Predicting the functional effect of compound heterozygous genotypes from large scale variant effect maps

Michael J. Xie, Gareth A. Cromie, Katherine Owens, Martin S. Timour, Michelle Tang, J. Nathan Kutz, Ayman W. El‐Hattab, Richard N. McLaughlin, Aimée M. DudleybioRxiv 2023 · DOI

TL;DR

This paper presents a high-throughput yeast-based assay to functionally assess thousands of genetic variants in the human PSAT1 gene, crucial for serine biosynthesis. By developing a predictive model, they can accurately estimate the functional impact of complex genetic combinations, paving the way for better diagnosis and understanding of rare genetic disorders.

What Problem Does This Paper Address?

Diagnosing rare genetic diseases, particularly those with a broad phenotypic spectrum like serine biosynthesis defects, is a significant challenge. While whole-exome sequencing can identify genetic variants, interpreting their functional impact, especially for rare variants or compound heterozygous genotypes (where an individual has two different variants in the same gene), is difficult. Current methods often rely on limited clinical data or computational predictions with high error rates. This paper aims to bridge this gap by creating a robust, scalable functional assay to assess the impact of PSAT1 variants and developing a predictive model for compound heterozygous genotypes, which are common in autosomal recessive diseases.

Methods

The core of this work is a yeast-based complementation assay. Human PSAT1's function was tested by expressing it in yeast cells that have a defective serine biosynthesis pathway (specifically, a ser1∆0 deletion strain). The yeast's ability to grow on minimal medium lacking serine serves as a quantitative readout of PSAT1 activity.

Key methodological advancements include:

  • Large-scale variant library construction: They synthesized over 2,182 unique amino acid substitutions in the human PSAT1 gene, accessible via single nucleotide variants (SNVs). These were introduced into yeast strains.
  • High-throughput phenotyping: An automated robotic system was used to grow and image thousands of yeast strains, allowing for efficient measurement of growth differences.
  • Variant calling and validation: A custom pipeline using Oxford Nanopore (MinION) sequencing was developed to identify the specific variant in each yeast strain, and this was validated against Illumina sequencing.
  • Diploid yeast models: To model human compound heterozygous genotypes, they created diploid yeast strains by mating haploid strains carrying different PSAT1 variants. This allowed for the direct assessment of allele pair function.
  • Predictive modeling: A linear model was developed to predict the functional outcome of diploid genotypes based on the experimentally determined function of individual alleles in haploid strains. This model was then used to predict the function of millions of potential human genotypes.

Datasets

  • Human PSAT1 variants: The study focused on 1,914 unique SNV-accessible missense amino acid substitutions in the human PSAT1 gene. This dataset represents approximately 88% of all possible SNV-accessible substitutions.
  • ClinVar database: Results were compared to clinical classifications of variants in the ClinVar database to validate the functional assay's relevance.
  • Disease literature: Data from published reports of patients with PSAT1-related serine biosynthesis defects (including Neu-Laxova syndrome and PSAT deficiency) were used to correlate genotype with disease severity.

Key Insights and Ideas

  • Yeast as a powerful model for human gene function: The study demonstrates that a yeast complementation assay can accurately reflect the functional impact of human gene variants, aligning well with clinical interpretations and protein structure-function relationships.
  • Functional stratification of variants: The assay revealed a clear bimodal distribution of variant effects, with a large group of variants showing little to no impairment and a smaller group exhibiting complete loss of function, mirroring null alleles. This allows for classification of variants into "unimpaired," "impaired," "amorphs" (complete loss), and "hypomorphs" (partial loss).
  • Structure-function correlation: Conserved regions of the PSAT1 protein, particularly those around the active site and involved in cofactor or substrate binding, are highly sensitive to amino acid substitutions, leading to significant functional impairment.
  • Predicting compound heterozygote function: The developed linear model can accurately predict the functional outcome of diploid genotypes from haploid measurements, offering a more efficient way to assess the impact of allele combinations than direct experimental testing of millions of combinations.
  • Clinical relevance of diploid modeling: The diploid yeast models successfully distinguished between genotypes associated with severe disease (NLS2) and milder forms (PSATD), aligning with clinical observations and disease severity.

Key Results

  • The study functionally characterized 1,914 unique PSAT1 missense amino acid substitutions.
  • Results showed excellent agreement (R²=0.94) with a previous study that assayed a subset of these variants.
  • A highly significant correlation (Spearman rank correlation=0.47, p < 7.2 x 10⁻²²) was found between residue conservation (ConSurf score) and median yeast assay score, indicating that conserved residues are more sensitive to mutations.
  • The developed pairwise additive model predicted diploid growth with 97% variance explained from haploid measurements.
  • This model achieved 73% accuracy in classifying genotypes as disease-causing or not, comparable to experimental diploid data (72% accuracy).
  • The study provides functional data for 77% of missense VUS (variants of uncertain significance) currently listed in ClinVar for PSAT1.
  • The model can predict the function of ~1.8 million potential human genotypes.

Limitations and Future Work

  • Model organism limitations: While yeast is a powerful model, it may not perfectly recapitulate all aspects of human PSAT1 function or cellular context.
  • Limited clinical data for rare variants: The interpretation of variants with scores between 82% and 91% remains uncertain due to a lack of corresponding clinical annotations.
  • Future work suggested:
    • As more clinical information becomes available, the ability to make clinical inferences from the dataset will improve.
  • The authors suggest that this approach of large-scale functional assays in model systems can be powerfully applied to the study of other rare diseases.
  • The predictive model for diploid genotypes can be further refined with more experimental data.
  • The generated dataset can inform future diagnostic efforts and research into the natural history of serine biosynthesis disorders.

Summary updated

Abstract

Taken together, our work provides an example of how large-scale functional assays in model systems can be powerfully applied to the study of a rare disease.
BibTeX
@misc{xie2023predicting,
  author = {Michael J. Xie and Gareth A. Cromie and Katherine Owens and Martin S. Timour and Michelle Tang and J. Nathan Kutz and Ayman W. El‐Hattab and Richard N. McLaughlin and Aimée M. Dudley},
  title = {{Predicting the functional effect of compound heterozygous genotypes from large scale variant effect maps}},
  year = {2023},
  doi = {10.1101/2023.01.11.523651},
  howpublished = {bioRxiv},
}
Figure