Molecular QTL Overlap as Functional Support

Why Molecular QTLs Matter at a Disease Locus

A disease-associated locus can contain many plausible variants, and annotation may suggest several possible target genes or molecular processes. A quantitative trait locus, or QTL, adds a useful kind of evidence: it identifies a genetic association with a measured molecular phenotype. For example, an expression QTL associates genotype with gene expression, while a protein QTL associates genotype with measured protein abundance.

A QTL association that falls at or near a disease credible variant can support the idea that the locus has a functional effect on a molecular phenotype. But “supports” is not the same as “identifies.” A nearby QTL signal might be caused by a different variant in linkage disequilibrium with a disease-associated variant. Even evidence that the same variant affects both traits does not, by itself, establish that the molecular phenotype mediates the disease effect.

Definition: A molecular QTL is a genetic variant associated with variation in a measured molecular phenotype. An eQTL is associated with gene expression; an sQTL is associated with a splicing phenotype, such as the relative use of a transcript or splice junction; and a pQTL is associated with measured protein abundance. A QTL association is evidence of a genotype–phenotype relationship under the study’s analysis; it is not, on its own, proof of a causal mechanism.

QTL studies commonly test many variants and molecular phenotypes. Their reported associations therefore need to be read in light of the study’s testing procedure, significance threshold, sample size, phenotype definition, and tissue or cell type. For instance, an eQTL measured in one tissue supports a claim about expression in that tissue more directly than a claim about expression in every tissue.

What “Overlap” Means

A credible set is a set of variants retained by a fine-mapping analysis as plausible candidates for explaining a locus-level association. A variant may have a posterior inclusion probability, or PIP, from that analysis. QTL evidence can be compared with a credible set at several levels, which should not be treated as interchangeable.

Exact variant overlap: a variant reported in a QTL analysis is also in the disease credible set. The variant identifiers, alleles, and genome build must be compatible; matching positions alone can be misleading if variant representation or reference alleles differ.
Regional overlap: the QTL association and the disease association occur in the same genomic region, but the reported variants may differ. This is evidence that both signals occur at the locus, not necessarily that they share a causal variant.
Proximity: a QTL-associated variant lies near a credible variant. Distance can be useful for describing the evidence, but no universal distance threshold makes two variants functionally equivalent.

A QTL lead variant is usually the variant with the strongest reported association in a tested region. It is not automatically the causal QTL variant: another variant may be correlated with it and may better explain the signal. Likewise, membership in a disease credible set indicates uncertainty over disease-associated variants, not proof that any member changes a molecular phenotype.

The strength of the comparison depends partly on whether the QTL signal itself has been fine-mapped. If only a lead QTL variant is available, an exact match to a disease credible variant is useful evidence, but alternative linked QTL variants may remain plausible. If both traits have been fine-mapped, methods that assess whether the signals share a causal variant can address this uncertainty more directly. Such a shared-signal result supports a common variant explanation; it still does not establish that the molecular phenotype is the route through which the variant affects disease.

Reading a QTL Association

For a simple illustration, suppose a quantitative molecular phenotype \(Y_i\) is measured for individual \(i\). Let \(G_i\) be the number of copies of a specified allele, coded 0, 1, or 2, and let \(C_i\) represent covariates such as age, sex, or technical factors. A common association model is linear regression:

Y_i = \beta_0 + \beta_G G_i + C_i^\mathsf{T}\boldsymbol{\gamma} + \varepsilon_i .

Here, \(\beta_G\) is the estimated difference in the molecular phenotype per additional copy of the coded allele, conditional on the included covariates. The interpretation and units depend on how \(Y_i\) was measured or transformed. For example, if expression was standardized, the effect may be in standard-deviation units; if protein abundance was log-transformed, the effect is on that transformed scale.

In this model, a test of no association at the variant is \(H_0:\beta_G=0\). When the estimate and its standard error are approximately normally distributed under the null, a two-sided test uses \(z=\hat{\beta}_G/SE\) and \(p=2[1-\Phi(|z|)]\), where \(\Phi\) is the standard normal cumulative distribution function. This calculation describes the association test, not the probability that the null hypothesis is true. In real QTL analyses, valid inference also depends on appropriate phenotype and genotype handling, covariate adjustment, and standard errors that account for the study design. Multiple testing must be addressed when many variants, genes, or molecular traits are tested.

Worked Example: An eQTL at a Credible Variant

Hypothetical data. A disease credible set includes variant v1 at position 42,100, with PIP 0.62. In a relevant tissue, an eQTL analysis reports the same harmonized variant for gene R. The estimated expression effect is \(\hat{\beta}_G=0.30\) expression standard deviations per copy of the coded allele, with \(SE=0.075\). For this illustration, assume the regression model is suitable, the reported standard error is valid, and the two-sided normal approximation is appropriate. The study has also applied a correction for its multiple QTL tests.

The association statistic is

z=\frac{0.30}{0.075}=4.00.

The two-sided normal-approximation p-value is \(2[1-\Phi(4.00)]\), approximately \(6.33\times 10^{-5}\). The estimated effect is positive: each additional copy of the coded allele is associated with an average increase of 0.30 expression standard deviations in this tissue, conditional on the model covariates. Given the stated multiple-testing procedure, this is evidence for an eQTL association.

The exact variant match makes the QTL association relevant to the disease credible set, and the PIP indicates that v1 is a plausible disease-associated variant. However, the QTL association alone does not show that v1 is the causal QTL variant if the QTL signal has not been fine-mapped. Nor does it show that altered expression of gene R mediates the disease association. The defensible conclusion is locus-level functional support for an expression relationship, not a proven disease mechanism.

Why Nearby Association Signals Can Differ

Linkage disequilibrium (LD) is correlation between alleles at different variants. Because of LD, a variant that is not causal for a molecular phenotype can still show an association: its genotype predicts the genotype at the causal variant. This is one reason that a QTL lead variant near a disease credible variant is not enough to establish that they are the same functional variant.

A short derivation makes the point. Suppose, for illustration, that one standardized genotype \(G_c\) affects a standardized molecular phenotype \(Y\), with \(Y=\beta_cG_c+\varepsilon\), and that the residual has mean zero conditional on the genotypes. A second standardized genotype \(G_j\) is correlated with \(G_c\). The marginal regression coefficient of \(Y\) on \(G_j\) is its covariance with \(Y\), because \(G_j\) has variance 1:

\beta_j^{\mathrm{marginal}} =\operatorname{Cov}(G_j,Y) =\operatorname{Cov}(G_j,\beta_cG_c+\varepsilon) =\beta_c\operatorname{Cov}(G_j,G_c) =r_{jc}\beta_c .

The last equality uses standardized genotypes, for which their covariance is their correlation \(r_{jc}\). Thus, if \(G_c\) has a genuine effect, a correlated \(G_j\) can have a nonzero marginal association even when it has no direct effect in this model. Real data can involve multiple causal variants, imperfect measurement, and other complications, but the simple result is enough to show why regional or lead-variant overlap must be interpreted cautiously.

Worked Example: A Nearby QTL Variant in Linkage Disequilibrium

Hypothetical data. A disease credible set contains variant D at position 80,000. A gene-expression study reports a QTL lead variant Q at position 80,700, and the variants have \(r=0.80\) in the relevant population. Assume, only for this illustration, that Q is the sole causal variant for the expression phenotype, that the standardized-genotype linear model above applies, and that its effect is 0.50 expression standard deviations per standard-deviation increase in genotype.

Under the single-causal-variant model, the marginal coefficient at D is expected to be

\beta_D^{\mathrm{marginal}}=r_{DQ}\beta_Q =0.80(0.50)=0.40.

So D could show an expression association of about 0.40 standard deviations on this scale even though the stipulated direct expression effect belongs to Q. The 700-base-pair distance and the LD correlation make the signals relevant to one another, but they do not demonstrate that D itself changes expression. This calculation is a model-based illustration, not a guarantee that a particular dataset will produce that exact estimate. The conclusion is regional functional support with unresolved variant identity; QTL fine-mapping or a suitable shared-signal analysis would be needed to assess the identity question more directly.

Different QTLs Support Different Molecular Claims

The molecular phenotype matters as much as the presence of a QTL association. An eQTL concerns measured expression, an sQTL concerns a splicing phenotype, and a pQTL concerns a protein measurement. These associations are not interchangeable: an expression association does not necessarily imply a change in protein abundance, and a protein association does not by itself identify whether the cause is transcription, translation, protein stability, or an assay-related effect.

Interpretation should also respect the measurement. Splicing phenotypes may describe relative transcript usage or a junction-level quantity rather than total expression. Protein abundance may be reported on a transformed or assay-specific scale. An effect estimate cannot be compared meaningfully across QTL types without checking those scales and the alleles’ direction of coding. The molecular phenotype’s tissue, cell type, and assay are part of the claim, not optional details.

Worked Example: Interpreting a pQTL Effect and Interval

Hypothetical data. A pQTL study tests a coded allele for association with a protein measured on a log-base-2 abundance scale. The estimated effect is \(\hat{\beta}_G=0.18\) log2 units per additional allele copy, with \(SE=0.05\). Assume the regression assumptions and standard error are adequate, the estimate is approximately normal, and the study’s multiple-testing procedure is accounted for.

The association statistic is \(z=0.18/0.05=3.60\), giving a two-sided normal-approximation p-value of about 0.00032. A 95% normal-approximation confidence interval is

0.18\pm1.96(0.05) =0.18\pm0.098 =[0.082,\,0.278]\text{ log2 units per allele copy}.

The interval describes values compatible with the data under the model and its assumptions; it is not a 95% probability statement about the fixed effect. The positive estimate indicates higher measured protein abundance with each additional copy of the coded allele on the log2 scale. If a disease credible variant is nearby or overlaps this QTL signal, this supports a protein-level association at the locus. It does not, without further evidence, show that the disease variant is the QTL causal variant or that protein abundance mediates disease risk.

How to Judge the Support

A careful interpretation moves from the reliability of the QTL association to the relationship between signals and then to the scope of the biological claim. It should distinguish the observed evidence from the explanation that remains uncertain.

Verify what was measured.
Identify the molecular phenotype, its scale, assay, tissue or cell type, and the tested allele. Confirm that variant coordinates and alleles are comparable to those in the disease credible set.

Check the association evidence.
Read the effect estimate, standard error, and significance result in context, including the study’s multiple-testing procedure, sample size, and analysis model.

Describe the overlap precisely.
State whether the evidence is an exact variant match, a regional overlap, or proximity. If relevant, report LD and whether the QTL signal has been fine-mapped.

Limit the conclusion to what is supported.
Describe the molecular association and its relevance to the locus. Do not infer a shared causal variant or a disease-mediating mechanism from proximity alone.

A tissue-matched QTL association that overlaps a disease credible variant is generally more informative for that tissue-specific molecular hypothesis than an association in an unrelated tissue. Still, a QTL can be absent because the study had limited power, the relevant cell type was not measured, or the phenotype was not captured by the assay. Conversely, a statistically strong QTL need not be relevant to the disease mechanism. Strength of association and biological relevance are related questions, not the same question.

Common Mistakes and a Key Distinction

“The lead QTL variant is near a credible variant, so they are the same causal variant.” Proximity does not establish identity. LD can produce nearby associations from distinct variants; assess the QTL signal and the relationship between signals directly.
“A QTL association proves that expression causes the disease.” It establishes an association with a measured molecular phenotype, under the study model. It does not establish mediation or rule out other explanations.
“No significant QTL means the variant has no molecular effect.” A nonsignificant result can reflect limited power, tissue mismatch, measurement limitations, or multiple-testing correction. It is not proof of no effect.
“An eQTL, sQTL, and pQTL are equivalent evidence.” They concern different measured phenotypes and support different molecular claims. State which phenotype was associated and on what scale.
“The smallest p-value identifies the causal variant.” A lead association is not automatically causal. Correlation among variants can make a noncausal variant’s association appear strong.

Key takeaway: Molecular QTL overlap can support a functional hypothesis at a disease locus, especially when the phenotype, tissue, and variant comparison are well matched. Exact overlap, regional overlap, and proximity are distinct observations. None alone proves that a particular variant changes the molecular phenotype, and a shared molecular association does not by itself prove disease mediation.

Check Your Understanding

For each question, distinguish what the QTL data show from what remains uncertain.

What molecular phenotype does an sQTL concern, and why is that not automatically equivalent to an eQTL?
A QTL lead variant is 500 base pairs from a disease credible variant. What additional information would help assess whether the signals share a causal variant?
In the standardized-genotype model, if \(r=0.70\) and the causal QTL effect is 0.40, what marginal effect is expected at the correlated variant under the stated assumptions?
Why does an exact match between a disease credible variant and a QTL lead variant not, by itself, prove mediation of the disease effect?
A QTL result is nonsignificant in one tissue. Give two reasons, other than a true absence of molecular effect, that could explain this result.