Prognostic value of CpG island methylator phenotype among colorectal cancer patients: a systematic review and meta-analysis

Review of the literature regarding prognostic value of CpG island methylator phenotype (CIMP) among colorectal cancer (CRC) patients showed CIMP to be associated with worse disease-free survival and overall survival. Future research into CIMP's most appropriate operational definition and its action mechanism would be of vital importance in helping us put past study results into proper perspective.

Abstract

Background

Divergent findings regarding the prognostic value of CpG island methylator phenotype (CIMP) in colorectal cancer (CRC) patients exist in current literature. We aim to review data from published studies in order to examine the association between CIMP and CRC prognosis.

Materials and methods

A comprehensive search for studies reporting disease-free survival (DFS), overall survival (OS), or cancer-specific mortality of CRC patients stratified by CIMP is carried out. Study findings are summarized descriptively and quantitatively, using adjusted hazard ratios (HRs) as summary statistics.

Results

Thirty-three studies reporting survival in 10 635 patients are included for review. Nineteen studies provide data suitable for meta-analysis. The definition of CIMP regarding gene panel, marker threshold, and laboratory method varies across studies. Pooled analysis shows that CIMP is significantly associated with shorter DFS (pooled HR estimate 1.45; 95% confidence interval (CI) 1.07–1.97, Q = 3.95, I² = 0%) and OS (pooled HR estimate 1.43; 95% CI 1.18–1.73, Q = 4.03, I² = 0%) among CRC patients irrespective of microsatellite instability (MSI) status. Subgroup analysis of microsatellite stable (MSS) CRC patients also shows significant association between shorter OS (pooled HR estimate 1.37; 95% CI 1.12–1.68, Q = 4.45, I² = 33%) and CIMP. Seven studies have explored CIMP's value as a predictive factor on stage II and III CRC patient's DFS after receiving adjuvant 5-fluorouracil (5-FU) therapy: of these, four studies showed that adjuvant chemotherapy conferred a DFS benefit among CIMP(+) patients, one concluded to the contrary, and two found no significant correlation. Insufficient data was present for statistical synthesis of CIMP's predictive value among CRC patients receiving adjuvant 5-FU therapy.

Conclusion

CIMP is independently associated with significantly worse prognosis in CRC patients. However, CIMP's value as a predictive factor in assessing whether adjuvant 5-FU therapy will confer additional survival benefit to CRC patients remained to be determined through future prospective randomized studies.

background

CpG islands are genomic regions that contain a high frequency of cytosine and guanine nucleotides, connected with a phosphodiester bond. They are typically found in or near promoter regions of the genome, where transcription is initiated, and are present in as high as 40%–50% of the human genes. Aberrant methylation of cytosine nucleotides within these CpG islands can lead to aberrant silencing of normal tumor-suppressor function and cancer formation [1, 2].

A distinct molecular subtype of cancer, characterized by high degrees of methylation, has been termed CpG island methylator phenotype (CIMP) [3]. CIMP cancers are characterized by unique molecular features, epidemiology, precursor lesions, and they represent ∼15% of sporadic cases of colorectal cancer (CRC).

However, the role CIMP plays in the pathogenesis of CRC is still poorly understood. It has been shown that hypermethylation secondary to CIMP leads to microsatellite instability (MSI) through the methylation of the MLH1 promoter and the consequent silencing of the MLH1 mismatch repair gene. Almost 70%–80% of MSI CRCs can be attributed to CIMP and associated MLH1 methylation [4]. CIMP, however, can exist with and without MSI.

Studies have shown CIMP to be an independent negative prognostic factor in disparate subgroups of CRC patients [5–7], although its influence is modified by other genetic factors including MSI [8, 9] and KRAS/BRAF status [10, 11]. It has been hypothesized that CIMP-high CRC patients might have a better response to 5-fluorouracil (5-FU) chemotherapy due to observations that suppression of gamma-glutamyl hydrolase [12], one of the putative effects of CpG island methylation, is associated with an increase in intracellular folate level [13], which magnifies the effect of 5-FU chemotherapy [14]. Several clinical studies have come to conflicting conclusions regarding the predictive value of CIMP [15, 16]. Precise estimates of the predictive value of CIMP can allow for refinement in clinical management, especially among stage II CRC patients, where the value of 5-FU chemotherapy is still controversial [17].

The goal of this systematic review is to summarize published studies using standardized techniques to gain a better understanding of the prognostic significance of CIMP in CRC patients with respect to their disease-free survival (DFS) and overall survival (OS).

methods

search strategy and study eligibility

Published studies were eligible if CRC patient survival was analyzed after stratification by CIMP status of primary tumor tissue. CIMP status was defined by the respective study authors with no restriction to laboratory method, gene panel, or marker threshold value. Primary outcome of the study was DFS, defined as the length of time since initial cancer treatment during which no evidence of cancer recurrence or death was found. Secondary outcome was OS, defined as the length of time since initial cancer treatment that patients were still alive. Cohort studies of either a prospective or a retrospective nature were included. Case reports and reviews were excluded. There was no restriction with respect to patient age, gender, ethnicity, comorbidity, tumor type, disease stage, length of follow-up, or treatment received (study eligibility summarized in Table 1).

Table 1.

Study eligibility criteria table

Study eligibility criteria
CIMP definition	Widespread CpG island methylation as defined by each study (no restriction regarding laboratory method, gene panel, or marker threshold value)
Outcome measure	Disease-free survival, overall survival, or cancer-specific mortality
Study design	Prospective or retrospective cohort studies
Anatomical site	Colon, rectal, or colorectal
Study tissue	Surgically resected primary tumor tissue
Stage	Any
Therapy	Any
Length of follow-up	Any

study identification

We followed Meta-analysis of Observational studies in Epidemiology (MOOSE) guidelines [18] to identify eligible studies. Combination of text and exploded controlled vocabulary terms were used to search PubMed, EMBASE, SCOPUS, and Web of Science electronic databases until 6 June 2013, relating to the following two concepts: (i) CIMP and (ii) CRC. Citations in relevant reviews and studies were hand-searched to capture additional potentially relevant studies. In addition to full publications, original studies in the form of conference abstracts and letters were included to capture grey literature. Each study was independently assessed for inclusion by at least two reviewers. Discrepancies within the reviewing pair were resolved via discussion. The flow chart of study identification was presented according to Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines in Figure 1 [19].

Figure 1.

Study identification flowchart. Using standardized protocol for a comprehensive search through four electronic databases, a total of 33 studies were included in this review for qualitative or quantitative analysis.

statistical methods

Study characteristics including study country, patient characteristics, and CIMP assessment methods were summarized in a consistent manner for easy examination. Patient number and percentages among different gender and age groups were calculated using available data when they were not directly reported.

Assignment of CIMP status into constitutive groups (CIMP-positive or CIMP-negative) or CIMP-high, CIMP-low, and CIMP-zero was carried out according to the grouping in each study. The term CIMP was used synonymously with CIMP-positive and CIMP-high, depending on whether the study authors have di- or trichotomized CIMP status. During our discussion, CIMP-low and CIMP-zero were treated equivalently with CIMP-negative, but we carried out separate statistical analyses for studies that dichotomized and studies that trichotomized CIMP status. Clinical and methodological sources of heterogeneity across studies were evaluated to assess the appropriateness of statistical synthesis. Factors taken into consideration included variation in patient disease stage, treatment received, and MSI status. Statistical heterogeneity was assessed by Breslow–Day test while Q and I² statistics were presented as measures of inconsistency of CIMP prognostic values across studies attributable to heterogeneity [20]. The Q statistic was a weighted sum of squares following a χ² distribution with k−1 (k = number of included studies) degrees of freedom. The P value calculated thereof gave an estimate of the deviation from total homogeneity. In our study, we employed a P value of 0.1 and an I² statistic of <50% as indicating acceptable statistical heterogeneity.

In order to summarize the association between CIMP status and survival across studies, a weighted average of the individual adjusted log hazard ratios (HRs) was used, with the weights inversely proportional to the variance of the log HR of each study. Adjusted HRs were sought among results of multivariate survival analysis using Cox proportional hazards regression model and the adjusted variables documented. When missing data regarding adjusted HR or variance was encountered, estimates were indirectly imputed from available numerical data where possible using confidence intervals (CIs), P values, and event numbers by methods described by Parmar et al. [21]. Recurrence-free survival (RFS) was interpreted as synonymous with DFS. Random-effects model was used in all numerical synthesis. Forest plots were presented with studies plotted in order of decreasing variance of the log HR. Horizontal lines represented 95% CIs. Each box represented the HR point estimate and its area was proportional to the weight of the study. The diamond on the bottom denoted the overall summary estimate, with CIs given by the width of the diamond. The vertical line signified the reference line with HR = 1.0. Sensitivity analyses were carried out to assess the impact of including conference abstracts, imputed estimates, and studies with more than one risk of bias marked as high or unclear risk. All numerical synthesis was carried out using Review Manager 5.2.

results

eligible studies

We initially identified 2414 studies for potential inclusion using only ‘CIMP’ and ‘CRC’ as keywords for searching (Table 1). (Flow chart of study identification is summarized in Figure 1.) The main reasons for exclusion of studies during the initial title/abstract screening were failure to examine disease prognosis, incorrect histology, or the studies were not original studies. Ninety-three studies remained for full-text inspection after title/abstract screening (Figure 1). Of these, 39 full-text studies were excluded because the primary outcomes of interest (OS and DFS) were not reported, 13 were excluded because no prognostic measure with regard to CIMP was reported, 6 were excluded due to overlapping patient population with included studies, and 2 were excluded because the study tissue was not primary tumor tissue. A total of 33 studies were included in the systematic review. (Study characteristics of these studies were summarized in Table 2.) Of these, only 19 studies were eligible for pooling hazard estimates in the meta-analysis. The reasons for exclusion were failure to present numerical HR estimates, failure to include CIMP in the multivariate analysis, and failure to present survival data with explicit numerical values.

Table 2.

Summary of included study characteristics

	Study characteristics		Patient population					CIMP assessment						Outcomes
Study	Publication type	Country	Sample size (# of patients), N	Stage	Male gender, N (%)	Median age, y/o (range)	Inclusion criteria	Specimen preservation	Gene panel	Lab method	CIMP classification	Marker threshold	CIMP prevalence, N (%)
Ahn et al. [10]	Full publication	Korea	161	III	93 (57.8%)	Mean 60.9 (31–84)	Sporadic colon CA	Cryopreservation	MINT1, MINT2, MINT31, hMLH1, p16, p14, SFRP1, SFRP2, WNT5A	Bisulfite pyrosequencing	CIMP −/+	CIMP+ ≥ 3/9 (plus additional genetic criteria)	29 (18.0%)	DFS
Barault et al. [11]	Full publication	France	582	I–IV	333 (57.2%)	≤65 y/o: N = 156 66–75 y/o: N = 184 >75 y/o: N = 242	Sporadic colon CA without IBD	Cryopreservation	Classic panel	MSP	CIMP 0/low/high	CIMP-high: 4–5/5 CIMP-low: 1–3/5	CIMP-high: 97 (16.7%) CIMP-low: 199 (34.2%) CIMP 0: 286 (49.1%)	OS
Dahlin et al. [12]	Full publication	Sweden	604	I–IV	315 (52.2%)	69.9 (58–79)	Sporadic colorectal CA	FFPE	CDKN2A, MLH1, CACNA1G, NEUROG1, RUNX3, SOCS1, IGF2, CRABP1	MethyLight	CIMP 0/low/high	CIMP-high: 6–8/8 CIMP-low: 1–5/8	CIMP-high: 74 (12.3%) CIMP-low: 215 (35.6%) CIMP 0: 301 (49.8%)	Cancer-specific mortality
Donada et al. [22]	Full Publication	Italy	120	II	57 (47.5%)	Mean 67.6	Colon CA	FFPE	Weisenberger	MSP	CIMP 0/low/high	CIMP-high: ≥3/5 CIMP-low: 1–2/5	CIMP-high: 22 (18.3%) CIMP-low: 36 (30.0%) CIMP 0: 62 (51.7%)	DFS, OS
Han et al. [23]	Full Publication	Korea	322	II–III	192 (59.6%)	61.0 (30–78)	Sporadic colorectal CA cases with complete resection and complete FOLFOX adjuvant treatment among stage II and III cases	NR	CACNA1G, CRABP1, IGF2, MLH1, NEUROG1, CDKN2A(p16), RUNX3, SOCS1	MethyLight	CIMP 0/low/high	CIMP-high: ≥5/8 CIMP-low: 1–4/8	CIMP-high: 25 (7.8%) CIMP-low: 125 (38.8%) CIMP 0: 172 (53.4%)	DFS
Jo et al. [24]	Full publication	Germany	150	NR	107 (71%)	Mean 62.4	Locally advanced rectal CA cases without preoperative chemotherapy	FFPE	Weisenberger	MSP	CIMP −/+	CIMP+: ≥3/5	15 (10%)	DFS, OS
Jover et al. [20]	Full publication	Spain	302	I–IV	185 (61.3%)	Mean 70.3	Sporadic colorectal CA	FFPE	CACNAG1, SOCS1, NEUROG1, RUNX3, MLH1	Bisulfite pyrosequencing	CIMP −/+	CIMP+: ≥3/5	89 (29.5%)	DFS
Ju et al. 2011 [41]	Full publication	Japan	78	I–IV	51 (65.4%)	Mean 63.9	Sporadic colorectal CA	Cryopreservation	Classic panel	MSP	CIMP −/+	CIMP+: ≥2/5	19 (24.4%)	DFS
Kakar et al. 2008 [37]	Full publication	USA	69	I–IV	39 (56.5%)	<60 y/o: N = 19 ≥60 y/o: N = 50	Sporadic colorectal CA	FFPE	hMLH1, p16, HIC1, RASSF2, ID4, MINT1, MINT31	MSP	CIMP −/+	CIMP+: ≥3/7	16 (23.2%)	OS
Kakar et al. 2012 [28]	Full publication	USA	33	I–IV	24 (72.7%)	≤26 y/o: N = 20 >60 y/o: N = 13	Signet ring cell colorectal carcinoma	FFPE	hMLH1, p16, HIC1, RASSF2, ID4, MINT1, MINT31	MSP	CIMP −/+	CIMP+: ≥3/7	16 (48.5%)	OS
Kalady et al. [25]	Full publication	USA	357	I–IV	193 (54.1%)	Mean 66.9	Sporadic colorectal CA	Cryopreservation	Weisenberger	MethyLight	CIMP −/+	CIMP+: ≥3/5	78 (21.8%)	DFS
Kim et al. [26]	Full publication	Korea	320	I–IV	187 (58.4%)	Mean 60.9	Sporadic colorectal CA cases without neoadjuvant therapy	FFPE	CACNA1G, CDKN2A (p16), CRABP1, IGF2, MLH1, NEUROG1, RUNX3, SOCS1	MethyLight	CIMP −/+	CIMP+: ≥5/8	37 (11.6%)	OS
Kim et al. [27]	Full publication	Korea	285	I–IV	168 (58.9%)	Mean 58.0	Sporadic colorectal CA cases without neoadjuvant therapy	NR	MLH1, MINT1, MINT2, MINT31, p16, p14, CACNA1G	Bisulfite pyrosequencing	CIMP −/+	CIMP+: ≥2/7 or ≥3/7	102 (35.8%) or 51 (17.9%)	DFS
Lee et al. [14]	Full Publication	Korea	134	I–IV	81 (60.4%)	<60 y/o: N = 57 ≥60 y/o: N = 77	Sporadic colorectal CA	FFPE	Classic panel	MSP	CIMP −/+	CIMP+: ≥2/5	42 (31.3%)	OS
Min et al. [28]	Full publication	Korea	245	I–IV	138 (56.3%)	Mean 65.0 (33–83)	Sporadic colorectal CA cases without neoadjuvant therapy	FFPE	Weisenberger	MethyLight	CIMP 0/low/high	CIMP-high: >2/5 CIMP-low: 1–2/5	CIMP-high 34 (13.9%)	DFS
Ogino et al. [29]	Full publication	USA	31	IV	22 (71.0%)	57.0 (31–81)	Advanced MSS colorectal CA cases without metastases	FFPE	CACNA1G, CDKN2A, CRABP1, IGF2, MLH1, NEUROG1, RUNX3, SOCS1, MINT1, MINT31, IGFBP3, MGMT, WRN	MethyLight	CIMP 0/low/high	CIMP-high: >8/13 CIMP-low: 1–8/13	CIMP-high: 3 (10.0%) CIMP-low: 22 (73.3%) CIMP 0: 5 (16.7%)	OS
Ogino et al. [15]	Full publication	USA	649	I–IV	283 (44%)	Mean 66.5	Primary colon CA cases	FFPE	CACNA1G, CDKN2A, CRABP1, IGF2, MLH1, NEUROG1, RUNX3, SOCS1	MethyLight	CIMP 0/low/high	CIMP-high: >5/8 CIMP-low: 1–5/8	CIMP-high: 126 (19.4%) CIMP-low: 252 (38.8%) CIMP 0: 271 (41.8%)	OS, cancer-specific mortality
Rhee et al. [30]	Full publication	Korea	207	I–IV	126 (60.9%)	≤55 y/o: N = 100 >55 y/o: N = 107	Microsatellite unstable colorectal cancers	FFPE	CACNA1G, CDKN2A, CRABP1, IGF2, MLH1, NEUROG1, RUNX3, SOCS1	MethyLight	CIMP 0/low/high	CIMP-high: ≥5/8 CIMP-low: 1–4/8	NR	OS
Rijnsoever et al. [31]	Full publication	Australia	275	II–III	138 (50.2%)	<71 y/o: N = 144 ≥71 y/o: N = 131	Colorectal CA	FFPE	P16, MDR1, MINT2	MSP	CIMP −/+	CIMP+: ≥2/3	128 (46.5%)	OS
Rijnsoever et al. [19]	Full publication	Australia	206	III	132 (64.1%)	Mean 60.9	Colorectal CA	83% FFPE; 17% cryopreservation	P16, MINT2, MDR1	MSP	CIMP −/+	CIMP+: ≥2/3	67 (32.5%)	DFS
Samowitz et al. [32]	Full Publication	USA	886	I–IV	473 (53.4%)	<55 y/o: N = 120 55–64 y/o: N = 222 65–70 y/o: N = 217 71–79 y/o: N = 327	Primary sporadic colon CA cases between 30 and 79 y/o without IBD	FFPE	Classic panel	MSP	CIMP-low/high	CIMP-high: ≥2/5	246 (27.8%)	OS
Samowitz et al. [33]	Full publication	USA	990	I–IV	NR	NR	Primary sporadic rectal CA cases between 30 and 79 y/o without IBD	FFPE	Classic panel	MSP	CIMP-low/high	CIMP-high: ≥2/5	103 (13.5%)	OS
Sanchez et al. [34]	Full publication	USA	391	I–IV	215 (55.0%)	Mean 66.7	Colorectal CA cases without IBD	Cryopreservation	Weisenberger	MethyLight	CIMP −/high	CIMP-high: ≥3/5	83 (21.2%)	DFS, OS
Shen et al. [9]	Full publication	USA	182	NR	120 (65.9%)	≤65 y/o: N = 91 >65 y/o: N = 97	Colorectal CA cases	FFPE	MINT1, MINT2, MINT31, hMLH1, p14, p16	MSP, COBRA	CIMP −/+	CIMP+: ≥2/6	28 (15.4%)	OS
Simons et al. [35]	Full publication	Netherlands	509	I–IV	277 (54.4%)	Mean 67.8	Colorectal CA cases	FFPE	Weisenberger	MSP	CIMP −/+	CIMP+: ≥3/5	95 (18.7%)	Cancer-specific mortality
Ward et al. [13]	Full publication	Australia	605	I–IV	323 (53.4%)	Mean 68.3 (29–99)	Sporadic colorectal CA cases without IBD	Cryopreservation	P16, MINT1, MINT2, MINT12, MINT31	MSP	CIMP −/+	CIMP+: >3/5	NR	OS
Zlobec et al. [36]	Full publication	Switzerland	337	NR	156 (46.3%)	Mean 69.9 (42–95)	Sporadic colorectal CA	FFPE	CRABP1, MLH1, p16INK4a, CACNA1G, NEUROG1	Bisulfite pyrosequencing	CIMP 0/low/high	CIMP-high: ≥4/5 CIMP-low: 1–3/5	CIMP-high: 24 (7.1%) CIMP-low: 145 (43.0%) CIMP 0: 168 (49.9%)	Cancer-specific mortality
Zanutto et al. [37]	Letter to the editor	Italy	42	II	NR	66.1 (30–83)	MSS colorectal CA	Cryopreservation	Classic panel	MSP	Methylation low/intermediate/high	Methylation high: >3/5 Intermediate 3/5	High: 13 (31.0%) Intermediate: 15 (35.7%) Low: 14 (33.3%)	DFS
Al-Sohaily et al. [38]	Conference abstract	Australia	281	NR	NR	NR	Rectal CA	FFPE	Weisenberger	MSP	CIMP 0/low/high	CIMP-high ≥3/5	13 (4.6%)	OS
Elsaleh et al. [39]	Conference abstract	USA	891	III	NR	NR	Colorectal CA	NR	NR	NR	NR	NR	NR	DFS
Kim et al. [40]	Conference abstract	Korea	151	NR	NR	NR	Colorectal CA	NR	Weisenberger	NR	CIMP −/+	CIMP+: ≥3/5	27 (17.9%)	OS
Kim et al. [41]	Conference abstract	Korea	49	NR	NR	NR	Colorectal CA cases with metastases treated by 5-fluorouracil, leucovorin, irinotecan, and cetuximab	FFPE	P16, p14, MINT1, MINT2, MINT31, hMLH1	Bisulfite pyrosequencing	CIMP −/+	NR	14 (28.6%)	DFS, OS
Koo et al. [42]	Conference abstract	Korea	191	III	NR	NR	Colorectal CA	NR	Weisenberger	NR	CIMP−/+	CIMP+: ≥3/5	63 (33.0%)	DFS

CA, cancer; IBD, inflammatory bowel disease; FFPE, formalin fixed paraffin embedded; NR, not reported; MSP, methylation-specific polymerase chain reaction; MSS, microsatellite stable; Classic panel: MINT1, MINT2, MINT31, CKKN2A(p16), hMLH1; Weisenberger, CACNA1G, IGF2, NEUROG1, RUNX3, SOCS; y/o: years old; DFS, disease-free survival; OS, overall survival; CIMP, CpG island methylator phenotype.

study characteristics

The 33 studies analyzed 10 635 patients for CIMP status and its relationship to disease prognosis. The mean number of patients for each study was 322 (range 33–990). Most studies included both colon and rectal cancer cases. Five studies [6, 7, 11, 43, 44] included colon cancers only and three studies [45–47] rectal cancers only. In most studies, a mixture of different disease stages was included; however, eight studies [6, 15, 22, 32, 38, 43, 48, 49] examined only stage II or III cancers, and one study [33] examined only stage IV cancers. In three studies [15, 23, 31, 48], only patients who received chemotherapy were included, while four other studies [32, 33, 37, 44] included only microsatellite stable (MSS) tumor tissues for analysis and 1 study [39] included only MSI tumor tissues for analysis.

The risk of bias of each included study is summarized in Table 3. While retrospective convenience sampling was employed for almost all studies, some studies stated their inclusion and exclusion criteria more explicitly than others and these studies were rated as low risk in selection bias. Almost all studies (31 of 33 studies), including conference abstracts, have explicitly stated their method of assessment of CIMP and genetic mutations such as KRAS or BRAF. Median follow-up length, range, and loss-to-follow-up rate were satisfactorily reported in about half of the studies. Known or commonly discussed confounders in the relationship between CIMP and survival, such as age, disease status, MSI status, KRAS/BRAF mutation, were adjusted for in two-thirds of the studies and, in these studies, the risk of confounding was rated as low risk.

Table 3.

Risk of bias of included studies

Green circles represent studies with high risk of bias; pink circles represent studies with low risk of bias; blue circles represent studies that did not provide sufficient information for assessing risk of bias.

heterogeneity in CIMP definition

Only studies that used primary tumor tissues were included in this review. Of the 33 included studies, 21 classified CIMP status in a dichotomized fashion (CIMP-positive versus negative) while CIMP was classified in a trichotomized fashion (CIMP-high/low/zero) in the remainder of the studies.

Significant variation existed between the studies regarding the gene panel, marker threshold, and laboratory method used for defining CIMP. The Classic panel (MINT1, MINT2, MINT31, CKKN2A(p16), hMLH1) and the 5-gene panel devised by Weisenberger et al. [42] in 2006 (CACNA1G, IGF2, NEUROG1, RUNX3, SOCS1) were the two most commonly employed gene panels, although 17 of the 33 included studies opted to use a different gene panel (Table 2). The number of genes assessed in each study ranged from 3 to 13 genes, with a median of five genes. Marker threshold for defining CIMP status was often chosen arbitrarily or based on previous literature; two studies [7, 29] chose their threshold based on association with known clinicopathological features of CIMP-positive tumors. One study [11] specifically compared the effect of using different CIMP panels and concluded that this variation could result in significantly different associations with survival outcomes. Sixteen [7, 9, 10, 15, 22, 28, 32, 37, 41, 43–47, 50] studies used methylation-specific PCR (MSP) as the laboratory method for methylation analysis, nine [8, 11, 29–31, 33, 48] used MethyLight, five [6, 16, 26, 39, 51, 52] used bisulfite pyrosequencing, one [5] used COBRA, and in three studies, laboratory method was not specified. In general, no specific gene panel or laboratory method was observed to produce a consistently higher or lower CIMP prevalence.

relationship between CIMP and prognosis

disease-free survival

The median prevalence of CIMP-positive or CIMP-high status amongst included studies was 18.2%, ranging from 4.6% to 46.5%. Overall median follow-up length was 45.3 months, ranging from 2.8 to 100.8 months.

Eleven studies examined the relationship between CIMP status and DFS amongst CRC patients. The majority [6, 16, 23, 30, 31, 41, 48, 51] of studies (8 of 11) found no significant relationship between CIMP and DFS, while 2 studies [26, 46] found CIMP to be associated with unfavorable DFS. In addition, one [6] found that CIMP, despite being an insignificant predictor for DFS in their overall study cohort, was associated with an unfavorable DFS among tumors of proximal colon. In contrast, Koo et al. [49] found a combination of CIMP(+)/KRAS-wild type to be associated with favorable DFS compared with other genetic/epigenetic profiles. Examining a subgroup of CRC patients who were MSS, Zanutto et al. [32] found that CIMP positivity was associated with a lower recurrence rate.

In our pooled analysis, CIMP was associated with an unfavorable DFS after adjusting for other relevant confounders in each contributing study. Insufficient data were present among studies that categorized CIMP status in a trichotomized fashion for pooled analysis. Six studies that categorized CIMP status in a dichotomized fashion provided satisfactory adjusted HR estimates suitable for numerical synthesis. This accounted for a total of 1454 patients. Their overall summary HR was 1.45 (95% CI 1.07–1.97), with a low heterogeneity (Q = 3.95, I² = 0%), indicating a worse DFS for CIMP-positive patients. From the forest plot (Figure 2), the CIs of adjusted HRs in five of six studies were insignificant, but the majority of them had the same trend. Sensitivity analysis was carried out to assess the effect of including imputed HR estimates by leaving out two studies with imputed HRs [46, 48]. The exclusion did not change our conclusion (pooled HR = 1.22, 95% CI 0.83–1.79; Q = 1.24, I² = 0%). In a separate analysis, a subgroup of studies [16, 41, 46] with ≥1 high risk of bias item were excluded. This also did not change our conclusion although the summary estimate did not attain statistical significance (pooled HR = 1.46, 95% CI 0.79–2.67; Q = 1.10, I² = 0%).

Figure 2.

Forest plots of hazard ratios (HRs) of disease-free survival (DFS) in studies of colorectal cancer (CRC) patients associated with CpG island methylator phenotype (CIMP) with no restriction to MSI status. CIMP is associated with a worse DFS among CRC patients after adjusting for other prognostic factors.

A subgroup of studies (7 of 33) also explored CIMP's value as a predictive factor on the effect of adjuvant 5-FU therapy on CRC patients. Seven studies examined the relationship between CIMP status and DFS among patients who received 5-FU-based adjuvant chemotherapy after surgical resection of the primary tumor. (Table 4 shows the individual study details). Elsaleh et al. [38] showed that CIMP(+) cases received survival benefit from 5-FU therapy independent of MSI and p53 status. Min et al. [31] found that stage II and III CRC patients with CIMP-high represented a cohort most likely to receive benefits from 5-FU therapy. Rijnsoever et al. [15], in a study restricted to stage III CRC cases, also showed that CIMP(+) status was associated with favorable response to 5-FU therapy. In addition, Rijnsoever et al. found that CIMP, despite being associated with worse prognosis among patients treated by surgery alone, was associated with a trend for better DFS among patients receiving both surgery and 5-FU therapy. [15] This finding was confirmed by Donada et al. [43] in their study of stage II colon cancer cases.

Table 4.

Summary of study conclusions regarding interaction of CIMP and adjuvant chemotherapy

Study	Study cohort	Chemotherapy regimen	DFS among patients receiving surgery alone	DFS among patients receiving surgery and adjuvant CT	DFS among CIMP (+) patients	DFS among CIMP (−) patients
Elsaleh et al. [39]	891 stage III CRC patients	5-FU-based chemotherapy	NR	NR	Surgery + CT > surgery alone (P = 0.03)	Surgery + CT–surgery alone
Min et al. [28]	124 stage II and III CRC patients	5-FU or capecitabine, leucovorin (Mayo regimen)	CIMP-H–CIMP-L/0	CIMP-H–CIMP-L/0 (P = 0.073).	Surgery + CT > surgery alone (P = 0.022)	Surgery + CT–surgery alone
Rijnsoever et al. [19]	206 stage III CRC patients	5-FU, leucovorin (Mayo regimen)	CIMP+ < CIMP− (P = 0.05)	CIMP+ > CIMP−	Surgery + CT > surgery alone (P = 0.002)	Surgery + CT–surgery alone (P = 0.6)
Donada et al. [22]	120 stage II colon cancer	5-FU, leucovorin	CIMP-H–CIMP-L/0 (P = 0.6)	CIMP-H > CIMP-L/0 (P = 0.05)	NR	Surgery + CT–surgery alone for CIMP-0, CIMP-low patients (P = 0.7, P = 0.9)
Jover et al. [20]	196 stage II and III CRC patients	5-FU-based chemotherapy	CIMP+ < CIMP− (P = 0.03)	CIMP+ to CIMP− (P = 0.1)	Surgery + CT–surgery alone (P = 0.6)	surgery + CT > surgery alone (P = 0.0001)
Han et al. [23]	322 stage II and III CRC who received surgery + CT	FOLFOX (5-FU, oxaliplatin, folinic acid)	NR	CIMP-H–CIMP-L/0 (P = 0.31)	NR	NR
Kim et al. [41]	49 metastatic CRC cases who received CT	FOLFIRI (5-FU, leucovorin, iriontecan) and cetuximab	NR	CIMP+ to CIMP−	NR	NR

CRC, colorectal cancer; NR, not reported; CT, chemotherapy; 5-FU, 5-fluorouracil; CIMP, CpG island methylator phenotype; CIMP-H, CIMP-high; CIMP-L, CIMP-low; CIMP-0, CIMP-zero; DFS, disease-free survival.

On the other hand, Jover et al. [16] found that among stage II and III CRC cases, CIMP(+) cases represented a cohort that was unresponsive to 5-FU therapy while CIMP(−) cases received significant survival benefit from 5-FU therapy. Han et al. [48] found both the predictive value of CIMP and the interaction between CIMP and adjuvant chemotherapy to be insignificant in their study of stage II and III CRC cases receiving leucovorin, fluorouracil, and oxaliplatin (FOLFOX) treatment. Kim et al. [23] found CIMP to be an insignificant predictor of DFS in their study of metastatic CRC cases treated with FOLFIRI. Insufficient (CIMP × adjuvant chemotherapy) interaction term estimates were available for pooling so no statistical summary was attempted. Overall, four out of seven studies agreed that adjuvant chemotherapy conferred a DFS benefit among CIMP(+) stage II and III CRC patients; of the remaining three studies, one concluded to the contrary and two found CIMP to be an insignificant predictive factor.

overall survival

Nineteen studies provided data regarding the effect of CIMP status on OS among CRC patients of different disease stages and genetic profiles. Of these, 13 studies [9–11, 22, 28–30, 37, 41, 44, 45, 47, 53] concluded that CIMP had no significant effect on OS. Six studies [5, 26, 33, 35, 39, 52] found that CIMP was associated with an unfavorable OS while no study found CIMP to be associated with a favorable OS.

Our pooled analysis showed that CIMP(+) status was associated with an unfavorable OS. A subset of 11 studies provided adjusted HR estimates with no restriction to MSI status suitable for pooling, accounting for a total of 3,559 patients. Of these, 7 studies have classified CIMP in a dichotomized fashion. Their overall summary HR estimate was 1.43 (95% CI: 1.18–1.73). No significant statistical heterogeneity was present (Q = 4.03, I² = 0%). Of the other four studies that have classified CIMP in a trichotomized fashion, both CIMP-high and CIMP-low were associated with a significantly worse OS in comparison with CIMP-zero (Summary HR 1.53 with 95% CI 1.11–2.12 and 1.33 with 95% CI 1.11–1.61, respectively). From the forest plot (Figure 3), we could infer that despite a lack of significance in many studies, they shared trends in the same direction which led to a significant overall summary HR. To assess the potential effect of including conference abstracts, a sensitivity analysis was carried out by excluding studies that was not a full publication [35]. The exclusion made no difference to our conclusion (pooled HR = 1.59, 95% CI, 1.16–2.19; Q = 2.22, I² = 0%). Sensitivity analysis was also carried out to assess the effect of including indirectly estimated HRs by leaving out three studies [7, 46, 52] with imputed log HR estimates. The exclusion made no difference to our conclusion (pooled HR = 1.69, 95% CI, 1.24–2.31; Q = 3.99, I² = 0%). In order to assess the impact of including studies of variable risk of bias, a subgroup analysis excluding studies [7, 9, 11, 30, 46] with ≥1 high risk of bias item showed essentially the same conclusions (pooled HR = 1.83, 95% CI, 1.23–2.73; Q = 3.49, I² = 14%).

Figure 3.

Forest plots of hazard ratios (HRs) of overall survival (OS) in studies of colorectal cancer (CRC) patients associated with CpG island methylator phenotype (CIMP) with no restriction to MSI status. CIMP is associated with a worse OS among CRC patients after adjusting for other prognostic factors.

Nine studies examined the relationship between CIMP and OS among MSS tumors. In this cohort, CIMP was reported to be associated with a significantly shorter OS in 6 studies [7, 9, 10, 29, 33, 39], while 3studies [11, 37, 44] found no significant association between CIMP and OS. Restricting the meta-analysis to the four studies with suitable summary HR estimates, accounting for a total of 1,112 MSS cases, produced a summary HR estimate of 1.37 (95% CI 1.12–1.68) with acceptable statistical heterogeneity (Q = 4.45, I² = 33%), suggesting that CIMP was associated with a significantly shorter OS amongst MSS tumors also (Figure 4).

Figure 4.

Forest plots of hazard ratios (HRs) of overall survival (OS) in studies of microsatellite stable (MSS) colorectal cancer (CRC) patients associated with CpG island methylator phenotype (CIMP). CIMP is associated with a worse OS after adjusting for other prognostic factors among a subgroup of CRC patients with MSS.

Four studies examined the relationship between CIMP and OS among MSI tumors. Rhee et al. [39] found CIMP(+) tumors to have shorter OS while Zlobec et al. [52], Ward et al. [9], and Barault et al. [7] all found no significant association between CIMP and OS. Insufficient numerical estimates were available for pooling.

discussion

Among patients included in our review, CIMP-positive accounted for almost one fifth (18.2%) of all sporadic CRCs. Understanding the implication of its presence on patient prognosis would be crucial in making management decisions for CRC patients.

One of the major confounding factors in a systematic review on topics relating to CIMP was the lack of a standardized operational definition of CIMP. No consensus existed regarding the most appropriate gene panel, marker threshold, or laboratory method for the assessment of CIMP [34]. Several studies carried out sensitivity analysis by employing different gene panels or marker thresholds and concluded that this could result in different conclusions regarding the prognostic value of CIMP [11, 26, 29, 33]. Despite the methodological heterogeneity, our review showed no gene panel, laboratory method, or preservation method to be consistently associated with a better or worse prognosis in patients with CIMP, thus a decision was made to carry on with the numerical synthesis. The results showed that CRC patients with CIMP were likely to experience both shorter DFS and OS after adjusting for their age, gender, disease stage and the treatment modalities they received.

A number of other genetic profiles have also been cited as being significant effect modifiers in the relationship between CIMP and CRC patient prognosis, such as MSI status, mutation status of BRAF, KRAS, or anti-p53 autoantibody. While most studies adjusted for established confounders such as age, gender, and tumor stage, the genetic markers such as KRAS/BRAF and MSI were not consistently assessed and adjusted in all studies. This was most likely due to conflicting results in previous literature regarding the prognostic value of these markers [25, 27, 36] and non-significant finding during univariate analysis in the respective studies. In our review, we were able to perform subgroup analysis for MSS CRC patients and we found that the survival disadvantage associated with CIMP persisted in this subgroup. There was not enough data among the included studies to allow data pooling for exploration of whether the same disadvantage held true among the MSI patients or other genetic subgroups. Of the included studies, Barault et al. [7], Dahlin et al. [8] and Ward et al. [9] all found MSI status to be a significant effect modifier in the relationship between CIMP and survival; Kakar et al. [28, 37] and Samowitz et al. [44] found BRAF mutation to be a predictive factor for poor prognosis among MSS cases independent of CIMP status; in contrast, Lee et al. [10] found KRAS/BRAF mutation to be an effect modifier in the relationship between CIMP and poor prognosis. Zlobec et al. [52] argued for a combination of BRAF/CIMP as the most appropriate marker combination for classifying distinct prognostic groups. The dilemma of wishing neither to leave out potential confounders from the regression model nor overfit the model with mutually correlated markers could not be solved unless the pathway between CIMP and survival or treatment response, along with the roles MSI, BRAF, and KRAS play in it, had been elucidated.

Besides the prognostic implication of CIMP among CRC patients, an even more important question was, should CIMP status influence the decision in giving adjuvant chemotherapy to CRC patients? According to National Comprehensive Cancer Network (NCCN) treatment guideline for colon cancer Version 3.2013 [24], neither the use of adjuvant chemotherapy outside of clinical trials for stage II (T2N0M0) colon cancer patients nor the use of multi-gene assay in the decision-making process was recommended. However, if we were able to identify a subgroup of stage II tumors that were more susceptible to 5-FU, we could delay disease recurrence with adjuvant chemotherapy in this subgroup. Three out of four studies in our review showed that patients with CIMP experienced significantly longer DFS with the addition of adjuvant chemotherapy compared with receiving surgery alone. On the other hand, four out of five studies in our review showed that patients without CIMP had equivalent outcomes whether they received adjuvant chemotherapy or not (Table 4). According to Reporting Recommendations for Tumor Marker Prognostic Studies (REMARK) [40], the recommended summary statistic for evaluating the interaction of treatment and a binary (or categorical) marker like CIMP was the interaction term. Of the seven studies in our review, interaction term was available in only one study [16], thus we were unable to perform analysis beyond descriptive summary across studies for evaluation of the impact of CIMP on treatment response. CIMP's value as a predictive factor in assessing whether adjuvant 5-FU therapy will confer additional survival benefit to CRC patients remained to be determined through future prospective randomized studies.

The main limitations of our review were the clinical and methodological heterogeneity across the included studies. While most studies excluded hereditary CRCs and patients with inflammatory bowel diseases, the distribution of cancer stage, treatments received, and genetic profiles varied between studies. In terms of the methodological heterogeneity, besides an inconsistency in CIMP definition, the length and completeness of follow-up, and the confounders assessed and adjusted for were also different from study to study. In this review, we have elected to use DFS and RFS synonymously, in order to include the maximal number of relevant clinical studies, acknowledging that we risk the bias that could be introduced through the small percentage of patients with second primary cancers.

In summary, this is the first comprehensive qualitative and quantitative synthesis of the currently available literature regarding CIMP's prognostic value among CRC patients. Our results showed significantly worse DFS and OS among CRC patients with CIMP compared with those without CIMP, and a potential survival benefit in CRC patients with CIMP who received adjuvant chemotherapy in comparison with those who received surgery alone. However, future research into the most appropriate operational definition of CIMP and the mechanism by which CIMP influenced disease prognosis would be of vital importance in helping us put past study results into proper perspective.

funding

This paper was supported by grants from National Cancer Institute K23 CA127141 (NA), the American College of Surgeons/ Society of University Surgeons (NA).

disclosure

The authors have declared no conflicts of interest.