Intermediate Molecular Phenotypes to Identify Genetic Markers of Anthracycline-Induced Cardiotoxicity Risk

Figures

• 

  Figure 1. Overview of the general approach. (A) This study investigates variations in susceptibility to Anthracycline-Induced Cardiotoxicity (CDA) within a backcrossed mouse cohort. CDA severity is quantified histopathologically using an Ariol slide scanner to assess both the subepicardial and subendocardial regions of the heart. Further, we quantify the myocardial levels of various intermediate phenotypes associated with cardiotoxicity. Following this, we identify genetic regions or Quantitative Trait Loci (QTL) correlated with CDA (cdaQTL), and other QTLs linked to intramyocardial levels of intermediate molecular phenotypes (ipQTL). (B) Subsequently, we examine whether certain ipQTLs significantly contribute to the phenotypic variation (susceptibility) of CDA, even if they lack a direct link. We employ genetic models to evaluate whether the phenotypic variation explained by the combined ipQTL and cdaQTL surpasses that explained solely by the cdaQTL. If true, the genetic determinant associated with myocardial levels of this intermediate molecular phenotype contributes to CDA susceptibility. (C) The genes responsible for ipQTLs remain unidentified. Nevertheless, any genetic determinant…

• 

  Figure 2. Genetic background, therapy regime, and age influence anthracycline-induced cardiotoxicity. (A) A backcrossed mouse cohort is used to evaluate Cardiotoxicity Due to Anthracyclines (CDA). Following chemotherapy treatment, cardiotoxicity is quantified histopathologically. (B,C) We compare cardiotoxicity between homogeneous genetic background mouse strains, FVB and F1. Following combined chemotherapy, FVB mice display reduced global heart fibrosis (B) and lessened subendocardial fibrosis (C) compared to F1 mice. No observable differences exist between the parental strains post-doxorubicin chemotherapy alone (Supplementary Table S1A). Please note the continuous distribution of heart fibrosis in F1BX mice (B,C). (D–F) For the genetically diverse F1BX mouse cohort, combined treatment induces greater cardiotoxicity than doxorubicin alone, as evidenced by the degree of global (D), subendocardial (E), and subepicardial (F) fibrosis. (G–J) In F1BX1 mice, CDA is more prevalent in the subendocardial zone than in the subepicardial zone, with respect to fibrosis area (G,H) and cardiomyocyte size (I,J), as analyzed via the Mann–Whitney U test. (K,L) The relationship between age, type of…

• 

  Figure 3. Association between intermediate molecular phenotypes and Anthracycline-Induced Cardiotoxicity (CDA) in F1BX mice. (A) Quantification of various intermediate molecular phenotypes in the myocardium of F1BX mice post-chemotherapy treatment; (B–I) Classification of mice into high and low CDA susceptibility groups via principal component analysis based on the myocardial levels of these phenotypes under different conditions: fibrosis following doxorubicin treatment (B), changes in cardiomyocyte size after combined therapy (C), subendocardial fibrosis post combined therapy (D), subepicardial fibrosis post combined therapy (E), subendocardial fibrosis following doxorubicin treatment (F), subepicardial fibrosis following doxorubicin treatment (G), changes in subendocardial cardiomyocyte size after combined therapy (H), and changes in subepicardial cardiomyocyte size after combined therapy (I). High (depicted in brown) and low (depicted in blue) levels of fibrosis and cardiomyocyte size, defined by the median, are used to differentiate the mice. This figure complements the data presented in Table S3.

• 

  Figure 4. Identification of quantitative trait loci associated with Intermediate Molecular Phenotypes (ipQTLs) of CDA susceptibility. (A) F1BX1 mice, generated by backcrossing, were genotyped using 1499 SNVs on an Illumina platform, followed by linkage analysis to detect ipQTLs. (B) The heatmap illustrates the global distribution of ipQTLs post-doxorubicin or combined therapy. Each square denotes a chromosome, numbered on the left. The intensity of each square indicates the degree of association (LOD score) between the genetic markers and the phenotype, as per the provided scale. The position of the marks within each square corresponds to their relative location within each chromosome (centromere and telomere positions above and below, respectively). Only linkages with a LOD score > 1.5 (suggestive) are represented. R/qtl software was utilized to identify ipQTLs. The precise location of each ipQTL on each chromosome and the associated genetic markers are detailed in Table S6A (doxorubicin therapy) and Table S6B (combined therapy).

• 

  Figure 5. Identification of Quantitative Trait Loci associated with Anthracycline-Induced Cardiotoxicity (cda-QTLs). (A) Genotyping of the F1BX mouse cohort facilitated the localization of cdaQTLs; QTLs intrinsically connected to CDA susceptibility as determined via histopathological analysis. (B,C) The heatmaps provide a comprehensive depiction of the distribution of cda-QTLs associated with cardiac fibrosis and cardiomyocyte size under diverse conditions (age and therapy type): post-doxorubicin chemotherapy (B) or post-combined therapy (C). For both panels, (B,C), only cdaQTLs possessing a LOD score > 1.5 were charted on each chromosome. Analyses were executed using R/qtl software. Each square symbolizes a chromosome, numbered to the left. The intensity of the markings within each square represents the level of association (LOD score) between the genetic markers and the phenotype. The mark’s position corresponds to its relative location within the chromosome, with centromere and telomere positions represented above and below, respectively. The precise locations of each cda-QTL and their corresponding genetic markers can be found in Table S6. (D) When identifying cdaQTLs in older mice…

• 

  Figure 6. Percentage of global fibrosis explained by cda-QTL6 or cda-QTL13 and genetic models. (A) Genetic models between ipQTLs and cdaQTL are created to assess whether an ipQTL contributes to the phenotypic variation of CDA. If the model’s variability significantly exceeds that of the cdaQTL, the ipQTL would contribute to the phenotypic variation of CDA and susceptibility. (B) The diagram above shows the percentage of global fibrosis explained by cda-QTL6 post-doxorubicin treatment (red). The circles on the right denote the name of the cda-QTL6 under these conditions, difQTL1 (doxorubicin-induces fibrosis QTL1), which appears linked to global fibrosis in all mice, and odifQTL2 (old mice difQTL2), which seems to be associated with global fibrosis in all mice. Below is the percentage of global fibrosis explained by cda-QTL13 following treatment with doxorubicin (red) or combined therapy (yellow). The circles on the right show the other names of the cda-QTL13 for those conditions: ydifQTL2 (young mice doxorubicin-induced fibrosis QTL2), ycifQTL3 (young mice combined therapy-induced fibrosis QTL3), and cifQTL4 (combined therapy-induced fibrosis QTL3), the latter linked to global fibrosis in…

• 

  Figure 7. Genetic models of CDA risk generated by bootstrapping 100 times and LASSO multivariate regression. Each human cohort was randomly divided into 80% training and 20% testing sets. During model construction, logistic regression was employed in a bootstrapping strategy with a fixed sampling rate (80% evaluation and 20% validation) and numerous iterations (100). After bootstrapping, the SNVs, significantly associated with CDA in evaluation and validation data at least five times across 100 iterations, were selected to construct the Least Absolute Shrinkage and Selection Operator (LASSO) regression model on the 80% training set. After training, the best cutoff on the Receiver Operating Characteristic (ROC) curve was optimized based on the maximum Youden’s index (sensitivity + specificity − 1), and the LASSO models, along with the corresponding optimal cutoff, were applied to the 20% testing set for independent evaluation of the model performance. (A–C) CDA risk model for the cohort of breast cancer patients. The genetic model (A) and the ROC curves of the training set (B) and the testing set (C) are shown. (D–F) Genetic risk model for the pediatric patient cohort. The genetic model (D)…