Author: Charleen D. Adams
| Var1 | Freq |
|---|---|
| bivalent | 6 |
| normal | 990 |
| rs10746563 | rs11648412 | rs28461437 | rs4027362 | rs634643 | rs640000 | |
|---|---|---|---|---|---|---|
| MRPL28 | 0 | 3 | 2 | 0 | 0 | 0 |
| MRPL32 | 0 | 0 | 0 | 0 | 2 | 0 |
| MRPL41 | 15 | 0 | 0 | 0 | 0 | 0 |
| MRPS34 | 0 | 0 | 0 | 7 | 0 | 0 |
| RPS16 | 0 | 0 | 0 | 0 | 0 | 7 |
| decrease | increase | |
|---|---|---|
| MRPL28 | 3 | 2 |
| MRPL32 | 1 | 1 |
| MRPL41 | 14 | 1 |
| MRPS34 | 4 | 3 |
| RPS16 | 1 | 6 |
SNPs that solely either increase or decrease expression across tissues account for more than 99% of the RP eQTLs.
There are a small number of SNPs (6), though, that are ‘bivalent’: they influence expression in both directions depending on the tissue.
Are SNPs that influence expression across tissues (in the same direction) likelier to be either causal or tagging the same underlying causal entity than SNPs that bidirectionally influence expression? I was thinking this might be so. When it comes to looking at disease associations, the SNPs that comport across tissues might be more likely to influence similar diseases.
Even though our cis-eQTLs were selected by GTEx to be within a +/- 1Mb window around transcription start sites (TSSs), only 70 of our 996 eQTLs (0.07%) are in CpG islands, per the SNPnexus annotation. 2 of the 6 ‘bivalents’ (rs4027362 and rs640000) are in CpG islands.
I realize there may be many other explanations, but SNPs that occur within CpG islands may increase methylation if a risk allele leads to a CpG dinucleotide that wouldn’t otherwise be there (thus maybe decreasing expression), or lead to less methylation if it eliminates a CpG dinucleotide (maybe increasing expression) [1]. Since methylation is tissue-specific, I wondered if maybe this partially explained some SNPs being bivalent across tissues. (But an easier explanation might be that the bivalent SNPs are false-positives in the GTex data.)