We thank Suay-Corredera and Alegre-Cebollada for their thoughtful correspondence regarding our recent publication.1 As described in their letter, we applied STRUM,2 a bioinformatics tool that predicts protein destabilization, to missense variants in cMyBP-C, the protein coded by MYBPC3, with the aim of identifying variants which lead to cMyBP-C loss of function via protein destabilization.3 We are pleased that they highlighted the utility of our approach in providing supportive evidence for variant pathogenicity and the value of the computational saturation mutagenesis as a reference for evaluation of novel variants.

The central assertion of their letter is that experimental analyses of variants of uncertain significance in MYBPC3 are important to complement and validate computational predictions made. We completely agree. In our paper we discuss the results of protein stability and incorporation into myofibrils when variants were studied within the context of full length cMyBP-C.3,4 While to date, we have only studied a subset of variants in this fashion, studies are ongoing with additional variants. It is also possible that subdomain misfolding, as predicted by STRUM, could result in smaller perturbations that would have effects on the function of the protein and its interactions with binding partners, without impacting its overall integrity and stability. For example, as the authors of this letter point out, cMyBP-C Trp792Arg and Arg810His, which are STRUM+ variants, incorporate normally in the sarcomere and have normal half-lives.3,4 While these findings may appear to be at odds with the computational predictions, experiments designed to study variants in isolated individual subdomains may yield different results.

Thus, we read with great interest Suay-Corredera and Alegre-Cebollada’s comparison of our computational predictions to their experimental studies. They expressed cMyBP-C subdomains carrying 33 MYBPC3 variants within E. coli (unpublished data).5 They observed that a number of the variant domains failed to express, which they interpreted to imply marked destabilization. When they combined these with variant domains that expressed but had reduced melting temperatures, they found that 60% of the variants that met their experimental criteria for domain destabilization were computationally predicted to cause domain destabilization (STRUM+). Conversely, there was strong concordance (89%) between STRUM- variants and their experimental data for variants that had no experimental evidence for domain instability. Based on these findings, the authors conclude that STRUM predictions yield a 40% false negative rate for detecting variants that cause protein domain destabilization. While possible, it is difficult for us to reconcile this with the fact that STRUM robustly enabled clinical risk stratification for patient outcomes. As such, definitive conclusions regarding the variants with conflicting experimental and computational results may be premature.

Experimental evidence using complementary methodologies to the ones employed by the authors could help to resolve these discrepancies. In addition to the cell biological studies of full-length MyBP-C we are currently conducting, cellular folding sensors have been developed within the field of directed evolution and could be applied to study MyBP-C subdomains.6 These systems are designed such that the cellular abundance of a selectable marker, proteins that fluoresce or confer antibiotic resistance, is influenced by the stability of your protein of interest. Combining these methods with landing pad technology which ensures only one copy or variant is expressed per cell may enable a robust analysis of MyBP-C subdomain stability within cells.7

We agree with Suay-Corredera and Alegre-Cebollada that coupling computational approaches with data acquired from complementary experimental methods has the power to provide support for or against pathogenicity, particularly for variants that result in loss of function. In aggregate, these data can be very helpful in the clinic for predicting disease course and outcomes in patients. Further, segregation within families and enrichment in disease versus nondisease populations still remain essential components of variant interpretation that must also be evaluated and considered.8