Mean-Variance QTL Mapping Identifies Novel QTL for Circadian Activity and Exploratory Behavior in Mice

Robert W. Corty, Vivek Kumar, Lisa M. Tarantino, Joseph S. Takahashi, William Valdar

TL;DR

This paper introduces "mean-variance QTL mapping," a novel approach that simultaneously models the effects of genetic variants on both the average trait value and its variability. By applying this method to existing mouse genetic datasets, the authors demonstrate its superior ability to detect quantitative trait loci (QTL) that traditional methods miss, particularly those influencing trait variance. This work advocates for the broader adoption of mean-variance QTL mapping as a standard practice in genetic research.

What Problem Does This Paper Address?

Traditional quantitative trait locus (QTL) mapping has been highly successful in identifying genetic variants that influence the average value of a trait. However, this approach typically assumes that the variability (or "noise") around the average is consistent across all individuals, regardless of their genotype. This assumption can lead to missed discoveries, as genetic variants can also influence how much a trait varies between individuals. The paper addresses this limitation by highlighting that genetic analyses often fail to capture the full complexity of genotype-phenotype relationships, particularly when dealing with variance heterogeneity. This heterogeneity can arise from various factors, including genetic background or environmental influences, and can obscure or even mask the effects of certain QTL on the mean trait value.

Methods

The core of this paper's methodological contribution is the application of mean-variance QTL mapping, which is based on the double generalized linear model (DGLM). This model extends the standard linear model (SLM) used in traditional QTL mapping.

Here's how it works:

  • Standard Linear Model (SLM): Models the phenotype ($y_i$) of individual $i$ as having a mean ($m_i$) and a constant residual variance ($\sigma^2$). The mean is typically a function of genetic effects (e.g., additive and dominance) and covariates (e.g., sex).
  • Double Generalized Linear Model (DGLM): Extends the SLM by allowing the residual variance ($\sigma_i^2$) to vary for each individual. It models both the mean ($m_i$) and the logarithm of the variance ($\log(\sigma_i^2)$) as linear predictors, incorporating genetic effects and covariates for both. This allows for the detection of:
  • mQTL: QTL affecting the mean trait value.
  • vQTL: QTL affecting the residual variance of the trait.
  • mvQTL: QTL affecting both the mean and the variance.

The paper also details the statistical framework for determining genomewide significance and controlling the family-wise error rate (FWER) using permutation-based approaches, adapted for the different DGLM tests.

Datasets

The study utilizes two publicly available mouse genetic datasets from the Mouse Phenome Database:

  1. Kumar et al. (2013) dataset: This dataset involves a reduced complexity intercross of C57BL/6J and C57BL/6N mice, focusing on circadian wheel running activity.
  2. Bailey et al. (2008) dataset: This dataset comes from an intercross between C57BL/6J and C58/J mice, examining anxiety-like behaviors, specifically rearing behavior.

The authors provide links to the raw data files and analysis scripts, making their work reproducible.

Key Insights and Ideas

  • Variance Matters: The most significant takeaway is that genetic variation can influence not just the average of a trait but also its variability. Ignoring variance heterogeneity can lead to missed QTL.
  • DGLM as a Unified Framework: The DGLM elegantly integrates the detection of mean-controlling (mQTL), variance-controlling (vQTL), and combined (mvQTL) loci within a single statistical framework, simplifying the interpretation of genetic architecture.
  • Increased Power: Mean-variance QTL mapping, particularly when variance heterogeneity is present, offers increased statistical power to detect QTL compared to traditional methods. This is demonstrated by the detection of novel QTL in both case studies.
  • Reduced Complexity Crosses Benefit: Even in "reduced complexity" crosses (where parental strains are closely related), mean-variance QTL mapping can still uncover novel associations that might be missed by standard approaches.
  • Practical Implications: The authors advocate for mean-variance QTL mapping as a "drop-in replacement" for standard QTL mapping, suggesting it should become a routine part of genetic analyses due to its enhanced capabilities and minimal added complexity when variance is homogeneous.

Key Results

  • Circadian Activity: In the reanalysis of the Kumar et al. data, traditional QTL mapping found no significant QTL for circadian wheel running activity. In contrast, mean-variance QTL mapping identified a novel mean-controlling QTL (mQTL) on chromosome 6. This QTL was significant across all three DGLM tests (mQTL, vQTL, mvQTL), with the mQTL test being the most significant. The authors attribute this discovery to the fact that mice homozygous for the C57BL/6N allele exhibited both higher activity and less residual variance.
  • Rearing Behavior: For the Bailey et al. data, mean-variance QTL mapping identified a novel variance-controlling QTL (vQTL) for rearing behavior on chromosome 2. Traditional QTL mapping did not detect this locus because it is not designed to identify variance-influencing QTL.
  • Power Improvement: Simulations indicated that mean-variance QTL mapping could achieve the same power as traditional mapping with approximately 100 fewer mice in the first case study, highlighting its efficiency.

Limitations and Future Work

  • Phenotype Transformation: The paper acknowledges concerns about spurious vQTL arising from inappropriate phenotype transformations. While they demonstrate that their approach is robust across various transformations (including Box-Cox), they emphasize the need to consider a wide range of transformations and models for accurate interpretation.
  • Covariate Modeling: The authors suggest that the DGLM framework can be extended to model variance covariates beyond just genetic effects, allowing for a more comprehensive understanding of factors influencing phenotypic variability. This is discussed in companion articles.
  • Software Availability: The paper points to companion articles that describe the R package vqtl, which implements mean-variance QTL mapping, making the method accessible to researchers.
  • Broader Application: The authors strongly advocate for the routine use of mean-variance QTL mapping, suggesting it should be a standard practice rather than an optional add-on, especially given the increasing availability of complex biological data.

Summary updated

Abstract

We illustrate, through two case studies, that "mean-variance QTL mapping"-QTL mapping that models effects on the mean and the variance simultaneously-can discover QTL that traditional interval mapping cannot. Mean-variance QTL mapping is based on the double generalized linear model, which extends the standard linear model used in interval mapping by incorporating not only a set of genetic and covariate effects for mean but also set of such effects for the residual variance. Its potential for use in QTL mapping has been described previously, but it remains underutilized, with certain key advantages undemonstrated until now. In the first case study, a reduced complexity intercross of C57BL/6J and C57BL/6N mice examining circadian behavior, our reanalysis detected a mean-controlling QTL for circadian wheel running activity that interval mapping did not; mean-variance QTL mapping was more powerful than interval mapping at the QTL because it accounted for the fact that mice homozygous for the C57BL/6N allele had less residual variance than other mice. In the second case study, an intercross between C57BL/6J and C58/J mice examining anxiety-like behaviors, our reanalysis detected a variance-controlling QTL for rearing behavior; interval mapping did not identify this QTL because it does not target variance QTL. We believe that the results of these reanalyses, which in other respects largely replicated the original findings, support the use of mean-variance QTL mapping as standard practice.
BibTeX
@inproceedings{corty2018mean,
  author = {Robert W. Corty and Vivek Kumar and Lisa M. Tarantino and Joseph S. Takahashi and William Valdar},
  title = {{Mean-Variance QTL Mapping Identifies Novel QTL for Circadian Activity and Exploratory Behavior in Mice}},
  booktitle = {G3},
  year = {2018},
  doi = {10.1534/g3.118.200194},
}
Figure