Oncogenic cooperation and coamplification of developmental transcription factor genes in lung cancer

We used ROMA, a high-resolution array comparative genomic hybridization (CGH) method (13), to profile the DNA copy number alterations of 77 human cancer cell lines and 184 primary human lung tumors. Supporting information (SI) Table 2 summarizes the overall histological classification of all samples. SI Tables 3 and 4 describe individual samples of cancer cell lines and tumors, respectively. Microarray measurements were converted to copy number estimates by using a segmentation algorithm based on Kolmogorov–Smirnov statistics (14). To restrict subsequent analysis to cancer-associated somatic genetic events, we used an automated procedure to mask common germ-line copy number variations (14). Next, we used an automated method, similar to the minimal common region method developed by Tonon et al. (15), to detect high-level focal amplicons (segmented DNA copy number ≥1.5 and ≤5 Mb) and determined their frequency and region of common overlap (see Materials and Methods). Of the 12 most-frequent focal amplicons, nine contained known oncogenes that are very likely to be the driver gene, and three amplicons did not associate with apparent driver genes (Table 1). The 8p12 amplicon contained FGFR1, although Tonon et al. (15) provided evidence that WHSC1L1 was the more likely candidate driver gene in lung cancer. Importantly, five of the five oncogenes that have been considered to be the best-established oncogenes in lung cancer [MYC, KRAS, EGFR, CCND1, and LMYC; reviewed by Minna et al. (8)] were found among the top 12 amplicons (Table 1). Thus it seems likely that the three recurring amplicons without apparent driver genes contain oncogenes important for lung cancer development. These three amplicons, located at 14q13.3, 5p15.33, and 8p11.21, contained between 5 and 11 known genes. We chose to pursue the 14q13.3 amplicon for further analysis, based on its smaller size, lowest gene content, and status as the second most-frequent high-amplitude amplicon (Table 1).

Gain of 14q13.3 was found in 19% of all samples, and high-level amplification of 14q13.3 was found in ≈4% of total samples, including three lung cancer cell lines and eight lung tumors. To visualize the common area of overlap, we plotted the segmented DNA copy number across a 10-Mb region that spanned the 14q13.3 amplicon for each of the highly amplified samples. Each of the individual amplicon structures was different, but two samples (647 and H661) were particularly informative, in that they defined the left and right boundaries of the 480- to 530-kb core amplified region segment shared by all 11 samples (Fig. 1A). The size range of 480–530 kb results from the uncertainty created by gaps in consecutive genome positions of the oligonucleotides on the array. To more precisely map the boundaries of the amplicons in the two informative samples, we used real-time quantitative PCR (QPCR) to measure DNA copy number at ≈15- to 20-kb intervals throughout the two ≈35-Mb boundary regions (Fig. 1 B and C). This analysis refined the boundaries of the commonly amplified core to a 413-kb region containing four protein-coding genes. Three genes, TTF1, NKX2-8, and PAX9, are fully contained within this region and thus are candidate driver genes of 14q13.3 amplification. The fourth gene, SLC25A21, is not a viable candidate, because the common amplified region contains only sequences downstream of its fourth exon.

In addition to these validated genes, there are four University of California, Santa Cruz, gene predictions within the common amplified region that correspond to EST sequences BX161496, AY102071, BC042093, and BC129834. None of these four candidate genes are predicted to contain coding sequences, nor do they have predicted protein orthologs in six other species, unlike the three validated genes that each clearly defined orthologs in three or more of the six species (http://genome.ucsc.edu). To explore the possibility these four predicted genes may possess oncogenic function but in a noncoding capacity, we examined expression of these four predicted genes by reverse transcriptase-dependent QPCR in a panel of RNA isolated from five normal lung tissues, 20 lung adenocarcinomas (AC), and 20 lung squamous-cell carcinomas (SCC) (which included six amplified tumors). None of the four predicted genes were expressed in any of these samples, effectively ruling out this last possibility for their candidacy as driver oncogenes (data not shown).

The genes TTF1 (also known as NKX2-1 or TITF1 for thyroid transcription factor 1) and NKX2-8 (NK2 transcription factor related locus 8) belong to the same NK2 homeodomain-containing transcription factor gene family (16). PAX9 is a member of the “paired box” (PAX)-containing transcription factor family in which the homeodomain is also present in most family members. However, PAX9 lacks the homeodomain but retains the PAX DNA-binding domain (17).

A basic premise of gene amplification as a mutational mechanism relevant to cancer is that the gene copy number gain leads to an increased dosage of messenger RNA and protein (18). To determine the impact of DNA copy number on RNA expression of candidate driver genes, we profiled global RNA expression of six amplified AC and 11 nonamplified lung tumors by using expression arrays. This study led to the detection of positive correlations of DNA copy number and RNA expression of TTF1 (0.067, P = 0.02), NKX2-8 (0.431, P = 0.003), and PAX9 (0.101, P = 0.01), demonstrating an influence of DNA copy number on RNA expression of all three genes (SI Fig. 4). Next, we determined by real-time QPCR the expression of these three genes as well as 14q13.13 DNA copy number in an independent set of 50 lung AC tumors. Using 3-fold DNA copy number increase as the cutoff, five AC samples (10%) were found to be amplified for the 14q13.3 region. NKX2-8 was up-regulated in the range of 4- to 15-fold in all five amplified samples (SI Fig. 5A). Clearly, NKX2-8 is a leading candidate driver gene based on its strong correlation of RNA expression and DNA copy number (Pearson correlation coefficient of 0.692, P = 0.0001). However, we could not rule out the driver gene candidacies of PAX9 or TTF1, because both genes also showed positive, albeit weaker, correlations of RNA expression and DNA copy number (Pearson correlation coefficients of 0.343 and 0.241, respectively, with P values = 0.001). Therefore, functional analysis of the three candidate driver genes was essential for identification of the driver gene(s).

To test the functional consequences of enforced overexpression of candidate driver genes, we tested whether overexpression of any of the candidate driver genes enhanced the tumorigenicity of lung cancer cell lines. We used two lung cancer cell lines, one with the 14q13.3 gain (NCI-H2170, referred to as “H2170” hereafter) and the other without the 14q13.3 amplicon (HCC15). Both of these non-SCLC cancer cell lines are of the SCC subtype. To create stable transfectants, we used retrovirus-mediated gene transfer to enforce expression of individual genes. Overexpression of transfected genes was confirmed by reverse transcriptase-dependent QPCR (data not shown). None of the three candidate genes when overexpressed further enhanced tumor formation of amplified H2170 cells. On the contrary, in nonamplified HCC15 cells, overexpression of either TTF1 or NKX2-8 (but not PAX9) enhanced tumorigenicity by increasing tumor take rate and tumor size (SI Fig. 6), indicating oncogenic function for both TTF1 and NKX2-8 in HCC15 cells. In addition, the observation that the prooncogenic activities of TTF1 and NKX2-8 were manifested only in nonamplified HCC15 cells suggests that functional advantage of increased gene dosages of TTF1 or NKX2-8 may have already been maximized in the amplified H2170 cells.

To explore the effects of these three genes on premalignant lung epithelial cells, we performed experiments with normal human lung epithelial cells immortalized by SV40 large T antigen, i.e., BEAS-2B cells (19). Individual and pairwise combinations of the three candidate driver genes were retrovirally transduced into BEAS-2B cells. We found that, when plated at low densities, BEAS-2B cells expressing individual candidate genes did not form colonies differently from empty vector controls (Fig. 2A). However, all three pairwise combinations (i.e., double transfectants) formed much larger colonies, ranked in the order of TTF1-NKX2-8 → TTF1-PAX9 → NKX2-8-PAX9 by visual inspection (Fig. 2A). To quantify the increased colony sizes, the crystal violet in the stained plates was eluted in a 0.1% SDS solution, and the absorbance at 595 nm was measured (20). Consistent with visual inspection, each of the three pairwise combinations showed 4- to 6-fold higher growth than all of the single-transfectant and double-vector control transfectant cells (Fig. 2B). To validate this synergistic growth effect, we repeated the experiment by using different vectors that contained different antibiotic selection marker genes (see Materials and Methods). With this second system, we were able to construct triple transfectants overexpressing all three genes. Significant enhancement of colony growth was detected at a level similar to what was observed with the double transfectants (SI Fig. 7), suggesting that the maximal synergy was reached in the double transfectants. These results confirmed that there is a growth-promoting synergy among the three genes in the background of premalignant lung epithelia.

The 14q13.3 amplicon does not appear to be limited to a specific subtype of lung cancer. This is particularly intriguing in that Ttf1 protein is rarely detected in the SCC type of lung cancer based on multiple studies in the literature (21–24). Consistent with the literature, in multiple squamous lung cancer cell lines (including the amplified H2170 cells), we have not been able to detect the protein expression of TTF1 by using a highly specific monoclonal antibody [clone 8G7G3/1 (25)] (Fig. 3A). Notably, the RNA message of TTF1 was undetectable in H2170 by reverse transcriptase-dependent QPCR, suggesting gene silencing at the transcription level (data not shown). Thus, TTF1 is ruled out as a driver gene in the amplified H2170 cells. However, that enforced expression of TTF1 enhanced the tumorigenicity of an amplification-free SCC cells (HCC15) implies that TTF1 has oncogenic activity and SCC cells have the physiological context needed to manifest the oncogenic activity of TTF1.

Because amplified SCC cells such as H2170 present a simplified case in which TTF1 is functionally irrelevant because of its lack of protein expression, we wished to use this cell system to address whether continued overexpression of the other two candidate driver genes play a tumor maintenance role. We screened small-hairpin RNA (shRNA) clones to identify the effective ones that could knock down the endogenous protein expression of NKX2-8 or PAX9 in H2170 cells. Preexisting anti-NKX2-8 shRNA clones were obtained from Open Biosystems (Huntsville, AL) (26), and the shRNA inserts were transferred into a RNA polymerase II-based retroviral expression vector (pMLP) that has been shown to consistently induce high suppression of target genes (27). With PAX9, we used the BIOPREDsi webtool (28) and identified three top RNAi sites in the ORF. These three sites were then specifically targeted by using the RNAi Codex webtool (29) to generate 101-mer oligonucleotides that we then cloned into the pMLP vector. After retroviral infection and drug selection to eliminate uninfected H2170 host cells, immunoblotting analysis of cell extracts of stable transfectants identified one anti-NKX2-8 (N2-shRNA; Fig. 3B) and two anti-PAX9 shRNAs (P1- and P2-shRNA; Fig. 3D) as the most potent in suppressing endogenous target protein expression. As judged by scanning densitometry, N2-shRNA conferred nearly 85% reduction of Nkx2-8 protein expression (relative to the empty vector control), and both P1- and P2-shRNA reduced the endogenous Pax9 protein expression to ≈7% of empty-vector control transfectants (Fig. 3 B and D). Each population of stable H2170 shRNA transfectants was evaluated for tumorigenicity, and we found that knocking down NKX2-8 or PAX9 severely compromised the tumorigenicity of H2170 cells. In the case of NKX2-8, the effective N2-shRNA reduced the tumor take rate to only one of five mice, and in the one mouse-forming tumor, the tumor was much smaller (Fig. 3C). With PAX9, the two effective shRNAs also resulted in significantly reduced tumor formation (Fig. 3E). We believe it is unlikely that nonspecific effects of the shRNAs caused the inhibition of tumor formation. First, the degree of tumorigenicity reduction correlates with the severity of knockdowns, with the most effective shRNAs (N2-, P1-, or P2-shRNA) giving the strongest suppression of tumor formation and take rate, whereas other less-effective shRNAs showed milder effects on tumorigenicity. Second, we have assessed the impact of the effective NKX2-8 shRNA (N2-shRNA) in a nonamplified SCC cell line (HCC15). We found that N2-shRNA did not influence the tumorigenicity of HCC15 as it did with H2170 (data not shown), indicating that (i) this shRNA does not have a generalized deleterious off-target effect, and (ii) only a subset of lung cancer cell lines depend on maintaining high levels of NKX2-8 expression. In sum, our studies suggest that both NKX2-8 and PAX9 play a pivotal role in the tumor maintenance of SCC cell with the 14q13.3 amplification.

In an independent genomic DNA copy number data set consisting of 91 clinically annotated lung AC profiled by Agilent's (Austin, TX) 44K platform (a full description of the data set is to be published elsewhere), 17 samples (16%) within this data set were found to harbor focal amplification (≤5 Mb) of 14q13.3 containing the three candidate driver genes (SI Table 5 summarizes the associated clinical and molecular parameters). Eight of these 17 samples (8.8% overall frequency) were considered highly amplified (log₂ of segmented DNA copy number >0.8). In addition to these 17 samples with focal amplification, 12 other samples contained low-level wide gain at 14q13 (SI Table 5). We analyzed the data set for potential correlations of the 14q13.3 amplicon status with the 17 clinical parameters by Spearman rank order correlation calculations and Fisher's exact tests. By stratifying the samples into two groups [early-stage lung cancer (stages I and II) vs. late stage (stage III cancers)], a significant positive correlation was observed between focal amplification of 14q13.3 and late-stage cancer (SI Table 6). Also, focal amplification of 14q13.3 correlates with recurrence (SI Table 5). Thus, the 14q13.3 focal amplification appears to confer more advanced malignant properties to lung cancer cells. In contrast, wide DNA copy number gain involving 14q13.3 did not correlate with late-stage lung cancer or recurrence, reinforcing the basic premise of this study that DNA copy number increases should be considered in the context of whether they are focal or wide.

Primary fresh-frozen lung tumors were obtained from the Cooperative Human Tissue Network (CHTN, Eastern Division, Philadelphia, PA) and from the laboratory of M. Wigler at Cold Spring Harbor Laboratory. Tumor samples, coded anonymously, were examined histologically to ensure at least 70% malignant tissue before CGH study. The majority of cancer cell lines were acquired from either American Type Culture Collection (Manassas, VA) or Deutsche Sammlung von Mikroorganismen und Zellkulturen (Braunschweig, Germany). Genomic DNA were prepared by using the proteinase-K method with at least three rounds of phenol/phenol-chloroform extraction to minimize protein contaminants (45). RNA of cell lines and tumors was prepared by using Qiagen (Valencia, CA) RNeasy mini kit.

CGH microarrays bearing 85,000 oligonucleotides designed to hybridize BglII restriction fragments of human genome were custom synthesized by Nimblegen Systems (Madison, WI) (13). The microarray hybridization was performed essentially as described (46). A more detailed description is presented in SI Text.

QPCR experiments to determine DNA copy number and RNA expression were conducted essentially as described by Mu et al. (47). The QPCR probes were designed by using Primer Express 2.0 software (Applied Biosystems, Foster City, CA), and the sequences are listed in SI Text. shRNA selection and creation of stable shRNA transfectants were as described (26, 27). The actual shRNA inserts cloned into the pMLP retroviral expression vector (27) are listed in SI Text. Retrovirus-mediated gene transfer was achieved with the murine stem cell virus-based expression vectors (see SI Text).