Different selective pressures shape the evolution of Toll-like receptors in human and African great ape populations

Abstract

The study of the genetic and selective landscape of immunity genes across primates can provide insight into the existing differences in susceptibility to infection observed between human and non-human primates. Here, we explored how selection has driven the evolution of a key family of innate immunity receptors, the Toll-like receptors (TLRs), in African great ape species. We sequenced the 10 TLRs in various populations of chimpanzees and gorillas, and analysed these data jointly with a human data set. We found that purifying selection has been more pervasive in great apes than in humans. Furthermore, in chimpanzees and gorillas, purifying selection has targeted TLRs irrespectively of whether they are endosomal or cell surface, in contrast to humans where strong selective constraints are restricted to endosomal TLRs. These observations suggest important differences in the relative importance of TLR-mediated pathogen sensing, such as that of recognition of flagellated bacteria by TLR5, between humans and great apes. Lastly, we used a population genetics-phylogenetics method that jointly analyses polymorphism and divergence data to detect fine-scale variation in selection pressures at specific codons within TLR genes. We identified different codons at different TLRs as being under positive selection in each species, highlighting that functional variation at these genes has conferred a selective advantage in immunity to infection to specific primate species. Overall, this study showed that the degree of selection driving the evolution of TLRs has largely differed between human and non-human primates, increasing our knowledge on their respective biological contribution to host defence in the natural setting.

INTRODUCTION

Important differences between human and non-human primates are observed in their susceptibility to disease, in particular in the incidence and severity of infectious diseases (1). Several medical conditions, such as HIV progression to AIDS, Plasmodium falciparum malaria, late complications in hepatitis B or C and influenza A symptomatology, affect humans more severely than other primates (2,3). These differences may be accounted for, at least partially, by interspecies genetic differences in immune responses, including those that have evolved adaptively. Indeed, different genome-wide scans for selection have shown that immunity-related proteins, compared with other protein classes, have been preferential targets of positive selection in humans, chimpanzees and, more generally, in the primate lineage (4–12). Furthermore, a study focusing on gene expression patterns in different primate species has revealed that immune responses are often lineage-specific, reflecting rapid host adaptation to varying microbial pressures (13). In this context, the study of the genetic and selective landscape of immunity genes across primate species may shed light on the existing differences in susceptibility to infection between human and non-human primates.

The innate immune system is a universal and evolutionarily ancient mechanism at the front line of host defence against invading pathogens (14–16). Among the different families of innate immunity receptors, the Toll-like receptors (TLRs) are by far the most comprehensively studied from immunological, clinical and epidemiological genetics standpoints (14,17,18). In primates, 10 distinct functional members of the TLR family exist and whose corresponding microbial ligands, with the exception of TLR10, have been identified (19–23). Depending on cellular sublocalization and ligand specificity, TLRs are subdivided into two groups. Cell-surfaced expressed TLRs, which are TLR1, TLR2, TLR4, TLR5, TLR6 and TLR10, generally detect products such as glycolipids, lipopeptides and flagellin, which are present in a wide variety of microorganisms. In turn, endosomal TLRs, including TLR3, TLR7, TLR8 and TLR9, are involved in the sensing of nucleic acids, principally from viruses (19,24). Upon ligand recognition, TLRs transduce the signalling responses that are required for the activation of innate immunity effector mechanisms and the subsequent development of adaptive immunity (17,19,24,25).

Given that TLRs are involved in the direct, early interaction between the host and the microbe, they provide an excellent model for the study of the selective pressures that are exerted by microorganisms on the host genome (4,18,26). In humans, a population genetics study indicates that TLRs can be divided into two distinct evolutionary groups: endosomal TLRs have been subject to strong purifying selection, whereas cell-surface TLRs experienced more relaxed constraints (27). With respect to non-human primates, in addition to a few inter-specific comparative studies where a single individual per species was analysed (28,29), only one study has analysed both inter-species divergence data and patterns of polymorphism within species, focusing on the chimpanzee subspecies Pan troglodytes verus (30). A signature of accelerated evolution was found across primate species for most TLRs, although, within species, the patterns of nucleotide variation were generally constrained. To date, however, no studies have investigated the intra-species evolution of TLRs in the various subspecies of chimpanzees, and no data are available on the diversity of the TLR gene family in gorillas.

In this study, we explored the evolutionary contribution of TLRs to host defence in great ape species. To do so, we generated sequence-based data from both multiple non-coding regions of the genome as well as from the 10 TLR family members in various population groups belonging to different subspecies of chimpanzees and gorillas. This approach, integrating non-coding and coding data, allowed us to first obtain a prediction for the expected diversity under neutrality, which serves as a basis for inferring how selection has targeted this family of microbial sensors in great apes. We analysed these newly generated data in conjunction with sequence data for the same non-coding regions and TLRs from a panel of human populations (27). This data set was used to estimate a number of population genetic parameters and summary statistics reflecting various aspects of the data, which provided information about evolutionary processes acting over different timescales (i.e. deep species-wide selective events versus more recent within-species, or subspecies, events). Furthermore, we used a recently developed population genetics-phylogenetic method that has increased power to detect fine-scale differences in the action of selection within genes (31), which we applied here for the first time to vertebrate data. We showed that the degree of natural selection driving the evolution of TLRs has differed between humans and African great apes, increasing our knowledge on the biological relevance of the TLR family in host defence in different primate species.

RESULTS

Sequence data and population structure

We generated sequence data from 52 African great apes, including 36 chimpanzees belonging to the three common Pan troglodytes (P.t.) subspecies—the central African P.t. troglodytes (12 individuals) and P.t. ellioti (8 individuals), and the western African P.t. verus (16 individuals)—and 16 gorillas (Gorilla gorilla). We resequenced 20 autosomal non-coding regions dispersed along the genome, to obtain a prediction for the expected levels of diversity at putatively neutrally evolving loci, together with the 10 TLR members. The mean size of each of the sequenced non-coding regions was ∼1.2 kb, for a total of 24.2 kb of diploid sequence information per individual. The 10 TLR genes accounted for 63.3 kb per individual, 42.2% of which corresponded to exonic regions. These data were then used to compute several summary statistics and neutrality tests (Table 1 and Supplementary Material, Tables S1 and S2).

Table 1.

Mean diversity indices and neutrality tests across TLR regions

	P.t. troglodytes (n = 20)a					P.t. ellioti (n = 14)a					P.t. verus (n = 26)a					G. gorilla (n = 24)a
	Sb	πc	TDd	FDe	Hf	Sb	πc	TDd	FDe	Hf	Sb	πc	TDd	FDe	Hf	Sb	πc	TDd	FDe	Hf
TLR1	49	22.5	−0.59	−0.53	0.18	24	12.0	−0.74	0.41	0.57	16	6.7	−0.58	0.84	−0.60	7	2.8	−0.93	−0.69	0.56
TLR2	17	8.1	−1.34	−2.13	0.57	16	8.1	−1.57*	−2.12*	−0.28	7	7.6	1.11	1.28	−0.47	18	14.8	0.61	0.01	0.66
TLR3	39	17.8	−0.60	−0.79	0.34	31	19.8	0.73	0.38	0.26	18	10.2	0.81	0.62	0.00	29	13.4	0.00	0.64	−1.64
TLR4	45	11.6	−1.20	−1.88	0.84	29	8.2	−1.41*	−1.83*	0.56	13	5.2	0.57	−0.61	−0.16	27	11.8	0.62	1.00	−1.46
TLR5	35	11.6	−0.99	−1.39	0.50	22	10.4	−0.18	0.09	0.56	16	8.5	0.65	1.20	−0.05	43	21.1	0.62	1.02	−0.31
TLR6	23	14.5	0.09	0.19	0.22	22	11.2	−1.07	−1.38	−0.45	10	5.4	−0.78	0.37	0.60	24	12.3	−0.46	1.40	−3.78*
TLR7	19	8.1	−0.60	−0.99	0.19	11	8.6	1.30	0.84	0.54	6	1.3	−1.74*	−2.26*	−0.91	13	3.4	−1.74*	−1.48	−0.51
TLR8	35	14.4	−0.65	−0.26	0.77	16	8.5	−0.50	−0.32	−0.65	11	3.7	−1.05	−0.76	0.77	6	1.4	−1.66*	−1.47	−0.82
TLR9	14	4.3	−1.25	−1.39	−0.81	7	3.6	0.18	0.21	0.61	16	7.6	0.41	1.20	0.00	12	3.7	−1.05	−0.28	−1.45
TLR10	24	9.5	−0.76	−0.65	0.47	9	3.6	−0.92	−1.35	0.79	12	5.2	−0.13	1.02	−1.93	10	4.1	−0.34	0.40	0.73

^aNumber of chromosomes analysed. Considering the X-linked TLR7 and TLR8, the numbers are 16, 9, 20 and 19 for P.t. troglodytes, P.t. ellioti, P.t. verus and G. Gorilla, respectively.

^bNumber of segregating sites.

^cNucleotide diversity π × 10⁻⁴.

^dTajima's D.

^eFu and Li's D*.

^fFay and Wu's H. Results that were significant without correction by demography and that remained significant after correction are in bold.

*0.01 < P-value < 0.05.

We first examined the presence of population structure and potential cryptic relatedness between individuals. We applied the Bayesian model-based clustering approach implemented in STRUCTURE (32) and used the sequence data obtained from the 20 non-coding regions and the 10 TLR genes. Our results revealed, as expected, genetic differentiation among the three chimpanzee subspecies at K = 3 (Supplementary Material, Fig. S1). To examine the degree of relatedness among individuals, we assessed the genetic relationships between individuals within each group of chimpanzees and gorillas, by running an Identity by State (IBS) method (33). A few pairs of individuals within each subspecies of chimpanzees and within gorillas displayed high levels of genetic identity, which are suggestive of cryptic relatedness (i.e. IBS distance > 0.85; see Section ‘Material and Methods’ for details). In light of this, we discarded one sample in each pair of individuals that presented such high IBS values (2 P.t. troglodytes, 1 P.t. ellioti, 3 P.t. verus and 4 Gorilla gorilla), and a final set of 42 samples was used in all subsequent analyses.

Sequence variation at non-coding regions and TLR genes

We investigated the genetic diversity of the 20 non-coding regions and the 10 TLR genes by estimating levels of nucleotide diversity (π) and other summary statistics (Table 1, Supplementary Material, Table S2). With respect to the 20 non-coding regions, the levels of nucleotide diversity in our great ape samples were slightly higher than previously reported, but the pairwise relationships between the different subspecies remained consistent with these studies based on different sets of autosomal intergenic regions (34–36). Among chimpanzees, P.t. verus presented the lowest level of nucleotide diversity, with a π value of 13.2 × 10⁻⁴, whereas the values of P.t. troglodytes and P.t. ellioti (π = 25.5 × 10⁻⁴ and 16.5 × 10⁻⁴, respectively) were found to be 1.9 and 1.3 times higher than that of P.t. verus. With respect to the group of gorillas, they displayed levels of nucleotide diversity in the same range of P.t. troglodytes (π = 24.5 × 10⁻⁴) (Fig. 1A).

Figure 1.

Global levels of nucleotide diversity (π) in TLRs among humans and African great apes. (A) Mean levels of nucleotide diversity for TLR regions (in red) with respect to those observed for 20 non-coding regions (‘neutral regions’, in blue). Student's test, *P = 1.9 × 10⁻² and **P = 1.0 × 10⁻⁴ with respect to the non-coding regions. (B) Nucleotide diversity levels for the individual TLR genes in great apes. The expected neutral diversity corresponds to the mean diversity levels observed for the 20 autosomal non-coding regions in each population.

The sequencing of the 10 TLR members allowed us to identify SNPs that are specific to the different chimpanzee subspecies: 211 SNPs were specific to P.t. troglodytes, 45 of which were non-synonymous; 102 SNPs were specific to P.t. ellioti, 24 of which were non-synonymous and 94 SNPs were observed only in P.t. verus, 16 of which were non-synonymous. The mean values of TLR nucleotide diversity were two times lower than those observed at the 20 non-coding regions, with P.t. troglodytes displaying the highest levels of TLR diversity and P.t. verus displaying the lowest (Fig. 1A). When considering the different TLR members individually, we observed large fluctuations in the levels of nucleotide diversity (Table 1, Fig. 1B). TLR9 in P.t. troglodytes was by far the least diverse gene, with a 6-fold decrease of diversity compared with the non-coding regions, and a similar pattern was observed for TLR7 in P.t. verus, presenting a 10-fold decrease of nucleotide diversity. For both genes, the drop of diversity was mainly accounted for by a decrease of nucleotide diversity in the exonic region (Supplementary Material, Table S3). In P.t. ellioti, TLR3 is the only gene presenting a higher nucleotide diversity level by comparison with neutral regions.

With respect to gorillas, we found a total of 200 SNPs, 34 of which were non-synonymous SNPs. The global nucleotide diversity levels of TLRs in gorillas were found to drop dramatically, representing a significant 2.8-fold decrease with respect to the neutral regions (Fig. 1A). This pattern is essentially accounted for by the virtual lack of diversity at TLR8, which displayed the lowest level of nucleotide diversity among all TLR genes (Table 1, Fig. 1B and Supplementary Material, Table S3). Altogether, these empirical observations suggest that although most TLR genes in African great apes have lower diversity than neutral regions, consistent with negative selection, they have undergone different evolutionary trajectories, an observation that was subsequently tested to assess the extent to which selection has acted upon this gene family.

Assessment of the strength of selection in TLRs from African great apes

To investigate whether and how natural selection has driven the observed heterogeneous patterns of TLR diversity in African great apes since the divergence from their common ancestor, we used tests that distinguish the putative functional impact of different types of mutations at the species-wide level, i.e. chimpanzees (the three subspecies merged) and gorillas. To do this, we first applied a statistical approach—the McDonald–Kreitman Poisson Random Field (MKPRF) test—that makes use of MK contingency tables (5,37,38). These tables summarize the number of non-synonymous and silent fixed differences between species and the common ancestor and the number of non-synonymous and silent polymorphisms within species. One of the parameters estimated by this method is ω, the ratio of the locus-scaled mutation rate at non-synonymous sites compared with that of silent sites. Under neutrality, ω is not significantly different from 1. Lower values are consistent with selection against non-synonymous variants (purifying selection), whereas higher values reflect selection favouring amino-acid substitutions (positive selection) (Material and Methods). In both chimpanzees and gorillas, most TLRs presented ω values <1, suggesting the action of purifying selection to different extents, with varying levels of statistical significance (Fig. 2). Specifically, in chimpanzees, three of the endosomal TLRs (TLR3, TLR8 and TLR9) and the cell-surface TLR5 presented ω values significantly <1 (0.27, 0.06, 0.32 and 0.29, respectively). Likewise, in gorillas, ω estimates were significantly <1 for two endosomal TLRs (TLR7 and TLR9) and three cell-surface TLRs (TLR1, TLR4 and TLR10), ranging from 0.05 to 0.1. Overall, our results indicated that strong purifying selection has been the predominant force shaping the TLR family members in chimpanzees and gorillas, irrespectively of whether they are endosomal or cell-surface located.

Figure 2.

Strength of purifying selection acting on individual TLR genes. The strength of purifying selection is measured in chimpanzees, gorillas and humans by estimating ω values. Bars indicate 95% confidence intervals and red dots indicate genes with ω estimates significantly <1. Only the names of significant genes are shown.

Purifying selection is stronger in African great apes than in humans

To compare the TLR patterns observed in chimpanzees and gorillas with those previously detected in humans (27), we reanalysed the human data set jointly with that from African great apes, and estimated ω using the divergence of the three species with respect to their last common ancestor. As previously observed, endosomal TLRs—TLR3, TLR7, TLR8 and TLR9—displayed the strongest signatures of purifying selection in humans, as revealed by their ω values, which were significantly <1 in all cases except for TLR3 (Fig. 2). These results confirmed that endosomal TLRs in humans have been overall subject to stronger purifying selection than the group of cell-surface TLRs. However, the sample size used in humans (n = 158) is much larger than those from chimpanzees (n = 30) and gorillas (n = 12), confounding inter-species comparisons. To circumvent this limitation, we generated random data sets of 30 and 12 individuals from the human data set, and re-estimated the ω parameter (Material and Methods). Our results showed that when considering 30 individuals, the ω values of TLR7, TLR8 and TLR9 remained significantly <1 in 92–100% of the resampled data sets. When resampling 12 human individuals, ω values of TLR7, TLR8 and TLR9 remained significantly <1 in 77, 28 and 100% of the resampled data sets, respectively, highlighting the loss of power owing to small sample sizes (Supplementary Material, Table S4). Despite the smaller sample sizes of chimpanzees and gorillas, we observed four and five TLRs displaying ω values significantly <1, compared with three in the much larger human data set, indicating that the extent of purifying selection has been overall stronger in African great apes in comparison with humans.

Episodes of adaptive evolution at the species-wide level

We subsequently explored the degree of intragenic variation of selective pressures in primate TLRs by using a recently developed population genetics-phylogenetics approach (31). This method allows the analysis of variation in selection pressures within and between species jointly and detects fine-scale differences in selective pressures within genes at specific codons. For comparison, the data from chimpanzees and gorillas were again analysed jointly with the human data set. We first estimated the variation in the selection coefficient (γ) along the 10 TLR genes in each of the three lineages. Although there was a large degree of uncertainty in the distribution of selection coefficients across the TLRs, reflecting the difficulty in teasing apart the relative frequency of similar selection coefficients, there was a clear preponderance of codons evolving under negative selection in all three lineages (−500 ≤ γ ≤ −1) (Fig. 3). In chimpanzees, as well as humans, most of the signal of negative selection was mainly accounted for by codons being under moderate negative selection (−50 ≤ γ ≤ −10), with a cumulative probability of 0.46 and 0.45, respectively. In gorillas, in agreement with the gene-level MKPRF results, we observed a clear shift towards codons targeted by strong negative selection (−500 ≤ γ ≤ −100), with a cumulative probability of 0.47 (Fig. 3).

Figure 3.

Distribution of selection coefficients across the ten TLR genes. Probability of codons for being under a selection coefficient (γ) ranging from −500 to 100 along the 10 TLRs in each of the three species.

Despite the general constraint observed in the three primate lineages, we also found signals of positively selected codons displaying intermediate selection coefficients (1 ≤ γ ≤ 10), with a cumulative probability for codons being weakly to moderately positively selected of 0.04 in humans, and 0.07 in chimpanzees and gorillas (Fig. 3). To further localize the signals of positive selection, we estimated the codon-specific posterior probabilities for each selection coefficient. To be conservative, we declared a codon site as being targeted by positive selection when the cumulative posterior probability of γ ≥ 1 was > 0.75. This threshold corresponds to 0.5, 0.1 and 0.2% of all codons across all TLRs for chimpanzees, gorillas and humans, respectively. Using these criteria, we identified 44, 9 and 19 positively selected codons in chimpanzees, gorillas and humans, respectively (Table 2). Some of these putatively selected codons have been previously detected as being targeted by positive selection using other codon-based analyses of positive selection (30). These results clearly showed that chimpanzees not only displayed a larger amount of positively selected codon sites with respect to gorillas and humans, but also the mean posterior probability for these positively selected codons (i.e. Pr > 0.75) was significantly higher (Student's t-test P = 2.6 × 10⁻⁶) in chimpanzees (mean Pr = 0.84) than in gorillas and humans (mean Pr = 0.79).

Table 2.

Top list of positively selected codons presenting the highest posterior probabilities

Chimpanzees					Gorillas					Humans
Gene	Codon	AA	Pra	Domainb	Gene	Codon	AA	Pra	Domainb	Gene	Codon	AA	Pra	Domainb
TLR1	2	Thr	0.82	SP	TLR1	542	Val	0.78	EC	TLR1	313*	Gly	0.81	EC
TLR1	93	Gln	0.81	EC	TLR1	547	Leu	0.78	EC	TLR1	318	Tyr	0.82	EC
TLR1	565	Thr	0.95	EC	TLR1	584*	Val	0.81	TM	TLR1	460	Ile	0.78	EC
TLR1	566	Leu	0.95	EC	TLR1	587	Val	0.81	TM	TLR1	533	Gly	0.78	EC
TLR2	238	His	0.83	EC	TLR6	552	Leu	0.80	EC	TLR1	725	Ser	0.78	CP
TLR2	372	Met	0.82	EC	TLR6	556	Pro	0.79	EC	TLR2	418	Ile	0.80	EC
TLR3	31	Ser	0.83	EC	TLR6	571	Pro	0.77	EC	TLR3	299	Gln	0.80	EC
TLR3	69	Ala	0.92	EC	TLR9	443*	Gly	0.78	EC	TLR3	555	Ile	0.79	EC
TLR3	306	Glu	0.82	EC	TLR9	446	Val	0.78	EC	TLR5	373	Ala	0.80	EC
TLR3	391	Thr	0.83	EC						TLR5	460	Phe	0.79	EC
TLR3	625	Val	0.82	EC						TLR5	564	Asn	0.78	EC
TLR3	726	Glu	0.83	CP						TLR6	79	Thr	0.79	EC
TLR3	826	Leu	0.83	CP						TLR8	452	Arg	0.75	EC
TLR4	321*	Glu	0.97	EC						TLR9	19	Met	0.77	SP
TLR4	322*	Arg	0.97	EC						TLR9	864*	Asp	0.80	CP
TLR4	781	Gln	0.84	CP						TLR10	17	Ala	0.78	SP
TLR5	37	Asn	0.82	EC						TLR10	178	Pro	0.78	EC
TLR5	108	Tyr	0.81	EC
TLR5	429	His	0.82	EC
TLR5	451	Gln	0.77	EC
TLR5	530	Gly	0.94	EC
TLR5	531	Leu	0.77	EC
TLR5	532	Ser	0.94	EC
TLR5	592	Asn	0.82	EC
TLR5	722	Gln	0.82	CP
TLR6	27	Gln	0.80	SP
TLR6	293*	Arg	0.88	EC
TLR6	296	Asp	0.88	EC
TLR6	308*	Thr	0.84	EC
TLR6	315*	Gln	0.83	EC
TLR6	470*	Val	0.80	EC
TLR6	539	Glu	0.76	EC
TLR6	543	Asn	0.76	EC
TLR6	589*	Val	0.78	TM
TLR6	658	Ser	0.81	CP
TLR7	693*	Lys	0.79	EC
TLR7	1032	Ala	0.78	CP
TLR8	55	Gln	0.79	EC
TLR9	278*	Asp	0.83	EC
TLR9	467*	Arg	0.83	EC
TLR9	667	Phe	0.78	EC
TLR9	734	Ala	0.83	EC
TLR10	246	Val	0.84	EC
TLR10	454	Gln	0.84	EC
TLR10	780	Leu	0.78	CP

^aPosterior probability. Only codons that presented a posterior probability >0.75 of being positively selected are shown.

^bProtein domains were defined on the basis of Swiss-Prot sequence annotation. SP, signal peptide; EC, extracellular; CP, cytoplasmic; TM, transmembrane.

*Codons found to be targeted by positive selection by other studies. Codons located in regions that have been associated with pathogen interactions are in bold.

Testing for more recent selection among African great ape subspecies

To investigate whether more recent events of selection had affected individual TLR genes within each of the three subspecies of chimpanzees separately and within gorillas, we next used several allele frequency spectrum-based neutrality tests (Tajima's D, Fu and Li's D* and Fay and Wu's H) [see (39–41) for extensive reviews on these tests]. Because these tests are known to be highly sensitive to the effects of demography, we accounted for this by using the patterns of diversity observed at the 20 non-coding regions sequenced in the same panel of individuals (Supplementary Material, Table S2). To approximate the demographic models of the studied African great apes, we adopted a simulation-based approach to fit simulated data sets to the empirical data sets based on the non-coding regions (Material and Methods). The parameters used to simulate the various demographic histories were randomly drawn from distributions based on previously inferred historical models (42,43) (Supplementary Material, Table S5 and Supplementary Material, Figs S2 and S3). Once the combinations of demographic parameters best fitting our empirical data sets were identified, we estimated the significance of neutrality tests for TLRs by generating neutral expectations under our estimated demographic model.

For three TLRs, we significantly rejected the null hypothesis of neutrality in specific chimpanzee subspecies (Table 1). TLR2 and TLR4 presented an excess of rare alleles in P.t. ellioti, as did TLR7 in P.t. verus, as attested to by the significantly negative values obtained for Tajima's D and Fu and Li's D*, a pattern indicative of weak negative selection or of a recent selective sweep. For TLR7, further support for the action of positive selection came from the analyses of the levels of population differentiation (F_ST) for each SNP. Indeed, the F_ST analyses identified the non-synonymous A542G mutation, which is located in the extracellular domain of TLR7, as being extremely differentiated in P.t. verus: the derived 542G allele reaches almost fixation in P.t. verus while it is totally absent among the other chimpanzee subspecies (F_ST = 0.96, Supplementary Material, Fig. S4). Furthermore, the network of TLR7 showed that the haplotype cluster defining the highly differentiated non-synonymous A542G mutation in P.t. verus is mainly accounted for by a single high-frequency haplotype (70%) (Supplementary Material, Fig. S5). It must be noted, however, that these results for TLR7 have to be interpreted with caution, as this gene is located on the X chromosome, which is expected to experience greater genetic drift because of a reduced effective population size (i.e. higher F_ST values between populations on the X chromosome than on autosomes).

In gorillas, we rejected neutrality for at least one statistical test for three TLRs. TLR6 showed an excess of high-frequency derived variants, as revealed by the significant negative value of Fay and Wu's H, and TLR7 and TLR8 presented an excess of rare alleles, as attested to by the significant negative values of Tajima's D and/or Fu and Li's D* (Table 1). Generally, these intra-species analyses suggest that recent positive selection has targeted a limited number of genes within specific subspecies of primates.

DISCUSSION

The identification of immunity genes evolving differently across primate species might help to understand the genetic basis underlying the observed differences in susceptibility to infectious diseases between humans and other primates. In this context, the evolutionary genetic dissection of TLRs represents an excellent model to investigate the way in which pathogens have exerted pressure on host genes and how the immune system of phylogenetically close species have adapted to their respective pathogen sets. Here, we have analysed the patterns of sequence variation of the 10 TLRs in four different population samples of primates, corresponding to three chimpanzee subspecies (P.t. troglodytes, P.t. ellioti and P.t. verus) and one subspecies of gorillas, and analysed these newly generated data in conjunction with a human data set.

Considering the TLR family as a whole, our analyses of divergence and polymorphism showed that purifying selection has been pervasive among primate TLRs, particularly in gorillas (Fig. 2). Indeed, TLRs appear to be under stronger constraints in gorillas than in humans, as revealed by the number of genes displaying significant signatures of purifying selection. Moreover, the proportion of constrained TLR genes in gorillas may be underestimated, owing to their smaller sample sizes with respect to humans, as suggested by our resampling analyses of humans where the power to detect significant signatures of selection was reduced. Our analyses also indicate that, although the number of TLRs evolving under purifying selection in chimpanzees is higher than in humans, the difference in much less pronounced. The stronger evolutionary constraints on non-human primate TLRs is further supported by the observation that no stop mutation has been detected in chimpanzees and gorillas, whereas the proportion of individuals carrying a stop mutation in at least one TLR (i.e. TLR2, TLR4, TLR5 or TLR10) in humans is collectively 16% (27,44). After correction by the sample sizes of African great apes, this difference remained significant for chimpanzees (P = 6.1 × 10⁻³) (Supplementary Material, Fig. S6). These data overall suggest that excessive amino acid-altering variation is not allowed among TLRs in great apes, attesting to their potentially deleterious nature in host survival. As a whole, the TLR system appears to have fulfilled a more essential, and less redundant, role in African great apes than in humans.

The relaxed constraints observed in humans, in terms of tests of selection and occurrence of stop mutations, are restricted to cell-surface TLRs (27,44), with endosomal TLRs presenting little amino acid-altering variation, no stop mutations and strong signatures of purifying selection. Although some overlap exist in the classes of ligands that they sense, cell-surface TLRs predominantly recognize products from bacteria, fungi and parasites, whereas endosomal TLRs sense nucleic acids, principally from viruses (19,24,25). The evolutionary dichotomy observed between the two classes of TLRs in humans, attesting to their different contributions to host defence, appears to be specific to our species, as purifying selection in African great apes targeted the two groups of TLRs indistinctively (Fig. 2). The case of the cell-surface TLR5—the receptor of bacterial flagellin—is particularly extreme. In chimpanzees, TLR5 evolves under strong purifying selection, highlighting its essential role in host survival. In humans, however, a stop mutation with a dominant-negative effect is present at high population frequencies (up to 23%) in Eurasians (27,44), and another mutation leading to reduced TLR5 signalling has recently been documented to be targeted by positive selection in Africans (45). Likewise, TLR10 is subject to strong purifying selection in gorillas, whereas two stop mutations reach a global frequency of 7% in humans of African descent (27). These examples clearly illustrate important differences in the relative importance of TLR-mediated pathogen recognition, such as that of flagellated bacteria for TLR5, between humans and African great apes.

The broad constraints described above can swamp signals of adaptive changes occurring at a limited number of sites. The advantage of the population genetics-phylogenetics method used here is that it models intragenic variation in selection coefficients and can thus detect signals of site-specific positive selection in otherwise constrained genes (31). We identified a number of codons presenting high posterior probabilities of being positively selected in specific primate species (Table 2). The signals of positive selection were virtually restricted to the extracellular domain of the TLRs, suggesting that the domain involved in pathogen detection has been evolving the most adaptively, as predicted by the arms race model of host–pathogen co-evolution. In humans, the wealth of functional data available lends further support to the action of positive selection, as some codons here detected as being positively selected lie in regions known to be involved in specific host–pathogen interactions or infectious disease susceptibility (Table 2). For example, the TLR1 313G is located two amino acid positions away, and within the same domain of bacterial interaction, than the P315L polymorphism, which impairs the sensing of microbial cell wall components (46). Likewise, the TLR3 299Q and 555I lie in the same extracellular leucine-rich repeat (LRR) domain of the F303S and P554S variants, which are involved in influenza-associated encephalopathy and herpes simplex encephalitis, respectively (47–49).

The event of positive selection detected at the 555I codon of TLR3, which mediates the recognition of the herpes simplex virus type-1 (HSV-1) (48), is particularly informative in this respect. Humans are naturally resistant to HSV-1 infection and the clinical course of this infection is usually benign. This contrasts with the high susceptibility to HSV-1 documented for non-human primates, such as gorillas, gibbons or marmosets, which develop systemic infections with a fatal outcome (50). In humans, however, clinical genetic studies have revealed that rare mutations in TLR3 underlie HSV-1 encephalitis in childhood (48,49,51). For example, the P554S amino acid substitution, which is located in the LRR20, confers autosomal dominant predisposition to HSV-1 encephalitis, suggesting the functional importance of this region in protective immunity (48,49). Interestingly, the codon V555I, which presents one of the highest probabilities of being under positive selection in humans (Table 2), is located one position away from the deleterious P554S. The isoleucine at codon 555 is specifically fixed in the human lineage, whereas the valine is conserved across 122 vertebrate species. In light of this, the human V555I substitution may have conferred the natural protection against HSV-1 observed in humans, highlighting the need for functional studies to substantiate this hypothesis.

Another interesting finding of this study is the large proportion of codons (>75%) located in the TLR1-TLR6-TLR10 cluster among those presenting the highest posterior probabilities of being positively selected (Table 2). This observation, which accounts for 36% in chimpanzees and for an overwhelming 78 and 53% in gorillas and humans, respectively, suggests that this gene cluster has been the main substrate of positive selection among all TLRs. Because a signature of positive selection targeting this region has been documented in humans (27,52), as well as in the orangutan and chimpanzee genomes, the TLR1-TLR6-TLR10 cluster has been proposed to be a hotspot of positive selection (53). In humans, functional analyses suggest that the underlying molecular phenotype is the avoidance of an excessive TLR-mediated inflammatory response (27), and genetic variation at TLR1 and TLR6 has been associated with susceptibility to leprosy and tuberculosis, respectively (54,55). Functional analyses of this genomic region in chimpanzees and gorillas are now needed to elucidate if such a shared hotspot of positive selection reflects a case of parallel/convergent evolution, or adaptations in different phenotypic directions involving the same locus.

In summary, despite the shared signature of positive selection at TLR1-TLR6-TLR10, our findings indicate that the selective landscapes characterizing human and great ape TLRs largely differ. Not only are great ape TLRs under stronger selective constraints than their human paralogues, but also such constraints are not restricted to endosomal TLRs, as observed in humans. This attests to the different evolutionary relevance in host defence of TLRs in humans and great apes and therefore to their variable roles in immunity to infection. These differences could reflect, at least partially, the pathogen burdens to which humans and African great apes may have been historically exposed, such as that imposed by the African forest, where chimpanzees and gorillas mostly live. Future studies exploring the extent and type of selection driving the evolution of TLRs in human populations living in similar habitats, such as forest-dwelling populations, should increase our understanding of how the environment to which different species have been exposed has influenced their genetic adaptation to pathogen pressures and, therefore, their present susceptibility or resistance to infection.

MATERIALS AND METHODS

DNA samples

DNA samples were obtained from a previously established DNA collection of 52 African great apes (56–58). This comprises 36 chimpanzees, including two groups of central African chimpanzees (Pan troglodytes troglodytes n = 12, and Pan troglodytes ellioti n = 8) and a group of western African chimpanzees (Pan troglodytes verus n = 16), and 16 gorillas (Gorilla gorilla). The human sequence data set, which corresponds to the same non-coding regions and 10 TLRs sequenced in great apes, has been described elsewhere and includes 63 sub-Saharan Africans, 47 Europeans and 48 East-Asians (27).

DNA resequencing

To obtain an empirical framework of the expected levels of diversity at putatively neutrally evolving loci, we resequenced in the entire primate collection 20 autosomal non-coding regions, of ∼1200 bp each, dispersed throughout the genome. These regions were chosen to be independent from each other and to be at least 200 kb away from any known gene, predicted gene or spliced expressed sequenced tag (Supplementary Material, Table S2) (59,60). With respect to the 10 TLR family members, the totality of the exonic region and at least an equivalent amount of non-exonic regions, including ∼1000 bp of their promoter regions (i.e. upstream of the first transcribed exon), were sequenced for each TLR (Supplementary Material, Table S1). Intronless genes, such as TLR6 and TLR9, were sequenced in their totality including ∼1000 bp of their promoter regions. Sequence files and chromatograms were inspected using the GENALYS software (61). Sequences have been deposited in GenBank under the following accession numbers: TLR1 (KF319302-KF319403, KF319404-KF319505, KF319506-KF319607), TLR2 (KF319608-KF319711, KF319712-KF319815), TLR3 (KF319816-KF319919, KF319920-KF320023, KF320024-KF320127, KF320128-KF320231, KF320232-KF320335), TLR4 (KF320336-KF320439, KF320440-KF320543, KF320544-KF320647), TLR5 (KF320648-KF320751, KF320752-KF320855), TLR6 (KF320856 - KF320959), TLR7 (KF320960-KF321033, KF321034-KF321107), TLR8 (KF321108-KF321181, KF321182-KF321255, KF321256-KF321329), TLR9 (KF321330 - KF321433) and TLR10 (KF321434-KF321537, KF321538-KF321641).

Population structure and individual relatedness

To investigate the presence of genetic structure in our different primate populations, we used the STRUCTURE v.2.1 software (32). We used the ‘correlations' and ‘admixture' models, without prior information about populations, 1 000 000 burn-in steps and 1 000 000 Monte Carlo Markov chain replications. Haplotype reconstruction was performed by means of the Bayesian statistical method implemented in Phase (v.2.1.1) (62). We applied the algorithm five times, using different, randomly generated seeds, and consistent results were obtained across runs. We recoded the 20 non-coding and the 10 TLR regions as microsatellites, considering each haplotype as an allele of a single multi-allelic locus. For each prior K value (2, 3 and 4), we ran 10 independent runs with different seeds and found likelihoods to be stable across runs. To check for potential cryptic relatedness between individuals, we examined the genetic relationship between individuals within each group by running an IBS method (33), which used the 875 and 422 SNPs identified in chimpanzees and gorillas, respectively, in the 20 non-coding and TLR regions. To control for cryptic relatedness among our samples, we removed in each species and subspecies one sample in each pair of individuals that presented an IBS distance >0.85 (i.e. 2 P.t. troglodytes, 1 P.t. ellioti, 3 P.t. verus and 4 Gorilla gorilla). The IBS distance corresponds to the mean IBS score across all SNPs divided by 2, as implemented in PLINK (33). The threshold value considered here has been classically used in humans as indicative of relatedness [see, for example, the Wellcome Trust Case Control Consortium (63)].

Estimation of the degree and intensity of selection

To model purifying selection operating on the different TLRs since the divergence of African great apes and their common ancestor, we first inferred the ancestral sequence for the 10 TLR genes by aligning the human, chimpanzee and gorilla sequences with genomes of other primates (Pongo pygmaeus and Nomascus leucogenys; UCSC database) and deduced by parsimony the ancestral state of each variant. We next investigated the effects of natural selection considering both inter-species divergence and within-species polymorphism by means of the MKPRF test (5,37,38). This extended version of the MK test (64) uses a Markov Chain Monte Carlo (MCMC) algorithm for the Bayesian analysis of polymorphism and divergence data under a Poisson random field setting (5,37). Different parameters are estimated by means of MK contingency tables comparing the levels of silent and non-synonymous polymorphisms within species and the number of fixed silent and non-synonymous differences between species and the ancestral sequence (i.e. common ancestor) (64). The ω parameter is calculated as follows: ω = θ_N/θ_S α ((d_N + p_N)/(d_S + p_S)), where θ_N and θ_S are estimates of the rate of non-synonymous and silent mutations. Under neutrality, ω is not significantly different from 1. Values <1 are consistent with purifying selection, which eliminates almost all new non-synonymous mutations from the population (θr << θs) because their occurrence is not tolerated (e.g. lethal or strongly deleterious mutations). Values >1 reflect an excess of amino acid changes and are consistent with the effects of positive selection.

Estimation of the intensity of positive selection using a codon-based approach

To evaluate the action of positive selection among the 10 TLR genes, we used gammaMap, a tool that jointly analyses polymorphism and divergence data to detect fine-scale variation in selection pressures within genes (31). This method, which models long-term phylogenetic patterns of allelic substitution between species as well as transient distributions of allele frequencies within species, allows for sophisticated, multi-allelic, codon-based mutation models with variable selection pressure.

GammaMap is a Bayesian method that requires the use of prior distributions on the values of the evolutionary parameters. We chose prior distributions intended to be weakly informative, in order to avoid imposing strong prior beliefs on the posterior parameter estimates. We assumed that the population-scaled mutation rate θ varied between genes and lineages according to a log-normal distribution with mean μ and standard deviation σ on a log scale. We assumed an improper uniform prior distribution on μ and a standard log-normal prior distribution on σ. For each lineage (human, chimpanzee and gorilla), we assumed independent values for the branch length T, the ratio of transitions to transversions κ and the probability that adjacent codons share the same selection coefficient p. We utilized improper log-uniform distributions on T and κ. For p we assumed a uniform distribution a priori, which is equivalent to assuming a uniform number of contiguous windows of codons with the same selection coefficient within a gene. We estimated the distribution of selection coefficients separately for each lineage, assuming a uniform Dirichlet distribution, which gives equal prior weight to each selection class (γ = −100, −50, −10, −5, −1, 0, 1, 10, 50 or 100). For TLR7 and TLR8, which are X-linked, we adjusted values of θ and γ to three-fourths of that in the other loci, i.e. we assumed no sexual selection. For each gene, we ran two Markov Chain Monte Carlo runs of 2 million iterations each, with a burn-in of 20 000 iterations and a thinning interval of 40 iterations. Runs were compared for convergence before merging them to obtain final results.

Intra-species sequence-based neutrality tests and population differentiation

To analyse different aspects of our data within populations, we calculated summary statistics and various sequence-based neutrality tests used to detect deviations of the allele frequency spectrum from neutral expectations, such as Tajima's D, Fu and Li's D* and Fay and Wu's H tests (reviewed in 39–41), with the DnaSP package v. 5.0 (65). To assess the levels of population differentiation between the different pairs of chimpanzee subspecies, we used the F_ST statistics (66), which is derived from the analysis of variance (67). We compared the observed F_ST values at the level of individual SNPs at TLRs (i.e. 523 SNPs) against the empirical F_ST distribution obtained from the 20 neutral regions (i.e. 352 SNPs) sequenced on the same panel of individuals. Evolutionary relationships between haplotypes were inferred from a median-joining network constructed with Network v.4.5 (68).

Best-fit models and correction for the mimicking effects of demography

To correct the intra-species neutrality tests for the confounding effects of demography, we incorporated into our neutral expectations the best-fit demographic models obtained separately for each of the three chimpanzee subspecies and the gorilla population studied here. To this end, we adopted a simulation-based approach to fit simulated genetic data sets to the empirical data set of 20 non-coding regions resequenced in the panel of 42 great apes, as previously described (69–71). The demographic parameters best fitting our empirical data can be then used to consider the effects of demography on the genetic diversity of each TLR region. Coalescent simulations were performed with SIMCOAL 2.0 (69). For each simulated data set, we simulated 20 independent regions (1200 bp each) under a finite-site mutation model with per generation per site mutation rate, gamma distributed with a mean of ∼2.5 × 10⁻⁸ (95% CI: 1.5 × 10⁻⁸–4 × 10⁻⁸). We considered a recombination rate per generation between two adjacent base pairs, which was gamma distributed with a mean ∼10⁻⁸ (95% CI: 0.5 × 10⁻⁸–1.4 × 10⁻⁸).

For chimpanzees, we compared the simulations to the empirical data set for each subspecies independently. We performed a total of 10⁶ simulations, according to the historical models specified in Hey (43) and Becquet et al. (42). The different demographic parameters that have been used for the simulations are summarized in Supplementary Material, Table S5. We set the divergence times between subspecies by randomly drawing values from uniform distributions [see confidence intervals reported in Table 3 of Hey et al. (43)]. The effective sizes of ancestral and current populations were drawn from uniform random distributions (varying between 1000 to 100 000 individuals). Successive variations of effective sizes, simulated at each divergence time, represented the successive waves of rapid growth and/or bottlenecks that occurred during the last 2MY of chimpanzee evolution (42,43). For each simulation, we computed a distance between the simulated summary statistics and the empirical summary statistics (averaged over the 20 non-coding regions) as previously described (59). The summary statistics used were the numbers of haplotypes, the numbers of segregating sites and the Tajima's D statistics, as well as the global and pairwise F_ST. We retained the combinations of demographic parameters that generated the simulated data sets exhibiting the 0.1% smallest distances, i.e. the demographic parameters that best fit our genetic data set of non-coding regions. We next used these demographic parameters to subsequently simulate, for each of the three chimpanzee subspecies, the genetic diversity expected in each TLR region (10⁴ simulations each).

With respect to gorillas, we performed a total of 10⁶ simulations. To investigate the evolutionary and demographic history of natural populations of gorillas based on our resequencing data, we simulated various demographic scenarios in accordance with previously inferred historical models (72,73). The effective size of the ancestral population N₀ was drawn from a uniform random distribution, similar to the prior distribution used for chimpanzees (N₀ can vary between 1000 and 100 000 individuals). We simulated a single demographic event occurring during the last 2 MY (80 000 generations in the past). Specifically, we performed 10⁶ simulations involving a single instantaneous variation of N₀, by drawing λ = N₀/N₁ (with N₀ and N₁ being the effective sizes before and after the demographic event, respectively) from a uniform random distribution; λ can vary between 0.2, corresponding to a 40-fold increase of N₀ (expansion) and 40, corresponding to a 40-fold decrease of N₀ (bottleneck). Note that we performed the same number of simulations for both the expansion and bottleneck models in order to give the same weight to each scenario. Consequently, even if the distribution of current effective size of gorilla (N₁) is uniformly distributed in each scenario separately, the distribution of current effective size of gorilla (N₁) remains not uniformly distributed when merging the two scenarios. We used the same approach, as described for chimpanzees, to retain the combination of parameters that best fit our empirical non-coding data set (i.e. 0.1% smallest distances). The summary statistics used were the numbers of haplotypes, the numbers of segregating sites and the Tajima's D statistics. We next used this combination of parameters to subsequently simulate, for each TLR region (10⁴ simulations each), the genetic diversity expected in the gorilla population.

Resampling procedure

To assess whether the sample size may affect the detection of purifying selection using the MKPRF test, we randomly resampled new data sets with a sample size similar to the ones used in great ape populations (i.e. 30 individuals for chimpanzees and 12 individuals for gorillas) from the panel of 158 humans previously sequenced for the 10 TLRs (27), and we estimated ω values for the 10 TLRs individually. We resampled random sets 10³ times, and so we obtained a distribution of 10³ ω values with their significance for each gene. To assess the significance of finding no stop mutation in chimpanzees and gorillas, we adopted the same resampling procedure as described above, and used 10⁴ random data sets. The P-value corresponds to the number of times that no stop mutation was found in the resampled data sets.

SUPPLEMENTARY MATERIAL

Supplementary Material is available at HMG online.

Conflict of Interest statement. None declared.

FUNDING

This work was supported by the Institut Pasteur, the ANR (ANR-08-MIEN-009-01), the Fondation pour la Recherche Médicale, the CNRS and a EPFL-Debiopharm Life Sciences Award to L.Q.-M., as well as by a France-Chicago Center grant from the University of Chicago and R01 GM72861 to M.P. The laboratory of L.Q.-M. has received funding from the French Government's Investissement d'Avenir program, Laboratoire d'Excellence ‘Integrative Biology of Emerging Infectious Diseases’ (grant no. ANR-10-LABX-62-IBEID), and from the European Research Council under the European Union's Seventh Framework Programme (FP/2007–2013)/ERC Grant Agreement No. 281297. L.B.B. is a scholar of the Fonds de la Recherche en Santé du Québec, and M.P. is a Howard Hughes Medical Institute Early Career Scientist.