Selection of Phototransduction Genes in Homo sapiens

Abstract

Purpose.: We investigated the evidence of recent positive selection in the human phototransduction system at single nucleotide polymorphism (SNP) and gene level.

Methods.: SNP genotyping data from the International HapMap Project for European, Eastern Asian, and African populations was used to discover differences in haplotype length and allele frequency between these populations. Numeric selection metrics were computed for each SNP and aggregated into gene-level metrics to measure evidence of recent positive selection. The level of recent positive selection in phototransduction genes was evaluated and compared to a set of genes shown previously to be under recent selection, and a set of highly conserved genes as positive and negative controls, respectively.

Results.: Six of 20 phototransduction genes evaluated had gene-level selection metrics above the 90th percentile: RGS9, GNB1, RHO, PDE6G, GNAT1, and SLC24A1. The selection signal across these genes was found to be of similar magnitude to the positive control genes and much greater than the negative control genes.

Conclusions.: There is evidence for selective pressure in the genes involved in retinal phototransduction, and traces of this selective pressure can be demonstrated using SNP-level and gene-level metrics of allelic variation. We hypothesize that the selective pressure on these genes was related to their role in low light vision and retinal adaptation to ambient light changes. Uncovering the underlying genetics of evolutionary adaptations in phototransduction not only allows greater understanding of vision and visual diseases, but also the development of patient-specific diagnostic and intervention strategies.

Introduction

Understanding the genetic basis of evolutionary adaptations in vision will help understanding vision and its diseases, as well as our understanding of selective processes. ¹ One approach to look for evidence of recent selection in humans is observation of the distortion in polymorphism patterns across the genome among different human populations. Even the patterns at neutral sites can reveal evidence of selection caused by selective pressure applied to the alleles at genetically linked sites. Thus, observation of selection is possible within haplotype blocks, regions of the genome where relatively little historic recombination has occurred. ² These regions are the hallmarks of the linkage disequilibrium, which has persisted throughout the lineage of anatomically modern humans, most likely arising less than 150 to 200 thousand years ago (ka). ³ In contrast, Homo sapiens only recently (within the last 100 ka, for the most likely out of Africa models) has distributed itself from its origin over the globe. ⁴ During this migration, selective pressure was present constantly and has been shown to persist to the last century. ^5,6 The new habitats that H. sapiens encountered were substantially different from the African habitat, which probably consisted of savannah-like environments. ^7,8 Recent evolutionary adaptations that are advantageous in these new habitats have been detected in non–sub-Saharan African populations, including in the lactose tolerance gene LCT, ⁹ salt regulation genes at the CYP3A cluster, ¹⁰ disease resistance genes, G6PD and CASP12, ^11,12 and pigmentation genes SLC24A5 and MATP. ^13,14  

Although some investigators have hypothesized that selective pressure has resulted in structural differences in visual organs, ¹⁵ to our knowledge relatively little work has been done to identify the specific set of vision-related genes that may be under selective pressure. The basic structure of the single-lens camera eye, its ciliary photoreceptor cells, and the process of phototransduction have been highly conserved for at least 400 million years (Ma), with only modest structural differences between gnathostome species. ¹⁶ Because of the visual system's role in overall fitness, many of the involved genes also have been highly conserved. ^16,17 However, the Asian and European habitats differed from the African in the lower mean illuminance levels, diurnal and seasonal illuminance cycles, cover provided in forested areas, changes in illuminance associated with the use of covered shelters, and increased variance in ground cover reflectivity. ^7,8,18–20 These new environments could have exerted selective pressure on the functional systems in the human eye. 

Specifically, we focus on evidence of selection within the genes involved in phototransduction. Visual phototransduction is the biologic process by which light is converted into an electric signal in the photoreceptor cells in the retina of the eye. It involves, at a minimum, absorption of a photon by an opsin holoprotein-associated chromophore, activation of a GTP binding protein that in turn activates phosphodiesterase, conversion of cGMP to GMP, and closing of the constitutively open cGMP gated channels. ²¹ The known set of interacting proteins responsible for phototransduction are encoded by the following genes: RHO, OPN1SW, OPN1MW, OPN1LW, SAG, GRK1, GRK7, RCVRN, GNAT1, GNAT2, GNGT1, GNGT2, GNB1, RGS9, RGS16, PDE6A, PDE6B, PDE6C, PDE6D, PDE6G, PDE6H, GUCY2D, GUCY2F, GUCA1A, GUCA1B, SLC24A1, CNGA1, CNGA3, CNGB1, CNGB3, and CACNA1F. See Figure 1 for an illustration of the interactions of phototransduction genes in rod photoreceptors. 

Figure 1

 

Phototransduction pathway in rod cells. The schematic illustrates the interactions between proteins during phototransduction. The genes encoding each protein in the pathway are overlaid on the schematic.

Because comprehensive functional data at the molecular and organismal scales on these genes currently are unavailable, we used the classic genetic determination of selection, that is, patterns of allelic variation that are not consistent with a neutral model of segregation. Specifically, we leveraged the data from the large-scale genotyping endeavors of the International HapMap Project and the 1000 Genomes project to identify patterns of allelic variation among populations. ^22,23  

Patterns of allelic variation can differ from the expectations of neutrality in multiple ways. Two important characteristics of allelic variation in the presence of selection are long haplotypes and large variance in allele frequencies among populations. ^24–26 Long haplotypes arise from selected variants quickly reaching high frequency before recombination breaks the associations with nearby polymorphisms. ²⁵ The variance in allele frequencies among populations is a result of the differential selection occurring in these populations—a variant and its linked polymorphisms rise in frequency in one population relative to others because of greater selective pressure in that population relative to the others. ²⁶ Here, we evaluated polymorphisms using two metrics that attempt to measure each of these aspects of allelic variation from neutrality. Extended haplotype homozygosity per site (EHHS) and its related metrics are used to quantify the length of haplotypes surrounding each polymorphism and compare these lengths across populations to find sites where differential selection is occurring. Fixation index (F_st) is used to quantify the level of differential selection by comparing directly the observed level of variance in allele frequency across the populations to the expected level. 

The hypotheses of this pilot study are that Asians and Europeans underwent selective pressure to adapt to changing visual environments, and that the selective pressure on the phototransduction system can be demonstrated using metrics of allelic variation at the gene-level. 

Materials and Methods

Single Nucleotide Polymorphism (SNP) Data and Quality Control

The analyses were performed using data generated by the international HapMap project. ²² Specifically, SNP genotypes generated during phase 3 (release 28) of the HapMap project for the European in Utah (CEU), Han Chinese in Beijing (CHB), and Africans in Yoruba (YRI) populations were used in this work. The HapMap data used here were generated using assays from 450 individuals (180 in CEU, 90 in CHB, and 180 in YRI). No human or animal testing was performed as part of this study and all analyses were performed on publicly available data. 

The HapMap SNPs were filtered based on the following criteria. First, to avoid sex-chromosome related complexities only autosomal SNPs were considered. Second, to maximize discriminatory power, only SNPs genotyped in all of the populations were considered. Finally, SNPs that were missing in a high percentage (>2%) of samples were excluded. The resulting filtered set of SNPs was used in all analyses. 

EHHS-Based Statistics

EHHS values were calculated for all SNPs in each of the three populations. ^25,27 For each SNP, its EHHS signal was calculated over the local region until the value decayed to 0.1. ^27–29 The integrated EHHS (iES) values for each SNP then were determined by integrating the SNP EHHS signal over the local region. SNP EHHS and iES were calculated using the publicly available REHH package for R. ³⁰  

Missing genotype values were filled in using a probabilistic algorithm described by Tang et al. before the EHHS was calculated. ²⁷ This algorithm assigns the missing genotype for an SNP based upon the genotype at the previous SNP and the conditional probability calculated using the nonmissing data at those two SNPs. ²⁷  

The ln(Rsb) statistic, which compares EHHS for the same SNP in two different populations, was calculated as the log ratio of the SNP iES value in the population of interest to the iES value in the YRI population for the CEU and CHB populations as this is assumed to be the ancestral population. ²⁷ Finally, the ln(Rsb) values for all SNPs within each population were normalized to have a median of zero and variance of one. The median, rather than the mean, was used for normalization to reduce the effects of outlier ln(Rsb) values. 

F_st Statistics

F_st was calculated from the genotype type data using a previously described method. ^31,32 Briefly, this method estimated F_st using the ratio of observed variance in allele frequency across populations to the expected variance as shown in Equation 1.  

 

The observed allele frequency variance for each SNP, σ ² , was calculated directly using HapMap genotype data. The expected variance for each SNP was estimated from the data using the weighted average of allele frequency, p̂, in the populations under comparison.  

F_st values were computed for each SNP in the CEU and CHB populations. In each case, the population of interest was compared to the YRI population. ²⁷  

Extending SNP Level Statistics to the Gene Level

Two procedures were used to extend the above SNP level statistics to summary measurements at the gene level. For ln(Rsb), a gene-level statistic was calculated as the sum of squared ln(Rsb) values for all SNPs within a gene, normalized by the number of SNPs within that gene.  

 

Here, the gene-level metric G_Rsb was computed for gene G based upon the ln(Rsb) value for each SNP i in G and the total number of SNPs in G, n.  

For F_st, the gene-level statistic G_Fst was calculated as the median F_st value of all SNPs within a gene G.  

 

Genomic bounds for known protein-coding genes retrieved from Ensembl (build 65) were used to assign SNP membership to genes. ³³ G_Rsb and G_Fst values were calculated for all known protein-coding genes in the CEU and CHB populations. Using the distributions of G_Rsb and G_Fst for all genes, the corresponding percentiles were calculated for each of the phototransduction genes in both populations. These percentiles serve as an estimate of the relative amount of positive selection occurring on each gene.  

Definition of Reference Positive and Negative Selection Gene Sets

We chose a set of positively selected, nonphototransduction genes that included EDAR, LCT, OCA2, and ABCB1, as these all have been widely reported to be undergoing recent positive selection. ^{28,29,34–39} We chose a set of negatively selected genes from human housekeeping genes, ⁴⁰ as these genes are constitutively active and any mutations will have systemic effects. Therefore, they are expected to be under negative selective pressure. A previously published meta-analysis of gene expression data across a wide variety of H. sapiens tissues, generated an ordered set of putative housekeeping genes. ⁴⁰ The top genes from this set according to FPEI, a homogeneity metric that expresses the ubiquity of that gene expression, and which contained at least 10 SNPs from the filtered set considered here. The resulting negatively selected genes were LDHA, CFL1, HEATR2, and ATP5L. 

Results

After quality control, a resulting set of 1,108,091 SNPs was used for EHHS and F_st analyses. The ln(Rsb) and F_st measurements of SNP-level selection had a low-level but consistent correlation of 0.227 in the CEU population and 0.225 in the CHB population. Heat maps showing the ln(Rsb) and F_st joint distributions for all SNPs are shown in Figure 2. The correlation observed is consistent with previous research that has indicated complementarity between these selection metrics. ^41,42 That is, while both metrics are useful for quantifying evidence for positive selection, they focus on different genetic signatures associated with selection. 

Figure 2

 

Joint distribution of ln(Rsb) and F_st for genome-wide SNPs. Heat maps indicating the density of the positive selection measures for 1,108,091 autosomal SNPs in the CEU and CHB populations. Top: distribution across the full range of Fst values. Bottom: distribution zoomed in on F_st values between 0.0 and 0.1.

The Table details the gene-level summary measurements for all autosomal phototransduction genes. The criterion applied to determine the genes under positive selection from the set of all phototransduction genes was the existence of a gene-level metric (G_Rsb or G_Fst) greater than the 90th percentile value in either the CEU or CHB population. This criterion was chosen instead of a more conservative one to increase the sensitivity for genes under positive selection. Several phototransduction genes showed evidence of recent positive selection in CHB and/or CEU. The genes PDE6G, GNAT1, SLC24A1, and RGS9 all resulted in gene-level selection measurements greater than 90% of all known protein-coding genes in the CHB population. For the CEU population, this set of genes consisted of RHO, GNB1, GNAT1, and RGS9. The union of these was considered to be the set phototransduction genes under recent positive selection, and consisted of the six genes: RGS9, GNB1, RHO, SLC24A1, PDE6G, and GNAT1. 

Table.

 

G_Rsb and G_Fst Statistics for Autosomal Phototransduction Genes for the CHB and CEU Populations

Gene	No. SNPs	CHB				CEU
		GRsb	%	GFst	%	GRsb	%	GFst	%
RGS9*	36	1.397	0.77	0.197	0.92	7.669	0.98	0.211	0.95
GNB1*	26	1.259	0.75	0.136	0.85	1.060	0.7	0.257	0.97
RHO*	2	0.312	0.28	0.108	0.79	6.381	0.97	0.370	0.99
SLC24A1*	10	4.955	0.95	0.105	0.78	1.791	0.83	0.014	0.15
PDE6G*	6	3.631	0.92	0.082	0.7	0.226	0.18	0.063	0.65
GNAT1*	2	0.497	0.44	0.576	0.99	0.109	0.07	0.193	0.94
PDE6D	18	2.553	0.88	0.017	0.17	1.417	0.77	0.050	0.56
CNGB1	59	0.333	0.3	0.087	0.73	0.467	0.41	0.077	0.72
RGS16	1	0.403	0.37	0.035	0.37	0.654	0.54	0.128	0.87
SAG	25	0.798	0.62	0.120	0.82	0.364	0.32	0.016	0.18
GNGT1	128	0.760	0.6	0.026	0.27	0.691	0.56	0.038	0.45
GNGT2	10	0.111	0.07	0.040	0.42	0.305	0.26	0.005	0.06
GNAT2	5	1.045	0.7	0.028	0.29	0.592	0.5	0.019	0.21
PDE6A	36	0.479	0.43	0.032	0.34	0.505	0.43	0.070	0.69
PDE6B	11	0.269	0.23	0.009	0.09	1.099	0.71	0.058	0.62
GUCY2D	7	2.272	0.86	0.035	0.37	0.294	0.25	0.032	0.38
GUCA1A	21	1.815	0.82	0.117	0.81	1.607	0.80	0.014	0.15
GUCA1B	10	2.009	0.84	0.147	0.86	0.681	0.55	0.093	0.77
OPN1SW	3	0.036	0.02	0.001	0.02	0.549	0.47	0.065	0.66
RCVRN	9	0.055	0.03	0.007	0.07	0.536	0.46	0.016	0.18
CNGA1	32	0.140	0.1	0.029	0.3	0.169	0.12	0.018	0.19
GRK7	10	0.132	0.09	0.093	0.75	0.136	0.11	0.051	0.56
PDE6H	9	0.821	0.63	0.006	0.06	0.049	0.04	0.006	0.07
PDE6C	42	0.877	0.65	0.030	0.31	0.379	0.33	0.074	0.71
CNGA3	26	2.055	0.84	0.025	0.26	2.130	0.86	0.032	0.38
CNGB3	95	0.700	0.56	0.063	0.57	0.549	0.47	0.037	0.44

 The raw values for each gene are shown along with the gene percentile within each population.

*   Achieved G_Rsb or G_Fst above 90th percentile in at least one of the populations.

SNP-level selection measurements of known positively-selected, negatively-selected, and phototransduction genes are shown in Figure 3. Here, ln(Rsb) and F_st values of each SNP in these genes are plotted against their position within the genes. The chromosomal location of each gene is shown along with SNP-level values for the CEU and CHB populations are shown. Plots of SNP-level values for the set of known positively-selected (Fig. 3A) and negatively-selected genes (Fig. 3B) are included for comparison. As expected, the known positively-selected genes exhibited high magnitude SNP-level selection metrics along their lengths relative to the negatively-selected genes and most of the phototransduction genes. 

Figure 3. 

 

Signal of positive selection across genes. SNP ln(Rsb) and F_st values across the length of some positively-selected (A), negatively-selected (B), and phototransduction genes (C).

Within the set of phototransduction genes under recent positive selection (Fig. 3C), SNP-level ln(Rsb) and F_st values revealed distinct levels and modes of selection acting on the genes. The SNP-level measurements for gene RGS9 showed elevated values across the length of the gene as well as distinct peaks along the length of the gene, covering several exons near the 3′ end of the gene. The SNP-level measurements of gene GNB1, on the other hand, revealed a lower magnitude, uniform selection across the entire length of the gene. Interpretation of SNP-level measurements for some of these genes (e.g., RHO, GNAT1) is limited by the lack of dense SNP coverage. 

Discussion

Our results indicated that there is evidence for selective pressure in the genes involved in retinal phototransduction, and that traces of this selective pressure can be demonstrated using SNP- and gene-level ln(Rsb) and F_st metrics of allelic variation. 

The joint distributions of ln(Rsb) and F_st measurements (Fig. 1) are consistent with the fact that these measure different aspects of selection. Specifically, ln(Rsb) measures the differences in haplotype length surrounding a SNP, while F_st measures the variance in allele variances at the SNP. Here, both were used to increase the overall sensitivity of finding phototransduction genes undergoing recent positive selection. Additionally, these SNP-level measurements were extended to simple gene-level metrics so that evidence for positive selection could be evaluated at the gene level. 

Genes shown previously to be under positive selection were used as positive controls. Three of these four genes (EDAR, OCA2, and LCT) corresponded to a G_Rsb or G_Fst value that was at the 90th percentile or higher in at least one the populations. Figure 3 also shows that the selection signals at SNPs in the genomic regions surrounding these genes were consistently high in most cases. The fourth gene, ABCB1, did not achieve G_Rsb or G_Fst above the 90th percentile, but has consistent, low-level evidence of selection across its length. In contrast, the SNPs in genes serving as negative controls (LDHA, CFL1, HEATR, and ATP5L) had greatly diminished signals. 

The set of retinal phototransduction genes with relatively strong evidence for recent positive selection at the gene level consists of RGS9, RHO, GNB1, PDE6G, GNAT1, and SLC24A1, of which the RGS9 has the most striking magnitude of F_st and ln(Rsb) metrics at the SNP level. GNB1 exhibited high magnitude SNP-level F_st signals, but a consistently negative ln(Rsb) signal, possibly indicating selected variants arose and quickly increased in frequency in the African population due to selective pressures not encountered by populations leaving Africa. 

Determining specific genes under positive selection by extending the SNP-level selection measurements to the gene-level summary metrics allows existing knowledge of gene functions and disease associations to be leveraged in developing hypotheses of the functional significance of the positive selection. Deleterious mutations in the selected retinal phototransduction genes have been associated with reduced visual acuity in low light environments and abnormalities in retinal adaptation to changes in ambient light. Specifically, mutations in RGS9 have been found to be associated with bradyopsia, a disease characterized by extreme delay in retinal adaptation to bright and dark light. ⁴³ Mutations in PDE6G in mouse models have been shown associated with delayed retinal adaptation to rapid changes in light level in mice. ⁴⁴ The genes GNAT1, RHO, and SLC24A1 have been associated with either dominant or recessive forms of congenital stationary night blindness. ^45–47 Among these selected genes, GNB1 is notable for its lack of specificity to phototransduction. It is expressed in a wide variety of tissue and is involved many biological processes. ⁴⁸ Selection acting on this gene may be caused by pressure on those processes, or could have functional consequences beyond phototransduction. It also should be noted that this selected set consists of genes active in phototransduction in rods only (RHO, GNAT1, GNB1, and PDE6G) or in rods and cones (RGS9 and SLC24A1). No cone-specific genes were found among the most highly selected set of phototransduction genes. 

Based on functional annotation and known disease associations of the selected genes, the proteins encoded by these genes are likely to have an important role in low light vision and retinal adaptation to ambient light changes. The rod-focused nature of the observed selection may suggest that improved acuity in low and intermediate light (scotopic and mesopic) rather than higher light (photopic) conditions was an important driver of the observed selection. Previous work on two of the selected genes (RGS9 and PDE6G) also shows that mutations in these genes have dramatic effects on retinal adaptation to changing light levels and more recent work also has found a signature of selection acting on RGS9. ⁴⁹ This suggests that an increased ability or immediacy in adapting to changing light levels also could have driven the selection. Based only on the observed levels of selection on these genes, we cannot distinguish between these hypotheses or preclude the possibility that some other functional consideration is driving the selection. 

Upon migrating to higher latitudes, H. sapiens would have encountered very different lighting environments than what was typical of their previous habitats. Lower mean illuminance levels, more pronounced diurnal and seasonal cycles, forest or foliage cover, and the use of covered shelters would have increased the importance of maintaining visual acuity in low light conditions. Additionally, an increased exposure to rapidly changing light levels would have resulted from the traversing through forested areas and snow-covered tundras as well as entering or exiting covered shelters. Changes in the environmental albedo (proportion of light reflected) also could have increased the exposure to rapidly changing light levels and visual scenes with a high dynamic range of illumination. The albedo values typical of higher latitude ground cover, from deep forests to light vegetation and tundra to complete snow cover (0.08–0.9), exhibit much higher variance compared to those of the savannah-like environments encountered by ancient humans in Africa (0.15–0.30). ^18–20,50 H. sapiens capable of maintaining high visual acuity in low light conditions and/or rapid retinal adaptation to changing light levels during hunting and predator avoidance would have a selective advantage in these habitats. 

The F_st and EHHS-based quantitative metrics used here are able to distinguish the observed allelic patterns that are likely due to selective pressure from those expected under neutral selection. While these results suggested that a subset of phototransduction genes have been under recent positive selection, they cannot confirm the functional significance of this selection. Therefore, this study is limited by the lack of ability to identify uniquely the predominant driving force behind the observed positive selection in the phototransduction system. Without direct functional evidence, it is impossible to determine the relative importance of increased visual acuity in low light conditions, more rapid adaptation to changing light levels, or even some other functional change in conferring a selective advantage. 

The gene-level selection metrics presented are subject to a few limitations that should be noted. The evaluation of the numeric metrics of selections could be performed only at known SNPs, and several genes (RHO, GNAT1, and PDE6G) contained only a very small number of SNPs. Any noise in the SNP-level measurements could result in serious biases in the gene-level metrics. In addition, the gene-level selection metrics presented may not be ideal for characterizing the most recently occurring selection. In general, more recent selection is associated with large regions (hundreds of thousands to millions of bases) exhibiting elevated selection metrics because insufficient time has passed for these regions to be fragmented thoroughly by recombination events. Because many of the genes considered span much smaller regions (tens of thousands of bases), it is not always clear based solely on the gene-level metrics that the genes with high gene-level metrics fall within a larger region of recent selection. These large regions also may overlap multiple genes, making it difficult to determine which are responsible for driving the selection. A qualitative assessment of the selected genes can help reveal their relationship with the larger regions of selection. Examination of the surrounding genomic region reveals that the genes RHO, RGS9, PDE6G, and SLC24A1 occur within large regions of selection spanning at least several hundred thousand base pairs (see Supplementary Figs. S1–S6). The region corresponding to RGS9, in particular, is centered on this gene and overlaps few others. 

The methods we use also lack the resolution to localize the selection to specific variants within the selected genes. The polymorphic variants (SNPs) used in this analysis are unlikely to be directly functional, but instead are in linkage disequilibrium with the functional variants. Identification and confirmation of the functional variants in these genes is the next logical step. A set of variants within the genes undergoing selection could be generated using sequence data from members of the three populations considered here. Requiring that sequence variants have substantially different frequencies across populations and application of commonly used variant filters would result in a set of putative selected variants. Functional study of these variants then could be performed to determine the impact of these variants on the observed positive selection. Functional evidence could be gathered either by defining the phenotype in human subjects using functional tests, such as low light acuity or photostress testing, ⁴³ or animal models engineered to harbor the same variants found in different human populations. This evidence would help to determine if the selected variants were likely to impart an advantage in low light vision, retinal adaptation to changing light levels, or have some other role. 

Preliminary analysis of the data from the Exome Sequencing Project (available in the public domain at http://evs.gs.washington.edu/EVS/) ⁵¹ has identified several plausibly functional variations in different prevalence between the European American and African American populations. The comprehensive identification of selected functional variants in phototransduction genes is the focus of ongoing investigation, the details of which are beyond the scope of this study. 

In summary, we have found evidence for selective pressure in the genes involved in retinal phototransduction in H. sapiens , at the SNP and gene levels, and outlined plausible hypotheses for the functional impact of this selective pressure. Further, we have proposed experiments for determining the selected genetic variants and functional impact of the selective pressure acting on the human phototransduction system. 

Supplementary Materials

pdf S1 

Acknowledgments

Supported by Grants NEI-EY017066, R01 EY018853, R01 EY019112 (MDA), NEI-EY017451 (RFM, TES), and NEI-EY018825 (TES) from the National Eye Institute and National Institutes of Health; BR-GE-0608-0457-UIA (TES) from the Foundation Fighting Blindness; and VA-5I01CX000119 (MDA, MC, TES) from the Veterans' Administration. 

Disclosure: M. Christopher, None; T.E. Scheetz, None; R.F. Mullins, None; M.D. Abràmoff, None