Copy number alterations in small intestinal neuroendocrine tumors determined by array comparative genomic hybridization

Fresh frozen samples from patients with sporadic SI-NETs were included in this study. All tumor samples were obtained from patients operated for SI-NETs at the Karolinska University Hospital-Solna since 1989 and had been collected and stored at the Karolinska tissue biobank. All samples were obtained with informed oral consent and ethical approval from the local ethical committee of Karolinska Institutet. A standardized procedure of tumor collection was employed including macroscopical dissection by a histopathologist, snap freezing and storage at −80°C until use. The histopathological diagnosis was established at routine examination according to published criteria [9, 18]. The immunohistochemical examination included staining for chromogranin A [19], and/or Grimelius silver staining and Masson staining as a marker for serotonin [12].

Totally 30 tumor samples (19 primary tumors and 11 metastases) from 29 patients (15 females and 14 males) were collected for the a-CGH screening. For the subsequent qPCR analyses of selected loci the tumor sample panel was extended to a total of 43 tumor samples from 32 patients. The clinical data concerning gender, age at diagnosis of the primary tumor, functioning tumor, previous SI-NET surgery, metastasis and follow-up are detailed in Additional file 1: Table S1. Eighteen tumors from 18 patients were from our previous publication (cases 1–18, Table 1) [12]. Twenty-four of the samples were primary tumors and 19 were metastases. Twelve of the metastases were regional (mesenterial, regional lymph nodes or omental) and 7 were distant metastases (5 liver and two ovarian). Matched primary and metastatic lesions were available from 9 out of 32 cases, and for two cases two metastatic samples were included. A sample of non-neoplastic tissue close to the primary tumor from a SI-NET case was included as normal control for quantitative real time PCR (qTR-PCR). Sixteen cases were female and 16 were male. The median age at the time of surgery was 67.5 years (mean 65 years; range 39–80 years). Patients were followed-up from diagnosis until death or last date of contact for a median of 88 months (mean 94 months; range 7–222). Seventeen patients were known to have a functioning tumor based on carcinoid syndrome and/or increased levels of 5-HIAA. Seven patients had been treated surgically before the operation at which the sample studied was collected. In addition, cases 1–8 were operated before the introduction of somatostatin analogues, which was subsequently the standard peroperative treatment concerning cases 9–32. MIB-1 proliferation index was available for cases 19–32 which all showed <2% except in case 25 that showed 2.5%. In cases 1–18 the primary tumor was operated in the period 1986–1997 i.e. before the introduction of MIB-1 analysis in the clinical workup of these tumors.

Genomic DNA was either available from our published series of SI-NETs [12] or was isolated from frozen tissue using DNeasy Blood and Tissue DNA isolation kit (Qiagen GmbH, Hilden, Germany) according to the manufacturer’s protocol. The DNA quality and concentration was assessed using a NanoDrop A100 Spectrophotometer (ND-1000, Thermo Scientific, USA).

A-CGH was applied on 30 SI-NETs from 29 patients using human BAC (Bacterial Artificial Chromosome) arrays with 1 Mb resolution for 6 samples (cases 1, 3, 4, 6, 7, and 9), or of tiling type for 27 samples from 26 cases (1, 2, 4, 5, 6, 8, 10–32). Initially, we used commercially available 1 Mb arrays (Spectral Genomics, Houston, TX USA currently Perkin Elmer) applying previously described methodology [20]. Briefly, a dye swap method was applied, and the Spectralware 2 software (Spectral Genomics) was used for data analysis with cut-off levels at 1.2 and 0.8 for identification of gain or loss, respectively. Subsequently, human tiling 33 K and 38 K BAC arrays, produced at the SCIBLU Genomics Centre at Lund University, Sweden (http://​www.​lu.​se/​sciblu) were used. The 33 K and 38 K array slides contained 33,370 and 38,000 BAC clones, respectively (CHORI BACPAC resources) (http://​bacpac.​chori.​org/​genomicRearrays.​php) giving a resolution of one clone per 50–100 kb. Information about experimental procedures and data analyses have previously been described in detail [21]. After hybridization, slides were scanned using a Genepix 4200A scanner (Axon instruments Inc., Union City, CA). The resulting Tiff images were quantified using the GenePix Pro 6.0 package analysis software (Axon instruments, Wheatherford TX, USA) and gene pix result (GPR) files were generated. The GPR files were then loaded onto BioArray Software Environment (BASE; http://​base.​thep.​lu.​se/​) [22] for further analyses and data processing such as filtering, normalization, smoothing and profiling.

Log₂ ratio thresholds were used to define gain (+0.25), loss (−0.25), amplification (+1) and homozygous loss (−1). Software based profiles were inspected manually for confirmation of CNAs. Recurrent regions of CNAs were defined as those observed in three or more a-CGH profiles [23]. Inside recurrent regions, minimal overlapping regions (MOR) were identified as the smallest region of overlapping loss or gain. CNA regions in telomeric and centromeric regions as well as involving few clones only were interpreted with caution. CNAs overlapping with known normal genomic variants according to the Database of Genomic Variants (http://​dgvbeta.​tcag.​ca/​dgv/​app/​home?​ref=​NCBI36/​hg18) were not reported.

For cluster analysis of recurrent regions of CNAs, a binary matrix was generated where the rows indicated altered chromosomal regions and columns represented the tumor samples. In each case copy number alterations for each case CN status was coded as “0” for normal copy number or “1” for CNA either gain or loss. The matrix was uploaded in Multi Experiment Viewer 4.7.4 software and unsupervised hierarchical clustering analysis was performed using Euclidean distance and average linkage [24]. CNA data for case 27 with paired primary-metastasis samples were pooled.

qPCR analysis of CNs was applied to all 43 SI-NETs for selected loci with frequent CN losses in 18p (EMILIN2), 18q (DCC, BCL2, CDH19), 16q (CDH1) and 11q (SDHD). All target and reference assays were purchased from Applied Biosystems. RNaseP (on chromosome 14) was used as endogenous control for normalization of analyzed loci in chromosomes 18, 16 and 11. The following assays were used: EMILIN2 (Hs01996822), DCC (Hs02317964, Hs02967342), BCL2 (Hs01500302), CDH19 (Hs02826809, Hs02956257), CDH1 Hs00934267), SDHD (Hs03794135) and RNaseP (part number 4403326). Assays were chosen to avoid overlap with known SNPs. The experimental procedure recommended by the manufacturer (Applied Biosystems) was followed. Five nanogram genomic DNA was used in the qPCR reaction, and water was analyzed in parallel as negative control. All qPCR reactions were run in quadruplicate in a Step One Plus qRT-PCR machine (Applied Biosystems) using standard cycling conditions of 10 min at 95°C, followed by 40 cycles of [95°C for 15 sec and at 60°C for 1 min]. Pooled normal blood DNA (Promega, Madison, WI, USA) was used as calibrator and a normal mucosal intestine DNA as normal control. CNs were predicted by Copy Caller v1.0 software (Applied Biosystems).

The follow-up period was calculated from the date of diagnosis of the primary tumor until the date of death or the last date of contact. The log rank test was used to calculate overall survival and illustrated by Kaplan-Meier plots concerning tumor groups, recurrent region of CNAs and clinical parameters. Moreover to evaluate the possible effect of confounding factors (e.g. gender or regional, distant and extra-hepatic metastases) multivariate analysis using Cox proportional hazards modeling was applied to those recurrent regions of CNAs which were significantly associated to survival. Associations between tumor groups and recurrent CNA or clinical parameters were evaluated by Fisher’s exact test and for age at diagnosis by Mann–Whitney U test. All statistical analyses were performed using the statistical Software SPSS v 16.0. P-values ≤ 0.05 were regarded as significant.

We determined genome-wide CNAs in 30 tumors representing 19 primary tumors and 11 metastases from 29 patients with SI-NET using a-CGH. All samples analyzed displayed CNAs, with the largest and smallest extent of total CNAs observed in tumors number 4 (550 Mb) and 29 (2.5 Mb), respectively. CN losses and gains were in many cases extensive and involved entire or almost entire chromosomes. Gains were more common than losses. Recurrent CN losses were found on chromosomes 18, 16, 11, 9 and 13 (Table 1) and gains on chromosomes 20, 14, 4, 5, and 7 (Table 2). Overall the CNAs detected by a-CGH in this study were consistent with those detected for the 17 cases analyzed by metaphase CGH in our previous study [12] as well as with those samples analyzed by the 1 Mb resolution array. However, a-CGH identified additional CNAs not previously detected by metaphase CGH such as loss of 18p in case 4, 10 and 15 or gains of chromosomes 4 and 5 in case 14.

Recurrent CN losses were most frequently found on chromosomes 18 (70%), 11q (23%), 9 (20%), 16 (17%) and 13 (13%). Fifteen of 30 tumors (50%) had entire or almost entire loss of chromosome 18. In addition, loss of chromosome 18 was the only detectable recurrent CNA in four tumors (5, 10, 12, and 17), three of which were primary tumors. Sub-chromosomal losses of chromosome 18 were observed in six tumors (Table 1, Figure 1). Two recurrent regions of loss were identified, one on 18p (pter-p11.21) and another on 18q (q12.1-q21.31) (Figure 1). A MOR of loss of 247 kb was observed within the 18pter-p11.21 interval at 18p11.32-p11.31 in tumor 31. This MOR encompasses three known genes, KIAA0650, LPIN2 and EMILIN2, and the loss of the latter was verified by genomic qPCR. Another MOR of 2 Mb loss was identified in tumor 16 at 18q22.1 which encompasses the genes CDH7 and CDH19. However, we could not confirm the loss of CDH19 on 18q22.1 in tumor 16 by qPCR.

Recurrent CN losses were observed on chromosome 16 in 5/30 (17%) tumors. A recurrent region of 34.5 Mb loss which maps to 16q12.2-qter was detected in 5 tumors (1, 6, 9, 21, and 27P) (Figure 2A and Additional file 2: Figure S1). This region encompasses tumor suppressor genes including CDH1, E2F4, CTCF and FOXF1. CN losses on the long arm of chromosome 11 were detected in 23% of the tumors (7/30). These tumors shared a recurrent region of 35.5 Mb at 11q22.1-qter (Table 1; Figure 2B and Additional file 2: Figure S1). This region encompasses several known tumor suppressor genes with a role in apoptosis such as SDHD and members of the cysteine-aspartic acid protease (caspase) family including CASP4 and CASP12. Whole chromosome 9 loss was found in two tumors and segmental losses in four tumors with recurrent regions at 9pter-p13.2 and 9p13.1-11.2. Loss of entire chromosome 13 was found in one metastasis and two primary tumors showed deletions at 13q11-ter and 13q13-q21.32.

Recurrent CN gains were found on chromosome 4 (27%), 5 (23%), 7 (13%), 14 (30%) and 20 (37%) (Table 2; Figure 2C and Additional file 2: Figure S1). Several genes involved in tumor development including a variety of growth factors are located on these chromosomes. These include for example FGF2, FGFB, FGFR3 on 4q; PDGFR, APC, TGFB1, SMAD5 on 5; BRAF, IGFBP3, MDR1 on chromosome 7; AKT1, BCL2L2, MAX, MMP14, DAD1 and DICER1 on 14, and E2F1, CDH4, LAMA5, TNFRSF6B on chromosome 20.

A MOR of 230 kb gain at 14q11.2 downstream of the DAD1 locus was observed in four tumors (20, 23, 27M, 28) which overlapped with partial or entire gains in five other tumors (1, 7, 25, 27P, 32). This region encompassed several genes among others DHRS4L2, REC8L1, IPO4, PSME1 and PCK2. One tumor (case 25) displayed gain of entire chromosome 14, and four tumors (1, 20, 23, 27M) carried partial or small gain of 14 without harboring copy number losses on chromosome 18. A MOR of 1.4 Mb gain at the telomeric region of chromosome 20 (q13.33) was detected in 9/30 cases (30%) (data not shown). The 20q13.33 region overlapped with several cancer related genes such as CDH4, LAMA5, RPS21, KIAA1510, TNFRSF6B (M68, DcR3), RTEL, EEF1A2 and PTK6.

All CNAs identified by a-CGH were reviewed to identify recurrent loss or gain regions. The selection of candidate genes for validation were based on the most common CNAs in the present study as well as candidate genes identified from the literature [5, 12, 14‐17, 25] and their potential role in tumorigenesis of SI-NETs. To validate CNAs in the selected recurrent regions, qPCR-based copy number analysis was used. For validation of CN losses in chromosome 18, we selected EMILIN2 (p11.32-31), DCC (q21.1-2), BCL2 (q21.33) and CDH19 (q22.1). CNs in chromosome 18 were verified in 12/19 (63%) of the tumors for EMILIN2, 11/18 (61%) for DCC, 11/17 (65%) for BCL2 and 11/18 (61%) for CDH19 (Additional file 3: Table S2). Consistent results were obtained when two different assays used for the same gene (DCC, CDH19).

For losses on chromosomes 16 and 11, we selected CDH1 on 16q22.1 and SDHD on 11q23.1 and they were verified respectively in 3/5 (60%) and 4/7 (57%) of tumors with corresponding losses detected by a-CGH (Additional file 3: Table S2). Taken together CNAs and CNs detected by a-CGH were confirmed by qPCR in 61% and 78% of tumors respectively (Additional file 3: Table S2).

In addition to the tumor panel screened by aCGH, qPCR-based CN profiling was performed in additional SI-NETs (Table 3). Overall, qPCR was carried out for 11 tumor pairs of which 9 were primary tumors with paired metastases from the same patients and two were paired metastases from the same individual (Table 3). CNAs for the target genes detected by a-CGH and qPCR were similar in primary tumors and corresponding metastases in most patients, however, small differences were observed at some loci.

The a-CGH data was subjected to hierarchical cluster analysis to detect tumor groups with similar patterns of chromosomal alterations based on recurrent regions of CN loss or gain. Two different tumor groups (I and II) and five chromosomal clusters -(a-e) were identified (Figure 3). Tumor group I consisted of 23 tumors (7 metastases, 30%) and group II contained 7 tumors (4 metastases, 56%). Losses in chromosomes 18, 16 and 11 and gain of chromosome 20 were found in both tumor groups. Tumor group II was significantly enriched for all alterations which were part of cluster-d including gains of chromosome 4 (P = 0.007), 5 (P = 0.003), 7p22.3 (P = 0.001), 7p22.2-22.1 (P = 0.001), 7q22.1 (P = 0.001), 7q22.3-qter (P = 0.006), 14q11.2 (P < 0.0005), and 14q32.2-32.31 (P < 0.0005). A significant correlation was also observed between group II and gain on 20pter-p11.21 (P = 0.014).

Cases with extra-hepatic metastases (10, 21, 25, 27 and 32; Additional file 1: Table S1), had a higher likelihood to segregate into group II (P = 0.016) and harbored gains on 7p22.3 (P = 0.018), 7p22.2-22.1 (P = 0.018), 7q22.1 (P = 0.018), 7q22.3-qter (P = 0.003), 14q11.2 (P = 0.049), and 14q32.2-32.31 (P = 0.016). Loss on 16q12.2-qter was more common in distant metastases compared with primary tumors (P = 0.003), and this loss along with gain on 7q22.3-qter were more common in metastases compared to primary tumors (P = 0.016 and P = 0.047, respectively), suggesting their involvement in tumor progression.

Next, we tested possible associations between CNAs in recurrent regions and patient survival. CN gain in the 20pter-p11.21 region was associated with shorter overall survival (P = 0.013) (Figure 3B). Cox proportional hazards modeling supported this finding as no confounders were found. No other significant associations were observed between recurrent CNAs and survival. Losses on 9p and 18p as well as gain on 7q were associated with younger age at diagnosis (P = 0.025-0.029). CN losses on chromosomes 11 and 16 and gain on chromosome 20 were more frequent in female patients involving 11q23.1-qter (P = 0.039), 16q12.2-qter (P = 0.019), 20pter-p11.21 (P = 0.017) and 20q11.1-11.21 (P = 0.039).