Article Text
PDF
Abstract
Background and aims Liver metastases are the leading cause of death in colorectal cancer. To gain better insight into the biology of metastasis and possibly identify new therapeutic targets we systematically investigated liver-metastasis-specific molecular aberrations.
Methods Primary colorectal cancer (pCRC) and matched liver metastases (LMs) from the same patients were analysed by microarray-based comparative genomic hybridisation in 21 pairs and gene expression profiling in 18 pairs. Publicly available databases were used to confirm findings in independent datasets.
Results Chromosome aberration patterns and expression profiles of pCRC and matched LMs were strikingly similar. Unsupervised cluster analysis of genomic data showed that 20/21 pairs were more similar to each other than to any other analysed tumour. A median of only 11 aberrations per patient was found to be different between pCRC and LM, and expression of only 16 genes was overall changed upon metastasis. One region on chromosome band 11p15.5 showed a characteristic gain in LMs in 6/21 patients. This gain could be confirmed in an independent dataset of LMs (n=50). Localised within this region, the growth factor IGF2 (p=0.003) and the intestinal stem cell specific transcription factor ASCL2 (p=0.029) were found to be over-expressed in affected LM. Several ASCL2 target genes were upregulated in this subgroup of LM, including the intestinal stem cell marker OLFM4 (p=0.013). The correlation between ASCL2 expression and four known direct transcriptional targets (LGR5, EPHB3, ETS2 and SOX9) could be confirmed in an independent expression dataset (n=50).
Conclusions With unprecedented resolution a striking conservation of genomic alterations was demonstrated in liver metastases, suggesting that metastasis typically occurs after the pCRC has fully matured. In addition, we characterised a subset of liver metastases with an ASCL2-related stem-cell signature likely to affect metastatic behaviour of tumour cells.
- Colorectal cancer
- liver metastasis
- ASCL2
- array-CGH
- expression profiling
Statistics from Altmetric.com
Significance of this study
What is already known about this subject?
Significant DNA-copy number and gene expression differences have been detected between unmatched colorectal cancer primary tumours and liver metastases.
Analyses of small numbers of matched primary tumour/metastasis pairs showed mixed results; while low-resolution genomic analysis did not detect differences, gene expression analyses are divergent in their conclusions.
What are the new findings?
Both chromosome aberration patterns and gene expression profiles are strikingly similar in primary colorectal carcinomas and matched liver metastases indicating that effective dissemination of metastatic cells occurs late in the molecular development of these tumours.
A subset of liver metastases is characterised by a DNA-copy number gain of chromosome band 11p15.5.
ASCL2 and IGF2 are possible driver genes for this gain.
An ASCL2 related stem cell signature was found to be upregulated in liver metastases with 11p15.5 gain.
How might it impact on clinical practice in the foreseeable future?
IGF2 and ASCL2 over-expression has the potential to shift the hierarchy of stem and progenitor cells within liver metastases resulting in more self renewal rather than differentiation and potentially affecting the clinical behaviour of these tumours. This is of special interest in the light of current clinical trials targeting the IGF receptor 1, of which IGF2 is a ligand.
Introduction
Colorectal cancer (CRC) is one of the most common malignancies in the Western world both in terms of incidence and mortality with an estimated lifetime risk of 5–6%.1 2 Distant metastasis occurs mainly in the liver and is the major cause of death of CRC patients. Depending on the stage of the primary tumour, occurrence of liver metastasis ranges from 20% (UICC Stage II) to 70% (UICC Stage IV).3 Recent therapeutic advances have already prolonged overall survival of patients with unresectable liver metastases4 5 but a better understanding of the molecular biology of metastases itself will be an essential prerequisite to developing targeted therapies specifically for patients suffering from metastatic disease.
In order to identify molecular alterations specific for liver metastases (LMs) previous studies used gene expression profiling or comparative genomic hybridisation (CGH) to identify characteristic differences between LMs and primary colorectal cancers (pCRCs), mostly using unmatched sample groups derived from different patients (reviewed by Nadal et al,6 Diep et al7 and Cardoso et al8). In such an unmatched-pair approach, the considerable inter-tumour heterogeneity inherent to colorectal cancer greatly limits the information value of these comparisons. This conceptual problem, however, can be overcome by using matched pCRCs and LMs from the same patient. In a recent meta-analysis of CGH data,7 comparing 260 pCRCs to 81 LM samples, the authors described significant differences between pCRCs and LMs and proposed a progression model of chromosome alterations for every step of the adenoma–carcinoma–metastasis sequence. When focusing exclusively on the included 25 matched pairs, these characteristic alterations could not be confirmed. A second meta-analysis including matched-pair data (n=43) found an increase of copy number aberrations in matched LMs similar to that of unmatched comparisons, but none proved to be statistically significant.9 Thus, chromosomal CGH studies could until now not pinpoint significant differences, possibly due to their low overall resolution. Studies analysing gene expression of matched pCRCs and LMs present a similar picture. Four of them, each including 12 or fewer pairs, revealed no or only slight differences.10–13 A fifth study by Ki et al compared 25 matched pairs and defined a 46-gene classifier that predicted a tumour as either primary or metastasis with an accuracy of 83%.14 Nevertheless, it is not clear whether expression of any of the genes analysed in this study was significantly different between pCRCs and matched LMs since no statistical analysis was performed.
To provide a more rigorous comparative analysis of pCRCs and LMs, we conducted both gene expression analysis and, for the first time, high-resolution array-CGH profiling of matched pairs. Even though the combination of these whole-genome analysis techniques has proven to be a fruitful approach for identifying candidate genes,15 16 this is the first study to apply such a comprehensive approach to colorectal cancer and its metastases.
Material and methods
Patient material
Fresh frozen material of pCRCs and LMs from 22 patients was obtained after informed consent and stored in a pseudo-anonymised fashion according to the approval from the local ethics committee of the University of Heidelberg. Histological examination of a subset of tumours by a pathologist confirmed a tumour cell content of >90%. As controls, DNA and RNA were isolated from seven normal colon and five normal liver samples. Extraction of DNA and RNA was performed via caesium chloride ultracentrifugation from the same material.17 For 21 samples high-quality DNA was available and subsequently used for genomic profiling, whereas high-quality RNA was available from 18 patients and subsequently used for expression profiling. Histopathological parameters are summarised in supplementary table 1.
Genomic profiling
Array-CGH was performed essentially as described previously18 19 and data analysed using R packages (http://www.r-project.org). Data were filtered according to signal/background ratio >3.0, mean/median spot intensity >0.3, replicate standard deviation <0.25 and normalised with a print-tip LOESS algorithm. Genomic regions of different copy number status were defined and copy number ratios within these regions smoothed using GLAD.20 Regions with log2 ratios of less than the median absolute deviation (MAD) of balanced regions were scored as losses and vice versa for gains. Normalised and raw data is available at GEO (http://www.ncbi.nlm.nih.gov/geo; GSE10179). For unsupervised hierarchical clustering the agglomeration rule ‘ward’ and distance metric ‘absolute correlation’ was used and confidence levels of the sample clustering was tested using multiscale bootstrap resampling technique.21 For confirmation purposes, the normalised array CGH data of Mehta et al22 including 50 LMs and Douglas et al23 containing genomic data of the colorectal cancer cell lines SW480 and SW620 were smoothed24 and re-analysed in the same way our data was processed.
Expression profiling
Illumina Sentrix-6 V2 BeadChips (Illumina Inc., San Diego, California, USA) were used for expression profiling. Labelling, hybridisation and washing were performed according to the manufacturer's instructions. Outliers were removed when their expression value dropped below a threshold of median expression +3*MAD of all negative control beads. Individual bead types were filtered when their bead replicate count dropped below 17. We discarded a bead type when the bead type's filter flag was set across all samples. Eighteen thousand, one hundred and eighty-four features passed quality controls and further data analysis was done by variance stabilisation and spline normalising the signals (lumi R package, 1.1.0). Normalised and raw data are deposited at GEO (GSE14297). Significance analysis of microarrays (samr, release 1.26) was performed implementing a permutation-based t-statistic for each oligo. A q-value of <0.05 as well as a log2 ratio of >1.58 (3-fold in linear scale) was used as level of significance. For unsupervised hierarchical clustering we used the agglomeration rule ‘average’ and distance metric ‘correlation’. Cluster confidence levels were tested as described above. Confirmation of differential expression of candidate genes in two publicly available expression array datasets from Koh et al13 and Ki et al14 was done using the Wilcoxon matched-pairs signed-ranks test. Affymetrix raw expression data for the cell lines SW480 and SW620 were taken from GEO dataset GSM274771 and GC-RMA normalised using the biocLite R package. Confirmation of co-expression data with ASCL2 was done using the Affymetrix expression data of Khambata-Ford et al (GSE5851).25
Quantitative real-time PCR (qRT-PCR)
cDNA was generated from all samples included in the expression profiling, except for HD555 and HD668, for which no high-quality RNA remained after the array analyses. pCRC and LM samples were analysed in an iCycler iQ real-time PCR detection system (Bio-Rad, Munich, Germany), using iQ Sybr green supermix (Bio-Rad). Specific primers were designed using Beacon Designer software (Premier Biosoft International, Palo Alto, USA). The primer pairs used are given in supplementary table 3. Relative expression differences between pCRC and LM were determined using the Pfaffl method with the average of ACTB and GAPDH as reference gene.
Results
Genomic aberrations are preserved upon metastasis
Twenty-one pairs of pCRCs and associated LMs from the same patients were analysed for DNA copy number aberrations using CGH arrays consisting of 8000 bacterial artificial chromosome clones. Six normal colon samples were hybridised as controls. A more or less complex aberration pattern was observed for all samples except those of patient HD662, which had a balanced genome both in the primary tumour and the metastasis. Expression of MLH1, MSH2 and MSH6 was detected by immunohistochemistry in pCRC of HD662, indicating that this tumour is microsatellite stable (supplementary figure 2). A frequency plot comparing the aberrations present in pCRC and LM revealed a strikingly high degree of similarity (figure 1a). Except for one small region covering chromosome band 11p15.5, all regions of frequent copy number change showed similar average frequencies and aberration patterns, with the breakpoints delimiting gains and losses being in close vicinity. 11p15.5 showed a frequency difference of 26%, with gains covering most of the chromosome arm occurring in three LMs and smaller gains of the most distal 2.4 Mbp in another three LMs. In unsupervised cluster analysis, 20/21 pCRC and LM pairs were joined pair wise (figure 1b). HD662, the only patient sample with a balanced genome, correctly joined the normal colon samples.
Overall analysis of genomic aberrations in pCRC and LM samples. (a) Average frequency of copy number gain and loss of genomic fragments in the pCRC and LM group, respectively (n=21 each), plotted against their chromosomal position. Chromosome arms of regions showing an aberration in more than 20% of tumours per group are indicated by name. (b) Unsupervised hierarchical clustering of genomic data of 21 pCRC and LM pairs plus six samples of normal colon (NC). All pairs except for HD536 cluster together at end branches and are more similar to each other than to any of the other samples. HD662 (balanced genome) correctly joins the NC samples. Probability values of bootstrap analysis are given at each branch division. AU: approximately unbiased, BP: bootstrap probability, edge #: no. of subdivision.
The striking overlap of the aberration pattern was even more obvious in whole genome plots of individual patients, with most gains, including high-level amplifications, and losses in liver metastases being already present in the primary tumours (figure 2a and supplementary figure 1). Detailed chromosome analyses revealed the preservation of complex aberration patterns down to the level of single genomic fragments (figure 2b and c). Overall, we detected a median of 53 aberrations in the 20 patients who showed at least one aberration, with a minimum of 29 and a maximum of 91 distinct aberrations per patient (table 1). A median of only 11 aberrations per patient was found to be different in pCRCs and LMs. Even in tumours with large numbers of aberrations, in which a further increase of genomic instability might be expected upon metastasis, the aberration pattern was conserved.
Individual genomic profiles of primary colorectal cancer (pCRC; red) and matched liver metastases (LMs; blue). (a) Example of one patient (HD469). Normalised ratio (tumour vs control) of all genomic fragments of pCRC and LMs plotted against chromosomal position. (b) Smoothed ratio (tumour vs control) of all genomic fragments on Chr. 20 of patient HD687 for pCRC and LMs plotted against their chromosomal position. Eight distinct regions with different copy number status and nearly perfect concordance between pCRC and LMs are indicated. (c) Smoothed ratio (tumour vs. control) of all genomic fragments on Chr. 17 of patient HD590 for pCRC and LMs plotted against their chromosomal position. Nine distinct regions with different copy number status and an amplicon with four regions (4–7) of nearly perfect concordance between pCRC and LMs are indicated.
Pair-wise comparisons of genomic aberrations in primary colorectal cancer (pCRC) and liver metastases (LMs)
Global conservation of gene expression patterns upon metastasis
Gene expression arrays were used to analyse the transcription patterns of the same tumour samples as for the genomic screen. High-quality RNA was available for 18 pairs. In addition, the expression profiles of seven normal colon and five normal liver samples were assessed. Unsupervised cluster analysis resulted in good separation of samples according to their origin (data not shown). SAM analysis was performed to identify the genes responsible for the split between most LM and pCRC samples into two separate groups. In metastases, 163 unique genes were significantly over-expressed, whereas expression of 15 genes was significantly lower (SAM analysis, q-value <0.05, minimum fold change of 3, supplementary table 2). Pathway analysis using Ingenuity software identified the most significant canonical pathways as: acute phase response signaling (40 genes, p=1.39e−44), coagulation system (14 genes, p=1.49e−19) and complement system (13 genes, p= 6.4e−18). The biological functions most prominent in the data set were metabolism (39 genes, p=2.33e−14) and inflammatory response (43 genes, p=9.11e−14). Thus, many genes found to be upregulated in LM are associated with normal liver function or code for proteins preferentially synthesised in the liver, such as cytochromes (eg, CYP2E1, CYP4A11), acute phase proteins (eg, CRP, ORM1, SAA1) or plasma proteins (eg, APOA1, FGA, FGB).
As only a few of these genes have been associated with cancer, their detected expression in liver metastases could possibly derive from normal liver cells present within the tumour samples, or, alternatively, might be an acquired phenotype of the metastatic cells. To distinguish these possibilities, protein expression of three of the most differentially regulated genes in LMs versus pCRCs (CYP2E1, ORM1 and SAA) was analysed by immunohistochemistry in normal colon, normal liver, pCRCs and LMs. Similar expression patterns were found in tumour cells of pCRC and LM pairs. Interestingly, liver metastases also contained islands of highly positive hepatocytes interspersed between colorectal cancer cells (supplementary figure 3). These contaminating hepatocyte islands very likely are the source of the detected expression differences between LMs and pCRCs. Consequently, we decided to exclude these genes from further analysis by subtracting liver specific genes (genes significantly higher expressed in normal liver compared to normal colon samples; 659 genes, SAM analysis, q-value <0.05, minimum fold change of 3, data not shown). Re-analysis of LM and pCRC samples by SAM using the reduced data set resulted in a drastically reduced number of significant genes. Only SPP1, the gene coding for osteopontin, was still significantly over-expressed in LM, whereas 15 genes where significantly downregulated (table 2). Of note, 11/15 downregulated genes are already lower expressed in pCRCs compared to normal colon, whereas three genes (SPINK4, MMP3 and REG4) showed only a downregulation in LMs after their expression was initially upregulated in pCRCs compared to normal colon. Overall, comparing the complete pCRC and LM cohorts, expression of 99.9% of all genes (17 509/17 525) in our dataset remained unchanged in colorectal cancer cells upon metastasis to the liver.
In order to validate the genes we find significantly changed in LMs, we evaluated their expression in two publicly available datasets containing gene expression data of matched pCRCs and LMs.13 14 The larger dataset by Ki et al (n=25) confirmed except for one all the genes in our list, whereas the smaller dataset by Koh et al (n=12) was only able to confirm our top five genes, with clear expression differences in three additional genes. Combined, a robust set of 11 genes could be defined which show expression differences between pCRCs and LMs in at least two out of three independent datasets.
ASCL2 related stem cell like phenotype in metastases with 11p15.5 gain
Although the genomic profiles of pCRCs and LMs showed an unexpectedly high degree of similarity, one genomic region with a distinct frequency difference was found on chromosome 11p15.5. This region was gained in six of 21 LMs, two with whole chromosome 11p gain, one with a gain of the telomeric 16 Mbp and three with a small gain of the telomeric 2.4 Mbp (figure 3a). Only one pCRC showed a gain of 0.5 Mbp within this region. Among the genes located within the 0.5 Mb minimal overlapping region gained by all tumours, two were found to be significantly over-expressed in LMs with 11p15.5 versus LMs without this gain (figure 3b): the growth factor IGF2 (insulin-like growth factor 2, p=0.0027, t-test) and the intestinal stem cell specific trascription factor ASCL2 (Achaete–Scute complex homologue 2, p=0.029, t-test).26 The significant differences for IGF2 and ASCL2 were confirmed by qRT-PCR (p=0.002 and p=0.036, respectively; t-test). Liver metastases without 11p15.5 gain showed the same average expression level for ASCL2 and IGF2 as pCRC. Of note, ASCL2 as a direct Wnt pathway target27 is already over-expressed in pCRC compared to normal colon (p<0.001), whereas no elevated expression level was found for IGF2 in pCRC versus normal colon. Further support for a transcriptional upregulation of ASCL2 in tumour cells with a copy number gain of 11p15.5 was found by analysing the genomic and expression data of two matched-pair colorectal cancer cell lines derived from the same patient, SW620 and SW480. The highly metastatic cell line SW620, derived from a metastasis of a patient whose primary cancer yielded the low metastatic cell line SW48028 shows a genomic gain of 11p15.5, whereas SW480 is balanced in this region (figure 3c). The expression level of ASCL2 in SW620 was found to be 1.8-fold higher than in SW480 (figure 3d), consistent with the expression data from LMs with 11p15.5 gain. Additionally, a gain of 11p15.5 was found in 28% of LMs in an independent array-CGH dataset re-analysed in the same way our data was processed (figure 4a, n=50),22 consistent with our data (26% of LMs). No matched data for pCRC is available for these LMs. Nevertheless, no 11p15.5 gain in more than 10% of cases is published in several array-CGH studies of primary cancers.29–32
Genomic aberration and expression pattern of the 11p15.5 region. (a) Average frequency of gain and loss of smoothed genomic fragments on chromosome arm 11p in the pCRC and LM group, respectively (n=21 each), plotted against their genomic position. (b) Average frequency of smoothed genomic fragments on chromosome arm 11p in LM (n=50) re-analysed from Mehta et al22 plotted against their genomic position. (c) Average gene expression level of LM with 11p15.5 gain and without is shown for all genes mapping within the minimally overlapping region of the genomic gain on 11p15.5. Only IGF2 and ASCL2 are significantly differentially expressed (p=0.0027 and p=0.029, respectively; t-test). (d) Smoothed copy number ratio (tumour vs control) of the metastatic and non-metastatic colorectal cancer cell line SW620 and SW480 for 11p15.5-p15.4, respectively. Re-analysed from Douglas et al.23 (e) Expression level of ASCL2 in SW620 and SW480 cell lines (p<0.05, t-test), GC-RMA normalised from GEO dataset GSM274771.
Correlation analysis of ASCL2 direct transcriptional target genes in LM. (a-d) Correlation analysis of expression of ASCL2 (X-axis) and four ASCL2 target genes (ETS2, SOX9, EPHB3 and LGR5) which show higher expression in LM with vs without genomic gain of 11p15.5, in an independent publicly available expression dataset of LM (n=50) [25]. All four genes are highly significantly correlated (p<0.001) with ASCL2 expression (Spearman Rank Correlation Coefficient).
Subsequent analysis of the ASCL2-induced transcription programme in LMs with 11p15.5 gain revealed induction of several downstream target genes. Six known direct transcriptional targets of ASCL2 (EphB3, LGR5, SOAT1, PTPRO, SOX9 and ETS2)26 were induced, albeit not significantly, in LMs with 11p15.5 gain compared to LMs with balanced 11p15.5. qRT-PCR could confirm the induction of 4/6 genes (EphB3 1.3-fold, LGR5 1.4-fold, PTPRO 1.6-fold, SOX9 2.1-fold). Results for the remaining two genes were inconclusive. Furthermore, the intestinal stem cell-specific gene OLFM4 (olfactomedin 4)33 was found to be 5.2-fold (array data) and 6.9-fold (qRT-PCR data) over-expressed in LMs with 11p15.5 gain compared to LMs without. In order to validate the association found in our data between upregulation of ASCL2 and downstream target genes in LMs with 11p15.5 gain, we analysed the largest available gene expression dataset of LMs from Khambata-Ford et al25 for correlation with ASCL2 expression. A significant correlation was found for SOX9, EPHB3, LGR5 (figure 4b–d) and ETS2 (r=0.7, p<0.001). To exclude the possibility that this correlation is primarily due to a higher level of Wnt pathway activation in these LMs, we analysed several known Wnt target genes for correlation with ASCL2. Most of them, including TCF4(TCF7L2), TCF1(TCF7), LEF1, MMP9 and CCND1 showed no correlation with ASCL2. Thus, a co-regulation of 4/6 ASCL2 direct target genes with ASCL2 could be confirmed in an independent set of LMs.
Discussion
An increase in number and complexity of genomic alteration has been observed along the progression of colorectal cancer from benign adenoma to invasive carcinoma, with the steepest increase in CRC specific aberrations during the step of invasion7 34 35 36 Similarly, expression profiling studies were able to distinguish different Dukes’ stages.37 38 However, whether there are specific alterations that characterise the last step of progression, from invasive carcinoma to metastatic disease, is still a matter of debate. Many studies addressing this question compared genomes of primary cancers to unrelated liver metastases using expression profiling37 39 40 41 or genomic profiling.42 43 Unless based on very large sample groups, this type of unmatched comparison is inherently unreliable due to the tremendous inter-individual differences between primary tumours. Consequently, many of the pCRC–LM differences that have been identified could not be confirmed in independent studies and are likely false-positives rather than aberrations that are truly associated with liver metastasis. In order to overcome these limitations, we compared matched pairs of pCRCs and LMs and showed that LMs exhibit a previously unrecognised high degree of similarity to their primary tumours on the genomic and expression level. In unsupervised cluster analysis, the genomic aberration pattern of LMs was in 20/21 cases more similar to their primary than to unrelated LMs or pCRCs. This is in contrast to prior studies comparing unmatched pairs and confirms, with improved resolution down to fewer than 100 kb, earlier matched-pair CGH studies which found no statistical significant differences between pCRCs and LMs.7 9 Our data suggest that, at the time of diagnosis, the pCRCs have ‘matured’ in the sense that they have reached an almost steady state of genomic alterations with the balance between growth supporting and growth inhibiting alterations being optimised for the survival of the cancer.44 45 At which stage of CRC development genome maturation is completed is unclear; however, this stage might already be reached in malignant polyps.36 46 This model provides the simplest explanation for our observation that pCRC and liver metastases, despite of years of independent growth, retain nearly identical aberration profiles. An in-parallel acquisition of aberrations to such a high degree independently in two growing tumours at different organ sites is in our view rather unlikely. Interestingly, this seems to stand in contrast to mammary carcinoma metastasis where an early dissemination and independent growth of metastases has been proposed.47
Accordingly, comparative gene expression profiling of matched pCRCs and LMs revealed a very low number (< 0.01%) of statistically significant, differentially expressed genes (n=16). The differential expression of these genes was confirmed for 11 genes in independent datasets. In particular, the dataset of Ki et al with 25 matched pairs conclusively confirmed our data. The only upregulated gene in our data was osteopontin (SPP1), one of the few firmly established markers for colon cancer progression.37 The upregulation of SPP1 in LMs was consistent in the two additionally analysed datasets and emerges as the only over-expressed gene in LMs compared to the primary tumour. Within the 15 genes downregulated in LMs, 11 showed already a marked reduction in expression in pCRC compared to normal colon. Furthermore, these genes are all described to be expressed in the differentiated compartment of the normal colonic crypt, suggesting a downregulation of lineage markers first upon transformation and secondly after leaving the intestinal niche. Low in LM but expressed at the same level between normal tissue and pCRC was ACTG2, an intestinal specific smooth muscle actin. This might be explained by expression of this form of smooth muscle actin in intestinal tissue and tumour stroma which typically are contained within pCRCs, but absent in the liver niche. Three genes showed a characteristic upregulation in pCRCs and a subsequent downregulation in LMs: SPINK4, MMP3 and REG4. Whereas SPINK4 and REG4 are expressed in differentiated cells (goblet cells and neuroendocrine cells, respectively48 49), MMP3 is mainly expressed in the connective tissue, implicated in the turnover of extracellular matrix to allow cell migration.50 Why these three genes confer a possible tumour promoting function in primary tumours, whereas their expression seems to be dispensable in liver metastases, is unclear. In particular, the downregulation of REG4 is intriguing, as it has been described as a factor predicting liver metastasis in primary tumours.51
Despite the high similarity on the genomic level, one small genomic region of 2.4 Mb on chromosome band 11p15.5 was found to be a distinct feature of a subset of LMs. We could confirm this gain in approximately 30% of LMs in an independent dataset, whereas this gain has not been described in pCRC.29–32 Using 11p15.5 as a marker for this subset, comparative analysis of the genomic and gene expression profiling data sets identified two genes, IGF2 and ASCL2, that were significantly over-expressed in tumours carrying the 11p15.5 gain compared to LMs or pCRCs without this aberration. ASCL2 is a transcription factor and known Wnt pathway target gene, found to be over-expressed in colorectal cancer also by others.27 52 Its mouse homologue recently was shown to be expressed exclusively in intestinal stem cells and to specifically control part of the stem cell specific transcription profile.26 Over-expression of ASCL2 appeared to induce this ‘stem cell’ expression pattern in the subset of LMs carrying the 11p15.5 gain as evidenced by higher expression of known target genes, including SOX9 and OLFM4. As no other study has up to now compared matched pairs with both high resolution array-CGH and expression profiling, a direct confirmation in a second dataset of the association between gain of 11p15.5 and over-expression of ASCL2 as well as downstream target genes was not possible. Nevertheless, we found that the metastasis-derived colorectal cancer cell line SW620 gained 11p15.5 and concordantly over-expressed ASCL2 when compared to SW480, a cell line derived from the pCRC of the same patient.28 Furthermore, we showed recently that ASCL2 expression should, in general, be correlated with the known downstream target genes. Using the expression data of Khambata-Ford et al25 (n=50 LMs) we could confirm the highly significant correlation of ASCL2 with 4/6 of the known direct target genes, namely SOX9, EPHB3, ETS2 and LGR5. This correlation is independent of a, generally, higher Wnt pathway activation in these tumours, as no correlation was found with several known ASCL2 independent target genes. High expression of ASCL2 therefore leads to or is correlated with an activation of its downstream stem cell related target genes in colorectal liver metastases.
The second gene apparently affected by the 11p15.5 gain was IGF2, an important autocrine growth factor in different tumour entities including colorectal cancer.53 Interestingly, loss of imprinting at the IGF2 locus leads to a similar increase of IGF2 expression (2- to 3-fold) as the one associated with gain of 11p15.5 in our study (3.3-fold), and results in increased expression of proliferation-related genes and an expansion of the intestinal crypt progenitor cell population in mice.54 55 Although not all cells in metastases with 11p15.5 gain will display stem-like properties, IGF2 and ASCL2 over-expression likely will shift the hierarchy of stem and progenitor cells resulting in more self-renewal rather than differentiation and potentially affecting the clinical behaviour of these tumours. This is of special interest also in the light of current clinical trials targeting the IGF receptor 1, of which IGF2 is a ligand.
In the present study, we demonstrated with unprecedented high resolution the striking conservation of molecular alterations in liver metastases compared to their origin. Despite these extensive similarities, in a combinatorial approach a stem cell specific transcription signature promoted by ASCL2 has been identified to be upregulated in a subset of liver metastases, those characterised by an 11p15.5 gain. Taken together, this systematic study laid the ground for a better understanding of the biology of liver metastasis in colorectal carcinoma. Generating mouse models specifically over-expressing or deleting Ascl2 in e.g. adenomas of Apcmin mice could help to unravel the in vivo function of Ascl2 in colorectal cancer cells in the near future.
Acknowledgments
The authors would like to thank Hans Clevers for fruitful discussions of the data. We furthermore gratefully acknowledge the excellent technical assistance of Stefanie Hofmann.
References
Supplementary materials
web only files
Files in this Data Supplement:
Footnotes
Linked articles 212241
Funding This work was supported by a grant from the Tumorzentrum Heidelberg–Mannheim to BR, TL and PS, and by the platform activities funded by grant 01GS0883 of the Deutsches Bundesministerium für Bildung und Forschung to BR and PL.
Competing interests None.
Ethics approval This study was conducted with the approval of the Ethics Committee of the University of Heidelberg.
Provenance and peer review Not commissioned; externally peer reviewed.
Linked Articles
- Commentary