p21 promotes oncolytic adenoviral activity in ovarian cancer and is a potential biomarker

Oncolytic viruses multiply selectively within infected cancer cells and cause death, with release of mature viruses that infect neighbouring cells. Upon infection, the first adenoviral protein to be expressed is E1A, which is required for the efficient transcription of other viral early genes [1]. Another function is to drive infected cells into S phase by disrupting the interaction between pRb and E2F [2], allowing transactivation of genes necessary for viral DNA replication. Two E1A conserved regions are responsible for this disruption: CR2 binds with high affinity to the B-domain of the pRb pocket whilst CR1 displaces E2F from the E1A CR2/pRb complex by low affinity binding with pRb directly at the E2F binding site [3].

We have shown that the E1A CR2 deleted adenovirus dl 922-947 has considerable activity in ovarian cancer and induces cell death through a non-apoptotic mechanism [4]. It is more potent than E1A wild-type adenoviruses and the E1B-55K mutant dl 1520 (Onyx-015, H101) [5, 6]. dl 922-947 replicates selectively in cells with abnormalities of the Rb pathway and consequent G1-S checkpoint, findings seen in over 90% of human cancers [7]. We also showed that dl 922-947 activity is associated with deregulation of multiple cell cycle checkpoints and that accelerated cell cycle progression enhances efficacy [8]. In ovarian cancer, multiple G1-S cell cycle abnormalities are observed [9, 10]. However, it is unclear which of these are most important for determining sensitivity to dl 922-947, nor is there a simple biomarker assay of virus activity. Clinical trials of E1A CR2-deleted adenoviruses are underway (http://​www.​clinicaltrials.​gov reference NCT00805376), so understanding these factors will aid identification of patients most likely to respond.

Our data indicate that infectivity is not the only determinant of cell sensitivity, so we have focussed on post-infection events. There is poor correlation between extent of viral replication and cell death when comparing different cell lines. Basal expression of p21 appears an important factor in identifying cells sensitive to adenovirus cytotoxicity and correlates with expression of E1A, death in vitro of malignant and transformed cells and also with anti-tumour activity in vivo. We also show that p21 is predominantly cytosolic and is targeted for proteasomal destruction after infection. Knockdown of p21 in high-expressing cells reduces E1A expression and adenovirus activity, whilst re-expression in p21^low cells increases E1A expression and the cytotoxicity of both dl 922-947 and wild-type adenovirus. Finally, we show that p21 stabilises cyclin D expression and thus promotes a cellular environment conducive to adenovirus replication.

A key step in development of novel cancer therapies is the identification of biomarkers that can predict which patients might respond to treatment. Our results suggest that expression of p21 may be a predictive biomarker for the oncolytic adenovirus dl 922-947. Many investigators have focused on expression of CAR (Coxsackie Adenovirus Receptor), the primary receptor for Group C adenovirus, as the most important determinant of adenovirus function [13]. Undoubtedly, the ability of virus to infect the host cell is vital for subsequent activity. However, the largest study of primary ovarian cancers reveals that the majority retain expression of CAR [14], whilst primary ascitic cells can show demonstrable CAR and α_vβ_3/5 integrin expression [15]. We show that infectivity alone is a poor predictor of cell sensitivity to virus-induced death. It is also clear that the process of adenovirus infection involves receptors and co-receptors other than classical CAR and α_vβ_3/5 integrins, [16, 17].

The first adenoviral gene to be expressed is E1A, which relies upon host cell transcription factors, including EF-1A, E2F and Sp1 [18, 19]. As E1A expression correlates directly with overall virus efficacy, the ability of the host cell to permit expression will profoundly influence all subsequent parts of the lifecycle. We show that p21 expression in uninfected cancer cells is associated with such a permissive state in cancer cells. One function of E1A is to induce infected cells into S phase, the phase most conducive to viral DNA replication. The two cells lines that were most sensitive to viral efficacy expressed p21 prior to infection and had the highest rate of S phase in uninfected asynchronous populations, although this did not correlate with rate of growth in vitro or in vivo or the state of pRb phosphorylation. pRb is phosphorylated on multiple sites by cyclin/cdk complexes [20], with individual complexes preferentially phosphorylating different residues to alter function [21‐23]. Hyperphosphorylation of pRb late in G1 heralds transcription of genes necessary for host cell (and viral) DNA replication. However, there is now evidence that pRb can be phosphorylated by kinases other than cdks [24] and that p21 itself can bind to pRb and alter phosphorylation [25], rendering patterns of pRb phosphorylation in asynchronous cells a poor predictor of virus activity. In addition, p21 binds to the same A/B pocket in pRb as E2F and may thus displace E2F independently of pRb phosphorylation [25].

p21 has a multiple roles within cells. Whilst it is a cdk inhibitor (CKI) and a single molecule can completely inhibit cyclin A/cdk1 activity [26], it functions as far more than a pure CKI; firstly, members of the p21 family act as assembly factors, promoting the formation of active cyclin D/cdk4 complexes at low concentrations, also stabilising cyclin D [27, 28]. Our results show that p21 knockdown causes loss of cyclin D expression in uninfected TOV21G and Hct116 cells, whilst its re-expression in ACP-p21 cells increases cyclin D levels. Conversely, knockdown of cyclin D causes a reduction in p21 levels, confirming the interdependency of the two in the absence of genotoxic stress. Secondly, binding of p21 to cyclin D/cdk4 complexes titrates p21 away from cyclin E/cdk2, thus promoting S phase entry [29]. Our data show that p21 is predominantly cytoplasmic, and thus is titrated away from cyclin E/cdk2 to facilitate viral activity. In addition, the two sensitive ovarian cell lines demonstrated greater phosphorylation of histone H1 in basal conditions. This histone is phosphorylated by cdk2 at G1/S transition [30] and is thus a marker of cdk2 kinase activity. Previous studies have indicated that Akt-mediated p21 phosphorylation can result in p21 localizing to the cytoplasm in Her2-positive cancer cells [31] and also inhibit p21 binding to cdk2 [12]. Our results indicate that the most sensitive line, TOV21G, had higher levels of basal AKT (S473) phosphorylation than the other lines, but there was no overall correlation between cell sensitivity and PI3kinase/AKT activity. Finally, p21, when phosphorylated at Thr57 by cdk2, has a role in promoting association between cdc2 (cdk1) and cyclin B and hence facilitates G2/M progression [32]. We have previously shown that dl 922-947 is capable of over-riding multiple cell cycle checkpoints in sensitive cells. This ability is augmented by nuclear expression of survivin, which acts to augment cyclin D activity [8].

In response to genotoxic stress, p21 translocates to the nucleus where it causes cell cycle arrest and promotes nucleotide excision repair following association with PCNA [33]. We show that p21 expression falls following adenovirus infection, with evidence of proteasomal degradation. Expression of E1A alone increases p21 expression through direct transactivation of the p21 promoter [34]. However, any increase in p21 significantly above basal levels would promote its ability to arrest the cell cycle, to the detriment of adenoviral function. Thus, p21 levels fall post-infection. The adenovirus proteins E1B-55K and E4orf6 form the core of an E3 ubiquitin ligase complex that targets cellular proteins, including p53, Mre11 and DNA ligase IV, for proteasomal destruction [35]: p21 may also be a target for this complex. Interestingly, expression of p21 fell only transiently in the ACP-p21 cells, where expression is under the control of a constitutive promoter, suggesting that some p21 loss in Hct116 and TOV21G cells may result from reduced transcription following the destruction of p53 by E1B55K/E4orf6.

Two recent publications have suggested that p21 expression might reduce oncolytic adenovirus activity in cancer cells, including Hct116. Like us, Höti et al [36] found that there was greater expression of E1A in Hct116 p21^+/+ cells than in p21^-/-. However, they found that treatment of cells with valproic acid, a pan-HDAC inhibitor, reduced oncolytic adenovirus activity and was associated with an increase in p21 expression. Recent evidence suggests that the number of genes responsive to valproic acid is at least 100 and may exceed 1000 [37, 38], whilst the pathways inhibited by valproic acid in myeloma cells include not only cell cycle progression, but also DNA replication and gene transcription [39], all of which are required for adenovirus function. In addition, valproic acid will inhibit HDAC3, which has a critical role in S phase progression [40]. Together, these findings suggest strongly that the effects of valproic acid are not mediated purely by p21. Finally, Höti et al indicated that co-infection of Hct116 p21^-/- cells with an oncolytic virus and a non-replicating virus expressing p21 reduced oncolytic virus replication compared to co-infection with a control virus. However, such treatment will force an increase in p21 expression after infection (and thus block cell cycle progression), which is the opposite of the natural expression pattern - like us, Höti et al and others [41] have shown that p21 expression falls after infection; we wish to show in our experiments that it is expression of p21 prior to infection that is relevant. In the second manuscript [42], Hct116 p21^+/+ and p21^-/- cells were infected with a variety of oncolytic adenoviruses, including another E1A CR2-deleted virus Δ24. Results suggested that there was greater anti-tumour efficacy in p21^-/- cells or in cells in which p21 was knocked down via siRNA. However, dose response experiments were performed only at a single early time point (72 hours) after infection and the lowest MOI employed is 1 pfu/cell: in our experiments, the difference in efficacy of both dl 922-947 and the wild-type viruses became evident 120-144 hours post-infection and the IC₅₀ values of 0.01-0.1 were seen; thus Shiina et al will have missed significant differences at lower doses and/or later time points. Also, following siRNA-mediated p21 knockdown, cells were re-seeded prior to infection with adenoviruses, which could significantly alter expression of a cell cycle-related gene even after RNAi; survival is then assessed after exposure to only a single dose of virus, rather than a formal dose response range. There is no assessment of virus protein expression, change in p21 expression following infection or in vivo assessment of the role of p21, nor any demonstration of the effect of p21 re-expression in cells with low endogenous expression. Finally, Δ24 was generated using the adenovirus plasmid pBHG10 [43], which lacks the entire E3 region, including E3-11.6 Adenovirus Death Protein (ADP), whilst dl 922-947 has intact ADP. Although the mechanism by which ADP promotes cell death late after infection is unclear, we and others have shown that deletions or mutations within ADP certainly alter the kinetics of adenovirus-induced death [4, 44] and also impair virus spread [45]. Thus we believe that our thorough examination both in vitro and in vivo does support a role for basal p21 expression in cancer cells prior to infection in promoting an environment conducive to viral replication.

There are conflicting data on p21 in ovarian cancer. In the serous sub-type, p21 expression is frequently lost [46] and this appears to be a poor prognostic factor [47]. However, in clear cell carcinoma, which has low response rates to chemotherapy and poor overall prognosis [48], p21 expression is frequently seen [46, 49]. It is noteworthy that TOV21G, our most sensitive line and which expresses high levels of p21, was derived from a clear cell tumour [50]. In addition, low malignant potential (borderline) ovarian tumours are characterized by p21 expression [51], and our results with TOSE cells, which are transformed but cannot form tumours in nude mice, are consistent with this finding.

Although there are likely to be many potential biomarkers of adenoviral activity, these results indicate that basal expression of p21 in ovarian cancer prior to infection is associated with an environment conducive to oncolytic adenoviral activity and might have use as a biomarker in future clinical trials.