Scaled multidimensional assays of variant effect identify sequence-function relationships in hypertrophic cardiomyopathy

Abstract

Background: An estimated 1 in 500 people live with hypertrophic cardiomyopathy (HCM), a disease for which genetic diagnosis can identify family members at risk, and increasingly guide therapy. Mutations in the myosin binding protein C3 (MYBPC3) gene account for a significant proportion of HCM cases. However, many of these variants are classified as variants of uncertain significance (VUS), complicating clinical decision-making. Scalable methods for variant interpretation in disease-specific cell types are crucial for understanding variant impact and uncovering disease mechanisms.

Methods: We developed a scaled multidimensional mapping strategy to evaluate the functional impact of variants across a critical domain of MYBPC3. We incorporate saturation base editing at the native MYBPC3 locus, a long-read RNA sequencing-enabled assay of variant splice effects, and measurements of HCM-relevant phenotypes, including MYBPC3 abundance, hypertrophic signaling, and ubiquitin-proteasome function in human induced pluripotent stem cell-derived cardiomyocytes (iPSC-CMs).

Results: Our multidimensional mapping strategy enabled high-resolution functional analysis of MYBPC3 variants in iPSC-CMs. Targeted transient base editing generated a comprehensive variant library at the native locus, capturing diverse variant effects on cellular HCM-relevant phenotypes. Our massively parallel splicing assay identified novel splice-disrupting variants. Integration of functional assays revealed that decreased MYBPC3 abundance is a key driver of HCM-related phenotypes. In parallel, downregulation of protein degradation was observed as a compensatory response to MYBPC3 loss of function, and novel disease mechanisms were identified for missense variants near a critical binding domain, underscoring their contribution to pathogenesis. Bayesian estimates of variant effects enable the reclassification of clinical variants.

Conclusions: This work provides a platform for extending genome engineering in iPSCs to multiplexed assays of variant effects across diverse disease-relevant cellular phenotypes, enhancing the understanding of variant pathogenicity and uncovering novel biological mechanisms that could inform therapeutic strategies.

Figure 1.. cMyBP-C M-domain and massively parallel splicing assay using minigene plasmid library

A, 3D structures (PDB: 7TIJ) of tri-helix bundle (THB, orange) and actin (purple). B, AlphaMissense prediction scores in cMyBP-C M-domain.The AlphaMissense prediction scores range from 0 (blue) to 1 (red), with higher scores indicating a greater likelihood of pathogenicity. C-terminal region of M-domain (p.319-p.356, encoded by exon 12), is a potential hotspot for pathogenic missense variants. C, Proportion of the canonical isoform of known pathogenic or likely pathogenic (P/LP), benign or likely benign (B/LB), and all other variants (P/LP; n = 7, B/LB; n = 9, and VUS; n = 386). Nonsense variants were excluded from classification as P/LP (see Methods). D, Landscape of single variant effects on canonical splicing (top). P/LP and B/LB variants are highlighted. The range of disrupting (red), intermediate (purple), and neutral (blue) are indicated by shading. Red and blue dashed lines indicate the mean proportion of canonical isoform (vs reference minigene) in P/LP and B/LB. The averaged proportion of the canonical isoform (vs reference minigene) in each position (middle). The number of RNA binding protein (RBP) motifs in each position (bottom). E, Number of variants classified based on proportion of the canonical isoform in each variant type. F, Variant effect map for splicing. X axis indicates the position in MYBPC3 c.927–50 – c.1020. Y axis indicates nucleotide identity. Box color denotes the variant classification based on proportion of the canonical isoform. Reference nucleotides were denoted with gray squares. Blank boxes indicate missing data or variants were filtered out due to low confidence.

Figure 2.. in situ mutagenesis using base editors

A, Schematic of in situ mutagenesis approach. CRISPR-X and Adenine base editing were performed separately. Created with BioRender.com. B, Enrichment scores in biological replicates (rho = 0.87, p = 5.14e-51). C, Mutagenesis efficiency of individual variants in the mutagenized iPSC library (CRISPR-X + ABE8e). Each plot represents the enrichment score for each variant. D, A total of 281 identified variants in mutagenized iPSC library and their enrichment scores. The color of boxes reflects the fold enrichment scores. Reference nucleotides were denoted with gray squares. Blank boxes indicate missing data or variants were filtered out due to low confidence. E, Proportion of nucleotide changes observed in CRISPR-X and Adenine base editor (ABE8e) library. A total of 145 and 65 variants were identified in the CRISPR-X and AEB8e libraries, respectively.

Figure 3.. Multiplexed functional assays using iPSC-derived cardiomyocytes

A, Immunofluorescence image of α-actinin (red) and cMyBP-C (green) in iPSC-derived cardiomyocytes (iPSC-CMs). Scale bar: 20 μm. B, Schematic of selection assays design using iPSC-CMs derived from mutagenized iPSC library. Created with BioRender.com. C through E, Functional scores in each variant type. F through H, Functional scores in pathogenic or likely pathogenic (P/LP), benign or likely benign (B/LB), and all other variants (VUS).

Figure 4.. Integrated analysis of multidimensional functional assays.

A, Variant effect map on cMyBP-C abundance (top), BNP expression (middle), and ubiquitin-proteasome system (UPS) function (bottom). The color of boxes reflects each functional score. Reference nucleotides were denoted with gray squares. Blank boxes indicate missing data or variants were filtered out due to low confidence. Nonsense variants are denoted by asterisks. Created with BioRender.com. B, Comparison of normalized proportion of the canonical splicing isoform and cMyBP-C reduction scores (n = 121, rho = −0.15, p = 0.094). C, cMyBP-C reduction scores in nonsense and splice-disrupting variants. D and E, Functional scores (hypertrophy and proteasome impairment) are plotted against for cMyBP-C reduction scores (BNP expression; n = 121, Proteasome impairment; n = 131).

Figure 5.. Multidimensional assay decodes missense variants of MYBPC3.

A, Mean functional scores of single amino acid substitutions in 1D structure of cMyBP-C. Each cell represents an amino acid position, with the background color and numerical value indicating the mean functional score. The fraction below (e.g., 2/6) represents the coverage of observed substitutions (number of the substitution was observed / total possible substitution for that position). Gray indicates missing data or variants were filtered out due to low confidence. B through D, Comparison of functional scores between variants within and outside of the tri-helix bundle (THB) region. E, Mean proteasome impairment score of amino acid substitutions were utilized to map onto 3D structure of the THB. F, Structure of cMyBP-C M-domain. G through I, Local atomic interaction in wild type and mutated structures for Glu319Asp, Asp320Glu, and Val321Leu. The p.Glu319Asp variant disrupts the salt bridge between p.Glu319 and p.Trp322, and creates a new salt bridge with the structure’s backbone at position 319. The p.Asp320Glu variant causes a loss of the salt bridge formed with p.Ser318. The missense variant p.Val321Leu introduces a much larger residue at position 321 which overlaps with the molecular surface of p.Val342 and thus will not fit without a conformational change.

Figure 6.. Functional assay evidence-based variant classification.

A, Decision tree for the variant classification for clinical interpretation using multidimensional functional assay data. We integrated data across assays by determining the assay with the strongest evidence for pathogenicity, and, failing that, the strongest evidence for benign classification as has been suggested for incorporation of Bayesian variant effects into clinical variant interpretation. The strength of evidence in each assay is based on Odds of pathogenicity (OddsPath). B, Classification of 17 variants (7 pathogenic, 1 likely pathogenic, 9 variants of uncertain significance (VUS)) identified in the Sarcomeric Human Cardiomyopathy Registry (SHaRe) based on our functional assay data. These variants were adjudicated by two independent cardiovascular genetics specialists with and without the functional data provided by our assays according to ACMG guidelines using a validated points system for reproducibility. To combine the data from our four assays, we assessed evidence for pathogenicity first. In the case of conflicting evidence, given the different disease mechanisms presented, the strongest level of evidence available was accepted. If none of the assays showed evidence of pathogenicity, the strongest available evidence for benign effect was accepted. C, Single-nucleotide resolution assay revealed the splicing disruption in the synonymous variant (c.1008C>T/lle336lle) identified in SHaRe registry.