TWIST1 polymorphisms predict survival in patients with metastatic colorectal cancer receiving first-line bevacizumab plus oxaliplatin-based chemotherapy

Abstract

The epithelial-mesenchymal transition (EMT) is an important mechanism of resistance to angiogenesis inhibition. The ability of EMT pathway genetic variants to predict the efficacy of anti-angiogenic therapy is unknown. We analyzed associations between functional single nucleotide polymorphisms (SNPs) in EMT-related genes and outcomes in metastatic colorectal cancer (mCRC) patients undergoing first-line bevacizumab-based chemotherapy. A total of 220 mCRC patients were included in this study: 143 patients treated with first-line bevacizumab-based chemotherapy (bevacizumab cohort) and 77 patients treated with cetuximab-based chemotherapy (cetuximab cohort). SNPs in TWIST1 (rs2285682, rs2285681), ZEB1 (rs10826943, rs2839658), SNAIL (rs1543442, rs4647958), and E-cadherin (rs16260) genes were analyzed by PCR-based direct sequencing. Patients carrying a TWIST1 rs2285682 G allele had a significantly longer median PFS of 18.1 months and OS of 44.1 months compared to those with the T/T genotype, who had a median PFS of 13.3 months (HR, 0.57; P=0.003) and OS of 29.2 months (HR, 0.53; P=0.001) in the bevacizumab cohort. In multivariate analysis, associations between TWIST1 rs2285682 and PFS and OS remained significant. Among women, the G allele of TWIST1 rs2285682 (PFS HR, 0.39, P=0.007; OS HR, 0.30 P=0.001) and TWIST1 rs2285681 (PFS HR, 0.27, P<0.001; OS HR, 0.25 P<0.001) was associated with improved survival. No significant associations were found in the cetuximab cohort. Our findings suggest that TWIST1 polymorphisms are associated with survival in mCRC patients treated with first-line bevacizumab-based chemotherapy and may serve as clinically useful biomarkers for anti-angiogenic therapy.

INTRODUCTION

The epithelial-mesenchymal transition (EMT) is a critical pathway of escape from angiogenesis inhibition. Bevacizumab-induced intratumoral hypoxia results in activation of the EMT proteins [1], SNAIL [2], TWIST1 [3], and ZEB1 [4], which orchestrate the loss of cell-cell adhesion, increased cell motility and evasion from cell cycle arrest [3, 4]. In preclinical models, anti-angiogenic therapy has been shown to increase the proportion of stem-like cancer cells [5], and bevacizumab-treated tumor tissue contains significantly more tumor cells which stain positive for these EMT markers [6].

Hypoxia also drives the differentiation of tumor cells into endothelial cells, which mediate tumor-induced angiogenesis and resistance to anti-VEGF therapy [7, 8]. EMT transcriptional repressors, particularly TWIST1, have been implicated in regulating the formation of tumor-derived endothelial cells [9]. While the role of EMT in promoting angiogenesis and tumor progression is known, the role of EMT polymorphisms to predict anti-angiogenic treatment efficacy has been largely unexplored.

We hypothesized that EMT-related genetic variants may predict bevacizumab efficacy in metastatic colorectal cancer (mCRC) patients. The goal of this study was to investigate the potential role of single nucleotide polymorphisms (SNPs) in EMT-related genes as outcome biomarkers in mCRC patients treated with bevacizumab-based chemotherapy.

MATERIALS AND METHODS

Patient population and study design

The bevacizumab cohort (training set) consisted of 143 mCRC patients treated with first-line FOLFOX or XELOX plus bevacizumab at the Cancer Institute Hospital in Japan. Within this cohort, 64% of patients received FOLFOX, with bevacizumab 5 mg/kg administered intravenously (day 1), followed by oxaliplatin 85 mg/m² administered with folinic acid (leucovorin) 200 mg/m² intravenously, 5-fluorouracil 400 mg/m² bolus infusion, and 5-fluorouracil 2400 mg/m² administered as a 48-h continuous infusion. This regimen was repeated at two-week intervals. The remaining 36% of patients in this cohort received XELOX, with bevacizumab 10 mg/kg administered intravenously (day 1), followed by oxaliplatin 150 mg/m² intravenously, and oral capecitabine (1000 mg/m²) given on days 1–14. This regimen was repeated at three-week intervals.

The cetuximab cohort consisted of 77 mCRC patients treated with first-line FOLFOX or SOX plus cetuximab in two clinical trials (JACCRO-CC05 and JACCRO-CC06 [10, 11]). Standard inclusion criteria were applied and have been previously described [10, 11]. All patients received cetuximab 400 mg/kg intravenously (day 1), followed by oxaliplatin 85 mg/m² administered with folinic acid (leucovorin) 200 mg/m² intravenously. Within this cohort, 36% of patients received FOLFOX, with 5-fluorouracil 400 mg/m² given as a bolus application, followed by 5-fluorouracil 2400 mg/m² as a 48-h continuous infusion. The remaining patients received SOX, with S-1 given orally on days 1–14.

All therapy was administered until the time of disease progression, intolerable toxicities, or early withdrawal. Responses were measured by intravenous contrast-enhanced computed tomography (CT) scans every 8 weeks according to response evaluation criteria in solid tumors (RECIST) 1.1. This study was approved by the Institutional Review Boards of each institute, and all patients signed informed consent for the analysis of molecular correlates.

Candidate polymorphisms

Polymorphisms were selected within EMT-related genes (TWIST1, ZEB1, SNAIL, E-cadherin), based on two specified criteria: 1) minor allele frequency (MAF) > 10% in Japanese according to the ENSEMBL database (http://www.ensembl.org/index.html), 2) reported or predicted functional relevance by in silico analysis. Functional significance was predicted based on information from the National Institute of Environmental Health Science SNP Function Prediction, Queen’s University F-SNP, and whether the SNP was located in a translated region of the gene (http://snpinfo.niehs.nih.gov/snpinfo/snptag.htm).

Genotyping

Genomic DNA was extracted from formalin-fixed paraffin-embedded (FFPE) tissue using the QIAmp Kit (Qiagen, Valencia, CA, USA) according to the manufacturer’s protocol (www.qiagen.com). PCR-based direct DNA sequence analysis was performed using the ABI 3100A Capillary Genetic Analyzer and Sequencing Scanner v1.0 (Applied Biosystems, Waltham, MA, USA). Extracted DNA was amplified using the primer sets in Supplementary Table 1.

Statistical analyses

The primary endpoint of this retrospective study was progression-free survival (PFS). Secondary endpoints were overall survival (OS) and tumor response. PFS was calculated from the first day of chemotherapy treatment to the first observation of disease progression or death from any cause. If progression or death was not observed, PFS was censored on the day of the last CT scan. OS was calculated from the first day of chemotherapy treatment to the date of death by any cause, or until the date of last follow-up, at which point OS was censored. Tumor responses were grouped into responders, including complete or partial response, and non-responders, including stable or progressive disease. Allelic distribution of the polymorphisms was tested for deviation from Hardy-Weinberg equilibrium using the exact test within each cohort.

Differences in baseline characteristics between the two cohorts were compared using X² test, Fisher’s exact test or the Wilcoxon rank sum test as deemed appropriate. Associations between SNPs and PFS and OS were examined in univariate analysis using Kaplan-Meier estimations and log-rank tests. In multivariable analysis, Cox regression models were adjusted for baseline characteristics that were significantly associated with PFS or OS in univariate analysis (log-rank P<0.10) within each cohort (Supplemental Table 2). Relationships between SNPs and response were assessed by Fisher’s exact test.

Recursive partitioning (RP), including cross validation, was used to explore and identify polymorphism profiles associated with PFS and OS using the rPart-function in R package [12, 13]. Case-wide deletion for missing polymorphisms was applied in univariate and multivariable analyses. In RP analysis, all patients with at least one polymorphism result available were included. SAS 9.4 (SAS Institute Inc, Cary, NC, USA) was used to conduct all analyses. All tests were two-sided at a significance level of 0.05.

RESULTS

Patient and tumor characteristics

Patient demographics and tumor characteristics are summarized in Table 1. The median follow-up periods were 5.2 and 2.1 years in the bevacizumab and cetuximab cohorts, respectively. The median PFS and OS were 1.3 and 3.2 years in the bevacizumab cohort, and 10.0 months and 2.5 years in the cetuximab cohort, respectively.

Table 1.

Baseline clinical characteristics of patients in the bevacizumab and cetuximab cohorts

	Bevacizumab Cohort (n=143)		Cetuximab Cohort (n=77)		P value †
	n	%	n	%
Gender
Male	80	56	44	57	0.98
Female	63	44	33	43
Age (years)
Median (range)	61 (27 – 75)		63 (39 – 79)		0.010
< 55	37	26	13	17
55 – 65	58	41	31	40	0.23
≥ 65	48	34	33	43
Tumor site
Right-sided colon	48	34	11	14	0.005
Left-sided colon	95	66	64	83
Unknown*	0		2	3
Liver metastasis
Yes	81	57	48	62	0.50
No	62	43	29	38
Number of metastases sites
< 2	55	38	33	43	0.62
≥2	88	62	44	57
Performance Status
ECOG 0	137	96	69	90	0.086
ECOG ≥1	6	4	8	10
Time to metastasis
Synchronous	59	41	59	77	< 0.001
Metachronous	84	59	18	23

^†P value was based on chi-square test, Fisher’s exact test or the Wilcoxon rank sum test when was appropriate.

^*Not included in the test.

Genetic variants and clinical outcomes in the bevacizumab cohort

In univariate analysis, TWIST1 rs2285682 was significantly associated with PFS and OS. Patients with any G allele had a significantly longer median PFS of 18.1 months than those with the T/T genotype, who had a median PFS of 13.3 months (HR, 0.57; 95% CI, 0.38, 0.85; log-rank P=0.003). Similarly, patients carrying a G allele had a significantly longer median OS of 44.1 months compared to those with the T/T genotype who had a median OS of 29.2 months (HR, 0.53; 95% CI, 0.35, 0.80; log-rank P=0.001; Table 2). In multivariate analysis adjusted for lung metastasis, primary tumor resection, liver metastasis, lymph nodes metastasis, peritoneal dissemination and ascites involvement, TWIST1 rs2285682 remained significantly associated with both PFS and OS (adjusted P=0.008 and P<0.001, respectively) (Table 2).

Table 2.

Association between EMT-related gene SNPs and clinical outcomes in the bevacizumab cohort

		Tumor Response			Progression-Free Survival					Overall Survival
	N	CR+PR	SD+PD	P value*	Median, months (95%CI)	HR (95%CI) †	P value*	HR (95%CI) ‡	P value*	Median, months (95%CI)	HR (95%CI) †	P value*	HR (95%CI) ‡	P value*
TWIST1 rs2285682				0.83			0.003		0.008			0.001		< 0.001
TT	69	53 (78%)	15 (22%)		13.3 (12.0, 15.7)	1 (Reference)		1 (Reference)		29.2 (24.9, 38.6)	1 (Reference)		1 (Reference)
GT a	53	42 (79%)	11 (21%)		18.1 (15.3, 22.5)	0.57 (0.38, 0.85)		0.56 (0.37, 0.86)		44.1 (36.8, 56.8)	0.53 (0.35, 0.80)		0.45 (0.29, 0.70)
GG a	8	5 (71%)	2 (29%)
TWIST1 rs2285681				0.46			0.28		0.042			0.13		0.001
CC	60	42 (71%)	17 (29%)		14.8 (11.8, 18.1)	1 (Reference)		1 (Reference)		34.2 (24.9, 45.7)	1 (Reference)		1 (Reference)
CG	43	34 (79%)	9 (21%)		15.3 (12.9, 18.1)	0.80 (0.50, 1.27)		0.78 (0.48, 1.28)		42.4 (29.2, 52.4)	0.63 (0.39, 1.02)		0.38 (0.23, 0.65)
GG	16	12 (86%)	2 (14%)		22.5 (9.9, 30.7)	0.62 (0.32, 1.21)		0.42 (0.21, 0.83)		43.8 (15.8, 71.1)	0.83 (0.43, 1.60)		0.54 (0.27, 1.08)
SNAIL rs1543442				0.42			0.51		0.064			0.52		0.15
GG	46	36 (78%)	10 (22%)		16.6 (13.4, 19.4)	1 (Reference)		1 (Reference)		43.4 (29.2, 47.3)	1 (Reference)		1 (Reference)
AG a	67	47 (71%)	19 (29%)		14.9 (12.9, 18.1)	1.16 (0.75, 1.78)		1.55 (0.98, 2.47)		35.9 (25.4, 42.8)	1.15 (0.74, 1.78)		1.40 (0.88, 2.23)
AA a	3	1 (50%)	1 (50%)
SNAIL rs4647958				1			0.77		0.73			0.85		0.49
TT	105	77 (75%)	25 (25%)		15.0 (12.9, 18.3)	1 (Reference)		1 (Reference)		36.8 (30.5, 45.6)	1 (Reference)		1 (Reference)
CT b	22	17 (77%)	5 (23%)		16.9 (10.5, 21.3)	1.08 (0.65, 1.79)		1.11 (0.62, 1.99)		43.4 (21.9, 55.1)	0.95 (0.57, 1.60)		0.81 (0.45, 1.46)
CC b	1	1 (100%)	0
ZEB1 rs2839658				1			0.96		0.74			0.75		0.86
CC	49	35 (73%)	13 (27%)		14.6 (11.8, 16.6)	1 (Reference)		1 (Reference)		35.9 (26.6, 45.7)	1 (Reference)		1 (Reference)
CT	50	36 (73%)	13 (27%)		16.3 (12.9, 18.1)	0.93 (0.59, 1.48)		0.94 (0.57, 1.54)		42.4 (27.8, 52.9)	0.84 (0.53, 1.33)		0.92 (0.56, 1.52)
TT	18	13 (72%)	5 (28%)		14.9 (8.4, 19.9)	0.98 (0.53, 1.82)		1.21 (0.64, 2.28)		29.2 (18.6, 57.3)	0.95 (0.49, 1.83)		1.10 (0.57, 2.15)
ZEB1 rs10826943				1			0.19		0.07			0.62		0.7
GG	61	44 (73%)	16 (27%)		16.3 (13.2, 19.0)	1 (Reference)		1 (Reference)		35.9 (23.4, 45.7)	1 (Reference)		1 (Reference)
AG c	31	23 (74%)	8 (26%)		12.9 (9.5, 17.0)	1.36 (0.85, 2.17)		1.62 (0.96, 2.72)		38.6 (23.4, 47.8)	1.13 (0.70, 1.84)		1.10 (0.66, 1.85)
AA c	4	3 (75%)	1 (25%)
E-cadherin rs16260				0.82			0.98		0.73			0.89		0.46
CC	88	68 (79%)	18 (21%)		16.9 (13.4, 19.0)	1 (Reference)		1 (Reference)		39.8 (31.0, 45.7)	1 (Reference)		1 (Reference)
AC	41	31 (78%)	9 (23%)		15.3 (11.8, 20.1)	1.00 (0.65, 1.53)		0.93 (0.60, 1.43)		36.1 (26.6, 52.8)	1.03 (0.67, 1.59)		0.84 (0.54, 1.33)

CI, confidence interval; HR, hazard ratio; CR, complete response; PR, partial response; SD, stable disease; PD, progressive disease.

^*P value was based on Fisher’s exact test for tumor response, log-rank test for PFS and OS in the univariate analysis (^†) and Wald test for PFS and OS in the multivariable Cox regression model adjusted for lung metastasis, primary tumor resection, liver metastasis, lymph nodes metastasis, peritoneal dissemination, and ascites (^‡).

^aCombined for estimates of HR.

In addition, the C allele of TWIST1 rs2285681 was associated with shorter PFS (14.8 vs. 16.8 months, HR, 0.64, adjusted P=0.0433) and OS (34.2 vs. 42.5 months, HR, 0.42, adjusted P=0.0003) in multivariable analysis. The remainder of the analyzed polymorphisms was not associated with clinical outcomes (Table 2).

Recursive partitioning trees were constructed to classify patients into high- and low-risk haplotype subgroups. In this model, TWIST1 rs2285682 genotype was the most important determinant of both PFS and OS. Patients with any G allele of TWIST1 rs2285682 (Group I) or the combination of TWIST1 rs2285682 T/T and SNAIL rs1543442 G/G (Group II) had the longest median PFS (Groups I and II, 17.1 months, Figure 1). In contrast, patients carrying the combination of TWIST1 rs2285682 T/T, any A allele of SNAIL rs1543442 and ZEB1 rs2839658 A/A had the shortest PFS (Group IV, 10.4 months, HR, 3.72, 95% CI, 2.01, 6.88, P<0.001; Figure 1). Similarly, patients carrying the TWIST1 rs2285682 T/T genotype had an inferior median OS of 29.0 months (Group II), compared to 44 months in those with any G allele (Group I, HR, 1.95, 95% CI, 1.30, 2.94; P=0.001; Figure 2).

Figure 1.

A, RP analysis of PFS. The end nodes of the tree model represent subgroups of low- and high-risk patients based on either a single gene variant or combination of gene variants. Fractions within the end nodes indicate patients who progressed/total patients with a given gene variant profile. B, PFS by tree model defined subgroups. Group 4 represents a high-risk subgroup based on a specific gene variant profile including TWIST1 rs2285682, SNAIL rs1543442, and ZEB1 rs2839658.

Figure 2.

A, RP analysis of OS. The end nodes of the tree model represent subgroups of low- and high-risk patients based on either a single gene variant or combination of gene variants. Fractions within the end nodes indicate patients who died/total patients with a given gene variant profile. B, OS by tree model defined subgroups. Group 2 represents a high-risk subgroup based on TWIST1 rs2285682.

In the Bevacizumab cohort, 76 patients had KRAS wild-type (exon 2) tumors, and 29 patients had KRAS mutant tumors. There were 39 patients whose KRAS status was unknown. In univariate analysis of the KRAS wild-type subgroup, TWIST1 rs2285682 genotype was significantly associated with OS. Patients with any G allele had a longer median OS of 44.1 months than those with the T/T genotype, who had a median OS of 26.6 months (HR, 0.60; 95% CI, 0.35, 1.04; log-rank P=0.05). In multivariate analysis, TWIST1 rs2285682 genotype was significantly associated with OS (HR, 0.37; 95% CI, 0.20, 0.69; P=0.002) (Supplementary Table 3). With regards to PFS, there was a similar trend favoring those with a TWIST1 rs2285682 G allele (18.3 vs. 13.4 months) in univariate analysis (HR, 0.66; 95% CI, 0.38, 1.13; log-rank P=0.096). In addition, patients carrying a TWIST1 rs2285682 G allele in the KRAS mutant subgroup had a longer PFS and OS compared those with the T/T genotype, though the sample size limits the power of this analysis (Supplementary Table 3).

Effect of gender and primary tumor location on associations between genetic variants and outcomes in the bevacizumab cohort

The effects of TWIST1 rs2285682 and rs2285681 genotypes on PFS and OS were then examined by gender. Women carrying a TWIST1 rs2285682 G allele had a significantly longer median PFS (20.1 vs. 12.0 months, HR 0.40, P=0.001) and OS (47.5 vs. 26.4 months, HR 0.44, P=0.005) compared to those with the T/T genotype (Supplementary Table 4). Women with a TWIST1 rs2285681 G allele had a significantly longer median PFS (18.8 vs. 12.0 months, HR 0.45, P<0.001) and OS (52.4 vs. 29.1 months, HR 0.48, P=0.019) compared to patients with the C/C genotype. In multivariate analysis including tumor location, both TWIST1 rs2285682 and rs2285681 variants remained significantly associated with PFS and OS.

In addition, women with a ZEB1 rs10826943 A allele had a significantly shorter median PFS compared to those with the G/G genotype in univariate and multivariable analyses (Supplementary Table 4). In men, there were no associations between EMT-related SNPs and clinical outcomes.

In patients with left-sided tumors, those with any TWIST1 rs2285682 G allele had a significantly longer median PFS and OS than patients with the T/T genotype in univariate and multivariate analyses (Supplementary Table 4). In patients with right-sided cancers, those with any ZEB1 rs10826943 A allele had a significantly shorter median PFS and OS compared to patients with the G/G genotype in univariate and multivariate analyses.

Genetic variants and clinical outcomes in the cetuximab cohort

In the cetuximab cohort, genotyping for TWIST1 rs2285682 was performed in 76 patients (99%). Genotyping was not successful in 1 patient due to limited quantity and quality of extracted genomic DNA. In the Cetuximab cohort, all patients had KRAS wild-type cancers. There were no significant associations between TWIST1 rs2285682 genotype and outcomes in this cohort (Supplementary Table 5).

DISCUSSION

Our study is the first to identify the predictive utility of TWIST1 genetic variants in mCRC patients receiving bevacizumab-based chemotherapy. The TWIST1 rs2285682 and rs2285681 G alleles conferred a statistically significant and clinically meaningful survival benefit in multivariate analysis with a more prominent effect in women. That no significant associations were found in the cetuximab cohort supports the hypothesis that the prognostic associations of these polymorphisms are specific to anti-angiogenic therapy.

TWIST1 is a transcriptional factor critical to embryonic development and tumor angiogenesis. In cancer cell lines, TWIST1 induces Jagged1 expression and Notch signaling [14], which mediates transdifferentiation of cancer cells into endothelium, de novo angiogenesis [9], as well as bevacizumab resistance [15]. Under hypoxic states, TWIST1 upregulation contributes to vasculogenic mimicry and may therefore serve as a resistance mechanism to anti-angiogenic therapy [16, 17]. In xenograft models, TWIST1 overexpression leads to increased VEGF synthesis, vascular volume and vessel permeability [18], partly as a result of thrombin stimulation [19][20]. Independent of VEGF regulation, TWIST1 prompts CCL2 production to recruit tumor-associated macrophages which also activate angiogenesis [21]. Clinically, increased intratumoral TWIST1 mRNA or protein expression has been correlated with adverse clinicopathological features and poor survival in CRC [22, 23]. To our knowledge, the influence of TWIST1 polymorphisms on angiogenesis and clinical outcomes in CRC has not been previously reported.

TWIST1 rs2285682 localizes to a transcription factor binding site on chromosome 7p21, and as a tag SNP may exert functional effects through linked polymorphisms at other loci. It is plausible that TWIST1 genetic variants may affect the establishment of tumor endothelium through modulation of VEGF or Jagged1/Notch signaling. Functional studies are needed to explore these hypotheses.

In addition to TWIST1, our recursive partitioning analysis revealed potential prognostic roles for ZEB1 and SNAIL polymorphisms. This is consistent with preclinical studies showing that VEGF inhibition and hypoxia provoke MET signaling and resultant EMT activation, marked by ZEB1 [20] and SNAIL [24] upregulation. How EMT haplotype influences these pathways on a molecular level to promote treatment resistance and disease progression, remains to be determined.

Notably, TWIST1 and ZEB1 genotypes seemed to have a stronger influence on survival among women. Gender disparities in CRC survival [25] and the protective role of estrogen receptors (ERs) on colorectal carcinogenesis [26][27] have been established, though the underlying mechanisms are not well understood. Preclinical data suggests that TWIST1 represses ERα mRNA and protein expression [28], and ERβ has been shown to affect ZEB1 levels by altering miR-200 expression [27]. Hence, the observed gender-genotype interactions may be mediated through the ER pathway.

Our study is limited by its retrospective design and small sample size, though the consistent associations across PFS and OS are promising and warrant further validation in larger prospective trials. Furthermore, the MAF of TWIST1 rs2285682 is at most 2% in Caucasians which limits the generalizability of our findings. However, TWIST1 rs2285681, ZEB1 rs10826943, and SNAIL rs1543442 minor alleles are much more prevalent across ethnicities. In addition, we could not account for tumor microsatellite instability status, particularly in exploring the effects of gender and primary tumor site on genotype and outcomes. Future investigations should also determine the influence of different cytotoxic and anti-angiogenic regimens on the prognostic impact of EMT polymorphisms.

Understanding how EMT facilitates anti-angiogenic resistance and integrating patient-specific markers is crucial to optimizing benefit from these agents. If confirmed in prospective studies, TWIST1 and other EMT-related genetic variants may guide personalized therapeutic decisions for mCRC patients undergoing first-line therapy.