Impact of chromosomal instability on colorectal cancer progression and outcome

Background It remains presently unclear whether disease progression in colorectal carcinoma (CRC), from early, to invasive and metastatic forms, is associated to a gradual increase in genetic instability and to a scheme of sequentially occurring Copy Number Alterations (CNAs). Methods In this work we set to determine the existence of such links between CRC progression and genetic instability and searched for associations with patient outcome. To this aim we analyzed a set of 162 Chromosomal Instable (CIN) CRCs comprising 131 primary carcinomas evenly distributed through stage 1 to 4, 31 metastases and 14 adenomas by array-CGH. CNA profiles were established according to disease stage and compared. We, also, asked whether the level of genomic instability was correlated to disease outcome in stage 2 and 3 CRCs. Two metrics of chromosomal instability were used; (i) Global Genomic Index (GGI), corresponding to the fraction of the genome involved in CNA, (ii) number of breakpoints (nbBP). Results Stage 1, 2, 3 and 4 tumors did not differ significantly at the level of their CNA profiles precluding the conventional definition of a progression scheme based on increasing levels of genetic instability. Combining GGI and nbBP,we classified genomic profiles into 5 groups presenting distinct patterns of chromosomal instability and defined two risk classes of tumors, showing strong differences in outcome and hazard risk (RFS: p = 0.012, HR = 3; OS: p < 0.001, HR = 9.7). While tumors of the high risk group were characterized by frequent fractional CNAs, low risk tumors presented predominantly whole chromosomal arm CNAs. Searching for CNAs correlating with negative outcome we found that losses at 16p13.3 and 19q13.3 observed in 10% (7/72) of stage 2–3 tumors showed strong association with early relapse (p < 0.001) and death (p < 0.007, p < 0.016). Both events showed frequent co-occurrence (p < 1x10-8) and could, therefore, mark for stage 2–3 CRC susceptible to negative outcome. Conclusions Our data show that CRC disease progression from stage 1 to stage 4 is not paralleled by increased levels of genetic instability. However, they suggest that stage 2–3 CRC with elevated genetic instability and particularly profiles with fractional CNA represent a subset of aggressive tumors.


Background
Genetic instability is a hallmark of cancer cells and has been proposed to act as a catalyst of cancer development from early stages on [1,2]. It is generally agreed that tumor progression occurs according to a scheme of gradual accumulation of genetic anomalies and that genetic instability is highest in most aggressive and metastatic forms of the disease. In colorectal cancer (CRC), genetic instability is subdivided into three classes; (i) mismatch repair deficiency (MIN), often of hereditary origin but also sporadically acquired, associated with base slippage mostly at poly(A) or poly(C) tracks and near diploid genomes, 15% of CRC (ii) chromosomal instability (CIN) resulting in severely rearranged karyotypes and aneuploidy, 65% of CRC (iii) non-MIN/non-CIN showing a methylator phenotype, 20% of CRC [3]. Major genetic mutations found and acting as key events in CRC, affect the WNT/APC/CTNNB1, KRAS/BRAF, FBXW7, PTEN, SMAD4, TGFBRII, and TP53 genes [4,5]. Interestingly, patterns of mutated genes vary according to the class of CRC. BRAF mutations seem prevalent in MIN, whereas TP53 mutations are essentially found in CIN. Interestingly, genes promoting DNA repair, DNA damage checkpoint as well as translesional DNA replication are mostly down-regulated in CRC tumors compared to proliferating normal adjacent tissues, probably favoring the overall genetic instability at the nucleotide level [6]. In addition to these functionally validated aberrations, CGH based studies have identified widespread copy number alterations (CNA), some of which highly recurrent. Typical CNA patterns in CRC involve gains at 8q, 13q and 20q as well as losses at 5q, 8p, 17p and 18q [7]. These observations were confirmed in higher resolution array-CGH analyses and the boundaries of these regions of CNA defined with greater precision. Moreover, a number of focal events were pointed out [8]. The number of genetic anomalies linked to CRC pathogenesis is elevated and has risen with recent large scale sequencing efforts [9]. However, questions remain as to the role of widespread chromosomal instability in the course of the disease, in particular how these relate to progression of CRCs and patient relapse.
Although the sequential order originally proposed by Fearon and Vogelstein for CRC progression has been disputed the overall model is still regarded as valid [10].
Stepwise progression from normal epithelium, through dysplasia to carcinoma builds on a gradual accumulation of genetic anomalies. Recent work showed that copy number alterations (CNA) set in early in adenomas and reached in progressed adenomas a level similar to that found in carcinomas [6,11]. It has also been suggested on the basis of a meta-analysis of chromosome CGH [7] and array-CGH [8] that progression from invasive cancer to metastasis was accompanied by an increase in the number of CNAs.
However, no clear cut results were proposed ascertaining the existence of a molecular progression scheme between early carcinoma (stage 1), invasive (stage 2 and 3) and metastatic (stage 4) CRC. In this work we wanted to verify whether we could relate CRC progression (from stage 1 to stage 4 and, eventually, to distal metastasis) to a gradual increase of genetic instability and sketch out a sequence of CNA increment. Moreover, we wanted to determine whether genetic instability correlated with patient outcome. To this aim we analyzed a set of 162 CIN CRCs comprising 131 primary carcinoma evenly distributed through stage 1 to 4 and 31 metastases (28/ 31 formed a primary-tumor/matched-metastasis pair) and 14 adenomas by array-CGH. Our data showed that stage 1, 2, 3 and 4 tumors did not differ significantly at the level of their CNA profiles. This led us to ask whether the level of genomic instability, as illustrated by array-CGH, was linked to disease outcome. Based on the Global Genomic Index (GGI), which corresponds to the fraction of the genome involved in CNA and the number of breakpoints (nbBP), which were determined as chromosomal sites where copy number shifts occurred, we defined two classes of tumors showing strong differences in outcome and hazard risk. CNAs correlating with early relapse or death in stage 2 or 3 patient were searched and two regions of copy number loss could be selected due to their strong association to negative outcome.

Patient and tumor samples
Genomic profiles were established on 176 samples: 14 adenomas, 131 primary carcinoma and 31 synchronous (9) or metachronous (20) metastases (among which 28 were paired to their primary tumor). Biological samples were collected in 4 clinical centers of south-west France: Bergonié Institute, Bordeaux; CHU Dupuytren, Limoges; CRLC Val d'Aurelle, Montpellier; Purpan Hospital, Toulouse between 1993 and 2008. Clinical data and follow-up information were collected. Data were anonymized. This project was submitted to the ethics committees of the respective clinical centers participating to the study and was approved by the National Institute of Cancer (INCa) following the recommendations of the French National Authority for Health (FNAH). Patient samples were processed according to French Public Health Code (law n°2004-800, articles L. 1243-4 and R. 1243-61) and the four biological resources center has received the agreement from the French authorities to deliver samples for scientific research. The authorization numbers were AC-2008-812 (Bergonié), AC-2007-34 (Dupuytren), AC-2008-700 (Val d'Aurelle), AC-2008-820 (Purpan). Before surgery patients are informed that their surgical specimens can possibly be used for research purposes. They can refuse this possibility by filling a form to express refusal and in this case tumor biopsies were destroyed.
Clinical characteristics are summarized in Table 1 and further detailed in Additional file 1. Adenomas and carcinomas were surgically removed and immediately frozen at − 80°C. Only samples containing more than 50% of tumor cells were included in the study. Samples were checked for microsatellite instability by microsatellite marker analysis and were all MIN negative. Four (4) patients (TNM stage 4) received a treatment prior to surgery.

Array-CGH
The 176 DNA samples were analyzed on two generations of Integragen BAC-arrays (Integragen, Evry, France) IgV6+ (5015 BACs), IgV7 (5878 BACs), with a median resolution of 0.6 Mb. BACs were spotted in quadruplicate. DNA labeling and hybridization, were done as previously described [12] with slight modifications: 600 ng of DNA were labeled with BioPrime Total Genomic Labeling System (Invitrogen SARL, Cergy Pontoise, France). Arrays were scanned using Axon 4000B scanner (Molecular Devices, CA, USA) and images were analyzed using Genepix 6.0. Data were analyzed in web-based platform for copy number array management and analysis (http://bioinfo-out.curie.fr/CAPweb/). Normalized and replicates filtered data were exported as text file for further analyses. In order to analyze all the data from different Integrachip versions, we used the Nexus 6.0 Software (Biodiscovery, El Segundo, CA, USA). Analysis settings for data segmentation and calling were the following: significant threshold for Rank Segmentation algorithm: 0.005, Max Continuous Probe Spacing: 6000, Min number of probes per segment: 6, high level gain: 0.485, gain, 0.138, loss:-0.153, homozygous copy loss:-0.73. Nexus 6.0 Software was used to calculate frequency plots, factor enrichment (significantly overrepresented factor values in a particular factor group identified using the two tailed Fisher's Exact test at a p-value of p < 0.05), significant chromosomal differences between two groups (comparison tool: two tailed Fisher's exact test with p-value < 0.005 and minimal frequency difference set at 10%) and Survival Predictive Power (log-rank test is used to identify genomic regions yielding a high degree of survival prediction; p-value is calculated by permuting the survival time for each sample and comparing the log-rank statistic for the permuted data to the original data; threshold used was p-value < 0.05).

Genomic quantitative variables calculation
An R script using Circular Binary Segmentation (CBS) algorithm implemented in DNAcopy (Bioconductor for R) and normalized/replicates filtered data as input, were used to determine genomic metrics such as gains, losses, high level gains, homozygous copy losses. For this purpose, the thresholds were as used in Nexus 6.0 analysis (high level gain: 0.485, gain, 0.138, loss:-0.153, homozygous copy loss:-0.73). The GGI was calculated at a probe level as follows: (number of probes gained + number of probes lost) /number of informative probes. The GGI corresponds to the fraction of the genome involved in CNA. The nbBP was determined as the number of transitions or breakpoints in the genomic profiles after smoothing and segmentation of the data. The R script is available upon request.

Statistical analyses
Continuous variables were presented as medians and range, and compared between populations with the Kruskal-Wallis test. Categorical variables were presented using contingency tables and compared with Pearson's chi-square test or Fisher's exact test. Differences were considered statistically significant when p < 0.05. Classes of genetic instability were defined using two quantitative variables as metrics: Global Genomic Index of alteration (GGI) and number of breakpoints (nbBP). First on the whole set of data (n = 176), GGI and nbBP were grouped into three classes using the 33th percentile (first tercile) and the 66th percentiles (second tercile). Then, for stage 2 and 3 set of data (n = 72), number of BP was grouped into two classes, low (<116) and high (>116). Using ROC curves (see Additional file 2) the optimal nbBP threshold was calculated to maximize the Youden's index (sensitivity and specificity minus 1) which induces the best discrimination according to vital status. Statistical associations between GGI or nbBP were calculated using the nonparametric test for trend across ordered groups. To account for multiple testing, the statistically significant threshold was set at 0.01.
Overall survival (OS) was the primary endpoint for this study and was calculated from the date of surgery until the date of death. Relapse-free survival (RFS) was the secondary endpoint and was calculated from the date of surgery until the date of relapse. Patients who died without relapse were censored at the time of death. Patients lost to follow-up were censored at the time of last visit. The Kaplan-Meier method was used to estimate OS and RFS. Survival rates were compared using log-rank test.
Genomic instability variables were significant in univariate analysis and were included into a multivariate Cox proportional hazards model. Using the model, a score was allocated proportional to the regression coefficients. The adjacent non-significant categories were regrouped in order to reduce the number of prognostic categories (see Additional file 3). Hazard rate (HR) and its 95% confidence interval (95% CI) were calculated using Cox model.

Outcomes in our colorectal cancer set
Median follow-up was 48.4 months (range: 1 to 115 months). Median overall survival was not reached.

Copy number alterations in our colorectal cancer set
Genomic profiles were established on our set of 176 colorectal tumors by CGH on BAC-arrays comprising 3000 to 5800 clones (mean resolution 1 to 0.6 Mb). Our sample set corresponded to 14 adenomas, 131 primary tumors and 31 distal metastases. All these CRC samples were selected as microsatellite stable. Overall CNA profiles in our set of tumors were in harmony with those described by others ( [8,13]) ( Figure 1A). Most commonly altered regions (gains or losses in > =35% of the samples) were gains at chromosomes 7p, 7q, 8q, 13q, 20 and losses at 8p, 17p and 18 ( Figure 1A, Additional file 4). High-level gains (HLG) (log2ratio > 0.485) were observed throughout the whole genome. However, only HLGs located at 7p21.3-p11.2, 8q11-q24.3, 13q11-q34 and 20p13-q13.33 occurred in more than 5% of the tumors.

Stratification of CNA profiles and genetic instability according to disease stages
We wanted to determine the existence of copy number changes correlated to disease progression from adenomas to carcinomas and from superficial (stage 1) to invasive (stage 2 and 3) and metastatic cancer (stage 4 and metastases). To this aim, we stratified CGH profiles according to disease stages and metastases ( Figure 1B). Adenomas clearly differed from carcinomas showing less rearranged profiles. This indicated that the transition from benign to malignant tumors was accompanied by a sharp increase in genetic instability. Contrastingly and interestingly, cumulative CNA profiles of stage 1, 2, 3, 4 carcinomas and distal metastases appeared globally similar. To identify regions of CNA associated to the progression from one disease stage to another, we performed pairwise comparisons (adenomas vs. stage 1 carcinomas, stage 1 vs. stage 2, 2 vs. 3, 3 vs. 4 and metastases vs. associated primary tumors) ( Figure 1C and Additional file 5). Most significant changes were seen between adenomas and stage 1 carcinomas, with gains at 8q, 13q, 20 and losses at 8p, 15p, 17p and 18q. Changes associated to stage transition (1 to 2, 2 to 3, 3 to 4) could be found, but were difficult to relate to a coherent scheme of progression. This was exemplified by losses at chromosome 14 and 15, associated to transition from stage 2 to 3 and from stage 3 to 4 ( Figure 1C). Both events were present in stage 2 and absent in stage 3. Strangely, their occurrence went back up in stage 4. This was not consistent with a cumulative progression scheme, in which tumors progress sequentially from stage 2 to stage 3 and end up progressing to stage 4.
Next, we verified whether the level of genetic instability increased according to disease stage using two metrics, Global Genomic Index (GGI), and the number of break-  (Figure 2A, B), but did not change significantly from stage 1 to 4 carcinomas ( Figure 2C, D). We delineated classes of genetic instability based on GGI or nbBP terciles and were intrigued to see that stage 2 and 4 presented a large proportion of GGI-high and/or nbBP-high tumors, while stage 1 showed a prevalence of low instability tumors (see Additional file 6A, B). Moreover, we noted that the nbBP was higher in younger patients (<60y, p = 0.013) and lower in rectal cancers (p = 0.032), irrespective of TNM stage ( Figure 2E-F). However, GGI and nbBP levels did not differ significantly between TP53 wild type and their mutated counterparts.

Genetic instability and outcome of the disease
While different stages of CRC could not be clearly distinguished by their cumulative CNA profiles, it was noticeable that individual tumors showed important differences in genetic instability, some tumors presenting highly rearranged genomes and others only limited numbers of anomalies. This prompted us to determine whether the level of genetic instability could be related to disease outcome. Because stage 1 tumors are associated to a very limited number of recurrence and stage 4 to negative outcome we focalized our analysis on the 72 stage 2 and 3 CRCs present in our dataset (17.5% stage 2 and 40.5% stage 3 CRC patients will eventually show disease progression at 5 years) [14] (see Additional file 7).
Using ROC curves, we determined threshold levels for nbBP that were best fitted to define a group of bad outcome. This defined 2 classes, nbBP-low (<116) and nbBP-high (> = 116). The latter was associated to bad outcome in both RFS (p = 0.02) and OS (p = 0.001) (see Additional file 8A, B). For GGI, three classes were defined (low, medium, high) according to terciles. Best prognosis was found with GGI-low (<0.25) and worst prognosis with GGI-median [0.25-0.41]. Unexpectedly, GGI-high (≥0.41) tumors presented an intermediate  Figure 3A). Using weights of the regression coefficients, we delimited 3 groups of high risk (G3, G4, G6) and 2 groups of low risk (G1 and G5). We combined all high risk and all low risk groups in one high and one low risk class, which showed clear differences in outcome (RFS: p = 0.012, OS: p < 0.001 Figure 3B, C) and Hazard Ratio using either relapse free (HR = 3) or overall survival (HR = 9.7) as an endpoint ( Table 2).

CNAs associated to bad outcome in stage 2 and stage 3 CRCs
The above described risk classes were based on quantitative criteria (GGI and nbBP) defining levels of genetic instability in CRC. The different subgroups that were defined thus presented different levels of genetic instability (see Additional file 9A). However, while G4 (nbBP-High/ GGI-median) and G5 (nbBP-Low/GGI-High) respectively belonged to the high and low risk class, their average numbers of gains, high level gains and losses were similar (Additional file 9B,C). This contrasted with G6 which bore distinctly higher numbers of CNAs.
These results prompted us to search for qualitative differences that may explain the differences in risk of relapse and death. We, thus, searched for specific copy number changes between high and low risk classes of CRCs, aiming at the definition of markers of relapse in stage 2 and 3 colorectal cancer. We used two complementary approaches. First we searched for significant differences in the high and low risk classes. Using the comparison tool of Nexus 6.0 software package we compared chromosomal regions of the high (n = 34) and low risk (n = 38) classes using twotailed Fisher's Exact test and identified 29 differentially represented genomic regions (see Additional file 10). Second, we used the Survival Predictive Power tool of Nexus in order to determine gains and losses significantly correlated with poor survival in the subset of stage 2 and 3 primary CRCs producing a list of 31 genomic regions (see Additional file 10). Losses at 16p13.3 and 19q13.3 were selected in both approaches and  presented the strongest correlation (p < 0.001) to RFS and OS (p = 0.007, p = 0.016 Table 3, Figure 4A-D). Losses at 16p13.3 and 19q13.3 showed significant cooccurrence. All samples with 16p13.3 loss showed concomitant 19q13.3 loss (p < 1×10-8), thus signing for a group of CRC with negative outcome (see Additional file 11). This correlation remained significant in a subset of stage 2 CRCs (not shown). It was noticeable that CNAs at 16p13.3 and 19q13.3 were not restricted to losses and included a sizeable fraction of gains. However, only losses were associated to negative outcome, whereas gains had either no impact on survival risk (16p13.3) or, on the contrary for 19q13.3, were associated to favorable outcome ( Figure 4C). This yin-yang correlation led us to investigate whether losses at 19q13.3 were enriched in high risk groups, while gains were more frequent in low risk groups. Indeed, we found that losses at 19q13.3 were enriched in the high risk group 6 (p = 0.0035), whereas gains were prevalent in low risk group 5 (p = 0.0025) (see Additional file 12). In a second time, we tested in our stage 2 and stage 3 dataset the prognostic significance of focal regions and genes that were previously described in the literature [8,15,16]. Gains and losses were determined for each region bearing these target genes and association with RFS and OS tested. Significant association with short survival was found for only 5/87 genes, namely SMAD4 (p = 0.0045), CCDC68 (p = 0.0054), TCF4 (p = 0.0054), RAX (p = 0.0047) located on chromosome 18q21.2-21.3 and TSKS (p = 0.005) located on chromosome 19q13.3. Previously proposed prognostic regions such as losses at 4p, 4q22-q35, 5q, 6q, 8p, 13q ,14q, 15q, 17p, and gains at 8q, 10q, 20q were not found significantly associated with either RFS or OS in our series of stage 2 and 3 tumors [8,[15][16][17][18].

Discussion
In this work, using array-CGH as an analytical approach, we aimed at determining whether genetic instability was related to progression of colorectal cancer and verify whether it could be used as a prognostic indicator. Colorectal cancer has served as a model of stepwise progression from normal epithelium, through benign growth, into malignant cells that eventually become invasive and acquire metastatic properties [5]. We were interested in determining the existence of a progression scheme between superficial carcinomas (stage 1), invasive carcinomas (stage 2 and 3) and metastatic carcinomas (stage 4) based on a gradual accumulation of genomic alterations, with alterations occurring specifically at each step of progression.
Main regions of gain and loss observed in this work were concordant with previously reported observations on CRC [8,13]. Our data showing a marked increase in the number of CNAs in the transition from adenomas to carcinomas are in concordance with previously reported array-CGH work on early and advanced colorectal adenomas [11]. Most significant changes between adenomas and stage 1 CRCs were the occurrence of gains at 8q, 13q, 20 and losses at 8p, 15p, 17p and 18q in keeping with published works [7,11,19]. However, in contrast to Diep and coworkers [7] who proposed that transition from Duke's B to C stage was associated to increased occurrence of gain at 1q and that Duke's C to D to that of gains at 20q and Xq and loss at 21q, we could not identify CNAs whose occurrence was assigned to the transition from one stage to the next. This was exemplified by loss at chromosomes 14 and 15 which were present in stage 2, absent in stage 3 and present again in stage 4. We found that the number of CNAs in metastases was higher than in primary tumors. Analyzing concomitantly 28 pairs of primary tumors and corresponding metastases, we detected small regions of gains on chromosome 1q, 6p21, 10p and 17q21 and loss at chromosome 8p12 that occurred more frequently in metastases than in primary tumors (see Additional file 5). However, we could not infer the existence of anomalies specifying metastatic invasion as previously proposed [20]. Along similar lines we could not determine differences in CNA profiles between TP53 wild type and mutated tumors [8]. The absence of specific changes associated to the TP53 status in our dataset may be related to sampling differences. As a matter of fact, our series was restricted to CIN CRCs, while that in the work by Sheffer and colleagues (2009) comprised MSI cases that show TP53 mutations at a lower frequency and present fewer CNAs and at different locations than in CIN tumors [21].
We could not associate any qualitative change of CNA to disease progression, but the increase of genetic instability between stage 1 and stage 4 tumors and between primary CRCs and metastases suggested that the global level of genetic instability could be of clinical or prognostic significance in CIN colorectal cancer. We were interested to note that whereas genetic instability was lowest in stage 1 and highest in stage 4 CRC, its level in stage 2, genetic instability was in keeping with that in stage 4. Taken together, our data were not consistent with a model where CRCs progress gradually from stage 1 to stage 4, because of the resemblance of CNA profiles and genetic instability levels in stage 2 and stage 4 tumors. This similarity suggested that a large part of stage 4 may arise directly from most unstable stage 2 tumors. The difficulty to define a progression scheme with increasing numbers of CNAs according to disease stages may be related to the fact that CRC staging is a clinical progression scale which takes local and distal invasion into account and no tumor intrinsic characteristics.
We, hence, asked the question of whether the level of genetic instability could be a prognostic indicator in stage 2 and 3 CRCs. Using the fraction of the genome involved in CNA (GGI) and the number of breakpoints (nbBP) detected in the array-CGH profiles, we defined 5 groups of CRC. These groups differed in their level of genetic instability, but also their profile of anomalies. Indeed, tumors with low nbBP presented CNA involving large chromosomal regions (whole chromosomes or chromosomal arms), whereas those with elevated nbBP showed fractional gains or losses. The group with the lowest instability (low-GGI/low-nbBP) was expectedly associated to good prognosis, whereas those with high-nbBP correlated with increased risk of relapse or shortened overall survival in stage 2 and 3 CRCs. Our results suggest that genetic instability could be an interesting tumor specific prognostic variable in CIN colorectal cancer. Along similar lines, Poulogiannis and coauthors [17] defined 4 groups of instability in CRC, with low levels of instability associated with good outcome and high levels with bad prognosis. It is of note that this study was performed on a series comprising both MSI and MSS CRC and it is likely that their group of low instability was largely composed of MSI cases, which are of better prognosis than MSS. Other studies have proposed to relate genetic CNA patterns to outcome [22] or response to chemotherapy [13]. Remarkably, in our study the group presenting an elevated fraction of the genome involved in CNA, but low breakpoint numbers (high-GGI/low-nbBP), was the other group associated with low risk. This group was representative of tumors with large regions of CNA or whole chromosomal arm copy variations and contrasted in terms of prognosis with tumors with high-nbBP which were of bad prognosis. These results were reminiscent of observations by Janoueix-Lerosey et al., [23], who showed that in neuroblastoma tumors with whole chromosome CNA displayed good survival, while those with fractionated CNA presented a high risk of relapse.
These data led us to search for specific copy number changes correlated to the bad outcome groups. To this aim we used two convergent strategies to identify regions of CNA correlated to adverse outcome in our stage 2 and 3 series. Copy number loss at 19q13.3 and/ or 16p13.3 was clearly associated to worsened disease course, as shown by strong correlation with either shortened RFS or OS in stage 2 and 3 tumors. The 16p13.3 region has already been described as a prognostic region in CRC [15]. This region bears a total of 167 known genes among which AXIN1 appears a serious candidate as it had been shown to be mutated in colorectal cancers and wild-type axin 1 can induce apoptosis in colorectal cancer cells [24,25]. The 19q13.3 region comprises 285 genes and among them the BAX pro-apoptotic gene and the polymerase delta gene POLD1. Low expression of BAX protein in stage 3 colorectal cancers has been linked to shorter RFS and 5-FU-based treatment resistance [26]. POLD1 deletion could be involved in impaired DNA replication generating breaks and high rate of mutations [27]. As 16p13.3 (AXIN1) and 19q13.3 (BAX) are frequently co-deleted in our high risk group, apoptotic pathway in these tumors could be severely impacted giving significant resistance to apoptosis and growth advantage in malignant cells of these tumors. It was of note that we could confirm the association with adverse outcome for only 5/87 genes whose gain or loss had been previously shown to be of prognostic significance. Similarly, a large number of prognostic gains or losses could not be confirmed in our dataset [8,[15][16][17][18].