**a**) Experimental design. Starting with 393 individuals that were successfully gene expression profiled in all 3 tissues (in the figure legends A, B, C denote OA, SA and liver respectively) and genotyped, we selected 20 sub samples with sizes from 50 to 393 equally spaced on the logarithmic scale such that

. For each sub sample we identified cis eQTLs in each tissue at 10% FDR.

**b**) Number of cis eQTLs

*q* as function of sample size

*n*. We plot

as a function of

*ln(n)* and show that total number of cis eQTLs identified in each tissue is linear in this range with

with

*R*^{2}>0.9 (see for linear regression coefficients). Looking at cis eQTLs across the 3 tissues we observ a tendency to plateau for cis eQTLs that are nonoverlapping (Only A, Only B and Only C) or partially overlapping in a pair of tissue (AB, BC and CA). No such behavior is observed for cis eQTLs common to all three tissues (ABC).

**c**) Mean genomic distance

measured in Mb between the eSNP and the transcript's position. Common cis eQTLs (ABC.A, ABC.B and ABC.C) have

significantly smaller (Mann–Whitney

*U* test p values ∼0) than any

computed for the nonoverlapping cis eQTLs (Only A, Only B, and Only C).

**d**) Mean association strength

. Overlapping cis eQTLs have on average a stronger association than any other sub set of cis eQTLs. Unique cis eQTLs (Only A, Only B, and Only C) are weakly associated when compared to overlapping cis eQTLs (ABC).

## PubMed Commons