Ann Surg Oncol (2013) 20:S560–S569 DOI 10.1245/s10434-013-2985-7
ORIGINAL ARTICLE – TRANSLATIONAL RESEARCH AND BIOMARKERS
Selection of Patients for Hepatic Surgery of Colorectal Cancer Liver Metastasis Based on Genomic Aberrations Sjoerd C. Bruin, MD1,6, Jorma J. de Ronde, MSc2, Bas Wiering, PhD3, Linde M. Braaf, Bsc1, J. H. W. de Wilt, MD, PhD3, Andrew D. Vincent, PhD4, Marie-Louise F. van Velthuysen, MD, PhD5, T. J. Ruers, MD, PhD6, Lodewyk F. A. Wessels, PhD2,7, and Laura J. van’t Veer, PhD1,5,8 1
Division of Experimental Therapy, The Netherlands Cancer Institute, Antoni van Leeuwenhoek Hospital, Amsterdam, The Netherlands; 2Department of Molecular Carcinogenesis, The Netherlands Cancer Institute, Antoni van Leeuwenhoek Hospital, Amsterdam, The Netherlands; 3Division of Surgical Oncology, Department of Surgery, Radboud University Nijmegen Medical Center, Nijmegen, The Netherlands; 4Department of Biometrics, The Netherlands Cancer Institute, Antoni van Leeuwenhoek Hospital, Amsterdam, The Netherlands; 5Department of Pathology, The Netherlands Cancer Institute, Antoni van Leeuwenhoek, Amsterdam, The Netherlands; 6Department of Surgery, The Netherlands Cancer Institute, Antoni van Leeuwenhoek Hospital, Amsterdam, The Netherlands; 7Delft Bioinformatics Laboratory, Delft University of Technology, Delft, The Netherlands; 8Present address: University of California San Francisco, Helen Diller Family Comprehensive Cancer Center, San Francisco, CA
ABSTRACT Background. We investigated whether genomic aberrations in primary colorectal cancer (CRC) can identify patients who are at increased risk of developing additional hepatic recurrence after colorectal liver metastases (CLM) resection. Methods. Primary tumour DNA from 79 CLM resected patients was analysed for recurrent copy number changes TM (12x135k NimbleGen aCGH). The cohort was divided into three groups: CLM patients with a recurrence-free survival after hepatic resection of at least 5 years (n = 21), patients who developed intra-hepatic recurrence (n = 32), and patients who developed extrahepatic recurrence (n = 26). By contrasting the primary tumour profiles of recurrence free and the extrahepatic recurrence CLM patients, a classifier, the extra-hepatic recurrence classifier (ERC1), predictive for subsequent extrahepatic-recurrence was developed.
S.C. Bruin and J.J. de Ronde contributed equally to this study Ó Society of Surgical Oncology 2013 First Received: 7 June 2012; Published Online: 20 June 2013 L. J. van’t Veer, PhD e-mail:
[email protected]
Results. The ERC1 had an accuracy of 70 % (95 % confidence interval (CI): 55–82 %, misclassification error 30 %, base error rate: 45 %). This analysis identified a region on Chromosome 12p13 as differentially aberrated between these two groups. The classifier was further optimized by contrasting the extrahepatic recurrence group with the combined group of intrahepatic and no recurrence group, resulting in an extrahepatic prognostic classifier (ERC2) able to classify patients with CLMs suitable for hepatic resection with 74 % accuracy (95 % CI: 62–83 %, misclassification error 26 %, base error rate: 32 %). Conclusions. Patients with CLM who will develop extrahepatic recurrence may be identified with ERCs based on information in the primary tumour. Risk estimates for the occurrence of extrahepatic metastases may allow a reduction of hepatic resections of colorectal liver metastases for those who are unlikely to develop extrahepatic metastases.
Colorectal cancer (CRC) is the second leading cause of cancer death in the Western world. The WHO estimates that 945,000 new cases occur yearly, with 492,000 deaths.1 The overall 5-year survival is 57 %, and up to 50 % of all patients will develop metastases. Distant metastases are responsible for the majority of CRC deaths, primarily due to colorectal liver metastases (CLM).2–4 Roughly 50 % of patients with stage III and 20 % of patients with stage II CRC will develop CLM.5 Of all patients who die of
Genomic Aberrations in Liver Metastases
advanced CRC, 60–70 % will have developed CLM.6 Even with the use of targeted drugs, the overall survival of patients with nonresectable CLM is only 2 years, and late detection of CLM often is fatal. In 15–25 % of patients with CLM, hepatic resection is possible and the only potentially curative treatment option.7 In these patients a 5-year survival of *58 % can be achieved and up to 20 % of this population will still be alive after 10 years.8 This survival benefit, however, has to be balanced against procedure-related morbidity and mortality rates of 15–35 and 1–4 %, respectively.9 Eligibility for hepatic surgery requires all metastases to be resectable while maintaining an adequate liver reserve. Furthermore, there should be no extensive extrahepatic disease, with the possible exception of a few resectable lung metastases.10 In patients with initially irresectable CLM, eligibility for hepatic surgery can occur after a multimodal approach, involving the surgeon, oncologist, radiotherapist, and radiologist. Unfortunately, within 5 years, two-thirds of patients will develop a recurrence despite optimal liver resection.11,12 In a well-selected group of patients with an intrahepatic recurrence (10–20 %), repeat resection is still an option.13–15 Several clinical pathologic models that predict outcome for individual patients with CLM have been investigated.16 For example, Fong et al.17 combined five clinical criteria in a clinical risk score (CRS) to classify patients who may benefit most from liver resection. Several years later they presented a model for predicting disease-specific survival after liver resection for CLM with a concordance index of 0.61. This model predicted disease-specific survival more accurately in the individual patient than CRS score.17,18 Besides these clinical pathologic models, molecular prognostic markers in patients with CLM show promising results.7 In 1990, Fearon and Vogelstein first described the multistep model of colorectal carcinogenesis.19,20 In this model, multiple mutations are considered necessary for CRC to develop. Two pathways leading to DNA alterations underlying the progression towards invasive CRC have been well described; the microsatellite instability (MSI) and the chromosomal instability (CIN) pathway. The majority of CRC (85 %) is CIN, characterized as an imbalance in chromosomal number, subchromosomal genomic amplifications and loss of heterozygosity. During the past decade, array comparative genomic hybridization (aCGH) has been employed to detect these chromosomal aberrations and to provide an overview of the extent of genomic instability throughout the whole genome. So far, aCGH analyses have resulted in several specific chromosomal loci that appear to play an important role in CRC progression.21–23 Recently, Mekenkamp et al.24 showed that the protein-encoding gene C20orf3 mapping at 20p11 is associated with hepatic-specific metastasis in CRC
S561
patients. However, none of these studies are evaluated for extra hepatic recurrence after resection of CLM. In this study, we investigated genomic instability in primary CRC and CLM to determine chromosomal loci specific for extrahepatic recurrence after hepatic resection of CLM. Furthermore, we evaluated the chromosomal aberrations in the context of progression from primary CRC to CLM. The understanding of both the molecular and cellular mechanisms underlying CRC progression and the prognosis after hepatic surgery is important when selecting patients for major surgery and to perform targeted follow-up. MATERIALS AND METHODS Patients and Tumour Samples Clinical characteristics were retrieved from digital medical registries and an existing prospective study database of patients operated for CLM between 1993 and 2005 at the Radboud University Nijmegen Medical Center. Pathological characteristics were scored by a single pathologist (MLFvV) on formalin-fixed, paraffin-embedded (FFPE) sections. All patients were selected for hepatic surgery with a standardized diagnostic protocol.25 The clinical and pathological data from the CLM patients are described in Table 1. The FFPE tumour blocks from primary and CLM were collected from 79 patients. Patients were divided into three groups: CLM patients with a recurrence-free survival after hepatic resection of at least 5 years (group I, n = 21), patients who developed intrahepatic recurrence (group II, n = 32), and patients who developed extrahepatic recurrence (group III, n = 26; Fig. 1a). A primary tumour and CLM from the same patient is defined as a tumour pair. In group I we analyzed 19 primary and 9 CLM samples (5 paired tumours), in group II 29 primary and 28 CLM samples (24 paired tumours), and in Group III 22 primary and 25 CLM samples (21 paired tumours). For the KC-SMART and PAM statistical analysis, only the primary tumour samples were analyzed (n = 19, 29, and 22 for groups I, II, and III, respectively; Table 1). Of the 79 patients included in the study, 42 patients were male and 37 were female with a median age of 59 years. The majority of patients had a primary rectosigmoid tumour (68 %). Nine percent (n = 7) of the patients received 1 week of neoadjuvant radiotherapy for their primary rectum tumour. Forty-two percent of the patients received adjuvant chemotherapy after resection of their primary tumour. The mean carcinoembryonic antigen (CEA) before treatment of the CLM was 7.5 (ng/ml) with a range between 1 and 220 (ng/ml). The CLM patients underwent one of the following three treatments: resection alone
S562
S. C. Bruin et al.
TABLE 1 Clinical, pathological, and tumour characteristics for each of the studied groups Group definition
No recurrence I (n = 21)
Recurrence liver II (n = 32)
Extrahepatic recurrence III (n = 26)
Total N = 79
p value
Primary
19
29
22
70
CLM
9
28
25
62
Pairs
5
24
21
50
63 (37–71)
58 (32–75)
60 (43–77)
59 (32–77)
0.76b
Female
10 (48 %)
13 (41 %)
14 (54 %)
37 (47 %)
0.59b
Male
11 (52 %)
19 (59 %)
12 (46 %)
42 (53 %)
Cecum
4 (19 %)
4 (12 %)
3 (12 %)
11 (14 %)
Col ascending
3 (14 %)
2 (6 %)
2 (8 %)
7 (9 %)
Col transcending
0 (0 %)
2 (6 %)
0 (0 %)
2 (3 %)
Col descending
2 (10 %)
2 (6 %)
0 (0 %)
4 (5 %)
Sigmo
7 (33 %)
12 (38 %)
12 (46 %)
31 (39 %)
Rectum
5 (24 %)
9 (28 %)
9 (35 %)
23 (29 %)
Dubbel tumor
0 (0 %)
1 (3 %)
1 2
1 (5 %) 2 (10 %)
0 (0 %) 5 (16 %)
1 (4 %) 3 (12 %)
2 (3 %) 10 (13 %)
3
15 (71 %)
26 (81 %)
18 (69 %)
59 (75 %)
4
2 (10 %)
1 (3 %)
4 (15 %)
7 (9 %)
Missing
1 (5 %)
0 (0 %)
0 (0 %)
1 (1 %)
0
8 (38 %)
12 (38 %)
12 (46 %)
32 (41 %)
1
12 (57 %)
18 (56 %)
12 (46 %)
42 (53 %)
2
0 (0 %)
2 (6 %)
2 (8 %)
4 (5 %)
Missing
1 (5 %)
0 (0 %)
0 (0 %)
1 (1 %)
0
3 (14 %)
7 (22 %)
3 (12 %)
13 (16 %)
1
9 (43 %)
11 (34 %)
8 (31 %)
28 (35 %)
X
9 (43 %)
14 (44 %)
15 (58 %)
38 (48 %)
Poor
0 (0 %)
2 (6 %)
2 (8 %)
4 (5%)
Moderately/poor Moderately
1 (5%) 17 (81%)
4 (12 %) 21 (66 %)
1 (4 %) 21 (81 %)
6 (8 %) 59 (75 %)
Analyzed tumors
Patient demographics Age Median (Range) Gender
Primary tumor characteristics Location primary 0.95b
pT 0.65a
pN 0.86a
pM 0.7b
Diff primary tumor
Moderately/well
1 (5%)
2 (6 %)
1 (4 %)
4 (5 %)
Well
2 (10 %)
3 (9 %)
0 (0 %)
5 (6%)
Missing
0 (0 %)
0 (0 %)
1 (4 %)
1 (1%)
11 (52 %)
13 (41 %)
8 (31 %)
32 (41 %)
2 (10%)
0.53b
Treatment history Adjuvant treat primary tumor Chemo Radiotherapy
4 (12 %)
1 (4 %)
7 (9 %)
Both
0 (0 %)
0 (0 %)
1 (4 %)
1 (1 %)
None
7 (33 %)
15 (47 %)
13 (50 %)
35 (44 %)
0.81b
Genomic Aberrations in Liver Metastases
S563
TABLE 1 continued Group definition
No recurrence I (n = 21)
Recurrence liver II (n = 32)
Extrahepatic recurrence III (n = 26)
Total N = 79
Missing
1 (5 %)
0 (0 %)
3 (12 %)
4 (5 %)
Median
3.1
9.6
12
7.5
(Range)
(1–74)
(1–220)
(1–220)
(1–220)
20 (62 %)
16 (62 %)
53 (67 %)
p value
0.01a
Treatment CLM Only resection
17 (81%)
0.04b
Ablation therapy
2 (10%)
2 (6 %)
0 (0 %)
4 (5 %)
Ablation and resection
2 (10%)
10 (31 %)
10 (38 %)
22 (28 %)
1 (1–7)
2 (1–7)
2 (1–10)
2 (1–10)
0.35a
Median
40
45
46
45
0.17a
(Range)
(8–100)
(10–150)
(9–90)
(8–150)
Median
70
70
60
(Range)
(30–90)
(55–80)
(50–90)
Median
70
70
70
(Range)
(60–90)
(50–95)
(50–95)
CLM characteristics Median (Range) Size lesion (mm)
Tumor percentage primary
Tumor percentage CLM
Group I: patients with a recurrence free survival after hepatic resection of at least 5 years Group II: patients who developed an intrahepatic recurrence within 5 years of surgery Group III: patients who developed an extrahepatic recurrence within 5 years of surgery a
Kruskal–Wallis test
b
Fisher exact test
(67 %), ablation (5 %), or a combination of resection and ablation (28 %). The mean number of lesions in the liver was two (range 1–10). The median size of the lesions was 45 mm (range 8–150). Tissue material was handled following standard operational procedures of the Radboud University Nijmegen Medical Center and respective primary treatment hospitals, as approved by the institutional review board. Tissue handling was anonymous following the Helsinki declaration. DNA Extraction DNA isolation was performed as described earlier.26 Briefly, genomic DNA was isolated by proteinaseK digestion after deparaffinization from 109 10-lm FFPE tissue sections containing at least 60 % tumour cells. DNA quality was qualified by measuring the maximum possible length of DNA to be amplified by a multiplex PCR as described elsewhere.27
Tumour DNA labelling was performed with ULS-Cy3 and ULS-Cy5 conjugates from the Enzo Agilent aCGH Labeling kit (http://www.enzolifesciences.com). Hybridisations on the arrays were performed at the NKI Central Microarray Facility using the NimbleGen Hybridization kit. Arrays were washed and subsequently scanned with an Agilent microarray scanner. Data processing of the scanned microarray slide included signal intensity measurement with the NimbleScan 2.5 software (Roche NimbleGen, Inc.) program. To prevent slide batch spotting bias, samples were hybridized in random order. Genomic DNA of 132 tumour samples (70 primary tumours and 62 CLM) was hybridized to 135k NimbleTM Gen arrays to obtain, for each probe on the array, a log2 ratio of the fluorescence intensity ratio of the sample versus a reference of pooled healthy male DNA was established. Statistical Analysis
Comparative Genomic Hybridization DNA copy number changes were investigated using the TM 12x135k NimbleGen (http://www.nimblegen.com) array comparative genomic hybridization (aCGH) platform.
Normalization and copy number analysis were performed using the process CGH-segMNT of the NimbleScan 2.5 software with default settings. Quality control measures consisted of checking the consistency of
S564
S. C. Bruin et al.
a
5-year recurrence free N = 21 (Group I)
Primary CRC N = 79
CLM N = 79
Intra hepatic recurrence N = 32 (Group II) Extra hepatic recurrence N = 26 (Group III)
b Group I Primary tumor
CLM
5-year recurrence free
DNA isolation and aCGH profiling
Classifier construction
Extra-hepatic recurrence classifier 1 (ERC1)
b FIG. 1 a Study groups. Formalin-fixed, paraffin-embedded (FFPE)
material from primary CRC and CLM were investigated. Patients who underwent surgery for CLM were divided into three different groups: patients with a recurrence-free survival of at least 5 years (group I, n = 21), patients who developed an intrahepatic recurrence within 5 years (group II, n = 32), and patients who developed an extrahepatic recurrence within 5 years (group III, n = 26). b and c Classifier construction. b aCGH profiles derived from DNA from primary tumours of patients with a recurrence-free survival of at least 5 years (group I) and patients who developed an extrahepatic recurrence (group III) within 5 years were employed to develop a classifier to separate these groups. The resulting classifier is denoted as ERC1. c aCGH profiles derived from DNA from primary tumours of patients with a recurrence-free survival of at least 5 years and patients who developed an intrahepatic recurrence (groups I and II, i.e., patients who are eligible for CLM resection) versus patients who developed an extrahepatic recurrence within 5 years. Group III were employed to develop a classifier to separate these groups. The resulting classifier is denoted as ERC2. Primary CRC primary colorectal cancer, resection CLM resection of colorectal liver metastases, ERC extrahepatic recurrence classifier
FGD distribution (%; 0.225 cut-off) 50
Mann-Whitney P-value = 0.026 Primary tumor
CLM
Extra hepatic recurrence
40
Group III 30
c Group I and II Primary tumor
CLM
5-year recurrence-free or intra-hepatice recurrence
20
10
DNA isolation and aCGH profiling
Classifier construction
Primary tumor
CLM
Extra-hepatic recurrence classifier 2 (ERC2)
Extra hepatic recurrence
Group III
signal distributions across samples, unsupervised clustering to check for outlier samples, sex matching by cross-referencing the X and Y chromosome readings with the clinical database, and evaluation of overall signal consistency between primary tumour and paired metastasis. Next, the data were smoothed using a running mean (window size of 10 probes) and a running median (window size of 11 probes) procedure. The running mean smoothing gave the best clustering results (i.e., primary tumour and metastasis pairs clustered together the most frequently in an unsupervised hierarchical clustering approach (data not shown)), and therefore the hierarchical clustering and the
Primary tumors
Liver metastases
FIG. 2 Fraction of genome altered between the primary colorectal tumours and the CLMs. The boxplots present the distribution of the fraction of genome altered (FGA) for all samples in the primary tumour material on the left and all CLM tumour material on the right. A cutoff of 0.225 was used to determine aberration status from the log-ratio aCGH data. Y-axis: percentage genome altered. Mann– Whitney–Wilcox, p = 0.026
PAM analyses were performed using the running mean smoothed data. As KC-SMART and DNAcopy segment algorithm already include a form of smoothing the raw data was used for these analyses. KC-SMART is a recently developed algorithm for identifying recurrent copy number changes. This technique was used to identify specific locations of genomic gains and losses recurring at significant levels within the primary tumours of the various groups, using the KC-SMART R package.28 We ran both the recurrent and the comparative analyses using a kernel width of 1 Mb, because this eliminates most probe induced noise while still retaining adequate resolution. Similarities
Genomic Aberrations in Liver Metastases FIG. 3 KC-SMART analysis of primary colorectal tumour material versus CLMs. KCSMART profiles of the primary tumours (black) and the CLMs (red). Although no significant differences were reported by the KC-SMART algorithm, some regions visually appear distinct, for example the region on chromosome 8. The x-axis represents the base pair position on the genome, and the y-axis represents the average smoothed log-ratio of the aCGH data
S565
KCSMART smoothed coverage 1.0
0.5
0.0
−0.5
Primary tumors CLMs 0
500
1000
1500
2000
2500
3000
Genomic position (in mb)
between KC-SMART profiles were calculated using the Pearson correlation. We computed the ‘‘fraction of the genome altered’’ (FGA) as described in Mehta et al.29 using a cutoff of 0.225 on a log2 ratio scale for amplified and deleted regions and a cutoff of 0.75 for highly aberrated regions. The data were segmented using the DNAcopy package’s ‘‘segment’’ method, an implementation of the Circular Binary Segmentation approach.30 The Mann–Whitney–Wilcox test was used to determine differences in FGA and the number of breaks between different groups. For this analysis, the sex chromosomes (X and Y) were removed because these could potentially bias the results, due to the difference in the malefemale ratio between groups. Hierarchical clusterings were run using the hclust package, using complete linkage and (1—Pearson correlation coefficient) as the distance measure. The PAM package (Prediction Analysis for Microarrays: PAM (http://www-stat.stanford.edu/*tibs/ PAM) was employed to derive classifiers on aCGH profiles derived from DNA extracted from primary tumour tissue. Leaveone-out cross-validation was applied to estimate the accuracies of the classifiers on new samples. Note that if one group only contains 10 % of the samples, then 90 % accuracy can be obtained by simply classifying all samples as belonging to the larger group. To account for potential underestimation of error rates, we also report the base error rate as a reference. The base error rate is 1—(the fraction of samples in the largest class), i.e., 0.1 base error rate in the above example. Confidence intervals were determined using the Wilson procedure for estimation of 95 % confidence intervals for proportions. The following PAM analyses on DNA from primary tumour tissue of patients was performed. Initially, a PAM classifier was constructed to distinguish between the two most extreme outcome groups: patients who had surgery
for CLM with a recurrence-free survival of at least 5 years (group I) versus CLM patients who developed an extrahepatic recurrence (group III; Fig. 1b). Second, a PAM classifier was developed that was predictive for the suitability of CLM resection in terms of whether a CLM will remain confined to the liver. We analyzed patients with a recurrence-free survival of at least 5 years together with patients who developed an intrahepatic recurrence (groups I and II) versus patients who developed an extrahepatic recurrence (group III; Fig. 1c). All analyses, except the normalisation with Nimblescan, were performed in the statistical scripting language R. We set the a-level of significance at 0.05 for all statistical analyses. RESULTS Patient Groups Seventy-nine patients who had undergone surgery for CLM were divided into three different groups: CLM patients with a recurrence-free survival of at least 5 years (group I, n = 21); patients who had developed within 5 years an intrahepatic recurrence (group II, n = 32); and patients who had developed within 5 years an extrahepatic recurrence (group III, n = 26; Fig. 1a). Formalin-fixed, paraffin-embedded (FFPE) material from 70 primary colorectal tumours and 62 liver metastases, including 50 paired samples, were investigated. Extent of Chromosomal Instability in Primary Tumours and CLM There was no significant pairwise differences (Mann– Whitney–Wilcox) in FGA between the primary tumour
S566
S. C. Bruin et al. KCSMART smoothed coverage
Primary tumors, Group I vs. III
1.0
0.5
0.0
−0.5 0
500
1000
1500
2000
2500
3000
2500
3000
Genomic position (in mb)
Primary tumors, Group I vs. III
∆ Values 2.5 2.0 1.5 1.0 0.5 0.0 0
500
1000
1500
2000
Genomic position (in mb)
FIG. 4 KC-SMART analysis for the primary tumour material of group I versus group III. The first panel presents the KC-SMART profiles of the primary tumours, grouped by their outcome: group I in black and group III in red. The second panel presents the actual differences between the two graphs, with red denoting higher signal (of group III relative to group I), and green denoting lower signal
(again, group III relative to group I). Although a difference can be observed on chromosome 12, none of the identified differences in signal reached statistical significance in the KC-SMART algorithm. The x-axis represents the base pair position on the genome, and the yaxis represents the average smoothed log-ratio of the aCGH data
samples of groups I (n = 19), II (n = 29), and III (n = 22) and no significant pairwise difference (Mann–Whitney– Wilcox) in FGA amongst the CLM samples of group I (n = 9), II (n = 28), and III (n = 25) (data not shown). To analyze the extent of CIN between primary tumours and the CLM, the FGA of the comparative hybridization between primary and CLM was compared. The CLMs exhibited a significantly higher level of FGA compared with that of the primary tumours (p = 0.03; Fig. 2). However, subsequent KC-SMART comparative analysis of the entire group of primary tumours versus all CLMs showed no overall specific significant differences in chromosomal aberrations between the primary tumours and the CLM. Likewise, the Pearson’s correlation coefficient between the KC-SMART profiles was very high (0.95; Fig. 3). Distinct features were seen in the CLM, for example on chromosome 8, but this was not significant in the comparative KC-SMART analysis. Hierarchical clustering of the data indicated that the majority of primary
colorectal tumours clustered with their paired CLM (data not shown). Despite differences in the tumour percentages between the primary tumour and the paired liver metastasis samples, we found an average Pearson correlation coefficient of 0.73 (median r = 0.75). By comparison, the average correlation of a primary tumour with all nonpaired metastasis samples was 0.43 (median r = 0.47). Primary Colorectal Classifier for CLM and Sequential Extrahepatic Recurrence To identify chromosomal aberrations specific for developing extrahepatic recurrence (ER) after CLM, a comparative KC-SMART analysis was performed on the primary tumour material from group I versus group III (5year DFS vs ER after resection of CLM; Fig. 1b). These extremes in outcome were initially taken, excluding the intermediate prognosis group II tumours, to maximize the signal to noise ratio. The KC-SMART algorithm identified
Genomic Aberrations in Liver Metastases
S567
a region on chromosome 12 to be most differentially altered between the two groups (Fig. 4). Although this was the highest ranking difference in terms of the comparative KCsmart statistic, it did not reach statistical significance (uncorrected p \ 0.0001; q-value = 0.82; Fig. 4). Samples in group III were amplified in this region, relative to group I. Based on information in the primary tumour, an ER classifier (ERC1) was able to identify patients at high risk for the development of ER after CLM resection with 70 % LOOCV accuracy (95 % CI: 55–82 %, misclassification error 30 %, base error rate: 45 %). The ERC1 for the primary tumours of group I versus group III included 18 probes, all of which were located in a single region on chromosome 12p13, harbouring nine genes. The KCSMART comparative plot illustrates the difference between the two groups in this region clearly (Fig. 5). The intermediate group (II) is a group of patients who developed an intrahepatic recurrence. Because these patients had not developed an extrahepatic recurrence, they may benefit from CLM surgery, just like the patients who are recurrence-free (i.e., group 1). To increase specificity in identifying patients who are at increased risk of developing extrahepatic recurrence, and all patients who may benefit from CLM surgery, a KC-SMART comparative analysis was performed comparing primary tumour material from group I and II combined versus group III. As before, this analysis indicated the region on chromosome 12p13 to be highly differentially aberrated. The corresponding classifier Group I Group III Group I SD Group III SD
KCSmart smoothed average 0.6
0.4
0.2
0.0
−0.2
0
25
50
75
100
125
Chromosomal position
FIG. 5 The KC-SMART comparative plot of chromosome 12 between groups I and III. A close up of the KC-SMART profiles of chromosome 12. The 12p13 region on the left presents the difference between groups I and III. The x-axis represents the base pair position on the genome, and the y-axis represents the average smoothed logratio of the aCGH data
(ERC2), which separates patients of groups I and II from group III (Fig. 1c) had an accuracy of 74 % (95 % CI: 62– 83%, misclassification error 26 %, base error rate: 32 %). The majority of the probes in the ERC2 were comparable to the probes from the ERC1, most notably those located on chromosome 12p13. DISCUSSION One of the challenges in CRC therapy lies in the early detection of CLM. The ability to identify patients who are at high risk for the development of extrahepatic metastases after surgery for CLM is the first step in this process. A classifier based on specific genomic aberrations may provide an important link in the process of selecting patients eligible for liver surgery or more tailored therapies. Moreover, elucidation of the molecular and cellular mechanisms related to the development of CLM is required for the development of future diagnostics and therapeutic interventions. In the present study, we examined chromosomal aberrations in primary CRC material to identify molecular markers predictive for the development of ER after resection of CLM. Our results indicate that CLMs have a significantly higher level of FGA compared with that of primary tumours (p = 0.03; Fig. 2). Mehta et al.29 reported that patients with elevated FGA had better survival after CLM resection. We were unable to confirm this finding in our study, because there were no apparent differences in FGA among the primary tumour samples from groups I, II, and III, nor among the CLM samples of these three groups. We believe that this is the first study to report a classifier based on genomic aberrations in primary colorectal tumours, specific for the development of ER after hepatic surgery for CLM. In our sample, the ERC1 was able to identify patients who would develop an ER within 5 years after hepatic surgery with an accuracy of 70 %. It has been shown that in patients who develop a CLM confined to the liver, repeated liver resection results in prolonged survival.13–15 That implies that not only patients in group I may benefit from surgery but also patients in group II. Our second classifier, the ERC2, identifies these patients with an accuracy of 74 %. In short, patients with poor clinical and pathological characteristics who are classified for low risk of ER are likely to benefit most from CLM, whereas those who are at a higher risk of the development of ER would probably benefit more from early and aggressive chemo- and targeted therapy and may potentially be withheld from major surgery. Further research is required to examine how genomic information can be combined with clinical and pathological characteristics to better improve the surgical selection of patients with CLM.
S568
S. C. Bruin et al.
Both classifiers were significantly enriched for probes located on chromosome 12p13, where amplification of this chromosomal region is associated with higher likelihood of extrahepatic recurrences after CLM. Genes located in this region include TSPAN9, FKBP4, NRIP2, and TULP3, all of which play a role in the regulation of cell development, activation, immunoregulation, transcription, signalling, growth, and motility. Another interesting gene located in this region is Forkhead box protein M1 (FOXM1), which is known to be related to tumour aggressiveness and tumourigenicity in CRC. Recently Uddin et al. found clinical evidence for a strong correlation among FoxM1-expressing CRC tumours and the upregulation of matrix metalloproteinase (MMP-2 and MMP-9).31 MMPs are crucial in the process of tumour invasion and metastasis and are directly linked to angiogenesis and metastasis development.32 They showed that down-regulation of FoxM1 with thiostrepton treatment caused inhibition of cell proliferation and invasion and migration of CRC cells. Furthermore, they showed that an altered FoxM1 expression significantly affected VEGF expression in CRC. This finding in combination with our results suggests that FOXM1 could be used in patients with CLM as a potential biomarker and novel therapeutic target. The primary limitation of this study is the absence of an external validation cohort. External validation of the ERC classifiers is essential for confirmation of accuracies and thereby ascertaining what role these tools may play in the selection of patients for CLM surgery. However, if confirmed, one might imagine patients classified for high risk of ER being be selected for novel therapeutic targeted therapy, and thereby limiting major liver surgery to patients who are most likely to benefit. ACKNOWLEDGMENT isolation.
I. Mikolajewska-Hanclich for the DNA
CONFLICTS OF INTEREST
None.
REFERENCES 1. Weitz J, Koch M, Debus J, Hohler T, Galle PR, Buchler MW. Colorectal cancer. Lancet. 2005;365(9454):153–65. 2. Welch JP, Donaldson GA. The clinical correlation of an autopsy study of recurrent colorectal cancer. Ann Surg. 1979;189(4):496– 502. 3. Sadahiro S, Suzuki T, Ishikawa K, et al. Recurrence patterns after curative resection of colorectal cancer in patients followed for a minimum of 10 years. Hepatogastroenterology. 2003;50(53): 1362–6. 4. Galandiuk S, Wieand HS, Moertel CG, et al. Patterns of recurrence after curative resection of carcinoma of the colon and rectum. Surg Gynecol Obstet. 1992;174(1):27–32. 5. Penna C, Nordlinger B. Colorectal metastasis (liver and lung). Surg Clin North Am. 2002;82(5):1075–90.
6. Weiss L, Grundmann E, Torhorst J, et al. Haematogenous metastatic patterns in colonic carcinoma: an analysis of 1,541 necropsies. J Pathol. 1986;150(3):195–203. 7. Neal CP, Garcea G, Doucas H, et al. Molecular prognostic markers in resectable colorectal liver metastases: a systematic review. Eur J Cancer. 2006;42(12):1728–43. 8. Gomez D, Sangha VK, Morris-Stiff G, et al. Outcomes of intensive surveillance after resection of hepatic colorectal metastases. Br J Surg. 2010;97(10):1552–60. 9. Farid SG, Aldouri A, Morris-Stiff G, et al. Correlation between postoperative infective complications and long-term outcomes after hepatic resection for colorectal liver metastasis. Ann Surg. 2010;251(1):91–100. 10. Rees M, Tekkis PP, Welsh FK, O’Rourke T, John TG. Evaluation of long-term survival after hepatic resection for metastatic colorectal cancer: a multifactorial model of 929 patients. Ann Surg. 2008;247(1):125–35. 11. Hughes KS, Simon R, Songhorabodi S, et al. Resection of the liver for colorectal carcinoma metastases: a multi-institutional study of patterns of recurrence. Surgery. 1986;100(2):278–84. 12. Nordlinger B, Quilichini MA, Parc R, Hannoun L, Delva E, Huguet C. Hepatic resection for colorectal liver metastases. Influence on survival of preoperative factors and surgery for recurrences in 80 patients. Ann Surg. 1987;205(3):256–63. 13. Lopez P, Marzano E, Piardi T, Pessaux P. Repeat hepatectomy for liver metastases from colorectal primary cancer: a review of the literature. J Visc Surg. 2012;149:97–103. 14. Petrowsky H, Gonen M, Jarnagin W, et al. Second liver resections are safe and effective treatment for recurrent hepatic metastases from colorectal cancer: a bi-institutional analysis. Ann Surg. 2002;235(6):863–71. 15. Yan TD, Sim J, Black D, Niu R, Morris DL. Systematic review on safety and efficacy of repeat hepatectomy for recurrent liver metastases from colorectal carcinoma. Ann Surg Oncol. 2007;14(7):2069-77. 16. Welsh FK, Tekkis PP, John TG, Rees M. Predictive models in colorectal liver metastases–can we personalize treatment and outcome? Dig Surg. 2008;25(6):406–12. 17. Fong Y, Fortner J, Sun RL, Brennan MF, Blumgart LH. Clinical score for predicting recurrence after hepatic resection for metastatic colorectal cancer: analysis of 1,001 consecutive cases. Ann Surg. 1999;230(3):309–18. 18. Kattan MW, Gonen M, Jarnagin WR, et al. A nomogram for predicting disease-specific survival after hepatic resection for metastatic colorectal cancer. Ann Surg. 2008;247(2):282–7. 19. Fearon ER, Vogelstein B. A genetic model for colorectal tumorigenesis. Cell. 1990;61(5):759–67. 20. Jass JR. Classification of colorectal cancer based on correlation of clinical, morphological and molecular features. Histopathology. 2007;50(1):113–30. 21. Ried T, Knutzen R, Steinbeck R, et al. Comparative genomic hybridization reveals a specific pattern of chromosomal gains and losses during the genesis of colorectal tumors. Genes Chromosomes Cancer. 1996;15(4):234–45. 22. Meijer GA, Hermsen MA, Baak JP, et al. Progression from colorectal adenoma to carcinoma is associated with non-random chromosomal gains as detected by comparative genomic hybridisation. J Clin Pathol. 1998;51(12):901–9. 23. Hermsen M, Postma C, Baak J, et al. Colorectal adenoma to carcinoma progression follows multiple pathways of chromosomal instability. Gastroenterology. 2002;123(4):1109–19. 24. Mekenkamp LJ, Haan JC, Koopman M, et al. Chromosome 20p11 gains are associated with liver-specific metastasis in patients with colorectal cancer. Gut. 2013;62(1):94–101. 25. Wiering B, Ruers TJ, Krabbe PF, Dekker HM, Oyen WJ. Comparison of multiphase CT, FDG-PET and intra-operative
Genomic Aberrations in Liver Metastases
26.
27.
28.
29.
ultrasound in patients with colorectal liver metastases selected for surgery. Ann Surg Oncol. 2007;14(2):818–26. Joosse SA, van Beers EH, Nederlof PM. Automated array-CGH optimized for archival formalin-fixed, paraffin-embedded tumor material. BMC Cancer. 2007;7:43. van Beers EH, Joosse SA, Ligtenberg MJ, et al. A multiplex PCR predictor for aCGH success of FFPE samples. Br J Cancer. 2006;94(2):333–7. de Ronde JJ, Klijn C, Velds A, et al. KC-SMARTR: An R package for detection of statistically significant aberrations in multi-experiment aCGH data. BMC Res Notes. 2010;3:298. Mehta KR, Nakao K, Zuraek MB, et al. Fractional genomic alteration detected by array-based comparative genomic
S569 hybridization independently predicts survival after hepatic resection for metastatic colorectal cancer. Clin Cancer Res. 2005;11(5):1791–7. 30. Venkatraman ES, Olshen AB. A faster circular binary segmentation algorithm for the analysis of array CGH data. Bioinformatics. 2007;23(6):657–63. 31. Uddin S, Ahmed M, Hussain A, et al. Genome-wide expression analysis of Middle Eastern colorectal cancer reveals FOXM1 as a novel target for cancer therapy. Am J Pathol. 2011;178(2):537– 47. 32. Curran S, Murray GI. Matrix metalloproteinases: molecular aspects of their roles in tumour invasion and metastasis. Eur J Cancer. 2000;36(13 Spec No):1621–30.
