Eight weeks after transgene induction, the lungs of TI-KM mice were nearly twice as large as those of TI-K mice (Fig. 1b) and total tumor area was significantly increased (Fig. 1c). However, tumor nodule number was not increased (Fig. 1e), indicating that tumor initiation is not affected by Mad2 overexpression. Adenocarcinomas in TI-KM mice exhibited a more aggressive morphology, with invasion into the pleura and higher proliferation rates (Fig. 1d-f) but no significant differences in apoptosis (Fig. 1d). Moreover, a statistically significant decrease in survival was seen in TI-KM versus TI-K mice. Whereas TI-K mice survived for 543 ± 32 days, TI-KM mice survived for 258 ± 11 days (P < 0.0001) (Fig. 1g). All mice died of lung tumors, as determined by postmortem examination. Although overexpression of Mad2 by itself under the CMV promoter induces lung tumors in 35% of animals³, only 1/40 Mad2 overexpressing mice (TI-M) in a Kras wild type background harbored a lung adenoma after 15 months on doxycycline (data not shown). This may be the result of Mad2 transgene expression being restricted to type II pneumocytes in the CCSP-rtTA model compared with ubiquitous overexpression in CMV-rtTA mice. Importantly, neither total nor exogenous levels of Mad2 in TI-KM animals are higher than in TI-M mice arguing that the increase in tumor burden is dependent on combined Kras and Mad2 overexpression (Supplementary Fig. 1a-c).

In both TI-K and TI-KM mice lung adenocarcinomas were SP-C positive, consistent with a type II-like classification (Supplementary Fig. 2a-b), and negative for CCSP⁸ (Supplementary Fig. 2c-d), which was seen in normal cells lining the airways (Supplementary Fig 2e-f). Hence, combined lung specific Mad2 and Kras overexpression leads to the formation of lung adenocarcinomas that are more aggressive than, though histologically similar to, those seen with Kras alone.

We have previously reported that ubiquitous overexpression of Mad2 leads to tumor initiation through the acquisition of a CIN phenotype³. To determine if the CIN generated by high levels of Mad2 could be contributing to the more severe phenotype in TI-KM mice, FISH analysis was performed on lung tissue using chromosome 12, 16 and 17 specific probes. TI-K lung tumors showed higher levels of aneuploidy (13%) compared to normal lung (2%, not shown). However, TI-KM tumors showed a three fold increase in aneuploid cells over TI-K (38% vs. 13%, p<0.0001) (Fig. 2a-b). TI-M animals developed very low levels of aneuploidy (5,7%), consistent with the observed lack of tumor formation. Despite high rates of aneuploidy seen in TI-KM primary tumors, array CGH analysis failed to detect any gross abnormalities (data not shown). This most likely reflects the absence of stable, clonally propagated events that significantly contribute to the CGH profile.

RNA microarray analysis confirmed the diversity of tumor subtypes in the Kras and Mad2 tumors. While the majority of Kras tumors clustered together, the TI-KM primary tumors were markedly different from each other and from the cluster of mutant Kras overexpressors. These results suggest that Mad2 overexpression induces a wide spectrum of chromosomal aberrations and expression changes which are likely contributing to enhanced tumorigenesis. The precise causative events have not been identified (Supplementary Fig. 3).

Continued expression of mutant Kras is required for tumor maintenance⁸. In contrast to what is observed with classical oncogene overexpression, continuous Mad2 overexpression is not required for tumor maintenance, as primary tumors induced after Mad2 overexpression in two different ubiquitous inducible model systems do not regress upon transgene de-induction³. To determine whether the CIN imparted by Mad2 overexpression affected Kras oncogene addiction, TI-K or TI-KM mice with lesions evident by Magnetic Resonance Imaging (MRI) were identified (Supplementary Table 1) and then switched to a regular diet. MRI 2 weeks after de-induction revealed a striking reduction in tumor volume both in TI-K and TI-KM mice (Fig. 2c and Supplementary Fig. 4). Examination of lungs from all mice confirmed normal histology with areas of pleural and parenchymal fibrosis. Thus, CIN associated with Mad2 overexpression is not sufficient to prevent tumor regression following abrogation of Kras and Mad2 expression. It is likely that these tumors regress in response to mutant Kras withdrawal and not Mad2, since, as stated above, de-induction of Mad2 in the CMV-rtTA and tTA systems never leads to tumor regression by itself³.

Next, we sought to determine whether CIN would affect the rate of tumor recurrence. 25 TI-K and 24 TI-KM mice maintained on doxycycline were monitored for tumor appearance by MRI (Supplementary Table 2). When solid tumors were identified (between 12-28 weeks), doxycycline was removed from the diet and the mice periodically screened for tumor re-growth. While TI-K tumors never relapsed after de-induction of the oncogene (even up to one year off doxycycline), tumors recurred in 11 out of 24 TI-KM mice sacrificed 4 to 11 months after doxycycline withdrawal (Fig. 3a-b). These nodules were solid or well defined proliferating papillary adenocarcinomas morphologically indistinguishable from the primary tumors (Fig. 3e). The presence of recurrences in TI-KM animals is not a result of the larger primary tumor size since even in TI-KM animals selected for a small tumor volume, 50% recurrence was observed (Fig. 3c and Supplementary Table 3). In fact, the average volume of primary KM tumors that went on to recur in this selected population was significantly smaller than that of K and of KM primaries that never recurred (Fig. 3c). Mad2 overexpression alone led only to a single tumor in 1 out of 40 animals after 15 months of induction, arguing that the KM relapses are dependent on having expressed both transgenes.

It is, however, formally possible that these recurrences are de novo tumors derived from cells that had expressed Kras and Mad2 but were not part of the initial tumor mass, a possibility that we consider unlikely for the following reasons: no such events have been reported in other inducible mouse models of oncogene addiction^9,10; TI-KM animals maintained off doxycycline never develop lung tumors in this model system; TI-KM mice that were on doxycycline for 5 weeks, during which time no tumors developed, and switched to a regular diet for 35 additional weeks did not develop tumors (Supplementary Fig. 5); and in 3 out of 3 informative cases where primary tumors were small enough to localize by MRI, recurrences appeared at the same anatomical site as the primary tumor (Fig. 3a).

Tumor relapse has been observed in several tetracycline inducible mouse models of oncogene addiction¹¹. Although many of these recurring tumors show doxycycline independent re-expression of the transgene, additional mechanisms responsible for tumor recurrence have also been identified. We performed HA specific immunohistochemistry and/or RT-PCR analysis to detect the Mad2 and Kras transgenes in the primary tumors, recurrent lesions and tumor free lung samples from TI-K and TI-KM mice (Fig. 3d-e and Supplementary Fig. 6b). None of the non-tumor-bearing lungs after doxycycline withdrawal re-expressed the inducible transgenes. Surprisingly, only one of the recurrences re-expressed the Kras transgene and 2/11 re-expressed the Mad2 transgene (which by itself is insufficient to induce tumor formation), suggesting modes of escape that are independent of the initiating oncogene. RT-PCR and sequencing of commonly mutated exons revealed that the recurrences were not a result of secondary point mutations in the endogenous Hras, Kras or Nras loci (data not shown), a common mechanism of acquired resistance to targeted therapies in other inducible mouse models¹² and in human cancer^13-15.

In parallel to what was seen with the primary tumors, recurrent tumors in TI-KM mice were highly aneuploid (39,4%). Lung tissue from TI-K animals after tumor regression showed very low levels of aneuploidy (2,7%), similar to wild type lung, suggesting that this baseline aneuploid level is insufficient to drive tumor recurrence. In contrast, elevated rates of aneuploidy were observed both in the TI-KM primary tumors (38%) and TI-KM lungs after regression (12,9 %) prior to tumor relapse. This suggests that high levels of aneuploidy in the primary tumor and in cells remaining after de-induction of Kras and Mad2 enable survival after oncogene withdrawal thereby facilitating relapse (Fig. 4a-b).

Kras^G12D lung tumors show positive staining for pErk, pStat3, pAkt and pS6, indicating that the MAPK, STAT3 and PI3K pathways are all stimulated by oncogenic Kras ¹⁶ (Supplementary Fig. 7). Only 5 out of 10 KM recurrences retained MAPK pathway activation and 4 out of 10 showed low level pStat3 activity (Fig. 4c and Supplementary Fig. 8). 10/10 recurrences scored positive for AKT, though the different levels and patterns seen by IHC make this unlikely to be the sole determinant of recurrence. To further characterize the signaling pathways implicated in tumor relapse, RNA microarray analysis was performed in 6 TI-KM recurrent tumors. Consistent (Supplementary Fig. 9) with the IHC analysis, recurrent tumors were very heterogeneous, different from each other and from the K or KM primary tumors. Thus, recurrent tumors employ alternative signaling pathways to escape oncogene addiction, a result consistent with the notion that Mad2-mediated CIN in the primary tumor provides extensive diversity and therefore an evolutionary advantage to oncogene driven solid tumors.

The role of aneuploidy in the initiation of human tumors has been modeled extensively in mice^5,17,18. Overexpression of the mitotic checkpoint gene Mad2 is a very common occurrence in human tumors likely due to its direct activation by E2F upon inhibition of the Retinoblastoma tumor suppressor pathway². Indeed, inducible overexpression of Mad2 in mice leads to modest levels of aneuploidy and robust tumor induction in multiple tissue types³. Extremely high levels of aneuploidy are observed in cells with complete loss of mitotic checkpoint function^19-21 but these cells are non-viable, a fact reflected in the observation that no mitotic checkpoint null tumors have ever been reported, regardless of p53 status. While loss of mitotic checkpoint function in some settings can suppress tumor formation in vivo^22-24, whether this is due to intolerable levels of aneuploidy and subsequent cell death or perhaps even non-cell autonomous effects on the tumor microenvironment has not yet been examined.

The requirement for continued expression of an initiating oncogenic stimulus for tumor cell survival (oncogene addiction¹) is a puzzling phenomenon since withdrawal of a deregulated activity might be predicted to lead to reversion of the cell to a normal state. Nonetheless, its occurrence suggests that targeting oncogenes therapeutically might lead to profound tumor regression. Indeed, inhibition of the deregulated kinase activity of BCR-ABL with the drug Imatinib (Gleevec) in chronic myelogenous leukemia is an extremely effective targeted therapy²⁵. Remarkably, patients in the accelerated phase of the disease, in which multiple chromosomal anomalies are observed, still show a strong clinical response albeit with significantly higher relapse rates^26-28. Here we show that a similar phenomenon might take place in solid tumors. CIN induced by overexpression of the mitotic checkpoint gene Mad2 (observed in a wide variety of human tumors) does not alleviate addiction to Kras in a lung tumor model, since profound tumor regression takes place when Kras expression is inhibited whether or not CIN is present in the primary lesion. However, a dramatic increase in tumor recurrence occurs in the chromosomally unstable setting with emergent independence from the original oncogenic stimulus. While targeted therapy resistance is often initially due to point mutations or amplifications of the intended target gene, CIN might provide an escape route even after sequential, successful inhibition of the altered target. Moreover, CIN in the primary tumor may accelerate disruption of DNA repair pathways thereby facilitating the generation of point mutations. Thus, early CIN in tumors, which in some cases directly results from the loss of the Rb tumor suppressor pathway and overactivation of the mitotic checkpoint, may provide the evolutionary fuel responsible for high relapse rates in human cancers that initially respond to targeted or perhaps even conventional chemotherapeutic regimens. Mouse models of cancer which often show benign levels of CIN relative to human cancers may thereby overestimate the efficacy of clinical drug candidates.

1

2