You are on page 1of 2

OUTLOOK PHYSICAL SCIENTISTS TAKE ON CANCER

ALEXANDER R. A. ANDERSON, WWW.MOFFITT.ORG/IMO


The inside of a tumour can be visualized on a computer by using mathematical models.

MAT H EMATICAL MOD ELLIN G

Forecasting cancer
Complex mathematical models are helping researchers understand cancer’s evolution
and providing clues on how to thwart drug resistance.

B Y K AT H A R I N E G A M M O N it is so difficult to treat — if it retains a diverse and computer modeller at the Moffitt centre
enough population, a few impervious tumour and one of the team who worked on the glioma

E
instein once called pure mathematics “the cells can survive chemotherapy or radiation and study, is using the model to investigate how
poetry of logical ideas”. It is a poetry that re-establish their hold on the body. So mathema- cancer moves and spreads1. His lab has built
has allowed astronomers to understand ticians are creating models to predict in which computer models that focus on the changes to
the movement of planets, and it has shaped our direction a mutating cancer cell will change, in individual cancer cells. The models integrate
understanding of the physical world around us. the hope of stopping evolution in its tracks. information from different in vitro experiments
Now it is set to reshape our view of cancer. From with factors from in vivo environments to gain a
speeding up clinical trials to predicting cellular ELIMINATING GUESSWORK better understanding of how cancer progresses
mutation and evolution, mathematical models Through modelling, researchers can now pre- one cell at a time. Anderson recently used clini-
are helping to transform fundamental ideas dict the behaviour of some of the toughest can- cal data from 650 prostate-cancer patients.
underlying the disease. cers. Historically, oncologists have evaluated the Each patient’s tumour was sliced multiple times
Mathematicians are making inroads in prognosis of glioma, an aggressive form of brain and stained for 250 markers. He then integrated
oncology with approaches that used to be the cancer, on the basis of imaging and histologi- that biopsy information into a model, creating
realm of biologists: building intricate models cal data. But that grading system didn’t match a digital version of the biopsy. When he and his
of cancer ecology and evolution. It’s a logical up with reality, leading researchers at the Uni- colleagues move the model forward in time and
transition. The forces of environmental selec- versity of Washington in Seattle and the Moffitt space, they can create a predictive measure of
tion and adaptation that act on organisms Cancer Center in Tampa, Florida, to develop whether a tumour will be aggressive.
also govern the way cancer cells develop, and a a mathematical model based on cell prolifera- This detailed ability to predict invasiveness
better understanding of how the disease evolves tion, invasion rates and changes in appearance. is particularly important in diseases such as
and adapts to environmental changes can The model marries data from imaging studies prostate cancer, in which some tumours are
help researchers find a way to stop it. “Evolu- with other models that incorporate blood-vessel removed unnecessarily and treated aggres-
tionary theory is the theory for cancer,” says growth patterns and tumour microenvironment, sively. To assess the model’s predictive accuracy,
Carlo Maley, a computational and evolution- and was able to reproduce the growth patterns of Anderson’s team, working with a biologist, is
ary biologist at the University of California, San each patient’s specific tumour, and also reliably testing it in mice. If the model works, it would
Francisco. “If we can engage with this evolution- predict how it would progress — validating the be the first to predict a tumour’s aggression
ary process, we can slow it down or shunt it in a method and improving on the existing grading. before the start of treatment, providing oncolo-
direction that’s more manageable.” Such models are starting to have an impact in gists with a guide for treatment options.
Cancer’s heterogeneity is the main reason the clinic. Alexander Anderson, a mathematical Beyond evolutionary dynamics, mathematical

S 6 6 | NAT U R E | VO L 4 9 1 | 2 2 NOV E M B E R 2 0 1 2
PHYSICAL SCIENTISTS TAKE ON CANCER OUTLOOK

models are helping to reshape the way some the cancer more vulnerable. The model lets resistance for different dosing strategies. “Even-
researchers approach cancer treatments. The researchers discard options that would increase tually we have to validate with mouse and
improved understanding of cancer evolution resistance, narrowing the range of options for human trials, but the pre-validation studies
provided by the models provide has led some treatments, doses and dosing frequency. can be on a computer,” Michor says. The com-
researchers to a counter-intuitive conclusion: The evolutionary game-theory model — in puter model can test just about every possible
for most cancers, targeted therapy just doesn’t which each treatment was a move in the game, combination in a few seconds — much faster
work. When you use drugs that target spe- made in anticipation of cancer’s response — sug- than doing clinical trials. “That’s the advantage
cific mutations, Anderson says, “our models gested the development of that vulnerability and of maths,” she says. “If you run an infinite num-
tell us that you’re destined to see drug resist- helped clarify results from a recent clinical trial2. ber of clinical trials, you will grow a really long
ance because there’s another route to the same beard. But on the computer, it works quickly.”
phenotype”. Cells that survive the first wave of In 2013, the Memorial Sloan-Kettering Can-
ZUMA WIRE SERVICE/ALAMY

therapy can evolve to have the same malignant cer Center in New York will run the first trial
characteristics but without the same vulner- based on Michor’s model. Researchers will test
ability to the drug. What’s more, the evolution- the dosing patterns the model deemed best to
ary dynamics of tumours suggest that trying to prevent or delay resistance in non-small-cell
eliminate cancers speeds up the development lung cancer. With this disease, resistance typi-
of drug resistance, reducing a patient’s chances cally appears a year after treatment ends, but
of survival. Models show that there might be a Michor’s model predicts that the best dosing
better way. strategy will delay resistance by another year3.
And there’s more. Mathematical models can
CHANGING THE GAME also help researchers understand, and even
Some researchers are looking for ways to hack prevent, tumour development. Larry Norton,
cancer’s programming and turn it against itself. an oncologist at the Memorial Sloan-Kettering
Robert Gatenby, an oncologist at the Moffitt centre, has devised a model of tumour growth
Cancer Center, has been seeking fundamental Prompting lung cancer to evolve one way might based on nineteenth-century mathematics. The
laws he can exploit. One way to attack tumours make traditional treatments more effective. S-curve model showed that while tiny tumours
is to make them similar to each other; the more barely grow at all, medium-sized tumours grow
homogenous they are, the lower the odds that Patients with recurring, aggressive lung cancer at a rapid pace, and large ones grow slowly.
they will become resistant to treatment, and who were treated with both traditional chemo- Norton and his colleagues hypothesized that
the greater the chances that therapy will be therapy and a vaccine targeting the tumour-sup- the rate at which a tumour shrinks would also
effective. In essence, he is looking for tricks to pressing p53 protein fared better than patients be proportional to its size, with large tumours
prevent the evolution of drug resistance. who didn’t receive the vaccine. The combination shrinking slowly and small ones rapidly.
Mathematical models allow Gatenby and of treatments increased survival time by about The conventional view of chemotherapy
others to determine the right combinations of 4 months and doubled the number of people dosing held that, because tumour growth was
drugs, and the optimal sequence in which to surviving more than a year. “The tumour suc- exponential, each treatment killed the same
give them, to best combat resistance. In one cessfully adapted to an initial therapy, but that amount of tumour. Norton, however, used
mouse study of triple-negative breast cancer adaptation rendered it more susceptible to his theory to argue that giving the same total
— a hard-to-treat subtype characterized by a another therapy,” Gatenby said. This is known as amount of chemotherapy over a shorter period
lack of receptors for oestrogen and other hor- an evolutionary double bind. “We don’t pretend of time should improve the cure rate. Because
mones — researchers used models to deter- that this could last forever, but if we could get an his model showed that young tumours have the
mine when they should use different drugs, extra five to ten years out of evolutionary tricks, fastest growth rate, he said, the less time they
and in what sequence. The results suggested it would be a huge improvement.” had to regrow between treatments, the better
that dosing the animals with extra oestrogen to the cure rate would be. Norton’s idea was vindi-
would encourage the disease to adapt to a high- DEVISING DOSING STRATEGIES cated in 2002, when a large breast-cancer trial
oestrogen environment. By guiding evolution’s Mathematical modelling is also helping to speed showed that giving chemotherapy every two
natural resistance mechanism, the cancers that up the clinical trials that test those drugs. Test- weeks instead of every three lowered the risk
emerged were oestrogen dependent — and ing all the possible permutations of drugs and of disease recurrence by 26% over three years,
were thus treatable with oestrogen-dependent doses to determine the best course of therapy even though the cumulative dose was the same.
cancer drugs, such as tamoxifen. might take millions of clinical trials — and there Norton makes the point that none of this
Treating cancer would be so much easier if are nowhere near enough patients or resources would be possible without integrating biolog-
oncologists could predict how it would react for that. So instead researchers are using math- ical growth models into treatment research.
to treatment. At the moment, cancer treat- ematical models that make it possible to derive “The biological sciences have become anti-
ment is an endless battle against an evolving the same information from only a handful of maths, and it’s just not working. Researchers
foe, Gatenby says. “Cancer comes up, and we experiments. Such models can describe the rate see lots and lots of facts, but they’re not con-
whack it and do another treatment.” But he sees of mutation, calculate the likelihood of resist- necting them together,” he says. “It’s like look-
a better way. “We need to plan strategically how ance, and even predict the chance that a certain ing up at a starry sky and wondering how it
we treat patients, so that as we give one therapy, tumour will respond to treatment. worked. Only when Kepler and Newton came
we’re producing an adaptive response to be Franziska Michor, a biostatistician at Har- along did things fall into place.” ■
anticipated with our second therapy. Instead vard School of Public Health in Boston, Mas-
of whac-a-mole, we need to be playing chess.” sachusetts, has built a model to predict the Katharine Gammon is a freelance science
Gatenby’s team uses a game-theory model optimal dosing scheme for lung-cancer ther- writer based in Santa Monica, California.
to show that cancers are good at adapting to apy. By accounting for the growth and death
their current situation but cannot anticipate rates of different types of cell, their mutation 1. Swanson, K. R. et al. Cancer Res. 71, 7366–7375
(2011).
the future. This insight gives oncologists a rates, and how quickly the body metabolizes a 2. Basanta, D., Gatenby, R. A. & Anderson, A. R. Mol.
fundamental advantage: they can manipulate drug, Michor’s mathematical framework calcu- Pharm. 9, 914–921 (2012).
the tumour environment in ways that leave lates the probability of the tumour developing 3. Chmielecki, J. et al. Sci. Trans. Med. 3, 90ra59 (2011).

2 2 NOV E M B E R 2 0 1 2 | VO L 4 9 1 | NAT U R E | S 6 7