Maspin is a marker for early recurrence in primary stage III and IV colorectal cancer.

BACKGROUND: Little is known about the factors that drive metastasis formation in colorectal cancer (CRC). Here, we set out to identify genes and proteins in patients with colorectal liver metastases that correlate with early disease recurrence. Such factors may predict a propensity for metastasis in earlier stages of CRC.
METHODS: Gene expression profiling and proteomics were used to identify differentially expressed genes/proteins in resected liver metastases that recurred within 6 months following liver surgery vs those that did not recur for >24 months. Expression of the identified genes/proteins in stage II (n=243) and III (n=176) tumours was analysed by immunohistochemistry on tissue microarrays. Correlation of protein levels with stage-specific outcome was assessed by uni- and multivariable analyses.
RESULTS: Both gene expression profiling and proteomics identified Maspin to be differentially expressed in colorectal liver metastases with early (<6 months) and prolonged (>24 months) time to recurrence. Immunohistochemical analysis of Maspin expression on tumour sections revealed that it was an independent predictor of time to recurrence (log-rank P=0.004) and CRC-specific survival (P=0.000) in stage III CRC. High Maspin expression was also correlated with mucinous differentiation. In stage II CRC patients, high Maspin expression did not correlate with survival but was correlated with a right-sided tumour location.
CONCLUSION: High Maspin expression correlates with poor outcome in CRC after spread to the local lymph nodes. Therefore, Maspin may have a stage-specific function possibly related to tumour cell dissemination and/or metastatic outgrowth.

Five-year survival rates in colorectal cancer (CRC) vary dramatically from ∼93% in patients with localised stage I tumours to ∼6% in patients with metastasised inoperable disease (Siegel ). The decision to administer adjuvant chemotherapy in CRC is predominantly stage-dependent. All stage III and high-risk stage II patients are recommended to receive adjuvant treatment. In contrast, patients with stage I and low-risk II disease generally do not receive adjuvant chemotherapy after resection of the primary tumour, as 5-year survival rates are high and there is only minimal benefit in unselected patients (O'Connell ; Edge and Compton, 2010). Recently, gene expression profiling has been used to identify high-risk stage II patients who can possibly benefit from adjuvant chemotherapy. Two published gene signatures have proven to be able to select stage II cancer patients for adjuvant treatment: the Oncotype DX Colon Cancer Assay (12 genes) and the ColoPrint (18 genes). Both signatures seem to perform better than most conventional risk factors (Salazar ; Webber ; Kelley ) and are currently being further validated. The discovery of novel biomarkers for predicting tumour behaviour is not only valuable from a prognostic viewpoint, but can also help to gain insight into the mechanisms that cause disease progression and metastasis formation. Differences in tumour cell survival, migration, invasion and extracellular matrix remodelling are all likely to have a role (Hanahan and Weinberg, 2000).

To turn a locally invasive colorectal tumour into a metastasising one very few, if any, additional genetic changes are required (Jones ). It is possible that the factors driving initial metastasis formation from the primary tumour also promote tumour recurrence at later stages of the disease, following surgical removal of liver metastases (Eberhard ). After resection of colorectal liver metastases, the vast majority of patients experience tumour recurrence within 2 years. However, the patients that remain disease-free beyond 2 years have a good chance of 5-year survival without additional therapy (Nordlinger ; Fong ; Simmonds ; de Jong ). This is most likely due to favourable tumour characteristics.

The rationale of this study was to obtain further insight into the mechanisms causing tumour progression and metastases formation. Gene expression profiling and proteomics were used to detect differences between stage IV CRC patients that recur within 6 months after resection vs those that fail to recur for >2 years. Subsequently, we analysed whether expression of the identified factors also predicts metastasis formation in early stages of CRC.

Materials and methods

Patients and samples

Microarray and mass-spectrometry analysis

Biopsies of 30 liver metastases were used to identify genes and proteins correlating with early recurrence in stage IV CRC patients. Patient, tumour and surgical characteristics were derived from our prospectively collected database. Thirty frozen tumour biopsies were collected between July 2003 and August 2008 at the University Medical Centre in Utrecht, the Netherlands (Data set 1) (Figure 1). Samples were included if patients were aged >18 years and received curative resection for histologically confirmed liver metastases from CRC. Patients with a history of non-colorectal malignancies, extra hepatic disease or microscopic residual disease (R1) after surgery and patients who received local ablation therapy or chemoembolization alone or in combination with resection were excluded. Only those specimens were included that were snap-frozen in liquid nitrogen within 30 min after resection and stored at −80 °C. The study protocol was approved by The Medical Ethical Committee (MEC) of the University Medical Center Utrecht as recognised by article 16 of the WMO (Dutch Law on Medical Research with human subjects). Written informed consent was obtained from all patients in Data set 1.

Figure 1

Study workflow

Tissue microarray study

Between 1996 and 2005, 419 patients underwent surgical colon cancer resection at the Kennemer Gasthuis Hospital in Haarlem, the Netherlands, that were classified as stage II (T3–4, N0, M0) or stage III (T1–4, N1–2, M0) according to the fourth edition of the TNM classification system (Data set 2) (Figure 1). These tumour samples were used to examine any stage-specific role of the biomarkers retrieved from Data set 1. Patient, tumour and surgical characteristics were retrospectively drafted from clinical and pathology reports. Collection, storage and use of tissue and patient data were performed in agreement with the ‘Code for Proper Secondary Use of Human Tissue in The Netherlands' (available at http://www.federa.org/codes-conduct) (Belt ).

Gene expression profiling

RNA isolation, labelling and hybridisation to whole-genome oligonucleotide high-density microarrays were performed as previously described (Snoeren ). In short, two expression profiles in dye-swap experiments were generated for each sample. The samples were compared against a commercial reference (Universal Human Reference RNA catalogue #740 000, Stratagene, La Jolla, CA, USA). The Human Array-Ready Oligo set (version 2.0) was purchased from Qiagen (Venlo, Netherlands) and spotted on Codelink slides (GE Healthcare, Little Chalfont, UK) in a dust-filtered and humidity controlled clean room. The microarrays contained 70-mer oligonucleotides representing 21.329 human genes and expressed sequence tags (ESTs), as well as 3871 additional spots for control purposes. We applied total RNA and cRNA quality control criteria in accordance with the Tumour Analysis Best Practices Working Group (2004). Gene expression in metastases showing recurrence within 6 months after resection was compared with metastases that did not show recurrence for >2 years by using ANOVA (Wu ). In a fixed effect analysis, sample, array and dye effects were modelled. P-values were determined by a permutation F2 test in which residuals were shuffled 5000 times globally. Genes with P <0.05 after family-wise error correction were considered significant.

Mass-spectrometry analysis

Tissue homogenisation

Frozen tumour tissue sections of five stage IV patients with short time to recurrence (<6 months) and five patients with prolonged time to recurrence (>24 months) were used for mass-spectrometry analysis. All patients were selected from Data set 1. The frozen tumour tissue sections were cut into pieces of ∼20 mg, after which they were solubilized in 800 μl SDS sample buffer (containing 62.5 mm Tris-HCl, 2% w/v SDS, 10% v/v glycerol, 0.0025% bromphenol blue, 100 mm DTT, pH 6.8) using a micro grinder. Denaturation was done by heating samples for 10 min at 100 °C. All remaining insoluble parts were removed by centrifugation (15 min; 14 000 r.p.m.).

Fractionation using gel electrophoresis

Equal amounts of protein (50 μg) were separated on NuPAGE Novex Bis-Tris Mini Gels (Invitrogen, Bleiswijk, Netherlands). Gels were stained with Coomassie brilliant blue G-250 (Pierce, Rockford, IL, USA, Bleiswijk, Netherlands), washed and each lane was sliced into 10 bands using a band pattern to guide the slicing. The gel slicing and in-gel digestion was performed in a laminar flow under keratin-free conditions.

In-gel digestion, NanoLC-MS/MS analysis

In-gel digestion and NanoLC-MS/MS analysis was performed as described by van Houdt .

Protein identification

MS/MS spectra were searched against IPI human 3.62 (83947 entries) using Sequest version 27, rev 12 (Thermo, San Jose, CA, USA). Cysteine carboxamidomethylation and methionine oxidation were treated as variable modifications. Peptides precursor ions were searched with a maximum mass deviation of 10 p.p.m. and fragment ions with a maximum mass deviation of 1 Da. Sequest output files were imported in Scaffold 2.06.1 (Proteome software, Portland, OR, USA) and search results of the 10 gel bands per biological sample were combined. A protein was considered identified when at least two unique peptides were identified in one of the samples. Peptides were identified with a PeptideProphet (Keller ) probability score of >95% and a ProteinProphet (Nesvizhskii ) probability score of >99%. Proteins were (label-free) quantified by spectral counting (Liu ; Pham ), that is, the sum of all MS/MS spectra for each identified protein. For each biological sample the spectral counts for each protein were normalised on the sum of the counts for that sample and multiplied by the average of the sums across samples. Subsequently, the beta-binominal test was applied to detect significantly different proteins between early and late recurrence samples.

Tissue microarray construction

Tissue microarrays (TMAs) were constructed from stage II and III colon cancer samples as described previously (Simon ; Belt ). In summary, formalin-fixed, paraffin-embedded specimens of resected colon cancer tumours were used as donor blocks. Three 0.6-mm cores were taken from the centre of the tumour and three cores from the periphery of the tumour and transferred into recipient TMA paraffin blocks, resulting in six cores per tumour. The maximum number of samples that was transferred to a single TMA was 264.

Immunohistochemistry

Formalin-fixed paraffin-embedded cores were deparaffinized with xylene and rehydrated in decreasing ethanol dilutions. Endogenous peroxidase activity was blocked with 3% hydrogen peroxide. Antigen retrieval was achieved by boiling slides in citrate buffer (pH 6) for 20 min. Slides were then incubated (overnight 4 °C) with the monoclonal anti-Maspin antibody (clone G167-70; Pharmingen, San Jose, CA, USA) at a 1 : 1000 dilution. For detection, goat anti mouse poly-HRP (Powervision, Immunologic, Immunovision Technologies, Brisbane, CA, USA) was used. All slides were developed with diaminobenzidine (DAB). Slides were counterstained with filtered hematoxylin, dehydrated through a graded series of ethanol, immersed in xylene and mounted.

Immunoreactivity for Maspin was assessed and scored by two independent investigators who were blinded to clinical, pathological or survival data. The presence of nuclear and cytoplasmic staining in the tumour cells was assessed separately in each sample and divided into high and low scores. The immunohistochemistry scoring system used is based on scoring systems used in previous studies (Bettstetter ; Dietmaier ). A combination of intensity (negative, 1+, 2+ or 3+) and percentage of stained cells was used to score each core biopsy. Intensity score of 2+ in >50% of cells or 3+ in >10% of cells was considered high (Figure 4). In case of discrepancies between the scoring by the two investigators, the slides were reviewed and a consensus was reached. For analytical purposes protein expression was divided into high and low scores combining cytoplasmic and nuclear scores (see Table 1).

Table 1

Immunohistochemical evaluation of nuclear and cytoplasmic staining in central and peripheral tumour cores

Score	Intensity	Percentage of cells staining
High	3+	>10%
	2+	>50%
Low	3+	<10%
	2+	<50%
	1+	All
	Negative	—
Combined scoring groups
Cytoplasmic low	AND	Nuclear low
Cytoplasmic low	AND	Nuclear high
Cytoplasmic high	AND	Nuclear low
Cytoplasmic high	AND	Nuclear high

Western blotting

The homogenised tumour tissue from all five patients with short time to recurrence (<6 months), which was used for mass-spectrometry analysis, was used for western blot validation. Equal amounts of protein was run out on NuPAGE Novex Tris-Acetate Mini Gel (Invitrogen) and was analysed by western blotting using primary antibodies directed against Maspin (clone G167-70; Pharmingen) and β-Actin (AC-15, Novus Biologicals, Littleton, CO, USA) in combination with secondary antibody peroxidase-conjugated anti-mouse IgG (Dako, Glostrup, Denmark). For detection of antibodies Amersham ECL Western Blotting Detection Reagent was used (GE Healthcare Life Sciences).

Follow-up and survival

All patients were subjected to routine follow-up. The follow-up data were updated by letters and telephone calls to referring physicians and general practitioners. The duration of the follow-up and the time between surgery and the detection of recurrence were obtained, in addition to CRC-specific survival.

Statistical analysis

Time-to-recurrence and CRC-specific survival data were calculated from the day of surgery to the day of the first recurrence or the day of death caused by CRC. Median time to recurrence and CRC-specific survival were estimated by the Kaplan–Meier method. To determine the influence of possible risk factors on time to recurrence and CRC-specific survival, a univariable Cox regression analysis was performed. A multivariable Cox proportional hazards model, containing the factors that displayed P-values <0.2 in univariable analysis, was used to determine the independent prognostic impact of all variables on time to recurrence and CRC-specific survival. Statistical significance was assumed for P-values <0.05. Statistical analyses were performed using SPSS for Windows version 15.0 (SPSS, Chicago, IL, USA).

Results

Gene expression profiling identifies Maspin as a marker for recurrence in stage IV CRC

Forty-six patients from Data set 1 fulfilled the inclusion- and quality control criteria. To identify genes that are differentially expressed, patients with metastases showing early (<6 months, n=17) and prolonged (>24 months, n=13) time to recurrence were selected (Table 2). Gene expression profiling of these 30 tumours revealed that 14 genes were significantly differentially expressed between the two groups: SERPINB5, THEM2, GDF15, LYPLA1, AMPD1, ARL6IP5, LAPTM4A, C13ORF3, COLEC11, FYTTD1, BNIP3, SMYD2, GTF3c3 AND ASAP2. Of these genes, 11 were upregulated in the group with time to recurrence of <6 months (Table 3). SERPINB5, also known as Maspin, was the gene most upregulated in patients with a short time to recurrence. Maspin expression was ∼2.1 times higher in tumours of patients with a short time to recurrence vs tumours in patients with a prolonged time to recurrence (Figure 2A; P=0.01).

Table 2

Clinical and pathological characteristics of patients included in microarray analysis

Variable	Characteristics	DFS <6 months (n=17)	DFS >24 months (n=13)	P-value
Sex
	Male	10 (58.8%)	7 (53.8%)	0.785
	Female	7 (41.2%)	6 (46.2%)	—
Age (years)
	(Mean; median; s.d.)	62.24; 65.00; 14.70	63.84; 64.00; 8.09	0.715
Tumour location primary
	Right-sided	5 (29.4%)	6 (46.2%)	0.636
	Left-sided	6 (35.5%)	3 (23.1%)	—
	Rectum	6 (35.5%)	4 (30.8%)	—
Histological grade
	Well	3 (17.6%)		—
	Moderate	12 (70.6%)	12 (92.3%)	0.714
	Poor	2 (11.8%)	1 (7.7%)	—
	Unknown	—	—	—
Nodal stage
	N0	8 (47.1%)	7 (53.8%)	0.713
	N1	9 (52.9%)	6 (46.2%)	—
	Unknown	—	—	—
Interval between primary and LM
	Synchronous	8 (47.1%)	3 (23.1%)	0.184
	Metachronous	9 (52.9%)	10 (76.9%)	—
Neoadjuvant chemotherapy
	Yes	5 (29.4%)	2 (15.4%)	0.375
	No	12 (70.6%)	11 (84.6%)	—
Adjuvant chemotherapy
	Yes	16 (94.1%)	6 (53.8%)	0.025
	No	1 (5.9%)	7 (46.2%)	—
No. of metastases
	(Mean; median; s.d.)	2.41; 2.00; 1.77	1.54; 1.00; 0.88	0.131
Size liver metastases
	(Mean; median; s.d.)	5.18; 4.21; 3.37	4.25; 3.70; 2.52	0.408
Preoperative CEA
	(Mean; median; s.d.)	111.46; 54.28; 134.77	37.26; 33.30; 30.81	0.125

Bold value denotes significant P-value (P<0.05).

Table 3

Differentially expressed genes found by microarray analysis in early (<6 months) vs late (>24 months) recurrence groups

Upregulated genes in early recurrence group
Gene	Description	Fold change	P-value
SERPINB5	Serpin B5	2.1	0.0118
COLEC11	Collectin-11 Precursor	1.4	0.0018
LAPTM4A	Lysosomal-associated transmembrane protein 4A	1.3	0.0000
C13orf3	Spindle and kinetochore-associated protein RAMA1	1.3	0.0016
LYPLAL1	Lysophospholipase-like protein 1	1.3	0.0000
GTF3C3	General transcription factor 3C polypeptide 3	1.3	0.0204
AMPD1	AMP deaminase 1	1.3	0.0000
ARL6IP5	PRA1 family protein 3	1.3	0.0000
SMYD2	SET and MYND domain-containing protein 2	1.2	0.0118
ASAP2	Arf-GAP with SH3 domain, ANK repeat and PH domain-containing protein 2	1.2	0.0356
FYTTD1	Forty-two-three domain-containing protein 1	1.2	0.0026
Downregulated genes in early recurrence group
Gene	Description	Fold change	P-value
THEM2	Thioesterase superfamily member 2	0.8	0.0000
BNIP3	BCL2/adenovirus E1B 19 kDa protein-interacting protein 3	0.7	0.0096
GDF15	Growth/differentiation factor 15 Precursor	0.6	0.0000

Figure 2

Microarray analysis of patients with short and prolonged time to recurrence identifies Maspin to be differentially expressed in stage IV CRC. (A) Median expression levels of Maspin in the early (<6 months, n=17) and prolonged (>24 months, n=13) time-to-recurrence groups (P=0.012). (B) Kaplan–Meier curves of 46 patients in Data set 1, illustrating time to recurrence in stage IV CRC liver metastasis patients with high Maspin expression (>median) and low Maspin expression (median, n=27) and low Maspin expression (median) and low Maspin expression (

The Kaplan–Meier survival curves of all 46 patients from Data set 1 (including the 30 patients used for gene expression profiling) (Figure 2B) show that the 2-year recurrence-free survival probability of patients with Maspin-high tumours (>median (168.6) is an estimated 0.10 (95% CI=0.00–0.20) compared with 0.39 (95% CI=0.18–0.60) in patients with Maspin-low tumours (Maspin is an independent predictor of early recurrence (P=0.02, HR=2.971, 95% CI=1.168–7558) in our training set. Adjuvant treatment was the only other independent predictor of early recurrence in multivariable analysis (P=0.05, HR=0.329, 95% CI=0.109–0.994) (Supplementary Table 1).

This result was validated using an independent microarray data set from a cohort of 53 patients with stage IV primary CRC (Smith ). Two Kaplan–Meier graphs, generated by the R2 microarray analysis and visualisation platform (http://r2.amc.nl) show that high levels (>median (259.6)) of Maspin are significantly correlated with disease-specific (Figure 2C; P=0.005) and overall survival (Figure 2D; P=0.029) in primary stage IV CRC patients.

Mass-spectrometry analysis confirms Maspin as marker for early recurrence in stage IV CRC

In parallel to gene expression profiling, we also analysed differentially expressed proteins in patients, from Data set 1, showing early (<6 months) and prolonged (>24 months) time to recurrence. To this end, mass-spectrometry analysis was performed on five tumours from each group (Supplementary Table 2). Protein-containing lysates of these tumour tissue samples were fractionated on an SDS–PAGE gel, followed by in-gel tryptic digestion (Figure 3A). Analysis of the extracted peptides was performed by Nano-LC-MS/MS, followed by database searching. In total, 2097 unique proteins were identified in all patient samples. Forty-six proteins were present in significantly different amounts in the two groups, of which 26 were overrepresented and 20 were underrepresented in the tumours of patients with early recurrence (Table 4). The sets of 14 genes and 46 proteins that showed a significant association with short or prolonged recurrence times contained only 1 overlapping factor: Maspin (Figure 3B). Validation of the mass-spectrometry results by western blotting showed that the number of spectral counts detected by mass-spectrometry analysis correlates very well with Maspin protein levels (Figure 3C). These results demonstrate that Maspin is differentially expressed on both mRNA and protein level in stage IV CRC patients with early and late recurrence.

Figure 3

Mass-spectrometry based analysis of proteome differences between patients with early and prolonged time to recurrence in stage IV CRC. (A) Coomassie-stained protein gradient gel loaded with protein samples from tumours form patients with short (<6 months) and prolonged (>24 months) time to recurrence. This gel was used for mass-spectrometry analysis. (B) Venn diagram showing overlap between differently expressed genes and proteins found by micro-array and mass-spectrometry analysis. (C) Western blot analysis of Maspin levels in patients with early time to recurrence (<6 months) and the corresponding amount of spectral counts detected by mass-spectrometry analysis.

Table 4

Differentially expressed proteins (P<0.05) found by mass-spectrometry analysis in early (<6 months) vs late recurrence (>24 months) groups

Upregulated proteins in early recurrence group
Protein	Description	Fold change	P-value
SERPINB5	Serpin peptidase inhibitor, clade B (ovalbumin), member 5	∞	0.01
RNASET2	Ribonuclease T2	∞	0.01
ICAM1	Intercellular adhesion molecule 1	∞	0.02
EIF2B1	Eukaryotic translation initiation factor 2B, subunit 1 alpha, 26 kDa	∞	0.02
TPSAB1	Tryptase alpha/beta 1	∞	0.02
SF3A3	Splicing factor 3a, subunit 3, 60 kDa	∞	0.05
OLFM4	Olfactomedin 4	31.9	0.05
IGFALS	Insulin-like growth factor binding protein, acid labile subunit	8.8	0.02
PTK7	PTK7 protein tyrosine kinase 7	8.3	0.02
LANCL1	LanC lantibiotic synthetase component C-like 1 (bacterial)	7.8	0.01
API5	Apoptosis inhibitor 5	5.8	0.05
XPOT	Exportin, tRNA (nuclear export receptor for tRNAs)	5.8	0.05
USP7	Ubiquitin specific peptidase 7 (herpes virus-associated)	5.3	0.04
ACSL3	Acyl-CoA synthetase long-chain family member 3	4.4	0.01
NUP155	Nucleoporin 155 kDa	4	0.01
CLIC4	Chloride intracellular channel 4	3.9	0.01
KRT6B	Keratin 6B	3.6	0.04
MYL9	Myosin, light chain 9, regulatory	3.3	0.03
VARS	Valyl-tRNA synthetase	3.2	0
CYFIP1	Cytoplasmic FMR1 interacting protein 1	3.1	0.03
SYNE2	Spectrin repeat containing, nuclear envelope 2	2.8	0.05
NUDT21	Nudix (nucleoside diphosphate linked moiety X)-type motif 21	2.6	0.05
CKAP4	Cytoskeleton-associated protein 4	2.3	0.02
VIL1	Villin 1	2.2	0.04
HSD17B12	Hydroxysteroid (17-beta) dehydrogenase 12	2	0.05
YWHAQ	Tyrosine 3-monooxygenase	2	0.04
Downregulated proteins in early recurrence group
Protein	Description	Fold change	P-value
PTTG1IP	Pituitary tumour-transforming 1 interacting protein	∞	0.011
DUT	Deoxyuridine triphosphatase	∞	0.013
CASP1	Caspase 1, apoptosis-related cysteine peptidase	∞	0.018
HSD17B4	Hydroxysteroid (17-beta) dehydrogenase 4	∞	0.024
VWF	von Willebrand factor	∞	0.024
RBM14	RNA binding motif protein 14	∞	0.026
LACTB2	lactamase, beta 2	∞	0.028
AMBP	Alpha-1-microglobulin/bikunin precursor	∞	0.042
ANKRD22	Ankyrin repeat domain 22	∞	0.046
ADCK3	AarF domain containing kinase 3	13.8	0.001
GYG1	Glycogenin 1	9.0	0.003
AKR1C3	Aldo-keto reductase family 1, member C3	6.2	0.053
PTGES2	Prostaglandin E synthase 2	5.4	0.053
SCCPDH	Saccharopine dehydrogenase (putative)	2.8	0.007
MRPL14	Mitochondrial ribosomal protein L14	2.7	0.011
CUL4A	Cullin 4A	2.7	0.051
SDCBP	Syndecan binding protein (syntenin)	2.6	0.029
CAST	Calpastatin	2.1	0.019
MCCC2	Methylcrotonoyl-CoA carboxylase 2 (beta)	1.9	0.004
HMCN1	Hemicentin 1	1.7	0.027

Tissue microarray analysis reveals that Maspin staining in the tumour centre is a stage-specific marker in CRC

The above results indicate that Maspin could be a prognostic marker for early recurrence in stage IV CRC patients. This prompted us to assess whether this marker had a similar prognostic power in stage II and III CRC patients. To this end, a total of 243 stage II and 166 stage III tumours on TMAs containing FFPE tissue cores were stained with specific monoclonal Maspin antibody (Bettstetter ; Dietmaier ; Fung ). The presence of nuclear and cytoplasmic staining in both central tumour cores and peripheral tumour cores was assessed in each sample and divided in high and low scores. A high score was defined as an intensity score of 2+ in >50% of cells or 3+ in >10% of cells (Table 1, Figure 4).

Figure 4

Representative immunohistochemical Maspin staining in CRC specimens. (A) Cytoplasmic immunoreactivity with Maspin. Upper panel representing low scores: negative (i) intensity 1+ in >50% of cells (ii). Lower panel representing high scores: intensity 2+ (iii) and 3+ (iv) in >50% of cells. (B) Nuclear immunoreactivity of Maspin. Upper panel representing low scores: intensity 2+ in<50% of cells (i/ii). Lower panel representing high scores: intensity 3+ in>10% of cells (iii/iv).

In stage II CRC tumours (n=243), Maspin staining in the central tumour cores showed no significant differences in time-to-recurrence and CRC-specific survival times (Table 5). Kaplan–Meier curves of Maspin are depicted in Figure 5A and B. High Maspin expression was found to be correlated with a right-sided tumour location (P=0.001) (Supplementary Table 3).

Table 5

Uni- and multivariate analysis of factors influencing disease-free survival in stage II CRC patients

Univariate analysis
	Disease-free survival			Colorectal-specific survival
Variable	Hazard ratio	95%CI	P-value	Hazard ratio	95% CI	P-value
Sex	0.854	0.498–1.464	0.566	0.803	0.434–1.484	0.483
Age (years)	1.008	0.984–1.032	0.535	1.021	0.991–1.051	0.169
Tumour size (mm)	0.990	0.975–1.006	0.215	0.996	0.979–1.013	0.605
Tumour location (rightsided/left sided/rectum)	1.456	0.830–2.553	0.190	1.271	0.678–2.380	0.455
Nodal stage (N0/N1/N2)	n.a.			n.a.
No. of lymph nodes retrieved	0.960	0.907–1.015	0.152	0.970	0.911–1.033	0.343
Histological grade (well/moderate/poor)	1.925	1.056–3.511	0.033	1.820	0.923–3.591	0.084
Mucinous differentiation (present/absent)	1.152	0.616–2.155	0.657	1.243	0.623–2.481	0.537
Ulceration (present/absent)	1.003	0.536–1.876	0.992	0.998	0.489–2.036	0.995
Angioinvasion (present/absent)	2.301	1.156–4.581	0.018	2.322	1.072–5.028	0.033
Adjuvant chemotherapy (yes/no)	1.289	0.943–1.762	0.112	1.259	0.869–1.768	0.184
Microsatelite stability status (MSS/MSI)	0.685	0.458–1.025	0.066	0.791	0.514–1.215	0.284
Central tumour cores
Maspin staining
Low nuclear, low cytoplasmic	1.015	0.733–1.405	0.930	0.937	0.639–1.375	0.740
Low nuclear, high cytoplasmic	0.792	0.282–2.224	0.658	0.999	0.307–3.255	0.999
High nuclear, high cytoplasmic	—	—	—	—	—	—
Peripheral tumour cores
Maspin staining
Low nuclear, low cytoplasmic	0.987	0.694–1.404	0.943	0.717	0.440–1.168	0.201
Low nuclear, high cytoplasmic	1.291	0.359–4.643	0.696	2.074	0.396–10.859	0.388
High nuclear, high cytoplasmic

Bold value denotes significant P-value (P<0.05).

Figure 5

In stage III CRC high Maspin expression in central cores is correlated with early time to recurrence and colorectal-specific death. (A) Kaplan–Meier curves illustrating the effects of different Maspin expression levels on time to recurrence in stage II CRC patients (P=0.98, log-rank test). (B) Kaplan–Meier curves illustrating the effect of different Maspin expression levels on colorectal-specific survival in stage II CRC patients (P=0.95, log-rank test). (C) Kaplan–Meier curves illustrating the effects of different Maspin expression levels on time to recurrence in stage III CRC patients (P=0.011, log-rank test). (D) Kaplan–Meier curves illustrating the effect of different Maspin expression levels on colorectal specific survival in stage III CRC patients (P=0.000, log-rank test). *The low cytoplasmic and high nuclear group were left out of the analysis due to the low number of tumours in this group (n=4 in stage II, n=2 in stage III).

In stage III CRC patients (n=166), Maspin staining in the central tumour cores was an independent predictor of time-to-recurrence and CRC-specific survival (Table 6, Figure 5C and D). Patients with combined low cytoplasmic and nuclear staining score had estimated 2-year recurrence-free and CRC-specific survival probabilities of 0.74 (95% CI=0.64–0.84) and 0.83 (95% CI=0.74–0.92; log-rank P=0.011), respectively. Patients with a combined high cytoplasmic and nuclear score did markedly worse with estimated 2-year recurrence-free and CRC-specific survival probabilities of 0.42 (95% CI=0.25–0.59) and 0.55 (0.38–0.72; log-rank P=0.000), respectively. The low-cytoplasmic and high-nuclear group was left out of the analysis due to the low number of tumours in this group (n=4 in stage II, n=2 in stage III).

Table 6

Uni- and multivariate analysis of factors influencing disease-free survival in stage III CRC patients

Univariate analysis
	Disease-free survival			Colorectal-specific survival
Variable	Hazard ratio	95%CI	P-value	Hazard ratio	95%CI	P-value
Sex	1.054	0.665–1.671	0.821	1.418	0.854–2.354	0.177
Age (years)	0.999	0.979–1.018	0.890	1.011	0.988–1.035	0.366
Tumour size (mm)	1.015	1.002–1.029	0.021	1.015	1.000–1.031	0.048
Tumour location (rightsided/left sided/rectum)	0.808	0.512–1.275	0.360	0.691	0.416–1.148	0.153
Nodal stage (N0/N1/N2)	2.110	1.328–3.353	0.002	2.262	1.355–3.777	0.002
No. of lymph nodes retrieved	0.989	0.942–1.038	0.651	0.992	0.940–1.047	0.765
Histological grade (well/moderate/poor)	0.673	0.412–1.099	0.113	0.643	0.367–1.127	0.123
Mucinous differentiation (present/absent)	1.093	0.628–1.902	0.752	1.131	0.612–2.092	0.694
Ulceration (present/absent)	0.807	0.463–1.404	0.447	0.560	0.320–1.007	0.053
Angioinvasion (present/absent)	3.226	2.029–5.130	0.000	3.107	1.860–5.189	0.000
Adjuvant chemotherapy (yes/no)	1.127	0.914–1.390	0.264	0.950	0.739–1.221	0.689
Microsatelite stability status (MSS/MSI)	1.025	0.713–1.472	0.895	0.365	0.523–1.269	0.365
Central tumour cores
Maspin staining
Low nuclear, low cytoplasmic	1.088	0.814–1.454	0.568	1.175	0.841–1.641	0.344
Low nuclear, high cytoplasmic	2.150	1.159–3.987	0.009	2.541	1.297–4.977	0.007
High nuclear, high cytoplasmic	—	—	—	—	—	—
Peripheral tumour cores
Maspin staining
Low nuclear, low cytoplasmic	0.869	0.653–1.156	0.334	1.013	0.730–1.405	0.940
Low nuclear, high cytoplasmic	1.754	0.822–3.742	0.146	2.074	0.909–4.732	0.083
High nuclear, high cytoplasmic	—	—	—	—	—	—

Bold value denotes significant P-value (P<0.05).

High Maspin expression in the central tumour cores was also associated with poor histological grade, mucinous differentiation and MSI-high status in stage III CRC patients (Supplementary Table 4). In multivariable analysis, mucinous differentiation was the only independent factor associated with high Maspin expression.

Maspin staining in peripheral cores showed no correlation with time to recurrence or CRC-specific survival in stage II patients or stage III patients (Tables 5 and 6 and Supplementary Tables 5 and 6). The low-cytoplasmic and high-nuclear group was left out of the analysis due to the low number of tumours in this group (n=4 in stage II and n=2 in stage III).

We next validated these results using the microarray data from a cohort of 75 patients with stage II CRC and 78 patients with stage III CRC (Jorissen ; Smith ). Kaplan–Meier graphs, generated by the R2 microarray analysis and visualisation platform (http//r2.amc.nl), show that high levels of Maspin (>median 486.2) are significantly (P=0.02) correlated with early recurrence in stage III but not in stage II CRC patients (Figure 6).

Figure 6

External validation, using the (A) Kaplan–Meier curves illustrating that high (>median) and low (median) Maspin expression have significant shorter time to recurrence than patients with low (

Discussion

In this report we identified Maspin as a marker for early recurrence after surgery for colorectal liver metastases. Maspin is a prognostic factor for early recurrence in stage III and IV CRC patients but not in stage II patients. Maspin is a member of the serine protease inhibitor (serpin) family. In breast cancer, Maspin acts as a tumour suppressor by inhibiting tumour cell motility, invasion and tumour growth (Zou ). In breast, ovarian and lung cancer Maspin expression is correlated with a relatively good prognosis (Marioni ; Zheng ). By contrast, studies in lung, breast, gastric and pancreatic cancer show that Maspin expression is associated with more aggressive disease (Umekita ; Cao ; Woenckhaus ; Yu ). The reasons for these discrepancies are presently unknown.

Mixed-stage studies in CRC have failed to reach consensus whether Maspin is associated with good or poor prognosis (Song ; Bettstetter ; Markl ). A study in patients with only stage III disease has shown that high nuclear Maspin is significantly correlated with poor overall survival (Dietmaier ), which is in line with the results presented in this report. We used a combined nuclear and cytoplasmic expression score. The group with high-cytoplasmic and high-nuclear staining is associated with early recurrence. By contrast, the groups with low-nuclear and either high- or low-cytoplasmic staining are associated with late recurrence. This suggests that high-nuclear Maspin staining is the major variable determining the association with early recurrence. This is in line with earlier reports (Dietmaier ; Markl ). In contrast to the results presented here, a previous study failed to detect a significant association of nuclear or cytoplasmic Maspin staining with outcome in a large cohort of stage III CRC patients (Fung ). As all three studies made use of the same antibody, the different outcome of the study of Fung may be related to the use of a different scoring system. The cutoffs used in this study resulted in a very unevenly distributed number of samples per group, in which the vast majority (75%) of samples were allocated to one group. Moreover, this study did not take the intensity of the staining into account, which has previously been associated with the amount of Maspin expression and clinical outcome (Dietmaier ). However, when we apply the Fung scoring system to our own data set, we still find that nuclear Maspin expression is associated with disease-free and CRC-specific survival in stage III CRC patients (P=0.031 and P=0.001; data not shown). The same significant association between high levels of Maspin and early-recurrence and colorectal-specific survival was seen when applying the Dietmaier scoring system (P=0.040 and P=0.019; data not shown).

Tumours with high Maspin expression respond significantly better to adjuvant 5-FU chemotherapy in stage III CRC (Dietmaier ). This trend was also observed in our study, although patient numbers were too low for this association to reach significance (Data not shown). Maspin may therefore not only have value as a prognostic marker but possibly also as a marker for predicting response to adjuvant treatment in advanced stage CRC. Whether Maspin itself could be a drug target depends on whether it is causally involved in facilitating the metastatic process.

In this study, Maspin was highly upregulated using two different high-throughput screens suggestive of a key role in predicting time to recurrence after surgery of colorectalliver metastasis. To our knowledge, this is the first study reporting a possible predictive role for Maspin in stage IV CRC. However, it must be noted that identification of Maspin and validation was performed in a relatively small cohort.

Experimental studies show that high Maspin expression is correlated with increased apoptosis resistance in CRC cells (Payne ). Furthermore, it has been shown that circulating tumour cells in peripheral blood of CRC patients express high levels of Maspin (Findeisen ). We did not find a difference in Maspin expression at the periphery of the tumour compared with that at the centre; however, Bettstetter did find that Maspin expression levels were higher at the invasive tumour font when compared with that at the more central parts of CRC tumours. This discrepancy can probably be explained by the fact that we scored peripheral cores containing mainly tumour bulk, but not necessarily cells invading the adjacent stroma, whereas Bettstetter only scored Maspin expression in cell clusters detaching from the peripheral tumour bulk.

These results combined with the results of our study, demonstrating that high Maspin expression is associated with poor prognosis in CRC that has metastasised to local lymph nodes or beyond, point to a potential role for Maspin in facilitating the metastatic process. Future work should therefore address whether (and how) high Maspin levels promote metastasis in advanced-stage CRC.