Impact of regulatory variation from RNA to protein
a 20 kb window around the corresponding gene. We used a shared standardization, normalization, regression, and per-mutation pipeline for all three phenotypes. At an FDR of 10% we detected 2,355 eQTLs, 939 rQTLs, and 278 pQTLs (Table 1, fig. S11).
There is substantial overlap among detected QTLs (fig. S12). Among the 4,322 genes quantified for all three pheno-types, 54% of the genes with pQTLs also have a significant rQTL and/or eQTL. Given the incomplete statistical power to detect QTLs in each dataset independently, we performed replication testing across datasets, using the specific SNP-gene pairs underlying each class of QTLs. This analysis is less sensitive to power limitations than genome-wide test-ing. The results confirm that many QTLs are shared across all three phenotypes (example in Fig. 1A). In particular, most (90%) genetic variants associated with ribosome occu-pancy are also associated with transcript levels (fig. S13). In contrast, eQTLs showed the lowest overlap with pQTLs (35%), as expected (Fig. 1B).
Our observation that many SNPs identified as eQTLs are not associated with differences in protein levels is consistent with the notion that, across species, protein levels diverge less than transcript levels (12–17). Yet some QTLs may not replicate at the protein level simply due to incomplete map-ping power. To address this, and to avoid over-estimation of effect sizes due to ascertainment bias at significant QTLs, we focused on eQTLs detected previously in European sam-ples by the GEUVADIS study (2). We then attempted to rep-licate the GEUVADIS eQTLs using our transcript, ribosome profiling, and protein data and considered the mean QTL effect size in each data type. Mean effect sizes calculated in this way are expected to be unbiased with respect to either technical or biological variance.
Using this approach, we observed a reduced mean effect size for the GEUVADIS eQTLs in the protein data compared to either the RNA-seq data (t-test P = 6.7×10 3) or ribosome data (P = 5.6×10 3; Fig. 1C). In contrast, the average effect sizes observed for the RNA-seq and ribosome data are not significantly different from each other, and their effect sizes are highly correlated across the tested eQTLs (Pearson c = 0.79, P < 10 96, fig. S14). The reduction in effect size observed in protein data is robust with respect to potential technical confounders, including iBAQ intensity level and transcript model complexity (fig. S15). We thus conclude that the ma-jority of genetic variants affecting transcript levels also alter ribosomal occupancy, typically with a similar magnitude of effect. Yet, both eQTL mapping and effect-size analyses in-dicate that many eQTLs have attenuated (or absent) effects on steady state protein levels (fig. S16).
In addition to the observation of generally attenuated ef-fect sizes in pQTLs compared with eQTLs, we identified a subset of variants that appear to affect levels of proteins but not mRNA, and hence are candidates to affect post-transcriptional gene regulation. To evaluate evidence for these, we tested each SNP for association with one regulato-ry phenotype, while treating one or both of the other pheno-types as covariates (conditional model). Considering protein levels, with RNA levels as a covariate, we identified 146 pro-tein-specific QTLs (psQTLs) at FDR = 10% (Fig. 2A). The identification of psQTLs is generally robust to the choice of technology used to characterize transcript expression (fig. S17).
Using an alternative approach, an interaction model, we identified 68 psQTLs with significantly larger effects in pro-tein than mRNA (LRT; FDR = 10%). We also used the inter-action model to identify 76 expression-specific QTLs (esQTLs, interaction model, LRT; FDR = 10%). We then con-sidered the ribosomal data. We found that the effect sizes for ribosomal occupancy are similar to the esQTL effect siz-es (Fig. 2B). Yet, for psQTLs, low ribosome effect sizes are observed. Thus, for QTLs with discordant effects between transcript and protein, the ribosome data usually tracked with levels of RNA. Put together, these results allow us to identify loci where genetic variants have specific impacts on protein levels that are not fully mediated by regulation of either transcription or translation and hence may affect rates of protein degradation.
Finally, we performed enrichment analysis in which we considered each tested gene-SNP pair separately, and evalu-ated the full distribution of P values from the conditional model (rather than choosing a significance threshold) for different genomic and functional annotations. SNPs within transcribed regions (exonic and UTR) are enriched for more significant psQTL effects, compared to intergenic or intronic SNPs, even within the narrow 20 kb windows tested (figs. S18, S19, and S20). In addition, psQTLs are further enriched for non-synonymous sites (compared to all exonic SNPs, Table 2).
Investigating additional annotations (Table 2, fig. S21), we found that non-synonymous SNPs near acetylation sites showed nominal enrichment for psQTLs. This possibly re-flects the functional role of lysine acetylation in modulating protein degradation (23). Overall, the enrichment results suggest that genetic variants involved in post-transcriptional regulation are functionally distinct from ge-netic variants that primarily affect transcription – they are more likely to fall within translated regions of the gene, and more likely to occur at non-synonymous sites.
In summary, we have shown that while a substantial fraction of regulatory genetic variants influence gene ex-pression at all levels from mRNA to steady state protein abundance, there are also a number of effects with specific impact on particular expression phenotypes. QTLs affecting mRNA levels are on average attenuated or buffered at the protein level, as has been observed between species (14). Our analysis indicates that this attenuation is not evident at the stage of translation. While the overall phenotypic simi-larity between ribosome occupancy and protein abundance is high, cis-regulatory genetic effects on ribosome occupancy appear to be more strongly shared with mRNA than with
In eukaryotes, most protein-coding genes are ...encoded short regulatory RNAs known as microRNAs. ...intrinsically sensitive to natural sequence variation...
MIRNA genes are transcribed by RNA POLYMERASE II (Pol II), similar to protein-coding genes [28,29]. In vertebrates and ?ies, most small regulatory ...
variation of the 28-kDa outer membrane protein gene (p28) of Ehrlichia ...RNA was puri?ed from E. chaffeensis isolates V1, V5, V6, V7, V9,...
aspects of sequence variation, regulation and ...pipeline, from protein expression to crystallization...of regulatory elements that control transcription ...
regulatory system with basic regulatory mechanisms ...Significant variation exists in the number of ...impact of a food source upon the selective ...
MicroRNA Target Recognition and Regulatory Functions...impact of miRNAs on both the expression and ...variation between individuals and how these ...
Protein-DNA Interaction Non-coding RNA regulation ...Genome structure variation detection Reference genome...As an approach for annotation of the regulatory ...
From RNA →Protein (Translation) Messenger RNA ...Why? 单林娜 制作 15 Regulatory gene Structural ...variation in sequence between different promoters, ...
regulatory variation._电子/电路_工程科技_专业资料...(b–f) With RNA, the SNP base will read out... Impact of selection, m... 暂无评价 26页 免费...
We used CTNNB1 shRNA directed against b-catenin to examine the impact of ...regulatory function of b-catenin, protein lysates were examined for the ...
热门文档