Week 7 - L2 - Comprehensive Study Notes: Genome-Wide Association Studies (GWAS)

Principles and Framework of Genome-Wide Association Studies

• Definition: A Genome-Wide Association Study (GWAS) is a hypothesis-free analytical method that tests single nucleotide polymorphisms (SNPs) across the entire genome to identify statistical associations with specific diseases or quantitative traits.

• Study Population Designs:

  • Binary Case-Control Cohorts: Populations divided into individuals presenting a specific clinical condition (cases) versus unaffected individuals (controls).

  • Continuous Trait Cohorts: Sampling whole populations to evaluate quantitative spectrum traits (e.g., measuring height variations across shorter to taller individuals, or measuring quantitative trait concentration levels).

• Primary Objective: Associate genetic trait variation or disease predisposition with specific SNPs across the genome to localize genomic regions that contribute to the underlying trait or condition.

• Genetic Architecture & Sample Size Requirements:

  • Most complex traits and common diseases are polygenic, caused by a large number of distinct genes each contributing a small individual effect size.

  • Because effect sizes at individual loci are small, GWAS requires extremely large sample sizes to achieve adequate statistical power.

• Underlying Biological Principle: GWAS relies on the concept of Linkage Disequilibrium (LD). LD allows fine mapping of causative genetic loci by testing tag SNPs that co-segregate with non-genotyped causative variants across the genome.

Data Collection Strategies and Global Biobanks

• Primary Data Requirements: Successful GWAS implementation requires obtaining both DNA samples and comprehensive phenotypic data from every individual in the study cohort.

• Sample Acquisition Strategies:

  • Targeted Recruitment: Direct recruitment of patients specifically presenting with a target disease or phenotypic state.

  • Population-Based Biobanks: Utilization of global biobanks established by health organizations, which store pre-collected DNA linked to deep phenotypic profile datasets.

• Population Bias Considerations:

  • Data selection must actively avoid population bias to ensure representative statistical findings.

  • UK Biobank Example:

    • Contains one of the largest human genome and genotype data collections in the world.

    • Recruits participants strictly on a volunteer basis without targeted population sampling.

    • Demographic Bias: UK Biobank volunteers are systematically wealthier, healthier, and more educated than the broader UK population. This inherent volunteer bias requires explicit analytical accounting during downstream study designs.

• Global Biobanking Statistics & Diversity Deficit:

  • As of 2022, global biobanks collectively contained over $2{,}200{,}000$ recorded genotypes, with continuous sample expansion.

  • Sourcing frameworks combine population-based screening, hospital encounter collection models (capturing patient DNA and phenotypic clinical data during medical visits), or hybrid models.

  • Ancestry Bias: The majority of international biobank genotypes and matched phenotypic data are derived from individuals of predominantly European ancestry, establishing a major systemic ancestral bias in genetic research.

Genotyping Methodologies: Microarrays vs. Next-Generation Sequencing

• SNP Microarrays:

  • Mechanism: Directly assays pre-selected tag SNPs (up to $900{,}000$ tag SNPs per array) that serve as proxy markers to infer surrounding unobserved variants in an individual's haplotype block.

  • Advantages: Highly inexpensive per sample compared to sequencing; significantly simplifies downstream bioinformatics and analytical pipelines.

  • Disadvantages:

    • Relies strictly on prior population haplotype knowledge.

    • Incapable of identifying ultra-rare variants that are absent from established population reference haplotypes.

    • Physical limits on the number of probe SNPs on an array bound the fine-mapping resolution.

• Next-Generation Sequencing (NGS) and Whole Genome Sequencing (WGS):

  • Mechanism: Employs target capture-based sequencing or full Whole Genome Sequencing (WGS) to directly sequence DNA across genomic regions or the complete genome.

  • Advantages:

    • Theoretically captures every SNP across an individual's genome.

    • Enables detection of ultra-rare and novel variants uncharacterized in standard population reference panels.

    • Provides maximum genomic data volume, enabling extremely fine mapping to pinpoint exact causative variants.

    • Allows cost-reduction strategies such as sample multiplexing when targeting sub-genomic capture regions.

  • Disadvantages:

    • Comparatively expensive per sample relative to microarrays for equivalent SNP counts.

    • Generates immense data volumes requiring substantial data storage infrastructure and complex bioinformatic processing pipelines.

Genotype Quality Control and Ancestry Stratification

• Primary Rationale for Quality Control (QC):

  • Assay Error Removal: Identifies and filters out inaccurate or miscalled genotypes resulting from technical laboratory or calling artifacts.

  • Low-Frequency Filtering: Sets statistical thresholds to exclude ultra-rare genotypes whose low frequency prevents meaningful statistical evaluation and causes false associations.

• Population Stratification Control:

  • Haplotype structures are shared among individuals with common ethnic backgrounds and ancestral lineages.

  • Systematic differences in ancestry proportions between cases and controls introduce severe confounding (population stratification).

  • Quality control procedures require explicit de-stratification of genotypic datasets based on individual genetic ancestry to prevent false-positive associations driven by population structure.

Haplotype Phasing Mechanics and Theoretical Framework

• Core Technical Limitations of Raw Genotypic Data:

  • Unphased Diploid Genotypes: Microarray and sequencing outputs yield diploid genotypes at individual loci without indicating which specific alleles reside on the same physical chromosome.

  • Incomplete Direct Genotyping: Assays target specific tag SNPs, requiring bioinformatic inference against population reference databases to deduce non-genotyped loci.

• Definition of Phasing: The process of determining the specific spatial cis/trans linear arrangement of alleles across multiple loci on individual physical chromosomes to reconstruct true haplotypes.

• Concrete Cis vs. Trans Configuration Example:

  • Consider three loci on a chromosome: Locus 1 ($A/A$ homozygote), Locus 2 ($T/C$ heterozygote), Locus 3 ($C/G$ heterozygote).

  • Cis Configuration: The $T$ allele at Locus 2 and $C$ allele at Locus 3 reside on the same physical chromosome.

    • Resulting Haplotypes: Haplotype 1 = $A-T-C$; Haplotype 2 = $A-C-G$.

  • Trans Configuration: The $T$ allele at Locus 2 and $C$ allele at Locus 3 reside on opposite homologous chromosomes.

    • Resulting Haplotypes: Haplotype 3 = $A-T-G$; Haplotype 4 = $A-C-C$.

  • Significance: Correct phase identification is essential for assigning accurate haplotypes prior to downstream genotype imputation.

• Methods for Phase Determination:

  • Pedigree / Parental Genotype Analysis: Evaluating parent-offspring trios directly resolves inherited parental chromosome allele arrangements; highly accurate but rarely available in large GWAS cohorts.

  • Long-Read Sequencing: Long individual sequencing reads span across multiple tag SNPs simultaneously, directly reading physical linkage along single DNA molecules (essential for rare variants lacking reference panels).

  • Statistical Phasing via Population Reference Panels:

    • Predicts the most probable physical phase by matching unphased genotypes against pre-characterized population haplotype frequencies.

    • Employs Hardy-Weinberg Equilibrium ($2\,p\,q$) calculations and Bayesian probability updating.

    • Mathematical Example Walkthrough:

      • Unphased Genotype: Locus 1 ($A/G$), Locus 2 ($T/C$), Locus 3 ($G/C$).

      • Reference Population Haplotype Frequencies: $A-T-C = 47\%$ ($0.47$), $G-C-G = 10\%$ ($0.10$), $A-T-G = 0\%$ ($0.00$), $A-C-C = 2\%$ ($0.02$), $G-T-G = 1\%$ ($0.01$), $A-C-G = 17\%$ ($0.17$), $G-T-C = 0\%$ ($0.00$).

      • Joint Frequency Calculations ($2 \times f(H_1) \times f(H_2)$):

        • Phase Pair 1 ($A-T-C$ / $G-C-G$): $2 \times 0.47 \times 0.10 = 0.094$ (yielding $4.6\%$ expected population frequency).

        • Phase Pair 2 ($A-C-C$ / $G-T-G$): $2 \times 0.02 \times 0.01 = 0.0004$ (yielding $0.04\%$ expected population frequency).

        • Phase Pairs containing $0\%$ frequency reference haplotypes resolve to $0$.

      • Bayesian Analysis Probability: Restricting probabilities exclusively to possible candidate pairs yields a $99\%$ posterior probability that Phase Pair 1 ($A-T-C$ / $G-C-G$) represents the true physical phase.

Genotype Imputation and Linkage Disequilibrium Analysis

• Definition of Imputation: A computational procedure that predicts ungenotyped SNPs in genomic regions surrounding tag SNPs by leveraging established reference panel haplotypes and Linkage Disequilibrium (LD).

• Imputation Mechanics:

  • Once tag SNP phases are resolved (e.g., establishing an $A-C-G$ haplotype on one chromosome), matching population reference panels fill in all intervening unassayed SNPs (e.g., inferring $T-T-G-T-G$ along the $A-C-G$ backbone).

  • Significantly increases available genetic data points across the genome without generating additional assay costs.

• Detailed Multi-Locus LD and Imputation Example:

  • Consider a genomic region with $6$ loci and potential alleles: Locus 1 ($A/T$), Locus 2 ($G/A$), Locus 3 ($G/C$), Locus 4 ($T/C$), Locus 5 ($G/C$), Locus 6 ($A/T$).

  • Population Reference Co-Segregation Analysis (Complete Linkage Disequilibrium, $r^2 = 1$):

    • Loci 1 and 2 co-segregate perfectly: $A$ at Locus 1 always co-occurs with $G$ at Locus 2; $T$ at Locus 1 always co-occurs with $A$ at Locus 2.

    • Loci 3 and 5 co-segregate perfectly: $G$ at Locus 3 always co-occurs with $G$ at Locus 5; $C$ at Locus 3 always co-occurs with $C$ at Locus 5.

    • Loci 6 and 4 co-segregate perfectly: $A$ at Locus 6 co-occurs with $T$ at Locus 4; $T$ at Locus 6 co-occurs with $C$ at Locus 4.

  • Tag SNP Efficiency: Assay testing can be reduced from $6$ loci down to $3$ tag loci (Loci 1, 3, and 6) while retaining $100\%$ statistical capacity to impute Loci 2, 4, and 5.

  • Further Haplotype Reduction: Because population haplotypes represent a restricted subset of permutations, testing only $2$ tag loci (Loci 1 and 3) fully resolves all $6$ loci:

    • Allele combination $A-G$ at Loci 1 & 3 uniquely correlates with $A$ at Locus 6 in population reference panels, filling in all remaining positions.

    • Allele combination $A-C$ at Loci 1 & 3 uniquely correlates with $C$ at Locus 6, fully resolving the haplotype.

Statistical Association Testing and Multiple Comparison Adjustments

• Single-SNP Association Testing Model:

  • Each genotyped and imputed SNP across the entire genome is modeled as an independent variable in a separate statistical test against the disease condition or quantitative trait value.

• The Multiple Testing Problem:

  • Standard statistical significance uses $p = 0.05$ ($5\%$ chance of false discovery).

  • Testing $100$ independent hypotheses at $p = 0.05$ yields an expected $5$ false positives.

  • A standard GWAS evaluates $\ge 1{,}000{,}000$ ($10^6$) SNPs across the genome.

  • Applying $p = 0.05$ across $1{,}000{,}000$ tests would generate approximately $50{,}000$ false positives, drowning true association signals in noise.

• Bonferroni Correction & Gold Standard Significance Threshold:

  • GWAS utilizes multiple comparison corrections, most notably the Bonferroni multiple comparison adjustment.

  • Gold Standard Genome-Wide Significance Threshold: $p < 5 \times 10^{-8}$.

  • Target Variant Class: Designed to detect common genetic variants with a Major Allele Frequency (or Minor Allele Frequency threshold) above $0.5$ (or common variants) in strong linkage disequilibrium with the phenotype.

  • Threshold Adjustments: Significance cut-offs must be dynamically recalibrated based on ancestral population structure, variant allele frequencies, and low statistical power when evaluating rare variants.

Manhattan Plot Visualization and Significance Thresholds

• Definition & Axes:

  • A Manhattan plot provides a full genomic visual representation of GWAS results, plotting genomic coordinates against statistical association significance.

  • X-Axis: Genomic coordinates arranged sequentially across autosomes (Chromosomes 1 through 22) and sex chromosomes.

    • Modern GWAS include sex chromosomes (X and Y) using analytical controls that adjust for sex chromosome dosage variations across individuals.

  • Y-Axis: Negative logarithm base 10 of the p-value: $-\log_{10}(p)$.

    • Higher vertical position on the plot corresponds to a smaller p-value and stronger statistical association.

  • Data Points: Each individual dot on the plot represents a single tested SNP.

• Significance Threshold Lines:

  • Genome-Wide Significance Threshold: Horizontal line drawn at $p = 5 \times 10^{-8}$ ($-\log_{10}(p) \approx 7.3$). Any SNP rising above this threshold demonstrates definitive genome-wide statistical association.

  • Suggestive Significance Threshold: Horizontal line drawn at a lower significance threshold below the genome-wide line. Highlights variants exhibiting moderate statistical association that failed to reach genome-wide significance due to limited sample power, marking them as candidates for priority follow-up study.

Post-GWAS Extensions: Meta-Analysis, Replication, and Functional Follow-Up

• GWAS Meta-Analysis:

  • Individual GWAS cohorts frequently lack sufficient sample size and statistical power to detect minor effect-size variants.

  • Meta-analysis combines summary statistics and genotypic datasets across multiple global biobanks and independent cohorts.

  • Significantly increases overall sample size, providing the statistical power needed to uncover true genetic associations.

• Replication Studies:

  • Essential validation step to confirm initial GWAS associations and eliminate false positives or cohort-specific artifacts.

  • Requires replicating candidate SNP associations in independent, larger sample cohorts or across distinct ancestral sub-populations.

• Distinction Between Association and Causality:

  • A GWAS establishes statistical association between a SNP marker and a trait or disease; it does not prove biological causality.

  • Associated tag SNPs are typically non-functional markers in linkage disequilibrium with true causative loci.

  • Downstream functional follow-up experiments (e.g., fine-mapping, expression studies, and biological assays) are required to identify the true causative genetic locus and underlying pathobiological mechanism.