Gene-expression patterns predict phenotypes of immune-mediated thrombosis

Abstract

Antiphospholipid antibody syndrome (APS) is a complex autoimmune thrombotic disorder with defined clinical phenotypes. Although not all patients with elevated antiphospholipid antibody (aPLA) levels develop complications, the severity of these potential events mandates aggressive and extended lifelong anti-thrombotic therapy. One hundred twenty-nine patients (57 patients with APS and venous thromboembolism [VTE], 32 patients with VTE without aPLA, 32 patients with aPLA only, and 8 healthy patients) were enrolled. RNA from peripheral-blood collection was used for DNA microarray analysis. Patterns of gene expression that characterize APS as well as thrombosis in the presence of aPLA were identified by hierarchical clustering and binary regression methods. Gene-expression profiles identify and predict individuals with APS from patients with VTE without aPLA. Importantly, similar methods identified expression profiles that accurately predicted those patients with aPLA at high risk for thrombotic events. All profiles were validated in independent cohorts of patients. The ability to predict APS, but more importantly, those patients at risk for venous thrombosis, represents a paradigm for a genomic approach that can be applied to other populations of patients with venous thrombosis, providing for more effective clinical management of disease, while also reflecting the possible underlying biologic processes.

Figure 1.

Patterns of gene expression that characterize clinical phenotypes. Hierarchical clustering of the initial patient samples based on gene-expression patterns. Each gene is represented by a single row, and each sample is represented by a single column. The color heat map represents genes in a graded fashion along a spectrum of activation, extending from strongly up-regulated genes in red to the down-regulated genes in blue.

Figure 2.

Gene-expression profiles that classify and predict APS phenotype. (A) Expression profiles of genes that discriminate between APS and non-APS patients with VTE. Image depicts a group of 50 genes selected to differentiate APS from control. Genes are ordered top to bottom according to regression coefficient. (B) Leave-one-out cross-validation (CV) probabilities of individual samples in our training cohort (n = 57) of APS patients (red) and control patients with VTE without aPLA (blue). The values on the horizontal axis are estimates of the overall metagene score in the regression analysis. The corresponding values on the vertical axis are estimated classification probabilities with corresponding 95% probability intervals marked to indicate uncertainty about these estimated values. The horizontal dashed line represents an arbitrary cutoff value to demonstrate the accuracy of prediction for any given probability of the conditions being compared. (C) Validation of the binary regression model in a blinded independent cohort of 32 subjects (21 patients with APS [red] and VTE, 11 controls [blue] with VTE but without aPLAs).

Figure 3.

Validations of predictions of thrombosis. (A) Expression profiles of genes that classify and predict thrombosis. Image depicts a group of 50 genes selected to differentiate patients with APS (aPLA + VTE) from patients with asymptomatic aPLAs in the training (n = 54) and validation (n = 35) sets. (B) Leave-one-out cross-validation (CV) analysis in the training cohort (n = 54) of APS patients (red) and asymptomatic patients with aPLAs (blue). Details are the same as described in Figure 2. (C) Validation of the binary regression model in a blinded independent cohort of 35 subjects (21 patients with APS [red] and VTE, 14 patients with asymptomatic aPLAs [blue]). Details are the same as in Figure 2.