Mechanism based therapies enable personalised treatment of hypertrophic cardiomyopathy

Cardiomyopathies have unresolved genotype–phenotype relationships and lack disease-specific treatments. Here we provide a framework to identify genotype-specific pathomechanisms and therapeutic targets to accelerate the development of precision medicine. We use human cardiac electromechanical in-silico modelling and simulation which we validate with experimental hiPSC-CM data and modelling in combination with clinical biomarkers. We select hypertrophic cardiomyopathy as a challenge for this approach and study genetic variations that mutate proteins of the thick (MYH7R403Q/+) and thin filaments (TNNT2R92Q/+, TNNI3R21C/+) of the cardiac sarcomere. Using in-silico techniques we show that the destabilisation of myosin super relaxation observed in hiPSC-CMs drives disease in virtual cells and ventricles carrying the MYH7R403Q/+ variant, and that secondary effects on thin filament activation are necessary to precipitate slowed relaxation of the cell and diastolic insufficiency in the chamber. In-silico modelling shows that Mavacamten corrects the MYH7R403Q/+ phenotype in agreement with hiPSC-CM experiments. Our in-silico model predicts that the thin filament variants TNNT2R92Q/+ and TNNI3R21C/+ display altered calcium regulation as central pathomechanism, for which Mavacamten provides incomplete salvage, which we have corroborated in TNNT2R92Q/+ and TNNI3R21C/+ hiPSC-CMs. We define the ideal characteristics of a novel thin filament-targeting compound and show its efficacy in-silico. We demonstrate that hybrid human-based hiPSC-CM and in-silico studies accelerate pathomechanism discovery and classification testing, improving clinical interpretation of genetic variants, and directing rational therapeutic targeting and design.

relaxed state DRX. We will refer to this proportion as the DRX:SRX ratio. Specifically, we Considering the ordinary differential equation that describes the time evolution of weaklybound crossbridges (W) reported above, in steady-state conditions, when 45 4$ = 0, the relationship between the steady-state occupancies and transition rates will be as no distortion ( 0/ ) is present in steady-state, and therefore 9 :

R
The system will therefore reach a steady-state configuration determined by 1 ( ) and ( ( ).
In particular, depending on , the ratio < 5 6 = 22 will be shifted towards the U or the W state in line with the hypothesis that changing the availability of myosin heads that can interact with actin will affect how many crossbridges can enter in the pre-powerstroke state W. When 1 ( ) = ( ( ) = 1, the original system is recovered (including the steady-state configuration). We therefore impose the constraint that 1 (1) = ( (1) = 1, so that = 1 represents the control condition where (DRX:SRX)=(DRX:SRX)control. Note that → 0 as DRX → 0 (no interaction with actin), which implies 1 (0) = 0 (no crossbridges in pre-power state W). Moreover, 1 ( ) must be strictly monotonic increasing in to signify a gradual increased transition to the prepowerstroke state W. Similarly, → ∞ ( ≫ 1) as SRX → 0 (no crossbridges in super relaxed state, so all contribute to force production). Therefore, ( ( ≫ 1) → 0, increasing the < To achieve this, we conducted a sensitivity analysis in which we varied the parameter to generate a population of models. We considered the two scenarios of myosin inhibition (R < 1) and enhancement (R > 1) separately. We considered a positive for 1 ( ) and a negative for ( ( ), since 1 ( ) and ( ( ) must be strictly monotonic increasing and decreasing in , respectively. We first determined the range of variation of by qualitatively evaluating the Mavacamten concentrations and R values. This necessitated that we determined this correspondence in silico, i.e. that we defined R values that, when used in our models, provide the same change in contractility in silico as it was observed with experimental data for 0.5 and 1µM Mavacamten concentrations.
As shown in Supplementary Fig. 2f, there exists a range of R values, one value for each model, for each of the two Mavacamten concentrations considered, that can satisfy that criterion, i.e.
replicate the ~30% and ~70% decrease in maximal steady-state tension under 0.5 and 1µM Mavacamten observed experimentally. Therefore, we constructed an ensemble of doseresponse curves where R values can be approximated from Mavacamten concentrations. We

In-silico evaluation of preclinically relevant biomarkers of human cardiac contractility
Two different protocols were simulated to obtain isometric twitch tension in human virtual intact cardiomyocytes and the steady-state tension-calcium relationship in human virtual skinned cardiomyocytes. Twitch tension is elicited in silico by the stimulus current that triggers an action potential through the excitation-contraction coupling process. The steady-state tension calcium curve is obtained by using fixed calcium values as input of the contractility model parametrised for skinned cardiomyocytes 5 . The tension produced by the cell quickly reaches a steady-state value that is saved and plotted with respect to the corresponding calcium value.
From these two protocols the relevant biomarkers were computed. The steady-state calciumtension curve was fitted to a Hill equation. From this, maximal tension, calcium sensitivity of tension production, and Hill coefficient were estimated. From the twitch tension, time to 50% contraction, time to peak, time 50 and 90% decay, total duration, tension baseline and amplitude were computed. The model was paced at 1 Hz until the steady-state was reached, and then biomarkers were computed.

In-silico modelling of myosin-based contribution to thin filament activation in human cardiomyocytes
A crossbridge-dependent binding of calcium to cardiac troponin C has been observed experimentally 16,17 and also modelled computationally [18][19][20][21] . Building upon these findings, in this study we considered a direct modulation of thin filament calcium sensitivity based on the availability of myosin to form crossbridges and develop tension. Myosin availability is

In-silico evaluation of drug repurposing for the thin filament HCM phenotype correction
In order to identify therapeutic targets for the phenotype resolution for the troponin variants, we also tested the effect of the following pharmacological strategies. We tested different levels of late sodium ( Supplementary Fig. 3a,b) and L-type calcium ( Supplementary Fig. 3c,d) currents block (20,40 We conducted a recalibration of our cellular model of Mavacamten with the goal of, given a certain drug dose administered in clinical trials, determining what is the free plasma concentration to be used in the simulations to get appropriate left ventricular ejection fraction (LVEF) reductions. Available data from early phase clinical trials is reported in Supplementary   Fig. 4a (data digitalised from 1 ). Data was separated between healthy subjects and HCM patients. Only data from healthy subjects was used for calibration, as our whole-ventricular simulations only considered pre-symptomatic patients before development of obstructive HCM.
Before calibration, our in-silico trial results using preclinical dose concentrations reported a much stronger effect of Mavacamten on the LVEF than the clinical data ( Supplementary Fig.   4b). To calibrate our cellular model of Mavacamten, we mapped simulated data onto clinical data to determine a correspondence between drug dose administered clinically ( !&3#3!>& ) and simulated free plasma concentrations ( 23@/&>$A4 ), as shown in Supplementary Fig. 4c. Clinical data can be written as !&3#3!>& = * !&3#3!>& + where !&3#3!>& represents the LVEF reduction and !&3#3!>& represents the corresponding plasma concentration measured from patients after drug administration. From fitting the clinical data of healthy volunteers reported in Supplementary Fig. 4a, and were estimated to be -0.0173 and 0.4732, respectively. Simulated data can be written as: where , , , were estimated to be 28.45, 117, -28.45, and 5.27, respectively (obtained from fitting the simulated data reported in Supplementary Fig. 4b