Cancer’s Game
New therapeutic approaches in oncology take aim at the very thing that
makes cancer so untreatable: its ability to evolve drug resistance.
BY CATHERINE OFFORD
I
n the early 2000s, back when biolo-
gist Olivia Rossanese worked as an
investigator at GlaxoSmithKline,
fighting cancer was an exercise in brute
force. Researchers at the company had
set their sights on developing inhibi-
tors of B-Raf, a protein kinase involved
in cell signaling that becomes dysfunc-
tional in many cancers, and “what we
were thinking was that we needed to hit
this . . . protein so hard,” says Rossanese.
“You had to inhibit it 99.999 percent to
shut down the signaling pathway.”
In 2008, Rossanese and her GSK col-
leagues discovered just the sort of com-
pound they were after: a small mole-
22 T H E SC I ENTIST | the-scientist.com
cule, dabrafenib, that potently inhibited
B-Raf and showed striking effects in mel-
anoma patients with certain mutations
in the BRAF gene. With dabrafenib, says
Rossanese, “we see really amazing
responses, and melanomas go away.” The
drug was approved by the US Food and
Drug Administration (FDA) in 2013.
But in a majority of patients, the
effect doesn’t last long. The cancer usu-
ally comes back within just six months
or so—and when it does, it’s resistant
to dabrafenib. GSK’s data showed that
less than half of metastatic melanoma
patients treated with dabrafenib alone
survived more than two years.
LUCY READING-IKKANDA

04 . 2020 | T H E S C IE N T IST 2 3
 It’s a familiar refrain. The vast major-
ity of cancer deaths in the US come
about not because of a lack of treatment,
but because the treatments themselves
stop working in the patients receiving
them. Accordingly, the emergence of
drug resistance is now widely regarded
by oncologists as the biggest challenge
in cancer therapy. Insensitivity to drugs
may arise due to changes in gene expres-
sion that allow a cancer cell to rewire its
metabolism to circumvent the targeted
pathway. Resistance also arises through
genetic mutations, which, provided they
offer a survival or growth advantage, can
come to dominate a population of repli-
cating cancer cells much as they would
a population of organisms undergoing
adaptive evolution in a new environ-
ment. (See “Resist or Desist,” The Scien-
tist, April 2017.)
Growing appreciation of cancer’s
capacity to evolve drug resistance is
revealing fatal weaknesses in the drug-
it-until-it-dies mentality that domi-
nates cancer treatment and drug discov-
ery efforts, says Mel Greaves, director
of the Centre for Evolution and Can-
cer at the Institute of Cancer Research
(ICR) in the UK. During traditional
drug screening, for example, oncolo-
gists “take a drug, put it in a test tube
with a cell culture or cell line of cancer,
and ask if it kills the cells,” says Greaves.
At the ICR, which launched its Centre
for Cancer Drug Discovery last summer,
“we’re saying, that’s just wrong.”
Instead, several groups of cancer
biologists are looking for therapies and
treatment strategies that target cancer
evolution itself, says Rossanese, now
head of biology at the new center, which
claims to have the “world’s first ‘Dar-
winian’ drug discovery program” specif-
ically designed to tackle drug resistance.
This can take the form of manipulating
the course of cancer evolution to clini-
cians’ advantage, or putting the brakes
on the processes that drive it in order
to limit a tumor’s capacity to adapt. In
taking this approach, researchers “just
assume resistance from the start,” Ros-
sanese says. “If you do that, and you
change your mindset that way, then how
would you design drugs? How would
you design trials?”
Cornering cancer
The emergence of resistance to potent
inhibitors such as dabrafenib isn’t sur-
prising, says Rossanese. By the time of
diagnosis, a typical tumor might already
comprise more than 1 billion cells, each
of which has the entire human genome
at its disposal. During the tumor’s
development up to that point, the accu-
mulation of mutations in replicating
cells will have led to heterogeneity, the
substrate for evolution, among different
subpopulations of cancer cells.
When a clinician administers high
doses of a drug that blocks an important

PLAYING TRICKS ON TUMORS

Treating cancers with high doses of tumor-targeting drugs often triggers the evolution of drug resistance,
which leads to tumor progression. Researchers are consequently exploring alternative treatment strategies that
manipulate tumor evolution to a patient’s advantage—by exploiting drug resistance instead of trying to avoid it.

EVOLUTIONARY DEAD END

Clinicians administer a drug and thus select for cells with resistance-conferring mutations. Then, having
narrowed the population down to just those resistant cells, they administer a second drug designed to
target a weakness, what researchers refer to as a “collateral sensitivity,” in those same cells.

Heterogeneous Strong selection for Resistant cells killed

tumour resistant cells thanks to collateral
sensitivity

First drug Second drug

added added
LUCY READING-IKKANDA

Just assume resistance from the start. If you
do that, and you change your mindset that
way, then how would you design drugs?
—Olivia Rossanese, Centre for Cancer Drug Discovery, Institute of Cancer Research
cellular pathway, “the pressure on the
cells to come up with a resistance mech-
anism is quite strong,” Rossanese says.
Any mutation conferring an advantage
in that scenario, even if it’s present in
just a few cells, offers an escape route,
and can quickly sweep through the
population to produce a drug-resistant
cancer that thwarts further treatment.
One way to try to block cancer’s evo-
lutionary escape routes is to use drug
combinations that target multiple onco-
genic pathways at once. For example,
the combination of dabrafenib and tra-
metinib—a drug that targets another
central protein in cellular signaling,
MEK—was approved in 2014 for cer-
tain types of melanoma and later for
other cancers after showing improve-
ment in survival rates compared with
dabrafenib treatment alone. How-
ever, many cancers eventually go on
to evolve multidrug resistance. There’s
also the issue of toxicity: generally, the
more drugs a clinician administers, the
higher a patient’s risk of side effects.
An alternative strategy is to set a
sort of evolutionary trap by administer-
ing a combination of drugs in a particu-
lar order. The aim is to select for resis-
tance to the first therapy before hitting
surviving cancer cells with a second
therapy designed to target a vulnerabil-
ity created by the very mutations that
conferred resistance to drug 1. Known
as evolutionary herding, the method
exploits the fact that any biological adap-
tation often involves trade-offs; being
better at surviving in one environment
may mean being worse at surviving in
another. As part of an announcement
last year about the ICR’s new drug dis-
covery center, computational biologist
Andrea Sottoriva likened the approach
to sending cancer “down dead ends and
to its own destruction.”
To turn the idea of evolutionary
herding into practical cancer therapies,
oncologists are using computational
and experimental techniques to pre-
dict which combinations of drugs, and
in what order, are most likely to work.
In one 2016 study, MIT researchers
exploited the evolution of drug resis-

HARNESSING COMPETITION
Researchers administer low levels of a drug, enough to kill most, but not all, of the vulnerable cells in the tumor population
while favoring the survival of drug-resistant lineages. Once the tumor has shrunk, clinicians stop administering the drug.
The drug-sensitive cells, which tend to have a competitive edge over cells that have invested in a costly drug-resistance
mechanism, can now begin to grow back. Competition between drug-sensitive and drug-resistant cells for resources in the
tumor microenvironment keeps the tumor size in check.

Low levels Low levels

Drug Drug
of drug of drug
withdrawn withdrawn
added added

Heterogeneous Regrowth of non-resistant Regrowth of non-resistant

tumour cells, control of resistant cells, control of resistant
subpopulation subpopulation

Weak selection for Reselection of resistant cells:

resistant cells tumour size under control
tance in a murine model of an aggressive
type of acute lymphoblastic leukemia
(ALL) called Philadelphia chromo-
some–positive ALL. 1 This cancer is
characterized by the fusion of two genes,
BCR and ABL1, and is often treated
with high doses of small-molecule
drugs that inhibit the resulting BCR-
ABL1 oncoprotein.
Blasting ALL cells in vitro with
dasatinib or bosutinib, two common
BCR-ABL1 inhibitors, resulted in the
emergence of resistance to both drugs,
regardless of which compound the cells
were exposed to, the researchers found.
But drug screening revealed that these
resistant cells showed increased vul-
nerability, or “collateral sensitivity,” to
a selection of other drugs. Sequencing
assays revealed a single base mutation,
a guanine-to-cytosine substitution, in
ABL1 that appeared to be responsible
both for protection against dasatinib
and bosutinib and for sensitivity to at
least four other small-molecule drugs—
a weakness that, the researchers write
in their paper, should be “therapeuti-
cally exploitable.”
More recently, Sottoriva and col-
leagues applied a similar approach to
manipulate the evolution of non-small
cell lung cancer (NSCLC) cells in vitro.2
The researchers first bombarded their
cell lines with trametinib, which, as
expected, caused major cell death fol-
lowed by the growth of drug-resistant
cells a few weeks later. Sequencing
revealed that these trametinib-resistant
cell lines had all lost functional copies of
the gene coding for CDKN2A, a protein
that helps regulate cyclin-dependent
kinases (CDK) and, consequently, cell
division. Because the loss of CDKN2A
is known to lead to increased produc-
tion of certain CDKs, the team predicted
that the cells would be particularly sen-
sitive to CDK inhibitors. Sure enough,
such drugs proved twice as lethal in the
trametinib-evolved lines as in control
NSCLC cells.
Efforts to target collateral sensitiv-
ity bring their own challenges, however.
For starters, collateral sensitivity may
be relatively rare, or at least difficult
to identify. In their study, Sottoriva
and colleagues described a second set
of NSCLC lines, this time treated with
gefitinib, which targets another protein
involved in cell signaling called epider-
mal growth factor receptor (EGFR).
The cells duly gained gefitinib resis-
tance, but they didn’t show collateral
sensitivity to any of the nearly 500 other
drugs the team hit them with, including
several that the researchers identified
as good candidates based on genetic
sequencing of resistant cell lines.
Even when researchers can iden-
tify collateral sensitivity, work by sev-
eral groups suggests that it often arises
unpredictably and may be tempo-
rary. Tumors continue to diversify as
cells replicate and accumulate muta-
tions, so cancers may eventually evolve
resistance to both the original treat-
ment and the therapy designed to take
advantage of resulting collateral sen-
sitivities. Charles Swanton, a clinician
scientist at the Francis Crick Institute
and University College London, notes
that in lung cancer, for example, tumor
sequencing data suggest that evolution
becomes “less constrained, not more
constrained” as cancer progresses. “In
terms of forcing tumors down cul-
de-sacs, I think perhaps in very early
stages of disease that might be a fruit-
ful approach,” he says. “But I think in
later stages of disease, the tumor is too
diverse for that to be possible.”

TRACKING CHANGES
To effectively manipulate a tumor’s evolution, researchers need a way of monitoring the various subpopulations of
cancer cells within that tumor. Standard tissue biopsies are impractical for many cancer types and tend to provide
poor measures of tumor heterogeneity: one study of renal carcinoma patients found that a single biopsy identified
on average just a little more than half of the mutations in each tumor (NEJM, 366:883–92, 2012). Many researchers
have consequently switched their attention to liquid biopsies, which pick up cancer-related biomarkers circulating in the
blood and may prove to be cheaper, less-invasive, and more-effective ways of monitoring within-tumor changes over time.
A proof-of-concept study by a team at Institut Curie in France a couple of years ago used whole-exome sequencing analyses
of circulating DNA to detect the rise of tumor subpopulations resistant to chemotherapy in 19 patients with neuroblastoma tumors
(Clin Cancer Res, 24:939–49, 2018), while a team at Asahikawa Medical University in Japan did the same with non-small cell lung
cancer patients receiving tyrosine kinase inhibitors (BMC Cancer, 18:1136, 2018). One analysis published last year found that, in nearly
80 percent of cases, this approach identified clinically relevant, resistance-inducing mutations that had been missed by tissue biopsies,
raising researchers’ hopes that the technique could be effective for monitoring tumor evolution (Nat Med, 25:1415–21, 2019). Other types of
liquid biopsies include analyses of circulating tumor cells and of genetic material (usually RNAs) in tumor-derived extracellular vesicles.
These approaches are being evaluated in multiple clinical trials as a way to monitor patient responses to cancer therapies, but it will be a while
before they’re ready for use in long-term, evolution-focused treatments, says Robert Gatenby of the Moffitt Cancer Center. Researchers don’t know,
for example, if the proportions of various types of circulating DNA directly mirror the proportions of each tumor subpopulation, or whether they “rep-
resent disproportionately the populations that are losing—or the populations that are winning—the evolutionary battles,” he says. “We’re hoping
[these techniques] will help us, but there’s a lot of work that has to be done.”
 Hunger games
About 15 years ago, while perusing
internet news, Robert Gatenby came
across an article about the diamond-
back moth, and the damage that this
pest species had been wreaking on
cabbages and other cruciferous plants
around the world for much of the last
century. Ecologists had realized, the
article explained, that by smothering
their crops in chemicals, farmers were
merely encouraging this fast-reproduc-
ing species to evolve insecticide resis-
tance, while killing off any competing
insects in the ecosystem that might
A cancer cell with a resistance mechanism
would have an advantage over other cancer
cells competing for space and resources—
but only when the relevant drug was present.
have helped keep moth populations in
check. To Gatenby, a radiologist and
codirector of the Center of Excellence
for Evolutionary Therapy at the Mof-
fitt Cancer Center, the parallels to can-
cer were obvious. High doses of can-
cer therapy are “the same as high-dose
insecticide,” he says. “You’re selecting
for resistance, and you’re taking away
competitors.”
He says the story made him won-
der whether oncologists might harness
competition within the cancer ecosys-
tem—that is, among the various clonal
subpopulations making up a tumor—
to stave off the evolution of resistance.
A cancer cell with a resistance mech-
anism would have an advantage over
other cancer cells competing for space
and resources, but only when the rel-
LUCY READING-IKKANDA
evant drug was present, Gatenby rea-
soned in theoretical papers during the
2000s and 2010s. In the drug’s absence,
it would have no such edge—indeed, the
resistance mechanism might even come
with a cost, if the cell directs resources
toward resistance instead of growth and
replication, for example.
This idea has been bolstered by years
of clinical data. For example, one small
2015 study of patients with colorec-
tal cancer who were receiving EGFR-
targeting drugs found that, soon after the
study started, several patients’ tumors
were taken over by cells with mutations
in KRAS, a well-known oncogene that
helps cancer cells bypass normal metab-
olism and confers drug resistance. Yet
when clinicians stopped administering
the drugs, that same KRAS-mutated sub-
population took a hit, with liquid biop-
sies showing declining numbers of these
cells with respect to other subpopula-
tions. The drug-resistant cells appeared
to be drug-dependent.
To turn the idea into a therapeu-
tic strategy, Gatenby proposed cycling
between providing a treatment and
withholding it, thus growing and
shrinking different cancer cell sub-
populations in a way that would limit
the overall size of the tumor and block
drug-resistant cells from taking over
completely. A team at Novartis tested
this strategy in the early 2010s using
patient-derived melanoma xenografts
in immunocompromised mice. The
researchers found that while a con-
tinuous high-dose treatment of mela-
noma tumors with the B-Raf inhibitor
vemurafenib led to lethal drug-resistant
disease in mice within a few months,
intermittent dosing on a four-weeks-
on, two-weeks-off schedule staved off
resistance for the full 200 days of the
study, and drastically slowed overall
tumor growth. 3
04 . 202 0 | T H E S C IE N T IST 27
 More recently, Gatenby’s group has
used the method in patients with meta-
static castration-resistant prostate can-
cer (mCRPC). Patients treated with the
hormone therapy abiraterone usually
progress to a drug-resistant and thus
more lethal form of the disease after
about 16 months. But instead of dosing
continuously, the Moffitt trial’s clini-
cians monitored 11 patients’ blood levels
of PSA, a biomarker for prostate can-
cer, and administered abiraterone only
until PSA had dropped to 50 percent of
its pretreatment level. Then, they sus-
pended treatment, waited a few weeks
or months until a patient’s PSA had
risen back to pretreatment levels, and
started over.
Interim findings published in 2017
indicated that, in patients exposed to
intermittent dosing, the average time
to progression to the more aggres-
sive form of the cancer was at least 27
months.4 Last year, after expanding the
We don’t try to remove the cancer, we try to
keep it at a low level that doesn’t kill you or
impact your life.
—Benjamin Roche, Center for Ecological and Evolutionary Cancer Research
group to 15 men, the team published
an update: six had progressed, but the
rest had not, boosting the median time
to progression to at least 30 months. 5
With the trial now wrapping up after
having accrued more than 20 patients
in total, “those patients, it looks like,
are going to have about a twenty-
month increase in their median time
to progression,” says Gatenby. Plus,
“they’re getting less than half the dose
of drugs that they would have received
otherwise”—a factor that could help
reduce costs and potential side effects.
The team is now testing this approach
in patients with castration-sensitive
prostate cancer, thyroid cancer, mela-
noma, and, as of later this year, meta-
static pediatric sarcoma.
Theoretical work by the Moffitt
researchers suggests that the approach
could be more effective if clinicians use
multiple drugs to control the propor-
tions of cell subpopulations—taking
advantage, for example, of any collat-
eral sensitivities. Simulations the team
published last year indicate that repeat-
edly alternating between abiraterone
and the chemotherapy drug docetaxel
could have significantly increased time-
to-progression in the mCRPC trial,
although effective implementation of
such a regime would require precise
techniques for monitoring the propor-
tion of various tumor subpopulations. 6
(See “Tracking Changes” on page 30.)
One implication of strategies that
exploit competition among tumor sub-
populations is that they shift the goal-
posts for treatment, notes Benjamin
Roche, codirector of the Center for
Ecological and Evolutionary Cancer
Research in Montpellier, France. By try-
ing to maintain “competition between
resistant and sensitive cancer cells . . .
we are completely changing our pri-
orities about the fight against cancer,”
says Roche, who has collaborated with
Gatenby on papers in the past. “We
don’t try to remove the cancer, we try
to keep it at a low level that doesn’t kill
you . . . or impact your life.”
LUCY READING-IKKANDA
This mentality has taken a while to
percolate through the oncology com-
munity, says Gatenby. A decade ago,
the idea that clinicians might chose to
manage, rather than obliterate, a cancer
“was a difficult sell.” Understandably, he
says, “physicians as a group did not like
it. [At conferences] the usual comment
was, ‘This is bullshit. Just give me better
drugs.’” Now, with clinical trials under-
way, Gatenby says, there’s more interest
in these sorts of approaches for cancers
that are all but untreatable.
Greaves says he agrees that con-
trolling, rather than curing, aggres-
sive cancers may be a sensible goal. But
there’s still “a bit of a gap between the
modeling and the clinical practice,” he
notes. Current theory can’t produce
clear treatment recommendations for
individual patients, he continues. “We
aren’t there yet.”
Blocking evolution
Efforts to herd or otherwise manipu-
late cancer evolution assume that the
emergence of drug resistance is inevi-
table, and is thus best directed in order
to achieve clinical benefit. But some
researchers are interested in how a can-
cer gets to be so adaptable in the first
place, and whether that process itself
might be blocked or slowed down. The
key, Rossanese explains, is the hetero-
geneity of the cancer cell population.
“When you increase heterogeneity,
you’re giving evolution a bigger sub-
strate to act upon,” she explains. “So
what if we could reduce some of the
ways cancers generate heterogeneity?”
Although research shows that some
cancer-related mutations arise as a
result of treatment itself, most DNA
errors are generated spontaneously as
cells in the tumor multiply. Data on
when and where these mutations usu-
ally arise are pouring in from research
projects, such as the Pan-Cancer Analy-
sis of Whole Genomes (PCAWG), that
sequence and analyze tumor DNA. For
example, researchers working on the
TRAcking Cancer Evolution through
therapy (TRACERx) project, which
Swanton leads, have sequenced tumors
from hundreds of NSCLC patients in the
UK to explore how mutation patterns
change as cancer progresses. (Swan-
ton also cofounded a company, Achilles
Therapeutics, to develop personalized
T cell therapies targeting antigens iden-
tified by mutations in patient sequenc-
ing data.) Among other things, such
data can help identify particular types of
mutations in a tumor that are associated
with elevated genetic diversity overall,
and thus identify potential targets for
evolution-stalling therapies.
One such target generating interest
among oncologists is the apolipoprotein
B mRNA-editing catalytic polypeptide-
like (APOBEC) protein family, a group
of enzymes that modify nucleic acids by
changing cytosine bases to uracil and
are thought to be involved in the innate
immune response. A 2019 study using
TRACERx and other datasets contain-
ing genomic information on various lung
and thoracic cancers found a strong cor-
relation between APOBEC-driven muta-
genesis and overall tumor heterogeneity,
suggesting that APOBEC activity may be
a significant contributor to the diversi-
fication of cancer cell subpopulations.7
Other research has linked the proteins
to tumor diversity and disease progres-
sion in head and neck, breast, and blad-
der cancer, among others. “We know it’s
an active process that’s driving heteroge-
neity in cancer,” says Rossanese.
APOBEC3B in particular appears
to be upregulated in at least half of all
known cancers, and ICR researchers are
currently in the early stages of devel-
oping small-molecule inhibitors of the
enzyme, Rossanese tells The Scientist.
“The idea is to test the hypothesis that
reducing mutational load and hetero-
geneity will in fact delay drug resis-
tance,” she explains. Researchers in
New Zealand, meanwhile, are target-
ing the protein using oligonucleotides.
In work published last fall, a team at
Massey University described an oligo-
nucleotide drug that selectively inhib-
ited APOBEC3B in vitro. 8
Like other evolution-based approaches
to therapy, the strategy has limitations.
At diagnosis, most cancer patients have
already accumulated substantial within-
tumor heterogeneity. While better cancer
screening could help tackle that problem
from a public health perspective through
earlier diagnosis, therapies targeting APO-
BEC or as-yet-undiscovered evolution-
driving proteins might have little effect
on late-stage disease, notes Gatenby. “I
think once the horse is out of the barn, it’s
going to be very hard to suppress evolu-
tion,” he says, adding that cancer cells may
well find new escape routes that research-
ers can’t predict. According to the axiom
attributed to evolutionary biologist Leslie
Orgel, “Evolution is cleverer than you are,”
Gatenby says. “I take that to heart.”
Rossanese isn’t dissuaded. She notes
that evolution-stalling therapies would
probably be used in conjunction with
more-traditional approaches. Even for a
patient who already has high tumor het-
erogeneity, “a lot of times, a primary ther-
apy is going to take out 99 percent of the
cancer cells, and that 1 percent that’s left is
going to have to adapt its new condition,”
she says. “We’re trying to hobble those
remaining cells as much as possible.” g
References
1. B. Zhao et al., “Exploiting temporal collateral
sensitivity in tumor clonal evolution,” Cell,
165:234–46, 2016.
2. Acar et al., “Exploiting evolutionary herding
to control drug resistance in cancer,” bioRxiv,
doi:10.1101/566950, 2019.
3. M. Das Thakur et al., “Modelling vemurafenib
resistance in melanoma reveals a strategy to
forestall drug resistance,” Nature, 494:251–55, 2013.
4. J. Zhang et al., “Integrating evolutionary
dynamics into treatment of metastatic
castrate-resistant prostate cancer,” Nat
Commun, 8:1816, 2017.
5. J. Zhang et al., “Integrating evolutionary
dynamics into treatment of metastatic
castrate-resistant prostate cancer (mCRPC):
Updated analysis of the adaptive abiraterone
(abi) study (NCT02415621),” J Clin Oncol,
37:5041, 2019.
6. J.B. West et al, “Multidrug cancer therapy in
metastatic castrate-resistant prostate cancer:
An evolution-based strategy,” Clin Cancer Res,
25:4413–21, 2019.
7. N. Roper et al., “APOBEC mutagenesis
and copy-number alterations are drivers
of proteogenomic tumor evolution and
heterogeneity in metastatic thoracic tumors,”
Cell Rep, 26:2651–66, 2019.
8. F.M. Barzak et al., “Selective inhibition of
APOBEC3 enzymes by single-stranded DNAs
containing 2 ′ -deoxyzebularine,” Org Biomol
Chem, 17:9435–41, 2019.
04 . 2020 | T H E S C IE N T IST 2 9