Quantitative Trait Locus (QTL)

QTL mapping relies on linear regression or mixed linear models to correlate genotype with phenotype across a population. For each genetic variant, the model tests whether allelic dosage significantly predicts the quantitative trait value.

• Null hypothesis: No association between genotype and phenotype
• Significance threshold: Corrected for multiple testing using Bonferroni or Benjamini-Hochberg procedures
• Effect size: Typically reported as beta coefficient or variance explained (R²)

In molecular QTL studies, the phenotype is a continuous measurement such as ChIP-seq read depth at a peak or ATAC-seq fragment count at a regulatory element.

QTLs are classified by the molecular phenotype they influence, each revealing distinct layers of regulatory biology:

• eQTL (expression QTL): Variants associated with gene transcript abundance, measured via RNA-seq
• bQTL (binding QTL): Variants affecting transcription factor binding intensity, measured via ChIP-seq
• caQTL (chromatin accessibility QTL): Variants altering chromatin openness, measured via ATAC-seq or DNase-seq
• hQTL (histone QTL): Variants associated with histone modification levels
• meQTL (methylation QTL): Variants linked to DNA methylation status at CpG sites

Each subtype captures a different regulatory mechanism, and integrative analysis across subtypes reveals causal chains from variant to phenotype.

QTLs are categorized by the genomic distance between the causal variant and the affected feature:

• Cis-QTLs: The variant lies within a defined window (typically ±1 Mb) of the target gene or regulatory element. These often act through direct disruption of local promoter or enhancer sequences.
• Trans-QTLs: The variant is located far from the affected feature, often on a different chromosome. These typically act through altered expression or activity of a diffusible regulatory factor such as a transcription factor.

Cis-QTLs generally exhibit larger effect sizes and are more replicable, while trans-QTLs require larger sample sizes for detection due to the multiple-testing burden of genome-wide searches.

In heterozygous individuals, QTL analysis can be refined to allele-specific measurements, providing an internal control that eliminates trans-acting and environmental confounders:

• Allele-specific binding (ASB): Compares ChIP-seq read counts mapping to the maternal vs. paternal allele at heterozygous SNPs within binding sites
• Allele-specific expression (ASE): Compares RNA-seq read counts per allele at transcribed heterozygous variants

This approach directly demonstrates that a regulatory variant alters binding or expression in cis, as both alleles reside in the same cellular environment. Statistical significance is assessed using binomial tests against the null expectation of 50:50 allelic balance.

Neural network models such as DeepSEA, Basenji, and Enformer predict molecular phenotypes directly from DNA sequence, enabling in silico dissection of QTL mechanisms:

• In silico mutagenesis: Systematically introduce each possible nucleotide at a variant position and measure the predicted change in binding or accessibility
• Effect size prediction: The difference between reference and alternate allele predictions quantifies the functional impact of a regulatory variant
• Variant prioritization: Rank candidate causal variants within a QTL by their predicted effect magnitude

These models learn regulatory grammar from thousands of genomic profiles, capturing nonlinear interactions that simple position weight matrix scanning misses.

A QTL interval often contains many correlated variants due to linkage disequilibrium (LD). Fine-mapping disentangles the causal variant from nearby passengers:

• Statistical fine-mapping: Bayesian methods such as CAVIAR, FINEMAP, or SuSiE compute posterior inclusion probabilities for each variant given the association signal and LD structure
• Functional fine-mapping: Overlay epigenomic annotations (chromatin state, TF binding, conservation) to prioritize variants in active regulatory elements
• Credible set: The minimal set of variants that collectively captures, for example, 95% of the posterior probability of containing the causal variant

Integrating statistical evidence with deep learning-based functional predictions dramatically narrows the search space for experimental validation.

A specific class of QTL where the quantitative trait is gene expression level. eQTLs link genetic variants to transcript abundance, revealing how non-coding polymorphisms regulate gene activity.

• cis-eQTLs: Variants near the target gene (within ~1 Mb) affecting its own expression
• trans-eQTLs: Distal variants, often in transcription factors, affecting expression of genes on other chromosomes
• Used to annotate GWAS hits with regulatory function

The phenomenon where a heterozygous variant causes differential transcription factor binding between maternal and paternal alleles. ASB provides direct functional evidence that a QTL-associated SNP alters protein-DNA interaction.

• Detected by quantifying allelic imbalance in ChIP-seq reads
• Resolves causality: the variant itself changes binding, not a linked proxy
• Critical for validating dsQTLs (DNase I sensitivity QTLs)

A computational perturbation method that systematically introduces virtual nucleotide substitutions into a DNA sequence and measures the resulting change in a neural network's binding prediction.

• Quantifies the predicted effect of every possible single-nucleotide variant
• Identifies causal regulatory variants within QTL credible sets
• Implemented in models like BPNet and Enformer for saturation mutagenesis

A QTL where the quantitative trait is chromatin accessibility measured by ATAC-seq or DNase-seq. caQTLs identify genetic variants that alter the physical openness of regulatory DNA.

• Variants in transcription factor binding motifs that change nucleosome positioning
• Overlap with eQTLs reveals regulatory cascades
• Provides mechanistic link between non-coding GWAS variants and gene regulation

A statistical process that refines a QTL signal to identify the most probable causal variant(s) within a linkage disequilibrium block.

• Uses Bayesian methods (e.g., CAVIAR, FINEMAP) to compute posterior inclusion probabilities
• Incorporates functional annotations like DeepSEA or Enformer predictions as priors
• Reduces the variant set for experimental validation via MPRA or CRISPR editing

A method that integrates eQTL data with GWAS to identify genes whose genetically predicted expression is associated with complex traits.

• Trains a predictor of gene expression from local SNPs using reference panels (e.g., GTEx)
• Imputes expression into large GWAS cohorts where RNA data is unavailable
• Prioritizes effector genes over merely associated loci