Transcriptome- and proteome-wide association studies identify genes associated with renal cell carcinoma
We performed a series of integrative analyses including transcriptome-wide association studies (TWASs) and proteome-wide association studies (PWASs) of renal cell carcinoma (RCC) to nominate and prioritize molecular targets for laboratory investigation. On the basis of a genome-wide association study (GWAS) of 29,020 affected individuals and 835,670 control individuals and prediction models trained in transcriptomic reference models, our TWAS across four kidney transcriptomes (GTEx kidney cortex, kidney tubules, TCGA-KIRC [The Cancer Genome Atlas kidney renal clear-cell carcinoma], and TCGA-KIRP [TCGA kidney renal papillary cell carcinoma]) identified 38 gene associations (false-discovery rate <5%) in at least two of four transcriptomic panels and identified 12 genes that were independent of GWAS susceptibility regions. Analyses combining TWAS associations across 48 tissues from GTEx identified associations that were replicable in tumor transcriptomes for 23 additional genes. Analyses by the two major histologic types (clear-cell RCC and papillary RCC) revealed subtype-specific associations, although at least three gene associations were common to both subtypes. PWAS identified 13 associated proteins, all mapping to GWAS-significant loci. TWAS-identified genes were enriched for active enhancer or promoter regions in RCC tumors and hypoxia-inducible factor binding sites in relevant cell lines. Using gene expression correlation, common cancers (breast and prostate) and RCC risk factors (e.g., hypertension and BMI) display genetic contributions shared with RCC. Our work identifies potential molecular targets for RCC susceptibility for downstream functional investigation.
Second author, with the NCI's Integrative Tumor Epidemiology Branch. My contribution was the analysis and interpretation of the data.
The largest GWAS of renal cell carcinoma — 29,020 cases against 835,670 controls — found 63 susceptibility regions. That is a good haul of coordinates and a poor haul of biology. Most of the variants sit in noncoding sequence, and a region is not a mechanism.
This study takes those summary statistics and asks which genes and which proteins carry the signal, for RCC overall and for its two major histologic subtypes.
Four kidneys and forty-eight tissues
Which reference transcriptome you use decides what you can find, so we used four: GTEx kidney cortex, kidney tubules, and the TCGA clear-cell and papillary tumor transcriptomes. Normal tissue and tumor tissue regulate genes differently, and running both makes the difference visible.
Individually the four panels returned 21, 48, 90 and 28 genes. Thirty-eight were significant in at least two panels — the set worth trusting.
Five genes came up in all four. Two of them, CASP9 and TMEM163, map to known GWAS regions. The other three do not: SULT1A2, PPIL3 and HLF have no genome-wide significant variant anywhere within a megabase of the transcription start site. They exist only as an accumulation of weak signals that TWAS aggregates and single-variant testing discards. HLF is a Tier-1 gene in the COSMIC cancer census.
We then meta-analyzed across 48 GTEx tissues using a Cauchy combination, which holds its type-I error under correlation, and cross-checked against UTMOST. This added 403 genes, and among them VHL and EPAS1 — the most frequently mutated gene in RCC and the gene encoding HIF-2α. Finding the two best-known genes in kidney cancer through an untargeted scan is the sanity check you want before believing anything else on the list.
Splitting the subtypes
Clear-cell RCC accounts for over 75% of cases, so results for "RCC" are largely results for ccRCC. Analyzing the subtypes separately is what exposes the differences.
PGAP3 associated with ccRCC and not with RCC overall. Papillary RCC is rarer and harder — 2,193 cases — and this is among the first studies to identify genes for it, including INTS8, a reported prognostic marker with no GWAS locus. Three genes, CASP9, IRF5 and PLEKHA6, are shared by both subtypes.
Adding proteins
This is the first PWAS for RCC. Using plasma protein prediction models from INTERVAL and ARIC, we found 13 protein associations, six of which also appeared in the TWAS.
Every one of them maps to a GWAS-significant locus — no protein-only discoveries. Plasma proteomic reference panels are far smaller than transcriptomic ones, and several of the prediction models explain little heritability, so these results carry more caution than the TWAS. The papillary analysis had too few cases to return anything.
Are the novel genes real
The genes found only by TWAS are the interesting ones and the ones most likely to be noise. They need independent support.
Using published ATAC-seq from RCC tumors and HIF ChIP-seq from RCC cell lines, we tested whether the cis-regulatory regions of the 145 RCC-associated genes are enriched for regulatory features, against a null built by shuffling regions across the genome 500,000 times.
They are. 78 of 145 had an active enhancer or promoter within 50 kb of the TSS, and 38 had a HIF binding site — HIF being the pathway RCC runs on. SULT1A2 and CDA, both TWAS-only, were among them.
That is the argument for these genes. Their regulatory variants never reached genome-wide significance individually, but they sit in genuine regulatory elements bound by the right transcription factors, and their cumulative effect on expression is what the TWAS picks up.
What RCC shares
Expression-mediated genetic correlation showed shared genetic contribution with breast cancer, and more weakly with bladder and prostate cancer. Ovarian cancer showed none.
A bi-directional regression supported causal effects of BMI and hypertension on RCC through genetically regulated expression, with no evidence in the reverse direction — consistent with the epidemiology, and now with a proposed molecular route. Smoking, systolic and diastolic blood pressure showed no such effect, which does not rule them out; it means this particular approach, working through gene expression, does not detect them.
Limits
Most participants in the GWAS and most donors of the reference tissue are of European ancestry, so ancestry-specific signals may be missed and the findings may not generalize.
The analysis covers cis effects only, within 500 kb of the TSS, so distal regulation is invisible to it.
And everything here is computational. These are prioritized candidates for laboratory work, which is what the title says and what the list is for.