Limits...
Dysregulation of the vascular endothelial growth factor and semaphorin ligand-receptor families in prostate cancer metastasis.

Bender RJ, Mac Gabhann F - BMC Syst Biol (2015)

Bottom Line: We found pro-lymphangiogenic signatures, including the genes encoding VEGFC and VEGFD, associated with primary tumors that ultimately became aggressive.To leverage our mechanistic understanding, and to link multigene expression changes to outcomes, we performed individualized computational simulations of competitive VEGF and Sema receptor binding across many tumor samples.Therapeutic inhibition of angiogenesis in metastatic prostate cancer should account for both of these routes.

View Article: PubMed Central - PubMed

Affiliation: Department of Biomedical Engineering, Johns Hopkins University, Baltimore, MD, 21218, USA. bender.rj@gmail.com.

ABSTRACT

Background: The vascular endothelial growth factor (VEGF) family is central to cancer angiogenesis. However, targeting VEGF as an anti-cancer therapeutic approach has shown success for some tumor types but not others. Here we examine the expression of the expanded VEGF family in prostate cancer, including the Semaphorin (Sema) family members that compete with VEGFs for Neuropilin binding and can themselves have pro- or anti-angiogenic activity.

Results: First, we used multivariate statistical methods, including partial least squares and clustering, to examine VEGF/Sema gene expression variability in previously published prostate cancer microarray datasets. We show that unlike some cancers, such as kidney cancer, primary prostate cancer is characterized by both a down-regulation of the pro-angiogenic members of the VEGF family and a down-regulation of anti-angiogenic members of the Sema family. We found pro-lymphangiogenic signatures, including the genes encoding VEGFC and VEGFD, associated with primary tumors that ultimately became aggressive. In contrast to primary prostate tumors, prostate cancer metastases showed increased expression of key pro-angiogenic VEGF family members and further repression of anti-angiogenic class III Sema family members. Given the lack of success of VEGF-targeting molecules so far in prostate cancer, this suggests that the reduction in anti-angiogenic Sema signaling may potentiate VEGF signaling and even promote resistance to VEGF-targeting therapies. Inhibition of the VEGF 'accelerator' may need to be accompanied by promotion of the Sema 'brake' to block cancer angiogenesis. To leverage our mechanistic understanding, and to link multigene expression changes to outcomes, we performed individualized computational simulations of competitive VEGF and Sema receptor binding across many tumor samples. The simulations suggest that loss of Sema expression promotes angiogenesis by lowering plexin signaling, not by potentiating VEGF signaling via relaxation of competition.

Conclusions: The combined analysis of bioinformatic data with computational modeling of ligand-receptor interactions demonstrated that enhancement of angiogenesis in prostate cancer metastases may occur through two different routes: elevation of VEGFA and reduction of class 3 Semaphorins. Therapeutic inhibition of angiogenesis in metastatic prostate cancer should account for both of these routes.

No MeSH data available.


Related in: MedlinePlus

VEGF/Sema expression signatures predicting biochemical recurrence (BCR). a-b: PLS-DA scores/loadings plots for the TCGA (a) and GSE21034 (b) datasets. The training accuracies and the discriminant line separating the two classes are displayed c-d: Survival curves show prognostic significance of VEGF/Sema signatures in the TCGA (c) and GSE21034 (d) datasets. The p-values are from log rank tests of the Kaplan-Meier survival estimators for each group
© Copyright Policy - OpenAccess
Related In: Results  -  Collection

License 1 - License 2
getmorefigures.php?uid=PMC4559909&req=5

Fig2: VEGF/Sema expression signatures predicting biochemical recurrence (BCR). a-b: PLS-DA scores/loadings plots for the TCGA (a) and GSE21034 (b) datasets. The training accuracies and the discriminant line separating the two classes are displayed c-d: Survival curves show prognostic significance of VEGF/Sema signatures in the TCGA (c) and GSE21034 (d) datasets. The p-values are from log rank tests of the Kaplan-Meier survival estimators for each group

Mentions: To assess whether some subsets of primary prostate tumors may have differing potential benefits from VEGF signaling inhibitors, we used partial least squares discriminant analysis (PLS-DA), a multivariate algorithm that allowed us to simultaneously consider both the pattern of VEGF/Sema gene expression and effects on an output variable. As an output, we used a binary variable indicating whether biochemical recurrence (BCR) eventually occurred. Data for BCR/follow-up times were available in the TCGA and GSE21034 datasets, with 10 and 27 BCR events, respectively (Additional file 1: Table S3). Since BCR-negative samples with short follow-up times may eventually undergo recurrence, we used only the ten and 27 BCR-negative samples with the longest follow-up times in the TCGA and GSE21034 datasets, respectively. These groups of indolent samples had follow-up times greater than 3.3 years in the TCGA dataset and greater than 5.2 years in the GSE21034 dataset. These minimum indolent follow-up times were greater than all but one of the times at which BCR occurred in the TCGA dataset and all but three in the GSE21034 dataset. The TCGA PLS-DA model was effective in differentiating aggressive and indolent tumors, with a 92 ± 3 % training accuracy (Fig. 2a). The GSE21034 PLS-DA model, on the other hand, was less effective with only a 66 ± 1 % training accuracy (Fig. 2b). This discrepancy in performance between TCGA and GSE21034 datasets was seen for other biomarkers meant to distinguish aggressive from indolent tumors (Additional file 1: Figure S7). Notably, the VEGF/Sema PLS-DA model (and PLS-DA models based on the genes from other biomarkers) lost some of its predictive ability when correcting for Gleason score, but still was significantly prognostic (Additional file 1: Figure S8). ROC curves from leave-one-out cross-validation barely deviated from a 45-degree line (Additional file 1: Figure S6), indicating that these models would likely be ineffective in distinguishing aggressive and indolent tumors in other datasets. ROC curves for PLS-DA models trained with the 478 genes of the “angiome”, a set of angiogenesis-related genes [26], improved but were still fairly poor: the AUCs for TCGA and GSE21034 were 0.72 and 0.66, respectively (Additional file 1: Table S6). The 1,233 genes of the extended angiome [26] also led to poor PLS-DA models, with AUCs of 0.74 and 0.69 (Additional file 1: Table S7). Nonetheless, the PLS-DA models provided gene expression signatures with potential prognostic significance: when the PLS discriminant scores were used as predictive variables in Kaplan-Meier survival analysis, aggressive and indolent tumors had significantly different outcomes (Fig. 2c, d).Fig. 2


Dysregulation of the vascular endothelial growth factor and semaphorin ligand-receptor families in prostate cancer metastasis.

Bender RJ, Mac Gabhann F - BMC Syst Biol (2015)

VEGF/Sema expression signatures predicting biochemical recurrence (BCR). a-b: PLS-DA scores/loadings plots for the TCGA (a) and GSE21034 (b) datasets. The training accuracies and the discriminant line separating the two classes are displayed c-d: Survival curves show prognostic significance of VEGF/Sema signatures in the TCGA (c) and GSE21034 (d) datasets. The p-values are from log rank tests of the Kaplan-Meier survival estimators for each group
© Copyright Policy - OpenAccess
Related In: Results  -  Collection

License 1 - License 2
Show All Figures
getmorefigures.php?uid=PMC4559909&req=5

Fig2: VEGF/Sema expression signatures predicting biochemical recurrence (BCR). a-b: PLS-DA scores/loadings plots for the TCGA (a) and GSE21034 (b) datasets. The training accuracies and the discriminant line separating the two classes are displayed c-d: Survival curves show prognostic significance of VEGF/Sema signatures in the TCGA (c) and GSE21034 (d) datasets. The p-values are from log rank tests of the Kaplan-Meier survival estimators for each group
Mentions: To assess whether some subsets of primary prostate tumors may have differing potential benefits from VEGF signaling inhibitors, we used partial least squares discriminant analysis (PLS-DA), a multivariate algorithm that allowed us to simultaneously consider both the pattern of VEGF/Sema gene expression and effects on an output variable. As an output, we used a binary variable indicating whether biochemical recurrence (BCR) eventually occurred. Data for BCR/follow-up times were available in the TCGA and GSE21034 datasets, with 10 and 27 BCR events, respectively (Additional file 1: Table S3). Since BCR-negative samples with short follow-up times may eventually undergo recurrence, we used only the ten and 27 BCR-negative samples with the longest follow-up times in the TCGA and GSE21034 datasets, respectively. These groups of indolent samples had follow-up times greater than 3.3 years in the TCGA dataset and greater than 5.2 years in the GSE21034 dataset. These minimum indolent follow-up times were greater than all but one of the times at which BCR occurred in the TCGA dataset and all but three in the GSE21034 dataset. The TCGA PLS-DA model was effective in differentiating aggressive and indolent tumors, with a 92 ± 3 % training accuracy (Fig. 2a). The GSE21034 PLS-DA model, on the other hand, was less effective with only a 66 ± 1 % training accuracy (Fig. 2b). This discrepancy in performance between TCGA and GSE21034 datasets was seen for other biomarkers meant to distinguish aggressive from indolent tumors (Additional file 1: Figure S7). Notably, the VEGF/Sema PLS-DA model (and PLS-DA models based on the genes from other biomarkers) lost some of its predictive ability when correcting for Gleason score, but still was significantly prognostic (Additional file 1: Figure S8). ROC curves from leave-one-out cross-validation barely deviated from a 45-degree line (Additional file 1: Figure S6), indicating that these models would likely be ineffective in distinguishing aggressive and indolent tumors in other datasets. ROC curves for PLS-DA models trained with the 478 genes of the “angiome”, a set of angiogenesis-related genes [26], improved but were still fairly poor: the AUCs for TCGA and GSE21034 were 0.72 and 0.66, respectively (Additional file 1: Table S6). The 1,233 genes of the extended angiome [26] also led to poor PLS-DA models, with AUCs of 0.74 and 0.69 (Additional file 1: Table S7). Nonetheless, the PLS-DA models provided gene expression signatures with potential prognostic significance: when the PLS discriminant scores were used as predictive variables in Kaplan-Meier survival analysis, aggressive and indolent tumors had significantly different outcomes (Fig. 2c, d).Fig. 2

Bottom Line: We found pro-lymphangiogenic signatures, including the genes encoding VEGFC and VEGFD, associated with primary tumors that ultimately became aggressive.To leverage our mechanistic understanding, and to link multigene expression changes to outcomes, we performed individualized computational simulations of competitive VEGF and Sema receptor binding across many tumor samples.Therapeutic inhibition of angiogenesis in metastatic prostate cancer should account for both of these routes.

View Article: PubMed Central - PubMed

Affiliation: Department of Biomedical Engineering, Johns Hopkins University, Baltimore, MD, 21218, USA. bender.rj@gmail.com.

ABSTRACT

Background: The vascular endothelial growth factor (VEGF) family is central to cancer angiogenesis. However, targeting VEGF as an anti-cancer therapeutic approach has shown success for some tumor types but not others. Here we examine the expression of the expanded VEGF family in prostate cancer, including the Semaphorin (Sema) family members that compete with VEGFs for Neuropilin binding and can themselves have pro- or anti-angiogenic activity.

Results: First, we used multivariate statistical methods, including partial least squares and clustering, to examine VEGF/Sema gene expression variability in previously published prostate cancer microarray datasets. We show that unlike some cancers, such as kidney cancer, primary prostate cancer is characterized by both a down-regulation of the pro-angiogenic members of the VEGF family and a down-regulation of anti-angiogenic members of the Sema family. We found pro-lymphangiogenic signatures, including the genes encoding VEGFC and VEGFD, associated with primary tumors that ultimately became aggressive. In contrast to primary prostate tumors, prostate cancer metastases showed increased expression of key pro-angiogenic VEGF family members and further repression of anti-angiogenic class III Sema family members. Given the lack of success of VEGF-targeting molecules so far in prostate cancer, this suggests that the reduction in anti-angiogenic Sema signaling may potentiate VEGF signaling and even promote resistance to VEGF-targeting therapies. Inhibition of the VEGF 'accelerator' may need to be accompanied by promotion of the Sema 'brake' to block cancer angiogenesis. To leverage our mechanistic understanding, and to link multigene expression changes to outcomes, we performed individualized computational simulations of competitive VEGF and Sema receptor binding across many tumor samples. The simulations suggest that loss of Sema expression promotes angiogenesis by lowering plexin signaling, not by potentiating VEGF signaling via relaxation of competition.

Conclusions: The combined analysis of bioinformatic data with computational modeling of ligand-receptor interactions demonstrated that enhancement of angiogenesis in prostate cancer metastases may occur through two different routes: elevation of VEGFA and reduction of class 3 Semaphorins. Therapeutic inhibition of angiogenesis in metastatic prostate cancer should account for both of these routes.

No MeSH data available.


Related in: MedlinePlus