The evolutionary landscape of cytosolic microbial sensors in humans

Abstract

Host-pathogen interactions are generally initiated by host recognition of microbial components or danger signals triggered by microbial invasion. This recognition involves germline-encoded microbial sensors or pattern-recognition receptors (PRRs). By studying the way in which natural selection has driven the evolution of these microbial sensors in humans, we can identify genes playing an essential role and distinguish them from other, more redundant genes. We characterized the sequence diversity of the NOD-like receptor family, including the NALP and NOD/IPAF subfamilies, in various populations worldwide and compared this diversity with that of other PRR families, such as Toll-like receptors (TLRs) and RIG-I-like receptors (RLRs). We found that most NALPs had evolved under strong selective constraints, suggesting that their functions are essential and possibly much broader than previously thought. Conversely, most NOD/IPAF subfamily members were subject to more relaxed selective constraints, suggesting greater redundancy. Furthermore, some NALP genes, including NLRP1, NLRP14, and CIITA, were found to have evolved adaptively. We identified those variants conferring a selective advantage on some human populations as the most likely targets of positive selection. More generally, the strength of selection differed considerably between the major families of microbial sensors. Endosomal TLRs and most NALPs were found to evolve under stronger purifying selection than most NOD/IPAF subfamily members and cell-surface TLRs and RLRs, suggesting some degree of redundancy in the signaling pathways triggered by these molecules. This study provides novel perspectives and experimentally testable hypotheses concerning the relative biological relevance of the various families of microbial sensors in humans.

Figure 1

Distribution of the Nonsynonymous and Nonsense Variants Identified across the NLRs in This Study The location of each nonsynonymous and nonsense variant within the different protein domains is shown. AD stands for activation domain. Concerning NLRX1, the N-terminal domain is neither a PYRIN nor a CARD.

Figure 2

Estimation of Purifying Selection Acting on Individual NLR Genes and Genes from Other Major Families of Microbial Sensors We assessed the strength of purifying selection by calculating ω via the MKPRF test. Scale bars indicate 95% confidence intervals, and red diamonds indicate genes with ω estimates significantly lower than 1. The results for the population selection parameter γ are presented in Figure S3. Note that nonsense mutations and coding indels were either absent or present at very low frequency (<1%) in NLRs (Table S2) and, more generally, in all considered families of PRRs other than the cell-surface molecules TLR10 and TLR5, for which 5% and up to 23%, respectively, of individuals from the general population carried a nonsense mutation.

Figure 3

Levels of Population Differentiation at the NLRs The F_ST statistic is presented as a function of heterozygosity for each SNP in (A) Africans versus Europeans, (B) Africans versus East-Asians, and (C) Europeans versus East-Asians. The 95^th and 99^th percentiles of the Human Genome Diversity Panel-Centre d'Etude du Polymorphisme Humain (HGDP-CEPH) genotyping dataset for the same individuals as those studied here are shown as dashed lines, whereas the blue area corresponds to the 99.9^th percentile. Black and red points represent silent and nonsynonymous SNPs, respectively. For each outlier SNP, the gene name, followed by its position respective to the ATG, is indicated. Outlier SNPs separated by a comma correspond to SNPs in complete LD, and nonsynonymous SNPs are underlined. Note that, in Asian populations, a group of SNPs in NLRP6, including the nonsynonymous SNP 1652C>A (Leu163Met), displayed some of the highest levels of differentiation of any of the SNPs studied.

Figure 4

Detection of Recent Positive Selection Acting on NLRP1 in Europe and on NLRP14 in Asia We plotted iπ_A/iπ_D values against derived allele frequencies (DAFs) for (A) NLRP1 in Europe and (B) NLRP14 in Asia. We obtained p values by comparing the iπ_A/iπ_D values for NLRP1 and NLRP14 with the expected values obtained from 10⁴ simulations by using a best-fitted demographic model of human populations; this model is the most conservative in the context of the detection of positive selection. The upper dashed line on the graph corresponds to the 99^th percentile, and the lower line corresponds to the 95^th percentile. Black and red points represent silent and nonsynonymous SNPs, respectively. Outlier SNPs separated by a comma correspond to SNPs in complete LD, and nonsynonymous SNPs are underlined. As for NLRP1, in addition to the selected SNP 51015G>A (Val1059Met), our analyses also identified another nonsynonymous SNP (SNP 62201G>A, Val1184Met) linked to an intronic variant (SNP 63236G>A). This signal might be a complex repercussion on the worldwide selective sweep. For the DIND analyses of all genes in all populations, see Figure S5.

Figure 5

Hierarchical Model Outlining the Evolutionary Dynamics and Biological Relevance of the Various Families of PRRs This representation is based on the intensity of the selective constraints (based on the MKPRF results) detected for the 34 PRRs. These analyses allowed us to distinguish three groups of genes: genes under purifying selection (ω < 1, in red), genes under weaker selective constraints (γ < 0, in yellow), and genes for which no deviation from neutrality was detected (in gray). Color intensity is proportional to the –log(p value) of ω or γ tests. Cellular sublocalization, protein domains, and ligands are given as an indication but are not exhaustive.