A SMAD4-modulated gene profile predicts disease-free survival in stage II and III colorectal cancer

Abstract

Background: Colorectal cancer is the second-leading cause of cancer-related mortality in the United States and a leading cause of cancer-related mortality worldwide. Loss of SMAD4, a critical tumor suppressor and the central node of the transforming growth factor-beta superfamily, is associated with worse outcomes for colorectal cancer patients; however, it is unknown whether an RNA-based profile associated with SMAD4 expression could be used to better identify high-risk colorectal cancer patients.

Aim: Identify a gene expression-based SMAD4-modulated profile and test its association with patient outcome.

Methods and results: Using a discovery dataset of 250 colorectal cancer patients, we analyzed expression of BMP/Wnt target genes for association with SMAD4 expression. Promoters of the BMP/Wnt genes were interrogated for SMAD-binding elements. Fifteen genes were implicated and three tested for modulation by SMAD4 in patient-derived colorectal cancer tumoroids. Expression of the 15 genes was used for unsupervised hierarchical clustering of a training dataset and two resulting clusters modeled in a centroid model. This model was applied to an independent validation dataset of stage II and III patients. Disease-free survival was analyzed by the Kaplan-Meier method. In vitro analysis of three genes identified in the SMAD4-modulated profile (JAG1, TCF7, and MYC) revealed modulation by SMAD4 consistent with the trend observed in the profile. In the training dataset (n = 553), the profile was not associated with outcome. However, among stage II and III patients (n = 461), distinct clusters were identified by unsupervised hierarchical clustering that were associated with disease-free survival (p = .02, log-rank test). The main model was applied to a validation dataset of stage II/III CRC patients (n = 257) which confirmed the association of clustering with disease-free survival (p = .013, log-rank test).

Conclusions: A SMAD4-modulated gene expression profile identified high-risk stage II and III colorectal cancer patients, can predict disease-free survival, and has prognostic potential for stage II and III colorectal cancer patients.

Keywords: SMAD4; cancer biology; colorectal cancer; gene expression profile; tumor suppressor genes.

FIGURE 1

Study flow diagram. A discovery cohort of 250 colorectal cancer (CRC) patients from Vanderbilt University Medical Center (VUMC) and Moffit Cancer Center (MCC) was used to identify SMAD4‐modulated genes among known target genes of BMP and Wnt (see Figure 2 and Methods). An independent training cohort of 553 CRC patients was then used to identify a high‐risk group of stage II and III CRC patients. A risk prognosis model was developed and then validated in a separate, independent cohort of 257 stage II and III CRC patients

FIGURE 2

Experimental approach used to identify the SMAD4‐modulated gene profile. Lists of BMP and Wnt target genes were generated, and the overlap (48 probes) is shown in a Venn diagram. These overlapping genes/probes were tested for correlation with SMAD4 expression levels (Spearman correlation p < .005) and for SMAD‐binding elements in their promoters (see Methods). The SMAD4‐modulated gene profile was defined as genes/probes that passed either of these tests

FIGURE 3

Validation of crosstalk between SMAD4 and target genes. (A) Tumoroid line derived from a SMAD4 wild‐type (WT) CRC patient and a tumoroid line derived from a SMAD4 mutant (MUT) CRC patient. Western blot shows SMAD4 status and inverse correlations between SMAD4 and three target genes: JAG1, TCF7, and c‐Myc. (B) Crispr/Cas9 technology was used to deplete SMAD4 in CRC patient‐derived tumor organoids (tumoroids). Compared to non‐targeted control (NT), western blot shows 80% SMAD4 knockdown (KD) and an inverse correlation between SMAD4, JAG1, TCF7, and c‐Myc with engineered depletion of SMAD4. For A and B, each quantification was normalized to the loading control β‐Actin. The respective fold‐change after normalization is shown for each condition in A and B to the right of the western blot

FIGURE 4

The SMAD4‐modulated gene profile is not associated with DFS in stage I‐IV CRC patients. (A) In the training dataset of 553 stage I‐IV CRC patients, two distinct patient clusters (Cluster a, red; Cluster b, blue) were identified via unsupervised hierarchical clustering. Each row on the heat map represents a single probe in the mean‐centered gene profile, and each column represents an individual patient in the training dataset. (B) Kaplan‐Meier analysis showed no significant difference in DFS between the clusters (DFS probability at 5 years: cluster a 63.8% and cluster b 66.5%; p = .68). (C) In the subset of 461 stage II and III CRC patients in the training dataset, two distinct patient clusters (Cluster a, red; Cluster b, blue) were identified in unsupervised hierarchical clustering. Each row on the heat map represents a single probe in the mean‐centered gene profile, and each column represents an individual patient. (D) Kaplan‐Meier analysis showed a significantly higher DFS in Cluster b patients than in Cluster a patients (75% vs. 58% at 100 months; p = .02)

FIGURE 5

SMAD4‐modulated gene profile predicts DFS in stage II/III CRC patients in the validation dataset. A prediction model based on the SMAD4‐modulated gene profile identified 60 low‐ and 197 high‐risk patients in the validation dataset of stage II and III CRC patients (probability of DFS at 5 years: 67.7% in the high‐risk cluster vs. 84.6% in the low‐risk cluster; p = .013)