Background

Colorectal cancer (CRC) is ranked third in cancer incidence and second in cancer-related mortality worldwide [1]. CRC is a heterogeneous disease, and recent data from gene expression profiling have classified CRC into four consensus molecular subtypes (CMSs) [2]. In addition, integrated analysis of somatic mutations, copy number change, and mRNA expression from the Cancer Genome Atlas (TCGA) revealed that the WNT, P53, RTK-RAS, TGF-β, and PI3K pathways are frequently altered in CRC [3, 4].

The current standard care in stage III CRC is complete resection of the tumor followed by adjuvant chemotherapy with 5-fluorouracil, leucovorin, and oxaliplatin (FOLFOX) or capecitabine plus oxaliplatin (XELOX) [5, 6]. However, approximately 30% of patients develop recurrence despite receiving adjuvant chemotherapy [7]. Several biomarkers, including microsatellite instability-high (MSI-H), the CpG island methylator phenotype (CIMP), KRAS mutation, and BRAF mutation, are associated with recurrence and survival in CRC patients [8,9,10,11]. Although the prognostic biomarkers associated with cancer recurrence are well studied, there is a paucity of data regarding biomarkers associated with survival after recurrence (SAR). Recently, from a study population of stage III CRC patients enrolled in a phase III adjuvant chemotherapy study, MSI-H was associated with improved SAR, and KRAS mutation and BRAF mutation were associated with poor SAR [12].

We have previously reported that mutation in genes within PI3K pathway is associated with better RFS and that mutation in genes within RTK-RAS pathway is associated with worse RFS in CRC treated with adjuvant chemotherapy [13]. The purpose of this study was to evaluate the association between five critical pathway mutations and SAR in CRC patients who had distant recurrence after the treatment of curative surgery followed by adjuvant fluoropyrimidine and oxaliplatin chemotherapy.

Methods

Patients and treatment

The study population consisted of patients with stage III or high-risk stage II CRC who developed distant recurrence after receiving curative surgery at Seoul National University Hospital (SNUH, Seoul, South Korea). All patients were included in the CRC patient cohort reported previously [13]. Patients received at least 6 cycles of adjuvant FOLFOX or 4 cycles of adjuvant XELOX chemotherapy. Upper rectal cancer patients were included if they did not receive pre- or post-operative radiation therapy. Exclusion criteria were the following: signet ring cell histology, anti-EGFR or anti-VEGF treatment for adjuvant chemotherapy, history of other malignancy within 5 years, and local recurrence only. Adjuvant FOLFOX was planned for 12 cycles and XELOX chemotherapy was planned for 8 cycles. 50 patients received FOLFOX-4, 27 received FOLFOX-6, and 10 received XELOX. Electronic database and electronic medical record system of SNUH was used to identify eligible patients and review their medical charts.

Molecular testing, including targeted sequencing of 40 genes associated with five critical pathways

Every exon of the 40 genes associated with the five critical pathways of CRC was sequenced [13]. Fourteen genes were selected from WNT pathway (ARID1A, AMER1, APC, AXIN2, CTNNB1, DKK1, DKK2, DKK3, DKK4, FBXW7, FZD10, LRP5, SOX9, TCF7L2), 2 genes from P53 pathway (ATM, TP53), 8 genes from RTK-RAS pathway (BRAF, EGFR, ERBB2, ERBB3, ERBB4, HRAS, KRAS, NRAS), 7 genes from TGF-β pathway (ACVR1B, ACVR2A, SMAD2, SMAD3, SMAD4, TGFBR1, TGFBR2), and 9 genes from PI3K pathway (IGF1R, IGF2, IRS2, MTOR, PDGFRA, PIK3CA, PIK3R1, PTEN, SRC). Detailed methods for targeted sequencing can be found in our previous article [13]. Briefly, genomic DNA (> 200 ng) samples were sheared and prepared according to routine library preparation. The captured library was amplified and sequenced using Hiseq 2500 (Illumina, USA). Sequencing data were filtered with a mean quality Q20 (Phred score) per read, and these filtered data were aligned to GRCh37 using bwa 0.7.5a. The aligned reads were processed with Picard Mark Duplicates and GATK base recalibration. After a series of processes, the aligned bases were piled up with SAM tools. Variant call and somatic analysis processes were performed by Varscan and were annotated with ANNOVAR. The pathways were defined as having a mutation if any gene included in the pathway had a mutation.

The microsatellite status was assessed using the following markers: BAT25, BAT26, D2S123, D5S346 and D17S250 [13]. Instability at 2 or more markers were defined as MSI-high, 1 as MSI-low, and 0 as MSS. MSI-H was regarded as having MSI, and MSI-L was grouped with MSS.

Statistical analysis

The primary objective of this study was to investigate the effect of mutations of critical pathways and clinico-pathologic characteristics on SAR. SAR was defined as the time from distant tumor recurrence to death from any cause. Categorical variables were compared by chi-square test or Fisher’s exact test. Continuous variables were compared using the independent-samples T test. SAR was calculated using the Kaplan-Meier method, and comparisons were made using log-rank tests. Hazard ratios (HRs) were calculated using the Cox proportional hazard model. To adjust for the baseline characteristics, Cox proportional hazard analysis of SAR included sex, age (continuous variable), tumor location [proximal (from the cecum to the transverse colon) vs. distal (from the descending colon to the rectum)], tumor histology (mucinous adenocarcinoma vs. non-mucinous adenocarcinoma), tumor (T) stage (continuous variable), lymph node (N) stage (continuous variable), curative operation (done vs. not done), recurrence before 1 year, microsatellite status (MSI-high vs. MSS and MSI-low), and five critical pathway mutations. Two-sided p-values less than 0.05 were considered statistically significant. Statistical analysis was performed using SPSS software for Windows, version 20.0 (SPSS, Chicago, IL, USA).

Results

Patient characteristics

At a median follow-up duration of 44.4 months, 87 patients had distant recurrence among our database of 516 patients [13]. The baseline characteristics of the 87 patients are summarized in Table 1. The primary tumor location was proximal (from the cecum to the transverse colon) in 28 patients (32.2%) and distal (from the descending colon to the rectum) in 59 patients (67.8%). Nine patients (10.3%) had MAC histology, and 3 patients (3.5%) showed MSI-H feature.

Table 1 Baseline characteristics
Full size table

After distant recurrence, palliative chemotherapy was introduced in 77 (88.5%) patients. The first-line chemotherapy regimens included FOLFIRI in 56 patients, capecitabine monotherapy in 8, FOLFIRI plus bevacizumab in 7, FOLFIRI plus cetuximab in 2, FOLFOX in 2, S-1 plus oxaliplatin in 1, and capecitabine plus irinotecan in 1. Second-line chemotherapy was administered to 47 (54.0%) patients. Only 1 patient received both bevacizumab and cetuximab during the course of treatment. In total, 9 (10.3%) patients received bevacizumab and 9 (10.3%) patients received cetuximab during their course of treatment.

Gene mutation and pathway mutation

In patients with recurrence, mutations were most frequently found in TP53 (65.5%) followed by APC (64.4%), KRAS (48.3%), FBXW7 (16.1%), and SMAD4 (10.3%) (Additional file 1: Table S1). The WNT pathway was mutated in 79.3%, P53 in 69.0%, RTK-RAS in 65.5%, TGF-β in 24.1%, and PI3K in 21.8% of patients. The mutation frequencies of most genes were similar between the patients with recurrence and without recurrence (Additional file 1: Table S1). The mutation frequency of PIK3CA and CTNNB1 was lower in the patients with recurrence, and APC showed a tendency of lower frequency in patients with recurrence. PI3K pathway mutation showed a tendency of lower frequency in patients with recurrence.

Patients with mutation in genes within TGF-β pathway showed distinct characteristics compared with patients without TGF-β pathway mutation (Table 1). Patients with TGF-β pathway mutation was associated with a younger age (median age of 54 years vs. 61 years, p = 0.020), and there was a tendency of a higher percentage of patients with initial high-risk stage II disease (28.6% vs. 12.1%, p = 0.074). Moreover, patients with TGF-β pathway mutation had a higher incidence of MSI-H and MAC. All 3 patients with recurrence in MSI-H had TGF-β pathway mutation. In addition, the incidence of TGF-β pathway mutation was 88.9% (8/9) in MAC patients compared with 16.7% (13/78) in non-MAC patients (p <  0.001) (Additional file 1: Table S2). The mutation incidences of other pathways were similar between MAC and non-MAC histology (Additional file 1: Table S2).

Prognostic role of five critical pathways

Among the 87 patients who had distant recurrence, 40 death events occurred with an estimated median SAR of 38.2 months. SAR was significantly worse in patients with TGF-β pathway mutation than in those with wild type (median SAR of 21.6 vs. 44.4 months, p = 0.021) (Fig. 1). In our previous report, mutation in genes within PI3K pathway was associated with a longer RFS, whereas mutation in genes within RTK-RAS pathway was associated with a shorter RFS [13]. However, there was no correlation between SAR and other critical pathway mutations (Additional file 2: Figure S1). In addition to TGF-β pathway mutation, curative resection after distant recurrence and MAC was associated with poor SAR, and N2 stage showed a tendency of poor SAR (Table 2). Tumor location and MSI were not associated with SAR. Multivariate analysis using the Cox proportional hazard model revealed that curative resection after distant recurrence [adj HR for SAR 0.27 (95% CI 0.13–0.56), p = 0.001] and MAC histology [adj HR for SAR 3.77 (95% CI 1.52–9.35), p = 0.004] were independent prognostic factors for SAR. In the present study, MAC histology showed a higher incidence of TGF-β pathway mutation compared to non-MAC histology (88.9% vs. 16.7%, p <  0.001) (Additional file 1: Table S2). Although statistically not significant, the negative prognostic role of TGF-β pathway mutation was maintained in patients with non-MAC histology (median SAR of 36.7 vs. 44.4 months, p = 0.127) (Fig. 2). After removing MAC histology as a covariate, multivariate analysis revealed TGF-β pathway mutation [adj HR for SAR 1.99 (95% CI 1.02–3.90), p = 0.044] and curative resection after distant recurrence [adj HR for SAR 0.28 (95% CI 0.13–0.59), p = 0.001] as independent prognostic factors for SAR.

Fig. 1
figure 1

TGF-β pathway mutation and survival after recurrence

Full size image
Table 2 Univariate analysis of survival after recurrence
Full size table
Fig. 2
figure 2

TGF-β pathway mutation and survival after recurrence in Non-MAC patients. Abbreviations: MAC, mucinous adenocarcinoma

Full size image

We next analyzed whether individual gene mutation is associated with SAR (Additional file 1: Table S2). Patients with SMAD4 mutation had worse SAR compared to wild type (SAR of 17.6 months vs. 38.7 months, p = 0031) and those with APC mutation had a tendency of favorable SAR compared to wild type (SAR of 45.6 months vs. 28.1 months, p = 0070). In the multivariate analysis, SMAD4 mutation was an independent negative prognostic factor for SAR [adjusted hazard ratio (adj HR) for SAR 3.54 (95% CI 1.13–11.03), p = 0.030]. The detailed mutation profile of TGF-β pathway is shown in Table S4 (Additional file 1: Table S4). Among 25 mutations, 5 were stop gain mutation, 4 were frame shift deletion mutation, and 2 were non-frameshift deletion mutation which would probably lead to loss of expression. Impact of 14 non-synonymous SNV in TGF-β pathway mutation was predicted using the PROVEAN, SIFT, and PolyPhen-2 (Additional file 1: Table S2) [14,15,16]. Among 14 non-synonymous SNV, 12(85.7%)-13(92.9%) were predicted to be damaging mutation (Additional file 1: Table S4). In total, 23(92.0%)-24(96.0%) mutations were predicted to be damaging mutation in the TGF-β pathway. We also predicted functional outcome of SMAD4 mutations using the OncoKB database (Additional file 1: Table S4) [17]. Among 9 SMAD4 mutations, 3 mutations were found to be loss-of-function and additional 2 mutations involved the same location with aforementioned loss-of-function mutation. ACVR1B and ACVR2A genes were not covered by the OncoKB and our mutation involving SMAD2, SMAD3, TGFBR1, and TGFBR2 was not reported in the OncoKB.

Discussion

This study revealed that mutation in genes within TGF-β pathway is associated with MAC and may have poor SAR in colorectal cancer patients. Other critical pathways were not associated with SAR. In our previous study, mutation in genes within TGF-β pathway was not associated with RFS [13]. This implicates that while TGF-β pathway mutation is not associated with recurrence in early-stage tumors, it may be fatal in later-stage tumors.

In the present study, patients with TGF-β pathway mutation had higher incidence of MAC. This is in line with data from the TCGA where TGF-β pathway is more frequently altered in MAC histology compared to non-MAC histology (67.9% vs. 30.6%, p <  0.001) [18]. The TGF-β pathway is associated with cell proliferation, cell differentiation, apoptosis, and epithelial-mesenchymal transition (EMT) [19, 20]. Although mutational inactivation of TGF- β pathway is frequently shown in colorectal cancer, the level of TGF-b production is somehow elevated in later stage tumor [21, 22]. One of the reasons is that TGF-b level is not only affected by tumor but is also affected by stromal cells [22, 23]. The correlation between TGF- β pathway mutation (which is usually loss-of function) and TGF-b level has not been clearly identified and maybe different among disease status. In terms of clinical relevance, there are reports showing that TGF-β pathway may have a prognostic role in CRC. CMS classifies colorectal cancer based on the gene expression profile [2]. CMS4, which is characterized by the upregulation of genes implicated in EMT, prominent TGF-β pathway activation, stromal invasion, and angiogenesis, is associated with a worse prognosis. In addition, preclinical studies show that activation of TGF-β pathway may be linked to chemotherapy resistance in CRC cell line [24, 25]. Recombinant TGF-β treatment caused chemotherapy resistance in CRC cell lines [24]. However, TGF-β pathway mutation was not associated with RFS in CRC patients treated with adjuvant FOLFOX/XELOX chemotherapy, which we have previously reported [13]. Discordance in the prognostic role of TGF-β pathway mutation among disease status could be partially explained by the fact that the response to TGF-β pathway activation is different according to cell type and status [20, 26]. There are opposite effects of the TGF-β pathway in that it may suppress the growth of tubular adenoma but may promote the tumor growth of sessile serrated adenomas by inducing EMT and metastasis [20, 26]. Apoptosis is the dominant feature after TGF-β pathway stimulation in classical tubular adenoma organoid culture [26]. By contrast, inducing EMT is the main outcome after TGF-β pathway stimulation in an organoid culture model for sessile serrated adenomas [26]. The activation of TGF-β may suppress the growth of tubular adenoma but may progress sessile serrated adenomas to CMS4 subtype cancer by inducing EMT [26]. Moreover, evidence show that TGF-b level is also affected by stromal cells [22]. While TGF-β pathway mutation is not associated with RFS in CRC patients treated with adjuvant FOLFOX/XELOX chemotherapy, it may be fatal in CRC patients who have recurred after curative surgery. Among the genes involved in the TGF-β pathway, SMAD4 mutation was associated with poor SAR. SMAD4 is a tumor suppressor gene and its mutation is involved in advanced stages, such as distant metastasis, in human colorectal carcinogenesis [27]. In the present study, we predicted outcome of TGF-β pathway mutation using the PROVEAN, SIFT, and PolyPhen-2. Among 14 non-synonymous SNV in TGF-β pathway mutation 12(85.7%)-13(92.9%) were predicted to be damaging mutation. Other 11 mutations consisted of 5 stop gain mutation, 4 frame shift deletion mutation, and 2 non-frameshift deletion mutation which would probably lead to loss of expression. In sum, 23(92.0%)-24(96.0%) mutations were predicted to be damaging mutation in the TGF-β pathway. In addition, all SMAD4 mutations reported in the OncoKB database were loss-of-function mutation. This shows that TGF-β pathway mutation may have contributed to poor SAR due to loss of function. The correlation between TGF-β pathway mutation, TGF-b expression, and chemotherapy resistance needs to be studied in the future.

Sinicrope et al. reported that MSI-H is associated with improved SAR and that KRAS mutation and BRAF mutation are associated with poor SAR in stage III colon cancer patients treated with adjuvant chemotherapy [12]. However, we could not evaluate the role of MSI-H and BRAF mutation in SAR because there were only 3 patients with MSI-H and 5 patients with BRAF V600E mutation. In contrast to a study by Sinicrope et al., RTK-RAS pathway mutation and KRAS exon 2 mutation were not associated with SAR in the present study. In a study by Sinicrope et al., the negative prognostic role of KRAS mutation was limited to patients who received adjuvant FOLFOX pus cetuximab but not in patients treated with adjuvant FOLFOX alone. In the present study, none of the patients received cetuximab in the adjuvant setting. This implicates that KRAS mutation may not affect SAR in patients treated with adjuvant FOLFOX or XELOX alone.

The major limitation of this study was that we did not evaluate genetic characteristics other than targeted sequencing of genes associated with 5 critical pathways. Although major genes included in the TGF-β pathway were sequenced, other alterations, including copy number variation or the expression level, were not analyzed. It is also important to analyze the TGF-β pathway induced by stromal cells [3, 23]. Therefore, comprehensive analysis of the TGF-β pathway could provide a better understanding of CRC biology in the future. Another limitation of the present study was that patients were treated differently after recurrence. However, this study was performed at a high-volume single center where the treatment plan was relatively uniform. The last limitation was the relatively small number of patients included in this cohort. However, this was the first study to evaluate the relationship between pathway mutation and SAR in CRC patients.

Conclusions

This study revealed that CRC patients with mutation in genes within TGF-β pathway may have poor SAR. Other pathway mutations were not associated with SAR. Future works on the relationship between TGF-β pathway mutation and survival according to disease status needs to be studied.