Background: Several studies have suggested that microsatellite instability (MSI) resulting from defective DNA mismatch repair confers a better prognosis in colorectal cancer (CRC). Recently, however, data have suggested this is secondary to the effects of ploidy/chromosomal instability (CIN). To estimate the prognostic significance of CIN for survival, data from published studies have been reviewed and pooled.
Methods: Studies stratifying survival in CRC by CIN status were identified by searching PubMed and hand-searching bibliographies of identified studies. Two reviewers confirmed study eligibility and extracted data independently, and data were pooled using a fixed-effects model. The principal outcome measure was the HR for death.
Results: 63 eligible studies reported outcome in 10 126 patients, 60.0% of whom had CIN+ (aneuploid/polyploid) tumours. The overall HR associated with CIN was 1.45 (95% CI 1.35 to 1.55, p<0.001). In patients with stage II–III CRCs, the HR was 1.45 (95% CI 1.27 to 1.65, p<0.001). The effect was similar for progression-free survival (HR = 1.71, 95% CI 1.51 to 1.94, p<0.001). There was no evidence of significant interstudy heterogeneity.
Conclusion: CIN is associated with a worse prognosis in CRC, and should be evaluated as a prognostic marker, together with MSI status, in all clinical trials, particularly those involving adjuvant therapies.
Statistics from Altmetric.com
Colorectal cancer (CRC) remains a major health burden, with over 1 million cases worldwide, mostly in the developed world.1 While treatment has advanced,2 the disease-specific mortality remains about 40%,3 and identifying patients who will benefit the most and least from therapy remains an important goal.
Two major types of genomic instability are recognised as alternative mechanisms of colorectal carcinogenesis. The more common, chromosomal instability (CIN), is present in about 65–70% of CRCs. CIN is poorly defined as the presence of multiple structural or numerical chromosome changes in tumour cells, and, in practice, often inferred from finding aneuploidy and/or polyploidy.4
Direct measurement of aneuploidy utilising flow cytometry is relatively crude, and CRCs thus assigned CIN+ status are likely to encompass a variety of chromosomal abnormalities. Generally, a more detailed assessment—for example, using array comparative genomic hybridisation (arrayCGH)—is impractical for large series. Nonetheless, many studies have reported that CIN+ measured by flow cytometry confers a worse prognosis; however, this observation is neither universal5 6 nor always significant.7–9 Consequently, it has been argued consistently that measuring CIN does not add further prognostic information to standard pathological and histological staging.10–13 A recent meta-analysis assessing the prognostic importance of the other major type of genomic instability (microsatellite instability or MSI) in >7500 patients found that MSI+ tumours had a better prognosis than MSI− tumours, lending weight to the assertion that genomic prognostication by MSI status determination alone should be performed.14
However, CRCs are not always positive for only one of either CIN or MSI. In addition to rare MSI+/CIN+ tumours, about a quarter of CRCs display neither form of genomic instability.15–18 It is therefore possible that determining MSI status alone does not capture all prognostic information, and it has recently been suggested that MSI-associated prognostic information is not independent of CIN status.18
The 2006 American Society of Clinical Oncology (ASCO) guidelines state that studies on CIN published since the last guidelines in 2000 are variable and advocate that measuring CIN in CRC is at best an experimental tool. The guidelines recommend that only MSI status should be investigated in large prospective series.19 We have reviewed all the published studies on CIN and used standard techniques of meta-analysis to derive a summary estimate of the prognostic significance of the CIN phenotype for survival.
Peer-reviewed studies of CIN in CRC were eligible if they reported overall survival (OS) stratified by CIN or ploidy status, and a summary statistic could be extracted as described by Parmar et al.20 Studies had to detail how CIN status was determined, and define aneuploidy/polyploidy as the presence of a second peak on the DNA histogram, the first peak corresponding to the diploid cell population. Studies had to be genetically non-selected, and could select patients only for stage or anatomical location (colon and rectum). Where data sets were overlapping or duplicated, only the most recent information was included. Studies only reporting progression-free survival (PFS) or equivalent were not included in the main analyses.
All identified studies were reviewed independently for eligibility by two authors (κ = 0.96). Studies not published in English were excluded after identification (see table 1).
We followed MOOSE (Meta-analysis Of Observational Studies in Epidemiology) guidelines21 to identify appropriate studies (see fig 1). A literature search of studies published up to September 2006 was performed using PubMed and Embase. The search terms were any of “colon cancer”, “rectal cancer” or “colorectal cancer” combined with any of “chromosomal instability”, “ploidy” or “aneuploidy”, combined with either “outcome” or “prognosis”, and the “all related articles” functionality of PubMed. Queries using equivalent terms in other languages did not add to the search in English. Studies thus identified, and all studies cited within, were examined for eligibility. We did not hand-search meeting abstracts, nor did we contact authors to identify unpublished data.
Survival data from eligible studies were summarised using a log hazard ratio (lnHRi) for comparison between CIN+ and CIN− groups. Data from individual studies were extracted by two independent reviewers, and pooled to generate the summary statistic lnHR and var(lnHR) using a fixed-effects model with inverse variance weighting.
If a trial reported observed and expected events in each group, the lnHRi and variance var(lnHRi) were calculated directly. If a trial reported hazards ratio (HR) and CI, these were converted to lnHRi and variance var(lnHRi). Where a direct calculation of lnHRi and var(lnHRi) was not possible, estimates were derived indirectly from other numerical data presented using the methods described by Parmar et al.20
If no numerical data for the estimation of summary statistics were given, data were extracted manually from Kaplan–Meier survival curves: survival rates were estimated at constant time points to reconstruct the lnHRi and var(lnHR), and patient censoring was assumed to be constant during follow-up, starting from the minimal follow-up period.20 If censoring data were presented, censored patients were allocated to the appropriate time interval. Survival curves were magnified to improve the accuracy of the reading.
In one study with 248 patients,22 no deaths occurred in the CIN− group, and 0.5 death was arbitrarily allocated in the last time interval as the resultant lnHRi and var(lnHRi) would otherwise have been uninterpretable. This had no effect on the overall HR and CI.
In six studies representing 674 patients,23–28 it was not possible to extract data by any of the methods described above. None reported a significant difference in outcome between CIN− and CIN+ CRC, and we assigned an lnHRi of 0 (corresponding to HRi = 1), and a var(lnHRi) of a similar sized study to avoid selection bias.29
Subgroup data were extracted as above.
Bias was assessed using the I2 and Q estimates. For values I2⩾50% (considered moderate heterogeneity30), a random-effects model would have been used. Heterogeneity was assessed with Egger’s bias coefficient31 and by funnel plot.32 Sensitivity analysis by meta-regression (empirical Bayes model) was performed to exclude a significant influence of other trial characteristics.33
All statistical analyses were conducted using Stata 9.2 statistical software (Stata Corp, College Station, Texas, USA)
We identified 123 potential studies: 10 were excluded as they were not in English,84–93 two as they were duplicates,94 95 10 as they did not report outcome data in CIN+ patients,96–105 two as they selected patients for age106 107 and one as it selected patients for relapse.108 Fourteen were subsequently superseded by other reports,109–122 seven were excluded as they used non-standard definitions of aneuploidy.123–129 Fourteen studies solely reporting PFS130–143 were only included in the PFS analysis.
The 63 studies included are summarised in table 2. The table does include six studies which did not report outcome data other than indicating that there was a non-significant trend towards worse survival in the CIN+ group.23–28
The 63 included studies analysed 10 126 patients for CIN status and OS. The mean number of patients was 161 per study, with a median number of 138 (range 24–565). Eight studies (1045 patients) were solely based on patients with colonic (non-rectal) carcinoma,7 24 54 61 71 75 79 81 and seven (968 patients) on those with rectal carcinoma.39 41 48 55 56 65 73 Most studies examined a mixture of tumour stages. Seven studies were, however, based solely on single-stage disease: two (273 patients) with stage II;59 66 two (140 patients) with stage III;54 83 and three (123 patients) with stage IV.27 47 52 Other studies also reported details for individual stages.9 18 34 37 40 43 46 57 60 61 64 67 72
Fifty-six studies were conducted in patients of Caucasian origin, six in patients of East Asian origin8 9 42 77 78 82 and one in Indian patients.58 Of the patients, 53.3% were male, based on 39 studies which provided this information.7 8 18 23 26 28 34–39 41 42 46–51 53 55 58–62 66–70 72 73 75–78 81
Determination of CIN status
CIN (aneuploidy/polyploidy) status of CRCs was determined using flow cytometry in 59 studies (9526 patients) and image analysis in four studies (600 patients).27 35 40 59 In addition to the standard definition of aneuploidy, the DNA index, defined as the modal channel position of the G0/G1 peak of the aneuploid cell population divided by the modal channel position of the G0/G1 peak of the diploid reference cells, needed to be above a cut-off point (range >1.0–1.2) in 35 studies.5 9 22 24 25 28 34–37 40–43 46–49 51 52 55 57–59 61 68 69 71–75 77 80 82 The frequency of CIN was 60.0%, the remaining CRCs being classified as CIN− (diploid/near-diploid).
Nineteen studies provided data for a direct estimation of lnHRi and var(lnHRi).7 18 34 38 44 49 51 53 56 59 61 68 70 74–77 80 83 In eight studies, other numerical data were used for an indirect estimation.5 8 36 42 43 55 57 79 Six studies had an HRi of 1.00 and a var(lnHRi) of a similar sized study allocated (see Methods).23–28 For all other studies, lnHRi and var(lnHRi) were estimated from Kaplan–Meier curves.
Except for nine studies, all HRi values were >1.0, indicating that patients with CIN+ cancers had a worse prognosis. Of these, 20 had a lower 95% CI >1, suggesting a significant effect.9 18 36 38 40 43 55–57 59 61 65 68 71–73 76 78 81 83 Of the nine studies where HRi was <1.0, seven presented Kaplan–Meier curves suggesting a worse prognosis for CIN+ tumours.37 39 46 52 60 63 67 These paradoxical findings were due to few remaining patients at the end of the study,39 46 low patient numbers,52 63 low event rate in the CIN− group37 60 and an unmatched drop in survival in the CIN− group during one time interval, skewing the resulting lnHRi.67 Two studies5 6 reported that CIN− tumours fared worse even by Kaplan–Meier analysis. All nine studies had an upper 95% CI >1.
The forest plot in fig 2 shows the HRi and 95% CI for all studies, and the summary HR of 1.45 (95% CI of 1.35 to 1.55, p<0.001). There was no evidence of heterogeneity between studies (Q = 69.10, I2 = 10.3%, p = 0.250). If the six studies23–28 which had an lnHRi of 1 allocated were excluded, the overall effect remained virtually unchanged (HR = 1.47, 95% CI 1.37 to 1.58, p<0.001), with no evidence of heterogeneity (Q = 64.12, I2 = 12.7%, p = 0.213). Analysis of Caucasian patients only gave HR = 1.45 (95% CI 1.34 to 1.56, p<0.001; Q = 63.21, I2 = 13.0%, p = 0.209).
A similar outcome was found for the colonic7 24 54 59 61 66 71 75 79 81 and rectal subgroups39 41 48 55 56 59 65 66 73: for colonic disease, HR = 1.67 (95% CI 1.32 to 2.11, p<0.001; Q = 12.93, I2 = 30.4, p = 0.166); for rectal disease, HR = 1.63 (95% CI 1.33 to 1.99, p<0.001; Q = 12.32, I2 = 35.0%, p = 0.138).
Impact of CIN on PFS
Assessing whether the above estimates were realistic, we analysed 2100 patients in 14 studies130–143 that only reported PFS (see table 3). Most patients who relapse will eventually die from CRC,144 145 and therefore the summary statistic for PFS studies should be similar to those of OS studies. As expected, this was the case (PFS HR = 1.56; 95% CI 1.30 to 1.87, p<0.001; Q = 13.62, I2 = 4.6%, p = 0.401).
Taking all reported PFS outcome in 4026 patients, including studies which also reported OS included above,18 35 38 41 42 60–62 68 74 75 81 the HR was 1.71 (95% CI 1.51 to 1.94, p<0.001, Q = 32.21, I2 = 22.4%, p = 0.152).
Survival by stage
CIN conferred a worse prognosis in 1179 stage II patients,9 18 34 37 43 57 59–61 64 66 67 HR = 1.68 (95% CI 1.25 to 2.25, p = 0.001; Q = 6.12, I2 = 0%, p = 0.865), and in 1177 stage III patients,9 18 34 40 43 54 57 60 61 67 72 83 HR = 1.38 (95% CI 1.14–1.67, p = 0.001; Q = 11.16, I2 = 1.4%, p = 0.430).
Considering those who might be offered adjuvant therapy, data on a further 738 stage II–III patients were available (outcome pooled in individual studies)7 48 51 63 68: in 3094 stage II–III patients the HR was 1.45 (95% CI 1.27 to 1.65, p<0.001; Q = 23.75, I2 = 0%, p = 0.750).
There were insufficient data for stage I patients to reach a meaningful estimate9 43 67; likewise in stage IV, where the HRi was also unreliable and not suitable for pooling, depending on data extracted from Kaplan–Meier curves in seven small cohorts.9 27 46 47 52 67 72
CIN and the effectiveness of therapy
To assess whether CIN+ tumours have inherently different outcomes from CIN− tumours, we analysed all studies in which patients did not receive systemic therapy,23 59 75 81 or which only included non-metastatic disease and enrolled before 1987, when patients rarely received adjuvant chemotherapy.8 54 56 57 66 76 78 79 This provided an indication of the underlying differences in tumour biology affecting outcome. Again, CIN+ patients fared worse (HR = 1.66; 95% CI 1.41 to 1.95, p<0.001; Q = 17.34, I2 = 36.6%, p = 0.098).
To determine if 5-fluorouracil (5-FU)-based adjuvant chemotherapy can modify the worse outcome of CIN+ patients with stage II–III CRC, we pooled the data from the only two studies reporting outcome in this setting.7 18 All patients received adjuvant 5-FU-based chemotherapy, and CIN+ patients had worse outcome compared with diploid patients (HR = 1.85; 95% CI 1.21 to 2.82, p = 0.004; Q = 0.19, I2 = 0%, p = 0.662). It was not possible to draw conclusions regarding the differences in outcome of receiving versus not receiving adjuvant chemotherapy therapy within the groups of diploid and aneuploid patients, respectively.
There were no studies that commented on the combined impact of therapy and CIN status on outcome in stage IV disease.
Publication bias and heterogeneity
Visual assessment of a funnel plot of studies provided no evidence of overt publication bias towards studies reporting a poorer OS associated with CIN (fig 3), nor did formal evaluation of publication bias using Begg’s and Egger’s tests (p = 0.735 and p = 0.101 respectively). On the assumption that significant heterogeneity might have been missed, all analyses were repeated using a random-effects model; this changed neither the direction nor the significance of our findings (overall HR = 1.47, 95% CI 1.36 to 1.58, p<0.001; Q = 69.10, I2 = 10.3%, p = 0.250). An influence analysis in which one study at a time was omitted from the summary estimate confirmed that no study significantly influenced the overall summary statistic (data not shown).
Publication bias introduced by researchers only reporting significant positive findings was a concern, even if over half of the studies included reported non-significant findings. We performed an analysis restricted to studies based on trial patients: four studies reported non-significant HRi7 34 48 49; two reported significant survival differences between CIN+ and CIN– patients.18 80 The summary statistic was very similar to that if all studies were considered (HR = 1.43, 95% CI 1.21 to 1.70, p<0.001, Q = 2.61, I2 = 0%, p = 0.759).
Other potential, non-quantitative sources of heterogeneity (different methods for determining ploidy, use of DNA index, ethnic background, variation in stage and anatomical location) were formally assessed by meta-regression and subgroup analysis: neither revealed any significant associations; study size, length of follow-up, method of data presentation and extraction (direct numerical vs indirect numerical vs graphic) and year of publication were also included and not found to be associated with outcome (see table 4). Ploidy measurement and definition varied very little between studies, and studies with non-standard definitions were excluded123–129; four studies used cytometric image analysis 27 35 40 59 and there is good concordance between this and flow cytometry.146 Exclusion of seven studies in Asian and Indian patients8 9 42 58 77 78 82 did not alter the overall finding; the summary HR and 95% CI of these seven studies alone were similar to the overall summary statistics for all studies (data not shown). PFS was analysed separately from OS.
We have shown that the published data support the view that CIN (ie, aneuploidy/polyploidy) is associated with a worse prognosis in CRC, and, it appears, can stratify CRC patients further after standard pathological staging. Patients with CIN+ CRC appear to have a poorer survival irrespective of ethnic background, anatomical location and treatment with 5-FU. The poorer outcome is found in terms of OS and PFS. CIN influences outcome in patients with stage II–III CRC, irrespective of whether these receive adjuvant therapy (see table 5). It was difficult to determine whether CIN in stage I and IV has prognostic value from the studies included as only around 8% of patients had either stage I or IV disease. While in stage I the data were consistent with an effect of the same direction and magnitude to those in stage II and III (data not shown), in stage IV the problem of low patient numbers was compounded by the highest heterogeneity encountered in our analyses, based on the unreliable data extraction in this subset. It is still possible that CIN has prognostic value in these stages, but requires further study.
Our findings are likely to be robust: they include large numbers of patients and have no evidence of statistical heterogeneity or bias. There was little evidence of qualitative heterogeneity, although some studies did come from different ethnic backgrounds or utilised different methods to detect CIN. For both, the number of studies which did not conform to the majority was small, and, importantly, their exclusion did not significantly alter the summary statistic. Further, the very similar HR for PFS and OS in non-overlapping sets of patients suggests that the finding of worse prognosis in CIN+ CRC is qualitatively and quantitatively correct.
The method of data extraction did not significantly influence the overall HR, as indirect numerical data extraction correlated well with direct methods; analysing outcome by method of data extraction produced very similar significant results (data not shown) and the sensitivity analysis was not significant (table 4).
All but two identified foreign language studies reported a significant decrease in survival in the CIN group in their English abstracts, and exclusion probably makes our estimate conservative. Non-significant findings may be more commonly reported in abstracts, and their exclusion may inflate our estimate. However, Egger et al found that omission of either has only small effects on the HR and CI, while the inability to assess study quality increases heterogeneity.147
It is not clear how CIN status relates to more sophisticated pathological staging beyond the AJCC (American Joint Committee on Cancer) staging employed in the studies analysed. This higher standard may capture some of the information contained in molecular staging, and may even complement it. However, uniform molecular staging should be relatively easy to achieve, while achieving the equivalent pathological staging uniformly may prove more difficult.148 149
Chromosomal instability (CIN) confers worse survival in stage II and III colorectal cancer
No clear relationship could be established in stage I and IV colorectal cancer
The predictive value of CIN for 5-fluorouracil-based chemotherapy could not be determined from the published data
While other publications have suggested that increasing chromosomal changes worsen prognosis, the data analysed were insufficient to establish CIN levels as a continuous prognostic variable
Our findings raise several questions: first, how is CIN measured by flow cytometry related to cancer biology? Assigning CIN+ status based on flow cytometry is a relatively blunt tool for assessing chromosomal changes, and does not distinguish stable and unstable chromosomal abnormalities, nor differentiate simple from complex changes. CRCs which constantly acquire new complex chromosomal abnormalities (unstable CIN) can thus be grouped with tumours which carry the same relatively minor changes in each cell. Disruptions to cell biology are likely to be varied depending on the level of CIN. However, all CIN+ CRCs must have abnormalities which impair faithful replication or segregation of sister chromatids, driving aneuploidy. As such, CIN status by flow cytometry is likely to assess accurately at least one aspect of tumour biology. Whether more sophisticated measures of global CIN, such as numerical and structural complexity and heterogeneity,150 or arrayCGH can refine and add to the CIN concept is not clear at present.
Kern et al151 found that increasing numbers of chromosomes showing loss of heterozygosity (LoH) correlated inversely with prognosis. Whilst we expect overall LoH to co-vary with CIN, LoH can result from several causes and is, at best, an indirect and time-consuming measure of CIN. An analogous analysis regarding the impact of levels of CIN on prognosis was not possible from the published data. Individual chromosomal abnormalities could act as markers for CIN,15 but how the prognostic information of, say, loss of chromosome 18q relates to that of CIN is poorly defined,63 152 even if it could be that 18q loss is the defining abnormality of the CIN−/MSI−group.15
Secondly, what is the prognostic relationship of CIN and MSI? Within individual CRCs, CIN and MSI status are not mutually exclusive: about a quarter of CRCs display neither, and there are rare cases of CIN+/MSI+ tumours.15–18 In line with this report on CIN, in a previous report on the prognostic value of MSI status, we have found that MSI is associated with outcome in stage II–IV disease.14 Unfortunately, neither data set allowed us to relate CIN to MSI status and to tease apart their relative contributions to prognosis. Only one published study has stratified survival by both CIN and MSI status, concluding that the univariate survival benefit in stage II–III CRC associated with MSI+ status was not independent of CIN status in multivariate analysis.18 It remains possible, however, that CIN−/MSI− CRCs differ from CIN−/MSI+ CRCs, with MSI+ affording a better prognosis independent of CIN−. Likewise, a third form of genomic instability, the CpG island methylator phenotype (CIMP), may carry prognostic information—and explain the existence of the CIN−/MSI− group—but its association with MSI may not render it an independent marker.153 There were insufficient data in the literature to try and assess the relationship of CIN and CIMP, but, if future studies assess all three forms of genomic instability, then this relationship may become clearer and lead to prospective trials including CIMP.
Lastly, should CIN status influence the type of chemotherapy given? No study consistently investigated the effectiveness of drugs other than 5-FU, and we cannot comment on these. It is conceivable that diploid patients in the adjuvant setting could be treated less aggressively than CIN+ patients. There is evidence that abnormalities of the spindle checkpoint drive CIN, and in turn promote taxane resistance.154 Given that most CRCs are CIN+, this may explain the poor response of CRC to taxanes observed in phase 1 trials,155 and diploid CRCs could show a better response to taxanes.
In the absence of clinical trials that address different treatment strategies, our findings should drive molecular stratification of patients within clinical trials to determine the contributions of CIN to treatment sensitivity and resistance. In stage II–III, where the published literature allows a firm conclusion regarding the prognostic value of CIN, it should be investigated as a predictive marker.
The association between genomic instability, outcome and benefit from systemic therapy makes it likely that determining the type(s) of genomic instability in CRCs is important. Contrary to current guidelines on prognostic markers in CRC,19 our systematic review of published data suggests that there is likely to be value in determining CIN prospectively, using flow cytometry in conjunction with more sensitive but prognostically less well defined methods. It remains possible that MSI+ status affords a better prognosis independently, and we favour MSI testing until the relationship between CIN and MSI is understood more fully. The precise contribution of each type of genomic instability to prognosis should be evaluated in clinical trials, particularly those involving adjuvant therapy, with an expectation that routine testing for one or both types of instability will be of benefit in clinical practice.
Funding: This study was supported by grants from Cancer Research UK (AW, IT) and the Association for International Cancer Research (RH)
Competing interests: None.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.