Genome-wide analysis of Ollier disease: Is it all in the genes?

Fresh frozen tissues from 37 tumors from 28 patients diagnosed with Ollier disease were collected from the EuroBoNet consortium (http://​www.​eurobonet.​eu) (Table 1): Leiden University Medical Center, The Netherlands (12 tumors), Leuven University, Belgium (5 tumors), Rizzoli Institute, Italy (6 tumors), Royal Orthopaedic Hospital, United Kingdom (7 tumors), Lund University, Sweden (2 tumors), Netherlands Committee on Bone Tumors, The Netherlands (2 tumors), Heidelberg University, Germany (1 tumor), University of Ghent, Belgium (1 tumor) and Groningen University Medical Center, The Netherlands (1 tumor). All samples were derived from primary tumors, not from recurrent tumors, and all were graded according to Evans et al [13]. Diagnoses were originally made in the multidisciplinary teams of the centers of origin. The histology was revised and representativity was assessed on the available paraffin or frozen tissue by one experienced bone tumor pathologist. For SNP array analysis, 14 Ollier ECs, and 23 Ollier CS (13 grade I, 8 grade II, 2 grade III) from 28 Ollier patients were used. As controls, normal DNA derived from fresh frozen muscle tissue (n = 3), peripheral blood lymphocytes (n = 4) or saliva (n = 4) was available for 11 Ollier patients and 3 patients with unrelated bone diseases. We used blood lymphocyte DNA from 12 healthy controls and 1 HapMap sample. We also isolated DNA from saliva for 3 of these controls to validate the use of saliva DNA in this study. Twenty eight of these thirty controls and DNA from 10 additional HapMap samples were used in MLPA. RNA from 4 articular cartilage, 2 growth plates and 7 ECs was used for expression array and 3 articular cartilage, 2 growth plates and 8 ECs were used in qRT-PCR.

Five tissue micro array (TMA) blocks containing 86 tumors were constructed, of which 65 were Ollier related and 21 were solitary central tumors (ECs and CS) from both the EuroBoNet and the European Musculoskeletal Oncology Society (EMSOS) networks (Table S1, Additional file 1): Leiden University Medical Center, The Netherlands (27 tumors), Rizzoli Institute, Italy (12 tumors), Copenhagen University, Denmark (9 tumors), University clinic of Orthopaedic Surgery and Medical university of Graz University, Austria (6 tumors), Bern University, Switzerland (5 tumors), University of Navarra, Spain (4 tumors), Netherlands Committee on Bone Tumors (21 tumors), Istanbul University, Turkey (2 tumors). All the samples were obtained according to the ethical guidelines of the host institution. Samples were coded and all procedures were performed according to the ethical guidelines "Code for Proper Secondary Use of Human Tissue in The Netherlands" (Dutch Federation of Medical Scientific Societies).

Tumor samples were selected that contained more than 80% of tumor cells, as estimated on haematoxylin and eosin-stained frozen sections. Most of the samples were macro dissected and L2099 sample was micro dissected to enrich the tumor percentage [14]. DNA from fresh frozen tissue was isolated using the wizard genomic DNA purification kit (Promega, Madison, WI), according to the manufacturer instructions. Blood DNA was isolated as described by Miller et al [15]. Saliva DNA was isolated using the Oragene DNA kit (DNA Genotek Inc Ontario, Canada) according to the protocol provided by the supplier and DNA was precipitated using sodium acetate precipitation. DNA concentration was quantified spectrophotometrically using Nanodrop ND-1000 (Thermo Fisher Scientific, Waltham, MA, USA) and the fragment sizes were determined on 1% agarose gel.

RNA isolation from the fresh frozen tissue was performed as described previously [16]. RNA concentration was measured spectrophotometrically and the fragment sizes were determined by RNA 6000 Nano LabChip kit using Agilent 2100 Bioanalyzer (Santa Clara, CA, USA). DNA and RNA from all the samples were good enough to continue with the experiment.

We used the Affymetrix Genome-Wide Human SNP Array 6.0. Genomic DNA preparation, labeling and hybridization were performed according to Affymetrix's recommended protocols (Affymetrix, Santa Clara, CA, USA). Then, arrays were scanned with GeneChip™ GSC3000 7G Whole-Genome Association System (Affymetrix). Overall hybridization quality was estimated by the genotype call rate using the Birdseed genotype calling algorithm in Genotyping Console (version 3.0.2, Affymetrix). Average call rate was 97.83%. To analyze the data we used statistical language R version 2.8 and Nexus software version 4.1 (BioDiscovery, CA, USA). We did not use HapMap samples as baseline in this study to avoid the bias for the experiment performance at different labs, batch effect and hybridization quality. The analysis was performed on a subset (30 controls, 14 ECs and 23 CS of Ollier patients, Table 1) of a larger experiment of 92 samples including samples unrelated to Ollier disease to achieve a larger set of common controls. Than we performed copy number analysis using 92 samples as a baseline in R software and only 29 control samples of high quality as a baseline in Nexus software. Results that we obtained using different softwares and different baselines were comparable. We used CEL files in R software. For the genomic analysis using R, we did genotyping using the CRLMM algorithm in the Oligo package [17], copy number analysis using the aroma affymetrix package [18], and we constructed LAIR plots to visualize regions of LOH and allelic imbalance [19]. Chromosome-X was not analyzed to avoid gender-related issues [20]. In Nexus, we performed copy number analysis using CNCHP log-ratio files generated by genotyping console using 29 controls as a baseline. Hidden Markov model (HMM) based SNP-FASST segmentation was used to identify aberrant genomic regions. Here we considered at least 5 probes for each segment. The data discussed in this publication have been deposited in NCBI's Gene Expression Omnibus (GEO) database (http://​www.​ncbi.​nlm.​nih.​gov/​geo/​ accession number GSE22965).

MLPA was used to confirm copy number gains and losses found within the selected candidate genes by SNP array. MLPA probes were designed using NCBI Build 36.1. We used two probes for TCRA and one probe for POU5F1, ANKS1B, FAM86D and PRKG1. Probe sequences can be obtained upon request. MLPA was performed as described previously [21]. Sample series of SNP array (Table 1) and in addition to that DNA from ten HapMap samples was used. Data analysis was done using SoftGenetics GeneMarker version 1.70. We set 1.2 and 0.8 as a threshold to detect the gains and losses respectively.

In total, we analyzed 7 ECs while 2 growth plate and 4 articular cartilage samples served as a control (Table 1). Expression array was performed using Illumina Human-6 v3.0 Expression BeadChips (Illumina Inc., San Diego, CA). For Illumina BeadArray assay, cRNA was synthesized with an Illumina RNA Amplification Kit (Ambion, Austin, TX, USA), purified, labeled and hybridized as per the manufacturer's instructions. Then, arrays were scanned with Illumina BeadArray reader (500GX, Illumina) and scanned images were imported in BeadStudio software version 3.1.3.0 (Illumina). For the expression array, unprocessed data were collected from BeadStudio and processed using VST [22] and quantile normalization with the R bead array package [23]. The processed data was analyzed using Linear Models for Microarray Data (LIMMA) analysis which uses moderated t-test to detect differentially expressed genes between two groups by taking into account natural variance within these groups and correcting for multiple testing using false discovery rate [24].

qRT-PCR was used to confirm the results obtained by the integration of SNP and expression array. Here we have used 5 controls (2 growth plates and 3 articular cartilage) and 8 ECs (Table 1). RNA used for expression array was also used for the qRT-PCR. cDNA was synthesized using 1 μg of total RNA with AMV reverse transcriptase (Roche Applied Science, Almere, The Netherlands) in combination with oligo-dT and random hexamer priming and qPCR was performed as described elsewhere [25]. Primer details can be provided upon request. Expression of the genes of interest (FAM86D, POU5F1 and ANKS1B) was normalized by geometric averaging of multiple internal control genes using the geNorm program [26]. Out of four normalization genes the best three were selected within this program: GPR108, SRPR and TBP. Relative quantification was performed using standard curves, followed by adjustment with the normalization factor calculated by geNorm program [25]. The average relative expression of gene of interest in ECs was compared to controls. To see the EC specific gene expression changes, relative expression of ECs with change in copy number was compared with ECs without change in copy number.

Five TMA blocks were constructed using TMA master (Zeiss, 3D Histech, Hungary) and each block contain maximum 70 cores including seven control tissues (growth plate, articular cartilage, breast carcinoma, prostate, colon, skin and tonsil) for orientation purpose. In total 5 TMA blocks contain 86 tumors, of which 65 Ollier related and 21 solitary ECs and CS (Table S1, Additional file 1). The TMA blocks contain 2-mm cores of each tumor in triplicate.

Tissue sections were cut from TMA blocks and dried overnight at 60˚C. Slides were kept in Xylol for 20 minutes. Immunohistochemistry was performed as described earlier [27]. Detailed information on the antibodies used to check protein expression of POU5F1 and NIPBL are given in table S2, additional file 1. We used power vision (poly-HRP-GAM/R/R, Immunologic) as a secondary antibody. Visualization was carried out with liquid Dab⁺ substrate chromogen system (DAKO, Glostrup, Denmark). As a negative control primary antibody was omitted. Immunostained TMA slides were scanned using a high resolution Mirax Desk instrument (Zeiss, Mirax 3D Histech, Hungary) and scored independently by two observers (JVMGB and TCP) with the Mirax viewer TMA module software version 1.1.12 (Zeiss) and discrepancies were discussed. In brief, the intensity (0 = no staining, 1 = weak, 2 = moderate, 3 = strong) and percentage of positive tumors cells (0 = 0%, 1 = 1-24%, 2 = 25-49%, 3 = 50-74%, 4 = 75-100%) were assessed. Cores with a negative internal control and loss of tissue were excluded from the analysis. A sum score ≥ 4 was considered positive. Statistical analysis was done using Pearson Chi-Square test in SPSS (version 16.0, Chicago, Illinois, USA).

Samples L1975, L1974, L1980, L813, L1083 and L286N were excluded from further analysis because of quality issues and low call rates. All samples including 29 controls and 32 tumors (14 ECs, 18CS) showed DNA copy number aberrations mainly restricted to known variable regions (Figure 1)[28, 29]. CNA were more frequently found in CS II and CS III as compared to ECs and CS I (Figure 1). The number of copy number changes in saliva and blood DNA were comparable.

To validate regions identified by SNP arrays, we performed MLPA on 37 Ollier tumors (Table 1) and 38 controls for the candidate genes FAM86D, TCRA, ANKS1B and PRKG1 (Table 4). In addition, a probe was designed at POUF51 which is at chromosome 6p21.31 in order to confirm loss of chromosome 6 in L206 and L1683. We have used 38 controls as a baseline. Results of validation experiments are summarized in Table 4. In short we confirmed three (FAM86D, PRKG1 and monosomy 6) out of five candidate regions. We confirmed the homozygous loss of FAM86D in two ECs of the same patient (Figure 2). Gain at PRKG1 in L1490 is confirmed which seems to be EC specific in this case but was not found in other ECs and CS. Loss at chromosome 6 using probe at POU5F1 was confirmed in L206 while loss was not confirmed in L1683 at given thresholds. TCRA region at chromosome 14q11.2 is known as a highly polymorphic region [31] and its frequent rearrangements are observed in blood lymphocytes [32]. We used majority of blood samples as a baseline and found copy number gain in all ECs at TCRA as a result of SNP array unpaired analysis. Paired analysis with SNP array showed copy number gain at TCRA in L1978 and was not confirmed with MLPA. Indeed this gain was due to loss in the corresponding normal DNA. Loss at ANKS1B in L2220 was not confirmed by MLPA at given thresholds however the peak was lower in L2220 compared to corresponding normal.

Expression array was performed using 7 ECs and 6 controls with Illumina Human-6 v3.0 (Table 1). We performed function based analysis by integrating the gene expression results with SNP array results. In total 1044 genes were differentially expressed in ECs compared to controls (adj. p-value < 0.01, Table S4, Additional file 1). We considered all up regulated genes (881 genes, adj. p-value < 0.01) with presence of gain in at least one EC and all down regulated genes (163 genes, adj. p-value < 0.01) with presence of loss in at least two ECs. We found NIPBL which was gained as well as up regulated while POU5F1 that was lost as well as down regulated in ECs compared to controls. The same pattern was found in CS.

Using qRT-PCR, we confirmed lower expression of ANKS1B, FAM86D and POU5F1 (Table 4). L2220 showed one copy loss of ANKS1B and its expression was decreased in this EC as compared to the average of relative expression of other 7 ECs however there is no difference in the expression comparing all ECs versus controls. SNP array results revealed homozygous loss in L206, L910 and one copy loss in L2103 for FAM86D. FAM86D expression was decreased in ECs compared to controls. There was no expression of POU5F1.

Using a TMA we demonstrated that POU5F1 protein expression was absent in all Ollier tumors which is in line with the expression array and qRT-PCR results. For NIPBL, 30% of Ollier tumors were positive (Figure 3, Table 4). Only 34/65 Ollier and 11/21 solitary tumors could be analyzed since cores from the rest of the tumors were lost during the immunohistochemistry. There was no significant difference in the expression of NIPBL within the tumors of different grades in Ollier disease (Pearson Chi-Square, p-value = 0.1).