A subset of platinum-containing chemotherapeutic agents kill
cells by inducing ribosome biogenesis stress rather than by
engaging a DNA damage response
Peter M. Bruno
1,2
, Yunpeng Liu
1,2
, Ga Young Park
3
, Junko Murai
4
, Catherine E. Koch
1,2
,
Timothy J. Eisen
2,5
, Justin R. Pritchard
1,2
, Yves Pommier
4
, Stephen J. Lippard
1,3
, and
Michael T. Hemann
1,2
1
The Koch Institute for Integrative Cancer Research at MIT, Cambridge, MA 02139, USA
2
Department of Biology, Massachusetts Institute of Technology, Cambridge, MA 02139, USA
3
Department of Chemistry, Massachusetts Institute of Technology, Cambridge, MA 02139, USA
4
Developmental Therapeutics Branch and Laboratory of Molecular Pharmacology, Center for
Cancer Research, National Institutes of Health, Bethesda, MD 20892, USA
5
Howard Hughes Medical Institute, Whitehead Institute for Biomedical Research, Cambridge, MA
02142, USA
Abstract
Cisplatin and its platinum analogues, carboplatin and oxaliplatin, are some of the most widely
used cancer chemotherapeutics. However, although cisplatin and carboplatin are primarily used in
germ cell, breast and lung malignancies, oxaliplatin is instead used almost exclusively in
colorectal and other gastrointestinal cancers. Here, we utilize a unique multi-platform genetic
approach to study the mechanism of action of these clinically established platinum anti-cancer
agents as well as more recently developed cisplatin analogues. We show that oxaliplatin, unlike
cisplatin and carboplatin, does not kill cells via the DNA damage response. Rather, oxaliplatin
kills cells by inducing ribosome biogenesis stress. This difference in drug mechanism explains the
distinct clinical implementation of oxaliplatin relative to cisplatin and may enable mechanistically
informed selection of distinct platinum drugs for distinct malignancies. These data highlight the
functional diversity of core components of front line cancer therapy and the potential benefits of
applying a mechanism-based rationale to the use of our current arsenal of anti-cancer drugs.
Corresponding author statement: Michael Hemann (hemann@mit.edu) and Stephen Lippard (lippard@mit.edu).
Data Availability Statement
The data that support the findings of this study are available from the corresponding author upon reasonable request.
Author Contributions
P.M.B., Y.L., G.Y.P, T.J.E., J.R.P., D.P.B., Y.P., S.J.L and M.T.H. conceived the idea for the research, designed experiments and
interpreted data. P.M.B., Y.L. and C.E.K. performed experiments. P.M.B. and Y.L. performed bioinformatic analyses. J.M. performed
DT40 sensitivity profiles. T.J.E. performed polysome gradient profiling. P.M.B., S.J.L. and M.T.H. wrote the paper.
Competing Financial Interests Statement
The authors declare no competing financial interests.
HHS Public Access
Author manuscript
Nat Med
. Author manuscript; available in PMC 2017 October 01.
Published in final edited form as:
Nat Med
. 2017 April ; 23(4): 461–471. doi:10.1038/nm.4291.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript
The use of cisplatin in the clinic began over 45 years ago in the absence of understanding of
the cellular and molecular mechanisms that underlie its efficacy
1
. Despite this, cisplatin has
become a component of treatment regimens for at least 18 distinct tumor types
2
. However,
cisplatin-induced side effects and the emergence of resistance to treatment led to the
development of two derivatives, carboplatin and oxaliplatin, which have also seen
considerable clinical use in a wide array of cancers. Interestingly, oxaliplatin has a different
side effect profile than cisplatin and carboplatin, and it is used in colorectal and other
gastrointestinal cancers where cisplatin and carboplatin have minimal efficacy. However, the
decision to use oxaliplatin to treat colorectal cancer was primarily motivated by its activity
against colorectal cancer cell lines in the human tumor cell line panel known as the NCI-60
and not due to a rationale involving its mechanism of action
3,4
. Although the assumption has
been that oxaliplatin, like cisplatin, kills cells by eliciting a DNA damage response, no
satisfactory explanation for oxaliplatin’s unique clinical use and side effect profile has been
identified. Here, we demonstrate that oxaliplatin acts through a fundamentally distinct
mechanism of action relative to cisplatin and we propose that these agents should be used in
a mechanism-targeted manner in the treatment of cancer.
Results
Diverse mechanisms of action for platinum compounds
To examine the mechanism of action of cisplatin and its platinum analogues we used an
RNAi-based functional genetic strategy to predict mechanism of cytotoxic drug action
5–7
.
This methodology has the advantages of being mammalian, isogenic and unbiased by dosage
effects resulting from export or metabolism. Additionally, it has previously been used to
characterize the mechanism of action of other metal based anti-cancer agents
8–15
. It is based
on a fluorescence competition assay using lymphoma cells that are partially infected with
eight short hairpin RNAs (shRNAs) that target distinct genes encoding proteins with known
or putative roles in cell death signaling pathways: p53 (
TP53
), Chk2 (
CHEK2
), Chk1
(
CHEK1
), ATR (
ATR
), ATX (
SMG1
), DNAPKcs (
PRKDC
), Bok (
BOK
) and Bim
(
BCL2L11
). The shRNA-bearing cells either enrich or deplete relative to the uninfected
population based on the survival advantage or disadvantage conferred by a given shRNA
(Fig. 1a). The combined responses of these cells to different drugs constitute drug
“signatures”. Signatures of all classes of clinically used cytotoxic agents have been
generated and assembled into a reference set separated into eight distinct categories of drugs
based on the constituents’ shared molecular mechanism of action (Supplementary Table S1).
A new drug signature can then be classified by a probabilistic K-nearest neighbors algorithm
to determine whether a new drug belongs to a class in the reference set or requires a new
category not represented therein (Fig. 1b).
To eliminate dosage or potency effects from confounding the RNAi signatures, all agents
were administered at a concentration that killed 80–90% (LD80–90) of the cells at 48 h.
These LD80–90 concentrations varied greatly from one compound to the next
(Supplementary Fig. 1a). However, via atomic absorption spectroscopy we determined that
for cisplatin analogues representing low, medium and high potency, the amount of platinum
required inside the cells for killing corresponded to their respective LD80–90 values
Bruno et al. Page 2
Nat Med
. Author manuscript; available in PMC 2017 October 01.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript
(Supplementary Fig. 1b). For instance, pyriplatin treatment at LD80–90 lead to much more
intracellular platinum than the other LD80–90 treatment, indicating each molecule is
relatively less toxic. Thus, differences in cellular uptake were effectively controlled using
LD80–90 concentrations.
Next, to examine the mechanism of action of cisplatin, carboplatin and oxaliplatin we
obtained their RNAi signatures. As expected, cisplatin and carboplatin were both predicted
to be DNA cross-linkers (Fig. 1c,d, Supplementary Table S2, and Supplementary Fig. S2a).
Previously, the three FDA-approved platinum based drugs had been understood to function
primarily as DNA damaging agents that form intra- and interstrand cross-links. These cross-
links are, in turn, removed largely by excision repair, generating single or double-stranded
breaks in the process
16
. However, despite the fact that oxaliplatin putatively forms 1,2-
intrastrand and other cross-links on DNA, oxaliplatin classified as being most similar to
compounds that inhibit transcription or translation (Fig. 1c,d, Supplementary Table S2, and
Supplementary Fig. S2a). These data may begin to explain why oxaliplatin has a different
cytotoxicity profile and clinical application. Additionally, RNAi signatures of cisplatin and
oxaliplatin taken in combination with 5-fluorouracil (5-FU), the primary drug with which
oxaliplatin is paired with, preserved these mechanistic differences (Supplementary Fig. 3).
Prior analysis of NCI-60 data similarly concluded that oxaliplatin acts distinctly relative to
cisplatin and carboplatin, although no cellular function was identified as being responsible
for the difference
17
. Notably, in this prior study, compounds clustered strictly by structure
which is indicative of the NCI-60 methodology’s emphasis on drug metabolism and
transport over mechanism of action
18
.
Interestingly, phenanthriplatin, a monofunctional and highly potent platinum(II) compound,
also classified as a transcription/translation inhibitor (Fig. 1c,d, Supplementary Table S2,
and Supplementary Fig. S2c). Because phenanthriplatin is incapable of making DNA cross-
links, yet also classifies as a transcription/translation inhibitor like oxaliplatin, it suggests
that the ability of oxaliplatin to form cross-links on the DNA is irrelevant to its mechanism
of action. We went on to characterize seven additional platinum compounds and discovered
that most also classified as DNA cross-linkers or transcription/translation inhibitors
(Supplementary Fig. S2b,c and Supplementary Table S2). Curiously, two monofunctional
platinum agents, acriplatin and pyriplatin, were found to have mechanisms of action not
represented in our reference set (Supplementary Fig. S2d and Supplementary Table S2). This
suggests that potential mechanisms of action for platinum compounds extend beyond the
scope of anti-cancer agents in current clinical use. Furthermore, these signature predictions
are maintained in all permutations of leave-one-out cross-validation of the drugs in the
reference set (Supplementary Table S3). Additionally, we used an indicator of structural
similarity, the Tanimoto coefficient
19
, to hierarchically cluster the compounds and found that
structural clustering was unable to recapitulate RNAi signature-based clustering, regardless
of whether the compounds were clustered by their native structure or their anticipated
structure once inside the cell (Supplementary Fig. S4). Thus, mechanism cannot be correctly
predicted based on structure alone.
To more thoroughly examine the differences in RNAi signatures used to classify these
molecules, we performed a detailed analysis of all of their signatures. The most notable
Bruno et al. Page 3
Nat Med
. Author manuscript; available in PMC 2017 October 01.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript
differences included decreased resistance with shChk2 and decreased sensitivity with
shChk1 for the transcription/translation inhibitor-like compounds relative to the DNA cross-
linker-like compounds (Fig. 1e and Supplementary Table 4). As another means of
visualizing the data, we utilized principal components analysis (PCA) to represent the
variance of our data in fewer dimensions. Plotting all of our tested platinum analogues with
canonical transcription/translation inhibitors, DNA cross-linking agents, and Top2 poisons,
we saw that the transcription/translation inhibitors separated from DNA cross-linkers along
the first principal component (PC1) (Fig. 1e). Upon examining the variable contributions
that made up PC1, we saw that shChk2 contributed most strongly among the hairpins. We
also identified Chk2 as the greatest contributor to the distinction between these two sets of
drugs in a p185+
BCR-Abl p19
arf
−/−
mouse model of acute lymphoblastic leukemia (ALL)
20
(Supplementary Fig. S2e,f).
DNA damage response affects response to cisplatin but not oxaliplatin
To confirm the RNAi signature data using a parallel approach, we examined drug response
in the avian DT40 cell line
21
. Here, 40 different DT40 cell lines, each with a different gene
knockout related to DNA damage repair and tolerance, were dosed with five different
platinum agents. In agreement with our RNAi signatures, the DT40 knockouts showed
distinct sensitivities to oxaliplatin and phenanthriplatin relative to the other three platinum
agents (Fig. 2a,b, and Supplementary Tables S5–6). In particular, loss of genes involved in
homologous recombination (HR) (
XRCC2
,
XRCC3
,
BRCA2
) and interstrand cross-link
repair (ICR) (
FANCC
,
FANCD2
,
FANCG
) showed the greatest differences between the two
categories of platinum agents. The relative lack of sensitivity of HR and ICR deficient cells
to oxaliplatin suggests that it, like phenanthriplatin, fails to form intra- and interstrand cross-
links. Interestingly, genes necessary for replication bypass (
POLZ
and
PCNA
) were critical
for all of the platinum derivatives. This result suggests that oxaliplatin and phenanthriplatin
treatment create lesions on the DNA that are only toxic in the absence of normal replication
bypass machinery. In addition, these results were recapitulated using RNAi against several
genes related to DNA damage repair and/or tolerance in
Eμ-Myc p19
Arf
−/−
lymphoma cells
and in
Eμ-Myc p53
−/−
lymphoma cells (Supplementary Fig. S5a,b). Thus, relative drug
sensitivities in the context of DT40 knockout cells support the RNAi-based category
classifications.
Dependency on checkpoint kinases stratifies platinum agents
Given the importance of the two cell cycle checkpoint kinases, Chk1 and Chk2, in
discriminating between the two mechanistic classes of platinum drug action and DNA
damage response signaling, we decided to first confirm that these distinctions are relevant
in
vivo.
To do this, we conducted a cell competition experiment using the
Eμ-Myc p19
Arf
−/−
lymphoma cells that were partially infected with GFP-tagged shChk2 and then tail-vein
injected into syngeneic recipient mice. Tumors from untreated, cisplatin, oxaliplatin or
phenanthriplatin treated mice were then analyzed for GFP percentage. As the
in vitro
data
predicted, shChk2-containing cells significantly enriched compared to uninfected cells in
mice treated with cisplatin but not in mice treated with oxaliplatin or phenanthriplatin (Fig.
3a). These results suggest that dependence on Chk2 activity, a key mediator of the canonical
Bruno et al. Page 4
Nat Med
. Author manuscript; available in PMC 2017 October 01.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript
DNA damage response, represents a primary distinction between the mechanistic
classifications of DNA cross-linkers and transcription/translation inhibitors.
Subsequently, we examined the cell cycle profiles of cells treated with phenanthriplatin,
oxaliplatin, and cisplatin treatment for 12 h at LD80–90. Oxaliplatin and phenanthriplatin
induced a G1 cell cycle arrest whereas cisplatin arrested cells in the S and G2/M phases
(Fig. 3b). We obtained similar results with the three drugs after 24 h of treatment in human
lung adenocarcinoma and colorectal cell lines, A549 and LoVo, respectively (Supplementary
Fig. S6). To determine the mechanistic basis for these cell cycle differences, we examined
signaling pathways that may be engaged following induction of the DNA damage response.
As shown by western blot, p21 (
CDKN1A
) protein is activated more quickly in response to
oxaliplatin and phenanthriplatin than cisplatin (Supplementary Fig. S7a). Additionally,
knockdown of p21 sensitizes cells to oxaliplatin and phenanthriplatin treatment but elicits
resistance to cisplatin (Supplementary Fig. S7b).
Next, to gain further insight into DNA damage signaling in response to these compounds,
we examined the p53 activating kinase, Chk2. Chk2 is activated in response to double-strand
breaks, whereby it goes on to phosphorylate p53 on serine 20 (serine 18 in mice), which
relieves MDM2 inhibition of p53
22
. Thus, treatment of cells with DNA damaging agents
selects for cells harboring Chk2 or p53 hairpins. Consequently, mechanistic characteristics
of DNA cross-linking agents should be discernible by examining canonical markers of DNA
damage. We therefore tested for γ-H2AX and phosphor-ser18 p53 by Western blot at 12 h
with and without hairpins targeting Chk1 and Chk2 for cisplatin, oxaliplatin and
phenanthriplatin. We observed that treatment with oxaliplatin and phenanthriplatin resulted
in γ-H2AX signal, but this was not dependent on Chk2, as it was for cisplatin (Fig. 3c). All
three drugs also elicited total p53 induction and phosphorylation of p53; however, this was
dependent on Chk2 for cisplatin (Fig. 3d). Additionally, the same was observed for both γ-
H2AX and p53 at 4 h comparing cisplatin and phenanthriplatin with and without shChk2
(Supplementary Fig. S8a,b). Moreover, we observed upregulation of p53 transcriptional
targets Puma and Noxa following phenanthriplatin treatment, concomitant with the increase
in p53 levels seen by western blot (Fig. 3e). Subsequently, we examined γ-H2AX and
phospho-ser18 p53 at and prior to 4 h. Phenanthriplatin, and to a lesser degree oxaliplatin,
induced γ-H2AX, phosphor-ser18 p53 and total p53 accumulation, sooner and to a greater
degree than cisplatin (Supplementary Fig. S8c,d). Phenanthriplatin and oxaliplatin also
caused more rapid cell death than cisplatin and doxorubicin (Supplementary Fig. S9). We
confirmed that cisplatin-induced phosphorylation of serine 20 of p53 was Chk2 dependent in
the human LoVo colorectal cell line (Supplementary Fig. S10a). Furthermore, the transcript
levels of the pro-apoptotic gene, Noxa, were increased following treatment with all platinum
agents tested in multiple human cell lines (Supplementary Fig. S10c–e). Taken together, the
early activation of apoptosis along with the early appearance, persistence, and Chk2
independence of γ-H2AX and phospho-ser18 p53 suggest a mechanism of cell death
induced by oxaliplatin and phenanthriplatin that does not rely on canonical DNA strand-
break signaling.
Bruno et al. Page 5
Nat Med
. Author manuscript; available in PMC 2017 October 01.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript