by standard mass spectrometry protocols as previously described [19] and performed by the Laboratory of Proteomics and Analytical Technologies (NCI-Frederick, Frederick, MD, USA).
Female BALB/c mice, 6–8 weeks of age, were obtained from CPqGM/FIOCRUZ animal facility where they were maintained under pathogen-free conditions. All animal work was conducted according to the Guidelines for Animal Experimentation of the Colégio Brasileiro de Experimentação Animal and of the Conselho Nacional de Controle de Experimentação Animal. The local Ethics Committee on Animal Care and Utilization (CEUA) approved all procedures involving animals (CEUA-L06508-CPqGM/FIOCRUZ).
Ten plasmids, Linb-1 (SP13 protein family), Linb-2 (SP13 family of proteins), Linb-7 (SP15-like protein), Linb-8 (SP15-like protein), Linb-11 (SP13 protein family), Linb-15 (C-type lectin family of proteins), Linb-19 (9.6-kDa protein), Linb-22 (C-type lectin family of proteins), Linb-24 (10-kDa protein), and Linb-28 (SP15-like protein)] encoding Lu. intermedia salivary gland-secreted proteins were cloned into VR2001-TOPO vector and purified as previously described [20]. To evaluate the immunogenic potential of proteins present in Lu. intermedia saliva, BALB/c mice were immunized intradermally in the right ear three times at two-week intervals with 10 µg of control DNA plasmid or DNA plasmids (recombinant)coding for salivary proteins in 10 µl of sterile water. For generation of immune sera, mice were exposed directly to the bites of Lu. intermedia sand flies. In this case, before each sand-fly exposure, female sand flies were left overnight without sugar or water and were used the following day. Ten healthy flies were placed in plastic vials, the upper surfaces of which were covered with a fine netting. Mice were anesthetized and a single ear from mice was pressed closely to the meshed surface of vials containing flies, secured by clamps designed for this purpose. Flies were allowed to feed in the dark for a period of 30 minutes. A minimum of five fully blood-fed flies per ear was required for each sensitization. After three exposures, with a two-week interval between each exposure, mice were euthanized for collection of immune sera.
ELISA microplates were coated overnight at 4°C with 50 µl SGH diluted to five pairs of SGs/ml in coating buffer (NaHCO3 0.45 M, Na2CO3 0.02 M, pH 9.6). After washing with PBS-Tween, wells were blocked with PBS-Tween plus 5% dried skim milk for one hour at 37°C. Wells were incubated overnight with sera from mice immunized with control or recombinant plasmids obtained two weeks after the last immunization, diluted (1∶50) in PBS-Tween. After further washings, wells were incubated with alkaline phosphatase-conjugated anti-mouse IgG antibody (Promega, Madison, WI, USA) diluted (1∶5000) in PBS-Tween for one hour at 37°C. Following another washing cycle, wells were developed with p-nitrophenylphosphate in sodium carbonate buffer pH9.6 with 1 mg/ml of MgCl2. Absorbance was recorded at 405 nm.
Following three intradermal inoculations with control (wild type) or with recombinant DNA plasmids (Linb-11 or Linb-7) in the right ear dermis, mice were inoculated with Lu. intermedia SGH (equivalent to 1 pair of SGs) in the left ear dermis. Twenty-four and forty-eight hours later, challenged ears were removed and fixed in 10% formaldehyde. Following fixation, tissues were processed, embedded in paraffin, and 5-µm sections were stained with hematoxylin and eosin (H & E) and analyzed by light microscopy. For morphometric analyses, inflammatory cells were counted in three fields/section using a 2005 magnification, covering a total area of 710 µm2.
Two weeks following the last immunization with control or with recombinant DNA plasmid (Linb-11) in the right ear dermis, mice were challenged in the left ear dermis by inoculation of stationary-phase promastigotes (105 parasites in 10 ul of saline) + SGH (equivalent to 1 pair of SGs). Lesion size was monitored weekly using a digital caliper (Thomas Scientific, Swedesboro, NJ, USA). L. braziliensis promastigotes (strain MHOM/BR/01/BA788) [21] were grown in Schneider medium (Sigma, St. Louis, MO, USA) supplemented with 100 U/ml of penicillin, 100 µg/ml of streptomycin, and 10% heat-inactivated fetal calf serum (all from Invitrogen).
Parasite load was determined using a quantitative limiting dilution assay and analyzed by the ELIDA program [22]. Briefly, infected ears and retromaxillar draining lymph nodes (dLNs) were aseptically excised at two and eight weeks post infection and homogenized in Schneider medium (Sigma). The homogenates were serially diluted in Schneider medium supplemented as before and seeded into 96-well plates containing biphasic blood agar (Novy-Nicolle-McNeal) medium. The number of viable parasites was determined from the highest dilution at which promastigotes could be grown out after up to two weeks of incubation at 25°C.
Reagents for staining cell surface markers and intracellular cytokines were purchased from BD Biosciences (San Diego, CA, USA). Measurement of in vitro cytokine production was performed as described elsewhere [21]. dLNs were aseptically excised at two and eight weeks post infection and homogenized in RPMI medium. Cells were resuspended in RPMI supplemented with 2 mM L-glutamine, 100 U/ml of penicillin, 100 µg/ml of streptomycin, 10% fetal calf serum (all from Invitrogen), and 0.05 M 2-mercaptoethanol. Cells were restimulated in the presence of anti-CD3 (10 µg/ml) and anti-CD28 (10 µg/ml) and were later incubated with Brefeldin A (Sigma) (10 µg/ml). Cells were blocked with anti-Fc receptor antibody (2.4G2) and stained with anti-mouse surface CD4 (L3T4) conjugated to FITC and Cy-Chrome. For intracellular staining of cytokines, cells were permeabilized using Cytofix/Cytoperm (BD Biosciences) and incubated with the anti-cytokine antibodies conjugated to PE:IFN-γ (XMG1.2), IL-4 (BVD4-1D11), and IL-10 (JES5-16E3). The isotype controls used were rat IgG2b (A95-1) and rat IgG2a (R35-95). Data were collected and analyzed using CELLQuest software and a FACSort flow cytometer (Becton Dickinson Immunocytometry System; Becton Dickinson and Company, Sunnyvale, CA, USA). The steady-state frequencies of cytokine-positive cells were determined using LN cells from PBS-inoculated mice.
Data are presented as means ± standard error of the mean. The significance of the results was determined by Kruskal-Wallis tests using Prism (Graph Pad Software, Inc., San Diego, CA, USA), and P values<0.05 were considered significant. To evaluate disease burden in mice, ear thickness of mice immunized with control or recombinant plasmids was recorded weekly for each individual mouse. The course of disease for experimental and control mice was plotted individually, and the area under each resulting curve was calculated using Prism (Graph Pad Software). The significance of the results was calculated by Kruskal-Wallis test.
Assembly of 1,395 high-quality transcript sequences from the cDNA library of Lu. intermedia SGs led to the identification of 278 contigs including 193 singletons. Annotation of these contigs—based on several database comparisons—indicated that 76% of the transcripts belong to the putative secreted (S) class, 9% to the housekeeping class (H), and 15% to the unknown (U) class (Table 1). The unknown class may derive from the 5′- incomplete mRNAs in the library or transcripts coding for novel proteins. Notably, the S class had on average 17 expressed sequence tags (ESTs) per contig, while the H and U classes had only 1.46 and 1.57 ESTs/contig, respectively, indicating high expression levels of secreted products in this cDNA library (Table 1). Transcripts coding for proteins associated with synthesis machinery, as expected, were the most abundant in the H class (Supplementary Table S1).
Inspection of S class contigs, deriving from 1,064 ESTs, identified the enzyme apyrase, 5′-nucleotidase, endonuclease, adenosine deaminase, hyaluroniadase, and glucosidase, all of these previously identified in other sand fly transcriptomes [1], [23], [24], [25] [12], [14], [26], [27], [28], [29](Table 2). Transcripts coding for proteins of ubiquitous distribution include members of the C-type lectin and Antigen 5 families. Insect-specific protein families are represented by the families of yellow proteins, D7 proteins, and SP15 proteins. Sand fly-specific families are also represented, including members of the SP13 family of proteins, anti-FactorXa protein (lufaxin), 10-kDa family, 30-kDa family, and 37–46-kDa family (these names were given in the review article [1]. One salivary protein present in Lu. longipalpis was deorphanized (is now referred as the 14.2 kDa salivary protein), and three Lu. intermedia orphan peptides were identified. Novel protein families—including a highly expressed family of small peptides accounting for nearly 50% of all ESTs—are part of the novelty of the salivary transcriptome of Lu. intermedia (Table 2). Additional analysis of these sequences and their clusterization by different degrees of similarity allowed further identification of divergent or novel protein families, some of which are described below in more detail.
We then analyzed the electrophoretic separation of Lu. intermedia salivary proteins (Figure 9A), followed by tryptic digestion of selected gel fractions, separation of peptides by reverse-phase HPLC, and subsequent mass spectrometry (Figure 9B). Together with the compiled database of coding sequences, we identified the proteins expressed in the SGs of Lu. intermedia. All the fractions displaying a signal from the mass spectrometer matched at least one transcript present in the Lu. intermedia cDNA library (Figure 9B). Accordingly, the enzyme apyrase, 5′- nucleotidase, endonuclease, adenosine deaminase, and hyaluronidase were identified at or near the predicted gel migration regions. The proteins Antigen-5, C-type lectin, D7 classical, short, and SP15 were also identified, as were members of the 33- and 30-kDa families of phlebotomines. The deorphanized Lutzomyia family member was also identified (Figure 9), as were three members of the putative orphan secreted proteins (not shown on Figure 9). We did not identify the largely expressed SP13 family of short peptides, as they may have migrated out of the gel.
To identify the immunogenic properties of a group of Lu. intermedia salivary proteins, mice were selected randomly and immunized intradermally in the ear with DNA plasmids coding for ten different transcripts identified in this cDNA library: Linb-1 (SP13 protein family), Linb-2 (SP13 family of proteins), Linb-7 (SP15-like protein), Linb-8 (SP15-like protein), Linb-11 (SP13 protein family), Linb-15 (C-type lectin family of proteins), Linb-19 (9.6-kDa protein), Linb-22 (C-type lectin family of proteins), Linb-24 (10-kDa protein), and Linb-28 (SP15-like protein). All recombinant DNA plasmids induced a significant humoral immune response against Lu. intermedia SGH when compared with sera obtained from mice immunized with control plasmid or naïve mice (Figure 10A). An exception was the DNA plasmid coding for Linb-11, which induced a low humoral response (Figure 10A). As expected, mice exposed to bites of Lu. intermedia sand flies developed a potent humoral response to the salivary proteins of this sand fly (Figure 10A).
A cellular immune response to salivary proteins is associated with protection in animal models of cutaneous leishmaniasis [6], [7]. Therefore, we examined whether Linb-11—a weak inducer of antibody response—could generate a cellular immune response in BALB/c mice. We also tested the response generated by Linb-7, a strong inducer of humoral response in BALB/c mice (Figure 10A). Morphometric analysis of the challenged ears showed a significant increase in cellular recruitment at 24 hours induced by Linb-11 and by Linb-7 (Figure 10B, top panel). Examination of ear sections confirmed this result. At 48 hours, the number of inflammatory cells recruited by Lu. intermedia SGH inoculation was significantly lower in Linb-11- compared with Linb-7-immunized mice (Figure 10B), suggesting that Linb-7 leads to a sustained cellular recruitment (Figure 10B).
Based on the finding that Linb-11 induces a low humoral immune response and a controlled cellular immune response, we tested whether this protein could protect mice against L. braziliensis infection. Linb-11-immunized mice challenged with L. braziliensis plus Lu. intermedia SGH had significantly smaller lesions (measured by ear thickness) when compared with control mice (Figure 11A). Disease burden, calculated as the area under the curves obtained from Figure 11A (as described in Materials and Methods), was significantly lower following immunization with Linb-11 (Figure 11B). Two weeks post infection, parasite load at the ear (Figure 11C) or in dLN (Figure 11D) were similar in Linb-11-immunized mice versus control mice; however, at eight weeks post challenge, we detected a significant reduction in parasite load in the ear (Figure 11C) and in dLN (Figure 11D) of Linb-11-immunized mice compared with control mice. This decrease in parasite load corroborated the lower ear thickness observed in Linb-11-immunized mice at this same this time point (Figure 11A).
Evaluation of the frequency of cytokine-secreting cells, at two and eight weeks post challenge with L. braziliensis plus Lu. intermedia SGH, indicated the presence of higher percentage of CD4+ IFN-γ+ T cells in Linb-11-immunized mice (Figure 12A). At this same time point, the percentage of CD4+IL-4+ (Figure 12B) or CD4+IL-10+ T cells was similar in immunized mice vs. controls (Figure 12C). At eight weeks post infection, the percentages of CD4+ IFN-γ+ and of CD4+ IL-4+ T cells were also similar (Figure 12A), whereas the percentage of CD4+ IL-10+ T cells was significantly lower in Linb-11-immunized mice. We may suggest that in mice immunized with Linb-11, the early (two weeks) predominance of CD4+IFN-γ+ cells (Figure 12A) results in a better control of lesion development (Figure 11A) and in parasite killing (Figure 11C–D).
Differently from the outcomes observed upon immunization with P. papatasi [13] or Lu. longipalpis [9] salivary protein, responses to Lu. intermedia salivary proteins did not result in protection against L. braziliensis infection [11]. We hypothesized that the composition of Lu. intermedia salivary proteins could explain these distinct outcomes. In the present work, we characterized the salivary transcripts and proteins present in Lu. intermedia SGs and we revealed the presence of i) novel sequences of proteins not found in Lu. longipalpis or other sand-fly species, including the toxin-like family of proteins, the ML domain containing proteins, and three families of small novel peptides; ii) a high degree of divergence between common family of proteins found in Lu. intermedia and Lu. longipalpis, including maxadilan, the SP13 family of proteins, the 10-kDa family of proteins, and the yellow-related proteins ; and iii) an expansion in the SP13, the 9.6-kDa-, the SL1, the C-type, and the ML protein families. Notoriously, the maxadilan homolog in Lu. intermedia is very divergent (only 34% identity to maxadilan from Lu. longipalpis) and displays low abundance when compared with Lu. longipalpis maxadilan. Interestingly, Lu. intermedia and Lutzomyia ayacuchensis, which lacks salivary maxadilan [27], are vectors of cutaneous leishmaniasis, while Lu. longipalpis, the vector for visceral leishmaniasis, has large amounts of maxadilan in the SGs. Moreover, the amount of maxadilan affects the outcome of infection as has been already shown by Warburg et al. [45].
Regarding the Lu. intermedia SG transcriptome, the gene expansion of protein families and the sequence divergence of these proteins, compared with homologs found in Lu. longipalpis, may explain the distinct immune responses generated by exposure to these two species. Importantly, the divergence is greater when comparing any molecule equal to or smaller than 15 kDa found in the SGs of Lu. longipalpis and Lu. intermedia. It may be possible that one or more of these divergent molecules in Lu. intermedia may generate an immune response that obscures or overrides the immune response of a putative protective protein in the SGs of Lu. intermedia.
We therefore tested whether individual proteins from Lu. intermedia could produce a protective immune response different from the one observed when using the whole SGH [11]. We tested ten available DNA plasmids coding for individual Lu. intermedia salivary proteins in BALB/c mice and identified one protein, Linb-11, that did not induce an antibody response, produced a transient cellular immune response and protected mice against L. braziliensis infection. We then hypothesized that this type of response might modulate disease development, following a live parasite challenge. Indeed, immunization with Linb-11 immunization resulted in lower parasite numbers. This outcome correlated with an early predominance of IFN-γ-secreting cells over IL-4+CD4+ and IL-10+CD4+ T cells. These data suggest that CD4+IFN-γ+ cells may have migrated to the lesion site, leading to macrophage activation, parasite killing and, in turn, smaller lesions. Additionally, the lower percentage of CD4+IL-10+ cells in Linb-11-immunized mice could also have contributed to the control in lesion development, exerting immunosuppressive functions [46]. We can speculate that the controlled immune response—paralleled by the lack of antibodies—could generate the proper environment to control L. braziliensis infection.
It remains to be investigated whether other proteins, such as those initially screened (Figure 10), may display immunomodulatory features similar to Linb-11 or whether other salivary proteins identified in Lu. intermedia but not yet tested—or even a combination of proteins—would improve the results obtained. Of note, immunization with plasmids coding for Linb-1 and Linb-2, members of the SP13 protein family, induced a strong humoral response, different from that of Linb-11. This outcome could be explained by the diversity in amino acid sequence of these three molecules. It is also important that candidate molecules, salivary or Leishmania-derived, are further tested in the context of natural transmission by infected sand flies. In this stringent scenario, some vaccine candidates have proven effective while others have failed to confer protection [9], [47].
In previous experiments using whole Lu. intermedia SGH, we detected a high IL-4:IFN-γ ratio, and this correlated with lack of protection and more severe Lu. braziliensis infection [11]. Animals immunized with Lu intermedia SGH and challenged with L. braziliensis plus SGH showed a significant decrease in CXCL10 expression paralleled by an increase in IL-10 expression [5]. We suggest that when using whole SGH, at least in the L. braziliensis experimental model, this salivary mixture overrides priming of the protective immune response induced by Linb-11. In this sense, Oliveira et al. [13] identified in the SGs of P. papatasi one molecule, PpSP15, that conferred protection against L. major infection, while another protein PpSP44 exacerbated disease. The current results also suggests that Lu. intermedia salivary proteins that induce a strong humoral immune response, such as Linb-7, may—to the contrary—exacerbate disease, as seen upon immunization with Lu. intermedia SGH [11]. This remains to be tested.
Linb-11 is a small molecule of 4.5 kDa so far found only in sand flies. This molecule belongs to the SP13 family of proteins found in other sand-fly species. Lu. intermedia displays an expansion of this protein family by producing six members, as reported in this cDNA library. Nonetheless, Linb-11 induced a mild but protective immune response in an experimental model of infection in the absence of an adjuvant (other than the CpG present in the DNA plasmid backbone). This may suggest that a protein that generates a more controlled immune response may be adequate to limit cutaneous leishmaniasis caused by Lu. braziliensis.
|
10.1371/journal.ppat.1002763 | Conditional Stat1 Ablation Reveals the Importance of Interferon Signaling for Immunity to Listeria monocytogenes Infection | Signal transducer and activator of transcription 1 (Stat1) is a key player in responses to interferons (IFN). Mutations of Stat1 cause severe immune deficiencies in humans and mice. Here we investigate the importance of Stat1 signaling for the innate and secondary immune response to the intracellular bacterial pathogen Listeria monocytogenes (Lm). Cell type-restricted ablation of the Stat1 gene in naïve animals revealed unique roles in three cell types: macrophage Stat1 signaling protected against lethal Lm infection, whereas Stat1 ablation in dendritic cells (DC) did not affect survival. T lymphocyte Stat1 reduced survival. Type I IFN (IFN-I) signaling in T lymphocytes reportedly weakens innate resistance to Lm. Surprisingly, the effect of Stat1 signaling was much more pronounced, indicating a contribution of Stat1 to pathways other than the IFN-I pathway. In stark contrast, Stat1 activity in both DC and T cells contributed positively to secondary immune responses against Lm in immunized animals, while macrophage Stat1 was dispensable. Our findings provide the first genetic evidence that Stat1 signaling in different cell types produces antagonistic effects on innate protection against Lm that are obscured in mice with complete Stat1 deficiency. They further demonstrate a drastic change in the cell type-dependent Stat1 requirement for memory responses to Lm infection.
| Signal transducer and activator of transcription 1 (Stat1) is an indispensable component of the cellular response to interferons (IFN) during immune reactions to pathogens. Stat1 deficiency leads to severe immune defects in humans and mice. The sensitivity of animals with complete Stat1 ablation to microbial pathogens prevented determining its contribution to various effector systems of the immune response. By way of tissue-restricted Stat1 ablation we now decipher the impact of Stat1 signaling in different cell populations on the innate and adaptive immune response to the intracellular pathogen Listeria monocytogenes. Our data highlight the importance of and requirement for IFNγ-activated macrophages for clearance of the pathogen during early phases of infection, and show a yet unanticipated detrimental role for T cell Stat1. During secondary responses the picture changes and Stat1 in T cells is crucial for proper clearance of L. monocytogenes. Likewise, Stat1 signaling in dendritic cells plays a fundamental role for adaptive immunity to L. monocytogenes. Exploring the local response to L. monocytogenes infection we reveal a role of Stat1 in shaping the cellular composition of inflammatory infiltrates. Furthermore, Stat1 deficiency in dendritic cells increases the proliferation of regulatory T cells, an effect likely to dampen the antibacterial response.
| Signal transducer and activator of transcription (Stat1) is a central mediator of interferon responses in the immune system. Signals from type I (IFNα/IFNβ; IFN-I), type II (IFNγ; IFN-II) and type III (IFNλ, IFN-III) interferons employ receptor-associated Janus kinases (Jaks) to activate Stats by tyrosine phosphorylation [1], [2]. Gene transcription is induced and leads to a range of cellular changes, including anti-viral properties, growth inhibition, apoptosis and differentiation. Depending on the cellular context Stat1 can act as either a tumour-suppressor or promote oncogenesis [3], [4], [5]. The central character of Stat1 in signal transduction by the IFN receptors results from the importance of Stat1 homodimers for transcriptional regulation by IFNγ. Moreover, Stat1 forms the ISGF3 complex together with Stat2 and interferon regulatory factor 9 (Irf9). ISGF3 is the main player in transcriptional responses to both IFN-I and IFN-III. Consistent with its central role, Stat1 deficiency in mice recapitulates the lack of IFN-I, IFN-III and IFNγ responses and leads to high susceptibility to viral and bacterial infections [6], [7], [8]. The critical importance of Stat1 for resistance to infection is emphasized by mutations of the Stat1 gene in humans. Patients with various degrees of Stat1 loss-of-function present clinically with recurrent and often lethal mycobacterial and viral infections [9], [10], [11].
Listeria monocytogenes (Lm) is the causative agent of human listeriosis and a serious threat for the health of immunocompromised individuals. It is also a well-studied model organism to analyse cell-mediated immunity to intracellular pathogens. Innate protection critically depends on the activities of the cytokines interleukin (IL) 12 and IFNγ [12], [13]. This most likely reflects NK cell activation, IFNγ production and subsequent clearance of the bacteria by activated macrophages. Sterile immunity and immunological memory result from the development of CD8+ T cells [14], [15]. Stat1-deficient mice succumb to Lm during the early, innate phase of infection, strongly suggesting a dominant role for Stat1 in IFNγ-mediated macrophage activation [7]. As even very low numbers of Lm, even if attenuated, rapidly kill Stat1−/− mice it is difficult to study attributes of the innate response. For example, Lm replicates in a variety of non-hematopoietic cell types such as epithelial cells or hepatocytes and the contribution of Stat1 to bacterial clearance in these cell types is not known. Moreover, the impact of Stat1 on the generation of adaptive immunity and immunological memory is unclear [16], [17], [18], [19], [20]. In this regard the potential role of both IFN-I and IFNγ in the maturation and activation of dendritic cells [21], [22], [23] and the impact of both IFN types on the development of effector and memory CTL is of particular interest. Moreover, it has not been possible to investigate a potential contribution of macrophage activation to Lm clearance in secondary immune responses of mice lacking Stat1 in all tissues.
Further interest in cell type-specific Stat1 activities derives from the opposing effects of IFNγ and IFN-I on innate resistance to Lm. IFNγ-deficient mice show a similar susceptibility as Stat1−/− mice [24]. By contrast IFN-I receptor (Ifnar) deficient mice are protected from lethal Lm infections [25], [26], [27]. Suppression of protective innate immunity by IFN-I was suggested to result from increased T lymphocyte apoptosis and subsequent IL10-mediated immunosuppression [28]. Furthermore, infection of macrophages and DCs with Lm causes IFN-I dependent downregulation of the IFNγ-receptor, hence unresponsiveness to IFNγ [29]. IFN-I also sensitize infected macrophages in vitro to die from infection with Lm [30], [31].
To overcome the limitations posed by the exquisite sensitivity of Stat1−/− mice to infections with Lm or other pathogens we generated mice with floxed Stat1 alleles. Here we report that cell type-restricted Stat1 ablation reveals a striking dichotomy of immunological effects. Macrophage Stat1 produces protective innate immunity whereas the opposite is true for T lymphocytes. In secondary immune responses to Lm T lymphocyte and dendritic cell Stat1 signaling becomes protective, but Stat1 in macrophages does not contribute to clearance of bacteria.
To decipher the importance of Stat1 signaling for protective immunity to Lm in the hematopoietic and non-hematopoietic cell compartments we conducted adoptive transfer experiments. WT and Stat1−/− mice were lethally irradiated and bone marrow of either Stat1−/− or WT background was implanted in these mice. After 8 weeks the chimerism was examined in blood, spleen and liver showing an efficient implantation of the transferred bone marrow (supplemental figure S1). Bone marrow-chimeric mice were subjected to intraperitoneal infections with sublethal doses of Lm and 72 hrs later the bacterial burden in spleen and liver was determined (figure 1A, 1B). Compared to WT mice reconstituted with WT bone marrow, mice lacking Stat1 in non-hematopoietic cells showed a minor reduction of bacterial clearance, hence minor contribution of non-hematopoietic Stat1 to innate resistance. This suggests that hepatocytes, although representing an important niche for Lm multiplication [32], [33], are not protected by Stat1 signaling. By contrast mice lacking Stat1 in bone marrow-derived cells displayed a clear loss of resistance.
In addition to pathogen clearance we tested the impact of Stat1 deficiency on the systemic cytokine response. WT mice which received Stat1−/− bone marrow responded to infection with a systemic cytokine storm, i.e elevated serum levels of almost all measured cytokines and chemokines (IL6, IL22, TNFα, MCP1, IL10, Rantes, IP10 and MCP3; figure 1C). This is likely to reflect the increase in bacterial burden, hence a higher intensity of the innate response. Intriguingly however, the highest levels of IL12p70 were determined in the group of Stat1−/− mice that received protective WT bone marrow and had very similar bacterial loads as WT mice. This suggests that Stat1 of non-hematopoietic cells participates in the negative regulation of IL12 synthesis. In line with increased IL12, IFNγ production was elevated compared to WT. Likewise IFNγ was increased in mice with Stat1−/− bone marrow although IL12 levels were normal. Therefore, IL12 and IFNγ levels are not strictly correlated. In this situation IFNγ synthesis is most likely part of the cytokine storm as a consequence of high bacterial burden.
To further study the contribution of individual immunecompetent cells for the innate phase of Lm infection we analysed resistance to lethal infection and bacterial clearance after tissue-restricted Stat1 ablation. To determine the importance of Stat1 signaling in myeloid cells we used LysMCreStat1flfl mice, which delete predominantly in macrophages and neutrophils [34], [35]. These mice display a significantly reduced ability to clear even a low dose of Lm from spleen and liver (figure 2A, 2B) and hence succumbed to infection, whereas all WT mice survived the intraperitoneal infection (figure 2C).
To test the involvement of Stat1 to the immune response to Lm in other cell types of the immune system, mice with Stat1 deficiency in CD11c-positive cells, predominantly dendritic cells, but also subpopulations of NK cells and alveolar macrophages, were generated (CD11cCreStat1flfl, figure S2); [36], [37]. Mice lacking Stat1 in T cells were obtained by crossing Stat1flfl to LckCre mice (LckCreStat1flfl; figure S2). DC- and T cell- deleted mouse strains were subjected to a sublethal dose of Lm by intraperitoneal injection and bacterial loads in spleen and liver were monitored for the next three days (figure S3A, S3B). These mice did not show elevated numbers of Lm in spleen and liver at day three (figure 2D, 2E) or a significantly altered susceptibility to sublethal infection (2C).
Intravenous infection with Lm lead to the same outcome as intraperitoneal infection. Three different doses of Lm, ranging from sublethal to lethal referred to WT mice, where chosen to determine the response of the Stat1-ablated mice (figure 2F–H). Increased sensitivity to infection was seen when myeloid cells lacked Stat1, whereas CD11cCreStat1flfl animals behaved similar to WT. Strikingly, LckCreStat1flfl mice displayed increased resistance to high infectious doses. This consequence of T cell specific Stat1 ablation was similarly observed following intraperitoneal infection with a higher than LD50 inoculum of Lm. Lack of activity of T cell Stat1 resulted in both increased survival of the animals and an enhanced clearance of the bacteria from spleen and liver (figure 2I, 2J).
IFN-I signaling in T cells was previously shown to reduce the clearance of Lm upon infection in the spleen [26]. To examine whether the enhanced survival in LckCreStat1flfl mice was due to a lack of IFN-I signaling, we analysed the bacterial load and survival in mice with IFN-I receptor deficiency in T cells (LckCreIfnarflfl). We noted a significantly lower number of splenic Lm in LckCreIfnarflfl mice compared to WT mice at high doses of infection, whereas the number of bacteria in the liver was not significantly reduced (figure 2K, 2L). Mice with complete Ifnar1 deficiency showed a clearly better ability to contain Lm infection than mice lacking Ifnar1 only in T cells. The increased ability of LckCreIfnarflfl mice to clear bacteria in the spleen did not result in a higher rate of survival compared to WT mice, whereas complete Ifnar1 deficiency did (figure 2M). This result suggests that the increase in resistance produced by the absence of T cell Stat1 cannot be entirely explained on the basis of the lack of IFN-I signaling in T cells.
TUNEL staining of spleen cells two days after i.p infection with Lm produced the expected large number of apoptotic cells in WT mice [26], which was strongly reduced in both LckCreStat1flfl and LckCreIfnar1flfl mice (figure 3). Since the decrease in cell apoptosis resulting from either Stat1 or Ifnar1 ablation was highly similar, the additional protection of LckCreStat1flfl mice from Lm infection is not due to a lesser rate of infection-induced apoptosis in Stat1-deficient T cells.
To further clarify the difference of LckCreStat1flfl and LckIfnar1flfl we isolated splenocytes of these genotypes and appropriate controls (WT, Stat1−/−, Ifnar1−/−) and infected them in vitro for two days with Lm at MOI 10. Subsequently we analysed the supernatant of these cultures for T cell cytokines. Stat1 deficiency in T cells lead to increased production of IL4 and IL17 and a clear suppression of IFNγ. In contrast, Ifnar1 deficiency in T cells did not decrease production of the signature cytokines under study, but increased the amounts of IFNγ, IL17 and IL10. The data indicate a Th population-independent negative regulation of T cell activation by IFN-I and demonstrate the strong influence of Stat1 on the generation of Th1 cells through its target gene T-bet [38] (figure S4). Thus, Ifnar or Stat1 ablation in T cells impact differently on the generation and function of Th cell populations in vitro.
Examination of systemic cytokine/chemokine levels demonstrated that mice lacking myeloid Stat1 signaling show increased levels of IL6, IL12p70, MCP1, MCP3, IL22, MIP1β, Rantes and IFNγ in their serum, similar to but not as dramatic as complete Stat1 deficiency (figure 4). As these mice have strongly elevated numbers of pathogens in their organs the increase in inflammatory cytokines may again reflect an increased activity of the innate immune system. Alternatively, increased cytokine production could also result from the loss of Stat1-mediated gene repression as reported for IL6 [39]. The function of Stat1 as both a transcriptional activator and repressor is well documented [40]. Both functions require binding to GAS sequences [41], but the detailed mechanisms are not understood.
Stat1 signaling in CD11c+ cells had a very selective impact on the levels of systemic cytokines, showing elevated levels of MCP1 compared to WT mice but interestingly, lower amounts of IL12p70. By contrast, higher levels of IL12p70 were detected in the serum of mice lacking Stat1 signaling in T cells, despite an equal bacterial load. The levels of TNFα and IL10 were too low to be detectable at this dose of infection.
Spleens (figure 5A) and livers (figure 5B) of infected animals with conditional Stat1 gene ablation were analysed using H&E staining to determine the severity of inflammation two days post-infection. In keeping with the loss of innate resistance, mice lacking Stat1 in myeloid cells showed a severe pathology of the spleen with increased lymphocyte depletion [42]. In the liver the radius of the inflammatory infiltrate area, classified as microabscess [43], correlated with the increase of bacteria found in this organ (figure 5B, 5C). In addition, the numbers of micro-abscesses correlated with bacterial burden, as lack of Stat1 in myeloid cells increased the area of infiltrates in the liver. Whereas CD11c+ cell-specific ablation of Stat1 led to significantly bigger areas of infiltrates compared to WT (figure 5C), the amount of infiltrates (figure 5D) and bacteria in the liver was not significantly enhanced compared to WT (figure 2E). Mice with T cell-restricted Stat1 gene deletion showed smaller infiltrate areas compared to the WT, again reflecting the protection of these mice from infection.
Liver failure may significantly contribute to the lethality of Lm infection [44]. To assess liver damage, we measured the amount of circulating amino aspartate transferase (AST), an enzyme released from damaged hepatocytes and readily measurable in serum samples [45] (figure 5E). LysMCreStat1flfl mice displayed strongly elevated levels of AST, indicating massive liver damage. No significant differences in AST were found in all other animals/genotypes compared to WT animals.
In addition to organ damage we examined the immune status of Stat1-ablated mice by analysing the composition of blood leukocytes. Mice lacking Stat1 in T cells had the highest numbers of circulating immune cells in their blood 72 h after infection (figure 5F, 5G, 5H). This is consistent with the notion that the reduced number of Lm in organs, coupled with reduced numbers of apoptotic cells led to a diminished recruitment of blood leukocytes. Additionally this result may indicate a defect in Stat1 regulated synthesis of T cell-derived chemokines.
The results shown in figures 1 and 2 emphasize the importance of Stat1 mediated macrophage activation. In spite of this, mice completely devoid of Stat1 cleared Lm less well than LysMCreStat1flfl animals. This result could be explained by incomplete ablation of the Stat1 gene in macrophages although our recent inspection of macrophages demonstrates deletion with very high efficiency [35]. Alternatively or additionally, therefore, the difference between LysMCreStat1flfl and Stat1−/− mice may reflect shaping of the innate immune response by Stat1 signaling in several different leukocyte populations. In our infection model the peritoneum is the site of immediate exposure of innate cells to the bacterial pathogen that initiates a local inflammatory response. To determine the degree to which cell type-specific Stat1 signaling determines this local immune response, we first analyzed the local chemokine/cytokine milieu in the peritoneal cavity over the course of the first three days of infection (figure S6). The most striking differences between genotypes were observed at day 2 (figure 6A). Absence of Stat1 in myeloid cells increased MIP1α MCP1 and Rantes amounts at day two and three compared to WT. MCP1 and Rantes were decreased upon CD11c-Cre-mediated Stat1 ablation at day two, but the levels of these chemokines recovered and exceeded WT levels at day three. T cell-specific Stat1 ablation lead to a decrease in Rantes levels at day two after infection, at day three the amount of the tested chemokines reached WT level. Examination of the pro- and anti-inflammatory cytokine gene expression patterns of adherent peritoneal macrophages isolated from infected mice indicated a small but significant role of myeloid cell Stat1 in the negative regulation of IL12 (figure 6B). Remarkably, Stat1 signaling in T cells was required for full IL12p40 expression. In keeping with the aforementioned negative regulation by Stat1 a more profound effect was noted with regard to IL6 production that was markedly upregulated upon STAT1 deficiency in either myeloid cells or the CD11c+ population. The CD11cCreStat1flfl genotype was unique in producing an adherent cell population with reduced IL10 production. Together with the systemic analyses shown in figure 1 and 2 our data suggest that peritoneal macrophages are major producers of IL6 and IL12. Lack of Stat1 signaling in CD11c+ dendritic cells or inflammatory monocytes may stimulate macrophages to produce excess amounts of IL6 and decreased amounts of IL10.
To determine whether the altered peritoneal chemokine/cytokine levels changed the cell recruitment, we isolated peritoneal exudate cells two days after intraperitoneal Lm infection and analysed the cell composition by Wright-Giemsa-stained cytospins and flow cytometry (figure 6C–E). Myeloid cells together constitute >95% of the peritoneal exudates in WT mice. In animals lacking Stat1 in DC reduced numbers of leukocytes were recruited, however neutrophils were increased at the expense of macrophages (figures 6C–E). Thus, Stat1 signaling in CD11c+ DC regulates monocyte/macrophage migration to the inflamed peritoneum. Mice with myeloid Stat1 ablation showed an increased influx of total peritoneal leukocytes with a similar tendency to reduce monocytes/macrophages and increase neutrophils. Finally, the absence of Stat1 signaling in T cells caused a strong increase in the amount of immune cells travelling to the peritoneum without altering their composition.
Together the data characterizing the peritoneal inflammatory response suggest a profound impact of Stat1 in different cell types on the cytokine milieu and on leukocyte composition. This may explain in part why myeloid cell-restricted Stat1 ablation does not fully reproduce the loss of bacterial clearance observed upon complete Stat1 gene deletion.
To analyse the impact of Stat1 signaling in different cell populations on establishing adaptive immunity against Lm, we applied an immunisation and challenge protocol to the respective conditional knockout mice. Under these conditions mice lacking Stat1 signaling in T cells failed to clear Lm from the spleen (figure 7A). Accordingly, an increased percentage of LckCreStat1flfl mice succumbed to infection compared to WT mice (figure 7B). Immunized mice lacking Stat1 in CD11c+ cells showed a slight impairment in clearing splenic Lm, yet the impact on survival was almost as pronounced as in mice lacking Stat1 in T cells. Myeloid Stat1 did not contribute to the establishment of adaptive immunity to Lm as bacterial clearance after immunisation was as strong as in WT mice.
Overall systemic cytokine levels were generally lower than those found after infection of naïve mice. Stat1 deficiency in CD11c+ cells caused a selective reduction of systemic IFNγ that may contribute to the reduced ability to raise adaptive immunity to Lm (figure 7C). The levels of IFNγ in mice lacking Stat1 signaling in T cells were equally high as in naïve mice. Given the reduced ability of Stat1−/− T cells to generate the Th1 lineage [38] (figure S4) this may reflect IFNγ production by cells other than Th1 or, alternatively, low numbers of Th1 cells developing in absence of Stat1 may produce higher IFNγ amounts due to the lack of the negative regulation Stat1 imposes on the IFNγ gene [46].
To further analyse the immunisation defects in CD11cCreStat1flfl mice, we investigated T cell responses after immunisation. Proliferation of splenic CD3+ T cells showed no significant differences (figure 8A). However, examination of the Treg population (CD4+Foxp3+) revealed an enhanced proliferative response in the spleens of mice with CD11c+-restricted Stat1 ablation (figure 8B). As regulatory T cells represent only a minor percentage of total splenic T cells it is not surprising that the difference in proliferation went unnoticed when analyzed in the context of total CD3+ T cell cells. The data suggest a contribution of DC Stat1 to the control of proliferation of a small proportion of antigen-specific Treg.
Studies in gene-modified mice and with cells from human patients suffering from recurrent infectious disease have unequivocally established the central importance of Stat1 for the establishment of protective innate immunity to viral and nonviral pathogens [6], [7], [47]. This includes Lm, the bacterial pathogen studied here. Conditional gene targeting allowed us to examine whether there is a uniform immunological impact of Stat1 across different cell types. Furthermore, we were able to investigate the importance of Stat1 signaling in the same cell types for the development of acquired antibacterial immunity.
Clearance of intracellular bacterial pathogens is caused either by a microbicidal effector mechanism of the infected cell or indirectly through CD8+ T cell-mediated cytolysis. Lm infects a variety of different cell types in vitro, either by active invasion or phagocytosis [48], [49]. In infected mice the pathogen replicates in both hematopoietic cells, predominantly macrophages, and non-hematopoietic cells amongst which the hepatocytes form a major niche [32], [33]. To our surprise Stat1 signaling provides non-hematopoietic cells with little effector potential, posing the question how Listeria are killed in these cell compartments particularly before the influx of antigen-specific CTL. One possibility is the death of infected hepatocytes and the subsequent phagocytosis of the cell contents including bacterial cargo by phagocytic cells of the innate immune system [32]. Subsequent sterile clearance most likely requires the development of CTL and active lysis of infected cells [50], [51], [52]. The importance of clearing Lm in the liver is underscored by our findings that the death of mice with different Stat1 genotypes correlated well with the inflammatory infiltrate in this organ and with the hepatotoxicity caused by infection.
LysMCre mediated gene deletion occurs predominantly in macrophages and granulocytes [34]. A recent report shows that the contribution of neutrophils to the immune response against Lm is surprisingly small [53] and, given the short half life of granulocytes, Stat1-dependent transcriptional response to IFN is unlikely to enhance their microbicidal activity. Therefore our data are consistent with the previous notion that IFNγ and Stat1 cause macrophage activation and that activated macrophages represent a dominant innate anti-listericidal effector mechanism of the innate immune response [54], [55]. Indeed, our data reveal the essential role of this cell type for clearance of Lm. That said the clearance deficit of mice lacking macrophage Stat1 was significantly lower than that of Stat1−/− mice. In this regard inspection of the local immune reaction elicited by intraperitoneal infection with Lm. showed that besides macrophages T cells and particularly CD11c+ cells shape the inflammatory environment by regulating chemokine production and cell influx. CD11c+ cells in this situation are likely to represent inflammatory DC that arise from inflammatory monocytes [56]. Although lack of Stat1 signaling in non-hematopoietic cells or DC does not per se affect survival and clearance of infection, it may synergize with the Stat1 deficiencies of other cell types to produce the more severe outcome of the complete Stat1 knockout. Increased Lm replication in Stat1−/− compared to LysMCreStat1flfl mice caused a more severe cytokine storm that is likely to be one cause of their accelerated death.
While the amount of serum cytokines generally followed the severity of infection in different Stat1 genotypes, IL12 was the exception because it was strongly increased in mice with Stat1-deficiency outside the hematopoietic compartment that were able to cope with infection nearly as well as WT mice. This finding reveals negative regulation of IL12 synthesis by a non-hematopoietic cell. IL10 is a negative regulator of IL12 production [57] and its synthesis can be suppressed by IFNγ [58]. However, IL10 is considered to be a product of hematopoietic cells [59]. The nature of the suppressive cell type and the mechanism of IL12 suppression will require further investigation.
The exacerbation of infection by Stat1 in T lymphocytes is particularly intriguing. Unanue and colleagues demonstrated the T cell response to IFN-I as a mechanism underlying their adverse effect [26], [28]. While our data confirm that the IFN-I response of T cells indeed reduces bacterial clearance, Ifnar1-deficiency in this cell type alone does not reproduce the consequences of the complete Ifnar1 knockout and it does not cause the robust effect of T cell-specific Stat1 deletion. Hence, T cells indeed inhibit protective innate immunity to Lm, but the effect of Stat1 goes beyond IFN-I signaling. Furthermore, additional cell compartments must contribute to the suppressive effect of IFN-I, as suggested by the comparison between infected Ifnar−/− and LckCreIfnar1flfl mice. At present we do not fully understand how Stat1 signaling in T cells reduces innate immunity to Lm. Clearly, it increases apoptosis of splenic cells, and the comparison to Ifnar-ablated cells suggests this results from the activity of type I interferons. However, T cell Stat1-mediated loss of innate protection goes beyond IFN-I effects on splenocyte apoptosis. In line with the results shown in figure S4, skewed CD4+ T cell differentiation causing a reduction of IFNγ-producing Th1 cells and a concomitant increase in Th2 and Th17 cells may be a contributing factor. Importantly, Ifnar1 deficiency in T cells increases IFNγ production when Lm antigens are presented by wt APC in vitro. At present we do not know whether T cell differentiation is a decisive factor within the first three days of infection in mice. Systemic cytokine profiles of LckCreStat1flfl animals showed little change with respect to WT mice in this period. Systemic IL10 levels were below the detection limit, leading us to assume that reduced clearance does not result from global immunosuppression. In addition, bacterial multiplication might be enhanced by the reported suppressive activity of IFN-I on IFNγ receptor expression [29].
Infection of immunized mice with Lm caused a drastic change in the consequences of Stat1 signaling in cell types of the immune system. Most importantly Stat1 activity in T cells was now required for protective immunity. Antigen-specific CTL are critical effectors for adaptive immunity to Lm [15], [52]. Our data provide the first genetic proof that Stat1 signaling in both T cells and DC is required for acquired resistance, in addition to showing that IFNγ-activated macrophages are dispensable once memory lymphocytes have been produced. The essential stimulus for Stat1 signaling in T cells is unclear. IFN-I appear to be dispensable for both CD4+ and CD8+ T cell development during Lm infection [60], [61]. Furthermore, a study analyzing CTL development in IFNγ-deficient animals infected with very low numbers of Lm shows that IFNγ is not essential for protective CTL-mediated immunity [51]. Possibly IFNγ contributes to protective CTL memory when infection occurs with a high infectious dose. Therefore, Stat1 may increase the efficacy of memory CD8+ T cell responses.
Defective Stat1 signaling in CD11c+ DC also reduced protection by the adaptive immune system upon secondary challenge with Lm. This is consistent with our recent finding that immunization with Stat1−/− DC caused a strongly diminished CTL response to Ova peptide [62] and with reports by others that Stat1−/− DC fail to elicit protective immunity to Leishmania major [63]. IFN-I and Stat1 reportedly support DC maturation and activation [64], [65]. Data in the literature thus suggest a defect of Stat1-deficient DC to present antigen to T cells. In line with this IFN-I were reported to stimulate the ability of DC to cross-present antigen [23]. In our experiments the lack of DC Stat1 affected survival of mice more than splenic clearance of bacteria, which argues against a general defect in generating effector CTL. Moreover, activation of both CD4+ and CD8+ T cells to the levels found with wt cells occurred in vitro when Stat1−/− DC were used as antigen presenters (Figure S5). In a mouse model of graft versus host disease Ma and colleagues noted an increased proliferation of FoxP3+ regulatory T cells upon transfer of Stat1−/− splenocytes into irradiated hosts [66]. Prompted by this finding we tested whether Stat1 deficiency in DC might similarly cause increased Treg proliferation in Listeria-infected mice. Indeed we noted that the proportion of proliferating FoxP3+ cells was about two-fold higher in spleens from infected mice with a CD11cCreStat1flfl genotype. It is therefore possible that an increased number of antigen-specific regulatory T cells suppresses effector T cells and thus reduces the immune response to Lm.
In summary, cell type-restricted ablation reveals a fascinating complexity of Stat1's regulatory power for the development of both innate and adaptive immune responses to Lm.
Animal experiments were discussed and approved by the University of Veterinary Medicine Vienna institutional ethics committee and carried out in accordance with protocols approved by the Austrian law (BMWF-68.205/0204-C/GT/2007; BMWF-68.205/0210-II/10b/2009, BMWF-68.205/0243-II/3b/2011). Bacteria were prepared for infection as described previously [67]. For infection, Lm LO28 were washed with PBS and injected intraperitoneally (i.p) or intravenously (i.v) of 8- to 10-week-old sex and age matched C57BL/6N (WT), Stat1flfl (B6.129P2-Stat1tmBiat, [35]), LysMCre Stat1flfl (B6.129P2-Lyz2tm1(cre)Ifo/J-Stat1tmBiat [34]), CD11cCre Stat1flfl (B6.Cg-Tg(CD11c-Cre)-Stat1tmBiat [68]), LckCre Stat1flfl (B6.129P2-Tg(LckCre)-Stat1tmBiat LckCre [69]), Ifnar1−/− (B6.129P2-IfnaR1tm1) [70], Stat∧−/− (B6.129P2-Stat1tm1), Ifnar1flflLckCre (B6.129P2-IfnaR1tm1-Tg (LckCre) [71] mice at the respective dose. The infectious dose was controlled by plating serial dilutions on Oxford agar plates. The survival of mice was monitored for 10 days, and data were displayed as Kaplan-Meier plots. For determination of bacterial loads of liver and spleen mice were killed at the indicated time points. The respective organs were isolated and homogenized in PBS. Serial dilutions of the homogenates were plated on BHI plates and incubated at 37°C for 24 h. For immunisation mice were injected i.p with 1×10∧6 attenuated Listeria (ΔActA). After 2–3 weeks mice were infected i.v with 1×10∧5 Lm.
For cytokine analysis mice were bled via the retro-orbital sinuses and serum was collected and stored at −80°C. Using the FlowCytomix system (ebioscience) concentrations of indicated cytokines (IFNγ, IL6, IL10, IL12p70, Mcp1, Mcp3, Rantes, GMCSF, Mip1α, Mip1β, IP10, IL22, TNF α) in 25 µl of serum were measured.
For isolation of peritoneal macrophages mice were infected for 48 h with 5×10∧6 LO28 i.p. Mice were sacrificed, the peritoneum was flushed with two times 10 ml of DMEM and cells were harvested by centrifugation and plated on 6 well plates. After 2 h adherent cells were washed with PBS and RNA was prepared for Real Time PCR analysis. The composition of total peritoneal exudate cells was examined using Wright-Giemsa staining of cytospins. Composition of adherent cells were analysed by flowcytometric analysis (F4/80-APC, CD11b-PE, CD3-FITC (BD biosciences), Ly6C-PerCP (ebioscience)). Chemokines and cytokines were measured using the FlowCytomix system after flushing the peritoneum with 1 ml of DMEM.
RNA was isolated using the Nucleospin II kit (Macherey and Nagel) according to protocol. Reverse transcription was accomplished using RevertAid (Fermentas). The Real Time PCRs were run on an Eppendorf cycler. After correction for the housekeeping gene Gapdh, every sample was calculated to the mean of WT mRNA levels. The following primer sequences were used (all 5′-3′): IL6: for TAGTCCTTCCTACCCCAATTTCC; rev TTGGTCCTTAGCCACTCCTTC; IL10: for GGTTGCCAAGCCTTATCGGA; rev ACCTGCTCCACTGCCTTGCT; IL12p40: for TGGTTTGCCATCGTTTTGCTG; rev ACAGGTGAGGTTCACTGTTTCT; GAPDH: for CATGGCCTTCCGTGTTCCTA; rev GCGGCACGTCAGATCCA.
Spleens were isolated after indicated timepoints and single cell suspensions were prepared using a 80 µm cell strainer. After red blood cell lysis cells were stained for CD3-PE, CD4-FITC, CD8-APC, CD11b-PerCP, Gr1-PE, FOXp3-APC, Ki67-PerCP (all BD bioscience). For intracellular staining cells were fixed and permeabilised using the FoxP3 staining kit (ebioscience)
Mouse organs were fixed with 4% paraformaldehyde over night, paraffin embedded and 4 µm sections were prepared using a microtome. Hematoxyline and eosin staining (H&E) were performed using standard protocols. The radius of the infiltrate was measured using the Zeiss Axioplan software. For TUNEL staining, sections of spleens were stained using the TUNEL-POD kit (Roche) according to protocol. Additionally, sections were blocked for endogenous peroxidase activity in methanol with H2O2 and after proteinase K treatment blocked with 5% normal goat serum to reduce background staining. TUNEL enzyme and POD conversion was applied as described, and AEC+ high sensitivity chromogen (Dako) was used as a HRP substrate. Subsequently, sections were counterstained with hematoxyline.
Aspartate amino transferase concentrations were measured in mouse serum using a COBASc11 analyzer (Roche).
WT (Ly5.1 and C57BL/6) and Stat1−/− animals were lethally irradiated with 8,2 Gy for 17 minutes and engrafted with 5×10∧6 bone marrow cells of respective genotypes by i.v injection. After 6 weeks engraftment was analysed in blood, spleen and liver by flow cytometry using the antibodies Ly5.1-FITC, Ly5.2-PE, CD11b-PerCP (BD biosciences).
Mice were bled via the retro-orbital sinuses in tubes coated with EDTA and the cellular composition was measured using a vet haematology analyzer V sight (A. Menarini diagnostics).
Bacterial loads of organs were compared using the Mann-Whitney test; mRNA expression data and cytokines levels were analysed with the Students t test. For both the GraphPad Software was used. Asterisks describe the significances as follows: * p≤0,05; **p≤0,01; ***p≤0,001
|
10.1371/journal.pgen.0030233 | Conserved Regulation of MAP Kinase Expression by PUF RNA-Binding Proteins | Mitogen-activated protein kinase (MAPK) and PUF (for Pumilio and FBF [fem-3 binding factor]) RNA-binding proteins control many cellular processes critical for animal development and tissue homeostasis. In the present work, we report that PUF proteins act directly on MAPK/ERK-encoding mRNAs to downregulate their expression in both the Caenorhabditis elegans germline and human embryonic stem cells. In C. elegans, FBF/PUF binds regulatory elements in the mpk-1 3′ untranslated region (3′ UTR) and coprecipitates with mpk-1 mRNA; moreover, mpk-1 expression increases dramatically in FBF mutants. In human embryonic stem cells, PUM2/PUF binds 3′UTR elements in both Erk2 and p38α mRNAs, and PUM2 represses reporter constructs carrying either Erk2 or p38α 3′ UTRs. Therefore, the PUF control of MAPK expression is conserved. Its biological function was explored in nematodes, where FBF promotes the self-renewal of germline stem cells, and MPK-1 promotes oocyte maturation and germ cell apoptosis. We found that FBF acts redundantly with LIP-1, the C. elegans homolog of MAPK phosphatase (MKP), to restrict MAPK activity and prevent apoptosis. In mammals, activated MAPK can promote apoptosis of cancer cells and restrict stem cell self-renewal, and MKP is upregulated in cancer cells. We propose that the dual negative regulation of MAPK by both PUF repression and MKP inhibition may be a conserved mechanism that influences both stem cell maintenance and tumor progression.
| The mitogen-activated protein (MAP) kinase (MAPK) enzyme is crucial for regulation of both stem cell maintenance and tumorigenesis. Two conserved controls of MAPK include its activation by RAS signaling and a kinase cascade as well as its inactivation by MAPK phosphatases (MKPs). We identify a third mode of conserved MAPK regulation. We demonstrate that PUF (for Pumilio and FBF [fem-3 binding factor]) RNA-binding proteins repress mRNAs encoding MAPK enzymes in both the Caenorhabditis elegans germline and human embryonic stem cells. PUF proteins have emerged as conserved regulators of germline stem cells in C. elegans, Drosophila, and probably vertebrates. Their molecular mode of action relies on binding to sequence elements in the 3′ untranslated region of target mRNAs. We report that PUF proteins bind and repress mRNAs encoding C. elegans MPK-1 as well as human ERK2 and p38α. We also report that PUF repression and MKP inactivation function redundantly in the C. elegans germline to restrict MPK-1/MAPK activity and prevent germ cell apoptosis. We suggest that this dual regulation of MAPK activity by PUF and MKP proteins may be a conserved mechanism for the control of growth and differentiation during animal development and tissue homeostasis.
| Mitogen-activated protein (MAP) kinases (MAPKs) control many aspects of animal development, including cell proliferation, differentiation, and survival [1]. Most relevant to this work are MPK-1, the primary Caenorhabditis elegans MAPK/ERK homolog [2,3], as well as ERK2 and p38α, two human MAPK homologs [1]. MAPK enzymes are activated by a class of dual specificity kinases that phosphorylate both threonine and tyrosine residues (e.g., [4]) and are inactivated by a class of dual specificity phosphatases, called MAPK phosphatases (MKPs) (e.g., [5,6]). Aberrant ERK2 activation contributes to human developmental disorders, such as Noonan syndrome, Costello syndrome, and cardiofaciocutaneous syndrome (reviewed in [7]); p38α, on the other hand, is thought to inhibit tumor initiation by inducing apoptosis in response to oxidative stress [8]. In mouse embryonic stem cells (mESCs), ERK2 and p38α MAPK signaling promotes differentiation and inhibits self-renewal [9,10].
The C. elegans germline provides a superb model for understanding the molecular controls of stem cells, proliferation, differentiation, and survival [11]. In this simple tissue, germline stem cells are restricted to the distal “mitotic region.” At a molecular level, germline stem cells are maintained by Notch signaling and two RNA-binding proteins, fem-3 binding factor (FBF)-1 and FBF-2. FBF-1 and FBF-2 (collectively called FBF) are nearly identical and largely redundant proteins that belong to the broadly conserved family of PUF RNA-binding proteins [12,13]. PUF proteins inhibit gene expression by binding regulatory elements in the 3′ untranslated region (3′UTR) of their target mRNAs, thereby controlling their translation or stability [14]. FBF maintains germline stem cells by repressing mRNAs that encode differentiation-promoting regulators. For example, FBF represses gld-1 and fog-1 mRNAs, which encode regulators that promote entry into meiosis or sperm differentiation, respectively [15,16]. The role of PUF proteins in stem cell maintenance appears to be a conserved and perhaps ancestral function [14,17], but the target mRNAs responsible for this function have not yet been identified. Here, we suggest that MAPK mRNA is one key target.
Once C. elegans germ cells have left the mitotic region, they move proximally and progress through meiosis and gametogenesis. Activated MAPK controls exit from meiotic pachytene and physiological apoptosis during oogenesis [18,19]. Normally, about half of the germ cells progress from pachytene into diakinesis and develop as oocytes, and the other half of the germ cells undergo apoptosis in the proximal pachytene region [19]; however, in mutants with blocked MAPK signaling, germ cells arrest in pachytene and fail to die [18,19]. An antibody that specifically detects activated MAPK, called α-DP-MAPK (for dually phosphorylated MAPK), reveals a dramatic increase in activated MPK-1 just prior to the pachytene to diplotene/diakinesis transition [20,21].
A key inhibitor of C. elegans MPK-1 is LIP-1, a homolog of the dual specificity phosphatase MKP [22]. LIP-1 has two roles in germline development. First, LIP-1 controls the extent of germline proliferation in the mitotic region: wild-type germlines contain significantly more mitotically dividing germ cells than do lip-1 null mutants [23]. Because the depletion of mpk-1 rescued the lip-1 proliferation defect, it seems likely that LIP-1 promotes germline proliferation by inhibiting MPK-1 activity. Second, LIP-1 promotes the G2/M arrest typical of diakinesis, apparently by inhibiting MPK-1 activity after germ cells have exited pachytene [21].
In this paper, we explore the regulatory relationship between PUF proteins and MAPK expression, both in the C. elegans germline and in human embryonic stem cells (hESCs). We find that mpk-1 mRNA is a direct target of FBF repression in C. elegans and that two human MAPK mRNAs, those encoding ERK2 and p38α, are repressed by human PUM2. We also demonstrate that FBF and LIP-1 function redundantly to inhibit germ cell apoptosis and suggest that this dual regulation of MAPK signaling, which occurs at post-transcriptional and post-translational levels, respectively, may be conserved during diverse cellular processes in animal development and tissue homeostasis.
The mpk-1 gene encodes two major transcripts, mpk-1a and mpk-1b, which produce MPK-1A and MPK-1B proteins, respectively [2,24] (Figure 1A). To identify which products were expressed in the germline, we performed RT-PCR of RNA prepared from adults that either contained a normal germline (GL+) or contained no germline (GL−). The mpk-1a mRNA is contained entirely within mpk-1b, but mpk-1b harbors a unique exon (Figure 1A). We therefore examined mpk-1 mRNAs using either mpk-1ab primers, which recognize both isoforms, or mpk-1b−specific primers. The mpk-1ab mRNA was abundant in both GL+ and GL− animals, but mpk-1b mRNA was very low or undetectable in GL− animals (Figure 1B). Therefore, mpk-1b appears to be enriched in the germline. To corroborate this result, we examined the two MPK-1 proteins in Western blots of protein prepared from wild-type (GL+), GL− mutants, and mpk-1(ga117) mutants. MPK-1A was present in both GL+ and GL− animals, but MPK-1B protein was found only in GL+ animals (Figure 1C). We conclude that mpk-1b RNA and its MPK-1B protein are predominantly expressed in the germline.
We next investigated the distribution of mpk-1 mRNA and MPK-1 protein in the germline. After in situ mRNA hybridization of extruded germlines, both mpk-1ab and mpk-1b−specific probes were low in the mitotic region, increased in the transition zone, and became abundant in the pachytene and oogenic regions (Figure 1D and 1E). No signal was detected with the control mpk-1 sense probe (Figure 1F). For immunohistochemistry, we used an anti-MAPK/ERK polyclonal antibody that cross-reacts with both MPK-1 isoforms in wild-type animals but is absent from mpk-1(ga117) mutants (Figure 1C). The distribution of MPK-1 protein was similar to that of mpk-1 mRNA: MPK-1 protein was low in the distal germline (e.g., mitotic region, transition zone), was increased in the proximal pachytene region, and became abundant in developing oocytes (Figure 1G and 1H). Essentially no signal was seen in mpk-1(ga117) mutant germlines (Figure 1I and 1J). To investigate the isoform expressed, we depleted mpk-1b mRNA by RNA interference (RNAi); the specific elimination of MPK-1B was verified by Western blots (unpublished data). MPK-1 protein was essentially absent from mpk-1b RNAi germlines, except for a low signal in developing oocytes (Figure 1K and 1L). We conclude that MPK-1B is the predominant MPK-1 isoform in the germline.
To determine if FBF might repress mpk-1 expression, we compared the abundance of MPK-1 protein in germlines that either had wild-type FBF (both fbf-1 and fbf-2) or no FBF (neither fbf-1 nor fbf-2). For this study, we could not examine a simple fbf-1 fbf-2 double mutant, because that animal does not maintain mitotically dividing germ cells [15]. Instead, we examined mpk-1 expression in tumorous (Tum) germlines that have robust germ cell proliferation both with and without FBF. In gld-1 mutants (Tum+FBF), MPK-1B was about 8-fold lower than in gld-1; fbf-1 fbf-2 mutants (Tum−FBF) (Figure 2A, lanes 2 and 3). MPK-1B was also lower in gld-1 gld-2 mutants (Tum+FBF) than in gld-1 gld-2; fbf-1 fbf-2 mutants (Tum−FBF) (Figure 2A). By contrast, MPK-1A levels were equivalent in these strains (Figure 2A). Therefore, FBF affects MPK-1B, but not MPK-1A, abundance.
To visualize where within the germline FBF affects MPK-1 expression, we stained dissected germlines with both the MPK-1 polyclonal antibody and DAPI, and we quantitated levels with ImageJ software. Consistent with the Western blot data, MPK-1 was lower in gld-1 germlines than in gld-1; fbf-1 fbf-2 germlines (Figure 2B–2F). This difference was particularly striking within the mitotic region, where MPK-1 was about 5-fold lower in Tum+FBF than in Tum−FBF germlines (Figure 2B–2F). We also stained fbf-1 single mutant germlines, which maintain a mitotic region but are compromised for full FBF activity; in about 20% of dissected germlines, MPK-1 protein was detected in the distal mitotic region (unpublished data). We conclude that FBF maintains a low level of MPK-1 protein in the distal germline.
FBF binds specifically to FBFbinding elements (FBEs) within the 3′UTR of its direct target mRNAs [12,13,15,16,23,25]; potential FBEs can be recognized by a consensus sequence (UGURHHAUW) [“R,” purine; “H,” not G; “W,” A or U] [26]. The mpk-1 3′UTR possesses two potential FBEs that conform to this sequence (Figure 3A). To assess FBF binding to these predicted mpk-1 FBEs, we used both yeast three-hybrid (Figure 3B and 3D) and gel retardation assays (Figure 3E). Yeast three-hybrid interactions were monitored by production of β-galactosidase from a lacZ reporter (Figure 3D). The mpk-1 FBEa and FBEb interacted with both FBF-1 and FBF-2 in three-hybrid assays (Figure 3C and 3D) and bound to purified recombinant FBF-2 in gel shift assays (Figure 3E). Furthermore, those interactions were specific: wild-type mpk-1 FBEa and FBEb bound FBF, but not PUF-8 (Figure 3D) or PUF-5 (unpublished data), and that binding was disrupted by mutations of the UGU in the consensus binding site (Figure 3C–3E, FBE* mutant changed UGU to aca). The apparent Kd values for mpk-1 FBEa and FBEb were about 93 nM and 320 nM, respectively. We conclude that the mpk-1 3′UTR bears two FBEs and that FBEa appears to have higher affinity for FBF than does FBEb.
We next asked whether FBF protein associates with mpk-1 mRNA in the nematode. Specifically, we prepared C. elegans extracts from animals carrying either a rescuing epitope-tagged GFP::FBF or a control GFP::tubulin (TUB), and incubated those extracts with immobilized GFP antibodies to immunoprecipitate (IP) associated mRNAs. We then used RT-PCR to assess either mpk-1 or control mRNAs (eft-3, negative control; gld-1, positive control). mpk-1 mRNA was reproducibly enriched in the IP from GFP::FBF-bearing animals compared to that from the GFP::TUB animals (Figure 3F). Therefore, FBF is likely to bind directly to the mpk-1 mRNA in vivo. Interestingly, the mpk-1 FBEa is conserved in three Caenorhabditis species: C. elegans, C. briggsae, and C. remanei (Figure 3G). We conclude that the mpk-1 3′UTR possesses FBEs and that FBF repression of mpk-1 expression is direct.
The C. elegans homolog of MAPK phosphatase, LIP-1, behaves genetically as an inhibitor of MAPK activity and is likely to inactivate MPK-1 in germ cells (see Introduction) [21,23]. Therefore, MAPK is negatively regulated in the germline by two distinct mechanisms: FBF represses mpk-1 expression (present work) and LIP-1 inhibits MAPK activity. To test the possibility that FBF and LIP-1 might function redundantly to control the distribution of activated MPK-1, we used the α-DP-MAPK monoclonal antibody, which recognizes the active form of MAPK by its dual phosphorylation (DP). In wild-type germlines, activated MPK-1 was not detected in the distal germline (e.g., mitotic region, transition zone) but became abundant in the proximal part of the pachytene region and in maturing oocytes (Figure 4A) [20,21]. A similar distribution was seen in fbf-1 and lip-1 single mutants (Figure 4A–4C). By contrast, activated MPK-1 was broadly distributed in fbf-1; lip-1 double mutant germlines, extending all the way to the distal tip (Figure 4D). We conclude that activated MPK-1 is subject to two redundant modes of downregulation: FBF acts post-transcriptionally to repress mpk-1 mRNA and LIP-1 is likely to act post-translationally to inhibit MPK-1 activity.
In wild-type C. elegans hermaphrodites, physiological germ cell apoptosis requires MPK-1 activation [19]. Strong loss-of-function mutations in any of the genes of the RAS/MPK-1 pathway interrupt germ cell apoptosis [19], but germ cell apoptosis does not increase in let-60/Ras gain-of-function (gf) mutants [19,27]. Although MPK-1 activity is much higher in let-60(gf) germlines than in wild-type, the distribution of activated MPK-1 is similar in let-60(gf) and wild-type germlines [21]. We hypothesized that germ cell apoptosis might be regulated by the distribution of activated MPK-1 rather than its quantity at the site of apoptosis. To test this idea, we counted the number of germ cell deaths in adult hermaphrodite germlines, using the vital dye SYTO 12 (Molecular Probes) to detect apoptotic corpses. In these experiments, we used fbf-1 mutants to deplete but not eliminate FBF activity: FBF-2 provides sufficient FBF to maintain germline stem cells. Wild-type hermaphrodite germlines had about 2.6 germ cell corpses per gonad arm (Figure 4E, 4F, and 4K), and a similar number was seen in fbf-1 and lip-1 single mutants (Figure 4K). However, in fbf-1; lip-1 double mutants, the number of germ cell corpses increased dramatically (Figure 4G, 4H, and 4K).
To determine if the increased germ cell apoptosis in fbf-1; lip-1 mutants depends on MPK-1 activity, we used mpk-1(ga111), a temperature-sensitive mutation. Specifically, we compared the number of germ cell corpses after shifting fbf-1; lip-1 double mutants and fbf-1; lip-1; mpk-1(ga111ts) triple mutants to restrictive temperature (25 °C). Whereas the fbf-1; lip-1 double mutant displayed excess germ cell death, the fbf-1; lip-1; mpk-1(ga111ts) triple mutant had far fewer corpses (Figure 4I, 4J, and 4K). Therefore, MPK-1 activity is required for the increased apoptosis in fbf-1; lip-1 double mutants. We conclude that FBF and LIP-1 proteins act redundantly to inhibit MPK-1 activity and promote germ cell survival.
We next investigated the possibility that PUF RNA-binding proteins might also control MAPK expression in humans. This idea was inspired in part by the knowledge that the human PUF protein, PUM2, is expressed abundantly in both human embryonic stem cells (hESCs) and human germline stem cells [28] and in part by the conserved PUF role in stem cell maintenance (see Introduction). Predicted Pumilio binding elements (known as NREs [nanos response elements]) were sought using UGUANAU as a core consensus [29]. The Erk2 3′UTR possesses a putative NRE immediately adjacent to the cleavage and polyadenylation hexanucleotide sequence (AAUAAA) (Figure 5A and 5B), and the p38α 3′UTR has four putative NREs (Figure 5A). In yeast three-hybrid assays, the Erk2 NRE, p38α NREa, and p38α NREb all interacted specifically with PUM2 (Figure 5C). Moreover, the wild-type NREs in Erk2 and p38α bound purified PUM2 protein in gel shift assays (Figure 5D), but mutant NREs (NRE*) with an altered consensus (Figure 5B) did not (Figure 5D). We next asked whether an Erk2 NRE is conserved in mouse Erk2 3′UTR. Intriguingly, the Erk2 NRE is conserved in human and mouse 3′UTRs, both being located next to the hexanucleotide sequence (Figure 5E). We conclude that PUM2 protein binds to Erk2 and p38α 3′UTRs and that PUF binding to MAPK 3′UTRs is highly conserved.
To test if PUM2 controls Erk2 and p38α expression, we performed a series of enhanced green fluorescent protein (EGFP)-based reporter assays in hESCs. Specifically, we fused an EGFP reporter to the Erk2 3′UTR that contained either a wild-type NRE, Erk2 3′UTR(wt), or a mutated NRE, Erk2 3′UTR(mut) (Figure 6A). We transfected these constructs along with a transfection control into hESCs and monitored EGFP expression 24 h later. We first observed EGFP using fluorescence microscopy and then determined expression levels by Western blot analysis (Figure 6B–6J). As a control, we used a reporter carrying a 3′UTR without any predicted NREs (EGFP::BGH [bovine growth hormone] 3′UTR). hESCs carrying the EGFP::BGH 3′UTR reporter expressed EGFP at a very high level (Figure 6J). By contrast, hESCs transfected with the Erk2 3′UTR(wt) reporter had much less EGFP (Figure 6B, 6C, and 6J). To ask if the NRE is critical for this low expression, we assayed Erk2 3′UTR(mut), a reporter with three altered nucleotides in the NRE consensus (UGU to aca) (Figure 6A). This Erk2 3′UTR(mut) reporter produced about 9-fold more EGFP than the Erk2 3′UTR(wt) reporter (Figure 6D, 6E, and 6J). We speculated that endogenous PUM2 might repress expression of the Erk2 3′UTR(wt) reporter but not the Erk2 3′UTR(mut) reporter. Attempts to use siRNA to silence endogenous PUM2 were not successful. We therefore cotransfected hESCs with the EGFP reporters and PUM2::FLAG (Figure 6A), and we found that PUM2::FLAG dramatically repressed Erk2 3′UTR(wt) expression (Figure 6F, 6G, and 6J) but did not repress Erk2 3′UTR(mut) expression (Figure 6H, 6I, and 6J).
We next asked if reporters carrying the p38α 3′UTR were also controlled in an NRE-dependent manner. To this end, we transfected hESCs with either of two EGFP-based reporter genes, p38α 3′UTR(wt) or p38α 3′UTR(mut) (Figure 6A). As found for the Erk2 reporter, the wild-type, but not the mutant, p38α 3′UTR was capable of efficiently repressing expression from the EGFP reporter gene in hESCs (Figure 6K–6O). In this case, expression from p38α 3′UTR(wt) was about 6-fold lower than that from p38α 3′UTR(mut) (Figure 6O). Taken together, we conclude that the PUM2 binding elements present in both Erk2 and p38α 3′UTRs mediate repression in hESCs.
The MAPK enzyme is controlled by several conserved pathways (Figure 7A). Best understood is its activation by RAS and a kinase cascade, a pathway that has been conserved in virtually all eukaryotic cells [4]. In addition, MAPK is inhibited by the conserved dual specificity MKPs [5,6]. Here, we show that the PUF RNA-binding proteins are another broadly conserved mechanism of MAPK control. We demonstrate that PUF proteins control the expression of MAPK-encoding mRNAs in both the C. elegans germline and hESCs. We also show that PUF repression and MKP inhibition are redundant in their ability to restrict activated MAPK and prevent apoptosis in the C. elegans germline. We propose that the dual regulation of MAPK signaling by PUF repression and MKP inhibition may be a conserved means of influencing both stem cells and tumor progression.
Both PUF RNA-binding proteins and MAPK enzymes are highly conserved from yeast to humans. In this paper, we show that PUF proteins directly bind to 3′UTR regulatory elements in MAPK-encoding mRNAs and thereby control the generation of MAPK protein. Specifically, C. elegans FBF binds and regulates mpk-1 expression in germ cells, and human PUM2 binds and regulates Erk2 and p38α 3′UTRs in hESCs. Similarly in yeast, the Mpt5 PUF protein inhibits Ste7/MAPKK expression to regulate the filamentation-specific MAPK pathway [30]. Therefore, an ancient relationship appears to exist between the PUF RNA-binding proteins and MAPK signaling. To our knowledge, our work provides the only direct link between PUF proteins and MAPK-encoding mRNAs. Because this direct connection exists in both C. elegans and humans, we suggest that it may represent a broadly conserved regulatory relationship among metazoans.
Figure 7A places PUF repression into a conserved pathway of MAPK control. Specifically, PUF proteins are responsible for the post-transcriptional repression of MAPK mRNAs; mechanistically, this could be achieved by controlling either their translation or stability. PUF proteins were originally thought to control mRNA stability in yeast but to control mRNA translation in animals [14], but as more examples of PUF-controlled mRNAs have surfaced, it has become clear that this generalization is too simple. For example, C. elegans FBF controls the stability of lip-1 mRNA [23], and yeast Mpt5 controls the translation of HO mRNA [31]. Regardless of mechanism, our work shows conclusively that PUF proteins are direct regulators of MAPK-encoding mRNAs.
MAPK is a key regulator of programmed cell death, among its other roles during animal development [1]. In this work, we investigated the function of PUF repression and MKP inhibition in the control of apoptosis in the C. elegans oogenic germline. In wild-type animals, about half of the germ cells die and the other half begin oocyte maturation (Figure 7B) [19]. Indeed, activated MPK-1 is most abundant where germ cells either die or begin oogenesis [20,21]. In mutants lacking either FBF-1/PUF or LIP-1/MKP, the distribution of activated MPK-1 is essentially normal and the number of germ cells that undergo cell death is also normal. By contrast, in double mutants lacking both FBF-1 and LIP-1, activated MPK-1 extends all the way to the distal end of the germline, where it is normally never seen, and apoptosis increases dramatically. This result suggests two things. First, because distribution of activated MAPK affects number of apoptotic germ cells, the decision to die may be programmed at a location distal to their actual site of death. Second, and perhaps most important for this work, the distribution of activated MAPK and the number of germ cell deaths are controlled redundantly by FBF repression and LIP-1 inhibition.
MAPK inhibition is ensured in the C. elegans germline, at least in part because FBF represses lip-1 mRNA in addition to its control of mpk-1 mRNA (Figure 7B) [23]. Therefore, when FBF/PUF activity is lowered in the distal germline (as germ cells leave the mitotic region and enter the transition zone), LIP-1/MKP abundance increases. The result of this extra step of regulation is that MAPK activity is kept low even when FBF levels decrease. Therefore, MAPK inhibition is ensured not only by redundant inhibitors but also by a well-buffered circuitry.
A key unanswered question is whether mammalian MAPK is also subject to homologous redundant controls. Clearly both exist in mammals: PUM2 represses both Erk2 and p38α mRNAs (present work), and MKPs negatively regulate ERK2, p38α and JNK members of the MAPK family [6]. But do they function in the same cells in a redundant fashion? The answer to this question will require removal of both PUF and MKP proteins in vertebrate cells, which remains a challenge for the future.
In the C. elegans germline, FBF is required for stem cell maintenance [15], and MPK-1 promotes differentiation (either oocyte maturation or apoptosis) [18,19]. Although a vertebrate role for PUF proteins in stem cell maintenance remains a matter of speculation [14,17], recent evidence has given this idea credence. Thus, PUM2 is enriched in germline stem cells and embryonic stem cells [28], and murine PUM2 mutant testes are smaller than normal and contain some agametic seminiferous tubules [32]. Therefore, the role of PUF proteins in stem cell maintenance may be conserved.
The roles of MAPK and MKP in vertebrates are reminiscent of those of MPK-1 and LIP-1 in the C. elegans germline. Vertebrate ERK2 and p38α MAPKs can antagonize stem cell self-renewal and promote differentiation [9,33–35]. In cancer cells, ERK2 and p38α MAPKs are thought to promote apoptosis [36,37]. Indeed, MKPs are often upregulated in human cancer cells, and the MKP inhibition of MAPK activity has been suggested to be critical for human cancer progression [38,39]. Therefore, MAPKs and MKPs affect both continued self-renewal and tumor progression.
In this work, we show that PUF RNA-binding proteins repress MAPK-encoding mRNAs in both C. elegans and hESCs. Indeed, to our knowledge, ERK2 and p38α mRNAs are the first PUM2 targets reported to date. The biological significance of this finding is not known. One simple idea is that PUF represses MAPK expression as part of a larger regulatory circuit designed to maintain stem cells in a naïve state. However, a more complete understanding will require learning the extent of PUM2 repression, the extent of MKP inhibition, and the biological readout of different levels of MAPK activity—all in the same cells. Although this more in-depth understanding remains a challenge for the future, we emphasize here that the PUF and MKP controls of MAPK signaling are broadly conserved and likely work together broadly to control stem cells and tumor progression.
All strains were maintained at 20 °C as described [40], unless noted otherwise. We used the wild-type Bristol strain N2 as well as the following mutants: LGI: gld-1(q485) [41], gld-2(q497) [42,43]; LGII: fbf-1(ok91) [15], fbf-2(q738) [13], gld-3(q730) [44], nos-3(q650) [45]; LGIII: glp-1(q224) [46], mpk-1(ga117) [2], mpk-1(ga111) [24]; and LGIV: lip-1(zh15) [22].
In situ hybridization was carried out using the protocol described [47], with minor modifications. Dissected adult hermaphrodite gonads were fixed (3% formaldehyde, 0.25% glutaraldehyde, 100 mM K2HPO4 [pH 7.2]) for 3 h at room temperature. After washing three times with PBT solution (1× PBS containing 0.1% Tween 20), gonads were treated with proteinase K (50 μg/ml) for 30 min at room temperature and then refixed in the same solution for 15 min at room temperature. DNA probes were synthesized with digoxigenin-11-dUTP by repeated primer extension. Fixed gonads were incubated for 24 h at 48 °C in a solution containing the DNA probe plus 5× SSC, 50% deionized formamide, 100 μg/ml herring sperm DNA, 50 μg/ml heparin, and 0.1% Tween 20. To visualize the probes, gonads were incubated with alkaline phosphatase−conjugated antidigoxigenin antibody (Roche, 1:2,000 dilution in PBT containing 0.1% BSA) at 4 °C for overnight. After washing several times in PBT (+ 0.1% BSA), staining was developed for 1 h in a solution (100 mM Tris Cl [pH 9.5], 100 mM NaCl, 5 mM MgCl2, 0.1% Tween 20, 1 mM Levamosole) containing 4-nitro blue tetrazolium chloride (0.23 mg/ml) and 5-bromo-4-chloro-3-indolyl-phosphatase (0.18 mg/ml) and then terminated in PBT containing 20 mM EDTA.
Dissected gonads were fixed with 3% formaldehyde, 100 mM K2HPO4 (pH 7.2) for 1 h, and postfixed with cold (−20 °C) 100% methanol for 5 min. Antibody incubations and washes were performed as described [47]. Polyclonal rabbit α-MAPK/ERK antibody (Sc94; Santa Cruz Biotechnology) was used at 1:400 dilution, and monoclonal mouse α-DP-MAPK antibody (Sigma) was used at 1:200 dilution. DAPI staining followed standard methods.
Blots were prepared by standard procedures. Protein samples were separated on 4%–20% gradient gels (Cambrex), and the blot was probed with polyclonal rabbit α-MAPK/ERK antibody (Sc94; Santa Cruz Biotechnology), α-GFP antibody (Molecular Probes), monoclonal mouse α-tubulin antibody (Sigma), α-actin antibody (MP Biomedicals), and α-FLAG antibody (Sigma).
Three-hybrid assays were performed as described [48]. For β-galactosidase assays, cells were grown in selective media to an OD600 of 1.0 and mixed with an equal volume of β-Glo (Promega) reagent. Luminescence was measured after 1 h. Gel shift assays were performed as described [49].
SYTO 12 (Molecular Probes) dye was used to estimate the relative numbers of germ cell corpses [19,50]. Animals were incubated in a 33 μM aqueous solution of SYTO 12 for 2 h at 20 °C and then transferred to seeded plates to purge stained bacteria from the intestine. After 30 min, animals were mounted on agarose pads and observed under a fluorescence microscope equipped with Nomarski optics to score SYTO 12−positive germ cells.
ORF and 3′ sequences from the fbf-1 genomic locus (from ATG to 317 bp downstream of the STOP codon) were PCR amplified with flanking attB1 and attB2 sequences and cloned into pDONR201 (Invitrogen) to create pCM3.06, which was sequence verified. A Gateway LR recombination reaction (Invitrogen) was performed between pCM3.06 (entry) and pCM2.03 (destination). pCM2.03 is a bombardment-ready vector containing the unc-119 rescuing fragment (used for transformant selection [51]), the pie-1 enhancer and promoter (to drive expression in the germline [52]), GFP with three synthetic introns (from pPD103.87, A. Fire, personal communication), and the attR1::Gateway Cassette B::attR2. The resulting plasmid pCM4.06 contains unc-119; pie-1 (enhancer + promoter)::GFP::attB1::fbf-1ORF+3′UTR::attB2. pCM4.06 was transformed into unc-119(ed3) worms by microparticle bombardment [51] to create line JH2012 (genotype: unc-119(ed3); axIs1459 [CM4.06]).
Age-synchronous adult Ppie-1::GFP-FBF-1 (JK4091) and Ppie-1::GFP-TUB (AZ224) transgenic animals were grown for 24 h after the L4 stage on NGM plates supplemented with concentrated OP50. Worms were harvested by rinsing plates with M9 buffer, and worms were washed with M9 buffer until the supernatant was clear. Worms were then washed twice with buffer A (20 mM Tris [pH 8.0], 150 mM NaCl, 10 mM EDTA [pH 8.0], 1.5 mM DTT, 0.1% NP-40, 0.02 mg/ml heparin), and worm pellets were frozen at −20 °C. Roughly 0.5 ml of worm pellets was used for affinity purification. Worm lysate was generated by grinding worms with a mortar and pestle under liquid nitrogen in lysis buffer (buffer A plus 1× Complete Protease Inhibitor Cocktail [Roche], 20 U/ml DNase I [Ambion], 100 U/ml RNase OUT [Invitrogen], 0.2 mg/ml heparin), followed by 30 passes with a glass dounce. The extract was then centrifuged twice at 10,000 g for 10 min at 4 °C to remove insoluble debris and fat. The protein concentrations of cleared extracts were determined by Bradford assay, and extracts were diluted to 10 mg/ml with lysis buffer. To limit nonspecific interactions with the affinity column, extracts were next precleared by incubation with 15 μl of Immobilized Protein A (Pierce) for 1 h at 4 °C. Then 1.6 μg of mouse α-GFP monoclonal antibody 3E6 (Q•Biogene) prebound to 16 μl of Immobilized Protein A (Pierce) was added to each precleared extract, and the resultant slurries were incubated at 4 °C for 2 h. Beads were then washed once with lysis buffer for 15 min at 4 °C and four times with wash buffer (20 mM Tris [pH 8.0], 150 mM NaCl, 1 mM EDTA [pH 8.0], 10% glycerol, 0.01% NP-40, 1 mM DTT, 10 U/ml RNase OUT) for 15 min at 4 °C. For protein analysis, beads were boiled in Laemmli buffer. To purify RNA, beads were treated with TRIzol reagent (Invitrogen) followed by RNeasy Mini Kit (Qiagen) to purify RNA, following manufacturer's instructions. Reverse transcription reactions were performed using 20 ng of Input or IP RNA in 10-μl reactions with an oligo(dT) primer and SuperScript III reverse transcriptase (Invitrogen). PCR was carried out on the cDNA template for eft-3, gld-1, and mpk-1 in the linear range (33 cycles) using gene-specific primers such that one primer spanned an exon-exon junction. PCR products were resolved on 1% agarose gel and stained with ethidium bromide.
H9 human ES cells with a stably transfected EBNA protein (H9-EBNA) that enables episomal replication of exogenous plasmid containing an Orip site were maintained in defined medium-TeSR medium [53,54] containing 50 ng/ml G418. For the transfection, H9-EBNA cells were dissociated by Dispase (Invitrogen) and seeded onto Matrigel-coated six-well plates. At 24 h after seeding, H9-EBNA cells were transfected with 1 μg of indicated reporter plasmids together with an equal amount of indicated effector plasmids by using Fugene6 reagents (Roche). At 24 h after transfection, the cells were photographed and lysed by RIPA buffer for the Western blot analysis. |
10.1371/journal.pntd.0001322 | Quantifying the Emergence of Dengue in Hanoi, Vietnam: 1998–2009 | An estimated 2.4 billion people live in areas at risk of dengue transmission, therefore the factors determining the establishment of endemic dengue in areas where transmission suitability is marginal is of considerable importance. Hanoi, Vietnam is such an area, and following a large dengue outbreak in 2009, we set out to determine if dengue is emerging in Hanoi.
We undertook a temporal and spatial analysis of 25,983 dengue cases notified in Hanoi between 1998 and 2009. Age standardized incidence rates, standardized age of infection, and Standardized Morbidity Ratios (SMR) were calculated. A quasi-Poisson regression model was used to determine if dengue incidence was increasing over time. Wavelet analysis was used to explore the periodicity of dengue transmission and the association with climate variables. After excluding the two major outbreak years of 1998 and 2009 and correcting for changes in population age structure, we identified a significant annual increase in the incidence of dengue cases over the period 1999–2008 (incidence rate ratio = 1.38, 95% confidence interval = 1.20–1.58, p value = 0.002). The age of notified dengue cases in Hanoi is high, with a median age of 23 years (mean 26.3 years). After adjusting for changes in population age structure, there was no statistically significant change in the median or mean age of dengue cases over the period studied. Districts in the central, highly urban, area of Hanoi have the highest incidence of dengue (SMR>3).
Hanoi is a low dengue transmission setting where dengue incidence has been increasing year on year since 1999. This trend needs to be confirmed with serological surveys, followed by studies to determine the underlying drivers of this emergence. Such studies can provide insights into the biological, demographic, and environmental changes associated with vulnerability to the establishment of endemic dengue.
| Dengue is the most common vector-borne viral disease of humans, causing an estimated 50 million cases per year. The number of countries affected by dengue has increased dramatically in the last 50 years and dengue is now a major public health problem in large parts of the tropical and subtropical world. It is of considerable importance to understand the factors that determine how dengue becomes newly established in areas where the risk of dengue was previously small. Hanoi in North Vietnam is a large city where dengue appears to be emerging. We analyzed 12 years of dengue surveillance data in order to characterize the temporal and spatial epidemiology of dengue in Hanoi and to establish if dengue incidence has been increasing. After excluding the two major outbreak years of 1998 and 2009 and correcting for changes in population age structure over time, we found there was a significant annual increase in the incidence of notified dengue cases over the period 1999–2008. Dengue cases were concentrated in young adults in the highly urban central areas of Hanoi. This study indicates that dengue transmission is increasing in Hanoi and provides a platform for further studies of the underlying drivers of this emergence.
| Dengue is caused by infection with one of four genetically related but serologically distinct dengue virus serotypes, which are transmitted by the bite of an infected female Aedes mosquito. It is the most common vector borne viral disease of humans with an estimated 50 million infections every year and around 3.6 billion people living in areas at risk [1], [2]. Over the past 50 years dengue has spread inexorably, with 9 countries reporting dengue transmission prior to 1970 compared to over 124 now, and incidence having increased 30 fold [3]. There are reasons to believe that the expansion of dengue will continue. Whilst the geographic range of A. aegypti, the principle urban vector of dengue, has shrunk in some areas, notably the Mediterranean, North America and Australia, it has expanded in Asia and has re-invaded large parts of South America following eradication attempts in the 1950's and 60's [4]. Meanwhile, the geographic range of Aedes. albopictus, a secondary vector of dengue, has expanded dramatically [5]. Both Aedes vectors are adapted to peridomestic urban habitats that are expected to burgeon over the next four decades, with the urban populations of Africa and Asia predicted to treble and double respectively [6]. A. albopictus is also well adapted to rural and temperate environments, and although dengue has been perceived as a predominantly urban disease, the scale and potential for rural dengue transmission is increasingly being recognized [7], [8]. Southeast Asia is at the epicenter of this global dengue outbreak, accounting for 70% of global dengue morbidity and mortality, and is a region with substantial potential for further expansion [8], [9]. Preventive interventions are currently limited largely to vector control but substantial efforts are being made to develop a vaccine.
Dengue epidemiology is a determined by a complex interaction of vector, pathogen, and host biology; macro and microclimate; the physical environment; and social factors [10], [11]. Dengue has been extensively studied in hyper-endemic areas but less work has been done in areas at the margins of transmission, where opportunities may exist to characterize the process of dengue emergence and assess the factors associated with transmission and disease severity in a less complex environment. The development of dengue vaccines is an added imperative to understand dengue epidemiology at various intensities of transmission in order to predict the impact of different vaccination strategies in different epidemiological contexts. Hanoi is the capital of Vietnam and has a sub-tropical climate, with distinct seasons and cool winters. Hanoi experiences annual seasonal dengue outbreaks with little or no transmission in the intervening months. Increasing numbers of notified dengue cases have led to concerns that dengue is ‘emerging’ in Hanoi, and in 2009 Hanoi experienced its largest ever-recorded outbreak of dengue. In this study, we set out to investigate if dengue is emerging in Hanoi by quantifying changes in disease incidence and the age of notified cases over the 12-year period 1998 to 2009 after adjusting for demographic changes. To explore disease dynamics we describe the temporal patterns of dengue incidence and its association with local climate variables. This work was conceived as the first step in studying the process of dengue emergence in Hanoi and as a framework for developing further studies of the determinants of dengue infection and disease risk in a low transmission setting.
The study used public health surveillance data routinely collected through the Ministry of Health notifiable diseases surveillance program. The study did not use patient medical records and all data were analyzed anonymously. The analysis of the surveillance data was approved by the National Institute of Hygiene and Epidemiology, a specialized institute of the Vietnam Ministry of Health.
Hanoi is Vietnam's capital and its second largest city. It is located in the north of the country in the low lying and densely populated Red River delta. In the 2009 decennial census, the population of Hanoi was estimated at nearly 6.5 million, with a population density of 1943 persons per km2. In contrast to South Vietnam, which is always hot but with dry and rainy seasons, North Vietnam has four distinct seasons with hot and humid summers, receiving the majority of rainfall, and cool and relatively dry winters.
Dengue is one of 24 infectious diseases in Vietnam for which there is monthly mandatory reporting from all administrative levels, which, in increasing size, are communes, districts, provinces, and regions. The detection and reporting of dengue cases follows the Ministry of Health 1999 guidelines on surveillance, diagnosis and treatment of dengue [12]. The case definition for suspected clinical dengue infection is: 1.) fever or history of acute fever, lasting from 2 to 7 days; and 2.) headache, myalgia, arthralgia, and rash. Prior to 1999 suspected dengue cases were reported to the MoH according to the WHO surveillance case definition. If a patient fulfills the clinical dengue case definition, a single blood sample may be taken and sent to the National Institute for Hygiene and Epidemiology for the detection of anti-dengue virus specific immunoglobulin (Ig) M antibodies by an in-house, M-antibody capture enzyme linked immunoassay (MAC-ELISA) as previously described [13]. ELISA results were only collected through the surveillance system from 2002 onwards.
Anonymous individual case reports of all clinically diagnosed dengue cases of any disease severity in Hanoi residents reported through the public health surveillance system were obtained from Hanoi Preventive Medicine Center (PMC) from1st January 1998 to 31st December 2009. Data available on every case were gender, commune and district of residence, and month and year of diagnosis. Age was missing for two subjects only.
Age stratified population estimates for Hanoi by district were derived as follows. Data for 1999 and 2009 were obtained directly from the decennial national Population and Housing Census' conducted by the General Statistic Office of Vietnam (GSO. http://www.gso.gov.vn). Age stratified population estimates for 2000–2008 were obtained from GSO and back-adjusted based on the results of the 2009 census. Since age stratified population estimates for Hanoi for 1998 were not available, we derived these by applying the age structure from the 1999 census to the estimated 1998 population. In 2008 a large province adjacent to Hanoi became a new district of Hanoi. This administrative change increased the population of Hanoi by more than 3 million from 3.2 to 6.5 million. All analyses presented here are restricted to the 14 Hanoi districts that remained unchanged throughout the 12-year study period.
Daily records of total rainfall; mean wind velocity; mean, maximum and minimum temperature; and relative humidity from a central Hanoi meteorology station were obtained from the National Centre for Hydrometeorological Forecasting (http://www.nchmf.gov.vn/web/en-US). Analysis of meteorological data used total monthly rainfall, the monthly mean of daily mean temperatures, the monthly mean of daily mean wind velocity, and the monthly mean of daily mean relative humidity.
To adjust for the potential confounding effect of changes in the age structure of the population of Hanoi over the study period, we applied age specific incidence rates for each year to a standard population (direct age standardization) [14]. The standard population was the population estimated by the 2009 Population and Housing Census. This provided age-adjusted annual estimates of the incidence and the median and mean age of notified dengue cases.
To determine if dengue incidence was increasing over time, a quasi-Poisson regression model was used with the annual count of dengue cases as the outcome variable, year as the independent variable, and (log-transformed) mid-year population size as an offset. The age adjusted count per year, rather than per month, was used since monthly counts are non-independent because of seasonal patterns in dengue incidence, and monthly counts would therefore not fulfill the Poisson assumption of independence of outcomes. A quasi-Poisson model was used as the data were over-dispersed. Linear regression was used to test whether there was a linear time trend in the median age of infection. Statistical analysis was conducted in R 2.9.0 (R Foundation for Statistical Computing, Vienna, Austria).
Standardized Morbidity Ratios (SMR) for the 2009 dengue outbreak were calculated using the age-stratified population distribution from the 2009 census and age-stratified dengue incidence in Ba Dinh District in 2009 as the reference population (SMR = 1). SMR's were displayed using the statistical software R 2.9.0 (R Foundation for Statistical Computing, Vienna, Austria).
Wavelet analysis was used before to explore the periodicity of dengue transmission and the association with climate variables. Wavelet analysis is especially suited to time series where the average, variance, or relationship with co-variates change over time (i.e. non-stationary), and provides the possibility to explore linkages between multiple non-stationary time series [15], [16]. Dengue periodicity was investigated using a continuous wavelet transformation, which decomposes the time series into time-frequency components. Wavelet coherency was performed to quantify associations between monthly dengue cases and local meteorological covariates. Wavelet phase analysis and cross-correlation function (CCF) were used to quantify the lag period over time and mean lag period of a specific frequency (i.e. annual cycle), respectively. Due to the large outbreaks at both ends of the time series (i.e.1998 and 2009) accompanied with low numbers of notified cases in the intervening years, the aggregated dengue time series were log transformed and normalized in order to homogenize the variance prior to analyses. The Morlet wavelet was used and performed in Matlab software. All significance levels were based on 1000 bootstrapped replications.
Between 1998 and 2009, 25983 dengue cases were notified to Hanoi PMC. Large outbreaks were observed in 1998 and 2009, with smaller annual outbreaks in intervening years (figure 1a). The annual outbreaks typically begin around July, peak in October, and then decrease towards the end of the year, corresponding with the wet and hot periods (figure 1b).
In the quasi-Poisson model there was no statistically significant yearly increase in age adjusted incidence when all years were included (table 1). Since dengue characteristically produces large epidemics interspersed by many years, these unusually large epidemic years may mask underlying trends in the data. Therefore we also ran the quasi-Poisson model with the two epidemic years of 1998 and 2009 excluded. In this analysis of years 1999-2008, the age adjusted dengue incidence increased 1.38 fold annually (table 1).
The unadjusted median and mean age of notified cases were 23 and 26.3 years respectively, and 53% of cases were male. The majority of notified cases (85%) occurred in individuals aged over 15 years (figure 2A). Although the average age of notified dengue cases increased over the study period, from a median (mean) age of 20 (22.4) years in 1998 to 24 (26.8) years in 2009, this was not statistically significant after adjusting for changes in the population age structure (figure 2B).
IgM ELISA results were available from 3678 cases notified between 2002 and 2009. The proportion of cases with an available ELISA result varied from 68.2% in 2002 to 4.5% in 2009. The overall, proportion of ELISA tests that were anti-dengue virus specific IgM positive was 23.7%, ranging from 13.6% in 2006 to 49.8% in 2003. The lowest annual test positivity rates were observed in 2006 (13.6%), 2007 (17.6%), and 2008 (16.1%).
Dengue SMRs were highest in the central districts of Hanoi, with the districts with an SMR greater than three all being highly urbanized and densely populated areas in the centre of the city (figure 3).
Continuous wavelet analysis of the aggregated time series showed a significant annual cycle of dengue transmission (figure 4C). Furthermore, a continuous multi-annual band (2–3 years period) was detected, however it did not reach statistical significance in the wavelet power spectrum (figure 4D). Similar dengue periodicity was observed among all districts (figure 4B).
All local climatic variables show a statistically significant association with dengue in the 1-year mode (figure 5, left panel). Although significant associations were also observed in the multi-annual band (2–3 years), it should be interpreted with caution since the 2–3 year cycle of dengue cases is itself not significant (figure 4C). The time series oscillations show variability over time in both the lag period and the strength of association (figure 5, right panel). On average, a delay of 1–2 months was observed for rainfall and mean temperature whereas a 4–5 months delay was seen for humidity and wind velocity (figure 5, right panel).
In Hanoi dengue transmission demonstrates clear annual cycles that are associated with a lag of around two months with seasonal increases in mean temperatures and rainfall. This pattern reflects climate driven changes in vector abundance, survival, and biting frequency; and the time it takes for a newly infected mosquito to become infectious to humans (the extrinsic incubation period) [17]–[20]. We observed a pattern of high wind speeds being associated with periods of low dengue notifications, an association noted by other authors [21], [22]. Although it is not established whether this association is causal, high wind speed could conceivably interfere with normal mosquito movements and biting behaviors.
Large outbreaks occurred in 1998 and 2009, and wavelet analysis indicated possible 2–3 years periodicity. A 2–4 years cycle has been described in Thailand, Cambodia, and South Vietnam. Moreover, a meta-cycle between 8–10 years has been observed in Thailand [23]–[28]. This cycle of larger multi-year outbreaks superimposed on an annual cycle is a characteristic feature of dengue epidemiology that has been attributed to various models of serotype-host interactions, and a possible influence of multi-year climate oscillations [10], [23]–[25], [29]–[33]. With a peak of 384 total notified cases per 100,000 persons in 2009, Hanoi has a much lower dengue incidence than other parts of Southeast Asia. Ho Chi Minh City in South Vietnam had around 199 hospitalized cases per 100,000 in 2009, whilst in Thailand and Cambodia the rates of symptomatic dengue infection in children <15 years are estimated at around 24/1000 and 41/1000 respectively [34], [35]. Yet despite these large differences in incidence, similar temporal patterns are observed in Hanoi, suggesting that the intensity of dengue transmission in Hanoi may be sufficient to replicate the inter-subtype dynamics observed in higher transmission areas [29]. This will have relevance for understanding how dengue vaccination may influence intrinsic pathogen-host dynamics in settings with different intensities of transmission [10], [32], [36].
Excluding the two major outbreaks (in 1998 and 2009), the annual age adjusted incidence of notified dengue increased significantly. However, over the same period the average age of infection has not decreased, as would be expected if the per capita rate of infection in susceptible people (the force of infection) was increasing [37]. This apparent discordance could arise through several mechanisms: 1) the recorded increase in incidence may be an artifact of improved case detection and reporting; 2) unmeasured demographic changes, such as an influx of unregistered adult workers, may increase the average age of the population beyond that captured by census and population estimates; 3) susceptible adult in-migration from areas of low dengue risk may increase the relative abundance of susceptible adult individuals; 4) the force of infection has declined resulting in fewer infections and an increase in the median age of infection, but this has resulted in an increase in the incidence of clinical cases (as opposed to infections) since adults are more likely than children to suffer clinically apparent illness as a result of primary dengue infection [38], [39]. It is not possible to distinguish between these scenarios on the basis of passively notified clinical dengue illness alone, and population-based studies with carefully monitored demographics and serial seroepidemiology are needed.
The average age of notified dengue cases in Hanoi is high and 85% are aged over 15 years. Although the average age of clinical cases in Thailand has increased over recent years, the clinical burden of dengue in much of Southeast Asia still primarily falls on children aged less than 15 years [27], [35], [37]. Dengue does however predominantly affect adults in several Southeast Asia countries including Malaysia, Sri Lanka, and Singapore, and dengue is an adult problem in much of the Americas [38], [40]–[45]. The most important influences on geographical differences in the age group predominantly affected are likely to be the local intensity of dengue transmission and the time since dengue was (re)introduced into the population [46]. The high average age of notified dengue cases in Hanoi may reflect a relatively low force of infection but may also reflect relatively recent introduction of dengue into Hanoi such that the adult population has not yet acquired multitypic immunity [46].
Our study is limited by its reliance on passively notified cases of clinically apparent dengue infection. Only a small proportion of all notified cases were dengue IgM positive and the lowest IgM test positivity rates were in 2006, 2007, and 2008. More cases were notified in each of these three years than in any other years except the outbreak years of 1998 and 2009. If the size of the 2006–2008 outbreaks were inflated by the reporting of an unusually large number of non-dengue illnesses, the increase in dengue incidence we have identified may be false. The data reported here are of illness episodes that met the surveillance case definition for dengue, which includes both fever and rash. Measles also presents with fever and rash, and although included in the Expanded Program of Immunization (EPI), there was a large national epidemic in 2009 [47]. However, the peak of this measles epidemic was in February 2009 and was clearly distinct in timing from the dengue epidemic that peaked in October 2009 [47]. Rubella is not included in the EPI and may also be confused for dengue. The occurrence of a large rubella outbreak in Hanoi in 2011 suggests that a large rubella outbreak was unlikely to have occurred in the 5 years prior to 2011 [48]. Chikungunya has not yet been detected in Vietnam. Since dengue often presents as a non-specific febrile illness that is difficult to distinguish clinically from a large range of other illnesses, it is also probable that the reported cases we have used in this analysis significantly under-estimate the true number of dengue infections.
Since anti-dengue virus IgM is not detectable until 3–5 days after infection, and is frequently negative in secondary dengue infections, asays to detect anti-dengue virus IgM antibodies have limited sensitivity (perhaps <20%) when used alone to diagnose dengue infections during the acute illness [49], [50]. Consequently a single negative anti-dengue virus IgM ELISA from an acutely ill patient with clinically suspected dengue is not very helpful in excluding dengue infection. There are also limitations to the specificity of a positive dengue IgM ELISA. IgM ELISA's can remain positive for three months or longer after initial dengue infection, therefore febrile illnesses following dengue infection may be incorrectly diagnosed as dengue on the basis of persisting anti-dengue IgM [50]. False positive results are also possible as a result of cross-reactivity with anti-bodies to Japanese Encephalitis (JE) following infection or vaccination. However JE in Vietnam is predominantly a low-incidence rural problem and the immunization programme has targeted children in higher risk rural Provinces [51]. Therefore it is unlikely that a significant number of dengue IgM positive cases in Hanoi are false positives due to cross-reaction with JE antibodies. Given these limitations, interpretation of single dengue IgM results is difficult, and other diagnostic approaches, such as combined NS1 antigen and IgM assays, should be considered to augment and aid interpretation of clinical surveillance data [52].
As dengue disease severity is modulated by age, prior dengue exposure history, and the infecting dengue serotype, the incidence of clinically apparent dengue cases is unlikely to have a linear relationship with transmission intensity [38], [39], [53]. Therefore data on the age specific incidence of clinical illness cannot be used to directly infer the age specific risk of infection. Seroepidemiological studies are required for this and should be conducted as part of national surveillance programs. Public health authorities in Vietnam have the impression that dengue is an emerging health problem in Hanoi. Although our analysis supports that conclusion, there is a need for population-based seroepidemiology studies that remove any residual confounding by unmeasured demographic factors, and give an unbiased estimate of the age and serotype specific force of infection. The interpretation of such studies will be improved by the recent development of assays that are able to distinguish the original infecting serotype in secondary dengue infections [36]. If the current extensive efforts to develop a dengue vaccine are successful, there exists the exciting potential that appropriately targeted and timed vaccination in areas of low and intermittent transmission may be a cost effective intervention to prevent annual outbreaks or even prevent the (re)establishment of dengue. Such interventions will require a thorough understanding of the dynamics of dengue in such areas.
|
10.1371/journal.ppat.1004732 | Escherichia coli α-Hemolysin Counteracts the Anti-Virulence Innate Immune Response Triggered by the Rho GTPase Activating Toxin CNF1 during Bacteremia | The detection of the activities of pathogen-encoded virulence factors by the innate immune system has emerged as a new paradigm of pathogen recognition. Much remains to be determined with regard to the molecular and cellular components contributing to this defense mechanism in mammals and importance during infection. Here, we reveal the central role of the IL-1β signaling axis and Gr1+ cells in controlling the Escherichia coli burden in the blood in response to the sensing of the Rho GTPase-activating toxin CNF1. Consistently, this innate immune response is abrogated in caspase-1/11-impaired mice or following the treatment of infected mice with an IL-1β antagonist. In vitro experiments further revealed the synergistic effects of CNF1 and LPS in promoting the maturation/secretion of IL-1β and establishing the roles of Rac, ASC and caspase-1 in this pathway. Furthermore, we found that the α-hemolysin toxin inhibits IL-1β secretion without affecting the recruitment of Gr1+ cells. Here, we report the first example of anti-virulence-triggered immunity counteracted by a pore-forming toxin during bacteremia.
| The pathogenic potentials of most microbes depend on a repertoire of virulence factors. Despite major progress in the understanding of the molecular mechanisms underlying the activities of bacterial effectors, little is known about how they cooperate during infection to overcome host immune defenses and promote microbial persistence. Here, we investigated the roles of two uropathogenic Escherichia coli (UPEC) effectors that are co-ordinately expressed, α-hemolysin (HlyA) and cytotoxic necrotizing factor 1 (CNF1). We demonstrated that the HlyA toxin is critical for bacterial stability in the blood and showed that one important role of HlyA is to inhibit the CNF1-induced host response. Collectively, these findings reveal why the coordinated activities of HlyA and CNF1 are necessary for the full virulence of UPEC. Moreover, they unravel a HlyA-driven counter-defense mechanism used by bacteria to facilitate their survival.
| Bacteremia caused by extraintestinal strains of pathogenic Escherichia coli is a leading cause of death worldwide [1,2]. Among these pathogens, uropathogenic E. coli (UPEC) is a major etiological agent of bacteremia [1,2]. Therefore, it is essential to define the mechanisms by which virulence factors of UPEC and innate immune signaling pathways control the bacterial burden in the blood.
The major virulence factors of UPEC have been characterized at the molecular level [3–6]. These factors include the presence of a specialized adhesive appendage and specific metabolic pathways as well as protein toxins; together, these features enable UPEC to efficiently colonize the urinary tract and translocate into the bloodstream of the host [7–9]. Two highly prevalent bacterial toxins, α-hemolysin (HlyA) and cytotoxic necrotizing factor-1 (CNF1), work together to damage and disrupt the cohesion of the uroepithelium, which additionally leads to the worsening of the host inflammatory reaction [10–12]. The high prevalences of hlyA and cnf1 in uroseptic strains of UPEC suggest the possible functions of both of these toxins during bacteremia [9,13,14]. In contrast, it has not been determined whether both toxins contribute to the pathogen burden during bacteremia. HlyA belongs to a group of pore-forming leukotoxins that contain RTX repeats [13–15]. Depending on its concentration and on the type of cell intoxicated, HlyA either displays cytolytic activity or hijacks innate immune signaling pathways [13,16–18]. However, its role during bacteremia remains to be determined. The gene encoding the CNF1 toxin is located downstream from the α-hemolysin operon and is co-expressed with HlyA [19,20]. All CNF1-positive uroseptic strains display a hemolytic phenotype [9]. The CNF1 toxin possesses an enzymatic activity that is responsible for the posttranslational deamidation of a specific glutamine residue in a subset of small Rho GTPases, namely, Rac, Cdc42 and RhoA [21]. This type of modification increases the flux of activated Rho proteins and augments signaling through their downstream signaling pathways [21].
The activation of small Rho GTPases by virulence factors is a common trait of various enteric and extraintestinal Gram-negative pathogens. This activation of Rho GTPases confers upon bacteria the property to invade epithelial cells and tissues as well as to hijack inflammatory cell responses [22–25]. Emerging studies have indicated that cells are capable of perceiving the abnormal activation of Rac/Cdc42 induced by virulence factors of pathogens and translating this information via NOD1 and RIP1/RIP2 kinase signaling pathways into danger signals [26,27]. This innate immune mechanism involving the sensing of pathogens is here referred to as anti-virulence immunity (AVI), and it shares similarities with effector-triggered immunity (ETI), the mechanism by which plants sense the activities of bacterial effectors [28,29]. It will be important to define whether and how AVI triggers pathogen destruction in collaboration with the recognition of conserved microbial-associated patterns by pattern recognition receptors (PRR).
Inflammatory caspases, such as caspase-1 and caspase-11, drive innate immune responses to a variety of bacterial stimuli, such as microbe-associated molecular patterns (MAMPs) [e.g., lipopolysaccharide (LPS) or muramyl dipeptide (MDP)], as well as toxin-driven membrane damage [30]. Pathogen perception by NOD-like receptors triggers the assembly of inflammatory caspases in an operational ASC-dependent inflammasome complex that carries out the processing and release of the pro-inflammatory cytokine IL-1β [31,32]. The activation of inflammatory caspases by various type III injected effectors of Salmonella, notably those activating the small GTPase Rac, largely accounts for the induction of inflammatory responses triggered by enteric epithelial cells [33–35]. The means by which pathogenic bacteria overcome inflammatory responses, notably those driven by caspases, and succeed in infecting their hosts largely remains to be elucidated.
Here, we investigate the manner by which virulence factors of UPEC and innate immune signaling pathways impact the outcome of bacteremia. To this end, we focused on the role of CNF1 and HlyA, two toxins produced by UPEC, on pathogen burden in the bloodstream and on animal survival.
We first assessed the role of the CNF1 toxin in determining E. coli burden during the course of bacteremia in the absence of interference from the other toxin, HlyA. For this purpose, we generated both an hlyA- deletion mutant (referred to as E. coli CNF1+) and an hlyA-cnf1- double deletion mutant (referred to as E. coli CNF1-) from E. coli WT UTI89. By characterizing the strains at the genetic and functional levels, we determined that the two mutants and the wild-type strain had identical growth properties (S1 and S2 Figs). BALB/c mice were then infected intravenously with E. coli CNF1+ or E. coli CNF1- isogenic strains, and the pathogen load was monitored by the serial dilution of blood samples and the enumeration of CFUs (Fig. 1A). We found that the kinetics of clearance from the bloodstream of the E. coli CNF1+ strain was very different (Fig. 1A). The E. coli CNF1+ strain was rapidly cleared, with no bacteria detectable at 48 h p.i. compared with E. coli WT and E. coli CNF1-, which produced 104 and 103 CFU/mouse, respectively, at 48 h p.i. (Fig. 1A). We next assessed whether the rapid clearance of the E. coli CNF1+ strain was actually due to the enzymatic activity of CNF1. We tested this hypothesis by complementing the E. coli CNF1- strain with either an expression vector of wild-type CNF1 (E. coli CNF1- pcnf1) or an expression vector of the catalytically inactive mutant CNF1 C866S (E. coli CNF1- pcnf1 C866S) (Fig. 1B). We found that E. coli CNF1- pcnf1 bacteria were cleared more rapidly from the blood than E. coli CNF1- pcnf1 C866S (Fig. 1B). Together, these results indicate that CNF1 activity promotes the eradication of bacteria from the bloodstream.
To discern whether there is a link between the effects of CNF1 on pathogen burden and strain virulence, we monitored the deaths of infected animals. To this end, E. coli CNF1- bacteria were injected at a dose sufficient to kill half of the mice by 48 h p.i. Animal survival following injection with the different isogenic mutants was monitored (Fig. 1C). We found that all of the mice infected with E. coli CNF1+ survived, whereas the group of mice infected with E. coli CNF1- displayed only 57% survival (Fig. 1C).
Taken together, our data establish that CNF1 activity has a detrimental effect on the bacterial burden in the blood and that it protects against pathogen-induced animal death.
We hypothesized that CNF1 activity has a negative impact on bacterial burden via the modulation of LPS-driven antimicrobial host responses. We assessed this conjecture by profiling the cytokines and chemokines secreted by primary monocytes isolated from the blood of mice after various experimental treatments. The monocytes were challenged with ultrapure LPS, with CNF1 alone, or with a combination of both factors. We used an unbiased approach that utilized an ELISArray semi-quantitative cytokine/chemokine screen to measure the levels of the following factors: IL-1β, TNFα, KC, IL-6, IL-1α, MIP1α, MIP1β, RANTES, MCP1, IL-12, MDC, MIG, IL17, IP10, TARC, EOTAXIN, IL-2, IL-4, IL-5, IL-10, IL-13, IL-23, INFγ, TNFβ1, GM-CSF, and G-CSF. Fig. 2A shows the panel of cytokines that were synergistically induced by LPS+CNF1 compared to ultrapure LPS or CNF1 alone. The results are presented as fold inductions compared to untreated monocytes for each treatment condition Figs. (2A and S3). The panel of cytokines produced by the monocytes treated with CNF1 alone is presented (S4A Fig). Other molecular mediators, such as the IL-4 and IL-10 anti-inflammatory cytokines, showed no significant induction in the monocytes treated with LPS+CNF1 (Fig. 2A). In support of our in vitro analysis results, we measured higher increases in these inflammatory mediators in the sera of mice infected with E. coli CNF1+compared with E. coli CNF1- (S4B Fig). Collectively, these results show that CNF1 activity potentiates the LPS-triggered secretion of the pro-inflammatory cytokines IL-1β, TNFα, and IL-6 primarily, as well as the secretion of the chemokines MCP1, MIP1α, MIP1β, and KC.
We next performed quantitative analysis of the impact of CNF1 activity on the monocyte responses to LPS. Primary monocytes isolated from the blood of naïve mice were treated with endotoxin-free CNF1 or with the catalytically inactive mutant CNF1 C866S. In the monocytes treated with recombinant purified CNF1, we observed the production of KC (75 +/- 5 pg/ml), which was strictly dependent upon the activity of CNF1 (Fig. 2B). Control experiments with ultrapure LPS alone or in combination with the catalytically inactive mutant CNF1 C866S triggered the secretion of KC (120 +/-10 pg/ml). Strikingly, we observed a 3-fold increase in the production of KC (350 +/-10 pg/ml) in the cells treated with both LPS and CNF1 compared with those treated with ultrapure LPS alone (Fig. 2B). The co-stimulation of monocytes with CNF1 and LPS resulted in a 12-fold increase in IL-6 secretion and a 2-fold increase in IL-1β secretion compared with stimulation with LPS alone (Fig. 2C and D). This synergy was detected with doses of CNF1 as low as 10 ng/ml and was of the same magnitude as that observed with the ATP treatment (S4C Fig). We also observed that CNF1 acts synergistically with other Toll-like receptors (TLR) ligands, such as Pam3CSK4 (TLR1/TLR2 agonist) and FSL-1 (TLR2/TLR6 agonist) (S4D Fig). IL-1β is an important mediator of inflammatory responses and is notably important in enabling the host to mount an efficient antibacterial immune response [36,37]. As a first approach, to evaluate the role of IL-1β in the elimination of pathogens, we treated infected mice with an IL-1β antagonist (KINERET). We found that this treatment dramatically antagonized the clearance of E. coli CNF1+ (Fig. 2E).
Collectively, our data pointed for the importance of the synergic induction of IL-1β by LPS+CNF1 in promoting efficient bacterial clearance from the bloodstream.
Given the critical role of IL-1β signaling in the elimination of bacteria, we next aimed to precisely determine the components required for IL-1β maturation. IL-1β is expressed as a proform (proIL-1β) that is processed by caspases-1/11 to generate the mature p17-secreted active form [38,39]. We further assessed the effects of the interplay between LPS and CNF1 on IL-1β maturation. This interaction was evaluated by immunoblotting to determine the level of secretion of p17 IL-1β into the medium of monocytes upon co-stimulation with LPS+CNF1 (Fig. 3A and B). Our results confirmed that CNF1 acts at the level of IL-1β maturation/secretion rather than at the level of IL-1β expression (Fig. 3A and B). We observed the complete inhibition of the release of p17 IL-1β in the monocytes treated with the pan-caspase inhibitor QVD as well as in the monocytes isolated from caspase-1/11 (C1-C11)-impaired mice (Fig. 3A and B). Notably, these results indicate that CNF1 plays a critical role in promoting the caspase-1/11-dependent maturation/secretion of IL-1β by monocytes challenged with LPS. We next assessed the interplay between inflammatory caspases and CNF1 during UPEC-induced bacteremia. To this end, we measured bacterial loads in the blood of C1-C11-impaired mice infected with E. coli CNF1+ or E. coli CNF1-. We compared the kinetics of the bacterial burden in these animals with those of their wild-type congenic C57BL/6 littermates. In the wild-type animals, we measured a decrease in the bacterial load in the animals infected with E. coli CNF1+, with no bacteria detectable in the blood at 48 h, compared with the presence of 107 CFU/animal in the blood of the mice infected with E. coli CNF1- (Fig. 3C). In contrast with the wild-type mice, the E. coli CNF1+ burden in the C1-C11-impaired mice remained high, reaching 105 CFU/animal at 48 h p.i. (Fig. 3D). These results indicate that inflammatory caspases-1/11 play major roles in the clearance of bacteria from the bloodstream as triggered by CNF1.
In an approach designed to be complementary to our functional approach, we analyzed the roles of inflammatory caspases in the initiation of the CNF1-dependent innate immune responses during bacteremia. To this end, we measured the levels of the IL-1β and KC cytokines in the sera of infected C1-C11-impaired mice and their congenic wild-type littermates. The wild-type infected with E. coli CNF1+ displayed higher levels of IL-1β and KC than the WT mice infected with E. coli CNF1- (Fig. 3E and F). Interestingly, in the C1–11-impaired mice infected with E. coli CNF1+, we measured dramatic decreases in the levels of KC and IL-1β in the sera compared with the wild-type mice (Fig. 3E and F). This finding is consistent with the fact that inflammatory caspases-1/11 are critical determinants in the CNF1-triggered cytokine response during bacteremia. Next, we aimed to determine whether the CNF1-triggered IL-1β secretion involved caspase-1 or caspase-11 using bone marrow-derived macrophages isolated from caspase-1 or caspase-11 single knock-out mice. Additionally, we analyzed the effect of CNF1 on macrophages of mice in which the inflammasome adaptor ASC was knocked out. We measured the inhibition of CNF1-triggered IL-1β secretion in macrophages isolated from caspase-1 or ASC but not caspase-11 knock-out mice (Fig. 4A). Thus, our analysis pinpointed the major roles of caspase-1 and ASC in the secretion of IL-1β triggered by CNF1+LPS. In addition, we measured that the activity of caspase-1 increased when monocytes were treated with LPS+CNF1 (Fig. 4B). We then investigated the role of the GTPase Rac in IL-1β secretion triggered by CNF1+LPS. To this aim, we expressed a mutant of Rac2 bearing the modification catalyzed by CNF1 (Rac2Q61E). We observed that the expression of Rac2Q61E in the macrophages challenged with LPS promoted the secretion of the p17 form of IL-1β (Fig. 4C). Finally, we were able to detect an association between Rac2 and caspase-1 upon CNF1+LPS treatment specifically by co-immunoprecipitation (Fig. 4D).
Our results identify the Rac/caspase-1/IL-1β signaling axis as a major component of the CNF1-induced elimination of Escherichia coli from the bloodstream.
We sought to determine the manner by which pathogenic bacteria cope with anti-virulence immunity. Although HlyA has been shown to interfere with innate immune responses that occur during urinary tract infections, its role during bacteremia is still unknown. We experimentally addressed this question by analyzing the effect of HlyA on CNF1-triggered protection against bacteremia in mice. We observed that all E. coli strains expressing HlyA displayed increased stability in the blood compared with the other strains, independent of the presence or absence of CNF1 (Fig. 5A). We noticed a reduced bacterial burden in the E. coli CNF1- strain compared with E. coli HLY+ CNF1-. This reduction in the bacterial burden was the greatest in the E. coli CNF1+ strain. We next complemented the E. coli CNF1+ strain with a plasmid encoding HlyA (E. coli CNF1+ phlyA). We found that the complementation of E. coli CNF1+ with the HlyA expression plasmid stabilized the bacterial load in the blood (Fig. 5B). These results show that HlyA protects E. coli against host responses, particularly those triggered by CNF1, thereby promoting bacterial stability in the blood.
We went on to determine at which level HlyA acts to prevent CNF1-triggered bacterial clearing. We first investigated whether HlyA has a blocking effect on the activation of the GTPase Rac by CNF1. We detected no modification in the CNF1-mediated activation of Rac when cells were co-incubated with HlyA (S5 Fig). Thus, we hypothesize that HlyA acts downstream of the activation of Rac at the level of cytokine production/secretion. Consistent with this hypothesis, we measured lower levels of IL-1β and KC in the bloodstream of the mice infected with HlyA-expressing E. coli strains (Fig. 5C and D).
Taken together, our data indicate that HlyA inhibits CNF1-induced pro-inflammatory cytokine responses.
Next, we aimed to further identify the key immune effector cells that control the rapid clearance of E. coli exacerbated by the Rho-activating toxin CNF1 during bacteremia. We monitored the levels of circulating innate immune cells in the blood at an early time period of infection with either the E. coli CNF1+ strain or the E. coli CNF1- strain. The data were analyzed as the percent of CD45-positive cells, a white blood cell marker, to exclude contamination by red blood cells. We first monitored circulating innate immune cells, including monocytes, neutrophils and granulocytes, using the CD11b marker. Interestingly, we measured higher percentages of CD11b+/CD45+ cells at both 3 h and 6 h p.i. in the blood of the mice infected with E. coli CNF1+ (3 h: 46% and 6 h: 64%) compared with E. coli CNF1- (3 h: 23% and 6 h: 43%). Control mice injected with PBS showed lower levels of CD11b+/CD45+ (3 h: 18% and 6 h: 22%) (Fig. 6A). We found that chemokines, such as MIP1α, MIP1β, MCP-1, and RANTES as well as KC were secreted by the monocytes treated with CNF1 in vitro and were increased in the sera of the mice at 3 h p.i. for the E. coli CNF1+ strain compared with E. coli CNF1- (S4B Fig). Because these chemokines are involved in chemotaxis as well as in the activation of neutrophils, we hypothesized that the clearance of the bacteria was due to cooperation between inflammatory monocytes and neutrophils. To test this hypothesis, we monitored the subpopulation of Gr1+ cells, including inflammatory monocytes and neutrophils, in the blood of infected mice. Mice infected with E. coli CNF1+ showed 37% and 58% Gr1+CD45+ cells at 3 h and 6 h, respectively, compared with only 21% and 40% when the mice were infected with E. coli CNF1- (Fig. 6B). These findings indicated that there was an increased recruitment of Gr1+ cells triggered by CNF1. The recruitment of Gr1+ cells was not reduced in the C1-C11 impaired mice (S6A Fig). Furthermore, the recruitment triggered by CNF1 was not impaired in HlyA-expressing E. coli (S6B Fig). To demonstrate the key role of Gr1+ cells in the clearance of E. coli expressing CNF1, we depleted this subpopulation, including inflammatory monocytes (Gr1+ F4/80+) and neutrophils (Gr1+ F4/80-), prior to infection. We measured an 80% reduction in the Gr1+ cell population following the injection of anti-Gr1+ (Ly-6G) monoclonal antibodies (RB6–8C5) (S7 Fig). We found that the depletion of Gr1+ cells was sufficient to block E. coli clearance during bacteremia and to prevent the rapid clearance of the E. coli CNF1+ strain (Fig. 6B).
Altogether, we show that Gr1+ recruitment is required but is not sufficient for the anti-virulence immunity triggered by the Rho GTPase-activating toxin CNF1. In addition, we demonstrate that Gr1+ cells are critical for the clearance of E. coli from the bloodstream.
The sensing of the activities of pathogen-encoded virulence factors is emerging as a paradigm of innate immune sensing. However, in vivo proof of the contribution of such sensing to mammalian immunity during infection is still scarce. Furthermore, the mechanisms by which pathogenic bacteria cope with the capacities of hosts to detect their virulence remain to be elucidated. As a major discovery, we demonstrated here the capacity of the host to control bacteremia through the exacerbation of LPS-driven IL-1β-mediated antimicrobial responses by CNF1 activity. This host feature relies on Rac/ASC/caspase-1 and the secretion of pro-inflammatory cytokines/chemokines, which in turn mobilize Gr1+ cells. Importantly, we described a yet unappreciated role of HlyA in impairing these innate immune responses downstream or in parallel with Gr1+ cell recruitment. We also established that pathogen burden and animal death were maximal for the HlyA-positive strains.
Inappropriate, excessive or absent innate immune responses have dramatic consequences on human health. Thus, it is critical to decipher the manner by which the host determines the pathogenic potential of a microbe and responds commensurately [40,41]. It is currently unclear how anti-virulence immunity (AVI) systems of detection work together with the recognition of MAMPs to control inflammation and bacterial virulence. Because CNF1 intoxicates cells without the involvement of additional bacterial factors, it can be used to address this critical question. In this work, we report that the detection of CNF1 activity amplifies the cellular LPS response to a large panel of pro-inflammatory cytokines, including IL-1β, by 2- to 12-fold, thereby producing a more potent immune response. Analysis of the IL-1β level in the sera of mice infected with E. coli CNF1+ indicated that they exhibited a 3-fold higher response and better resistance to infection than those infected with isogenic E. coli CNF1-. Our study provides both in vitro and in vivo evidence that AVI works in concert with MAMP-triggered responses to amplify the innate immune response and ultimately improve host viability. We speculate that this cooperation between AVI and MAMP-triggered immunity is a means by which the host gauges the pathogenic potential of a microbe and tailors a response commensurate with the estimated threat level.
The sensing of the activities of bacterial virulence factors has recently emerged as a conserved means of detecting pathogens. Rho GTPases are targeted by various virulence factors encoded by pathogenic bacteria. These virulence factors either post-translationally modify Rho GTPases by deamidation, glucosylation, adenylylation, or ADP-ribosylation or mimic exchange factors or GTPase-activating proteins, thereby hijacking the GTP/GDP cycle and causing inappropriate activation or inactivation of the critical regulators of these cycles, which include Rho, Rac and Cdc42 GTPases [21]. Interestingly, recent studies have indicated that animal hosts have evolved dedicated strategies for detecting the activities of these virulence factors [26,27,42,43]. Indeed, based on our work focusing on CNF1 and studies of Salmonella typhimurium SopE/E2 virulence factors, we can speculate that the abnormal activation of Rac/Cdc42 triggers the assembly of an anti-virulence immune complex involving NOD1 and RIP kinases to promote NF-κB activation and in parallel to assemble a Rac/ASC/caspase-1 complex for the maturation of IL-1β during infections [27,35]. In addition, a recent study has implicated the NLR pyrin as a sensor of the inactivation of the Rho GTPase RhoA by virulence factors via a mechanism that leads to the activation of the pyrin inflammasome and inflammatory caspase-1 [42]. Taken together, these studies indicate that, in parallel with the PRR detection of MAMPS, the host monitors changes in the GTP/GDP cycles of Rho GTPases rather than monitoring each post-translational modification individually, a process that would require a large repertoire of receptors.
Our observation that CNF1 induces an immune response that is detrimental to bacteria raises the question of why CNF1 has been evolutionarily conserved in the UPEC genome. Several reports have established that the CNF1 toxin can trigger the disruption of epithelial cell junctions, promote cell migration and induce the internalization of bacteria into epithelial cells [25]. One hypothesis is that CNF1 has been evolutionarily conserved as an invasion factor to help bacteria cross epithelia during the early stages of infection. Our results led us to speculate that CNF1 has become genetically associated with HlyA to protect the bacteria from an otherwise detrimental CNF1-induced innate immune response. Indeed, the CNF1 and HlyA toxins are co-transcribed within a highly conserved PAI, and epidemiological studies have established that CNF1 is always expressed in association with HlyA [12,19,44,45]. Interestingly, the functional relationship between these toxins described here offers a new framework through which to understand the molecular basis of their tight genetic link and explains why E. coli that express both of these toxins are pathogenic to mammals. Consistent with this idea, our work sheds light on the manner by which pathogenic bacteria cope with AVI. We report a yet unappreciated role of HlyA in the impairment of innate immune responses. In this role, HlyA has major influences on bacterial burden and host viability. Our genetic analysis revealed that HlyA protects microbes from both CNF1-dependent and CNF1-independent detrimental effects. Mice infected with E. coli expressing CNF1, but not HlyA, showed an increase in the IL-1β and KC proinflammatory cytokine levels. Given that the bacterial load is minimal under these conditions, this finding cannot be ascribed to increased LPS exposure. It is possible that HlyA targets host immune cell signaling to prevent the production of inflammatory cytokines. This idea is supported by our findings that HlyA did not impair the CNF1-induced activation of Rac, although it blocked the secretion of IL-1β. Further, we showed that bacteria expressing HlyA, but not CNF1, showed an increase in persistence of one log unit in the blood compared with bacteria that were deficient in both toxins. Although HlyA acted primarily to counteract the host recognition of CNF1 activity in our model, it most likely has additional effects on other components of the innate immune response to E. coli, including phagocytosis or detection by the immune system of other bacterial components [18]. A common feature of pathogenic bacteria is the production of a wide range of pore-forming toxins of various sizes, which have specific ionic and molecular selectivities. It will be important to establish which types of pore-forming toxins are able to block innate immunity and to what extent HlyA blocks the recognition of other factors produced by E. coli.
Multicellular organisms have evolved sophisticated defense mechanisms to counter microbial attack. In turn, successful microbial pathogens have evolved strategies to overcome host defenses, leading to the occurrence of diseases or chronic infections [46]. In plants, a similar system of detection of the activities of virulence factors has been termed “effector-triggered immunity” [28]. Interestingly, in this model, the pathogen-evolved mechanism counteracting the innate immune defense response has been called a “counter-defense mechanism” [47]. In our model, HlyA counteracts the CNF1-induced host cytokine response. By analogy, if we consider that CNF1 is sensed by the innate immune defense system, HlyA must be considered as a counter-defense effector used by E. coli to counteract the CNF1-induced host response. The data presented in the present work support a model in which HlyA acts as a major virulence factor that protects microbes from both CNF1-dependent and CNF1-independent innate immune defenses during bacteremia. Based on this model, pore-forming toxins might represent viable drug targets for the treatment of UPEC bacteremia.
This study was carried out in strict accordance with the guidelines of the Council of the European Union (Directive 86/609/EEC) regarding the protection of animals used for experimental and other scientific purposes. The protocol was approved by the Institutional Animal Care and Use Committee on the Ethics of Animal Experiments of Nice, France (reference: NCE/2012–64).
The E. coli UTI89 clinical isolate was originally obtained from a patient with cystitis [48] and was a kind gift from E. Oswald. The UTI89 streptomycin-resistant (SmR) evolved strain (WT) and isogenic mutants were grown in Luria-Bertani (LB) medium supplemented with streptomycin (200 μg/ml). The CNF1 strain was transformed with a pQE30 plasmid (Qiagen) (E. coli CNF1- pempty), with pQE30-CNF1 (E. coli CNF1- pcnf1 WT) or with pQE30-CNF1 C866S (E. coli CNF1- pcnf1 C866S) and grown in LB supplemented with ampicillin (100 μg/ml) plus IPTG (200 μM) for the infection experiments. The E. coli CNF1+ strain was transformed with pBR322 (E. coli CNF1+ pcontrol) or with pEK50 (a plasmid bearing an operon encoding HlyA (hlyCABD)) (E. coli CNF1+ phlyA) and grown in LB supplemented with ampicillin (100 μg/ml). The DH10B K12 E. coli strain (Life technologies) was transformed with either pEK50 (K12-pHlyA) (a plasmid bearing the operon encoding HlyA (hlyCABD)) or the pCR2.1 (K12-pLacZ). The pEK50 plasmid and anti-HlyA antibody were a kind gift from V. Koronakis. The anti-TolC antibody was a kind gift from C. Wandersman. For the infections, a 1/50 dilution of an overnight culture of each strain was inoculated and grown to OD600 = 1.2. Bacteria were either washed in culture medium and diluted to obtain the corresponding MOI for the cell culture infection experiments or were harvested by centrifugation and washed twice in PBS before dilution in PBS to obtain the desired bacterial concentrations for the mouse infection experiments. Recombinant wild-type cytotoxic necrotizing factor-1 (CNF1) and its catalytically inactive form (CNF1-C866S; CNF1 CS) were produced and purified as previously reported [49]. The recombinant proteins were passed through a polymyxin B column (AffinityPak Detoxi-Gel, Pierce). The lack of endotoxin content was verified using a colorimetric LAL assay (LAL QCL-1000, Cambrex). Each stock of the CNF1 preparation (2 mg/ml) was shown to contain less than 0.5 endotoxin units/ml.
The multi-step procedure used to substitute the hlyA and cnf1 genes in the bacterial chromosome was performed as previously described [50]. Briefly, the pMLM135 plasmid (cat, rpsl+) was used to transform the UTI89 streptomycin-resistant (SmR) evolved strain. The integration of pMLM135 into the chromosome was selected by plating cells on chloramphenicol-containing medium at 42°C. Excision of the hlyA or cnf1 gene from the chromosome was selected by plating cells on medium containing streptomycin (200 μg/ml). The chromosomal deletions were verified by PCR and by the monitoring of the loss of HlyA and/or CNF1 activity in the deleted strains (S1 Fig). We verified that the isogenic mutant strains had growth properties that were identical to those of the UTI89 strain (S2 Fig). The sequences of the primers used in this study are available upon request.
Murine monocytic cells were obtained from pooled blood from 5–10 mice. Monocytes were isolated using a Ficoll-Paque (GE Healthcare) gradient technique. Adherent cells were maintained in M medium [RPMI 1640 medium supplemented with 10% FCS (Lonza), 2 mmol/L L-glutamine, 1 mM pyruvate, 10 mM HEPES, penicillin (100 U/ml), and streptomycin (100 μg/ml)]. When indicated, M-CSF was added as previously described [51]. Cells were transfected using the Amaxa mouse macrophage nucleofector kit according to the manufacturer’s instructions with pCDNA3.1-HA-Rac2Q61E, pRK5-Myc-Rac2, pCAGGS-Caspase-1, pCDNA3.1 or pRK5. Monocyte isolation was confirmed by flow cytometry analysis using F4/80 and CD11b antibodies (Cedarlane). HEp-2 cells and HEK 293T cells were obtained from ATCC (CCL-23 and CRL-3216) and maintained according to ATCC instructions. HEK 293T cells were transfected using Lipofectamine 2000 (Life technologies) according to the manufacturer’s instructions.
Female BALB/c and C57BL/6 mice (6–8 weeks old) were purchased from Janvier (Le Genest St Isle, France). Caspase-1/11-impaired (also designated as ICE KO) and congenic C57BL/6 mice have been previously described and were kindly provided by R. Flavell [52]. These mice are genetically identical to mice that are now also available from Jackson Laboratories (Stock #016621). Caspase-11 knock-out mice have been previously described by VM Dixit (Genentech). Caspase-1 knock-out mice were generated by D. Dieter and M. Lamkanfy. Their generation will be described elsewhere. Mice were injected i.v. with 107 CFUs of E. coli as previously described [8,53]. For the determination of bacteremia, blood was collected from the tail vein at the indicated times post-infection, serially diluted in sterile PBS and plated on LB plates containing streptomycin (200 μg/ml) or ampicillin (100 μg/ml) for the strains transformed with pQE30- or the pBR322-derived plasmids, and the plates were incubated for 16 h at 37°C. Injection quality was controlled by plating blood samples obtained from the mice at 5 min after injection. Note that the kinetics for the experiments using the transformed strains were terminated after 24 h because we observed that without selective pressure, the plasmid was stable for up to 24 h. For cytokine analysis, plasma was collected (1200×g, 4°C, 5 min) and stored at −20°C.
Mice were injected intraperitoneally with a monoclonal anti-Gr1 antibody (RB6–8C5, 100 μg/20 g body weight). After 48 h, the depletion of Gr1+ cells was verified in four mice by analyzing F4/80 and/or Gr1-stained white blood cells by flow cytometry. The anti-Gr1-injected mice were then infected with either UTI89 or UTI89 isogenic mutants.
ELISArrays were performed according to the manufacturer’s instructions (Qiagen, MEM-003A, MEM-004A, MEM-006A, MEM-008A, and MEM-009A). Cytokine concentrations were determined by ELISA and by IL-1β maturation and visualized by western blotting according to the manufacturer’s instructions (KC, TNFα and IL-6, R&D Systems, USA; IL-1β, Raybiotech, USA). Caspase-1 activity was measured using FAM-YVAD-FMK caspase-1 assay kit (ImmunoChemistry Technologies) according to the manufacturer’s instructions. Antibodies used in this study are: anti-IL-1β/IL-1F2 (R&D systems), anti-Caspase-1 (M-20, SantaCruz), anti-Caspase-11 (17D9, Novus), anti-ASC (AL177, Adipogen), anti-HA (16B12, Covance), anti-Myc (9E10, Roche), anti-Rac (102/Rac1, BD Biosciences), anti-GAPDH (FL335, SantaCruz), anti-β-actin (AC-74, Sigma).
Primary monocytes (5x106 cells per condition) were treated with 1 μg/ml of CNF1 toxin for 6 h with or without addition at identical MOI (MOI of 0.5) of live K12 E. coli expressing HlyA (K12-pHlyA) or LacZ (K12-pLacZ) as a control. Cells were lysed at 4°C using a lysis buffer (Tris 25mM pH 7.5, NaCl 150mM, MgCl2 5mM, EGTA 0.5mM, TritonX100 0.5%, glycerol 4%) and Pull-down assays were performed using 50 μg of GST-PAK70–106. Proteins were resolved on 12% SDS-PAGE followed by transfer on PVDF membranes. Equal amount of proteins engaged in the Pull-down assays was confirmed by immunoblotting anti-β-actin.
Statistical analyses were performed using Prism V5.0b software (GraphPad, La Jolla, CA). Unless stated otherwise, comparisons between two groups were performed using the Mann-Whitney nonparametric test, and comparisons among three or more groups were conducted with the Kruskal-Wallis test with Dunn's post-test. P-values<0.05 (*) and P-values<0.01 (**) were considered statistically significant.
|
10.1371/journal.pgen.1002469 | Familial Identification: Population Structure and Relationship Distinguishability | With the expansion of offender/arrestee DNA profile databases, genetic forensic identification has become commonplace in the United States criminal justice system. Implementation of familial searching has been proposed to extend forensic identification to family members of individuals with profiles in offender/arrestee DNA databases. In familial searching, a partial genetic profile match between a database entrant and a crime scene sample is used to implicate genetic relatives of the database entrant as potential sources of the crime scene sample. In addition to concerns regarding civil liberties, familial searching poses unanswered statistical questions. In this study, we define confidence intervals on estimated likelihood ratios for familial identification. Using these confidence intervals, we consider familial searching in a structured population. We show that relatives and unrelated individuals from population samples with lower gene diversity over the loci considered are less distinguishable. We also consider cases where the most appropriate population sample for individuals considered is unknown. We find that as a less appropriate population sample, and thus allele frequency distribution, is assumed, relatives and unrelated individuals become more difficult to distinguish. In addition, we show that relationship distinguishability increases with the number of markers considered, but decreases for more distant genetic familial relationships. All of these results indicate that caution is warranted in the application of familial searching in structured populations, such as in the United States.
| The forensic identification of criminal suspects through DNA profiling is now common in the United States. Indirect identification by familial DNA profiling is increasingly proposed to extend the utility of DNA databases. In familial searching, a DNA profile from a crime scene partially matches a database profile entry, implicating close relatives of the partial match. While the basic principles behind familial searching methods are simple and elegant, statistical confidence that a partially matched profile belongs to a true genetic relative has not been fully explored. Here, we derive relative identification likelihood ratio statistics and consider how the ability of familial searching to distinguish relatives from unrelated individuals varies over population samples and is affected by inaccurately assumed population background. We observe lower relationship distinguishability for population samples with less identifying information in the genetic loci considered. Additionally, we show that relationship distinguishability decreases with discordance between true and assumed population samples. These results indicate that, if an inappropriate genetic population group is assumed, individuals from certain marginalized groups may be disproportionately more often subject to false familial identification. Our results suggest that care is warranted in the use and interpretation of familial searching forensic techniques.
| Forensic identification via exact genetic profile matching has become common practice in the United States [1]. In exact genetic identification, genetic markers found in a crime scene sample are genotyped and exactly matched to a suspect or database entry, suggesting that the sample originates from the matched individual. In some cases, a database search yields no exact genetic profile matches, but does reveal partial matches where some, but not all, alleles match. A partial match could result from a genetic familial relationship between the individual who left the sample and the database entrant. If the database entrant has relatives, they might be investigated to determine if any of their genetic profiles exactly match the sample.
Familial searching is now used fairly frequently in the United Kingdom and was instrumental in the identification of suspects of violent crimes for 20 cases lacking other evidence as of 2008 [2]. Its use in the United States has been more limited due to concerns regarding civil liberty infringement, racial bias, and efficacy [3]–[6]. However, in July 2010, familial searching was used in a highly publicized California case to identify a suspect serial killer (the “Grim Sleeper”) [7]–[10].
Despite the increasing use of familial searching in the United States, important questions about the method remain on both social and scientific grounds. In order to understand these concerns, we must appreciate that familial searching is most useful as a database mining method in cases with no suspects. In the United States, the Combined DNA Index System (CODIS) is the Federally administered system for National DNA Index System (NDIS), the national offender/arrestee database, which includes entries from State DNA Index Systems [11]. CODIS has standardized the use of genotypes at 13 particular short tandem repeats (STRs) (the CODIS loci) in forensic identification. The CODIS loci were chosen based on several criteria including reliable multiplexed PCR amplification, availability of commercial genotyping kits, clearly distinguishable alleles, linkage equilibrium, Hardy-Weinberg equilibrium, and high polymorphism in examined population samples [12]–[15]. An NDIS entry contains CODIS loci genotypes and a traceable index number, without other identifying information (e.g. location, race, or ethnicity) [16]. In September 2011, NDIS included over 10 million genotype profiles and continues to grow through new cases and expanded inclusion criteria [1].
These features of the forensic testing landscape matter because, unlike exact DNA identification, a typical database search for familial matches prospectively identifies candidate suspects who, while closesly genetically related to database entrants, are not in themselves in the database, provoking complex privacy concerns [4], [5], [9], [17], [18]. Additionally, social groups which both share genetic relationships and are over-represented in the database would experience a disproportionate increase in genetic surveillance if familial matching were routinely implemented, further exacerbating their over-representation in these databases [6], [12], [17]–[19].
The question of relative inference has been well-studied in other contexts with varying marker types, relationships, and numbers of individuals [20]–[28]. Here we focus on statistical and population genetic assumptions underpinning the familial searching methodology in the forensic context. Specifically, we consider the effects of both uncertainty in allele frequency estimation and population structure. First note that allele frequency estimates calculated within socially-defined population groups (e.g. African American, European American, Latino) are used to estimate the probability of an observed partial match, assuming a particular genetic relationship. Match probabilities for some individuals may not be accurately estimated using the available categorical socially-defined population group model and sample allele frequency data, particularly individuals with genetic ancestry outside of typically studied groups or individuals whose socially-defined population group does not inform their genetic ancestry. In exact identification, the probability of observing two individuals with identical specific 13-locus genotypes is astronomically low, with the exception of monozygotic twins. With these extremely low probabilities, differences or inaccuracies in allele frequency estimates are almost inconsequential, possibly changing the probability of an observed genotype a few orders of magnitude, but unlikely to alter the conclusion of the statistical analysis [29]. However, in familial identification, the probability of observing a coincidental partial match is much higher (e.g. for a parent-offspring relationship exactly one allele is shared by descent per locus). With these higher probabilities, population genetic differences in marker informativeness and errors in allele frequency estimation can perturb match probability estimations to such a degree as to affect the interpretation outcome.
In this study, we aim to examine some of these concerns by exploring how familial searching techniques behave on populations with varying allele frequency distributions and varying accuracy of allele frequency specification. We formulate and calculate confidence intervals for familial identification likelihood ratio (LR) estimates, and investigate how well siblings and unrelated individuals can be distinguished over different population samples with varying allele frequency distributions and under accurately and inaccurately assumed allele frequency distributions. We show that population samples vary in the amount of identifying information encoded in the CODIS loci and, therefore, in relationship distinguishability, even with correctly specified allele frequencies. Since completely accurate allele frequency specification is not guaranteed and the most appropriate population sample may not be known or available, we are also interested in the systematic effects of assuming allele frequencies which are appropriate for one population, but which are not appropriate for the individuals investigated. We show that relationship distinguishability decreases with the accuracy of allele frequency estimates, potentially resulting in high rates of coincidental familial identification for some groups. These results are especially pertinent in the multiple testing context of large database searching. In addition, we explore the relationships between relationship distinguishability, the number and type of markers used for identification, the relationship considered, and the true and assumed coancestry coefficient parameter value.
To determine if a partial genotype match is better explained by a genetic familial relationship or stochasticity, we used the ratio of the likelihood of the observed partial match assuming the individuals share a given genetic familial relationship, to the likelihood of the observed partial match assuming the individuals are unrelated. With the data available, this LR is the most powerful statistic to separate relatives from unrelated individuals [30]. So even though the exact methodology used by forensic agencies for familial forensic identification is not readily publicly available, our use of the LR optimistically assumes the most powerful method using the CODIS loci. In the first part of this analysis, only sibling relationships are evaluated to reduce dimensionality. Other genetic familial relationships were explored and are reported below.
Unrelated individuals were simulated in a randomly mating population by independently drawing alleles from allele frequency distributions, similarly to Bieber et al. [31]. Siblings were then simulated by dropping alleles through a pedigree with unrelated parents. We simulated both unrelated individuals and siblings using allele frequency distributions from five socially-defined population samples, Vietnamese, African American, European American, Latino, and Navajo. Using both unrelated individuals and siblings, we calculated the sibling relationship and 95% confidence interval of that estimate, assuming allele frequencies from each population sample. We simulated siblings and unrelated individuals under each of the five allele frequency distributions and calculated and 95% confidence interval of that estimate assuming each of the five allele frequency distributions 10,000 times for each pair of population samples. As a result, we have with confidence intervals for sibling relationships between unrelated individuals and siblings simulated from every population sample, assuming allele frequencies from every population sample. In most of the analyses presented here, we focus specifically on the lower 95% confidence limit of (LCL) to account for sampling and biological variance in allele frequency estimation and to conservatively identify relationships. We refer to the population sample used to simulate the individuals as the true population sample, as opposed to the assumed population sample used to calculate the LR for their relationship. Figure S1 shows the 95% confidence intervals for 100 simulations of unrelated individuals, where individuals were simulated based on each population sample and confidence intervals were computed assuming the allele frequency distribution of each population sample.
Note that across all of these simulations specific parameter values were chosen and kept constant, specifically, sibling relationships, the assumed coancestry coefficient (probability of two alleles being identical by descent (IBD) between two individuals not recently related) used in calculations of , confidence interval length parameterized by significance level as , and the use of the 13 CODIS STRs. Regardless of the values of these parameters, the relative trends across true and assumed population samples will be maintained, although the scale may vary with parameter value choice.
We observed lower distinguishability when the true and assumed allele frequency distributions differ more. The degree of difference between population sample allele frequency distributions at the CODIS alleles is quantified for every population pair using (Table 2). To account for multiple alleles at multiple loci and varying sample sizes, we estimate with the method of Weir and Cockerham [33]. Note that s reported here were calculated using the only CODIS loci, as is appropriate for an analysis of forensic methods. For a thorough investigation of the population genetics of these samples, more loci would be required, producing different results than those shown here, as reported in other studies [34], [35].
To explore the relationship between distinguishability and the genetic distance between true and assumed population samples, in Figure 4, is plotted against for each pair of true and assumed population samples. and are significantly correlated (), supporting the hypothesis that incorrectly assuming allele frequencies leads to low distinguishability and high false positive rates. In particular, we observe low distinguishability when Navajo, or to a lesser extent Vietnamese, is the true population sample, correlating with higher with the other assumed samples.
Intuitively, when allele frequencies are misspecified, the most likely error is assuming that common alleles are more rare simply because truly common alleles are more likely to be observed than truly rare alleles. In the same way, rare alleles are assumed to be common, but by definition, rare alleles are less likely to be observed shared between individuals, so overall the misspecification of common alleles as rare dominates. When misspecifying common alleles as rare, observing the same common alleles in multiple individuals seems surprising, so a genetic relationship model is favored over a model of no relationship. That is, the probability of a partial match assuming a relationship is inflated and the probability of a partial match assuming no relationship is deflated. In this way, allele frequency misspecification results in an increase in false positive relative identifications.
Although the relationship between distinguishability and allele frequency misspecification has not yet been deeply considered in the context of genetic familial identification (but see [36]), it has been discussed in the forensic literature for exact matching and it is well-known in the linkage analysis community. For exact forensic identification using the 13 CODIS loci, discrepancies between assumed and true allele frequencies affect the computed match probabilities, but seldom change the ultimate outcome [37]–[40]. In linkage analysis, when inaccurate population allele frequencies are used to calculate genotype probabilities, false linkage signals between genotype and phenotype are common [41], [42].
In the analysis presented thus far, we showed how distinguishability varies over true and assumed population samples with varying allele frequency distributions. To maintain manageable dimensionality, some key parameters likely to vary in forensic analyses were kept constant. Here we explore the relationships between these parameters, particularly different genetic relationships, varying marker data, and varying the true and assumed coancestry coefficients ( and ). To focus on the relationships between these parameters, in these analyses the correct known allele frequencies were used.
Pairs of individuals were simulated taking into account the true coancestry coefficient, , using the genotype probabilities described in the Text S1, for the following genetic relationships: parent-offspring, sibling, half-sibling, first cousin, second cousin, and unrelated. Note that in contrast with the analyses presented above, here is used to model background relatedness. LRs were computed comparing the probabilities of the simulated data assuming the true relationship and assuming the individuals are unrelated. This analysis was repeated over varying numbers and types markers and a variety of assumed values.
The analysis presented here confirms and quantifies the intuition from population genetics that for particular loci, groups with comparatively low-variance allele frequency distributions have less identifying information encoded in genotypes. Decreased identifying information results in lower relationship distinguishability, even when the correct allele frequency estimates are used (Figure 2, Figure S2). This is abundantly apparent for the Native American samples considered in this analysis.
With a basic understanding of population genetics, it is clear that socially defined groups, like Navajo, Latino, or European American, have very different underlying population structures reflecting distinct demographic history, degrees of genetic diversity, and admixture. It is hardly surprising that a group which has undergone multiple population size reductions, like the Navajo, has a lower-variance allele frequency distribution than a group with a history of genetic diversity and social inclusion, like African Americans. This is particularly evident at the CODIS loci, which were chosen in part because of their broad allele frequency distributions in a few studied populations, without considering all relevant populations [13]–[15].
These population differences in allele frequency distributions are key when considering a potential source of error: inappropriately assumed allele frequency distributions. When the allele frequency distributions for an inaccurately specified population group are assumed, the probabilities of the observed data under a sibling relationship and under no close genetic relationship become less distinct, so relationship distinguishability decreases. We found that distinguishability decreases with increased distance between assumed and true allele frequency distributions, as measured through . Specifically, both Navajo and Vietnamese samples are more genetically distant to the other three samples considered and show decreased distinguishability when allele frequencies of one of those three samples are assumed.
The results of this analysis indicate that when a decision threshold is chosen so that the power to identify siblings is reasonably high, population samples with allele frequencies which differ from those assumed would experience disproportionately higher rates of false positive familial identification (Figure 3). This could be exacerbated by unknown population-based differences in genotyping which would distort allele frequencies, for example, population-specific mutations in PCR primer binding sites [45]–[51]. More extensive genotyping of genetically diverse populations may make available more appropriate allele frequency distributions. However, it is not clear how or if the most appropriate allele frequency distribution for a pair of samples can be determined. Population-based differential distinguishability will persist, regardless of additional population-specific allele frequency distributions or uniformly applied corrections. One possible correction would be increasing the value of the parameter , however, in Figure S6 we see that even when the true allele frequencies are assumed, increasing decreases distinguishability. If more genetic data were used, particularly markers on the Y chromosome or mitochondrial DNA, as are in some states but not Federally, profile informativeness could be increased to the point where allele frequency approximations made little difference in the ultimate outcome (Figure S5) [10], [52]. However, additional Y chromosome and mitochondrial markers will only inform matrilinial or patrilinial relationships and any additional markers will be subject to similar population-specific variation, and will be limited by practical genotyping constraints and the need to avoid medically-associated regions. Additionally, it is not clear if more distant relationships (cousins, second cousins, etc) would be confidently identified, even with more independent genetic loci (Figure S5) [53], [54]. As it is, the core 13 CODIS loci, or the minimum 10 loci recommended by the Scientific Working Group on DNA Analysis Methods Ad Hoc Committee on Partial Matches (SWGDAM), seem inadequate to implement sibling matching with low false positive rate and high power in structured populations [52], [55]. More complex situations, like mixed or low-template DNA samples, require further study and may not be feasible with the 13 CODIS loci [55], [56].
Motivated by the question of forensic familial searching, in this analysis we focus on distinguishing relatives with a specified relationship and unrelated individuals. In other contexts, it may be more appropriate to distinguish different kinds of relatives (e.g. siblings and parent-offspring) or relatives with an unspecified relationship and unrelated individuals. In the former case, the ratio of LRs for the relationships of interest versus unrelated individuals reduces to the LR comparing the two specified relationships. In the later case, models allowing IBD sharing probabilities to vary can be formulated and incorporated into the LR. For example, when comparing a null model with set IBD sharing probabilities for unrelated individuals and an alternative where the likelihood of data is maximized over any IBD sharing probabilities, a LR test can be formulated which follows a distribution under the null hypothesis.
This analysis considers familial identification in a forensic context, but is applicable to tests for relatedness applied in the various contexts especially when considering unlinked genetic markers as in paternity investigation, ecological surveys, and conservation biology. When more extensive genotype or sequence data are available, it is appropriate to use more sophisticated tests for relatedness considering linkage or shared haplotype length [28], [57], [58].
The population genetic model used in forensic identification is remarkably coarse. In direct identification, the CODIS loci provide ample data to determine identity and non-identity, even with the coarse population genetic model of a small number of discrete homogenous genetic groups corresponding to social racial groups. We have shown that under this model, new concerns arise with familial searching. However, the model itself requires some scrutiny. It is clear that human genetic population structure is complex and humans are not easily split into a small number of discrete homogenous genetic groups [59]–[62]. Even with carefully chosen and defined population samples, it is practically impossible to account for human genetic variation and the discrete population group model fails to account for individuals with mixed ancestry. Additionally, individuals are typically assigned to genetic population groups based on social race. While there is correlation between genetic ancestry and social race, one does not determine the other [63]. As a result, in the discrete population group model, some individuals may not be grouped with the most similar genetic group.
Forensic familial searching will most likely be implemented in the context of a large offender/arrestee database, introducing questions of multiple testing over both database entrants, and the number of genetic familial relationships considered. Because forensic methodology practice varies over jurisdictions, it is not clear how these multiple testing issues have been, or will be, addressed. However, it is reasonable to assume that familial searching will result in a list of partial database matches with for genetic familial relationships. The parameter values used in the calculations must be conservative to keep the number of high partial matches manageably short, but the parameters also must allow enough leniency so that a true match will appear in the list considered. Ideally, parameter values used in practice should be tuned using simulations based on real genotype data representing realistic cryptic relatedness and population structure appropriate to the database and relevant population. When tuning parameters, as power increases, false positive rate will as well. Both of these values must be considered in deciding on appropriate parameter values. However across parameter values, some groups may have higher rates of false identification, as we have shown here, raising questions about the practicality of familial searching. Without access to accurate database or population information, or to a clear decision procedure practice, we refrain from making specific recommendations about parameter choice or methodology in this analysis.
Individual and population genotype information is necessary to determine the extent to which inaccurately assumed allele frequencies cause high false positive rate in familial matching in practice. For instance, in this study, we considered unrelated individuals, conforming to exactly one of five allele frequency distributions, in completely randomly mating populations. However the use of familial searching rests on the premise that relative groups are in the database and population structure is undeniably present in most databases [64]. Access to suitably secure and encrypted database information would enable analyses with an accurate portrayal of relatedness and population substructure. As recommended by Krane et al., increased transparency in database makeup, search procedure, and database access are required for rigorous analyses of forensic methodology [65].
If implemented with the core CODIS loci, familial searching may result in low distinguishability and potentially high false positive rates among certain groups, especially if only African American, European American, Southeastern Latino, and Southwestern Latino allele frequency distributions are in assumed LR calculations, as recommended by SWGDAM [55]. Because some of these groups (Native Americans and some immigrant groups) are correlated with social groups already over-represented in the criminal justice system, group members would be more likely to have a relative in the database, and that relative would be more likely to have a coincidental partial match with a crime scene sample [3]–[6], [9], [17], [18], [66]–[68]. Cumulatively, members of these groups are more likely to be investigated as a familial match due to over-represention in the database, and an unusually high false positive familial identification rate.
Our analysis makes use of allele frequency data for the 13 CODIS loci over different population samples socially defined by race. Note that alternate schemes to group individuals will also produce genetic differences between groups [56], [63], [69]. Here, we consider genetic differences between socially-determined groups which are relevant to the practice of genetic familial forensic identification. To do so, we used the allele frequencies reported by Budowle and Moretti [29] for samples from ‘Vietnamese,’ ‘African American,’ ‘Caucasian,’ ‘Hispanic,’ and ‘Navajo’ populations. In this manuscript, these same samples are refered to with the following labels: Vietnamese, African American, European American, Latino, and Navajo. As short hand, we refer samples derived from individuals from each sample as the sample name, for example ‘the Latino sample.’ The number of individuals genotyped to estimate allele frequencies for each sample varied, with , and individuals sampled for Vietnamese, African American, European American, Latino, and Navajo samples, respectively.
The consent and population grouping procedures used in obtaining these data are not clear. In the time since these data were collected, dominant cultural ethics regarding informed consent process have changed considerably, motivated largely by several cases of severe misuse of samples provided by Indigenous communities [70]–[73]. As a result, today it is becoming less acceptable to gather data in the same way [74]–[78]. We use the data because of its public availability, however we look forward to working with data collected using transparent informed consent methodology.
LRs are used to compare the probability of observed genotypes for two individuals under two different hypotheses: the individuals are unrelated () and the individuals share a specified genetic familial relationship () [79]. The LR is defined as [79]where is the observed pair of genotypes. When , the observed data are more likely for unrelated individuals and when , the observed data are more likely for individuals with the specified genetic relationship.
By assuming independence between all CODIS loci, can be broken down aswhere is the observed genotype for each individual at locus .
Relationships between individuals can be described using the identical by descent (IBD) sharing probabilities , , and , which are the probabilities that individuals with the specified relationship share 0, 1, and 2 alleles IBD, respectively [79]. For example, for a parent/offspring relationship , , and and for a sibling relationship , , and .
Using these IBD sharing probabilities, the LR becomeswhere the IBD sharing probabilities in the numerator are specified by the specific genetic relationship considered. The probability of the observed genotype combinations given IBD sharing probabilities depends on the specific combination of alleles observed. The probabilities of all observed genotypes, given IBD sharing probabilities, are defined in Text S1. These probabilities include a correction for expected background relatedness using the coancestry coefficient . In the first part of this study, we use the value of based on standard methodology in population genetics and as recommended by SWGDAM [55], [80].
The LR described above provides information about whether the observed data are more likely for unrelated or related individuals. However, the true population allele frequencies () are unknown, so needs to be estimated with the observed allele frequencies. Available sample allele frequencies are subject to sampling variation and variation due to demographic history [81]. Observed allele frequencies follow directly from observed genotype frequencies. Using , the probability of the data is calculated under different IBD sharing schemes, so the estimate of the likelihood ratio () can be computed. By considering the distribution of , we can find the distribution of and calculate confidence intervals on reported values.
Sampling variation is inherent in allele frequency estimation since a random sample must be chosen for the estimate. By their nature, different random samples vary in their representation of specific alleles, resulting in different allele frequency estimates. Additionally, random genetic sampling exists in the historical differentiation of populations, resulting in population groups with distinct allele frequencies. Since all present-day human population groups descend from a common ancestral population, the alleles present in each present-day population group reflect a sample of the alleles from the common ancestral population.
Under evolutionary equilibrium and a simple model of demographic history, the relationship between population group allele frequencies () can be modeled using a Dirichlet distribution informed by the coancestry coefficient (), accounting for genetic and sampling variation in estimated allele frequencies [81], [82]. With this model, we define the confidence interval in order to express uncertainty conferred by allele frequency estimate.
Using the same approach as Beecham and Weir [81], we note that the total is the sum of the for each locus . The central limit theorem indicates that, for even as few as 13 independent loci, this sum will be approximately normally distributed [81]. Thus, the confidence interval for is [81]where is the variance of and is the standard normal value for the given , in this study and so . While the typical arbitrary value of is used in this study, the trends explored will be maintained with different values of . Also note that a one-sided confidence interval can be derrived similarly with . This confidence interval is in space, so we can exponentiate the results to get the confidence interval of . The value of (derived in Text S1) depends on the variances of the observed allele frequencies. These, in turn, depend on to accommodate evolutionary variation over populations and this is why numerical techniques such as bootstrapping cannot be used to calculate likelihood ratios, as explained by Beecham and Weir [81].
Using the data provided by Budowle and Moretti [29], individuals were simulated based on the allele frequencies reported for each of the five population samples. For the population structure analysis, individuals are simulated from a given population sample by independently drawing two alleles from the appropriate allele frequency distribution for every locus. Note that the total independence between drawn alleles implicitly creates a population with a coancestry coefficient of zero (). Independently generated individuals are unrelated. Related individuals are simulated by generating unrelated individuals and randomly dropping alleles through a pedigree to achieve the desired relationship. In this way, we simulate pairs of both unrelated and related individuals from each population sample.
The total lack of population structure or cryptic relatedness () in our simulated populations causes unrelated individuals to share fewer alleles than would be expected in a real population. This contrasts with our use of the correction in calculations, conservatively lowering our calculated . This is consistent with forensic applications, where a conservatively high value for is chosen for the anticipated populations. Specifically, and have been suggested for use with populations primarily of European and Native American descent, respectively [43], [83].
In the second part of this analysis, when we consider the interplay between various parameters, it is necessary to simulate unrelated individuals from a population with a given non-zero coancestry coefficient (). To simulate unrelated and related individuals from a population with , random alleles are drawn using the probabilities of two-individual genotypes, given and a specified relationship, as written in Text S1.
We are interested in comparing LCL distributions generated with different parameters, particularly LCL distributions for truly unrelated individuals and truly related individuals. If the relationship perfectly distinguished relatives and unrelated individuals, these two distributions would be totally separate. The degree of overlap between the related and unrelated distributions roughly indicates the degree of genetic similarity of relatives and unrelated individuals, and so, how well distinguishes the two.
To quantify distinguishability, we use an empirical version of the measure proposed by Visscher and Hill [56]where and are the sample means of for the simulations of related and unrelated individuals, respectively, and and are the sample variances of for the simulations of related and unrelated individuals, respectively. Note that is analogous to the non-centrality parameter of the LR test statistic distribution under the alternative hypothesis. Higher indicates greater LR distribution differentiation and more distinguishability, while lower indicates more overlap and less distinguishability. The statistic accurately describes the differentiation in LR distributions, and is particularly appealing because it describes the difference in distributions, so it does not rely on a parameterized decision procedure to discretely determine relationship status.
|
10.1371/journal.ppat.1000553 | The QseC Adrenergic Signaling Cascade in Enterohemorrhagic E. coli (EHEC) | The ability to respond to stress is at the core of an organism's survival. The hormones epinephrine and norepinephrine play a central role in stress responses in mammals, which require the synchronized interaction of the whole neuroendocrine system. Mammalian adrenergic receptors are G-coupled protein receptors (GPCRs); bacteria, however, sense these hormones through histidine sensor kinases (HKs). HKs autophosphorylate in response to signals and transfer this phosphate to response regulators (RRs). Two bacterial adrenergic receptors have been identified in EHEC, QseC and QseE, with QseE being downstream of QseC in this signaling cascade. Here we mapped the QseC signaling cascade in the deadly pathogen enterohemorrhagic E. coli (EHEC), which exploits this signaling system to promote disease. Through QseC, EHEC activates expression of metabolic, virulence and stress response genes, synchronizing the cell response to these stress hormones. Coordination of these responses is achieved by QseC phosphorylating three of the thirty-two EHEC RRs. The QseB RR, which is QseC's cognate RR, activates the flagella regulon which controls bacteria motility and chemotaxis. The QseF RR, which is also phosphorylated by the QseE adrenergic sensor, coordinates expression of virulence genes involved in formation of lesions in the intestinal epithelia by EHEC, and the bacterial SOS stress response. The third RR, KdpE, controls potassium uptake, osmolarity, and also the formation of lesions in the intestine. Adrenergic regulation of bacterial gene expression shares several parallels with mammalian adrenergic signaling having profound effects in the whole organism. Understanding adrenergic regulation of a bacterial cell is a powerful approach for studying the underlying mechanisms of stress and cellular survival.
| Bacterial cells respond to the human stress hormones epinephrine (adrenaline) and norepinephrine (noradrenaline). These hormones are sensed by a bacterial receptor named QseC, which is a sensor kinase in the membrane that increases its autophosphorylation upon binding to these host signals. In addition to recognizing these signals, QseC also responds to a bacterial hormone-like molecule named autoinducer-3 (AI-3) that is produced by the human intestinal microbial flora. In this manuscript we have mapped genetically and biochemically the QseC signaling cascade in the deadly pathogen enterohemorrhagic E. coli (EHEC) O157:H7. EHEC uses this signaling system to activate expression of virulence genes. We show that the QseC signaling cascade is very complex so it can precisely modulate when different virulence traits are expressed. Because these sensor kinases are being evaluated as drug targets, a profound understanding of this signaling pathway is important for the development of novel therapeutic strategies to combat bacterial infections.
| The survival of an organism lies within its intrinsic ability to detect and efficiently respond to stress cues. Stress responses play a key role in adaptation to environmental, psychosocial, and physical insults. Hence it comes as no surprise that stress responses require synchronization and coordination of an organism's resources to ensure that metabolic substrates are available to meet the increasing energy demands of an effective stress response. Stress responses are generally termed “fight or flight” responses in higher animals, because they rely in the ability of an organism's to assess whether its better chance of survival relies on facing or avoiding an environmental insult. The hormones epinephrine and norepinephrine are at the core of stress responses [1].
In mammalian cells epinephrine and norepinephrine are recognized by GPCRs, which are membrane receptors coupled to heterotrimeric guanine-binding proteins (G-proteins). These proteins consist of three subunits α, β and γ. The binding of these signals to GPCRs result in a conformational change that activates the G-protein through the exchange of GDP for GTP. The activated G-protein dissociates from the receptor, the α, β, and γ subunits then dissociate and activate their intracellular targets. The GPCR specificity is controlled by the type of G-protein associated with the receptor. G-proteins are divided in four families according to their association with effector proteins. Three of these signaling pathways, Gαs, Gαi and Gαq, have been extensively studied, with Gαs activating adenylate cyclase, Gαi inhibiting adenylate cyclase, and Gαq activating phospholipoase C [1].
Most of the knowledge of epinephrine/norepinephrine-mediated signaling has been derived from studies in mammalian systems. However, although bacterial cells sense and respond to epinephrine and norepinephrine, the signaling pathways regulated by these mammalian hormones in bacteria have not been mapped [2],[3]. Bacteria do not express homologues of mammalian adrenergic receptors. These signals are sensed through histidine sensor kinases (HKs) [4],[5]. HKs constitute the predominant family of signaling proteins in bacteria. HKs usually act in concert with a response regulator (RR) protein constituting a two-component system. Upon sensing a defined environmental cue the HK autophosphorylates a conserved histidine residue, and then transfers this phosphate to an aspartate residue in the receiver domain of a cognate RR. The majority of the RRs are transcription factors, which are activated upon phosphorylation [6].
Two HKs, QseC and QseE, characterized in E. coli have been reported to sense epinephrine and norepinephrine [4],[5]. QseC binds to and increases its autophosphorylation in response to epinephrine, norepinephrine, and a bacterial signaling molecule termed autoinducer-3 (AI-3) [4]. QseE increases its autophosphorylation in response to epinephrine, phosphate and sulfate [5]. QseC acts upstream of QseE, given that transcription of qseE is activated by QseC [7]. The cognate RR for QseC is QseB [4], and the genes encoding this two-component system are co-transcribed constituting an operon [8]. The cognate RR for QseE is QseF, with the qseF gene also being co-transcribed with qseE within the same operon [8]. QseF, however, is also phosphorylated by four other non-cognate HKs: UhpB, BaeS, EnvZ and RstB [9]. QseC homologues exist in at least 25 bacterial species [10], while QseE homologues can only be found in enterics. This distribution of receptors may play a role in colonization or virulence with increased levels of epinephrine/norepinephrine.
The majority of the studies assessing adrenergic regulation of bacterial gene expression, have been conducted in bacteria that inhabit the human gastrointestinal (GI) tract [2],[4],[11],[12],[13],[14],[15]. Norepinephrine is present in the GI tract, being synthesized by adrenergic neurons of the enteric nervous system (ENS) [16]. Epinephrine is synthesized in the central nervous system and the adrenal medulla, and reaches the intestine in a systemic manner after being released into the bloodstream [17]. Norepinephrine is found at a nanomolar range in sera, while it is at a micromolar range in the intestine [18]. Both hormones have important roles in intestinal homeostasis regulating peristalsis, blood flow, chloride and potassium secretion [17],[19]. Both epinephrine and norepinephrine are recognized by the same adrenergic GPCRs in mammalian cells, and the ligand-binding site for these hormones is largely similar [20].
Enterohemorrhagic Escherichia coli (EHEC) O157:H7 is a GI pathogen that exploits adrenergic signaling to regulate virulence gene expression [2]. EHEC colonizes the human intestine and leads to the development of hemorrhagic colitis and hemolytic uremic syndrome (HUS). In the colon, EHEC forms attachment and effacement (AE) lesions on the intestinal epithelial cells, which cause extensive rearrangement of the host cell cytoskeleton resulting in the formation of a pedestal-like structure underneath the bacterial cell [21]. The genes required for AE lesion formation are located in the chromosomal pathogenicity island termed the locus of enterocyte effacement (LEE) [22]. The first operon in the island (named LEE1), encodes Ler, the master regulator of the LEE genes [23]. The remaining genes encode the type-three secretion system (TTSS) [24], which forms a syringe-like apparatus that the bacteria use to translocate effector molecules to the host cells. Many of these effectors mimic mammalian signaling proteins having profound effects in the host cell signal transduction culminating in diarrheal disease [25]. Seven of these effectors are encoded within the LEE region [25], while many others are scattered throughout the genome [26],[27]. The first secreted effector discovered outside of the LEE was NleA [28]. NleA is known to inhibit cellular protein secretion by disrupting mammalian COPII function and mutation of the nleA gene resulted in attenuation in mouse model of infection [28],[29]. EHEC also produces a potent Shiga toxin (Stx) that is responsible for the major symptoms of hemorrhagic colitis and HUS [30].
Expression of LEE, Shiga toxin and the flagella and motility genes in EHEC are regulated by the signals AI-3, epinephrine and norepinephrine through QseC [4],[10]. This regulation is important for EHEC virulence, given that qseC mutants are attenuated for infection in animal models of disease [4],[10]. QseC activates transcription of the flhDC genes, which encode the master regulators of the flagellar regulon, directly through QseB binding to the flhDC promoter. Importantly, this interaction is dependent on QseB's phosphorylation state [31], whereas, expression of the LEE and Shiga toxin genes are not regulated by QseB. Here we report a global analysis of EHEC gene expression in response to adrenergic signals, and map the QseC signaling cascade. In this study we unravel the adrenergic response of a bacterial cell at the genetic and biochemical levels, and demonstrate that adrenergic signaling has a profound effect on cell homeostasis, cell-to-cell signaling, and bacterial pathogenesis.
We had previously reported that inactivation of the qseC gene results in reduced flagella expression and motility, and reduced auto-activation [8],[31]. To further characterize the role of QseC in EHEC, Affymetrix E. coli 2.0 microarrays were used to compare expression profiles of the WT and ΔqseC strains in the presence and absence of the signals AI-3 and epinephrine in Dulbecco's modified eagle media (DMEM), which is optimal for expression of the LEE genes, and LB, which is optimal for expression of the flagella regulon. These arrays contain ∼10,000 probe sets (array genes), covering all genes in the genomes of the two sequenced EHEC strains (EDL933 and Sakai), K-12 strain MG1655, uropathogenic E. coli (UPEC) strain CFT073, and 700 probes to intergenic regions (which can encode non-annotated small ORFs, or small regulatory RNAs). Expression data can be accessed using accession number (GSE15050) at the NCBI GEO database. During growth in LB, 126 probe sets were down-regulated (28 specific to EHEC), and 708 were up-regulated (232 EHEC specific) in the qseC mutant (Table 1). The majority of the genes with an altered profile were derived from the E. coli K-12 strain MG1655 (68%), which represent a common E. coli backbone conserved among all E. coli pathovars [32]. Many of these genes are associated with metabolism, and they also include the flagella regulon (Figure 1B and Figure 2D and 2E). The EHEC specific genes (32%) include several prophage-encoded genes and stxAB encoding Shiga toxin. These studies revealed that QseC not only activates transcription of the flagella regulon, but also of the genes encoding Shiga toxin.
Transcriptome comparisons between WT and the qseC mutant grown in DMEM, a condition conducive to LEE and virulence gene expression, in the presence of AI-3 alone (both WT and the qseC mutant produce AI-3 when grown to late exponential phase in DMEM) or AI-3 plus epinephrine also revealed a global role for QseC regulation of virulence genes (Table 2). In the presence of AI-3 alone, expression of 106 genes was increased and 273 decreased in the qseC mutant compared to WT. In the presence of AI-3 plus epinephrine expression of 70 genes was increased and 311 decreased in the qseC mutant compared to WT. AI-3 and epinephrine have been reported to act as agonistic signals [33]. This agonistic relationship in signaling can be further illustrated by the observation that while AI-3 is only sensed through QseC, epinephrine is sensed by both QseC and QseE [4],[5]. However, it is worth mentioning that QseC acts upstream of QseE, given that transcription of qseE is activated by QseC [7]. These data suggest that both signals tend to activate global gene expression in a qseC-dependent fashion more frequently than repress expression. Among the genes activated in a qseC-dependent manner are the LEE (through activation of ler transcription, within the LEE1 operon, encoding the Ler activator of all other LEE genes) and stxAB (Shiga toxin) genes (Figure 1B, 1C and 1D). The genes encoding Stx are located within the late genes of a λ- bacteriophage and are transcribed when the phage enters its lytic cycle upon induction of an SOS response in the bacterial cell [34]. Upon the induction of an SOS response, recA is upregulated and cleaves the λ cI repressor allowing transcription of the middle and late genes to proceed, and together with them the stxAB genes. QseC-induction of stxAB transcription occurs through induction of recA expression (Figure 1D), suggesting that QseC mediates SOS induction in bacterial cells. In addition to activating expression of the LEE-encoded TTSS, the majority of the genes encoding effectors translocated through this TTSS are also regulated by QseC (Figure 1B). Of note, transcription of the gene encoding the NleA effector is strongly repressed by QseC in LB, while its expression is slightly (non-statistically significant) decreased in the qseC mutant in DMEM (Figure 1E and 1F).These analyses confirmed QseC's activation of the flagellar genes and revealed several new regulatory targets, including: LEE (through ler), nleA, genes of the SOS response and Shiga toxin. Altogether, these data suggest that QseC is at the top of the signaling cascade activated by AI-3, epinephrine and norepinephrine, initiating regulation of all EHEC virulence genes.
Through QseC, EHEC senses AI-3, epinephrine and norepinephrine to activate flagella and motility, AE lesion formation and Shiga toxin expression. Given that these are expensive biological processes that have to occur in concert, the kinetics of expression of these genes has to be exquisitely fine-tuned. We have previously reported that a ΔqseC EHEC had reduced motility, expressed less flagella, and presented reduced transcription of the flagella regulon [35]. The cognate RR of the QseC HK is QseB, which is phosphorylated at a conserved aspartate residue by QseC [4] (Figure 2A). In this study we deleted the cognate response regulator qseB. Since we had previously shown that QseC regulated the flagellar genes through a direct interaction of QseB and the flhDC promoter (FlhDC are the master activators of the flagella regulon) [31], we hypothesized that mutation of qseB would result in decreased motility. However, a ΔqseB mutant has no motility defect (Figure 2B), and expresses flagella at the same levels as the WT strain (Figure 2C). To confirm these results, we assessed transcription of flhD by real-time RT-PCR in WT, ΔqseC, and ΔqseB mutants. Relative expression levels of flhD in these three strains indicated that transcription of flhD is decreased in ΔqseC but is unaltered in ΔqseB (Figure 2D). We then performed β-galactosidase assays with the −900 to +50 bp region of the flhDC promoter fused to a promoterless lacZ gene as a reporter. We found that in ΔqseC there was five-fold less β-galactosidase activity as compared to WT (Figure 2E), but there was no difference in β-galactosidase activity between the WT and ΔqseB. Because QseB and QseC constitute a cognate two-component system, we expected that the qseC and qseB mutants would have similar phenotypes. However, while the qseC mutant has decreased motility and expression of the flagellar regulon, the qseB mutant shows similar levels of flhDC expression and motility as the WT strain. These results led us to develop two potential hypotheses for the differential effects of knocking an HK (QseC) and its cognate RR (QseB) on flhDC transcription. First, QseB can bind to different DNA sequences according to its phosphorylation state, acting as a repressor or activator depending on which site it is bound to. Second, QseC could be a promiscuous HK and can phosphorylate non-cognate RRs that acts on the flhDC promoter.
To test the first hypothesis we overexpressed QseB in a ΔqseC background. We assumed that this strain would have an overabundance of unphosphorylated QseB. We found that this strain was less motile than ΔqseC, indicating that unphosphorylated QseB can act as a repressor of the flagellar gene expression (Figure 3A). We also complemented the ΔqseB strain with a plasmid expressing QseB, and observed that the complemented strain had decreased motility; again suggesting that overabundance of unphosphorylated QseB has a repressive role in motility (Figure 3B). However, when we complemented the ΔqseB strain with a plasmid expressing qseBC (Figure 2), we did not observe any differences in motility, probably because the levels of QseB and QseC were balanced in this strain. Next, we overexpressed qseB, in a strain containing the −900 to +50 bp region of the flhDC promoter upstream of a promoterless lacZ. We found that in the strain overexpressing qseB there was a five-fold decrease in β-galactosidase activity (Figure 3C). We also observed decreased flhDC transcription in a strain overexpressing a QseB site-directed mutant (QseB D51A) that cannot be phosphorylated (the conserved aspartate phosphorylated residue has been changed to an alanine) (Figure 3C), further indicating that an abundance of unphosphorylated QseB represses expression of flhDC.
We had previously shown that QseB can bind to two regions of the flhDC promoter, −300 to +50 bp and −900 to −650 bp [31]. We demonstrated that this binding required QseB to be phosphorylated [31] (Figure 3C), which can be achieved by providing a small phosphate donor, acetyl phosphate, to QseB in vitro. QseB will only bind to the −300 to +50 bp flhDC region in the presence of acetyl phosphate (Figure 3D), and the QseB D51A mutant is also unable to bind to this region of flhDC (Figure 3D). We have discovered a new QseB binding site in the flhDC promoter from −650 to −300 bp to which QseB can bind in the absence of phosphorylation. QseB binds to this −650 to −300 bp site in the absence of acetyl phosphate, and QseB D51A can also bind to this site (Figure 3E and 3F). The presence of this new binding site provides further evidence for a dual role of QseB in the regulation of the flhDC promoter. At low signal concentration there is low QseC activation and thus low QseB phosphorylation. In this case only the −650 to −300 bp site of the flhDC promoter will be occupied by non-phosphorylated-QseB and this binding may lead to repression. When the signal is high the opposite is true. The −300 to +50 bp and −900 to −650 bp sites will be occupied by phosphorylated QseB and flhDC will be activated (Figure 3H). In further support of this model, a nested deletion analyses of the flhDC promoter fused to lacZ shows that the full length fusion (−900 to +50 bp) is activated by QseC (Figure 3G). This fusion contains all three QseB binding sites, and in the presence of QseC, phosphorylated QseB will occupy the activating sites from −950 to −650 bp and −300 to +50 bp, increasing transcription. In the −650 to +50 bp fusion, transcription of flhDC is repressed in the absence or presence of QseC, probably because of non-phosphorylated QseB binding to the −650 to −300 bp site, which represses flhDC transcription. Non-phospho-QseB binding to the −650 to −300 bp region is probably “locked” in the absence of the upstream (−900 to −650) site. When both upstream sites are removed (−300 to +50 bp fusion), phospho-QseB bound to this proximal site will activate flhDC transcription (Figure 3F). In the complete absence of QseB, as in a qseB null strain, there will be QseC-independent expression of flhDC transcription, without any repression or activation (de-repression) by QseB (Figure 2). These data indicate that regulation of flhDC transcription by QseC occurs through its cognate RR QseB, and that QseB plays a dual role in this regulation according to its phosphorylated state.
QseB, however, does not seem to play a role in QseC-dependent activation of LEE and stxAB transcription (Figure 4), suggesting that this regulation may occur through phosphorylation of other RRs. In addition to QseB there are at least 31 other RR in E. coli that could be activated via QseC [9]. There is minimal cross-talk (cross-phosphorylation) between different two-component systems ensuring faithful transmission of information through distinct signaling pathways [36],[37]. Indeed, the incidence of cross-phosphorylation between non-cognate HKs and RRs is low in E. coli, Yamamoto et al. showed that phosphorylation of non-cognate response regulators by HKs is rare and occurs in only 22 of 692 possible combinations [9]. However, in this same study, Yamamoto noticed that a distinct few HKs are more prone to also signal through non-congate RRs.
We have previously reported that QseC autophosphorylates in response to AI-3, epinephrine and norepinephrine in an in vitro liposome assay and can phosphotransfer onto its cognate RR, QseB [4]. In order to test QseC's ability to phosphotransfer onto non-cognate RRs, we purified 31 E. coli RRs and performed phosphotransfer assays with QseC in liposomes. Of note all of these RRs were soluble and correctly folded upon purification, and have been previously shown by Yamamoto et al. to be active in phosphotransfer reactions with their cognate HKs [9]. Through this assay, we found only two additional QseC phosphorylation targets: KdpE and QseF (Table 3, Figure 5A and 5B). KdpE has been shown to regulate potassium uptake and medium osmolarity [38]. We found that kdpA, one of the genes regulated by KdpE, is also down-regulated in the ΔqseC (Figure 6A), indicating that cross-phosphorylation between QseC and KdpE results in QseC regulation of KdpE-dependent targets. To assess the contribution of KdpE to QseC's signaling transduction pathway, we deleted kdpE but found no motility defect (Figure 6C) or decreased flhDC expression (Figure 6B) in the kdpE mutant, indicating that KdpE is not regulating flhDC. When we assessed transcription of ler (LEE) and stx, we observed that KdpE activates transcription of the LEE genes, but not stx, suggesting that through the KdpE RR, QseC activates expression of the LEE genes (Figure 6D).
The second non-cognate RR phosphorylated by QseC, QseF, is responsible for aiding in AE lesion formation by activating expression of the phage-encoded gene espFu [7]. EspFu is a secreted effector, translocated to epithelial cells by the LEE-encoded TTSS, and it is involved in host actin nucleation and polymerization for AE lesion formation [39],[40]. QseF, however is not involved in regulation of LEE gene expression (Figure 6E) [7], nor in flagella and motility regulation [7]. However, a qseF knockout presented diminished expression of the stx gene (Figure 6E), suggesting that QseC activation of Shiga toxin expression occurs through the QseF RR. The QseF cognate HK is QseE [9], which is a second bacterial adrenergic receptor that senses epinephrine, phosphate and sulfate [11]. The addition of epinephrine to EHEC activates expression of qseEF, and this regulation is eliminated in the ΔqseC mutant, indicating that QseC activates transcription of qseEF [7]. Transcriptional regulation of qseEF by QseC, in addition to cross-phosphorylation of QseF by QseC and QseE may fine tune the timing for switching from motility, to AE lesion formation to Shiga toxin production during infection.
QseC phosphorylates three RRs: QseB, KdpE and QseF (Figure 7A). Through QseB the flagella regulon is regulated. KdpE activates expression of ler, and consequently of all LEE genes. QseF plays a role in inducing an SOS response and Shiga toxin production, as well as activating expression of espFu [7], which encodes an effector essential for AE lesion formation. To search globally which sets of QseC-dependent genes are regulated through each RR we performed transcriptome assays (GEO series GSE15050). These comparisons were performed with gene arrays hybridized with cDNA from RNA extracted from WT, ΔqseC, ΔqseB, ΔkdpE and ΔqseF strains grown in DMEM to an OD600 of 1.0, conditions known to yield maximal endogenous AI-3 production in these strains [41]. Given that AI-3 is only sensed through QseC, and QseC will phosphorylate in the presence of either AI-3 or epinephrine [4],[11], by working under these conditions we would detect only QseC-dependent genes. We avoided using epinephrine in these comparisons, because epinephrine is also sensed by the QseE HK [4],[11]. Transcription of 324 genes was increased, and 344 decreased in the ΔqseC mutant compared to WT (Figure 7B). Of the 324 genes increased in the ΔqseC, 15 were also increased in ΔqseB, 13 in ΔqseF, and 63 in ΔkdpE (Figure 7B). These data suggest that 91 of these 324 genes repressed by QseC are under the control of the QseB, KdpE and QseF RRs. These leaves 233 genes repressed through QseC unaccounted for. A possible explanation could be that these genes may be activated and repressed by QseB in a similar fashion to flhDC (Figure 3), and these genes would not appear as transcriptionally regulated through QseB using gene arrays. QseC activates transcription of 344 genes, with 205 being activated through QseB, 44 through QseF and 87 through KdpE (Figure 7B). These three RRs activate transcription of 336 of the 344 QseC-dependent genes, giving almost 100% coverage of QseC-activated genes.
Chemical signaling between cells underlies the basis of multi-cellularity. Although bacteria are unicellular, bacterial populations also utilize chemical signaling, through hormone-like compounds named autoinducers, to achieve cell-cell communication and coordination of behavior [42]. Chemical signaling is also essential for an organism to survive, successfully adapt to ever changing environments and protect themselves from insults, which can be collectively considered stress. Successful stress responses require energy input, and the coordination of many complex signaling pathways within the cell. Co-evolution of prokaryotic species and their respective eukaryotic host have exposed bacteria to hormones and eukaryotic cells to autoinducers. Therefore, it is not surprising that bacteria can respond to host hormones, and that some pathogenic species have high-jacked these signaling systems to promote disease states [43].
One example of a pathogen that senses host hormones to regulate virulence is EHEC [2]. Upon reaching the human colon, EHEC senses the autoinducer-3 (AI-3) produced by the microbial gastrointestinal flora, and epinephrine and norepinephrine produced by the host through the HK QseC [2],[4]. This signal transduction activates transcription of virulence genes in a coordinated fashion leading to the formation of AE lesions on intestinal cells by the locus of enterocyte effacement (LEE) genes, the flagella regulon for enhanced motility, and Shiga toxin production which is responsible for HUS. EHEC probably first encounters the AI-3 signal produced by the microbial flora that inhabits the intestinal lumen [2]. Because the infectious dose of EHEC is very low (estimated to be 50 CFUs) [21], it is unlikely that it responds to self-produced signal to initiate infection. Upon sensing AI-3, QseC initiates the signaling cascade that will activate the flagella regulon leading to swimming motility, which may aid EHEC to come closer to the intestinal epithelial layer. As EHEC approaches the epithelium and starts forming AE lesions it is probably then exposed to epinephrine and/or norepinephrine. Norepinephrine is synthesized within the adrenergic neurons of the enteric nervous system (ENS) that innervates the basolateral layer of the intestine [16]. Epinephrine is synthesized in the central nervous system (CNS) and in the adrenal medulla; it acts systemically after being released into the bloodstream, when it can reach the intestine [17]. AE lesion formation and the commencement of bloody diarrhea may increase EHEC exposure to epinephrine and norepinephrine, further upregulating expression of virulence genes in EHEC. This coordinated regulation involves a number of two-component regulatory systems composed of HKs and RRs that result in cascades of gene expression.
Recognition of AI-3/epinephrine/NE by QseC can be specifically blocked by the administration of the α-adrenergic antagonist phentolamine [4], and a synthetic compound called LED209 [10]. Using two different rabbit infection models it has been demonstrated that QseC plays an important role in pathogenesis in vivo, since qseC mutants were attenuated for virulence in these animals [4],[10]. Recently, a novel two-component system, the QseEF system [7], where QseE is the HK and QseF is the RR was shown to also regulate virulence in EHEC. QseE can also respond to the host hormone epinephrine like QseC, but in contrast, does not sense the bacterial signal AI-3. QseE is downstream from QseC in this signaling cascade, given that qseEF transcription is activated by epinephrine via QseC. The QseEF system is not involved in regulation of flagella and motility, but plays an important role in activating genes necessary for AE lesion formation [7] and also activates expression of Shiga toxin (Figure 6).
The AI-3/epinephrine/NE signaling system is not restricted to EHEC. In silico analysis showed homologues of QseC in other bacterial species such as Salmonella sp, Shigella flexneri, Francisella tularensis, Haemophilus influenzae, Erwinia carotovora, and many others [10]. In vivo studies provided evidence that the QseC HK is important in Salmonella typhimurium [10],[44] and Francisella tularensis [45] pathogenesis, since qseC mutants of these strains are attenuated in animal models of infection and in vivo inhibition of QseC by LED209 results in attenuation of infection by these organisms [10].
Because QseC is central for sensing adrenergic signals, and the effect these signals have in basic biological processes, a complete understanding of the QseC signaling transduction pathway in bacteria will offer clues on how eukaryotic stress responses affect a prokaryotic cell. We demonstrate that QseC acts promiscuously through three RRs (Figures 5 and 7) to initiate a complex signaling cascade that affects both metabolism and pathogenesis (Figure 8). QseC controls the expression of all of these features, either directly or indirectly and must be considered to be at or near the top of the signaling cascade. The fact that more that one kinase can activate multiple response regulators suggests that there is a hierarchy of signaling, beginning with QseC. It is currently unclear if the regulation by the associated HK and RR overrides the signal employed by a non-cognate HK or if they work in synergy to amplify the initial signal. This additional level of control may be the fine-tuning that is observed in EHEC where the motility, formation of lesions and secretion of toxin must be exquisitely choreographed to have an effective infection occur.
An additional level of complexity included in this signaling cascade is that QseB, binds to different sites in the target promoters according to its phosphorylation state (Figure 3). This allows further modulation of gene expression by the spatial arrangement of these sites in the regulatory region of genes, allowing the same RR to both repress and activate transcription of the same gene. In the non-activated form (non-phosphorylated) QseB forms an additional regulatory barrier to the expression of flhDC. Only under conditions where QseB is both phosphorylated and in sufficient concentration is there full activation of the flagella regulon. Thus this two-step process provides additional levels of control for this energetically expensive appendage. These types of mechanisms ensure that only under conditions which are favorable the resources are devoted to this response. The DNA binding domain of QseB shares similarities with the DNA binding domain of the OmpR RR, which also recognizes different sites on DNA according to its phosphorylation state [46],[47].
Because epinephrine and norepinephrine exert a profound effect in the host physiology and immune system, the ability to sense these hormones by bacteria may facilitate gauging the fitness of the host. Inter-kingdom chemical signaling plays an important role in the relationships forged between bacteria and animals. Chemical communication within kingdoms has been studied for many decades, however, the interception of these languages between different kingdoms has been appreciated only more recently. As this field expands, more and more examples will be described, and many questions answered.
All bacterial strains and plasmids utilized in this study are listed in Table S1. E. coli strains were grown aerobically in LB or DMEM (Invitrogen) medium at 37°C unless otherwise stated. Antibiotics were added at the following concentrations: 100 µg ml−1 ampicillin and 30 µg ml−1 chloramphenicol.
Standard methods were used to perform plasmid purification, PCR, ligation, restriction digests, transformation and gel electrophoresis [48].
Construction of isogenic kdpE (DH11) and qseB (MC474) mutants was carried out as previously described [49]. Briefly, 86-24 cells containing pKD46 were prepared for electroporation. A kdpE PCR product was generated using primers kdpEλRed-F and kdpEλRed-R (Table S2) and pKD3 as a template and PCR-purified (Qiagen). A qseB PCR product was generated using primers qseBλRed-F and qseBλRed-R (Table S2) and pKD3 as a template and PCR-purified (Qiagen). Electroporation of the PCR products into these cells was performed; cells were incubated at 22°C for 16 h in SOC, and plated on media containing 30 µg ml−1 chloramphenicol overnight at 42°C. Resulting colonies were patched for chloramphenicol resistance and ampicillin sensitivity, and PCR-verified for the absence of the gene. The chloramphenicol cassette was then resolved from the mutants in order to create non-polar, isogenic kdpE and qseB mutants. Plasmid pCP20, encoding a resolvase, was electroporated into the mutant strains, and resulting colonies were patched for chloramphenicol sensitivity. Construction of qseC and qseF mutants has been previously published [7],[35].
Site-directed mutagenesis was carried out using the Quick Change II site-directed mutagenesis kit (Stratagene). Mutagenesis PCR primers were constructed using the Primer X software (http://www.bioinformatics.org/primerx/) and are listed in Table 1 (qseBD51AF and qseBD51AR). The plasmid pVS154 was PCR amplified with the mutagenesis primers according to Stratagene's PCR protocol, generating the plasmid pDH12 (86-24 qseB D51A in pBADMycHis). The PCR product was digested with DpnI for 3 h at 37°C in order to remove the template plasmid. After digestion, the PCR product was transformed into XL-1 Blue supercompetent cells (Stratagene) and plated on selective media. The next day, plasmid DNA was isolated and sequenced to determine if the mutation was present.
Cultures were grown aerobically in LB medium at 37°C overnight, diluted 1∶100 in LB or DMEM (in the presence of self produced AI-3 and in the absence or presence of 10 µM epinephrine) and grown aerobically at 37°C. 0.2% arabinose was added to the media when induction was required. RNA from three biological replicate cultures of each strain was extracted at the late exponential growth phase (OD600 of 1.0) using the RiboPure Bacteria RNA isolation kit (Ambion) according to the manufacturer's guidelines. The primers used in the real-time assays were designed using Primer Express v1.5 (Applied Biosystems) (Table S2). Real-time reverse transcription-PCR (RT-PCR) was performed in a one-step reaction using an ABI 7500 sequence detection system (Applied Biosystems). For each 20-µl reaction mixture, 10 µl 2× SYBR master mix, 0.1 µl Multi-Scribe reverse transcriptase (Applied Biosystems), and 0.1 µl RNase inhibitor (Applied Biosystems) were added. Amplification efficiency of each of the primer pairs was verified using standard curves of known RNA concentrations. Melting-curve analysis was used to ensure template specificity by heating products to 95°C for 15 s, followed by cooling to 60°C and heating to 95°C while monitoring fluorescence. Once the amplification efficiency and template specificity were determined for each primer pair, relative quantification analysis was used to analyze the unknown samples using the following conditions for cDNA generation and amplification: 1 cycle at 48°C for 30 min, 1 cycle at 95°C for 10 min, and 40 cycles at 95°C for 15 s and 60°C for 1 min. The rpoA (RNA polymerase subunit A) gene was used as the endogenous control. Real-time RT-PCR primers for the LEE genes and rpoA have been previously described [33].
In order to study the binding of QseB to the flhDC promoter EMSAs were performed using the purified QseB protein and the flhDC promoter. DNA probes were then end-labeled with [γ-32P]-ATP (NEB) using T4 polynucleotide kinase using standard procedures [48]. End-labeled fragments were run on a 5% polyacrylamide gel, excised and purified using the Qiagen PCR purification kit. Electrophoretic mobility shift assays were performed by adding increasing amounts of purified QseB or QseBD51A protein (0–20 µM) to end-labeled probe (10 ng) in binding buffer [500 µg ml−1 BSA (NEB), 50 ng µl−1 poly-dIdC, 60 mM HEPES pH 7.5, 5 mM EDTA, 3 mM DTT, 300 mM KCl, 25 mM MgCl2] with or without 0.1 M acetyl phosphate for 20 min at 4°C. Immediately before loading, a 5% ficol solution was added to the mixtures. The reactions were electrophoresed for approximately 14 h at 65 V on a 5% polyacrylamide gel, dried and exposed to KODAK X-OMAT film.
Assays were performed as previously described [31]. Briefly, motility assays were performed at 37°C on 0.3% agar plates containing Tryptone media (1% tryptone and 0.25% NaCl). The motility halos were measured at 4 h and 8 h.
One liter of LB media was inoculated at 1∶100 and grown to O.D. 0.6 at 30°C. The culture temperatures were reduced to 25°C, induced with 400 µM IPTG (Sigma) or 0.2% arabinose, and grown for either 3 h or 18 h. Cells were harvested, suspended in lysis buffer (50 mM phosphate buffer pH 8, 300 mM NaCl, and 20 mM imidazole) and lysed by homogenization. The lysed cells were centrifuged and the lysates were loaded onto to a Ni2+- NTA-agarose gravity column (Qiagen). The column was washed with lysis buffer and protein was eluted with elution buffer (50 mM phosphate buffer pH 8, 300 mM NaCl, 250 mM imidazole). Fractions containing purified protein were confirmed by SDS-PAGE and concentrated for further use.
Liposomes were reconstituted as described previously [4],[51]. Briefly, 50 mg of E. coli phospholipids (20 mg/ml in chloroform; Avanti Polar Lipids) were evaporated and then dissolved into 5 ml of potassium phosphate buffer containing 80 mg of N-octyl-β-d-glucopyranoside. The solution was dialyzed overnight against potassium phosphate buffer. The resulting liposome suspension was subjected to freeze–thaw in liquid N2. Liposomes were then destabilized by the addition of 26.1 mg of dodecylmaltoside, and 0.625 mg of QseC-MycHis was added, followed by stirring at room temperature for 10 min. Two hundred-sixty milligrams of Biobeads (Biorad) were then added to remove the detergent, and the resulting solution was allowed to incubate at 4°C for 16 h. The supernatant was then incubated with fresh Biobeads for 1 h at 22°C the next day. The resulting liposomes containing reconstituted QseC-MycHis were frozen in liquid N2 and stored at −80°C until used.
Assays were performed as previously described [4]. Briefly, twenty microliters of the liposomes containing QseC-MycHis were adjusted to 10 mM MgCl2 and 1 mM DTT, and 10 µM epinephrine, frozen and thawed rapidly in liquid N2, and kept at room temperature for 1 h (this allows for the signals to be loaded within the liposomes). [γ32P]dATP (0.625 µl) (110 TBq/mmol) was added to each reaction. To some reactions, 12.5 µg of response regulator was added. At each time point (0, 10, 30 min), 10 µl of SDS loading buffer (with 20% SDS, to completely denature the liposome) was added. For all experiments involving QseC alone, a time point of 10 min was used. The samples were run on SDS/PAGE without boiling and visualized via PhosphorImager. The bands were quantitated by using imagequant version 5.0 software (Amersham Pharmacia).
Assays were performed as previously described [31]. Briefly, bacteria containing lacZ fusions were grown overnight at 37°C in LB containing the appropriate selective antibiotic. Cultures were diluted 1∶100 and grown in LB, and when necessary supplemented with 0.2% arabinose, to an OD600 of 1.0 at 37°C. These cultures were then assayed for β-galactosidase activity using o-nitrophenyl-beta-d-galactopyranoside (ONPG) as a substrate as described previously [52].
|
10.1371/journal.pntd.0003920 | Human IgG1 Responses to Surface Localised Schistosoma mansoni Ly6 Family Members Drop following Praziquantel Treatment | The heptalaminate-covered, syncytial tegument is an important anatomical adaptation that enables schistosome parasites to maintain long-term, intravascular residence in definitive hosts. Investigation of the proteins present in this surface layer and the immune responses elicited by them during infection is crucial to our understanding of host/parasite interactions. Recent studies have revealed a number of novel tegumental surface proteins including three (SmCD59a, SmCD59b and Sm29) containing uPAR/Ly6 domains (renamed SmLy6A SmLy6B and SmLy6D in this study). While vaccination with SmLy6A (SmCD59a) and SmLy6D (Sm29) induces protective immunity in experimental models, human immunoglobulin responses to representative SmLy6 family members have yet to be thoroughly explored.
Using a PSI-BLAST-based search, we present a comprehensive reanalysis of the Schistosoma mansoni Ly6 family (SmLy6A-K). Our examination extends the number of members to eleven (including three novel proteins) and provides strong evidence that the previously identified vaccine candidate Sm29 (renamed SmLy6D) is a unique double uPAR/Ly6 domain-containing representative. Presence of canonical cysteine residues, signal peptides and GPI-anchor sites strongly suggest that all SmLy6 proteins are cell surface-bound. To provide evidence that SmLy6 members are immunogenic in human populations, we report IgG1 (as well as IgG4 and IgE) responses against two surface-bound representatives (SmLy6A and SmLy6B) within a cohort of S. mansoni-infected Ugandan males before and after praziquantel treatment. While pre-treatment IgG1 prevalence for SmLy6A and SmLy6B differs amongst the studied population (7.4% and 25.3% of the cohort, respectively), these values are both higher than IgG1 prevalence (2.7%) for a sub-surface tegumental antigen, SmTAL1. Further, post-treatment IgG1 levels against surface-associated SmLy6A and SmLy6B significantly drop (p = 0.020 and p < 0.001, respectively) when compared to rising IgG1 levels against sub-surface SmTAL1.
Collectively, these results expand the number of SmLy6 proteins found within S. mansoni and specifically demonstrate that surface-associated SmLy6A and SmLy6B elicit immunological responses during infection in endemic communities.
| Adult schistosome parasites can live in the human bloodstream for years without being adversely affected by the host immune response. Identifying which proteins are on the surface of the parasite and understanding how they contribute to long-term host/parasite relationships is an essential step in developing novel intervention strategies. Here, utilising a comprehensive bioinformatics approach to identify Schistosoma mansoni gene products sharing distinct surface-associated features including signal peptides, hydrophobic C-termini, disulfide bonds and uPAR/Ly6 domains, we identified eleven proteins of interest. These proteins, reassuringly, include three representatives previously found associated with the schistosome surface (here termed SmLy6A, SmLy6B and SmLy6D) as well as three novel members (SmLy6G, SmLy6H and SmLy6J). To identify if surface-associated SmLy6 members are recognized by S. mansoni infected individuals, we specifically examined antibody responses to SmLy6A and SmLy6B in an endemic human population. Our work expands the number of putative cell surface associated schistosome proteins and provides a greater understanding of the dynamics of antibody responses in endemic communities against two representatives.
| Human schistosomiasis is estimated to affect more than 200 million people living in developing countries, with 120 million people symptomatic and 20 million suffering severe illness [1]. With a further 600 million individuals at risk of infection from Schistosoma mansoni, Schistosoma haematobium and Schistosoma japonicum (the three main human-infective species) and up to 70 million disability-adjusted life years (DALYs) lost annually, this neglected tropical disease (NTD) is one of the most important on the planet [2]. Schistosomiasis control is predominantly facilitated by mass drug administration (MDA) of praziquantel, a safe and potent chemotherapy developed in the late 1960’s [3]. However, mono-chemotherapy control of schistosomiasis raises the spectre of drug resistance [4] and highlights the need for developing an immunoprophylactic anti-schistosomal vaccine. Throughout the past several decades, progress in schistosome vaccine development has proved disappointing, with six promising candidates failing to induce sufficient protection in independent experimental models [5]. However, with the increased use of genomic, transcriptomic and proteomic methodologies to characterize schistosome biology, the number of novel and, perhaps more potent, vaccine candidates has vastly expanded [6–12].
Recent DNA microarray analyses [10] and sub-proteomic studies [11] have contributed to these efforts and led to the prioritization of several S. mansoni immunoprophylactic candidates. Amongst these, three (smp_081900, smp_019350 and smp_105220 –GeneDB.org identifiers) contained weak homology to Lymphocyte Antigen 6 (Ly6) proteins. Ly6 family members, such as the CD59 protein and urokinase/plasminogen activator receptor (uPAR) in humans, are extracellular, cysteine-rich proteins commonly tethered to the plasma membrane via a glycosylphosphatidylinositol (GPI) moiety [13,14]. All Ly6 family members contain one or more uPAR/Ly6 domains (Pfam identification code: PF00021; [15]), a 60–110 amino acid region that forms a characteristic “three-fingered protein domain” (TFPD) fold and contains eight or ten conserved cysteines [16]. This TFPD fold places Ly6 proteins as part of the larger TFPD protein superfamily, which also includes the transforming growth factor beta receptor (TGFβR) family and the snake α-neurotoxins family [16]. While each of these three protein families share the same core fold, they are differentiated by protein features (e.g. presence of C-terminal intracellular kinase domain in TGFβR family; [17]) and cysteine spacing (e.g. absolute conservation of C1-X-X-C2 spacing in the Ly6 family; [18]). Unlike the restricted phylogenetic representation of the snake α-neurotoxins, Ly6 genes have been found across both vertebrate and invertebrate phyla (Homo sapiens CD59; [19], Drosophila melanogaster rtv; [20] and Caenorhadbditis elegans odr-2 [21]). Functionally, Ly6 proteins have been involved in diverse roles ranging from septate junction formation in D. melanogaster [18], odorant response in C. elegans [21] and limb regeneration in salamanders [22]. In humans, the majority of the 27 Ly6 proteins [23] have unknown functions except for a few notable exceptions including the CD59 protein and uPAR [16]. CD59 is a cell surface glycoprotein found on most cells and functions to prevent complement-mediated lysis by binding C8α and C9 to prevent the formation of the complement membrane attack complex [24]. The triple uPAR/Ly6 domain-containing uPAR protein is a cell surface receptor for urokinase plasminogen activator (uPA), binding uPA to regulate its activity in extracellular matrix remodeling [25]. In schistosomes, proteomic analyses of the parasite tegument have identified two uPAR/Ly6 domain-containing proteins on the surface of the adult (initially named SmCD59a and SmCD59b; [11]) and another member identified within the tegumental fraction (smp_081900; [26]). However, subsequent analysis by Farias et al. has identified that these proteins are not functional orthologs of the human CD59 protein and are, rather, part of a larger set of S. mansoni uPAR/Ly6 domain-containing proteins containing seven members with unknown functions [27].
Here, we comprehensively re-analyse the S. mansoni genome to identify an expanded repertoire of eleven SmLy6 family members, SmLy6A-K, and describe the serological immune responses elicited by the two S. mansoni Ly6 family members initially named SmCD59a and SmCD59b (renamed in this study SmLy6A and SmLy6B) during infection in endemic communities. Specifically, we compare IgG1, IgG4 and IgE responses against SmLy6A and SmLy6B to SmTAL1 (a characterized non-surface tegumental protein) in a cohort of S. mansoni infected Ugandan males. Here, we detect evidence of anti-SmLy6 IgG1 (>IgG4>IgE) specific immune responses with IgG1 titers reflecting the differential abundance of SmLy6A and SmLy6B found throughout the schistosome lifecycle. Finally, we show that IgG1-specific SmLy6A and SmLy6B titers drop after praziquantel treatment. Together, our data collectively expand the number of putative cell-surface Ly6 family members found in S. mansoni and demonstrate that infected humans mount an immunoglobulin response against two representative members, which is strongly supportive of their surface localisation. These findings may contribute to future studies investigating the immunoprophylactic potential of the SmLy6 family.
Unless otherwise stated, all chemicals and reagents were purchased from Sigma-Aldrich, United Kingdom.
All procedures performed on mice and rats adhered to the European Union Animals Directive 2010/63/EU and were approved by both Aberystwyth University’s animal welfare and ethical review body (AWERB, project license PPL 40/3700) and the French local ethics committee (CEEA Nord Pas de Calais). Ethical clearance for the Musoli cohort was obtained from the Uganda National Council of Science and Technology (ethics committee for Vector Control Division, Ugandan Ministry of Health) who approved the age of consent as 15 yr at the time of sample collection (2004/2005). Consent forms were translated into the local language and informed written consent was obtained from all adults and from the parents/legal guardians of all children. Parental consent was not sought for individuals between 15–18 yr old.
The life cycle of a Puerto Rican strain (NMRI) of S. mansoni was maintained via routine passage through Biomphalaria glabrata snails and C57BL/6 mice (Harlan, United Kingdom) [28]. Mixed-sex schistosomula (cultured for 24-hr), miracidia and adult worms (7-week old) were obtained as previously described [29].
PSI-BLAST [30] searches were performed with five separate representative Ly6 proteins obtained from NCBI—Homo sapiens CD59 ([AAA60957]; [24]), H. sapiens uPAR ([Q03405.1]; [25]), C. elegans ODR-2 ([NP_001024089]; [21]), D. melanogaster Boudin ([NP_572373]; [18]) and D. melanogaster Rtv ([NP_572693]; [20]) against the S. mansoni RefSeq database (genome version 5.0) [31]. Selection of potential S. mansoni Ly6 homologues during PSI-BLAST searches required the protein sequence to contain three characteristic features of uPAR/Ly6 domains used previously to identify Ly6 proteins [18]: 1) presence of ten or more cysteine residues within a 60–110 aas region (lower and upper amino acid length defined using existing uPAR/Ly6 domains (Pfam domain: PF00021; [15])), 2) presence of an N-terminal motif C1-X-X-C2 (where C1 represents the first cysteine residue, X represents any amino acid and C2 represents the second cysteine residue) and 3) presence of a C-terminal C10-N motif (where C10 represents the tenth cysteine residue and N represents asparagine). Newly identified Ly6 homologues were incorporated into the search matrix until no more members could be identified, typically after five to six rounds of iterative searching. Then, these sequences were used as novel queries to identify any further homologues remaining in the S. mansoni genome. To identify putative orthologues for SmLy6A-K in the S. haematobium and S. japonicum genomes, tBLASTn searches using each of the SmLy6 family members was performed against the latest publically available gene predictions (searches performed 17-08-2013) from those genomes (downloaded from SchistoDB.net and GeneDB.org respectively).
Alignment of the SmLy6 family members and HsCD59 protein sequences was performed using MUSCLE software [32] with additional manual editing based on canonical residues implemented as necessary. The prediction of signal peptides was performed using the software SignalP 3.0 [33] and presence/absence of signal peptides was defined by the default Neural Network Dscore threshold of 0.43. Predicted hydrophobic regions in SmLy6 proteins were assessed using TMpred software [34] and potential GPI anchor sites were determined using the BIGPI server [35] with the site possessing the lowest e-value indicated.
S. mansoni total RNA was isolated [10] and used as templates to synthesise cDNA by reverse transcription as previously described [36]. SmLy6 sequences were confirmed by PCR amplification of cDNA obtained from either 7-week adult mixed-sex worms (SmLy6A-I) or miracidia (SmLy6J, SmLy6K) using Phusion proof-reading polymerase (Finnzymes). Nucleotide sequences for each SmLy6 family member are deposited in Genbank (accession numbers provided in Table 1) and primers used to amplify SmLy6 sequences by PCR are listed in S1 Table.
Data from the 37,632 element S. mansoni long-oligonucleotide DNA microarray studies of Fitzpatrick et al. were interrogated to find the transcription pattern of SmLy6 family members in 14 life-cycle stages [10]. A full set of raw and normalized data is available via Array express under the experimental accession number E-MEXP-2094.
Tertiary structural modeling of SmLy6 family members was performed using the predicted mature protein sequences of SmLy6A-K produced by translating the sequenced transcripts and removing the amino acids encoding signal peptides and regions C-terminal to the GPI anchor.
Given the lack of significant sequence identity to any known structure and the small size, the structure of the protein was predicted using Rosetta [37] following the protocol described elsewhere [38] with the modification that only 1000 models were taken forward for full-atom minimization. Structural models were ranked using Rosetta’s energy function and the top 20 models were visually inspected to assess the convergence of the models towards a similar fold. Models were visualized using PyMOL (DeLano Scientific LLC).
Oligonucleotide primers incorporating XbaI and XhoI restriction sites were used to amplify SmLy6A (forward primer: 5'–TCTAGAATGCATCGTTGTTATGTG—3'; reverse primer 5'–CTCGAGTGTTTGTGTACCAGAA—3') and SmLy6B (forward primer: 5'—TCTAGAATGATAAAAAATAAGAAAGTC—3'; reverse primer 5'—CTCGAGATGTTTAGGTGATGC—3') using Platinum Taq DNA high fidelity polymerase (Invitrogen) and a pJET1.2/blunt plasmid (Fermentas) containing either SmLy6A or SmLy6B full-length open reading frames as templates. The primers were designed to amplify the SmLy6 sequences encoding the predicted mature protein of SmLy6A (His28—Thr104) and SmLy6B (Ile20—His101). These products were inserted into the XhoI and XbaI sites of a modified pET30a expression vector (Novagen) designed to add a C-terminal 6xHis tag. Both pET30a/SmLy6A and pET30a/SmLy6B were sequenced (IBERS sequencing facility, Aberystwyth University) before use to confirm correct reading frame and encoded amino acids.
As both recombinant (r)SmLy6A and B were expressed, purified and dialysed using the same protocols, we list here the common set of methods used but only refer to rSmLy6A production to avoid confusion. The pET30a/SmLy6A plasmid was transformed into chemically competent E. coli BL21 star (DE3) cells (Invitrogen) and expression of rSmLy6A followed the protocols listed in the BL21 star (DE3) manual. Four hours after isopropyl β-D-1-thiogalactopyranoside (IPTG) induction (0.5mM final concentration), bacteria were pelleted, lysed and the resultant insoluble fraction was then used to purify the recombinant protein. The pelleted insoluble fraction containing rSmLy6A was first re-suspended in a wash buffer consisting of 500mM NaCl, 50mM Tris-HCl, 10mM EDTA, 0.5% Triton-X-100, pH 8. After one hour on a rocking platform, the sample was pelleted and a second wash (wash buffer + 3M urea) was performed using the same procedure as above. After these wash steps, purification of rSmLy6A was performed under denaturing conditions using Ni-NTA agarose beads (Qiagen) according to the manufacturer’s instructions. The resultant purified protein had urea removed by stepwise dialysis against ≥20 vols of 100mM NaH2PO4, 10mM Tris-HCl, pH 6.3 buffer containing 6M, 3 M, 1 M, then two buffer changes without urea. Each dialysis step was performed for a minimum of two hours. The final sample was clarified by centrifugation at 21, 000 g for 15 min at 4°C.
The identity of rSmLy6A and rSmLy6B as the sole purified product was confirmed by in-gel trypsin digestion [39], followed by mass spectrometric analysis of extracted peptides (LC-MS/MS for rSmLy6A; protocol as listed in [39], MALDI-TOF for rSmLy6B; protocol as listed in [40]) and subsequent sequence identification by a Mascot (version 2.2.1; Matrix Science) database search of the S. mansoni predicted proteins from genome assembly 4.0. Recombinant SmTAL1 was expressed and purified as previously described [41].
Antisera against rSmLy6A or rSmLy6B emulsified with Alum adjuvant (Alu-Gel-S, Serva, Germany) were raised in male 8-week-old Fischer rats (Charles River). Intraperitoneal injections of rSmLy6A or rSmLy6B were performed on days 0 (50μg), 21 (30μg) and 35 (30μg) with terminal bleeds performed at day 56.
Soluble parasite proteins were prepared from mixed-sex 24-hr schistosomula and 7-week adult worms by the addition of an extraction buffer (5mM Tris, 400mM KCl, 1% Triton-X100, 10mM EDTA, 1% Protease Inhibitors (Sigma-Aldrich) pH 9.2), sonication, two freeze/thaw cycles and centrifugation at 17, 000 x g for 15 minutes. Soluble parasite proteins were separated using SDS–PAGE, transferred to a polyvinylidene difluoride (PVDF) membrane and then probed as previously described [42]. The polyclonal anti-rSmLy6A rat antiserum was used at a 1:5,000 dilution while the polyclonal anti-rSmLy6B rat antiserum was used at a 1:2,000 dilution. Rabbit anti-rat IgG peroxidase-conjugated secondary antibody (Sigma-Aldrich) (1:5000) and chemiluminescence (Chemiluminescent Peroxidase Substrate-3) were used according to the manufacturer’s recommendation (Sigma-Aldrich). Western blots of rSmLy6A and rSmLy6B were performed using the same method and reagents as listed above with the replacement of both primary antiserum and secondary antibody incubations with a single HisProbe-HRP (Pierce) incubation at 1:4,000 dilution.
The study cohort included inhabitants of Musoli, a fishing community on Lake Victoria, Uganda, Africa. Descriptions of cohort selection, quantitative parasitology and treatment regimes for this study are communicated in a previous publication [43]. In this report we focused on 216 members of the cohort who were under 60 years of age and who donated blood before and 9 weeks after they received praziquantel (PZQ).
IgE, IgG1 and IgG4 levels for recombinant Tegumental Allergen-Like protein 1 (rSmTAL1) were measured by isotype-specific ELISA as described previously [43]. IgE, IgG1 and IgG4 levels against rSmLy6A and rSmLy6B were measured using a similar protocol with the following modifications. Coating concentrations were 6.25 and 9.6μg/ml for rSmLy6A and rSmLy6B respectively. To measure IgE, plasma was diluted 1:20 with 10% (v/v) fetal calf serum (FCS) and to measure IgG1 or IgG4 plasma was diluted 1:200 with 1% (v/v) FCS. Wells were incubated overnight at 4°C, washed and then incubated for 4 h with 0.5 μg/ml biotinylated mouse anti-human IgE (Clone G7-18, Pharmingen), biotinylated mouse anti-human IgG1 (Clone G17-7, Pharmingen) or biotinylated mouse anti-human IgG4 (Clone JDC-14, Pharmingen), followed by 1:3000 streptavidin/biotinylated-HRP complex (Mast Group Ltd.). The assay was then developed with o-phenylenediamine substrate solution (Sigma) and stopped with 2 M sulphuric acid. Standard curves were generated by control immunoglobulins including IgE (Calbiochem), IgG1 (Sigma) or IgG4 (Sigma) as appropriate. Plasma samples from 26 uninfected European/North American donors were included in each assay.
Age was divided into the following groups to account for the non-linear relationship between infection intensity and age: 7–9yrs (n = 39), 10–13yrs (n = 41), 14–23yrs (n = 42), 24–32yrs (n = 44), 38–50yrs (n = 45). Pre-treatment antibody responses were classed as a binomial variable: "responders" and "non-responders". Responders were individuals with Ab response greater than the mean + 3 standard deviations of the response of a plasma panel donated by European/North American individuals. Pre-treatment and post-treatment antibodies levels of positive pre-treatment responders to recombinant proteins were compared using a Wilcoxon signed-rank test.
As Ly6 proteins share little sequence identity [16,18], we performed a systematic search of the S. mansoni genome using position-specific iterated BLAST search (PSI-BLAST; [30]). First iteration searches were performed using five representative Ly6 proteins (HsCD59, HsuPAR, DmBoudin, DmRtv and CeODR-2) as query sequences. Sequences for use in subsequent iterations were selected by identifying regions between 55–115aa in length containing 10 cysteines, where Cys1 and Cys2 were separated by any two residues (C1-X-X-C2) and Cys10 was followed by an Asn residue (C10N) (see Materials and Methods for detailed description of PSI-BLAST parameters). This comprehensive search of the S. mansoni genome, using canonical characteristics of the Ly6 family, resulted in the identification of eleven S. mansoni Ly6 members named SmLy6A-K (see Table 1). All eleven SmLy6 transcript sequences were verified by PCR (PCR primers listed in S1 Table) from parasite cDNA, with complete open reading frames (ORF) verified for nine transcripts (SmLy6A-H and J) and partial ORFs confirmed for two (SmLy6I, K). Of these eleven Ly6 members, eight have been previously reported—SmLy6A and SmLy6B in a tegumental proteomic study ([11]; SmCD59a and b respectively), SmLy6D, which is the tegumental vaccine candidate Sm29 [44,45] and seven SmLy6 members (SmLy6A-F, I and K) in a recent publication also characterizing this family [27]. SmLy6G, Ly6H and Ly6J represent novel S. mansoni family members.
Examination of SmLy6 gene positions within the S. mansoni genome finds seven of the eleven family members present on the chromosome 1 with four present in a 41,000bp contiguous cluster (SmLy6I, SmLy6K, SmLy6F and SmLy6C; Table 1). Putative orthologues for SmLy6 genes were also identified in the S. haematobium [12] and S. japonicum [46] genomes using BLAST searches against gene predictions, with SmLy6A, SmLy6B, SmLy6H, SmLy6J and SmLy6K possessing particularly high levels of sequence identity to S. haematobium orthologs (≥80% at the protein sequence level; Table 1).
Alignment of the amino acid sequences encoded by SmLy6A-K revealed low levels of identity (average 28%) across the putative uPAR/Ly6 domains, but did show conservation of the ten canonical cysteine residues across all SmLy6 proteins (Fig 1; HsCD59 sequence included as a representative Ly6 protein). A single uPAR/Ly6 domain was identified in all but one of the SmLy6 protein, with SmLy6D containing two. Examination of the two SmLy6D uPAR/Ly6 domains finds that the two cysteines (C7 and C8) which form the fifth disulphide bond in the uPAR/Ly6 domain are absent in the 1st domain but present in the second (Fig 1; yellow residues). For all eleven family members, the spacing between the C-terminal cysteines—C8 to C9 (0 or 1 residues) and C9 to C10 (3–5 residues)—is consistent with the inclusion of these proteins into the uPAR/Ly6 superfamily (Fig 1).
All eleven SmLy6 proteins also contained a predicted hydrophobic signal peptide sequence (SignalP program; [33]) as well as a C-terminal hydrophobic region (TMpred program; [34]), both of which are characteristic of Ly6 proteins (Fig 1; signal peptide in red, hydrophobic region in blue). Predicted GPI-anchor sites (BIGPI server; [35]) were found to be highly conserved in all SmLy6 proteins with eight of the sequences (SmLy6A, B, C, E, F, G, H, I) most likely to be modified at the conserved N residue, while the GPI-attachment sites for SmLy6D, J and K located nearby (Fig 1; green residues). Examination of the SmLy6 amino acid sequences fails to find conservation of the four residues (Fig 1, boxes) known to be important in HsCD59 function [47] in any of the SmLy6 family members.
To explore the tertiary structural characteristics of the mature SmLy6 proteins (signal peptide and residues C-terminal to the GPI anchor removed), ab initio modeling was performed (Fig 2) as homology modeling is an unsuitable technique for proteins with limited sequence similarity to homologs [48]. Importantly, all eleven SmLy6 structure models produced by these analyses possessed a conserved core fold, consisting of four beta strands (yellow arrows) forming a beta sheet (Fig 2). These four beta strands (e.g. SmLy6A) match the characteristic ‘three-fingered’ fold of the Ly6 family members such as HsCD59 [19].
Using information available from a DNA microarray database [10], the mRNA abundance for 8 of the 11 SmLy6 genes (SmLy6A-D and F-I) across 14 different schistosome life-stages was deduced (S1 Fig). The results for SmLy6A (Fig 3A) and SmLy6B (Fig 3B) showed a similar trend (low/no expression in egg/snail-residing parasite stages, but rising abundance within mammalian-residing parasite stages). However, SmLy6B was more abundantly transcribed (~ 2–4X) in the earlier schistosomula stages analysed (3 hr—3 day) when compared to SmLy6A.
Recombinant SmLy6A and B were expressed in E. coli cells (BL21 star), purified via the C-terminal 6xHis tag under denaturing conditions using Ni-NTA (nickel-nitrilotriacetic acid) affinity chromatography and sequenced to confirm identity (rSmLy6A - Fig 3C, rSmLy6B - Fig 3D). Rat anti-sera raised against both rSmLy6A and rSmLy6B detected native proteins in soluble worm antigen preparations of the correct approximate molecular mass for both native SmLy6A (8.03kDa; Fig 3E) and SmLy6B (8.85kDa; Fig 3F). Anti-rSmLy6B also recognized an appropriate molecular mass protein corresponding to native SmLy6B in soluble 24-hr schistosomula protein extracts (Fig 3F).
Antibody responses against rSmLy6A and rSmLy6B were measured in the plasma of a S. mansoni infected male (aged 7–70 yrs) cohort [43] from a high transmission area in Uganda at two time-points—before and 9 weeks after PZQ treatment (Fig 4). To assess the age profiles of the responses, anti-rSmLy6A and rSmLy6B IgG1, IgG4 and IgE levels were plotted for five age groups (7–9, 10–13, 14–23, 24–32 and 33+ years) and positive responders (seropositive) were defined as those individuals exhibiting mean anti-rSmLy6 titers above the mean level observed from uninfected European/North American samples + 3x standard deviation (Fig 4).
Before PZQ treatment (Fig 4) detectable IgG1 responses to SmLy6A were rare, present in only 7% of the cohort (16 individuals, Fig 4A). In contrast, a sizable percentage (25%) of the cohort contained seropositive rSmLy6B IgG1 levels before treatment (Fig 4D). No association was observed between infection intensity and rSmLy6A or rSmLy6B IgG1 responses (S2 Table). Similarly, pre-treatment IgG4 responses in the cohort also showed that responses to rSmLy6B were more abundant (2.63μg/ml geometric mean) and more common (21%, Fig 4E) than rSmLy6A—IgG4 responses (0.04μg/ml geometric mean, 10% of the cohort, Fig 4B). Finally, IgE responses to rSmLy6A and rSmLy6B (Fig 4C and 4F) were rarer but detectable in some individuals (9 and 4% respectively).
To further analyse the endemic human antibody responses to rSmLy6A and rSmLy6B in this infected cohort, pre- and post-PZQ treatment antibody levels were measured. In addition to rSmLy6A and rSmLy6B, antibody levels against rSmTAL1 (a non-surface tegumental antigen previously known as Sm22.6), were compared to assess whether protein localization within the tegument may influence antibody responses. For IgG1 responses, anti-rSmLy6A and rSmLy6B showed no age-associated profile either pre- or post-PZQ treatment, with prevalence in all age groups comparable (Fig 5A and 5B). In contrast, SmTAL1-IgG1 responses showed clear age-associated profiles of increasing prevalence with age in both pre- and post-PZQ treatment (Fig 5C). This finding was also observed in the IgG4 and IgE responses where no age-associated relationships were observed pre- or post-PZQ treatment to SmLy6A and B, when compared to a positive association detected for rSmTAL1 (see S2 Table).
In those males that were seropositive pre-treatment for rSmLy6B, levels of rSmLy6B-IgG1 decreased significantly at 9 weeks post-treatment (p<0.001 n = 50; Fig 5E) with a similar trend observed for SmLy6A-IgG1 (p = 0.02, n = 16; Fig 5D). This is the reverse of the SmTAL1-IgG1 response dynamics, where the response increased significantly at 9 weeks post-treatment (p<0.001 n = 26; Fig 5F). For IgG4, the relatively low levels of rSmLy6A-IgG4 did not significantly change between pre-treatment and 9 weeks post-treatment (S2 Fig), but there was a significant drop in rSmLy6B-IgG4 (p<0.001, n = 193; S2 Fig). The minimal IgE responses to rSmLy6A and rSmLy6B did not change significantly following treatment (S2 Fig).
Investigating the human immune responses directed against schistosome surface proteins is a logical step in the progression of next-generation vaccine candidates to combat schistosomiasis [49]. Here we have focused our investigation on the Ly6 protein family as previously studies have indicated that three members are found on the schistosome surface tethered by a GPI-anchor (SmLy6A, SmLy6B and SmLy6D; [11]), two are capable of inducing protective immunity in the mouse model (SmLy6B [50] and SmLy6D [44]) and one is recognised during human schistosomiasis (SmLy6D; [45]).
Our genome analysis of this family identifies eleven members (SmLy6A-K), each possessing the characteristic features of Ly6 proteins including a signal peptide, a uPAR/Ly6 domain in a TFPD fold with conserved cysteine spacing, a potential GPI-anchor site and a C-terminal hydrophobic region (see Table 1 and Fig 1). Positioning in the genome suggests recent gene duplication events, but the lack of highly similar (>95% identity) orthologues in the S. japonicum and S. haematobium genomes likely reflects rapid divergence of these genes as observed in Ly6 genes found in other species [14,16,18]. With the identification of SmLy6A, B, C and D in the adult worm tegument [11,26] and Ly6 family members also present in the tegument of Fasciola hepatica adults [51], it is clear that a sub-set of these schistosome proteins function at the host/parasite interface. Farias et al. have clearly demonstrated that SmLy6 members do not function as CD59 orthologs [27], but whether these schistosome proteins interact with other host molecules or help maintain the tegumental barrier (potentially related to the septate junction function of Dm-boudin; [18]) is currently unknown.
The nomenclature used to describe these proteins within the schistosome scientific community is currently inconsistent with SmCD59a, SmCD59b, Dif-5 and Sm29 all previously being used (see Table 1). Additionally, during the completion of this study, a report by Farias et al. described a sub-set of this Ly6 family in which the authors explicitly named them SmCD59.1-.7. [27]. We have chosen to use the name SmLy6 for these proteins in our study as: (a) sequence analysis does not support greater similarity to CD59 than other specific Ly6 members, (b) functional orthology to the CD59 protein was also comprehensively disproved by Farias et al. [27] and (c) the most commonly used designation for this class of protein is Ly6 (signal peptide-containing, GPI-anchored proteins with a TFPD fold) [21,23]. The discrepancy between the number of Ly6 members identified in this study (eleven members) and that described by Farias et al. (seven members) reflects the different search methods used. Our PSI-BLAST methodology has been used previously to identify the D. melanogaster Ly6 family and is particularly powerful where sequence similarity is low overall, but specific protein motifs are conserved [18]. In this way we have successfully extended the family membership by over 50% from the search performed by Farias et al., however it remains possible that other divergent members were not identified in this study (specifically smp_202630 and smp_064430 as they bear similarities to the family such as the presence of the C1-X-X-C2 and C10N motif). Perhaps the most surprising new member of the Ly6 family is the potential vaccine candidate Sm29 (Table 1; SmLy6D).
Sm29 was first described in 2006 as a tegumental protein of unknown function and sharing no sequence similarity to any known protein family [45]. However, our PSI-BLAST analysis of the S. mansoni genome demonstrated that Sm29 contained all major Ly6 features (Fig 1; SmLy6D) with ab initio modeling clearly showing the TFPD fold characteristic of the Ly6 family (Fig 2; SmLy6D). The presence of two uPAR/Ly6 domains in SmLy6D makes it unique in the SmLy6 family, as all the other members have only one domain (Fig 1). This particular feature is similar to human uPAR protein, which contains three tandem uPAR/Ly6 domains. SmLy6D is also comparable to uPAR by lacking the C7-C8 disulphide bond within the first uPAR/Ly6 domain (Fig 1). uPAR’s binding with uPA has been shown by site-directed mutagenesis to be coordinated by four residues present in the first uPAR/Ly6 domain (Arg53, Leu55, Tyr57, and Leu66) between C6 and C9 [52]. Intriguingly, in the same region of its second uPAR/Ly6 domain, SmLy6D possesses an Arg, Leu, Tyr, Ile tetrapeptide sequence (Fig 1). Whether these four residues truly form the functional residues for Sm29/SmLy6D is beyond the scope of this study, however, we believe it warrants further investigation.
Our co-measurement of SmTAL1 and SmLy6A/SmLy6B serological responses in the same infection cohort provided an opportunity to compare antibody isotype responses to proteins found on (SmLy6A/SmLy6B; [11,27]) or below (SmTAL1; [43]) the schistosome surface. Here, we observed clear differences in both the age-associated profiles as well as the predominant serotypes between SmTAL1 and the studied SmLy6 proteins (Fig 5).
As shown in Fitzsimmons et al., almost half of this cohort was seropositive for SmTAL1-IgE and SmTAL1-IgG4 (45% in both cases), with fewer individuals seropositive for SmTAL1-IgG1 (12%; [43]). These strongly IgE/IgG4-dominated serotype responses have also been observed against other non-surface (cryptic) S. mansoni proteins such as Cathepsin B1 (SmCB1—a gut associated peptidase; [53]) and Glutathione-S-transferase (Sm28-GST—a parenchymal protein [54]). The IgE/IgG4 dominated response against SmTAL1 and other non-surface proteins [55] contrasts with the results observed for SmLy6A and SmLy6B, which comprise mainly of IgG1 antibodies, with lower levels of IgG4 and IgE (Fig 4). These findings are consistent with other vaccine candidates found on or near the surface of the parasite [56] such as Tetraspanin-2 (SmTSP-2) [57], Sm23 [58] and Glyceraldehyde 3-Phosphate Dehydrogenase (SG3PDH) [59,60]. Amongst the SmLy6 family, serological responses within a small Brazilian cohort of schistosome-infected humans against the surface-exposed SmLy6D (Sm29) have previously been investigated [45]. Here, the authors identified high anti-SmLy6D IgG1 titres, with no anti-SmLy6D IgG4 or IgE component detected in infected individuals. This is broadly similar to our study. However, we have also identified a sizable percentage of low titre IgG4 responders and a few detectable IgE responses to both SmLy6A and SmLy6B (Fig 4). These data indicate that single Ly6 (SmLy6A and SmLy6B) versus multiple Ly6 (SmLy6D)-domain containing family members can induce different immunological responses in endemic populations. Whether these isotypic differences between single and double domain Ly6 family members are found in other schistosome-infected cohorts, are influenced by differential SmLy6 protein abundance or whether other single Ly6-domain containing family members also elicit an IgG4/IgE response is currently unknown, but requires further investigation for future immunoprophylaxis consideration of the SmLy6 family.
SmLy6A and SmLy6B serological responses also differ from SmTAL1 in terms of age-association profiles (Fig 5A–5C) and post-treatment PZQ effects (Fig 5D–5F). Collectively, and regardless of prevalence (SmLy6B – 25% > SmLy6A – 7%), anti-SmLy6 IgG1 immune responses did not significantly change with age and dropped upon PZQ treatment (Fig 5A and 5B) in the studied population. In contrast, IgG1 responses against SmTAL1 demonstrated a positive association with age group analysed and were boosted by chemotherapy (Fig 5C). These data suggest that, while sub-surface SmTAL1 only becomes exposed to host antibody responses upon natural or drug mediated parasite death [43], SmLy6 proteins (SmLy6B>SmLy6A) are more likely to be recognised in endemic populations due to their surface localisation in live parasites. However, the anti-Ly6 antibody response may also be influenced by the difference in surface exposure of these antigens during parasite development as recently demonstrated for SmLy6A (SmCD59a) [61]. Here, Reimers et al. illustrated that SmLy6A was not abundantly found on the surface of both 7 and 14 day schistosomula (despite being detectable in soluble protein extracts), indicating that this particular Ly6 member is inaccessible to antibodies unless damaged [61]. The prevalence/magnitude of anti-SmLy6 IgG1 responses may, therefore, be linked to antigen abundance (surface or sub-surface) in the schistosomula stages (SmLy6B is expressed in earlier lifecycle stages than SmLy6A; Fig 3) and/or low-level cross reactivity (SmLy6A and SmLy6B share 42% sequence identity at the amino acid level). Support for cross-reactivity is present in this study, with all of the SmLy6A IgG1 responders (n = 16) also IgG1 responders to SmLy6B. However, between isotypes, there is minimal overlap in individuals responding with no individuals positive for all three isotypes tested (IgG1, IgG4 and IgE) against SmLy6A and only three of the eighty-five individuals positive for all three antibody isotypes against SmLy6B.
In this study, we have comprehensively characterized the S. mansoni Ly6 family, finding eleven members (three novel proteins and the new classification of Sm29 as a Ly6 family member) that possess protein features and tertiary fold structures distinctive of membrane-associated localisation. Comparative isotypic, age-association and post-PZQ treatment antibody responses to two representative members (SmLy6A and SmLy6B) in a cohort of S. mansoni infected males from a Ugandan fishing community strongly support their membrane association at the adult schistosome surface (supporting [11] and [27]), when viewed in light of the immunological results generated for sub-surface SmTAL1. This unique immunological comparison within the same cohort provides clear evidence for the plasticity of the human immune response to different parasite proteins depending on lifecycle expression and tegumental localisation. We, therefore, contend that both temporal and spatial expression of SmLy6 members are equally important in initiating immune responses; a factor that may have relevance in the progression of this family as next generation immunoprophylactic candidates.
|
10.1371/journal.pbio.2006926 | Differential and convergent utilization of autophagy components by positive-strand RNA viruses | Many viruses interface with the autophagy pathway, a highly conserved process for recycling cellular components. For three viral infections in which autophagy constituents are proviral (poliovirus, dengue, and Zika), we developed a panel of knockouts (KOs) of autophagy-related genes to test which components of the canonical pathway are utilized. We discovered that each virus uses a distinct set of initiation components; however, all three viruses utilize autophagy-related gene 9 (ATG9), a lipid scavenging protein, and LC3 (light-chain 3), which is involved in membrane curvature. These results show that viruses use noncanonical routes for membrane sculpting and LC3 recruitment. By measuring viral RNA abundance, we also found that poliovirus utilizes these autophagy components for intracellular growth, while dengue and Zika virus only use autophagy components for post-RNA replication processes. Comparing how RNA viruses manipulate the autophagy pathway reveals new noncanonical autophagy routes, explains the exacerbation of disease by starvation, and uncovers common targets for antiviral drugs.
| Viruses often co-opt host cellular processes to replicate their genomes and spread to other cells. Many of these cellular pathways provide good targets for antiviral drugs, as they are less likely to develop resistance since they are encoded in the host and not the fast-evolving viral genome. The autophagy pathway is an important stress response pathway that allows cells to recycle cellular components for energy conservation by sequestering cytoplasmic molecules and organelles in double-membraned vesicles (DMVs) and by degrading the contents into reusable elements. Many RNA viruses induce this pathway to provide membrane surfaces for replication and as a source of vesicles for maturation and exit from cells. We developed a panel of CRISPR (Clustered Regularly Interspaced Short Palindromic Repeats) knockout (KO) human cells lacking individual components of the autophagy pathway to assess what aspects of the pathway diverse RNA viruses utilized. We discovered that poliovirus, dengue virus, and Zika virus all use different initiation components of the autophagy pathway but similar downstream components. Additionally, we found that poliovirus uses autophagy components for genome replication, while dengue and Zika viruses use autophagy components for postreplication processes. Ultimately, we uncovered potential drug targets for multiple RNA viruses.
| Manipulation of the autophagy pathway is a burgeoning field of research, providing many potential targets for antiviral, anticancer, and neuro-preservation therapies. The autophagy pathway is a highly conserved cellular response that is induced upon nutrient deprivation, other stresses, and developmental cues. Canonical autophagy (“self-eating”) proceeds through a series of distinct steps that nucleate and expand membranous structures, termed autophagosomes, that enclose cytoplasmic contents [1] (Fig 1A). The resulting double-membraned vesicles (DMVs) then fuse with lysosomes, in which hydrolases promote degradation of the cytoplasmic contents for reuse by the cell. The autophagy pathway utilizes large amount of lipids to accomplish the formation of autophagosomes. The origins of these membranes are debated but are likely to derive both from the endoplasmic reticulum (ER) and lipids scavenged from a variety of membranes throughout the cell [2–6].
There are over 30 autophagy-related genes (ATGs) that were first identified in yeast, most of which have mammalian homologs [1]. Autophagy gene products take part in several distinct complexes (Fig 1A). In human cells, the specific kinase activity of the complex that contains Unc-like autophagy-activating kinase (ULK1), PTK2/FAK family interacting protein of 200 kDa (FIP200), ATG13, and ATG101 complex is normally repressed by mammalian target of rapamycin (mTOR) under basal conditions. Upon mTOR activation, ULK1/FIP200/ATG101 is activated, thus activating the downstream autophagy pathway [7]. mTOR repression of the ULK1 complex activity can be bypassed by another kinase, AMP-activated protein kinase (AMPK). Thus, ULK1 activation can be achieved by two different routes of autophagy induction [8]. This complex is essential for the downstream formation of autophagosomes via the kinase activity of ULK1 [9]. The VPS34/ beclin-1(BECN1)/ATG14L complex, also essential for the initiation and maturation of autophagosomes, relies on ULK1 kinase activity, with BECN1 serving as a direct substrate [10]. The kinase activity of VPS34 is in turn required for the production of phosphatidylinositol lipids needed to recruit downstream autophagy proteins [11,12]. The activation of ATG9, the only known intrinsically membrane-bound protein in the pathway, also requires ULK1. ATG9 is important for scavenging lipids from various sites throughout the cell and trafficking them back to growing autophagosomes [13–17]. These events lead to ATG5-mediated covalent attachment of phosphatidylethanolamine (PE) to light-chain 3 (LC3) [18], the most characteristic component of immature and mature autophagosomes. Membrane-associated LC3 is needed for recruitment of cargo, stabilization of negative curvature, closure of the autophagosome, and fusion with lysosomes [19–22]. While each of these proteins is considered essential for the canonical autophagy pathway, several instances of mammalian autophagy that lack one or more of these components have been reported, including ULK1-independent, BECN1-independent, and ATG5-independent formation of DMVs and degradation of cytoplasmic contents [23–25]. Such processes are termed “noncanonical” autophagy. This also includes secretory autophagy, in which cytoplasmic components are released undegraded into the extracellular milieu [26].
Many viruses interact with the autophagy pathway, which can play both antiviral and proviral roles, sometimes even during the same viral infections. Antiviral activities of autophagy include xenophagy, in which intracellular viruses are targeted for degradation as well as the triggering and facilitation of immune responses [27]. As a testament to the power of autophagy as an arm of innate immunity, many successful viruses such as herpesviruses [28], Sindbis virus [29], and vesicular stomatitis virus [30] have evolved strategies to block the induction of the autophagy pathway. On the other hand, utilization of components of the autophagy pathway by many viruses serves proviral functions. The unique ability of the autophagy pathway to generate membranous structures de novo and to allow or disallow their acidification may be desirable features for any virus that survives by manipulating intracellular membranes. All positive-strand RNA viruses associate with cytoplasmic membranes to replicate their genomes. Virally associated membranes come from a variety of cellular compartments, including ER, Golgi, mitochondria, and lysosomes [31–34]. Many viral proteins are membrane-bound, and RNA amplification occurs on the topologically cytoplasmic surfaces of vesicles. Viruses such as dengue, for instance, induce ER-membrane invaginations to form isolated pockets in which viral proteins congregate for RNA replication outside the reach of cellular antiviral factors.
For poliovirus (PV), dengue virus (DENV), and Zika virus (ZIKV), proviral roles of the autophagy pathway and its components have been documented. PV and other picornaviruses have been shown to induce the formation of autophagosome-like membranes for purposes of RNA replication, virion maturation, and nonlytic viral spread [34–40]. Coxsackievirus B3 (CVB3) has also been shown to use the autophagy pathway for viral spread and to induce the formation of DMVs [37,41]. DENV has been shown to induce the proliferation of LC3-containing membranes [31,42–44]. Inhibiting PI 3-kinases, including VPS34, with 3-methyladenine (3-MA) decreases DENV yield [45], and a specific autophagy inhibitor, spautin-1, deranges the maturation of DENV particles [46]. ZIKV has also been shown to induce the formation of LC3-containing membranes [47,48]. Furthermore, one study suggests that noncanonical secretory autophagy may contribute to the spread of ZIKV [49], as it does with PV and CVB3.
To determine which components of the autophagy pathway are used by these RNA viruses and whether there are any shared principles, we generated a panel of CRISPR (Clustered Regularly Interspaced Short Palindromic Repeats)-Cas9 knockout (KO) human epithelial-derived cell lines (HeLa). Previously, most research investigating viruses and cellular autophagy has involved targeting single genes genetically or pharmaceutically. This does not take into account the potential utilization of noncanonical autophagy pathways or gain-of-function effects of drugs [50]. We found that PV, DENV, and ZIKV all utilize multiple components of the autophagy pathway while bypassing others and that each virus uses a unique set of initiation components. A common feature is that all of the tested viruses require the LC3 protein but bypass its canonical cellular lipidation process, using other means to recruit LC3 to virally induced membranes. This study highlights the importance of assessing the full autophagy pathway when seeking to understand how pathogens manipulate this pathway for purposes of genome replication and spread, as well as potential common drug targets.
To provide an in-depth characterization of which autophagy components are utilized by the three different viruses, we targeted either one or two genes from distinct complexes required for canonical autophagy (Fig 1A). Sequencing confirmed the genomic ablation of the targeted genes in ΔFIP200, ΔATG9, ΔBECN1, ΔVPS34, ΔLC3, and ΔATG5 (S1A Fig). For GC-rich ULK1, specific antibodies confirmed a functional loss of protein, as they did for the targeted gene products in all the CRISPR-Cas9 KO cell lines (S1B Fig). To test whether canonical autophagy was inactivated in these KO cell lines, the accumulation of p62/SQSTM (p62), which is degraded in wild-type (WT) cells via constitutive autophagy, was monitored (Fig 1B). Accumulated p62 was present at low abundance in the parental HeLa cells (WT) and increased in cells treated with chloroquine (CQ) to prevent the formation of autolysosomes and thus the degradation of autophagosomal contents. All KO cell lines showed enhanced constitutive p62 accumulation, whether or not they were treated with CQ, indicating that the canonical basal autophagy pathway was successfully inhibited (Fig 1B). Additionally, to confirm that the canonical, starvation-induced autophagy pathway was obstructed in all of the KO cell lines, we quantified the presence of endogenous LC3 puncta upon starvation. WT cells displayed significantly more puncta upon starvation than any of the KO lines (S1C Fig).
We then took advantage of these generated KO cell lines to determine which components of the autophagy pathway affect viral growth for a picornavirus (PV) and a flavivirus (DENV). PV is a nonenveloped virus that matures intracellularly. The intracellular amplification of PV in WT and KO cell lines was monitored in a single infectious cycle, which comprises cell entry, translation, RNA replication, and packaging but not cell-to-cell spread. As shown in Fig 1C, cells ablated for ULK1, FIP200, ATG9, and LC3B had significant viral growth defects at this time point. However, viral titers in cells ablated for BECN1, VPS34, and ATG5 were indistinguishable from WT. More detailed time courses of a single PV infectious cycle confirmed the reduction in virus yield in ΔULK1, ΔFIP200, ΔATG9, and ΔLC3 cells (Fig 1C). The fact that ATG5 is not required is particularly curious, given that viral growth was drastically reduced in the ΔLC3B line (Fig 1C). These observations argue that the virus still utilizes the LC3 protein, though perhaps not its lipidated form. We confirmed that adding back autophagy components to our KO cell lines restored viral titers to WT levels for ULK1 and FIP200 (S2A Fig). Additionally, adding back a kinase dead ULK1 mutant (K46I) reduced viral yield in WT cells, consistent with its previously reported dominant negative effects on the autophagy pathway [51]. ULK1–K46I also failed to rescue the viral defect in ULK1 KO cells (S2A Fig).
DENV is an enveloped virus that matures upon release from cells after budding into the ER and traversing the canonical downstream secretion pathway. We monitored a single cycle of DENV growth by quantifying the release of infectious virus into the extracellular medium. All CRISPR-Cas9 KO cells were infected with DENV for a single 24-hour replication cycle at a low multiplicity of infection (MOI) (Fig 1D). We found that DENV does not require the same initiation complex as PV, as viral titers were unaffected in ΔULK1 cells. We additionally tested whether a double knock down of ULK1 and ULK2 affected DENV growth, given that ULK1 and ULK2 can be redundant in some cell types [9,52]. We saw no defect in DENV titers when small interfering RNAs (siRNAs) against ULK2 were used in the ΔULK1 cell line or when ULK1 and ULK2 siRNAs were used in the ΔFIP200 cell line (S2B Fig), suggesting that DENV does not require this complex. Similar to PV, DENV titers were reduced in ΔATG9 cells, perhaps highlighting a common need for lipid scavenging in HeLa cells. DENV utilized VPS34 but not BECN1, both part of the same complex, arguing that DENV might specifically require VPS34 but not the rest of the complex. Additionally, adding back VPS34 to the ΔVPS34 cell line restored DENV titers to WT levels (S2C Fig). Downstream in the autophagy pathway, DENV utilized LC3 but not ATG5, similarly to PV.
DENV and PV infection in cell culture benefit from the induction of the canonical autophagy pathway when treated with autophagy inducers such as rapamycin [46] or loperamide [39]. Recently, we showed that induction of the autophagy pathway by starvation led to an enhancement of PV amplification that was dependent on the canonical pathway, as ΔATG5 and ΔBECN1 cell lines failed to show enhanced viral infection upon starvation [53]. To test whether in vivo PV infection is also enhanced by starvation as well as whether this exacerbation depends on the canonical autophagy pathway, we bred C57BL/6 mice that were homozygous for a transgene expressing the poliovirus receptor (PVR) [54] as well as the Atg5 gene flanked by Lox sites, the targets of Cre recombinase. These mice were bred either to express or not express Cre recombinase under the control of a tamoxifen-inducible promoter [55]. The conditional deletion of floxed Atg5 upon tamoxifen treatment was confirmed by quantitative PCR (qPCR) of a DNA junction expected to be generated during the Cre-mediated excision (S2D Fig). The decrease of Atg5 function in the Cre-expressing mice was confirmed at the protein level by the large increase in abundance of LC3-I, which normally would be converted to LC3-II by Atg5 (S2E Fig). Although a small amount of residual LC3-II remained in the Cre-expressing mice, this is typical of the nonabsolute nature of “floxed” gene removal by Cre recombinase [55]. Mice were fasted for 48 hours before intramuscular infection with PV. Fasted mice showed increased PV titers in mice with an intact Atg5 gene (Fig 1E). However, in the absence of Atg5, the enhanced viral titer seen with fasting was not observed (Fig 1E). Additionally, we treated mice with loperamide to induce the autophagy pathway by a different mechanism [56,57] and found a similar response of enhanced viral titers in WT mice treated with the drug but no difference between untreated and drug-treated mice lacking Atg5 (Fig 1F). These data argue that the exacerbation of viral infection by starvation or loperamide-induced autophagy requires ATG5 and most likely the canonical pathway, even though poliovirus growth per se uses only portions of this pathway. We speculate that preinduction of the canonical autophagy pathway increases the concentrations of proteins or lipids that facilitate subsequent viral infection.
Previous experiments have argued that in the absence of cellular autophagy or its components, intracellular RNA amplification as well as later steps are defective in PV-infected cells [36,39]. On the other hand, DENV-infected cells lacking an intact autophagy pathway synthesize viral RNA, but virion maturation is defective [46]. To distinguish between amplification in the first round of infection and subsequent spread, intracellular RNA accumulation was monitored by reverse transcription quantitative PCR (RT-qPCR) at early and late time points (Fig 2A). Even in the first cycle of PV infection in the KO cell lines, the decline in RNA accumulation was significant in ΔULK1, ΔFIP200, ΔATG9, and ΔLC3 cells, and this effect continued upon subsequent rounds (Fig 2B). To test whether viral entry was affected in the KO cell lines, we performed an entry assay. Virus was allowed to infect cells for 30 minutes, after which virions that had not entered were stripped from the cell surface with an acid wash. RT-qPCR analysis of PV RNA that successfully entered the cell showed no difference between WT and KO cell lines (S3A Fig), indicating that the RNA accumulation defect lies downstream of entry. Intracellular accumulation of viral protein 2C was reduced in the KO cell lines shown (S3B Fig), consistent with a defect in translation, RNA replication, or both. Similarly, transfection of a PV replicon that expressed firefly luciferase showed reduced protein accumulation in the KO cell lines (S3C Fig). To distinguish between defects in viral translation and RNA replication, we monitored translation at a 2-hour time point in the absence and presence of guanidine, a specific inhibitor of RNA replication. Only WT cells differed in their accumulation of luciferase in the absence or presence of guanidine (S3D Fig). Therefore, it is RNA replication that is specifically inhibited in the KO lines.
To determine whether autophagy components influence DENV translation and RNA synthesis, viral packaging and spread, or both, viral RNA synthesis was monitored after single (24 hpi [hours post infection]) and multiple (48 hpi) rounds of infection. In the absence of ATG9, VPS34, or LC3, no reduction in DENV RNA was observed in the first cycle (Fig 2B), even though the production of infectious virus was significantly reduced at this time point (Fig 1D). This is consistent with previous observations that in the presence of autophagy inhibitor spautin-1, packaged virions were defective but RNA accumulation was not [46]. Upon additional rounds of infection, the accumulation of DENV viral RNA was greatly reduced (Fig 2B), consistent with the poor infectivity of virions from the first round. In accordance with these data, we observed similar amounts of viral protein accumulation at early time points in WT and ΔATG9 cells but reduced accumulation at later time points (S3E Fig). We conclude that in the absence of a noncanonical autophagy pathway, initial DENV translation and RNA synthesis are not affected, but infection of subsequent rounds is greatly reduced as a result of defective assembly, maturation, or egress.
ZIKV is a flavivirus that is similar to DENV but infects a larger number of cell types in humans [47]. ZIKV RNA was monitored during infection, and like DENV, viral RNA abundance remained similar to WT in the first infectious cycle (24 hpi). However, after additional infectious cycles, both RNA (Fig 2B) and virus production (Fig 2C) were greatly reduced. These data argue that for ZIKV, like DENV, autophagy components are only necessary for a postreplication process such as packaging or particle maturation. However, unlike either PV or DENV, the accumulation of infectious ZIKV was reduced in the entire CRISPR-Cas9 KO panel except the ΔATG5 line (Fig 2C). This argues that ZIKV utilizes both upstream initiation complexes ULK1/FIP200 and BECN1/VPS34, perhaps feeding into many aspects of the canonical autophagy pathway. Nonetheless, the fact that all three viruses bypass the need for ATG5-mediated lipidation of LC3B highlights a shared strategy.
To monitor the accumulation of viral RNAs on a single-cell basis, we measured the number of infected cells during multiple cycles of infection by flow cytometry. Virus-specific probes were designed to hybridize to each positive-strand viral genome. These probes can subsequently be bound by branched DNA structures, with each DNA “tree” containing thousands of fluorophores as a means to amplify the signal, which can allow the detection of single RNA molecules [58,59]. The low background for these probes from uninfected cells was validated in WT HeLa cells by confocal microscopy (Fig 2D). These probes were then used to determine the percentage of cells positive for viral RNA by flow cytometry, using a low MOI (0.1 plaque forming units [PFU]/cell) and multiple cycles of infection. In the cell lines that showed significant defects in growth for each virus, fewer cells were positive for PV, DENV, or ZIKV viral RNA (Fig 2D).
Autophagy can be induced by two different arms of upstream signaling: mTOR inactivation, leading to dephosphorylation of ULK1 and thus its activation, or AMPK activation, leading to a distinct phosphorylation of ULK1 in the absence of mTOR repression [8,60,61]. mTOR typically responds to nutrient signals while AMPK responds to the energy status of the cell. Other viruses have been shown previously to activate the autophagy pathway via AMPK activation [62]. The differential utilization of upstream autophagy components by PV, DENV, and ZIKV could result from the activation of different upstream signaling cascades. To test this possibility, we looked at the phosphorylation status of S6K, which is phosphorylated when mTOR is active and autophagy is repressed [63]. We saw that while rapamycin treatment led to the loss of S6K phosphorylation, PV, DENV, and ZIKV infection did not alter its phosphorylation status (Fig 3A). This suggests that the activation of autophagy during viral infection is independent of mTOR inactivation and thus, by definition, noncanonical. We next looked at AMPK phosphorylation status and discovered that PV and ZIKV led to AMPK phosphorylation and thus activation but not DENV (Fig 3B). Since AMPK is known to directly activate ULK1, these data support the idea that PV and ZIKV utilize the ULK1 complex in activating the autophagy pathway, while DENV bypasses both mTOR and AMPK, perhaps avoiding ULK1 activation entirely.
To determine whether ULK1 and FIP200 function in the same way in stimulating PV growth as they do in canonical autophagy, we used a small molecule inhibitor of ULK1 (MBL56), which specifically blocks its kinase activity [64]. In WT HeLa cells, addition of MBL56 decreased viral titers after 6 hours of infection in a dose-dependent manner (Fig 3C). However, MBL56 had no additional effect on viral titers in infected ΔFIP200, arguing that ULK1 and FIP200 function together just as they do in canonical autophagy.
DENV bypasses the upstream ULK1 complex and instead requires VPS34. To determine if DENV requires a catalytically functional VPS34 protein, we used VPS34-specific inhibitor SAR405 [65]. Addition of SAR405 led to reduced DENV titers at 24 hpi in a dose-dependent manner (Fig 3D). We see reduced ZIKV titers with addition of SAR405 as well, showing that ZIKV also requires a catalytically functional VPS34 (Fig 3D). Interestingly, we did not see any reduction in PV titers with addition of SAR405 (Fig 3D), consistent with the previous conclusion that PV does not require VPS34 activity. Therefore, PV and ZIKV induce the autophagy pathway upstream of AMPK, while DENV uses an alternate route to induce autophagy pathway, perhaps through the direct activation of VPS34.
To test whether the extensive membrane rearrangements that occur in cells infected with PV and DENV are altered in cell lines lacking particular autophagy components that affected viral growth, we studied the ultrastructure of infected WT and KO cells. WT, ΔULK1, and ΔATG9 cells were infected with PV at a high MOI (10 PFU/cell) for 6 hours, followed by high-pressure freezing and freeze substitution to preserve membrane structures. Images were visualized by electron microscopy (EM) (Fig 4A). PV-infected WT cells showed the characteristic DMVs shown previously to be induced either by infection or coexpression of PV proteins 2BC and 3A [34,35]. These DMV structures were not present when cells were treated with guanidine, a potent inhibitor of viral replication (Fig 4A and 4B). In ΔULK1 and ΔATG9 cells, qualitatively different membrane rearrangements were observed. Infection of ΔULK1 cells led to an increase in single-membraned vesicles, while infection of ΔATG9 cells led to an increased number of electron dense single-membraned vesicles, both of which diminished when cells were treated with guanidine (Fig 4A and 4B). Single-membraned vesicles can be the precursors of DMVs in canonical autophagy; this may also be the case in PV-infected cells [66]. Therefore, the single-membraned vesicles observed in ΔULK1 and ΔATG9 cells could represent precursors or off-pathway structures that accumulate during arrest of the functions of the ULK1 complex. The presence of such precursors could also be increased by the lower viral protein abundance in these KO cell lines (S3B Fig). No alterations in membrane structures were seen in any cells in the presence of guanidine, arguing that translation from input RNA is not sufficient to induce these changes.
DENV growth does not use the canonical ULK1 complex but instead requires both VPS34 and ATG9 (Fig 1D). To identify any alteration in membrane structure under these conditions, we visualized DENV-infected WT, ΔVPS34, and ΔATG9 cells by EM. Typically, during a DENV infection, DMVs and large convoluted membrane structures thought to be extensions of the ER are formed by 24 hpi (Fig 4C) [67]. These structures were notably absent in ΔVPS34 and ΔATG9 cells (Fig 4C and 4D), even though comparable amounts of viral RNA and protein were observed in these cell lines at this time point (Fig 2B and S3E Fig), arguing that VPS34 and ATG9 play crucial roles in the architecture of these membrane structures.
ATG5 is often considered the hallmark of canonical autophagy due to its essential function in the lipidation of LC3. Considering that we observed no defects in the growth of PV, DENV, or ZIKV in ΔATG5 cells but significant dependence on LC3, we were curious about the mechanism of LC3 utilization during these viral infections. To observe the membrane-associated sequestration of LC3 during canonical autophagy and viral infection, we monitored the relocalization of a plasmid expressing a green fluorescent protein (GFP)–LC3 fusion protein [20,68]. Under normal conditions, GFP–LC3 is dispersed throughout the cytoplasm, as shown in the top panel of Fig 5A. Upon induction of the canonical autophagy pathway, GFP–LC3 forms distinct puncta that represent membrane-bound, lipidated LC3. Under starvation conditions, GFP–LC3 puncta form in WT cells but not in ΔATG5 cells. Infection with PV, DENV, or ZIKV also led to the formation of GFP–LC3 puncta in both the presence and absence of ATG5 (Fig 5A and 5C).
To test further whether the induction of LC3 puncta by PV, DENV, and ZIKV was truly independent of LC3 lipidation, we expressed a version of GFP–LC3 that contained a G120A mutation, which lacks the essential glycine residue needed for the attachment of PE [69]. As expected, starvation did not induce the formation of GFP–LC3–G120A puncta in either the presence or absence of ATG5 (Fig 5B, upper panels). The fact that very few GFP–LC3–G120A puncta were observed in WT cells might indicate that the virus preferentially induces puncta formation with lipidated LC3, which is still present as untagged endogenous LC3. However, GFP–LC3–G120A puncta formation was observed in ΔATG5 cells upon infection with PV, DENV, or ZIKV (Fig 5B and 5C). These results suggest that all three viruses can either recruit unlipidated LC3 to membranes or can lead to lipidation of LC3 independently of ATG5.
To determine whether or not LC3 was lipidated during infection, we enriched for membrane-associated proteins and determined the lipidation status of LC3 by immunoblot. Cells were treated with rapamycin, CQ, or infected with PV. As expected, no LC3-II was observed in ΔATG5 cells, and treatment with CQ increased LC3-II abundance in WT but not ΔATG5 cells when whole-cell lysates were tested (Fig 5D). Proteins most tightly associated with membranes can be enriched by treatment of extracts with saponin [70]. LC3-I was retained in the membrane fraction in PV-infected cells in both WT and ΔATG5 cells (Fig 5D). Thus, PV is capable of recruiting LC3-I to membranes in the absence of lipidation and the canonical lipidation machinery. In agreement with the above data, we observed by EM that PV infection still leads to the formation of DMVs in ΔATG5 cells (Fig 5E and 5F) and that these structures are dependent on active viral replication since they are absent in cells treated with guanidine. However, no DMVs were seen upon PV infection of ΔLC3B cells, but an increase in single-membraned vesicles was observed. Therefore, PV is capable of forming DMVs without lipidated LC3 but still appears to require the LC3 protein itself. Interestingly, PV infection of ΔATG5 cells resulted in the formation of many large vesicles that resemble multivesicular bodies (MVBs) (Fig 5E and 5F). These might represent an altered curvature capacity of unlipidated LC3 alone [22], as compared to the increased curvature in PV-infected WT cells, which contain both LC3-I and LC3-II (Fig 5D).
To determine how PV leads to recruitment of unlipidated LC3 to membranes, we used an anti-GFP antibody to capture GFP–LC3 and tightly associated, detergent-resistant proteins during infection. Mass spectrometry (MS) was then used to identify the co-immunoprecipitated (IP) proteins. To confirm that our GFP–LC3 IP was specific, we compared several cellular proteins known to interact with both LC3-I and LC3-II or only with LC3-II (Fig 6A). The LC3 lipidation process starts with a C-terminal cleavage by the proteinase ATG4B and covalent attachment first to ATG7 and then to ATG3 [71–73]. The ATG5/12/16L complex then acts as an E3 ligase, transferring the covalent attachment of LC3 from ATG3 to PE [74]. LC3 binds to p62 whether it is lipidated or not [75,76], and, indeed, we observe equal binding to p62 in WT and ΔATG5 cells and with GFP–LC3 or GFP–LC3–G120. However, autophagy-associated protein FYCO1 only binds to LC3-II [77], as there were fewer peptide reads in ΔATG5 cells or when GFP–LC3–G120A was used. We saw little to no binding of ATG4B, which binds the terminal glycine of LC3 [72,73], when GFP–LC3–G120A was used. Interesting, ATG4B–LC3 binding was significantly reduced in ΔATG5 cells, perhaps because LC3-II is also one of its substrates [20,78]. ATG7 and ATG3, which participate in the covalent conjugation of LC3 upstream of the function of ATG5, bound to GFP–LC3 in the presence or absence of ATG5. However, the G120A substitution cannot initiate the cascade of covalent attachments and is not recognized by either ATG7 or ATG3 [79]. Thus, the recovery of peptides from all five of these LC3-interacting proteins exhibits differential dependence on the lipidation status of LC3 in accordance with their biological roles.
To identify viral proteins that directly bind to LC3 as well as those that are found on the same membranous surface, we prepared samples that did or did not contain detergent to disrupt membranes. Plasmids that expressed either GFP–LC3 or GFP–LC3–G120A were transfected into WT or ΔATG5 cells and infected with PV. The viral peptides identified by IP-MS were mapped across the viral genome to visualize the viral proteins pulled down with LC3 (Fig 6B). In the absence of detergent, peptides from almost every viral protein were recovered (S4A Fig). However, in the presence of detergent, most peptides pulled down from WT cells were derived from capsid proteins and 2C (Fig 6B). When we analyzed PV peptide reads as percentage of total reads, capsid proteins formed a large percentage of the total peptide reads in all conditions, suggesting that these viral proteins bind LC3 or LC3-associated membranes (Fig 6C). The ratio of reads in the presence of detergent to those in its absence showed that only peptides in 2C in WT cells remained associated with LC3 in the presence of detergent (Fig 6D). We hypothesize that LC3 binding to viral protein 2C, a known hydrophobic protein, is stabilized by LC3 lipidation.
To determine which form of protein 2C immunoprecipitated with GFP–LC3, we visualized the collected proteins by SDS PAGE and immunoblotted with anti-2C antibody (Fig 6E). Interestingly, precursor 2BC but not 2C was pulled down, arguing that the peptides identified by MS actually derived from 2BC. We additionally found 3A pulled down with LC3, an interesting observation because 2BC and 3A together can induce membrane rearrangements similar to those induced by viral infection [35], and lead to LC3-II accumulation outside of viral infection [80]. We did not find 2B or 3A peptides pulled down in our MS data, perhaps because these are small proteins and thus difficult to detect by MS.
A consensus LC3-interacting region (LIR) has been identified with a common WxxL motif, and contacts between this motif and LC3-binding protein p62 have been visualized by X-ray crystallography [81,82]. Only two potential LIR motifs were identified in the PV genome: WWKL in viral capsid VP2 and WQWL in viroporin 2B, marked in the genome schematic (Fig 6B). Their high conservation in picornaviruses is shown in S3B Fig. The existence of these sites is consistent with the specific recruitment of LC3 by viral proteins. The destruction of cellular protein p62 during PV infection was observed in the presence or absence of canonical autophagy, which normally accomplishes its degradation (S4C Fig), perhaps ensuring LC3 availability for viral protein binding during infection.
GFP–LC3 IP-MS was also performed on DENV-infected bovine hamster kidney (BHK) cells with WT GFP–LC3 and GFP–LC3–G120A. To confirm that the MS data were specific to LC3 lipidation conditions, we saw both GFP–LC3 and GFP–LC3–G120A bound to p62 but not to ATG3 or ATG7 when LC3–G120A was used (S5A Fig). All DENV proteins were pulled down with WT GFP–LC3 and GFP–LC3–G120A (S5B Fig). This is not surprising, given that DENV proteins are known to directly interact with each other [67,83,84], and suggests that one or more DENV proteins likely bind to LC3. Possible LIR motifs found in NS1, NS2A, and NS5 could potentially be sites of interaction with LC3.
By comparing how different RNA viruses manipulate autophagy factors, we show here that there are numerous ways by which a virus can hijack this complex cellular pathway. Each virus, even closely related viruses like DENV and ZIKV, uses slightly different components, although they might ultimately use the subsequent membranes and degradative processes for similar purposes. Each virus requires a different subset of initiation components for efficient viral growth, although all three viruses utilize lipid scavenger ATG9 and recruit LC3 directly to membranes, bypassing the need for ATG5-mediated lipidation (Fig 7). We believe that this comparative study addresses some of the inconsistencies seen with viral induction of the autophagy pathway.
The fact that the ULK1/FIP200 and BECN1/VPS34 complexes can be bypassed by individual viruses is consistent with the idea that noncanonical autophagy pathways can be induced in response to different stress stimuli [85]. ULK1-independent autophagy can be induced by glucose deprivation or excessive ammonia [23,86]. BECN1-independent autophagy can be induced by proapoptotic compounds such as Z18 [24,87]. There is also evidence for a noncanonical autophagy pathway that is VPS34-dependent and BECN1-independent upon induction by arsenic trioxide [88], providing precedent for the DENV-induced noncanonical pathway. These data highlight that functional cellular autophagy is not entirely reliant on every component of the canonical pathway. Perhaps many viruses take advantage of these alternate and possibly redundant pathways to ensure a productive infection [89].
One of the striking findings described here is the dependence of PV, DENV, and ZIKV infection on LC3 but not on its lipidation. It has been shown previously that unlipidated LC3-I is found on viral-associated vesicles during infection with murine hepatitis virus, equine arteritis virus, and Japanese encephalitis virus [90–92]. These data, in addition to LC3-dependent and ATG7-independent viral amplification, led these authors to conclude that the virally induced membranes were not derived from the autophagy pathway. Instead, colocalization of ER degradation enhanced by alpha-mannosidase (EDEM1) and LC3 suggested that the virally induced vesicles were ER-associated protein degradation (ERAD) associated [90]. We argue that direct recruitment of LC3 to membranes by viruses can perform autophagosome-like functions. Colocalization with EDEM1 could be a result of the known degradation of EDEM1-containing vesicles by an autophagy-like process [93]. We specifically found unlipidated LC3 associated with membranes during viral infection as well as direct binding of LC3 to PV proteins by IP-MS. A few cases of cellular LC3 lipidation–independent autophagy have been described [25,94]. In these cases, when either ATG5 or ATG7 is ablated, the autophagy pathway can still proceed, leading to the formation of autophagosomes in response to unusual stimuli, such as the drug etoposide [25]. LC3 is not lipidated under these conditions. It remains unknown how these autophagosomes fuse without lipidated LC3. LC3-I has been found associated with membranes during cellular secretory processes, although how LC3-I is recruited to membranes was not tested [93]. It is possible that in these lipidation-independent cases protein-mediated recruitment of LC3-I occurs, as we observed during viral infection.
Recently, the influenza M2 protein was found to contain a functional LIR motif. This LIR motif was shown to be important for viral-mediated relocalization of LC3 to the plasma membrane, presumably to facilitate viral budding [95]. This represents a striking example of autophagosome relocalization by a virus. We have identified potentially functional viral LIR motifs in the capsid protein VP2 and 2BC protein of PV, as well as in the polyproteins of DENV and ZIKV. One hypothesis is that PV proteins bind LC3 as a means to localize PV-replication complexes and capsids specifically to autophagosome-like membranes for assembly purposes. 2BC is localized to replication complexes and is a membrane-associated protein [96], so an interaction with LC3 at the outer surface of the double membrane is possible. The binding of capsid proteins to LC3 was unanticipated but of potential functional interest. Immature capsid precursors are known to be specifically associated with membranes [97] and intact virions can be released nonlytically by a process similar to secretory autophagy [39,40,98]. We propose that LC3 association of 2BC and immature capsids facilitate RNA replication, RNA packaging, and intercellular spread.
Autophagosome-dependent unconventional secretion was first demonstrated in yeast [99–101]. In human cells, autophagy-dependent secretion has been demonstrated for interleukin 1β [102] and increasing numbers of other proteins and complexes [26,103]. This suggests that autophagosomes can be rerouted to the plasma membrane and the contents released, providing precedent for virally induced double-membraned structures exiting a cell nonlytically. Infectious picornaviruses such as PV, Coxsackievirus B, and Hepatitis A virus have been found to be released in membrane-bound vesicles, each containing many viral particles [40,41,104]. For both unconventional secretion and the assembly of intracellular membranous components, the cellular autophagy pathway, with its unusual topologies and induction of membrane curvatures, is a rich source of host components for many viral infections.
We observed in mice that fasting could enhance viral growth of PV. We also show that the increase in PV titers is dependent upon the presence of Atg5, linking the starvation phenotype to its role in inducing the autophagy pathway. The data presented here show that PV benefits from the induction of the canonical autophagy pathway but do not require all of the components when virally inducing the pathway. We hypothesize that pre-existing membrane structures from starvation-induced autophagy can be utilized by viruses for growth and maturation, leading to quicker exit from cells and overall more virus production. Malnutrition has been linked to increased death from infectious diseases in children [105], and, although exacerbation of disease by malnutrition is often attributed to confounding effects, another outcome of malnutrition is induction of cellular autophagy. The implications of these data are broad: if viral infection is exacerbated by induction of the autophagy pathway, then starvation and many over-the-counter drugs that induce autophagy could lead to greater disease severity. It also implies that simple supplementation of nutrients to block induction of the autophagy pathway may help abrogate increased viral amplification.
Human HeLa cells and BHK21 cells were cultured in Dulbucco’s modified eagle’s medium (Gibco) supplemented with 10% bovine serum and 1% penicillin/streptomycin. PV type 1 Mahoney was grown from infectious cDNA, as previously described [106]. DENV type 2 (16681) was propagated from an infectious cDNA clone (pD2/IC) in C6/36 mosquito cells. The ZIKV strain used was PRVABC59 (Puerto Rico). The virus was purchased from BEI Resources and propagated on C6/36 mosquito cells.
CRISPR-Cas9 plasmid px458 (Addgene #48138) was cut with BbsI and ligated to annealed sgRNAs (see S1 Table), as described previously [107]. HeLa cells were transfected with resulting plasmids (Lipofectamine 3000, Invitrogen). Forty-eight hours later, cells were single-cell sorted for GFP+ cells using a BD Aria II sorter (Stanford FACS facility) in a 96-well format. Cells that propagated were tested for gene disruption by harvesting genomic DNA, PCR amplifying the region of interest, and sequencing to look for the presence of frameshift mutations. Clones were also tested for reduced protein expression by western blot. For the ATG9, BECN1, VPS34, and LC3B genes, multiple KO lines were obtained and tested, although subsequent experiments were carried out using one representative line.
Cells were transfected with 50 ng/well of GFP–LC3 in a 24-well plate. After 24 hours, cells were seeded on coverslips and infected or treated the following day. Cells were fixed in 4% paraformaldehyde, stained with DAPI (Invitrogen), and mounted on slides with PermaFluor (Thermo Fisher Scientific). Slides were imaged using a Leica SP8 confocal microscope (Stanford imaging facility). GFP–LC3 puncta were counted per cell.
Infected cells from a 24-well plate were fixed and stained using the PrimeFlow RNA assay according to the manufacturer’s protocol (Invitrogen). Viral RNA-specific probes were used for the positive strand of PV (VF1-10252), DENV2 (VF1-15158), or ZIKV (VF1-20236). Cells were run on a flow cytometer (Scanford, Stanford FACS facility) and analyzed using FlowJo software (v. 10.4).
Total RNA was isolated from cells using the RNeasy mini kit (Qiagen). RT-qPCR was performed on an Applied Biosystems 7300 machine using the QuantiTect SYBR Green RT-qPCR Kit (Qiagen). Primers specific to PV, DENV, or ZIKV were used (see S2 Table).
Cells were plated in a 96-well plate and transfected with 8 ng of PV replicon RNA (kind gift from Raul Andino). A subset of cells were treated with 2 mM guanidine (Sigma) prior to transfection to inhibit viral replication. Cell lysates were collected at 0, 2, 4, and 6 hours post transfection, and firefly luciferase activity was measured using the Firefly Luciferase Assay System (1000) in a 96-well format with a BioTek Neo2 luciferase reader.
Protein lysates were harvested with RSB buffer (10 mM NaCl, 10 mM Tris pH 7.5, 1.5 mM MgCl2, and 1% NP-40) with added EDTA-free protease inhibitor (Roche). Lysates were quantified using a DC protein assay (Bio-Rad), run on an SDS PAGE gel, and transferred to PVDV membranes. Immunoblots were blocked in 5% milk or BSA for phospho antibodies and incubated overnight with anti-LC3 antibody (Novus biologicals); PV 2C and 3A antibodies (kindly provided by Ellie Erhenfeld and Kurt Bienz); DENV NS3 antibody (GeneTex); p62 antibody (Sigma); FIP200 antibody (Abcam); ULK1, BECN1, VPS34, ATG9, and ATG5 antibodies (Cell Signaling); GAPDH antibody (Santa Cruz); phospho-AMPK alpha-1 (Thermo Fisher); and phospho-p70 S6K (Cell Signaling) at 1:1,000 dilution. Secondary antibodies conjugated to HRP (Invitrogen) were used at 1:10,000 dilution. Immunoblots were imaged on a ChemiDoc (Bio-Rad).
DNA transfections were done using a Lipofectamine 3000 kit (Invitrogen). The ULK1 plasmid was acquired from Addgene (#31963). The K46I mutation was made by site-directed mutagenesis (QuickChange Lightening kit, Agilent). Lentiviral vectors were cloned using Gibson cloning (NEB), inserting the gene of interest into the pLenti-CMV-Puro-DEST vector, as described previously [108]. 293T cells were transfected with the pLenti vector containing the addback gene, Gag-Pol, VSV-G, and p-Advent. Supernatants were collected 24 hours later and treated with protamine sulphate. Supernatants were added to KO cell lines and allowed to infect for 24–48 hours. Successful transductions were selected by 2 μg/ml puromycin for up to 1 week. Resulting cell lines were tested for rescue of protein levels by western blot. For siRNA knock-down of ULK1 and ULK2, 4 pooled siRNAs targeting human ULK1 or ULK2 (siGENOME SMARTpool, Dharmacon #005049 and #005396) were transfected into HeLa cells using Lipofectamine 3000 (Life Technologies). Knockdown was tested by qPCR at 72 hours post transfection.
WT and KO cell lines were plated in 10-cm plates and infected with PV (6 hours) or DENV (24 hours) at an MOI of 10 PFU/cell. A subset of samples were infected and treated with 2 mM guanidine (Sigma) simultaneously. Cells were collected and fixed with 4% PFA for 10 minutes. Cells were washed twice with DMEM and resuspended in 50 ul of 20% BSA in DMEM. Samples were subjected to high-pressure freezing using a Leica EmPACT High Pressure Freezer and freeze substitution to replace water with acetone (Stanford Imaging Facility). Samples were sectioned and imaged on a transmission electron microscope (JEOL JEM1400, Stanford Imaging Facility). Per condition, 10–20 cells were analyzed for cellular structures on blinded images.
ULK1 inhibitor MBL56 (kind gift from Kevin Shokat) was used at 0.5, 5, and 50 μM. Rapamycin was used at 1 μM in DMSO (Sigma); Chloroquine (CQ) was used at 50 μM (Sigma); Spautin-1 was used at 10 μM (Sigma); SAR405 was used at 1, 5, and 10 μM (Sigma); and Hanks’ Balanced Salt Solution (HBSS; Gibco) plus 10% FBS was used for starvation media.
GFP immunoprecipitations were performed with GFP-Trap magnetic beads (Chromotek). Cells were harvested with lysis buffer (10 mM Tris pH 7.5, 150 mM NaCl, 0.5 mM EDTA, and 0.5% NP-40) containing EDTA-free protease inhibitor (Roche). Magnetic beads were added and allowed to bind for 2 hours at 4°C with rotation. Beads were washed 3 times with wash buffer (10mM Tris pH 7.5, 150 mM NaCl, and 0.5 mM EDTA) and eluted with 2 M glycine pH 2.5, followed by neutralization with 1 M Tris pH 10.4, for mass spectrometry. Samples analyzed by immunoblot were eluted by boiling in 2X SDS sample buffer (120 mM Tris pH 6.8, 20% glycerol, 4% SDS, 0.04% bromophenol blue, and 10% BME) and run on an SDS PAGE gel. Mass spectrometry was performed and analyzed by the Stanford University Mass Spectrometry facility (SUMS).
Embryos of conditional Atg5 flox/+ mice B6.129S (Hara et al., 2006) were obtained from Dr. Noboru Mizushima through the RIKEN BioResource Center (Ibaraki, Japan; stock RBRC02975) and cryorecovered at Stanford University. Atg5 flox/+ mice were further backcrossed to C57BL/6J to N10. B6.Cg-Ndor1Tg(UBC-cre/ERT2)1Ejb/J, a lentitransgenic line harboring a single-copy integrant of a fusion gene consisting of Cre recombinase and a mutant human estrogen receptor (ERT2) under control of the human ubiquitin C promotor (UBC), was obtained from the Jackson Laboratory (Bar Harbor, Maine; stock 008085) at N6 and further backcrossed to C57BL/6J to N10.
C57BL/6 mice expressing the human PVR, Atg5fl/fl, and Cre+/- were treated with 75 mg/kg of tamoxifen for 5 days, followed by 2–4 weeks of recovery to induce the expression of Cre and subsequent excision of the floxed Atg5 exon 2. Ten to twelve-week-old mice were fasted for 48 hours prior to infection or given 25 mg/kg loperamide (Sigma) in saline by i.p. injection every 12 hours for the duration of the infection. Mice were infected intramuscularly with PV (1 x 107 PFU/50 ul in PBS). Muscle tissue was harvested 4 dpi and processed using the Bullet Blender BBX24 (Next Advance). DNA and protein lysate were also extracted from tissues, using the DNeasy Blood and Tissue kit (Qiagen) and RSB buffer, respectively.
The mouse experiments in this study were approved by the Administrative Panel on Laboratory Animal Care at Stanford University (approval number 9296). This committee is accredited by the Association for the Assessment and Accreditation of Laboratory Animal Care (AAALAC International).
|
10.1371/journal.ppat.1000240 | MyD88 Is Required for Protection from Lethal Infection with a Mouse-Adapted SARS-CoV | A novel human coronavirus, SARS-CoV, emerged suddenly in 2003, causing approximately 8000 human cases and more than 700 deaths worldwide. Since most animal models fail to faithfully recapitulate the clinical course of SARS-CoV in humans, the virus and host factors that mediate disease pathogenesis remain unclear. Recently, our laboratory and others developed a recombinant mouse-adapted SARS-CoV (rMA15) that was lethal in BALB/c mice. In contrast, intranasal infection of young 10-week-old C57BL/6 mice with rMA15 results in a nonlethal infection characterized by high titer replication within the lungs, lung inflammation, destruction of lung tissue, and loss of body weight, thus providing a useful model to identify host mediators of protection. Here, we report that mice deficient in MyD88 (MyD88−/−), an adapter protein that mediates Toll-like receptor (TLR), IL-1R, and IL-18R signaling, are far more susceptible to rMA15 infection. The genetic absence of MyD88 resulted in enhanced pulmonary pathology and greater than 90% mortality by day 6 post-infection. MyD88−/− mice had significantly higher viral loads in lung tissue throughout the course of infection. Despite increased viral loads, the expression of multiple proinflammatory cytokines and chemokines within lung tissue and recruitment of inflammatory monocytes/macrophages to the lung was severely impaired in MyD88−/− mice compared to wild-type mice. Furthermore, mice deficient in chemokine receptors that contribute to monocyte recruitment to the lung were more susceptible to rMA15-induced disease and exhibited severe lung pathology similar to that seen in MyD88−/−mice. These data suggest that MyD88-mediated innate immune signaling and inflammatory cell recruitment to the lung are required for protection from lethal rMA15 infection.
| In 2002, a new human coronavirus (CoV), termed SARS-CoV, emerged in southern China from coronaviruses circulating within live animals sold for food. Due to the ease and speed of human global travel, this new respiratory virus rapidly spread worldwide, illustrating the need to better understand how these viruses cause disease and how the immune system responds to infection. SARS-CoV infection of the human lower respiratory tract caused an atypical pneumonia characterized by viral replication in lung tissue and lung inflammation visible by chest X-ray. To identify how the immune system responds to and provides protection from SARS-CoV infection, we have developed a mouse model that mimics many aspects of SARS-CoV disease in humans. Utilizing this mouse model, we discovered that a host gene, termed MyD88, is required to control SARS-CoV replication and spread in lung tissue and for protection from death. In addition, MyD88-dependent functions were required for early immune and inflammatory responses in the lung following SARS-CoV infection, and the absence of these early responses correlated with severe SARS-CoV-induced disease and death. Our studies identify host immune responses that provide protection from SARS-CoV infection and provide valuable insight toward the development of successful antiviral therapies.
| In 2003, a novel coronavirus, SARS-CoV, emerged from zoonotic pools of virus in China to cause a global outbreak of Severe and Acute Respiratory Syndrome (SARS) affecting 29 countries, causing over 8000 human cases and greater than 700 deaths [1]–[3]. The clinical course of SARS-CoV disease in humans is characterized by fever, non-productive cough, and malaise culminating in lung infiltrates visible by X-ray and an atypical pneumonia [4]–[8]. Immunologically, SARS-CoV infection of humans generates a cytokine/chemokine storm where elevated levels of IP-10, MIP1-α, and MCP-1 are detected within the blood [9]. Histological examination of lung tissue in terminal SARS-CoV cases revealed SARS antigen primarily within bronchiolar epithelium, Type I and II alveolar pneumocytes, and less frequently within macrophages and lymphocytes in the lung, suggesting a roll for multiple cell types in SARS-CoV pathogenesis [10],[11].
Though clinical and epidemiological data from the epidemic and reemergence has provided insight into the molecular pathogenesis of SARS-CoV infection, thorough studies of virus and host interactions have been hampered by the lack of animal models that fully recapitulate human disease. C57BL/6 mice infected with the epidemic strain, SARS Urbani, do not show any overt signs of disease but there is virus replication in the lung (107TCID50/g 3dpi), induction of a number of proinflammatory chemokines, and viral clearance even in the absence of T, B, and NK cells, suggesting that innate immunity alone is required for the clearance of SARS Urbani within this acute model of SARS-CoV replication [12]. The newly developed mouse adapted SARS-CoV, MA15, differs from Urbani in 6 amino acids and infection of young or senescent BALB/c mice with either MA15 or recombinant MA15 (rMA15) results in high virus titers in the lung, pulmonary pathology, and 100% mortality resembling the pathogenesis of the most severe human cases of SARS-CoV [5],[10],[13]. Unfortunately, a SARS-CoV mouse model does not yet exist that recapitulates the less severe pathogenesis and recovery seen in a majority of the human cases. Moreover, a model of SARS-CoV pathogenesis with both disease and convalescence would allow for the elucidation of pathways involved in the innate or adaptive protective response to infection.
Toll-like receptors (TLRs) are cellular receptors that recognize molecular signatures of pathogens and initiate an inflammatory signaling cascade that is critical to the innate immune response [14]. Myeloid differentiation primary response gene 88 (MyD88) is a key adaptor protein for most TLR-dependent inflammatory signaling pathways as well as IL-1R1, IL-18R1 and IFNγR1 signaling pathways [14] MyD88 interacts with a variety of cellular proteins leading to the activation of NF-κB, JNK, and p38 and the induction of inflammatory cytokines, chemokines, and type I interferons [14]. The role of MyD88 in the host response to viral infection has been investigated for a number of viral pathogens. These studies have indicated that MyD88 is crucial for the response to some viral infections, while it appears dispensable for others. For example, MyD88 signaling is not required for clearance of reovirus infection after peroral inoculation of mice [15]. In contrast, MyD88−/− mice infected with respiratory syncytial virus (RSV), vesicular stomatitis virus (VSV), or lymphocytic choriomeningitis virus (LCMV) results in more severe disease [16]–[19]. Though TLR7/MyD88/IFNα dependent signaling has been implicated as important in the pathogenesis of a related coronavirus, mouse hepatitis virus (MHV), the role of MyD88 signaling in SARS-CoV pathogenesis has not yet been investigated [20].
In this study, we describe a novel C57BL/6 mouse model of rMA15 acute pathogenesis characterized by high titer virus replication within the lung, induction of inflammatory cytokines and chemokines, and immune cell infiltration within the lung. WT mice display signs of disease that include 12–15% loss of body weight by 3 dpi and lung pathology, however, these mice recover from infection by 6 dpi. Furthermore, we demonstrate a protective role for MyD88-dependent regulation of innate and inflammatory immune responses in this model of rMA15 pathogenesis. MyD88−/− mice infected with rMA15 have significantly higher and prolonged virus titers in the lung, exhibit a severe delay in host immune and inflammatory responses, including monocyte/macrophage recruitment to the lung, and ultimately succumb to infection. In addition, mice deficient in chemokine receptors that regulate the recruitment of inflammatory leukocytes to the lung were also more susceptible to rMA15 infection. These data suggest that a failure or delay in MyD88 inflammatory signaling and a concurrent delay in inflammatory monocyte/macrophage recruitment to the lung during acute infection results in exacerbated SARS-CoV disease. This novel mouse model of acute SARS-CoV pathogenesis could be extended to the investigation of many other components of the innate immune response in order to form a more comprehensive view of SARS-CoV pathogenesis, which may guide the rational design of antiviral therapies.
Since previous studies suggested MyD88-dependent inflammatory signaling was important for protection from severe disease caused by VSV, LCMV, and RSV, we evaluated the importance of MyD88 signaling in SARS-CoV pathogenesis [16],[18],[19],[21]. To assess the contribution of innate and adaptive immune responses in SARS-CoV-induced disease, age matched C57BL/6 (WT) (n = 14), congenic RAG-1−/− (n = 21), and congenic MyD88−/− (n = 16) mice were infected intranasally with 105 pfu of recombinant mouse-adapted SARS-CoV (rMA15) and monitored for virus-induced morbidity and mortality. Infection of WT, RAG-1−/−, or MyD88−/− mice resulted in weight loss beginning at day 2 post-infection (Fig. 1A). Infected WT and RAG-1−/− mice lost 14±4% and 9±4% of starting body weight by 3 dpi, respectively, and returned to starting body weights by 5–6 dpi, indicating that the mice had recovered from rMA15 induced disease. In contrast to WT and RAG-1−/− mice, which began to recover weight after 3 dpi, infected MyD88−/− mice continued to lose weight after 3 dpi creating a significant weight disparity between WT and MyD88−/− mice (WT vs. MyD88 percent weight p = <0.05 at 4, 5, and 6 dpi). While 100% of WT and 86% of RAG-1−/− mice survived the infection, >90% of MyD88−/− mice (n = 16) succumbed to infection by 6 dpi (Fig. 1B). MyD88 plays a critical role in proinflammatory signaling following stimulation of all known TLR's, except TLR3, as well as IL1R and IL18R. Interestingly, 100% of IL-1R1- or IL-18R1-deficient mice survived infection by rMA15 (data not shown) indicating that the genetic absence of either receptor did not recapitulate the lethal phenotype observed following infection of MyD88−/− mice. These findings indicate that i) adaptive immunity does not play a major role in protection from lethal SARS-CoV infection in mice, ii) the genetic absence of MyD88 significantly enhances SARS-CoV-induced morbidity and mortality , and iii) MyD88-dependent signaling through a receptor(s) other than IL-1R1 or IL-18R1 is responsible for protection from rMA15.
To determine if lethal infection of MyD88-deficient mice was due to enhanced and/or prolonged virus replication, a kinetic analysis of rMA15 viral loads within the lungs of WT, RAG-1−/−, and MyD88−/− mice was performed. Viral loads in lung tissue of rMA15-infected MyD88−/− mice were significantly higher than WT mice at 2, 3, 4, and 6 dpi and RAG-1−/− mice at both 2 and 4 dpi (Fig. 1C). Interestingly, despite the absence of significant signs of disease at late time points, viral lung titers remained elevated in lung tissue of rMA15-infected RAG-1−/− mice at 7 and 9 dpi (Fig. 1C). In fact, no new or relapsing signs of disease were observed in rMA15-infected RAG-1−/− mice as late as 5 weeks post-infection (data not shown). In addition to lung tissue, viral titers were also determined for the brain, liver, kidney, and spleen. Infectious virus was not detected in the brain, liver, or kidney of WT or MyD88−/− mice at 1, 3, or 4 dpi (limit of detection = 250 pfu/gram of tissue; n = 4–5 mice per time point). Sporadic viral titers were detected in the spleens of WT and MyD88−/− mice (data not shown). These findings indicate that the greater mortality observed in rMA15-infected MyD88−/− mice was not due to enhanced replication at extrapulmonary sites.
To further assess the role of MyD88 in controlling SARS-CoV replication within the lung, in situ hybridization was performed on tissue sections using an 35S-labeled riboprobe complementary for the N gene of SARS-CoV. As shown in Fig. 2, in situ signal was not observed in lung sections derived from mice that received intranasal administration of PBS alone (top panels). At both 1 and 2 dpi, intense rMA15-specific in situ signal was observed throughout the lung tissue, including lung airway epithelia, in rMA15-infected WT and MyD88−/− mice. However, by 3 to 4 dpi, the distribution and intensity of rMA15-specific in situ signal had greatly diminished in lung tissue of WT mice, while the rMA15-specific signal in MyD88−/− mice was more intense and much more broadly distributed (Fig. 2). By 6 dpi, though diminished, rMA15-specific in situ signal was still readily detectable in lung tissue of MyD88−/− mice, whereas only very rare rMA15-specific in situ signal could be detected in lung tissue of WT mice. In sum, these findings suggest that MyD88 is required for control of MA15 replication in pulmonary tissue at early times post-infection and that the inability to control or clear this early replication is associated with increased lethality.
Infection of WT mice with rMA15 results in a rapid inflammatory response in the lungs. This virus-induced inflammatory response likely has both protective and pathologic consequences. To investigate the importance of MyD88 in SARS-CoV-induced lung inflammation, we employed quantitative RT-PCR (qRT-PCR) to assess mRNA levels of antiviral and proinflammatory cytokine/chemokine gene expression in the infected mouse lung at various times post-infection. MyD88 is required for the induction of type I IFN in mouse cells following stimulation of TLR7 and TLR9 [14]. However, similar to previous reports, we were unable to detect significant induction of type I IFN in lung tissue of either WT or MyD88−/− mice following SARS-CoV infection by qRT-PCR (Fig. 3A) or in serum using a type I IFN bioassay (data not shown) compared to mock-infected control mice [22]. Type III IFNs, which are induced by viral infection and TLR ligands, also have direct antiviral affects [23]. Recently, lung tissue and epithelial cells were found to be responsive to type III interferon in vivo, suggesting that type III IFN may function to prevent viral infection at mucosal surfaces [24],[25]. Interestingly, SARS-CoV infection of either WT or MyD88−/− mice resulted in similar induction of type III IFN over mock-infected mice which peaked at 2 dpi and declined thereafter (Fig 3A). In fact, induction of type III IFN was slightly higher, although not statistically significant, in infected MyD88−/− mice at all time points analyzed. Taken together, these findings suggest that the enhanced susceptibility of MyD88−/− mice is not due to a failure to induce a protective type I IFN response, which was undetectable in both strains of mice, or type III IFN response, which was similar in both strains of mice.
Infection of WT mice with rMA15 resulted in a significant induction of proinflammatory chemokines including CCL2 (MCP-1), CCL3 (MIP-1α), and CCL5 (RANTES) as compared to mock-infected control mice (Fig. 3B). In contrast to type I and type III IFNs, induction of CCL2 was dramatically reduced in rMA15-infected MyD88−/− mice compared to WT mice at 2 dpi (14 fold difference, P<0.005). Statistically significant differences in the abundance of CCL2 mRNA were not detected at 4 or 6 dpi (Fig. 3B). Similarly, the induction of CCL3 (22 fold difference at 2 dpi, P<0.005; 5 fold difference at 4 dpi, p<0.005) and CCL5 (8 fold difference at 2 dpi, P<0.005; 8 fold difference at 4 dpi, P<0.01) were dramatically reduced in infected MyD88-/- compared to infected WT mice (Fig. 3B). In addition to proinflammatory chemokines, virus-induced expression of several proinflammatory cytokines, including TNF-α (9 fold difference at 2 dpi, P<0.005; 3 fold difference at 4 dpi, p<0.005), IL-1β (3.8 fold difference at 2 dpi, P<0.01), and IL-6 (11 fold difference at 2 dpi, P<0.005); , was severely impaired in rMA15-infected MyD88−/− mice compared to infected WT mice (Fig. 3C). These data indicate that MyD88 is required for the early induction of proinflammatory chemokines and cytokines within pulmonary tissues of SARS-CoV-infected mice and suggest that some aspect(s) of this inflammatory response is required for protection from lethal disease.
The levels of inflammatory chemokine and cytokine transcription suggested that the innate immune response was severely delayed in MyD88-deficient mice as compared to WT mice. To assess lung damage and pulmonary inflammation throughout the course of virus infection in WT and MyD88−/− mice, we evaluated hematoxylin and eosin stained lung tissue sections from 2, 4 and 6 dpi (Fig. 4). At 2 dpi, MyD88−/− mice exhibited a denuding bronchiolitis characterized by an extrusion of airway epithelial cells into the lumen of the airway and epithelial/endothelial atypia (vacuolization and disruption of normal epithelium and endothelium) but did not exhibit any obvious signs of inflammatory cell infiltration including peribronchivascular (PBV) or peri-venular immune cell infiltration (“cuffing”). In contrast, WT mice at 2 dpi exhibited pronounced lung inflammation characterized by perivascular cuffing, endothelial and epithelial atypia, and peribronchivascular immune cell infiltration, without the severe denuding bronchiolitis seen in MyD88−/− mice (Fig. 4). At 4 dpi, MyD88−/− mice continued to exhibit a denuding bronchiolitis, epithelial/endothelial atypia, and the added phenotype of PBV edema without immune cell infiltration around the airway though perivascular infiltration of immune cells was observed (Fig. 4). Similar to what was seen at early times post infection, WT mice at 4 dpi had an exacerbation of the inflammatory infiltrate seen at 2 dpi, but denuding bronchiolitis and PBV edema were not observed. Interestingly, by 6 dpi, the severity of PBV edema and denuding bronchiolitis in MyD88−/− mice had waned, and signs of lung inflammation were evident, with marked PBV infiltrates and perivascular cuffing even more severe than that seen in WT mice at similar times post infection. These findings suggest that MyD88 is essential for the early induction of the host inflammatory response and the timely recruitment of inflammatory leukocytes in the SARS-infected lung.
The expression analyses of proinflammatory chemokines/cytokines and the lung pathology suggested that MyD88 is critical for early immune/inflammatory responses in lung tissue following SARS-CoV infection. To investigate whether the impaired chemokine and cytokine responses in MyD88−/− mice impacted the cellular composition within the lung, total leukocytes were isolated from enzymatically digested pulmonary tissue and the cell surface phenotypes of the isolated cells were determined by flow cytometry.
At 2 dpi, no significant differences were detected in the number of natural killer cells (NK1.1+/CD3−) or T lymphocytes (CD4+/CD3+/NK1.1− or CD8+/CD3+/NK1.1−) isolated from the lung tissue of mock-infected or SARS-CoV infected WT mice (data not shown). These findings suggested that the differences in inflammation in rMA15-infected WT and MyD88−/− mice observed in the histological analyses of lung tissue were likely due to differences in myeloid cell populations. Therefore, anti-CD11b, anti-CD11c, anti-Gr-1, and anti-F4/80 antibodies were used to define the following cell types by cell surface antigen staining: alveolar macrophages (CD11c+/F4/80+/CD11blow/−), dendritic cells (CD11c+/CD11b− or CD11b+/F4/80−/Gr-1−), inflammatory monocytes/macrophages (CD11b+/F4/80+/Gr-1int/CD11c−), and neutrophils (Gr-1high/CD11b+/F4/80−/CD11c−) [26]–[28]. As shown in Fig. 5A, cell surface staining of lung leukocytes with anti-Gr-1, anti-F4/80, and anti-CD11b antibodies revealed two distinct cell populations, defined by region 3 (R3) and R4/R5 in the histograms, that were significantly increased in both percentages (Fig. 5B) and total numbers (Fig. 5C) in rMA15-infected WT mice as compared to mock-infected mice. The cells defined by R3 in our analyses have a Gr-1high/F4/80−/CD11b+ cell surface phenotype (Fig. 5A and data not shown), which is consistent with that of neutrophils, and were modestly increased in the lung tissue of rMA15-infected WT mice at 2 dpi (Fig. 5B and 5C). The most dramatic differences detected in percentages and total numbers were cells with a Gr-1int/F4/80+/CD11b+ cell surface phenotype defined in R4 and R5 (Fig 5A, B, and C). Additional analyses demonstrated that these cells were Ly6C+ and CD11c− (data not shown). This cell surface staining pattern, including the Gr-1int and F4/80low staining, has been well characterized by a number of studies as that of inflammatory monocytes [26]–[35]. Strikingly, both the Gr-1high/F4/80− population (R3) and the Gr-1int/F4/80+/CD11b+ inflammatory monocyte/macrophage population (R4 and R5) were dramatically reduced in lung tissue of rMA15-infected MyD88−/− mice compared to infected WT mice (Fig. 5A, B, and C). In fact, at 2 dpi, similar numbers of inflammatory monocytes were detected in infected MyD88−/− and mock-infected control mice (Fig. 5C). To determine if the failure to recruit inflammatory monocytes/macrophages was sustained at later times post infection in MyD88−/− mice, we performed similar cell isolation experiments at 4 dpi. In contrast to 2 dpi and consistent with our histological analysis of lung tissue, at 4 dpi, a similar percentage and total number of Gr-1int/F4/80+/CD11b+/CD11c− inflammatory monocytes/macrophages were isolated from the lung tissue of WT and MyD88−/− mice (Fig. 5D). Similar to 2 dpi, we did not detect significant numbers of CD3+ T lymphocytes within the lung tissue of rMA15-infected mice compared to PBS-inoculated control mice on 4 dpi, indicating that T lymphocytes were not a major component of the inflammatory response at these times post-infection (data not shown). These results further indicate that MyD88 is critical for early host immune and inflammatory responses, which include the initial recruitment of inflammatory monocytes/macrophages to pulmonary sites, in response to rMA15 infection.
The histological and flow cytometric analyses outlined above suggested that monocytes/macrophages are i) the major cell population increased in the SARS-infected lung at early times post-infection, ii) increased in the lung by a MyD88-dependent mechanism, and iii) critical for protection against severe rMA15-induced disease. In addition, in response to rMA15 infection, MyD88−/− mice failed to upregulate expression of a number of proinflammatory chemokines that promote monocyte recruitment. Therefore, we hypothesized that mice deficient in monocytes or specific chemokine receptors important for recruitment of inflammatory monocytes may be more susceptible to rMA15-induced disease. We attempted to deplete circulating monocytes by IP injection of clodronate liposomes or alveolar monocytes/macrophages by IN administration of clodronate liposomes in C57BL/6 mice prior to rMA15 infection. Though we were able to deplete alveolar macrophages by IN administration of clodronate liposomes, both IP and IN administration of clodronate liposomes failed to alter morbidity and mortality and failed to prevent the recruitment of inflammatory monocytes to the infected lung at 2 dpi (data not shown). Additionally, the intranasal administration of clodronate liposomes 2 days post rMA15 infection of C57BL/6 mice failed to induce more severe disease or mortality. Due to inconclusive results from our clodronate depletion studies and to continue to explore the importance of monocyte recruitment in SARS-CoV disease, we infected mice deficient in chemokine receptors known to be important for monocyte recruitment. As shown in Fig. 6A, mice deficient in CCR1, CCR2, or CCR5 developed more severe and prolonged disease as compared to WT mice. Between 3 and 14 dpi, CCR1 and CCR2 percent weight differed significantly from WT mice, while CCR5 weight differed significantly from WT between 2 and 13 dpi (Fig. 6A). Unlike CCR2 and CCR5 deficient mice, 40% of infected CCR1 deficient mice succumbed to infection by 7 dpi (Fig. 6B). To assess the lung damage and degree of pulmonary inflammation in chemokine receptor deficient mice, we evaluated hematoxylin and eosin stained lung sections from 2 dpi (Fig. 6C). Signs of inflammation and virus induced lung pathology are evident in wild-type mice on 2dpi with PBV cuffing caused by infiltrating immune cells, apoptosis of the airway epithelium, and a mild denuding bronchiolitis. In contrast, mice deficient in either CCR1, CCR2 or CCR5 exhibited more prominent airway epithelial cell apoptosis, a severe denuding bronchiolitis with an accumulation of cohesive apoptotic debris within the airway, and perivenular/periarterial cuffing but there was a distinct lack of cuffing around the affected airways. In direct correlation with the increased morbidity and mortality of rMA15 infected CCR1 deficient mice, the lung pathological conditions described above were the most severe in CCR1 deficient mice.
Since human clinical SARS data is complicated by host genetic variation, disease exacerbating comorbidities, age variation, and variable drug treatment regimens, animal models provide a more homogenous and controlled environment within which to ask questions related to the mechanisms of disease pathogenesis. Prior to the generation of a mouse adapted SARS-CoV (MA15) which causes 100% mortality in BALB/c mice, previous SARS-CoV BALB/c or C57BL/6 animal models using the epidemic strain, SARS Urbani, were purely models of in vivo virus replication without overt signs of disease [12],[13],[36],[37]. In contrast to previous models, our novel C57BL/6 model of SARS-CoV pathogenesis recapitulates disease similar to non-severe human SARS-CoV cases with high virus titer replication in the lung, significant weight loss, elevated inflammatory cytokine/chemokines, the recruitment of inflammatory cells to the lung, viral clearance and subsequent convalescence [4]–[6],[9]. Furthermore, the recovery from rMA15 disease is dependent on MyD88 but does not seem to be dependent on the presence of functional T or B cells (Fig. 1A). In contrast to a previous report where RAG-1−/− mice were demonstrated to clear SARS Urbani (dose: 1×104 TCID50) with similar kinetics as compared to WT mice, we demonstrate that RAG-1−/− mice recover from disease signs with similar kinetics as WT mice but are unable to clear (dose: 105 pfu) the more robust rMA15 [12]. The discrepancy regarding clearance of virus in RAG-1−/− mice may be due to the differing doses used and/or the enhanced pathogenesis of the mouse adapted virus. The disease observed in our rMA15 C57BL/6 disease model has also been observed with a second independently derived mouse adapted SARS-CoV suggesting that the disease phenotype is not simply an artifact of the rMA15 mutational spectra but that both sets of mouse adapting mutations enhance the intrinsic pathogenic potential of the epidemic strain (data not shown). Taken together, the morbidity and mortality data for rMA15 infected WT, RAG-1−/−, and MyD88−/− mice suggest that early MyD88 dependent innate signals are required for protection from rMA15 induced mortality.
Serological and pathological data from the SARS-CoV epidemic suggests that the innate immune response plays a crucial role in the control of SARS-CoV infection but the molecular mechanisms of innate immune activation, protection from severe disease, and the contribution of the innate response to immune pathology remain unknown [5],[9],[10],[38]. MyD88 is a key signaling adaptor protein for most all TLRs, IL-1R1, IL-18R1, and IFNγ-R1 [14]. Contrary to previous virological studies [16]–[19],[21],[39], we have demonstrated MyD88 plays a crucial role in protection from SARS-CoV infection independent of Type I (α/β) and III (IL-28/29 or interferon lambda) interferon, and the adaptive immune response. Though MyD88 mediated proinflammatory signaling has been implicated in the protection from numerous bacteria and parasitic infections, few in vivo studies have implicated MyD88 in protection from viral diseases [16]–[19], [21], [40]–[44]. Intranasal infection of MyD88-deficient mice with RSV or VSV produces more severe disease that was correlated with a failure to recruit immune cells to the sites of infection [17],[18]. For RSV, MyD88-dependent induction of type I interferon correlated with the recruitment of eosinophils to the lung and efficient virus clearance [17]. In contrast, MyD88-dependent protection from lethal VSV infection occurred independent of type I interferon, correlated with the recruitment of monocytes to the site of infection and was dependent on IL-1R1 signaling [18]. In the C57BL/6 mouse model of SARS-CoV pathogenesis reported here, we demonstrate MyD88-mediated protection from SARS-CoV infection in the absence of detectable induction of type I interferon. Furthermore, infection of IFNα/β receptor deficient mice with rMA15 results in moderate weight loss and complete recovery with kinetics that is indistinguishable from those of WT mice (personal communication, Frieman and Baric, manuscript in preparation). Unlike RSV and VSV, we found that WT C57BL/6 mice are protected from lethal SARS-CoV infection by a MyD88-dependent mechanism that does not involve adaptive immunity, the induction of type I/III interferon, or IL1-R/IL-18R signaling (data not shown) suggesting that SARS-CoV is interfacing with the innate immune system in a potentially novel manner.
Human cases of SARS-CoV, mouse models, and in vitro data suggest inflammatory chemokines and cytokines and the recruitment of inflammatory cells are important in SARS-CoV pathogenesis [5],[10]. Our studies indicate that protection from SARS-CoV infection correlates with MyD88-dependent induction of IL1-β, TNF-α, IL-6, MCP-1, MIP-1α, and RANTES at early times post infection and many of these cytokines/chemokines were upregulated in human SARS-CoV cases [9],[45]. At early times post rMA15 infection, the MyD88-dependent chemokine/cytokine response occurs with the recruitment of inflammatory monocytes/macrophages to the lung at 2 dpi and is coincident with the control of virus replication in WT animals. Days 2, 3, 4 and 6 post infection, virus titers are significantly lower in WT mice as compared to MyD88−/− animals and these data are supported by the dramatic loss of in situ hybridization signal in WT mice by 3 dpi. Furthermore, the lung pathology and flow cytometry results suggest that the absence of inflammation in MyD88−/− mice at early times post infection (2 dpi) correlates with much more severe lung damage and by the time the host mounts an adequate inflammatory response (4 dpi), lung damage is too severe and mortality ensues. The importance of macrophages in SARS-CoV pathogenesis has been noted in the past where SARS antigen was frequently detected in macrophages in the pathological evaluation of post mortem lung tissues from human SARS-CoV cases [11]. Interestingly, in vitro data suggests that macrophages are not productively infected by SARS-CoV, however, these cells secrete inflammatory cytokines like IP-10 and MCP-1 in response to the virus [46]. We have yet to determine the cell type responsible for the induction of the MyD88-dependent protective inflammatory response though we have demonstrated the recruitment of inflammatory monocytes/macrophages occurs even if alveolar macrophages are depleted in WT mice (data not shown). In the future, bone marrow chimeras between WT and MyD88−/− mice may help deduce if myeloid derived cells are responsible for the initial induction of the protective inflammatory response.
Chemokine receptors play a crucial role in directing inflammatory leukocytes to the sites of infection in order to mount an effective immune response [28],[47]. CCR1, CCR2 and CCR5 each bind a unique repertoire of chemokine ligands but all are able to guide the trafficking of monocytes and other leukocytes to sites of inflammation [28],[47]. Previous reports have implicated that that chemokine receptors, CCR1, CCR2 and CCR5 can both promote protection (CCR1,CCR2) and progression (CCR5) of disease caused by a neurovirulent coronavirus and the chemokine receptor dependent alteration of disease correlated with the recruitment of inflammatory leukocytes to the sites of infection [48]–[50]. Our studies demonstrate the importance of chemokine receptors in protection from rMA15 disease where CCR1, CCR2 and CCR5 deficient mice experienced a significantly more severe disease and associated mortality as compared to WT mice. Furthermore, infected CCR deficient mice suffered from severe lung pathology (denuding bronchiolitis, epithelial apoptosis, etc.) and defects in inflammatory cell recruitment to the airway that were very similar to those seen in MyD88−/− mice. CCR1 (MCP-1), CCR2 (MIP-1α) and CCR5 (MIP-1α and RANTES) bind chemokines upregulated in the lungs of rMA15 infected WT mice whose expression are coincident with the recruitment of inflammatory leukocytes to the lung and protection from mortality. Recent data from Glass et al. suggest that CCR5 dependent recruitment of monocytes, T cells and NK cells to the brains of West Nile virus infected WT mice are required for the control of virus replication in the CNS and protection from mortality [51]. CCRs can also guide immunopathogenesis during virus infection where CCR2 deficient mice are protected from a lethal influenza virus infection due to the failure to recruit inflammatory monocytes to the infected lung [27]. Taken together, the above data suggests an important role for MyD88-dependent inflammation, the innate immune response, and chemokine recruitment of inflammatory cells in both the prevention and progression of severe SARS-CoV disease.
Viral pathogenesis is a complex process where interactions between the virus and the host determine the outcome of virus-induced disease. Many of the pathogenic mechanisms of SARS-CoV disease remain unknown and the existence of a robust mouse model of SARS-CoV pathogenesis will allow for the detailed analysis of virus-host interactions. We have developed a novel model of acute SARS-CoV pathogenesis. Using this model, we discovered a critical role for MyD88-dependent inflammation in the protection from SARS-CoV induced mortality suggesting that the innate immune response plays a key role in the early control of SARS-CoV in the lung. Our future studies are aimed at understanding the mode of MyD88 dependent innate immune activation and the molecular mechanisms of inflammatory monocyte clearance of SARS-CoV from the lung. In the future, our studies may guide epidemiological studies in human populations in order to deduce if MyD88 related inflammatory genes or CCRs contributed to protection or prevention of severe SARS-CoV disease. Importantly, our SARS-CoV disease model can be employed to study the contributions of various innate immune genes in the protection from severe SARS-CoV disease which eventually may help clarify the current view of SARS-CoV pathogenesis and guide the development of intelligently designed antiviral therapies.
Vero E6 cells were grown in MEM (Invitrogen, Carlsbad, CA) supplemented with 10% FCII (Hyclone, South Logan, UT) and gentamycin/kanamycin (UNC Tissue Culture Facility). Stocks of the recombinant mouse-adapted SARS-CoV (rMA15) were propagated and titered on Vero E6 cells and cryopreserved at −80°C until use as described [52]. All viral and animal experiments were performed in a Class II biological safety cabinet in a certified biosafety level 3 laboratory containing redundant exhaust fans while wearing personnel protective equipment including Tyvek suits, hoods, and HEPA-filtered powered air-purifying respirators (PAPRs) as described [52].
C57BL/6J (stock# 000664) , RAG-1−/− (stock# 002216), IL-1R1−/− (stock# 003245), and IL-18R−/− (stock# 004131) mice were obtained from The Jackson Laboratory (Bar Harbor, Maine) and bred in house. MyD88−/− mice were obtained from Shizou Akira (Osaka University) and backcrossed 11 generations to the C57BL/6 background. CCR1−/−, CCR2−/−, CCR5−/−, and control C57BL/6 were obtained from Taconic Laboratories. Animal housing and care were in accordance with all UNC-Chapel Hill Institutional Animal Care and Use Committee guidelines. 10 week old mice were anaesthetized with a mixture of ketamine/xylazine and intranasally infected with either DPBS alone or 105pfu/50 µl rMA15 or the recombinant epidemic strain, icSARS, in DPBS (Invitrogen, Carlsbad, CA). Mice were monitored at 24 h intervals for virus-induced morbidity and mortality.
To quantify the amount of infectious virus in tissues, lung, liver, kidney, spleen, and brain tissue were weighed, placed in 0.5 ml DPBS, homogenized, and titered via plaque assay on Vero E6 cells as previously described [53]. Whole blood was harvested via cardiac puncture and collected in BD microtainer tubes for serum separation. Serum was titered via plaque assay as described above.
Lung tissues were fixed in PBS/4% paraformaldehyde, pH 7.3, tissues were embedded in paraffin, and 5 µm sections were prepared by the UNC histopathology core facility. To determine the extent of inflammation, sections were stained with hematoxylin and eosin (H & E) and scored in a blinded manner.
35S-UTP-labeled riboprobes specific to the N gene of SARS-CoV (Urbani) or to the EBER2 gene from Epstein-Barr virus (negative control probe) were generated with an SP6-specific MAXIscript in vitro transcription kit (Ambion) and in situ hybridization was performed as described previously [37]. Briefly, deparaffinized tissue sections were hybridized with 5×104 cpm/µl of 35S-labeled riboprobes overnight. Tissues were washed, dehydrated through graded ethanol, coated in NTB autoradiography emulsion (Kodak), and incubated at −80°C for 7 days. Following development, sections were counterstained with hematoxylin and silver grain deposition was analyzed by light microscopy. rMA15-specific signal was determined by comparing silver grain deposition on parallel sections hybridized with the 35S-labeled riboprobe complementary for the EBER2 gene of Epstein-Barr virus.
Lungs from mock- or rMA15-infected mice were removed and homogenized directly in 1 ml of Trizol reagent (Invitrogen) and total RNA was isolated following the manufacturer's instructions. Complementary DNA was generated from 0.25–1 ug of total RNA using 250 ng random primers (Invitrogen) and superscript III reverse transcriptase (Invitrogen). Real-time PCR experiments were performed using Taqman© gene expression assays and an AB Prism 7300 (Applied Biosystems). 18S rRNA was used as an endogenous control to normalize for input amounts of cDNA. The relative fold induction of amplified mRNA were determined by using the Ct method.
Mice were inoculated as described above, sacrificed by exsanguination at 2 and 4 days post-infection, and lungs were perfused via cardiac puncture with 1× PBS. Lungs were dissected, minced, and incubated for 2 hours with vigorous shaking at 37°C in digestion buffer [RPMI, 10% FBS, 15 mM HEPES, 2.5 mg/ml collagenase A (Roche), 1.7 mg/ml DNase I (Sigma)]. Cells were passed through a 40 micron cell strainer, resuspended in RPMI media, layered on 5 ml lympholyte-M (Cedarlane), and centrifuged 30 minutes at 2500 rpm. Banded cells were collected, washed in wash buffer (1× HBSS, 15 mM HEPES), and total viable cells were determined by trypan blue exclusion. Isolated cells were incubated with anti-mouse FcγRII/III (2.4G2; BD Pharmingen) for 20 min. on ice and then stained in FACS staining buffer (1× HBSS, 1% FBS, 2% normal rabbit serum) with the following antibodies from eBioscience: anti-F4/80-FITC, anti-Gr-1-PE, anti-CD11b-APC, anti-CD11c-PE, anti-Ly-6C-FITC, anti-CD3-FITC, anti-CD8-APC, anti-CD4-PerCP, and anti-NK1.1-PE. Cells were fixed overnight in 2% paraformaldehyde and analyzed on a Cyan cytometer (Dako) using Summit software.
Percent starting weights, viral titers and inflammatory cell numbers were evaluated for statistically significant differences by the non-parametric Mann-Whitney test within GraphPad Prism or unpaired t-tests using GraphPad InStat3 software. P values of ≤0.05 were considered significant.
|
10.1371/journal.pgen.1004906 | Nicotiana Small RNA Sequences Support a Host Genome Origin of Cucumber Mosaic Virus Satellite RNA | Satellite RNAs (satRNAs) are small noncoding subviral RNA pathogens in plants that depend on helper viruses for replication and spread. Despite many decades of research, the origin of satRNAs remains unknown. In this study we show that a β-glucuronidase (GUS) transgene fused with a Cucumber mosaic virus (CMV) Y satellite RNA (Y-Sat) sequence (35S-GUS:Sat) was transcriptionally repressed in N. tabacum in comparison to a 35S-GUS transgene that did not contain the Y-Sat sequence. This repression was not due to DNA methylation at the 35S promoter, but was associated with specific DNA methylation at the Y-Sat sequence. Both northern blot hybridization and small RNA deep sequencing detected 24-nt siRNAs in wild-type Nicotiana plants with sequence homology to Y-Sat, suggesting that the N. tabacum genome contains Y-Sat-like sequences that give rise to 24-nt sRNAs capable of guiding RNA-directed DNA methylation (RdDM) to the Y-Sat sequence in the 35S-GUS:Sat transgene. Consistent with this, Southern blot hybridization detected multiple DNA bands in Nicotiana plants that had sequence homology to Y-Sat, suggesting that Y-Sat-like sequences exist in the Nicotiana genome as repetitive DNA, a DNA feature associated with 24-nt sRNAs. Our results point to a host genome origin for CMV satRNAs, and suggest novel approach of using small RNA sequences for finding the origin of other satRNAs.
| Satellite RNAs (satRNAs) are small RNA pathogens in plants that depend on associated viruses for replication and spread. While much is known about the replication and pathogenicity of satRNAs, their origin remains a mystery. We report evidence for a host genome origin of the Cucumber mosaic virus (CMV) satRNA. We show that only the CMV Y-satRNA (Y-Sat) sequence region of a fusion transgene was methylated in Nicotiana tabacum, indicating that the Y-Sat sequence is subject to 24-nt small RNA (sRNA)-directed DNA methylation. 24-nt sRNAs as well as multiple genomic DNA fragments, with sequence homology to Y-Sat, were detected in Nicotiana plants, suggesting that the Nicotiana genome contains Y-Sat-like repetitive DNA sequences, a genomic feature associated with 24-nt sRNAs. Our results suggest that CMV satRNAs have originated from repetitive DNA in the Nicotiana plant genome, and highlight the possibility that small RNA sequences can be used to identify the origin of other satRNAs.
| Satellite RNAs (satRNAs) are among the smallest RNA pathogens in plants and depend on associated viruses (helper viruses) for replication, encapsidation and movement inside the host plant [1], [2]. Their RNA genomes range from 220 to 1500 nucleotides (nt) in size and can form compact secondary structures by intra-molecular base-pairing that can be resistant to degradation by ribonucleases. SatRNAs are classified into three classes [3]. Class 1 satRNAs include large mRNA satellites that are 800 to 1500 nt in length and contain a single open reading frame that encodes at least one non-structural protein. SatRNAs belonging to class 2 are linear, less than 700 nt in size and possess no mRNA activity so do not encode any protein. SatRNAs of this class, including the Cucumber mosaic virus (CMV) satRNAs [4], occur most frequently. SatRNAs of class 3 are circular, around 350 to 400 nt in length and also do not exhibit mRNA activity. SatRNAs normally accumulate at high levels in infected host plants relative to their helper viruses, presumably because of the small size and ribonuclease-resistant structure of their RNA genome. A previous study shows that a CMV satRNA, unlike the CMV helper virus, is resistant to host RNA-dependent RNA polymerase-mediated antiviral silencing in Arabidopsis [5], which may also contribute to the high level accumulation of satRNAs. Whereas high-level replication and systemic infection of satRNAs depend on helper virus-encoded proteins, recent studies on CMV satRNAs indicate that satRNAs can be imported into the nucleus and transcribed there by host plant proteins independently of helper viruses [6], [7]. satRNAs are not required for the life cycle of their helper viruses, but participate in helper virus-host interactions by modulating the level of helper virus accumulation and the severity of helper virus-induced symptoms [8]. In addition, satRNAs can induce disease symptoms in the host plants that are distinct from helper virus-caused symptoms [4]. Recent studies indicate that such satRNA-induced symptoms are due to silencing of host genes directed by satRNA-derived small interfering RNAs (siRNA) [9], [10].
Like all plant viruses and subviral agents, the origin of satRNAs remains unclear. Two main origins of satRNA have been suggested: the genome of the helper virus or that of the host plant. However, unlike defective interfering RNAs, a group of subviral RNAs derived from truncated forms of the helper virus genome, satRNAs usually possess little or no sequence homology with their helper viruses [1], which argues against the helper virus genome as their origin. One exception is the virulent satRNA strain of Turnip crinkle virus, which contains a long (166-nt) segment that is homologous to the 3' end of the helper virus genome [11]. A number of studies have suggested satRNA emergence from the host genome. Sequence similarity has been observed between nucleotide stretches of the Arabidopsis genome and CMV satRNAs [1]. SatRNAs, such as CMV satRNAs that occur widely in Nicotiana species and some other Solanaceae species, are more commonly detected in experimental systems than in the wild or nature [1]. A number of studies have reported de novo emergence of satRNAs on serial passaging plants with the helper virus under controlled environmental conditions [12], implicating the host genome as the origin of satRNAs. Another report showed structural similarities of Peanut stunt virus (PSV) satRNAs with cellular introns of nucleus, mitochondria and plant viroids [13]. However, in spite of these suggestions, no intact plant genome sequence with similarities to satRNAs has been reported.
RNA silencing is an evolutionarily conserved gene regulation mechanism in eukaryotes mediated by 20-25-nt small RNAs (sRNAs) [14], [15]. These sRNA are processed from double-stranded (ds) or hairpin (hp) RNA by Dicer or Dicer-like (DCL) protein. To induce silencing, one strand of a sRNA is loaded into an Argonaute (AGO) protein to form the RNA-induced silencing complex (RISC), and guides the RISC to bind to complementary single-stranded RNA and cleave the RNA. Plants have three basic types of sRNA, 20-24-nt microRNA (miRNA), 21-22-nt siRNA, and 24-nt repeat-associated siRNA (rasiRNA) [16], [17]. MiRNAs induce posttranscriptional degradation or translational repression of mRNAs that encode regulatory proteins, such as transcription factors, and play a key role in plant development [17], [18]. The 21-22-nt siRNAs direct degradation of viral RNA and some endogenous mRNA, and are important in plant defence against viruses and in the control of some endogenous genes [17], [19]. The 24-nt rasiRNAs are unique to plants, and are involved in RNA-directed DNA methylation (RdDM) which is important for maintaining genome stability by silencing transposons and repetitive DNA sequences [17], [20], [21]. RdDM is highly sequence specific, and can be induced by both endogenous siRNAs and siRNAs derived from infecting viral agents including satRNAs [20], [22].
During the analysis of a transgene (35S-GUS:Sat) containing a β-glucuronidase (GUS) sequence transcriptionally fused at the 3′ end with the CMV Y-satellite RNA (Y-Sat) sequence in Nicotiana tabacum, we observed that the Y-Sat sequence was specifically methylated. This led us to hypothesize that 24-nt siRNAs homologous to Y-Sat may exist in N. tabacum inducing RdDM at the Y-Sat sequence of the transgene. Subsequent analyses revealed the existence of both 24-nt sRNAs and multiple DNA fragments in Nicotiana plants that showed sequence homology to Y-Sat, suggesting that CMV satRNAs originate from repetitive regions in the Nicotiana genome.
Three 35S promoter-driven GUS constructs were created, two of which had a 3′ fusion of a full-length 369-nt Y-Sat sequence [23] in either the sense (sSat) or antisense (asSat) orientation (Fig. 1A). These were transformed into tobacco and multiple independent transgenic lines obtained for each.
Plants transformed with the 35S-GUS:Sat fusion constructs showed reduced levels of GUS protein in comparison to those transformed with just the 35S-GUS construct lacking the Y-Sat sequence (Fig. 1B, 1C). This reduction in GUS activity occurred for both the sense and antisense orientations of the Y-Sat sequence (Fig. 1B). The low level of GUS activity was relatively uniform across the independent primary (T0) transformants (Fig. 1B) and persisted in the subsequent (T1 and T2) generations (Fig. 1C). The MUG assay results were confirmed by northern blot hybridization, which showed that the GUS:sSat and GUS:asSat transcripts accumulated at a much lower level than the GUS transcript in the respective transgenic plants (Fig. 1D).
The repressed expression of the 35S-GUS:Sat transgenes could either be due to transcriptional (TGS) or posttranscriptional (PTGS) gene silencing or to transcript instability caused by the fused Y-Sat sequence. Since both the 35S-GUS:sSat and 35S-GUS:asSat transgenes showed similar reduction in expression, the secondary structures of the fusion sequence were not likely to be responsible for the repression, as sense and antisense Y-Sat sequences are predicted to form different secondary structures. It also implied that RNA instability was not the main cause for the repressed GUS:Sat trangene expression. This was supported by results from an Agrobacterium infiltration (agro-infiltration) assay where both TGS and PTGS were negated, which showed that the large difference in expression levels between the 35S-GUS and 35S-GUS:sSat transgenes observed in stably transformed plants (∼6 fold; Fig. 1) was dramatically reduced in the agro-infiltrated tissues (∼1.0–2.4 fold; S1 Fig.).
We next investigated if PTGS or TGS was responsible for the repressed expression of the 35S-GUS:Sat transgenes. PTGS of a transgene is associated with 21-nt siRNAs corresponding to the transcribed region [17]. Northern blot hybridization failed to detect GUS- or Y-Sat-specific 21-nt siRNAs in three of the four 35S-GUS:Sat transgenic lines analysed (Fig. 2A), suggesting that TGS, but not PTGS, was the main cause of transgene repression. Consistent with this, nuclear run-on assay showed that the repressed 35S-GUS:sSat and 35S-GUS:asSat transgenes generated much reduced RNA signals in comparison to the highly expressed 35S-GUS transgene (Fig. 2B, C), indicating that they are transcriptionally repressed.
As DNA methylation at promoter sequences can cause TGS [24], we investigated if DNA methylation occurred in the 35S promoter of the 35S-GUS:sSat transgene using McrBC digestion-PCR. McrBC is a methylation-dependent restriction enzyme that recognizes DNA containing two or more methylated cytosine residues, separated by 30–2000 base pairs, and cleaves the DNA at multiple sites close to one of the methylated cytosines. Differences in PCR-amplified McrBC-digested and undigested DNA can provide a measurement of DNA methylation levels. McrBC digestion did not result in clear reduction in the amplification of a 380-bp amplicon, derived from the the 35S promoter near the transcription start site, which contains a total of 175 cytosines from both strands (Fig. 3). This indicated that the 35S promoter in the 35S-GUS:sSat transgene was not methylated, and that the reduced transgene expression was not due to promoter methylation. We extended the DNA methylation analysis to the GUS and Y-Sat sequence of the transgene. Similar to the 35S promoter sequence, four different regions of the GUS coding sequence showed no clear methylation, as indicated by strong amplification of McrBC-digested DNA (Fig. 3).
In contrast to the 35S and GUS sequences, McrBC digestion strongly reduced PCR amplification of the Y-Sat sequence in the 35S-GUS:sSat transgene (Fig. 3), indicating that it was highly methylated. Furthermore, the degree of this methylation, as judged by the extent of reduction in PCR amplification upon McrBC digestion (Fig. 3B and C, left), appeared to be inversely correlated with the level of GUS activity in the 35S-GUS:sSat plants (Fig. 3C, right). McrBC PCR also indicated Y-Sat-specific DNA methylation in the 35S-GUS:asSat transgene that contains an antisense Y-Sat sequence (S2 Fig.). Both the 35S and GUS sequences showed no difference in PCR amplification between McrBC-digested and undigested DNA, whereas the Y-Sat sequence showed a clear reduction in amplification upon McrBC digestion. Furthermore, the expression level of the 35S-GUS:asSat transgene also appeared to be inversely correlated with the extent of Y-Sat sequence methylation (S2 Fig.).
The McrBC PCR results were validated using bisulfite sequencing, which determines DNA methylation at a single cytosine nucleotide level due to the ability of bisulfite to convert unmethylated, but not methylated, cytosines to uracils [25]. Three regions of the 35S-GUS:sSat transgene were amplified from bisulfite-converted DNA, including the 35S promoter, the 35S-GUS junction, and the Y-Sat sequence (S3A Fig.). We sequenced the bisulfite PCR product as a mixed DNA population, and determined the DNA methylation level based on the ratio between the peak heights of cytosine (C) and thymine (T) residues in the sequencing trace files, which has proven to be an effective way for measuring overall DNA methylation levels in a specific plant sample [26]. Consistent with the McrBC PCR result, the 35S and GUS sequences showed no significant methylation of cytosine residues as indicated by the lack of cytosine (blue) peaks at the cytosine positions in the trace files, which were instead replaced by thymine (red) peaks (S3B-C Fig.). In contrast, the Y-Sat sequence showed strong methylation in all four DNA samples analyzed, especially at the CG and CHG sites (H = A, C or T nucleotides) (Fig. 4). Two pairs of 35S-GUS:sSat transgenic lines were analyzed by bisulfite sequencing, and in each pair the plants that showed lower GUS expression level displayed a higher degree of cytosine methylation, particularly at the CHG and CHH sites (Fig. 4). Taken together, the DNA methylation analyses indicated that the Y-Sat sequence of the 35S-GUS:Sat transgenes was specifically targeted for methylation in transgenic N. tabacum plants, and that this methylation appeared to correlate with the repression of the transgenes.
The sequence-specific DNA methylation detected in the Y-Sat sequence of the 35S-GUS:Sat transgenes raised the possibility that the Y-Sat sequence might be subject to RNA-directed DNA methylation (RdDM). RdDM can occur at all cytosine contexts (CG, CHG and CHH), and is directed by the 24-nt size class of siRNAs [17], [20], [21]. sRNAs of 24 nt, but not of 21–22 nt, were readily detectable in both transgenic and wild-type Nicotiana plants by northern blot hybridization using the Y-Sat sequence as a probe, especially in the flowers (Fig. 5A), a tissue known to contain relatively high abundance of 24-nt siRNAs [27], [28]. Importantly, the 24-nt sRNA signals were not affected by the presence of the 35S-GUS:Sat fusion transgenes (Fig. 5A and 2A), indicating that they are generated by the host plant genome and not by the transgene. Northern blot hybridization also showed that these 24-nt Y-Sat-like sRNAs are present in all three Nicotiana species analysed (Fig. 5B). These results indicated that Nicotiana species generate 24-nt sRNAs with sequence homology to the Y-Sat, and suggested that the DNA methylation of the Y-Sat sequence in the fusion transgenes was likely induced by these 24-nt sRNAs.
Detection of 24-nt Y-Sat-like sRNAs in wild-type Nicotiana plants using northern blot hybridization prompted us to identify the nucleotide sequences of these sRNAs using Illumina sequencing. To avoid possible contamination by CMV Y-Sat (used in our Canberra laboratory), leaf samples from uninfected N. tabacum cv. Xanthi nc (Nt-Xanthi) grown under insect-proof conditions (in our Beijing laboratory where the CMV Shandong strain (SD-CMV) was used) were collected for sRNA extraction and sequencing. To investigate if CMV infection might affect the accumulation of Y-Sat-like sRNAs, we also sequenced sRNAs isolated from Nt-Xanthi infected with SD-CMVΔsatR, an infectious CMV clone devoid of SD-satRNA [29]. Approximately 17 and 23 million clean reads of sRNAs were obtained from the uninfected and SD-CMVΔsatR-infected plants, respectively, with 73.6% and 39% mapping perfectly to the uncompleted N. tabacum genome (ftp.sgn.cornel.edu) (Table 1). These N. tabacum-matching sRNAs were dominated by the 21 and 24-nt size classes (Table 1), consistent with the size distribution expected for plant sRNAs [30]. A large number of sRNAs matching the SD-CMV genome (27%) were identified in the sRNA sequencing data from SD-CMVΔsatR-infected plants, but none from the uninfected plants (Table 1). The majority (85.1%) of these SD-CMV-derived sRNAs was 21-22-nt in size, consistent with previous reports of sRNA distribution patterns from RNA viruses [19].
To identify Y-Sat-like sRNAs produced from the tobacco genome, the total sRNA reads were compared against the Y-Sat genome using BLASTN with varying statistical significance determined by E-values (i.e. the lower the E-value the greater the statistical significance of the match). Two additional CMV satRNA sequences (satCMV110 and SD-satRNA) were also used for comparison along with three randomly chosen similar sized (∼360 nt) non-satRNA sequences (SD-CMV RNA1, Influenza A virus subtype H1N1 and Rice stripe virus RNA3 genome sequences) as negative controls.
The number of matching sRNAs was generally increased in the SD-CMVΔsatR-infected sample compared to the uninfected sample (Table 2 and Fig. 6). At a low-stringency E-value (le−2), there was no significant difference in the number of sRNAs aligning to the three CMV satRNAs versus the three control sequences. However, using greater E-value stringencies (i.e. le−3, le−4 and le−5) to improve the alignment quality led to a higher frequency of sRNAs aligning to the Y-Sat genome than the control sequences in both the uninfected and particularly the SD-CMV△satRNA-infected samples (Table 2 and Fig. 6A, 6B). A large proportion of these Y-Sat-matching sRNAs were 24 nt in size (Table 2). The satCMV110 and SD-satRNA also had more aligning sRNAs in the SD-CMV△satRNA-infected sample than the control sequences when using the higher alignment stringency (Table 2). In fact, no sRNA aligned to the relevant control sequences when using an E-value of le−5, with the exception of sRNAs aligning to the SD-CMV control sequence, which is a result of the SD-CMV infection (Table 2). The 14 unique Y-Sat-matching sRNA sequences aligned with both the 5′ and 3′ regions of the Y-Sat genome (Fig. 6C), suggesting the possible existence of long stretches of Y-Sat-like sequences in the N. tabacum genome. The five SD-satRNA-matching sRNAs identified in the SD-CMVΔsatR-infected sample (1e−5) showed perfect sequence identity to the SD-satRNA genome, and had a size range of 23–26 nt (one 23-nt, one 26-nt and three 24-nt; Fig. 6B, the bottom panel; the colour-coded sequences). Three of these SD-matching sRNAs also aligned with the nt. 45–74 region of the Y-Sat and satCMV110 genomes (Fig. 6B, the top and middle panels; the colour-coded sequences). Taken together, the sRNA sequencing data indicated the presence of 24-nt sRNAs in the N. tabacum genome with sequence homologies to CMV satRNAs, particularly to Y-Sat, which could account for the 24-nt sRNA signals detected by northern blot (Fig. 5) and explain the DNA methylation of the Y-Sat sequence in the 35S-GUS:Sat transgenes (Fig. 3, 4 and S2 Fig.). It is noteworthy that the majority of the CMV satRNA-matching sRNAs could not be mapped to the uncompleted N. tabacum genome (S1 Table), indicating that these sRNAs are derived from the unannotated regions of the genome.
The CMV satRNA-matching sRNAs were more frequently aligning to certain regions of the satRNA genomes (Fig. 6 and S4 Fig.). Interestingly, these sRNA “hotspots” were relatively conserved among the three different satRNAs (boxed in Fig. 6 and S4 Fig.) and corresponded to conserved sequence regions among all CMV satRNA genomes (S5 Fig.). Among the three CMV satRNAs, SD-satRNA showed the most divergent distribution pattern of aligning sRNAs (Fig. 6A, B). Phylogenetic analysis of all published CMV satRNA sequences revealed that SD-satRNA, Y-Sat and satCMV110 are grouped into three distinct clusters, with SD-satRNA being the most ancient among the three (S6 Fig.).
In plants, 24-nt siRNAs are usually derived from repetitive DNA regions such as transposable element (TE) sequences in the genome [30]. Multiple hybridizing DNA bands were detected in N. tabacum, N. clevelandii, and N. benthamiana plants by Southern blotting using the Y-Sat sequence as a probe (Fig. 7A). The different band patterns among the three Nicotiana species ruled out plasmid DNA contamination in the DNA samples. We also observed differences in the intensity of the hybridization signals, which suggested that Y-Sat-like DNA sequences exist in the Nicotiana species but with different copy numbers. Digestion of N. tabacum DNA with various restriction enzymes confirmed the existence of multiple copies of the Y-Sat-like DNA sequences (Fig. 7B).
In this study we observed that a GUS transgene fused to the CMV Y-Sat sequence was transcriptionally repressed in N. tabacum plants in comparison to a non-fusion GUS transgene. Transcriptional gene silencing of a transgene is usually caused by promoter methylation and characterized by complete or near-complete silencing of the transgene in a subset of transgenic lines [31], [32]. However, the 35S-GUS:Sat transgenes showed low but consistent levels of GUS expression across the independent transgenic lines. Furthermore, the 35S promoter was not methylated, suggesting that the repression of the 35S-GUS:Sat transgene is not due to conventional TGS.
Both McrBC PCR and bisulfite sequencing detected high levels of DNA methylation in the transcribed region of the 35S-GUS:Sat transgene, with the methylation restricted to the Y-Sat sequence. The extent of methylation in the Y-Sat sequence appeared to be inversely correlated to the level of 35S-GUS:Sat transgene expression, suggesting that this methylation may play a direct role in the repression of the transgene. How DNA methylation in the transcribed region represses the transcription of the 35S-GUS:Sat transgene remains to be investigated. However, our finding provides the first example of transgene repression associated with host-induced methylation targeted to a specific sequence in the transcribed region. The Y-Sat-specific DNA methylation is similar to that observed for the Cereal yellow dwarf virus (CYDV) satRNA sequence of a GUS:CYDV-satRNA fusion transgene in N. tabacum [22]. The sequence-specific methylation of the CYDV sequence is caused by RdDM directed by sRNAs derived from replicating CYDV satRNA [22], also implicating RdDM as the cause of Y-Sat-specific methylation. However, methylation of the Y-Sat sequence occurred in the absence of replicating Y-Sat, suggesting that it was directed by host sRNA-induced RdDM.
Consistent with host-derived sRNA-induced RdDM being responsible for the methylation of the Y-Sat sequence, sRNAs with sequence homology to Y-Sat were readily detected by both northern blot hybridization and sRNA deep sequencing. Importantly, these sRNAs were primarily 24-nt in size, which are known to direct RdDM. This size distribution is different to that of sRNAs derived from replicating satRNAs, which are dominated by 21-22-nt size classes [19], ruling out contaminating satRNAs as the source of these sRNAs. In plants, 24-nt sRNAs are derived primarily from repetitive DNA sequences including TE sequences. Indeed, Southern blot analysis detected multiple DNA fragments in the Nicotiana genome with sequence homology to the Y-Sat sequence. Furthermore, the majority of the Y-Sat-matching sRNAs could not be mapped to the published N. tabacum genome sequence, suggesting that they are likely derived from highly repetitive regions of the genome that are usually difficult to assemble during genome sequencing and hence poorly annotated. Consistent with this possibility, a BLASTN search of the published genome sequences of Nicotiana species and two other Solanaceae species Solanum lycopersicum (tomato) and S. tuberosum (potato) did not yield any long (>20 bp) stretches of Y-Sat-like sequences, again implicating unassembled, repetitive regions as the source of the Y-Sat-matching sRNAs. Taken together, our results suggest that the Nicotiana genome contains Y-Sat-like DNA sequences, and that these sequences exist as repetitive DNA giving rise to 24-nt rasiRNA-like sRNAs that induce RdDM against the Y-Sat sequence in the 35S-GUS:Sat transgene. In addition to RdDM, 24-nt siRNAs have previously been shown to be capable of directing RNA degradation [33]. It would be interesting to examine if these host-derived 24-nt siRNAs play a role in the trilateral host-CMV-satRNA interaction by targeting satRNAs affecting satRNA accumulation.
Our results raise the possibility that CMV satRNAs originate from repetitive DNA elements in the Nicotiana genome. It has previously been speculated that satRNAs could be generated from the host plant under some unique conditions, such as helper virus infection [1]. Our sRNA sequencing showed that Y-Sat-like sRNAs accumulated at a much higher level in CMV-infected than uninfected N. tabacum plants. This implies that the Y-Sat-like repetitive DNA is transcriptionally repressed under normal conditions by DNA methylation, but is activated by CMV infection, possibly via the function of the 2b silencing suppressor that can repress DNA methylation in plants [34]. This would result in increased levels of transcript from repetitive DNA regions, including the Y-Sat-like DNA repeats, which could serve not only as substrate for sRNA production, but also as potential progenitor of CMV satRNAs. This scenario is consistent with the CMV-infected samples containing a much larger proportion of sRNAs that could not be mapped to the published N. tabacum genome than the uninfected samples (S1 Table). Viral replicases or RNA-dependent RNA polymerases have relatively high error rates [35], so replication of viral RNAs including satRNAs can be accompanied by a high rate of nucleotide mutations. Furthermore, for the host-derived progenitor RNAs to become satRNAs, they may need to undergo sequence changes to gain nucleotide motifs for efficient replication and encapsidation, and to gain stable secondary structures for resistance to nucleases. Thus, satRNAs, originating from the host, are likely to have substantial sequence variations from the original DNA or RNA sequences of the host genome. This would explain why most of the Y-Sat-matching N. tabacum sRNAs did not share perfect sequence identities with the Y-Sat genome. Our results do not rule out the possibility that the Y-Sat-like sequences in the N. tabacum genome was originally acquired from a viral genome like the non-retroviral RNA virus elements recently discovered in both plants and animals [36], [37]. However, the fact that CMV satRNAs lack the ability to self-replicate and possess no sequence homology with their helper viruses, together with the previous observation of de novo emergence of satRNAs during serial passaging plants with CMV infection under controlled environmental conditions [12], favours the view that the satRNAs originate from the sequence of the Nicotiana genome.
Our sRNA sequencing data indicated that different CMV satRNAs have different levels of sequence homology to the N. tabacum sRNAs. One possible explanation for this is that the different satRNAs originated from different Nicotiana species, or from different copies of repetitive DNA in one Nicotiana species, that have nucleotide variations. Another possibility is that these satRNAs originated from the Nicotiana genome at different times, with the more ancient ones having greater sequence divergence from the host genome than the more recent ones. This possibility is supported by the phylogenetic analysis of CMV satRNAs, showing the SD-satRNA to be more ancient than Y-Sat, which coincides with N. tabacum sRNAs matching better to Y-Sat than SD-satRNA.
In conclusion, by studying the abnormally repressed expression pattern of the 35S-GUS:Sat transgenes in Nicotiana plants, we have generated evidence indicating that CMV satRNAs have originated from repetitive DNA regions in the Nicotiana genome. Our study suggests that a small RNA sequence-based approach can be used to find the origin of satRNAs.
The 35S-GUS-Ocs cassette, prepared using the pART7 plasmid [38], was previously described [39]. The cassette was excised using NotI and inserted into the NotI site of pART27 that contains a NPTII kanamycin resistance gene as the selectable marker for plant transformation [38], resulting in the 35S-GUS construct. For the 35S-GUS:sSat and 35S-GUS:asGUS constructs, the CMV 369-nt Y-Sat sequence was assembled using four long overlapping oligonucleotides (Y-Sat2, 3, 4 and 5; S2 Table), then PCR-amplified using Y-Sat1 and Y-Sat6 that contained a HindIII restriction site, and cloned into pGEMT-Easy vector (Promega). The full-length Y-Sat fragment was then excised with HindIII and inserted into the HindIII site between the GUS and Ocs sequence in the 35S-GUS-Ocs cassette in either the sense or antisense orientation. The resulting 35S-GUS:sSat-Ocs and 35S-GUS:asSat-Ocs cassettes were then excised with NotI and inserted into pART27 at the NotI site, forming the 35S-GUS:sSat and 35S-GUS:asSat constructs, respectively.
For transformation of N. tabacum Wisconsin 38 (W38), the constructs were introduced into Agrobacterium tumefaciens LBA4404 via triparental mating. For Agrobacterium infiltration assay, the constructs were transformed into A. tumefaciens GV3101.
Transformation of N. tabacum W38 was performed as previously described [10] using 50 mg/L kanamycin as the selective agent plus 150 mg/L timentin to inhibit Agrobacterium growth. Transformed plants with established roots were transferred to soil and grown at 25°C under natural light.
Agro-infiltration of N. benthamiana leaves was carried out essentially as described previously [40] with minor modifications. Basically, A. tumefaciens strains containing respective plant expression constructs, including a green florescent protein (GFP) construct as a visual marker and a P19 construct as the RNA silencing suppressor [41], were grown overnight at 28°C in Luria-Bertani medium (LB) containing appropriate antibiotics. After centrifugation, Agrobacterium cells were re-suspended in buffer containing 10 mM MgCl2 and 150 µM acetosyringone, to a final optical density at 600 nm (OD600) of 1.0. Agrobacterium cell suspension containing either the 35S-GUS or the 35S-GUS:sSat construct was mixed with the GFP and P19 cell suspensions at a 1∶1∶1 ratio, incubated at room temperature for ∼3 h, and then infiltrated into expanded N. benthamiana leaves using a flat-pointed syringe. For protein and RNA isolation, agro-infiltrated leaf sections at 4 or 5 days post agroinfiltration (dpa) were visualized under a blue light torch (NightSea™, DFP-1™, for exciting green light emission), excised with a pair of scissors and immediately frozen in liquid nitrogen.
Genomic DNA used for Southern blot hybridization and bisulfite conversion was isolated using the CTAB method described by Draper and Scott [42]. Restriction digestion of DNA (∼20 µg), purification, agarose gel electrophoresis and Southern blot hybridization were essentially as previously described [39]. A HindIII fragment containing the full-length Y-Sat sequence was excised from the pGEM-T Easy clone described above, and used for preparing a 32P-labelled probe using the Megaprime DNA Labelling System (Amersham Biosciences). Hybridized membranes were washed with 1×SSC first at room temperature for 20 min, then at 50°C for 20 min, and finally at 60–65°C for 30 min. The blots were visualized using a phosphorimager (FLA-5000, Fuji Photo Film).
RNA used for small RNA northern blot hybridization was isolated from agro-infiltrated N. benthamiana leaf tissues or transgenic N. tabacum leaves using Trizol Reagent (Invitrogen). For northern blot detection of small RNAs from N. tabacum flowers, total RNAs was isolated using Trizol reagent, high-molecular-weight RNA removed by precipitation with 5% polyethylenglycol 8000 (PEG 8000) and 0.5 M NaCl on ice for 20 min, and small RNAs recovered from the supernatant by precipitation with 3 volumes of ethanol at −20°C for 1 hr. RNA used for northern blot analysis of GUS or GUS:Sat fusion mRNA was isolated using the hot-phenol extraction method [43]. sRNA northern blot hybridization was performed as described using 32P-labelled in-vitro antisense transcript of full-length Y-Sat or GUS coding sequence that were fragmented with Na2CO3 treatment [10]. mRNA northern blot hybridization was carried out as previously described [39] using the same full-length antisense GUS RNA (unfragmented) as probe.
GUS activities were determined using the kinetic MUG (4-methylumbelliferyl-β- glucuronide) assay according to Chen et al. [39]. GUS activities for Fig. 1B was calculated according to Chen at al. [39], with the Y-axis as pmol MU per min per 5 µg protein. GUS activities for the remaining figures were presented as slope value per 5 µg protein [44].
Nuclei were isolated from approximately 4 g of fresh N. tabacum leaves and used for nuclear run-on assay according to the procedure described in Meng and Lemaux [45]. The full-length GUS gene, the elongation factor 1α (EF1α) gene (used as internal reference), and the Fusarium oxysporum FKS1 gene (used as negative control) sequences, were amplified by PCR using the following primers (primer sequences are shown in S2 Table): M13-F and M13-R (from a pGEM-GUS plasmid), EF1α-F and EF1α-R, and FKS1-F and FKS1-R, respectively. The PCR product was separated in 1% agarose gel, denatured, neutralized and then blotted onto Hybond N+ membrane using 20× SSC following the manufacturer's instruction. The membrane was hybridized with the nuclear run-on transcript according to Meng and Lemaux [45] and the hybridizing signals were visualized using a phosphorimager (FLA-5000, Fuji Photo Film).
Genomic DNA (∼2 µg) from transgenic 35S-GUS:sSat N. tabacum plants, was mixed with 1×NEB buffer 2, 0.1 µg/µl BSA, 1 mM GTP in 94 µl volume, which was divided in two equal 47 µl aliquots. To one aliquot 3 µl McrBC enzyme (NEB, 10 units/µl) was added, to the other 3 µl of H2O. Both were incubated at 37°C overnight and then diluted with H2O to 100 µl. Four µl was used for each PCR reaction, which was performed using the following cycles: 1 cycle of 95°C for 3 min, 33 cycles of 95°C for 30 sec, 56°C for 45 sec, and 72°C for 90 sec, followed by one cycle of 72°C for 10 min. Sequences of the McrBC PCR primers are listed in S2 Table. Real-time PCR was performed using the Rotor-Gene 6000 (Corbett Life Science, San Francisco, USA) real-time rotary analyser using SYBR Green reagent and Platinum Taq polymerase (Invitrogen) in four technical replicates for each sample.
Bisulfite conversion of transgenic N. tabacum genomic DNA (∼4 µg) was performed as previously described [33]. The bisulfite-treated DNA was purified using Qiagen PCR Purification kit. Primer design (sequences shown in S2 Table), nested PCR and direct sequencing of PCR products were as previously described [44].
Total RNA was isolated from wild-type Nt-Xanthi plants and SD-CMVΔSat-infected plants using hot-phenol extraction and small RNAs were prepared as described previously [46]. Small RNA library construction and Illumina sequencing was performed by BGI (http://www.genomics.cn/en/index). For data analysis, adapter and low-quality sequences were removed and cleaned reads used for length distribution and subsequent analysis. Overlapping sequences between different libraries were identified using Perl script (http://www.perl.org/). All clean reads were mapped to the uncompleted Nt-Xanthi genome (ftp.sgn.cornel.edu) using SOAP (http://soap.genomics.org.cn/), and all perfectly mapped sequences were used to determine the mapping ratio in different samples.
All clean reads matching to satRNA and control sequences was examined using BLASTN (http://blast.ncbi.nlm.nih.gov/Blast.cgi) with different E-values (1e−2, 1e−3, 1e−4 and 1e−5) to determine different degrees of similarity. ClustalX (http://www.clustal.org/) was used for sequence alignments and phylogenetic analysis of satRNA genomes.
|
10.1371/journal.pcbi.1005509 | phyC: Clustering cancer evolutionary trees | Multi-regional sequencing provides new opportunities to investigate genetic heterogeneity within or between common tumors from an evolutionary perspective. Several state-of-the-art methods have been proposed for reconstructing cancer evolutionary trees based on multi-regional sequencing data to develop models of cancer evolution. However, there have been few studies on comparisons of a set of cancer evolutionary trees. We propose a clustering method (phyC) for cancer evolutionary trees, in which sub-groups of the trees are identified based on topology and edge length attributes. For interpretation, we also propose a method for evaluating the sub-clonal diversity of trees in the clusters, which provides insight into the acceleration of sub-clonal expansion. Simulation showed that the proposed method can detect true clusters with sufficient accuracy. Application of the method to actual multi-regional sequencing data of clear cell renal carcinoma and non-small cell lung cancer allowed for the detection of clusters related to cancer type or phenotype. phyC is implemented with R(≥3.2.2) and is available from https://github.com/ymatts/phyC.
| Elucidating the differences between cancer evolutionary patterns among patients is valuable in personalized medicine, since therapeutic response mostly depends on cancer evolution process. Recently, computational methods have been extensively studied to reconstruct a cancer evolutionary pattern within a patient, which is visualized as a so-called “cancer evolutionary tree” constructed from multi-regional sequencing data. However, there have been few studies on comparisons of a set of cancer evolutionary trees to better understand the relationship between a set of cancer evolutionary patterns and patient phenotypes. Given a set of tree objects for multiple patients, we propose an unsupervised learning approach to identify subgroups of patients through clustering the respective cancer evolutionary trees. Using this approach, we effectively identified the patterns of different evolutionary modes in a simulation analysis, and also successfully detected the phenotype-related and cancer type-related subgroups to characterize tree structures within subgroups using actual datasets. We believe that the value and impact of our work will grow as more and more datasets for the cancer evolution of patients become available.
| Cancer is a heterogeneous disease. The high genetic diversity is driven by several evolutionary processes such as somatic mutation, genetic drift, migration, and natural selection. The clonal theory of cancer [1] is based on Darwinian models of natural selection in which genetically unstable cells acquire a somatic single nucleotide variant (SSNV), and selective pressure results in tumors with a biological fitness advantage for survival.
The development of multi-regional sequencing techniques has provided new perspectives of genetic heterogeneity within or between common tumors [2–6]. The read counts from multi-region tumor and matched normal tissue sequences from each patient are then used to infer the tumor composition and evolutionary structure from variant allele frequencies (VAFs); i.e., the proportion of reads containing the variant allele. Using the VAF, the cancer evolutionary histories can be reconstructed as a tree, termed a cancer evolutionary tree, which reflects the accumulation patterns of the identified SSNVs for each patient.
A variety of cancer evolutionary trees can be considered as the consequence of the evolutionary principle for the underlying tumor, which may lead to resistance to chemotherapeutics and targeted therapies [19, 20]. Therefore, characterizing inter-tumor heterogeneity according to the patterns of evolutionary trees is an important strategy for developing new targeted therapies and for preventing the emergence of drug resistance. There are currently two types of cancer evolutionary trees: sample tree and sub-clonal tree. A sample tree regards each multi-region sample as being equivalent to a species in a classical tree of taxonomic phylogenetic relationships, and infers the evolutionary trees from the binary VAF profiles using classical phylogenetic algorithms such as the maximum parsimony method. A sub-clonal tree clusters SSNVs into sets of mutations with common frequency and reconstructs the lineage based on the following two assumptions [7–16]: (i) a mutation cannot recur during the course of cancer evolution, and (ii) no mutation can be lost [17] (Fig 1A). In these trees, the root and its subsequent node represent a normal cell and a founder cell, respectively. Descendant sub-clones are represented as nodes below the founder cell, and edge lengths indicate the number of SSNVs that are newly accumulated in the descendant nodes (Fig 1B). For reviews, see [18].
Several studies have suggested specific evolutionary patterns of tumors with various, and at times conflicting, results. For example, Gerlinger et al. [21] identified the parallel evolution of sub-clones in clear cell renal cell carcinomas (ccRCCs), whereas no such parallel evolution was evident in studies on non-small cell lung cancer (NSCLC) [22, 23]. Zhang et al. [23] also showed that in a relapsed group of patients, the fraction of SSNVs in sub-clones was significantly larger than that of founder cells. These studies indicate that both the branching patterns and fraction of SSNVs in sub-clones are important factors for identifying the cancer type or phenotype of related subgroups.
Although the reconstruction methods developed thus far have revealed intra-tumor heterogeneity by reconstructing individual evolutionary trees, there are currently no standardized tree comparison methods for obtaining a detailed understanding of inter-tumor heterogeneity according to evolutionary patterns with a set of reconstructed trees. Comparison of phylogenetic trees has long been discussed in the context of the evolution of species, and several comparative analytical methods have been developed, including nearest-neighbor interchanging [25], subtree transfer distance [26], quartet distance [27], Robinson-Foulds distance [28], path length metrics [29], branch length scores [30], tree edit distance [31–33], and Billera-Holmes-Vogtmann (BHV) distance [34]. However, these distances are defined for phylogenetic trees with the same set of leaves, and therefore cannot accurately deal with the following problems, (p1)–(p4), that are specific to the context of cancer evolutionary trees.
(p1) and (p2) imply that a tree structure is not always binary, and the number of sub-clones within the tree differs among patients. Various types of new sub-clones can be produced from a common ancestral sub-clone by acquiring new sets of SSNVs, which results in complex tree structures and tree sizes. (p3) and (p4) indicate that sub-clones are rarely identical among patients since the SSNVs within a patient are quite different from those of other patients, and it is thus hard to match sub-clones between patients. In addition, the total number of SSNVs can vary substantially among patients. Therefore, according to experimental conditions, it is important to adjust for these effects to effectively compare the trees. These problems motivated us to develop a method for the effective comparison among tree via transformation of the tree topologies and edge attributes, a procedure we refer to as tree registration.
In this paper, we propose a new clustering method for cancer evolutionary trees based on tree topologies and edge attributes that describe the relationships of sub-clones and the number of SSNVs that accumulate in the sub-clones. Our conceptual framework is based on object-oriented data analysis [24], in which the observation units are non-numeric objects such as functions and trees. The main contributions of this paper are development of (i) a tree registration method for cancer evolutionary trees, (ii) a clustering method of the registered trees, and (iii) an evaluation method of the clusters, which can be applied using our software phyC in the R environment.
In the registration, we resolve the issues raised in (p1)–(p4) through development of a method for transforming tree objects by mapping tree topologies and their attributes to make the trees comparable (Fig 1C). The registered trees are embedded in Euclidean space, which enables defining the distance between the cancer evolutionary trees. Based on this distance, we divide a set of the trees into several sub-groups with a clustering method (Fig 1D). We developed two tools for interpretation of the clusters: multidimensional scaling (MDS) and a sub-clonal diversity plot.
We evaluated the performance of phyC using simulated data that mimic the actual scenarios. We also demonstrate the applicability of phyC using two actual datasets from patients with ccRCCs [21] and NSCLC [23], respectively, to show the interpretability of the clustering results. phyC is implemented with R(≥3.2.2) and is available from https://github.com/ymatts/phyC.
We denote n reconstructed cancer evolutionary trees as X = {xi; i = 1, 2, …, n}, and the edges and edge lengths are denoted as {eij; i = 1, 2, …, n, j = 1, 2, …, mi} and {|eij|; i = 1, 2, …, n j = 1, 2, …, mi}, respectively. Without loss of generality, {ei1; i = 1, 2, …, n} indicates the edge from the normal cell to the founder cell. Given the number of terminal nodes Ni; i = 1, 2, …, mi, we set depth (i.e., the number of edges in the path from the root to the terminal node) as dik(i = 1, 2, …, n; k = 1, 2, …, Ni).
We developed a registration method for the cancer evolutionary trees. The goal of the registration is to transform the observed trees such that dissimilarities can be defined with consideration of the tree topologies and edge attributes. To solve the problems (p1)–(p4), we provide the following approaches, (q1) and (q2):
To account for different tree structures and sizes as raised in (p1), we consider a reference tree-encoding approach that is similar to [35]. In this approach, we prepare a very large bifurcated tree called a reference tree (corresponding to the maximum tree in [36] and encode the observed tree topologies and edge lengths onto the reference. Zero-length edges are regarded as degenerated edges (Fig 1C). The advantage of this approach is that once we encode the observed tree onto a reference tree, the comparison can be simply achieved for trees of the same structures and sizes. To account for the issue (p4), we developed a method for normalization of the edge length to remove the bias in the detected number of SSNVs.
Here, we describe the details of the registration method. First, we set the maximum depth in X as dmax(X) = max {dik; i = 1, 2, …, n, k = 1, 2, …, Ni} and define the reference tree as follows.
Definition 1 (Reference tree) The reference tree is a bifurcated tree with the minimum depth of dmax(X).
Thus, the reference tree has m = 2(2dmax(X) − 1) edges (Fig 1C). We denote the reference tree as Xref with edges and edge lengths Ek; k = 1, 2, …, m and |Ek|; k = 1, 2, …, m, respectively. The registration can then be defined with the reference tree.
Definition 2 (Registration) Registration is a mapping f: X ↦ Xref.
We define the mapped trees as Y = {f(xi); i = 1, 2, …, n}, and more specifically, the mapped edge and edge length are set to {Eik; i = 1, 2, …, n, k = 1, 2, …, m} and {|Eik|; i = 1, 2, …, n, k = 1, 2, …, m}, respectively. The number of edges differs between the observed tree and the reference tree, and we also need to account for any unmapped edges. Since the degenerated edges can be regarded as the zero-length edge when considering the distance of trees [36], we can define |Eik′| = |eij|; k′ ∈ A for the mapped edge index set A ⊆ {1, 2, …, m}, and define |Eik′| = 0; k′ ∈ B for the unmapped edge index B = {1, 2, …, m}\A.
To resolve (p3), we developed the mapping rule eij ↦ Eik for j = 1, 2, …, mi, k = 1, 2, …, m, such that the observed trees are mapped onto the reference tree beginning with sub-trees with the largest depths and moving on to those with the smallest depths (Fig 1C). When the depths are the same among the sub-trees, we use the edge length and map the sub-trees beginning with those with the largest edge lengths (Fig G in S1 Text).
In the last step of the registration, we perform normalization for the edge length. Zhang et al. [23] importantly suggested in NSCLC study that patients with relapsed disease had larger fractions in their primary tumors (average 41% in patients with relapse versus 24% in patients without relapse, p = 0.045 by t test). Therefore, we consider that the ratio of the number of accumulated SSNVs is an important factor to characterize and compare the cancer evolutionary trees, and we divided each edge length by the total number of SSNVs within patients.
To define the dissimilarity between the registered trees, we begin with the space of the set of the registered trees. Billera et al. [34] proposed the concept of a continuous tree space-associated geodesic distance metric as a natural way to embed and compare phylogenetic trees. This tree space consists of a set of Euclidean regions, called orthants, one for each tree topology. Orthants are joined together whenever one tree topology can be made into another by exchanging edges between the trees. Within an orthant, the coordinates of each point represent the edge lengths for a particular tree with the topology associated with that orthant. Since we only encode the observed trees onto the reference tree with the same topology, the registered trees do indeed lie in the same orthant as a special case of BHV space.
Corollary 1 (Euclidean embedding) The registered trees lie in Euclidean space.
We represent the registered tree as a vector, whose elements correspond to each edge length as z i ′ = ( z i 1 , z i 2 , … , z i m ) ; z i j ∈ R. Note that zero length edges are regarded as degenerated edges. Thus, n registered trees are represented as the n × m matrix Z ′ = ( z 1 , z 2 ,…, z n ).
We define the dissimilarity as follows:
s ( x i , x j ) ≔( z i - z j ) ′ ( z i - z j ) . (2)
The basic statistics of the cancer evolutionary trees can also be defined. The tree average is defined as μ = 1 n ∑ i = 1 n z i and the tree variance is defined as σ 2 = 1 n ∑ i = 1 n ( z i - μ ) ′ ( z i - μ ).
Based on the tree representation with Z, which can be regarded as n observations with an m features matrix, we can simply apply standard clustering algorithms and divide the n trees into subgroups. Hierarchical clustering was then implemented using phyC. To determine the number of clusters automatically, we applied the gap statistics criterion [37] with the NbClust R package [38].
Interpreting clustering results is a key issue for tree comparison, which requires understanding the features of the cancer evolutionary trees in clusters. In particular, visual representation can be a powerful tool for such interpretation. Therefore, we developed two computational tools for comparing trees and understanding the cluster features.
We evaluated the performance of the proposed method using simulation data that reflects real situations. The main purpose of simulation is to show the effectiveness of the registration process for cancer evolutionary tree classification when compared to methods without the registration. Moreover, we investigated whether the differences in tree topology or edge length are more important when comparing tree objects.
There are two types of tree comparison methods: one is based on only tree topologies, and the other is based on both tree topologies and edge attributes. We examined phyC from the viewpoint of classification performance for tree topologies, edge length, and both. We conducted the following three simulations:
Simulation I was conducted to examine whether phyC can classify tree topologies, i.e., the edge lengths are all the same. In simulation II, we examined the classification performance of phyC for edge length differences. Simulation III was designed to examine the performance of phyC for both tree topology and edge length differences between tree objects.
To evaluate the clustering results, we adopted three external clustering validation indices [40], which are described in S1 Text: purity (PR), normalized mutual information (NMI), and Rand index (RI). In the following three simulations, we created 100 replicates of each dataset and evaluated the mean and standard deviation of PR, NMI, and RI.
We here demonstrate the application of our proposed method using an actual ccRCC dataset [21] and an NSCLC dataset [23], consisting of 8 and 11 multi-regional tumor samples with VAFs collected among 587 and 7,026 SSNVs, respectively. Since both studies used the maximum parsimony method to reconstruct the cancer evolutionary trees, we also adopted this method to analyze the datasets with our approach. We binarized the VAF profiles with VAF ≥ 0.05 as one and otherwise zero. Using the binary profile, we estimated the phylogenetic trees using the function acctran in the R package phangorn [42] and we obtained 19 cancer evolutionary trees.
We developed phyC, which was designed for clustering a set of cancer evolutionary trees to characterize cancer evolutionary patterns according to tree shape, based on analysis of tree topologies and edge attributes. Using this approach, we effectively identified the evolutionary patterns with different degrees of heterogeneity in a simulation study. We also successfully detected the phenotype-related and cancer type-related subgroups when applying this method to actual ccRCC and NSCLC data.
Considering the generally high level of inter-tumor heterogeneity, it is important to be able to identify phenotype- or cancer type-related evolutionary patterns. Previous studies have classified and interpreted the branching patterns of such sub-clones with manual methods, and then separately analyzed the compositions of SSNVs in each sub-clone. However, development of a quantitative analysis method is required to deal with datasets containing a large number of patients with cancer evolutionary trees to characterize and interpret the evolutionary patterns.
Our approach relies on reconstruction methods of evolutionary trees, and we used a parsimony approach that is widely adopted in studies of multi-regional sequencing. The proposed method only requires knowledge of the edges and edge attributes of rooted trees, and is therefore widely applicable to outputs of other recently developed state-of-the-art reconstruction methods, which allowed us to consider the heterogeneous mixture of cells within a sample.
There are several limitations of the present method that are worth mentioning, which should be tackled in further investigations. First, we have ignored the specific content of SSNVs in the sub-clones. We believe that this is a reasonable assumption to some extent, since the variation of SSNVs is too large to yield an effective comparison. However, the effects and consequences of different types of SSNVs can also vary, such as driver mutations or passenger mutations. Thus, when comparing edges with the same lengths from different trees, the two edges may not actually be equivalent if driver genes are included in one edge but not in the other. The first cut distinction between driver and passenger mutations could also simplify the algorithm and improve its running time. Therefore, a method that can incorporate the effect of driver genes in the sub-clones should be explored in future work.
Second, we have here only considered the SSNVs accumulating in the evolutionary trees, ignoring potential copy number or epigenetic aberrations; however, these factors may also affect heterogeneity within a tumor. Multi-regional sequence analysis has been performed using exome sequencing as well as copy number, methylation, and mRNA expression array profiling, providing an integrated interpretation of cancer evolution [43]. To determine the evolutionary patterns from these integrated data, our method can be extended to the case of multivariate edge attributes, including copy number variations and hyper- or hypo-methylation, as well as other genetic and epigenetic aberrations.
Finally, we did not take into account the potential effects of regional sampling biases and individual variations among tumors or patients. Gerlinger et al. [21] pointed out that increasing the sequenced regions of samples might lead to additional detection of sub-clones, and thus the complexity of inferred evolutionary trees might be affected by the sampling strategy. Therefore, a method for sampling bias reduction is needed to improve the clustering accuracy and plausible interpretation.
Our proposed approach represents the first practical method to quantitatively and accurately compare a variety of evolutionary trees with different structures, sizes, and labels, and with biases of edge length, while further allowing for biological interpretation. Our results imply that this approach has potential applications for personalized medicine such as predicting the outcomes of chemotherapeutics and targeted therapies, e.g., drug-resistance, based on evolutionary trees. We believe that the value and impact of our work will grow as more and more multi-regional sequencing datasets of patients become available.
|
10.1371/journal.pgen.1007992 | A DNA repair protein and histone methyltransferase interact to promote genome stability in the Caenorhabditis elegans germ line | Histone modifications regulate gene expression and chromosomal events, yet how histone-modifying enzymes are targeted is poorly understood. Here we report that a conserved DNA repair protein, SMRC-1, associates with MET-2, the C. elegans histone methyltransferase responsible for H3K9me1 and me2 deposition. We used molecular, genetic, and biochemical methods to investigate the biological role of SMRC-1 and to explore its relationship with MET-2. SMRC-1, like its mammalian ortholog SMARCAL1, provides protection from DNA replication stress. SMRC-1 limits accumulation of DNA damage and promotes germline and embryonic viability. MET-2 and SMRC-1 localize to mitotic and meiotic germline nuclei, and SMRC-1 promotes an increase in MET-2 abundance in mitotic germline nuclei upon replication stress. In the absence of SMRC-1, germline H3K9me2 generally decreases after multiple generations at high culture temperature. Genetic data are consistent with MET-2 and SMRC-1 functioning together to limit replication stress in the germ line and in parallel to promote other germline processes. We hypothesize that loss of SMRC-1 activity causes chronic replication stress, in part because of insufficient recruitment of MET-2 to nuclei.
| Post-translation modifications to histone proteins are known to regulate gene expression and chromosomal events such as recombination. Histone modifications are highly dynamic and are deposited by large number of histone-modifying enzymes. Little is known about how these enzymes are regulated. Using a model system, the nematode Caenorhabditis elegans, we show that a conserved histone-modifying enzyme, MET-2, associates with a conserved DNA repair protein, SMRC-1. In mammals, the SMRC-1 homolog, SMARCAL1, participates in repairing DNA that is damaged during replication. Focusing on the tissue responsible for production of sperm and eggs, the germ line, we find that SMRC-1 protects cells from DNA replication stress and promotes the accumulation of nuclear MET-2. Moreover, SMRC-1 affects MET-2 germline activity (as measured by histone modification state) in populations grown for multiple generations at stressful culture temperatures. Genetic analysis indicates that MET-2 and SMRC-1 participate in a common mechanism to limit DNA damage in the germ line. We propose that histone modifications are regulated to promote DNA replication and DNA repair.
| Repetitive sequences pose challenges to genome integrity during DNA replication, DNA repair, and transcription. In eukaryotes, repetitive genomic regions typically adopt a condensed chromatin structure that is thought to limit potentially harmful consequences of repetitive sequences by limiting transcription, stabilizing DNA to promote correct repair of DNA breaks, and limiting formation of secondary structures that would otherwise impede DNA replication [1–4]. Inappropriate transcription of repetitive regions leads to DNA:RNA hybrids (R-loops), which can prevent replication fork progression. DNA break repair is particularly important at repetitive regions because homologous recombination between non-allelic repetitive sequences causes duplication/deletion of genomic regions [5, 6]. Replication of heterochromatic regions also requires modification of histones within newly incorporated nucleosomes; histone chaperones and some DNA replication factors recruit histone methyltransferases for this purpose [7, 8]. Beyond regulation at repetitive sequences, replication and chromatin state are interdependent throughout the genome, e.g., chromatin compaction influences replication fork progression [9], and conversely, impaired replication can affect chromatin modification status and reduce the accuracy of histone incorporation at sites across the genome [10, 11]. Thus, the interplay among histone modifications, DNA replication, and repetitive sequences is complex.
H3K9 methylation is a histone modification widely associated with heterochromatin [12, 13]. In C. elegans, different repetitive sequences accumulate H3K9me2 and/or H3K9me3 [14–18], and loss of these marks correlates with increased susceptibility to DNA replication stress [17]. H3K9me1 and me2 are deposited primarily by MET-2 (methyltransferase-2), the sole C. elegans member of the SETDB1 family important for heterochromatin establishment and maintenance in numerous species [19–22]. MET-2 also promotes H3K9me3 formation, perhaps indicating that H3K9me1/me2 are substrates for H3K9 trimethylation [22]. SET-25 (SET domain proteins) is responsible for H3K9me3 at other sites [21, 22], and SET-32 is required for H3K9me3 in the initiation of heritable RNA-based transcriptional silencing [23–25]. Genetic studies indicate that MET-2, alone or together with SET-25, promotes germline viability and is critical for fertility in strains maintained at elevated culture temperatures over numerous generations [17, 18, 21, 26]. Moreover, during meiosis, H3K9me2 is enriched on non-synapsed chromosomes, e.g., the male X chromosome, characteristic of a process termed meiotic silencing [27, 28]. Overall, H3K9 methylation at repetitive sequences appears to ensure long-term stability of the genome and production of viable gametes and offspring.
Known SETDB1 interactors include co-factors as well as proteins required for stable interaction with chromatin or for re-establishing H3K9 methylation following DNA replication. Members of the ATF7IP (activating transcription factor 7-interacting protein; also called mAM/MCAF1, Mbd-1 chromatin associated factor) protein family are SETDB1 co-factors in vertebrates [29, 30] and Drosophila [31]. C. elegans LIN-65 is a structurally related (but not orthologous) protein necessary for H3K9me2 deposition [32, 33] and for MET-2 nuclear import in the embryo [33]. C. elegans ARLE-14 promotes MET-2 association with chromatin [33] as do vertebrate KAP1 (KRAB-associated protein 1; also called TRIM28) and hnRNP K (heterogeneous nuclear ribonucleoprotein K) [20, 21, 34]. SETDB1 also associates with a member of the SWI/SNF ATPase family, BAF155/SMARCC1 [34], and inactivation of BAF155, or any of several other BAFs, impairs SETDB1 activity at retroviral elements [35, 36]. During replication, CAF-1 (chromatin-associated factor–1) and MBD1 (methyl-CpG binding domain protein) recruit SETDB1 to re-establish H3K9 methylation behind the replication fork [7, 37, 38]. Thus, numerous factors ensure H3K9 methylation in different contexts.
To better understand how MET-2 activity is targeted in the C. elegans germ line, we sought to identify MET-2-interacting proteins. Here we describe SMRC-1, the sole C. elegans ortholog of vertebrate SMARCAL1 (SWI/SNF-related, matrix associated, actin-dependent regulator of chromatin, subfamily A-like 1). SMARCAL1-related proteins comprise a distinct subfamily of SWI/SNF ATPases and are thought to protect genome integrity by promoting the repair and restart of stalled DNA replication forks [39–41]. In vitro, SMARCAL1 proteins bind single stranded (ss) DNA and can rewind DNA substrates, e.g., replication forks and D-loops at Holliday junctions, and RNA:DNA substrates, meaning they might act on R-loops [39, 42–44]. Studies in human cultured cells showed that telomere maintenance, an endogenous source of replication stress, requires SMARCAL1 activity [45]. The MET-2—SMRC-1 association is interesting given that MET-2 provides some protection against lethality caused by DNA replication stress [17, 18].
We demonstrate that SMRC-1 protects against DNA replication stress, limits accumulation of DNA breaks and mutations, and promotes germline and embryonic viability and development. Moreover, SMRC-1 promotes H3K9me2 deposition and an increase in nuclear MET-2 abundance under conditions of replication stress. Genetic data suggest MET-2 and SMRC-1 function in a common mechanism in the germ line to limit DNA damage caused by replication stress and in parallel to promote other germline processes. Our data suggest SET-25 does not promote SMRC-1-mediated processes and has a minimal role in limiting replication stress. Taken together, our data suggest that SMRC-1 recruits MET-2 to limit the adverse effects of replication stress.
To facilitate our study of SMRC-1, we generated loss-of-function smrc-1 mutations using CRISPR-Cas9 genome editing. These included nonsense (om136), frameshift (om138, ea8), and deletion (ea46, ea73) alleles, each predicted to be severe loss-of-function (S1A Fig) [46–48] (see Methods). Each allele was outcrossed and maintained as a balanced heterozygote. We also epitope-tagged the endogenous smrc-1 locus in order to analyze SMRC-1 protein expression (S1A Fig).
We evaluated the smrc-1 phenotype in order to determine the importance of SMRC-1 during development. smrc-1 mutants had reduced fertility, an increased frequency of male offspring, and reduced embryonic viability (Table 1). These phenotypes resulted primarily from the loss of maternal smrc-1(+) product and were more severe at elevated culture temperature. At 25°C, smrc-1(om136) M+Z- F1 hermaphrodites (the progeny of smrc-1(+/-) mothers) were viable and fertile, although they produced fewer embryos than did wildtype controls (S1 Table, Table 1). Some smrc-1(om136) M-Z- F2 individuals died as embryos, and survivors included a high proportion of males (a Him phenotype, typically due to X chromosome nondisjunction) (Table 1). Most viable smrc-1(om136) M-Z- F2 hermaphrodites were fertile but produced fewer embryos than the F1 generation (Tables 1 and 2). Most F3 embryos were non-viable (Table 1).
We investigated the developmental defects underlying impaired smrc-1 fertility by DAPI staining smrc-1(om136) M-Z- F2 adult hermaphrodites and evaluating their germ lines. Fertile F2 adults typically had normal germline organization, whereas sterile F2 adults had obvious germline defects such as abnormal nuclear morphology, reduced numbers of germ cells compared with wildtype, and/or failure to produce sperm, oocytes, or both gamete types (S2 Fig). In rare cases, germ cells were not at all visible in the adult gonad. When both sperm and oocytes were present, one or both gamete types were presumably fertilization-defective. A subset of sterile hermaphrodites had polyploid (endomitotic) oocytes in the oviduct, a phenotype that most commonly arises due to impaired ovulation [49, 50]. We conclude that SMRC-1 function promotes multiple aspects of germline development.
We addressed the sensitivity of smrc-1 mutants to DNA replication stress by exposing them to hydroxyurea (HU), a treatment that causes replication fork stalling. We initially examined treated smrc-1(om136) M-Z- F2 individuals, and later examined M+Z- F1 individuals for comparison with low-fertility genotypes. We treated larvae with HU beginning at L1 stage and monitored their survival and fertility (see Methods). Survival of L1 larvae post-HU exposure reflects their ability to resolve DNA lesions and resume development; fertility of surviving adults specifically reflects the ability of mitotic germ cells to resolve DNA lesions. HU treatment had a significantly more severe effect on viability and fertility of smrc-1 M-Z- F2 mutants than wildtype (Fig 1A). The replication stress hypersensitivity of the smrc-1 mutant suggests a prominent role for SMRC-1 in limiting replication-associated DNA damage.
We hypothesized that smrc-1 mutants might accumulate mutations over successive generations which would reduce survival and fertility as a result of errors due to replication stress, and possibly other sources of DNA damage [42, 51–54]. We evaluated this possibility by serially passaging 16 smrc-1(ea8) mutant lines at 25°C and recording brood sizes in each generation (see Methods). To eliminate bias, we passaged the first L4 larva at each generation; if that animal developed as a sterile adult, we rescued the line by passaging a fertile sibling. We observed a broad range of brood sizes at each generation among the 16 serial lines (0 to >100 offspring) (Fig 2). Eleven lines had to be rescued by siblings at least once in the course of 30 generations. Overall, there was a trend toward reduced fecundity in successive generations and populations appeared to become sicker.
Germ cell apoptosis is elevated in many C. elegans DNA damage response mutants [55], and thus we considered that elevated apoptosis might contribute to the reduced smrc-1 fertility. We evaluated apoptosis by monitoring expression of CED-1::GFP, a protein expressed on the surface of phagocytic cells as they engulf apoptotic cells [56], and by staining with the vital dye, acridine orange. Here, CED-1 is expressed by sheath cells, components of the somatic gonad that engulf apoptotic germ cells. Both assays revealed elevated levels of apoptosis in smrc-1 germ lines compared to controls (Fig 3A, S2 Table). In the C. elegans hermaphrodite gonad, CED-1 is expressed by somatic sheath cells adjacent to the germ line [56]. As expected based on the literature, we observed rare CED-1::GFP -positive cells at the loop region of wildtype gonad (Fig 3A). smrc-1 mutants contained significantly more CED-1-positive cells, often present throughout the germ line (Fig 3A). smrc-1 apoptosis was significantly reduced in the absence of CEP-1/p53 (Fig 3A, S2 Table), an essential component of the DNA damage checkpoint machinery active at the late pachytene stage [57]. In contrast, apoptosis was not significantly suppressed by inactivation of PCH-2 (S2 Table), a component of the machinery that monitors chromosome pairing [58]. We conclude that smrc-1 mutants accumulate unrepaired DNA damage that, in turn, triggers the DNA damage checkpoint and results in elevated germline apoptosis.
C. elegans mutations that cause DNA damage to accumulate, due to either an increased number of DNA lesions or impaired DNA damage repair machinery, are classified as “mutators” [59]. We hypothesized that SMRC-1 might limit accumulation of mutations. We investigated this possibility in two ways. First, we screened for reversion of the dominant unc-58(e665) phenotype; this assay allows detection of intragenic and extragenic suppressors and is commonly used to quantify mutator activity [60]. We observed a 3- to 5-fold increase in unc-58 reversion in the smrc-1 mutant background compared to wildtype at both 20°C and 25°C (Fig 3B). For comparison, reversion increased 8- to 15-fold in DNA damage response mutants clk-2, hus-1, and mrt-2 [60]. Second, we assayed for enhancement of the dog-1 phenotype. DOG-1 (deletions of G-rich DNA) helicase is related to human FANCJ and essential for proper replication of poly G/C tracts in C. elegans [61]. Poly G/C tracts can assume a DNA secondary structure that is a natural source of replication stress [62]. Proteins that function in the DNA damage checkpoint or homologous recombination, e.g., CEP-1 or XPF-1 and SWS-1, respectively, are implicated in maintaining poly G/C tract integrity in the absence of dog-1 activity [63–65]. We evaluated poly G/C tract integrity by assessing the deletion rate within G/C-rich exon 5 of the vab-1 gene. vab-1 microsatellite deletions or insertions did not accumulate in smrc-1(ea8) or smrc-1(ea46) single mutants (Fig 3C). In contrast, deletion frequency was two-fold higher in dog-1;smrc-1 double mutants compared to dog-1 single mutants at both 20°C and 25°C (Fig 3C). For comparison, deletion frequency was increased 2.7-fold in cep-1, 3.2-fold in xpf/him-9, and 1.5-fold in cep-1 mutants compared to wildtype [63, 65]. We conclude that SMRC-1 limits the accumulation of deletions within poly G/C regions when DOG-1 activity is absent.
In vitro studies suggest that mammalian and Drosophila SMARCAL1 may act on DNA:RNA hybrids (43), and R-loops cause DNA replication stress in vivo (5). We were interested in determining if SMRC-1 might limit accumulation of R-loops. We evaluated R-loop abundance in wildtype and smrc-1(om138) M-Z- F2 mutants by immunolabeling with a DNA:RNA hybrid-specific antibody, S9.6 [66]. We quantified the proportion of nuclei with S9.6 foci (the S9.6 labeling index) in the proliferative, leptotene/zygotene, and pachytene regions of the germ line. smrc-1 germ lines had a significantly greater S9.6 labeling index than wild type in each of these regions (Fig 3D). Moreover, the S9.6 -positive smrc-1 germ line nuclei had more S9.6 foci on average than the positive wildtype nuclei (Fig 3D). We interpret these data to indicate that SMRC-1 activity limits R-loop abundance.
We considered whether the loss of SMRC-1 function in proliferating germ cells might contribute to formation of DSBs that could impact genome integrity. In early C. elegans meiosis, SPO-11 endonuclease initiates DSB formation at multiple sites along each chromosome; in most nuclei, only one DSB per chromosome is repaired as a crossover (CO) and others are repaired as noncrossovers (NCOs) [67, 68]. COs do not occur and homologs prematurely dissociate at diakinesis if SPO-11 is absent [67, 68]. Introduction of DSBs from exogenous sources, such as ionizing radiation (IR), can partially rescue COs in spo-11 mutants [67, 68].
We took advantage of the spo-11 univalent phenotype to test whether unregulated (non-SPO-11-mediated) DSBs might arise in smrc-1 mutants and contribute to the loss of germline viability. First, we generated smrc-1; spo-11 double mutants and evaluated diakinesis chromosomes in the most proximal oocyte (at the -1 position) in each gonad arm. In smrc-1(ea8) and smrc-1(om138) single mutants, we observed 4–7 DAPI-bright bodies in the -1 oocyte (Fig 3E). Faint links occasionally were visible between what appeared to be distinct chromosomes in smrc-1 mutant oocytes, suggesting aberrant connections between non-homologous chromosomes (Fig 3E). The presence of 7 DAPI-bright bodies in some nuclei is consistent with five synapsed autosomal pairs and two non-synapsed X chromosomes, which would lead to nullo-X gametes and subsequent production of male offspring, as observed. In smrc-1(om138);spo-11(ok79) double mutants raised at 25°C, we observed striking evidence of additional DNA damage. We observed 5–12 DAPI-bright bodies, and more frequent faint links between chromosomes (at least one linkage observed in 22% of nuclei, N = 41) (Fig 3E). Therefore, SMRC-1 appears to limit production of aberrant DNA damage that could be carried into meiosis and allow inappropriate connections between chromosomes.
RAD-51 is a single-strand DNA binding protein that associates with ssDNA adjacent to DSBs and facilitates the homology search and strand engagement in homologous recombination [69]. We performed anti-RAD-51 labeling to evaluate the distribution of DSBs in the smrc-1(om138) M-Z- F2 germ line. We observed RAD-51 foci primarily in meiotic nuclei and more rarely in mitotic nuclei, similar to wild-type controls (S4A Fig). Several differences were noted, however. Specifically, while DSBs occurred in leptotene stage in both wild type and smrc-1 mutants, RAD-51 foci formation was delayed in smrc-1. One possibility is that SPO-11-induced breaks occurred over a more protracted period of time in smrc-1 mutants. Also, a large number of RAD-51 foci persisted into late pachytene (zone 6) in smrc-1, suggesting that DSB repair is delayed, as might be expected for the non-SPO-11 mediated DNA damage (described above). Finally, the total number of RAD-51 foci was elevated in smrc-1 mutants compared with wild type. This increase is consistent with the presence of both SPO-11-mediated and pre-meiotic DNA damage-associated DSBs in the smrc-1 germ line.
We next evaluated RAD-51 foci in smrc-1;spo-11 meiotic nuclei as a means of visualizing aberrant DSBs. Few RAD-51 foci were observed in the spo-11(ok79) negative control, as expected [70]. RAD-51 foci were substantially more abundant in smrc-1(om138);spo-11(ok79) and smrc-1(om136);spo-11(me44) germ lines, particularly in leptotene-pachytene nuclei (S3A Fig). These foci may represent DSBs that formed due to DNA damage during mitosis or pre-meiotic S phase. We note that RAD-51 foci were more abundant at late pachytene nuclei in smrc-1 single mutants, which presumably contain both SPO-11-induced and aberrant DSBs. This result raises the possibility that smrc-1 is required for the normal processing/repair of meiotic DSBs.
The presence of elevated RAD-51 foci prompted us to ask whether meiotic CO events might have an altered distribution in smrc-1 mutants. For C. elegans, CO frequency is significantly greater on the autosomal arms than in the chromosome centers [71]. We first assayed CO frequency in control and smrc-1 animals in two small intervals within the central region of chromosome I. Our data indicated a several-fold increase in recombination between visible marker mutations in these intervals in smrc-1 mutants compared to controls (S3B Fig). We next mapped CO distribution along the length of chromosome I by assaying single nucleotide polymorphisms (SNPs). We generated a smrc-1 allele in the polymorphic CB4256 strain background and evaluated SNPs distributed across chromosome I (S1A Fig; see Materials and methods). This strategy allowed us to measure recombination within five large intervals along the chromosome (S3C and S3D Fig). The recombination frequency in these large intervals was not statistically different from controls except for a significant decrease in recombination frequency within interval 4 (P<0.03). Strikingly, we observed a >7-fold increase in the frequency of double CO events in the smrc-1 background relative to the control, which indicates impaired CO homeostasis. The differences obtained with the two mapping strategies could be explained by the size of the regions assayed; the domains with the genetic markers are both contained within the -8.5cM– 5cM region assayed by the chromosome-wide analysis. By interrogating a large domain with out cytological assay, fluctuations in recombination at the local level may be buffered by compensatory changes in nearby regions. Alternatively, the locally elevated CO rates between the visible markers might be explained by the presence of sequences that are prone to breakage, e.g., microsatellite repeats, are located between the pairs of visible markers. Indeed, numerous microsatellite repeat sequences are located between unc-11 and dpy-5, including a ~14.8 kb cluster (at chromosomal position 4280037–4294876); several smaller microsatellite repeat clusters are located between dpy-5 and unc-13 (www.wormbase.org). Such sequences could also account for the increased frequency of double COs.
Our attention was originally drawn to SMRC-1 as a consequence of our co-immunoprecipitation (co-IP) studies designed to identify MET-2-associated proteins. In these studies, we performed IPs using anti-MET-2 polyclonal antibody (described in Mutlu et al., 2018) and consistently recovered a protein of the expected size, ~150 kD, that was absent from met-2(n4256) negative controls (Fig 4A). SMRC-1 was recovered in these assays. To validate the association, we 3xflag-tagged the endogenous smrc-1 gene using CRISPR-Cas9 genome editing (S1A Fig) and performed anti-FLAG co-IP. We consistently recovered MET-2 in the 3xFLAG::SMRC-1 co-IP (Fig 4B).
We investigated SMRC-1 and MET-2 distribution in the germ line to identify where they are co-expressed. We visualized SMRC-1 by immunolabeling dissected 3xflag::smrc-1 gonads. We note that 3xflag::smrc-1 animals developed normally and had brood sizes similar to controls, suggesting the epitope tag did not substantially impact SMRC-1 function (Table 1). We detected 3xFLAG::SMRC-1 in proliferative and meiotic germ cell nuclei in XO males and XX hermaphrodites (Fig 5A and 5B). In males and hermaphrodites, labeling intensity decreased as nuclei transitioned from the proliferative region into early meiosis (leptotene-zygotene stages) and then increased again as nuclei moved through pachytene and diplotene stages. In males, signal decreased again during the condensation phase of spermatogenesis and moved to the nuclear periphery, well apart from chromatin (Fig 5A). In hermaphrodites, signal intensity was strongest in diakinesis stage oocytes, consistent with embryos inheriting substantial SMRC-1 protein (Fig 5B).
We visualized MET-2 with several reagents, including anti-MET-2 antibody, epitope-tagged transgene generated by mosI-mediated single copy insertion (mosSCI), and endogenously-tagged MET-2 generated by CRISPR-Cas9 editing (S1A Fig; Materials and methods) [33] These reagents detected nuclear MET-2 throughout the germ line (Fig 6, S1B–S1D Fig), consistent with both our previous observation of nuclear MET-2 in embryonic nuclei using the same reagents [33] and also examination of adult somatic tissue [72]. We observed MET-2 puncta superimposed on a more diffuse signal in germline nuclei and, to a lesser extent, cytoplasm in both male (Fig 6A) and hermaphrodite (Fig 6B) germ lines. The MET-2 distribution appeared to shift as germ cells moved from the proliferative region into and through meiosis; nuclear puncta were more obvious in mitotic and leptotene-zygotene nuclei, and the signal became more evenly distributed as nuclei entered and progressed through pachytene stage (Fig 6). We conclude that germline MET-2 comprises nuclear and cytoplasmic pools. Nuclear puncta resemble the nuclear hubs observed in embryos which are thought to be sites of methyltransferase activity [33].
We co-visualized SMRC-1 and MET-2 using a 3xmyc::smrc-1 3xflag::met-2 strain generated by CRISPR-Cas9 genome editing (S1A Fig). Labeling this strain verified that SMRC-1 and MET-2 are expressed in the same nuclei (Fig 6C, S4 Fig). We observed partial overlap between 3xMYC::SMRC-1 and 3xFLAG::MET-2 signals within nuclei as would be expected for proteins that physically associate (Fig 6C, S4 Fig).
Human SMARCAL1 localizes to stalled replication forks and forms foci in response to HU treatment of cultured cells [41]. We tested the impact of stalled replication on SMRC-1 distribution by comparing the 3xFLAG::SMRC-1 signal in germ cells with and without HU treatment. For this assay, we treated L4 larvae with 25mM HU for 24 hours at 22°C and then dissected and immunolabeled the adult gonads. Germ cell nuclei located distal to the leptotene/zygotene region were enlarged and appeared to have ceased mitosis, consistent with robust activation of the replication checkpoint (Fig 7A). We quantified the anti-FLAG signal and normalized it relative to (i) DAPI, (ii) mCherry-tagged histone H2B included in the strain background, and (iii) histone H3. We consistently observed elevated SMRC-1 abundance in distal nuclei of HU-treated animals (Fig 7A, S5 Fig), suggesting that DNA replication stress triggered an increased SMRC-1 abundance during mitosis.
Zeller et al. (2016) reported that met-2 set-25 double mutants have reduced viability following HU treatment, suggesting that H3K9 methylation offers protection from replication stress [17]. DNA damage has been shown in other systems to increase H3K9me2 levels in other systems [3]. Regulated nuclear import of MET-2 is one way in which its activity is controlled in the embryo [33]. Given these observations, we hypothesized that nuclear MET-2 abundance in the proliferative germ line might increase under conditions of replication stress. We tested this idea by treating 3xflag::met-2 L4 larva with HU (as described above for 3xflag::smrc-1, see Methods) and visualizing 3xFLAG::MET-2 by immunolabeling. We reproducibly observed elevated MET-2 levels in mitotic germ cell nuclei of HU-treated animals compared to untreated controls (Fig 7B). We conclude that replication stress triggers an increased MET-2 accumulation in nuclei of proliferative germ cells.
We performed anti-H3K9me2 labeling to determine if the increase in nuclear MET-2 correlates with increased activity. As previously reported, we observed weak or no H3K9me2 signal in the proliferative germ line and any signal that was present tended to be punctate and located near the nuclear periphery ([21, 27, 73]; this study). In HU-treated germ lines, we observed weak, diffuse labeling that tended to be located more centrally (Fig 7D). To compare the H3K9me2 signal in these two sets of nuclei, we modified the Corrected Total Cell Fluorescence calculation previously developed to compare cellular immunolabeling signal to calculate specifically a Corrected Total Nuclear Fluorescence (CTNF) value (Fig 7D) (see Materials and methods). The CTNF was significantly greater for distal nuclei that had received the HU treatment, suggesting that replication stress led to an increase in H3K9me2 levels.
Since nuclear SMRC-1 and MET-2 levels increase in the distal germline upon replication stress, we asked if SMRC-1 promotes the MET-2 increase. For this purpose, we generated a strain carrying the smrc-1(om138) mutation in a 3xflag::met-2 background and assayed the impact of HU treatment. We observed a significant increase in nuclear MET-2 abundance, however the increase was less pronounced and more variable than in smrc-1(+) controls (Fig 7C). These results are consistent with SMRC-1-dependent and -independent regulation of nuclear MET-2 accumulation during replication stress.
In the course of these experiments, we noted that smrc-1 and wildtype germ cells responded differently to HU treatment. In wildtype, distal germline nuclei became notably enlarged and decreased in number, as reported in the literature (e.g., [74]). The size increase and number reduction were less pronounced in smrc-1 nuclei, although still significant (Fig 8A). To investigate if smrc-1 mutants were resistant to mitotic arrest, we repeated the HU treatment and performed anti-H3S10phos (histone H3 phosphorylated on serine 10) labeling to detect mitotic nuclei [75]. Untreated wildtype and smrc-1(om138) mutants had the same mitotic index, whereas HU-treated smrc-1 mutants had a significantly higher mitotic index than HU-treated wildtype controls (Fig 8B). Hence, smrc-1 germ cells appear to be resistant to mitotic arrest. We note that the failure to elicit a cell cycle arrest is not due to an inability to respond to HU, as there was a decrease in mitotic nuclei numbers of HU exposure in smrc-1 mutants. Failure of mitotic arrest may explain why MET-2 abundance does not increase as much in the distal germ line of HU-treated smrc-1 mutants as it does in HU-treated wildtype. Mitotic arrest occurs when the mitotic DNA damage checkpoint has been tripped; this checkpoint is HUS-1-dependent and distinct from the later DNA damage checkpoint that triggers apoptosis [76]. Perhaps SMRC-1 promotes the mitotic DNA damage checkpoint, and hence the resistance to mitotic arrest observed in smrc-1 mutants.
We asked whether SMRC-1 promotes germline H3K9me2 by immunolabeling smrc-1 mutants passaged for either two or 30 generations at 25°C. We evaluated smrc-1(om136) XX hermaphrodite and XO male germlines in the M-Z- F2 generation and in five smrc-1(ea8) lines in the F30 generation. In M-Z- F2 animals, the H3K9me2 labeling pattern appeared comparable to wild type in both smrc-1 XO and XX germ cells (Fig 9A). Among F30 germ lines, the average H3K9me2 signal was weaker than wildtype in a majority of gonads evaluated (Fig 9B). In a small subset, H3K9me2 signal was comparable to or greater than controls (Fig 9B). H3K9me2 signal was similar among nuclei within individual germ lines, suggesting a systemic change in H3K9me2 regulation throughout the tissue.
To investigate the genetic relationship between smrc-1 and met-2, we generated a smrc-1 met-2 double knockout strain (S1A Fig) and assayed the smrc-1 met-2 phenotype in parallel with met-2 and smrc-1 single mutants (Tables 1 and 2). At 25°C, met-2 and smrc-1 homozygotes remained fertile for numerous generations. In contrast, smrc-1 met-2 double mutant fertility dropped to near zero by the F2 generation. The clutch size of smrc-1 met-2 M+Z- F1 double mutants was similar to smrc-1 M+Z- F1 single mutants, but a lower proportion of offspring were viable (Table 1). Only 8% of the viable smrc-1 met-2 M-Z- offspring were fertile (Table 2), and they produced very few embryos, only ~6% of which were viable (Table 1). We also observed a protruding vulva (Pvl) phenotype in ~43% of smrc-1 met-2 M-Z- animals (Table 2) that may reflect DNA damage in the vulval precursor cells [77]. Overall, the smrc-1 met-2 phenotype is consistent with SMRC-1 and MET-2 acting redundantly to promote one or more essential germline process(es).
Given that MET-2 and SET-25 modify some common regions of the genome, we considered that SET-25 activity might contribute to SMRC-1-related processes. To address this question, we first investigated the impact of SET-25 loss on the smrc-1 developmental phenotype at 25°C. 100% of set-25 single mutants were fertile. ~8% of smrc-1 set-25 double mutants were sterile, similar to smrc-1 single mutants, and the two genotypes had similar developmental defects (Table 2, S2B Fig). We next investigated the impact of SET-25 loss on HU sensitivity. The response set-25 single mutants to HU resembled wild type and response of smrc-1 set-25 double mutants resembled smrc-1 single mutants (Fig 1). We conclude that SET-25 activity is not essential for protection from DNA replication stress, and loss of SET-25 activity does not impact the smrc-1 developmental phenotype.
The met-2(n4256) set-25(tm5021) double mutant was previously described as slow growing with substantial embryonic lethality, elevated HU sensitivity, and increased CEP-1-dependent germline apoptosis at 25°C [17]. met-2 set-25 double mutants also produce some abnormal oocytes and have elevated apoptosis [18]. We regenerated the met-2(n4256) set-25(tm5021) double mutant and grew it in parallel with smrc-1 met-2 and smrc-1 set-25 to compare germline development of animals grown together under the same conditions. At 25°C, embryonic lethality was very high in met-2 set-25 double mutants and all adult escapers were fertile, as reported (Table 2).
We also evaluated HU sensitivity in the met-2 smrc-1 set-25 triple mutant. Since sterility was high in the met-2 smrc-1 set-25 M-Z- F2 individuals, we analyzed the responsiveness of M+Z- F1 individuals to HU exposure (Fig 1B). At the highest doses of HU, met-2 smrc-1 set-25 M+Z- sensitivity was significantly elevated compared to either smrc-1 met-2 or smrc-1 set-25 M+Z- double mutants and the effect on fertility (i.e., the effect in the germ line) was particularly striking (Fig 1B). Hence, H3K9 methylation per se combined with SMRC-1 together provide substantial protection from replication stress.
Chromatin structure is carefully modulated to limit DNA damage and maintain genome integrity. Our analysis identifies a link between conserved components of the DNA repair and chromatin regulatory machineries in C. elegans: SMRC-1, a member of the SMARCAL1 annealing helicase family known to promote DNA repair during replication; and MET-2, a member of the SETDB1 histone methyltransferase family responsible for H3K9 methylation. By a variety of measures, we show that SMRC-1 promotes germline viability and limits DNA damage. Nuclear SMRC-1 abundance increases in the proliferative germ line under conditions of replication stress, and SMRC-1 promotes a concomitant increase in nuclear MET-2 accumulation. There is a small, but statistically significant increase in detectable H3K9me2 signal in these nuclei. When SMRC-1-deficient mutants are maintained long-term at elevated culture temperatures, H3K9me2 deposition becomes unregulated, and in most cases reduced. Fertility defects arise in these serially passaged smrc-1 mutants, which we hypothesize result from accumulated DNA damage over multiple generations due to chronic replication stress and reduced H3K9me2 deposition.
Two observations suggest that H3K9me2 limits the negative consequences of replication stress in the germ line: HU treatment causes an increase in nuclear MET-2 accumulation; and met-2 mutants are hypersensitive to HU with respect to fertility. In our assays, MET-2 limits replication stress independently of SET-25, suggesting H3K9 mono- and di-methyl marks are more important than trimethyl marks in this context. We also demonstrate a previously unappreciated role for MET-2 in stabilizing poly G/C tracts.
SMRC-1 promotes fertility, likely as a consequence of its roles in DNA repair and limiting DNA damage. We hypothesize that the increased sensitivity to replication stress and reduced ability to repair DNA lesions contribute to the smrc-1 developmental phenotypes. SMRC-1 is particularly important at elevated culture temperatures, and we note that the loss-of-function phenotypes of Drosophila Marcal1 and mouse SMARCAL are also more severe at elevated culture temperature [78]. SMRC-1 buffers against replication stress, limits R-loop accumulation, limits SPO-11-independent DSBs, promotes MET-2 accumulation in the nucleus under conditions of replication stress, and affects the distribution of meiotic crossovers. At stressful temperatures, SMRC-1 activity affects H3K9me2 accumulation throughout the germ line. The association with SMRC-1 may recruit MET-2 to the nucleus where they may function at the replication fork. SMARCAL1 family proteins are hypothesized to function outside of S phase to promote DSB repair [79, 80]. and SMRC-1 may recruit MET-2 to help stabilize chromatin for repair in this context.
Based on genetic data, SMRC-1 and MET-2 appear to both shared and distinct functions in the germ line. We hypothesize that SMRC-1 and MET-2 act together to limit germline sensitivity to replication stress. In contrast, the met-2 smrc-1 synthetic sterility may be the cumulative effect of severely reduced H3K9 methylation in combination with DNA damage beyond that at replication forks.
We consider two non-mutually exclusive models for how the MET-2 –SMRC-1 association may promote genome integrity. First, SMARCAL1 family proteins associate with ssDNA at the replication fork, hence SMRC-1 may be well-positioned to recruit MET-2 for re-establishment of H3K9me2 marks on nascent chromatin (Fig 10A). Reestablishing heterochromatin at repetitive sequences after DNA replication is important for maintaining genome stability [81–83]. Second, SMRC-1 may recruit MET-2 to DNA breaks, thereby stabilizing DNA and facilitating repair. The interaction could be important when breaks arise during replication and/or at another point in the cell cycle (Fig 10B). SETDB1 is recruited to DNA damage sites directly and specifically in mammalian systems and SETDB1 enrichment is essential for proper repair of the DNA lesions [84]. Consistent with this finding, Checchi et al. (2011) observed elevated germline apoptosis and sensitivity to cep-1 loss in animals treated with met-2 RNAi, which may indicate a role for MET-2 in mediating the DNA damage response [85]. At repetitive regions, MET-2-mediated H3K9me2 deposition may have an additional beneficial effect of reducing the DNA replication rate to assure the complete and accurate replication of error-prone repetitive sequences. In this scenario, MET-2 may function in a positive feed-back loop to attract more SMRC-1, thus reinforcing replication fidelity. It has been proposed that MET-2-mediated H3K9me2 deposition limits transcription of repetitive regions and thereby limits RNA:DNA hybrid formation at those sites [17]. Perhaps one way in which MET-2 limits RNA:DNA hybrid stability is by recruiting SMRC-1, which may have a role in resolving the hybrids.
Little is known about possible meiotic functions for SMARCAL1 family proteins as functional studies have only been performed in mitotic cells. Mammalian SMARCAL1 associates with the ssDNA binding protein, RPA, during DNA replication and catalyzes replication fork regression, ultimately promoting branch migration [38, 43, 86, 87]. Holliday junctions, which resemble replication forks, are present during meiotic recombination, and C. elegans RPA (RPA-1) is present in pachytene nuclei and promotes meiotic DSB repair [88, 89]. The meiotic recombination pattern observed in smrc-1 mutants may therefore have multiple underlying causes.
C. elegans meiotic DSBs are enriched on chromosomal arms where they inversely correlate with repetitive sequences and H3K9me2 enrichment [71, 90]. These data fit with observations from a number of species that DSBs–and therefore COs–tend not to occur at repetitive sequences, perhaps in part due to H3K9me2 [2]. Our RAD-51 labeling data and diakinesis chromosome analyses indicate that SMRC-1 protects the genome from aberrant DSBs and inaccurate repair. In smrc-1 mutants, RAD-51 foci persisted late into pachytene, consistent with delayed DSB repair in the absence of SMRC-1. In C. elegans, the process of CO homeostasis ensures that most DSBs are repaired via a non-crossover (NCO) mechanism and only one DSB per chromosome is resolved via CO [67]. Our mapping data indicate that SMRC-1 activity promotes CO homeostasis. One explanation for the loss of CO homeostasis in smrc-1 mutants may be that aberrant DSBs in the smrc-1 proliferative germ line are not subject to the same strict regulatory controls as SPO-11-induced breaks. An alternative hypothesis is that SMRC-1 activity limits CO frequency.
Human SMARCAL1 promotes DSB repair via non-homologous end joining (NHEJ) in cultured cells [80] and Drosophila Marcal1 mediates the synthesis-dependent strand annealing (SDSA) step in DSB DNA repair [79]. SMRC-1 activity may limit meiotic recombination by promoting NCO repair, perhaps by recruiting/stabilizing MET-2 at repetitive regions. The association between MET-2 and SMRC-1 could serve as a surveillance system to prevent DSB formation at repetitive regions, thus limiting the occurrence of CO at these sequences.
Syracuse University issued an IACUC number to E.M.M. for the custom anti-MET-2 antibody generation, which was performed by Yenzym Antibodies LLC. The Syracuse University IACUC number is #09 = 021.
C. elegans were maintained according to standard methods [91]. Details of nematode strains, mutant construction by CRISPR, and epitope tagging can be found in S1 Text.
Protein blots and immunohistochemistry were performed using standard methods. Detailed procedures, including antibodies used and quantification methods, can be found in S1 Text.
MET-2 IP was performed with nuclear extract prepared from him-8(e1489) adults. 3xFLAG::SMRC-1 IP was performed with whole extract from endogenously-tagged 3xflag::smrc-1 adults. Detailed procedures can be found in S1 Text.
Assays were carried put as previously described [92]. L1 larvae of different genotypes were treated with HU for a pulse of 16 hr at 25°C and then cultured using standard conditions. L4 larvae were treated with HU for 16 hr at room temperature (~22°C) until adulthood, and then immunolabeled. Detailed HU treatment protocols can be found in S1 Text.
We assayed suppression/reversion of the unc-58(e665) phenotype as described [59] in unc-58 control and smrc-1(ea8);unc-58 mutants raised at 20°C. To detect dog-1 enhancement, we assayed for deletions in vab-3 exon 5 as described [62]. Details are included in S1 Text.
Six lines of balanced smrc-1(ea8)/qC1 were maintained at 25°C for three generations and then expanded to 16 unbalanced founders. Strains were maintained by serial passaging as described in the S1 Text.
|
10.1371/journal.pntd.0003474 | Control of Trachoma in Australia: A Model Based Evaluation of Current Interventions | Australia is the only high-income country in which endemic trachoma persists. In response, the Australian Government has recently invested heavily towards the nationwide control of the disease.
A novel simulation model was developed to reflect the trachoma epidemic in Australian Aboriginal communities. The model, which incorporates demographic, migration, mixing, and biological heterogeneities, was used to evaluate recent intervention measures against counterfactual past scenarios, and also to assess the potential impact of a series of hypothesized future intervention measures relative to the current national strategy and intensity. The model simulations indicate that, under the current intervention strategy and intensity, the likelihood of controlling trachoma to less than 5% prevalence among 5–9 year-old children in hyperendemic communities by 2020 is 31% (19%–43%). By shifting intervention priorities such that large increases in the facial cleanliness of children are observed, this likelihood of controlling trachoma in hyperendemic communities is increased to 64% (53%–76%). The most effective intervention strategy incorporated large-scale antibiotic distribution programs whilst attaining ambitious yet feasible screening, treatment, facial cleanliness and housing construction targets. Accordingly, the estimated likelihood of controlling trachoma in these communities is increased to 86% (76%–95%).
Maintaining the current intervention strategy and intensity is unlikely to be sufficient to control trachoma across Australia by 2020. However, by shifting the intervention strategy and increasing intensity, the likelihood of controlling trachoma nationwide can be significantly increased.
| Australia is the only remaining high-income country reporting endemic levels of trachoma, with infections occurring predominantly within rural and remote Indigenous communities. Although the Australian government has recently invested large sums of money to combat the disease, it remains unclear whether the national goal of controlling trachoma by 2020 will be achieved. Here, we use a novel individual-based simulation model to estimate the impact of numerous potential future invention strategies and intensities. Our model is the most sophisticated trachoma transmission model to date, and the first to specifically represent trachoma in Australian Indigenous communities. Model projections suggest that although the current intervention strategy and intensity are unlikely to achieve the target of national control by 2020, the likelihood of achieving this goal can be significantly increased by shifting the intervention strategy and increasing the intensity of key intervention components such as screening, treatment and facial cleanliness activities. Our findings that the most resource rich country with endemic trachoma may require a more intensive intervention effort to control the disease suggest that challenges may remain in the fight for the global control and eventual elimination and eradication of trachoma.
| Australia is the only high-income country in which trachoma, the worldwide leading cause of preventable blindness [1], remains endemic [2]. In remote Aboriginal communities deemed to be at-risk of trachoma, an estimated 4% of adults suffer severely impaired vision or blindness [3] due to many years of repeated re-infection with the bacterium Chlamydia trachomatis—the infectious agent from which trachoma disease develops [4]. In 2009, the Australian government pledged AUS$16 million over an initial four-year period towards the national goal of controlling trachoma by 2020 [3]. That is, to reduce the prevalence of trachomatous inflammation follicular (TF) to less than 5% amongst 5–9 year-old children within a community. This target closely aligns with the Global Elimination of Trachoma by 2020 (GET 2020) initiative [5] developed by the World Health Organisation (WHO). The Australian trachoma intervention effort combines annual surveillance activities with a Surgery, Antibiotics, Facial cleanliness and Environmental improvement (SAFE) control policy recommended by the WHO [3,6]. This four-component policy incorporates treatment for those with clinically detected disease and long-term solutions for reducing infection incidence and disease prevalence [7]. The WHO offers recommendations for the frequency and intensity of the screening and treatment programs integrated into the SAFE policy [8]; however, the Australian intervention effort involves a greater intensity of screening and treatment due to larger resource availability compared to other trachoma-endemic countries. Despite this, the prevalence of trachoma remains high in many Aboriginal communities [3] whilst several developing countries prepare to announce the national control or eradication of the disease [9].
In this paper, we assess the progress of recent trachoma intervention efforts in Australia and evaluate the possibility of achieving national control by 2020. This is achieved by addressing the following three questions: (i) have past trachoma intervention efforts been effective in reducing infection incidence and disease prevalence? (ii) what epidemiological impact can be expected if the current intervention strategy and intensity is maintained until 2020? (iii) how can a shift in strategy or increase in intensity improve this impact? These questions are addressed through the development and analysis of a novel simulation model of trachoma transmission in remote Australia.
Previous models of trachoma transmission have typically implemented population-based methods [10–14]. These traditional models can be useful for extracting general principles but often lead to an over-simplification of disease dynamics [15]. Recent studies have indicated that transmission between two individuals is influenced by factors such as age, with children younger than 10 years being the typical reservoir of infection [12,16], and the presence of nasal or ocular discharge, i.e. a dirty face [17]. Demographic factors such as household overcrowding [10,11] and inter-community migration are also believed to contribute to trachoma persistence [18]. The temporary migration of individuals between communities is believed to be of particular importance in sustaining endemic trachoma in Australia [18]. Here, an individual-based simulation model is developed to incorporate these complexities. This is the most sophisticated trachoma transmission model to date, and the first model to specifically represent endemic trachoma in Australian Aboriginal communities. The parameters of the model are informed by the best available Australian and international data (see S1 Table).
The model developed for this study simulates a population of Aboriginal persons within a remote Australian region. Each individual represented in the model is a member of an at-risk community encompassed by the region, and is also a resident of a household within a community (Fig 1A). The temporary migration of individuals (and potentially other members of their household) is simulated based on rates of movement between communities [19].
The model is characterised by a five-state natural history structure (Fig 1B). Upon infection with C. trachomatis an individual enters a short latent period where infection load is such that the newly infected individual is not yet infectious, whilst active disease (trachomatous inflammation follicular/intense: TF/TI) has yet to develop [20]. An immunopathological response then develops and the infected individual progresses to an infectious state where clinical disease appears [4]. Following the clearance of infection, the inflammatory disease state resolves slowly in the absence of reinfection. Individuals in this state are partially immune to re-infection, but if re-challenged will experience prolonged disease [21]. In the event that no re-infection occurs during this episode, the individual fully recovers to the susceptible state. The duration of each infection and disease state is dependent upon exposure to repeated episodes of infection. Exposure to reinfection is assumed to decrease with age[16].
The transmission of infection and the subsequent development of disease are stochastically determined at the individual-level [22]. That is, the probability of infection transmission between an infectious individual and a susceptible individual is calculated and a random number generated to determine whether transmission will occur at the relevant time point. The probability of transmission between two individuals is assumed to be influenced by the ‘clean face’ status of both the susceptible individual and the infectious individual, where facial cleanliness is assumed to reduce the probability of transmitting and contracting infection [17]. Age-stratified community-level facial cleanliness prevalence data has been recorded across remote Australia as a process of trachoma screening events since 2007, and is directly entered into the model [23]. The probability of transmission is also affected by the infectiousness of the infected individual, assumed to be proportional to the bacterial load, which in-turn is assumed to be dependent upon the number of previous infections [12]. Two distinct settings for human interaction, and thus transmission potential, are considered in the model. The primary setting for transmission is the household, although transmission can also occur in the wider community. See S1 Text for further details and model equations.
The model described was independently calibrated to empirical age-stratified community-level disease prevalence data from three Australian regions through first-order Monte Carlo filtering methods [24]. See S1 Table and S1 File for model parameters and further details of the model and the calibration process. The model source code is also available on-line [25]. Each modelled community was classified as hyperendemic (≥ 20% active trachoma disease prevalence in 5–9 year olds), mesoendemic (≥ 10% but < 20%) or hypoendemic (≥ 5% but < 10%) dependent on the mean community disease prevalence observed from 2007 to 2011. The three modelled regions were selected to form a representative sample of trachoma-endemic remote Australia, with selection based on the endemicity of the communities within each region as well as the quantity and quality of surveillance data available. Throughout this paper, the simulated regions are de-identified and referred to as ‘predominantly hyperendemic’, ‘predominantly mesoendemic’ and ‘predominantly hypoendemic’ based on the prevailing endemicity of the communities within the region. Communities with a consistent 5–9 year old disease prevalence of less than 5% are considered not-at-risk, with 5% also considered the threshold for control [8]. The timing and age-stratified intensity of 2007–2011 screening and treatment events were directly entered into the model according to programmatic monitoring data.
The calibrated model was utilised to evaluate the impact of recent intervention efforts. This was achieved by simulating the model in the absence of past intervention efforts such as screening programs, treatment events, housing development initiatives and improvements in facial cleanliness prevalence. A direct comparison was made between the model calibrated to reflect observed conditions and the model output under the hypothetical scenario of no past intervention efforts. The model was then used to project the future impact of a series of potential intervention scenarios. A base-case future intervention scenario was compiled by extrapolating the trends from previously observed trachoma intervention events. The values obtained through this analysis are presented in S2 Table, whilst a description of the current National Guidelines for Trachoma Control in Australia are presented within S3 Table. This base case scenario was then analysed against a series of alternative intervention scenarios. The results obtained from a selection of these alternative scenarios, which are described in Table 1, are presented in this paper.
Each of the future intervention scenarios were simulated until 2020 and the community-level age-stratified prevalence of infection and disease were recorded in each modelled community. The likelihood of controlling trachoma was then calculated as the proportion of model simulations, for each community, in which the control criterion was satisfied by 2020. To produce representative outputs which accounted for the stochasticity of the model, 1,000 simulations were produced using 1,000 distinct parameter sets. These parameter sets were sampled from the realistic range of plausible parameter estimates obtained through the model calibration process (described in S1 File). These results were aggregated to form control likelihood estimates for communities of specific endemicity under a given intervention scenario. All numerical computation was performed using MATLAB [26].
Model-based evaluations of the interventions implemented between 2007 and 2011 suggest that disease prevalence has generally been reduced through trachoma intervention efforts. However, the scale of impact of the past intervention measures was found to vary between regions. The greatest reductions were observed in the predominantly hyperendemic regions, where trachoma prevalence among 5–9 year old children was estimated to have been 23.5% (mean from 1,000 simulations, with range 18.5%–30.7%) in 2011 in the absence of interventions compared with 14.3% (10.5%–18.5%) with interventions; in the predominantly mesoendemic region, trachoma prevalence was estimated to have reduced from 14.8% (10.3%–19.7%) to 5.8% (3.2%–8.0%) due to intervention efforts (Fig 2). However, the impact of intervention measures in the predominantly hypoendemic region is more modest: disease prevalence in 2011 was estimated to have reduced from 5.1% (2.2%–8.9%) to 4.3% (2.3%–6.5%) (Fig 2). This occurs despite a comparable, if not stronger, screening and treatment effort being observed in the hypoendemic region. Indeed, the mean 5–9 year old screening coverage from 2007 to 2011 in the predominantly mesoendemic and predominantly hypoendemic regions are 70.5% and 78.9%, respectively. The corresponding values for treatment coverage were 87.6% and 86.1%, respectively. A sensitivity analysis of model input parameters (see S1 File) suggests that this finding may be influenced by a higher baseline prevalence of child facial cleanliness in the predominantly hypoendemic region.
Future projection simulations estimate the likelihood of achieving trachoma control in hypoendemic and mesoendemic communities by 2020 to be 85% (77%–89%) and 70% (60%–79%), respectively, should current trachoma intervention efforts be maintained (Fig 3). However, the likelihood of satisfying the trachoma control criteria in hyperendemic communities under this scenario was calculated to be only 31% (19%–43%). The estimated likelihoods of controlling trachoma in hypoendemic communities by 2020 were found to be consistently high across each of the considered intervention scenarios. However, large differences were found in control likelihoods in the mesoendemic and, in particular, hyperendemic communities across the modelled future scenarios. This suggests that by optimising the intervention strategy and intensity, the likelihood of achieving trachoma control in highly endemic communities can be greatly increased.
Increasing housing construction, and therefore easing the burden of household overcrowding, in addition to maintaining the current intervention strategy increased the estimated likelihood of achieving trachoma control in hyperendemic communities from 31% (19%–43%) to 38% (30%–47%). Alternatively, assuming a 1.5-2-fold reduction in both infectiousness and susceptibility due to facial cleanliness[18], achieving a consistently high facial cleanliness prevalence (90%) amongst children was found to increase the likelihood of controlling trachoma from these worst-affected communities by 2020 to 64% (53%–76%). By additionally attaining consistently large screening and treatment coverages, this likelihood of control was estimated to be 75% (63%–85%). The epidemiological effect of this combination of interventions was found to be greater than the sum of the individual interventions, suggesting that a synergistic effect exists between screening, treatment and attaining high levels of facial cleanliness amongst children. The greatest likelihood for achieving control in the worst-affected communities occurred when these intervention intensities were further coupled with alterations in the treatment strategy; by introducing bi-annual mass drug administration (MDA) in hyperendemic communities, the likelihood of achieving trachoma control increased to 86% (76%–95%), with a corresponding control likelihood of 96% (92%–100%) in mesoendemic communities. This most-effective future scenario was projected out until 2030, and no rebounding of the epidemic was observed.
The resources required to achieve control is an important consideration. Here, crude estimations of the resources required across the predominantly hyperendemic region were calculated for each scenario by the total number of people receiving treatment (Fig 4). An estimated total of 26,088 antibiotic doses are to be distributed between 2012 and 2020 under the current intervention strategy and intensity. This value compares with 12,855 treatments under the scenario in which ambitious yet feasible child facial cleanliness prevalence targets were also consistently satisfied. The significantly smaller number of treatments reflects the lower incidence rates attained when a substantially larger proportion of the young population had clean faces. Under the future scenario where ambitious yet feasible screening, treatment and facial cleanliness targets were consistently met, 15,312 antibiotic doses would be distributed. Despite this slight increase in antibiotic distribution compared with the previous scenario, the large increase in control likelihood that can be attained by implementing such a control policy and intensity, particularly in hyperendemic communities, makes a solid case for implementing such a control effort. A bi-annual MDA program would result in 25,989 antibiotic doses distributed. A large proportion of this total would be distributed within the first three years of implementing the strategy; however, the treatment effort required following this initial peak would decline over time to be less than that required under the current control strategy (Fig 4). The MDA strategy, with substantially greater control likelihood, emphasises the ‘hit hard, hit early’ principle for greatest effectiveness and cost-effectiveness.
Australia is the only high-income country to have endemic trachoma. Whilst being a signatory to the WHO’s GET 2020 initiative, the Australian government has responded to the health issue in recent years with large investment. However, as several developing countries with histories of trachoma prepare to announce the national control or eradication of the disease [9], high prevalence levels of trachoma are still observed in remote Australian Aboriginal communities. Since 2006, Australia has implemented national surveillance activities to collect age-segregated community-level data describing the timing, frequency and intensity of screening and treatment programs as well as disease prevalence, facial cleanliness prevalence, and more recently environmental conditions that may affect trachoma incidence and persistence [3,23,27]. Increases in community screening and treatment, along with recorded increases in facial cleanliness among children has correlated with declines in trachoma prevalence in Australia.
Although our model estimates that current strategies and intensities of programs are unlikely to lead to national control, alternate scenarios appear to be feasible and effective means of achieving this goal. Our results suggest that to achieve control in the worst-affected communities, a more intense intervention effort may be required. For hypoendemic communities (prevalence 5–10%), the model output indicates that continuing the current intervention strategy and intensity will likely be sufficient to control trachoma by 2020. By assuming that facial hygiene programs can reduce transmission potential over 2-fold, our model estimates that a substantial increase in community facial cleanliness prevalence may be sufficient to control trachoma by 2020 in mesoendemic communities (prevalence 10–20%). The model results indicate that annual screening events are appropriate in mesoendemic communities, whilst the long-term gain of implementing mass drug administration (MDA) as opposed to treating only the household contacts of index cases was found to have a negligible impact. The model predicts that the future intervention effort required to considerably raise the likelihood of achieving control in hyperendemic communities (prevalence >20%) by 2020 is more difficult. An increase in screening, treatment and facial cleanliness prevalence should be combined with an enhanced housing construction program. Continuous bi-annual MDA is also recommended for three consecutive years before resuming screening events to significantly raise the likelihood of controlling trachoma in these most endemic communities. However, it should also be noted that it may be possible for active disease amongst children to reach below 5% by 2020 even if all trachoma-specific interventions were discontinued immediately. Such decreases are not unusual in far less wealthy regions of the world where programmatic coverage has been poor. But without specific interventions in the past, the rates of trachoma have changed very little in these communities over many years. Thus, we believe concerted and targeted approaches—informed by this analysis—will increase the chance of trachoma control.
Previous models of trachoma transmission have described the natural history of trachoma infection and disease using simple population-based systems of differential equations [10–14]. Although useful for extracting general principles, population-based models are limited by the level of complexity that they are able to incorporate. In the context of trachoma control in Australia, the range of complexities one can consider—such as an individual’s age, facial cleanliness status, usual residence and the environmental state of their home community—lends itself to a more flexible model with finer granularity. As such, this study is based on a detailed individual-based simulation model, informed by and calibrated to relatively large amounts of data for remote Australian Aboriginal communities experiencing trachoma. The model developed here attempts to accurately represent the natural history of trachoma infection and disease whilst also assimilating the demographic, cultural and biological factors that influence ocular C. trachomatis transmission. Whilst no model may ever be sophisticated enough to capture all of the heterogeneities involved in the transmission of an infectious disease, the usefulness of the output hinges on the optimality of the model’s balance between complexity and accuracy [22]. Extensive collaboration was sought to ensure that the model described in this paper achieved such a balance. However, it is imperative that the results presented must be consumed with perspective.
Other limitations that should be addressed when assessing the validity of modelling results regard the empirical data that are used to inform the model parameters. For the purpose of this research, a large volume of nationally collated surveillance data was employed but these data are also potentially limited in completeness and representativeness. There exists a certain degree of uncertainty in the estimates of trachoma prevalence, particularly in the early years of data collection, as there was some degree of variation in screening coverage rates. However these coverages have progressively improved along with the accuracy of trachoma grading. Australian treatment coverage data can be difficult to interpret as the method of distribution has varied and has not always been clearly specified; these methods have included MDA, household contact based treatment and treating only affected children. Equally, the definition of the denominator and hence the coverage achieved have also varied somewhat. The model also assumes a steady-state equilibrium at baseline, which may be inaccurate as a result of previous trachoma treatment efforts. Although these are important limitations and have an impact on the precision of the forward estimates of effectiveness of the interventions, they do not influence the comparisons of the relative effectiveness of the different strategies. This is a strength of this research as it shows the greater effectiveness of a combined more intensive strategy compared with that currently employed.
The world has a goal of eliminating blinding trachoma as a public health concern. The countries which have done so or are on the verge of announcing such success should be commended for their excellent public health efforts. However, current strategies may not be sufficient in other contexts and as we have demonstrated in this study, they may be insufficient in the trachoma-endemic country with the greatest amount of resources. Through detailed simulation modelling we have suggested some slight shifts in strategies and changes in intensities in the short-term which have the potential to yield substantial returns in the future in order to achieve this ultimate goal.
|
10.1371/journal.pgen.1003992 | Sumoylated NHR-25/NR5A Regulates Cell Fate during C. elegans Vulval Development | Individual metazoan transcription factors (TFs) regulate distinct sets of genes depending on cell type and developmental or physiological context. The precise mechanisms by which regulatory information from ligands, genomic sequence elements, co-factors, and post-translational modifications are integrated by TFs remain challenging questions. Here, we examine how a single regulatory input, sumoylation, differentially modulates the activity of a conserved C. elegans nuclear hormone receptor, NHR-25, in different cell types. Through a combination of yeast two-hybrid analysis and in vitro biochemistry we identified the single C. elegans SUMO (SMO-1) as an NHR-25 interacting protein, and showed that NHR-25 is sumoylated on at least four lysines. Some of the sumoylation acceptor sites are in common with those of the NHR-25 mammalian orthologs SF-1 and LRH-1, demonstrating that sumoylation has been strongly conserved within the NR5A family. We showed that NHR-25 bound canonical SF-1 binding sequences to regulate transcription, and that NHR-25 activity was enhanced in vivo upon loss of sumoylation. Knockdown of smo-1 mimicked NHR-25 overexpression with respect to maintenance of the 3° cell fate in vulval precursor cells (VPCs) during development. Importantly, however, overexpression of unsumoylatable alleles of NHR-25 revealed that NHR-25 sumoylation is critical for maintaining 3° cell fate. Moreover, SUMO also conferred formation of a developmental time-dependent NHR-25 concentration gradient across the VPCs. That is, accumulation of GFP-tagged NHR-25 was uniform across VPCs at the beginning of development, but as cells began dividing, a smo-1-dependent NHR-25 gradient formed with highest levels in 1° fated VPCs, intermediate levels in 2° fated VPCs, and low levels in 3° fated VPCs. We conclude that sumoylation operates at multiple levels to affect NHR-25 activity in a highly coordinated spatial and temporal manner.
| Animals precisely control when and where genes are expressed; failure to do so can cause severe developmental defects and pathology. Transcription factors must display extraordinary functional flexibility, controlling very different sets of genes in different cell and tissue types. To do so, they integrate information from signaling pathways, chromatin, and cofactors to ensure that the correct ensemble of genes is orchestrated in any given context. The number of regulatory inputs, and the complex physiology and large numbers of cell and tissue types in most experimentally tractable metazoans have rendered combinatorial regulation of transcription nearly impenetrable. We used the powerful genetics and simple biology of the model nematode, C. elegans, to examine how a single post-translational modification (sumoylation) affected the activity of a conserved TF (NHR-25) in different cell types during animal development. Our work suggests that sumoylation constrains NHR-25 activity in order to maintain proper cell fate during development of the reproductive organ.
| Tissue-specific and cell type-specific transcriptional networks underlie virtually every aspect of metazoan development and homeostasis. Single TFs, operating within gene-specific regulatory complexes, govern distinct gene regulatory networks in different cells and tissues; thus, combinatorial regulation underpins tissue- and cell type-specific transcription. Determining the precise mechanisms whereby such specificity arises and how networks nevertheless remain flexible in responding to environmental and physiological fluctuations is an interesting challenge. TFs integrate signaling information from co-factors, chromatin, post-translational modifications, and, in the case of nuclear hormone receptors, small molecule ligands, to establish transcription networks of remarkable complexity.
Here, we approach this problem by studying a covalent modification of a nuclear hormone receptor (NHR) in C. elegans, a simple metazoan with powerful genetic tools, a compact genome, and an invariant cell lineage leading to well-defined tissues. NHRs are DNA-binding TFs characterized by a zinc-finger DNA binding domain (DBD) and a structurally conserved ligand binding domain (LBD) [1]. The genome of C. elegans encodes 284 NHRs while humans only have 48 NHRs [1]. Of the 284 NHRs, 269 evolved from an HNF4α-like gene [2], and 15 have clear orthologs in other species. NHR-25 is the single C. elegans ortholog of vertebrate SF-1/NR5A1and LRH-1/NR5A2, and arthropod Ftz-F1 and fulfills many criteria for the study of tissue-specific transcriptional networks [1]. NHR-25 is broadly expressed in embryos and in epithelial cells throughout development [3], [4]. It is involved in a range of biological functions such as molting [3]–[5], heterochrony [6], and organogenesis [7]. Furthermore, both NHR-25 and its vertebrate orthologs regulate similar processes. SF-1 and NHR-25 promote gonadal development and fertility [8], [9], while NHR-25 and LRH-1 both play roles in embryonic development and fat metabolism [4], [10]–[12]. The pleiotropic phenotypes seen following RNAi or mutation of nhr-25 highlight the broad roles of the receptor, and its genetic interaction with numerous signaling pathways (β-catenin, Hox, heterochronic network) [6]–[8] make it an excellent model to study combinatorial gene regulation by NHRs.
SUMO (small ubiquitin-like modifier) proteins serve as post-translational modifiers and are related to but distinct from ubiquitin [13]; we show here that NHR-25 is sumoylated. Sumoylation uses similar enzymology as ubiquitination to conjugate the SUMO protein onto substrate lysines [13]. Briefly, SUMO is produced as an inactive precursor. A SUMO protease activates SUMO by cleaving residues off the C-terminus to expose a di-glycine [13]. A heterodimeric E1 protein consisting of UBA2 and AOS1 forms a thioester bond with the exposed diglycine and then transfers SUMO to an E2 enzyme (UBC9), also through a thioester bond [14]. The E2 enzyme then either directly conjugates SUMO onto a target lysine, or an E3 ligase can enhance the rate of sumoylation; that is, unlike in ubiquitination, E3 ligases are not always required. Like many post-translational modifications, sumoylation is reversible and highly dynamic. The same SUMO protease that initially activated SUMO cleaves the isopeptide linkage that covalently attaches SUMO to the target protein [14]. Indeed, global failure to remove SUMO from substrates compromises viability in mice and S. pombe [15], [16].
The extent of sumoylation of a given target can be regulated by varying the expression, localization, stability or activity of components of the sumoylation machinery in response to external and internal cellular cues [14]. SUMO-regulated processes include nuclear-cytosolic transport, DNA repair, transcriptional regulation, chromosome segregation and many others [14]. For example, sumoylation of the glucocorticoid receptor prevents synergy between two GR dimers bound at a single response element [17]. In this sense, SUMO is analogous to the small hydrophobic hormones and metabolites that serve as noncovalent ligands for nuclear receptors, except it associates both covalently and non-covalently with its targets. Sumoylation modulates the activities of multiple classes of cellular proteins, such as transcriptional regulators, DNA replication factors and chromatin modifiers.
Elucidating how a single nematode NHR integrates cellular signals to regulate specific genes in distinct tissues will advance our understanding of metazoan transcription networks. To this end, we examined how sumoylation regulates the C. elegans nuclear hormone receptor NHR-25, and the physiological relevance of this nuclear hormone receptor-SUMO interaction. Using a combination of genetics, cell biology, and in vitro biochemistry we sought to understand how signaling through sumoylation impacts NHR-25's role in animal development, and how sumoylation affects the NHR-25 transcriptional network.
We identified an interaction between NHR-25 and the single C. elegans SUMO homolog (SMO-1) in a genome-wide Y2H screen using the normalized AD-Orfeome library, which contains 11,984 of the predicted 20,800 C. elegans open reading frames [18]. SMO-1 was the strongest interactor in the screen on the basis of two selection criteria, staining for β-galactosidase activity and growth on media containing 3-aminotriazole (Figure 1A). To assess the selectivity of the SMO-1–NHR-25 interaction, we tested pairwise combinations of SMO-1 with full-length NHR-25, an NHR-25 isoform β that lacks the DNA-binding domain, and each of seven additional NHRs: NHR-2, NHR-10, NHR-31, NHR-91, NHR-105, FAX-1, and ODR-1 (Figure S1A). The NHR-25-SMO-1 interaction proved to be selective, as SMO-1 failed to bind the other NHRs tested. NHR-25 also interacted with the GCNF homolog, NHR-91 (Figure S1A).
SMO-1 was an enticing NHR-25 interacting partner to pursue. SUMO in C. elegans and other eukaryotes regulates TFs and chromatin, thus is well positioned to impact NHR-25 gene regulatory networks. Furthermore, spatial and temporal expression patterns of smo-1 and nhr-25 during development largely overlap [3], [4], [19]. SUMO interacts with the mammalian homologs of NHR-25, suggesting that the interaction is likely evolutionarily conserved [20], [21]. Among its many phenotypes, smo-1 loss-of-function (lf) mutants display a fully penetrant protruding vulva (Pvl) phenotype, reflecting disconnection of the vulva from the uterus [19] (Figure 1B, C). smo-1 RNAi or mutation also cause low penetrance of ectopic induction of vulval cells, which can generate non-functional vulval-like structures known as multivulva (Muv) [22] (Figure 1B, C). Similar to smo-1 mutants, nhr-25 reduction-of-function leads to a Pvl phenotype, but does not cause Muv [7]. This nhr-25 Pvl phenotype results from defects in cell cycle progression, aberrant division axes of 1° and 2° cell lineages, and altered vulval cell migration (Table 1, Figure 2, Bojanala et al., manuscript in preparation). Because at an earlier stage NHR-25 is also necessary for establishing the anchor cell (AC) [8], which secretes the EGF signal that initiates vulval precursor cell (VPC) patterning, our RNAi treatments were timed to allow AC formation and examination of the effect of nhr-25 depletion on later developmental events.
When smo-1 and nhr-25 were simultaneously inactivated, animals exhibited a fully penetrant vulvaless (Vul) phenotype and an exacerbated Muv phenotype (Figure 1B, C). The ectopically induced vulval cells expressed an egl-17::YFP reporter, indicating that 3°-fated cells aberrantly adopted 1° and 2° fates in these animals (Figure S2B). This egl-17::YFP reporter allowed us to monitor 1°/2° fate induction despite the cell division arrest phenotypes of nhr-25(RNAi) and smo-1(lf);nhr-25(RNAi) animals. Lineage analyses showed that following simultaneous inactivation of both smo-1 and nhr-25, daughters of all VPCs normally responsible for vulva formation, (P5.p, P6.p and P7.p) failed to undergo the third round of vulval cell division (Table 1) resulting in premature cell division arrest and the Vul phenotype. Although P5.p, P6.p and P7.p VPCs were induced, the execution of 2° fate was abnormal: in both smo-1(ok359) and smo-1(ok359);nhr-25(RNAi) backgrounds, the expression of the 1° marker, egl-17::YFP exhibited ectopically high expression in P5.p and/or P7.p (Figure S2A) at the 4-cell stage. Moreover, in smo-1;nhr-25(RNAi) animals, the P(3,4,8).p cell, which normally divides only once and fuses into the hypodermal syncytium, kept dividing (Table 1). This continued division enhanced the Muv induction phenotype seen in smo-1 mutants. Thus, reduction of SMO-1 activity enhanced cell division defects in 1° and 2° nhr-25 mutant VPCs, while reduction of NHR-25 activity enhanced the smo-1 mutant Muv phenotype in 3° fated cells.
NHR-25 and SMO-1 interact physically in Y2H assays and genetically in vivo, consistent with their overlapping expression patterns [4], [19]. Furthermore, the mammalian NHR-25 homologs are sumoylated, suggesting that SMO-1-NHR-25 interactions are conserved and physiologically important. Y2H interactions with SUMO can reflect non-covalent binding, or covalent sumoylation where the SUMO protein is coupled onto the substrate through an isopeptide bond. These two possibilities can be distinguished genetically. Mutations in the β-sheet of SUMO interfere with non-covalent binding, whereas deletion of the terminal di-glycine in SUMO selectively compromises covalent sumoylation [23]. As can be seen in Figure 3A, deletion of the terminal di-glycine residues of SMO-1 (ΔGG) completely abrogated the interaction with NHR-25. The SMO-1 V31K mutation predicted to disrupt the conserved β-sheet of SMO-1 hampered the Y2H interaction between NHR-25 and SMO-1, although not as severely as the SMO-1 ΔGG mutation (Figure 3A). These findings are similar to those with DNA thymine glycosylase and the Daxx transcriptional corepressor, both of which bind SUMO non-covalently and are also sumoylated [24], [25]. The V31K β-sheet mutant was competent to bind the C. elegans SUMO E2 enzyme, UBC-9, confirming its correct folding (Figure S3A). Together, these results suggested that NHR-25 is both sumoylated and binds SMO-1 non-covalently; conceivably, the two modes of interaction confer distinct regulatory outcomes.
As our Y2H data suggested that NHR-25 was sumoylated, we identified candidate sumoylation sites within NHR-25 using the SUMOsp2.0 prediction program [26]. The sumoylation consensus motif is ψ-K-X-D/E, where ψ is any hydrophobic amino acid, K is the lysine conjugated to SUMO, X is any amino acid, and D or E is an acidic residue [14]. Three high scoring sites reside in the hinge region of the protein: two are proximal to the DBD (K165 and K170) and one (K236) is near the LBD (Figure 3C). We mutated these sites, conservatively converting the putative SUMO acceptor lysine residues to arginine to block sumoylation. Single mutation of any of the three candidate lysines had no apparent effect on the NHR-25 interaction with SMO-1 in Y2H assays, whereas the three double mutants had modest effects, and the NHR-25 3KR triple mutant (K165R K170R K236R) abrogated binding (Figure 3D). A fourth candidate sumoylation site (K84) located in the DBD was completely dispensable for the Y2H interaction (data not shown). To verify that the 3KR mutations blocked the interaction with SMO-1 specifically, rather than causing NHR-25 misfolding or degradation, we confirmed that NHR-25 3KR retained the capacity to bind NHR-91 (Figures S1, Figure 3B). These data suggested that either non-covalent binding is dispensable for the SMO-1-NHR-25 interaction and that this was a rare case in which the SUMO β-sheet mutation impaired sumoylation, or that the three lysines in NHR-25 were important for both the covalent and non-covalent interaction with SMO-1.
To ensure that our Y2H results indeed reflected NHR-25 sumoylation, we turned to in vitro sumoylation assays. As both human and C. elegans sumoylation enzymes were used in these experiments, we distinguish them with prefixes “h” and “Ce”. As a positive control, we expressed and purified recombinant hE1, hUBC9, hSUMO1, and hSENP1 from E. coli. We also purified a recombinant partial hinge-LBD fragment of mouse SF-1 from E. coli; this fragment contains a single sumoylation site in the hinge region. SF-1 is a vertebrate ortholog of NHR-25 and the fragment that we used is a robust sumoylation substrate (Figure S4A) [27]. We then purified an N-terminally hexahistidine-Maltose Binding Protein (6×His-MBP) tagged fragment of NHR-25 (amino acids 161–541) containing most of the hinge region and ligand-binding domain, including all three candidate SUMO acceptor lysines. Coomassie staining and immunoblotting revealed three slower-migrating species, which were collapsed by the addition of the SUMO protease, hSENP1 (Figure 4A, S5A). We detected sumoylation of the same 6×HisMBP-NHR-25 fragment when it was expressed in rabbit reticulocyte lysates, followed by incubation with hE1, hE2 and hSUMO1 (Figure 4B).
We further tested NHR-25 substrates containing two (2KR; K170R K236R) or three arginine substitutions (NHR-25 3KR). When only one predicted acceptor lysine was available (2KR), we detected a single dominant sumoylated species, whereas for NHR-25 3KR, sumoylation was abrogated (Figure S5B). We performed sumoylation reactions on in vitro transcribed and translated wild type NHR-25, NHR-25 3KR, and NHR-25 3EA. In NHR-25 3EA (E167A E172A E238A) the acidic glutamic acid residues within the three consensus sumoylation sites were mutated to alanine. NHR-25 3EA leaves the acceptor lysines available, but is predicted to inhibit sumoylation by impairing interaction with UBC9. While wild type NHR-25 was clearly sumoylated, the 3EA mutation severely impaired sumoylation (Figure 4B).
When sumoylation reaction times were extended 5–20 fold, additional species of sumoylated NHR-25 were generated (Figure S6A). These species could reflect sumoylation of NHR-25 on other sites or formation of hSUMO1 chains. To distinguish between these possibilities, we used methyl-hSUMO1, which can be conjugated onto a substrate lysine, but chain formation is blocked by methylation. Long incubations with methyl-hSUMO1 resulted in only three sumoylated NHR-25 bands, as determined by NHR-25 immunoblotting, indicating that there are indeed only the three major acceptor lysines (Figure S6A). hSUMO2, which readily forms polySUMO chains, was included as a control in this experiment. Even with extended incubation times, we observed only three dominant sumoylated forms of NHR-25, suggesting that additional bands in reactions using hSUMO1 or CeSMO-1 reflect inefficient chaining. We conclude that NHR-25 is sumoylated in vitro on three lysines and that C. elegans SMO-1 does not readily form polySUMO chains, unlike yeast SMT3 and mammalian SUMO2.
All studies of C. elegans sumoylation to date have used hE1, hUBC9, and hSUMO proteins [19], [28], [29]. We purified recombinant CeE1, CeUBC-9 and CeSMO-1 from E. coli and tested their activity in in vitro sumoylation assays. Our CeE1 preparation was inactive, but was effectively substituted by hE1. Under those conditions, our CeUBC-9 and CeSMO-1 catalyzed sumoylation of the SF-1 hinge-LBD fragment (Figure S4B). Similar to hUBC9 and hSUMO1, recombinant CeUBC-9 and CeSMO-1 yielded three sumoylated species using the 6×His-MBP-NHR-25 substrate (Figure 4C, S4C).
To determine the kinetics of the three SUMO modifications of NHR-25, we performed a time course of standard sumoylation reactions with hUBC9/CeUBC-9 and hSUMO1/CeSMO-1 proteins. In both cases, we detected a single band by 15 minutes, followed by two and then three sumoylated species as the reaction progressed (Figure S6B–E). These data imply that the three sumoylation sites are modified sequentially, in a particular order.
All of our reactions were performed without addition of an E3 ligase. The high efficiency of SF-1 sumoylation in the absence of E3 ligase is in part due to a direct interaction with UBC9 [30]. Surprisingly, we failed to detect an interaction between NHR-25 and CeUBC-9 either by Y2H assays or through immunoprecipitation of purified proteins (Figure S3B; data not shown). However, when we performed a yeast three-hybrid assay, where untagged CeSMO-1 was added to the system, we observed a weak interaction between NHR-25 and CeUBC-9, suggesting either that CeSMO-1 bridges NHR-25 and CeUBC-9 or that NHR-25 recognizes a CeSMO-1-bound CeUBC-9 species (Figure S3B).
To begin to investigate how sumoylation affects NHR-25-dependent transcriptional activity, we employed a HEK293T cell-based assay. We used a luciferase reporter driven by four tandem Ftz-F1 (Drosophila homolog of NHR-25) consensus sites, previously shown to be responsive to NHR-25 [8]. When Myc-tagged wild type NHR-25 was transfected, reporter expression was enhanced (Figure 5A), and the sumoylation-defective mutant NHR-25 (3KR) activated the reporter more strongly (Figure 5A). Anti-Myc immunostaining indicated no detectable increase in protein level or nuclear localization (Figure 5B).
To better characterize NHR-25-dependent transcriptional activity and generate reporters that could subsequently be used for in vivo assays, we generated a construct based on the canonical, high affinity SF-1 regulatory elements derived from the Mullerian inhibiting substance (MIS) and CYP11A1 (CYP) genes. We assessed NHR-25 binding to these elements using yeast one-hybrid (Y1H) and electrophoretic mobility shift assays (EMSAs). The Y1H assays indicated that NHR-25 bound the MIS and CYP11A1 elements (Figure S7A, B). Mutations in the MIS binding site that block SF-1 binding (MIS MUT) [27] prevented NHR-25 binding (Figure S7B). Moreover, the NHR-25 L32F (ku217) mutant, which has impaired DNA binding in vitro [7], displayed reduced activity in the Y1H experiment (Figure S7B). Consistent with the Y1H data, we found that a 6×His-MBP tagged fragment of NHR-25 (amino acids 1–173) purified from E. coli clearly bound MIS and CYP11A1 sites singly (Figure S7C) or in combination (2×NR5RE WT, for nuclear receptor NR5 family Response Element; Figure S7D) but only weakly to the mutant sites (Figure S7C–D, 7A).
Sumoylation of SF-1 regulates binding to specific DNA sequences [27]. Therefore, we asked whether sumoylation could similarly affect DNA binding capacity of the 6×His-MBP tagged fragment of NHR-25. We found that this fragment, which encompasses the DBD and part of the hinge region of NHR-25 (amino acids 1–173), was an even more potent sumoylation substrate than the hinge-LBD fragment, as almost all of the DBD substrate could be sumoylated (Figure 6A). Unlike SF-1 [27], NHR-25 DNA binding did not inhibit sumoylation (data not shown). Use of methyl-hSUMO1 in our in vitro sumoylation assays indicated that there were three sumoylation sites within the 6×His-MBP tagged fragment of NHR-25 DBD substrate (Figure 6B). These corresponded to the hinge region K165 and K170 acceptor lysines, which are analogous to the SF-1 fragment used by Campbell et al. (2008), and a third SUMO acceptor lysine (K84) within the DBD region between the second zinc finger and the conserved Ftz-F1 box (Figure 6C). This acceptor lysine is conserved in D. melanogaster Ftz-F1 as well as the mammalian LRH-1 (Figure 6C) [31]. EMSAs indicated that sumoylation diminished binding of the NHR-25 DBD fragment to the MIS and CYP derived binding sites (Figure S7D). Modifying the EMSAs such that the sumoylation reaction preceded incubation with the 2×NR5RE oligos severely impaired binding (Figure S7E). These in vitro findings are consistent with the notion that, as in mammals, sumoylation could diminish NHR-25 DNA binding.
We next wanted to assess the effects of sumoylation on NHR-25-dependent transcription in vivo. To enhance the sensitivity of our assays, we constructed a reporter carrying four tandem repeats derived from each of MIS and CYP genes (Figure 7A, eight SF-1/NHR-25 binding sites designated as 8×NR5RE). The binding sites were spaced ten base-pairs apart to facilitate potential cooperative binding [32]. We generated transgenic C. elegans carrying the 8×NR5RE positioned upstream of a pes-10 minimal promoter and driving a 3×Venus fluorophore bearing an N-terminal nuclear localization signal. In wild type animals, reporter expression was not detected (Figure 7B), whereas after smo-1 RNAi, strong expression was detected in developing vulval cells, the hypodermis, seam cells, the anchor cell (Figure 7B) and embryos (not shown), tissues in which NHR-25 is known to be expressed (Figures 7F) and functional [4], [7], [33]. Reporter expression was especially prominent during the L3 and L4 stages. Mutation of the binding consensus, 8×NR5RE(MUT) abolished reporter expression in a smo-1 (RNAi) background (Figure 7E), as expected for NHR-25-dependent reporter expression. Moreover, genetic inactivation of nhr-25 either by RNAi (smo-1, nhr-25 double RNAi) or by use of nhr-25(ku217), a reduction-of-function allele of nhr-25, abrogated reporter expression even in smo-1 knockdown animals (Figure 7C, D). We conclude that sumoylation of NHR-25 strongly reduces its transcriptional activity in vivo.
To examine functionally the consequences of NHR-25 sumoylation, we returned to the roles of nhr-25 and smo-1 in vulval organogenesis. Noting that smo-1 mutants but not nhr-25 reduction-of-function mutants display a Muv phenotype, we investigated whether this might reflect enhanced NHR-25 activity due to its reduced sumoylation. We therefore generated transgenic animals expressing tissue-specific NHR-25 and/or SMO-1 driven by three different promoters; egl-17 for the VPCs, grl-21 for the hypodermal hyp7 syncytium, and wrt-2 for the seam cells. These transgenes included (i) wild type NHR-25; (ii) NHR-25 3KR; or (iii) SMO-1 alone. Although egl-17 is typically used as a 1° and 2° cell fate marker during vulva development, it is expressed in all VPCs in earlier stages [34](Figure S2C). We used the egl-17 promoter rather than commonly used VPC driver, lin-31, because the heterodimeric partner of LIN-31 is sumoylated and directly involved in vulva development [28].
Muv induction was scored by observing cell divisions of the six VPCs with the potential to respond to the LIN-3/EGF signal, which promotes differentiation. Normally, only P5.p, P6.p, and P7.p are induced while P3.p, P4.p and P8.p each produce no more than two cells as they are destined to fuse with the surrounding hyp7 syncytium (Figure 2). In wild type animals, overexpression of NHR-25 in the VPCs (egl-17 promoter) but not in hyp7 or seam cells (grl-21 and wrt-2 promoters, respectively) drove Muv induction at the P8.p position, mimicking smo-1 RNAi (Figure 8, Table S1). Thus, high level NHR-25 acted cell-autonomously to produce a Muv phenotype. Overexpression of the NHR-25 3KR mutant in the VPCs resulted in an even more penetrant Muv phenotype and greater induction of P3.p, P4.p, and P8.p (Figure 8A). In contrast, overexpression of SMO-1 alone did not produce the Muv phenotype.
These overexpression experiments implied that excess unsumoylated NHR-25 altered 3° VPC fate, permitting extra divisions that produce the Muv phenotype. If sumoylation of NHR-25 normally constrains its activity, animals with decreased sumoylation activity would be expected to enhance the Muv phenotype. To test this hypothesis, we assessed the effect of smo-1 RNAi in animals expressing a low-copy, integrated transgene expressing C-terminally GFP-tagged NHR-25 [35]. This transgene likely recapitulates the expression pattern of endogenous nhr-25, since the construct includes the complete 20 kb intergenic region upstream of nhr-25, and the entire nhr-25 gene and 3′-UTR; the animals display normal vulvas. However, exposure to smo-1 RNAi caused the Muv phenotype in about 30% of animals carrying the nhr-25::gfp transgene, which exceeded the 12% Muv frequency in smo-1 RNAi controls (Figure 8). This extra vulva induction was seen in the P4.p. lineage in addition to P8.p. Together, our findings strongly suggest that in wild type animals, NHR-25 sumoylation prevents ectopic vulva induction in 3° fated cells.
One interpretation of our genetic and biochemical data is that the in vivo ratio of sumoylated to non-sumoylated NHR-25 specifies or maintains the 3° VPC fate. We were therefore interested in how NHR-25 sumoylation was regulated. SMO-1 is expressed at constant levels throughout vulval development [19], so we examined whether NHR-25 levels were regulated in VPCs during development. The low-copy, integrated NHR-25::GFP translational fusion allowed us to examine the developmental pattern of NHR-25 expression. NHR-25::GFP was evenly distributed prior to the first division in all VPCs, whereas after the first division the pattern became graded: highest in 1° P6.p daughters, lower in 2° P5.p and P7.p daughters, and lowest in 3° P(3,4,8).px (Figure 9A, B). After the third round of cell divisions NHR-25::GFP expression continued in all 22 P(5–7).pxxx cells and remained high during early vulva morphogenesis (Figure 9D) until it temporarily disappeared by the “Christmas tree stage” (data not shown).
smo-1 RNAi caused ectopic NHR-25::GFP expression in P(4,8).pxx cells (Figure 9E), which displayed the strongest Muv induction in NHR-25::GFP;smo-1(RNAi), and Pegl-17::NHR-25(3KR) backgrounds (Figure 8). In wild type animals, NHR-25::GFP was normally expressed in the anchor cell at the time of the first VPC divisions, and subsequently decreased (Figure 9D). Interestingly, we noted that in nine of ten smo-1(RNAi) animals NHR-25::GFP was re-expressed in the AC at the “bell stage” (Figure 9F). Subsequently, no AC invasion occurred and the AC remained unfused. Therefore, in addition to restricting NHR-25 activity in 3° cells (previous section), sumoylation also limits NHR-25 accumulation in cells that are destined to assume the 3° fate. The resultant NHR-25 gradient combined with constant levels of SMO-1 may account for the observed pattern of NHR-25 sumoylation.
The capacity of TFs to specify expression of precise networks of genes in a given context, yet remain flexible to govern dramatically different sets of genes in different cell or physiologic contexts, likely involves combinatorial regulation of transcription. In this study, we show that sumoylation represses bulk NHR-25 activity in multiple C. elegans tissues. In addition, our findings suggest that particular fractional sumoylation states of NHR-25 govern the appropriate course of cell divisions and the 3° fate decision of vulval precursor cells, thereby determining morphogenesis of the entire organ.
Supporting the notion that sumoylation can constrain NHR-25 activity, we found that a reporter fusion responsive to NHR-25 was strongly upregulated upon depletion of smo-1 by RNAi (Figure 7B). Our in vitro findings suggested that sumoylation of NHR-25 diminished DNA binding (Figure S7), while our in vivo studies suggested that reduction of smo-1 caused ectopic accumulation of NHR-25 (either synthesis or impaired degradation) in VPCs P4.p and P8.p (Figure 9). These data suggest two modes, not mutually exclusive, through which sumoylation can regulate NHR-25. Moreover, overexpression of either NHR-25 or its sumoylation-defective form (NHR-25 3KR) led to multivulva induction in cells that normally adopt the 3° fate (Figure 8).
Together, our data support a model in which proper differentiation of VPCs depends on the appropriate balance of sumoylated and unsumoylated NHR-25 (Figure 10). Importantly, NHR-25 affects VPC specification cell-autonomously, as overexpression of NHR-25 in other epidermal cells, such as the seam cells or hyp7, did not cause a Muv phenotype (Table S1). Furthermore, NHR-25 appears to form a gradient across the VPC array, accumulating to high levels in 1° fated cells, intermediate levels in 2° fated cells and low levels in 3° fated cells (Figure 9). Our findings indicate that sumoylation promotes a specific pattern of NHR-25 activity in differentially fated VPCs and the relative level of NHR-25 sumoylation is critical for promotion and/or maintenance of the 3° cell fate (Figure 10).
The role(s) of NHR-25 and SMO-1 in vulval induction are likely pleiotropic. Multiple vulval development factors are sumoylated [22], [28], [29], [36], including LIN-11, which is responsible in part for promoting vulval-uterine fusion [19]. Based on expression pattern and phenotypes, NHR-25 likely acts in other cell-types (hyp7, 1°/2° VPCs, or AC) and at different developmental time points to regulate vulval induction. The Muv phenotype of smo-1-deficient animals was enhanced by nhr-25 RNAi (Figure 1). Synthetic multivulva (synMuv) genes inhibit lin-3 activity in the syncytial hyp7 cell to prevent aberrant vulva induction in the neighboring 3° cells [37]. Yet, overexpression of NHR-25 in the hyp7 syncytium did not cause Muv induction (Table S1), thus it is unlikely that NHR-25 acts through this pathway. Our overexpression data indicates that NHR-25 acts cell-autonomously in the VPCs (Figure 8), and likely interacts with canonical signaling pathways that promote VPC fate. The NHR-25 expression gradient is reminiscent of the LIN-3/EGF gradient which promotes vulval induction through Ras activation and subsequent Notch signaling [38]. nhr-25 appears to act downstream of LET-60/Ras signaling, as gain-of-function LET-60/Ras causes elevated NHR-25 expression (data not shown). However, regulation of lin-3 by NHR-25 in the anchor cell has also been suggested [39]. Ectopic expression of NHR-25 in the AC following smo-1 RNAi is unlikely to cause Muv induction since, developmentally, this expression occurs much later than VPC fate determination. In wild type animals, NHR-25 levels are therefore downregulated in the AC, which may be required for proper completion of AC invasion and/or fusion. Additionally, the cell division arrest seen in nhr-25 RNAi leading to the Pvl phenotype was enhanced by inactivation of smo-1 (Figure 1). For instance, the Pvl phenotype can arise from nhr-25 reduction of function, which causes defective 1° and 2° cell divisions (Figure 1, Table 1), or from smo-1(lf), which impairs uterine-vulval connections [19]. Thus, an exquisite interplay between various sumoylated targets as well as the balance between sumoylated and unsumoylated NHR-25 collaborate to ensure proper vulval formation.
How could unsumo∶sumo NHR-25 balance regulate 3° cell fate? Sumoylation might alter NHR-25 levels or activity in a manner that shifts the unsumo∶sumo NHR-25 ratio, which in turn acts as a switch to determine NHR-25 output. The activities of a mammalian nuclear hormone receptor have been shown to shift dramatically with signal-driven changes in levels of receptor activity [40]. Another possibility is that the sumoylated and unsumoylated versions of NHR-25 regulate distinct targets, and the unsumo∶sumo ratio in different cells thereby determines the network of NHR-25-regulated genes. Indeed, sumoylation appears to affect the genomic occupancy of the NHR-25 ortholog SF-1 [27]. We note that NHR-25 sumoylation could be context-dependent. Sumoylation could increase NHR-25 activity at particular response elements. Accordingly, sumoylation positive regulates the activity of the nuclear hormone receptors RORα and ER [41], [42].
The finding that overexpression of NHR-25 strongly provoked a Muv phenotype suggests that sumoylation state of NHR-25 in VPCs is exquisitely regulated. Such regulation might be accomplished by subtle changes in availability of SUMO in different VPCs, not detected by our assays, or by the relative activities of the sumoylation machinery and the SUMO proteases. A similar competition for constant levels of SUMO regulates Epstein-Barr virus infections, where the viral BZLF protein competes with the host PML protein for limiting amounts of SUMO1 [43].
It is intriguing to consider SMO-1 as an NHR-25 ligand parallel to hormones or metabolites bound noncovalently nuclear hormone receptors in other metazoans, and by the C. elegans DAF-12 receptor. Indeed, such expansion of the concept of signaling ligands could “de-orphan” many or all of the 283 C. elegans nuclear hormone receptors for which no traditional ligands have been identified. Detection of noncovalent ligands is very challenging; numerous mammalian NHRs remain “orphans” despite intensive efforts to find candidate ligands and evidence that the ancestral NHR was liganded [44]. In principle, SUMO can be conjugated to its target sequence motif anywhere on the surface of any protein, whereas classic NHR ligands bind only stereotyped pockets within cognate NHR LBDs. Viewed in this way, SUMO may directly regulate many NHRs (and other factors as well), whereas classical NHR ligands act more selectively on only one or a few NHRs. The multifactorial regulation of NHRs would provide ample opportunity for gene-, cell- or temporal-specificity to be established in cooperation with the SUMO ligand.
There are three ways in which SUMO can potentially interact with target proteins: i) non-covalent binding, where a protein binds either free SUMO or SUMO conjugated onto another protein; ii) sumoylation, where SUMO associates covalently with a target protein through an isopeptide linkage; and iii) poly-sumoylation, where chains of SUMO are built up from an initially monosumoylated substrate. In C. elegans, SMO-1 can bind proteins non-covalently [45] or can be covalently linked to substrates (Figure 4). Polysumoylation occurs through SUMO modification of acceptor lysines within SUMO proteins [46]. In our assays, we saw no robust polyCeSMO-1 chains compared to the hSUMO2 control, even after prolonged reaction times (Figure S6). Consistent with this result, sumoylation motifs were predicted within hSUMO1, 2 and 3, and yeast SMT3 but not in CeSMO-1. PolySUMO chains in yeast and vertebrates can be recognized by SUMO targeted ubiquitin ligases (STUbLs) that polyubquitinate the polySUMO chain and direct it for degradation by the 26S proteasome [46]. Judging from BLAST analysis, there are no evident homologs of the known STUbLs hsRNF4 or yeast SLX5–8 in C. elegans. As both S. cerevisiae SUMO (SMT3) and vertebrate SUMO2 and SUMO3 form polySUMO chains, it appears that C. elegans has lost the ability to form polySUMO chains.
The mammalian homologs of NHR-25 (SF-1 and LRH-1) are sumoylated on two sites within the hinge region of the protein, between the DBD and LBD [21], [27], [47]. These SUMO acceptor sites occur at corresponding positions in NHR-25, with the site near the DBD being duplicated (Figure 6C). Additionally, our DBD sumoylation experiments suggest the presence of a fourth sumoylation site in NHR-25, conserved with D. melanogaster Ftz-F1 and mammalian LRH-1 (Figure 6C) [7], [31]. Thus, NHR-25 appears to have sumoylation sites that are conserved in both SF-1 and LRH-1 as well as at least one site that is only conserved in LRH-1. Similarly, NHR-25 seems to combine regulation of processes that in mammals are either regulated by SF-1 only or LRH-1 only. Additionally, human SUMO1 can be conjugated onto NHR-25 and C. elegans SMO-1 can be conjugated onto SF-1 (Figure 4, S4). Therefore, despite the 600–1200 million years of divergence since the common ancestor of humans and nematodes, regulation of NR5A family by sumoylation appears to be incredibly ancient. There are also, however, notable differences. For instance, while LRH-1 and SF-1 strongly interact with UBC9, providing a mechanism for robust, E3 ligase-independent sumoylation [20], this did not appear to be the case for NHR-25. As indicated above, we also did not find evidence for polysumoylation of NHR-25.
Having established SUMO as an NHR-25 signal that regulates cell fate, it will be exciting to further explore how sumoylation affects the NHR-25 gene regulatory network. It will be essential in future work to identify direct NHR-25 target genes by ChIP-seq, to determine how sumoylation impacts NHR-25 response element occupancy, and to mutate sumoylation sites and response elements with genome editing technologies, such as CRISPR/Cas9 [48]. The compact C. elegans genome facilitates unambiguous assignment of putative response elements to regulated genes, a daunting challenge in vertebrate systems. Further, the extensive gene expression and phenotypic data accessible to the C. elegans community will allow identification of candidate NHR-25 target genes directly responsible for regulating animal development and physiology. Understanding how NHR-25 sumoylation regulates specific genes, and how this information is integrated into developmental circuits will advance our understanding of combinatorial regulation in metazoan gene regulatory networks.
cDNAs and promoters/binding sites were Gateway cloned (Invitrogen) into pDONR221 and pDONR-P4P1r, respectively. Mutations were introduced into the nhr-25 cDNA using site-directed mutagenesis with oligonucleotides carrying the mutation of interest and Phusion polymerase (NEB). cDNAs and promoters were then moved by Gateway cloning into destination vectors. NHR-25 (amino acids 161–541) and NHR-25 (amino acids 1–173) were moved into the bacterial expression vector pETG-41A, which contains an N-terminal 6×His-MBP tag. CeUBC-9 and CeSMO-1 cDNAs were moved into the bacterial expression vector pETG-10A, which contains an N-terminal 6×His tag. The CeUBC-9 construct also carried an N-terminal tobacco etch virus (TEV) cleavage site for removal of the 6×His tag, similar to the hUBC9 bacterial expression construct. For Y1H experiments, 2×SF-1 binding sites were Gateway cloned into pMW2 and pMW3 [49]. For Y2H experiments, cDNAs were moved into pAD-dest and pDB-dest [18], which contain the Gal4 activation domain and DNA binding domain, respectively. For Y3H, smo-1 was moved into pAG416-GPD-ccdB-HA [50], which results in constitutive expression. For luciferase experiments, cDNAs were moved into pDEST-CMV-Myc. For our C. elegans expression experiments, cDNA constructs were Gateway cloned into pKA921 along with either the egl-17, wrt-2, or grl-21 promoter. The egl-17 promoter was PCR cloned from N2 genomic DNA. The wrt-2 and grl-21 promoters (pKA279 and pKA416, respectively) were previously cloned [12]. pKA921 contains a polycistronic mCherry cassette to allow monitoring of construct expression. For our 3×Venus reporters, three-fragment Gateway cloning into pCFJ150 [51] was performed. The 8×NR5RE-pes-10Δ promoter fragments were cloned into pDONR-P4P1r. C. elegans codon optimized 3×Venus was cloned from Prnr::CYB-1DesBox::3×Venus [52] and an NLS was added on the 5′ end of the gene and NLS-3×Venus was Gateway cloned into pDONR221. The unc-54 3′-UTR in pDONR-P2rP3 was a gift from the Lehner lab. Primer sequences are provided in Table S2. Plasmids generated for this study are listed in Table S3.
Yeast transformations and Y2H assays were carried out as described by Deplancke et al. [53]. For the Y2H screen, S. cerevisiae strain MaV103 carrying a pDB-nhr-25 construct was transformed with 100 ng of the AD-Orfeome cDNA library, in which 58% of the known C. elegans open reading frames are fused to the Gal4 activation domain [18]. Six transformations were performed per screen and 149,800 interactions were screened, representing 12.5-fold coverage of the library. Positive interactions were selected for by growth on SC dropout plates lacking leucine, tryptophan, and histidine; these plates were supplemented with 20 mM of the histidine analog 3-aminotriazole. Interactions were confirmed by β-galactosidase staining. We identified 42 candidate interactors, but only smo-1 was recovered multiple times (seven independent isolations). Moreover, upon cloning and retesting the candidate interactor cDNAs, only smo-1 was confirmed as an interactor. The screen identified no other components of the SUMO machinery or known SUMO binding proteins. Generation of Y1H bait strains and Y1H analyses were performed as described [53]. pDB constructs carrying NHR-2, NHR-10, NHR-31, NHR-91, NHR-105, FAX-1, and ODR-1 cDNAs were a gift from Marian Walhout.
Recombinant hE1, hUBC9, hSUMO1, hSUMO2, hSENP1, and murine SF-1 LBD were purified as described [27], [54]–[56]. 6×His-CeSMO-1 and 6×His-TEV-CeUBC-9 were expressed in BL21(λDE3) E. coli and purified using a similar scheme as used to purify their human counterparts [55], [56]. 6×His-MBP-NHR-25 (amino acids 161–541) was freshly transformed into BL21(λDE3) E. coli. A 1 L culture was grown to an OD600 of ∼0.8, induced with 0.2 mM isopropylthio-β-galactoside (IPTG), and shaken at 16°C for four hours. Bacteria were lysed using a microfluidizer in 20 mM Tris-HCl pH 8.0, 350 mM NaCl, 20 mM imidazole containing EDTA-free Protease Inhibitor Cocktail III (EMD Millipore). 6×His-MBP-NHR-25 was then purified using nickel affinity chromatography (5 ml His Trap FF column, GE Healthcare). Peak fractions were pooled, dialyzed into 20 mM HEPES (pH 7.5), 1 mM EDTA, and 2 mM CHAPS {3-[(3-cholamidopropyl)-dimethylammonio]-1-propanesulfonate}, and purified by anion-exchange chromatography using a MonoQ column (GE Healthcare) and eluted with a 1 M ammonium acetate gradient. Peak fractions were pooled, concentrated and 6×His-MBP-NHR-25 was purified by size-exclusion chromatography using an S200 column (GE Healthcare). Peak fractions containing 6×His-MBP-NHR-25 were pooled, concentrated, dialyzed into 20 mM Tris pH 7.5, 50 mM NaCl, 10% glycerol, flash frozen in liquid nitrogen, and stored at −80°C. Later purifications used only nickel affinity chromatography. Using this preparation in sumoylation assays produced results similar to those obtained using the preparations purified over the three aforementioned columns. 6×His-MBP-NHR-25 (amino acids 1–173) was expressed and purified using a single nickel affinity chromatography step, as described above for the 6×His-MBP-NHR-25 (amino acids 161–541) fragment.
Reactions were performed as described by Campbell et al. [27]. Briefly, 50 µl sumoylation reactions were set up with 0.1 µM E1, 10 µM UBC9, and 30 µM SUMO in a buffer containing 50 mM Tris-HCl (pH 8.0), 100 mM NaCl, 10 mM MgCl2, 10 mM ATP, and 2 mM DTT. Substrates were added at 1 µM and when required, 2.5 µg of hSENP1 SUMO protease was added. When in vitro transcribed proteins were used as substrates, 50 µl reactions were generated using a TnT T7 Quick Coupled Transcription/Translation System (Promega). 16 µl of this reaction was then used as a substrate in a 25 µl sumoylation reaction using the same molarities as described above. When SUMO protease was required, 1.25 µg of hSENP1 was added. Reactions were incubated at 37°C for the desired time, and stopped by boiling in protein sample buffer (10% Glycerol, 60 mM Tris/HCl pH 6.8, 2% SDS, 0.01% bromophenol blue, 1.25% beta-mercaptoethanol). Proteins were resolved by SDS-PAGE on either 4–12% Bis-Tris gradient gels (Invitrogen) or 3–8% Tris acetate gels (Invitrogen) followed by either Coomassie staining or immunoblotting. For immunoblotting, anti-NHR-25, anti-guinea pig-HRP (Santa Cruz), and anti-guinea pig-IR800 (Li-Cor) antibodies were used. Blots were developed using a LAS500 imager (GE Healthcare) or an Odyssey laser scanner (Li-Cor).
Reactions were performed as described by Campbell et al. [27] with the following alterations. We added 400 µg/ml of bovine serum albumin to the EMSA buffer (50 mM Tris (pH 8.0), 150 mM NaCl, 10 mM MgCl2, 10 mM DTT, 10 mM ATP, and a 1 µM concentration of double-stranded oligonucleotide). Sequences of oligonucleotides are provided in Table S2. Oligonucleotides were annealed and then centrifuged in an Amicon Ultra 0.5 ml centrifugal filter (MWCO 50). Sumoylation reactions were set up on ice and added directly to the annealed oligonucleotides (20 µl final volume). Standard reactions used 500 nM of unmodified NHR-25 substrate, titration experiments added NHR-25 in 100 nM increments from 200–700 nM. At this point SENP1 (0.5 µl) was added when appropriate. We incubated these reactions at room temperature for 30 minutes to allow both sumoylation and DNA binding to occur. Half of the EMSA reaction (10 µl) was removed and added to 2 µl of 4× protein sample buffer and denatured by boiling for five minutes. Sumoylation products in the input were analyzed by immunoblotting using anti-MBP (NEB) and anti-mouse-IR800 (LiCor) antibodies. Blots were imaged using an Odyssey laser scanner. The remaining EMSA reaction was resolved on a 4–20% TBE polyacrylamide gel (Invitrogen) at 200 volts and stained with 1× SYBR Gold (Molecular Probes) in 0.5× TBE. Gels were then imaged using a Typhoon laser scanner (GE Healthcare).
C. elegans was cultured at 20°C according to standard protocols and the wild type strain is the N2 Bristol strain [57]. The following mutant and transgenic strains were used in this study: PS3972 unc-119(ed4) syIs90 [egl-17::YFP+unc-119(+)], OP33 unc119 (ed3); wgIs33 [nhr-25::TY1::EGFP::3×FLAG(92C12)+unc-119(+)], VC186 smo-1(ok359)/szT1[lon-2(e678)]; +/szT1, MH1955 nhr-25(ku217). The following transgenic strains were generated for this study: HL102 jmEx102[Pegl-17::Myc::NHR-25_mCherry+rol-6(su1006)], HL107, HL108, HL110 are independent lines carrying jmEx107[Pegl-17::Myc::NHR-25(3KR)_mCherry+rol-6(su1006)], HL117 jmEx118 [Pegl-17::Myc::SMO-1_mCherry+rol-6(su1006)], HL111 and HL112 are independent lines carrying jmEx111[Pgrl-21::Myc::NHR-25_mCherry+rol-6(su1006)], HL121 jmEx121[Pgrl-21::Myc::SMO-1_mCherry+rol-6(su1006)], HL113 and HL114 are independent lines carrying jmEx113[Pwrt-2::Myc::NHR-25_mCherry+rol-6(su1006)], HL115 and HL116 are independent lines carrying jmEx115[Pwrt-2::Myc::SMO-1_mCherry+rol-6(su1006)], HL153 jmEx153[8×NR5RE (WT):pes-10Δ:NLS-3×Venus:unc-54 3′-UTR+Pmyo-2::tdTomato], HL155 jmEx155[8×NR5RE (MUT):pes-10Δ:NLS-3×Venus::unc-54 3′-UTR+Pmyo-2::tdTomato], HL170 nhr-25(ku217); jmEx153.
The following Gateway-based constructs were generated in pKA921: pJW522[Pegl-17(1914 bp)::Myc::NHR-25_polycistronic_mCherry], pJW774 [Pegl-17(1914 bp)::Myc:: NHR-25(3KR)_polycistronic_mCherry], pJW773 [Pegl-17(1914 bp)::Myc::SMO-1_polycistronic_mCherry], pJW526 [Pgrl-21(746 bp)::Myc::NHR-25_polycistronic_mCherry], pJW775 [Pgrl-21(746 bp)::Myc::SMO-1_polycistronic_mCherry], pJW524[Pwrt-2(1380 bp)::Myc::NHR-25_polycistronic_mCherry], pJW776[Pwrt-2(1380 bp)::Myc::SMO-1_polycistronic_mCherry]. The following Gateway-based constructs were generated in pCFJ150 [51]: pJW1109 [8×NR5RE(WT):pes-10Δ:NLS-3×Venus:unc-54 3′-UTR] and pJW1110 [8×NR5RE(MUT):pes-10Δ:NLS-3×Venus::unc-54 3′-UTR]. Plasmids were prepared using a PureYield Plasmid Midiprep System (Promega) followed by ethanol precipitation, or a Qiagen Plasmid Midi kit (Qiagen). Transgenic strains were generated by injecting 50 ng/µl of each plasmid into the C. elegans gonad [58] with the co-injection marker pRF4 [59]. For 8×NR5RE reporter strain generation, N2 animals were injected with 30 ng/µl of the reporter plasmid and 5 ng/µl of co-injection marker Pmyo-2::tdTomato [60].
Feeding RNAi was performed as described, with the indicated alterations to the protocol [61]. dsRNA was initially induced for four hours in liquid culture using 0.4 mM IPTG, before bacteria were concentrated and seeded on plates also containing 0.4 mM IPTG. Bacteria carrying pPD129.36 without an insert were used for control RNAi. For nhr-25 RNAi, synchronized L2 larvae (19–20 hours after hatching) were fed on bacteria expressing nhr-25 dsRNA to bypass the anchor cell (AC) defect. smo-1 RNAi was performed on late L4 or young adults. For in vivo reporter assays, sodium hypochlorite-treated eggs were placed on RNAi plates seeded with dsRNA induced bacteria.
To score vulva induction, nematodes were anesthetized in 10 mM levamisole, mounted onto 5% agar pads (Noble agar, Difco) and the number of daughter cells for each VPC were counted under differential interference contrast (DIC) optics. For lineaging analyses, the division pattern was followed under DIC from the two to eight cell stages [62]. Animals were mounted onto 5% agar pad with bacteria in S-basal medium without anesthesia. Olympus Fluoview FV1000 and Zeiss Axioplan microscopes were used for observation and imaging.
A peptide-based anti-NHR-25 antibody was raised in guinea pig (Peptide Specialty Laboratories, GmbH, Germany). Animals were immunized against four short peptides in the hinge and LBD regions: PEHQVSSSTTDQNNQINYFDQTKC (24 a.a. 141–163); SLHDYPTYTSNTTNC (15 a.a. 250–263); TSSTTTGRMTEASSC (15 a.a. 283–296) RYLWNLHSNXPTNWEC (16 a.a. 507–521).
Human embryonic kidney (HEK) cell line 293T was maintained in Dulbecco's modified Eagle's medium (DMEM, Gibco), supplemented with 10% fetal bovine serum. Transfections were performed with polyethyleneimine (25 kDa, Sigma). The transcriptional activity of NHR-25 was tested with a luciferase vector carrying a CMV basic promoter driven by two copies of the Ftz-F1 binding consensus sequences TGAAGGTCA and TCAAGGTCA (total of four binding sites, 2×TGA-TCA::Luc) [8], [63]. Cells were seeded onto 24-well plates and the next day were transfected for three hours with a polyethylenimine mixture containing 50 ng of pTK-Renilla plasmid (Promega) as an internal control, 300 ng of the luciferase reporter plasmid, and 150 ng of the appropriate expression vector. The total amount of DNA was kept constant (1 µg) by adding empty expression vector where necessary. Forty hours post-transfection, the cells were harvested and processed using the Dual Luciferase Reporter Assay System (Promega). Eight independent biological replicates from three independent experiments were assayed, and data were presented as average values with standard deviations after normalization against the Renilla luciferase activities. For immunocytochemistry, transfected cells were fixed with 4% formaldehyde (Sigma) for 10 min. After washing with PBS, cells were permeabilized with PBS containing 0.2% TritonX-100 in (PBST), washed with TBST buffer (25 mM Tris-HCl, pH 7.5, 136 mM NaCl, 2.7 mM KCl and 0.1% TritonX-100), incubated in blocking solution (2.5% skim milk and 2.5% BSA in TBST). Anti-Myc 9E10 antibody (Sigma; 1∶2000 dilution) was added and incubated for overnight at 4°C. Following washing, goat-anti-mouse-TRITC conjugated 2° antibody (Jackson ImmunoResearch; 1∶2000 dilution) was added and incubated at room temperature for two hours. Cells were counterstained with DAPI (1 µg/ml) to visualize the nucleus.
|
10.1371/journal.pgen.1004134 | High Risk Population Isolate Reveals Low Frequency Variants Predisposing to Intracranial Aneurysms | 3% of the population develops saccular intracranial aneurysms (sIAs), a complex trait, with a sporadic and a familial form. Subarachnoid hemorrhage from sIA (sIA-SAH) is a devastating form of stroke. Certain rare genetic variants are enriched in the Finns, a population isolate with a small founder population and bottleneck events. As the sIA-SAH incidence in Finland is >2× increased, such variants may associate with sIA in the Finnish population. We tested 9.4 million variants for association in 760 Finnish sIA patients (enriched for familial sIA), and in 2,513 matched controls with case-control status and with the number of sIAs. The most promising loci (p<5E-6) were replicated in 858 Finnish sIA patients and 4,048 controls. The frequencies and effect sizes of the replicated variants were compared to a continental European population using 717 Dutch cases and 3,004 controls. We discovered four new high-risk loci with low frequency lead variants. Three were associated with the case-control status: 2q23.3 (MAF 2.1%, OR 1.89, p 1.42×10-9); 5q31.3 (MAF 2.7%, OR 1.66, p 3.17×10-8); 6q24.2 (MAF 2.6%, OR 1.87, p 1.87×10-11) and one with the number of sIAs: 7p22.1 (MAF 3.3%, RR 1.59, p 6.08×-9). Two of the associations (5q31.3, 6q24.2) replicated in the Dutch sample. The 7p22.1 locus was strongly differentiated; the lead variant was more frequent in Finland (4.6%) than in the Netherlands (0.3%). Additionally, we replicated a previously inconclusive locus on 2q33.1 in all samples tested (OR 1.27, p 1.87×10-12). The five loci explain 2.1% of the sIA heritability in Finland, and may relate to, but not explain, the increased incidence of sIA-SAH in Finland. This study illustrates the utility of population isolates, familial enrichment, dense genotype imputation and alternate phenotyping in search for variants associated with complex diseases.
| Genome-wide association studies (GWAS) have been extensively used to identify common genetic variants associated with complex diseases. As common genetic variants have explained only a small fraction of the heritability of most complex diseases, there is a growing interest in the role of how low frequency and rare variants contribute to the susceptibility. Low frequency variants are more often specific to populations of distinct ancestries. Saccular intracranial aneurysms (sIA) are balloon-like dilatations in the arteries on the surface of the brain. The rupture of sIA causes life-threatening intracranial bleeding. sIA is a complex disease, which is known to sometimes run in families. Here, we utilize the recent advancements in knowledge of genetic variation in different populations to examine the role of low-frequency variants in sIA disease in the isolated population of Finland where sIA related strokes are more common than in most other populations. By studying >8000 Finns we identify four low-frequency variants associated with the sIA disease. We also show that the association of two of the variants are seen in other European populations as well. Our findings demonstrate that multiple study designs are needed to uncover more comprehensively their genetic background, including population isolates.
| About 3% of the population develops saccular intracranial aneurysms (sIAs) during life [1], [2]. Some 95% of subarachnoid hemorrhages are caused by ruptured sIA (sIA-SAH), a devastating form of stroke affecting individuals mainly in the sixth decade of life [3]. The annual incidence of SAH is 4–9 per 100 000 worldwide [4] but over twice as high in Finland and in Japan [5]. The sIA disease is a complex trait, the risk of which is affected by age, sex, smoking, hypertension, excess drinking [6], and in over 10% of the cases family history of sIA disease [7]–[9].
To date, genome wide association (GWA) studies have identified six definite and one probable loci with common variants associated to sIA: 4q31.23 (OR 1.22) [10], [11]; 8q11.23–q12.1 (OR 1.28); 9p21.3 (OR 1.31); 10q24.32 (OR 1.29); 12q22 (OR 1.16) [10]; 13q13.1 (OR 1.20); 18q11.2 (OR 1.22) [12] (Table S5). These seven loci were estimated to explain 6.1%, 4.4% and 4.1% of the four-fold sibling recurrence risk in Finland, Europe and Japan respectively [10]. In these previous GWA studies, results on 2q33.1 locus were inconsistent: the locus was significant in the first GWAS [13], not significant in the enlarged follow-up GWAS [12], and in the third GWAS the risk allele was reversed in the Japanese replication sample [10].
The population of Finland is one of the most thoroughly characterized genetic isolates. Due to the small size of the founder population, subsequent bottleneck effects and genetic drift, the Finnish population is enriched for rare and low frequency variants that are almost absent in other European populations and some variants rare elsewhere are increased in frequency [14]. This is best illustrated by the increased prevalence of 36 rare Mendelian, mostly recessive, disorders in Finland (www.findis.org); the so called Finnish disease heritage (FDH) [15]. We hypothesized that some of the enriched rare or low frequency variants could contribute to the increased sIA-SAH susceptibility in Finland.
In this GWA study we combined the power of 1000 Genomes imputation, the special benefits of a population isolate and enrichment of familial cases in the discovery cohort. Familial sIA patients more often carry multiple sIAs as compared to sporadic sIA patients, which may confer additional genetic burden to the sIA formation [8], [16], [17]. Therefore, in addition to the case vs. control analysis, we also analyzed the number of sIAs per individual as an intermediate phenotype. We conducted an association analysis in a discovery sample of 760 Finnish sIA cases and 2,513 matched controls followed by replication in an additional sample of 858 Finnish sIA cases and 4048 controls. The successfully replicated loci in Finland were further studied in a Dutch cohort of 717 sIA cases and 3004 controls to assess the extent to which the allele frequencies and risk effect sizes match between the isolate of Finland and a continental European population (Figure 1). In addition, we hypothesized that a previously inconclusive locus on 2q33.1 [10], [13], [18] is a true sIA risk locus at least in Finland and aimed to replicate the best discovery associations in the locus in this study in the Finnish and in the Dutch samples.
We successfully identified associations with low frequency variants in three novel loci in the case vs. control analysis and one in the aneurysm count analysis. Two of the case vs. control loci replicated also in the Dutch cohort with similar allele frequencies and comparable risk effect sizes. The variant in the aneurysm count locus demonstrated a strong bottleneck effect by being 15 times more frequent in the Finnish than in the Dutch controls. We also successfully replicated the previously inconclusive 2q33.1 locus.
To increase the potential genetic load in the study sample, our discovery sample consisted of 760 cases from the isolated, high-risk Finnish population, purposefully enriched for familial sIA (40%) patients and 2513 genetically matched Finnish controls. The imputation of the 304,399 previously genotyped variants [12] against the 1000 Genomes Project reference panel (v3, March 2012 release) increased the number of common and low frequency variants available for the association analysis to 9,359,231. Quantile-quantile (QQ) plots of association p-values did not indicate substantial inflation (λ = 1.04) (Figure S1). The discovery association analysis revealed one locus at 12p11.1 driven by rs653464 at conventional genome-wide significance (p<5×10−8) and 14 other loci at p<5×10−6 (Table S1; Manhattan plot in Figure S3).
We chose 17 SNPs representing the 15 promising loci (p<5×10−6) above for replication in an independent sample of 858 Finnish sIA cases and 4,048 controls (Table 1). Four SNPs and one deletion were associated at p<0.05 with the sIA disease (Table S1), two of them in the previously reported sIA loci 9p21.3 (rs1333042; OR 1.3, p = 6.3×10−7) and 13q13.1 (rs113124623; OR 0.88, p = 0.01). The genome-wide significant 12p11.1 locus in the discovery sample did not replicate (p = 0.29).
In the meta-analysis of the two Finnish samples, four SNPs reached the commonly used level of genome-wide significance at p<5×10−8 (Table 2). Three were novel: 2q23.3 (rs74972714; OR 2.1, 95% CI 1.68–2.63, p = 7.4×10−11, control allele frequency or CAF 2.35%), 5q31.3 (rs113816216; OR 1.92, CI 1.53–2.40, p = 1.74×10−8, CAF 2.09%) and 6q24.2 (rs75018213; OR 1.97, CI 1.6–2.43, p = 2.25×10−10, CAF 2.53%). One was previously reported at 9p21.3 (rs1333042; OR 1.31, CI 1.21–1.42, p = 1.8×10−11, CAF 42.3%) (Table 2). We assessed the robustness of the associations controlling also for age and the effect sizes and p-values were almost identical (data not shown).
To assess how the allele frequencies and effect sizes of variants identified in the Finnish population compare to other European populations, we studied those variants in a Dutch sample consisting of 717 sIA cases and 3,004 controls (Table 1). All three variants tagging the novel loci at 2q23.3, 5q31.3 and 6q24.2 had a similar low minor allele frequency (1.6–3.9%) in Finland and the Netherlands (Table 2). Two of them had similar effect sizes and were also replicated: 5q31.3 (rs113816216; OR 1.3, CI 0.98–1.75, p = 0.045, CAF 3.87%) and 6q24.2 (rs75018213; OR 1.5, CI 0.98–2.3 p = 0.034, CAF 2.3%). The previously reported 9p21.3 locus also replicated in the Dutch sample (rs1333042; OR 1.32, CI 1.17–1.49, p = 3.42×10−6, CAF 47.86%).
In the meta-analysis of the Finnish and Dutch samples, all three novel loci 2q23.3 (rs74972714; OR 1.89, p = 1.42×10−9), 5q31.3 (rs113816216; OR 1.66, p = 3.17×10−8) and 6q24.2 (rs75018213; 1.87, p = 7.1×10−11) were significantly associated to the sIA disease at genome-wide significance (Table 2; see Table S7 for imputation accuracy statistics). Some heterogeneity in effect sizes exists between samples (Table S9).
As the standard genome-wide significance 5×10−8 is estimated to correct for independent tests of common variants (MAF> = 5%) and we tested also a set of low-frequency variants, the common significance level may be too liberal. Based on Europeans of the 1000 Genomes project we estimated the significance level to be 3.82×10−8 (See Materials and Methods). All of the reported variants are below this level.
Some 20–30% of the sIA patients carry multiple sIAs, a phenomenon more commonly seen in familial sIA disease [8], [16], [17]. We hypothesized that an increased number of sIAs (≥2) in a given patient would reflect a higher underlying genetic load, motivating us to use aneurysm count as an intermediate phenotype to increase statistical power. The number of sIAs was used as a count data using the negative binomial regression analysis in the discovery sample of 760 Finnish sIA cases (1–8 sIAs per patient) and 2,513 controls. The QQ plot (Figure S2) and the genomic inflation factor (1.05) did not indicate substantial population stratification.
Nine loci had variants at p<5E-6 (Table S2; Manhattan plot in Figure S4). The most significant variant of each locus was selected for replication in the new Finnish sample of 858 sIA cases (1–6 sIAs per patient) and 4,048 controls. Two loci were replicated at p<0.05: 7p22.1 (rs150927513; RR 1.39, p = 8.36×10−4, CAF 5.24%) and 16p13.3 (rs144159053; rate ratio (RR) 1.66, p = 4.4×10−3, CAF 1.27%) (Table S2). rs10802056 on 1p12 had a significant association p-value but the effect direction was different and thus was not considered as replicated. We assessed the robustness of the associations controlling also for age and the effect sizes and p-values were almost identical (data not shown).
In the meta-analysis of the Finnish samples, 7p22.1 was genome-wide significant (rs150927513; RR 1.6, CI 1.37–1.88, p = 4.92×10−9, CAF 4.61%);Table 3; See genotype to aneurysm count distribution in Table S3). The rate ratio (RR) estimate is the relative rate of aneurysm formation (i.e. change in expected number of aneurysms) per allele as compared to minor allele homozygotes.
To compare the allele frequency and effect size of rs150927513 identified in the Finnish population to those of continental European populations, we studied the variant also in the Dutch, but the imputation quality (Impute info 0.38) and estimated allele frequency (0.29%) were too low to obtain reliable estimates (RR 0.97; 95% CI 0.17–4.03, p = 0.97). We additionally checked the minor allele frequency of rs150927513 in 498 whole-genome sequenced Dutch individuals of GENOMEoftheNETHERLANDS-project (http://www.nlgenome.nl/). Only two individuals were heterozygous and the rest were major allele homozygotes (MAF 0.2%), which is in agreement with our imputation results of the Dutch sample.
Previously published results on the 2q33.1 locus are inconsistent, being significant in the first GWAS [13], not significant in the enlarged follow-up GWAS [12], and uncertain in the third GWAS [10]. We aimed to study if the 2q33.1 would replicate in Finland, even though no variant in this region reached p<5E-6 in the discovery sample. We chose two of the most significant SNPs (in this study) at 2q33.1 for replication in the new Finnish replication sample, which was not used in the previous studies (rs12472355; OR 1.21, p = 2.23×10−4, CAF 44.3%, and rs919433; OR 1.18, p = 1.01×10−3, CAF 44.6%). They are in LD with the three previously investigated SNPs (rs787994, rs1429412, rs700651; LD r2 0.75–0.96). The variants rs12472355 (OR 1.23, CI 1.13–1.33, p = 4.84×10−7) and rs919433 (OR 1.21, CI 1.12–1.31, p = 2.15×10−6) did not reach genome-wide significance in the combined Finnish samples (Table 2). They were highly significant in the Dutch sample (rs12472355; OR, 1.39, CI 1.23–1.57, p = 1.05×10−7 and rs919433; OR 1.43, CI 1.26–1.61, p 9.77×10−9), and in the meta-analysis of all three samples they reached genome-wide significance (Table 2). The allele frequencies were notably higher in the Finnish samples (44% and 43.7%) than in the Dutch samples (33.2% and 31%).
We estimated the heritability explained by the reported variants. The four novel loci on 2q23.3, 5q31.3, 6q24.2 and 7p22.1 were estimated to explain 1.7% of the heritability in the combined Finnish samples. Adding the previously inconclusive 2q33.1 locus increases the heritability explained to 2.1%.
For validating the imputation accuracy, we genotyped 87 individuals of the discovery sample using Sequenom genotyping. The concordance rates range from 96–99% except rs74972714 was slightly lower at 94% (Table S8). We did additional validation by Sanger sequencing 10 individuals per variant who were predicted to carry minor alleles. The imputation was near perfect in all other SNPs except rs75018213 had discrepancies between major allele homozygote and heterozygotes (Table S11). We further estimated by simulation, how likely it would be to get the observed OR for rs75018213 in the discovery sample just by change, given the imputation accuracy (See Text S1 for details). The probability of chance finding was very low (p: 0.0001) even if assuming that the minor allele would be over-imputed by 20% in cases (p: 0.004).
Some individuals were genotyped by both Sanger sequencing and Sequenom and the concordance between the two methods was perfect (Table S11). Finally, we estimated, in silico, the imputation efficiency of reported SNPs in Dutch population. 96 individuals of the Genome of the Netherlands project had both high coverage whole-genome sequencing (40×) data as well as GWA chip genotyping data available. We imputed the genotypes of reported SNPs using the same imputation methods, 1000 Genomes reference panel and set of SNPs in GWA chips as was done in the discovery and Dutch comparison analyses. The genotype concordance rates were excellent (Table S13). It is noteworthy that the imputation quality measure reported by the Impute2 program was higher in all of the SNPs in our Dutch replication cohort (Table S7) than in the in silico validation experiment. This indicates excellent imputation quality in the Dutch replication.
We attempted to identify putative causative variants from whole exome sequencing data of 583 Finnish individuals. We focused on variants within 1 MB of the lead SNPs with high impact on protein product (i.e. variants affecting splice site, losing or gaining stop/start codon, altering reading frame) or non-synonymous coding SNPs. We additionally filtered variants if they were not in LD with the lead SNPs (r2<0.4, Europeans of 1000 Genomes if available). 254 variants were identified, most of which were rare. However 15 variants were enriched to low-frequency range (MAF>1%) (Table S12). The impact of these variants needs to be evaluated in follow-up studies.
The UCSC Genome Browser and HaploReg version 2 [19] were used to search for ENCODE regulatory elements at the five genome-wide significant variants.
rs74972714 at 2q23.3 and rs150927513 at 7p22.1 reside within a DNAse hypersensitivity peak. The rs75018213 at 6q24.2 resides on an ENCODE GATA2 transcription factor binding site peak (Table S4).
Using genome-wide Chip-SEQ analysis, Ernst et al. constructed a predicted cell-type specific regulatory region map of nine chromatin marks in nine cell lines [20]. rs113816216 at 5q31.3 resides on a predicted erythroleukemia cell specific (K562) strong enhancer and rs75018213 at 6q24.2 on a predicted lymphoblastoid cell (GM12878) weak enhancer (Table S4).
We searched for putative transcription factor binding sites affected by the four variants, based on position weight matrices from Transfac, Jaspar and ENCODE (top 3 enriched motifs for each transcription factor, identified in transcription factor Chip-SEQ peaks [19]). rs74972714 at 2q23.3 affects putative binding sites for EBF1 (ENCODE), HDAC2 (ENCODE), RXRA:PPARG complex (Transfac), ZNF423 (Jaspar) and ZIC3 (Jaspar). rs113816216 at 5q31.3 affects the putative binding sites for RFX1 (Transfac), SREBP1 (ENCODE), STAT3 (Transfac) and IKZF3 (Transfac). rs150927513 at 7p22.1 affects putative binding sites of T (brachyury) (Transfac), CEBPB (Transfac) and P300 (ENCODE). rs75018213 at 6q24.2 is not directly on any putative transcription factor binding site. (Table S4).
At the 2q33.1 locus neither of the studied variants (rs919433, rs12472355) are on ENCODE DNAse hypersensitivity or transcription factor binding site peaks. However, rs919433 is on a predicted lymphoblastoid (GM12878) cell enhancer whereas rs12472355 is not directly on any regulatory region. rs919433 disrupts a putative transcription factor binding sites for RUNX2 (OSF2,Transfac) and the MYC:MAX complex (Transfac).
To study the potential effects of the variants in the five significant loci on the transcripts of nearby genes, we correlated the variants to expression levels of exons of nearby genes (expression quantitative trait locus (eQTL) analysis) obtained using RNA-sequencing in lymphoblasts of genotyped European individuals from the 1000 Genomes Project (Finnish, British, Toscani and CEPH populations, n = 373; www.geuvadis.org, [21]). Each variant was correlated to transcripts residing within 1 MB. There were 55 genes in 586 exons available for analysis (see Materials and Methods) and in total 748 tests were performed corresponding to Bonferroni corrected significance threshold of 8.7×10−5. Strongest association for each variant are reported below and all eQTL results in Table S6.
The most significant eQTL associations were observed at the 2q33.1 locus: rs12472355 associated significantly to the closest gene ANKRD44 (per allele fold change (FC) 0.94, p = 1.83×10−5) and also to HSPD1 (FC 0.94, p = 1.6×10−4), whereas rs919433 was associated to the same genes but in different order of significance; HSPD1 (FC 0.94, p = 3.8×10−5) and ANKRD44 (FC 0.95, p = 1.4×10−4). Among the novel low-frequency variants, only rs150927513 at 7p22.1 was significantly associated to TNRC18 (FC 1.23, p = 5.1×10−5). Nominal associations were observed for two other novel low frequency variants: rs113816216 at 5q31.3 to VDAC1 (FC 2.12, p 4.6E-4); rs74972714 at 2q23.3 to EPC2 (FC 0.75, p = 3.9×10−2). rs75018213 at 6q24.2 did not have any association even at nominal p<0.05 (Table S6).
We additionally investigated the eQTL landscape of identified loci by pairwise comparison of p-values from eQTL (MAF>0.05 p<0.001) and sIA analyses (Figure S5) and by plotting eQTL associations (p<0.001) in the implicated loci (Supplementary Figure S6 A–E). Only few loci show strong (p<1E-5) association in eQTL and at least nominal (p<0.05) association to sIA (Table S10). There does not seem to be stronger eQTL associations in LD with the lead SNPs. In the 2q33.1, where the lead SNPs were significantly associated to transcript levels, there seems to be a lot of regulatory potential in the same locus, even though not in direct LD with the lead variants (Figure S6 E).
In this study, we used three approaches to improve the power to identify new loci associated to the sIA disease. First, we focused on the Finnish population isolate with increased risk for subarachnoid haemorrhage from ruptured sIAs (sIA-SAH) [5]. Second, we enriched the proportion of familial sIA patients in the discovery sample, thus possibly increasing the prevalence of risk alleles. Third, we increased genome-wide coverage through imputing ungenotyped variants based on 1000 Genomes Project data. The used 1000 Genomes Project imputation reference panel included 93 Finns, which made it well suited for discovery of enriched sIA associated variants in the Finnish population. Using this combination of strategies, we were able to identify three new loci associated with sIA disease, and one locus associated with the number of aneurysms. Additionally we replicated a locus where the evidence so far was inconclusive. Together these five loci account for 2.1% of the heritability in the Finnish samples. In comparison, the six previously identified SNPs explain 2.5% of the heritability in the discovery sample of the current study. Our results likely reflect the varying genetic background of complex traits, such as sIA, in different populations.
The lead SNPs in the four novel loci all have a low frequency (<5%) in the general population and could not have been identified without imputing the genotype data against the 1000 Genomes reference. One of the variants, rs150927513 at 7p22.1 that was associated with the number of sIAs, indicates a strong bottleneck effect, for it was 15 times more frequent in the controls of combined Finnish samples (4.6%) than in the Dutch sample (0.3%), and it is virtually non-existent in other populations (1000 Genomes). The three other loci had similar frequencies in Finland and other European populations (1000 Genomes). These four novel loci explain 1.7% of the heritability in the Finnish samples.
The four sIA loci had higher effect sizes (point estimates ranging from 1.59 to 1.88) than the lead SNPs identified by previous GWA studies. We cannot yet conclude whether relatively high ORs of low frequency risk alleles are a typical feature of sIA disease. Similar, and higher, odds ratios for low frequency and rare variants have been reported in isolates for other traits [22], [23]. It is likely that this first wave of low frequency and rare susceptibility variants represent “low hanging” fruits that do not allow general conclusions about the susceptibility landscape of sIA or other complex traits.
The variant rs74972714 at 2q23.3 has a frequency of about 2% in European populations, including Finns. It was significantly associated to sIA in the Finnish samples but did not show a trend for being associated in the Dutch sample despite having a similar allele frequency. Further studies are required to find out whether this variant tags a risk allele specific to Finnish sIA patients. The variant is located 40 kb downstream of LYPD6 and 55 kb upstream of MMADHC (Figure 2 A). LYPD6 has recently been characterized as a member of the Ly-6 protein superfamily [24]. LYPD6 is ubiquitously expressed with highest levels in heart and brain. GPI-anchored Ly-6 proteins such as PLAUR function, e.g., in cellular adhesion [24]. LYPD6 overexpression can inhibit transcriptional activity of the AP1 transcription factor complex [24], a key inflammation mediator activated, e.g., in endothelial cells in atherogenic disturbed blood flow conditions, leading in turn to upregulation of pro-inflammatory molecules [25]. Similar transcriptional changes have been found in the ruptured human sIA wall [26]. MMADHC is an intracellular vitamin B12 trafficking gene. Mutations in this gene can cause methylmalonic aciduria or homocystinuria, or both [27].
The variant rs113816216 at 5q31.3 has a frequency of 1–3% in Finland and most other European populations, except in Spain (7%). It was significantly associated to the sIA disease in the Finnish samples and was also significant in the Dutch sample but had a somewhat lower OR there (Table 2). The meta-analysis of all combined samples exceeded the genome wide significance threshold. The variant is located in the intron of FSTL4 (Figure 2 B), a poorly characterized gene. FSTL1, a paralog of FSTL4, codes a protein inducing innate immunity as TLR4 agonist [28]. Increased tissue levels of FSTL1 were associated to the severity of heart failure [29] and to the coronary artery aneurysm formation in Kawasaki disease [30]. Variants in FSTL4 were modestly associated to human ischemic stroke [31], and a variant 70 kb from FSTL4 nominally to hypertension [32].
The variant rs75018213 at 6q24.2 has similar frequencies (2%) in European populations, including Finns. It was significantly associated to the sIA disease in the Finnish samples and was also significant in the Dutch sample but had a somewhat lower OR there (Table 2). It is located in the intron of EPM2A. The LD spans over 300 kb downstream covering FBXO30, LOC100507557, SHPRH and GRM1 (Figure 2 C). In the ENCODE data, rs75018213 is located in a GATA2 transcription factor binding site RNA-seq peak. Homozygous deletions in the EPM2A gene result in progressive myoclonus epilepsy (PME) with Lafora bodies (OMIM 254780) [33]. No vascular anomalies have been reported in EPM2 deletion patients with a PME phenotype or their heterozygote parents. EPM2A encodes a phosphatase, which dephosphorylates glycogen, but it is likely that EPM2A has broader functions in regulation of glycogen biosynthesis, endoplasmic reticulum stress, autophagy, and possibly also cell cycle [34].
The variant rs150927513 at 7p22.1 was significantly associated to sIA count per individual in the Finnish population (Table 1). Its frequency was 4.6% in the Finnish samples but only 0.3%, in the Dutch sample, in line with most European populations. This variant would therefore likely not have been identified if a sufficient number of Finnish individuals had not been included in the reference panel.
The variant is located in the intron of RADIL (Figure 2 D), a rap GTPase interactor, an essential effector of RAP1 in activation of integrins in cell-adhesive signalling by G protein-coupled receptors [35]. RADIL has also been shown to control, together with RAP1, neutrophil adhesion and chemotaxis [36]. Neutrophils seem to have a role in the formation and rupture of intracranial and abdominal aortic aneurysm [26], [37], [38]. The strongest eQTL association was to an exon of TNRC18 (FC 1.23, p = 5.1×10−5), a functionally uncharacterized gene.
As we analysed the number of sIAs as a count variable from 0–8, the inherent assumption was that the same variant would increase the risk of the first and the subsequent sIA formation. Thus, any variant associated to the number of sIAs will to some extent be associated in the case vs. control analysis. Indeed, in the analysis of combined Finnish cohorts rs150927513 was associated in the case-control analysis (OR 1.54, p = 6.5×10−7) and consistently also in the analysis of multiple vs. single sIA patients (OR 1.65, p = 8.4×10−4). The association of this variant, should be interpreted as reflecting the tendency of sIA formation, rather than considering multiple sIAs as a completely different dichotomous end point.
The 9p21.3 locus has been robustly associated to the sIA disease [12] as well as to cardiovascular, metabolic and cancer traits [39], [40], and it has been extensively studied by others [41]. The allele frequency and effect size in the current study, although with a different lead SNP (r2 = 0.7 to previous lead SNP rs1333040), are in strong agreement with the previous study [12]. This locus is not therefore discussed further here.
Two common variants, rs12472355 and rs919433 at 2q33.1 were significantly associated to the sIA disease in the Finnish and Dutch samples (Table 2), rs919433 intronic and rs12472355 upstream 30 kb from ANKRD44 (Figure 2 E). The allele frequencies were somewhat higher in the Finnish samples (rs919433, 44%; rs12472355 43.7%) than in the Dutch samples (33.2%; 31%) or in the Japanese population according to 1000 Genomes Project (28.1%; 27.5%). In this locus, the risk allele was reversed in the Japanese cohort of the previous sIA GWA study [10]. ANKRD44 is likely a subunit of protein phosphatase 6 [42] that functions, e.g., in cell cycle control [43] and in inhibition of NF-κB activation [44]. NF-κB is a significant mediator in experimental sIA formation in rats, highly expressed in human sIA wall [45], and it was associated to human sIA wall rupture in transcriptomic profiling [26]. In eQTL analysis rs12472355 was significantly associated to ANKRD44 (FC 0.94, p = 1.83×10−5) and rs919433 to HSPD1 (FC 0.94, p = 3.8×10−5)
In conclusion, we identified four novel loci associated to sIA disease and confirmed one additional locus with previously inconclusive evidence, together explaining 2.1% of the sIA heritability in Finland. Our data illustrates the utility of high-risk population isolates, familial disease history, and dense genotype imputation in search for low-frequency variants associated to complex human diseases. The inclusion of Finnish individuals in the imputation reference panel and especially the highly differentiated variant in 7p22.1 would likely not have been identified
The identification of the four novel low frequency variants would likely have required much larger sample sizes in more mixed populations. Further studies of the identified five loci are needed to explain their functional mechanisms in the pathogenesis of sIA disease.
For all of the Finnish and Dutch samples, the local ethics committees approved the study and all patients gave written informed consent.
From both of the analyses (the case vs. controls and the number of sIAs) the best independent SNPs were taken to replication if p<5E-6. Additional significant independent SNPs in a locus was tested by analyzing each SNP within 1 MB from the top SNP while adding the top SNP as a covariate. Additionally the most significant SNP in the current study in 2q33.1 region with uncertain evidence in previous sIA GWASs was taken to replication. Variant was considered replicated if it reached one-tailed significance of p<0.05 and was consistent in terms of risk allele. In all of the results, one-tailed p-values are given for the Finnish replication and in Dutch results.
Genomic DNA was extracted from peripheral blood and genotyped by Illumina arrays: the Finnish discovery sample and the Dutch replication cases by CNV370k DUO chip; the HBCS and YFS controls by Illumina Human670K customBeadChip; and the H2000 controls by Illumina Infinium HDHuman610-Quad BeadChip.
In the Finnish replication sample, DNA was genotyped using Sequenom MassARRAY system and iPLEX Gold assays (Sequenom Inc., San Diego, USA). The data was collected using the MassARRAY Compact System (Sequenom) and the genotypes were called using TyperAnalyzer software (Sequenom). Genotyping quality was examined by a detailed QC procedure consisting of success rate checks, duplicates, water controls and Hardy-Weinberg Equilibrium (HWE) testing. SNPs were filtered if genotype missingness >0.05 or if HWE p<0.001.
For imputation of additional genotypes in the discovery sample, the Young Finns replication cohort and in the 2nd Dutch replication sample the genotypes were first pre-phased [55] using the Shape-IT [56] phasing software and the pre-phased haplotypes were subjected to imputation. The Impute version 2.2.2 software [57] with 1,000 Genomes Phase I integrated variant set release (v3) reference panel (05 Mar 2012 release downloaded from http://mathgen.stats.ox.ac.uk/impute/data_download_1000G_phase1_integrated.html) was used. Imputed genotypes were filtered if the Impute info measure was <0.5 or minor allele frequency <0.01 in the Finnish discovery sample.
We analyzed whether the identified genome-wide significant SNPs might affect gene expression by using the European samples of the Geuvadis RNA-sequencing data set, with mRNA sequencing data from LCLs of 373 samples from the FIN, CEU, GBR and TSI populations of 1000 Genomes project (for details, see [21]).
We did eQTL analysis for each of the associating variants and all the genes within a 1 MB window that were expressed in >50% of the individuals (Table eQTL). We used exon quantifications based on individual read counts per exon, after correction by the total number of mapped reads per sample and PEER normalization to remove technical variation. For each exon, we calculated linear regression between these expression values and genotype dosage of the associating variants in the 1000 Genomes data.
Regional association plots were generated using LocusZoom with LD data from European populations of 1000 Genomes project (Hg19/March 2012) [58].
The UCSC Genome Browser and HaploReg version 2 [19] were used to search for ENCODE regulatory element regions located at the five genome-wide significant variants. HaploReg database also annotates if SNP resides on a putative transcription-factor binding site (TFBS) according to Transfac or Jaspar TFBS profiles and also 10 most enriched TFBS profiles identified in ENCODE TF Chip-Seq peaks. We used all the Jaspar and Transfac annotations and three most enriched ENCODE based TFBS annotations for each TF.
GWA was performed against two complementary phenotypes: the case vs. control status and the number of sIAs.
|
10.1371/journal.pgen.1006266 | TCS1, a Microtubule-Binding Protein, Interacts with KCBP/ZWICHEL to Regulate Trichome Cell Shape in Arabidopsis thaliana | How cell shape is controlled is a fundamental question in developmental biology, but the genetic and molecular mechanisms that determine cell shape are largely unknown. Arabidopsis trichomes have been used as a good model system to investigate cell shape at the single-cell level. Here we describe the trichome cell shape 1 (tcs1) mutants with the reduced trichome branch number in Arabidopsis. TCS1 encodes a coiled-coil domain-containing protein. Pharmacological analyses and observations of microtubule dynamics show that TCS1 influences the stability of microtubules. Biochemical analyses and live-cell imaging indicate that TCS1 binds to microtubules and promotes the assembly of microtubules. Further results reveal that TCS1 physically associates with KCBP/ZWICHEL, a microtubule motor involved in the regulation of trichome branch number. Genetic analyses indicate that kcbp/zwi is epistatic to tcs1 with respect to trichome branch number. Thus, our findings define a novel genetic and molecular mechanism by which TCS1 interacts with KCBP to regulate trichome cell shape by influencing the stability of microtubules.
| The particular shape of plant cells is not only crucial for their biological functions but also affects the overall shape of organs. How cell shape is controlled is a fundamental question in developmental biology, and the study of plant cell shape regulation is an interesting part of plant biology. Arabidopsis trichomes have been used as a good model system to investigate cell shape at the single-cell level. In this study, we use Arabidopsis trichomes as a model to identify the trichome cell shape 1 (tcs1) mutants with the reduced trichome branch number. TCS1 encodes a microtubule binding protein, which is required for the stability of microtubules. We further find that TCS1 physically interacts with a microtubule motor involved in the regulation of trichome branch number. TCS1 acts genetically with this microtubule motor to control trichome branch number. Thus, our findings provide important insights into how the microtubule cytoskeleton determines cell shape.
| The particular shape of plant cells not only relates to their functions but also influences the overall shape of organs. Arabidopsis trichomes are well established as a system for studying cell shape at the single-cell level [1–3]. Arabidopsis trichomes differentiate from single epidermal cells, which stop proliferating and begin endoreduplication cycle or endocycle. After three or four endoreduplication cycles, trichome cells have two successive branching events and morphological changes, and then form mature trichomes [1]. Trichomes on Arabidopsis leaves are regularly spaced and exhibit a distinctive shape with a stalk and three or four branches. The cytoskeletons appear to be important for establishing and maintaining the branching pattern of trichomes [4–6]. It is generally accepted that mutations in genes involved in the regulation of actin cytoskeleton often cause distorted trichomes, while the disruption of genes regulating the microtubule cytoskeleton usually influences the number of trichome branches [4,5,7–12]. However, the genetic and molecular mechanisms by which the cytoskeletons determine trichome cell shape remain largely unknown in plants.
In trichomes, microtubules, a major component of the plant cytoskeletons, not only regulate anisotropic cell expansion but also control cell branching. Several factors that regulate trichome branch number by influencing the microtubule cytoskeleton have been described in Arabidopsis. Arabidopsis TUBULIN FOLDING COFACTOR (TCF) C and TCFA have been suggested to be required for microtubule biogenesis, and their loss-of-function mutants show the reduced trichome branch number and shape as well as multiple growth defects [13,14], suggesting that the formation of new microtubules is likely to be important for the formation of new branches. KINESIN-13A has the microtubule-depolymerizing activity in vitro and in vivo, and kinesin-13a mutants produce trichomes with more branches [15]. Kinesin-like calmodulin-binding protein (KCBP/ZWICHEL) is involved in the regulation of microtubule stability and trichome morphogenesis in plants [4,16]. Trichomes on kcbp/zwichel (zwi) leaves have a short stalk and only one or two branches compared with wild-type trichomes with three or four branches [16]. KCBP-interacting Ca2+ binding protein (KIC) represses the activity of KCBP in response to Ca2+ and regulates trichome branching [17]. Plants overexpressing KIC produce trichomes with reduced branch number [17]. KCBP also physically interacts with ANGUSTIFOLIA (AN) in yeast cells, which is involved in the regulation of the microtubule cytoskeleton [18]. Trichomes on an leaves have one or two branches, indicating AN is required for normal trichome branching [18,19]. KCBP has been suggested to function with suppressors of zwi (SUZ) in a complex to control the number of trichome branches, but the SUZ genes remain to be cloned in Arabidopsis [20]. KCBP has also been recently reported to interact with both microtubules and F-actin to affect trichome branch initiation and elongation, respectively [21]. These studies imply that KCBP acts as an important node linking cytoskeletons with trichome cell shape.
To further understand the genetic and molecular mechanisms of cell shape control, we characterize tcs1 mutants, which form trichomes with the reduced branch number. Mutations in TCS1 influence the stability of microtubules. TCS1 encodes a coiled-coil domain-containing protein, which binds to microtubules in vitro and in vivo and promotes the assembly of microtubules. Further results reveal that TCS1 interacts physically and genetically with KCBP/ZWI to control the number of trichome branches. Thus, our findings reveal a novel genetic and molecular mechanism of TCS1 and KCBP in trichome cell shape control.
We isolated the trichome cell shape 1 (tcs1) mutants in a screen of publicly available T-DNA mutant collections of Arabidopsis thaliana. The tcs1-1, tcs1-2 and tcs1-3 trichomes had the reduced branch number compared with wild-type trichomes (Fig 1). By contrast, the tcs1 mutants did not show any obvious defects in plant growth. Progeny of crosses of the three lines indicated that they are allelic. We further measured the number of trichome branches using the first pair of leaves. In wild-type leaves, trichomes normally had two branching points with three branches (92%), although trichomes with four branches were occasionally found (Fig 1C). By contrast, about 70% and 25% of trichomes on tcs1 leaves had two and three branches, respectively (Fig 1C). The tips of tcs1 trichome branches were sharp, as those observed in wild-type trichome branches (Fig 1B). Thus, these results show that TCS1 influences the number of trichome branches in Arabidopsis.
In Arabidopsis, the reduced branch number of trichomes is often correlated with a decrease in the level of endoreduplication or the destabilization of microtubules [18,22]. We firstly investigated whether TCS1 could affect endoreduplication in trichome cells. As the nuclear size is often associated with the ploidy level, we measured the nuclear size of Col-0 and tcs1-1 trichomes. The average nuclear size of tcs1-1 trichomes was similar to that of wild-type trichomes (S1A and S1B Fig). The ploidy levels in tcs1-1 leaves were comparable with those in wild-type leaves (S1 Fig). These results suggest that TCS1 may not regulate endoreduplication. We then asked whether TCS1 could influence the microtubule cytoskeleton. The microtubule-disrupting drug oryzalin has been shown to destabilize microtubules, leading to a decrease in the number of trichome branches in Arabidopsis [23]. We therefore treated 4-day-old seedlings of Col-0 and tcs1-1 with 20 μM oryzalin for 2 hours. After a 10-day recovery on ½ MS medium, we examined the branch number of Col-0 and tcs1-1 trichomes. As shown in Fig 2A and 2B, the oryzalin treatment caused a 7.7% decrease in the average number of Col-0 trichome branches, while the oryzalin treatment resulted in an 18.9% reduction in the average number of tcs1-1 trichome branches. The microtubule-stabilizing drug paclitaxel (taxol) has been reported to stabilize microtubules [23]. We asked whether taxol could rescue the trichome branch phenotype of tcs1. Four-day-old seedlings of Col-0 and tcs1-1 were treated with 20 μM taxol for 2 hours. After a 10-day recovery on ½ MS medium, we examined the branch number of Col-0 and tcs1-1 trichomes. In our growth condition, the taxol treatment caused a 2.6% increase in the average number of Col-0 trichome branches, while the taxol treatment resulted in a 9.3% increase in the average number of tcs1-1 trichome branches (Fig 2C and 2D), suggesting that taxol partially rescues the phenotype of tcs1-1 trichome branches.
As the microtubule is crucial for hypocotyl elongation [24], we asked whether TCS1 affects hypocotyl growth. As shown in Fig 2E and 2F, the average length and width of dark-grown tcs1-1 hypocotyls was comparable with that of dark-grown Col-0 hypocotyls. We then treated dark-grown Col-0 and tcs1-1 seedlings with oryzalin and measured their hypocotyl length and width. After oryzalin treatment, hypocotyls of tcs1-1 were significantly shorter and wider than those of the wild type (Fig 2E and 2F). Epidermal cells in tcs1-1 hypocotyls were short and wide in comparison with those in wild-type hypocotyls (S2 Fig). These results show that hypocotyls of tcs1-1 are hypersensitive to oryzalin treatment than wild-type hypocotyls.
As tcs1 trichomes had the reduced branch number and were hypersensitive to oryzalin and taxol, we asked whether TCS1 affects the stability of microtubules in trichome cells. We therefore crossed GFP-TUB6 transgenic plants with the tcs1-1 mutant and generated GFP-TUB6;tcs1-1 plants. Cortical microtubule arrays in tcs1-1 trichome cells were similar to those in wild-type trichome cells (Fig 3A). We then applied the microtubule-disrupting drug oryzalin to trichome cells of GFP-TUB6 and GFP-TUB6;tcs1-1 leaves. As shown in Fig 3A and 3B, cortical microtubule arrays disappeared faster in tcs1-1 trichome cells than those in wild-type trichome cells. We counted the number of cortical microtubules in the trichome branch junction. Microtubules were similar in density before oryzalin treatment. However, more cortical microtubules were disrupted in tcs1-1 than in the wild type after drug treatment. These results indicate that TCS1 influences the stability of microtubules in trichomes. Similarly, we observed that microtubule arrays disappeared relatively faster in epidermal cells of tcs1-1 cotyledons than those in epidermal cells of wild-type cotyledons after oryzalin treatment (S3 Fig).
As tcs1 hypocotyls were hypersensitive to the microtubule-disrupting drug oryzalin, we investigated whether TCS1 is required for the stability of microtubules in hypocotyl cells. Cortical microtubule arrays in epidermal cells of GFP-TUB6;tcs1 hypocotyls were comparable with those of GFP-TUB6 hypocotyls (Fig 3C and 3E). We then applied the microtubule-disrupting drug oryzalin to epidermal cells of etiolated GFP-TUB6 and GFP-TUB6;tcs1-1 hypocotyls. Cortical microtubule arrays disappeared relatively faster in epidermal cells of tcs1-1 hypocotyls than those in epidermal cells of wild-type hypocotyls (Fig 3D and 3E). When oryzalin was washed off after the treatment, the recovery of cortical microtubules in epidermal cells of tcs1-1 hypocotyls was slower than that in epidermal cells of wild-type hypocotyls (Fig 3D and 3E). Taken together, these results indicate that TCS1 influences the stability of microtubules.
The tcs1-1, tcs1-2 and tcs1-3 mutants were identified from the T-DNA insertions in the fourth exon and the sixth exon of the gene At1g19835, respectively (Fig 4A). T-DNA insertions were confirmed using T-DNA specific and flanking primers (S4A–S4C Fig). We further investigated the expression level of At1g19385 in tcs1-1, tcs1-2 and tcs1-3 mutant seedlings. As shown in S4D Fig, the full length transcript of At1g19835 was not detected in tcs1 mutants, suggesting that tcs1 mutants are loss-of-function alleles. A plasmid containing wild-type At1g19835 cDNA driven by a 35S promoter was introduced into the tcs1-2 mutant. Transgenic plant exhibited complementation of tcs1-2 phenotypes (Fig 4C and 4D). In addition, transformation of tcs1-1 with TCS1-GFP fusion protein under the control of the TCS1 promoter (pTCS1:TCS1-GFP) restored a wild-type phenotype (S4E Fig). Therefore, these results indicate that At1g19835 is the TCS1 gene.
TCS1 encodes a 982-amino-acid protein that contains four coiled-coil domains, which belongs to a family of long coiled-coil protein that consists of 7 members in Arabidopsis [25] (Fig 4B; S5 Fig). Although the family members have been named as filament-like plant proteins (AtFPP), their biochemical and biological functions are totally unknown in Arabidopsis [25]. By performing a BLAST search in the databases, we identified TCS1 homologs in Brassica rapa, Gossypium raimondii, Sorghum bicolor, Zea mays, and Oryza sativa, but we did not find convincing homologs from animals and yeasts (S5 Fig), suggesting that TCS1 and its homologs might have evolved to control cell morphogenesis in plants.
To determine the expression pattern of TCS1, RNA from roots, flowers, seedlings and leaves were investigated by RT-PCR analysis. TCS1 mRNA was detected in all plant organs tested (S6 Fig). Tissue-specific expression pattern of TCS1 was examined using histochemical assay of GUS activity of transgenic plants containing the TCS1 promoter:GUS fusion (pTCS1:GUS). GUS activity was detected in cotyledons, leaves, inflorescences and developing etiolated hypocotyls (Fig 4E–4H). GUS activity was also observed in trichomes (Fig 4I), consistent with the role of TCS1 in trichome morphogenesis.
As tcs1 affects the stability of microtubules, we asked whether TCS1 could directly bind to the microtubules. A cosedimentation assay was used to analyze the binding of TCS1 to taxol-stabilized microtubules. TCS1 was expressed as a maltose binding protein (MBP) fusion protein (MBP-TCS1) in E.coli. As shown in Fig 5A and S7A Fig, MBP-TCS1 was cosedimented with the microtubules. The binding of TCS1 to microtubules was saturated at a stoichiometry of about 0.38 M MBP-TCS1 per mole of tubulin dimers (S7C Fig). The binding of the positive control (AUGMIN subunit 8, AUG8) to microtubules was saturated at a stoichiometry of about 0.22 M His-AUG8 per mole of tubulin dimers in our experimental conditions (S7B and S7C Fig) [26]. We then asked whether TCS1 could directly interact with tubulins. To test this, we conducted pull-down experiments. As shown in Fig 5B, MBP-TCS1 bound to tubulins, while the negative control (MBP-TCP14) did not interact with tubulins. Thus, these results indicate that TCS1 physically interacts with tubulins in vitro.
We further performed co-immunoprecipitation analyses to detect the interaction of TCS1 with tubulins in Arabidopsis. Total proteins from pTCS1:TCS1-GFP or 35S:GFP plants were isolated and incubated with GFP-Trap-A agarose beads to immunoprecipitate TCS1-GFP and GFP. The anti-GFP and anti-tubulin antibodies were used to examine immunoprecipitated proteins, respectively. As shown in Fig 5C, tubulins were found in the immunoprecipitated TCS1-GFP complex but not in the negative control (GFP), indicating that TCS1 physically associates with tubulins in Arabidopsis.
To further investigate whether TCS1 localizes to cortical microtubules, we conducted live-cell imaging using a functional TCS1-GFP fusion under the control of TCS1 promoter. As shown in Fig 5D–5F, TCS1-GFP localizes to puncta along cortical microtubules (mCherry-TUB6) in pavement cells, indicating that TCS1 binds to the microtubules. We then investigated the co-localization of TCS1-GFP and microtubules in developing trichomes. We have previously showed that it is difficult to observe the signal of mCherry labeled-microtubules in trichomes [21]. We therefore used pTCS1:TCS1-GFP and GFP-TUB6 -expressing lines to compare TCS1 with microtubules. In GFP-TUB6 trichomes, transverse microtubule arrays formed rings encircling the elongating branches, without the signal at the extreme apex (Fig 5G) [21]. Similarly, we observed that TCS1-GFP was present in elongating trichome branches, but leave a TCS1-depleted zone at the extreme apex (Fig 5H). These results indicate that TCS1 and microtubules exhibit similar organization patterns in trichomes, further suggesting that TCS1 is a microtubule-binding protein.
As TCS1 directly interacts with microtubules, we asked whether TCS1 could affect microtubule assembly. We therefore added various concentrations of MBP-TCS1 (0, 0.25, 0.5 and 1 μM) and 1 μM MBP to a 20 μM tubulin solution, and tubulin polymerization was investigated by measuring turbidity. As shown in Fig 5I, the presence of MBP-TCS1 increased turbidity, indicating that MBP-TCS1 increases microtubule mass. The assembly rate of tubulins was increased in a dosage-dependent manner with the addition of MBP-TCS1. To confirm this result, we observed the assembly of rhodamine-labeled tubulins incubated with MBP and MBP-TCS1 under confocal microscopy. As shown in Fig 5J, the assembly of microtubules was detected in the presence of MBP-TCS1 rather than MBP. Taken together, these results indicate that TCS1 promotes microtubule assembly.
To further understand the molecular mechanism of TCS1 in the regulation of trichome branch number, we performed a yeast two-hybrid screen to identify putative TCS1-binding proteins. TCS1 was fused to the GAL4 DNA binding domain (BD) and used as a bait. In this screen, KCBP/ZWI was identified as a putative TCS1-interacting protein. KCBP/ZWI has been shown to affect microtubules and trichome branches [16], suggesting that TCS1 could interact with KCBP/ZWI to control trichome branches. We tested the interactions between TCS1 and the full length KCBP in yeast cells. As shown in Fig 6A, TCS1 interacted with KCBP in a yeast two-hybrid assay. We then investigated the interaction of TCS1 with KCBP using in vitro pull-down experiments. TCS1 was expressed as a maltose binding protein (MBP) fusion protein (MBP-TCS1), while KCBP was expressed as a glutathione S-transferase (GST) fusion protein (GST-KCBP). As shown in Fig 6B, MBP-TCS1 bound to GST-KCBP, while the negative control (MBP) did not bind to GST-KCBP. This result indicates that TCS1 physically interacts with KCBP in vitro.
We further performed co-immunoprecipitation analysis to investigate the association of TCS1 with KCBP in Arabidopsis. We generated 35S:Myc-KCBP transgenic plants. We crossed the pTCS1:TCS1-GFP and 35S:GFP transgenic lines with 35S:Myc-KCBP transgenic plants to generate pTCS1:TCS1-GFP;35S:Myc-KCBP and 35S:GFP;35S:Myc-KCBP plants, respectively. Total proteins were isolated and incubated with GFP-Trap-A agarose beads to immunoprecipitate TCS1-GFP and GFP. The anti-GFP and anti-Myc antibodies were used to detect immunoprecipitated proteins, respectively. Myc-KCBP was found in the immunoprecipitated TCS1-GFP complex but not in the negative control (GFP) (Fig 6C), indicating that TCS1 physically associates with KCBP in Arabidopsis.
As TCS1 physically interacts with KCBP, and tcs1 mutants showed similar trichome branching phenotypes to kcbp/zwi mutants, we sought to establish genetic relationships between TCS1 and KCBP in the regulation of trichome branch number. We obtained the zwi-101 mutant (SALK_017886) harboring the T-DNA insertion in the KCBP/ZWI gene (S8 Fig). The full length mRNA of KCBP could not be detected in zwi-101, suggesting that zwi-101 is a loss-of-function allele. The zwi-101 trichomes exhibited the reduced number of branches (Fig 6D), consistent with previous results [16]. We then generated a zwi-101 tcs1-2 double mutant and investigated its trichome branch number. As shown in Fig 6D and S9 Fig, the branch number of zwi-101 tcs1-2 double mutant trichomes was comparable to that of zwi-101 single mutant trichomes, suggesting that zwi-101 is epistatic to tcs1-2 with respect to the number of trichome branches. Considering that both TCS1 and KCBP affect the stability of microtubules, we asked whether genetic interactions between TCS1 and KCBP in trichome branch number are related to microtubule stability. We therefore treated 4-day-old seedlings of Col-0, zwi-101, tcs1-2 and zwi-101 tcs1-2 with 20 μM oryzalin for 2 hours. After a 10-day recovery on ½ MS medium, the number of Col-0, zwi-101, tcs1-2 and zwi-101 tcs1-2 trichome branches was investigated. After oryzalin treatment, the number of zwi-101 tcs1-2 trichome branches was similar to that of zwi-101 trichome branches (Fig 6E). The oryzalin treatment caused a similar decrease in the average number of zwi-101 tcs1-2 and zwi-101 trichome branches. These results suggest that TCS1 acts genetically with KCBP to regulate the number of trichome branches by influencing the stability of microtubules.
KCBP was reported to physically interact with AN in yeast cells [18]. The an mutants showed the reduced branches of trichomes in leaves [18,19]. We asked whether TCS1 and AN could function in a common pathway to control trichome branches. To test this, we obtained the an-101 mutant (SALK_026489) harboring the T-DNA insertion in the AN gene (S10 Fig). The full length mRNA of AN could not be detected in an-101, suggesting that an-101 is a loss-of-function allele. The an-101 mutant trichomes mainly had one or two branches (S11 Fig), consistent with previous results [18]. We then generated the an-101 tcs1-2 double mutant and investigated its trichome branches. The number of an-101 tcs1-2 trichome branches was similar to that of an-101 trichome branches (S11 Fig), suggesting an epistatic genetic interaction. We further tested whether TCS1 could physically interact with AN. As shown in S12 Fig, TCS1 did not directly interact with AN in vitro (S12 Fig).
A fundamental question in developmental biology is how cell shape is controlled. In plants, cell shape is crucial not only for the function of the individual cell, but also for its role in organ shape and size control. However, the genetic and molecular mechanisms that determine cell shape remain largely unknown in plants. In this study, we report that the TCS1 gene, which encodes a microtubule binding protein with long coiled-coil domains, is required for trichome cell shape in Arabidopsis. TCS1 directly binds to microtubules and promotes microtubule assembly. TCS1 physically and genetically interacts with KCBP/ZWICHEL to regulate the number of trichome branches by influencing microtubule stability. Thus, our findings reveal a novel genetic and molecular mechanism of TCS1 and KCBP in trichome cell shape control.
The tcs1 trichomes showed the reduced branch number (Fig 1), although tcs1 plants appear to be similar to wild-type plants. Trichome branching is a complicated process, which is regulated by a number of factors. In Arabidopsis, DNA replication (endoreduplication) in trichome cells influences the number of trichome branches [22,27,28]. However, mutations in TCS1 did not affect ploidy levels in leaves and nuclear size in trichome cells (S1 Fig). Thus, it is unlikely that TCS1 regulates trichome branch number by influencing DNA replication events. After endoreduplication, a cytoskeleton-dependent polarization event happens during trichome morphogenesis [23], resulting in a total of three to four branches in the mature trichome on leaves. Molecular-genetic and pharmacological studies have established that microtubules are essential for trichome branching in Arabidopsis [23,29]. Interestingly, the trichomes of tcs1 were hypersensitive to the microtubule-disrupting drug oryzalin in comparison with those of the wild type (Fig 2A and 2B). Similarly, tcs1 hypocotyls were more sensitive to oryzalin than wild-type hypocotyls (Fig 2E, 2F, and S2 Fig). By contrast, the microtubule-stabilizing drug taxol treatment partially rescued the branch number of tcs1 trichomes (Fig 2C and 2D). These results suggest that TCS1 may affect the stability of microtubules, which are crucial for trichome cell morphogenesis. Consistent with this notion, we observed that microtubules in tcs1 cells disappeared faster than those in wild-type cells when treated with oryzalin (Fig 3 and S3 Fig). Thus, these results support that mutations in TCS1 influence the stability of microtubules, resulting in the altered trichome cell shape in Arabidopsis.
The TCS1 gene encodes a coiled-coil domain-containing protein, which belongs to a family of long coiled-coil protein that consists of 7 members in Arabidopsis [25]. However, the biological functions of the TCS1 family members are totally unknown in Arabidopsis [25]. Therefore, TCS1 is a novel regulator of trichome cell shape in Arabidopsis. Sequence analyses show that TCS1 homologs are plant-specific proteins (S5 Fig), suggesting that TCS1 and its homologs might have evolved to regulate cell shape in plants. Expression of TCS1 was detected in all tested tissues (S6 Fig), although the only visible phenotype in tcs1 mutants was found in trichomes. It is possible that TCS1 might function redundantly with other proteins to influence cell growth in other tissues or cell types.
Several microtubule binding proteins have been known to influence the branch number of trichomes in Arabidopsis [30,31]. Our biochemical analyses showed that TCS1 physically interacts with microtubules in vitro and in vivo (Fig 5A–5C). Live-cell imaging assay found that TCS1 directly bound to the microtubule in Arabidopsis cells (Fig 5D–5F). In addition, TCS1-GFP and GFP-TUB6 showed similar organization patterns in elongating trichome branches (Fig 5G and 5H). These results support that TCS1 is a microtubule binding protein. It is possible that TCS1 directly binds to microtubules during trichome development and stabilizes microtubules, thereby influencing the formation of trichome branches in Arabidopsis. Biochemical analyses showed that TCS1 promotes microtubule assembly, consistent with the function of TCS1 in stabilizing microtubules (Fig 5I and 5J). The microtubule assembly has been known to influence trichome branch number. For example, mutations in the TUBULIN FOLDING COFACTOR (TCF) C and TCFA result in the unbranched trichome phenotype [13,14]. These mutants were proposed to affect the making of assembly component α/β tubulin dimmers and possibly decrease the assembly of new microtubules. Mutations in KINESIN-13A, which promotes microtubule depolymerization, resulted in the increased number of trichome branches [15,32]. Thus, it is possible that TCS1 promotes microtubule assembly and increases the stability of microtubules, thereby influencing trichome branch number in Arabidopsis.
KCBP, a microtubule motor, regulates cell division and trichome cell shape in Arabidopsis [16,17]. Trichomes on zwi leaves had one or two branches with blunt tips. It has been suggested that KCBP participates in the trichome morphogenesis by regulating the local reorientation and stability of microtubules [30,31]. Similarly, TCS1 regulates trichome branch number by influencing the stability of microtubules. The tcs1 mutants showed similar trichome branch number phenotype to kcbp/zwi mutants, suggesting that TCS1 could genetically interact with KCBP to control the branch number of trichomes. Consistent with this idea, our genetic analyses show that zwi is epistatic to tcs1 with respect to trichome branch number. Further results demonstrated that TCS1 physically interacted with KCBP in vitro and in vivo (Fig 6A–6C). As both TCS1 and KCBP influence the stability of microtubules, it is possible that TCS1 functions with KCBP to control trichome branch number by affecting the dynamics and stability of microtubules in Arabidopsis. Supporting this notion, zwi-101 tcs1-2 and zwi-101 trichomes showed a similar level of hypersensitivity to oryzalin (Fig 6E). A recent study have shown that KCBP interacts with both microtubules and actin cytoskeleton to regulate trichome branching and elongation in Arabidopsis [21]. The zwi trichomes had the reduced number of branches, shortened stalks and stunted branches [16]. The reduced number of branches in zwi trichomes is likely caused by defects in microtubules. By contrast, the transverse cortical F-actin cap at the trichome branch apex has been proposed to regulate polarized branch elongation and tip sharpening [21]. tcs1 mutants only affected the trichome branch number and had normal stalks and trichome branch tips (Fig 1A and 1B), suggesting that KCBP and TCS1 may have the overlapped function in the regulation of microtubule cytoskeleton rather than actin cytoskeleton.
Genetic studies suggested that KCBP may interact with multiple factors, such as SUZ1, SUZ2 and SUZ3 and might function as a complex, although the SUZ genes have not been cloned in Arabidopsis [20]. KCBP has been shown to physically interact with a plant-specific protein kinase termed KCBP-interacting protein kinase (KIPK), a calcium binding protein (KIC) and AN [17,18,33]. KIC regulates trichome morphogenesis by influencing microtubule binding and microtubule-stimulated ATPase activities of KCBP, although the genetic interactions between KCBP and KIC remain unknown [17]. AN is also required for normal trichome morphogenesis in Arabidopsis although AN has been suggested to indirectly regulate microtubules [18,19]. It has been proposed that the level of each protein in the KCBP complex is likely to be crucial for trichome morphogenesis [17]. As TCS1 physically and genetically interacted with KCBP, it is possible that TCS1 and other members in the KCBP complex may have genetic interactions in trichome branching. Supporting this notion, we found an epistatic interaction between AN and TCS1 with respect to the number of trichome branches, although TCS1 does not physically interact with AN (S12 Fig). It will be a worthwhile challenge to build up the genetic and molecular interactions between TCS1 and other members of the KCBP complex in the future. Taken together, our findings reveal a novel genetic and molecular mechanism by which TCS1 interacts with KCBP to control trichome cell shape by influencing the stability of microtubules.
The tcs1-1 (SAIL_403_D02), tcs1-2 (SALK_040648), tcs1-3 (SALK_078664), zwi-101 (SALK_017886), and an1-101 (SALK_026489) mutants were obtained from the Nottingham Arabidopsis Stock Centre (NASC). The T-DNA insertions were verified by PCR and sequencing using the primers described in S1 Table. Arabidopsis seeds were sterilized with 100% isopropanol for 2 min and 10% NaClO (v/v) for 10 min and then washed six times with sterile water. Arabidopsis seeds were dispersed on ½ Murashige and Skoog (MS) medium containing 0.9% agar and 1% glucose and then stored at 4°C for 3 days in the darkness. Plants were grown at 22°C under long-day conditions (a 16-h-light /8-h-dark cycle). To observe etiolated hypocotyls, we grow plants in dark at 22°C.
A PCR-based Gateway system was used to generate 35S:TCS1, pTCS1:TCS1-GFP and pTCS1:GUS constructs. The TCS1 CDS was amplified using the primers TCS1-CDS-LP and TCS1-CDS-RP (S1 Table). PCR product was subcloned into the pCR8/GW/TOPO TA cloning vector (Invitrogen) using TOPO enzyme. The TCS1 CDS was then subcloned into the Gateway binary vector pMDC32 to generate the 35S:TCS1 construct. The TCS1 genomic sequence containing a 2012-bp promoter sequence and 3298bp gene was amplified using the primers gTCS1-GFP-LP and gTCS1-GFP-RP. PCR products were firstly cloned into the pCR8/GW/TOPO TA cloning vector (Invitrogen) using TOPO enzyme. The TCS1 genomic sequence was then subcloned into the pMDC107 vector to generate the construct pTCS1:TCS1-GFP. The 2164bp promoter sequence of TCS1 was amplified using the primers TCS1pro-LP and TCS1pro-RP. PCR products were cloned into the pCR8/GW/TOPO TA cloning vector (Invitrogen) using TOPO enzyme. The TCS1 promoter was then subcloned into the pMDC164 vector to generate the transformation plasmid pTCS1:GUS. The plasmids 35S:TCS1, pTCS1:TCS1-GFP and pTCS1:GUS were transferred into tcs1-2 or Col-0 plants using Agrobacterium GV3101, and medium with hygromycin (30μg/mL) was used to select transgenic plants.
Trichome branches on the first pair of Col-0 and tcs1-1 leaves were counted at 15 days after germination (DAG). Leaves and etiolated hypocotyls of wild-type and tcs1-1 mutants were fixed in a solution (formalin, acetic acid, ethanol and H2O in a ratio of 1: 0.5: 4.75: 3.75) for 24 hours, dehydrated with a graded ethanol series and dried at critical point in liquid CO2. Samples were coated with gold and then observed in an S-4160 Field Emission Scanning Electron Microscope (SEM) (Hitachi).
To determine the effect of TCS1 on cortical microtubules, the microtubule-disrupting drug oryzalin (3,5-dinitro-N4, N4-dipropylsulfanilamide; Sigma-Aldrich) was applied to trichomes and hypocotyl epidermal cells of the wild type (Col-0) and tcs1-1 for specific times. The microtubule-stabilizing drug taxol (paclitaxel, Sigma-Aldrich) was used to treat trichomes of the wild type and tcs1-1.
To quantify the numbers of cortical microtubules in trichome and hypocotyl cells, the ImageJ software was employed. A line of fixed length (10 μm or 20 μm) perpendicular to the orientation of the most cortical microtubules was drawn, and the number of cortical microtubules across the line was counted. At least 10 cells from each treatment were used, and four lines of fixed length were drawn for each cell. The average number of cortical microtubules before and after treatments was calculated. The Student’s test was used to analyze the significance of the difference.
Samples (pTCS1:GUS) were putted into a GUS staining solution [0.1% Nonidet P-40, 1 mM 5-bromo-4-chloro-3-indolyl-b-D-glucuronic acid, 10 mM EDTA, 100 mM Na3PO4 buffer, and 3 mM each K3Fe(CN)6/K4Fe(CN)6] and incubated at room temperature for 6 hours. After GUS staining, 70% ethanol was used to remove chlorophyll.
GFP fluorescence in cells of trichomes and hypocotyls was detected using a Zeiss LSM710 META confocal microscope. GFP was observed using wave lengths of 510 to 530 nm. To study the co-localization of TCS1 and microtubules, we crossed pTCS1:TCS1-GFP transgenic plants with mCherry-TUB6 expressing plants. Seeds were germinated on ½ Murashige and Skoog (MS) medium supplemented with 0.9% agar and 1% glucose. Leaves of 6-day-old mCherry-TUB6;pTCS1:TCS1-GFP seedlings were observed under a spinning disk confocal microscope equipped with lasers for GFP and mCherry (Intelligent Design).
Leaves, stems, cotyledons and roots from 12-day-old seedlings were collected to isolate total RNAs using an RNeasy Plant Mini kit (TIANGEN). Reverse transcription (RT)-PCR was performed using Superscript III reverse transcriptase (Invitrogen). ACTIN2 mRNA was an internal control. The specific primers used for RT-PCR are shown in S1 Table.
The coding sequence of TCS1 was cloned into NotI and SalI sites of the bait vector pDBleu (Invitrogen) and the prey vector pEXP-AD502 (Invitrogen) to generate TCS1-BD and TCS1-AD constructs, respectively. The specific primers for TCS1-BD and TCS1-AD were TCS1-Y2H-SalI-LP and TCS1-Y2H-NotI-RP (S1 Table). The coding sequence of KCBP was cloned into NotI and SalI sites of the bait vector pDBleu (Invitrogen) and the prey vector pEXP-AD502 (Invitrogen) to generate KCBP-BD and KCBP-AD constructs. The specific primers for KCBP-BD and KCBP-AD constructs were KCBP-Y2H-SalI-LP and KCBP-Y2H-NotI-RP (S1 Table). The prey and bait plasmids were co-transformed into the yeast strain PT69-4A to investigate their interactions.
The coding sequence of TCS1 was cloned into the vector pMAL-C2 to generate the MBP-TCS1 construct. The specific primers for the MBP-TCS1 construct were MBP-TCS1-LP and MBP-TCS1-RP (S1 Table). The coding sequence of KCBP was cloned into the vector pGEX-4T-1 (Amersham-Pharmacia) to generate the GST-KCBP construct. The specific primers for the GST-KCBP constructs were GST-KCBP-LP and GST-KCBP-RP (S1 Table). The coding sequence of AN was subcloned into the vector pET-NT to generate the AN-His construct. The specific primers for the AN-His construct were AN-His-LP and AN-His-RP (S1 Table). The MBP-TCS1, GST-KCBP and AN-His plasmids were transformed and expressed in E. coli Rosetta (DE3).
To investigate protein-protein interaction, we performed the pull-down assay. Bacterial lysates containing ~15 μg of MBP-TCS1 or MBP-TCP14 fusion proteins were mixed with ~20 μg of tubulins. Bacterial lysates containing ~15 μg of MBP-TCS1 or MBP proteins were mixed with `~20 μg of AN-His fusion proteins. Amylose resin (30 μL; New England Biolabs) was added into each combined solution with gently rocking at 4°C for 1 h. Bacterial lysate containing ~20 μg of GST-KCBP was mixed with lysate containing ~15 μg of MBP-TCS1 or MBP. Glutathione Sepharose 4B (30 μL;GE Healthcare) was added into each combined solution with gently rocking at 4°C for 1 h. The TGH buffer (1× Complete Protease Inhibitor cocktail [Roche], 150 mM NaCl, 10% glycerol, 1.5 mM MgCl2, 50 mM HEPES, pH 7.5, 1 mM phenylmethylsulfonyl fluoride, 1 mM EGTA, pH 8.0, and 1% Triton X-100) was used to wash beads for five times. The isolated proteins were then analyzed by 10% SDS-polyacrylamide gels and determined by immunoblot analysis with anti-MBP, anti-tubulin, anti-GST, and anti-His antibodies (Abmart), respectively.
For the microtubule cosedimentation assay, different concentrations of MBP-TCS1 were added to paclitaxel-stabilized microtubules in the PEMT buffer (1 mM MgCl2, 1 mM EGTA, 100 mM PIPES, and 20 μM taxol, pH 6.9). After incubation at 25°C for 30 min, the samples were centrifuged at 100,000g at 25°C for 30 min to separate supernatants and pellets. They were then analyzed by 10% SDS-PAGE and determined by staining the gels with Coomassie Brilliant Blue R 250.
For the microtubule polymerization assay, different concentrations of MBP-TCS1 were added to 20 μM tubulin solution in the PEM buffer (1 mM MgCl2, 1 mM EGTA, 1 mM GTP, and 100 mM PIPES, pH6.9). The polymerization was investigated turbidimetrically by absorbance at 350 nm with a 0.4-cm path quartz cell at 37°C in a DU-640 spectrophotometer (Beckman Coulter, Fullerton, CA). 1 μM MBP was used as a negative control. The data was recorded from time 0 to 35min, when the turbidity in all samples did not increase any more.
For the observation of microtubule assembly, 1μM MBP-TCS1, 20μM Rhodamine labeled tubulins and 1mM GTP were incubated at 37°C for 30 min. The microtubule polymerization was then stopped using 1% glutaraldehyde. Spinning-Disc Confocal Microscopy Imaging was performed on an Olympus IX81 inverted microscope equipped with a Yokogawa spinning-disc confocal head (Yokogawa Electric) and an Andor iXon charge-coupled device camera (Andor Technology). 1μM MBP was used as a negative control. Images were captured using Andor iQ software, version 1.1 (Andor Technology), and processed using ImageJ software.
The coding sequence of KCBP was cloned into the XmaI and SpeI sites of the pCAMBIA1300-221-Myc vector to generate the transformation plasmid 35S:Myc-KCBP. The specific primers used for 35S:Myc-KCBP construct were MYC-KCBP-XmaI-LP and MYC-KCBP-SpeI-RP (S1 Table). The 35S:Myc-KCBP plasmid was transferred into tcs1-2 plants using Agrobacterium GV3101, and hygromycin (30μg/mL)-containing medium was used to select transformants. Total proteins from pTCS1:TCS1-GFP;35S:Myc-KCBP and 35S:GFP;35S:Myc-KCBP were isolated with the extraction buffer (1× Complete protease inhibitor cocktail, 50 mM Tris/HCl, pH 7.5, 2% Triton X-100, 1 mM EDTA, 150 mM NaCl, 20 mg/mL MG132, and 20% glycerol) and mixed with GFP-Trap-A (Chromotek) for 1 h at 4°C. Beads were washed four times with the wash buffer (1×Complete protease inhibitor cocktail, 50mMTris/HCl, pH7.5, 150mM NaCl, and 0.1% Triton X-100). The immunoprecipitates were analyzed by 10% SDS-polyacrylamide gel and determined by immunoblot analysis with anti-GFP (Abmart) and anti-Myc (Abmart) antibodies, respectively.
Arabidopsis Genome Initiative locus identifiers for the genes mentioned in this article are as follows: AT1G19835 (TCS1), AT5G65930 (KCBP), and AT1G01510 (AN).
|
10.1371/journal.pmed.1002714 | Effectiveness and treatment moderators of internet interventions for adult problem drinking: An individual patient data meta-analysis of 19 randomised controlled trials | Face-to-face brief interventions for problem drinking are effective, but they have found limited implementation in routine care and the community. Internet-based interventions could overcome this treatment gap. We investigated effectiveness and moderators of treatment outcomes in internet-based interventions for adult problem drinking (iAIs).
Systematic searches were performed in medical and psychological databases to 31 December 2016. A one-stage individual patient data meta-analysis (IPDMA) was conducted with a linear mixed model complete-case approach, using baseline and first follow-up data. The primary outcome measure was mean weekly alcohol consumption in standard units (SUs, 10 grams of ethanol). Secondary outcome was treatment response (TR), defined as less than 14/21 SUs for women/men weekly. Putative participant, intervention, and study moderators were included. Robustness was verified in three sensitivity analyses: a two-stage IPDMA, a one-stage IPDMA using multiple imputation, and a missing-not-at-random (MNAR) analysis. We obtained baseline data for 14,198 adult participants (19 randomised controlled trials [RCTs], mean age 40.7 [SD = 13.2], 47.6% women). Their baseline mean weekly alcohol consumption was 38.1 SUs (SD = 26.9). Most were regular problem drinkers (80.1%, SUs 44.7, SD = 26.4) and 19.9% (SUs 11.9, SD = 4.1) were binge-only drinkers. About one third were heavy drinkers, meaning that women/men consumed, respectively, more than 35/50 SUs of alcohol at baseline (34.2%, SUs 65.9, SD = 27.1). Post-intervention data were available for 8,095 participants. Compared with controls, iAI participants showed a greater mean weekly decrease at follow-up of 5.02 SUs (95% CI −7.57 to −2.48, p < 0.001) and a higher rate of TR (odds ratio [OR] 2.20, 95% CI 1.63–2.95, p < 0.001, number needed to treat [NNT] = 4.15, 95% CI 3.06–6.62). Persons above age 55 showed higher TR than their younger counterparts (OR = 1.66, 95% CI 1.21–2.27, p = 0.002). Drinking profiles were not significantly associated with treatment outcomes. Human-supported interventions were superior to fully automated ones on both outcome measures (comparative reduction: −6.78 SUs, 95% CI −12.11 to −1.45, p = 0.013; TR: OR = 2.23, 95% CI 1.22–4.08, p = 0.009). Participants treated in iAIs based on personalised normative feedback (PNF) alone were significantly less likely to sustain low-risk drinking at follow-up than those in iAIs based on integrated therapeutic principles (OR = 0.52, 95% CI 0.29–0.93, p = 0.029). The use of waitlist control in RCTs was associated with significantly better treatment outcomes than the use of other types of control (comparative reduction: −9.27 SUs, 95% CI −13.97 to −4.57, p < 0.001; TR: OR = 3.74, 95% CI 2.13–6.53, p < 0.001). The overall quality of the RCTs was high; a major limitation included high study dropout (43%). Sensitivity analyses confirmed the robustness of our primary analyses.
To our knowledge, this is the first IPDMA on internet-based interventions that has shown them to be effective in curbing various patterns of adult problem drinking in both community and healthcare settings. Waitlist control may be conducive to inflation of treatment outcomes.
| Global estimations continue to show increasing morbidity, mortality, and social harm caused by all types of problem drinking.
Face-to-face brief interventions for problem drinking are effective but rarely used.
Internet-based interventions could overcome this treatment gap.
We investigated effectiveness and moderators of treatment outcomes in internet-based interventions for adult problem drinking.
We conducted a one-stage individual patient data meta-analysis (IPDMA). This is, to our knowledge, the first study to identify moderators at the participant, intervention, and study design levels that are associated with treatment outcomes in internet-based interventions for adult problem drinking.
Our IPDMA included 14,198 adults at baseline from 19 randomised controlled trials who exhibited various profiles of problem drinking. We obtained posttreatment data for 8,095 participants.
Our results show that internet-based alcohol interventions in both community and healthcare populations are effective in reducing mean weekly alcohol consumption and in achieving adherence to low-risk drinking limits.
We did not find differences in impact related to drinking profiles, meaning that people exceeding risk limits to a smaller or a larger degree benefited from the interventions, as did binge-only drinkers. Human-guided interventions showed a stronger impact on treatment outcome than fully automated ones, but waitlist design controls may inflate outcomes.
The health gains of internet-based alcohol interventions could be substantial, because such programmes can reach high numbers of problem drinkers by virtue of their swift entry procedures and their easy scalability.
Future research should seek to identify categories of people for whom such interventions work best, to analyse how the interventions work and to determine what delivery contexts are most favourable. It should explore which patient populations could benefit most from referral to unguided forms and which would be more amenable to guidance by GPs or other professionals.
| Global estimations continue to show increasing physical and psychological morbidity, all-cause and specific-cause mortality, and social harm deriving from all types of alcohol misuse. Usually, a positive and linear association is seen between increased consumption and related health risks [1]. A number of factors underlie this mounting health burden. These include increases in the prevalence of alcohol consumers due to population growth and societal ageing, an absolute increase in adult alcohol consumption due to greater wealth and wider acceptance of alcohol use, and escalating alcohol use amongst women and the elderly. At the same time, there are growing insights into health risks connected with even minimal levels of alcohol consumption [2,3].
Brief alcohol interventions (BAIs) in primary care and community settings have been found clinically and cost-effective, with effect sizes in the small to moderate range, for reducing both hazardous drinking (which increases the risk of physical or psychological harm) and harmful drinking (which has already caused some damage) [4]. Together, their target groups are referred to as ‘problem drinkers’ to distinguish them from drinkers with alcohol use disorders, for whom more intensive treatments are recommended [5]. Problem drinkers account for the highest prevalence of alcohol misuse. Based on accumulated evidence, many national and professional guidelines now recommend brief interventions for problem drinkers in primary care settings and among community populations [6]. These interventions are comprised mostly of brief single or multiple sessions (up to six) and are based on personalised normative feedback (PNF) [7] or combinations of PNF, motivational interviewing (MI) [8], cognitive-behavioural therapy (CBT) [9], or behavioural self-control (BSC) principles [10]. Despite the ample evidence available, the actual impact of BAIs on curbing the prevalence of problem drinking in the wider population has been disappointingly low. The main factors in the weak impact include problems with implementation, as relatively few healthcare professionals actually administer BAIs; in addition, only a small proportion of patients who might benefit are actually offered BAIs, and even fewer accept the offer [11].
Internet-based alcohol interventions (iAIs) may overcome some of these problems by virtue of their low-threshold accessibility, their high scalability, and their acceptability to problem drinkers, as was recently echoed by McCambridge and Saitz [11]. Major advantages of iAIs, as perceived by many problem drinkers, are reduced stigma and greater comfort about disclosing drinking problems. The majority of iAIs are based on manualised therapeutic principles similar to those in BAIs. They are offered in unguided and guided formats. Unguided iAIs are fully automated interventions that participants can perform without human guidance. Guided interventions provide human support to guide participants through the intervention, mainly via asynchronous secure email contact [12]. The support may come from health professionals or trained volunteers. Meta-analytic studies have shown that unguided iAIs, in particular, are now used on a wider scale than conventional BAIs [13]. They have been found clinically effective (small effects) in reducing mean weekly adult alcohol consumption as compared with controls [14]. As a result, iAIs have been incorporated into some clinical guidelines for treating problem drinking in primary care [15].
All this notwithstanding, various uncertainties still surround the evidence base for iAIs. First of all, still little is known about whether women and older people derive benefits comparable to those seen for male and younger problem drinkers. Such knowledge is important in view of the rising prevalence rates of problem drinking among women and the elderly and their underrepresentation in many intervention studies [16]. Secondly, problem drinking actually embraces several different drinking profiles, and only a few iAI studies have investigated whether these might moderate treatment outcomes [17]. Such profiles include exceeding the advised weekly alcohol limits to a moderate (‘regular drinking’) or a serious degree (‘heavy drinking’) and ‘binge-only drinking’, whereby alcohol users episodically exceed the maximum advised intakes per drinking occasion. Such divergent drinking profiles may or may not necessitate different interventions. Thirdly, there is the question of whether guided iAIs are more effective than unguided ones—a finding reported for CBT-based internet interventions for common mental disorders such as depression [18]. A related question is whether iAI treatment outcomes might vary according to the therapeutic orientation of the intervention.
The few moderator analyses conducted to date had a common limitation: they were statistically underpowered to properly address such questions [14]. To overcome this major problem, we conducted an individual patient data meta-analysis (IPDMA) that boosted the number of participants studied and thereby the statistical power. That enabled us to better evaluate the overall effectiveness of iAIs in reducing alcohol consumption, as well as to explore statistically significant differences within the data by performing moderator analyses on treatment outcomes, with a focus on participant, intervention, and study design characteristics.
PsycINFO, Science Citation Index Expanded, Social Sciences Citation Index, Arts and Humanities Citation Index, CINAHL, PubMed, and EMBASE were searched up to 31 December 2016. All papers retrieved were evaluated by independent assessors (HR, EK, or NBo) (for search string, see S1 Data).
Randomised controlled trials (RCTs) were eligible if they (1) studied people aged ≥18 with quantifiable levels of alcohol consumption that exceeded recommendations for low-risk drinking; (2) compared an iAI with a control condition (e.g., assessment only, waitlist, or minimal intervention); (3) studied an iAI based on therapeutic principles such as PNF, BSC, CBT, MI, or combinations thereof; and (4) studied either an unguided or a guided intervention or both. RCTs in populations of students or pregnant women were excluded. Primary authors of identified trials were asked to provide their raw RCT data for a set of pre-identified variables (HR/EK, S2 Data and see S4 Data for data access contact list of original studies) and were queried as to whether they were aware of ongoing RCTs that met our inclusion criteria; two more RCTs were thus identified [19,20]. No study protocol for this study has been developed.
Five criteria from the Cochrane Collaboration risk-of-bias assessment tool were applied (by EK, HR, and NBo): (1) adequate random sequence allocation, (2) concealment of allocation to the different conditions, (3) blinding of participants and therapists to the study condition, (4) blinding of assessors to outcomes, and (5) handling of missing data [21].
The potential differences in treatment outcomes between the trials included and those that could not be included in preparing our IPDMA were assessed with a conventional meta-analysis (Comprehensive Meta-Analysis, version 3.3.070; S3 Data).
Results are reported in accordance with the Preferred Reporting Items for Systematic Review and Meta-Analyses for IPD (S1 PRISMA Checklist) [34].
Fig 1 illustrates the selection process for the trials included in our IPDMA. We identified 183 full papers, from which 24 eligible RCTS were found, five of which [35–39] could not be included (all involving unguided iAIs) because authors did not respond to our invitation.
Table 1 shows the characteristics of the 19 included RCTs (26 comparisons). Most trials applied the full Alcohol Use Disorders Identification Test (AUDIT) (n = 9, cutoff ≥8) [40] or AUDIT-C scales (n = 4, cutoff ≥4 or ≥5) [23] as inclusion criteria. Four RCTs used cutoff thresholds based on daily or weekly low-risk drinking recommendations; the Fast Alcohol Screening Test (FAST) was applied in two trials [41]. Participants were recruited either directly from the community (n = 12 trials), from healthcare settings (n = 4), or from work settings (n = 3). Eight trials employed a minimal-intervention control design, six trials applied assessment-only control, and five included a waitlist-control comparator. Eleven trials estimated the effects of multiple-session iAIs, seven studied single-session iAIs, and one study included both types. Twelve investigated effects of therapeutically integrated iAIs and seven studied PNF-only interventions. Most comparisons (n = 19) involved unguided iAIs; eight involved human-guided interventions. The first post-intervention assessment occurred in most trials (n = 15) between 1 and 3 months after treatment, in three trials at 6 months, and in one study at 12 months. A total of N = 14,198 participants was included, out of the 17,545 participants in the 24 identified trials (a 79.77% inclusion rate).
Of the total of 14,198 enrolled participants, 8,095 provided post-intervention outcome data (complete cases, Table 2). The mean age of the overall sample was 40.7 (SD = 13.2) and the sample was rather evenly divided by gender (47.6% women, 52.4% men). Some 51.9% of participants had tertiary education, 74.8% had paid employment, and 56.7% were in partner relationships. The mean weekly SU level at baseline was 38.1 (SD = 26.9). Most problem drinkers (80.1%, SUs 44.7, SD = 26.4) could be categorised as regular drinkers and 19.9% (SUs 11.9, SD = 4.1) as binge-only drinkers. Regular drinkers could be distinguished into heavy drinkers (34.2%, SUs 65.9, SD = 27.1) and non-heavy drinkers (65.8%, SUs 23.7, SD = 10.6). Heavy drinkers were found in both unguided and guided iAIs (34% and 30%, respectively). The mean full AUDIT score (n = 9 trials) was 15.0 (SD = 6.8), indicating hazardous or harmful alcohol use [40]. Of the participants for which a full AUDIT score was available, 22.2% (n = 678) scored above 20, indicating a risk of alcohol dependence. Missing SU scores at baseline were virtually nil (0.4%). Missing data at the first post-intervention assessment for the primary outcome were considerable (43%), predominantly resulting from study dropout, which was not entirely random: participants under age 55 and those with baseline heavy-drinking profiles dropped out significantly more than others.
The quality of the RCTs was relatively high (Fig 2 and S1 Table). All but one scored high-risk on the blinding of participants, which was expected, as this criterion is difficult to meet for behavioural change trials. All trials included ITT analyses, but seven had a high bias risk in terms of high study dropout (over 30%).
The overall difference in mean weekly alcohol reduction was significant and in favour of the iAI condition (b = −5.02 SUs, 95% CI −7.57 to −2.48, p < 0.001; see Table 3). We identified four outliers (RCTs in which the 95% CI did not overlap with that of our pooled effect size) [43,52,53,59]. Removal of these outliers altered the result only slightly (b = −4.81 SUs, 95% CI −6.69 to −2.93, p < 0.001). For two trials (Khadjesari 2014, N = 1,330, and Sinadinovic 2014, N = 633), we had estimated the mean weekly SUs on the basis of the first two questions of the AUDIT-C; removal of those trials likewise only slightly altered the result (b = −5.74 SUs, 95% CI −8.55 to −2.92, p < 0.001).
iAI participants also had a significantly greater likelihood of TR than controls (OR = 2.20, 95% CI 1.63–2.95, p < 0.001, NNT = 4.15, 95% CI 3.06–6.62), which remained after removal of the outliers (OR = 2.15, 95% CI 1.67–2.77, p < 0.001; see Table 4) or the two AUDIT-estimated trials (OR = 2.50, 95% CI 1.81–3.45, p < 0.001; see Table 4). Follow-up periods in the analysis were different, but they were not associated with outcomes (primary p = 0.41, secondary p = 0.12).
In the first sensitivity analysis, we checked the extent to which the results would be different if we used a two-stage approach instead of a one-stage approach. The second sensitivity analysis involved the inclusion of all participants according to the ITT principle by use of a multiple imputation strategy.
The third sensitivity analysis concerned the MAR assumption that is commonly used to deal with missing outcome data. All three of our sensitivity analyses confirmed the results of our main analysis for the overall effect and for most of the moderating effects of participant-, intervention-, and study-level characteristics for the primary and secondary outcomes. This appears to verify the robustness of our findings (see Tables 3 and 4 for the results of the two-stage approach and S2 Table and S3 Table, in which the results of the multiple imputation analyses are presented). Some minimal differences for moderators were seen in the multiple imputation analyses. The moderating role of gender and education for the primary outcome lost significance after multiple imputation. For the secondary outcome, the moderating role of single versus multiple sessions became significantly different in favour of multiple sessions, while intervention in the work setting became effective (p = 0.041), as was the case for assessment-only interventions. The contrast between PNF versus integrated iAIs became nonsignificant, as was the case for the contrast between unguided iAIs with WLCs versus other types of control conditions. Thus, in some cases these made our moderator analysis appear more conservative, while in some other cases the MI was more conservative.
Fig 3 depicts the results of the third, MNAR sensitivity analysis, which assessed departure from the MAR assumption. The figure shows estimates (and 95% CIs) of the overall intervention effect on our primary outcome variable, SU, for differing values of δ. The value of δ = 0 corresponds to the MAR assumption, on which the results displayed in Table 3 are based. Positive (or negative) values of δ correspond to situations in which—in each study included in the IPDMA and in both the intervention and the control arms—the mean of unobserved scores for post-intervention SUs would be higher (or lower) than the observed post-intervention SUs, after adjustment for pre-intervention SUs. If MAR holds, the overall effect is estimated in the two-stage method at −4.80 SU. Fig 3 shows that if the post-intervention SUs of dropouts, adjusted for the pre-intervention SUs, were to be 35 SUs higher on average than the post-intervention SUs of participants (being about 1.4 SD above the pre-intervention SUs shown in Table 2), then the estimate of the overall effect would be −4.06 SUs (95% CI −6.25–1.87). If the mean post-intervention SU level of dropouts were to be lower than those of participants (negative value of δ), then the overall effect would be stronger; for instance, if δ = −20, then the estimated overall effect would be −5.32 SUs (95% CI −7.64 to −3.01). This sensitivity analysis leads us to conclude that our results would remain rather stable, even in the event of substantial deviations from the MAR assumption (see S1 Text).
Heterogeneity for the overall primary outcome was high and significant (I2 = 89.6%, CI 78.4%–95.2%, p < 0.001) and for the secondary outcome as well (I2 = 78.2%, CI 56.3%–89.9%, p < 0.001). It could be partly explained by the identified outliers, as it dropped from high to moderate for the primary outcome (I2 = 55.5%, CI 16.2%–80.3%, p < 0.001) and from high to small for the secondary outcome (I2 = 30%, CI 0%–69.1%, p < 0.001) after removal of the outliers from the analyses (Tables 3 and 4).
The conventional meta-analysis (24 trials, 34 comparisons) was based on our search up to 31 December 2016 and included additional data from two RCTs published in 2017 [20,47]. It revealed a small significant difference in mean weekly SUs at the first follow-up in favour of iAI participants as compared with controls (Hedges’ g = 0.26, 95% CI 0.17–0.34, p < 0.001; Fig 4, forest plot of results of conventional meta-analysis). There was significant, moderate heterogeneity, indicating that the effect was greater in some trials than in others (I2 = 65%, p < 0.001; 95% CI 49–75). No significant difference in effect size was observed between the included and non-included RCTs in the IPDMA in terms of the primary outcome (SU reduction).
There were indications of publication bias, based on a visual inspection of the funnel plot (see S1 Fig) and Egger test (intercept 1.559, p < 0.05), but there was no publication bias observed on the basis of Duval and Tweedie’s trim-and-fill procedure (random-effects model). We could not conduct a conventional meta-analysis for our secondary outcome, as only a limited number of studies reported on it. In S3 Data, this conventional meta-analysis has been expanded with two further eligible studies published between 1 January 2017 and 30 May 2018 that could not be included in our IPDMA. Our aim here was to explore whether more recent studies could potentially alter our IPDMA results; as they did not significantly alter the effect size in our conventional analysis, we believe this confirms the robustness of our analysis.
This study found that participants treated in iAIs showed a higher mean weekly decrease of 5.02 SUs of alcohol consumption and a greater likelihood of favourable TR (OR 2.20) than controls. Women decreased their mean weekly alcohol consumption significantly less than men (around 2 SUs). Our sensitivity analysis confirmed our assumption that this difference was not an artefact of the higher cutoff thresholds for men than for women at study inclusion (leaving women less space for alcohol reduction) [60]. More highly educated participants reduced their mean weekly consumption significantly less than lesser educated ones (around 2 SUs). This result differs from the few studies that have reported on education as a moderator of iAI treatment outcomes; these showed either improved outcomes for more educated participants [61] or no such impact [62]. For gender and education as moderators of the primary outcome, our sensitivity analyses pointed in similar directions to the outcomes of our main analysis, although the results were no longer significant. In our study, age was found to have moderated TR, with participants above 55 showing greater likelihood of post-intervention adherence to low-risk drinking recommendations than younger people. None of the other participant characteristics moderated treatment outcomes. Internet interventions appear effective when applied in community and healthcare settings, but effectiveness in work settings is still inconclusive.
Guided iAIs yielded significantly better results than unguided ones for both treatment outcomes. iAIs based solely on PNF showed a lower likelihood of TR than iAIs based on integrated therapeutic principles. Waitlist control moderated both types of treatment outcomes, with iAIs in WLC studies showing significantly better outcomes in terms of both SU reduction and TR than those in otherwise-controlled studies. It thus appears that iAI treatment outcomes could have been overestimated in studies in which WLC groups were applied as comparators. One possible explanation for such higher effect sizes in WLC studies would be that problem drinkers allocated to waiting lists might delay their alcohol reduction because they anticipate treatment soon. In contrast, people in other types of control groups might have already found alternative support by the time of the follow-up assessment, thus potentially reducing their alcohol consumption more than WLC controls. By the same token, such tendencies could deflate effect sizes in non-WLC studies [63].
The overall greater reduction of 5.02 SUs of alcohol consumption seen here in iAI treatment participants as compared with controls was higher than the 2.2 SUs we found in our earlier, conventional meta-analysis [14]. One potential explanation for that difference is the higher number of guided iAI studies included in the present IPDMA; these showed higher treatment outcomes than unguided ones. Our current finding is comparable to the 5.61-SU reduction by adult iAI participants over controls reported in the conventional meta-analysis by Kaner and colleagues [16]. Our results compare quite favourably with outcomes of patients treated in primary care settings with brief guided face-to-face interventions, who showed decreases from 2 to 4 SUs [64,65]. We were also able to assess TR in terms of NNT (4.15). Due to data limitations, conventional meta-analyses have not been able to report on NNTs or on potential moderators of iAI treatment such as gender, age, and drinking profiles [16].
To the best of our knowledge, this is the first IPDMA to test the impact of iAIs and their moderators on treatment outcomes with adequate statistical power. The included RCTs had a low overall risk of methodological bias. Our results appear robust after comparison with our two-stage IPDMA results, as well as with those from our multiple imputation analysis and those from our conventional meta-analysis. The ANCOVA model that underlies our IPDMA implicitly relies on the MAR assumption, allowing dropout, which was 43% in our study, to depend on baseline consumption level. Although it cannot be ruled out that dropout was actually attributable to characteristics not included in the model, our MNAR sensitivity analysis suggested that the estimate of the overall effect would be reasonably stable against moderate deviations from the MAR assumption. The generalisability of our results to people in real-life settings might be hampered by poor assessment of ethnicity and by the focus on studies from high-income countries. In addition, only a small number of studies addressed effects of iAIs administered in care settings other than the community (such as in primary care practices, emergency departments, or workplaces). Another limitation is that all studies applied self-reported alcohol consumption measures, which is possibly a source of social desirability bias [66]. We also observed high heterogeneity in our analyses, and it could be explained only partly by excluding outliers or by some of the subgroup analyses that we conducted. Hence, the moderating factors we identified offer only partial clarification of moderating influences on treatment outcome. We were bound, of course, by the available data. Other moderators, such as self-efficacy or participants’ preference for iAIs over other types of interventions, cannot be ruled out [67]. Longer-term outcomes of iAIs could not be assessed, as few studies addressed them.
Both men and women from different age groups and with different drinking profiles, including heavy drinking and binge-only drinking, can benefit from iAIs, and in particular from the therapeutically integrated ones as opposed to PNF-only interventions. Participants in iAIs reduced their mean alcohol consumption from 38.1 to 32.9 SUs per week, and they had a substantially higher probability of posttreatment adherence to low-risk drinking recommendations. The fact that heavy drinkers decreased their alcohol consumption by amounts similar to those of non-heavy drinkers has favourable implications, as the health impact of a given reduction is greater at higher levels of alcohol consumption [68]. Despite the finding that many participants were still consuming beyond low-risk limits at posttreatment, the population health gains could nevertheless be substantial, in view of the high number of participants that can be reached with iAIs and the positive relationship between decreased alcohol consumption and the lower risks of physical and mental health disorders in the long term. These include earlier-onset dementia [69], several types of cancer [70], cardiovascular diseases [3] (Wood 2018), and depression and anxiety [68,69,71]. iAIs have great scaling potential, partly by virtue of their swift entry procedures for patients and the relatively low cost of repeated reuse, especially if unguided. For many people, iAIs could serve as a first step towards changing their problem-drinking behaviours and towards more intensive treatment, if needed.
In view of the constraints experienced with face-to-face BAIs in primary care settings, future studies should also explore various types of brief interventions, in order to gauge how problem drinkers in such settings can best be targeted. Those could be either face-to-face BAIs or iAIs, and the latter could be guided by general practitioners (GPs) or other professionals. For some patient populations, referral to unguided forms could be more beneficial [72]. More primary care studies are needed, however, including head-to-head comparisons of unguided versus guided versus face-to-face interventions. The same applies to the optimum treatment orientations and levels of intensity and duration [73,74]. As we have seen, not all treatment participants benefited from iAIs. We therefore need to better understand for which people such interventions work, how they work, and in what contexts (an approach also highlighted by Babor in 2008) [75,76]. A final observation is that some countries have now substantially lowered the advised limits for daily and weekly alcohol consumption, in response to mounting epidemiological evidence of health risks inherent in the conventional limits [77]. A threshold not exceeding 10 SUs of weekly alcohol consumption for both men and women has been proposed [3]. Future studies should correspondingly adjust their sample inclusion criteria based on units of alcohol consumption.
|
10.1371/journal.pgen.1004065 | A PNPase Dependent CRISPR System in Listeria | The human bacterial pathogen Listeria monocytogenes is emerging as a model organism to study RNA-mediated regulation in pathogenic bacteria. A class of non-coding RNAs called CRISPRs (clustered regularly interspaced short palindromic repeats) has been described to confer bacterial resistance against invading bacteriophages and conjugative plasmids. CRISPR function relies on the activity of CRISPR associated (cas) genes that encode a large family of proteins with nuclease or helicase activities and DNA and RNA binding domains. Here, we characterized a CRISPR element (RliB) that is expressed and processed in the L. monocytogenes strain EGD-e, which is completely devoid of cas genes. Structural probing revealed that RliB has an unexpected secondary structure comprising basepair interactions between the repeats and the adjacent spacers in place of canonical hairpins formed by the palindromic repeats. Moreover, in contrast to other CRISPR-Cas systems identified in Listeria, RliB-CRISPR is ubiquitously present among Listeria genomes at the same genomic locus and is never associated with the cas genes. We showed that RliB-CRISPR is a substrate for the endogenously encoded polynucleotide phosphorylase (PNPase) enzyme. The spacers of the different Listeria RliB-CRISPRs share many sequences with temperate and virulent phages. Furthermore, we show that a cas-less RliB-CRISPR lowers the acquisition frequency of a plasmid carrying the matching protospacer, provided that trans encoded cas genes of a second CRISPR-Cas system are present in the genome. Importantly, we show that PNPase is required for RliB-CRISPR mediated DNA interference. Altogether, our data reveal a yet undescribed CRISPR system whose both processing and activity depend on PNPase, highlighting a new and unexpected function for PNPase in “CRISPRology”.
| CRISPR-Cas systems confer to bacteria and archaea an adaptive immunity that protects them against invading bacteriophages and plasmids. In this study, we characterize a CRISPR (RliB-CRISPR) that is present in all L. monocytogenes strains at the same genomic locus but is never associated with a cas operon. It is an unusual CRISPR that, as we demonstrate, has a secondary structure consisting of basepair interactions between the repeat sequence and the adjacent spacer. We show that the RliB-CRISPR is processed by the endogenously encoded polynucleotide phosphorylase enzyme (PNPase). In addition, we show that the RliB-CRISPR system requires PNPase and presence of trans encoded cas genes of a second CRISPR-Cas system, to mediate DNA interference directed against a plasmid carrying a matching protospacer. Altogether, our data reveal a novel type of CRISPR system in bacteria that requires endogenously encoded PNPase enzyme for its processing and interference activity.
| Listeria monocytogenes is a gram-positive foodborne pathogenic bacterium that has evolved two distinct lifestyles: a saprophytic one, primarily in decaying vegetation and a parasitic one in the tissues of mammals and birds, causing a disease known as listeriosis. Infection in humans starts by the ingestion of contaminated food products that deliver the bacteria in the intestinal lumen. In the course of the infection of susceptible individuals e.g. elderly and pregnant women, Listeria can cross three barriers of the organism: the intestinal, blood-brain and feto-placental barriers, causing meningitis, encephalitis and abortion. The main and best studied regulator that orchestrates the Listeria infectious process is PrfA (Positive regulatory factor A), a transcription factor that activates expression of the major known virulence genes [1]. In addition to protein determinants contributing to infection, Listeria possesses a virulence gene repertoire that expands to non-coding RNA (ncRNAs) molecules [2]–[4].
Bacterial ncRNAs are key regulatory molecules of metabolic, physiological and pathogenic processes and can be generally classified in four groups: a) the RNA regulatory elements located in the 5′ untranslated regions (5′UTRs) which regulate the expression of the corresponding mRNAs through the binding of various factors, like proteins (e.g. CsrA) and small metabolites (riboswitches) or by sensing environmental cues like temperature (thermosensors); b) the trans-acting small RNAs (sRNAs) regulating one or several target mRNAs located elsewhere on the chromosome; c) the sRNAs that sequester RNA-binding proteins; and d) the antisense transcripts (asRNAs), which overlap and are complementary to their target genes in the same genomic locus [5]. A novel class of non-coding RNAs, named CRISPRs (clustered regularly interspaced short palindromic repeats) has been shown to mediate bacterial adaptive immunity against invading bacteriophages and conjugative plasmids. A CRISPR is defined by the alternating array of identical 20–40 nucleotides (nt) long repeat sequences, interspaced by non-repetitive spacer sequences. In the proximity of the locus, are usually found gene clusters called CRISPR-associated (cas) genes. Cas genes form 23–45 different gene families (depending on the classification), encoding diverse proteins with nuclease, helicase, integrase, polymerase or nucleotide-binding activities, which are involved in the different steps of CRISPR generation, maintenance, processing and the interference mechanism. Analysis of the various sets of cas genes has revealed that CRISPR-Cas system generally cluster into three basic types (Type-I, Type-II and Type-III) which are further divided into at least ten subtypes (Types IA–F, Types IIA–C and Types IIIA–B) [6], [7]. The first clue about CRISPR function was brought about by the discovery that the different spacers were homologous to bacteriophage and plasmid sequences [8]–[10]. It was thus hypothesized that CRISPRs could play a role in immunity against invading genetic elements, which was later experimentally demonstrated in several elegant studies, e.g. in Streptococcus thermophilus [11], Escherichia coli [12] and Staphylococcus epidermidis [13]. The mechanism of action underlying the whole process is still not entirely understood, but can be roughly divided in three major stages: i) CRISPR adaptation that occurs when bacteria first encounter the foreign invader after transformation, conjugation or transduction. The CRISPR system recognizes the foreign element and incorporates parts of its DNA into what becomes a new spacer in the CRISPR locus; ii) CRISPR expression that generates a long poly-spacer precursor RNA, which is then cleaved by Cas proteins producing smaller, mature RNAs (crRNAs). Each crRNA generally contains part of the repeat and a single spacer that serves as a guide for the sequence specific recognition of the foreign invader; iii) CRISPR interference mediated by mature target-specific crRNAs that, with the help of cas gene products, inactivate the foreign, bacteriophage or plasmid nucleic acid [14], [15].
In Listeria, 14 plasmids [16] and 11 bacteriophages [17] have been sequenced so far. Bacteriophages infecting Listeria belong to the Siphoviridae and Myoviridae families in the Caudovirales order. They are either temperate integrating in the host genome by site-specific recombination or virulent actively replicating and forming virion particles that subsequently lyse the host cell. Comparative genomic analysis of Listeria bacteriophages revealed that their genomes are highly mosaic, characterized by interspecies homology as well as homology to bacteriophages infecting Bacillus, Enterococcus, Clostridium and Staphylococcus [17], [18]. Prophages are considered as the major source of diversity within the Listeria genus [18] and can constitute up to 7% of the Listeria coding genes [19], [20].
Recently, CRISPR-Cas systems have started to be analyzed in Listeria [18], [21]. We previously described in L. monocytogenes strain EGD-e, a small CRISPR RNA (RliB) exhibiting five identical repeats interspaced by non-related spacer sequences of similar size. Strikingly, no cas genes were found either in the proximity of RliB or elsewhere in the L. monocytogenes EGD-e genome [22]. Despite the absence of Cas proteins, RliB is expressed and significantly upregulated in bacteria isolated from the intestinal lumen of gnotobiotic mice, in bacteria grown in the human blood, or bacteria exposed to hypoxia. More importantly, we showed that RliB is involved in L. monocytogenes virulence [4].
Here, we characterized the cas-less RliB-CRISPR by first determining its secondary structure and analyzing its processing. Furthermore, we undertook a search for RliB protein ligands, to address the molecular machinery underlying RliB processing in the absence of Cas proteins. By using two different protein affinity purification approaches, we showed that RliB binds and is a substrate for polynucleotide phosphorylase (PNPase), a bi-functional enzyme harboring a 3′ to 5′ exoribonuclease and 3′ polymerase activities [23]. Furthermore, we performed a global analysis of CRISPR-Cas systems in all sequenced Listeria genomes, revealing a striking ubiquity of the RliB-CRISPRs in L. monocytogenes strains. Surprisingly, RliB-CRISPRs are never associated with cas gene clusters and we could demonstrate that even in Listeria strains harboring a complete set of cas genes, RliB-CRISPRs are processed by PNPase. Finally, we carried out a functional assay for RliB-CRISPR and demonstrated it requires presence of the cas genes of a second CRISPR system to lower the acquisition frequency of a plasmid carrying the matching protospacer. Moreover, we show that PNPase is required for this DNA interference activity. Together, our data highlight a novel type of CRISPR system that relies on the activity of PNPase, highlighting a new role for this enzyme in bacteria.
In L. monocytogenes EGD-e, RliB is located between the genes lmo0509 and lmo0510 that encode a protein similar to phosphoribosyl pyrophosphatase and a hypothetical protein, respectively. Its primary sequence resembles a typical CRISPR. It is composed of 5 identical 29 nt repeat sequences (GTTTTAGTTACTTATTGTGAAATGTAAAT) interspaced by four 35–37 nt long spacer sequences (S1, S2, S3 and S4 in the Figure 1A). The spacer 3 (S3) has identity with the Listeria temperate bacteriophage B054 sequence and spacer 4 (S4) identity to Listeria virulent bacteriophage P70 sequence [17], [24]. We analyzed the secondary structure of the full length RliB that is detectable in vivo, using RNase V1 (specific for helical regions), RNase T2 (specific for unpaired nucleotides with a preference for adenines) and dimethylsulfate (which methylates N1 of adenine and N3 of cytosine) (Figure S1). The secondary structure of RliB, that explains most of the probing data, involves six stem-loop structures among which five contain a GUUUU motif within the loops, followed by a hairpin terminator at the 3′ end (Figure 1B). In contrast to CRISPR systems where the repeat sequences form independent and stable palindromic structures [25], the RliB hairpin structures are mostly formed by base pairings between the repeat sequences and the adjacent spacer sequence. These data suggest that RliB structure largely depends on the nature of the incoming spacer DNA.
We had previously noticed that RliB in L. monocytogenes EGD-e is processed to a smaller fragment [22]. Here, we examined this processing by Northern blot analysis of total RNA isolated from the wild type (WT) L. monocytogenes EGD-e and bacteria expressing RliB from a constitutive promoter (Phyper-RliB). We used probes complementary to the repeat (R), to each unique spacer (S1, S2, S3 and S4) and to the 3′ end of the molecule including the terminator region (T) (Figure 1C). All the probes allowed detection of a 400 nt fragment, which corresponds to the full length RliB molecule. The probes for S1, S2, S3 and R regions detected an additional 280 nt long fragment. The probes for S3 and S4 regions showed a minor 100 nt long fragment and the probe for S3 region a fragment smaller than 50 nt.
Altogether, our data show that the RliB-CRISPR has a secondary structure largely determined by the interactions between each repeat and the adjacent spacer and, despite the absence of Cas proteins, it is processed and exists under two major forms: i) a 400 nt full length RliB molecule and ii) a shorter form of RliB molecule, approximately 280 nt long, comprising spacers S1, S2 and S3.
Considering the complete absence of cas genes in L. monocytogenes EGD-e strain, we hypothesized that the RliB-CRISPR processing is governed by another bacterial ribonuclease. To identify which enzymes are involved in this process, we first searched for proteins that interact with RliB using the affinity purification method with a tagged RliB molecule. Given that addition of a tag may perturb the folding of the bait-RNA molecule and/or change its accessibility, resulting in a loss of interaction with its binding partners, we used two strategies using two different tags added either at the 5′ or at the 3′ end of the bait-RNA. The first affinity purification was performed with the 3′-biotinylated full length RliB (RliB-B) and a control RNA, the quorum sensing induced RNAIII from Staphylococcus aureus (Figure 2A). In the second approach, we used as a bait RliB tagged at the 5′ end with two hairpin structures constituting the “MS2 binding sequence” (RliB-MS2), i.e. the RNA binding sequence of bacteriophage MS2 coat protein (MS2) (Figure 2B). Structure probing using enzymes show that the MS2-tag did not change the structure of RliB (Figure S1). The RliB-B or RliB-MS2 RNAs were bound to streptavidin or MBP-MS2 coated beads, respectively and incubated with total Listeria cell extracts. After extensive washing of unspecific proteins, the bound fraction was eluted and loaded on SDS-polyacrylamide denaturing gel. To verify the integrity of the bait-RNA, we also analyzed the eluted tagged RNA using polyacrylamide-urea gel electrophoresis. For both experiments, we detected a single and major protein band of approximately 78 kDa, specific to RliB-bound elution fractions (RliB-B and RliB-MS2) (Figures 2A,B). The protein was identified by mass spectrometry to be the Listeria polynucleotide phosphorylase (PNPase) encoded by gene lmo1331 (pnpA), a bi-functional enzyme that acts as 3′-5′ exoribonuclease and a 3′-terminal polymerase [23], [26].
We then analyzed whether the interaction between RliB and PNPase is direct or requires another binding partner. The L. monocytogenes PNPase protein carrying 6 histidines at its C-terminal end was purified and binding experiments were carried out using gel retardation assays with in vitro transcribed P32-labeled full length RliB and increasing amount of the purified PNPase. Formation of a complex between RliB and PNPase was observed with 400 nM PNPase, showing that the interaction is direct and does not require another binding partner. To demonstrate the specificity of PNPase binding, competition experiments were done with various non-labelled RNAs. The addition of non-labelled RliB outcompeted the interaction between PNPase and P32-labeled RliB in contrast to the addition of a non-labelled control RNA from S. aureus (RsaA) that did not affect the complex formation (Figure 2C). Altogether, our results show that PNPase specifically interacts with RliB.
PNPase is a bifunctional enzyme, which in vivo acts primarily as a 3′ to 5′ exoribonuclease of single stranded target RNAs [26]. We investigated if RliB is a substrate of PNPase. We first verified the activity of the purified PNPase protein and performed an in vitro assay where a 37 nt P32-end labeled substrate RNA (RNA37) was incubated alone or with 200 nM purified PNPase (Figure 3A). The presence of PNPase resulted in the degradation of the 5′ end P32-labeled RNA37 while no cleavage reaction was observed using a P32-pCp 3′ end labeled RNA37 (results not shown). These data demonstrate that the purified protein is active and able to degrade single-stranded RNA substrates. Three non-labeled competitor RNAs were then added to the reaction; i) a non-labeled RliB; ii) a non-labeled control RNA (RsaA); iii) and the non-labeled RNA37 substrate. As expected, the addition of non-labeled RNA37 substrate decreased the cleavage reaction. Strikingly, the addition of 100 nM non-labeled RliB resulted in the loss of PNPase mediated RNA37 degradation, whereas addition of RsaA did not alter the degradation of RNA37, indicating that RliB acts as a competitive inhibitor of PNPase.
To investigate further the activity of PNPase on RliB, we incubated the full length 5′ end-labeled RliB with increasing concentrations of purified PNPase (Figure 3B). We observed on the gel the appearance of a band migrating around 270 nt, an RliB processing product generated by the PNPase-mediated degradation up to the stem-loop IV. This cleavage reaction was inhibited by the addition of the full length non-labeled RliB. Altogether, these data suggest that in vitro, PNPase is processing RliB until its exoribonuclease activity is stalled in the S4 repeat region.
To study the effect of PNPase on RliB in vivo, we constructed a pnpA deletion mutant (ΔpnpA) and compared by Northern blot the size of the RliB transcript in the ΔpnpA mutant and WT bacteria. In the absence of PNPase, two major bands migrating as 300 and 330 nt long RNAs, were observed. Upon complementation (ΔpnpA-pnpA), the RliB processing was restored, identical to that observed in the WT strain (Figure 3C). Our results thus strongly suggest that RliB is a substrate for PNPase in vivo.
CRISPR arrays are thought to evolve rapidly in prokaryotic genomes [14], [27], [28]. Therefore, we investigated the presence of RliB in other Listeria strains. For this, we searched for CRISPRs in 29 complete and 17 draft Listeria genomes (Table S1). As mentioned in the introduction, the highly diverse CRISPR-Cas systems are classified into three main types (I, II and III) each including several subtypes [6]. In Listeria, we found two types of CRISPR-Cas systems: i) CRISPR-Cas systems type-I (subtype I-B) with the cas operon composed of cas6-cas8a1-cas7-cas5-cas3-cas1, including also in some cases cas4, associated with the repeat sequence GTTTTAGTTACTTATTGTGAAATGTAAAT that is almost identical to the repeat of RliB-CRISPR; ii) CRISPR-Cas systems type-II (subtype II-A) associated with csn2-cas2-cas1-cas9 operon and the repeat sequence GTTTTGTTAGCATTCAAAATAACATAGCTCTAAAAC (Figure 4A).
CRISPR-I is present at the locus between lmo0517 and lmo0510 in 7 complete L. monocytogenes genomes, 10 draft L. monocytogenes genomes, in Listeria seeligeri and Listeria ivanovii (Figure 4B and Table S1) and it is always associated with a type-I cas operon located in close proximity. The CRISPR-II was detected between lmo2591 and lmo2596 in 6 complete L. monocytogenes genomes, 9 draft genomes and in Listeria innocua (Figure 4B, Table S1). The CRISPR-II is also found exclusively associated with type-II cas operon. The tight association of CRISPR-I and CRISPR-II with type-I and type-II cas genes, suggests that the function of those CRISPRs is dependent on the activity of the corresponding Cas protein machinery.
In contrast to CRISPR-I and CRISPR-II that are found in about 30% of the complete Listeria genomes, the RliB-CRISPR is present at the same genomic locus in all analyzed complete and draft L. monocytogenes genomes as well as in other Listeria species (Figure 4B, Table S1). This suggests a stronger selective pressure on this element relative to the cas-associated CRISPRs. In silico structure prediction performed on three representative RliB-CRISPRs carrying different number of repeats revealed a putative secondary structure that is highly similar to that experimentally determined in L. monocytogenes EGD-e strain (Figure S2). Cas operons have not been detected in the close proximity to the RliB-CRISPRs. Furthermore, 14 complete Listeria genomes completely lack cas genes. The number of repeats in RliB-CRISPRs range from 1 to 11 and does not correlate with the presence or absence of cas genes elsewhere in the genome (Table S3). Together, the conservation of RliB-CRISPRs among Listeria strains suggests that they may have a function despite the absence of Listeria Cas proteins.
Although RliB-CRISPR and CRISPR-I have different pattern of conservation the two systems share almost identical repeat sequences (Figure 4A). To investigate the correlation between the two systems, we compared their putative leader and upstream sequences. Multiple alignments of the DNA fragment preceding the identified RliB-CRISPR and CRISPR-I systems revealed a striking homology (Figure S3A). More interestingly, the putative leader sequences harbour a highly conserved sequence homologous to RpoD dependent promoter, that was previously reported upstream of RliB in the L. monocytogenes EGD-e strain [22] (Figure S3B). High homology of the repeats and the leader sequences of RliB-CRISPR and CRISPR-I systems suggest a possible close relationship between the two systems.
To investigate if the PNPase-mediated processing of RliB-CRISPR is specific to L. monocytogenes EGD-e strain or is more general, we examined if PNPase is also involved in RliB-CRISPR processing in Listeria strains containing a complete set of cas genes. We constructed a pnpA deletion mutant in the L. monocytogenes Finland strain (ΔpnpA-Fin), which carries a complete CRISPR-Cas system type I and in the L. monocytogenes EGD strain (ΔpnpA-EGD), which has a complete CRISPR-Cas system type II. We also constructed the deletion mutants for the RliB-CRISPR in the same strains (ΔrliB-CRISPR-Fin and ΔrliB-CRISPR-EGD, respectively). The RliB-CRISPR processing was examined by northern blot in the corresponding strains (Figure 5).
The RliB-CRISPR in EGD strain (RliB-CRISPR-EGD) is composed of 11 identical repeats and 10 spacer sequences among which spacers S2, S6, S7 and S8 share similarity to Listeria temperate bacteriophages B054, B025 and A006 (Figure 5A, 6). In the WT EGD strain, RliB-CRISPR is expressed as a 750 nt long RNA that is processed into shorter fragments with the major processed form being 280 nt long. In the absence of PNPase, the total amount of full-length RliB-CRISPR-EGD increased and the transcript processing changed compared to the WT bacteria, i.e. we observed additional bands with a major one of 700 nt (Figure 5B).
The RliB-CRISPR in the Finland strain (RliB-CRISPR-Fin) is composed of 12 identical repeats and 11 spacer sequences among which 8 spacers are shared with RliB-CRISPR-EGD. Spacers S1, S3, S5, S7, S8 and S9 show high similarity to sequences in Listeria bacteriophages P70, B025, B054 and A006 (Figure 5A, 6). The full-length RliB-CRISPR-Fin is expressed in the WT bacteria, as a 780 nt long RNA that it is processed to several shorter fragments with the 280 nt being again the most abundant form. In the absence of PNPase, RliB-CRISPR-Fin processing changed as additional bands are observed compared to the WT bacteria (Figure 5B).
Together, our results suggest that PNPase contributes to the RliB-CRISPR processing in vivo, independently of the presence of either CRISPR-Cas system type I in the L. monocytogenes Finland strain, or the presence of CRISPR-Cas system type II in the L. monocytogenes EGD strain.
We analyzed CRISPR spacers to compare the putative functions of cas-less RliB-CRISPR and cas-associated CRISPR-I and CRISPR-II. In total, we identified 978 spacers that correspond to 348 unique sequences (Table S2). These were used to search for the similarity with the sequences of all complete prokaryote, plasmid and virus genomes available in the Genbank as well as the sequences of integrated temperate bacteriophages (prophages) identified in complete Listeria genomes (Figure 6, Table S3).
We identified 142 (41%) spacers that share identity to bacteriophages known to infect Listeria species (Figure 6). They match sequences detected in 6 temperate (B054, B052, A118, A500, A006, PSA), 4 virulent phages (A115, P35, P70, P100) as well as 35 distinct prophages found in complete Listeria genomes (Figure S4). Overall, we found matching protospacers for 33% RliB-CRISPR spacers, 41% CRISPR-I spacers and 42% CRISPR-II spacers (Figure 6, Table S3). RliB-CRISPR and CRISPR-I systems share an identical protospacer adjacent motif (PAM) CCA at the 5′ of the protospacer, in contrast to CRISPR-II harboring NGG at the 3′ of the protospacer (Figure S5). Numerous spacers showed 100% identity with viral sequences (14% RliB-CRISPR spacers, 15% CRISPR-I spacers and 24% CRISPR-II spacers). None of the spacers matched bacterial (excluding prophages) or plasmid sequences. The high abundance of spacers perfectly matching bacteriophages in the RliB-CRISPRs and in the CRISPR-I and CRISPR-II, suggests that both cas-less and cas-associated CRISPR-Cas systems have a role in the immunity against bacteriophages.
To investigate the nature of the phage nucleic acid potentially targeted by the RliB-CRISPR, CRISPR-I and CRISPR-II systems, we first examined the orientation of the protospacers in respect to the corresponding spacers and then the function of the phage genes where the protospacers are located. Protospacers targeted by CRISPR-I and CRISPR-II systems originate both from sense and antisense DNA strand and are equally distributed along the phage genome, which suggests that these systems target phage DNA (Table S4, S5, Figure S6). Among 13 protospacers targeted by RliB-CRISPR, 9 protospacers are in the antisense orientation, 3 are positioned in intergenic regions and 2 are in the sense orientation. Moreover, among 11 protospacers for which the function of the targeted gene is known, 10 protospacers are located in the late phage genes encoding DNA packaging and structural proteins (Table S6, Figure S6). The occurrence of both sense and antisense oriented protospacers suggests RliB-CRISPR most probably targets DNA. However, higher occurrence of the antisense oriented protospacers and more interestingly, specificity for the function of the targeted bacteriophage gene suggests that RliB-CRISPR could potentially have a function in RNA interference.
Furthermore, we identified genomes containing a number of spacers matching their own prophages. For example, L. monocytogenes strain EGD has prophage B025 (our unpublished data) and carries one RliB-CRISPR spacer (S7) and three CRISPR-II spacers (S21, S22, S23) that match the same prophage with up to 97% identity (Figure S4A,B). We also identified two RliB-CRISPR spacers, three CRISPR-I spacers and two CRISPR-II spacers in 9 Listeria genomes that correspond to prophages in the same genome with 100% identity. In two cases, the strains (L. monocytogenes 08-5578 and 08-5923) lack the cas genes and in one case the repeat flanking the self-targeting spacer carries a point mutation (L. monocytogenes J0161), suggesting that in three instances the spacers are presumably inactive (Table S7, Figure 6). The remaining spacers are either in cas-associated CRISPRs or in the RliB-CRISPR. Furthermore, the PAMS corresponding to self-targeted protospacers do not significantly deviate from the consensus (Table S7). These results show that spacers matching the protospacer located in the same bacterial chromosome do not necessarily have strong negative fitness effects.
To test if the cas-less RliB-CRISPR might provide Listeria with DNA interference activity, we designed an experiment using a conjugation system and two plasmids that differ in the presence or absence of protospacer: i) a protospacer plasmid (P) and ii) the control plasmid (C), as previously done by Almendros et al. [29] (Figure 7). The plasmid P carries a protospacer matching spacer 3 (S3) of the RliB-CRISPR in the L. monocytogenes EGD-e strain and spacer 5 (S5) of the RliB-CRISPR in the Finland strain. The plasmid C is identical to the plasmid P, but the protospacer sequence is shuffled in silico (C) and does not correspond to any known sequence in the NCBI database.
Listeria is not naturally competent and the plasmid transformation efficiency in this bacterium is very low in comparison to other bacteria such as Bacillus or Streptococcus. Moreover, the Δpnp genetic background has a severe effect on bacterial growth, and plasmid transformation is even more difficult than in the WT strain. Therefore, the plasmids P and C were conjugated simultaneously via Escherichia coli S17 strains to L. monocytogenes EGD-e and Finland WT strains and their isogenic mutants deleted for RliB (ΔrliB) and PNPase (ΔpnpA). Quantitative PCR (Q-PCR) was used to determine the identity of the plasmids distributed among the transformants (see materials and methods). We then calculated for each individual strain the ratio (R) of the number of colonies carrying plasmid P and the number of colonies carrying plasmid C (R = nP/nC) (Figure 7B). The proportion of the transformants carrying the plasmid P for each experiment is an indication of the interference activity driven by the spacer, a lower proportion of transformants with the plasmid P (R<1) suggesting interference activity.
In the L. monocytogenes EGD-e in which no cas genes was identified, there is no significant difference in the R values between the strains carrying the RliB-CRISPR (WT) and the strains lacking either the RliB-CRISPR (ΔrliB-EGDe) or PNPase (ΔpnpA-EGDe), demonstrating that both plasmids are equally acquired and that the system is not able to provide any detectable DNA interference in the tested experimental conditions (Figure 7B). In contrast, in the L. monocytogenes Finland that bears an additional CRISPR-Cas Type-I system, the R ratio is significantly smaller than 1 in the WT strain whereas it reaches 1 in the strains lacking either the RliB-CRISPR (ΔRliB-Fin) or the PNPase (ΔpnpA-Fin). Thus, RliB-CRISPR can lower the plasmid P acquisition in the strain carrying the CRISPR-I system, suggesting that the RliB-CRISPR is able to use the trans encoded Cas proteins encoded by the CRISPR-I and confer to Listeria a DNA interference activity. Interestingly, PNPase is required for this process.
Since the initial discovery that CRISPR-Cas systems function as an adaptable prokaryotic immune system, the CRISPR research has been flourishing and biochemical insights into the CRISPR-Cas systems have increased dramatically over the past few years. However, these systems are extremely diverse and the function and molecular mechanism of many of them are still unknown. In particular, little is known about CRISPR-Cas type I-B, I-C, and I-D systems [30]. Here, we studied RliB-CRISPR that has an unusual secondary structure comprised of basepair interactions between the repeat sequence and the adjacent spacer. We showed that RliB-CRISPR is expressed and processed even in the complete absence of Cas proteins, and demonstrated this event occurs under the guidance of PNPase. The RliB-CRISPR spacers match numerous temperate and virulent Listeria bacteriophages and in the presence of CRISPR-Cas system Type-I, RliB-CRISPR lowers the frequency of a plasmid carrying a matching protospacer. In addition, we show that this DNA interference is dependent on the presence of PNPase. Overall, our data demonstrate that PNPase is involved in Listeria RliB-CRISPR processing and its DNA interference activity.
In silico analysis of the secondary structures of CRISPR repeats across bacterial and archaeal CRISPR-Cas systems suggested that some CRISPR repeats can form stable stem-loops due to the palindromic nature of their repeats, but that other lack any detectable conserved structure [25]. RliB repeats are only weakly palindromic and unlikely form a stable stem-loop structure. Here, we experimentally determined the secondary structure of RliB and surprisingly, discovered that RliB contained 6 hairpin structures formed mostly by base-pair interactions between the spacer sequences and the adjacent repeats, and with GUUU-rich apical loops (Figure 1B). The structure of RliB is thus dependent on the nature of the acquired spacer. In silico analysis of other representative RliB-CRISPRs showed their putative structures rely on the same principle, suggesting that base-pair interactions between the repeat and the spacer could be a common structural motif among RliB-CRISPRs (Figure S2). It is tempting to hypothesize that successful acquisition of a new spacer requires some degree of complementarity with the repeat. Spacer acquisition is the least understood step of CRISPR-Cas system function [31] and our data potentially highlight new aspects of the integration mechanism via homology with the repeat sequence.
It is generally accepted that CRISPR arrays require Cas proteins for their processing and activity. A first example of CRISPR-Cas system that does not rely solely on the Cas proteins but requires also the activity of endogenously encoded enzymes has been recently reported in Streptococcus pyogenes. In this case, a CRISPR array type-II is processed by the widely conserved endoribonuclease III and a small trans-acting RNA tracrRNA [32]. In addition, a recent study of Zhang et al [33] revealed a CRISPR in Neisseria meningitidis, where crRNAs are transcribed from promoters that are present within each repeat and require RNase III and trans-encoded tracrRNA-mediated processing for their maturation. Surprisingly, the maturation processing is dispensable for the CRISPR interference [33].
Here, we characterized a CRISPR array that is processed in a bacterium completely devoid of cas genes. In contrast to other CRISPRs, RliB-CRISPR is present in all sequenced strains of L. monocytogenes, even in other members of the genus and never co-localizes with cas operons. We demonstrated that RliB binds to and is a substrate for endogenously encoded PNPase, both in cas-less Listeria strains and in those encoding a complete set of cas genes elsewhere in the genome. Generally, PNPase degrades single-stranded RNA in a processive manner along the substrate until it stops, stalled by a stable RNA structure [23]. For instance, a hairpin structure in a bacteriophage mRNA can block the processivity of PNPase to protect the RNA against degradation [34]. It remains to be understood at the molecular level how PNPase specifically recognizes the RliB-CRISPR and how the progression of the enzyme stops.
In the three analyzed L. monocytogenes strains, RliB-CRISPRs is expressed as a full length molecule that is processed to several fragments out of which a 280 nt fragment is the most abundant form. The consistency of processing that is independent of the number of the repeat/spacer units suggests that the molecular mechanisms guiding the processing in the tested CRISPRs is conserved. Interestingly, in the bacteria deleted for pnpA (Δpnp-EGDe, Δpnp -EGD, Δpnp -Fin), some processing still occurs, indicating there are other endogenously encoded ribonucleases contributing to this mechanism, particularly in the EGD-e strain that is devoid of cas genes (Figures 3C and 5B). Listeria encodes at least 17 different putative RNases identified by homology with closely related Bacillus subtilis [35]. Future work will have to determine which enzymes might function together with PNPase and also contribute to the CRISPR processing.
We showed that cas-less RliB-CRISPRs are rich in spacers matching virulent and temperate bacteriophages. In addition, a large fraction of those spacers have 100% matches with phages, strongly suggesting a function for RliB-CRISPR even in the absence of cas (Figure 6, Table S3). Accordingly, we showed that cas-less RliB-CRISPR lowers the acquisition of a plasmid carrying the corresponding protospacer, provided a CRISPR-I system is present (Figure 7). The RliB-CRISPR and CRISPR-I share many similar features; almost identical repeat sequence (Figure 4), homologous putative leader sequences (Figure S3) and identical PAM motifs (Figure S5), indicating that these two systems are closely related and are possibly functionally linked. It is thus not surprising that RliB-CRISPR can share the Cas machinery with the CRISPR-I to acquire the DNA interference activity, however future analysis will be required to establish the exact mechanism by which this crosstalk occurs.
More interestingly, the DNA interference activity of the RliB-CRISPR is also dependent on the presence of PNPase (Figure 7), indicating that the processing by this enzyme is important for the activity of the RliB-CRISPR. PNPase is a highly complex enzyme with 3′ to 5′ exoribonuclase and RNA polymerase activities being the most studied up to date. However, it was recently shown that PNPase can degrade single stranded DNA (ssDNA) and also catalyze template independent polymerization of dNDPs into 3′ends of ssDNA, which established a molecular model for the role of PNPase in DNA repair [36], [37]. In Escherichia coli, PNPase affects the stability of several regulatory sRNAs [38], [39]. Here, we hypothesize that Listeria PNPase, potentially with other endogenously encoded enzymes, may contribute to the RliB-CRISPR maturation. Alternatively, PNPase may affect the RliB-CRISPR RNA stability and turnover, and hence, regulate the levels of its mature form. Finally, PNPase dependent processing of the RliB-CRISPR and the DNA interference might be uncoupled activities. Hence, this complex enzyme could use different enzymatic activities to contribute to different processes. It will be also important to determine if PNPase is involved in other CRISPR-Cas system activities, such as new spacer acquisition. Currently, our data do not provide evidence on which form of RliB molecule is active in the DNA interference. These and other mechanistic details, such as the role of PAMs are to be determined in the future.
Noticeably, the RliB-CRISPR mediated DNA interference is not 100% effective. This might be the consequence of our experimental design or this CRISPR-Cas system did not evolve to eradicate the bacteriophage from a population but rather to fine-tune its copy number in the bacterial cytoplasm.
Our functional assay showed that the RliB-CRISPR in the L.monocytogenes EGD-e strain that completely lacks cas genes, although processed by PNPase, is not able to provide DNA interference activity against a plasmid carrying a matching protospacer. This lack of activity is probably due to the absence of trans encoded CRISPR-I system required for RliB-CRISPR DNA interference activity, as shown in L. monocytogenes Finland strain. However, the conservation of the RliB-CRISPRs in Listeria is independent on the presence of CRISPR-I, strongly suggesting that it is a functional element with an important function even in the absence of Cas Type-I, as sequences lacking selection pressure for their maintenance are quickly lost in bacterial genomes [40]. Interestingly, RliB-CRISPRs in average possess a smaller number of repeats and the variability of their spacers is lower compared to the spacer content of the CRISPR-I and CRISPR-II. Have they evolved a new function? It is to be kept in mind the remarkable finding that all RliB-CRISPRs accumulate as a 280 nt fragment, which might be the functional form. In support for a functional role of RliB, our recent RNA-seq analysis has shown that RliB-CRISPR is not only conserved but also expressed in the more distant L. innocua species that also lacks CRISPR-I [41].
Although RliB-CRISPRs share many similarities with cas-associated CRISPR-I system, the identified RliB-CRISPR protospacers are more often in the antisense orientation with respect to the corresponding spacer and in addition they are mostly located in the late phage genes encoding DNA packaging and envelope proteins. It is tempting to speculate that “the” RliB-CRISPRs cas-independent activity might be RNA interference. In this scenario RliB-CRISPR would not destroy the bacteriophage DNA but would rather control the bacteriophage late gene expression i.e., it would prevent the formation of viral particles and lysis of the bacterial cell. RliB-CRISPR interference could be also based on transcription-dependent DNA targeting, as recently described in Sulfolobus islandicus REY15A [42]. Alternatively, RliB-CRISPR might have evolved a broader function relevant for Listeria physiology that is not related to the immunity. Such examples have been described in Pseudomonas aeruginosa, where a CRISPR appear to be involved in lysogeny dependent biofilm formation [43], in myxobacteria where CRISPR has been implicated in swarming motility [44] and more recently in Francisella novicida where a tracrRNA was shown to regulate an endogenous transcript encoding a lipoprotein important for the bacterial infection [45].
The interaction between bacteriophages and bacteria is mostly seen as a parasitic interaction where the virus exploits the host resources for its own benefit. However, there are some viruses that have a beneficial effect on their host [46]. In case of pathogenic bacteria, bacteriophages often carry virulence factors required for successful infection [47]. More recently, a study by Rabinovich et al. (2012) showed that during Listeria intracellular infection, a temperate prophage is excised, which reconstitutes a function of the gene where the bacteriophage was integrated, and promotes bacterial escape from macrophage phagosomes. Remarkably, the excision event does not lead to propagation and release of the progeny virions neither to the subsequent lysis of the bacterial cell. Hence, the virion production is actively aborted [48]. This example highlights an important crosstalk between the phage and the pathogenic bacteria during the infection of the mammalian cell, and more importantly, it emphasizes the conditional advantage for a bacterium to maintain a bacteriophage and control its virulence. Our previous studies have shown that RliB expression is upregulated in the bacteria grown in human blood and in the intestine of gnotobiotic mice and is important for Listeria virulence [4]. Whether RliB-CRISPR expression and prophage excision followed by aborted virion production are linked processes, remains to be examined. Our study thus paves the way for new regulatory studies on the interactions between bacteriophages and bacteria during saprophytic life or during infection.
Strains used in this study are L. monocytogenes EGD-e (BUG1600) and its isogenic mutants ΔrliB(BUG2621) and ΔpnpA(CMA751), L. monocytogenes EGD (BUG600) and its isogenic mutant, ΔpnpA-EGD (BUG3415) and ΔrliB-EGD (BUG3243), L. monocytogenes Finland 1998 (BUG3297, CLIP2012/00396, FE49845/IHD42536) and its isogenic mutants ΔpnpA-Fin (BUG3465) and ΔrliB-Fin (BUG3466). Mutants were obtained by deletion of the corresponding ORF or non-coding RNA by PCR-ligation and amplicon cloning in the suicide vector pMAD as previously described [49]. Overexpression of RliB was obtained by cloning the rliB gene into the pAD vector carrying Phyper constitute promoter [50], resulting in the strain Phyper-RliB (BUG2987). PNPase complementation was obtained by cloning the PnpA ORF into the pPl2 vector [51] resulting in the strain ΔpnpA+pnpA (CMA752).
Bacteria were grown overnight in Brain heart infusion (BHI) medium (Difco) at 37°C with shaking at 200 rpm. Cultures were subsequently diluted 1/500 into 100 ml BHI and grown at 37°C until mid-exponential phase (OD600 = 1.0). When required, erythromycin and chloramphenicol were used at 5 µg/ml and 20 µg/ml, respectively as final concentration. For RNA extraction, bacteria were pelleted, by centrifugation at 10,000 X G for five minutes, flash frozen in liquid nitrogen and stored at −80°C.
Bacterial pellets were resuspended in 400 µl solution A (½ volume Glucose 20%+½ volume Tris 25 mM pH 7.6+EDTA 10 mM) to which an additional 60 µl of 0.5M EDTA was added. Bacteria were lysed in FastPrep homogenizer (Bio101) and RNA was subsequently extracted using TRI reagent (Invitrogen) as described previously [4]. RNA integrity was verified using the Experion Automated Electrophoresis system (Biorad).
10–20 µg of total RNA was mixed with two volumes of Formaldehyde Loading Buffer (Ambion) followed by denaturation at 65°C for 15 min. Samples were separated by electrophoresis on 5% TBE-Urea polyacrylamide gels (Criterion-Biorad) at 100 V for 2 hours in 1× TBE running buffer at RT, followed by an overnight transfer at 4°C/100 mA to Nytran membranes (Sigma). Membranes were UV-crosslinked and probed with RNA probes or DNA oligo probes. Briefly, RNA probes were synthesized and α32P-UTP labelled using the Maxiscript T7RNA polymerase kit (Ambion) with PCR generated templates according to the manufacturer's instructions. Oligonucleotide DNA probes were 5′ labelled with γ32P-ATP using the T4 Polynucleotide Kinase according to the manufacturer's protocol (New England Biolabs). Membranes were prehybridized for 60 min in Ultrahyb buffer (Ambion) and hybridizations were performed overnight at 64°C for RNA probes and at 37°C for oligonucleotide probes. Following hybridization, membranes were washed twice for 5 min with 2× SSC, 0.1% SDS at room temperature. When hybridized with RNA probes, membranes were additionally washed twice for 15 min in 0.1× SSC, 0.1% SDS at 60°C. The size marker was a 50-bp ladder (Invitrogen), which was 5′ end labelled with γ32P-ATP.
We first have optimized an affinity purification assay using 3′-biotinylated RliB and streptavidin sepharose modified as described in Jestin et al. [52]. As a negative control, we used the regulatory RNAIII from S. aureus. Total cell extract prepared from 500 ml of culture of L. monocytogenes ΔrliB mutant strain was first incubated with streptavidin sepharose beads to remove proteins unspecifically bound to the beads. The beads were first incubated with the 3′ biotinylated RNA and the pre-cleaned crude extract was passed through the column and washed with the binding buffer containing 50 mM Hepes-NaOH pH 7,5, 5 mM MgCl2, 1 mM DTT, and 150 mM KCl. The elution of the proteins was done with the same buffer containing 6 M urea, 2 M thiourea and 30 mM d-biotin. The fractions were then analyzed by 4–15% gradient SDS-PAGE, and the proteins were identified by mass spectrometry.
A second approach was used to purify proteins associated with RliB carrying at its 5′ end two hairpin motifs recognized by the coat protein of the MS2 bacteriophage. As a control we used the untagged RliB. Both RNAs were transcribed in vitro using homemade T7 RNA polymerase. The experimental conditions were as previously described [53]. The MS2 coat protein fused to Maltose binding protein was expressed in E. coli and purified on an amylose column followed by a monoQ column. The MS2-MBP coat protein was first immobilized on the amylose resin, and the tagged-RNA was loaded on the column, which was washed with 2 ml of the Binding Buffer. Subsequently, the pre-cleared bacterial lysate was loaded onto the column, followed by three washes with 2 ml Binding Buffer, and the proteins were eluted with the binding buffer containing 10 mM maltose. The fractions were loaded on a SDS-PAGE and the proteins were identified by mass spectrometry.
Enzymatic hydrolysis was performed with 1 pmol of RliB in 10 µl of a buffer containing 50 mM NaOH-Hepes pH 7.5, 10 mM MgCl2, 150 mM KCl, in the presence of 1 µg carrier tRNA at 20°C for 5 min: RNase T2 (0.01 units), RNase V1 (0.5 units). Chemical modifications were performed on 2 pmol of RliB at 20°C in 20 µl of the same buffer containing 2 µg of carrier tRNA. Methylation of C(N3) and A(N1) positions was done with 1 µl DMS (diluted 1/8 and 1/16 in ethanol) for 2 min at 20°C. Modification of U(N3) and G(N1) was performed with 2,5 µl and 5 µl of CMCT (40 mg/ml) for 20 min at 20°C. The cleavage or modification sites of unlabeled RNAs were detected by primer extension. Details for hybridization conditions, primer extension, and analysis of the data have been previously described [54].
PNPase cleavage assays were done using a 5′ end-labelled RliB or RNA37. Reaction was performed in 10 µl of TMK buffer containing 20 mM Tris-HCl pH 7.5, 10 mM magnesium-acetate, 100 mM KCl, 1 mM DTT at 37°C for 15 min in the presence of PNPase 200 nM in the presence of 1 µg of carrier tRNA. Competition experiments were carried out in the presence of 200 nM, 400 nM of cold RliB and its derivatives (RliB-3′ domain or RliB-5′ domain). Reactions were stopped by phenol extraction followed by RNA precipitation. The assays were loaded on a denaturing 12% polyacrylamide-urea gel electrophoresis. The PNPase cleavage sites were assigned by running in parallel RNase T1 ladder and an alkaline ladder on a denatured end-labelled RNA [54].
To perform gel retardation assays, 5′ end-labelled transcript (20000 cpm, <1 nM) was incubated in the presence of increasing concentrations of PNPase (100 to 800 nM) in TMK buffer containing 20 mM Tris-HCl pH 7.5, 10 mM magnesium-acetate, 100 mM KCl at 37°C for 15 min. At the end of the binding reaction 6X loading dye (30% glycerol, 0.25% bromophenol blue and 0.25% xylene cyanol) was added to the samples and they were analyzed on a 6% polyacrylamide gel under non-denaturing conditions.
We analyzed 45 Listeria genomes taken from GenBank, available at the time of analysis. These include 28 complete genomes and 17 draft genomes with less than 800 contigs (Table S1). We also added the genome of L. monocytogenes EGD strain (unpublished data). We used GenBank annotations, excluded genes with stops in phase and with lengths not multiple of three. We re-annotated the prophages in the genomes using a methodology described previously [55].
|
10.1371/journal.pcbi.1002578 | Shaping the Dynamics of a Bidirectional Neural Interface | Progress in decoding neural signals has enabled the development of interfaces that translate cortical brain activities into commands for operating robotic arms and other devices. The electrical stimulation of sensory areas provides a means to create artificial sensory information about the state of a device. Taken together, neural activity recording and microstimulation techniques allow us to embed a portion of the central nervous system within a closed-loop system, whose behavior emerges from the combined dynamical properties of its neural and artificial components. In this study we asked if it is possible to concurrently regulate this bidirectional brain-machine interaction so as to shape a desired dynamical behavior of the combined system. To this end, we followed a well-known biological pathway. In vertebrates, the communications between brain and limb mechanics are mediated by the spinal cord, which combines brain instructions with sensory information and organizes coordinated patterns of muscle forces driving the limbs along dynamically stable trajectories. We report the creation and testing of the first neural interface that emulates this sensory-motor interaction. The interface organizes a bidirectional communication between sensory and motor areas of the brain of anaesthetized rats and an external dynamical object with programmable properties. The system includes (a) a motor interface decoding signals from a motor cortical area, and (b) a sensory interface encoding the state of the external object into electrical stimuli to a somatosensory area. The interactions between brain activities and the state of the external object generate a family of trajectories converging upon a selected equilibrium point from arbitrary starting locations. Thus, the bidirectional interface establishes the possibility to specify not only a particular movement trajectory but an entire family of motions, which includes the prescribed reactions to unexpected perturbations.
| Brain-machine interfaces establish new communication channels between the brain and the external world with the goal of restoring sensory and motor functions for people with severe paralysis or sensory impairments. Current methodologies are based on decoding the motor intent from the recorded neural activity and transforming the extracted information into motor commands to control external devices as robotic arms. We developed a novel computational approach, based on the concept of programming dynamical behaviors trough the bi-directional sensory-motor interaction between the brain and the connected external device. This approach is based on the emulation of some control features of a biological interface, the spinal cord. The first prototype of our interface controls the state of motion of a simulated point mass in a viscous medium. The position of the point mass is encoded into a stimulus to the somatosensory cortex of an anesthetized rat. The evoked activity of a population of motor cortical neurons is decoded into a force vector applied to the point mass. The parameters of the encoder and of the decoder are set to approximate a desired force field. In the first test of the interface, we obtained a family of trajectories that converged upon a stable attractor.
| In a recent demonstration [1], Schwartz and coworkers decoded neural activities from the motor area of a monkey's cerebral cortex to control the movement of a robotic arm. The monkey learned to activate the recorded neurons and to guide the arm for transporting food to the mouth. This is an undisputed milestone in Neural Engineering, highlighting the potential of neural interfaces (NIs) as a means to restore a connection with the world for people with severe paralysis. In addition to their clinical impact, NIs have the potential to revolutionize our ways to study the nervous system, by connecting live neural populations with external devices, both physical and simulated. This constitutes a leap forward with respect to current paradigms, in which physiological experiments and computational analyses are carried out separately.
Both the clinical and the basic science applications of NIs call for the possibility to close the sensory-motor loop, by combining a decoding interface – mapping neural activities into inputs to the external device – with an encoding interface – mapping the state of the device into a direct input to the brain, such as an electrical stimulus. In this study we addressed the challenge to create a coordinated bidirectional brain-machine interaction by concurrently setting up a decoding and an encoding interface, which combined generate a dynamic control policy in the form of a force field. In this approach, we aimed at emulating the operation of the spinal cord, as the prime biological interface between the brain and the musculoskeletal apparatus.
Ascending tracts of the spinal cord inform the brain about the state of motion of the limbs and about physical properties of the environment. Descending tracts distribute motor commands across groups of muscles both by direct connections with the motoneuronal pools and by connections with spinal interneurons that activate multiple muscles spanning one or more joints [2], [3]. Earlier studies in frogs [4]–[6], rats [7], and cats [8] have revealed that the electrical stimulation of the grey matter in the lumbar spinal cord results in a field of forces acting on the ipsilateral hind limb. This finding has a simple biomechanical basis. The force generated by a muscle varies depending on the state of motion of the muscle – i.e. its instantaneous length and shortening rate. In addition, variety of other factors, such as fatigue and hysteresis, and environmental variable, such as temperature, affect muscle force. While the detailed analysis of these factors is beyond the scope of this work, we may simply state that when the spinal cord activates an ensemble of muscles in response to a cortical command, the net mechanical outcome is a spatial pattern of forces – a force field – that sets the limb in motion. The above mentioned studies have highlighted the presence of convergent patterns of forces, but evidence from other investigations [9] have suggested more complex spatio-temporal structures of the underlying force fields.
Our study aimed at reproducing in an artificial interface this basic control mechanism. We considered the problem of generating by function approximation a force field that converges to a central equilibrium point. This is a very particular instantiation out of a much larger repertoire of possible mechanical behaviors, which may be represented as a functional map from the state of motion of a limb, i.e. its position and velocity, and the ensuing force generated by the musculoskeletal apparatus.
In the language of control theory, the spinal cord establishes a policy [10] by specifying the forces to be generated throughout the reachable space in response to unexpected perturbations. We have adopted this perspective for developing a new type of neural interface called dynamic Neural Interface (dNI), which borrows a local portion of cortical tissue to emulate the generation of force fields by the spinal cord [4].
The dNI has 4 components, as illustrated in Figure 1. We performed all tests on anesthetized Long-Evans rats. The rats' brain interacted with a dynamical system through a sensory interface and a motor interface. On the brain side, one microwire array delivered the microstimulation to the vibrissal representation of primary somatosensory cortex (S1) and a second microwire array recorded the neural signals from vibrissal motor cortex (M1). On the other side of the interface there was a simple and well-understood dynamical system: a simulated point mass moving over a horizontal plane within a viscous medium.
We began each experiment by collecting a “training” set of neural population responses to repeated presentations of different electrical stimulation patterns. We used these training data to implement a calibration procedure for establishing concurrently the encoding function of the sensory interface and the decoding function of the motor interface. Following the calibration procedure, we tested the competence of the interface (test phase) to drive the simulated point mass towards a goal location, which was defined by the central equilibrium point of a radial force field.
The purpose of the sensory-motor mapping is to set the parameters of the sensory and motor interfaces so as to approximate the desired force field. While force fields are continuous maps from position to force, the interface has a finite number of stimuli. Therefore, the mapping procedure must construct an approximation of the desired field with only a small number of vectors. To this end, we construct a cascade of three mappings: 1) a mapping from the position of the external device to one of selected stimuli; 2) a mapping from each stimulus to the evoked neural activity, and 3) a mapping from the evoked neural activity to the force acting on the external device. The first and last mappings are established by the interface software (i.e. sensory and motor interfaces), the second mapping is established by the properties of the neural structures that connect the stimulation and recording arrays.
In this first implementation, the sensory interface established a map from the position of the point mass to one of 4 stimulation electrodes (Figure 1A). The sensory mapping procedure (as detailed below) divided the workspace into 4 contiguous regions corresponding to a small “vocabulary” of 4 stimuli. At each iteration step, the interface algorithm selected the stimulus based on the region in which the point mass was located. The electrode delivered a train of 10 biphasic pulses (150 µA, 100 µs/phase) at 333 Hz [11], [12]. Larger vocabularies of stimuli can be generated by using a greater number of electrodes and by including electrode combinations. With a greater number of distinct stimuli, the workspace would be divided into smaller and denser regions, thus increasing the quality of the approximation of the desired continuous field. In a physiological system, the region of space that can activate a sensory neuron is called a “receptive field”. Here, the workspace of the sensory interface is divided into regions that are analogous to receptive fields: the mechanical system triggers an electrode when it passes by the region corresponding to that electrode.
The motor interface transformed recorded neural activities into force vectors applied to the simulated point mass (Figure 1B–D). A commercial spike-sorting algorithm (Rasputin, Plexon Inc.) decomposed the recorded neural signals into single-unit activities. We sorted 5–20 single units in each session from a 16 channel microwire array (average ± SEM across sessions was 13.69±0.48 units). The single trial responses of each neuron to the stimulation pattern consisted of a time series of spike counts computed in time bins of size Δt over a window of duration T·Δt, starting from the end of the stimulus. The neural population response was quantified as an array of such binned spike sequences. We found that post-stimulus windows of duration in the range between 100 and 600 ms binned at a resolution of Δt = 5 ms led to best performance of the interface (see below). Unless otherwise stated, in the following we present results obtained by running the interface using Δt = 5 ms and T·Δt = 600 ms. In this case, the input to the motor interface was a matrix with N rows and 120 (i.e. 600/5) columns (Figure 1B). During the test phase, the single-trial neural population response matrix was linearly mapped into the two components of a planar force vector.
In the following we describe the “dynamic shaping” algorithm for the concurrent calibration of the sensory and motor maps. The algorithm is defined by a set of 4 key parameters:
During the calibration each stimulation patterns was repeated R times and, accordingly, R×N neural responses were recorded. Each response was an array of T values: the number of spikes in each bin. The calibration responses were then represented as S×R N-dimensional vector functions:(1)From these calibration responses, we averaged the responses obtained from the repetitions of each stimulus, to extract S mean responses(2)Following the same notation, a neural response vector is an N-dimensional vector function(3)The inner product of two neural responses is defined by extension over time bins and units of the Euclidean inner product:(4)The S mean calibration responses form a set of basis fields – a direct extension of the concept of basis vectors – that were used to approximate all recorded neural responses. In particular, each calibration response was approximated as a sum of mean responses:(5)To derive the combination coefficients , one takes the inner product of each side of Equation (5) with each basis function. This leads to S vector/matrix equations(6)where(7)Equation (7) can be solved for provided that (if the projection matrix is singular, one can use a pseudo-inverse. But this does not seem to be a likely situation and was not encountered with any of our datasets).
With this, each calibration response was mapped respectively into an S-dimensional vector(8)Each response corresponds to a d-vector and vice-versa, each d-vector corresponds to a unique approximation of the response (the likelihood that two distinct signals map onto the same d-vector is vanishingly small). Therefore, we took the S-dimensional vectors as representations of the individual neural responses obtained after applying each stimulus.
To calibrate the motor interface, we used principal component analysis (PCA) and extracted the two principal components that capture the greatest amount of variance in the set of the S×R calibration vectors, . These two components are two S-dimensional arrays that form the rows of the 2×S projection matrix(9)This operator defines the two-dimensional plane with maximum variance over the set of S stimuli. The next step of the calibration procedure involved stretching the matrix so as to match the range of variation of the x and y components of the force vectors over the desired force field domain:(10)The gain is a 2×2 diagonal matrix that scales the two-dimensional projections of the calibration recordings to cover the range of the desired force field, . The field establishes a correspondence between the position, , of the controlled object – in this first implementation a point mass – and a resulting force . Here, we make the additional hypothesis that this field is invertible, which means that there is a function mapping force vectors to corresponding positions. This is obviously the case if the field is linear, as in and the “stiffness” matrix is non-singular. The requirement of invertibility can be relaxed to a local and continuous form.
The two projection matrices, and , and the mean calibration responses, , to all the stimuli generate a map from the data collected during the experiment to a corresponding two-dimensional force vector(11)This is a linear filter that operates in real time.
The sensory interface maps the instantaneous position of the controlled object onto one of the stimulation patterns in the calibration vocabulary.
This sensory interface performs a look-up operation:(12)that picks up the stimulus, , corresponding to the “calibration site” that is closest to the current position ρ of the controlled object. The calibration sites are the S locations:(13)where is the force derived by Equation (11) from the average response, , to the i-th stimulus in the vocabulary.
In this first implementation, there were 4 distinct electrical stimuli, s1, s2, s3 and s4 and 4 mean corresponding neural responses, r1, r2 r3 and r4 (Figure 2A). Each mean neural response was a high-dimensional collection of spiking activities, which was reduced by the motor interface to the two coordinates of a force vector. Principal component analysis (PCA) performed this dimensionality reduction by extracting from each of the 4 mean neural responses recorded during the calibration phase the two principal components that capture the highest amount of signal variance. We scaled these two components so as to span the variance of the force vectors over the desired force field. This process resulted in a simple linear mapping, i.e. a gain matrix and an offset vector that, when applied to the neural response produced a force vector (Equation 11). In particular, the 4 mean responses collected during the calibration mapped to 4 template force vectors F1, F2, F3, and F4 (Figure 2B). The desired force field established a relationship between these template force vectors and 4 positions, x1, x2, x3 and x4 (Figure 2C). These 4 positions partitioned the space of the external device into 4 contiguous regions, A1, A2, A3 and A4, based on a nearest-neighbor map: a generic point x was associated to the region Ai if xi was the nearest calibration position (Figure 2D). In this case, the sensory interface triggered the stimulus si. It is straightforward to extend this procedure to an arbitrary number of stimuli for generating denser approximations of the desired force field.
The concurrent operation of the sensory and the motor interfaces resulted in the realization of a force field that approximated a desired radial force field converging towards a central equilibrium point (Figure 2C). If one might assume that the recorded neural activity elicited by each stimulus remained invariant through time, then the field generated by the interface would be a piecewise constant approximation of the desired field. However, the inherent variability of neural activities observed after each repetition of an electrical stimulation pattern violated this assumption. This variability was mostly caused by background activities that interacted with the activities induced by the stimulus. In the anesthetized preparation, the background activities can be considered as random noise. In the alert animal, these activities may also contain a voluntary component. In this way the actual field is an additive superposition of the field approximation established by the interface with a random component induced by background neural noise. Extracting as much information about the stimulus as possible from the recorded signals is a key technical challenge for generating a controlled desired dynamical behavior with the bidirectional interface.
During the test phase, we probed the ability of the dNI to drive the simulated point mass towards a goal location, corresponding by design to the central equilibrium point of the desired force field. This is a simplified representation of a reaching movement, where the interface emulates the generation of a convergent force field similar to those observed after microstimulation of the spinal grey matter [4]–[6].
The dNI generated a movement of the simulated point mass (Figure 3D) by the following procedure:
Because of the cortico-cortical pathways between stimulated and recorded populations [13], the neural population responses were clearly modulated by the stimuli (Figure 3C). However, the actual behavior of the interface contained a stochastic component due to the fact that each stimulation pattern, when repeated over different trials, caused a variable response in the recorded motor cortex. Part of the response variability in our anaesthetized preparation likely arose from trial to trial fluctuations in ongoing internal activity unrelated to the stimuli [14]. These trial to trial response variations resulted in a random time-varying component of the force field.
The performance of the dNI likely depends upon information that the neurons make available for communication with the dynamical system, which in turn likely depends upon the temporal precision at which spike trains are considered [15], [16]. In particular, previous studies of neural encoding suggest that more information may be extracted from neural responses if they are examined with a relatively fine precision of the order of few to few tens of ms [17], [18] and that the optimal precision to extract information from neural activity may vary depending on the specific task or condition [19], [20].
In this study we therefore determined empirically the range of response parameters that maximized some measures of the quality by which the neurons can communicate with the rest of the system. The neural response r following the electrical stimulation was quantified as a time series of spike counts for each of the N neurons computed in T small time intervals of size Δt post-stimulation. The size of the bins Δt (corresponding to the temporal precision used to evaluate neural responses) and the parameters defining the time window duration (the number of time bins T and the offset of the post-stimulus window) are all arbitrary parameters that we attempted to set optimal according to some quantitative criterion. To study systematically how the performance of the dNI depends on the temporal parameters defining the neural response, we generated a set of “off-line” trajectories according to the following simulation procedure. At each step of the simulation, the position of the point mass was paired with the stimulation pattern associated with its nearest neighbor, as in the actual on-line experiment. Then, a recorded pattern was randomly drawn from an additional collection of neural responses to the 4 electrical stimulation patterns stored in the sensory interface.
Using the off-line trajectories, we estimated the amount of information that the neural population makes available to communicate with the dynamical system. This information was evaluated as the Mutual Information between the force vector expected to be generated by the electrical stimulation in a given trial (a template force vector corresponding to the mean force vector established during the calibration trials in response to the considered electrical stimulation, Figure 3A blue arrows) and the actual force vector obtained from the neural response in that trial.
We found that the really critical response parameter was the temporal precision Δt at which spikes are sampled (Figures 4C and 4D). A fine temporal precision Δt≈5–10 ms was needed to obtain high Information values. Using coarser temporal precisions of 50 or 100 ms led to dramatic decreases of the Information values (Figure 5A). Figure 5B reports the results of how the Information , averaged over all sessions and calculated using a sampling precision Δt = 5 ms, depended upon the windows duration T·Δt and upon the offset value defining the response window. Information was very stable in the range T·Δt≈25–600 ms. The fact that the interface performs well also for decoding windows as short as few tens of ms encourages us to believe that it will be possible to push the dNI technology towards implementing feedback which is rapid enough to control real life motor tasks.
Moreover, there was a highly significant correlation (p<10e-9) between the Information and both the convergence rate (the percentage of trajectories that converge into the target) and the inverse of the mean number of steps to convergence of the off-line dNI trajectories (Figure 4E–F). As a result, the performance of the dNI was maximal for fine temporal precisions: the convergence rate peaked for Δt≈5–10 ms (Figure 4C). At Δt = 5 ms, the convergence rate of the dNI was on average 6 times higher than the convergence rate obtained with a purely random choice of the electrical stimulus to be applied (Figure 4D), demonstrating that the neural information had a sizeable impact on the dNI dynamics. These results suggest that precise spike timing is not only crucial for communication within the nervous system [16], but it is also important for bidirectional communication between external effectors and the nervous systems.
The impact of the Mutual Information provided by the neurons participating in the dNI upon the performance of the dNI was further investigated by studying the relationship between and the convergence speed of the dNI on the off-line simulated trajectories. For each set of possible response parameter and experimental session, we computed the mean number of steps needed for the trajectory to converge and the probability of reaching convergence to the center of the force field (averaged over 100 off-line-generated trajectories) with these response parameters and we correlated it with the Information computed in the same conditions. In sum, the empirical finding was that higher Information values corresponded to faster and more reliable convergence of the dynamical behavior and all measures pointed to the same range of neural response parameters being optimally efficient for dNI operation.
We also evaluated how the performance of the interface depended upon the population size by comparing the convergence rates when using all the neurons of each datasets with those using only half or one quarter of the units. The average number of recorded neurons during each experimental session was 13.69±0.48 (mean±SEM over all sessions). For each dataset, we randomly selected (out of nA recorded units) nH and nQ units for the calculation of the performance with half and one quarter units, with nH and nQ being the approximation to the closest integer of nA/2 and nA/4, respectively. For each selection of the subpopulation, we subtracted the obtained convergence rate by that obtained from a random choice of the stimulation patterns (as we did when analyzing the performance of the entire population). Figure 5C shows that a decrease in performance is observed only when reducing the population size to one quarter of the recorded one. Convergence rates with one quarter neurons are statistically different from the rates in the other two cases (p = 3.3552e-006, ANOVA), while the performances with all and half neurons were not statistically different (p>0.1, ANOVA). This suggests that using multi electrode recording arrays is useful for the performance of the system.
Finally we used different performance metrics to compare on-line trajectories with off-line simulated trajectories to evaluate if the off-line dataset could be used to simulate and study in more detail on-line behavior. To perform this comparison we selected 70 converging on-line trajectories selected from 13 rats and 70 corresponding off-line trajectories. In particular we calculated the root mean square error (RMSE), the mean integrated distance to target (MIDT) and the number of steps to convergence. For the calculation of RMSE, we first computed for each trial i the ideal trajectory as the one sharing the initial point with the actual trajectory, but evolving with a force . Then, for each trial i we computed the root mean square error as with T being the maximum duration of the trial and the actual position of the point-mass at time t. We computed MIDT as the average distance from the target. For each trial i, being the position of the point mass at time t and the position of the target we define: . Because the target corresponds to the origin of the plane, MIDT is simply the length of the trajectory normalized by its converging time.
As reported in Figure 4B, we found no significant differences in the computation of RMSE, MIDT or number of steps to convergence between on-line and off-line data (t-test, with p = 0.17 for RMSE, p = 0.41 for MIDT, p = 0.5 for number of steps). The consistency between the off-line open-loop simulated trajectories and the actual closed-loop trajectories recorded on-line during the experiment suggests that the parameters set optimally by generating offline simulated trajectories from calibration data will be optimal also for running the same interface online. In this respect Mutual Information is an advantageous optimization metrics during calibration, because the corresponding evaluation of the inverse number of steps requires running a larger number of simulated trajectories and would thus be computationally slower.
With few notable exceptions [21]–[25], the development of neural interfaces has proceeded along two separate tracks. There are sensory interfaces, such as the cochlear implants [26] that transform external physical events into neural stimuli for the brain and there are motor interfaces that decode activities from cortical regions to generate commands for external devices [1], [27]–[29]. However, the efficiency of biological motor behavior rests upon the seamless integration of sensory information and motor commands. This integration occurs both in our deliberate and conscious responses to external stimuli and in hardwired reflex responses organized by the neural circuitry of the spinal cord. In fact, the voluntary motor commands originating from the highest brain centers operate upon the world by modulating the activities and the response properties of the spinal networks. Here, we have taken a first step towards the design of a brain-machine interface that emulates the same basic principle: the interface has a sensory and a motor component whose direct interaction generates a system of automatic responses, which are to be modulated by volitional activities. In this sense, our proposed architecture draws inspiration from the natural “neural interface” that all vertebrates are endowed with: the spinal cord.
Unlike its biological counterpart however, the proposed interface is not connected to a musculoskeletal system, but can act over a broader family of dynamical systems. In this example, we chose a simulated point-mass moving within a viscous fluid. The interface generates position-dependent forces converging to a stable equilibrium point. This simple framework highlights an important issue in the design of brain-machine interface: the boundary between neural and artificial control. The parameters of the external system – in this case the viscous and inertial matrices – may result from a combination of passive physical elements and feedback control components. There is therefore an important role of the engineering design in establishing the dynamical properties of the external device, as it is seen by the neural system through the interface.
The concept that force fields afford a representation of the motor output in the spinal cord was first expressed in the aforementioned stimulation studies [4]–[8]. However, the mechanistic concept behind this representation can equally well characterize a variety of other observations, including some of the most classical ones. The stretch reflex first described by Sherrington [30] is one the clearest examples. Another example is spinal pattern generators that produce a different type of field, a field inducing a cyclical motion of the limbs. Grillner and coworkers [9] offered a compelling model of locomotion pattern in the lamprey, and in both cases the rhythmic activity is sustained by a phase-shift between the state of motion and the consequent forces. While the experimental tests in the current paper have been focused on the enforcement of equilibrium-seeking behavior, different behaviors are programmable through the approximation of different force-fields.
The description of the bidirectional neural interface as a force-field has a conceptual rationale in the causality of mechanical interactions between a control system and its environment [31], [32]. At the interface with the environment, a control system may act either as a generalized admittance, determining a state of motion in response to an applied force, or as a generalized impedance, determining a force in response to an applied state. Considerations about neuromuscular mechanics suggest the second case as more appropriate, because the mapping from state (position and velocity) to force is typically well defined but not invertible. In this sense too, the architecture of the interface reflects the organization of the biological motor system. However the extent of the similarity may vary depending on the structure that is being controlled. The dynamical parameters – for example the mass and viscosity – may be characteristics of the physical system that is been controlled by the interface. But they also may be – at least partially – introduced in the interface algorithms to shape a desired behavior. For example a virtual mass and a virtual viscosity can be added in parallel to the physical system to increase stability and modify the resulting trajectories.
Intelligent and purposeful motor behavior involves the ability to react to unexpected perturbations and to change planning goals. In this respect, the study presented in this report represents a preliminary step towards the development of an interface that facilitates exploration and adaptation providing its users with the possibility to modulate a field of forces. Even if the concept of controlling a limb by shifting its equilibrium position is not new [33]–[35], in the context of BMIs this is a radically new platform compared to current approaches based on decoding – instant by instant – the desired state of motion of the connected device, such as, for example, a robotic arm. Consider a reaching movement with a prosthetic hand. As the hand moves towards the target an obstacle is encountered that triggers a correction. The standard decoding method requires recreating an entire path that circumvents the obstacle and reaches the final target. In contrast, a field-based approach, reprogramming the path may be limited to shifting the hand position to a point that is clear of the obstacle and then let the field guide the hand towards the target without further reprogramming.
While early BMI studies were mostly focused on decoding motor cortical activities [29], [36], more recently there has been a growing interest for evoking somatosensory perception by electrical stimulation. For example Weber and co-workers are pursuing the stimulation of dorsal root ganglia, recreating patterns of evoked responses in somatosensory-area [37]. Recently, Venkatraman and Carmena [38] were able to stimulate neurons in the rat barrel cortex and to produce the sensation of an object being swiped by the whiskers. More recently yet, Nicolelis and coworkers were able to integrate in BMI motor cortical decoding with artificial tactile sensing elicited by microstimulation of S1 [21]. These results are consistent with earlier observations by Romo and coworkers who demonstrated the possibility to induce tactile sensation analogous to finger touch in monkeys [39]. Based on the available evidences, we expect the electrical stimuli generated by our interface to be adequate to induce somatosensory perception in the alert animal. Since we are stimulating in the barrel cortex, we predict – after Venkatraman and Carmena [38] – that the stimuli would induce perceptions analogous to whisking an object. However, in a brain-machine interface the ultimate goal would be to produce sensations corresponding to the state of an artificial device, such as a food feeder, whose structure may or may not resemble that of a biological limb. Understanding how the somatosensory system may adapt the perceptual correlate of electrical stimuli is a future research goal, beyond the scope and reach of the present study. Here, we focused on the production of automatic responses in the form of preprogrammed force fields, in the perspective that these responses may be both accessible and modifiable by volitional commands. Studies of current interfaces provide ample evidence demonstrating the ability of the mammalian brain to modulate the activities of populations of cortical neurons in different brain areas [27]–[29], [40], [41]. To the extent that this circuitry can be accessed and purposefully modulated by voluntary neural commands, the dNI will offer its user with the possibility to achieve motor goals in a stable manner and without the need for constant on-line supervision. At this time, however, the possibility that the force field produced by the interface may be accessible to volitional control remains to be demonstrated by additional experiments with alert animals. In particular, it will be critical establishing what field parameters may be modified by volitional inputs converging upon the neural structures that determine the output of the interface. We need to stress that the particular case of a convergent field is not the only that can be implemented and that has functional relevance. For example, a force field can be programmed to have rotational structure so as to induce cyclical motions of the controlled objects. Parallel pattern of forces, on the other hand, may approximate the control of a contact force. The simple case of the viscoelastic force field in our task provides the mathematical basis for generating stable trajectories – i.e. trajectories that converge to a nominal path in exponential time if displaced by an unexpected perturbation. In addition to expanding the behavioral repertoire of NIs, the bidirectional interface establishes a new venue for investigating the mechanisms of neural plasticity through a controlled exchange between cortical structures and a virtually unlimited repertoire of dynamical systems implemented either in hardware or by computer simulation.
This study was carried out in strict accordance with the Italian law regarding the care and use of experimental animals (DL116/92) and approved by the institutional review board of the University of Ferrara and by the Italian Ministry of Health (73/2008-B). For all experimental procedures, rats were anaesthetized with a mixture of Zoletil (30 mg/kg) and Xylazine (5 mg/kg) delivered intraperitoneally and all efforts were made to minimize suffering.
The experiments were carried out on 13 male Long-Evans rats, weighting 350–400 g and for the entire duration of the experiment, anesthesia was maintained with supplementary doses of anesthetic (intra-peritoneal or intra-muscular) such that a long-latency, sluggish hind limb withdrawal was sometimes achieved only with severe pinching of the hind foot. The anesthetized animal was placed in a stereotaxic apparatus (Myneurolab). A craniotomy was made, using a micro drill, over the primary somatosensory cortex (S1) and primary motor cortex (M1) whisker representations of the same hemisphere. To place the stimulation array, a small craniotomy (2×2 mm) was made in the parietal bone to expose the barrel cortex, which was identified according to vascular landmarks and stereotaxic coordinates [42]–[44]. The dura mater was not removed because the electrodes were sufficiently rigid to pass through it. The placement of the electrodes was tested and confirmed by recording the neuronal responses to manual whisker stimulation. The arrays were lowered perpendicular through the cortical surface using a hydraulic microdrive (2650, Kopf) at depth between 500 and 900 µm from the pia (granular layer) [45]–[47].
To insert the recording array, the frontal cortex was uncovered at 0.5 mm rostral and 0.5 mm lateral to bregma, and the vibrissal representation was exposed, at coordinates consistent with previous maps of the M1 whisker representations [12], [43], [48]–[50]. In preliminary experiments, we conducted intracortical microstimulation (monophasic cathodal pulses, 30 ms train duration at 300 Hz, 200 µs pulse duration with a minimum interval of 2.5 s) to evoke whisker twitches, at high threshold intensities, between 1.5–1.8 mm below the cortical surface. This depth was found to correspond to the layer V of granular cortex. The microwire array was lowered perpendicularly into the cortex to layer V at sites ranging from 1.0 to 2.5 mm lateral and 1.0 to 3.0 mm rostral to bregma. Also in this case the dura was not removed and was kept moist with a 0.9% saline solution. The effectiveness of the placement of stimulation and recording arrays was verified by computing peri-stimulus time histograms of neural responses to the different stimulation patterns (see Figure 6A–D for an example).
For both recording and stimulation procedure we used 16 polyimide-insulated tungsten electrodes microwire arrays (50 µm wire diameter, Tucker-Davis Technologies), configured in two rows of 8 electrodes each (250 µm electrode spacing and 375 µm rows separation) and placed over the primary somatosensory cortex (S1) and primary motor cortex (M1) whisker representations of the same hemisphere. Placement of electrodes was later confirmed by histological section.
The intracortical microstimulation (ICMS) consisted of trains of 10 biphasic pulses, each phase lasting 100 µs, delivered at 333 Hz with amplitude of 150 µA. Each stimulation train was delivered throughout two adjacent electrodes of the stimulation array using a programmable 8 channel stimulus generator (Stg4008, Multichannel Systems) built with a stimulus isolation unit for each output channel. Software-generated TTL triggers were used both to start the stimulation pattern and to store the stimulus timing in the recorded neural signals.
The recording microwire array was lowered perpendicularly into the cortex using a hydraulic microdrive (2650, Kopf) and extracellular neuronal discharges were recorded using a multichannel recording system (Map system, Plexon Inc.) with a sampling frequency of 40 KHz per channel.
During the experimental sessions an on-line PCA-based sorting procedure (illustrated in Figure 6C) was performed using commercially available software (Rasputin, Plexon Inc.). Time stamps of identified units were sent in real-time via local LAN to custom-made software developed in Matlab (Mathworks®) to translate the input neural signal into output stimulation triggers according to the behavior of the simulated controlled system.
We ensured that the neural responses used to guide the interface did not contain a component which reflected an electrical stimulation artifact rather than true neural response by the following steps: (i) we used only responses collected after the stimulation artifact had ended (i.e. the onset of neural response activity in each calibration trial and test trial started after the stimulation artifact ended) (ii) the templates of the on-line spike sorting procedure were established without including data collected during electrical stimulation (iii) we further verified by visual inspection that spikes identified near the onset had the same amplitude and shape of that identified far from the electrical stimulation (Figure 6A–C).
At the end of electrophysiological session, DC of 5 µA for 10 s was passed through electrodes placed both at the beginning and at the end of the array, to mark its position. The current produced a lesion that was easily seen in cytochrome oxidase-stained histological sections. When the acute experimental phase was completed, the animals were deeply anesthetized with Isoflurane and transcardially perfused with 500 ml of 0.1 M-phosphate buffered saline (PBS) with 0.9% NaCl at 37°C followed by a 1l cold buffered solution of 2.0% paraformaldehyde, 1.25% glutaraldehyde and 2.0% sucrose (pH 7.4). The brains were removed from their skulls, coronally transected at the level of bregma and then postfixed overnight at 4°C. The caudal portion, including S1, was saturated in 20% sucrose, then 30% sucrose until it sank. Coronal sections of frozen brain (60 µm thick) were cut on a sliding microtome (SM2000R, Leica) to determine the depth of microelectrodes tip. The sections were processed for cytochrome oxidase (CO) according to previous reports [51], [52] to identify layer IV. Sections were washed three times in a 0.1 M PB solution and then incubated at 37°C in a cytochrome-C oxidase staining solution containing 4% sucrose, 0.05% DAB, and 0.05% cytochrome C (Sigma Laboratories), until barrels were clearly delineated. Then sections were washed in PBS and mounted on slides. Mounted sections were dehydrated in a series of alcohols, defatted in xylene and coverslipped.
CO stained sections were observed under brightfield illumination with Olympus BX51 microscope (Olympus) interfaced with a color video camera (CX-9000) and with a NeuroLucida system (MicroBrightField) (Figure 6F). Using a 10× objective, live color images of the histological material were displayed on a high-resolution video monitor. The boundaries of the barrels were drawn using the image on the screen and the depth of the electrolytic lesions was measured by the Neurolucida software.
In this implementation, the device interacting bidirectionally with neural activity is a simulated point mass in a viscous medium. Typical values for the mass (M) and viscosity (B) were 10 Kg and 15 N•s/m. A linear force field results in the linear differential equation(14)with an isotropic stiffness (K) of 4 N/m, the ideal system driven by the noiseless linear field was slightly over-damped (damping ratio ). While the choice of these parameters is arbitrary, in a practical implementation, the parameters of the viscoelastic field (here, K and B) should be selected based on the desired time constant of the payload's motion. As the interface implements a piecewise constant approximation of the linear field, , corrupted by random background activity, the stability properties afforded by the desired continuous field can only be considered as an optimal limit. This first realization of the interface has some notable limitations. One is that the control law generates an output force in response to a position input. In a more complete system, the input should convey not only position, but state information, that is position and velocity. Here, the derivative component of the controller is a fixed property, expressed by the term in the dynamics equation. Another obvious simplification is in the choice of a point mass () for controlled object. A mechanical arm is generally characterized by a non-linear differential equation. However, the second order linear ordinary differential equation (Equation 14) is used in robotics to represent the error dynamics of non-linear systems controlled by proportional-derivative (PD) methods [53]:(15)with ( is a desired trajectory). In our framework, this PD control law can be reformulated as(16)where is a time varying function to be supplied by the voluntary input to the interface. In this case, the dNI would provide stability to a desired movement in a way analogous to the combined influence on limb movements of muscle mechanics and feedback mechanisms of the spinal cord. Therefore, while the form of Equation 14 is quite simple, it also expresses a fundamental mathematical representation for control.
By tuning the sensory and motor interfaces to approximate a predetermined force field, the dNI establishes an automatic behavior. The neural connections between the stimulated and the recorded populations determine the force to be generated at each position in the field. However, the recorded activities are also affected by inputs from other brain areas. In the alert brain, these additional inputs provide a pathway for the volitional commands to modulate the dynamics of the interface. To see this, suppose that the output of the interface is the programmed force field, (where ρ indicates the radial distance from the origin of the plane upon which the point mass moves) plus a force component, generated by a volitional command. The net force is then(17)This can be re-written as(18)where(19)is a time-varying equilibrium point. Thus, the dNI architecture provides a way to integrate voluntary commands with preprogrammed automatic responses so as to generate dynamically stable movements. A computer simulation study of the relationship between Information in neural activity, the mechanical parameters of the dynamical system and the performance of the neural interface is reported in [54].
As explained in Results, we considered the Mutual Information that the recorded neurons provide to guide the dynamic system. The latter was evaluated as the Mutual Information between the force vector expected to be generated by the electrical stimulation in a given trial (corresponding to the template force vector established during the calibration trials in response to the considered electrical stimulation) and the actual force vector obtained from the neural response using the algorithm described in the above Section:(20)where is the probability of presenting an electrical stimulation that leads to an expected force , is the probability of obtaining in a given trial a force vector when presenting an electrical stimulation that leads to an expected force , and is the probability of obtaining in a given trial a force vector unconditional to the type of electrical stimulation applied. High (respectively low) values of indicate instead a near-deterministic (respectively near-random) relationship between the force provided by the neurons and the one needed for guiding the dynamic system.
was computed from the data as follows. Since there is a one-to-one correspondence between and the type of electrical stimulation pattern, and since Mutual Information is invariant to monotonic transformations or relabeling of the variables, the patterns were labeled with the same index s (s = 1, …S) that indexes the electrical stimulation patterns. Then, the conditional probabilities of to each stimulation pattern s were computed as frequency-of-occurrence histograms from the trials to stimulus s. The values of the components and of the force were discretized into five equipopulated bins in order to facilitate the sampling of the empirical probability histograms. Then, the probability histograms were plugged into the above equation for and its value was computed numerically. It is well known that, because the empirical probabilities are estimated from a limited number of trials, these empirically obtained Information measures still suffer from an upward systematic error (bias) due to limited sampling [55]. We corrected for this bias as follows. First, we used a simple analytical procedure [56] to estimate and subtract out the bias of each Information quantity. We then applied the “shuffling procedure” described in [55]–[57], which greatly reduces the bias of multidimensional Information estimates. We then checked for residual bias by a “bootstrap procedure”: stimuli and responses were paired at random, and the Information for these random pairings was computed. Because in this random case the Information should be zero, the resulting value is an indication of a residual error. In this study we found (data not shown) that the bootstrap estimate of this residual error was very small and much smaller than the Information values obtained for optimal neural response parameters, indicating that our estimates of were reliable.
|
10.1371/journal.ppat.1006456 | The ApaH-like phosphatase TbALPH1 is the major mRNA decapping enzyme of trypanosomes | 5’-3’ decay is the major mRNA decay pathway in many eukaryotes, including trypanosomes. After deadenylation, mRNAs are decapped by the nudix hydrolase DCP2 of the decapping complex and finally degraded by the 5’-3’ exoribonuclease. Uniquely, trypanosomes lack homologues to all subunits of the decapping complex, while deadenylation and 5’-3’ degradation are conserved. Here, I show that the parasites use an ApaH-like phosphatase (ALPH1) as their major mRNA decapping enzyme. The protein was recently identified as a novel trypanosome stress granule protein and as involved in mRNA binding. A fraction of ALPH1 co-localises exclusively with the trypanosome 5’-3’ exoribonuclease XRNA to a special granule at the posterior pole of the cell, indicating a connection between the two enzymes. RNAi depletion of ALPH1 is lethal and causes a massive increase in total mRNAs that are deadenylated, but have not yet started 5’-3’ decay. These data suggest that ALPH1 acts downstream of deadenylation and upstream of mRNA degradation, consistent with a function in mRNA decapping. In vitro experiments show that recombinant, N-terminally truncated ALHP1 protein, but not a catalytically inactive mutant, sensitises the capped trypanosome spliced leader RNA to yeast Xrn1, but only if an RNA 5’ polyphosphatase is included. This indicates that the decapping mechanism of ALPH1 differs from the decapping mechanism of Dcp2 by leaving more than one phosphate group at the mRNA’s 5’ end. This is the first reported function of a eukaryotic ApaH-like phosphatase, a bacterial-derived class of enzymes present in all phylogenetic super-groups of the eukaryotic kingdom. The substrates of eukaryotic ApaH-like phosphatases are unknown. However, the substrate of the related bacterial enzyme ApaH, diadenosine tetraphosphate, is highly reminiscent of a eukaryotic mRNA cap.
| Eukaryotic mRNAs are stabilised by a 5’ cap and one important step in mRNA decay is the removal of this cap by the nudix domain protein Dcp2 of the decapping complex. The decapping complex is highly conserved throughout eukaryotes, with the exception of trypanosomes that lack the entire complex. Here, I show that trypanosomes have evolved to use an ApaH-like phosphatase instead of a nudix domain protein as their major decapping enzyme. This work closes an important gap in the knowledge of trypanosome mRNA metabolism. Moreover, this is the first reported function of an ApaH-like phosphatase, a bacterial derived class of enzymes that are widespread throughout eukaryotes.
| One hallmark of eukaryotic mRNAs is the mRNA cap, a 7-methyl-guanosine bound 5’-5’ to the mRNA’s 5’ end by a triphosphate bridge. Together with the poly(A) tail that is connected to the cap via the poly(A) binding protein and the eIF4F complex, the cap mediates mRNA circularisation and contributes to mRNA stabilisation. For mRNA degradation, the circular structure is resolved by the removal of the poly(A) tail by the deadenylase of the Caf1/Ccr4/Not complex. Deadenylated mRNAs are then targets for one of two alternative decay pathways. mRNA can be degraded 3’-5’ by the exosome followed by the hydrolysis of the remaining capped di- or oligo-nucleotides by the pyrophosphatase DcpS. Alternatively, mRNA is decapped by the Dcp1/Dcp2 complex, followed by 5’-3’ exonucleolytic decay by the major cytoplasmic exoribonuclease XRN1. In many eukaryotes, including yeast and trypanosomes, 5’-3’ decay is the dominant mRNA decay pathway. Enzymes of this decay pathway localise to RNA granules, cytoplasmic RNA protein aggregates of largely unknown function. Best studied are P-bodies, which are constitutively present and contain the mRNA decay machinery and stress granules, which are induced by exposure to cellular stress and contain translation initiation factors [1,2]. RNA granules are present in all eukaryotes, including trypanosomes [3].
The nudix hydrolase Dcp2 is the catalytic component of the decapping complex [4–7] Dcp2 cleaves the mRNA cap between the α and β phosphate, releasing m7GDP and 5’-end monophosphate RNA [4,7]. Dcp1 binds to Dcp2 and acts as an activator [8,9]. The activity of the Dcp1/Dcp2 complex is further increased by several decapping enhancers; the ones conserved from yeast to human are Edc3, Pat1, Dhh1, Scd6 and the Lsm1-7 complex [10]. Edc3 and Pat1 bind and stimulate Dcp2 directly [11,12]. Pat1 binds and recruits the Lsm1-7 complex which mediates the selective binding to deadenylated mRNA substrates [13]. The effects of the DEAD-box RNA helicase Dhh1 and the Lsm domain protein Scd6 on decapping are probably only indirectly: both repress translation which increases the amount of decay-competent mRNA substrates [14–16]. At least in mammals, two additional nudix domain proteins are involved in decapping a subset of mRNAs [10].
To date, mRNA decapping by nudix domain proteins, in particular by Dcp2, is the only known mechanism of mRNA cap removal in eukaryotes. The enzyme is widespread throughout the eukaryotic kingdom: it was described in yeast [17], mammals [7] and plants [18], but also in several deep-branching protozoa, for example Entamoeba histolytica [19] and Giardia lamblia [20]. One marked exception are Kinetoplastida, a heterogenic group of flagellated protozoa that include some prominent human pathogens such as Trypanosoma brucei, Trypanosoma cruzi and Leishmania. Kinetoplastida have neither homologues to Dcp1 and Dcp2 nor to the major decapping enhancers Pat1 or Edc3 [21] and they do not possess a cytoplasmic Lsm1-7 complex [22,23]. Only the translational repressors Dhh1 [24–26] and Scd6 [27,28] are conserved. Despite of the absence of the decapping complex, the release of m7GDP from mRNAs is detectable in vitro [29]: trypanosomes must have evolved a mechanism for mRNA decapping, that is different to the one of other eukaryotes. One reason for the trypanosome’s need of a different decapping mechanism could be the unusual mRNA cap that in Kinetoplastida is transferred from the spliced leader RNA to all polycistronically transcribed mRNAs in a trans-splicing reaction [30]. The cap is of the heavily methylated type 4: the first four transcribed nucleotides (AACU) have ribose 2’-O methylations and there are additional base methylations on the first (m62A) and fourth (m3U) position [31,32]. Importantly, the differences in mRNA decay between Kinetoplastids and other eukaryotes are restricted to the decapping complex. Trypanosomes have conserved homologues to all other mRNA decay components, including the exosome [33], the Xrn1 homologue XRNA [34,35] and the CAF1/NOT complex [36,37].
We have recently purified starvation stress granules from trypanosomes [38]. One of the newly identified granule proteins, here renamed ALPH1, is a strong candidate for the long-wanted trypanosome decapping enzyme, based on its predicted substrate specificities and on its exclusive co-localisation with XRNA to a special granule at the posterior pole of the cell. RNAi experiments confirm that TbALPH1 has all features required of an mRNA decapping enzyme: ALPH1 depletion is lethal and results in a massive, global increase in mRNAs, that are deadenylated but have not yet started degradation. In vitro experiments confirm that ALPH1 can sensitise a capped RNA to Xrn1 in the presence of an RNA 5’ polyphosphatase.
The protein with the Gene ID number Tb927.6.640 was identified both in stress granules [38] and within the mRNA-bound interactome [39]. Interestingly, the protein did not only localise to stress granules, but also to the posterior pole granule [38], a localisation that was so far only observed for the 5’-3’ exoribonuclease XRNA [40]. This observation indicates that Tb927.6.640 might belong to the same pathway as XRNA and perhaps is the long-wanted decapping enzyme.
Tb927.6.640 is annotated as a kinetoplastid-specific phospho-protein phosphatase (PPP) in the TriTryp database [21]. A closer analysis of the sequence reveals two changes in the conserved PPP signature motif GDXXDRG: the second aspartate is replaced by asparagine and arginine is replaced by lysine (Fig 1A). These changes are characteristic for an ApaH-like phosphatase (Alph), a subgroup of the PPP family that is closely related to the bacterial enzyme ApaH [41,42]. I will refer to Tb927.6.640 as TbALPH1. There are two further ApaH-like phosphatases in the T. brucei genome: Tb927.4.4330 (TbALPH2) and Tb927.8.8040 (TbALPH3) (S1 Fig), but neither was identified as stress granule component [38] or as involved in mRNA binding [39]. The substrates of eukaryotic ApaH-like phosphatases are unknown, even though these proteins are widespread throughout the eukaryotic kingdom [41]. The substrate of the bacterial ApaH protein, however, is diadenosine tetraphosphate [43,44], a molecule that resembles an eukaryotic mRNA cap (Fig 1B).
TbALPH1 has 734 amino acids and is almost equally split into a C-terminal domain, the catalytic domain and an N-terminal domain (Fig 1C). ALPH1 of T. brucei and T. evansi also has an asparagine-rich region within the N-terminal domain. The protein appears to be unique to Kinetoplastida; there are similarities to both Alph’s of other eukaryotes and to bacterial ApaH proteins, but these are restricted to the catalytic domain. When all available ALPH1 sequences of the Kinetoplastida are aligned, there are 14.6%, 57.6% and 31.9% minimal identity between any two sequences in the C-terminal part, catalytic domain and N-terminal part of the protein, respectively (Fig 1C). Thus, the catalytic domain is best conserved, followed by the C-terminal domain and a rather poorly conserved N-terminal domain. When the ALPH1 sequences of Trypanosomes and Leishmania/Leptomonas were aligned separately, the percentages of minimal identities increased, but the N-terminal domain remained the least- and the catalytic domain the best-conserved domain (Fig 1C).
If ALPH1 is involved in decapping of bulk mRNAs, its depletion should be lethal and cause stabilisation of total mRNAs. RNAi depletion of TbALPH1 was performed with the tetracycline inducible system described in [45]. Three populations of cells of clonal origin were analysed. The reduction in TbALPH1 mRNA was controlled by Affymetrix single molecule FISH with a green fluorescent probe antisense to ALPH1 and a red fluorescent probe antisense to a control mRNA (DBP1). There was a clear reduction in the number of ALPH1 mRNAs (from 6 to 1 molecule per cell), but not in the number of DBP1 mRNAs within 24 hours of tetracycline induction in all three clones, indicating that the RNAi was successful (Fig 2A). Next, growth was monitored over a time-course of RNAi induction. Cells stopped growth within 48 and 72 hours of RNAi induction (Fig 2B), indicating that TbALPH1 is an essential protein.
The effects of ALPH1 RNAi depletion on mRNA levels were examined. Total RNA was isolated over a time-course of ALPH1 RNAi induction and analysed by quantitative northern blots (Fig 2C–2E). The blots were probed for total mRNAs with an oligo antisense to the miniexon sequence that is trans-spliced to all trypanosome mRNAs, and for two individual mRNAs. The two individual mRNAs, GPI-phospholipase C (GPI-PLC) and phosphoglycerate kinase C (PGKC) were chosen because they are unstable in the procyclic life cycle stage, in which the experiments were performed [46–49]; unstable mRNAs show a larger increase in steady state levels at inhibition of RNA decay. Total RNA from bloodstream form (BSF) trypanosomes served as a control. RNAi depletion of ALPH1 caused a significant increase in mRNA levels for all mRNAs analysed. Total mRNA levels increased 1.7/2.2 fold after 48/72 hours of ALPH1 RNAi induction. For the two developmentally regulated mRNAs the increase was higher: GPI-PLC mRNA increased 4.2/4-6 fold and PGKC mRNA increased 16/22 fold after 48/72 hours of RNAi induction.
The data suggest that ALPH1 has an essential role in trypanosome mRNA metabolism. The RNAi phenotype of ALPH1 is very similar to the phenotype observed after RNAi depletion of XRNA: cells stop growth [50], total mRNA levels increase about 2-fold (S2 Fig) and there is a particular pronounced effect on unstable, developmentally regulated mRNAs [35,51].
If ALPH1 is the decapping enzyme, it should act downstream of mRNA deadenylation. Thus, the mRNAs that accumulate after ALPH1 depletion should have shorter poly(A) tails. Such a decrease in poly(A) tail length was observed after depletion of several yeast enzymes acting downstream of deadenylation, including Dcp2 [17], Xrn1 [52], Dcp1 [53] and Lsm1 [54].
In trypanosomes, the histone H4 mRNA is frequently used to report changes in poly(A) tail lengths [34,36]. The genes encoding histone H4 are organised as ten tandem repeats, producing mRNAs with an average size of 520 nucleotides (range 469–589), excluding the poly(A) tail. The small size of this mRNA allows the visualisation of changes in poly(A) tail lengths as band shifts on a northern blot, because the poly(A) tail occupies a large fraction of the total mRNA size [34,36].
When a northern blot with RNA harvested over a time-course of ALPH1 RNAi was probed for histone H4, there was a clear decrease in mRNA size starting at 48 hours of tetracycline induction (Fig 3A). Two experiments served to control that the band-shifted mRNA is indeed deadenylated (Fig 3B): First, when the RNA samples were treated with RNAse H in the presence of oligo (dT), the band shift was lost. Second, no band shift is detectable in RNAs without a poly(A) tail, namely the 5.8S rRNA or the SL RNA, even though both are significantly smaller than histone H4. As a negative control, I performed RNAi specific to the deadenylase CAF1 (S3 Fig). There was a mild increase in histone H4 mRNA size (Fig 3C), as previously reported [36]. As a positive control, I performed RNAi specific to XRNA [50]. I observed a decrease in histone H4 mRNA size (Fig 3D), as previously reported [34], similar to the effect seen after ALPH1 RNAi. The data are consistent with ALPH1 acting downstream of the mRNA deadenylase CAF1 on deadenylated mRNA substrates.
So far, the RNAi phenotypes of XRNA and ALPH1 are identical. Both are essential proteins and RNAi depletion causes an accumulation of deadenylated mRNAs, suggesting that both enzymes act downstream of deadenylation. However, if ALPH1 is the decapping enzyme, it should be involved in the initiation of mRNA decay, while XRNA is involved in the actual mRNA degradation process.
I have recently developed a method that enables to distinguish between these two enzyme functions [55]. Briefly, an endogenous very long mRNA (Tb427.01.1740) is used as a reporter to visualise mRNA degradation intermediates. The extreme 5’ and 3’ ends of this mRNA are simultaneously probed with a red and green fluorescent single molecule mRNA FISH probe (Affymetrix), respectively. This way, intact mRNA molecules appear as yellow spots, mRNAs in 5’-3’ decay are green and mRNAs in 3’-5’ decay or in transcription appear as red spots (Fig 4A). Due to the large size and short half-life of the reporter mRNA, there are only 21% of intact mRNAs, but 48% are in 3’-5’ decay [55]. RNAi depletion of XRNA for 48 hours resulted in a 5.6 fold increase in yellow spots and a 4.3 fold increase in green spots [55] (S4 Fig). Thus, a reduced level of XRNA does not only cause an increase in intact mRNA molecules, but also an increase in 5’-3’ decay intermediates; the most likely explanation for this is that XRNA is not 100% processive over the extreme length of the reporter mRNA. Any early release of XRNA from its substrate thus causes an increase in 5’-3’ decay intermediates because the time of decay-reinitiation is increased due to the shortage in XRNA molecules. In contrast to the RNAi phenotype of XRNA, RNAi depletion of any enzyme involved in the initiation of mRNA decay should cause an increase in intact mRNA molecules (yellow spots) only.
To examine whether ALPH1 functions in mRNA decay initiation or mRNA degradation, cells were harvested over a time-course of ALPH1 RNAi and probed for the mRNA metabolism reporter Tb427.01.1740 [55]. The numbers of red, green and yellow spots were counted. There was a 1.5/3.5 fold increase in yellow spots (intact mRNA molecules) after 24/48 hours of RNAi induction, but no change in green or red spots (Fig 4B and 4C) (S5 Fig). The spots were further classified as either nuclear or cytoplasmic based on their co-localisation with the DAPI-stained nucleus on a Z-stack projection image. The increase in yellow spots was entirely due to an increase in cytoplasmic yellow spots (Fig 4B), as expected as a result of a decapping arrest.
These data now show a clear difference between the RNAi phenotypes of XRNA and ALPH1: depletion of XRNA RNAi results in an increase in both decay intermediates and intact mRNAs, while depletion of ALPH1 only causes an increase in intact mRNAs. The data suggest that ALPH1 acts upstream of XRNA in the initiation of mRNA decay, but downstream of deadenylation. The only known enzyme that fulfils these criteria is the decapping enzyme.
As part of a screen, we have previously shown that TbALPH1 localises to both the posterior pole of the cell as well as to starvation stress granules and P-bodies [38]. A localisation to the posterior pole was previously observed only for XRNA but for no other proteins involved in mRNA metabolism [40].
To investigate a possible co-localisation between TbALPH1 and TbXRNA, I expressed ALPH1 with a C-terminal eYFP tag from the endogenous locus, together with XRNA-mChFP. Importantly, a C-terminally tagged ALPH1 is fully functional, as cell lines entirely dependent on the fusion protein could be obtained by deleting the second ALPH1 allele and showed no reduction in growth as well as the correct localization of the fusion protein (S6 Fig). As a control, I also co-expressed TbALPH1-eYFP with an N-terminal mChFP fusion of the DEAD box RNA helicase DHH1, a P-body marker protein that is absent from the posterior pole [40].
I observed full co-localisation between XRNA and ALPH1 to the posterior pole granule at heat shock as well as to both starvation stress granules and the posterior pole granule when cells were treated with PBS (Fig 5A and S7 Fig). In contrast, the co-localisation between ALPH1-eYFP and mChFP-DHH1 was only partial, as previously observed [38]. ALPH1 and DHH1 co-localised to P-bodies in untreated cells and to most starvation stress granules in starved cells (Fig 5B). DHH1 was absent from the posterior pole granule and ALPH1 was absent from some starvation stress granules, in particular from the ones localised closest to the posterior pole granule (Fig 5B and S8 and S9 Figs).
The exclusive co-localisation between ALPH1 and XRNA is further evidence for both proteins being members of the same pathway.
The above data provide strong in vivo evidence for ALPH1 being the trypanosome decapping enzyme. To directly show the decapping activity in vitro, recombinant ALPH1 protein was purified from E. coli. The full length ALPH1 protein could not be expressed in a soluble form, and a truncated version lacking the N-terminal 221 amino acids (ALPH1ΔN) was therefore used. As a control, the same truncated version was expressed as an inactive mutant (Alph1ΔN*). For this, the highly conserved GDVHG motif involved in metal ion binding was mutated to GNVHG [56]. A Coomassie-stained gel with both purified proteins is shown in S10 Fig.
Next, the purified proteins were incubated with total trypanosome mRNA in the absence and presence of yeast Xrn1. If ALPH1 is the decapping enzyme, it should sensitise capped mRNAs to Xrn1. All RNA samples were analysed by northern blots probed for the capped SL RNA and, as a positive control, for the uncapped 5.8S rRNA (Fig 6). While the 5.8S rRNA disappeared in the presence of Xrn1, the ALPH1ΔN-treated SL RNA did not, indicating that ALPH1ΔN by itself does not produce RNA substrates for Xrn1 (Fig 6, lane 4). However, the addition of ALPH1ΔN, but not of the inactive Alph1ΔN* mutant, reproducibly caused a small band shift of the SL RNA that was absent in the similar-sized uncapped 5.8S rRNA (for example lane 3 and 4 in Fig 6). This indicated that ALPH1 targets the SL RNA, but without producing a substrate for Xrn1. One explanation is that ALPH1 does not produce a monophosphorylated mRNA, but a di- or triphosphorylated mRNA, which would be resistant to Xrn1 degradation. To test this, ALPH1-treated samples were treated with an RNA 5’ polyphosphatase (5’PP), which transforms tri- or diphosphorylated RNA into monophosphorylated RNA. When this phosphatase was included, subsequent Xrn1 treatment caused disappearance of the SL RNA band (Fig 6, lane 5). This was not observed with the catalytically inactive mutant Alph1ΔN* or with any other combinations of the enzymes (Fig 6, lanes 6–10). Thus, ALPH1 sensitises a capped RNA to yeast Xrn1 if an RNA 5’ polyphosphatase is included. This is consistent with ALPH1 being the decapping enzyme, producing a di- or triphosphorylated mRNA.
Recently, one of the five trypanosome nudix domain proteins was renamed TbDCP2 because it has in vitro decapping activity [57]. The similarity to Dcp2 of other eukaryotes is low and the decapping activity is very poor, when the mRNA substrate had a type 4 cap [57]. Moreover, the protein was not purified with trypanosome RNA granules [38] and not identified as mRNA binding protein or protein involved in mRNA regulation in recent screens [39,58]. If TbDCP2 is responsible for the decapping of trypanosome mRNAs, its RNAi phenotype should resemble the RNAi phenotype of ALPH1. For three independent clonal cell lines, we found that RNAi depletion of TbDCP2 did not affect trypanosome growth, caused no increase in total mRNA levels and no change in the adenylation stage of the histone H4 mRNA (S11 Fig). Together, the data provide no evidence for TbDCP2 being the major trypanosome decapping enzyme and thus the functional orthologue to Dcp2.
I show that the trypanosome ApaH-like phosphatase ALPH1 has all characteristics of an mRNA decapping enzyme. The enzyme is essential, localises to P-bodies and its depletion causes the accumulation of deadenylated mRNA molecules that have not yet started degradation. Its exclusive co-localisation with the 5’-3’ exoribonuclease XRNA suggests that both enzymes are members of the same pathway. Importantly, total mRNA levels increase to the same extent upon TbALPH1 and XRNA RNAi (compare Fig 2C and S2 Fig), ruling out that TbALPH1 only targets a small sub-group of mRNA molecules. In vitro experiments show that ALPH1ΔN, but not an inactive mutant of ALPH1 can sensitise the capped SL RNA to yeast Xrn1, as long as an RNA 5’ polyphosphatase is included. In contrast, depletion of the nudix domain protein TbDCP2 that was previously suggested to be the decapping enzyme, did not affect growth, mRNA levels or mRNA adenylation state. This is consistent with its in vitro activity being very low when a mature type 4 cap was used as a substrate [57]. I propose that TbALPH1 is the major trypanosome decapping enzyme and thus the functional orthologue to DCP2 of other eukaryotes.
The in vitro decapping experiments provide evidence for the decapping mechanism of ALPH1 being fundamentally different from the DCP2-mediated decapping. The data are consistent with ALPH1 leaving either a tri- or a diphosphate at the 5’ end of the decapped mRNA, instead of a monophosphate. A diphosphate is more likely than a triphosphate, as the bacterial ApaH protein cleaves phosphoanhydrid bonds rather than phosphoester bonds. In previous in vitro decapping assays using cell extracts of the Trypanosomatid Leptomonas, three products were released from an artificial RNA substrate with a type 0 cap: m7GDP, m7GMP and m7GpppG [29]. The release of m7GMP was interpreted as the result of a cap scavenger activity [29], the degradation of the cap structure after 3’-5’ exosomal decay [59]. However, more recent data provide little evidence for a cytoplasmic function of the trypanosome exosome [50] and it is possible that the detected m7GMP is in fact the product of ALPH1. It remains an unsolved question, how the uncapped, poly-phosphorylated mRNA is further degraded. One possibility is that the trypanosome 5’-3’ exoribonuclease, unlike yeast Xrn1, accepts poly-phosphorylated mRNAs as a substrate. If this is not the case, the trypanosome decapping complex must contain an RNA phosphatase that produces monophosphorylated mRNA substrates for XRNA. Notably, all in vitro experiments were performed with an N-terminally truncated ALPH1 protein. Thus, there also remains the possibility that the N-terminus of ALPH1 affects the catalytic mechanism.
To date, all reported decapping enzymes that act on intact mRNA caps are nudix domain proteins. This includes DCP2, long believed to be the only eukaryotic decapping enzyme, as well as the more recently discovered enzymes Nudt3 and Nudt16 that are responsible for decapping a subset of mRNAs in mammals [10,60–63]. Even in bacteria, mRNA decay is initiated by the nudix hydrolase RppH, which cleaves pyrophosphate from the mRNA’s 5’ terminal triphosphate [64]. Homologues to DCP2 are readily detectable in representative organisms of all major eukaryotic super-groups defined by [65]: Opisthokonta (yeast, human), Amoebozoa (Dictyostelium), Archaeplastida (Arabidopsis), SAR (Plasmodium), CCTH (Cryptophyta, Haptophyta) and Excavata (Naegleria, Trichomonas). Loss of DCP2 is selectively observed in (at least) two subgroups of the Excavata, the Euglenozoa (Kintoplastida and Euglena gracilis) and Fornicata (Giardia intestinalis and Spironucleus salmonicida). Only the kinetoplastid Perkinsela has a DCP2 homologue; it may be the product of a lateral gene transfer event as it has closest homology to DCP2 of plants and Perkinsela lacks all other components of the decapping complex. Does the loss of DCP2 correlate with the gain of ALPH1? All Kinetoplastida and Euglena do have a homologue of TbALPH1. The ALPH1 homologue of Euglena gracilis constitutes almost solely of the catalytic domain, but is still more similar to TbALPH1 than to TbALPH2 or TbALPH3. In contrast, neither Giardia intestinalis nor Spironucleus salmonicida has a homologue to TbALPH1. Thus, ALPH1 has replaced DCP2 in Euglenozoa, but Fornicata must compensate the absence of DCP2 in a different way. Notably, there are many eukaryotes that have ApaH like phosphatases in addition to a DCP2 homologue; these include eukaryotes of all branches, with the marked exception of vertebrates, insects or land plants [41,66]. The homology to ALPH1 is restricted to the catalytic domain and whether these are decapping enzymes, acting in addition to DCP2, or have evolved a different function, remains to be investigated. To my knowledge, no function of an ApaH-like phosphatase has yet been reported.
The described changes in mRNA decapping of Euglenozoa are not restricted to the replacement of DCP2 by an ApaH like phosphatase: the genomes of Kinetoplastids and Euglena gracilis also lack readily identifiable homologues to the decapping enhancers DCP1, EDC3 and Pat1. This indicates that the mechanistic differences between the mRNA decapping reactions of Euglenozoa and all other eukaryotes are major. Why have Euglenozoa evolved a mechanism for mRNA decapping that is fundamentally distinct to mRNA decapping present in the rest of the eukaryotes? SL RNA based trans-splicing cannot be responsible, as this is also present in several other eukaryotes that do have a DCP2 homologue, for example Nematodes and Dinoflagellates. However, to the best of my knowledge, only SL RNAs of Kinetoplastids have been found to be of the heavily methylated type 4. A recent publication shows that mammalian mRNAs that have a N6,2′-O-dimethyladenosine as the first nucleotide following the m7G cap are resistant to decapping by Dcp2 [67]; the corresponding nucleotide of the trypanosome type 4 cap is with three methyl groups even heavier methylated (Fig 1B). Thus, the unusual mRNA cap structure of Euglenozoa may require an unusual decapping enzyme.
Uniquely, a fraction of TbALPH1 and XRNA co-localise to a special granule at the posterior pole of the cell that is devoid of any other RNA binding proteins. I never observed any accumulation of mRNA degradation intermediates at the posterior pole (Fig 4 and [55], indicating that it is not the place of mRNA decay. I propose that instead, the granule separates the decay enzymes ALPH1 and XRNA from their mRNA targets and thus flexibly regulates the amount of mRNA decay as needed by the cell. The fraction of XRNA at the posterior pole is highly variable and depends on the cellular state of mRNA metabolism [40] and I found the same for ALPH1 (S. Kramer, manuscript in preparation).
ApaH-like phosphatases are present in all branches of the eukaryotic kingdom. With ALPH1, I report the first function for a eukaryotic ApaH-like phosphatase. In the absence of DCP2, Trypanosoma brucei, and possibly all Euglenoids use ALPH1 as their major mRNA decapping enzyme. This is an example of convergent evolution. TbALPH1 is essential, has no homologue in vertebrates and there is a crystal structure available for the closely related TbALPH3 [68]. These are some requirements for the employment of ALPH1 as a drug target against human diseases caused by Kinetoplastida, such as Leishmaniasis or Trypanosomiasis.
Trypanosoma brucei Lister 427 procyclic cells were used for most experiments. All RNAi- and overexpression experiments were done in Lister 427 pSPR2.1 cells that express a TET repressor [45]. RNAi or overexpression transcripts are under the control of a TET operator and expression is thus inducible with tetracycline. Cells were cultured in SDM-79 [69] at 27°C and 5% CO2. Transgenic trypanosomes were generated using standard procedures [70]. All experiments used logarithmically growing trypanosomes. For starvation, one volume of cells was washed once in one volume PBS and cultured for 2 hours in one volume PBS. Heat shock was done for two hours at 41°C either in a thermoblock or waterbath.
ALPH1 (SK335): ALPH1 open reading frame without stop codon in pJET1.2 (Thermo-Fisher Scientific) with N-terminal BspMI site to allow HindIII-compatible overhangs (sequence upstream of start codon: ACCTGCactAAGCTTCCGCCACC) and C-terminal BglII site (sequence downstream of last codon: GGTTCTagatctTGATCA) in frame to allow HindIII/BamHI compatible cloning into plasmids based on [71] or [45].
ALPH1 RNAi (SK309). The C-terminal 706 nts of the ALPH1 open reading frame were cloned head to head into p3666 [45].
XRNA RNAi (3862): as described in [50].
TbDCP2 RNAi (SK435). The C-terminal 698 nts of the DCP2 open reading frame were cloned head to head into p3666 [45].
CAF1 RNAi (SK431). The C-terminal 827 nts of the CAF1 open reading frame were cloned head to head into p3666 [45].
XRNA-mChFP, endogenous expression (SK346): The C-terminal 1107 nts of the XRNA ORF were cloned into p2705 [71]. The plasmid was linearized with NheI.
ΔALPH1 (SK390): A blasticidine resistance cassette was flanked by the 560 nts upstream of the ALPH1 ORF and 607 nts downstream of the ALPH1 ORF to allow deletion of ALPH1 by homologous recombination.
ALPH1-eYFP-4Ty1, endogenous expression (SK169): The C-terminal 706 nts of ALPH1 were cloned into SK141 [38], which is p2710 [71] with 4-Ty1 tags downstream of the eYFP. The plasmid was linearized with SalI. In the text, this plasmid is often referred to as ALPH1-eYFP to avoid confusion, as the Ty1 tag is not needed for the experiments shown here.
ALPH1-eYFP, endogenous expression (SK348): The C-terminal 706 nts of ALPH1 were cloned into p2948, which is p2710 [71] with a hygromycin resistance instead of neomycin resistance. This plasmid was used instead of SK169 to allow co-expression of ALPH1-eYFP with XRNA-mChFP (SK346). The plasmid was linearized with MscI.
mChFP-DHH1, endogenous expression (p2845): as described in [40].
His-ALPH1ΔN for bacterial expression (SK391): The ALPH1 open reading frame lacking the N-terminal 663 nucleotides was cloned into a modified pET15b plasmid for expression in E. coli. This resulted in IPTG inducible expression of an ALPH1ΔN protein with an N-terminal extension of MGSSHHHHHHSSGLVPRGSHMELYFQEASAT.
His-Alph1ΔN* for bacterial expression (SK468): As SK391, but the GDVHG encoding sequence was mutated to the GNVHG encoding sequence (D:N; GAC:AAC).
Affmyetrix single molecule mRNA FISH was done as described previously [55] using the QuantiGene ViewRNA ISH Cell Assay (Affymetrix), protocol for glass slide format. Affymetrix probe sets were: ALPH1 (green, type 4): The N-terminal 1500 nts antisense to the ALPH1 ORF. DBP1 (red, type 1): the first 1260 nts antisense to the DBP1 ORF. CAF1 (red, type 1): antisense to the full ORF of CAF1. 5’ probe set (red, type 1, AF19): antisense to 288 nts upstream of the Tb427.01.1740 ORF followed by the first 812 nts of the ORF [55]. 3’ probe set (green, type 4): antisense to the last 656 nts of the Tb427.01.1740 ORF followed by the 444 nts downstream of the stop codon [55].
Z-stack images (100 stacks at 100 nm distance) were taken with a custom build TILL Photonics iMic microscope equipped with a sensicam camera (PCO), deconvolved using Huygens Essential software (Scientific Volume Imaging B. V., Hilversum, The Netherlands). All images are presented as Z-projections (method sum sliced) produced by ImageJ software. eYFP was monitored with the FRET-CFP/YFP-B-000 filter, mCherry and type 1 Affymetrix probes with the ET-mCherry-Texas-Red filter, type 4 Affymetrix probes with the ET-GFP filter and DNA with the DAPI filter (Chroma Technology CORP, Bellows Falls, VT).
The organisms used for the sequence analysis and alignments in Fig 1C were: Trypanosoma brucei TREU927, Trypanosoma brucei Lister strain 427, Trypanosoma evansi strain STIB 805, Trypanosoma brucei gambiense DAL 972, Trypanosoma congolense IL3000, Trypanosoma vivax Y486, Trypanosoma cruzi marinkellei strain B7, Trypanosoma cruzi Sylvio X10/1, Trypanosoma cruzi CL Brener Esmeraldo-like, Trypanosoma cruzi Dm28c, Trypanosoma cruzi CL Brener Non-Esmeraldo-like, Trypanosoma rangeli SC58, Trypanosoma grayi ANR4, Leptomonas seymouri ATCC 30220, Leishmania tarentolae Parrot-TarII, Leishmania sp. MAR LEM2494, Leishmania aethiopica L147, Leishmania tropica L590, Leishmania mexicana MHOM/GT/2001/U1103, Leishmania gerbilli strain LEM452. The ALPH1 sequence of Trypanosoma grayi ANR4 lacks the N-terminal domain, most likely due to a sequencing/database error; this organism was excluded from the N-terminal alignments.
Northern blots were done as previously described [40]. mRNA was prepared with the miRNeasy kit instead of the RNeasy kit (both Qiagen), if the detection of small RNAs (SL-RNA or 5.8S rRNA) was aimed. 18S rRNA and 5.8S rRNA were detected with antisense oligos coupled to IRDye 800, namely 5′ -CCTTCGCTGTAGTTCGTCTTGGTGCGGTCTAAGAATTTC-3′ and 5’-ACTTTGCTGCGTTCTTCAACGAAATAGGAAGCCAAGTC-3’, respectively. Total mRNA and SL RNA were detected by an oligo antisense to the miniexon sequence (5′-CAATATAGTACAGAAACTGTTCTA ATAATAGCGTT-3′), coupled to IRDye 700.
Blot images were collected with the Odyssey Infrared Imaging System (LI-COR Biosciences, Lincoln, NE) and quantified with ImageJ. The histone H4, GPI-PLC and PGKC probes were radioactively labelled DNAs of the entire ORFs; for the DCP2 probe the C-terminal 697 nts were used. Blot images were collected with a phosphorimager and analysed with ImageJ.
For the first step (decapping), each 100 μl reaction contained 10 μl NEB buffer 3 (100 mM NaCl, 50 mM Tris-HCl, 10mM MgCl2, 1mM DTT, pH 7.9), 2 μl Ribolock (Thermo Fisher Scientific), 20 μg total trypanosome RNA prepared with the miRNeasy kit (Qiagen) and if required 2 or 20 μl recombinant ALPH1ΔN or Alph1ΔN* (corresponding to about 800 ng protein), respectively. After one hour incubation at 37°C, 200 μl water was added to increase the volume and the RNA was cleaned by extraction with 300 μl phenol and subsequent ethanol precipitation in the presence of 2 μl RNA grade glycogen (Thermo Fisher Scientific). The pellet was resuspended in the amount of water needed for the 20 μl reaction of the second step (RNA 5’ polyphosphatase treatment). To each reaction, 2 μl RNA 5’ polyphosphatase buffer (Epicentre) and 1 μl Ribolock were added and if required, 1 μl RNA 5’ polyphosphatase (Epicentre). After one hour incubation at 37°C, 280 μl water was added and the RNA was cleared and precipitated as above, but without adding additional glycogen.
The pellet was resuspended in the amount of water needed for the 100 μl reaction of the third step (Xrn1 treatment). To each reaction, 10 μl NEB buffer 3 and 2 μl Ribolock were added and, if required, 2 μl recombinant yeast Xrn1 (NEB). The samples were incubated at 37° for one hour and the RNA was cleared and precipitated as above. The pellet was resuspended in 6 μl water and the entire sample was analysed by northern blot.
One litre of RosettaBlue competent bacteria cells (Novagen) carrying the plasmids SK391 or SK468 was started from an over night culture. At an OD of 0.8, protein expression was induced with 0.02% Isopropyl-β-D-thiogalactopyranosid (IPTG) for 2 hours. Cells were harvested (15 min, 10.000 g), resuspended in 40 ml buffer A (300 mM NaCl, 50 mM NaH2PO4, pH 8) and disrupted by ultrasonication (Sonifier B-12, Branson). The non-soluble material was pelleted by centrifugation (15 min, 20.000 g) and the soluble material (supernatant) was incubated with 3 ml Nickel-NTA agarose (Qiagen) equilibrated in buffer A for 60 min while rotating. The agarose was washed 3 times for 5 min in 50 ml buffer A and poured into a 2 ml Pierce centrifuge column. For washing, the column was loaded successively with 5 ml buffer A, 5 ml 10 mM imidazole and 5 ml 20 mM imidazole. Elution was done in two steps with 5 ml 100 mM imidazole and 5 ml 250 mM imidazole. Both eluates were rebuffered into PBS with PD10 columns (GE healthcare). The majority of the proteins eluted with 100 mM imidazole, but the 250 mM eluate contained less contaminants and was therefore used for the enzyme assays.
4 μg total RNA was incubated with or without 2 μl oligo (dT) (100 μM) in 16 μl volume at 65°C for 10 minutes. After slowly cooling down to room temperature, 2 μl RNAse H buffer, 1 μl Ribolock (ThermoScientific) and 1 μl RNAse H (NEB) or water (control) was added. The reactions were incubated at 37°C for one hour. RNA was purified by phenol-extraction followed by ethanol precipitation in the presence of 1 μl glycogen and 1 μl tRNA (25 mg/ml). RNA pellets were resuspended in 6 μl water for northern blot analysis.
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.