Ovarian cancer is a histologically and genomically complex disease (4
). Morphologically, ovarian cancers can be divided into Type 1 low-grade, low-malignant potential tumors and Type II high-grade, serous carcinomas, carcinosarcomas, and undifferentiated carcinomas. Although outcome has improved recently for patients in the latter group with 5-year survival rates now approaching 50%, the cure rate remains low at approximately 30% (33
). Genomic characterization of Type II tumors suggests that alterations in the TP53
genes occur early in their pathogenesis and cooperate to promote genomic instability. This genomic instability results in diverse subsequent events that are believed to drive ovarian tumor growth and metastatic progression, including alterations that activate the PI3K/AKT pathway (3
As phosphorylated AKT is expressed at high levels in the majority of high-grade, serous ovarian cancers, we sought to define the AKT dependence of ovarian cancer cell lines with the goal of identifying genomic signatures predictive of drug sensitivity. Using an integrative approach, we were able to define four classes of ovarian cancer cells: cells with 1) PI3K/AKT pathway alterations, 2) RAS/RAF/MEK1 alterations, 3) RB1 loss and 4) those wild-type for all the preceding pathways and genes. Although PI3K/AKT pathway activation was common and correlated with AKT dependence, pathway activation was the result of diverse underlying molecular events and pathway activation alone was not sufficient to confer AKT inhibitor sensitivity. Notably, all cell lines with RAS/RAF alterations and those with RB1 loss, including those expressing high levels of phosphorylated AKT, exhibited intermediate or high levels of resistance to AKT inhibition. These results support the testing of AKT pathway inhibitors in patients with serous ovarian cancer, but suggest that AKT inhibition alone will be effective in only a subset of patients.
Given the central role of AKT signaling in normal cellular physiology, there is particular concern that inhibitors of this pathway may exhibit a narrow therapeutic index. One potential approach to minimizing toxicity when targeting this pathway is to selectively inhibit only those AKT isoforms within a specific tumor that promote transformation and/or progression. Each of the three AKT isoforms has been implicated as playing a dominant role in select cancer types (8
). Our analysis of the ovarian cancer cell line panel revealed that AKT1 and AKT2 were ubiquitously expressed whereas AKT3 expression was detectable in only a subset of cell lines. Moreover, only a subset of the TCGA tumors expressed high level of AKT3 mRNA. Based upon these data, we hypothesized that AKT3 inhibition may not be required in some ovarian tumors for maximal antitumor effect. To address this question, we used two highly selective, allosteric inhibitors of AKT that differed only in their potency for AKT3. In AKT3-deficient models such as the PTEN
-null IGROV-1 cell line, the effects of the pan- and AKT1/2-selective inhibitors were identical. Furthermore, knockdown studies using isoform-selective siRNA suggested that AKT1 was the dominant AKT isoform driving proliferation in these cells and that AKT3 inhibition was dispensable. In contrast, a subset of cells expressing AKT3 were sensitive to the pan-AKT inhibitor MK2206 but resistant to the AKTi-1/2. In sum, the data suggest that an AKT-isoform selective approach may be of utility in a subset of patients, but that pan-AKT inhibition will be required in others.
One limitation of cell lines is that they may not accurately reflect the genomic profile of the cancer lineage that they purport to model and thus may not be predictive of clinical efficacy. Such cell line bias may arise as some genetic lesions (e.g., KRAS/BRAF
mutation) may provide a selective advantage to growth in culture (43
). Through serial passage, cell lines may also drift or acquire additional genetic changes that were not present in the initial tumor. To address these issues, we compared the genomic profile of our ovarian cancer cell line panel to that of 316 high-grade, serous ovarian cancers within the TCGA dataset (10
). Our analysis indicated that while PI3K
alterations were common in both the cell line and tumor panels, the specific molecular alterations present within the tumors were only loosely recapitulated within the cell lines. For example, while PIK3CA
amplification was common (17%) and PIK3CA
mutations were rare (1%) in serous ovarian tumors, consistent with other ovarian cancer cell line studies, PIK3CA
mutations were overrepresented within the cell line panel whereas not a single ovarian cancer cell line harbored focal PIK3CA
). Similarly, KRAS
amplification was common in the tumors (11%) but only present in a single cell line, SKOV-8. SKOV-8 cells did express high levels of RAS-GTP and were MEK-dependent, and their response to MEK and AKT inhibitors was similar to those of the OVCAR-5 cell line, which expresses a KRAS
G12V allele, a mutation found in less than 1% of serous ovarian cancers. Differences between KRAS
amplification and mutation, however, may become apparent with further study and thus it would be inappropriate to consider OVCAR-5 as a representative model for the larger cohort of RAS
-altered ovarian tumors, most of which exhibit amplification of wild-type KRAS
. In summary, the data suggest that the currently available ovarian cancer cell lines only modestly reflect the genomic complexity of the human disease and that a richer panel of ovarian cancer cell lines with multiple representative examples derived from each genetic class is needed.
Our integrated analysis of the cell line and tumor panel also highlights the difficulty of using array-based copy number data to identify those patients with functional gene amplifications and deletions. In the case of PTEN, copy number status as scored by either the GISTIC or RAE algorithms correlated strongly with PTEN mRNA expression. Further, PTEN copy-number neutral (diploid) or homozygous deletion calls were good predictors of the presence or loss of PTEN protein and levels of p-AKT expression by immunohistochemistry and reverse-phase protein arrays. However, hemizygous loss of the PTEN gene did not reliably correlate with functional loss of PTEN protein expression by IHC or downregulation of PTEN mRNA expression. These results suggest that in absence of homozygous deletion, copy number data alone was inadequate to accurately characterize PTEN status. A heterogeneous pattern of PTEN expression by IHC was also common suggesting that clonal heterogeneity will prove to be an additional hurdle to the use of array based platforms to accurately identify tumors with functional loss of PTEN.
In summary, our data suggest that the activity of AKT inhibitors will be restricted to tumors harboring genomic alterations within the pathway and that combination therapy will be required to elicit a tumor response or regression in most tumors. On the basis of these data, we predict a low response rate with selective AKT pathway inhibitors when such agents are used alone in ovarian cancers. This reality may necessitate the development of such compounds initially in cohorts of patients from other tumor lineages in which the frequency of defined PI3K/AKT pathway alterations is high. Agents shown to potently inhibit pathway activity in such tumors with an acceptable therapeutic index could then be tested in a combinatorial fashion in ovarian cancer using a truly individualized approach based upon real-time, detailed genomic and proteomic characterization of individual tumors.