Postoperative serum methylation levels of TAC1 and SEPT9 are independent predictors of recurrence and survival of patients with colorectal cancer


  • We thank H. H. Li and S. F. Chong for their help in statistical analyses. We are also grateful to C. T. T. Loi for accrual of clinical data.



Serum carcinoembryonic antigen (CEA) is the only marker recommended for surveillance of colorectal cancer (CRC) recurrence; its sensitivity and specificity, however, are suboptimal. This study sought to evaluate the values of postoperative serum methylation levels of 7 genes for prognostication and especially for recurrence detection after curative resection.


This prospective cohort study included 150 patients with stage I-III CRC from whom 3 consecutive blood sampling was taken 1 week before, and 6 months and 1 year after operation. Methylation levels of 7 genes were evaluated via quantitative methylation-specific polymerase chain reaction. Serum CEA was measured in parallel. Univariate and multivariate survival analyses were followed by construction of receiver operating characteristic curves for recurrence detection.


After a median follow-up of 59 months, 43 patients (28.7%) developed recurrent lesions. High serum methylation levels of TAC1 in serum at 6-month follow-up (6M-FU), and SEPT9 at 1-year follow-up (1Y-FU) were independent predictors for tumor recurrence and unfavorable cancer-specific survival (CSS) (P < .05 in all tests). Serum NELL1 methylation levels were significant alone for CSS at both 6M-FU and 1Y-FU, but not for disease-free survival. Dynamic changes of TAC1 and SEPT9 with methylation increment were also independently predictive for recurrence (P < .05 in all tests). More importantly, TAC1 at 6M-FU and SEPT9 at 1Y-FU exhibited earlier detection of potential recurrences compared with concurrent serum CEA.


Levels of TAC1 and SEPT9 methylation detected in postoperative sera of patients with CRC appear to be novel promising prognostic markers and may probably be considered for monitoring of CRC recurrence. Cancer 2014;120:3131–3141. © 2014 American Cancer Society.


Approximately 25% to 40% of patients who undergo curative resection of colorectal cancer (CRC) develop tumor recurrence with eventual demise.[1] Resection rates for recurrent lesions remain low (17.4%-54.8%) [2] although survival benefits have been described.[3, 4] Resection rates may potentially be increased if recurrence can be diagnosed earlier, where lesions are small, localized, and clear resection margins are achievable. Alternatively, early institution of chemo or radiotherapy may maintain recurrent disease at small volumes with minimal or no symptoms.

Current modalities for postoperative CRC surveillance include thorax/abdominal/pelvic CT scans, PET scans, colonoscopies and serum carcinoembryonic antigen (CEA) measurement. CEA, the only blood marker recommended in established guidelines, however, has poor sensitivity or specificity.[5] Normal CEA values may be found in almost 50% of cancers before surgical resection and often do not rise during recurrences.[5] CEA elevation also has a slow lead time predating the clinically identifiable recurrence by approximately 5 months.[6] The other modalities have various limitations of costs, radiation exposure, and invasiveness with potential complications thus restricting repeated use.

Other putative recurrence markers present in postoperative blood have been suggested. These include circulating tumor cells (CTC),[7] carbohydrate antigen 19-9 (CA 19-9), mannan-binding lectin-associated serine protease-28 and S100A4 messenger RNA (mRNA).[9] None of these markers, however, are recommended in established guidelines[10] due to limitations of small sample size, presence of inconsistent results, or lack of verification in large, independent sample series.[7-9]

Epigenetic silencing of tumor-related genes by promoter hypermethylation is a common event in various cancers including CRC. Although a small number of tumor-specific methylated genes detected in plasma or serum of CRC patients demonstrate diagnostic potentials for CRC, far fewer of these circulating features are investigated for prognostic relevance. The only promising candidate, serum methylated HLTF,[11, 12] failed to replicate the prediction of CRC recurrence in a recently completed validation study.[13]

We have previously identified 7 genes (tachykinin-1 [TAC1], MAL, septin 9 [SEPT9], nel-like type 1 [NELL1], somatostatin [SST], cellular retinoic acid-binding protein 1 [CRABP1], and eyes absent homolog 4 [EYA4]) with higher methylation levels or frequencies in preoperative sera of CRC patients compared to age-matched healthy controls.[14] These genes demonstrate cancer-specific methylation in tumor tissues,[14] but their utility as a surveillance tool remains unknown. The aim of this study was to evaluate these 7 genes as CRC prognostic markers and their ability to detect cancer recurrence.


Patients and Sample Collection

This prospective study included 150 consecutive patients with sporadic stage I-III CRCs in a single institution (Singapore General Hospital) between October 2003 and June 2005. Patients with inflammatory bowel disease, recurrent colorectal cancer, family history suggestive of Lynch Syndrome defined by Amsterdam criteria, or familial adenomatous polyposis were excluded. Curative surgical treatment was defined as absence of gross residual tumor after resection and negative margins confirmed pathologically. One of 150 patients had neoadjuvant treatment before surgery and other 45 patients received adjuvant chemotherapy and radiotherapy (Table 1). Informed consent was obtained and the study was approved by the Institutional Review Board of the Singapore General Hospital.

Table 1. Summary of Neoadjuvant or Adjuvant Therapy in 46 Colorectal Cancer Patients
PatientChemo AimPrimary Tx TypeChemo RegimenaChemo Duration (mo)RT MomentRT Duration (wk)RT Site
  1. Abbreviations: chemo, chemotherapy; RT, radiotherapy; Tx, treatment.

  2. Chemotherapy treatments—CI 5FU: continuous infusion of 5-fluorouracil; bolus 5FU: bolus 5-fluorouracil; FOLFOX: a regimen made up of folinic acid (leucovorin), fluorouracil (5FU), and oxaliplatin; XELOX: a regimen made up of Xeloda and oxaliplatin.

  3. a

    In statistical analysis, patients who received regimen of CI 5FU, bolus 5FU, or Xeloda were reclassified as one group, whereas those with FOLFOX or XELOX as another group.

1adjuvantchemoRTCI 5FU6Post-op RT6Rectum
2adjuvantchemoXeloda6No RT0 
3adjuvantchemoFOLFOX6No RT0 
4neoadjchemoRTXELOX8pre-op RT6Rectum
5adjuvantchemoXeloda6No RT0 
6adjuvantchemoRTXELOX6Post-op RT6Rectum
7adjuvantchemoRTCI 5FU6Post-op RT6Rectum
8adjuvantchemoBolus 5FU6No RT0 
9adjuvantchemoXeloda2 (Stop due to toxicity)No RT0 
10adjuvantchemoXeloda6No RT0 
11adjuvantchemoXELOX6No RT0 
12adjuvantchemoRTXELOX6Post-op RT6Rectum
13adjuvantchemoXeloda6No RT0 
14adjuvantchemoRTUnknown6Post-op RT6Rectum
15adjuvantchemoBolus 5FU0No RT0 
16adjuvantchemoXeloda6No RT0 
17adjuvantchemoRTXeloda6Post-op RT6Rectum
18adjuvantchemoRTCI 5FU6Post-op RT6Rectum
19adjuvantchemoBolus 5FU6No RT0 
20adjuvantchemoXELOX6No RT0 
21adjuvantchemoRTFOLFOX6Post-op RT6Rectum
22adjuvantchemoXeloda6No RT0 
23adjuvantchemoXeloda6No RT0 
24adjuvantchemoXELOX6No RT0 
25adjuvantchemoBolus 5FU6No RT0 
26adjuvantchemoRTBolus 5FU6Post-op RT6Rectum
27adjuvantchemoRTBolus 5FU6Post-op RT6Rectum
28adjuvantchemoRTXeloda6Post-op RT6Rectum
29adjuvantchemoRTCI 5FU5 (Stop due to toxicity)Post-op RT6Rectum
30adjuvantchemoRTCI 5FU6Post-op RT6Rectum
31adjuvantchemoRTXeloda6Post-op RT6Colon
32adjuvantchemoXeloda6No RT0 
33adjuvantchemoXeloda6No RT0 
34adjuvantchemoBolus 5FU6No RT0 
35adjuvantchemoRTXeloda6Post-op RT6Rectum
36adjuvantchemoXeloda6No RT0 
37adjuvantchemoRTCI 5FU6Post-op RT0Rectum
38adjuvantchemoXeloda5 (Stop due to toxicity)No RT0 
39adjuvantchemoXeloda1 (Stop due to toxicity)No RT0 
40adjuvantchemoRTXeloda1 (Stop due to toxicity)Post-op RT6Rectum
41adjuvantchemoRTBolus 5FU6Post-op RT6Rectum
42adjuvantchemoBolus 5FU6No RT0 
43adjuvantchemoRTFOLFOX2 (Stop due to other reason)Post-op RT3 (Stop due to toxicity)Rectum
44adjuvantchemoXELOX6No RT0 
45adjuvantchemoXeloda6No RT0 
46adjuvantchemoXeloda6No RT0 

Postoperative CRC surveillance was via an established protocol. Briefly, postoperative patients were followed up at 3-month intervals for the first 2 years, 6-monthly for the next 2 years and then yearly thereafter. At each consultation, CEA levels were measured and full history and physical examination (including digital rectal examination) were performed. Colonoscopy was performed within 6 months of surgery if initial complete evaluation was not possible preoperatively due to tumor obstruction or stenosis. Those who had an initial complete evaluation underwent colonoscopy at first year follow-up, and again at 3-year intervals postoperatively. Patients with suspicious symptoms and signs or rising CEA trend on follow-up will be evaluated earlier with colonoscopy and/or radiological imaging including computerized tomography of the chest, abdomen and pelvis, bone scan and positron emission tomography scans where necessary.

Local recurrence was defined as clinical, radiological, and/or pathologically evident tumor of the same histological type at or in the region of anastomosis. Distant recurrence was defined as clinical or radiological evidence of systematic spread outside the primary tumor basin. Mortality dates and causes of death were obtained from the Singapore Cancer Registry.

Sample Processing and DNA Isolation

Blood samples are collected longitudinally at 3 time points: 1 week before surgery, 6-month follow-up (6M-FU), and 1-year follow-up (1Y-FU). Fresh tumor tissues were collected upon surgical removal. Sample processing and DNA isolation were performed as described.[14]

Methylation Analysis

Genomic DNA underwent bisulfite conversion by using Epi-Tech kit according to manufacturer's protocol (Qiagen, Hilden, Germany). Bisulfite-converted DNA was subjected to subsequent methylation-specific quantitative PCR (qPCR) as described.[14] Quantities of 7 target genes and 1 control gene β-actin (ACTB) were interpolated from respective standard curves constructed from 5 to 6 serial dilutions of a methylated DNA standard.[14] Each reaction was run in duplicate or triplicate. Every plate included a positive control and a no-template control. Methylation levels of genes of interest were normalized by dividing the gene/ACTB ratio of a sample by the gene/ACTB ratio of a positive control and multiplying by 1000. Normalized methylation value (NMV) was used as a measure representing relative levels of methylation in each sample.

Serum CEA Measurement

Serum CEA was determined by a microparticle enzyme immunoassay on the Abbott Axsym analyzer according to manufacturer's instructions (Abbott Laboratories, Abbott Park, Ill).

Statistical Analysis

Serum methylation levels of all 7 genes and CEA were examined as both continuous and dichotomous variables. Differences of methylation magnitude between groups were compared via Mann-Whitney U test followed by Bonferroni correction. Disease free survival (DFS) time was calculated from the date of surgery to presentation of clinical or pathological evidence of disease recurrence, or to the last contact on or before January 31, 2010 (if no distant recurrence was recorded during this follow-up period). Cancer-specific survival (CSS) was calculated from the date of surgery to the date of death or the last contact on or before January 31, 2010 (if no death was recorded during this follow-up period). Kaplan-Meier survival curves were compared using the log-rank test. Hazards ratio (HR) was computed by univariate Cox proportional hazards regression model followed by multivariate Cox model analysis with stepwise backward procedure to remove variables from the regression model. ROC curves for recurrence detection were constructed and optimal cutoff values of serum methylation markers were determined based on Youden index. Assay sensitivity was defined as the proportion of recurrent cases that had serum levels above cutoff values. Sensitivity differences were examined by McNemar's test. Association between methylation levels in sera and tumors was evaluated using Spearman's correlation test. Statistical analysis was performed using the Statistical Package for Social Sciences, version 17.0 (SPSS Inc, Chicago, Ill). All statistical tests were 2-sided and P values less than .05 were considered statistically significant.


Characteristics of Clinicopathological Factors and Serum Markers

Clinicopathological characteristics of these 150 patients were obtained from a prospectively maintained computerized database and are summarized in Table 2. Median age at diagnosis was 67 years (range, 33-88 years).

Table 2. Clinicopathological Characteristics of 150 Colorectal Cancer Patients
ParameterCases N (%)
  1. a

    Based on American Joint Commission on Cancer guidelines, 5th edition.

  2. b

    Right colon includes cecum through transverse colon, whereas left colon includes splenic flexure, descending colon, and sigmoid colon.

SexMale85 (56.7)
 Female65 (43.3)
TNM stagingaI26 (17.3)
 II62 (41.3)
 III62 (41.3)
Depth of tumor invasionT17 (4.7)
 T224 (16.0)
 T3107 (71.3)
 T412 (8.0)
Lymph nodal statusNo lymph node involved8 (58.7)
 1-3 lymph node involved36 (24.0)
 ≥4 lymph node involved26 (17.3)
Tumor differentiationWell25 (16.7)
 Moderate114 (76.0)
 Poor11 (7.3)
Histological typeAdenocarcinoma145 (96.7)
 Mucinous carcinoma5 (3.3)
Perineural invasionYes18 (12.0)
 No128 (85.3)
 Not recorded4 (2.7)
Vascular embolismYes25 (16.7)
 No120 (80.0)
 Not recorded5 (3.3)
Tumor sitebRight colon19 (12.7)
 Left colon71 (47.3)
 Rectum60 (40.0
Preoperative serum CEA≤3.5 ng/mL66 (44.0)
 >3.5 ng/mL83 (55.3)
 Not tested1 (0.7)
Neoadjuvant therapyYes1 (0.7)
Adjuvant therapyYes45 (30.0)
 No104 (69.3)

After a median follow-up period of 59 months (range, 5-79 months), 43 patients (28.7%) developed either local or distant recurrence (8 and 35 cases, respectively) as defined by clinical diagnostic criteria. Forty (90.9%) patients had died from disease progression.

All serum markers (methylation markers and CEA) were non-normally distributed at any of the 3 sampling time points. Similar to CEA, the overall methylation levels of all 7 genes decreased with time (the highest before surgery and the lowest at 1Y-FU; Fig. 1).

Figure 1.

Serum normalized methylation values (NMVs) of 7 genes in healthy subjects (N = 26), and stage I-III colorectal cancer patients before surgery (N = 150), at 6-month follow-up (6M-FU) and 1-year follow-up (1Y-FU) stratified by recurrence status. Box plots represent 25th and 75th percentiles with midline inside representing median. Upper and lower bars represent 10th and 90th percentiles dots beyond which were outliers. Mann-Whitney U test was run to examine differences of serum methylation levels between healthy controls (reported previously[14]) and preoperative colorectal cancer patients, between subgroups with and without recurrence with blood sampling at 6M-FU or 1Y-FU. Only significantly different pairs are denoted. Pre-op, preoperative; Rlps, relapse; Non-Rlps, nonrelapse.

Association of Serum Methylation Levels With Cancer-Specific Survival

Serum methylation levels were dichotomized as “high” and “low” according to median levels at respective follow-up time-points. Dichotomized CEA status was generated according to the reference cutoff value of 3.5 ng/mL. At the preoperative time points, age, TNM stage, and vascular embolism were significantly associated with cancer-specific survival (CSS, P = .012, P = .043, and P = .020, respectively). Except for SST gene, none of other serum methylation markers or CEA significantly affected CSS. In the follow-up period, patients with high methylation levels of TAC1 and NELL1 measured at 6M-FU experienced a significantly higher risk for cancer-specific death (P < .001, P < .05, respectively, Table 3). At 1Y-FU, SEPT9 and NELL1 were significant and independent prognostic factors of CSS (P < .01, P < .001, respectively, Table 3).

Table 3. Adjusted Hazard Ratios (95% Confidence Interval) of Independent Prognostic Factors for Cancer-Specific Survival by Multivariate Cox Proportional Hazards Analysis
FactorAt 6-Month Follow-Up (N = 144)aAt 1-Year Follow-Up (N = 137)b
  1. a

    At the end of follow-up (median: 59 months), 39 of these 144 cases were deceased.

  2. b

    At the end of follow-up (median: 59 months), 34 of these 137 cases were deceased.

  3. c

    Serum methylation markers were dichotomized by medians of methylation levels at respective follow-up time points.

  4. d

    A sample was classified under “high methylation” if any component gene alone was highly methylated.

  5. *P ≤ .05; **P ≤ .01; ***P ≤ .001.

  6. NA indicates not analyzed, because statistical significance was not reached in the initial univariate survival analysis. NS indicates not significant.

TNM stage (AJCC 5th edition)  
III vs I3.29 (1.01-10.72)*6.78 (1.64-28.10)**
Perineural invasion  
Yes vs No2.73 (1.16-6.41)*NA
Histological type  
Mucinous carcinoma vs adenocarcinomaNA7.65 (1.96-29.90)**
Serum CEA  
>3.5 ng/mL vs ≤3.5 ng/mL2.92 (1.47-5.79)**14.17 (6.41-31.33)***
Methylation of single gene  
TAC1 (High vs Low)c4.12 (1.76-9.61)***NS
SEPT9 (High vs Low)cNS2.69 (1.26-5.73)*
NELL1 (High vs Low)c2.51 (1.05-5.98)*4.41 (1.90- 10.22)***
Combined methylation  
TAC1 and/or SEPT9 (High vs Low)d4.07 (1.69-9.82)**NS
TAC1 and/or NELL1 (High vs Low)d4.84 (2.00-11.67)***2.41 (1.05- 5.55)*
SEPT9 and/or NELL1 (High vs Low)dNS3.72 (1.67- 8.26)***
TAC1 and/or SEPT9 and/or NELL1 (High vs Low)d4.67 (1.81-12.04)***NS

The combined efficacy of TAC1, SEPT9, or NELL1 was also studied. A sample was classified under “high methylation” if any component gene alone was highly methylated. Multivariate analyses revealed that high levels of the combination variable consisting of all 3 genes measured at 6M-FU predicted a higher risk for cancer-specific death (HR = 4.67, 95% CI = 1.81-12.04, P = .001, Table 3) compared to individual genes alone.

Association of Serum Methylation Levels With Disease-Free Survival

At preoperative time points, vascular embolism, perineural invasion, serum CEA, and methylation levels of SST were significant prognostic factors for cancer recurrence (P = .010-.035). At postoperative time points, TAC1 at 6M-FU and SEPT9 at 1Y-FU were significant factors. The incidence of recurrence was significantly higher in the group with high TAC1 methylation levels at 6M-FU compared with lower level group (44.0% versus 13.3%, P < .001). Disease-free survival was also significantly inferior for patients with high TAC1 methylation (Fig. 2). This was similarly observed in methylated SEPT9 at 1Y-FU (50.0% versus 19.4%, P < .001). After adjustment for other significant factors, patients with high serum methylation of TAC1 at 6M-FU (HR = 5.72, 95% CI = 2.67-12.28, P < .001, see Table 4 for univariate and multivariate Cox analysis results) and high serum methylation of SEPT9 at 1Y-FU (HR = 3.50, 95% CI = 1.67-7.32, P = 0.001, Table 5) had higher recurrence risk.

Figure 2.

Disease-free survival probabilities in CRC patients stratified by serum methylation levels of TAC1 at 6 months, and SEPT9 at 1 year after tumor resection. Methylation levels were dichotomized according to medians at respective follow-up time-points. For assessment of gene combination, a sample was classified under “high methylation” if any component gene alone was highly methylated. 6M-FU, 6-month follow-up; 1Y-FU, 1-year follow-up. The difference in survival curves was examined by log-rank test.

Table 4. Cox Analysis of Disease-Free Survival (DFS) at 6-Month Follow-Up in 150 Colorectal Cancer Patients With Stage I-III Tumors
FactorUnivariate AnalysisMultivariate Analysis
No. of PatientsNo. of EventsHR (95% CI)PaHR (95% CI)Pa
  1. Abbreviations: CI, confidence interval; HR, hazard ratio; NS, not significant.

  2. a

    Bold values indicate statistical significance.

  3. b

    Right colon includes cecum through transverse colon, while left colon includes splenic flexure, descending colon and sigmoid colon.

  4. c

    Methylation levels were dichotomized by medians at 6-month follow-up time point.

  5. d

    A sample was classified under “high methylation” if any component gene alone was highly methylated.

Age at diagnosis150431.02 (0.99, 1.05).138  
Sex   .669  
Female65170.88 (0.48, 1.61)   
TNM stage (AJCC 5th edition)   .005 .042
I2631.0 1.0 
II62152.43 (0.70, 8.40) 3.22 (0.92, 11.30) 
III62254.73 (1.43, 15.69) 3.79 (1.12, 12.84) 
Differentiation   .098  
Moderate114311.21 (0.51, 2.90)   
Poor1163.37 (1.08, 10.49)   
Histological type   .529  
Mucinous521.58 (0.38, 6.52)   
Tumor siteb   .982  
Right colon1951.0   
Left colon71201.00 (0.38, 2.68)   
Rectum60181.07 (0.40, 2.87)   
Perineural invasion   .002 NS
No infiltration128321.0   
Perineural infiltration18103.12 (1.53, 6.36)   
Vascular embolism   .002 .003
No embolism120291.0 1.0 
Embolism25132.76 (1.43, 5.32) 2.92 (1.45, 5.88) 
Serum methylation of NELL1c   .090  
High1772.01 (0.89, 4.53)   
Serum methylation of SEPT9c   .035 NS
High34141.99 (1.05, 3.77)   
Serum methylation of EYA4c   .424  
High75231.28 (0.70, 2.33)   
Serum methylation of CRABPc   .529  
High52171.22 (0.66, 2.24)   
Serum methylation of MALc   .309  
High75231.37 (0.75, 2.49)   
Serum methylation of TAC1c   <.001 <.001
Low75101.0 1.0 
High75334.42 (2.18, 8.99) 5.72 (2.67, 12.28) 
Serum methylation of SSTc   .952  
High75211.02 (0.56, 1.85)   
Combined methylationd of TAC1 and/or NELL1   <.001 <.001
Low7091.0 1.0 
High80344.32 (2.07, 9.01) 5.39 (2.45, 11.85) 
Combined methylationd of TAC1 and/or SEPT9   <.001 <.001
Low6271.0 1.0 
High88364.75 (2.11, 10.94) 6.75 (2.76, 16.94) 
Combined methylationd of NELL1 and/or SEPT9   .088  
High46171.70 (0.92, 3.14)   
Combined methylationd of TAC1 and/or NELL1 and/or SEPT9   <.001 <.001
Low5761.0 1.0 
High93374.86 (2.05, 11.54) 6.92 (2.66, 18.04) 
Serum CEA   <.001 .001
≤3.5 ng/mL121291.0 1.0 
>3.5 ng/mL29143.18 (1.67, 2.04) 3.18 (1.62, 6.27) 
Table 5. Cox Analysis of Disease-Free Survival (DFS) at 1-Year Follow-Up in 127 Colorectal Cancer Patients With Stage I-III Tumors
FactorUnivariate AnalysisMultivariate Analysis
No. of PatientsNo. of EventsHR (95% CI)PaHR (95% CI)Pa
  1. Abbreviations: CI, confidence interval; HR, hazard ratio.

  2. a

    Bold values indicate statistical significance.

  3. b

    Right colon includes cecum through transverse colon, whereas left colon includes splenic flexure, descending colon, and sigmoid colon.

  4. c

    Methylation levels were dichotomized by medians at 1-year follow-up time point.

  5. d

    A sample was classified under “high methylation” if any component gene alone was highly methylated.

Age at diagnosis127301.02 (0.99, 1.06).171  
Sex   .120  
Female5490.54 (0.25, 1.18)   
TNM stage (AJCC 5th edition)   .021 .003
II5391.51 (0.41, 5.59) 1.84 (0.50, 6.83) 
III50183.63 (1.07, 12.35) 5.29 (1.52, 18.47) 
Differentiation   .282  
Moderate96210.86 (0.35, 2.13)   
Poor732.30 (0.57, 9.22)   
Histological type   .940  
Mucinous carcinoma411.08 (0.15, 7.92)   
Tumor siteb   .985  
Right colon1431.0   
Left colon61141.02 (0.29, 3.56)   
Rectum52131.05 (0.31, 3.81)   
Perineural invasion   .107  
No infiltration112241.0   
Perineural infiltration1252.21 (0.84, 5.79)   
Vascular embolism   .176  
No embolism107231.0   
Embolism1661.86 (0.76, 4.58)   
Serum methylation of NELL1c   .110  
High1972.00 (0.86, 4.67)   
Serum methylation of SEPT9c   <.001 .001
Low88131.0 1.0 
High39173.62 (1.76, 7.46) 3.50 (1.67, 7.32) 
Serum methylation of EYA4c   .994  
High64151.00 (0.49, 2.04)   
Serum methylation of CRABPc   .442  
High37101.35 (0.63, 2.88)   
Serum methylation of MALc   .521  
High62130.79 (0.38, 1.63)   
Serum methylation of TAC1c   .382  
High61171.38 (0.67, 2.84)   
Serum methylation of SSTc   .900  
High62151.05 (0.51, 2.15)   
Combined methylationd of TAC1 and/or NELL1   .171  
High69201.70 (0.80, 3.64)   
Combined methylationd of TAC1 and/or SEPT9   .109  
High72211.90 (0.87, 4.14)   
Combined methylationd of NELL1 and/or SEPT9   .009 .006
Low77121.0 1.0 
High50182.66 (1.28, 5.53) 2.89 (1.37, 6.11) 
Combined methylationd of TAC1 and/or NELL1 and/or SEPT9   .155  
High78221.80 (0.80, 4.04)   
Serum CEA   <.001 <.001
≤3.5 ng/mL107181.0 1.0 
>3.5 ng/mL20125.86 (2.81, 12.23) 7.29 (3.33, 15.93) 

The combined efficacy of 3 methylation markers was also analyzed for DFS. At 6M-FU, high combined methylation of TAC1 and/or SEPT9, or TAC1 and/or NELL1 was a significant predictor for recurrence. At 1Y-FU, impact on DFS remained significant when SEPT9 and NELL1 were analyzed in combination (Fig. 3).

Figure 3.

ROC curves of serum markers for recurrence detection. (A) TAC1 methylation and CEA measurement at 6-month follow-up, and (B) SEPT9 methylation and CEA measurement at 1-year follow-up. AUC, area under ROC curve; CI, confidence interval.

Association of Dynamic Changes of Serum Methylation Levels With Disease-Free Survival

Dynamic changes of serum methylation levels of the 7 genes were expressed by Δ values. Three intervals defined included the first half-year interval (between preoperation and 6M-FU), second half-year interval (between 6M-FU and 1Y-FU) and 1-year interval (between preoperation and 1Y-FU). For survival analysis, we assigned “methylation increase” to the Δ variables if serum methylation levels increased during any interval, and “methylation decrease” with reduced or unchanged methylation. Multivariate Cox analyses showed that ΔTAC1 during the first half-year interval and ΔSEPT9 during the second half-year and 1-year intervals were independent factors for tumor recurrence (Table 6). It was observed that patients with failed transition from high preoperative to lower or undetectable methylation levels of TAC1 at 6M-FU experienced higher risk for recurrence when compared with the other group (HR = 4.71; Table 6). It is also interesting to note that ΔCEA was only significant within the second half-year interval (Table 6).

Table 6. Adjusted Hazard Ratios (95% Confidence Interval) of Independent Prognostic Factors for Disease-Free Survival by Multivariate Cox Proportional Hazards Analysis
FactorFirst Half-Year Intervala (N = 144)dSecond Half-Year Intervalb (N = 127)e1-Year Intervalc(N = 127)f
  1. a

    From preoperation to 6-month follow-up; bFrom 6-month follow-up to 1-year follow-up; cFrom preoperation to 1-year follow-up.

  2. d,e,fAt end of follow-up (median, 59 months), recurrence diagnosed by clinical criteria was established in 42, 30, and 30 patients, respectively.

  3. g

    Changes of serum levels within an interval with the levels at the initial time point set as the baseline levels.

  4. *P ≤ .05; **P ≤ .01; ***P ≤ .001.

  5. NA, not analyzed as statistical significance was not reached in the initial univariate survival analysis; NS, not significant.

TNM stage (AJCC 5th edition)   
III vs INSNS3.53 (1.04-12.00)*
Perineural invasion  NA
Yes vs No4.81 (2.14-10.78)***NSNA
ΔCEA (Increase vs Decrease)gNS3.14 (1.43-6.90)**NS
ΔTAC1 (Increase vs Decrease)g4.71 (2.30-9.63)***NSNS
ΔSEPT9 (Increase vs Decrease)gNS2.58 (1.23-5.41)*3.35 (1.56-7.19)**

Diagnostic Values of Serum Methylation Markers and CEA for Recurrence Detection

ROC curves were constructed based on postoperative serum levels of methylated TAC1 and CEA at 6M-FU, as well as methylated SEPT9 and CEA at 1Y-FU (P ≤ .001 for all; Fig. 3). Optimal cutoff values yielded 88% specificity for TAC1 at 6M-FU and 80% specificity for SEPT9 at 1Y-FU. Sensitivity of TAC1 at 6M-FU for recurrence detection was significantly higher than that of concurrent CEA concentrations (58.1% versus 32.6%, P = .019; Table 7). The sensitivity differences between serum SEPT9 and CEA levels at 1Y-FU however did not reach statistical significance (Table 7).

Table 7. Sensitivity, Specificity, and Lead Time of Postoperative Serum Markers for Detection of Recurrence
 At 6-Month Follow-UpAt 1-Year Follow-Up
Sensitivity (%)Specificity (%)Lead time (mo)aSensitivity (%)Specificity (%)Lead Time (mo)a
  1. a

    Lead time is defined as a period between detection of high levels of serum markers and definite recurrence established by clinical diagnostic criteria.

  2. *P = .019 by McNemar's test between serum CEA and methylated TAC1.

  3. NA, not analyzed, because statistical significance was not reached in the upstream ROC test.

CEA (Cutoff at 3.5 ng/mL)14/43 (32.6)91/107 (85)5.90-11.812/30 (40.0)87/97 (90)2.60-39.5
TAC1 (Optimal cutoff)25/43 (58.1)*94/107 (88)8.10-46.0NANA  
SEPT9 (Optimal cutoff)NANA  17/30 (56.7)78/97 (80)5.10-40.0

The lead time (ie, period between detection of high levels of serum markers and definite recurrence established by clinical diagnostic criteria) of all serum markers in our series was non-normally distributed. The median lead time of serum TAC1 at 6M-FU was 2.2 months earlier as compared with CEA (Table 7). Similarly, at 1Y-FU time-point, the median lead time of serum methylated SEPT9 was 2.5 months earlier than that of serum CEA (Table 7).

Comparisons of Methylation Levels in Sera and Matched Tumors

The vast majority of tumors harbored methylation in the 3 genes of interest (100%, 99% and 95% for TAC1, SEPT9, and NELL1 respectively, Table 8). Magnitude of methylation was more prominent in tumors than in matched sera (P < .05 in all tests). Correlation between methylation levels in tumor and sera was however not present (P > .05 for all gene pairs). On the other hand, methylated TAC1 and SEPT9 in sera could always be traced back to corresponding tumors that were methylated as well; Unmethylated SEPT9 observed in only 1 tumor was also concordant with unmethylated SEPT9 in the matched serum, demonstrating no false-positive detection of these 2 markers in sera (Table 8). One exception was observed in the analysis of NELL1. Of 8 unmethylated tumors, 1 was paired up with methylated serum, whereas the other 7 remained unmethylated (Table 8).

Table 8. Methylation Levels of TAC1, SEPT9, and NELL1 in Paired Tumor and Serum Samples
Case No. in SerumCase No. in Tumor
  1. NMV indicates “normalized methylation value” generated by dividing the gene/ACTB ratio of a sample by the gene/ACTB ratio of a positive control and multiplying by 1000.

 NMV ≤ 0NMV > 0TotalNMV ≤ 0NMV > 0TotalNMV ≤ 0NMV > 0Total
NMV ≤ 003317172794101
NMV > 001451450767614647


The aim of CRC surveillance protocols after curative resection is early detection of recurrences. Disease recurrence is highest within the first 2 years,[1] but traditional serum markers and tools used have multiple limitations often leading to late detection thus limiting treatment options. In this study, high postoperative serum methylation levels of TAC1 at 6M-FU, SEPT9 at 1Y-FU were independent predictors of CRC recurrence and unfavorable cancer-specific survival (P < .05). Serum methylation levels of NELL1 were significant alone for CSS at both 6M-FU and 1Y-FU but not for DFS. In further analysis, dynamic changes of TAC1 and SEPT9 with methylation increment were also independently predictive of cancer recurrence (Table 6). More importantly, serum methylation of TAC1 at 6M-FU and SEPT9 at 1Y-FU revealed an earlier lead time advantage of more than 2 months compared to concurrent serum CEA. The observed median lead time (5.1-8.1 months) of these 2 markers was also earlier than that of 2 to 4 months reported in a rectal cancer cohort by protein markers of CEA, CA 19-9 and tissue plasminogen activator,[15] and in agreement with median lead time of 7 months (range, 4-10 months) conferred by an mRNA panel consisting of hTERT, CK-19, CK-20 and CEA.[16] To our knowledge, this is the first study demonstrating superiority of circulating methylation markers in earlier detection. Greater sensitivity and earlier detection of recurrences may thus allow a repeat of curative surgical intervention or prompt initiation of more aggressive chemotherapeutic regimens with suppression of further tumor dissemination. Recurrent disease can be thus kept at low volumes with few symptoms conferring a superior quality of life for these patients. Other key advantages of serum methylation assay include that similar to serum CEA, it is minimally invasive, rapid, simple and repeatable. Certainly, further prospective and large studies to validate their full utility and survival benefits are still required.

TAC1 plays multiple biological functions in CRC pathophysiology. TAC1 encodes a neuroendocrine gastrointestinal peptide that is a precursor for hormones including substance P and neurokinin A, affecting secretion, motility and inflammatory reactions of the gastrointestinal tract. Neurokinin A has also been reported to exert antiproliferative effects.[17] SEPT9 is a member of the septin family involved in cytokinesis and cell cycle control. This gene is a candidate tumor suppressor gene also responsible for cancers such as ovarian cancer.

Notably, TAC1 methylation was a significant prognostic factor at 6M-FU, but significance was replaced by methylated SEPT9 at 1Y-FU. We further confirmed that these were not due to effects of adjuvant therapy and noted no correlation of adjuvant treatment with serum methylation levels of TAC1 at 1Y-FU, nor associated with changes during the second half-year interval (P > .05). It has been described that TAC1 methylation intensity is significantly higher in early-stage tumors.[18] SEPT9 methylation in contrast is more frequent in tumors with advanced stage III-IV (64% versus 20% stage I-II).[19] The reasons are not clear at this point in time but indicate the heterogeneity in tumorigenesis. We thus hypothesize that disease recurrence similar to tumorigenesis, may occur and develop via various pathways. Different methylation markers become apparent at different time frames. The use of single methylation markers may thus be not suitable and a combination panel for surveillance is required.

In this study, there were no false-positive methylation of either TAC1 or SEPT9 observed in serum, but methylation of tumor DNA was not always accompanied by parallel detection in serum DNA (Table 8). One possibility is the profound dilution effect of the circulatory system on tumor DNA resulting in very low and undetectable levels with current technologies. Furthermore, there may be insufficient amounts, or shedding of detectable neoplastic DNA may not have occurred during blood sampling. The low abundance of methylation variants present in excess amount of background DNA also confers technical challenges to qPCR further compromising assay sensitivity.

One of the limitations in this study was the relative long intervals between blood sampling. In contrast to the current clinical surveillance protocol of 3-monthly intervals, the study protocol of 6-monthly intervals unfortunately led to 10 cases developing recurrences that preceded the scheduled date for 1Y-FU evaluation. There were thus fewer cases available for DFS analysis and computation of lead time pertinent to SEPT9 methylation measured at 1Y-FU. Furthermore, if the objectives are for earlier detection, a more frequent sampling, for example, at 2-month intervals may be required. We await further data of our ongoing validation study in a larger and independent cohort.

In summary, we have identified 2 novel prognostic markers that may be beneficial for surveillance after curative resection. We have demonstrated their concordance in detecting tumor recurrence as well as a lead time advantage over serum CEA. Additional validation studies are required to fully define the utility of these markers in recurrence surveillance and to assess whether subsequent modification of treatments will result in improved survival.


This work was supported by the National Medical Research Council, Singapore.


The authors made no disclosures.