Genomic events associated with progression of pleural malignant mesothelioma

INTRODUCTION

Malignant mesothelioma (MM) of the pleura is an aggressive cancer often associated with exposure to asbestos. MM has a very long median latency of 43.6 years ¹ and a very short median survival of approximately 11 months ². MM cytogenetics reveals a highly variable karyotype with multiple rearrangements. The most frequents events observed in MM are deletions in 1p21-22, 3p21, 4, 6q14-25, 9p21, 13q, 14q, 15q15, 17p13, and 22 ³. Monosomy 22 is the most frequent numerical change. A more detailed insight into chromosomal rearrangements associated with cancer became possible with the advent of the comparative genomic hybridization (CGH) fluorescent technology that permits high resolution analysis of DNA copy number variation ⁴5. CGH analysis of cell lines and tumor specimens derived from MM patients has confirmed previous karyotypic abnormalities and has revealed other genomic alterations, such as losses of 15q11.1-15, 14q24.2-qter, and 13q12-14, as well as gains on 5p ⁶7. Recently, the development of microarray-based versions of CGH has greatly increased resolution and throughput, enabling the analysis of genomic imbalances of less than 100 kb in size ⁸. Representational oligonucleotide microarray analysis (ROMA) is one such technology, which uses complexity reducing representations to increase probe signal to noise for the comparison of tumor to normal cells ⁹. In this study, we report the results of ROMA analysis performed on 22 tumors from MM patients who had surgery for the disease, with the goal to define chromosomal regions affected during disease progression.

MATERIALS AND METHODS

Mesothelioma patients and specimens

Patients having resection of mesothelioma by the senior author (HIP) while he was at Wayne State University, Detroit Michigan from 1998-2005 gave written consent for procurement of tumor and normal peritoneum, as approved by the Wayne State University Institutional Review Board (D1420: Collection of Serum and Tissue Samples for Patients with Biopsy-Prove or Suspected Malignant Disease). These specimens upon transfer to NYU School of Medicine through a cooperative materials transfer agreement were approved for discovery of prognostic and early detection biomarkers by the NYU School of Medicine Institutional Review Board. 27 snap-frozen specimens (22 tumors and 5 unaffected peritonea harvested from the peritoneum of patients undergoing extrapleural pneumonectomy) derived from 22 patients with MM were used to produce 27 DNA samples for ROMA analysis. 4 out of 5 normal samples were tumor-matched. As seen in Table 1, 9 patients showed disease longer than 12 months after surgery, and 13 progressed within 1 year of surgery.

DNA isolation and analysis

DNA from snap-frozen tissues was isolated with QIAamp mini-kit (QIAGEN, Valencia, CA) and assayed for concentration and quality using Victor³ microplate reader (PerkinElmer, Waltham, MA) and agarose electrophoresis. BglII genomic representations, and Cy3-dCTP (reference DNA) or Cy5-dCTP (DNA of interest) incorporation, hybridization, and washing conditions were done as described recently ¹⁰. Hybridizations were carried out on arrays bearing 85,000 oligonucleotides (NimbleGen, Systems, Madison, WI). Slides were scanned with an AxonGenePix 4000B scanner (Axon Instruments, Union City, CA). Enzymes used in the study were BglII and T4 DNA ligase (New England Biolabs, Ipswich, MA). Primers were supplied by Sigma Genosys (St. Louis, MO). Cot1 DNA and tRNA were obtained from Invitrogen (Carlsbad, CA). The Megaprime labeling kit, Cy3-conjugated dCTP, and Cy5-conjugated dCTP were purchased from Amersham Biosciences (GE Healthcare, Piscataway, NJ). Taq polymerase Mastermix was supplied by Eppendorf (Westbury, NY). NimbleGen photoprint arrays were synthesized by NimbleGen Systems Inc. (Madison, WI).

ROMA

Arrays were described previously ¹⁰. In brief, the arrays are based on representational techniques and, thus, all oligonucleotides map to BglII fragments that are within the representations size range of 200-1000 bp. The array was designed to NCBI Build 30 and the coordinates have been updated to NCBI Build 35. Sample preparation, BglII representations, has been described previously ¹¹. DNA was labeled as described with minor changes ¹¹. Procedures for hybridization were followed as reported previously ¹². Arrays were scanned with an Axon GenePix 4000B scanner set at a pixel size of 5 μm. GenePix Pro 4.0 software was used to quantify the intensity for the arrays. Array data were imported into S-PLUS for further analysis. Measured intensities without background subtraction were used to calculate ratios. Data were normalized using an intensity-based lowest curve fitting algorithm.

ROMA data processing

Signal intensity ratios were normalized against standard ROMA control (normal DNA specimen isolated from human skin fibroblasts) for each of 27 probes (including 5 normal pleura specimens) used in the study. The data file was imported into CGH-Explorer v. 2.55 available from http://www.ifi.uio.no/forskning/grupper/bioinf/Papers/CGH/ , log2-transformed, meancentered, and used for visualization and statistical analysis. Statistical evaluation was done using the ACE (analysis of copy errors) algorithm available with the software ¹³. To control the error of multiple comparison without losing discovery power, the expected proportion of false discoveries was estimated using the false discovery rate (FDR) ¹⁴.

Affymetrix genome analysis

In the validation studies, Affymetrix GeneChip® Human Mapping 250K Nsp arrays containing ∼262,000 SNPs and Genome-Wide Human SNP Arrays 6.0 containing more than 906,600 single nucleotide polymorphisms (SNPs) and more than 946,000 probes for the detection of copy number variation were used as the hybridization targets. The array probe datasheets can be found at the web sites http://www.affymetrix.com/support/technical/datasheets/500k_datasheet.pdf and http://www.affymetrix.com/support/technical/datasheets/genomewide_snp6_datasheet.pdf, respectively. Briefly, total genomic DNA (250-500 ng) from each test sample was digested with NspI or StyI restriction enzyme and ligated to an adapter that recognizes cohesive 4-basepair (bp) overhangs. A generic primer that recognizes the adapter sequence was used to amplify the adapter-ligated DNA fragments. PCR conditions were optimized to preferentially amplify fragments in the 200- to 1,100-bp size range. After purification using Microcon Ultracel filters, the amplified DNA was fragmented with DNase I, biotin-labeled, and hybridized to the GeneChip according to the manufacturer’s recommendations. For the 250K NspI arrays, the intensities of probe hybridization were analyzed using Affymetrix GCOS 1.4 software, and the genotyping was performed using GTYPE 4.1 with Dynamic Model Mapping Analysis by default settings at 0.33 for both homozygote and heterozygote call thresholds. Copy number analyses were carried out using Affymetrix Chromosome Copy Number Analysis Tool (CNAT) version 4.0.1. For the 6.0 arrays, the genotyping and copy number analyses were performed using Affymetrix Genotyping Console 2.1 with Birdseed genotype calling algorithm and Chromosome Copy Number Analysis Tool (CNAT) version 5.

Gene expression profiling

RNA from 22 tumor specimens used in our ROMA study were isolated and analyzed for differential expression as described ¹⁵ using the Affymetrix U133A platform. For independent validation of candidate genes with regard to their expression, we used an additional, completely different set of MM (n=30) and normal pleura (n=7) specimens for comparative gene expression study on the U133 2.0 Plus platform.

RT-PCR analysis

RT-PCR premixes were prepared using a Super Script One-Step RT-PCR kit (Invitrogen, Carlsbad, CA). Primers for RT-PCR assessment of gene expression were designed as described ¹⁶. Conditions for the reaction were as follows: 50°C — 30 min., 35 cycles of (94°C — 15s, 56°C- 15s, 72°C — 1 min.), 72°C — 5 min. Oligonucleotide sequences for RT-PCR were designed to ensure RNA-specific synthesis across introns: OSM: 5′-CTTGGAGAAGCTGCAGATGG-3′, 5′-AGCCTCTAACTCCCTAGCTTC-3′, FUS1: 5′- AACTCCCAGGCTCAATCAAG-3′, 5′-CAGACTCTGCCACGACATC-3′; PL6: 5′-GCTTGACTCGGGTACAGAAC-3′, -5′-CTATGGCGCTGGTAGAAGC-3′; CDH2: 5′- CTGCTTCAGGCGTCTGTAG-3′, 5′-ATTGCCTTCCATGTCTGTAG-3′; DNAJA1: 5′-CGGGTTCGGCTACAAAAGAG-3′, 5′-GATGACGATGGTTCGGTTGTC-3′; MKI67: 5′-GATGTGGAAGTTCTGCCTACG-3′, 5′-GCGGTTGCTCCTTCACTG-3′. RT-PCR with primers to invariantly expressed PPIA (5′-TCTGAGCACTGGAGAGAAAGG-3′, 5′-GGAAAACATGGAACCCAAAGG-3′) was used as a loading control. RT-PCR was performed on a set of matched normal-tumor specimens from 8 MM patients. Band intensities were measured using Kodak 4000 Image Station. Repeated experiments showed consistency of our RT-PCR analysis.

RESULTS

Copy number alterations in patients with long-and short-term recurrences

MM patients used for ROMA were represented by two groups: the one with short-term recurrence after surgery (less than 12 months, STR) and the other one with longer time to relapse (LTR). Five normal tissue specimens were assembled into the third group (NORM). The total number of events for each specimen was calculated using the ACE algorithm available from the CGH-Explorer package ¹³. Table 1 shows demographics of the patients as well as total CNA counts (gains and losses estimated by the total amount of differentially hybridized probes) for each tumor specimen with the false discovery rate of 0.023 (∼2% false positive discoveries). In both tumor cohorts combined we observed a ∼10-fold increase in CNA as compared to the controls (means 5042 and 519 oligonucleotide probes, respectively, p≤0.0004). The LTR and STR groups differed significantly (Fig. 1A, respective means 1864 and 7242, p<0.05) and showed a statistically significant positive correlation between the groups and total CNAs (Fig. 1B, p=0.0014; r=0.58), implying that the adverse development of the disease may be linked to increase in chromosomal aberrations. A multivariate analysis was performed in order to distinguish whether CNA was an independent predictor of survival compared to other demographics including age, stage, sex, asbestos exposure, platelet count, and histology of the mesothelioma. Two models were considered: one in which stage was used as one of the variables and another in which stage was not. The second model was explored because in a number of cases where the mesothelioma would not have surgical therapy as in all of these cases, the stage would be unknown. We found that when stage was incorporated into the multivariate analysis it was by far the most important independent predictor of survival (p=0.0004) with a 3.6 fold increase in relative risk of death for patients with Stages III/IV. In this model, platelets, histology, gender, age, histology, and a copy number abnormality greater than 5000 were not significant. However, in the model where stage is not incorporated, the CNA >5000 and platelet count > 350 were independent predictors of survival (p=0.01 and p=0.003, respectively) with a relative risk of death 3.7 fold and 6.9 fold higher than for lower CNAs and platelet counts. Age, gender, histology, and lymph node status were not significant factors.

Fig. 1

Accumulation of CNAs in MM

Differentiation of three groups of specimens by total CNA number (A) and by correlation analysis (B).

Mapping of most frequent chromosomal alterations and defining their minimal common regions

To increase sensitivity of the analysis the combined ROMA data for all 22 tumor specimens were plotted across human chromosomes 1-22 (Fig. 2). The resulting ACE profile was consistent with chromosomal instability (CIN) that affects most chromosomes. Some chromosomal regions, however, were more frequently involved than others as shown by prominent peaks in red (gains) and green (losses). Losses in chromosomes 22q12.2, 19, and 1p36 appeared to be the most frequent events (55-74%) in pleural MM followed by gains in 5p14-p15 and deletions in 17p and 9q (41-46%). Deletions in 1p13-21, 9p21, 3p14-21, and 17q21 as well as gains in 8q and 18q were observed in 27-36% of cases. Minimal common areas (MCAs) for these recurrent events were established through overlapping of individual deletions or gains (Fig. 3 and Table 2). Genes residing in the MCAs were identified using probe coordinates and Human Genome Database (NCBI Builds 36.2 and 35).

Fig. 2

Chromosomal localization and frequencies of imbalances in 22 MM tumors as assessed with ACE (FDR=0.023)

Losses in tumor versus normal control are shown in green and gains are in red. Sex chromosomes are omitted from the study.

Fig. 3

Fine mapping of losses in 9p21.3 (A) and gains in 18q12.1 (B) regions

A, From top to bottom: STR specimens R140, R249, R367, R312, R318, R342, and R142 show overlapping deletions (in green). Minimal common area (MCA) and genes located therein are shown. B, STR specimens R249, R367, R312, R318, R342 and R142, and LTR specimens R219 and R298 show overlapping gains (in red). LTR sample R140 shows loss in the same region, depicted in green.

Cluster analysis of the STR and LTR group

The dramatic difference in the total number of genomic events observed between the STR and LTR groups prompted us to examine the relationship between and within these groups in more detail. Using ACE with an FDR of 0.023, we were able to isolate 85 probes, which discriminated significantly (p<0.01) between the STR and LTR groups and used them for cluster analysis (Fig. 4). The resulting dendrogram produced a clear distinction between the STR and LTR groups with only one patient, R166, misplaced. Inside the groups two samples, R158 and R258, did not cluster tightly within the respective group members. Notably, these two patients appear to be the only cases without a convincing history of asbestos exposure (Table 1). No discrimination between biphasic and epithelial types of MM was observed, and R125, the only sarcomatoid tumor in the study, clustered loosely with the rest of the group (Fig. 4).

Fig. 4

Hierarchical clustering of MM specimens based on 85 probes that distinguish between STR and LTR

Gene expression studies identify OSM, FUS1, PL6, CDH2, and DNAJA1 as differentially expressed during MM progression

Loss of chromosome 22, by far the most frequent event in our study (∼74%, Table 2), was observed in both LTR and STR groups, suggesting that this is a common event in MM tumorigenesis. The MCA defined for 22q12.2 spans ∼2 Mb and contains the oncostatin M gene (OSM), which encodes a proliferation-inhibiting cytokine. To find out if this gene’s activity is related to MM progression, we performed RT-PCR analysis on additional tumor and matched normal peritoneum specimens derived from 8 patients (Fig. 5A). OSM expression was seen in all normal samples and was not found in 5 tumor samples suggesting that it is, indeed, frequently suppressed in MM. Affymetrix expression array studies on an independent cohort of 30 MM and 7 normal peritoneum samples also confirmed our observation (Fig. 5B).

Fig. 5

Validation of gene candidates by RT-PCR (A) and Affymetrix expression analysis (B)

A, Matched normal (N) and tumor (T) specimens derived from eight MM patients were used for RT-PCR. MKI67, encoding Ki-67, was used as a proliferation marker and invariantly expressed PPIA as a loading control. B, Expression array validation on an additional set of MM (n=30) and normal pleura (n=7) specimens for on the U133 2.0 Plus platform.

We also analyzed the 3p21.3 chromosomal area, which showed loss in 32% of cases. Our interest to this area was stimulated by its association with lung and other common cancers ¹⁷18. We reported previously that loss in this area, although less precisely determined, can be observed in MM at ∼40% frequency. In this study, we were able to discriminate between the 3p21.31 (∼2.2 Mb) and 3p14.3-p14.2 (∼1.1 Mb) areas and report that at least two candidate tumor suppressor genes from the 3p21.3 region, FUS1(TUSC2) and PL6(TMEM115), were ∼2-fold down-regulated in 6 out of 8 tumornormal pleura matched specimens (Fig. 5A). Immuhnohistochemistry for the FUS1 product using a novel antibody confirmed down-regulation of the gene (courtesy Ignatio Wistuba, MD Anderson Cancer Center, Houston, Texasm, data not shown, manuscript in preparation).

Among the recurrent gains that we were able to identify in our series, the one in 18q12.1 contained the cadherin N gene (CDH2). Since cadherins play important roles in tumor progression, we studied CDH2 expression using the same approach as for OSM. Up-regulation of CDH2 was observed in a substantial fraction of additional sets of tumors (38%) by both RT-PCR and Affymetrix assays (Fig. 5AB) in agreement with the estimated frequency for this genomic gain (∼36%, Table 2). 6 out of 8 ROMA specimens that showed the CDH2 gain belonged to the STR group (Fig. 3B).

We also focused on DNAJA1, a member of the Hsp40 family of heat shock proteins. This gene is located in the smallest deletion detected in our study 9p21.1 microdeletion area (pos.32780-33292 kb, Fig. 6A) that was established as the best discriminator between LTR and STR in a probe-by-probe statistical evaluation (p<0.007). We have chosen DNAJA1 as a cancer-related candidate since another gene from the same family, DNAJB4, was recently extensively studied in relation to non-small cell lung carcinoma and proved to be a tumor suppressor that shows LOH and transcriptional down-regulation in tumors ¹⁹. Using RNA samples isolated from tumor specimens used for ROMA we found that DNAJA1 showed reduced expression in the STR cohort, which correlated with its allelic loss (4 out of 5 specimens with the deletions also showed down-regulation or lack of gene expression) pointing at the possible link between gene silencing and adverse clinical outcome (Fig. 6B). Affymetrix expression array analysis performed on the same set of tumor specimens (Fig. 6C) matched perfectly with the RT-PCR assessment.

Fig. 6

Association of DNAJA1 with STR MM by array-CGH (A), RT-PCR on ROMA specimens (B), and Affymetrix expression study (C)

DNAJA1 location in the 9p21.3 MCA is shown. Deletions in this area are shown in green, and the respective samples on the gel are marked with “d”.

Copy number alterations in patients with long-and short-term recurrences

MM patients used for ROMA were represented by two groups: the one with short-term recurrence after surgery (less than 12 months, STR) and the other one with longer time to relapse (LTR). Five normal tissue specimens were assembled into the third group (NORM). The total number of events for each specimen was calculated using the ACE algorithm available from the CGH-Explorer package ¹³. Table 1 shows demographics of the patients as well as total CNA counts (gains and losses estimated by the total amount of differentially hybridized probes) for each tumor specimen with the false discovery rate of 0.023 (∼2% false positive discoveries). In both tumor cohorts combined we observed a ∼10-fold increase in CNA as compared to the controls (means 5042 and 519 oligonucleotide probes, respectively, p≤0.0004). The LTR and STR groups differed significantly (Fig. 1A, respective means 1864 and 7242, p<0.05) and showed a statistically significant positive correlation between the groups and total CNAs (Fig. 1B, p=0.0014; r=0.58), implying that the adverse development of the disease may be linked to increase in chromosomal aberrations. A multivariate analysis was performed in order to distinguish whether CNA was an independent predictor of survival compared to other demographics including age, stage, sex, asbestos exposure, platelet count, and histology of the mesothelioma. Two models were considered: one in which stage was used as one of the variables and another in which stage was not. The second model was explored because in a number of cases where the mesothelioma would not have surgical therapy as in all of these cases, the stage would be unknown. We found that when stage was incorporated into the multivariate analysis it was by far the most important independent predictor of survival (p=0.0004) with a 3.6 fold increase in relative risk of death for patients with Stages III/IV. In this model, platelets, histology, gender, age, histology, and a copy number abnormality greater than 5000 were not significant. However, in the model where stage is not incorporated, the CNA >5000 and platelet count > 350 were independent predictors of survival (p=0.01 and p=0.003, respectively) with a relative risk of death 3.7 fold and 6.9 fold higher than for lower CNAs and platelet counts. Age, gender, histology, and lymph node status were not significant factors.

Fig. 1

Accumulation of CNAs in MM

Differentiation of three groups of specimens by total CNA number (A) and by correlation analysis (B).

Validation of ROMA data

The capability of ROMA to detect copy number alterations in individual tumor specimens was validated using Affymetrix 250K NspI arrays. Five out of 22 DNA specimens used for ROMA were randomly chosen for a blinded CGH experiment. The ROMA and Affymetrix data were then compared sample-by-sample using moving average graphical output. The comparison showed a remarkable consistency in classification and mapping of the vast majority of the genomic imbalances, confirming the accuracy and reliability of ROMA (Suppl. Table 1). Among the five validated samples, several recurrent alterations were identified including losses in 1p and 9p (R342, R318, and R142), 3p (R342, R318), and 17pter (R318, R142, R342). All of these losses were classified by ROMA/ACE as highly frequent in our series of 22 cases (Table 2).

Mapping of most frequent chromosomal alterations and defining their minimal common regions

To increase sensitivity of the analysis the combined ROMA data for all 22 tumor specimens were plotted across human chromosomes 1-22 (Fig. 2). The resulting ACE profile was consistent with chromosomal instability (CIN) that affects most chromosomes. Some chromosomal regions, however, were more frequently involved than others as shown by prominent peaks in red (gains) and green (losses). Losses in chromosomes 22q12.2, 19, and 1p36 appeared to be the most frequent events (55-74%) in pleural MM followed by gains in 5p14-p15 and deletions in 17p and 9q (41-46%). Deletions in 1p13-21, 9p21, 3p14-21, and 17q21 as well as gains in 8q and 18q were observed in 27-36% of cases. Minimal common areas (MCAs) for these recurrent events were established through overlapping of individual deletions or gains (Fig. 3 and Table 2). Genes residing in the MCAs were identified using probe coordinates and Human Genome Database (NCBI Builds 36.2 and 35).

Fig. 2

Chromosomal localization and frequencies of imbalances in 22 MM tumors as assessed with ACE (FDR=0.023)

Losses in tumor versus normal control are shown in green and gains are in red. Sex chromosomes are omitted from the study.

Fig. 3

Fine mapping of losses in 9p21.3 (A) and gains in 18q12.1 (B) regions

A, From top to bottom: STR specimens R140, R249, R367, R312, R318, R342, and R142 show overlapping deletions (in green). Minimal common area (MCA) and genes located therein are shown. B, STR specimens R249, R367, R312, R318, R342 and R142, and LTR specimens R219 and R298 show overlapping gains (in red). LTR sample R140 shows loss in the same region, depicted in green.

Cluster analysis of the STR and LTR group

The dramatic difference in the total number of genomic events observed between the STR and LTR groups prompted us to examine the relationship between and within these groups in more detail. Using ACE with an FDR of 0.023, we were able to isolate 85 probes, which discriminated significantly (p<0.01) between the STR and LTR groups and used them for cluster analysis (Fig. 4). The resulting dendrogram produced a clear distinction between the STR and LTR groups with only one patient, R166, misplaced. Inside the groups two samples, R158 and R258, did not cluster tightly within the respective group members. Notably, these two patients appear to be the only cases without a convincing history of asbestos exposure (Table 1). No discrimination between biphasic and epithelial types of MM was observed, and R125, the only sarcomatoid tumor in the study, clustered loosely with the rest of the group (Fig. 4).

Fig. 4

Hierarchical clustering of MM specimens based on 85 probes that distinguish between STR and LTR

Gene expression studies identify OSM, FUS1, PL6, CDH2, and DNAJA1 as differentially expressed during MM progression

Loss of chromosome 22, by far the most frequent event in our study (∼74%, Table 2), was observed in both LTR and STR groups, suggesting that this is a common event in MM tumorigenesis. The MCA defined for 22q12.2 spans ∼2 Mb and contains the oncostatin M gene (OSM), which encodes a proliferation-inhibiting cytokine. To find out if this gene’s activity is related to MM progression, we performed RT-PCR analysis on additional tumor and matched normal peritoneum specimens derived from 8 patients (Fig. 5A). OSM expression was seen in all normal samples and was not found in 5 tumor samples suggesting that it is, indeed, frequently suppressed in MM. Affymetrix expression array studies on an independent cohort of 30 MM and 7 normal peritoneum samples also confirmed our observation (Fig. 5B).

Fig. 5

Validation of gene candidates by RT-PCR (A) and Affymetrix expression analysis (B)

A, Matched normal (N) and tumor (T) specimens derived from eight MM patients were used for RT-PCR. MKI67, encoding Ki-67, was used as a proliferation marker and invariantly expressed PPIA as a loading control. B, Expression array validation on an additional set of MM (n=30) and normal pleura (n=7) specimens for on the U133 2.0 Plus platform.

We also analyzed the 3p21.3 chromosomal area, which showed loss in 32% of cases. Our interest to this area was stimulated by its association with lung and other common cancers ¹⁷18. We reported previously that loss in this area, although less precisely determined, can be observed in MM at ∼40% frequency. In this study, we were able to discriminate between the 3p21.31 (∼2.2 Mb) and 3p14.3-p14.2 (∼1.1 Mb) areas and report that at least two candidate tumor suppressor genes from the 3p21.3 region, FUS1(TUSC2) and PL6(TMEM115), were ∼2-fold down-regulated in 6 out of 8 tumornormal pleura matched specimens (Fig. 5A). Immuhnohistochemistry for the FUS1 product using a novel antibody confirmed down-regulation of the gene (courtesy Ignatio Wistuba, MD Anderson Cancer Center, Houston, Texasm, data not shown, manuscript in preparation).

Among the recurrent gains that we were able to identify in our series, the one in 18q12.1 contained the cadherin N gene (CDH2). Since cadherins play important roles in tumor progression, we studied CDH2 expression using the same approach as for OSM. Up-regulation of CDH2 was observed in a substantial fraction of additional sets of tumors (38%) by both RT-PCR and Affymetrix assays (Fig. 5AB) in agreement with the estimated frequency for this genomic gain (∼36%, Table 2). 6 out of 8 ROMA specimens that showed the CDH2 gain belonged to the STR group (Fig. 3B).

We also focused on DNAJA1, a member of the Hsp40 family of heat shock proteins. This gene is located in the smallest deletion detected in our study 9p21.1 microdeletion area (pos.32780-33292 kb, Fig. 6A) that was established as the best discriminator between LTR and STR in a probe-by-probe statistical evaluation (p<0.007). We have chosen DNAJA1 as a cancer-related candidate since another gene from the same family, DNAJB4, was recently extensively studied in relation to non-small cell lung carcinoma and proved to be a tumor suppressor that shows LOH and transcriptional down-regulation in tumors ¹⁹. Using RNA samples isolated from tumor specimens used for ROMA we found that DNAJA1 showed reduced expression in the STR cohort, which correlated with its allelic loss (4 out of 5 specimens with the deletions also showed down-regulation or lack of gene expression) pointing at the possible link between gene silencing and adverse clinical outcome (Fig. 6B). Affymetrix expression array analysis performed on the same set of tumor specimens (Fig. 6C) matched perfectly with the RT-PCR assessment.

Fig. 6

Association of DNAJA1 with STR MM by array-CGH (A), RT-PCR on ROMA specimens (B), and Affymetrix expression study (C)

DNAJA1 location in the 9p21.3 MCA is shown. Deletions in this area are shown in green, and the respective samples on the gel are marked with “d”.

DISCUSSION

Array CGH technologies provide new means for identification of most frequent (recurrent) DNA copy number imbalances in clinical specimens that contain a mix of normal and heterogeneous tumor genomes. In this study, we applied ROMA to clinical MM specimens to identify genetic events associated with disease progression. Cross-platform validation confirmed the reliability and accuracy of ROMA. To identify the most frequent aberrations in MM, we compiled and analyzed data generated on all 22 tumor specimens using ACE, a highly useful method of statistical processing of array-CGH data ²⁰. This analysis showed multiple aberrations in the tumors from patients with fast relapse after surgery, estimated their frequencies, and accurately mapped them. The observed pattern suggests that adverse prognosis in MM is linked with a highly increased CIN. Fine mapping of the recurrent chromosomal events allowed us to identify genes localized within these areas, and associate them with the disease by expression analysis.

The critical MCAs shown in Table 2 are in agreement with previous karyotypic and LOH studies on MM (reviewed in ³). Loss of 9p21.3, observed by us in 32% of MM cases (Table 2), is one of the most thoroughly studied abnormalities in cancer and is linked to the tumor suppressors p16^{and p14 (encoded by alternative reading frames of} CDKN2A), and p15 ^{(encoded by} CDKN2B) ²¹23. About 85% of MM cell lines and more than 20% of MM tumor specimens show homozygous deletions in this chromosomal region ²⁴. Mice deficient for all three 9p21.3 tumor suppressor genes are more tumor-prone and develop a wider spectrum of tumors than Cdkn2a mutant mice, with a preponderance of skin tumors and soft tissue sarcomas frequently composed of mixed cell types and often showing biphasic differentiation ²¹. In our study, the 9p21.3 deletion was observed exclusively in the STR cohort, substantiating the reported association with poor outcome ²⁵. The genes were localized in the middle of the MCA for the region, underlining the high accuracy of our mapping (Fig. 2).

The reported high rate of chromosome 22 loss (∼74%, Table 2) is consistent with chromosome 22 monosomy being the most frequent numerical change in MM ³. Allelic loss and mutations in NF2, located at 22q12.2, have been observed in ∼50% of MMs ²⁷. Our MCA established for chromosome 22 points at a neighboring gene, OSM, in the same area (Table 2). OSM encodes a cytokine known as oncostatin M, which possesses growth suppressor propensity. Notably, OSM expression was silenced in 5 out of 8 paired tumor/normal specimens derived from MM patients (Fig. 5A). Growth suppressive effect of OSM was documented for melanoma, glioblastoma, breast, lung, and prostatic cancer cell lines ²⁶30 but has not been previously implicated in MM. We hypothesize that co-deletion of OSM with NF2 may be advantageous for tumor progression.

Loss of chromosome 17p is also a common and important event in MM progression (our estimation of 46% is similar to the 40% reported in the literature ³) since the p53 gene TP53 is localized in 17p13.1. The MCA generated for the 17p (4791-5367 Kb, Table 2) by ACE, however, does not include TP53, suggesting the existence of a yet unknown tumor suppressor gene.

The 3p21.3 deletion is involved in many malignancies including asbestos-associated lung cancers ³¹ suggesting its significance for common tumorigenesis ¹⁷18. The region is densely packed with genes and may contain several potential tumor suppressors ³². One of the genes located in the area, RASSF1A, has been already established to encode a pro-apoptotic tumor suppressor ³³ and its loss is associated with many cancers including MM ³⁴. According to the most recent studies, at least two more genes from the same area, FUS1 (TUSC2) and PL6 (TMEM115), may be also associated with cancer. Forced expression of FUS1 resulted in tumor growth suppression ³⁵, and its disruption in a mouse model produced a compromised immune response and increased susceptibility to spontaneous cancers ³⁶. Association of FUS1 expression with stimulation of p53-regulated apoptosis provided an important insight into its tumor suppressor activity ³⁷. Silencing of PL6 observed in all clinical specimens of renal clear cell carcinoma (n=57) and other VHL-dependent malignancies suggests its involvement in early stages of VHL-deficient cancers ³⁸. Here we show that both FUS1 and PL6 are consistently down regulated in MM at the mRNA level (Fig. 5A), in agreement with their roles as tumor suppressors.

We also identified three chromosomal areas that, to our knowledge, have never been reported in connection with MM. One of these events, loss of chromosome 19, was the second most frequent numerical change after chromosome 22 (55 versus 74%, Table 2) implying its critical importance for MM. Remarkably, deletions that coincide with our MCA area (19p13) were recently found in asbestos-induced lung cancer and associated with the genotoxic effect of asbestos ³¹. The region we defined contains at least 15 potential cancer-associated genes (Table 2). Further research is needed to provide clues as to which of them are most likely asbestos exposure-associated tumor suppressors.

Loss in chromosome 9q34.11 (41%, Table 2) is yet another novel finding that allowed us to consider 3 of 9 genes identified in the MCA area, STXBP1, SH2D3C and ENG, as plausible tumor suppressor candidates. Experiments performed on a mouse skin cancer model suggested that ENG, which encodes the TGF-β coreceptor endoglin, may indeed act as a tumor suppressor during the late stages of carcinogenesis ³⁹. Among the recurrent gains that we identified, those in chromosomes 5p, 8q, and 17q have been already characterized ³40, while the remaining one in 18q12.1 appeared to be novel. Assessment of the 18q12.1 gain that we found predominantly in the STR patients (6 STR versus 2 LTR) identified one gene (CDH2) in the area. CDH2 encodes N-cadherin, which was upregulated in MM samples (Fig. 5A). Expression of N-cadherin in MM was reported previously at the protein level and used for discrimination of MM from lung adenocarcinoma ⁴¹44. Overexpression of adhesion-associated N-cadherin was also observed in hepatocellular carcinoma ⁴⁵, renal cancer ⁴⁶, invasive retinoblastoma ⁴⁷, and melanoma ⁴⁸49. In the latter case, N-cadherin expression was casually linked with transendothelial migration of melanoma cells and associated with Notch1 signaling. Studies on pancreatic cancer, one of the most aggressive malignancies, suggest that N-cadherin plays a principal role in epithelial-to-mesenchymal transition, which, in turn, is linked with invasion and metastasis ⁵⁰. A peptide that blocks N-cadherin prevented tumor growth and spread in a mouse skin cancer model ⁵¹. Overexpression of CDH2 in MM and a high correlation observed between expression of CDH2 and MKI57, a proliferation-specific marker that encodes Ki-67 ⁵² (Fig. 5A), is consistent with the gain observed in 18q12.1 and suggests an oncogenic role for N-cadherin in MM.

In conclusion, by combining a highly sensitive ROMA assay with novel noise-reducing statistical software, we have been able to precisely profile DNA copy number changes typical of MM and have identified recurrent genomic imbalances implicating genes that may be associated with early stages of the disease and its adverse prognosis.

Supplementary Material

1

ACKNOWLEDGEMENTS