Introduction: : Aortic valve stenosis (AVS) is a progressive cardiovascular disease characterized by the fibrosis and calcification of the aortic valve leaflet, leading to heart failure if untreated. Importantly, AVS is sexually dimorphic with increased fibrosis in female patients and increased calcification in male patients. Sex-specific AVS pathophysiology is regulated by valvular interstitial cells (VICs), which are the resident fibroblast present in the aortic valve. During normal wound healing, VICs activate to a myofibroblast phenotype to facilitate tissue repair, then return to a quiescent state. However, during AVS progression, VICs remain persistently activated to a myofibroblast phenotype and secrete excessive extracellular matrix leading to valvular fibrosis. Currently, AVS is treated through valve replacement procedures, but this is an imperfect treatment due to patients frequently experiencing restenosis of the newly implanted valve. Additionally, to date, no drug therapies have successfully treated AVS, likely due to an incomplete understanding of sex-specific AVS pathophysiology. Indeed, prior work has shown that female VICs are more prone to activating to a myofibroblast phenotype and less responsive to subsets of anti-fibrotic drugs relative to male VICs. Given these findings, we believe that effective drug therapies for AVS will need to be customized based on patient sex. Here, we harnessed IDentif.AI, an artificial intelligence-derived drug optimization platform, to develop sex-specific predictive models of anti-fibrotic drug combination efficacy. We used the models to identify and validate subsets of drug combinations and synergistic drug interactions that are sexually dimorphic.
Materials and
Methods: : Male and female VICs were isolated from 6-8-month-old male and female adult pigs using collagenase digestion. Leaflets from a minimum of 4 sex-matched hearts were pooled together for each isolation, representing one biological replicate. After expansion, VICs were seeded on RGD-functionalized poly(ethylene glycol) norbornene (PEG-nb) hydrogels that were optimized to recapitulate healthy (soft hydrogels, E ~ 7 kPa) or diseased (stiff hydrogels, E ~ 51 kPa) aortic valve tissue. For drug screening assays, male and female VICs were cultured on stiff hydrogels with optimized doses of each inhibitor for two days then immunostained for alpha smooth muscle actin (αSMA) expression and analyzed using an automated MATLAB code to quantify the αSMA gradient mean intensity for each cell4. Drug efficacy was quantified using the average αSMA gradient mean intensities of the control relative to the drug condition to calculate the percent αSMA reduction using the formula (100*(αSMAcontrol – αSMAcondition)/αSMAcontrol). For drug combination optimization, we harnessed the IDentif.AI platform to pinpoint combinatorial designs that significantly decrease alpha smooth muscle actin (αSMA) expression in male and female VICs (Fig. 1A). Male or female VICs cultured on stiff hydrogels were exposed to a set of 59 combinations (chosen from 8 drugs at 3 doses) following an orthogonal array composite design to interrogate the drug-drug interaction space via a second order quadratic series derived using stepwise regression.
Results, Conclusions, and Discussions:: Sex-specific drug response curves for eight anti-fibrotic drugs were generated based on reductions in male and female VIC αSMA expression. After calibrating sex-specific doses of each inhibitor, a dataset of 59 drug combinations were tested separately in male and female VICs to inform the IDentif.AI platform. Sex-specific IDentif.AI models generated a ranked list of all drug combinations alongside predicted percent αSMA reduction values. We prioritized three combinations from the top 2-drug, 3-drug, and 4-drug combinations for male VICs for subsequent experimental validation (M1-M9) alongside two predicted ineffective combinations (M10 and M11). We found that eight out of the nine IDentif.AI-designed predicted effective combinations were indeed highly effective in male VICs, whereas the predicted ineffective combinations had no meaningful impact on αSMA expression (Fig. 1B, 1C). The IDentif.AI workflow was then repeated with female VICs cultured on stiff hydrogels, revealing six of nine predicted effective combinations (F1-F9) reduced αSMA expression, whereas predicted ineffective combinations (F10-F12) did not (Fig. 1B, 1D). Next, we leveraged our separate male and female predictive models to identify sex-biased drug combinations. To confirm our predictions, we measured the percent αSMA reduction of three predicted male-biased and three predicted female-biased drug combinations. We found that these combinations are indeed sex-biased as predicted (Fig. 1E). Moreover, we found that synergistic drug interactions for subsets of two-drug combinations (male: Losartan/SD-208, female: LY294002/H1152) were sex-dependent (Fig 1F). After that, we tested our nine male (M1-M9) and female (F1-F9) identified combinations on soft hydrogels and tissue-culture polystyrene (TCPS) to investigate the impact of microenvironmental stiffness on drug efficacy. We found that the majority of optimized combinations were significantly more effective in VICs cultured on soft or stiff hydrogels relative to TCPS, motivating the need for hydrogel biomaterials to study sex differences in drug responses (Fig. 1G).
Collectively, our work identified and validated drug combinations to inhibit VIC myofibroblast activation using computational drug optimization tools, demonstrating how predictive models may be leveraged to accelerate sex-specific precision medicine. In sum, our workflow may help accelerate AVS drug development for male and female patients and address health disparities in AVS treatment outcomes.
