Array comparative genomic hybridization-based characterization of genetic alterations in pulmonary neuroendocrine tumors

Abstract

The goal of this study was to characterize and classify pulmonary neuroendocrine tumors based on array comparative genomic hybridization (aCGH). Using aCGH, we performed karyotype analysis of 33 small cell lung cancer (SCLC) tumors, 13 SCLC cell lines, 19 bronchial carcinoids, and 9 gastrointestinal carcinoids. In contrast to the relatively conserved karyotypes of carcinoid tumors, the karyotypes of SCLC tumors and cell lines were highly aberrant. High copy number (CN) gains were detected in SCLC tumors and cell lines in cytogenetic bands encoding JAK2, FGFR1, and MYC family members. In some of those samples, the CN of these genes exceeded 100, suggesting that they could represent driver alterations and potential drug targets in subgroups of SCLC patients. In SCLC tumors, as well as bronchial carcinoids and carcinoids of gastrointestinal origin, recurrent CN alterations were observed in 203 genes, including the RB1 gene and 59 microRNAs of which 51 locate in the DLK1-DIO3 domain. These findings suggest the existence of partially shared CN alterations in these tumor types. In contrast, CN alterations of the TP53 gene and the MYC family members were predominantly observed in SCLC. Furthermore, we demonstrated that the aCGH profile of SCLC cell lines highly resembles that of clinical SCLC specimens. Finally, by analyzing potential drug targets, we provide a genomics-based rationale for targeting the AKT-mTOR and apoptosis pathways in SCLC.

Neuroendocrine tumors of the lung compose a spectrum of tumors ranging from low-grade typical carcinoids to small cell lung cancer (SCLC) (1). SCLC and bronchial carcinoids are characterized by neuroendocrine features and are thought to be derived from the Kultschitzky cells of the bronchial tree (2). These tumors have distinct histological morphologies and very different clinical behaviors (1, 3). The 10-y overall survival rate of patients with a typical bronchial carcinoid has been reported at 87%, whereas the median overall survival rate of SCLC is only 2–3 mo if left untreated and approximately 1 y with treatment. Whereas SCLC is typically caused by cigarette smoke, the etiopathogenesis of carcinoids is obscure, and no clear correlation with smoking has been found (1). Although both bronchial carcinoids and carcinoids of gastrointestinal (GI) origin are well-differentiated neuroendocrine tumors, there are site-specific differences between the two tumors in terms of clinicopathological features, disease course, and prognosis (4, 5). It is presumed that bronchial carcinoids are embryologically derived from the foregut, whereas most carcinoids of GI origin are derived from the midgut or hindgut. The cells of origin of these three tumors are possibly different, and the carcinogenetic events are likely to be distinct (6). It is not clear whether a similar sequence of early oncogenic events may occur at the origin of bronchial carcinoids, carcinoids of GI origin, and SCLC.

Traditional cytogenetic studies and conventional comparative genomic hybridization (CGH) analyses have been performed in both SCLC [7–10; reviewed by Balsara et al. (11)] and bronchial carcinoids [12, 13; reviewed by Leotlela et al. (14)]. These studies indicated that SCLC and bronchial carcinoids display distinct chromosomal imbalances and that, in general, bronchial carcinoids tend to have fewer chromosomal imbalances (15). Further advances of these studies, however, are hampered by small case numbers as well as by low resolution attained by conventional CGH analysis. Array-based CGH (aCGH) analysis has the advantages of high resolution and high-throughput genome-wide screening of genomic imbalances (16–18) and has been used to successfully characterize several malignant disorders (19, 20).

Many lung-cancer cell lines have been developed and extensively used as research tools to study cancer biology and to aid in drug development (21–23). Whether cancer cell lines are reliable representatives of in vivo cancer biology is, however, still a matter of debate. Cancer cell lines are developed under selection pressure and, after many passages in vitro, may no longer be representative of the original primary culture from which the cell line was derived (21, 24). In line with this, changes in gene expression were observed during the establishment of SCLC cell lines (25). Thus far, little is known about gene copy number alterations (CNA) between primary SCLC and cell lines derived from SCLC.

Using aCGH analysis, we explored genomic alterations in SCLC, bronchial carcinoids, and carcinoids of GI origin. We also compared to what degree SCLC cell lines are similar to clinical SCLC specimens. We were able to demonstrate that SCLC cell lines and SCLC tumors displayed very similar aCGH patterns. Furthermore, we identified genetic alterations that could define potential targets for treatment of SCLC and bronchial carcinoids.

Results

Recurrent Copy Number Alterations in Pulmonary Neuroendocrine Tumors.

Fig. 1 depicts the karyotypes of SCLC tumors, SCLC cell lines (Fig. 1A), bronchial carcinoids (Fig. 1B), and carcinoids of GI origin (Fig. 1C). As expected, the karyotype of bronchial carcinoids was more conserved than that of SCLC tumors or SCLC cell lines. Whereas recurrent CNA tended to involve the whole arm of some chromosomes in SCLC tumors or cell lines, the alterations were more often located in narrower gene regions in bronchial carcinoids. Fewer cytogenetic bands and fewer genes in these bands with recurrent CNA were observed in bronchial carcinoids compared with SCLC tumors or cell lines (Table 1). In SCLC tumors, there were recurrent copy number (CN) gains on chromosomes 1, 3q, 5p, 6p, 12, 14, 17q, 18, 19, and 20 and recurrent CN losses on chromosomes 3p, 4, 5q, 10, 13, 16q, and 17p. This pattern is comparable to what was found in prior cytogenetic studies or by conventional CGH analysis (9–11). In bronchial carcinoids, recurrent CN gains were observed on chromosomes 5, 7, and 14, and recurrent CN losses were observed on chromosomes 3, 11, and 22q. In carcinoids of GI origin, regions of recurrent CNA were disseminated across autosomes with the exception of CN gain of the chromosome 20. Three of the 9 carcinoids of GI origin harbored CN gain of chromosome 5, which was observed recurrently in bronchial carcinoids.

Fig. 1.

Frequency of genomic alterations in neuroendocrine tumors and cell lines. Genome-wide frequency of copy number alterations in (A) SCLC tumor samples (n = 33) and SCLC cell lines (n = 13), (B) bronchial carcinoids (n = 19), and (C) carcinoids of GI origin (n = 9). (D) Comparison of the frequency of copy number alterations in SCLC tumors obtained from lung (n = 19) or from metastatic sites (n = 14). Gains and losses are shown in green and red, respectively.

Table 1.

Baseline characteristics of all samples and cell lines

	SCLC tumors	SCLC cell lines	Bronchial carcinoids	Carcinoids of GI origin
Number	33	13	19	9
Age (y)
Median	69		56	63
Range	52–94		30–83	30–85
Gender
Male	15	12	3	4
Female	18	1	16	5
Origin of samples
Lung	19		19
Mediastinum	6
Lymph node	2
Liver	3			1
Soft tissue	2	1
Brain	1
Pleural effusion		8
Bone marrow		4
Small intestine				6
Rectum				2
Cytogenetic bands with recurrent CNA
Gains	122	98	86	92
Losses	48	71	45	89
Genes in cytogenetic bands with recurrent CNA
Gains	8459	6851	536	3406
Losses	5085	7232	1021	1178
MicroRNAs in cytogenetic bands with recurrent CN aberrancy
Gains	320	201	69	196
Losses	164	255	37	34
Cytogenetic bands with high CN gain	4	11	0	0
Genes in cytogenetic bands with high CN gain	41	39	0	0
MicroRNAs in cytogenetic bands with high CN gain	1	6	0	0

CN, copy number; CNA, copy number alternation, referring to cytogenetic band with log₂ ratio > 0.2 (gains) or < −0.2 (losses); high CN gains are cytogenetic bands with log₂ ratio >3.

High Copy Number Gains Are Associated with SCLCs but Not with Carcinoids.

Considering that gene amplification is common in cancer and often related to activation of specific genes and pathways with oncogenic properties (26), we studied cytogenetic bands or genes with high CN gain (log₂ ratio >3). The nonprotein coding plasmacytoma variant translocation (PVT1) gene was of interest because it is immediately downstream of the MYC gene and thought to be oncogenic (27, 28), and PVT1-CHD7 fusions were found in the NCI-H2171 and LU-135 SCLC cell lines (29). In our aCGH study, PVT1 intragenic CN gain was observed (SI Appendix, Fig. S1). By using real-time PCR to study the CN of the PVT1 gene in SCLC cell lines, we found that the CN could be higher than 100 for specimens with high CN gain in the PVT1 gene, and intragenic CNA in the NCI-H82 and NCI-H620 cells was confirmed (SI Appendix, Table S1). High CN gains were not observed in bronchial carcinoids or carcinoids of GI origin (Table 1). High CN gains were found in four distinct cytogenetic bands in SCLC tumors (SI Appendix, Table S2), encoding 41 genes (SI Appendix, Table S3). Nine SCLC cell lines were observed to have high CN gains of 11 cytogenetic bands, encoding a total of 39 genes (SI Appendix, Table S4).

High CN gains were observed for the fibroblast growth factor receptor 1 (FGFR1) gene in one SCLC tumor sample (SI Appendix, Fig. S2A) and for the Janus kinase 2 (JAK2) gene in another sample (SI Appendix, Fig. S2B), suggesting that amplification of these tyrosine kinases could have been driving the malignant behavior of these individual tumors. A high CN gain of the multiple PDZ (MPDZ) gene was detected in one sample; the MPDZ protein (also known as MUPP-1) is a component of tight-junctions (30). Another high CN gain in one sample was detected in the chromosomal region 19q13.12, which contains 10 genes encoding factors involved in transcriptional regulation (SI Appendix, Table S3). Eight (61.5%) SCLC cell lines carried high CN gains of genes of the MYC family (SI Appendix, Table S5). CN gains (log₂ ratio >0.2) of at least one MYC family member were detected in 27 of 33 SCLC tumors (81.8%) and in 1 of 19 bronchial carcinoids (4.3%, P < 0.001), which is in agreement with the notion that alteration of MYC family genes correlates with tumor aggressiveness.

Genetic Alterations Shared by SCLC and Carcinoid Tumors.

Because both SCLC and carcinoids share neuroendocrine features (2, 31), we hypothesized that they may share common genetic alterations during the process of carcinogenesis. The number of genes and microRNAs affected by CNA in SCLC tumors, bronchial carcinoids, and carcinoids of GI origin is depicted in Fig. 2. In total, 203 genes and 59 microRNAs were found of which CNA were common for SCLC tumors, bronchial carcinoids, and carcinoids of GI origin (Fig. 2; Table 2; and SI Appendix, Table S6). None of these genes or microRNAs maps to a known chromosome fragile site (32).

Fig. 2.

Common copy number alterations across three types of neuroendocrine tumors. Venn diagram of (A) genes and (B) microRNAs with recurrent CNA shared by SCLC tumors, bronchial carcinoids, and carcinoids of GI origin. G: gains; L: losses.

Table 2.

Cytogenetic bands with recurrent CNA in SCLC tumors, bronchial carcinoids, and carcinoids of GI origin

Cytogenetic band	No. of genes	Region length (bp)	Selected genes*
Gains
1 q25.1	2	237,106
2 p11.1	1	213,248
5 p15.33-p15.31	16	5,580,620	IRX2, IRX1, ADAMTS16, MED10, SRD5A1, POLS, FASTKD3
5 p15.2	7	3,704,524	DAP, CTNND2
5 p15.1	5	1,188,763
7 q11.22	2	29,091
7 q11.21	3	144,657	PMS2L3
7 q11.21–11.22	60	3,531,411	HIP1, ELN, FZD9, LIMK1, BAZ1B, GTF2IRD1, GTF2I, GTF2IRD2, GTF2IRD2B
12 q24.22-q24.23	1	622,164
14 q32.2-q32.31	49	1,099,667	RTL1
17 q24.3	1	1,594,088	SOX9
17 q25.1	1	63,059
17 q25.1	3	77,123
17 q25.3	2	765,774	CBX2
Losses
11 q21	8	698,251	MAML2
13 q13.3-q14.11	18	2,882,639	FOXO1, ELF1
13 q14.2	4	300,043	ITM2B, RB1
13 q32.3	6	673,267
15 q22.31	2	350,716	DAPK2
22 q13.1	12	1,185,129	SGSM3, EP300, RBX1

*Genes classified by Gene Ontology (GO) as related to cell proliferation, cell differentiation, cell cycle regulation, apoptosis, or DNA damage and repair are presented.

The retinoblastoma 1 (RB1) tumor suppressor gene was among the genes encoded on chromosome 13q that was commonly lost (SI Appendix, Table S6). Notably, the whole cluster of imprinted genes delineated by the delta-like homolog 1 gene and the type III iodothyronine deiodinase gene (DLK1-DIO3), located on chromosome 14q32 where many small nucleolar RNAs (snoRNAs) and 51 microRNAs are encoded, was on the list of shared genes (details of the snoRNAs are listed in the SI Appendix, Table S6). The delta-like homolog 1 (Dlk1) protein is expressed in neuroendocrine-lineage tumors, such as adrenal gland tumors, neuroblastoma, pheochromcytoma, and some SCLC cell lines (33, 34). Eight microRNAs in the DLK1-DIO3 domain are potential tumor suppressors (35). This suggests that genomic alterations of the DLK1-DIO3 domain may play a role in the development of tumors with neuroendocrine features.

Similarity of Copy Number Alterations Between SCLC Tumors and Cell Lines.

SCLC cell lines shared similar karyotype patterns with SCLC tumors, with the exception of more CN losses observed on chromosome 2 in cell lines (Fig. 1A). Among ~20,000 genes analyzed, only 771 genes (~4%) showed significantly different frequencies of CNA (P < 0.01) between SCLC tumors and cell lines. In contrast, 7,868 genes (~39%) were observed to have significantly different frequencies of CNA between SCLC tumors and bronchial carcinoids and 4,189 genes (~21%) between SCLC tumors and carcinoids of GI origin. It should be noted that most SCLC cell lines were derived from malignant pleural effusions or bone marrow cultures rather than from the primary tumors (Table 1). However, among 33 SCLC tumors, samples obtained from the primary lung (n = 19) and metastatic sites (n = 14) demonstrated similar karyotype patterns (Fig. 1D), and CN losses on chromosome 2 were not observed in SCLC tumors obtained from metastatic sites.

We further selected cytogenetic bands in which the difference of frequencies of CNA between SCLC tumors and cell lines was greater than 50% and statistically significant (Table 3). In total, only 74 genes were identified in these cytogenetic bands (SI Appendix, Table S7). Cytogenetic band 7q34 was the only band for which a CN gain was observed more frequently in cell lines. This band contains two genes, PRSS1 and PRSS2, encoding serine protease trypsin 1 and trypsin 2, respectively.

Table 3.

Cytogenetic bands in which the difference of CNA frequencies between SCLC tumors and cell lines were more than 50%

Cytogenetic band	No. of genes	Region length (bp)	Frequency in tumors (%)	Frequency in cell lines (%)	Selected genes*
Gains
1 p36.32	6	172,645	69.7	15.4	TNFRSF14, HES5, MMEL1
1 p36.32	1	417,368	60.6	7.7	PRDM16
1 p36.31	10	404,420	69.7	15.4	HES2, HES3, ZBTB48, CAMTA1
1 p36.31-p36.23	6	1,066,347	66.7	15.4
7 q34	3	228,766	18.2	69.2	PRSS1, PRSS2
18 p11.21-q11.1	1	1,143,946	54.5	0.0
Losses
2 q13	1	166,560	15.2	69.2
2 q13	2	172,569	18.2	69.2
2 q14.1-q14.2	4	1,229,775	18.2	69.2
2 q14.2	2	125,682	15.2	69.2	STEAP3, EN1
2 q14.2	1	52,720	9.1	61.5	CLASP1
2 q21.1	24	1,490,881	12.1	69.2
2 q21.3	5	645,481	18.2	69.2	ACMSD, CCNT2
2 q22.1	1	618,210	18.2	69.2
2 q22.1	1	396,539	24.2	76.9
2 q22.1	1	229,309	24.2	76.9
2 q22.3	1	285,232	15.2	69.2
2 q31.1	1	146,828	15.2	69.2
2 q33.3	3	232,140	9.1	61.5

P < 0.001 for all cytogenetic bands.

*Genes classified by Gene Ontology (GO) as related to cell proliferation, cell differentiation, cell cycle regulation, apoptosis, or DNA damage and repair are presented.

Potential Drug Targets of Bronchial Carcinoids and SCLC Identified by aCGH.

To identify genes that may serve as predictive biomarkers for anticancer therapies and/or constitute potential targets of approved drugs and drugs under development, we performed real-time PCR to confirm the accuracy of CNA of 10 selected genes detected by aCGH assay (SI Appendix, Fig. S3). We observed a highly significant correlation (P < 0.001) between the ratios, obtained by real-time PCR and by aCGH, of the CN of selected genes to the endogenous control RPPH1 gene (SI Appendix, Table S8 and Fig. S4). We demonstrated a trend for higher mRNA expression of JAK2 or FGFR1 in SCLC cells with CN gains (SI Appendix, Fig. S5). A total of 31 druggable genes were identified on the basis of our aCGH and real-time PCR analyses (Table 4). The frequencies of the observed CNA in our study were comparable to data obtained from the Tumorscape (http://www.broadinstitute.org/tumorscape/pages/portalHome.jsf) (36).

Table 4.

Frequency of CNA among genes encoding drug target proteins in bronchial carcinoids and SCLC tumors

Genes	Protein	Drug	Event	Tumorscape: all cancers(%)	Tumorscape: SCLC (%)	SCLC tumors (%)	SCLC cells (%)	Bronchial carcinoids (%)
IGF1R	igf1r	MK0646	Gain	16.4	20	21.2	15.4	10.5
EGFR	egfr	Gefitinib	Gain	31.6	37.5	30.3	15.4	26.3
ERBB2	her2	BIBW-2992	Gain	25.7	40	39.4	30.8	10.5
ERBB3	her3		Gain	19	27.5	36.4	15.4	0
ERBB4	her4	Lapatinib	Gain	13.1	15	21.2	0	21.1
KDR	vegfr2	Sorafenib	Gain	10.3	10	24.2	15.4	0
KIT	c-kit	Imatinib	Gain	10.5	10	36.4	15.4	0
MET	met	XL-184, MK8033	Gain	27.6	52.5	18.2	53.8	26.3
FGFR1	fgfr	PD-173074	Gain	19.9	27.5	33.3	30.8	15.8
RET	ret	XL-184	Gain	11.2	7.5	3.0	0	5.3
PDGFRB	pdgfr2	Imatinib	Gain	16.9	12.5	6.1	7.7	26.3
JAK2	jak2	INCB018424	Gain	11.4	42.5	27.3	30.8	15.8
PIK3CA	PI3K	BGT-226	Gain	21.6	57.5	75.8	53.8	21.1
AKT1	akt	GSK690693	Gain	16.3	47.5	63.6	38.5	26.3
PTEN	pten		Loss	24.2	62.5	75.8	69.2	21.1
FRAP1	mtor	Everolimus	Gain	13.4	45	54.5	15.4	26.3
SRC	src	Dasartinib	Gain	29.4	37.5	60.6	38.5	15.8
ALK	alk	PF-02341066	Gain	16.5	27.5	30.3	23.1	10.5
KRAS	k-ras	PD-325901	Gain	24.8	45	21.2	30.8	26.3
PTCH1	patched	GDC-0449	Loss	21.8	27.5	30.3	38.5	0
VHL	vhl	Sunitinib	Loss	23.6	80	75.8	61.5	21.1
CDKN1A	p21	Flavopiridol	Loss	9.7	15	30.3	30.8	21.1
CDKN2A	p16	Flavopiridol	Loss	40	22.5	30.3	30.8	5.3
BRCA1	brca1	PARP inhibitor	Loss	11.8	10	6.1	15.4	5.3
BRCA2	brca2	PARP inhibitor	Loss	27.2	57.5	63.6	69.2	15.8
BCL2	bcl-2	Obatoclax	Gain	11.9	47.5	51.5	61.5	10.5
MCL1	Mcl1	Obatoclax	Gain	36.7	57.5	75.8	61.5	5.3
PMAIP1	noxa	ABT-263	Gain	11.7	45	66.5	61.5	5.3
TP53	p53	Flavopiridol	Loss	34.7	7.5	51.5	61.5	10.5
ERCC1	ercc1	Cisplatin	Gain	13.1	40	51.5	30.8	15.8
RRM1	rrm1	Gemcitabine	Gain	9.4	10	15.2	7.6	0

Also included in this table are data downloaded from Tumorscape (http://www.broadinstitute.org/tumorscape/pages/portalHome.jsf) (36).

In bronchial carcinoids, genes encoding epidermal growth factor receptor (EGFR), hepatocyte growth factor receptor (MET), platelet-derived growth factor receptor 2 (PDGFRB), protein kinase B (AKT1/PKB), mammalian target of rapamycin (FRAP1), and Kirsten rat sarcoma viral oncogene homolog (KRAS) exhibited the most frequent CNA; however, the frequencies of CNA of candidate genes were below 30%. In contrast, the frequencies of CNA of selected drug target genes reached more than 75%. Frequent CNA were observed for genes encoding proteins involved in the PI3K-AKT (PIK3CA, AKT1, and PTEN) and apoptosis (BCL2, MCL1, and PMAIP1) pathways for SCLC, indicating that both pathways may be important for SCLC carcinogenesis or tumor maintenance and are potential drug targets.

Discussion

Genomic CN imbalances can be studied by cytogenetic analysis, conventional chromosome-based CGH techniques, or the recently developed array-based CGH assay. To perform traditional cytogenetic studies, metaphase chromosomes of cancer cells are used for evaluation (7, 8). Whereas conventional CGH studies in SCLC detect genomic alterations involving extensive regions of the genome, i.e., at the megabase level as shown by Levin et al. (9), aCGH analysis identifies genomic alterations at the kilobase level, readily defining the boundaries and thus the genes carried with regions of CNA. [reviewed by Feuk et al. (37)]. Molecular studies of SCLC are often hampered by the difficulty to obtain tumor samples because surgical interventions are relatively rare in the treatment course of this tumor type. Because of this, formalin fixed, paraffin-embedded (FFPE) samples are often the only resource available for further molecular analysis. Our analysis is the largest array-based CGH study performed on pulmonary neuroendocrine tumors, focusing on FFPE SCLC specimens obtained up to two decades ago (38), and is the most comprehensive screening of CNA in the SCLC genome. The karyotype of SCLC samples in our analysis is comparable to what was found in prior cytogenetic studies or by CGH analysis (9–11, 39), which may help open the door for a wider application of aCGH analysis using FFPE SCLC samples. We observed that, in SCLC tumors, recurrent CNA were observed in more than half of all autosomal genes. The intricate and abundant CNA are important hallmarks of SCLC as they may be associated with the higher proliferation rate of SCLC (40).

Using conventional CGH techniques, comparative analyses of the genomic differences among SCLC, bronchial carcinoids, and GI carcinoids have been conducted (41–43). Through comparison of bronchial carcinoids and SCLCs, Ullmann et al. (41) found that carcinoids carried fewer chromosomal aberrancies than SCLC tumors and that SCLC tumors were characterized by many gains and losses in their genome. Zhao et al. (42) reported on marked genomic imbalance differences between bronchial carcinoids and carcinoids of GI origin. The chromosome 11q13 deletion is a hallmark of multiple endocrine neoplasia type 1 (MEN-1)-related foregut carcinoids and was observed in 5 of 19 bronchial carcinoids in our analysis (13, 14). Because carcinoids of GI origin may originate from midgut or hindgut embryologically (5), the chromosome 11q13 deletion was observed in only one of nine carcinoids of GI origin in our analysis. Our analysis was in agreement with the hypothesis of site-specific difference in the carcinogenesis of bronchial carcinoids and carcinoids of GI origin (14).

By comparing somatic CNA shared by several cancer types from a large collection of cancer samples, Beroukhim et al. (36) explored key genes with causal roles in oncogenesis. Considering that the three tumor types in our analysis share neuroendocrine characteristics, we hypothesized that certain common molecular events may be related to the potential common origin of these tumor types. Our analysis addressed what these three tumor types genetically have in common. For example, the RB1 tumor suppressor gene was found to display a recurrent copy number loss in all three tumor types and is one of the leading deleted genes in other cancer types (36). In contrast, recurrent copy number gain of the MYC family members and recurrent copy number loss of the TP53 gene were observed in SCLC tumors as well as in other cancer types (36) but were seldom observed in carcinoids. Amplification of the DLK1-DIO3 domain was observed rarely in cancers other than SCLC and esophageal squamous cell carcinoma (36). We assumed that deletion of the RB1 gene may be an early event of all benign and malignant neoplasms; deletion of the TP53 gene and amplification of the MYC genes may be related to malignant neoplasms only; frequent DLK1-DIO3 domain amplification tends to be a pattern of neuroendocrine tumors.

Cancer cell lines are widely used in research, but their ability to represent primary tumors has been questioned, and only a few studies have addressed this issue. Jones et al. (44) reported that few new mutations were observed during the development of colon cancer cell lines from primary tumors. Others suggested that the genomes of a panel of 84 human non-small cell lung cancer (NSCLC) cell lines are highly representative of the original primary NSCLC tumors from which they were derived (45). By comparing the results of previously reported preclinical studies and clinical trials, Voskoglou-Nomikos et al. (46) suggested that cancer cell lines can be useful in predicting the phase II clinical trial performance of anticancer drugs. Our analysis demonstrated that copy number aberrancies of most genes, some of which encoded potential drug targets, were common in SCLC tumors and SCLC cell lines.

By comparing the CNA of SCLC tumors and cell lines, we found 74 genes locating in cytogenetic bands for which the difference of CNA frequencies between the two groups was greater than 50%. Among the 74 genes, CN gains of the PRSS1 and PRSS2 genes, which encode trypsin 1 and trypsin 2, respectively, are of interest. Trypsin expression is increased in several cancer types and may mediate cancer cell proliferation, metastasis, and invasion through the interaction with matrix metalloproteinases and the proteinase-activated receptor 2 (47). It is possible that trypsin 1 and trypsin 2 may account for the aggressiveness of SCLC in patients and for rapid expansion in culture.

There has been little progress in the medical treatment of SCLC in the past two decades (48). We identified several potential drug target genes on the basis of our aCGH data (Table 4). It is notable that genes encoding members of the PI3K-AKT-mTOR pathway and apoptotic regulating proteins had relatively high frequencies of CNA in SCLC tumors, and codeletion and/or coamplification of different members in the same pathway were observed. Our study lays a strong foundation for further functional characterization of the affected genes in the PI3K-AKT-mTOR and apoptotic pathways in SCLCs, which may help to define multitarget-specific strategies for the treatment of SCLC patients in the future. In addition, we observed high CN gain of the FGFR1 gene and the JAK2 gene in two individual SCLC tumors. Although the frequencies of CN gains of the FGFR gene and the JAK2 gene in SCLC tumors were only 33.3% and 27.3%, respectively, it is possible that subgroups of SCLC may respond to FGFR1 inhibitors or JAK2 inhibitors, given the very high level of amplification of this gene in individual samples. The frequencies of CNA in the drug target list in Table 4 in bronchial carcinoids were less than 30%, and high CN gain was not observed; this is consistent with the low response rate of targeted therapies, including everolimus and sunitinib, in the treatment of bronchial carcinoids (1).

In conclusion, our analyses suggest that CNA of some genes, including the RB1 tumor suppressor gene, were common in SCLC, bronchial carcinoids, and carcinoids of GI origin. We confirmed that SCLC cell lines represent SCLC tumors well karyotypically, but found that they do not contain the full spectrum of high copy number amplifications found in tumors. On the basis of our analysis, we provide a genomics-based rationale for targeting the AKT-mTOR and apoptosis pathways in SCLC.

Materials and Methods

Sample Acquisition.

Thirteen SCLC cell lines were selected for this study. The SCLC cells were GLC4 and GLC4-CDDP (kindly provided by S. de Jong, Department of Medical Oncology, University Medical Center Groningen, 9713 GZ Groningen, The Netherlands), NCI-H69, NCI-H82, NCI-H128, NCI-H146, NCI-H187, NCI-H526, NCI-N592, NCI-H620, NCI-H678, NCI-H792, NCI-H1173. Sixty-one FFPE or cytology specimens of bronchial carcinoids, carcinoids of GI origin, or SCLC cases were included (SI Appendix, Table S9). Samples originated from the VU University Medical Center, Amsterdam, The Netherlands (38); Suburban Hospital; the National Cancer Institute, National Institutes of Health (NIH); Johns Hopkins Hospital; and the University of Pisa. Use of human samples was approved by institutional review boards according to the legal regulations of the participating countries. All tissue samples were reviewed at the National Cancer Institute, NIH. Diagnosis was verified, and tissue sections hosting more than 50% tumor materials were selected for aCGH.

DNA and RNA Extraction.

Total genomic DNA was isolated from cell lines, cytology specimens, or deparaffinated FFPE specimens using reagents of the DNeasy Blood and Tissue Kit, QIAquick PCR Purification Kit (Qiagen) as well as Dako Target Retrieval Solution in protocols optimized for maximum yield from the various types of samples. Total RNA from cell lines was extracted by TRIzol (Invitrogen) in accordance with the manufacturer's instructions. Further details are available upon request.

Array CGH Analysis.

aCGH was performed using the SurePrint G3 Human CGH Microarray Kit 180K or 105K (Agilent Technologies) as well as reference genomic male DNA (Promega). For Cy3/Cy5 labeling, the Genomic DNA ULS labeling kit was used (Agilent Technologies). Slides were scanned on an Agilent Microarray Scanner, followed by data extraction and normalization by Feature Extraction v10.5 software (Agilent Technologies). Data analysis was carried out using Nexus 4.0 software (Biodiscovery). Sex chromosomes were excluded from analysis. The thresholds of log₂ ratio values for gain and loss were 0.2 and −0.2, respectively; the thresholds for high CN gain and homozygous deletion were 3.0 and −1.0, respectively. CNA was recurrent if more than 35% of the samples of a specific tumor type carried the same CNA with a P value less than 0.05.

Real-Time PCR.

Total RNA and genomic DNA from SCLC cell lines were used for mRNA expression and CN determination, respectively. Total RNA was reverse transcribed using High Capacity cDNA Reverse Transcription Kit, and then the mRNA expression was determined using TaqMan Gene Expression Assays (Applied Biosystems). The CN of genes of interest were studied by TaqMan copy number assay (Applied Biosystems), in which the probes for the gene of interest and the endogenous control gene were labeled by FAM and VIC reporters, respectively, and were measured in the same well. Primers for mRNA expression study and for gene CN study are available upon request. The GAPDH gene and the RPPH1 gene were used as endogenous reference controls for mRNA expression study and gene CN study, respectively. Real-time PCR were operated on theABI 7900HT fast real-time PCR system (Applied Biosystems). mRNA expression was analyzed by the 2^−ΔΔCt method, gene CN was analyzed by CopyCaller software v1.0 (Applied Biosystems), and the CN of a gene in a sample was calibrated to the CN of reference genomic DNA (Promega), which was supposed to be two.

Statistical Analysis.

Comparisons of CGH patterns between cancer cell lines and clinical samples and between different histotypes were analyzed by Fisher's exact test. P < 0.01 was regarded as significant.