Prediction of Adrenocortical Carcinoma Relapse and Prognosis with a Set of Novel Multigene Panels

Abstract

Effective assessment of adrenocortical carcinoma (ACC) prognosis is critical in patient management. We report four novel and robust prognostic multigene panels. Sig27var25, SigIQvar8, SigCmbnvar5, and SigCmbn_B predict ACC relapse at area under the curve (AUC) of 0.89, 0.79, 0.78, and 0.80, respectively, and fatality at AUC of 0.91, 0.88, 0.85, and 0.87, respectively. Among their 33 component genes, 31 are novel. They could be differentially expressed in ACCs from normal tissues, tumors with different severity (stages and lymph node metastasis), ACCs with TP53 mutations, and tumors with differentially expressed immune checkpoints (CTLA4, PD1, TGFBR1, and others). All panels correlate with reductions of ACC-associated CD8+ and/or NK cells. Furthermore, we provide the first evidence for the association of mesenchymal stem cells (MSCs) with ACC relapse (p = 2 × 10^-6) and prognosis (p = 2 × 10^-8). Sig27var25, SigIQvar8, SigCmbnvar5, and SigCmbn_B correlate with MSC (spearman r ≥ 0.53, p ≤ 1.38 × 10^-5). Sig27var25 and SigIQvar8 were derived from a prostate cancer (PC) and clear cell renal cell carcinoma (ccRCC) multigene signature, respectively; SigCmbnvar5 and SigCmbn_B are combinations of both panels, revealing close relationships of ACC with PC and ccRCC. The origin of these four panels from PC and ccRCC favors their prognostic potential towards ACC.

Keywords: adrenocortical carcinoma; disease-free survival; immune checkpoint proteins; mesenchymal stem cells; overall survival; prognostic biomarkers.

Figure 1

Effective prediction of poor OS of ACC by Sig27var25 and SigIQvar8. (A) Hazard ratio (HR), 95% confidence interval (CI), and p values for the indicated multigene signatures in predicting poor OS. (B) Stratification of fatality risk by Sig27 and SigIQGAP1NW. Cutoff points were estimated by Maximally Selected Rank Statistics. Kaplan–Meier curves were constructed using the R survival package. Statistical analyses were performed using logrank test. (C,D) Separation of ACCs with high fatality risk from those with low risk by SigIQvar8 (C) and Sig27var25 (D). Cutoff points were estimated by Maximally Selected Rank Statistics. (E,F) Stratification of fatality risk with the indicated gene panels using cutoff points estimated by maxstat (Maximally Selected Rank Statistics), empirical, kernel, and normal methods using the R cutpointr package. (G,H) ROC and PR curves to evaluate the performance of Sig27var25 and SigIQvar8 in fatality risk stratification. The curves were produced using the PRROC package in R. (I) Time-dependent ROC-AUC for the indicated multigene panels. Error bands are for standard error (SE). Time-dependent ROC-AUC values were obtained using the R timeROC package.

Figure 2

Estimation of ACC progression by Sig27var25 and SigIQvar8. (A,B) Waterfall plots for Sig27var25 (A) and SigIQvar8 in stratification of ACC progression risk. The progression status, sensitivity, and specificity of the risk separation are indicated. Cutoff points were estimated using the empirical methods with n = 1000 bootstraps and used as the baselines for waterfall plot generation using R. (C,D) Separation of ACCs with a high risk of progression from those with low risk by the indicated signatures. Methods used in cutpoint estimation, cutpoints and the respective sensitivity and specificity are indicated. (E,F) ROC and PR curves for the indicated multigene panels.

Figure 3

Clinical factor independent prediction of ACC progression and poor prognosis by Sig27var25 and SigIQvar8. UV: univariate Cox analysis; MV: multivariate Cox analysis. MV included age at diagnosis and stages with stages I and II were grouped into “0” and stages III and IV were classified as “1”.

Figure 4

Estimation of ACC progression and fatality risks by SigCmbnvar5. (A) HR, 95% CI, and the respective p-values for SigCmbnvar5-derived prediction of OS and PFS under both univariate (UV) and multivariate (MV) settings. MV includes age at diagnosis and tumor stages. (B,C) Stratification of ACCs fatality (B) and progression risk (C) with the indicated cutoff points. (D,E) ROC (D) and PR curves (E) for discrimination of OS and PFS.

Figure 5

Comparison of progression and fatality risk stratification among Sig27var25, SigIQvar8, SigCmbnvar5, SigCmbn_B, and Sigpub_BP. (A,B) Cutoff points for the indicated signature scores were estimated by the empirical method. The individual survival curves, p-values, sensitivities, and specificities for OS and PFS are shown. (C,D) OS ROC and PR curves (C) and PFS ROC and PR curves (D) for the indicated signatures.

Figure 6

Differential expression of signature component genes. Gene expression was determined using RNA-seq data within the TCGA database organized by the GEPIA2 website [28]. * p < 0.05.

Figure 7

Associations of component gene expressions with tumor stage (A), lymph node metastasis (B), and TP53 mutations (C). Analyses were performed using the TCGA data organized by the UALCAN platform [29]. * p < 0.05; ** p < 0.01; *** p < 0.001 in comparison to stage 1 (A), N0 (B), and TP53-Nonmutant tumors. $ p < 0.05; $$$ p < 0.001 in comparison to stage 3 ACCs (A).

Figure 8

Correlations of the indicated signature component genes (x axis) with the indicated immunosuppressive factors (y axis).

Figure 9

Correlations of multigene signatures with reductions of CD8+ and NK cells in ACC. (A) Immune cells were profiled using Xcell within the MSDIC R package, followed by the determination of the correlation of CD8+ T cells with Sig27var25 scores using the ggpubr R package (left panel). The Sig27var25 negative and positive ACC (boxplot, right panel) were defined with the cutoff point estimated using an empirical method (right panel). (B,C) NK CD56bright and cytotoxic cells in ACCs were profiled using ssGSEA within the MSDIC R package. (D) Scores from the indicated multigene signatures, NK CD56bright cells, CD8+ T cells, and cytotoxic cells were used to construct the Spearman correlation image with the corrplot R package. Correlations with p < 0.01 are included. * p < 0.05; ** p < 0.01.

Figure 10

Correlation of multigene signatures with MSC. (A) MSCs in ACC were profiled using xCell within the MSDIC R package. Cutoff points for OS and PFS were estimated using Maximally Selected Rank Statistics, which were used to construct the survival curves. (B) Correlation of MSC with Sig27var25 score (left panel) and enrichment of MSCs in Sig25var25 positive ACCs (right panel). The positive and negative statuses were defined according to the empirically derived cutoff point. (C) Correlations of MSC with the indicated signatures were determined by Spearman correlation; all correlations are at p < 0.01.