Interpreting Functional Impact of Genetic Variations by Network QTL for Genotype–Phenotype Association Study

Kai Yuan, Tao Zeng, Luonan ChenFrontiers in Cell and Developmental Biology 2022 · DOI

TL;DR

This paper introduces "network QTL" (nQTL), a novel framework that moves beyond traditional genotype-to-gene expression (eQTL) analysis. By considering genotype-phenotype associations at the network level (genotype → network → phenotype), nQTL can uncover more comprehensive functional impacts of genetic variations, leading to better disease subtyping, prognosis, and drug response prediction.

What Problem Does This Paper Address?

The post-genome era faces a significant challenge: understanding how genetic variations translate into observable traits (phenotypes), especially for complex diseases. While Genome-Wide Association Studies (GWAS) identify potential genetic loci, and Expression Quantitative Trait Loci (eQTL) link genetic variants to gene expression, these methods often overlook the systemic role of genes and their interactions. Conventional eQTLs typically model genotype → expression → phenotype, treating genes in isolation. This approach misses crucial network-level effects where multiple genes and their interactions, influenced by a single genetic variant, collectively shape a phenotype. The authors aim to bridge this gap by developing a method that captures these network-level genotype-phenotype associations.

Methods

The core of this paper is the network QTL (nQTL) framework. It operates in several stages:

  1. Single-Sample Network (SSN) Construction: Transcriptome data from individual samples is transformed into gene-pair correlation data using the SSN theory. This captures sample-specific network interactions, a key departure from population-level analyses.
  2. Network Trait Identification: The MatrixEQTL tool is adapted to identify "edge traits" – significant associations between Single Nucleotide Polymorphisms (SNPs) and gene-pairs (edges in the network) across samples. This is analogous to eQTLs identifying "gene traits" (SNP-gene associations).
  3. Network Signature Screening: "Hot SNPs" (those with many interacting gene-pairs) and "hot gene-pairs" are used to filter and identify significant network signatures. These signatures are essentially clusters of gene-pairs that are strongly associated with genotypes.
  4. Phenotype Association: Finally, the links between these network signatures and targeted phenotypes (e.g., disease status, drug response) are inferred using methods like Partial Least Squares Discriminant Analysis (PLSDA) and Canonical Correlation Analysis (CCA).

This approach allows for a cascade analysis: genotype → network → phenotype.

Datasets

The authors evaluated their nQTL framework using several datasets:

  • Simulated Datasets: Eight sets of synthetic data were generated to assess the framework's performance in identifying both gene traits (eQTL) and edge traits (nQTL) under various parameter settings.
  • Lung Cancer Dataset (GSE28571): Publicly available data from surgically treated lung cancer patients, including matched expression and SNP data, was used to identify genotype-phenotype associations related to lung cancer prognosis and subtypes.
  • Cancer Cell Line Encyclopedia (CCLE) Hematopoietic Data: Data from 947 human cancer cell lines, including gene expression, copy number, and drug sensitivity profiles, was used to study genotype-phenotype associations in hematopoietic and lymphoid tissues, particularly for drug response.
  • Human Peripheral Blood Mononuclear Cell (PBMC) Dataset: Single-cell RNA sequencing (scRNA-seq) data from PBMCs was used to validate findings from bulk expression data and to study healthy immunity phenotypes.
  • Pan-Cancer Datasets (SurvExpress, TCGA): Independent datasets were used for survival analysis to validate the prognostic power of identified network signatures.

Key Insights and Ideas

  • Beyond Gene-Level Associations: The most significant insight is that focusing solely on individual gene expression (eQTL) misses crucial network-level effects. nQTL captures how a single genetic variant can rewire entire gene interaction networks, leading to a more holistic understanding of phenotype determination.
  • Sample-Specific Networks are Key: The use of single-sample networks (SSNs) is critical. It allows the framework to capture individual variations in gene interactions, which are often masked in population-averaged analyses. This is particularly important for understanding complex diseases where individual responses can vary significantly.
  • Network Signatures as Biomarkers: The identified "network signatures" (clusters of interacting gene-pairs) show strong potential as biomarkers for disease subtyping, prognosis, and even predicting drug response. They offer a more robust and biologically relevant representation than individual genes.
  • Unsupervised Discovery: The nQTL framework can identify associations in an unsupervised manner, meaning it doesn't require pre-defined hypotheses about specific genes or pathways, making it powerful for exploring novel biological mechanisms.

Key Results

  • Synthetic Data Performance: On synthetic datasets with preset edge traits, nQTL analysis demonstrated significantly higher AUC values compared to eQTL analysis, indicating its superior ability to detect network-level associations. For datasets with preset expression traits, nQTL performance was comparable to eQTL.
  • Lung Cancer Analysis: nQTL identified more significant discoveries on edge traits than eQTL. Network signatures derived from nQTL analysis showed significant phenotype discrimination for lung cancer subtypes and provided satisfactory discrimination power in survival analysis, outperforming conventional methods like WGCNA in some aspects.
  • CCLE Hematopoietic Data: nQTL analysis effectively identified network signatures associated with lymphoid cancer and leukemia, and these signatures showed significant phenotype discrimination and survival prediction capabilities. The study also revealed associations between network signatures and drug-target genes.
  • PBMC Data: nQTL analysis successfully identified network signatures related to immunity, viruses, and antigens. These signatures showed significant relevance to immune factors and inflammatory factors, and their genes demonstrated significant survival analysis across multiple cancer datasets.
  • Validation with Single-Cell Data: The nQTL framework's findings from bulk expression data were validated using single-cell RNA-seq data, demonstrating its ability to detect associations without requiring individual-specific single-cell data.

Limitations and Future Work

  • Data Requirements: The nQTL framework, particularly the SSN construction, relies on having sufficient gene expression data for individual samples. While it can work with bulk RNA-seq, the ideal scenario for capturing fine-grained network dynamics might involve single-cell data.
  • Causality Inference: While nQTL identifies associations, inferring direct causal relationships within the genotype-network-phenotype cascade remains a challenge and an area for future research.
  • Integration with GWAS: The authors suggest that integrating nQTL with GWAS summary data could be a fruitful avenue for future research, potentially improving the functional interpretation of GWAS-identified loci.
  • Network Reconstruction Methods: The paper acknowledges that the choice of sample-specific network construction method can influence results. Future work could explore the integration of other advanced network inference and module decomposition approaches to further enhance nQTL's power and robustness.
  • MAF Filter: For future studies focusing on specific variants (like GWAS outcomes), incorporating a Minor Allele Frequency (MAF) filter could refine genotype analysis.
  • Quality Control and Significance Correction: Further development in quality control and significance correction strategies for nQTL analysis is recommended.

Summary updated

Abstract

An enormous challenge in the post-genome era is to annotate and resolve the consequences of genetic variation on diverse phenotypes. The genome-wide association study (GWAS) is a well-known method to identify potential genetic loci for complex traits from huge genetic variations, following which it is crucial to identify expression quantitative trait loci (eQTL). However, the conventional eQTL methods usually disregard the systematical role of single-nucleotide polymorphisms (SNPs) or genes, thereby overlooking many network-associated phenotypic determinates. Such a problem motivates us to recognize the network-based quantitative trait loci (QTL), i.e., network QTL (nQTL), which is to detect the cascade association as genotype → network → phenotype rather than conventional genotype → expression → phenotype in eQTL. Specifically, we develop the nQTL framework on the theory and approach of single-sample networks, which can identify not only network traits (e.g., the gene subnetwork associated with genotype) for analyzing complex biological processes but also network signatures (e.g., the interactive gene biomarker candidates screened from network traits) for characterizing targeted phenotype and corresponding subtypes. Our results show that the nQTL framework can efficiently capture associations between SNPs and network traits (i.e., edge traits) in various simulated data scenarios, compared with traditional eQTL methods. Furthermore, we have carried out nQTL analysis on diverse biological and biomedical datasets. Our analysis is effective in detecting network traits for various biological problems and can discover many network signatures for discriminating phenotypes, which can help interpret the influence of nQTL on disease subtyping, disease prognosis, drug response, and pathogen factor association. Particularly, in contrast to the conventional approaches, the nQTL framework could also identify many network traits from human bulk expression data, validated by matched single-cell RNA-seq data in an independent or unsupervised manner. All these results strongly support that nQTL and its detection framework can simultaneously explore the global genotype-network-phenotype associations and the underlying network traits or network signatures with functional impact and importance.
BibTeX
@inproceedings{yuan2022interpreting,
  author = {Kai Yuan and Tao Zeng and Luonan Chen},
  title = {{Interpreting Functional Impact of Genetic Variations by Network QTL for Genotype–Phenotype Association Study}},
  booktitle = {Frontiers in Cell and Developmental Biology},
  year = {2022},
  doi = {10.3389/fcell.2021.720321},
}
Figure