AbstractPurpose: To use gene transcript profiling to identify cancer-associated gene expression.
Experimental Design: Methods included (i) marker discovery using laser capture microdissection
(LCM)-assisted specimen preparation and antisense RNA-long serial analysis of gene expression (aRNA-
LongSAGE) on matched colon cancer and uninvolved colon tissue specimens (n ¼ 5). Candidate tumor-
associated genes were selected by combining the LongSAGE libraries reported herein with our previous
colon cancer LCM-microarray transcript profiling data; (ii) marker selection and validation by quantitative
real-time PCR (n¼ 15) and immunohistochemistry (n¼ 31); and (iii) independent validation onmultiple
tissue microarray (n ¼ 203).
Results: Among 30 upregulated and 73 downregulated genes, upregulation of fermitin familymember 1
(FERMT1), adenosylhomocysteinase (AHCY), secernin 1 (SCRN1), and SAC3 domain-containing protein
1 (SAC3D1) expression and downregulation of IgJ andMALL expression in colon cancer were confirmed by
quantitative PCR. FERMT1 and AHCY protein expression was also upregulated in colon cancer compared
with uninvolved colon mucosa, and FERMT1 expression showed upregulation in colon adenoma. Patients
with moderate/strong tumor FERMT1 protein expression (n ¼ 122) showed significantly poorer overall
survival (OS; P¼ 0.011) and disease-free survival (DFS; P¼ 0.005) than patients with negative/weak tumor
FERMT1 protein expression (n ¼ 81). Multivariate Cox regression analysis showed that FERMT1 protein
expression was also an independent prognostic factor for OS (P¼ 0.018) andDFS (P¼ 0.009). In addition,
upregulated FERMT1 protein expression appeared to have some specificity among alimentary tract tumors.
Conclusions: FERMT1 is a novel prognostic factor for colon carcinoma. Clin Cancer Res; 17(9);
2908–18. �2011 AACR.
Introduction
Colorectal cancer is the fourth most common cancer inmen and the third most common cancer in women world-wide. In China and other economically transitioning coun-tries, colon cancer incidence rates have been rising rapidly,most likely owing to changes in lifestyle and nutritionalhabits (1). Laboratory investigations of differences in geneexpression between normal cells and their corresponding
carcinoma cells are crucial to understanding how onco-genes and tumor suppressor genes alter complex cellularfunctions and thus drive tumor progression (2). In coloncancer, a large number of disease biomarkers have beenassociated with clinical outcomes, including adenomatouspolyposis coli (APC), beta-catenin, DNA mismatch repairgenes, growth factors and their receptors, cell adhesionmolecules, and more recently microRNAs and colonepithelial stem cell markers (3–9). Despite this rapidlyaccumulating knowledge, there remains a need for bio-markers of disease progression that can be utilized in apreventive strategy to stratify patients into appropriatescreening, surveillance, treatment, and preventionprograms.
At present, cDNA microarrays and serial analysis ofgene expression (SAGE) are the most widely used tech-niques for determining gene expression levels and ratiosin different disease states and in cells under differentphysiologic conditions (10). cDNA microarrays are usedto measure the relative gene expression levels of thou-sands of known transcripts in different cell and tissuesamples (11–14). SAGE is based on the high-throughputsequencing of concatemers of short sequence tags thatoriginate from a known position within a transcript and
Authors' Affiliations: Departments of 1General Surgery and 2Pathology,Shanghai First People's Hospital, Shanghai Jiaotong University, Shanghai;and 3Department of Hepatobiliary Pancreatic Surgery, Shandong Qian-foshan Hospital, Shandong University, Jinan, China
Note: Supplementary data for this article are available at Clinical CancerResearch Online (http://clincancerres.aacrjournals.org/).
J. Fan, D. Yan, and M. Teng contributed equally to this work.
Corresponding Author: Zhihai Peng, Department of General Surgery,Shanghai First People's Hospital, Shanghai Jiaotong University, Shanghai200080, China. Phone: 86-13601996229; 86-21-63240090 ext. 3132;Fax: 86-21-63240825; E-mail: [email protected] or Huamei Tang,Department of Pathology, Shanghai First People's Hospital, ShanghaiJiaotong University, Shanghai 200080, China.
therefore theoretically contain sufficient information foridentification of a unique transcript. SAGE data andtechniques have been utilized by a number of investiga-tors to identify genes with differential regulation andpotential contributions to colon cancer progression(15–18). However, annotation of the short, 10-bpsequence tags may identify more than 1 transcript (19,20). This lack of specificity can be overcome by usingLongSAGE libraries that contain longer, 17-bp tags, thusgenerating more reliable mapping to Unigene clusters orthe complete genome sequence (21–23).Cellular heterogeneity within patient samples very often
presents a challenge to accurate tissue- or tumor-specificgene expression profiling. Gene expression measurementscan become skewed owing to tissue heterogeneity, which inturn significantly confounds attempts at statistical analyses.This challenge can be overcome via the use of laser capturemicrodissection (LCM) to obtain purified cell populationsfrom heterogeneous tissue, resulting in the derivation ofprecise information on the gene expression profiles ofdefined cell types (2, 24). Tissue microarrays (TMA) canthen be used to confirm gene expression in the clinicalcontext.In the present study, we first present a LCM-LongSAGE
protocol for investigating global gene expression profilingof parenchymal cells in colon carcinoma. This protocolallowed for the rapid, quantitative measurement of gen-ome-wide gene expression in specific cell populations.Differential regulation of a subset of genes was validatedby quantitative real-time PCR (qRT-PCR), immunohisto-chemistry, and clinical correlations with patient character-istics, tumor histopathology, and patient outcomes.
Materials and Methods
Construction of LongSAGE librariesUse of amplified antisense RNA (aRNA) for transcript
profile analysis of colon cancer cells and correspondinguninvolved colon epithelium by LCM-microarray wasdescribed previously (25); this was used for the generationof LongSAGE libraries. aRNA was adjusted to a finalvolume of 10 mL with diethylpyrocarbonate-treated H2O.A total of 2 mL of SAGE-random primer (50-NNNNNNCAT
G-30; Eurofins MWG Operon) and 1 mL of dNTP mix (10mmol/L; Promega) were added to the aRNA, followed by65�C incubation for 5 minutes and then frozen on dry ice.A total of 4 mL first-strand buffer, 1 mL 0.1 mol/L dithio-threitol, 1 mL RNaseOUT, and 1 mL SuperScript III RNaseH� reverse transcriptase were added to complete the reversetranscription reaction (Invitrogen). The reaction was incu-bated for 5, 60, and 15 minutes at 37�C, 50�C, and 70�C,respectively. RNase H� (2 U/mL; USB Corp.) was thenadded before a final incubation at 37�C for 20 minutes(2). LongSAGE libraries were then generated with an I-SAGE Long Kit according to themanufacturer’s instructions(Invitrogen).
A total of 2,422 concatemer clones for colon cancer cellsand 2,600 concatemer clones for uninvolved colon epithe-liumwere subjected to direct sequencing (ABI 3730 system;Applied Biosystems) according to the manufacturer’sinstructions. All sequence files were processed withSAGE2000 software. Tags were matched to the humanreliable tag database SAGEmap_Hs_NlaIII_17_best.gz(ftp://ftp.ncbi.nih.gov/pub/sage/mappings/).
Comparison of gene expression profiles between 2platforms
Differentially expressed genes from the microarray plat-form were screened according to the protocol in the Affy-metrix Statistical Algorithms Descriptions Document. Thepopulation of genes displaying overlapping differentialexpression as revealed by microarray and LongSAGE ana-lysis (P < 0.05) was selected with Microsoft Office Access2003. Expression trends of these overlapping genes werecompared between the 2 platforms with criteria for screen-ing candidate tumor-associated genes for differentialexpression of a value of P < 0.05 in the LongSAGE librariesand a 4-fold or more change in the cDNA microarray data.
Validation by quantitative PCRMatched colon cancer and uninvolved colon specimens
(N¼ 15) were collected, snap frozen, and stored at�80�C.Histopathology confirmed that each tumor sample con-tained at least 70% tumor cells and that the uninvolvedcolon samples contained more than 90% epithelial cells.Total RNA was extracted with an RNeasy kit (Qiagen), andsingle-stranded cDNAs were synthesized with a High Capa-city cDNA Reverse Transcription kit (Applied Biosystems).qPCRs were carried out with SYBR Green PCR Master Mix(Applied Biosystems) and Mastercycler ep realplex (Eppen-dorf). Primers used for qPCR are listed in SupplementaryTable S1. 18S rRNA was used as an internal control. EachqPCR for each gene was carried out in triplicate. Foldchanges in expression were calculated and transformedto log2 with the following calculations: ratio ¼ 2DDCt;log2 ratio ¼ DDCt.
Tissue microarray for validation of protein expressionfor select genes
A total of 203 radical colectomy specimens from 86 menand 117 women 22 to 95 years of age were obtained from
Translational Relevance
Upregulation of Fermitin familymember 1 (FERMT1)in colon carcinomas, compared with matched unin-volved epithelium was identified via laser capturemicrodissection–assisted specimen preparation andantisense RNA-long serial analysis of gene expression.High FERMT1 protein expression was found to beassociated with lymph node metastasis, late AJCC stage,poor differentiation, positive vascular invasion, andpoor patient survival. Therefore, FERMT1 may be usedas a prognostic biomarker for identifying the poorprognosis subset of patients with colon cancer.
LCM-LongSAGE Analysis of Colon Cancer Biomarkers
www.aacrjournals.org Clin Cancer Res; 17(9) May 1, 2011 2909
the Department of Surgery, Shanghai Jiao Tong UniversityAffiliated First People’s Hospital, and used for constructingthe TMA, which included 24 stage I, 81 stage II, and 80 stageIII, and 18 stage IV cancers; 99 well-differentiated, 74moderately differentiated, and 30 poorly differentiatedcancers; and 95 specimens with positive lymph nodemetastasis and 108 specimens with negative lymph nodemetastasis. All tumors were adenocarcinomas and weregraded according to World Health Organization criteria(26). Tumor staging was conducted according to the Amer-ican Joint Committee on Cancer (AJCC) sixth editioncancer staging system (27). All patients with stage III orIV disease were administered standard chemotherapeuticprotocols with 5-fluorouracil after operation at our institu-tion. Clinical data were collected via medical record review.The final patient follow-up date was June 29, 2008, and themedian observation time for survivors was 61 months(range, 9–89 months). All patients provided informedconsent according to a protocol approved by the Institu-tional Review Board of Shanghai First People’s Hospital.None of the patients received any treatment for coloncancer prior to surgery. From each colectomy specimen,2 cores each of primary cancer and uninvolved tissueadjacent to the tumor at a distance of at least 2 mm fromthe tumor were arrayed. TMAs were created using a TissueMicroarrayer (Beecher Instruments).
Additional TMAs were purchased from Outdo Biotech(Shanghai).The standard tissue section OD-CT-DgCol03-001 TMA included 31-matched colon carcinoma and unin-volved colon tissues, and the OD-CT-DgCol01-001 TMAincluded 22 colon carcinoma and 20 colon adenomaspecimens. The OD-CT-Dg03-001 TMA included 5 pairsof cancer tissue and corresponding uninvolved tissue fromliver carcinoma, esophageal carcinoma, gastric carcinoma,colon carcinoma, and rectal carcinoma, and 6-paired sam-ples of pancreatic carcinoma and corresponding unin-volved pancreas tissue. A flow chart of the screeningstrategy is presented in Supplementary Figure S1.
ImmunohistochemistryFermitin family member 1 (FERMT1) protein expression
was detected on the OD-CT-DgCol03-001, OD-CT-DgCol01-001, and OD-CT-Dg03-001 TMAs and our 203-case colon cancer TMA with a polyclonal rabbit antibody(U1610-01; United States Biological; diluted 1:1,000). Thisantibody showed a single protein band by Western blottingat this dilution (Supplementary Fig. S2). Adenosylhomocys-teinase (AHCY) protein expression was detected on OD-CT-DgCol03-001 with a polyclonal rabbit antibody (10757-2-AP; ProteintechGroup; diluted1:100). A goat anti-rabbit IgG(Dako) was used as a secondary antibody. Tissue sectionswere counterstained withMayer hematoxylin. Positive stain-ing was scored by 2 independent investigators withoutknowledgeofpatientoutcomes [double-blinded;H.T. (coau-thor) and H.H. (Department of Pathology, Shanghai FirstPeople’s Hospital)] and then classified into 4 groups: nega-tive, weakly positive, moderately positive, and strongly posi-tive according to percentage of positive cells and staining
intensity. A mean value of the 2 independent scores wascalculated for each sample. When different values werereported for a given sample, the samples were re-evaluated,and if they were still different, the investigators discusseduntil a unanimous agreement was reached.
Statistical analysisContinuous variables are presented as the mean � SD.
Categorical variables are expressed as the number (n) andpercentage and were compared with the Fisher exact test(28). FERMT1 and AHCY staining for the 31 matcheduninvolved and carcinoma tissues were compared withthe signed-rank test (29). Overall survival (OS) and dis-ease-free survival (DFS) curves were calculated with theKaplan–Meier method with the log-rank test (30). Toinvestigate independent risk factors for death and lymphnode metastasis, Cox proportional hazard models wereused (31) and expressed as the HR with 95% CIs. Signifi-cant factors in the univariate Cox proportional hazardmodels were selected for the final multivariate Cox propor-tional hazard model with the forward conditional method(32). All statistical analyses were set with a significancelevel of 0.05 and were carried out with SPSS 15.0 statisticalsoftware (SPSS Inc.).
Results
LongSAGE data librariesPure cell populations were collected from 203 paired
colon carcinoma and uninvolved mucosa specimens byLCM and were used to create a LongSAGE library. A total of100,666 tags were generated including 45,560 unique tags(Supplementary Table S2). The duplication ditag propor-tions were 6.47 in the uninvolved colon LongSAGE libraryand 6.15 in the colon carcinoma LongSAGE library. Thesetags were divided into 4 groups according to their abun-dance in the libraries. The 848 transcripts with a value P <0.05 were screened for differential gene expression betweenuninvolved colon and tumor tissues.
When the results from LCM-LongSAGE and TMAs werecompared, 404 genes showed overlapping differential tran-scripts in the 2 platforms. Among these transcripts, theexpression trend of 388 transcripts representing 326unique genes was identical between the 2 platforms. Atotal of 30 upregulated genes and 73 downregulated genesmet our criteria for screening candidate tumor-associatedgenes. The top 20 upregulated and downregulated genesare presented in Supplementary Tables S3 and S4, respec-tively. Eight of these upregulated genes, including FERMT1and AHCY (33–37), have been previously reported as beingupregulated in colon cancer, whereas 12 represented newlyidentified upregulated genes. Likewise, differential regula-tion of 8 of the downregulated genes has been previouslyreported in colon cancer.
qPCR for validation of altered gene expressionqPCRwas carried out to validate the expression of 6 differ-
entially expressed genes, 4 with significantly upregulated
Fan et al.
Clin Cancer Res; 17(9) May 1, 2011 Clinical Cancer Research2910
expressionand2withsignificantlydownregulatedexpression.Our rationale for focusing on these genes was based on folddifference in expression, the potential functionof these genesintumordevelopment,andthefactthattherehavebeenfeworno detailed studies on these genes. Upregulation of FREMT1,AHCY, secernin 1 (SCRN1), and SAC3 domain-containingprotein 1 (SAC3D1) expression and downregulation ofimmunoglobulinJ(IgJ)andMAL-likeprotein(MALL)expres-sion in colon carcinomas were confirmed by qPCR (Supple-mentary Fig. S3).
Immunohistochemical staining of matched cancer anduninvolved colon tissues for AHCY and FERMT1protein expressionAmong FERMT1, AHCY, SCRN1, and SAC3D1 there
were no commercially available antibodies for SAC3D1,and antibody for SCRN1 resulted in ambiguous immuno-
histochemical staining. Antibodies for FERMT1 and AHCYwere found to be useful for titration and staining, butpreliminary results indicated that AHCY protein expressionwas not a significant prognostic factor for tumor outcome.Therefore, we focused on FERMT1. For further confirma-tion of altered gene expression leading to changes inprotein levels, AHCY and FERMT1 protein expressionwas detected on a commercial TMA containing 31matchedcolon carcinoma and uninvolved colon tissues. Semiquan-titative immunohistochemical results are presented inFigure 1A. Significantly stronger AHCY and FERMT1 stain-ing was observed in carcinomas than in uninvolved colontissue (P < 0.001). Among the 31 matched tissues, 30(96.8%) showed stronger FERMT1 staining in carcinomatissues than in uninvolved colon tissues, and no matchedtissue showed stronger FERMT1 staining in uninvolvedcolon tissues than in carcinoma tissues. Similar results
Figure 1. Immunohistochemicalvalidation of AHCY and FERMT1protein expression in multipleTMAs. A, immunohistochemicalstaining of 31 matched coloncarcinoma and uninvolved colontissue specimens was carried outwith specific antibodies againstAHCY and FERMT1. AHCY andFERMT1 protein expressionwas confirmed to be upregulatedin colon cancer comparedwith uninvolved colonmucosa (P <0.001). B,immunohistochemical staining of22 colon carcinomas and 20 colonadenomas was carried out with aFERMT1-specific antibody.FERMT1 protein expression wasupregulated in colon cancercompared with colon adenoma(P ¼ 0.0002). Representativephotomicrographs of FERMT1protein expression in colon cancer(C) and uninvolved colon mucosa(D). Representativephotomicrographs of AHCYprotein expression in colon cancer(E) and uninvolved colon mucosa(F). Representativephotomicrographs of FERMT1protein expression in liver cancer(G) and uninvolved liver tissue (H).Representative photomicrographsof FERMT1 protein expression inpancreatic cancer (I) anduninvolved pancreas tissue (J).Representative photomicrographsof FERMT1 protein expression inesophageal cancer (K) anduninvolved esophageal tissue (L).Representative photomicrographsof FERMT1 protein expression ingastric cancer (M) and uninvolvedgastric tissue (N).
were found for AHCY staining, with 24 (77.4%) matchedtissues showing stronger AHCY staining in carcinomatissues than in uninvolved colon tissues, and again nomatched uninvolved tissues showed stronger AHCY stain-ing than in respective carcinoma tissues (Table 1).
FERMT1 protein expression detected in multiple TMAsDifferential expression of FERMT1 in colon cancer has
been shown at the mRNA level (33), but changes inFERMT1 protein expression in colon cancer have notbeen reported. To explore changes in FERMT1 proteinexpression during the progression of colon cancer, FERMT1protein was detected on a TMA that included 22 coloncarcinoma and 20 colon adenoma specimens. Strong ormoderate FERMT1 protein–positive staining was detectedin 72.7% (16/22) of the colon carcinomas and 15% (3/20)of the colon adenomas (P ¼ 0.0002, Fig. 1B). Thus,FERMT1 may be progressively upregulated during coloncancer development.
Immunohistochemical detection of FERMT1 proteinexpression was also done on a TMA comprising multipletypes of alimentary tumors (Fig. 1C–N). All esophagealcarcinomas and corresponding uninvolved esophagealtissues showed weak/negative FERMT1 protein expres-sion. Likewise, similar moderate FERMT1 protein expres-sion was detected in 2 pairs of matched liver carcinomaand uninvolved liver tissue, but FERMT1 protein wasdownregulated in the other 3 cases of liver carcinoma.FERMT1 protein was also downregulated in 3 cases ofpancreatic carcinoma and had similar expression betweencarcinoma and uninvolved tissue in the other 3 cases.Weak FERMT1 protein staining was detected in all of theuninvolved gastric tissue and 2 gastric carcinoma speci-mens, whereas 3 gastric carcinoma tissues showed mod-erate FERMT1 protein staining. In contrast, FERMT1
protein was upregulated in all of the colon and rectalcarcinomas on the TMA. Thus, upregulated FERMT1 pro-tein expression seemed to support the tissue-specificnature of FERMT1 expression.
Clinicopathologic significance of FERMT1 and AHCYprotein expression in colon cancer
The clinicopathologic significance of FERMT1 and AHCYis summarized in Table 2. Overexpression of FERMT1 wassignificantly associated with lymph node metastasis (P ¼0.006), AJCC stage (P¼ 0.012), tumor differentiation (P ¼0.036), and vascular invasion (P ¼ 0.049). Interestingly,AJCC stage III and moderately differentiated tumorsshowed the highest FERMT1 expression. Increased AHCYprotein expression was also associated with advanced AJCCstage (P ¼ 0.021), but unlike FERMT1, AHCY was asso-ciated with advanced tumor T stage (P ¼ 0.002) andmetastasis (P ¼ 0.004).
Patient overall and disease-free survivalGiven that 8 patients with stage IV disease underwent
noncurative surgery, a total of 195 patients were includedin the following survival analyses to avoid the possibleconfounding effects of unresectable metastatic tumors.Results of the OS analyses are presented in Figure 2 andTable 3. Patients with negative/weak and moderate/strongtumor FERMT1 protein expression had a 5-year survivalrate of 77.6% and 60.8%, respectively, with significantlyworse OS for patients with moderate/strong FERMT1expression (P ¼ 0.011). However, there was no significantdifference between the 2 AHCY protein expression groupsin OS.
T stage, AJCC stage, differentiation, and FERMT1 expres-sion showed significant effects on OS in multivariate Coxmodels and were included into the final multivariate Cox
Table 1. FERMT1 and AHCY immunohistochemical staining for protein expression in matched coloncancer and uninvolved colon tissues
proportional hazard model. As seen in Table 3, the risk ofdeath for patients with T3 stage disease was significantlylower than for patients with T4 stage disease [HR (95%CI):0.38 (0.21–0.67)]. The risk of death for those with AJCCstage IV was significantly higher than those with AJCC stageI [HR (95% CI): 14.68 (1.33–161.83)], and the risk ofdeath for those with moderate or poor tumor differentia-tion was significantly higher than for patients with well-differentiated tumors [HR (95% CI): 1.81 (1.01–3.24) formoderate differentiation and 3.05 (1.52–6.14) forpoor differentiation]. Finally, the risk of death for patientswith moderate/strong tumor FERMT1 expression wassignificantly higher than for patients with negative/weakFERMT1 expression [HR (95% CI): 1.92 (1.12–3.30)].
The results of the DFS analyses are presented in Figure 3and Table 3. Patients with negative/weak and moderate/strong tumor FERMT1 protein expression had 5-year DFSrates of 69.8% and 54.7%, respectively. Kaplan–Meiersurvival curves showed that patients with moderate/strongtumor FERMT1 protein expression had significantly worseDFS than patients with negative/weak tumor FERMT1protein expression (P ¼ 0.013).
Similar to the Cox models for OS, T stage, AJCC stage,differentiation, and FERMT1 expression showed indepen-dent effects on the risk of disease and were included intothe final multivariate Cox proportional hazard model. Asseen from Table 3, the risk of recurrent disease for patientswith T3 stage tumors was significantly lower than for
Table 2. Association between clinicopathologic features and FERMT1 or AHCY protein expression
patients with T4 stage tumors [HR (95% CI): 0.42 (0.24–0.71)]. The risk of recurrent disease for patients with AJCCstage IV tumors was significantly higher than for patientswith AJCC stage I tumors [HR (95% CI): 9.06 (1.01–81.34)]; the risk of disease for patients with moderate orpoor tumor differentiation was significantly higher than forpatients with well-differentiated tumors [HR (95% CI):1.87 (1.11–3.16) for moderate differentiation and 2.56(1.27–5.17) for poor differentiation]. Finally, the risk ofrecurrent disease for patients with moderate/strongFERMT1 expression were significantly higher than forpatients with negative/weak FERMT1 expression [HR(95% CI): 1.99 (1.19–3.35)].
Discussion
In the present study, LCM-assisted tissue specimen pre-paration was combined with cDNA microarray and Long-SAGE analysis for the rapid identification and validation oftumor-associated genes in colon cancer. Combined ana-lyses revealed significant upregulation of 30 genes and
downregulation of 73 genes in colon carcinoma. Upregu-lation of FERMT1, AHCY, SCRN1, and SAC3D1 expressionand downregulation of IgJ and MALL expression in coloncancer were confirmed by RT-PCR. Upregulated FERMT1protein expression in colon carcinoma was associated withsignificantly poorer DFS and OS and was confirmed as anindependent prognostic factor for DFS.
The FERMT1 protein is expressed predominantly in skin,intestine, and kidney. It localizes to cell junctions andregulates integrin function to facilitate the linkage of celladhesion structures to the actin cytoskeleton (38). As acrucial connector between cytoskeletal structures and theextracellular matrix, FERMT1 may also be involved in theassembly and stabilization of actin filaments and likelyplays a role in modulating cell adhesion, morphology, andmotility (38). FERMT1mutations causes Kindler syndromein humans, an autosomal recessive form of genodermatosis(39). Interestingly, FERMT1 may also have roles in neo-natal intestinal development andmaintenance of epithelialintestinal barrier function (40).
FERMT1 has also been suggested to have a role in humancancer following a report of significant upregulation ofFERMT1 mRNA in lung (60-fold) and colon (6-fold) car-cinomas as identified by microarray analysis and con-firmed by qRT-PCR (33). In that report, FERMT1 wasoverexpressed in 70% of the colon carcinomas and 60%of the lung carcinomas tested. In the present study, 66.7%of colon carcinomas showed an increase in FERMT1 expres-sion by 3- to 28-fold, with a mean fold increase of 7.9-fold.
TGF-b1 contributes to tumor invasion and cancer pro-gression by increasing the motility of tumor cells (41–43).Interestingly, FERMT1 is regulated by TGF-b1, and a TGF-b1–induced increase in FERMT1 expression results inincreased cell spreading correlated with epithelial tomesenchymal cell transition, an important step in carci-nogenesis (44). It has also been reported that FERMT1protein is predictive of breast cancer lung metastasis (45).In the present study, the majority of uninvolved colontissues and colon adenoma tissues were negative forFERMT1 expression whereas moderate or strong FERMT1protein expression was detected in the majority of coloncarcinomas. High FERMT1 protein expression was alsoassociated with aggressive tumor phenotypes, includingpositive lymph node metastasis, late AJCC stage, poordifferentiation, positive vascular invasion, and poor patientsurvival. Interestingly, upregulated FERMT1 protein expres-sion showed some specificity amongst alimentary tracttumors and may be a useful prognostic biomarker in coloncancer. Although a definitive mechanistic role for FERMT1in colon cancer progression will need to be confirmed byadditional experimentation with techniques such as RNAinterference inmodel systems, existing data suggest a strongbiological basis for the association reported herein.
AHCY catalyzes the reversible hydrolysis of S-adenosyl-homocysteine to adenosine and L-homocysteine. Thus, itregulates the intracellular S-adenosylhomocysteine concen-tration, which is thought to be important for transmethyla-tion reactions (GeneCards v.3. http://www.genecards.org/
0.0
0 12 24 36 48
Survival time
FERMT1 expression
Negative and weak
Moderate and strong
Censored
60 70 84 96
0.2
0.4
0.6
0.8
1.0
A
B
By FERMT1 expression
By ACHY expression
Cum
mula
tive
surv
ival
Log-rank test, P = 0.011
0.0
0 12 24 36 48
Survival time
ACHY expression
Negative and weak
Moderate and strong
Censored
60 70 84 96
0.2
0.4
0.6
0.8
1.0
Cum
mula
tive
surv
ival
Log-rank test, P = 0.082
Figure 2. Kaplan–Meier survival curves for OS as determined byexpression of FERMT1 and AHCY. A, patients with moderate/strongFERMT1 expression (n ¼ 122) showed significantly poorer OS (P ¼ 0.011)than patients with negative/weak FERMT1 expression (n ¼ 81). B, nosignificant difference in OSwas detected on the basis of AHCY expression(P ¼ 0.082).
Fan et al.
Clin Cancer Res; 17(9) May 1, 2011 Clinical Cancer Research2914
cgi-bin/carddisp.pl?gene¼AHCY&search¼AHCY). Upregu-lation of AHCY gene expression in colorectal carcinomacompared with uninvolved colon mucosa has beenreported in 3 independent studies of transcript profileanalysis with cDNAmicroarrays (34–37). However, valida-tion of differential AHCYprotein expression in colon cancerhas not been reported. In the present study, we report thefirst confirmation of elevated AHCY mRNA and proteinexpression in colon cancer compared with uninvolvedcolon tissue. Mechanistic studies of AHCY contributionsto colon cancer development remain a topic for futureinvestigations.
Comparison of research results obtained with micro-arrays and SAGE is very complicated because of thechallenges associated with selecting differentiallyexpressed genes from microarray and SAGE data. Foldchange and P values are 2 commonly used criteria forselecting differentially expressed genes. Use of these 2ranking criteria often produces different lists of differen-tially expressed genes (46, 47). In the present study, 9.4%(404/4,300) of the differentially expressed transcriptsidentified in the microarray data and 47.6% (404/848)of the differentially expressed transcripts in the SAGE data
were identical. These data indicate that the intraplatformcorrelation was modest. Most of the overlapping tran-scripts (388/404, 96.0%) between the 2 platformsshowed identical gene expression trends, showing thereliability of LongSAGE data. Among the top 20 upregu-lated genes and downregulated genes represented by theoverlapping transcripts, the differential expression of40.0% (16/40) of the genes have been confirmed in otherreports. The present study validated the differentialexpression of 6 of these genes by qPCR.
It has been reported that antibody concentration canaffect the apparent relation between biomarker expres-sion and outcome (48). A potential limitation of thepresent study is that although we carried out antibodytitration for each antibody used for immunohistochem-istry at the beginning of the study, and a single FERMT1protein band was obtained by Western blotting at anantibody dilution of 1:1,000 (the same dilution as thatused for immunohistochemistry), the fact that we used asingle antibody dilution for immunohistochemistry mayhave led to results that may have differed from those thatwe used for multiple dilutions. Further studies will clarifythis issue.
Table 4. The effect of FERMT1 and AHCY protein expression on overall survival time of colon cancerpatients
Time, mo 0 12 24 36 48 60 72 84 96
FERMT1Negative and weak No. at risk 81 79 72 70 63 46 31 3 0
No. of events 0 2 9 11 16 18 20 21 21Moderate and strong No. at risk 122 118 111 99 80 53 24 3 0
No. of events 0 4 11 22 40 47 51 53 53ACHY
Negative and weak No. at risk 134 131 123 116 99 72 41 6 0No. of events 0 3 11 17 31 37 42 44 44
Moderate and strong No. at risk 69 66 60 53 44 27 14 0 0No. of events 0 3 9 16 25 28 29 30 30
Table 5. The effect of FERMT1 and AHCY protein expression on disease free survival time of colon cancerpatients
Time, mo 0 12 24 36 48 60 72 84 96
FERMT1Negative and weak No. at risk 77 74 70 63 54 40 30 3 0
No. of events 0 3 7 14 20 23 23 23 23Moderate and strong No. at risk ns 105 93 74 65 46 21 3 0
No. of events 0 13 25 42 51 53 55 55 55ACHY
Negative and weak No. at risk 131 122 112 95 81 63 38 6 0No. of events 0 9 19 34 46 49 50 50 50
Moderate and strong No. at risk 64 57 51 42 38 23 13 0 0No. of events 0 7 13 22 25 27 28 28 28
Fan et al.
Clin Cancer Res; 17(9) May 1, 2011 Clinical Cancer Research2916
In conclusion, we present the first report using LCM-LongSAGE for quantitative measurement of genome-widegene expression in paired colon carcinomas and unin-volved colon mucosa. Gene profiling results were verifiedvia evaluation of differential protein expression in morethan 200 patients, together with histopathologic and clin-ical correlation with different stages of disease and patientoutcomes. This new, comprehensive approach to the accu-rate and rapid identification and validation of tumor-associated genes in colon cancer may serve as a modelfor future investigations. Identification and validation ofupregulated FERMT1 protein expression as a molecularevent in late-stage tumor progression showed that FERMT1is a new prognostic biomarker in colon cancer for predict-ing patient outcomes.
Disclosure of Potential Conflicts of Interest
No conflicts of interest were declared.
Acknowledgments
The authors thank Dr. H Hu, Department of Pathology, Shanghai FirstPeople’s Hospital, for review of stained samples.
Grant Support
This research was supported by the Key Basic Research Project of the Scienceand Technology Commission of Shanghai Municipality (05JC14029), the Pro-gram for Outstanding Medical Academic Leader of Shanghai Municipality(LJ06024), the National High-Technology Research and Development Program("863" Program) of China (2007AA022003), and the Leading Project of theScience and Technology Commission of Shanghai Municipality (09411963500).
The costs of publication of this article were defrayed in part by thepayment of page charges. This article must therefore be hereby markedadvertisement in accordance with 18 U.S.C. Section 1734 solely to indicatethis fact.
Received September 22, 2010; revised December 14, 2010; acceptedDecember 29, 2010; published OnlineFirst January 10, 2011.
References1. Li M, Gu J. Changing patterns of colorectal cancer in China over a
period of 20 years. World J Gastroenterol 2005;11:4685–8.2. Heidenblut AM, L€uttges J, Buchholz M, Heinitz C, Emmersen J,
Nielsen KL, et al. aRNA-LongSAGE: a new approach to generateSAGE libraries from microdissected cells. Nucleic Acids Res 2004;32:e131.
3. Willis ND, Przyborski SA, Hutchison CJ, Wilson RG. Colonic andcolorectal cancer stem cells: progress in the search for putativebiomarkers. J Anat 2008;213:59–65.
4. Misra S, Hascall VC, Berger FG, Markwald RR, Ghatak S. Hyaluronan,CD44, and cyclooxygenase-2 in colon cancer. Connect Tissue Res2008;49:219–24.
5. Shi Q, Sargent DJ. Meta-analysis for the evaluation of surrogateendpoints in cancer clinical trials. Int J Clin Oncol 2009;14:102–11.
6. Jankowski JA, Odze RD. Biomarkers in gastroenterology: betweenhope and hype comes histopathology. Am J Gastroenterol2009;104:1093–6.
7. Modjtahedi H, Essapen S. Epidermal growth factor receptor inhibitorsin cancer treatment: advances, challenges and opportunities. Antic-ancer Drugs 2009;20:851–5.
8. Aslam MI, Taylor K, Pringle JH, Jameson JS. MicroRNAs are novelbiomarkers of colorectal cancer. Br J Surg 2009;96:702–10.
9. Thomas SN, Tong Z, Stebe KJ, Konstantopoulos K. Identification,characterization and utilization of tumor cell selectin ligands in thedesign of colon cancer diagnostics. Biorheology 2009;46:207–25.
10. van Ruissen F, Ruijter JM, Schaaf GJ, Asgharnegad L, ZwijnenburgDA, Kool M, et al. Evaluation of the similarity of gene expression dataestimated with SAGE and Affymetrix GeneChips. BMC Genomics2005;6:91.
11. Bianchini M, Levy E, Zucchini C, Pinski V, Macagno C, De Sanctis P,et al. Comparative study of gene expression by cDNA microarray inhuman colorectal cancer tissues and normal mucosa. Int J Oncol2006;29:83–94.
12. Komori T, Takemasa I, Higuchi H, Yamasaki M, IkedaM, Yamamoto H,et al. Identification of differentially expressed genes involved in color-ectal carcinogenesis using a cDNA microarray. J Exp Clin Cancer Res2004;23:521–7.
13. Joyce T, Pintzas A. Microarray analysis to reveal genes involved incolon carcinogenesis. Expert Opin Pharmacother 2007;8:895–900.
0.0
0 12 24 36 48
DFS time
FERMT1 expression
Negative and weak
Moderate and strong
Censored
60 72 84 96
0.2
0.4
0.6
0.8
1.0
A
B
By FERMT1 expression
By ACHY expression
Cum
mula
tive
surv
ival
Log-rank test, P = 0.013
0.0
0 12 24 36 48
DFS time
ACHY expression
Negative and weak
Moderate and strong
Censored
60 72 84 96
0.2
0.4
0.6
0.8
1.0
Cum
mula
tive
surv
ival
Log-rank test, P = 0.397
Figure 3. Kaplan–Meier survival curves for DFS as determined byexpression of FERMT1 and AHCY. A, patients with moderate/strongFERMT1 expression (n¼ 122) showed significantly poorer DFS (P¼ 0.013)than patients with negative/weak FERMT1 expression (n ¼ 81).B, no significant difference in DFS was detected on the basis of AHCYexpression (P ¼ 0.397).
LCM-LongSAGE Analysis of Colon Cancer Biomarkers
www.aacrjournals.org Clin Cancer Res; 17(9) May 1, 2011 2917
14. Ki DH, Jeung HC, Park CH, Kang SH, Lee GY, Lee WS, et al. Wholegenome analysis for liver metastasis gene signatures in colorectalcancer. Int J Cancer 2007;121:2005–12.
15. Parle-McDermott A, McWilliam P, Tighe O, Dunican D, Croke DT.Serial analysis of gene expression identifies putative metastasis-associated transcripts in colon tumour cell lines. Br J Cancer2000;83:725–8.
16. Yousef GM, Borgoño CA, Popalis C, Yacoub GM, Polymeris ME,Soosaipillai A, et al. In-silico analysis of kallikrein gene expression inpancreatic and colon cancers. Anticancer Res 2004;24:43–51.
17. Stein WD, Litman T, Fojo T, Bates SE. A Serial Analysis of GeneExpression (SAGE) database analysis of chemosensitivity: comparingsolid tumors with cell lines and comparing solid tumors from differenttissue origins. Cancer Res 2004;64:2805–16.
18. Huang ZG, Ran ZH, LuW, Xiao SD. Analysis of gene expression profilein colon cancer using the Cancer Genome Anatomy Project and RNAinterference. Chin J Dig Dis 2006;7:97–102.
19. Velculescu VE, Zhang L, Vogelstein B, Kinzler KW. Serial analysis ofgene expression. Science 1995;270:484–7.
20. Pylouster J, Senamaud-Beaufort C, Saison-Behmoaras TE. WEB-SAGE: a web tool for visual analysis of differentially expressed humanSAGE tags. Nucleic Acids Res 2005;33:W693–5.
21. Saha S, Sparks AB, Rago C, Akmaev V, Wang CJ, Vogelstein B, et al.Using the transcriptome to annotate the genome. Nat Biotechnol2002;20:508–12.
22. Wahl MB, Heinzmann U, Imai K. LongSAGE analysis significantlyimproves genome annotation: identifications of novel genes andalternative transcripts in the mouse. Bioinformatics 2005;21:1393–400.
23. Hu M, Polyak K. Serial analysis of gene expression. Nat Protoc2006;1:1743–60.
24. Caretti E, Devarajan K, Coudry R, Ross E, Clapper ML, Cooper HS,et al. Comparison of RNA amplification methods and chip platformsfor microarray analysis of samples processed by laser capture micro-dissection. J Cell Biochem 2008;103:556–63.
25. Fan J, Peng Z, Zhou C, Qiu G, Tang H, Sun Y, et al. Gene-expressionprofiling in Chinese patients with colon cancer by coupling experi-mental and bioinformatic genomewide gene-expression analyses:identification and validation of IFITM3 as a biomarker of early coloncarcinogenesis. Cancer 2008;113:266–75.
26. Hamilton SR, Aaltonen LA, editors. World Health Organization Clas-sification of Tumours. Pathology and Genetics of Tumours of theDigestive System. Lyon, France:IARC Press; 2001.
27. O’Connell JB, Maggard MA, Ko CY. Colon cancer survival rates withthe new American Joint Committee on Cancer sixth edition staging. JNatl Cancer Inst 2004;96:1420–5.
28. Fisher RA. On the interpretation of c2 from contingency tables, and thecalculation of P. J R Stat Soc 1922;85:87–94.
29. Wilcoxon F. Individual comparisons by ranking methods. BiometricsBull 1945;1:80–3.
30. Kaplan EL, Meier P. Nonparametric estimation from incompleteobservations. J Am Stat Assoc 1958;53:457–81.
31. Cox DR. Regression models and life tables. J R Stat Soc Ser B1972;34:187–220.
32. Christensen E. Multivariate survival analysis using Cox's regressionmodel. Hepatology 1987;7:1346–58.
33. Weinstein EJ, Bourner M, Head R, Zakeri H, Bauer C, Mazzarella R.URP1: a member of a novel family of PH and FERM domain-contain-ing membrane-associated proteins is significantly over-expressed inlung and colon carcinomas. Biochim Biophys Acta 2003;1637:207–16.
34. Cardoso J, Boer J, Morreau H, Fodde R. Expression and genomicprofiling of colorectal cancer. Biochim Biophys Acta 2007;1775:103–37.
35. Notterman DA, Alon U, Sierk AJ, Levine AJ. Transcriptional geneexpression profiles of colorectal adenoma, adenocarcinoma, andnormal tissue examined by oligonucleotide arrays. Cancer Res2001;61:3124–30.
36. Birkenkamp-Demtroder K, Christensen LL, Olesen SH, FrederiksenCM, Laiho P, Aaltonen LA, et al. Gene expression in colorectal cancer.Cancer Res 2002;62:4352–63.
37. Lechner S, M€uller-Ladner U, Renke B, Sch€olmerich J, R€uschoff J,Kullmann F. Gene expression pattern of laser microdissected coloniccrypts of adenomas with low grade dysplasia. Gut 2003;52:1148–53.
38. Lai-Cheong JE, Parsons M, McGrath JA. The role of kindlins in cellbiology and relevance to human disease. Int J Biochem Cell Biol2010;42:595–603.
39. Siegel DH, Ashton GH, Penagos HG, Lee JV, Feiler HS, WilhelmsenKC, et al. Loss of kindlin-1, a human homolog of the Caenorhabditiselegans actin-extracellular-matrix linker protein UNC-112, causesKindler syndrome. Am J Hum Genet 2003;73:174–87.
40. Ussar S, Moser M, Widmaier M, Rognoni E, Harrer C, Genzel-Bor-oviczeny O, et al. Loss of Kindlin-1 causes skin atrophy and lethalneonatal intestinal epithelial dysfunction. PLoS Genet 2008;4:e1000289.
41. TianM, Neil JR, SchiemannWP. Transforming growth factor-b and thehallmarks of cancer. Cell Signal 2010 Nov 6. [Epub ahead of print].
42. Meulmeester E, Ten Dijke P. They dynamic roles of TGF-b in cancer. JPathol 2010 Sep 27. [Epub ahead of print].
43. Slattery ML, Herrick J, Lundgreen A, Wolff RK. Genetic variation in theTGF{beta}-signaling pathway and colon and rectal cancer risk. CancerEpidemiol Biomarkers Prev 2010 Nov 10. [Epub ahead of print].
44. Kloeker S, Major MB, Calderwood DA, Ginsberg MH, Jones DA,Beckerle MC. The Kindler syndrome protein is regulated by trans-forming growth factor-beta and involved in integrin-mediated adhe-sion. J Biol Chem 2004;279:6824–33.
45. Landemaine T, Jackson A, Bellahc�ene A, Rucci N, Sin S, Abad BM,et al. A six-gene signature predicting breast cancer lung metastasis.Cancer Res 2008;68:6092–9.
46. Chen JJ, Tsai CA, Tzeng S, Chen CH. Gene selection with multipleordering criteria. BMC Bioinforma 2007;8:74.
47. Mukherjee S, Roberts SJ. A theoretical analysis of the selection ofdifferentially expressed genes. J Bioinform Comput Biol 2005;3:627–43.
48. McCabe A, Dolled-Filhart M, Camp RL, Rimm DL. Automated quan-titative analysis (AQUA) of in situ protein expression, antibody con-centration, and prognosis. J Natl Cancer Inst 2005;97:1808–15.
Fan et al.
Clin Cancer Res; 17(9) May 1, 2011 Clinical Cancer Research2918
