Carcinogenesis vol.27 no.3 pp.392–404, 2006 doi:10.1093/carcin/bgi237 Advance Access publication October 11, 2005

Oligonucleotide microarray analysis of distinct expression patterns in colorectal cancer tissues harboring BRAF and K-ras mutations

Il-Jin Kim1,y, Hio Chung Kang1,2,y, Sang-Geun Jang1, Introduction Kun Kim1, Sun-A Ahn1, Hyun-Ju Yoon1, Sang Nam Yoon3 and Jae-Gahb Park1,2,3, Colorectal cancer (CRC) is an important human cancer, and incidence rates of this cancer are increasing in Asian countries 1 Korean Hereditary Tumor Registry, Cancer Research Institute and such as Korea (1). CRC development is a multi-step process Downloaded from https://academic.oup.com/carcin/article/27/3/392/2476120 by guest on 27 September 2021 Cancer Research Center, Seoul National University, Seoul, Korea, (2) involving microsatellite instability (MSI), mutations in 2Research Institute and Hospital, National Cancer Center, 809 Madu-dong, Ilsan-gu, Goyang, Gyeonggi 411-764, Korea mismatch repair (MMR) of such as MLH1 and MSH2, and 3Department of Surgery, Seoul National University College of and mutations in APC, SMAD4,K-ras, TP53 and b-catenin Medicine, Seoul, Korea (2–9). Somatic BRAF mutations have been reported in To whom correspondence should be addressed. 5.1–18% of CRCs (10–12). BRAF is one of three serine/ E-mail: [email protected] threonine (ARAF, BRAF, CRAF/RAF1) that act within Various types of human cancers harbor BRAF somatic the RAS–RAF–MEK–ERK–MAPK signaling pathway (13). mutations, leading researchers to seek molecular targets BRAF mutations have also been reported in 80% of primary for BRAF inhibitors. A mutually exclusive relationship melanomas (10), 68% of metastatic melanomas (14) and has been observed between the BRAF-V600E mutation 14–33% of ovarian carcinomas (10,15). Interestingly, a mutu- and K-ras mutations, suggesting that the BRAF-V600E ally exclusive relationship between the BRAF-V600E muta- mutation may differ from the other BRAF mutant types. tion and K-ras mutations has been found in most human Here, we used microarray analysis to examine differences cancers (10,12,16). In CRCs, BRAF-V600E mutations between the BRAF and K-ras mutant colorectal samples appeared to be associated with MMR deficiency and were and within the BRAF group (V600E versus non-V600E), not found in samples with K-ras mutations (12). This suggests in the hope that the identified gene sets could form the that the BRAF-V600E and K-ras mutations may have equival- basis for new target development. Eleven colorectal can- ent effects on tumorigenesis (12). However, other studies have cers (CRCs) with BRAF mutations and nine with K-ras shown that BRAF mutations were associated with MLH1 mutations were examined by high-density microarray promoter methylation but not MMR deficiency (17–19). analysis. We also tested whether other significant genetic Moreover, BRAF mutations were not found in MMR-deficient or clinical status involved in CRC development, such as hereditary non-polyposis colorectal cancer (HNPCC) samples APC and TP53 mutations, MSI and TNM-Duke’s staging, (18,19) but were found in sporadic CRCs, suggesting that were related with the observed BRAF-orK-ras associated BRAF may be involved in the carcinogenesis of non- expression profiles. Unsupervised two-way hierarchical inheritable CRCs (18,19). This is unexpected because it was clustering and multidimensional scaling revealed that believed that MMR-deficient phenotypes were associated with the differentially expressed genes clustered according to both sporadic and hereditary CRCs (18). Thus, researchers the mutation status of BRAF and K-ras, and that postulated that HNPCC patients could be screened for BRAF samples with the BRAF-V600E and non-V600E mutants mutations prior to MMR gene screening, as a weed-out tech- could be distinguished from each other by gene profiling. nique (18). Finally, BRAF mutations have been closely correl- Examination of TNM–Duke’s staging, MSI and mutations ated with methylator phenotypes in several genes including in APC and TP53 revealed that these significant muta- MLH1 (20), and BRAF mutations have been associated with tions could not account for the hierarchical clustering longer disease-free survival and a shorter duration of response results observed in our study. We herein identified to treatment was reported (21). Collectively, these observa- distinct patterns and gene sets that may tions seem to indicate that BRAF mutations could be used as form the basis for identification of BRAF-targeting therapeutic markers (18,20,21) and/or could be molecularly molecules or provide researchers with a better understand- targeted for development of new anticancer strategies. ing of the molecular pathogenesis underlying RAS–RAF A good example for molecularly targeted anti-cancer drug is signaling. imatinib (formerly STI-571), which inhibits the BCR–ABL in chronic myeloid leukemia and KIT in gastrointestinal stromal tumors (GIST) (22). In fact, several candidate molecu- lar BRAF inhibitors have entered clinical trials (23,24). However, the V600E BRAF mutation clearly differs from the non-V600E BRAF mutations (10) in that V600E is independ- Abbreviations: CRC, colorectal cancer; GIST, gastrointestinal stromal ent of Ras signaling and elevates basal kinase activity without tumors; HNPCC, hereditary non-polyposis colorectal cancer; LOOCV, leave-one-out cross validation; MDS, multidimensional scaling; MMP1, K-ras mutations (10), while the non-V600E BRAF mutations matrix metalloproteinase 1; MMR, mismatch repair; MSI, microsatellite are Ras-dependent and are not mutually exclusive with K-ras instability; PAM, prediction analysis of microarrays; ROC, receiver operating mutations (10). As the V600E mutation accounts for 80% of characteristic; SPRED2, sprouty-related, EVH1 domain containing 2. all BRAF mutations (10), it would seem logical to search for yThese authors contributed equally to this work. V600E-specific molecular inhibitors.

Carcinogenesis vol.27 no.3 # Oxford University Press 2005; all rights reserved. 392 Microarray analysis of CRCs with BRAF and K-ras mutations

Microarray analysis is commonly used to screen genome- samples were run on an ABI 3100 sequencer (Applied Biosystems), and the wide gene expression profiles in human diseases, including Genescan software (Genotyper 2.1, ABI, Foster City, CA) was used to calcu- cancers (25–27). This technique can also be used to identify late the size of each fluorescent PCR product for MSI determination (38). new cancer subgroups (class discovery), to classify samples Sample hybridization using oligonucleotide microarrays into known cancer subgroups (class prediction) (28) and to The microarray experiments were performed according to our previous experi- predict the prognosis or therapeutic responsiveness of cancer mental protocol (39) with the addition of 10% DMSO (dimethyl sulfoxide) to patients (29,30). A previous microarray study revealed distinct the hybridization mixture. Briefly, the extracted total RNA was quantified by spectrophotometer (Beckman Coulter, Fullerton, CA), checked by 1% agarose gene expression patterns in BRCA1/2 mutant tissues sampled gel electrophoresis and then purified with an RNeasy kit (Qiagen, Valencia, from hereditary ovarian/breast cancer patients versus the non- CA). An aliquot of 40 ml of purified RNA was ethanol-precipitated with 3 M mutant group (BRCAx) (31,32). In CRCs, microarray analysis sodium acetate (pH 5.2), 1 ml of glycogen (5 mg/ml) and 100 ml of 100% was used to divide samples into two groups based on the ice-cold ethanol, and 20 mg of purified RNA was used to synthesize double- existence of MSI, which is an important marker in CRC (33). strand cDNAs using SuperScript II reverse transcriptase (Life Technologies, Rockville, MD) and an HPLC-purified T7-(dT)24 primer (Metabion, Germany). Recently, BRAF mutant melanoma samples were distinguished The synthesized double-strand cDNA was purified with a Qiagen DNA puri- from BRAF wild-type samples by supervised microarray ana- fication kit (Qiagen) and ethanol-precipitated with 1 ml of glycogen, 20 ml Downloaded from https://academic.oup.com/carcin/article/27/3/392/2476120 by guest on 27 September 2021 lysis (34), suggesting that gene expression profiling according of 7.5 M of ammonium acetate, and 100 ml of 100% ethanol. Biotinylated to BRAF status might be useful for the identification of cRNA was synthesized from the double-stranded cDNA using the GeneChip Expression 30-Amplification Reagents (Affymetrix), and then purified and molecular markers involved in RAS–RAS–MEK–ERK– fragmented. The fragmented cRNA was quantified, and 10 mg of cRNA was MAPK signaling. However, no previous work has used hybridized to the oligonucleotide microarray, which was subsequently washed microarray analysis to investigate the molecular differences and stained with streptavidin-phycoerythrin. Scanning was performed with an among samples with BRAF-V600E and non-V600E mutations, Affymetrix Scanner (Affymetrix). and K-ras mutations. Microarray data analysis Here, we examined whether microarray analysis was The scanned GeneChip data were analyzed with the GCOS 1.1 (GeneChip capable of distinguishing colorectal samples according to Operating Software, Affymetrix) and DMT 3.0 (Data Mining Tool, Affymet- BRAF and K-ras mutation status. Once these divisions were rix) programs. All genes present on the GeneChip were globally normalized established, we tested whether other significant genetic and adjusted to a user-specific target signal value (500). The Universal Human Reference RNA (Stratagene) sample was also hybridized to the microarray, changes involved in CRC development, such as APC and and its values served as the base line (control) for calculating the signal log TP53 mutations, MSI and TNM-Duke’s staging, were related ratios in the 20 CRC samples. After normalization, the present calls (P-calls) with the observed BRAF-orK-ras-associated expression generated by the DMT software were calculated for each gene in the 20 profiles. samples. Probes without a P-call from any of 20 samples were excluded from the analysis and 16 332 probes were left. For the unsupervised hierarchical clustering, two-way hierarchical cluster- ing was applied to both genes and arrays, using the Cluster and Treeview Materials and methods programs (40). The two-way median center was selected for adjusting data, and the uncentered correlation was used for average linkage clustering. A total CRC tissue samples of 11 310 probes showing 450% expression (410 P-calls) in the 20 samples A total of 20 sporadic CRC cases were selected, including 11 samples with selected and the signal–log ratio values of 11 310 probes were used for BRAF mutations and 9 samples with K-ras mutations, as previously described hierarchical clustering. To clearly identify differentially expressed genes, the (9). Clinical characteristics, including age, sex, preoperative CEA level, tumor Student’s t-test was performed using DMT. This allowed identification of 2526 location, differentiation, lymphatic invasion, venous invasion, neural invasion, probes (P 5 0.05) differentially expressed between the BRAF and K-ras A–C stage, Duke’s stage and TNM stage were noted. CRC tissues were groups, and 1703 probes (P 5 0.05) differentially expressed between the collected from the Seoul National University Hospital and the National Cancer BRAF-V600E and non-V600E groups. These two gene sets were further ana- Center, Korea. The fresh cancer tissues were stored at 70C in a liquid lyzed by hierarchical clustering to identify distinct gene expression patterns. nitrogen tank. DNAs and total RNAs from tumor samples (portions with Multidimensional scaling (MDS) was performed (SPSS, Chicago, IL) and 460% cancer cells) were extracted using the Trizol reagent (Invitrogen, the SigmaPlot (SPSS) program utilized. The signal–log ratio values of all 20 Carlsbad, CA) according to the manufacturer’s instructions. The extracted samples were entered in the SPSS, the multidimensional scaling factors (X, Y genomic DNA was used for mutational analysis of BRAF,K-ras, APC, TP53 and Z-values) were calculated for each sample, and these values were used for and MSI. The extracted total RNAs were used for gene expression analysis on MDS analysis with the signal plot software. The initial MDS was performed an Affymetrix U133A 2.0 GeneChip (Affymetrix, Santa Clara, CA) containing using 11 310 probes in all 20 samples, with further analyses performed using 22 277 probes. The Universal Human Reference RNA (Stratagene, CA) was the 2526 and 1703 probes identified above. Prediction analysis of microarrays used as a control for comparison with the gene expressions in the 20 CRCs. (PAM) analysis was performed as described previously (41), and was used to identify a gene set for classification by ranking genes with a penalized Mutational analysis of BRAF, K-ras, TP53, APC and MSI t-statistic and user-applied threshold (41). Leave-one-out cross validation Exons 11 and 15 of the BRAF gene were screened by oligonucleotide microar- (LOOCV) analysis was performed using GeneCluster2.0 software (http:// ray and direct sequencing with the previously described primer sets (16). PCR www.broad.mit.edu/cancer/software/software.html). ‘S2N (Signal-to-Noise)’ reactions were carried out in a volume of 25 ml containing 100 ng genomic value was used for ‘distance function’ and 100 top ranked genes (number of DNA, 10 pmol of each primer, 250 mM each dNTP, 0.5 U of Taq polymerase neighbors) were selected from each BRAF and K-ras group. A permutation and the reaction buffer provided by the supplier (Qiagen, Hilden, Germany). analysis was performed 1000 times using ‘median’ class estimate (P ¼ 0.05) in Samples were denatured for 5 min at 94C in a GeneAmp PCR system 9700 the top ranked 100 genes. The LOOCV results were compared using different (Applied Biosystems, Foster City, CA), and then amplified by 35 cycles of gene numbers from 10 to 100 genes (10, 20, 30, 40, 50, 60, 70, 80, 90 and 100) 94C for 30 s, 55C for 30 s and 72C for 1 min, with a final elongation of and the best LOOCV results and ‘predictor’ gene set were obtained. The best 10 min at 72C. Codons 12 and 13 of the K-ras gene were screened by LOOCV results were judged by the lowest receiver operating characteristic oligonucleotide microarray as described previously (9) and bidirectionally (ROC) error value. The selected ‘predictor’ was statistically confirmed one sequenced using the Taq dideoxy terminator cycle sequencing kit and an more time using the Fisher test. and function analyses were ABI 3730 DNA sequencer (Applied Biosystems). performed with the Affymetrix NetAffx software package (Affymetrix). All the coding regions (all 15 exons) (35) of APC and exons 2–11 of TP53 (36,37) were entirely screened by an automatic bidirectional sequencing Statistical analysis method. Five microsatellite markers (BAT-25, BAT-26, D2S123, D5S346 Statistical analyses were performed using the x2 or Fisher’s exact tests to and D17S250) were used for determining MSI status using DHPLC (denatur- determine the strength of the correlations between the mutations in BRAF, ing high performance liquid chromatography) and the capillary-based method K-ras, APC, TP53, MSI, TNM–Duke’s stage and the two clusters (left and (38). MSI-H cancers are defined as having MSI in 2 of the 5 Bethesda panel right clusters). P ¼ 0.05 was set as the significance level using the SPSS markers and MSI-L cancers show MSI in only 1 of the 5 markers. Labeled software (SPSS). Tests for associations between BRAF and K-ras quantitative 393 I.-J.Kim et al.

Table I. Mutations of BRAF,K-ras, APC, TP53 and MSI in 20 CRCs

Sample Location TNM Duke’s BRAF K-ras APC TP53 MSI BAT-25 BAT-26 D2S123 D17S250 D5S346

172 P 3 C V600E W W W H I I I I I 194 P 2 B V600E W 4666_4667insA W H I I I I I E190X 296 D 3 C V600E W 4666_4667insA W L S S S S I 497 D 4 D V600E W W R273H S S S S S S R282W 590 D 2 B V600E W W W H I I S S I 626 P 2 B V600E W 677_678insA W H I I S S S 653 P 3 C V600E W W R282W S S S S S S 656 P 4 D V600E W W W S S S S S S 734 P 4 D V600E W 4608_4741delinsG W S S S S S S

481 D 4 D D594G W W G286K S S S S S S Downloaded from https://academic.oup.com/carcin/article/27/3/392/2476120 by guest on 27 September 2021 640 D 4 D G464V W W P151S S S S S S S 218 P 4 D W G12D K1370X R282W S S S S S S 231 D 4 D W G12D W W S S S S S S 271 P 4 D W G12D W W S S S S S S 346 D 3 C W G12V W W S S S S S S 359 D 2 B W G13D 4216delC W S S S S S S 406 P 2 B W G13D W W S S S S S S 418 D 1 A W G13D W IVS4-1G4AH I I S I I 456 P 2 B W G12D 550-551insA W S S S S S S 475 D 3 C W G13D E853X W S S S S S S

P, proximal colon cancer; D, distal CRC; W, wild-type; I, microsatellite instability; S, microsatellite stability; Duke’s, Duke’s stage; TNM, TNM stage; H, MSI-high; L, MSI-low.

RT–PCR (qRT–PCR) expression were performed with the one-way ANOVA that eight samples (four from the BRAF mutation group and and Wilcoxon–Mann–Whitney tests (SPSS). four from the K-ras mutation group) harbored truncating APC Quantitative RT–PCR mutations, six samples (four from the BRAF mutation group We selected four genes (IL8, MMP1, TUBA1 and PTS) for real-time qRT–PCR and two from the K-ras mutation group) had TP53 somatic for validation of the microarray data. Using the SuperScript Preamplification mutations, and five samples (four from the BRAF mutation System for first strand cDNA synthesis, 5 mg of total RNA was used for group and one from the K-ras mutation group) exhibited MSI- creation of single-stranded cDNA (Life Technologies). The cDNA was diluted and quantitatively equalized for PCR amplification. For real-time qRT–PCR, H and one showed MSI-L (from the BRAF group). Detailed the ABI Prism 7900 sequence detection system (Applied Biosystems) was mutational data for the 20 samples are presented in Table I. used. SYBRÒ Premix Ex TaqTM (Takara, Kyoto Japan) were used for each PCR reaction and GAPDH gene was simultaneously run as a control and used Unsupervised hierarchical clustering of CRC tissues with for normalization. Non-template-control wells without cDNA were included BRAF and K-ras mutations as negative controls. Each test sample was run in triplicate. The primer sets for PCR amplification were designed as follows: IL8-F: 50-CATACTCCAA- We analyzed 20 CRC tissues with either BRAF or K-ras ACCTTTCCAC-30, IL8-R: 50-AGCCCTCTTCAAAAACTTCT-30, MMP1-F: mutations by two-way hierarchical clustering to see whether 50-AAATCTTGCTCATGCTTTTC-30, MMP1-R: 50-CACTGAAGGTGT- the microarray analysis-derived gene expression profiles AGCTAGGG-30, PTS-F: 50-TACGGGAATGGTTATGAATC-30, PTS-R: would cluster into two groups according to the mutation status 50-CTCACCACATCTGCAAAGTA-30, TUBA1-F: 50-CAACCTACACCAA- CCTCAAT-30, TUBA1-R: 50-GAACTCTGTCAGGTCCACAT-30. Following of BRAF and K-ras. We selected 11 310 probes showing the standard curve method, the expression quantities of the examined genes 450% expression (410 P-calls) in the 20 samples and per- were determined using the standard curves and the CT values and normalized formed unsupervised hierarchical clustering, which revealed using GAPDH expression quantities. that CRC samples harboring BRAF and K-ras mutations dif- fered in terms of their gene expression profiles (P ¼ 0.002) Results (Figure 1A). Samples with K-ras mutations tended toward the left of the hierarchical cluster, with the exception of samples BRAF, K-ras, APC and TP53 mutation screening and MSI 456 and 218, while samples with BRAF mutations clustered to analysis in CRCs the right, with the exception of sample 640. The latter sample We previously screened CRC tissues for BRAF and K-ras (640), which clustered with the K-ras group, harbored the mutations using oligonucleotide microarrays and direct G464V mutation in exon 11. Similarly, K-ras sample 456 sequencing (9). Here, we selected 11 CRC tissues with BRAF (mutation at codon 12) clustered with sample with BRAF mutations; 9 harbored the V600E (codon 600, Val ! Glu, GTG sample 481, which was the other non-V600E mutation ! GAG, exon 15) mutation, 1 harbored the G464V (codon 464, included in the study (D594G). Glu ! Val, GGA ! GTA, exon 11) and 1 harbored the Next, we analyzed the 11 samples with BRAF mutations by D594G (codon 594, Asp ! Gly, GAT ! GGT, exon 15) two-way hierarchical clustering using 11 310 probes, and found mutation. We also selected nine CRC samples with K-ras that the non-V600E mutant samples (481 and 640) clustered mutations for comparison purposes; five harbored mutations very near each other (Figure 1B). We also developed a clus- at codon 12 and the other four had mutations at codon 13. After tering dendrogram for APC, TP53 and MSI status, to examine we analyzed the gene expression profiling of these 20 CRC whether these genetic alterations affected the unsupervised samples, we further screened the specimens for APC and TP53 gene clustering results (Figure 1C). There was no observable mutations and for MSI (via five Bethesda panel). We found relationship between mutations in APC (P ¼ 0.264) or TP53 394 Microarray analysis of CRCs with BRAF and K-ras mutations

A B

C Downloaded from https://academic.oup.com/carcin/article/27/3/392/2476120 by guest on 27 September 2021

Fig. 1. Unsupervised hierarchical clustering of 20 CRCs with BRAF and K-ras mutations. (A) Unsupervised two-way hierarchical clustering was performed using 11 310 probes showing 450% expression (410 P-calls) in the 20 CRC samples. For each gene, red indicates higher gene expression relative to the control and green indicates lower gene expression. Red circle, K-ras mutant group; blue circle, BRAF mutant group; triangle, non-V600E BRAF mutant samples. The numbers indicate sample names. (B) Unsupervised clustering result of the 11 BRAF mutant samples. Clustering was performed using the 11 310 probes used in (A). Arrows indicates non-V600E BRAF mutant samples, and the remaining are BRAF-V600E samples. (C) Mutation status of APC, TP53, MSI and TNM–Duke’s stage was analyzed and correlations were examined by unsupervised clustering. Blue circles indicate mutation or MSI in each gene, while empty boxes indicate no mutation.

(P ¼ 0.690) and the observed gene expression patterns. In BRAF-V600E samples and non-V600E by t-test and identified terms of MSI, although five of six samples tended to the 1703 probes. When we used these probes for MDS, the V600E right panel, there was no statistically significant association and non-V600E groups were more clearly separated between clustering and MSI (P ¼ 0.163). No significant rela- (Figure 2C). tionship between TNM–Duke’s staging and gene expression patterns was also observed. Supervised hierarchical clustering of CRC tissues with genes showing significant differences between the BRAF and K-ras MDS of the 20 CRC samples with the full complement of groups 11 310 expressed probes and those identified by t-test Having observed distinct gene expression patterns between the MDS was used to investigate differences in the gene expres- BRAF and K-ras groups by unsupervised clustering and MDS, sion patterns between BRAF and K-ras mutated CRC tissues. we performed a supervised hierarchical clustering analysis of The 11 310 probes selected above were used for MDS, which the 20 samples using the 2526 differentially expressed genes revealed distinct patterns between the BRAF and K-ras groups identified by t-test. This analysis yielded distinct clustering (Figure 2A), consistent with the results of our hierarchical dendrograms, with the 20 samples clearly separated into the clustering. Unlike the hierarchical clustering results, non- BRAF and K-ras groups (data not shown). Furthermore, the V600E samples 481 and 640 were generally located between V600E and non-V600E BRAF mutation groups yielded the K-ras and BRAF-V600E groups in MDS. However, con- dendrograms in which the 11 BRAF samples were divided sistent with the clustering results, K-ras sample 218 was correctly into two groups (data not shown). placed outside the K-ras mutation group. We then sought to examine a detailed MDS by analyzing the BRAF and K-ras groups by t-test using the DMT 3.0 (Data Mining Tool, PAM-based identification of 98 genes capable of distinguishing Affymetrix) program. We identified 2526 probes showing the BRAF and K-ras groups statistically significant differences in expression (P 5 0.05) PAM analysis was used to identify genes capable of distin- between the two groups, and used these probes for a second guishing between the BRAF and K-ras groups. Cross-validated round of MDS. A more clear distinction was observed between probabilities were used to set the threshold level at 2.19. We BRAF and K-ras groups (Figure 2B). We then compared identified 98 genes that were differentially expressed between 395 I.-J.Kim et al.

A B Downloaded from https://academic.oup.com/carcin/article/27/3/392/2476120 by guest on 27 September 2021

C

Fig. 2. MDS (multidimensional scaling) of 20 CRC samples with BRAF or K-ras mutations. (A) MDS analysis of the 20 samples was performed using 11 310 probes to confirm and compare the unsupervised hierarchical clustering results. Red circle, K-ras mutant group; blue square, BRAF-V600E group; dark triangle, BRAF non-V600E group; arrow (diamond), sample 218. The letters X, Y and Z indicate the multidimensional scaling values calculated by the SPSS software. (B) MDS analysis of all 20 samples using 2526 probes identified by t-test (P 50.05) as being differentially expressed in the BRAF group versus the K-ras group. Red circle, K-ras mutant group; blue square, BRAF-V600E group; dark triangle, BRAF non-V600E group; arrow (diamond), sample 218. (C) Eleven BRAF mutant samples were analyzed by MDS using 1703 probes which significantly (P 50.05) distinguished the BRAF-V600E and non-V600E groups. Blue circle, BRAF mutant group; dark triangle, BRAF non-V600E mutant samples. two groups, indicating that they could be used as classifiers with microarray data (Figure 4). Statistical differences (Table II). between BRAF and K-ras group were found in all four genes, IL8 (P ¼ 0.0004), MMP1 (P ¼ 0.0006), PTS (P ¼ 0.0003) and LOOCV analysis TUBA1 (P ¼ 0.0018). LOOCV analysis was used to confirm the unsupervised hier- archical clustering analysis and 80 genes were selected as a Clinical characterization of 20 CRCs classifier (Table III). The identified 80 classifiers correctly No clinicopathological differences were found between the predicted 18 out of 20 samples (90%). Samples 218 and BRAF and K-ras mutant samples, with one exception. Lymph- 640 which were misclassified in the unsupervised hierarchical atic invasion of the tumor was found in 6 of 11 BRAF mutant clustering were also wrongly predicted by the LOOCV. samples, but in 0 of 9 K-ras mutant samples (P ¼ 0.014). The ROC error obtained by LOOCV was 0.101. By the 1000 permutation test, all 80 genes but 1 (TUBA1) showed statist- Discussion ically significant LOOCV score than that of 1% of permuta- tion. The identified 80 genes were confirmed by Fisher test BRAF somatic mutations have been found in most human (P ¼ 0.0009) (Figure 3). cancers, with the exception of gastric cancers (10,16,42,43). The V600E mutation accounts for 480% of the identified Quantitative real-time RT–PCR for validating microarray BRAF mutations regardless of cancer type, and the remainder results of the identified BRAF mutations are restricted to exons 11 and To validate the oligonucleotide microarray results, we 15. The V600E BRAF mutation appears to have a mutually performed a SYBR green-based real-time quantitative PCR exclusive relationship with K-ras mutations, which are an assay of four genes, IL8, MMP1, PTS and TUBA1 that showed important component of the multi-step development of significantly different expressions between the BRAF and CRC (2). As K-ras somatic mutations are found in many K-ras groups. The quantitative PCR results were consistent human cancers (9,10), the mutually exclusive relationship 396 Microarray analysis of CRCs with BRAF and K-ras mutations

Table II. List of 98 genes identified by PAM as distinguishing the BRAF and K-ras mutant groups

No. Gene BRAF K-ras GO biological process GO molecular function GO cellular component symbol score score description description description

1 SENP7 0.3137 0.3834 sumoylation Cysteine-type peptidase activity Nucleus 2 DACH1 0.2046 0.2501 Cell growth and/or maintenance DNA binding Nucleus 3 NAT2 0.2027 0.2477 Metabolism Acetyltransferase activity 4 RERE 0.1862 0.2276 Regulation of transcription, Protein binding Nucleus DNA-dependent 5 MLLT4 0.1842 0.2252 Cell adhesion Motor activity Intercellular junction 6 IL8 0.184 0.2249 G-protein coupled receptor Chemokine activity Extracellular space protein signaling pathway 7 DST 0.1615 0.1974 Cell adhesion Actin binding Basal plasma membrane 8 PRDM2 0.1543 0.1886 Regulation of transcription, Metal ion binding Nucleus

DNA-dependent Downloaded from https://academic.oup.com/carcin/article/27/3/392/2476120 by guest on 27 September 2021 9 IRF7 0.1495 0.1827 Inflammatory response Specific RNA polymerase II Cytoplasm transcription factor activity 10 BTG2 0.1473 0.18 DNA repair Transcription factor activity 11 TI-227H 0.1452 0.1775 12 SCAMP1 0.1368 0.1672 Post-Golgi transport Integral to membrane 13 TPR 0.1275 0.1558 Protein-nucleus import Cytoplasm 14 TNFRSF10B 0.122 0.1491 Activation of NF-kappa TRAIL binding Integral to membrane B-inducing kinase 15 ZNF36 0.1137 0.139 Regulation of transcription, Transcription factor activity Nucleus DNA-dependent 16 TUBA1 0.1126 0.1376 Microtubule polymerization GTP binding Tubulin 17 MARCH-VI 0.1119 0.1368 18 LY6E 0.1099 0.1343 Cell surface receptor linked Integral to plasma membrane signal transduction 19 RCOR3 0.1054 0.1288 20 GPR161 0.1036 0.1266 G-protein coupled Rhodopsin-like receptor activity Integral to membrane receptor protein signaling pathway 21 CFTR 0.1032 0.1261 Ion transport ATP binding Apical plasma membrane 22 PDE3A 0.1022 0.1249 Lipid metabolism CGMP-inhibited cyclic-nucleotide Membrane phosphodiesterase 23 NRGN 0.0975 0.1192 Neurogenesis Calmodulin binding 24 MVK 0.0859 0.105 Biosynthesis ATP binding Cytoplasm 25 SPRR2A 0.0847 0.1036 Keratinocyte differentiation Structural molecule activity Cornified envelope 26 CXCL5 0.0842 0.1029 Cell-cell signaling Chemokine activity Extracellular 27 TDGF1 0.078 0.0954 Activation of MAPK Growth factor activity Cell surface 28 SRRM2 0.0778 0.0951 29 TNFAIP6 0.0753 0.092 Cell adhesion Hyaluronic acid binding Extracellular 30 COL7A1 0.0714 0.0872 Cell adhesion Protein binding Basement membrane 31 WIG1 0.0678 0.0828 Nucleic acid binding Nucleus 32 NCOA2 0.0671 0.082 Regulation of transcription, Signal transducer activity Nucleus DNA-dependent 33 NOX1 0.0664 0.0811 FADH2 metabolism activity Integral to membrane 34 RAB3B 0.066 0.0807 Protein transport GTP binding 35 G1P2 0.0636 0.0778 Cell-cell signaling Protein binding Cytoplasm 36 SEMA3F 0.0628 0.0767 Development Receptor activity Extracellular space 37 SNX1 0.0625 0.0764 Endocytosis Protein transporter activity Golgi apparatus 38 CEL 0.0618 0.0755 Cell differentiation Cholinesterase activity Golgi apparatus 39 ARSD 0.0607 0.0741 Metabolism Arylsulfatase activity Lysosome 40 IMP-3 0.0596 0.0729 RNA processing RNA binding Cytoplasm 41 MGC8902 0.059 0.0722 42 RANBP3 0.0586 0.0716 Protein transport RNA protein binding Nuclear pore 43 TNFSF7 0.0579 0.0708 Tumor necrosis factor Integral to plasma membrane receptor binding 44 H2AFX 0.0578 0.0707 organization and DNA binding Chromosome biogenesis (sensu Eukaryota) 45 SPRED2 0.0566 0.0692 Development Membrane 46 CUGBP2 0.0552 0.0674 RNA processing RNA binding 47 NPIP 0.0549 0.0672 48 FAM12A 0.0541 0.0661 Sperm displacement Nucleic acid binding Extracellular space 49 CAPN3 0.052 0.0635 Muscle development Calcium ion binding Intracellular 50 ABLIM2 0.0486 0.0594 Cytoskeleton organization Actin binding and biogenesis 51 EZH1 0.0483 0.059 G-protein coupled receptor Chromatin binding Heterotrimeric G-protein protein signaling pathway complex 52 MMP1 0.0476 0.0581 Collagen catabolism Calcium ion binding Extracellular matrix (sensu Metazoa) 53 IGHD 0.046 0.0562 Immune response Antigen binding

397 I.-J.Kim et al.

Table II. Continued

No. Gene BRAF K-ras GO biological process GO molecular function GO cellular component symbol score score description description description

54 FZD2 0.0458 0.056 G-protein coupled receptor G-protein coupled Integral to plasma membrane protein signaling pathway receptor activity 55 CTH 0.0455 0.0556 Amino acid metabolism Cystathionine gamma- activity 56 PSD3 0.0442 0.054 57 CTAG2 0.0442 0.054 Integral to membrane 58 G0S2 0.0441 0.0539 Regulation of cell cycle 59 WSB1 0.0431 0.0526 Intracellular signaling cascade 60 IL6ST 0.0417 0.051 Cell surface receptor linked Interleukin-6 receptor activity Integral to plasma membrane signal transduction 61 BAG2 0.0412 0.0503 Apoptosis Unfolded protein binding Downloaded from https://academic.oup.com/carcin/article/27/3/392/2476120 by guest on 27 September 2021 62 HOXB9 0.0389 0.0476 Development Transcription factor activity Nucleus 63 SLC7A6 0.037 0.0453 Amino acid metabolism Amino acid-polyamine Integral to plasma membrane transporter activity 64 COPA 0.0368 0.045 ER to Golgi transport Hormone activity Golgi apparatus 65 AOAH 0.0364 0.0445 Inflammatory response Acyloxyacyl activity Integral to membrane 66 TM4SF6 0.0352 0.043 Cell motility Signal transducer activity Integral to membrane 67 CASK 0.035 0.0428 Cell adhesion ATP binding Actin cytoskeleton 68 BAT1 0.0332 0.0406 ATP binding Nucleus 69 GOLGA4 0.0328 0.0401 Vesicle-mediated transport Golgi trans face 70 BST2 0.0327 0.04 Cell proliferation Signal transducer activity Integral to plasma membrane 71 LONP 0.0314 0.0383 ATP-dependent proteolysis ATP binding 72 FOXJ3 0.0302 0.037 73 IQGAP1 0.0301 0.0368 Signal transduction GTPase inhibitor activity Actin filament 74 GALNT4 0.0301 0.0368 Manganese ion binding Golgi apparatus 75 TNRC18 0.0292 0.0357 76 RBM5 0.028 0.0342 RNA processing DNA binding Nucleus 77 GPR109B 0.0262 0.032 G-protein coupled receptor Purinergic nucleotide Integral to plasma membrane protein signaling pathway receptor activity, G-protein 78 UTX 0.0259 0.0317 Nucleus 79 TRIM2 0.0231 0.0283 Protein ubiquitination Myosin binding Cytoplasm 80 CCL3 0.021 0.0256 G-protein coupled receptor Chemokine activity Extracellular protein signaling pathway 81 NEDD4L 0.0197 0.024 Excretion activity Intracellular 82 HNRPD 0.0149 0.0182 RNA catabolism DNA binding Chromosome, telomeric region 83 IFIT1 0.0143 0.0175 Immune response Cytoplasm 84 MLL 0.0142 0.0174 Cell growth and/or maintenance RNA polymerase II transcription Nucleus factor activity 85 CHD1L 0.013 0.0159 ATP binding 86 TFPI 0.0105 0.0129 Blood coagulation Serine-type endopeptidase Extracellular inhibitor activity 87 DEFA6 0.0099 0.0121 Antimicrobial humoral Extracellular response (sensu Vertebrata) 88 S100A12 0.0092 0.0113 Defense response to bacteria Calcium ion binding Cytosol 89 DAPK2 0.0091 0.0111 Apoptosis ATP binding Cytoplasm 90 IL6 0.0088 0.0107 Acute-phase response Cytokine activity Extracellular space 91 TGFBR2 0.0049 0.006 Positive regulation of ATP binding Integral to membrane cell proliferation 92 SLAMF8 0.0046 0.0057 93 SULT1B1 0.0031 0.0038 Biogenic amine metabolism Sulfotransferase activity 94 AMY2A 0.0027 0.0033 Carbohydrate metabolism Alpha-amylase activity Extracellular space 95 SLC7A8 0.0026 0.0032 Amino acid metabolism Amino acid permease activity Integral to membrane 96 SKD3 0.0021 0.0025 ATP binding 97 RRBP1 0.0015 0.0018 Protein targeting Ribosome receptor activity Integral to endoplasmic reticulum 98 ANKRD11 7.00E-04 9.00E-04 Nucleus

with BRAF-V600E may have large value in molecular and group), and furthermore whether we could distinguish the clinical research. BRAF is a serine/threonine kinase; since BRAF-V600E group from the non-V600E group. We obtained are considered good candidate molecular targets for 20 CRC samples, 11 with BRAF mutations and 9 with K-ras drug development (23,24), it is possible that BRAF or BRAF- mutations. Clinicopathological examination revealed that related genes could be good targets for anti-cancer drugs lymphatic invasion of the tumor was significantly associated (23,24). with the BRAF group (P ¼ 0.014), but further study with a In a first step towards identifying some of these molecular larger sample size will be required to confirm this result. We targets, we herein examined whether a gene set could be then used microarray analysis followed by unsupervised hier- identified as distinguishing BRAF mutated CRC samples archical clustering and MDS (multidimensional scaling) meth- (BRAF group) from those harboring K-ras mutations (K-ras ods. Two-way hierarchical clustering, revealed that the K-ras 398 Microarray analysis of CRCs with BRAF and K-ras mutations

Table III. List of 80 genes identified by LOOCV as distinguishing the BRAF and K-ras mutant groups

No Gene GO biological process GO molecular function GO cellular component Change symbol description description description

1 TUBA1 Microtubule-based movement GTPase activity Microtubule Up 2 PTS Amino acid metabolism 6-pyruvoyltetrahydropterin Up synthase activity 3 CCT5 Protein folding ATP binding Up 4 SARA1 Intracellular protein transport GTPase activity Endoplasmic reticulum Up 5 CBR1 Metabolism Carbonyl reductase (NADPH) activity Cytosol Up 6 TXN Electron transport Electron transporter activity Up 7 TAF10 Transcription Transcription factor activity Nucleus Up 8 FADD Cell surface receptor linked Signal transducer activity Cytoplasm Up signal transduction

9 ZWINT Nucleus Up Downloaded from https://academic.oup.com/carcin/article/27/3/392/2476120 by guest on 27 September 2021 10 GLUD1 Amino acid metabolism Glutamate dehydrogenase Mitochondrion Up [NAD(P)þ] activity 11 MGC2494 Up 12 FGF1 Regulation of cell cycle Growth factor activity Extracellular space Up 13 TALDO1 Carbohydrate metabolism Transaldolase activity Cytoplasm Up 14 PELO Protein biosynthesis Nucleus Up 15 RAB3B Small GTPase mediated GTPase activity Up signal transduction 16 C14orf122 Up 17 FTH1 Iron ion transport Ferroxidase activity Plasma membrane Up 18 OAZ1 Polyamine biosynthesis Ornithine decarboxylase Up inhibitor activity 19 C17orf35 Signal transduction G-protein coupled receptor activity Integral to plasma membrane Up 20 EIF2S2 Protein biosynthesis DNA binding Ribosome Up 21 SLC31A2 Ion transport Copper ion transporter activity Integral to plasma membrane Up 22 BST2 Humoral immune response Signal transducer activity Integral to plasma membrane Up 23 NRGN Signal transduction Calmodulin binding Up Up 24 INHBC 25 FZD2 Establishment of tissue polarity Non-G-protein coupled 7TM Integral to plasma membrane Up receptor activity 26 IL17R Cell surface receptor linked Receptor activity Integral to plasma membrane Up signal transduction 27 H2AFX Nucleosome assembly DNA binding Nucleosome Up 28 ANKRD11 Nucleus Up 29 RDBP Transcription Nucleotide binding Nucleus Up 30 LHX6 Regulation of transcription, Transcription factor activity Nucleus Up DNA-dependent 31 PSMB3 Ubiquitin-dependent protein Threonine endopeptidase activity Proteasome core complex Up catabolism 32 LEREPO4 Nucleic acid binding Up 33 MCM4 DNA replication DNA binding Nucleus Up Up 34 RECQL4 35 RNASEH2A DNA replication RNA binding Up 36 FKBP6 Protein folding Peptidyl-prolyl cis-trans Up activity 37 CALM1 Up 38 C10orf70 Mitochondrion Up 39 SPR Tetrahydrobiopterin biosynthesis Nitric-oxide synthase activity Up 40 SPHK1 activation Magnesium ion binding Membrane fraction Up 41 PTMA Regulation of cell cycle Nucleus Up Up 42 ENO1 43 RRM1 DNA replication Ribonucleoside-diphosphate Ribonucleoside-diphosphate Up reductase activity reductase 44 WBSCR16 Ran guanyl-nucleotide Up exchange factor activity 45 ITGB1BP1 Cell-matrix adhesion Protein C-terminus binding Membrane Up 46 RAB2 Intracellular protein transport GTP binding Endoplasmic reticulum Up 47 BAG2 Protein folding Unfolded protein binding Up 48 RAB8A Intracellular protein transport GTP binding Up 49 ARHGAP10 GTPase activator activity Up Up 50 RARG-1 51 TI-227H Down 52 LOC94431 Down 53 NPIP Neuropeptide signaling pathway Membrane Down 54 SRRM2 Down 55 BTG2 DNA repair Transcription factor activity Down

399 I.-J.Kim et al.

Table III. Continued

No Gene GO biological process GO molecular function GO cellular component Change symbol description description description

56 LONP ATP-dependent proteolysis Nucleotide binding Down 57 ARIH1 Ubiquitin-dependent protein Ubiquitin-protein ligase activity Ubiquitin ligase complex Down catabolism 58 CASK Protein amino acid Guanylate kinase activity Plasma membrane Down phosphorylation 59 ZNF262 Development DNA binding Nucleus Down 60 SENP7 Proteolysis and peptidolysis Cysteine-type peptidase activity Nucleus Down 61 TNFRSF10B Induction of apoptosis Receptor activity Integral to plasma membrane Down 62 RPP14 TRNA processing RNA binding Nucleus Down 63 IQGAP1 Signal transduction GTPase inhibitor activity Actin filament Down 64 SLC7A6 Protein complex assembly Amino acid-polyamine Plasma membrane Down transporter activity Downloaded from https://academic.oup.com/carcin/article/27/3/392/2476120 by guest on 27 September 2021 65 RPL23 Protein biosynthesis Structural constituent of ribosome Ribosome Down 66 LOC399491 Integral to membrane Down 67 WSB1 Intracellular signaling cascade Down 68 VPS13C Down 69 HOXB9 Regulation of transcription, Transcription factor activity Nucleus Down DNA-dependent 70 WDR42A Down 71 NPIP Neuropeptide signaling pathway Membrane Down 72 EZH1 Transcription Chromatin binding Nucleus Down 73 PUM2 Regulation of translation RNA binding Down 74 DACH1 Transcription DNA binding Nucleus Down 75 MLL Transcription DNA binding Nucleus Down 76 HNRPA3 Nucleotide binding Nucleus Down 77 NCSTN Membrane protein ectodomain Protein binding Endoplasmic reticulum Down proteolysis 78 FOXJ3 Transcription Transcription factor activity Nucleus Down 79 SNX1 Intracellular protein transport Protein transporter activity Golgi apparatus Down 80 SETDB1 Chromatin modification DNA binding Nucleus Down

Up, upregulated in BRAF group; down, downregulated in BRAF group. group clustered to the left and the BRAF group clustered to the our unsupervised clustering results (Figure 1C). Although right (P ¼ 0.002) (Figure 1A). There were a few exceptions to high frequencies of APC and TP53 mutations have been repor- this trend. K-ras sample 218 (G12D) clustered closer to the ted in CRCs (2,4,5), we did not observe any relationship BRAF group. This sample showed no clinicopathological between the clustering data and mutations in APC or TP53. differences with the other K-ras samples; the reason for its While we did observe a tendency for MSI-positive samples to outlier status is not clear, but may be related to the heterogen- cluster to the right along with the BRAF mutant group, this eity of CRCs. More notably, BRAF sample 640 clustered with association was not statistically significant (P ¼ 0.163). Thus, the K-ras samples; sample 640 harbored the BRAF G464V it appears as though the observed associations were related to mutation in exon 11 and is one of two non-V600E samples BRAF or K-ras status rather than to genetic alterations in APC, examined in this study. The second of these, sample 481 TP53 or MSI. (D594G), was found to cluster with another K-ras outlier, We then used MDS to confirm the apparent differences in sample 456. These results seem to suggest that the non- the gene expression patterns between the BRAF and K-ras V600E group was more similar to the K-ras group than the groups and between the V600E and non-V600E BRAF groups V600E group. Furthermore, when we used unsupervised hier- (Figure 2). In contrast to clustering, MDS can uncover multiple archical clustering of the 11 BRAF samples to confirm that the layers of meaning within microarray data, permit sample clas- non-V600E samples had similar gene expression patterns sification in multiple independent dimensions (components) (Figure 1B), samples 481 and 640 clustered together. These and provide a quantitative measure of the sample variance results strongly suggested that the V600E samples could be generated by each component versus that across the entire distinguished from the non-V600E samples based on their dataset (44). Our MDS revealed that while the BRAF and gene expression profiles. Thus, it is possible that drugs or K-ras groups were clearly distinct from each other, K-ras inhibitors targeting BRAF should potentially be divided into sample 218 (previously mentioned as an outlier in the cluster BRAF-V600E- and non-V600E-targeting molecules. The arms analysis) was far from both groups, though more closely in cluster dendrogram between BRAF and K-ras groups are aligned with the BRAF group (Figure 2A). The distinct differ- relatively short and these might have resulted from low sample ences between the BRAF and K-ras groups were even more numbers. Additional studies with larger sample sets will be pronounced when MDS was performed using 2526 probes required to confirm this possibility. identified by t-test (P 5 0.05) as being significantly different To exclude the possibility that our clustering results were between the two groups. Similarly, the BRAF-V600E group affected by genetic alterations in other cancer-related genes, was also clearly distinguished from the non-V600E when we examined the 20 CRC samples for MSI and APC and TP53 MDS was performed using 1703 probes identified by t-test mutations, and then tested these results for associations with (P 5 0.05) as being significantly different between the two 400 Microarray analysis of CRCs with BRAF and K-ras mutations

A B Downloaded from https://academic.oup.com/carcin/article/27/3/392/2476120 by guest on 27 September 2021

Fig. 3. LOOCV analysis. (A) LOOCV results in all 20 BRAF and K-ras group samples. Eighty genes were selected as a classifier with 0.101 ROC error value. All but two samples (218 and 640) were correctly predicted. (B) All 20 samples were analyzed by a hierarchical clustering using 80 genes identified by the LOOCV.

A B

C D

Fig. 4. Real-time qRT–PCR for validating microarray results. Four genes (IL8, MMP1, PTS and TUBA1) that showed significantly different expressions between the BRAF and K-ras groups were used for SYBR green-based real-time quantitative PCR assay. Significantly different expressions between the BRAF and K-ras groups were found in IL8 (A), MMP1 (B), PTS (C) and TUBA1 (D). Statistical P-value was calculated with one-way ANOVA method. 401 I.-J.Kim et al. groups (Figure 2C). We further used hierarchical clustering activity has been shown to inhibit melanoma cell proliferation (supervised) to analyze the 2526 statistically significant probes and metastatic potential (51), MMP1 is expected to be an in the BRAF versus K-ras samples and the 1703 significant important target for BRAF inhibitors. IQGAP1 (IQ motif con- probes differentiating the V600E and non-V559E groups. taining GTPase activating protein 1) was also significantly Throughout these analyses, the BRAF and K-ras groups downregulated in the BRAF group (P 50.001). IQGAP1 was could be distinguished from each other, as could the V600E reported to bind with ERK2, suggesting that it may modulate and non-V600E groups. the RAS–MAPK signaling cascade (52). An LOOCV analysis was performed to validate clustering Having identified the obvious molecular targets for BRAF results. We performed an LOOCV in which one sample is revealed by our expression profiling, we next compared our held, a predictor is trained on the remaining samples, the left differentially expressed genes (P 50.05) with those identified out sample is classified by this predictor and the process by Pavey et al. (34) as being differentially expressed in BRAF is repeated iteratively (http://www.broad.mit.edu/cancer/ mutant and wild-type melanoma cell lines. A number of genes software/software.html). All but two samples were correctly appeared on both lists, including SENP7, SKD3, CDH1, HSF1, predicted by LOOCV (90% accuracy, 18/20). Samples 218 and MBD2, UBE3B, DACH, TIA1, ANXA7 and CLTA. Of these, Downloaded from https://academic.oup.com/carcin/article/27/3/392/2476120 by guest on 27 September 2021 640 were wrongly predicted and these two samples were also HSF1 (heat shock transcription factor 1) was reportedly phos- misclassified in the unsupervised hierarchical clustering phorylated by other members of the MAPK family in a (Figure 1A). The LOOCV data confirmed the unsupervised ras-dependent manner (53). In addition, MBD2 (methyl-CpG hierarchical clustering data and suggested that gene expression binding domain protein 2) is involved in the methylator pheno- profiles of the BRAF and K-ras groups could be distinguished types of many genes (54). As BRAF is associated with MLH1 from each other. We then examined the differentially methylation and methylator phenotypes (17–20), MBD2 expressed probes to identify possible molecular targets for appears to be a likely candidate for a BRAF-related epigenetic BRAF or BRAF-related pathways. PAM analysis identified target. Out of 98 genes from PAM and 80 genes from LOOCV, 123 probes capable of classifying the BRAF and K-ras groups. only 24 genes were overlapped. The low numbers of over- Exclusion of the redundant and hypothetical genes revealed a lapped genes between PAM and LOOCV may be due to the total of 98 possible classifier genes, including several reported differences of data analysis algorithm in these two methods. to play important roles in the RAS–RAF–MEK–ERK–MAPK In sum, we herein showed that in sporadic CRCs, the gene pathway, including IL-8, TGFBR2, SPRED2, MMP1 and expression profiles of the BRAF and K-ras groups could be IQGAP1. IL-8, which was generally upregulated in the BRAF distinguished from each other, as could those of the BRAF- group (P 50.001), was reportedly upregulated by activation of V600E and non-V600E groups. Although the sample size is RAF, MEK and ERK (45), and has been identified as a tran- limited in this work, the identified expression patterns and scriptional target of RAS-RAF signaling, suggesting that it may gene sets will hopefully form the basis for future development be involved with BRAF (46). IL-8 secretion was required for of molecular targets for BRAF. the initiation of tumor-associated inflammation and neovascu- larization (47), and a previous study suggested the possibility of treating melanoma patients with a combination of inhibitors Acknowledgements against IL-8 and MEK (45). This work was supported by a research grant from the National Cancer Center, TGFBR2 (transforming growth factor, beta receptor II) was Korea and the BK21 project for Medicine, Dentistry and Pharmacy. significantly downregulated in the BRAF group (P ¼ 0.003). TGFBR2 frameshift mutations are commonly found in MSI Conflict of Interest Statement: None declared. CRC samples. Notably, TGFBR2 was significantly downregu- lated in MSI-H CRCs, while IL-8, IL-1b, ICAM1 and CD68 were all reportedly upregulated (48). These findings were References completely concordant with our findings in the BRAF group. 1.Shin,H.R., Jung,K.W., Won,Y.J. and Park,J.G. (2004) 2002 Annual report Another interesting gene is SPRED2 (sprouty-related, EVH1 of the Korea Central Cancer Registry: based on registered data from 139 domain containing 2, P 5 0.001), which was significantly hospitals. Cancer Res. Treat., 36, 103–114. downregulated in the BRAF group and upregulated in the 2.Fearon, E.R. and Vogelstein, B. (1990) A genetic model for colorectal K-ras group. SPREDs have been shown to inhibit ERK activa- tumorigenesis. Cell, 61, 759–767. 3.Forrester,K., Almoguera,C., Han,K., Grizzle,W.E. and Perucho,M. (1987) tion, and researchers have suggested that SPRED induction Detection of high incidence of K-ras oncogenes during human colon could be a novel strategy for preventing cancer cell metastasis tumorigenesis. Nature, 327, 298–303. (49). Furthermore, SPREDs modulate the RAS–RAF inter- 4.Rodrigues,N.R., Rowan,A., Smith,M.E., Kerr,I.B., Bodmer,W.F., action and MAPK signaling; the latter effect may be due to Gannon,J.V. and Lane,D.P. (1990) p53 mutations in colorectal cancer. Proc. Natl Acad. Sci. USA, 87, 7555–7559. suppression of RAF phosphorylation (50). 5.Miyoshi,Y., Nagase,H., Ando,H., Horii,A., Ichii,S., Nakatsuru,S., Aoki,T., Another gene of interest is MMP1 (matrix metalloproteinase Miki,Y., Mori,T. and Nakamura,Y. (1992) Somatic mutations of the APC 1), which was upregulated in the BRAF group. MMP1 expres- gene in colorectal tumors: mutation cluster region in the APC gene. Hum. sion is known to increase during melanoma progression, and Mol. Genet., 1, 229–233. has been associated with shorter disease-free survival (51). 6.Fishel,R., Lescoe,M.K., Rao,M.R., Copeland,N.G., Jenkins,N.A., Garber,J., Kane,M. and Kolodner,R. (1993) The human mutator gene Furthermore, high MMP1 levels in melanoma have been asso- homolog MSH2 and its association with hereditary nonpolyposis colon ciated with the constitutive activation of the RAS–RAF– cancer. Cell, 75, 1027–1038. MEK–ERK pathway due to activating BRAF mutations (51). 7.Bronner,C.E., Baker,S.M., Morrison,P.T. et al. (1994) Mutation in the Consistent with these findings, we observed that MMP1 was DNA mismatch repair gene homologue hMLH1 is associated with hereditary non-polyposis colon cancer. Nature, 368, 258–261. more highly expressed in the BRAF mutant group than in the 8.Kim,I.J., Kang,H.C., Park,J.H., Shin,Y., Ku,J.L., Lim,S.B., Park,S.Y., K-ras group (P ¼ 0.006). As BRAF mutations are found Jung,S.Y., Kim,H.K. and Park,J.G. (2003) Development and applications in 80% of melanoma cases, and blockade of MEK–ERK of a beta-catenin oligonucleotide microarray: beta-catenin mutations are 402 Microarray analysis of CRCs with BRAF and K-ras mutations

dominantly found in the proximal colon cancers with microsatellite 30.Mariadason,J.M., Arango,D., Shi,Q., Wilson,A.J., Corner,G.A., instability. Clin. Cancer Res., 9, 2920–2925. Nicholas,C., Aranes,M.J., Lesser,M., Schwartz,E.L. and Augenlicht,L.H. 9.Park,J.H., Kim,I.J., Kang,H.C., Shin,Y., Park,H.W., Jang,S.G., Ku,J.L., (2003) Gene expression profiling-based prediction of response of colon Lim,S.B., Jeong,S.Y. and Park,J.G. (2004) Oligonucleotide carcinoma cells to 5-fluorouracil and camptothecin. Cancer Res., 63, microarray-based mutation detection of the K-ras gene in colorectal 8791–8812. cancers with use of competitive DNA hybridization. Clin. Chem., 50, 31.Hedenfalk,I., Duggan,D., Chen,Y. et al. (2001) Gene-expression profiles 1688–1691. in hereditary breast cancer. N. Engl. J. Med., 344, 539–548. 10.Davies,H., Bignell,G.R., Cox,C. et al. (2002) Mutations of the BRAF gene 32.Hedenfalk,I., Ringner,M., Ben-Dor,A. et al. (2003) Molecular classifica- in human cancer. Nature, 417, 949–954. tion of familial non-BRCA1/BRCA2 breast cancer. Proc. Natl Acad. Sci. 11.Yuen,S.T., Davies,H., Chan,T.L. et al. (2002) Similarity of the phenotypic USA, 100, 2532–2537. patterns associated with BRAF and KRAS mutations in colorectal 33.Dunican,D.S., McWilliam,P., Tighe,O., Parle-McDermott,A. and neoplasia. Cancer Res., 62, 6451–6455. Croke,D.T. (2002) Gene expression differences between the microsatellite 12.Rajagopalan,H., Bardelli,A., Lengauer,C., Kinzler,K.W., Vogelstein,B. instability (MIN) and chromosomal instability (CIN) phenotypes in and Velculescu,V.E. (2002) Tumorigenesis: RAF/RAS oncogenes and colorectal cancer revealed by high-density cDNA array hybridization. mismatch-repair status. Nature, 418, 934. Oncogene, 21, 3253–3257. 13.Kolch,W. (2000) Meaningful relationships: the regulation of the 34.Pavey,S., Johansson,P., Packer,L. et al. (2004) Microarray expression Ras/Raf/MEK/ERK pathway by protein interactions. Biochem, J., 351, profiling in melanoma reveals a BRAF mutation signature. Oncogene, 23, Downloaded from https://academic.oup.com/carcin/article/27/3/392/2476120 by guest on 27 September 2021 289–305. 4060–4067. 14.Pollock,P.M., Harper,U.L., Hansen,K.S. et al. (2003) High frequency of 35.Won,Y.J., Park,K.J., Kwon,H.J., Lee,J.H., Kim,J.H., Kim,Y.J., Chun,S.H., BRAF mutations in nevi. Nat. Genet., 33, 19–20. Han,H.J. and Park,J.G. (1999) Germline mutations of the APC gene in 15.Singer,G., Oldt,R.,III, Cohen,Y., Wang,B.G., Sidransky,D., Kurman,R.J. Korean familial adenomatous polyposis patients. J. Hum. Genet., 44, and Shih,IeM. (2003) Mutations in BRAF and KRAS characterize the 103–108. development of low-grade ovarian serous carcinoma. J. Natl Cancer Inst., 36.Bakkar,A.A., Wallerand,H., Radvanyi,F. et al. (2003) FGFR3 and TP53 95, 484–486. gene mutations define two distinct pathways in urothelial cell carcinoma 16.Kim,I.J., Park,J.H., Kang,H.C., Shin,Y., Park,H.W., Park,H.R., Ku,J.L., of the bladder. Cancer Res., 63, 8108–8112. Lim,S.B. and Park,J.G. (2003) Mutational analysis of BRAF and K-ras in 37.Kim,I.J., Kang,H.C., Shin,Y. et al. (2004) A TP53-truncating germline gastric cancers: absence of BRAF mutations in gastric cancers. Hum. mutation (E287X) in a family with characteristics of both hereditary Genet., 114, 118–120. diffuse gastric cancer and Li-Fraumeni syndrome. J. Hum. Genet., 49, 17.Wang,L., Cunningham,J.M., Winters,J.L. et al. (2003) BRAF mutations in 591–595. colon cancer are not likely attributable to defective DNA mismatch repair. 38.Kim,I.J., Shin,Y., Kang,H.C. et al. (2003) Robust microsatellite instability Cancer Res., 63, 5209–5212. (MSI) analysis by denaturing high-performance liquid chromatography 18.Domingo,E., Laiho,P., Ollikainen,M. et al. (2004) BRAF screening as a (DHPLC). J. Hum. Genet., 48, 525–530. low-cost effective strategy for simplifying HNPCC genetic testing. J. Med. 39.Kang,H.C., Kim,I.J., Park,J.H., Shin,Y., Ku,J.L., Jung,M.S., Yoo,B.C., Genet., 41, 664–668. Kim,H.K. and Park,J.G. (2004) Identification of genes with 19.Deng,G., Bell,I., Crawley,S., Gum,J., Terdiman,J.P., Allen,B.A., Truta,B., differential expression in acquired drug-resistant gastric cancer cells Sleisenger,M.H. and Kim,Y.S. (2004) BRAF mutation is frequently using high-density oligonucleotide microarrays. Clin. Cancer Res., 10, present in sporadic colorectal cancer with methylated hMLH1, but not in 272–284. hereditary nonpolyposis colorectal cancer. Clin. Cancer Res., 10, 40.Eisen,M.B., Spellman,P.T., Brown,P.O. and Botstein,D. (1998) Cluster 191–195. analysis and display of genome-wide expression patterns. Proc. Natl Acad. 20.Nagasaka,T., Sasamoto,H., Notohara,K. et al. (2004) Colorectal cancer Sci. USA, 95, 14863–14868. with mutation in BRAF, KRAS, and wild-type with respect to both 41.Korkola,J.E., DeVries,S., Fridlyand,J., Hwang,E.S., Estep,A.L., oncogenes showing different patterns of DNA methylation. J. Clin. Chen,Y.Y., Chew,K.L., Dairkee,S.H., Jensen,R.M. and Waldman,F.M. Oncol., 22, 4584–4594. (2003) Differentiation of lobular versus ductal breast carcinomas by 21.Kumar,R., Angelini,S., Czene,K., Sauroja,I., Hahka-Kemppinen,M., expression microarray analysis. Cancer Res., 63, 7167–7175. Pyrhonen,S. and Hemminki,K. (2003) BRAF mutations in metastatic 42.Zhao,W., Chan,T.L., Chu,K.M., Chan,A.S., Stratton,M.R., Yuen,S.T. and melanoma: a possible association with clinical outcome. Clin. Cancer Leung,S.Y. (2004) Mutations of BRAF and KRAS in gastric cancer Res., 9, 3362–3368. and their association with microsatellite instability. Int. J. Cancer, 108, 22.Peggs,K. (2004) Imatinib mesylate—gold standards and silver linings. 167–169. Clin. Exp. Med., 4, 1–9. 43.Ikenoue,T., Hikiba,Y., Kanai,F. et al. (2004) Rapid detection of mutations 23.Cripps,M.C., Figueredo,A.T., Oza,A.M., Taylor,M.J., Fields,A.L., in the BRAF gene using real-time polymerase chain reaction and melting Holmlund,J.T., McIntosh,L.W., Geary,R.S. and Eisenhauer,E.A. (2002) curve analysis. Cancer Genet. Cytogenet., 149, 68–71. Phase II randomized study of ISIS 3521 and ISIS 5132 in patients with 44.Mori,Y., Selaru,F.M., Sato,F. et al. (2003) The impact of microsatellite locally advanced or metastatic colorectal cancer: a National Cancer instability on the molecular phenotype of colorectal tumors. Cancer Res., Institute of Canada clinical trials group study. Clin. Cancer Res., 8, 63, 4577–4582. 2188–2192. 45.Lev,D.C., Onn,A., Melinkova,V.O., Miller,C., Stone,V., Ruiz,M., 24.Coudert,B., Anthoney,A., Fiedler,W. et al. (2001) Phase II trial with ISIS McGary,E.C., Ananthaswamy,H.N., Price,J.E. and Bar-Eli,M. (2004) 5132 in patients with small-cell (SCLC) and non-small cell (NSCLC) lung Exposure of melanoma cells to dacarbazine results in enhanced tumor cancer. A European Organization for Research and Treatment of Cancer growth and metastasis in vivo. J. Clin. Oncol., 22, 2092–2100. (EORTC) Early Clinical Studies Group report. Eur. J. Cancer, 37, 46.Knall,C., Young,S., Nick,J.A., Buhl,A.M., Worthen,G.S. and 2194–2198. Johnson,G.L. (1996) Interleukin-8 regulation of the Ras/Raf/mitogen- 25.Schena,M., Shalon,D., Davis,R.W. and Brown,P.O. (1995) Quantitative activated protein kinase pathway in human neutrophils. J. Biol. Chem., monitoring of gene expression patterns with a complementary DNA 271, 2832–2838. microarray. Science, 270, 467–470. 47.Sparmann,A. and Bar-Sagi,D. (2004) Ras-induced interleukin-8 expres- 26.DeRisi,J., Penland,L., Brown,P.O., Bittner,M.L., Meltzer,P.S., Ray,M., sion plays a critical role in tumor growth and angiogenesis. Cancer Cell, 6, Chen,Y., Su,Y.A. and Trent,J.M. (1996) Use of a cDNA microarray 447–458. to analyse gene expression patterns in human cancer. Nat. Genet., 14, 48.Banerjea,A., Ahmed,S., Hands,R.E., Huang,F., Han,X., Shaw,P.M., 457–460. Feakins,R., Bustin,S.A. and Dorudi,S. (2004) Colorectal cancers with 27.Ramaswamy,S. and Golub,T.R. (2002) DNA microarrays in clinical microsatellite instability display mRNA expression signatures character- oncology. J. Clin. Oncol., 20, 1932–1941. istic of increased immunogenicity. Mol. Cancer, 3, 21. 28.Golub,T.R., Slonim,D.K., Tamayo,P. et al. (1999) Molecular classification 49.Miyoshi,K., Wakioka,T., Nishinakamura,H., Kamio,M., Yang,L., of cancer: class discovery and class prediction by gene expression Inoue,M., Hasegawa,M., Yonemitsu,Y., Komiya,S. and Yoshimura,A. monitoring. Science, 286, 531–537. (2004) The Sprouty-related protein, Spred, inhibits cell motility, 29.Kitahara,O., Katagiri,T., Tsunoda,T., Harima,Y. and Nakamura,Y. (2002) metastasis, and Rho-mediated actin reorganization. Oncogene, 23, Classification of sensitivity or resistance of cervical cancers to ionizing 5567–5576. radiation according to expression profiles of 62 genes selected by cDNA 50.Wakioka,T., Sasaki,A., Kato,R., Shouda,T., Matsumoto,A., microarray analysis. Neoplasia, 4, 295–303. Miyoshi,K., Tsuneoka,M., Komiya,S., Baron,R. and Yoshimura,A.

403 I.-J.Kim et al.

(2001) Spred is a Sprouty-related suppressor of Ras signalling. Nature, 53.Kim,J., Nueda,A., Meng,Y.H., Dynan,W.S., Mivechi,N.F. (1997) Analysis 412, 647–651. of the phosphorylation of human heat shock transcription factor-1 by MAP 51.Huntington,J.T., Shields,J.M., Der,C.J., Wyatt,C.A., Benbow,U., kinase family members. J. Cell Biochem., 67, 43–54. Slingluff,C.L.Jr and Brinckerhoff,C.E. (2004) Overexpression of colla- 54.Magdinier,F. and Wolffe,A.P. (2001) Selective association of the genase 1 (MMP-1) is mediated by the ERK pathway in invasive melanoma methyl-CpG binding protein MBD2 with the silent p14/p16 locus cells: role of BRAF mutation and fibroblast growth factor signaling. in human neoplasia. Proc. Natl Acad. Sci. USA, 98, 4990–4995. J. Biol. Chem., 279, 33168–33176. 52.Roy,M., Li,Z. and Sacks,D.B. (2004) IQGAP1 binds ERK2 and modulates Received June 3, 2005; revised August 15, 2005; its activity. J. Biol. Chem., 279, 17329–17337. accepted September 28, 2005 Downloaded from https://academic.oup.com/carcin/article/27/3/392/2476120 by guest on 27 September 2021

404