(E-168) A Quantitative Approach to Classify Aortic Valve Interstitial Cell Phenotypes in vitro

Introduction:: Aortic valve interstitial cells (AVICs) are fibroblast-like cells that reside within aortic valve (AV) leaflet tissues and function to maintain the structure of the AV through turning over extracellular matrix components (ECM) [1]. In periods of growth or disease, AVICs can undergo phenotypic activation into myofibroblasts which are highly secretory and contractile in nature. Notably, activated myofibroblasts are larger and less spindle-like in shape than quiescent fibroblasts. Misregulation and persistence of the activated AVIC phenotype can lead to excessive ECM deposition causing scarring or fibrosis. If this process persists, the AV can stiffen to a pathological extent that impedes proper opening and closing behavior, a condition known clinically as AV stenosis [2]. The predominant hallmark of the activated AVIC phenotype is expression of discrete alpha-smooth muscle actin (αSMA) stress fibers. Traditionally, researchers and pathologists have used immunostaining to visualize αSMA; however, grading the severity and discreteness of the fibers has remained largely subjective. Previous investigations have relied on absolute or normalized fluorescence intensity measurements which can vary due to differences in microscopes, sources of antibodies, staining protocols, and imaging parameters such as laser power and gain. Therefore, there is currently no standard quantitative method to consistently assess the activated AVIC phenotype, leading to inconsistencies across the literature. In the present study, we have developed a quantitative image analysis approach that assesses both 2D morphological features and expression of discrete αSMA stress fibers in a manner independent of fluorescence intensity to demarcate the quiescent and activated AVIC phenotypes.


Materials and Methods:: Porcine AVICs seeded on soft, quiescent-promoting (Fig1A) and stiff, activation-promoting (Fig1C) hydrogels were fixed, stained, and imaged for αSMA, plasma membrane, and the nucleus [3]. The AVIC body and nucleus were segmented using a water shed algorithm. Fourier series fits of the AVIC body and nuclear boundaries were conducted to obtain coefficients that summarized their shapes (Fig1B&D). A total of 15 series (62 coefficients total) were used to fit the AVIC body, whereas 5 (22 coefficients total) were used for the nucleus. Next, we assessed the discreteness of αSMA fibers by computing coherency [4], which depends on fluorescence intensity gradient and not magnitude and represents the level of anisotropy at pixel-resolution. First, we computed the intensity gradient of the αSMA image in the x and y directions (I_x and I_y). Then, on a pixel-by-pixel basis, the structure tensor was assembled: [W*(I_x²), W*(I_xI_y) ; W*(I_xI_y), W*(I_y²)] where W is a Gaussian smoothing kernel. Note that the structure tensor is second rank and symmetric. Next, the Eigenvalues (λ₁,λ₂) and Eigenvectors (e₁, e₂) of the structure tensor were obtained. Finally, coherency was computed as (λ₁-λ₂)/(λ₁+λ₂). Note that the vector e₂ represents the local αSMA fiber orientation. The median coherency value of each AVIC was used to represent its degree of anisotropy (DOA). Vectors comprising the Fourier coefficients (84 total) and the DOA for every AVIC were assembled into a matrix and subjected to principal component (PC) analysis to assess the greatest variations among quiescent and activated AVICs. >1000 total AVICs were analyzed.


Results, Conclusions, and Discussions:: AVICs on soft hydrogels were smaller and showed diffuse αSMA expression compared to AVICs on stiff hydrogels (Fig1A&C). AVIC body and nuclear shapes were adequately represented by Fourier series (Fig1B&D). PC analysis revealed that the greatest variation among AVICs seeded on soft and stiff hydrogels consisted of variations in both morphology and DOA (Fig1E&F). AVIC body (Fig1E) and nuclear (Fig1F) shape and size and the corresponding DOA at the extremes (+/-2 STD) of the first PC axis were visualized. We observed that AVICs with smaller body and nuclear area were associated with a low DOA whereas AVICs with larger body and nuclear area were associated with a high DOA. This analysis revealed that AVIC area, nuclear area, and DOA accounted for significant variation among AVICs.

Next, ROC curves were generated to assess the diagnostic ability of the identified variables in grading AVIC phenotype (Fig1G). First, we manually graded the AVICs as quiescent or activated using standards commonly employed in the literature (e.g., presence of visually discrete αSMA fibers) and then computed the true positive rate (TPR) and false positive rate (FPR) at ever-increasing threshold values of DOA, cell area, and nuclear area. DOA had the largest area under the curve (AUC=0.88), signifying that it is a better predictor of AVIC activation than cell (AUC=0.79) and nuclear (AUC=0.77) area. However, cell and nuclear area also showed excellent diagnostic ability as evidenced by high AUC values. All variables were considerably superior over a random classifier (dashed line, Fig1G). Logistic regression revealed that a DOA of 0.72 denotes an activation probability of 50% (Fig1H), suggesting that AVICs with a DOA above 0.72 are likely activated. A DOA of 0.72 resulted in a high TPR (0.72) and low FPR (0.13) (red star, Fig1G) further supporting that a DOA of 0.72 is an appropriate quantitative threshold to grade AVIC activation.

Looking forward, we will incorporate the identified quantitative variables in a multi-variate logistic regression model to grade AVIC activation based on shape and αSMA fibers. This model will be used to characterize AVICs in normal and diseased states both in vitro and in vivo.


Acknowledgements (Optional): : We acknowledge funding from National Institutes of Health T32 HL007822 awarded to AK, American heart association 20PRE35200068 awarded to DB, and National Institutes of Health R01 HL142935 awarded to KSA.


References (Optional): : [1] Taylor et al., Biochemistry & Cell Biology 2003, 35(2): 113-118.
[2] Rutkovskiy et al., Journal of the American Heart Association 2017, 6(9):e006339.
[3] Batan et al., The Faseb Journal 2022, 36:e22306.
[4] Püspöski et al., Advances in Anatomy, Embryology and Cell Biology, vol. 219, ch.3, 2016.