Genome-wide array comparative genomic hybridization analysis reveals distinct amplifications in osteosarcoma

Background Osteosarcoma is a highly malignant bone neoplasm of children and young adults. It is characterized by extremely complex karyotypes and high frequency of chromosomal amplifications. Currently, only the histological response (degree of necrosis) to therapy represent gold standard for predicting the outcome in a patient with non-metastatic osteosarcoma at the time of definitive surgery. Patients with lower degree of necrosis have a higher risk of relapse and poor outcome even after chemotherapy and complete resection of the primary tumor. Therefore, a better understanding of the underlying molecular genetic events leading to tumor initiation and progression could result in the identification of potential diagnostic and therapeutic targets. Methods We used a genome-wide screening method – array based comparative genomic hybridization (array-CGH) to identify DNA copy number changes in 48 patients with osteosarcoma. We applied fluorescence in situ hybridization (FISH) to validate some of amplified clones in this study. Results Clones showing gains (79%) were more frequent than losses (66%). High-level amplifications and homozygous deletions constitute 28.6% and 3.8% of tumor genome respectively. High-level amplifications were present in 238 clones, of which about 37% of them showed recurrent amplification. Most frequently amplified clones were mapped to 1p36.32 (PRDM16), 6p21.1 (CDC5L, HSPCB, NFKBIE), 8q24, 12q14.3 (IFNG), 16p13 (MGRN1), and 17p11.2 (PMP22 MYCD, SOX1,ELAC27). We validated some of the amplified clones by FISH from 6p12-p21, 8q23-q24, and 17p11.2 amplicons. Homozygous deletions were noted for 32 clones and only 7 clones showed in more than one case. These 7 clones were mapped to 1q25.1 (4 cases), 3p14.1 (4 cases), 13q12.2 (2 cases), 4p15.1 (2 cases), 6q12 (2 cases), 6q12 (2 cases) and 6q16.3 (2 cases). Conclusions This study clearly demonstrates the utility of array CGH in defining high-resolution DNA copy number changes and refining amplifications. The resolution of array CGH technology combined with human genome database suggested the possible target genes present in the gained or lost clones.


Background
Osteosarcoma (OS) is a primary malignant tumor of bone arising from primitive bone-forming mesenchymal cells and it accounts for approximately 60% of malignant bone tumors in the first two decades of life [1]. These tumors typically arise in the metaphyseal regions of long bones, with the distal femur, proximal tibia and proximal humerus. A significant number of osteosarcomas are of conventional type which can be subdivided into three major categories based on their predominant differentiation of tumor cells: osteoblastic, chondroblastic, and fibroblastic. Currently, only the histological response (degree of necrosis) to therapy represent gold standard for predicting the outcome in a patient with non-metastatic osteosarcoma at the time of definitive surgery [2]. Patients with lower degree of necrosis have a higher risk of relapse and poor outcome even after chemotherapy and complete resection of the primary tumor. Therefore, a better understanding of the underlying molecular genetic events leading to tumor initiation and progression could result in the identification of potential diagnostic and therapeutic targets.
Chromosomal aberrations in osteosarcoma are highly complex and characterized by high frequency of amplifications. These amplifications may result in the overexpression of genes and contribute to the genomic instability in osteosarcoma. The identification of genes within the amplified sites is crucial for understanding the biology and clinical behavior of osteosarcoma. Until, recently gene amplification has been detected by PCR, southern blot analysis or FISH-based approach using gene specific probes. These techniques are inherently restricted to the previously known amplified genes in the genome. In contrast, genome-wide screening of amplified chromosomal regions with CGH has become an important tool for the detection of amplified regions in the tumor genome. So far published chromosomal CGH studies in osteosarcoma have identified several high-level chromosomal amplifications at 1p22, 1p31, 1q21, 1q23, 2q24, 3p25, 3q26, 6q24.3, 4q12, 5p14-p15, 5q33, 6p12-p21, 6q24.3, 7p21-p22, 8q12-q23, 10p21, 10q11.1, 10q22, 11q13, 11q23, 12p13, 12q12-q15, 17p11.2, 17q21, 18q22, 19p13.1 and 20p11.2 [3][4][5][6][7]. However, conventional CGH has limited sensitivity and resolution (~10-15 megabases) because of its dependence on the morphology of metaphase chromosomes. In addition, extensive follow-up work is required to identify candidate genes after regions of gain or loss have been identified. Recently, novel method termed as array-based comparative genomic hybridization (array CGH) has been described, which enables high throughput quantitative measurement of high-resolution DNA copy number changes throughout the genome [8]. This method is based on hybridization of differentially labeled test and reference DNAs to an array of mapped human genomic DNA fragments (~100-200 kb) and has been recently applied to human and mouse tumors [9][10][11][12][13][14]. To identify high-resolution copy number, we used array CGH to the panel of 48 tumors. The resolution of array CGH technology combined with human genome database not only allowed a precise identification of amplicons but also suggested the possible target genes within the amplicons.  15, 24, 25, 27, 29, 32, 34, 40, 48, 68, 76, 78, 79, 80, 82, 83, 85, 88, 95, 98, 99, 102, 123, 423, 425, and 474). All tissues in this study were obtained after IRB approved informed consents were signed. The age at diagnosis ranged from 5 years to 71 years at diagnosis. The histological information of 42 patients is presented in Table 1.

Array CGH
The array used in this study consists of 967 human BACs, which were spaced approximately 3 megabase across the whole genome. These arrays were obtained from Spectral Genomics, Houston, TX. The experiments were performed according to the manufacturer's protocol. Arrays were prehybridized with human Cot-I DNA (GIBCO Invitrogen, Carlsbad, CA) and salmon testes DNA to block the repetitive sequences on BACs. One microgram of normal DNA (reference) and tumor DNA (test) was labeled with cy5-dUTP and cy3-dUTP respectively, by random priming. To avoid dye bias, we performed dye swap experiments for each sample. The probe mixture is dissolved in hybridization mixture, denatured, cooled, and mounted with 22 × 60 mm coverslip. Hybridizations were performed in sealed chambers for 16-20 hours at 60°C. After post hybridization washes, arrays were rinsed, dried with compressed air, and scanned into two 16-bit TIFF image files using Gene Pix 4000A two-color fluorescent scanner (Axon Instruments, Inc., Union City, CA) and quantitated using GenePix software (Axon Instruments, Union City, CA).

Data processing and analysis
After scanning of the slide, the fluorescent intensities of the green and red channels were background subtracted.
The resulting values were normalized by intensity based local weighted regression method (Lowess) to correct for systematic bias in dye labeling and fluorescent intensity [15]. Then the ratio of the red/green channel of each clone was calculated and log base 2 transformed (log ratios). Each experiment was repeated once with dye reversal to addressing the confounding effect of the dye and experiment. The average of the dye-reversal experiment pair was calculated by reversing the sign of one experiment so that the log ratio reflects the tumor versus normal ratio.
We developed a new analytical method called invariant analysis to define the significant copy number changes. This method is designed to: i) increase the power of the analysis by combining all the cases in our dataset to define an invariant population (unchanged population); and, ii) to address the signal to noise differences among individual cases due to sample and hybridization variability. Our goal is to define a set of unchanged clones that can be used to calculate the upper and lower bound thresholds of the log ratios for the unchanged population in each experiment. First, we calculated the variance of each clone from all the experiments. We computed the p-values of the each clone by comparing to the clone with median variance using chi-square distribution http://linus.nci.nih.gov/ brb/TechReport.htm. The clones that have p-value greater than preset cutoff 0.9 were considered as invariant clone set, i.e. clones that do not vary significantly in all experiments. Then the mean and standard deviation of the log ratios of these invariant clones in each experiment were calculated. The clones with log ratios that exceed mean +/-2 × SD of the invariant set were considered gains and losses, respectively. For amplification and homozygous deletions, clones were defined to have at least 2 fold of the upper bound threshold and 4-fold of lower bound threshold, respectively. The gene(s) present in the clones were identified using UCSC browser http:// genome.ucsc.edu/cgi-bin/hgGateway ?org=human by downloading gene table (refFlat) from human gene assembly, July 2003. We search the candidate genes based on linear mapping position, which include 100 kb up and downstream from the clone center position. The supplemental data for this article is available at: https://big file.bcm.tmc.edu/outbound/prao/supplemen-tal%20data%20Man%20et%20al.xls

Statistical analysis
Significant clones in 6p, 8q, 12q and 17p amplicons were calculated using 2-sample t-test with randomized variance model http://linus.nci.nih.gov/~brb/TechReport.htm. The experiments in each of the two groups, amplification and normal, used for comparison were defined based on the invariant analysis (see above). The clones that have p <0.001 were considered as significant. We chose a stringent cutoff to minimize the multiple testing problem.

Results
To define the gains and losses in our experiments, we used invariant analysis for the first time to describe genomic changes by array CGH. In this method, we defined an invariant clone set that has low variance of log ratios among all the array experiments. After the mean and standard deviation of the log ratios in the invariant set of each experiment were calculated, clones that have higher or lower log ratios than the mean +/-2SD of the invariant set (upper bound and lower bound) were used to define gains and losses. We chose to use this method because it addresses some of the shortcomings of the modeling method, such as using all information provided in a set of experiment to determine the unchanged population instead of using one experiment at a time. However, the variation of each experiment is accounted for because the thresholds are calculated using the invariant set from each experiment. It also does not require a separate reference set for comparison. Finally, it provides an adjustable cut-Frequency of DNA copy number changes identified by array CGH in 48 osteosarcomas The amplified and homozygously deleted clones were defined to have at least 2 fold of the upper bound and 4fold of lower bound, respectively. Figure 1 summarizes the high-resolution DNA copy number changes identified by array CGH in 48 osteosarcomas derived from 42 patients. Copy number changes were detected involving small genomic regions, whole chromosomes, and chromosomal arms showing homozygous deletions and highlevel amplifications.

Overview of genomic profiles
Copy number changes excluding clones from sex chromosomes were involved in a significant fraction of most tumor genome. The estimated average genomic distance between clones was ~3-4 Mb. The frequency of clones showing gains (79%) was greater than losses (66%). High-level amplifications and homozygous deletions constitute 28.6% and 3.8% of tumor genome respectively. The most frequently deleted clones were identified from the chromosomal regions 2q31. 1 (Table 3). We explored the possible statistical relationship between copy number alterations and histological and clinical parameters. We found no significant relationship between copy number changes and primary/ metastatic disease, or histological type or histological response. This may be due to the involvement of large number of genomic loci and insufficient sample size.

Amplification is a frequent phenomenon in osteosarcoma
Previous studies using CGH have identified several chromosomal amplification sites in osteosarcoma. Because of the limitation of the method, it fails to pinpoint the precise site of amplicon. However, the present study by array CGH has identified 238 clones (28.6%) with highlevel amplifications. Recurrent amplifications were noted in ~37% of the total amplified clones ( Figure 3). These amplified clones were mapped to 1p22, 1p31. Gain of clones from 6p12-p21 regions was noted in 33/48 (~65%) cases analyzed. High-level amplification of the clones from same region was noted in 25% of the cases by array CGH. We found that most of the cases with amplification of 6p12-p21 displayed either increased or slightly varying degree of copy number increase across the 6p12-p21 region. The combined log ratios from all the cases defined the boundaries of amplification between RP3-329A5 and RP11-79F13. The amplicon spans approximately 9.4 Mb with amplification peak for clone RP11-81F7. Further, we used FISH to validate 6p amplicon on tumor metaphase and interphase cells from cases 274, 364, 426 and 527. Increased copy numbers for clones RP11-91E11, AL391415, RP11-81F7, RP11-79I2, RP11-90H17 and RP11-79F13 were noted in interphase cells with maximum copy number increase for clone RP11-81F7 ( Figure 4A). This was consistent with amplification peak for clone RP11-81F7 in the tumors profiled by array CGH ( Figure 2B). In addition, we used 2-sample t-test with randomized variance model to define significant clones from 6p12-p21 amplicon. By this method, we identified RP11-79F13 (p = 0.00000007), RP11-79I2 (p = 0.00000007) and RP11-81F7 (p = 0.00000007) as statistically significant clones.

Discussion
This study represents the first application of genome-wide copy number changes by array CGH in osteosarcoma. Recent studies in breast, renal and bladder cancer showed the potential assessment of this technology in detecting high-resolution copy number changes [9,11,14]. This approach will augment the identification of cancer causing genes by relating the clone information directly with sequence information from human genome database. In this study, we used array CGH to screen for high-resolution DNA copy number changes and precise identification of amplifications in a panel of 48 osteosarcomas.
Gene amplification is an important genetic mechanism in human cancers, as it clearly associated with tumor progression and has a prognostic significance and has even provided a target for therapeutics [17,18]. These amplifi- cations are often seen at the cytogenetic level as homozygously staining regions (hsrs) or double minute chromosomes (dms). However, cytogenetic recognition of amplifications doesn't contribute to the mapping and identification of amplified DNA sequences. The advent of CGH points an ever-increasing number of chromosomal amplifications in various tumors. These amplifications contribute to the genomic instability in tumors. We have recently shown that the mutation of p53 significantly correlates with genome-wide DNA instability and seems to represent a major genetic factor contributing to the extremely high levels of genomic instability found in high-grade osteosarcomas [19].
Our analysis have identified frequently amplified clones from 6p11.2-p21, 8q12.1, 8q24.13, 12q12, 12q13-q15, 12q21-q21.33, 16p13 and 17p11.2-p12. Amplification of clones from 6p12-p21 region was noted in 25% of the cases analyzed. This was consistent with the previously published results by CGH. By array CGH, we refined the 6p amplicon to 9.4 Mb with amplification peak for clone RP11-81F7. We recently demonstrated the origin of 6p amplicon as consequence of tandem duplication of clones RP11-81F7 and RP11-79F13 [7]. Based on combined array CGH and FISH analysis suggest CDC5L, HSPCB, and NFKBIE, and HGNC and MRPL14 are the target genes from 6p12-p21 amplicon. Of these genes, CDC5L may be an important gene in cancer because of its role as a posi-tive cell cycle regulator for G2/M transition [20]. Consistent with our analysis, overexpression of HSPCB was shown recently by cDNA microarray studies on osteosarcoma [21]. This protein was shown to play an important role in assemble/disassembly of tubulin by inhibiting tubulin polymerization.
High-level amplification of clones on 12q revealed three distinct sites of amplifications -AMP1 (12q12), AMP2 (12q14.1) and AMP3 (12q21.33). Pervious studies have shown the amplification GLI, CHOP, SAS, HMGI-C, CDK4, HDM2, and PRIM1 from 12q13-q15 region in osteosarcoma [22,23]. The present array CGH analysis identified a possible target gene IFNG from AMP2 (RP11-298M11; p = 0.0000001), which is physically mapped close to the HDM2 oncogene locus [24]. Previous studies demonstrated that T-cell production of IFNG strongly suppresses osteoclastogenesis by interfering with the RANKL-RANK signaling pathway. IFNG induces rapid degradation of the RANK adaptor protein, TRAF6, result- ing in strong inhibition of the RANKL-induced activation of the transcription factor NFKB and JNK [25]. The AMP3, which was 23.3 Mb distal to AMP2 containing RP11-89F6. Our analysis from AMP3 revealed two interesting candidate genes: transcription factor ELK3 and PCTAIRE protein kinase 2 (PCTK2). ELK3 is a member of the ETSdomain transcription factor family and the protein is activated by signal-induced phosphorylation [26]. The protein encoded by PCTK2 belongs to the cdc2/cdkx subfamily of the ser/thr family of protein kinases and play an important role in the regulation of the mammalian cell cycle [27]. High-level amplification of three clones from 12p13 was noted in case 27 and the amplicon span 4.6 Mb with peak amplification for clone RP11-89D16. No candidate genes contained with in this BAC. Amplification 12p has been reported previously in 9/19 high-grade osteosarcomas by CGH. Recent FISH analysis has identified the amplification of CCND2, ETV6, and KRAS2 from 12p region [28].
Amplification of 17p11.2 was noted in 27% of the cases analyzed by array CGH. Our array CGH analysis has identified three clones with high-level amplifications that spans ~3.7 Mb region on 17p11.2. Several candidate genes were identified within these clones (TPP3A, SMCR5, DRG2, FL11, MYCD, SOX 17, ELAC2, and PMP22). Recent studies have shown the amplification of some of the genes identified in the present study (PMP22, and TOP3A) from 17p11.2-p12 in high-grade OS by semiquantitative PCR and cDNA microarrays [29,30].

Conclusions
In summary, high resolution array-based CGH revealed large number of chromosomal aberrations previously identified in osteosarcoma by chromosomal CGH and conventional cytogenetic methods. The present study allowed precise identification of smaller DNA copy number alterations, which suggest the presence of specific target genes in osteosarcoma. Although this study suggested several possible target genes from amplified regions from 6p, 8q, 12q and 17p, but these genes should be validated by other molecular and immunohistochemical approaches on well-defined large patient samples. Further, interaction or association studies between small genomic losses and gains will facilitate the identification of new genetic pathways in the pathogenesis of osteosarcoma.