Oncogene Expression and Genetic Background Influence the Frequency of DNA Copy Number Abnormalities in Mouse Pancreatic Islet Cell Carcinomas Free

Human solid tumors are remarkably heterogeneous in genome abnormalities (1, 2). Variation in the spectrum of recurrent genomic abnormalities between tumors arising in different anatomical locations is not surprising given that different organ microenvironments likely pose distinct barriers to incipient neoplasias as a consequence of their particular biological functions and homeostasis. By contrast, the degree of genomic heterogeneity observed in histologically similar tumors arising in the same anatomical site is surprising. One interpretation is that ostensibly similar tumors in an organ arise from cells that subtly differ in maturity, location, or function and, as such, have distinctive genome-wide expression profiles that are benefited by distinct genomic alterations. This notion is to some extent supported by gene expression microarray analyses showing that cancers originating in the same anatomical site can cluster into distinct groups (3, 4); however, tumors within a distinctive expression cluster can still show considerable genomic heterogeneity (5). One source of such heterogeneity may be passage through telomere crisis, which introduces a substantial stochastic component to the genomic aberration composition (6, 7). In addition, the particular genomic aberrations selected in tumors of ostensibly similar histological type may also be influenced by polymorphic variation of modifier genes between individuals and/or variation between individuals in the timing and type of carcinogenic insult. Our goal in this study was to test these latter possibilities. We used the RIP-Tag model of pancreatic islet cell carcinogenesis (8) because our earlier studies indicated that the composition of recurrent aberrations in these tumors was influenced by genetic background and carcinogen exposure (9).

RIP-Tag mice carry the SV40 early region encoding the large T (Tag) and small t oncoproteins under the control of the rat insulin gene regulatory region (RIP; Ref. 8). Tag abrogates the functions of the retinoblastoma and p53 tumor suppressors (10) in the pancreatic islet β cells, leading to the formation of β cell tumors (insulinomas) in every mouse inheriting the hybrid oncogene. We used two different RIP-Tag lines in these studies. In one line, designated RIP1-Tag2, transgene expression begins at about E8.5 in progenitor cells of the endocrine pancreas and rare cells of the thymus, and expression is maintained in their descendants thereafter. Mice in this class develop islet tumors by 10 weeks of age, preceded by neoplastic lesions that are inferred to be stages on a multistep pathway to solid tumors. In the other line, designated RIP1-Tag5, the transgene is not expressed during prenatal pancreatic development or at any time in the thymus. Rather, Tag expression is sporadically activated in individual islet β cells beginning 9–11 weeks after birth (11). These delayed onset RIP-Tag mice develop islet tumors at 20–22 weeks of age (12).

In the present study, we compared recurrent aberrations in the delayed onset RIP1-Tag5 model and early onset RIP1-Tag2 model, both bred into the C3H background, to determine whether the timing of oncogene activation influences the tumor genome aberration pattern. We also compared three sublines of RIP1-Tag2, inbred into three different genetic backgrounds, C57Bl/6 (B6), C3Heb/Fe (C3H), and FVB/N (FVB), to determine whether genetic background influences the tumor genotype.

The RIP1-Tag2;C57B6/J and RIP1Tag2C3HeFeb/J were derived from the same founder animal and subsequently backcrossed to either C57B6 or C3HeFeb/J for >40 generations (n > 40). The RIP1Tag2;FVB/N line was derived by backcrossing the RIP1-Tag2;C57B6/J line (n > 40) to FVBN/J mice for three generations (N3). The RIP1-Tag5 line harbors the same transgene as that of the RIP1-Tag2 lines but was derived independently via injection of C3HebFe/J embryos; the founder animal was bred and line subsequently maintained in the C3HebFe/J strain background.

Pancreases were removed and fixed overnight in zinc-buffered formalin. Fixed tissue was rinsed in 1 × PBS, subsequently dehydrated in 35, 50, and 70% ethanol, and embedded in paraffin using a Leica TP1050 tissue processor. Sections (5 μm) were deparrafinized in xylene and rehydrated in 100, 95, 70, and 50% ethanol and 1 × PBS. H&E staining of sections was carried out using standard reagents (Surgipath) and protocols.

Array CGH was performed as described (14) Gained and deleted clones were identified in each tumor as those with fluorescence ratios above and below the tumor-specific threshold (see Supplemental Information for additional details). For comparison of a given pair of the three lines or sublines, each bacterial artificial chromosome clone was tested for differences in frequencies of copy number, and the adjusted Ps were computed using a permutation-based approach (13).

Tissue sections were incubated in 2% H₂O₂ in methanol for 15 min to block endogenous peroxidase activity. Sections were rinsed in 1 × PBS, blocked in serum-free blocking reagent (DAKO) for 1 h, and then incubated with a rabbit anti-Tag antibody (1:2000 diluted in 0.5 × serum-free blocking reagent) at 4°C overnight. Sections were washed in 1 × PBS for 3 min, repeated four times, then incubated in a 1:200 dilution of Cy3-conjugated AffiniPure F(ab′2) goat antirabbit for 45 min, washed 4 × 3 min in 1 × PBS, and finally mounted in Vectashield Mounting Medium with 4′,6-diamidino-2-phenylindole. Tag levels were accessed using a Zeiss Axioscope-2 microscope and the appropriate light filters. Images were captured using Openlab software and subsequently manipulated in PhotoShop 7 (Adobe).

Although both RIP1-Tag2 and RIP1-Tag5 transgenic mice have been extensively characterized during the past 10–15 years, we elected to assess current representatives of each family by histopathological and biochemical techniques over periods spanning the evident stepwise progression of tumorigenesis (4–16 weeks for the RIP1-Tag2 lines and 10–34 weeks for RIP1-Tag5). Events in the two models are summarized in Fig. 1. Both models showed distinct stages of hyperplastic/dysplastic islets, angiogenic islets, and tumors that arose in a stochastic and temporally protracted fashion. Hyperproliferative islets with quiescent vasculature and angiogenic (dysplastic) islets appeared at 4 and 7 weeks of age, respectively, for both RIP1-Tag2 sublines and at 12–16 weeks, respectively, for RIP1-Tag5/C3H. Nascent solid tumors formed in both models 10–14 weeks after the onset of oncogene expression, with mice eventually succumbing to hypoglycemia as a result of a vast increase in β cell mass. In both models, the more advanced lesions appeared in progressively lower fractions of the islets at risk. Thus, regardless of when oncogene expression commences, the pathway to tumorigenesis is multistep, and Tag expression is necessary but not sufficient to elicit tumor growth.

We evaluated using array CGH the possibility that timing of oncogenesis and/or genetic background influenced recurrent aberrations, comparing genomic profiles of tumors that arose in the delayed onset RIP1-Tag5 line (inbred in C3Heb/Fe) and in three sublines of the early onset RIP1-Tag2 line, each inbred into a different genetic background: (a) C3Heb/Fe; (b) C57B/6; and (c) FVB/N. CGH arrays used in these studies consisted of 1056 bacterial artificial chromosomes distributed at 4–5 Mb intervals across most of the mouse genome and at higher resolution across regions of copy number abnormalities (CNAs) defined previously in RIP-Tag tumors (9, 14) on chromosomes 2, 4, 6, 9, and 16. We analyzed ∼40 end-stage tumors isolated from mice from each of the RIP1-Tag5/C3H line, RIP1-Tag2/C3H subline, and RIP1-Tag2/B6 subline; 12 tumors were profiled from the RIP1-Tag2/FVB subline (Fig. 2).

The recurrent CNAs of the three inbred (B6, C3H, and FVB) RIP1-Tag2 sublines were similar to those detected in an earlier study (9). These included copy number gains on chromosomes 2 and 4 (CNG2 and CNG4) and copy number losses on chromosomes 6, 8, 9, 14, and 16 (CNL6, CNL8, CNL9, CNL14, and CNL16; Fig. 2, A–C). CNL9 was detected in a significant fraction of tumors derived from all three RIP1-Tag2 sublines, with 20% of C3H and B6 tumors exhibiting loss and 83% of the FVB tumors, the difference between C3H and FVB and B6 and FVB being significant (Fig. 2, F and G; P = 0.01). Likewise, CNL16 was detected in 20% of RIP1-Tag2/C3H tumors, 50% of RIP1Tag2/FVB tumors, and 60% of RIP1-Tag2/B6 tumors, with the difference in frequency between B6 and C3H being significant (P = 0.01; Fig. 2 E). Thus, frequency of loss of specific regions of both chromosomes 9 and 16 was associated with genetic background.

A different picture emerges in comparing the CGH profiles of the early onset RIP1-Tag2/C3H and late onset RIP1-Tag5/C3H lines (Fig. 2, B, D, and H). CNG2, CNG4, and CNL8 were detected in both lines. As was the case for two of the three RIP-1Tag2 sublines, CNL14 was not detected in the RIP1-Tag5 line, further supporting the possibility that it conveys a growth advantage only in particular genetic milieus. CNG2, CNL9, and CNL16 were detected at somewhat lower rates in the delayed onset R1T5/C3H tumors, although these differences were not quite statistically significant. However, the frequency of CNL16 was significantly different between the early onset RIP-1Tag2 and late onset Rip1-Tag5 tumors (P < 0.01). CNL6 was detected in >80% of RIP1-Tag5/C3H tumors, whereas only ∼20% of the RIP-1Tag2/C3H tumors exhibited this loss. The region of most significant difference involved the distal portion where we believe at least one novel tumor suppressor gene resides.⁴

Although the most likely mediator of differences in CNA frequency between RIP1-Tag5/C3H and RIP-1Tag2/C3H tumors is the temporal onset and initial pattern of Tag expression, another formal possibility is that Tag oncoprotein levels are significantly different in the two lines. Recent data from a variety of cell culture experiments indicate that responses elicited by various signaling pathways are significantly different in the context of distinct levels of oncogene expression (15, 16, 17), suggesting that differing levels of Tag oncoprotein expression could give rise to the observed differences in the frequency of recurrent CNAs. To address this question, we assessed Tag expression in pancreatic sections from end-stage mice in two of the RIP1-Tag2 transgenic lines (C3H and B6) and RIP1Tag5/C3H transgenic line, using immunofluorescent labeling with a high-specificity rabbit polyclonal anti-T-antigen antibody (Fig. 3). There was considerable variation in Tag staining within a particular type of lesion in each of the transgenic lines, with sporadic cells exhibiting high-level staining in hyperplastic/dysplastic islets (Fig. 3, A–C), angiogenic islets (data not shown), and invasive carcinomas (Fig. 3, D–F). The neoplastic lesions of RIP1-Tag5 line appeared to have a greater frequency of these high-expressing cells than either RIP1-Tag2 subline. However, the degree of variability among all three lines was similar. Notably, the cells in large tumors and at the invasive fronts of carcinomas in both R1T2 lines and the R1T5 line exhibited Tag protein levels that were indistinguishable from those in early progenitor lesions and noninvasive tumors (Fig. 3, D–F). We infer, therefore, that progression to the invasive phenotype in this model is not simply the selection and outgrowth of tumor cells that have high levels of Tag expression but rather represents a distinct stage of malignant progression defined by secondary events, the molecular components of which have, at least in part, been characterized (18, 19).

Cancers are thought to be multifactorial diseases that result from a selective evolution so as to overcome organismic barriers to uncontrolled cell growth. Considerable effort is now being devoted to the elucidation of how individual genomes and the environment interact to influence this tumor development in the face of such barriers (20). We show in this study that aspects of these questions can be investigated in genetically engineered mouse models of cancer that vary in the timing and pattern of oncogene expression and their genetic composition. Specifically, we asked whether recurrent alterations in the genome during tumorigenesis varied as a function of temporal onset and pattern of oncogene expression in the insulin-producing β cells of the pancreatic islets or as a consequence of genetic background in otherwise identical sublines.

The RIP1Tag2 and RIP1-Tag5 lines of oncomice differ markedly in the temporal onset and initial pattern of oncogene expression in the pancreatic islets. In the former, transgene expression commences rather uniformly in all β cells during pancreatic organogenesis, whereas in the latter, expression begins sporadically, in scattered β cells, beginning at 9–11 weeks of age (11). Despite these substantive initial differences in time and pattern of oncogene expression, the pathways of tumor progression are similar and include analogous premalignant stages and time course from incipient neoplasia to end-stage tumors. Given these similarities in pathway, but distinct differences in the dynamics of oncogene expression, we postulated that these transgenic lines could be instructive about whether such differences in the initiating oncogenic event influence the spectrum of genetic changes in the tumors that eventually arise. Our analyses reveal the frequency of CNL6 to be the most significant difference between the prenatal onset RIP1-Tag2/C3H and adult onset RIP1-Tag5/C3H lines, 20 versus 80%, respectively. Conversely, there is a trend toward higher rates of loss at CNL9 and CNL16 in the RIP1-Tag2C3H (and, to a greater extent, RIP1-Tag2/B6 and RIP1-Tag2/FVB) tumors than in RIP1Tag5/C3H tumors. Thus, it would appear that the microenvironment and/or physiology of the oncogene-expressing β cells impart different constraints on the developing tumors that are relieved by distinct genomic alterations, with CNL6 favoring the tumors whose pathway initiated in isolated oncogene-expressing cells amid fields of normal pancreatic islet cells in adults, whereas CNL16 is favored in the context of relatively uniform expression of the oncogene beginning in midgestation pancreatic differentiation. This latter pathway is characterized by a phenotypic latency lasting for 5–6 weeks, from E8.5 when Tag expression ensues to the first appearance of hyperplastic islets with aberrant proliferation and apoptotic indices at 4 weeks of age. Thus, by using the criteria of recurrent genomic abnormalities, arising in ostensibly similar multistage pathways to islet carcinoma induced in the same cell type by the same oncogene, we have found evidence for differing secondary genetic events. These differences may prove to be analogous to the genetic diversity seen in similar human tumors (20), suggesting that human tumorigenesis may also be biased by subtle differences in the patterns and expression timing of initiating oncogenic events.

We also observed differences in the frequency of CNL9 and CNL16 in the RIP1-Tag2 sublines, indicating that genetic background also influences tumor genotype. The rate of CNL16 was significantly higher in the B6 background than C3H background, whereas CNL9 was higher in the FVB background than in C3H and B6 backgrounds. This phenomenon is analogous to tumor susceptibility loci described in the human population, where allelic status of a particular gene, or sets of genes, determines an individual’s predisposition to a particular cancer and/or the genetic abnormalities that arise within (20). With respect to CNL16, the phenomenon described here suggests that in the allelic milieu of the C3H strain, the selective growth advantage imparted by loss of CNL16 is less than that in the B6 strain, thus rationalizing its reduced frequency in RIP1-Tag2C3H tumors. The actual number of distinct loci contributing to this phenomenon is not clear, but one critical allelic difference between these two strains may reside in CNL16 itself. The prediction that follows is that the C3H allele is a less potent tumor suppressor than the B6 allele, thereby reducing or obviating its genetic loss. A similar phenomenon may be functioning with respect to CNL9, and thus, the FVB allele of CNL9 may be a more robust tumor suppressor gene than the B6 and C3H alleles and, therefore, is selected against in most of the tumors of this subline.

The current study demonstrates an unexpected diversity in genetic secondary events that likely result from selective pressures faced by developing neoplasias. Our studies suggest that the factors limiting progression within a tissue may vary in subtle yet important ways, contributing to the genetic diversity of histologically similar cancers. Knowledge of such diversity in secondary genetic events may prove relevant not only to understanding mechanisms of human tumorigenesis but also differential responses to therapy. It is interesting that both CNL6 and CNL16 study map to regions found to be significantly associated with cancer susceptibility using genetic mapping approaches (21). This correspondence suggests the possibility that genomic analyses of recurrent aberrations in oncogene-induced tumors in mice will contribute to identification of human cancer susceptibility genes. Indeed, a recent report that polymorphic genes influencing cancer susceptibility in mice also correlate with cancer risk in humans (22) supports the proposition that genes discovered through analyses of murine tumors will prove relevant to mechanisms of human tumorigenesis.

We thank Ehud Drori, Susan Cacacho, and Marina Vayner for mouse breeding and genotyping and Cherry Concengo for histology.