A previous comparison
42 demonstrated that the genes within 1 Mb of CREAM refractive error loci share only four commonalities with the genes differentially expressed in two animal transcriptome studies of refractive error.
22,24 In the present study, however, we demonstrated that expanding this initial comparison to include all available exploratory human and animal studies implicates a further 99 genes (i.e., 103 in total) and 8 proteins that are both differentially expressed in animal models and located within 1 Mb of human GWAS loci. We further show that, although a very large number of genes and proteins have been implicated across these studies, the overlap between human GWAS candidate genes and the genes downregulated during early myopia induction in animal models is statistically significant (i.e., more than expected by chance). These results extend the findings of previous targeted studies,
12,40,41 to provide systematic evidence that many similar genes are associated with refractive error loci in humans and optically induced refractive change in animal models.
Further research is needed to elucidate the biological mechanisms underlying these cross-species commonalities. Our results are consistent with a model in which genetic and environmental factors control ocular growth via similar biological pathways, and/or a model in which genetic variants alter susceptibility to environmental factors rather than being causative in and of themselves (as previously suggested).
56 Notably, there was greater concordance between GWAS candidate genes and the genes and proteins that were downregulated (rather than upregulated) in animal studies. This finding is difficult to explain given the heterogeneous data being compared (i.e., genetic variation versus environmentally driven expression changes), but could plausibly reflect a bias for genetic variants that decrease gene expression (e.g., by disrupting transcription factor binding).
57
Although proteomics studies showed some overlap with both transcriptome studies and GWAS candidate genes, only the proteomic–transcriptome overlap was statistically significant. The lower concordance between GWAS and proteomics studies could reflect several factors. Firstly, some degree of discordance is expected given that the ratio of genes to proteins is not one-to-one, and there are an array of intermediate transcriptional and posttranscriptional regulatory mechanisms.
58,59 Indeed, correlation is generally low even at the mRNA and protein level
60 (an observation concordant with the present results where fold changes across transcriptome and proteomics studies were often inconsistent). Secondly, most of the proteomic studies used two-dimensional gel electrophoresis (2D-E) combined with large fold-change cutoffs (ranging from ≥1.3–3). Two-dimensional gel electrophoresis has difficulty separating hydrophobic membrane proteins,
61 which were enriched in the GWAS–transcriptome commonalities (see
Fig. 4C in the Results section). Because the 2D-E approach is limited to the detection of proteins showing large fold changes, it may also decrease concordance with GWAS that profile the entire genome (particularly if selection pressures create a bias for genetic variants linked to modest changes in gene and/or protein abundance).
62,63 Indeed, all of the GWAS–proteomic overlaps originated from the two proteomics studies
32,37 with arguably higher sensitivity (based on their statistical parameters and the number of proteins implicated). Finally, it seems likely that some of the environmentally driven expression responses involved in human myopia are not mediated by underlying genetic variation.
7,64 The GWAS data do not necessarily capture these “environment-only” responses, presumably lowering the overall concordance with animal studies of environmentally induced refractive error.
In addition to assessing the overall significance of cross-methodology overlaps, we considered the impact of several animal study methodologic parameters (species, tissue, time-point, and optical manipulation) on the pattern of results. These latter comparisons demonstrated that phylogenetically distant model species show similar gene (e.g., chick and primate) and protein (e.g., tilapia and tree shrew) expression changes, suggesting that expression responses are well conserved across animal models of environmentally driven refractive error. Moreover, although most genes associated with each optical manipulation were unique, most models of myopia induction and growth slowing showed significant overlap at the gene and/or protein level, suggesting that a subset of similar biological mechanisms is activated across growth paradigms. Lastly, we observed commonalities between the genes and proteins differentially expressed in the retina/RPE/choroid and the proteins differentially expressed in the sclera. These results are consistent with theories of visually driven ocular growth that postulate a signal (or cascade of signals) propagating from the retina through to the sclera,
19,65–67 as any such signaling mechanism would presumably elicit related expression responses across multiple tissue layers. The cross-tissue transcriptome–proteome and proteome–proteome commonalities were primarily composed of genes involved in mediating cellular and extracellular structure including cytoskeletal proteins (
MYH8,
TPM3,
GMFB,
ACTB,
GSN,
VIM), collagen (
COL14A1), and transcripts that stimulate cell growth and proliferation (
ANGPTL7,
CDC42). Molecular chaperones (
CRYBA1,
CCT8), albumin (
ALB), and apolipoprotein A-I (
APOA1) were also implicated. Notably, these cross-tissue commonalities included the two genes (
GMFB,
COL14A1) implicated by all methods (GWAS, transcriptome, and proteomics), providing strong support for their role in inducing structural change across multiple posterior ocular layers.
Our assessment of methodologic parameters also identified several conditions that provided increased concordance with human GWAS candidate genes. Although these findings contribute to the existing debate around the relative efficacy of different animal models for understanding human myopia,
19,39,43,68–71 they should be broadly interpreted with caution given the heterogeneous designs of the studies included in each methodologic list. Here, we found that GWAS candidates showed greater concordance with the genes differentially expressed at early (relative to late) induction time-points. A similar pattern was not seen for the proteomic data where only one study (with two differentially expressed proteins) met inclusion criteria for early expression changes. Notably, however, five of the eight GWAS–proteomic overlaps originated from the study by Frost and Norton
32 of LIM in the slower tree shrew model. Although this dataset was included in the “late” induction category, the animals had only achieved moderate refractive shifts (−3.4 D). The greater concordance with animal studies at early time-points and/or when refractive shifts were small to moderate is particularly interesting in light of recent findings that many of the CREAM GWAS loci show early onset effects in children that remain stable or progress further with age.
72 In this context, our findings support the biological validity of the proposal by Stone and Khurana,
56 and others (e.g., Guo et al.
73 and He et al.
74) that different genes are involved in the onset versus the persistence and progression of refractive change.
Genome-wide association study candidate genes also showed greater concordance with the genes differentially expressed in LIM (relative to other optical manipulations). This finding is less convincing, as the effect size was weak and other methodologic factors varied considerably across the different optical manipulation groups (e.g., induction time, species, and the number of studies). Similarly, our analysis demonstrated that the genes differentially expressed in the retina/RPE/choroid (i.e., all of the genes implicated in transcriptome studies) overlapped significantly with the genes near human GWAS loci, while the proteins differentially expressed in the retina/RPE/choroid and sclera do not. In the absence of transcriptome studies profiling sclera, it seems likely that this finding reflects the overall greater concordance of the transcriptome results with GWAS candidate genes.
It is notable that, although four model species (chick, mouse, tree shrew, and primate) showed some commonalities with GWAS candidate genes, there was no evidence of more significant overlap in similar species to humans (e.g., primates or mammals). In combination with the significant cross-species overlap at the gene and protein level discussed above, these findings suggest that model species is not an important factor for determining the degree of overlap between environmentally mediated expression changes in animals and the genes near human refractive error loci. However, it should be noted that data in higher-order animal species such as primate are currently limited and, as such, future studies may well uncover evidence for a species effect.
Heterogeneous experimental designs and incomplete data availability necessitated a few important limitations in our analysis. We identified candidate genes on the basis only of their proximity to refractive error peaks,
47 and thus they are not necessarily those by which a given SNP affects refractive development. Moreover, our analysis was limited to protein-coding genes and may have missed more complex effects mediated by transcriptional and posttranscriptional regulators such as microRNAs
75 (Tedja MS, et al.
IOVS 2016;57:ARVO E-Abstract 4791). Many of the animal studies included in our analysis used expression profiling techniques with limited gene or protein coverage, and most also imposed fold-change cutoffs (see
Supplementary Table S1). These factors presumably decrease concordance with GWAS that profile the entire genome. In addition to using varied statistical criteria, the included animal studies encompassed a wide range of experimental designs (i.e., animal age and circadian timing, measurement platform, tissue, induction time-point, optical manipulation, and species). As mentioned previously, these methodologic variations made it difficult to compare equivalent conditions across transcriptome and proteome methodologies, and to segregate out balanced datasets representing different methodologic parameters. Moreover, these methodologic variations precluded finer evaluation of variables such as tissue type and induction time (which were instead assessed in crude groupings). As such, although our primary finding regarding significant overlap between GWAS candidate genes and the genes differentially expressed in animal models is robust, our remaining findings regarding the effects of different animal study methodologic parameters should be interpreted with caution.
To improve the systems-level understanding of ocular growth control, future studies need to integrate multiple levels of omics data collected under comparable experimental conditions. In particular, expression quantitative trait locus (eQTLs) studies combining genotype and gene expression data from posterior ocular tissues are needed to improve candidate gene predictions for GWAS loci.
76 Studies collecting concurrent transcriptome and proteome measurements in identical animal models may also help to elucidate whether the fold-change discordance we identified across these study types was due to methodologic factors (such as the timing of expression profiling) or posttranslational events. Given the recent emergence of next-generation transcriptome and proteomic technologies,
77,78 such future studies are also likely to generate better-quality datasets in terms of reproducibility, sensitivity, and coverage. It is also notable that large-scale exploratory transcriptome studies of gene expression in the sclera during refractive error induction are currently unavailable (although some preliminary results have been reported in mouse).
79 As noted above, our analysis identified significant overlap of proteins differentially expressed in the retina/RPE/choroid and sclera. Given that transcriptome and proteomics results also showed greater within- than across-method concordance, it is reasonable to expect that future scleral transcriptome studies would identify additional commonalities with retinal transcriptome responses that may help to elucidate the signaling mechanisms propagating across the posterior eye.
In summary, we conducted the first large-scale comparison of genetic, transcriptome, and proteomic studies of refractive error. We showed that gene and protein expression changes are well conserved across animal models of environmentally driven refractive error, and that the genes implicated in these animal models overlap significantly with the genes near human GWAS refractive error loci. These findings provide strong support for the continued use of animals for investigating the biological basis of human myopia and developing novel therapeutic approaches to control eye growth.