Metastatic transcriptional pattern revealed by gene expression profiling in primary colorectal carcinoma

Authors


Abstract

Metastatic spread to the liver is the major contributor to mortality in patients with colorectal carcinoma (CRC). In order to seek for gene expression patterns associated with metastatic potential in primary CRC, we compared the transcriptional profiles of 10 radically resected primary CRCs from patients who did not develop distant metastases within a 5-year follow-up period with those of 10 primary/metastatic tumor pairs from patients with synchronous liver metastases. To focus selectively on neoplastic cells, the study was conducted on laser-microdissected bioptic tissues. Arrays of 7,864 human cDNAs were utilized. While a striking transcriptional similarity was observed between the primary tumors and their distant metastases, the nonmetastasizing primary tumors were clearly distinct from the primary/metastatic tumor pairs. Of 37 gene expression differences found between the 2 groups of primary tumors, 29 also distinguished nonmetastasizing tumors from metastases. The gene encoding for mannosyl (α-1,3-)-glycoprotein β-1,4-N-acetyl-glucosaminyl-transferase (GnT-IV) became significantly upregulated in primary/metastatic tumor pairs (p < 0.001). GnT-IV upregulation was confirmed by RT-PCR. These data support the existence of a specific transcriptional signature distinguishing primary colon adenocarcinomas with different metastatic potential, the further pursuit of which may lead to relevant clinical and therapeutic applications. © 2005 Wiley-Liss, Inc.

Colorectal carcinoma (CRC) is the second leading cause of cancer deaths in developed countries.1 Metastatic spread, mostly to the liver, is the major cause of mortality in patients with this disease. Yet while the molecular events involved in the CRC adenoma-carcinoma sequence have been largely characterized,2 little information is available on the mechanisms responsible for the metastatic phenotype. The identification in the primary tumors of molecular factors predictive of metastatic risk may have a relevant clinical impact, allowing for more personalized and effective treatment of CRC patients.

Recent evidence based on gene expression profiling has shown that a metastatic fingerprint is detectable in the bulk of primary lung and breast tumors.3, 4 On the other hand, no such metastatic signature has been demonstrated in CRC. A general problem related to gene expression studies in the oncologic field is the heterogeneous histology of the bioptic specimens that may contain, in addition to cancer cells, significant fractions of nonneoplastic tissue. Due to this heterogeneity, relevant changes in cancer cell gene expression may be easily overlooked.

In order to seek for gene expression patterns specific to metastasis in CRC carcinoma, avoiding interference from nontumoral tissue, we combined laser microdissection of cancer cells and DNA microarray to compare the transcriptional profile of primary colon carcinomas both with that of patient-matched liver metastases and with that of a group of primary tumors of metastasis-free patients at presentation and during a 5-year follow-up.

Material and methods

Patients and tissue samples

The study population consisted of 20 fully informed consent patients who underwent surgery for colorectal adenocarcinoma at our institution between 1992 and 2001. Cases were selected from our prospective CRC database on the basis of their presenting characteristics and outcome (10 without evidence of distant metastasis for at least a 5-year follow-up period, and 10 with synchronous liver metastasis). Tumors were staged according to the international tumor node metastasis (TNM) staging system and the histologic grade was assessed according to World Health Organization (WHO) criteria. The group of patients without distant metastasis was comprised of 6 men and 4 women (mean age, 69 years; range, 55–92). The tumor was located in the colon in 3 cases (30%) and in the rectum in 7 cases (70%). All tumors were stage II (T3, N0); 3 of the tumors (30%) were well differentiated, 5 (50%) were moderately differentiated and 2 (20%) were poorly differentiated. The group of patients with synchronous liver metastasis was comprised of 5 men and 5 women (mean age, 61 years; range, 42–78). The tumor was located in the colon in 6 cases (60%) and in the rectum in 4 cases (40%). One of the tumors (10%) was T2, 7 (70%) were T3 and 2 (20%) T4; 5 of the tumors (50%) were N1 and 5 (50%) N2. One of the tumors (10%) was well differentiated, 7 (70%) were moderately differentiated and 2 (20%) were poorly differentiated. Patients were evaluated every 6 months until the end of the second year and yearly thereafter. Follow-up included physical examination, hepatic enzymes and serum carcinoembryonic antigen assay, liver ultrasound or computed tomography, chest X-ray (yearly) and colonoscopy (yearly). Bioptic specimens were snap-frozen in liquid nitrogen immediately after excision using RNase-free vials without other protective solutions or optimum cutting temperature (OCT) embedding. The samples were therefore stored in liquid nitrogen until being further processed.

Array construction

A 7,864-cDNA array comprising a duplicated panel of human genes from Incyte Genomics (Palo Alto, CA) was assembled onto Seven Star metal-coated slides (Amersham-Pharmacia Biotech, Little Chalfont, U.K.) by using an Amersham-Pharmacia/Molecular Dynamics Gen III Spotter. Target genes were in 50% DMSO (list of genes available in Supplementary Table II).

Table II. Transcriptional Changes Distinguishing PM Tumors from both PM+ Tumors and from Metastases
GeneUnigeneLog 2 ratioVariationp
PM+MetPM
  1. The averages of tumor-to-reference log2 ratios of fluorescence are reported for the 3 groups of tumors. Up and down variation indicate induction and repression in PM+ and metastases with respect to PM tumors.

Tetraspanin 3Hs. 1000900.370.44−0.05Up0.0387
ESTHs. 3427800.180.11−0.87Up0.0216
ESTHs. 318404−0.18−0.18−0.88Up0.0182
ESTHs. 1934820.520.61−0.05Up0.0132
Homo sapiens TNFR-related death receptor-6 (DR6)Hs. 1596510.020.04−0.91Up0.0101
ESTHs. 29700−0.60−0.72−1.52Up0.0118
Phosphotriesterase-relatedHs. 129915−0.83−0.88−0.14Up0.0067
PSME4: proteasome (prosome, macropain) activator subunit 4Hs. 112396−0.01−0.01−0.79Up0.0058
ESTHs. 2954480.190.19−0.69Up0.0214
ESTHs. 2841420.871.05−0.31Up0.0276
Serine (or cysteine) proteinase inhibitor, clade B (ovalbumin), member 9Hs. 104879−0.030.11−0.44Up0.0332
Myogenin (myogenic factor 4)Hs. 28301.081.010.44Up0.0082
ATP-binding cassette, subfamily D (ALD), member 2Hs. 1178520.220.34−0.39Up0.0477
Nucleoporin 50 kDHs. 271623−0.08−0.11−0.63Up0.0037
ESTHs. 324179−0.25−0.23−0.73Up0.0287
CENTD2: centaurin, delta 2Hs. 212641.151.000.47Up0.0127
Metastasis-associated 1-like 1Hs. 1730430.100.14−0.36Up0.0172
Gn-T IV: mannosyl (alpha-1, 3-)-glycoprotein beta-1, 4-N-acetylglucosaminyltransferase, isoenzyme AHs. 1775760.900.83−1.13Up0.0006
Glycoprotein, synaptic 2Hs. 3061220.490.18−0.55Up0.0063
Bladder cancer overexpressed proteinHs. 125830−0.24−0.290.29Down0.0187
ESTHs. 1460700.170.170.77Down0.0243
Protein kinase, cAMP-dependent, regulatory, type I, betaHs. 1519−0.16−0.110.27Down0.0249
Voltage-dependent anion channel 3Hs.7381−0.15−0.070.82Down0.0365
Growth arrest-specific 7Hs. 226133−0.49−0.580.05Down0.0272
Molecule possessing ankyrin repeats induced by lipopolysaccharide (MAIL); similarity with the I kappa B family of proteinsHs. 301183−0.34−0.390.48Down0.0251
rcd 1 (required for cell differentiation, S. pombe) homolog 1Hs. 94211−0.30−0.140.53Down0.0328
Eukaryotic translation initiation factor 3, subunit 8 (110 KD)Hs. 4835−1.37−1.53−0.06Down0.0068
ESTHs.1065340.110.160.48Down0.0493
Solute carrier family 25 (mitochondrial carrier; Graves disease autoantigen), member 16Hs. 180408−0.30−0.220.94Down0.0018

Laser microdissection

Laser microdissection was performed on frozen samples of 10 primary colorectal adenocarcinomas without metastasis, 10 primary colorectal metastatic adenocarcinomas and their synchronous metastases using AS-LMD Laser Microdissection System (Leica Microsystems, Wetzlar, Germany). Approximately 10,000 neoplastic cells from each specimen were collected in RNAse-free conditions. Briefly, 7 μm cryostat sections were obtained from snap-frozen tumor specimens maintained at −80°C by using a disposable blade; the samples were anchored on cryostat supports using diethylpyrocarbonate (DEPC)-water (without OCT embedding). The sections were then mounted on polyethylennaphtalate (PEN)-foiled slides (Leica Microsystems), consisting of microscope glass slides that support a thin plastic (PEN) film designed for use with AS-LMD. Immediately after the cut, the sections were fixed for 1 min in 70% alcohol, slightly stained with Mayer's hematoxylin (30 sec), washed in 2 baths of DEPC-water (5 min), dehydrated in increasing alcohols (80%, 95%, 100%), placed at 37°C for 30 min and microdissectioned at once. To suppress RNase activity, DEPC-water was used for alcohol dilutions, too. Moreover, a hematoxylin-eosin-stained section was prepared for histologic evaluation of each sample. The areas of interest were selected on a computer screen using the mouse and excised by using a pulsed UV laser (337.1 nm wavelength) cutting the plastic film along the drawn line. The excised cells were collected in a PCR tube located beneath the slide.

RNA extraction and amplification

Total RNAs were extracted from each sample (∼1 × 104 cells) of laser-captured cells into 800 μl of TRIzol Reagent (Invitrogen, Carlsbad, CA) in the presence of 200 μg RNase-free glycogen as a carrier. Genomic DNA was degraded by the DNA-free system (Ambion, Woodward Austin, TX). The purified RNAs were stored at −80°C in the presence of 1.4 U/ml RNase inhibitor (RNaseOUT; catalog number 10777; Invitrogen). All the RNAs were subjected to one round of T7-based linear amplification (MessageAmptm aRNA kit; catalog number 1750; Ambion). The quality of the starting total RNA and of the amplified mRNA (aRNA) was assessed by an Agilent 2100 Bioanalyzer (Agilent Technologies, Palo Alto, CA).

Probe preparation, hybridization and laser scanning

A total of 2 μg of the amplified pool of messenger RNAs was used as a template to prepare reversely labeled cDNA probes by the CyScribe FirstStrand cDNA labeling kit (Amersham-Pharmacia). In each hybridization, a Cy5-labeled cDNA probe from Universal Human Reference RNA (Stratagene, La Jolla, CA) was mixed with each of the Cy3-labeled tumor probes.

The reference consisted of a pool of RNAs from 10 different human cell lines. An amount of each probe containing 10 pmoles of Cy dye was used in each hybridization. The probe mixture was diluted using a formamide containing hybridization buffer and 30 μl of each mixture were applied on each arrayed slide and incubated 14–18 hr at 42°C in a humidified chamber. The slides were washed in buffers preheated at 55°C unless otherwise specified (3 × 5 min in 1 × SSC/0.1% SDS, 4 × 5 min in 0.1 × SSC/0.1% SDS, 2 × 5 min in room temperature 0.1 × SSC, 2 × 10 sec in room temperature H2O). Each slide was rapidly dried by centrifugation and then laser-scanned with a Gen III Amersham/Molecular Dynamics instrument.

Data analysis

An Array Vision software package (Imaging Research, St. Catharine, Ontario, Canada) was used to visualize the images and for filtration and normalization of raw data. Data were expressed as sample-to-reference log2 ratios. After within-slide LOWESS normalization, scale normalization was performed to correct for differences in signal intensity between slides. Only spots with intensities above background in both duplicate arrays of each slide and in at least 9 out of 10 tumors in each group were considered for further analysis. In the comparisons between independent groups, i.e., primary tumors without metastases (PM) vs. primary tumors with synchronous metastases (PM+) and PMvs. metastases, 2-sample Welch t-statistics was applied. In the paired comparison (i.e., PM+vs. metastases), paired t-statistic was applied. p-values were obtained by permutations and adjusted by the minP step-down multiple-testing procedure.5 Genes were ranked on the basis of their adjusted p-values. We defined as informative genes those with unadjusted p < 0.05 and as differentially expressed genes those with a significant expression difference (adjusted p < 0.05). In order to represent the detected transcriptional differences graphically, Sammon's nonlinear mapping based on informative hits was applied.6 All computation and graphics were made in R environment (www.r-project.org) and Bioconductor (www.bioconductor.org).

Data validation by RT-PCR

In order to validate the microarray results, we applied RT-PCR. Reagents for reverse transcription were from Promega (Madison, WI). For each PCR reaction, 1.25 units of Taq polymerase were utilized (AB Analitica, Padua, Italy). The primers were GTAAAATGTTTCAAGCGCCG (forward) and GAAGGTCTGAAGCGAATTCG (reverse) for GnT-IV, CTTCCAGTAAAGCTTGTGTCG (forward) and CTTCCCTGTTCGCGTACA (reverse) for Nup50, TCATGAAGTGTGACGTTGACATCCGT (forward) and CCTAGAAGCATTTGCGGTGCACGATG (reverse) for β-actin. PCR cycles for GnT-IV and Nup50 were as follows: 2 min at 94°C, 30 cycles of 30 sec at 94°C, 1 min at 58°C and 1 min at 72°C, followed by 10 min at 72°C; for β-actin, they were 2 min at 94°C, 26 cycles of 30 sec at 94°C, 1 min at 65°C and 2 min at 68°C, followed by 7 min at 72°C.

Results

In order to identify gene expression changes of relevance for metastasis, the transcriptional profiles of 10 primary colorectal adenocarcinomas from patients with synchronous liver metastasis (PM+) were first compared with those of 10 primary tumors from patients without distant metastasis at the time of surgery and free from recurrence after a 5-year follow-up (PM; Table I). Virtually pure populations of cancer cells were obtained by laser microdissection of the tumor biopsies (Fig. 1) and cDNA arrays of 4,224 human genes were used to analyze gene expression profiles. Present genes were defined as those the spots of which were above background in at least 90% of the samples in each group. A panel of 37 informative genes out of 200 present genes distinguished the PM+ from the PM tumors. Of these, 23 were upregulated in PM+ relatively to PM tumors and 14 were downregulated (Supplementary Table I).

Figure 1.

Selective microdissection of neoplastic cells from frozen sections of primary colon cancer stained with hematoxylin. The red line encircles the area that will be microdissected (original magnification 20×).

Table I. Patient Characteristics
NumberSexAgeLocationTNMG
Patients devoid of metastasis
 579F64Rectum3001
 633M70Colon3001
 767F68Rectum3001
 1002M61Rectum3002
 1311M66Rectum3002
 1400M55Colon3002
 442M92Colon3002
 603M71Rectum3003
 1115F70Rectum3003
 1543F69Rectum3002
Patients with metastasis
 5647M55Colon3112
 4230M42Rectum4112
 6236F76Colon3213
 5808F78Colon3213
 5527M60Rectum3212
 4216F58Colon2212
 5007M65Colon4212
 654M55Rectum3112
 2099F52Rectum3111
 6661F66Colon3112

However, when the gene expression profiles of PM+ tumors were compared to those of their metastases, a substantial similarity was observed between the 2 groups, and only 6 informative changes out of 773 present genes were detected. For 5 of these genes, no significant differences were found between the metastases and the PM tumors, suggesting that they are not part of a metastatic gene expression pattern. In contrast, the expression level of thrombin was significantly higher in the metastatic cells with respect to both PM and PM+ primary tumors.

Next, to evaluate the significance of the transcriptional changes found between PM+ and PM tumors in the context of metastasis formation, the gene expression profiles of PM tumors were compared with those of the metastases. Of 253 present genes, 75 were informative genes. Of these, 39 were upregulated in the metastases and 36 were downregulated. Notably, of the 37 transcriptional differences found between PM+ and PM primary tumors, 29 also distinguished the metastases from PM tumors (Table II). Thus, 29 changes were shared by PM+ tumors and by their distant metastases, distinguishing both groups from PM primary tumors.

A global representation of the gene expression comparisons in a reduced dimension space, based on Sammon's nonlinear mapping and expressing the similarity between tumor samples as relative distances, is shown in Figure 3. As only 6 informative changes occurred between PM+ tumors and the metastases, the graphs were all obtained on the basis of the same number of most informative differences in each of the comparisons. However, similar graphs were obtained when the same analysis was performed on the basis of all the informative genes in each comparison (not shown). While PM+ tumors and metastases displayed a substantially overlapping distribution, PM tumors were clearly distinct from both PM+ and the metastases.

Figure 3.

Representation of the gene expression data. (ac) Sammon's nonlinear mapping of samples referred to the 3 pairwise comparisons. Open circles represent PM tumors; filled squares, the PM+ tumors; filled triangles, the metastases. Distances are calculated on the basis of the 6 most informative genes in each of the comparisons. (d) Percentage of the present genes displaying significant variations.

Among the genes distinguishing PM+ tumors and their metastases from PM tumors (Table II), 9 corresponded to EST sequences of unknown function. Among the others, we found the gene encoding for mannosyl (α-1,3-)-glycoprotein β-1,4-N-acetyl-glucosaminyltransferase (GnT-IV). This gene, of potential relevance for cancer progression,7 was induced in primary/metastatic pairs in comparison to PM tumors. The statistical significance of the GnT-IV upregulation was high enough to define this gene as differentially expressed. The gene encoding for Nup50 was also upregulated in the PM+ and the metastases. NUP50 encodes for a nucleoporin complex component involved in nucleocytoplasmic protein trafficking.8, 9 Thus, in order to verify the microarray data by an alternative approach, 6 tumor specimens were randomly selected from each group of samples and the expression levels of GnT-IV and Nup50 were compared to the housekeeping gene, β-actin, by reverse transcription followed by PCR (Fig. 2). The RT-PCR clearly confirmed the upregulation of both genes in the primary/metastatic tumor pairs relatively to nonmetastasizing tumors, supporting the microarray results.

Figure 2.

Confirmation of microarray data by RT-PCR on 6 RNAs from each group of tumors; 100 ng of each linearly amplified RNA were reversely transcribed and used as PCR templates. While β-actin was expressed at comparable levels in all groups of tumors, GnT-IV and Nup50 were dramatically upregulated in the metastasizing primary tumors and in the correspondent metastases.

Discussion

The gene expression profile of colon adenocarcinomas with metastatic ability was compared both with that of the correspondent synchronous liver metastases and with that of tumors that did not develop metastases during follow-up. To focus on tumoral tissue, the study was conducted on virtually pure cancer cell populations obtained by laser microdissection from the bioptic specimens. Our results show that nonmetastasizing tumors display transcriptional profiles significantly different from those of the primary/metastatic tumor pairs. These findings suggest that specific gene expression signatures may be of utility in correctly identifying clinically relevant subclasses of cancer patients. Furthermore, given the close transcriptional resemblance observed between the metastatic cells and their original tumor cells, it is likely that major cell fractions rather than rare sporadic cells of the primary tumor possess metastatic competence. These data are consistent with recent microarray studies conducted on lung and breast cancer. Indeed, by comparing a variety of metastatic and primary solid tumors, Ramaswamy et al.3 identified a transcriptional panel distinguishing lung primary tumors with different metastatic abilities and prognosis. Moreover, Weigelt et al.4 observed a high similarity in the gene expression profiles between primary breast tumors and their distant metastases.

To date, microarray-based studies in CRC have mainly focused on comparison of primary tumor to normal samples or of primary tumor samples at different stages of disease.10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21 However, few studies have directly addressed the mechanisms of metastasis formation or supplied information of prognostic value. Recently, Bertucci et al.22 were able to identify a gene set distinguishing groups of CRC patients with significantly different 5-year survival, while in another report a 23-gene expression cluster predicting a 13-fold risk of disease recurrence in stage II CRC patients was identified.23 These findings are of considerable clinical value as the recognition of a metastatic fingerprint in subsets of stage II CRC may help identify those patients who could benefit from adjuvant chemotherapy, a treatment that has already proven to be highly effective in stage III patients.24, 25, 26 Moreover, most recently, a study comparing the gene expression profile of colorectal tumors at different stages showed that many stage IV tumors tend to cluster with metastases in hierarchical clustering analyses.27 These results are in accord with the substantial transcriptional similarity we found between metastases and those of their original tumors.

Most studies on gene expression in CRC have been conducted on tissue specimens containing both tumoral and stromal cells, suggesting that tissue microdissection is a dispensible procedure for detecting relevant cancer-related changes. However, as our surgical specimens contained variable fractions of nontumoral cells (10% to more than 50% of total tissue; data not shown), we preferred to include a laser microdissection step to reduce this potential source of variability. Here, by focusing on the transcriptional profiles of primary tumors with clear-cut clinical differences in metastatic potential, we provide evidence that the malignancy of human CRC adenocarcinoma cells is correlated with a unique gene expression pattern. A subset of 29 genes showed a differential expression in the primary tumor/metastasis pairs compared to the primary nonmetastasizing tumors. Of these, 9 were EST sequences of unknown function and at least 10 were genes of potential oncologic relevance (Table II). Nineteen genes were upregulated in the metastases and in the correspondent primary tumors, whereas 10 genes were downregulated compared to nonmetastasizing tumors. Among the upregulated genes, we found MTA1L1, an homolog of the metastasis-associated MTA1. MTA1 is involved in chromatin remodeling and might function as a transcriptional regulator. This gene was reported to be overexpressed in breast and gastrointestinal metastatic tissue.28, 29 This, together with our data, supports a possible function of MTA1L1 in metastatic progression. Of special interest in the context of colon carcinoma is the gene encoding for mannosyl (α-1,3-)-glycoprotein β-1,4-N-acetylglucosaminyltransferase (GnT-IV). GnT enzymes catalyze N-acetylglucosamine branching on asparagine-linked oligosaccharides of cell proteins.30, 31GnT-IV transcription and protein activity was found by others to be considerably high in choriocarcinoma tissue,7 and recent reports relate the level of expression of another gene of the same family, GnT-V, to metastases and poor prognosis in CRC.32 The increased GnT-IV expression that we observed in primary tumor/metastases pairs could reflect a role of the encoded protein in the formation of metastasis. Yet, another gene we found upregulated in primary/metastatic pairs is nucleoporin-50 (NUP50), a component of the nuclear pores. NUP50 has been found to interact with p27Kip1, which in turn inhibits the activity of cyclin-dependent kinases in response to antiproliferative stimuli. The downregulation of p27Kip1 requires its interaction with components of the nuclear pore.8, 9 Thus, an increased expression of Nup50 might affect p27Kip1 degradation, thereby altering the control of cell proliferation. Among the downregulated genes, we found the growth arrest-specific gas7 gene. Although the exact function of gas7 remains to be elucidated, it is known that the expression of gas proteins is enhanced in cells entering a quiescent state.33 Therefore, the downregulation of gas genes is expected in tumor cells with high metastatic ability.

As the repertoire of genes we have analyzed was relatively limited, it is likely that several other genes not included in our arrays may contribute to tumor progression. Nevertheless, the selection of a cancer-related subset of genes from an array of unbiased composition suggests that the variations we observed are indeed part of a transcriptional pattern required for metastatic development. The identified changes may either be directly involved in the primary/metastatic transition or, alternatively, contribute to the acquisition of gene expression features predisposing to metastasis, albeit not directly required for tumor spreading. In this context, the increased expression of tetraspanin-3 in PM+ tumors and in the metastases may affect the repertoire of cell adhesion molecules, with possible consequences on the activation of an invasive tumor phenotype. On the other hand, variations of genes such as Nup50 or the growth arrest-associated gas7, which are likely to affect proliferation and apoptosis, may be related to the enhanced ability of tumor cells to survive under the unfavorable conditions encountered in their route to distant sites of invasion. It is also noteworthy that while PM tumors were transcriptionally distinct from both PM+ and metastatic tumors (18% and 30% of all present genes changed, respectively; Fig. 3), the latter 2 groups were barely discernible on a molecular level. Only the gene encoding for thrombin was found to be overexpressed in the metastatic cells in comparison with both groups of primary tumors, possibly reflecting a role in tumor-induced angiogenesis.

Although a molecular dissection of metastatic development is not available yet, the hints so far supplied by both gene expression profiling and other approaches make it possible to advance some hypotheses. Primary tumors with different metastatic abilities could evolve independently and once the appropriate global gene expression pattern has been activated in a subset of tumors, minor changes are required for metastatic spreading. Certain primary tumors may become primed for metastasis very early in carcinogenesis (i.e., they start off with the “bad foot”), although this may not be sufficient for full metastatic competence. Clonal expansion events will thereafter occur and be needed to acquire the final metastatic phenotype. Further studies addressing both gene expression at early stages of carcinogenesis and the influence of the genetic background on metastasis formation34 will be necessary to clarify these issues.

In conclusion, our data are consistent with the existence of distinct gene expressional features, intrinsic to cancer cells and not attributable to the adjacent tissues, associated with high metastatic potential in primary colorectal tumors. These findings may have profound clinical implications. More specifically, the identification and validation of a metastatic fingerprint early in tumor development may help identify stage II CRC patients with high risk of developing distant metastases during the 5-year follow-up, who will therefore benefit from adjuvant therapy, and spare unnecessary treatments for the low-risk patients.

Acknowledgements

The authors are extremely grateful to Gavin Hardy from Amersham Biosciences for his substantial technical assistance, Katia Tavella of the Department of Oncologic and Surgical Sciences, University of Padova, for her assistance in the preparation of RNA samples, as well as the Association “Antonio R. Cananzi” for Headache and Cerebrovascular Diseases. Supported in part by grant 3933 (to Research and Innovation) from the Italian Ministry of Education, University and Research.

Ancillary