In this study, we generated a quantitative signal–response data set for human U2OS osteosarcoma cells in response to doxorubicin-induced DNA damage, and used relational modeling techniques to examine the flow of information between signaling pathways and cellular phenotypic responses. Our data set revealed a series of complex and non-obvious dynamic cellular phenotypic and molecular signaling events whose magnitudes, kinetics, and inter-relationships were highly dependent on both the extent of DNA damage and the surrounding cytokine microenvironment.
We observed that TNFα synergized with doxorubicin to increase cell death after DNA damage, tipping the balance between cell-cycle arrest and apoptosis. This effect was most pronounced at lower levels of DNA damage, and has potential implications for understanding both the response of individual tumor cells to intrinsic DNA damage, and the heterogeneous response of whole tumors to exogenous DNA damaging therapies, depending on local cytokine profiles within and immediately adjacent to the tumor. There have been several reports that exogenously administered TNFα can synergize with genotoxic chemotherapy to improve tumor regression in both human patients and animal models of cancer, but the effect has been largely attributed to the effects of TNFα on tumor vasculature and on the innate and adaptive immune system (Mocellin and Nitti, 2008
). Our data, however, indicate that TNFα can dramatically enhance chemotherapy-induced cell death through modulation of signaling pathways within the tumor cells themselves. The detailed molecular mechanism underlying this enhancement of doxorubicin-induced cell death by TNFα remains unclear, but since TNFα was found to elicit an increase in early JNK and p53 signaling and a loss of late AKT signaling, interplay between these pathways may be implicated in the observed enhancement of cell death. Consistent with this, we observed that the specific inhibition of robust late AKT activity, but not early AKT activity, following high-dose doxorubicin resulted in a >2.5-fold increase in apoptotic cell death (Supplementary Figure S6
), indicating that late Akt signaling is a pro-survival signal under these treatment conditions. The loss of this signal in cells co-treated with TNFα may therefore at least partially explain the dramatically higher levels of apoptotic cell death seen in that treatment condition. Interestingly, the U2OS cell line used in this investigation contains two wild-type copies of a functional NAD(P)H:quinine oxidoreductase 1 gene (NQO1*
1) gene (Fagerholm et al, 2008
). Recent findings have implicated NQO1 in regulating cell death in response to both anthracyclines (i.e., epirubicin and doxorubicin) (Fagerholm et al, 2008
; Jamshidi et al, 2011
), and TNFα (Ahn et al, 2006
) through effects on NF-κB, p53, and cellular anti-oxidant activity. These findings suggest that NQO1 may be a key integrator of the cellular response to these inputs, potentially playing a role in the observed synergistic response to TNFα and doxorubicin co-treatment.
In contrast to the early cell death observed following treatment of cells with high levels of doxorubicin, low-dose doxorubicin treatment resulted in the accumulation and synchronization of a population of cells in late G1 or at the G1/S boundary at early times after damage. This cell-cycle arrest in G1 or at G1/S, which could also include some cells that were damaged during M when checkpoints are relatively inactive (Giunta et al, 2010
; van Vugt et al, 2010
), serves as a cell-fate decision point, with cells either dying actively by apoptosis from this arrested state, or re-entering the cell cycle by proceeding into S-phase in a more or less synchronous manner. While we cannot completely exclude the possibility that a small amount of DNA replication had occurred prior to initiation of apoptosis in the cells that die from this arrested state, we observed similar amounts of cell death in doxorubicin-treated G1 or G1/S-arrested cells when DNA replication was prevented by inhibiting the major replicative DNA polymerases with aphidocolin (). This result indicates that doxorubicin is capable of inducing apoptosis in G1 or G1/S-arrested cells in the absence of bulk DNA replication, and may explain why even slowly proliferating tumors with a substantial population of G1 cells are sensitive to killing by doxorubicin (Grdina et al, 1980
; Colly et al, 1984
; Ling et al, 1996
; Wartenberg and Acker, 1996
). Importantly, the cells in our experiments were freely cycling at the time of DNA damage. Whether doxorubicin induces a similar extent of apoptosis in quiescent G0-arrested cells, was not investigated, although some data indicate that this may occur (Ritch et al, 1982
; Wartenberg and Acker, 1996
). Finally, our treatment protocol in which cells were exposed to doxorubicin for 4 h followed by drug removal is likely to at least partially mimic the bolus administration of doxorubicin that is used clinically. Intriguingly in this regard, the early G1/S arrest followed by synchronous S-phase entry that we observed following low-dose doxorubicin treatment in our in vitro
cell culture model has also been observed in a rat model study of human acute myelogenous leukemia following doxorubicin administration in vivo
(Colly et al, 1984
Some molecules that are directly activated by DNA DSBs displayed the expected response, with high-dose doxorubicin treatment leading to greater activation than low-dose treatment. Other molecules, however, displayed a reversal of this relationship over some or the entire portion of the time course. Unexpectedly complex dynamics were observed even for signaling molecules in the same ‘linear' pathway. For example, the kinetics of ATM activation/autophosphorylation on Ser-1981 was closely correlated with ATM-mediated phosphorylation of Nbs1 on Ser-343 after low-dose DNA damage, but not after high-dose damage, where instead we observed enhanced ATM autophosphorylation, but muted Nbs1 phosphorylation. Likewise, both low- and high-dose damage resulted in a biphasic profile of ATM activation, but phosphorylation of p53 on its ATM site, Ser-15, tracked only with the late phase of ATM activity. These types of observations indicate complex coupling of kinases and substrates with phosphatases and events controlling protein production and turnover. Interestingly, a transcriptional modulator, homeo-domain interacting protein kinase 2 (HIPK2), has emerged as a player in the core DDR response to DSBs, providing an additional level of regulation between ATM and p53 activities (Rinaldo et al, 2007
; Winter et al, 2008
; Puca et al, 2010
). HIPK2 will be an important molecule to include in future studies investigating the complex regulatory interplay between p53 and ATM and life–death decision following DNA damage.
We made use of two distinct mathematical modeling approaches, PLSR and TI-SWR, to facilitate the identification of significant relationships between measured signaling events and phenotypic responses. PLSR is an established relational modeling approach that we have used previously to identify signaling pathways and autocrine cascades involved in cytokine-induced apoptosis in HT-29 colon carcinoma cells (Janes et al, 2004
). TI-SWR, a new, complementary data-driven relational modeling approach explicitly elucidates temporal relationships between particular signaling events and phenotypic responses (see Supplementary information
). In these analyses, PLSR showed that the majority of the variance in the signaling data set was captured by two PCs that respectively associated with cell survival versus death or with cell-cycle arrest versus progression, based on plots of the cellular-response loadings in this same PC space. This PLSR analysis surprisingly implicated ERK1/2 activity in both the G1/S arrest and apoptotic cell death phenotype following DNA damage. This indication was then buttressed by the TI-SWR analysis that discerned a strong role for ERK activity at 2, 4, and 8 h following DNA damage in maintenance of the early, transient cell-cycle arrest in G1/S, and in promotion of apoptotic cell death from G1/S arrest following DNA damage. These predictions from the models were verified experimentally, confirming a role for ERK in cell-cycle arrest and programmed cell death (). While we cannot rule out the possibility that ERK accomplishes this function indirectly by affecting the activity of other DDR molecules, a computational analysis does not suggest this to be the case (Supplementary Table S2
These results are unanticipated in light of ERK1/2's well-established role in promoting survival and progression into the cell cycle in the absence of DNA damage (Ballif and Blenis, 2001
; Chambard et al, 2007
; Meloche and Pouyssegur, 2007
; Junttila et al, 2008
). Furthermore, several studies have implicated ERK activity in these same pro-survival and cell-cycle progression responses in cells under DNA damaging conditions. Tsakiridis et al (2008)
, for example, reported that in non-small cell lung tumors from patients undergoing high-dose radiotherapy with or without chemotherapy, ERK activation was directly correlated with a poor response to treatment, while Kumar et al (2007)
found that inhibition of ERK resulted in greater apoptotic cell death after γ-irradiation in human endothelial cells that over-expressed Bcl-2. Work from Nishioka et al (2009)
showed that inhibition of ERK caused greater growth arrest and apoptosis in NB4, HL60 and freshly isolated AML cells following administration of the DNA-damaging chemotherapeutic agent cytarabine, while Hayakawa et al (1999)
found that Erk inhibition enhanced cell death following cis
-platinum treatment of human ovarian carcinoma cell lines.
Other studies, however, have provided findings consistent with at least part of our results. A small but growing number of studies have indicated roles for Erk in promotion of cell death under a variety of conditions including following DNA damage (Zhuang and Schnellmann, 2006
; Cagnol and Chambard, 2010
). Inhibition of ERK, for example, has been reported to improve survival in A172 human glioma cells exposed to cisplatin (Choi et al, 2004
), and in human hepatocarcinoma cell lines exposed to doxorubicin (Alexia et al, 2004
). Similarly, Liu et al (2008
) reported that Erk inhibition reduced the extent of apoptosis in rat cardiomyocytes exposed to doxorubicin, while Woessmann et al (2002)
found that Erk inhibition reduced apoptosis in osteosarcoma and neuroblastoma cell lines treated with cis
-platinum. Furthermore, a few reports have indicated that Erk may be involved in G2/M control following DNA damage. Tang et al (2002)
have suggested that ERK mediates G2/M arrest following low-dose etoposide treatment and promotes apoptosis following high-dose etoposide. Similarly, Yan et al (2007)
concluded that ERK is necessary for the induction of a G2/M checkpoint following irradiation in MCF-7 cells, where it is apparently required for activation of ATR, Chk1, and Wee1 in response to ionizing radiation.
Our study extends these findings by demonstrating that ERK1/2 is required for the maintenance of a G1/S arrest—a point of regulation that has not been previously reported
—which is induced following exposure to low doses of doxorubicin, and that apoptotic cell death from this induced arrest is ERK dependent. These unexpected observations regarding the role of ERK in cell-cycle arrest and death following genotoxic injury, as revealed through PLSR and TI-SWR analysis of a systems-level, time-dependent signal–response data set, could have important implications for the application of MEK inhibitors in cancer therapy. It is important to note that these findings were made and validated in the U2OS cell line. While U2OS contain two wild-type copies of the p53 gene, this cell line is known to be hypermethlyated at the INK4a/ARF locus, leading to lack of expression of both p16ARF and p14ARF (Park et al, 2002
). G1/S cell-cycle arrest in response to DNA damage may be expected to be less robust in the absence of p16/p14 molecules to reinforce this response. It is possible that a role for ERK in the maintenance of the G1/S arrest response, and death from this arrest, represents a ‘fail-safe' mechanism, and that a role for ERK in this response may be expected to be less pronounced in cell types that express p16/p14. Furthermore, the absence of a robust G1/S checkpoint in these U2OS cells may partially account for the unexpected doxorubicin-induced death in the G1 state. Importantly, hypermethylation and mutations at the INK4a/ARF locus are found in many human cancers, including a large fraction of melanomas and carcinomas (Tannapfel et al, 2001
; Sharpless and Chin, 2003
; Cheung et al, 2009
). Thus, Erk may fulfill this role in a variety of other tumor contexts. Our results imply that ERK pathway inhibitors should be used cautiously in combination chemotherapeutic treatments.
In future work, it will be crucial to determine under what genetic background and DNA-damaging conditions ERK1/2 is pro-survival and under what conditions it can promote apoptosis in order to effectively implement chemotherapeutic regimens that combine DNA damaging chemotherapeutic agents with MEK inhibitors. Computational approaches such as those used and developed in this work will aid in the unraveling of these complex dependencies. Our complementary use of PLSR and TI-SWR for the analysis of a quantitative signal–response data set and identification of novel relationships between molecular signaling and cellular response represents a useful extension of established computational analysis methods. TI-SWR is not limited to exploring signal–response relationships, but could potentially also be used to explore direct relationships between different molecular signals at the same or different points in time (c.f. Supplementary Table S2
). Furthermore, this method seems to offer generalizability to any quantitative, time-dependent signal–response data set and will be useful in the continuing quest to elucidate connections between signaling pathways and cell fate such as life–death decisions following exposure to DNA damaging chemotherapeutic agents.