Research Highlights

Expression Quantitative Trait Loci Detected in Cell Lines are often Present in Primary Tissues

Evaluation of: Bullaughey K, Chavarria CI, Coop G, Gilad Y: Expression quantitative trait loci detected in cell lines are often present in primary tissues. Hum. Mol. Genet. 18, 4296–4303 (2009).

DNA sequence polymorphisms that influence mRNA expression represent a substantial portion, if not the vast majority, of functional genetic variation that impacts cellular phenotypes and contributes to disease pathogenesis. However, unlike amino acid-altering variants that are easily recognized within the coding sequence, regulatory variants have been more difficult to identify. Expression quantitative trait loci (eQTL) mapping studies represent one of several genome-wide approaches for the identification of such variation. In such studies, DNA sequence variants (typically genotyped using high-density oligonucleotide arrays) are screened for genetic association with measures of mRNA transcript abundances, as measured by expression microarray, where these measures of gene expression are considered as distinct continuous phenotypes (reviewed in [1]). Such studies are often conducted in model organisms and in human cell types and have confirmed the feasibility of the eQTL approach. Although many of the initial studies performed in humans used Ebstein–Barr virus transformed lymphoblastoid cell lines (LCLs) characterized through the International HapMap Project [2,3], studies across a broad range of cell types and tissues have been reported, and data on thousands of candidate regulatory variants are now available for further characterization and correlation with clinical traits.

One ongoing controversy is in regard to the reproducibility of identified eQTL across tissues. Given the distinct patterns of expression across cell types, substantial differences in identified eQTL are not surprising. However, a recent report from Dimas and colleagues suggests that the overlap of eQTL between LCLs and primary cell types obtained from the same individuals was less than 20% [4], questioning the generalizability of eQTL results derived from LCLs. Unlike primary cell types, LCLs are far easier to maintain and represent a convenient resource for initial genetic surveys, more detailed functional follow-up, and other uses, including pharmacogenetic studies [5]. To clarify the extent to which eQTL results can be generalized to other tissues, Kevin Bullaughey and colleagues at the University of Chicago (IL, USA) performed a small eQTL survey of LCL-identified variants in five primary tissues: heart (n = 13), kidney (n = 23), livers (n = 18), lung (n = 20) and testis (n = 10) [6]. In light of the small sample size, the investigators selected a set of candidate eQTL a priori for which they had adequate statistical power to detect eQTL in their sample. They first selected 206 eQTL that had been previously observed in the HapMap LCL samples that explained at least 19% of the variation in gene expression; had a minor allele frequency of at least 0.20 in Caucasians; and were observed in all three HapMap populations. They next genotyped 196 of 206 SNPs in their samples and filtered the list to 21 candidates with adequate statistical power (at least 80% in one cell type, as estimated using the study sample genotype data in conjunction with the HapMap expression data). Expression was subsequently measured by real-time PCR for these 21 candidates across all tissue samples, and linear regression was used to test for eQTL association. eQTLs were detected in at least one primary tissue for 11 of the genes (52%; p-value = 0.003), while five genes had significant eQTL in more than one primary tissue. With a higher pretest probability for association (given that these loci had been previously identified in LCLs), using a less stringent p-value of 0.10, evidence for significant results were observed for 67% of the LCL-identified eQTL in at least one primary tissue. Thus, in contrast to the prior report by Dimas and colleagues, these results reported by Bullaughey et al. suggest that a more substantial proportion of eQTL identified in LCL can be reliably identified in other cell types. Owing to the methods for candidate eQTL selection in this study, it is possible that the observed rates of reproducibility are overestimated, as only candidates with both high frequency and high effect size (those most likely to be robust to tissue-specific effects) were studied. Final resolution of this issue will require comparisons of less common variants of more modest effect in considerably larger samples.

Financial & competing interests disclosure

The author has no relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript. This includes employment, consultancies, honoraria, stock ownership or options, expert testimony, grants or patents received or pending, or royalties.

No writing assistance was utilized in the production of this manuscript.

Expression Quantitative Trait Loci are Highly Sensitive to Cellular Differentiation State

Evaluation of: Gerrits A, Li Y, Tesson BM et al.: Expression quantitative trait loci are highly sensitive to cellular differentiation state. PLoS Genet. 5, E1000692 (2009).

Cell differentiation from pluripotent stem cells to committed lineage is achieved through sequential activation of specific gene networks in a tightly regulated manner. It is therefore possible that specific genetic variants may have important regulatory function during one stage of cell differentiation but not another. Regulatory variants that are active during early stages of differentiation may profoundly influence cellular (and, subsequently, clinical) phenotype. However, because most eQTL studies to identify regulatory variants have been performed in fully differentiated cell types, the existence and overall importance of such variants are, to date, unknown. Gerrits and colleagues from the University of Groningen, the Netherlands attempted to address this issue by evaluating the stability of eQTL signals across closely related yet distinct bone marrow-derived hematopoetic cell types (stem cells, progenitor cells and committed myeloid and erythroid cell types) using 25 female inbred mice from the BXD recombinant panel [1]. Using multivariable models to simultaneously test for both main and cell-type-specific eQTL effects, a total of 1648 eQTL were identified at a false-discovery rate of approximately 0.02. Of these, in only a minority of cases (22%) was the eQTL detected across all four cell types (so-called ‘static eQTL‘). Of the remaining ‘dynamic eQTL‘, 78.9% were observed in only one cell type. The relative contribution of local (defined as being within 10 Mb) and distal regulatory variants appeared to be cell-type dependent as well, as the proportion of total eQTL that was local ranged from 71% in progenitor cells to 24.7% in erythroid cells.

The picture that emerges from these data is that the annotation of regulatory sequence variation may turn out to be far more challenging then perhaps first anticipated. At first glance, the observed evidence of an impressive temporal effect (even among such closely related stages of cell development) suggests that it will be difficult, if not impossible, to dismiss any variant as a regulatory candidate without performing eQTL mapping in all cell types at all stages of differentiation. While this is likely to be true in some instances, careful inspection of the results as presented would suggest that the extent of the temporal heterogeneity may have been substantially overestimated owing to a failure on the part of the investigators to consider the impact of cell-specific gene expression (which is variable across transcripts) on the statistical power to detect eQTL. In any given test of genetic association, statistical power is determined by sample size (fixed at 25 mice in this study), allele frequency (which is variable across markers, but fixed for any given marker across cell type and transcript), and the variance in transcript abundance. This last attribute is highly variable across transcripts and also across cell types. For transcripts whose expression is widely variable across cell types, genetic associations will be much easier to detect in cell types where expression is high (and the coefficient of variance is low), as compared with cell types where the range of expression is lower and the observed variance approaches the limits of the array platform. Thus, even for genetic associations that are truly static across both cell types, a larger sample size would be required to detect the association in the latter cell type. This dynamic is very possibly at play here, as Gerrits and colleagues remark: “in the majority of cases the genes with a cell-type-specific eQTL were also most highly expressed in that particular cell type”. Further supporting this notion, upon review of the supplementary data, one finds that transcripts with evidence of dynamic eQTL in more than one cell type (representing 22% of all dynamic eQTL) had significantly higher means (and higher mean strain differences) of transcript abundance as compared with those transcripts with eQTL restricted to only one cell type (p < 10^-10 on two-sided t-tests). Therefore, it is unclear from these data whether the observed differences are due to true patterns of temporal specificity of eQTL function (as argued by Gerrits), or whether they are largely owing to higher false-negative rates of eQTL detection. Additional analyses of this dataset (for instance, using Bayesian approaches to identify eQTL in more weakly expressing cell types by weighting on the presence of an eQTL in the strongly expressing cell types) and in other datasets will help to resolve this important issue.

Population Genomics in a Disease Targeted Primary Cell Model

Evaluation of: Grundberg E, Kwan T, Ge B et al.: Population genomics in a disease targeted primary cell model. Genome Res. 19, 1942–1952 (2009).

Genome-wide association studies (GWAS) have been published for more than 100 human traits and diseases; however, the more than 500 disease-associated variants identified to date typically explain only a small fraction of the total estimated risk attributable to genetic factors. Among the many explanations for this ‘missing heritability‘ is the focus in each study on the ‘top hits‘ – those variants with the most extreme p-values meeting a nominal level of significance – to limit the false-positive rates that result from multiple testing. Undoubtably, additional variants with modest effects but less impressive statistical significance are present in these datasets, yet methods for prioritizing these for validation and replication from among the potentially tens of thousands of unimportant variants with similar levels of significance remains a daunting challenge. One possible approach is to give priority to those variants that influence gene expression, as these functional variants are more likely to influence clinical phenotype. Indeed, such information has been used successfully to identify disease-associated traits. [1,2] These studies (and similar) provide encouraging motivation for genome-wide efforts aimed at mapping regulatory variants, and results from a variety of eQTL mapping studies are now available to assist in the interpretation of GWAS studies. These studies have been performed using immortalized cell lines and in a variety of primary cell types and tissues. In light of recent observations that regulatory variants often demonstrate cell-specific patterns of association, one pressing issue regarding the utility of such data is whether or not the choice of cell type for the regulatory mapping studies greatly impacts subsequent ability to detect specific disease-associated variation.

In the November 2009 issue of Genetic Research, Elin Grundberg and colleagues from McGill University (QC, Canada) address this issue of tissue specificity by comparing the predictive power to identify regulatory variants associated with bone mineral density (BMD) of two separate eQTL studies [3]: a primary human osteoblasts dataset (obtained from 94 individuals, 44.6% female, who underwent total hip replacement surgery), and an immortalized LCLs, (obtained from previously published data on 60 unrelated individuals of Western European ancestry from the HapMap project [Stranger et al.] [4]). The sets of cis-acting eQTL from each analysis were intersected with the 1000 most significant SNPs associated with BMD from a recently published GWAS study [5], and fold enrichment for eQTL SNPs within the BMD data was determined by comparing results with those from 10 reordered datasets. Grundberg and colleagues noted that among the most significant 0.5% of HOb eQTLs there was a threefold enrichment of SNPs with a BMD-association p-value of less than 0.001 (nine observed overlaps in the real data compared with three observed by permutation). In contrast, no overlap was noted between the most significant LCL SNP and the top-associated BMD SNP. Sequential BMD-association testing of these SNPs resulted in replication of the association of one variant (rs1885987) with hip BMD in two independent male cohorts (p = 4.7 × 10^-3 and 9.3 × 10^-4). A nonsignificant trend of association with BMD was also noted among females (p = 0.14). Importantly, this newly identified BMD-associated SNP ranked 499 in the original BMD GWAS study, further illustrating the potential utility of eQTL data for disease gene identification. The lack of enrichment for BMD-associated SNPs in the LCL data argues for the use of eQTL data from disease-relevant tissues. However, it is notable that the two eQTL datasets were derived in different cohorts of differing sample size. It is unclear whether or not the results would be similar if LCL data had been generated in the same 94 subjects as the primary human osteoblasts data. Nonetheless, these results further support the use of primary cell types from disease-relevant tissues for eQTL/disease-association mapping studies. Given that more than one cell type is typically implicated for most complex diseases (some examples include islet cells, hepatocytes and adipose in Type II diabetes; immune cells, airway epithelium and airway smooth muscle in asthma; and both osteoblasts and osteoclasts in osteoporosis), integration of results across multiple relevant datasets will be required to maximize the detection of disease-susceptibility variants.