Conquering Cancer: Personalized Cancer Care

Search transcript...

YAFFE: The birthplace of engineering and science education in North America. I'm Michael Yaffe, Professor of Biology and Biological Engineering at the Koch Institute for Integrative Cancer Research, here at MIT. And it's my pleasure to chair this final session on personalized medicine. I thought I would take a few minutes to just introduce the topic.

When I graduated from medical school, our commencement speaker was a very erudite man of letters. A great humanist with an obvious disdain for science. And in his commencement address he implored us not to clot our minds by thinking too deeply about science when we were taking care of patients. After all, for most diseases at that time, understanding scientific details did not translate into any type of therapeutic advantage. Instead, he implored us to focus exclusively on the humanistic aspects of the doctor-patient relationship. After all, he said, we didn't become doctors because we like to fix things. And people are interesting things to fix.

I was trained as an engineer and a physicist and I was rather taken aback by what he had to say. And so now, almost a quarter of a century later, I can take great pride in saying to you, that's exactly why we should become doctors and scientists. Because we like to fix things . And there's nothing more rewarding or satisfying than fixing another human being. And what better place to discover how to fix people, than here at MIT?

These days we're beginning-- just beginning-- to learn how to take knowledge from the laboratory and translate that directly into the clinic. Particularly in the area of cancer research. In a few moments, you'll hear three vignettes from some very distinguished scientists and clinicians about personalized medicine. Personalized because it means no longer treating all lung tumors, or all breast cancers, or all leukemias, or are all lymphomas the same way-- based on what organ the tumor arose in. Or some gross histological or cytogenetic features of the tumor.

Instead, it means being able to understand what makes each person's cancer unique and designing a custom strategy. To try and turn that knowledge into cure. Now, for many of us, personalized medicine means sequencing the genome of the tumor. But I would call that Personalized Medicine 101.

And here are some common mutations listed on this slide that are found in the tumor that Eric told us about. Human glioblastoma from the tumor cell Genome Atlas. And the mutant genes, that are listed here are genes that are parts of pathways that scientists have been studying for the last three decades. Quite a bit of that work having been done here at MIT.

But just knowing the mutation or which methods your RNAs are expressed doesn't tell us anything about what drug to give a particular patient. Instead, I would argue did this information is very important, but incomplete. And in order to understand how the tumor cell works, we have to understand how it's put together.

We have to go beyond Personalized Medicine 101, to Personalized Medicine 202 and 303. We have to learn how to read the integrated circuit of the cell. It keeps the cancer cell alive and growing. And in this wonderful slide it I've stolen from my colleague, Bob Weinberg, you can see a glimpse of the wiring. A shot under the hood-- so to speak-- of how the tumor cell works.

The great challenge and personalized medicine, is to understand how these systems are wired together. All of the feedback loops. Functionally, who is talking to who within this complex network. And break these down into pathways and processes that we know are required for tumor cells to survive.

And then figure out exactly which connections there are that are not working in a particular person's tumors. And then from that information, that should allow us to design the right drug, the right treatment, at the right time, and the right dose, to maximize chances of killing the tumor without killing the patient. I think in the next three vignettes you'll see direct examples of this.

I don't know exactly what the next three speakers will be telling us about, but I suspect that Professor Hemann will tell us how even a partial understanding of the status of these networks can be used to predict tumor responses in drug resistance. And I suspect a Professor Livingston may tell us how defects in one pathway or another can be leveraged with specific drugs to improve tumor responses. And Professor Haber is likely to tell us it maybe we don't even need to look in the tumor at all. Maybe we should look at the cells that the tumor has shed.

And with that, it's my great pleasure to introduce our first speaker, Professor Michael Hemann from MIT. Mike is the Latham Family Career Development Assistant Professor. He did his postdoctoral work at Cold Spring Harbor. And he's one of our most impressive junior faculty members. He's one of our homegrown up-and-coming stars. And with that, I will give you Professor Hemann. I would just remind you if you have questions, please write them on the cards and pass them to the young lady who should be walking around during the session.

HEMANN: Thank you, Mike, for the introduction.

I do have the distinct pleasure of being a new faculty member in our fantastic new building. And I should probably get an actual picture of it instead of an artist's rendition. But my lab is interested in cancer therapy. In particular how tumors acquire resistance the frontline chemotherapies. So I'm going to start with really, a classic problem that faces clinical oncologists.

So these are two lymphomas. And they are pathologically indistinguishable. And because they're pathologically indistinguishable, they're treated with the same frontline chemotherapy. Unfortunately, these tumors may have decidedly different responses to the same therapy. So Tumor A may respond very well, while Tumor B may show no response whatsoever.

So these patients are subjected to all the side effects of frontline chemotherapy, yet incur none of the benefits. And this is true of the vast majority of malignancies. That a subset of patients, statistically, will fail to respond to frontline chemotherapy.

So what's the difference between Tumor A and Tumor B. Well, I think from all of the discussion today, that's the difference is genetic or epigenetic. So there's a change or a set of changes that distinguishes one tumor from another tumor. So all we have to do is identify what the relevant alteration is, and we can start stratifying patients based on likelihood of therapeutic outcome.

Unfortunately, as Eric Lander pointed out better than I think I never could, tumor cells are enormously complex. So this is the karyotype a normal cell. It has 46 well-behaved chromosome. This is the karyotype of a particularly chaotic cancer cell. And what you can see is that there are a vast number of abnormalities here. There are chromosome translocations, there are deletions, there are huge gains in chromosome number, they're also losses in chromosome number.

And if we look at even more fine detail, just the DNA copy number on two chromosomes, where everything above the line represents a gain in copy number and everything below represents a loss in copy number, there's still a massive number of alterations just on these two chromosomes. So just by looking at this, it's very difficult to tell which of these alterations were relevant for tumor development, which are going to impact therapeutic outcome, and if we want to design new drugs, which are going to represent meaningful drug targets.

So it's a lot like the situation that's described here on this slide, these two pictures of the same bridge. Now, I'm not an engineer, but I do know that this bridge has a lot of problems at this point. And it's very difficult just by looking at this bridge, at this point, to recognize which were the problems that actually caused it to fall to begin with. And which were simply a consequence of it falling down. And so like any good engineer, our approach is to try to model this process. So we can harvest tumors, and we can treat them in cell culture, we can transplant them into recipient mice, and we can do chemotherapy studies. Importantly these studies are done outside of the patients, so we aren't experimenting directly on human patients.

Additionally, we can modify these tumors by retroviral infection. So we can introduce new DNA, like oncogenes, as Jackie Lee talked about earlier. So we can look at the defined effect of these alterations on therapeutic response.

And additionally, we can suppress genes using RNA interference. Now as most of you know, RNAI was the technology that was developed by Andy Fire, who was a postdoc at MIT along with Craig Mello that allows one to essentially rapidly inactivate any gene in the genome. So we can recapitulate a lot of the alterations that are present in cancer. So essentially, we have a tool kit with which we can start reconstructing cancer genomes.

So for example, using retroviruses, we can overexpress genes that recapitulates this gain in copy numbers seen in many cancers. And using RNAI, we can model this loss in copy number that is seen on this chromosome end.

So I thought I'd give an example of this kind of engineering that we can perform to look at modulators of therapeutic response. And this is a project that was done in conjunction with Mike Yaffe, looking at two proteins, ATM and P53. Now we chose to look at these proteins for a couple reasons. The first of which is that these proteins are highly mutated in human cancers. And so, P53 is mutated in over 40% of all human cancers. Whereas ATM is mutated in over 10% of human cancer. So these represent some of the most highly mutated genes in human tumors.

Additionally, these genes are known to play very important roles in the response to DNA damage. And most conventional chemotherapeutic exert their effects by inducing DNA damage. Now in response to DNA damage, these proteins are known to do induce essentially two terminal phenotypes. Either cell death in target cells, or cell cycle arrest. And both of these processes turn out to be critical in terms of overall outcome in response to frontline chemotherapy.

So cell cycle or rest can be thought of as sort of a survival mechanism. So, cells that undergo cell cycle or arrest can repair their DNA and eventually, can start growing again. Whereas cells the die, obviously that's a terminal phenotype. And that's really what we want to achieve in treating tumors with frontline chemotherapy.

And so we're interested in what effect suppressing ATM and P53 had on the balance of cell death versus cell cycle arrest in tumors. And so again, to do this we tried to model this process. We started with RNAI vectors targeting either ATM, or P53, or the combination of both of these, and used them to infect target cells. And then transplanted them into recipient mice and looked at the effect on overall therapeutic outcome. And what we saw as a therapeutic outcome actually varied quite considerably depending on just the status of these two genes.

So if P53 an ATM were intact, the tumors had a very good response, there's cell death and we had tumor regression. Whereas if tumors were silenced for either ATM or P53, instead of dying, these tumors underwent cell cycle arrest. They repaired their DNA and they eventually relapsed.

Finally, and I think interestingly, if we suppressed both P53 and ATM, we induced an interesting form of cell death called Mitotic Catastrophe. But again, it is a form of cell death. So these tumors regressed and there was an overall good response.

So these data, or this modeling, led to a very defined set of predictions related to human cancers. And those predictions were that if the patient had either an ATM or a P53 mutation, it should correlate with a poor prognosis. However, if they had both ATM and P53 mutations, it should correlate with good prognosis. Similarly, if they were wild type for both of these genes, it should correlate with good prognosis.

And we were able to examine this in a cohort of breast cancer patients. And what we saw is-- similar to our prediction-- breast cancers that were ATM deficient, but had wild type P53, showed a poor prognosis. Whereas those that were wild type for both of these proteins showed good prognosis, overall. Similarly if tumors were mutant for P53, they showed poor prognosis. And in rare cases where there is the co-occurrence of ATM and P53 alterations, there was always very good prognosis.

just looking at two genes or two proteins, we could start to make reasonably decisive predictions about overall therapeutic outcome. We could also use this information to start devising strategies to overcome drug resistance in these tumor. So for example, if we know that the combined loss of ATM and P53 correlates with good prognosis and mitotic catastrophe, then one strategy to sensitize this tumor to chemotherapy, is simply to silence ATM to make a tumor that is drug-resistant now drug-sensitive. And in fact, there are a number of ATM inhibitors that are currently in clinical development.

So for this 40% of patients that have P53 alterations, targeting ATM is actually a very good therapy. It turns drug-resistant tumors into drug-sensitive tumors, or at least supports that process. However, this kind of agent would have to be used with an extreme amount of caution. Because as I told you earlier, tumors that have normal levels of P53 and ATM show good overall prognosis and cell death in response to therapy. So, if we actually inhibited ATM in these tumors we would actually promote cell cycle arrest and we would actually inhibit the action of the drug. It would result in poor prognosis overall.

So if you just treated patients with an ATM inhibitor, without regard to P53 status, this would look like a terrible drug. It actually has adverse effects in over 50% of all patients that get it. And this is a classic problem in drug development. That if you have a drug that shows adverse effects in 50% of your target patients, it looks like a terrible drug and it won't pass clinical trials. But if you can actually start targeting based on P53 status, you turn a terrible drug into a very good drug. And I think this is precisely what we mean when we say personalized medicine. It's providing a therapy to a subset of patients that we know we're going to respond very well to that therapy, and avoid those patients where we know we're going to have an adverse effect.

So of course, these are just two genes. But the combination of these two start to provide a molecular signature for these tumors, where previously we had pathological data that was completely meaningless. Now of course, we'd like to start expanding this approach to start looking at the whole genome scale analysis of alterations that are present in cancer. Many of the alterations, or the 240 alterations described by Eric Lander, we'd like to have systems for modeling these on a very large scale. And in fact, our lab and others have developed strategies where we can actually introduce thousands of RNAI vectors into tumors and look at their effect on therapeutic response.

So instead of looking at a single gene, we can look at large collections of genes, and large collections of therapeutics, and have functional response profiles. So that ultimately looking at these two tumors, we actually not only have a gene signature or a mutation signature, but we know how those mutations correlate with therapeutic response. So we can use these combined signatures to choose the appropriate therapies in these cancers.

And so just to finish, we now really have a tool set where we can start taking all this genomic data, and sequencing data, and use it to model therapeutic response prior to administration of these drugs to patients. And so importantly, this information can be used to predict patient outcome. But I think more importantly, it can also used to optimize strategies for treating drug resistance and overcoming drug resistance, even using our current our current arsenal a frontline chemotherapeutics. So I'll end there and thank you for your attention.

YAFFE: Thank you, Mike. It gives me great pleasure to introduce our second speaker, Professor David Livingston. So it's an honor to introduce David. In the last 15 years, he's been both a very special friend to me and in many ways a role model. David is the Emil Frei Professor of Genetics and Medicine at Harvard Medical School, and the Deputy Director of the Dana Farber/Harvard Cancer Center. He received his undergraduate degree from Harvard in 1961 and his medical degree from Tufts Medical School in 1965. He undertook his clinical training in internal medicine at the Peter Bent Brigham Hospital, and did scientific training at both the National Cancer Institute and Harvard Medical School.

He joined the faculty of Harvard Medical School was an Assistant Professor of Medicine in 1973 and he's been a faculty member at Harvard Medical School in the Dana Farber Cancer Institute continuously since that time. Doctor Livingston served as the Director and Physician-in-Chief of the Dana Farber Cancer Institute from 1991 to 1995. He was the Chair for the Board of Scientific Advisers at the National Cancer Institute from 1995 to 1999.

And David's lab pioneered many of the early advances that led to our current understanding of oncogenes and tumor suppressor genes that you heard about this morning. He was elected to the Institute of Medicine in 1990 and to the National Academy of Sciences in 1995. Ladies and gentleman, David Livingston.

LIVINGSTON: I don't think we're here to talk today about infectious illness. But after listening to Michael's beautiful presentation, I don't know this Livingston and I have a fever, both. And I know Phil Sharp doesn't because first time I met him, he met me when I stole his change from the bureau at a dormitory, wasn't it-- at the Weizmann Institute-- so I have a felonious background, as well.

But I want to talk to you about a problem that transcends felony. And that's about breast and ovarian cancer. And some of the themes you'll hear about are themes that Mike Hemann touched on more than in-passing in the past 20 minutes or so. And in particular, I want to talk to you about an interesting vulnerability that's been uncovered in the work of a number of laboratories in recent years. Two in particular. To the development of a very serious lesion. And that's the presence of the DNA double strand break.

And I do want to tell you that I have the privilege of working every day at Dana Farber next to a young colleague, Dan Silver, who got his PhD here, with David Baltimore, a number of years ago. And who, not only thinks about the kind of problem I'm going to talk to you about very deeply, but he actually sees patients with breast cancer. And thinks in scientific terms, just like the antithesis of the doctor that Mike told us his medical school valedictorian speaker told us not to be. Dan Silver is the opposite of that. And it's his wisdom that's helped me think about this talk today.

So, let's consider a normal cell-- in fact-- that experiences DNA damage. And is accompanied by a very deep system of DNA repair that successfully removes most of the damage. And what damage is not dealt with perfectly, is dealt with imperfectly, but in a way in which life goes on, until there's a failure of DNA repair, for example. A failure DNA repair at the hands of the loss of function of one, of two, inherited breast cancer, high penetrance breast cancer susceptibility genes BRCA1 and 2. Both of which play a major role in dealing with a variety forms of DNA damage, but most especially DNA damage in the form of a double strand DNA break.

Now interestingly, in this setting, where there is a chronic problem in dealing with DNA damage, mutations in the global sense, including aneuploidy, can emerge. And a number of interesting outcomes, each of which facilitates the development of the neoplasm arise. For example, the activation of certain growth promoting genes, or oncogenes. The presence of loss of heterozygosity, and therefore the activation of certain tumor suppressor genes. And the alterations in genes that regulate cell death. All of these can emerge. They're selected for over time. As we heard about before, selection for life is the powerful force afoot here. And not long thereafter, years, but not forever, the tumors may emerge.

And in that setting, I want to introduce to those of you who don't think about this regularly, and apologize to those of you do, a therapeutic concept. First, a genetic concept applied in prokaryotes. Later, in simple eukaryotes. And I'm going to show you a clinical example of its application here.

So the term is synthetic lethality. And by synthetic lethality I refer to what happens to an organism when two genes, when individually eliminated, give rise to no negative, no toxic effect on that organism. But when eliminated together results in cell death. So we have Genes X and Y. Eliminate either of them, that is minus, means loss of function. There's no major biological effect. Eliminate them together, cell death emergencies. One of the genes is let's say is BRCA1. And the other is a putative inhibitor. And I'm going to tell you about a real inhibitor, of a real gene, in a couple of minutes.

In both of these functions, the target of Y and BRCA1 are intact. A cancer cell for example, would survive. Certain cancer cells would survive. BRCA1 function is gone. The target of Y still functions, no biological effect. The opposite, no biological effect. In the absence of BRCA1 function, the target of Y loses function, and now the cell dies.

So, by death I want to refer to a very special setting-- that in which a double strand break develops in the genome. And a very small number of double strand breaks, in cells for example that have lost BRCA1 or 2 function. Which are actually required to repair that break by a complex process in which the broken genome uses the healthy genome, the unbroken gene, as a template for the synthesis of new DNA sequence across the break. Basically, the sister chromatid, as we call it, serves as the template for the perfect repair of this double strand break. The end product of which is typically, an the intact DNA strand.

Now BRCA1 and 2 are not only proteins that are engaged in double strand break repair. They're professional tumor suppressor genes. And by that I mean, the cancers that arise in women, and in the case of BRCA2, sometimes the men, that carry germline loss of function mutations in these genes, have actually lost both functioning copies of the gene here. Unlike their normal cells, and certainly in their germline, where there's one good copy and one mutant copy.

And so, the loss of function of either of these predisposes to the development at very high risk, and very high penetrance, in both of these cases. Approaching 85% in the case of BRCA1, and 60% in the case of BRCA2. The development of breast cancer and ovarian cancer in the case of BRCA1. Breast and ovarian and a host of other tumors in the case of BRCA2.

In fact, this is not an uncommon development, even on the streets of Boston, where some 2 to 6% of all the breast and almost 10% of all the ovarian cancer that is seen is acquired on an inherited basis. And a lot of the genetic misbehavior is a result of misbehavior at these two, low [? side. ?] And in fact, as I told you, both of these characters are required for the repair of a double strand break. Now, many double strand breaks-- not all-- arise at the hands of an antecedent single stranded break. And you'll why I tell you about this in a minute.

And a single stranded DNA break is shown up here, on the left. All hell breaks loose. All form of molecular hell breaks loose in the form of a process called base excision repair. Twenty plus proteins, each of which knows its place in society. It is called upon at the right time to participate in the repair of this. An early step of which is, the acquisition of the services of a pair of enzymes called poly ADP-ribose polymerases, of which there are distinguished experts in this institution. These are enzymes that are capable of converting NAD into a polymer, which is coupled covalently to one or more sites in the primary structure of the protein of interest. And this very complex process is one of the early steps that is required to repair a single strand break. What happens thereafter is a series of reactions, the end product of which is this single strand break now becomes repaired with the right nucleotide in the appropriate complementarity and life goes on.

Now, if it turns out that this single strand break were to arise, because this replication fork here stalls, what's left behind is a single stranded gap. Which is like a break. And it turns out the biologically, this single stranded gap is much more susceptible to undergoing a second break, here in this strand, resulting in a double strand break. Which for a cell, is like a medical emergency. So in the absence of base excision repair, this single strand break is more likely to become a double strand break. And that's where BRCA1 and 2 function come in to the game. And in fact, to show you that there is a consequence of the interposition chemical inhibitor of one of the PARP enzymes, especially PARP 1.

What you can see here is that in the chronic presence of a PARP inhibitor in these tissue culture cells, you can see these red dots. These red dots are an accumulation a protein, Rad51, that's absolutely required for the homologous recombination-based repair of a double strand break. As if a single strand break that existed spontaneously in the absence of a PARP inhibitor, in the presence of a PARP inhibitor, becomes a double strand break. This cell-- that's with those red dots mean-- has the capacity to repair that now double-strand break by homologous recombination.

So to summarize what I think is the key thought here, a single-strand break in the absence of basic scission repair, and most specifically in the absence in the case we're going to talk about of PARP1 function. And in the absence of homologous recombination, actually becomes a double strand break that's not repairable by standard homologous recombination means. And if there are enough of these double-strand breaks that the backup systems that do repair can't handle the job, a cell can die.

And in fact, two groups, one headed by Alan Ashworth, and the other by Thomas Helleday in Britain, hypothesize based on real evidence, that this kind of synthetic, lethal effect had clinical implications that might work in women with BRCA1 and 2 breast and ovarian cancer. And in fact, some other evidence is shown here. This is taken from one of their papers. You can see that if you have an inhibitor, a rather specific high-potency inhibitor PARP1, less of it is required.

In each case where there's a BRCA1 or 2 mutation, to kill these tissue culture cells that are genetically defined, compared to wild type sales over here. And this is true in animal models as well.

And so now we know in this clinical experiment that was carried out by the next generation of students in the line of Alan Ashworth and Steve Jackson-- who were involved in one of the papers-- in this report that appear in the Lancet this past summer, in these rather traditional kinds of clinical so-called waterfall plots, where we analyze going to the right clinical response at two doses. 100 milligrams given twice a day. 400 milligrams given twice a day. So, a greater dose of drug over here. This is a highly specific PARP1 high-potency small molecule inhibitor.

Please note that the deeper the bar, the greater the clinical response is measured, radiographically. So at lower and higher dose, there's lesser, but still significant. And greater, and of course significant, clinical effects in a significant fraction of the patients. As if the drug is having a tumor recital effect, depending on dose, just what a pharmacologist would expect of a reasonable drug. And the effect is manifest by loss of tumor bulk.

So, what have I told you? I've told you that inhibitors of PARP1, or promising new drug treatment that target's a defect-- loss of DNA repair, loss of a specific kind of DNA repair-- present in tumors. And not generally present in their normal counterparts. And I should point out that a recent PNAS paper from Phil Sharp and his coworkers has made this point using RNAI in an in vivo setting. A very powerful confirmation of these observations.

As monotherapy employ one of these agents alone, it may be effective primarily for certain tumors in which there is a defect in homologous recombination. That's the message. And this, of course, is a manifestation of synthetic lethality.

So what happens when, as was predicted before, resistance sets in? And it does. And it does after not too long a while in most, if not all of the patients. And here's a model setting, brought to us by Tatayusa Tanaguchi, and his co-workers who is one of two scientists who brought to us the concept I'm going to talk about and finish up with.

So he took this particular models setting in which a pre-malignant cell eventually becomes a tumor cell. And then that tumor cell is confronted with, in this case, not a PARP inhibitor, but with a cisplatin or cisplatin analog, which induces cross-links between the two DNA strands. The repair of which, just like the repair of a double-strand break, requires the repair of a double-strand break that arises when you remove the crosslink.

And so the argument went something like this. You have a tumor, like an ovarian cancer, that's highly sensitive. But there's one cell there in the red that's not from the outset. And after progressive treatment with platinum, the tumor becomes resistant. What Tanaguchi and independently, Alan Ashworth showed, was the striking observation that if these were tumor cells that arose from a BRCA2 BRCA1 donor, in which there's one mutant copy, and one wild type copy to start, there is, when resistance develops, a significant reversion. Either directly, or by second site mutagenesis, naturally occurring mutagenesis to a wild-type phenotype, and a genotype that's still abnormal, but the second site revertant now behaves with respect to homologous recombination normally.

So basically, let the genome reorder the gene under selective conditions. The gene may not be perfect, or it can be, but it is now functional. It can now we express homologous recombination and it becomes resistance, either to a platinum analog, or two PARP inhibitor,

And so for example, in this cartoon from Alan Ashworth, the argument is-- That at the very beginning, when a response does occur, there is a dramatic anti-tumor effect of either platinum or PARP inhibitor, there may be a rare cell that has undergone spontaneous reversion of the original lesion at BRCA1 and 2. And after the response is carried forward with the progress of therapy, increasingly resistant cells appear. and a significant number of these, up to 30% for BRCA1 tumors, maybe for BRCA2 as well, are actually tumor cells that have acquired, on a clonal basis, reversion of the initial genetic lesion. Something we thought we really didn't occur clinically in a significant way.

So I want to end up with this model clinical story. It's actually a real clinical story taken from Dr. Tanaguchi's paper. And it probably has occurred repeatedly. A breast cancer patient, who is also a BRCA1 mutant undergoes therapy with surgery and a long period of adjuvant chemotherapy. Chemotherapy, three drugs, six months or more, maybe even a year, in an effort to be sure that if there are micro metastasizes in that patient's body, the chemotherapy has a chance at eliminating them enough so that recurrent disease is suppressed.

And then there were years of disease free history. And in this patient, a new ovarian cancer developed. And when indeed, the genotype of BRCA1 was analyzed in this individual, the tumor, turned out to be clinically, totally platinum-resistant, unlike 80 to 90% of sporadic ovarian cancers, which are very sensitive to platinum analogs. And the only detectable BRCA1 allele was wild type.

As if, maybe, during this long period of exposure to DNA-damaging agents, a clone or more was selected at some point during the evolution of what would eventually be a BRCA1 ovarian cancer. But this now, mutant-containing BRCA1 allele reverted and the tumor that arose, unfortunately, was resistant to its drugs.

So in this case, homologous recombination deficiency, a commonly encountered phenotype that represents a pathway to chemotherapy sensitivity. If that were true, just as Mike is predicted in another setting, it too, might be a predictor of clinical outcome. A way to personalize, maybe on a more global basis. A possible future therapy. And here's a plea for the most urgent need of all, which is some sort of an analysis method. A test, a biomarker that's robust, intractable and that works in human tissues, and that could report these cells, in this tumor, are HR competent, homologous recombination competent, and these are incompetent. Without that kind of information, It's still flying on instruments, and I fear the outcomes are still not rosy. Thank you very much for your time and attention.

YAFFE: Great honor to introduce our last speaker in this session. And that's Professor Daniel Haber, who's the director of the Massachusetts General Hospital Cancer Center, and the Issellbacher/Schwartz Professor of Oncology at Harvard Medical School. I'll pronounced his name Haber, in the German style, even though I know he spent his entire youth in Paris, before he matriculated at MIT, where he earned his Bachelor's and Master's degrees.

He continued his studies at Stanford University in the MSTP program, where he earned his MD PhD degrees. And then subsequently, completed a medical internship and residency at Massachusetts General Hospital, followed by a clinical fellowship in medical oncology at the Dana Farber Cancer Institute in 1986.

Doctor Haber pursued post-doctoral research training with Dr. David Houseman, here at MIT, where he characterized the Wilms' Tumor suppressor gene. He was appointed an Assistant Professor of Medicine at Harvard Medical School in 1991. Promoted to Associate Professor in '96 and Professor in 2001.

Doctor Haber serves the Associate Chief for research at the Hematology Oncology Unit at MGH. He's chair of the Cancer Genetics Program for the Dana Farber Harvard Comprehensive Cancer Center, and he's the director of the MGH Center for Cancer Risk Analysis. He was elected into the American Society for Clinical Investigation in 1995, and the Institute of Medicine 2009. Professor Haber.

HABER: Thanks very much, Mike. Coming to Mass General, I don't know how many people know in the audience, we can wish you a happy 150th anniversary to MIT. We're celebrating our 200th anniversary this year, at exactly the same year. So to quote my primary care physician, we both look pretty good for our age.

That said, as the last speaker, I thought I would focus on bringing some of the lessons home to clinical practice with a focus on circulating tumor cells that are in the bloodstream. But also how we use these to try to optimize biomarkers and apply these to the clinical care patients.

So to that, this is obviously the challenge. Which is that we've talked about the wiring diagrams, we've seen these in some of the previous talks. And there's this whole array of novel therapeutics, which now target different points in this wiring diagram. The big question of course, is what in the world do we do with all this information.

If you look at this, and this is actually relatively old, this is from McKinsey-- this is the number and the types of drugs that are coming down the pipeline in drug companies. And you can see as far back as 2001, most of these was cytotoxic chemotherapy drugs. If you look even as recently as 2006, and it's even greater now, there's a huge array of targeted therapies. Targeted drugs that attack particular proteins. And again, we have absolutely no idea how to apply these in the clinic and how to test them, let alone in combination.

So we do have a couple successes. These are the ones I've selected that we're relatively recent and focus on epithelial cancers. There's a certain type of lung cancer that arises in non-smokers. It's about 10% of all cancers. More common in Asia than in the US. And as you can see, these cancers have specific genetic causes. About half of all of these non-smoking lung cancers are associated with mutations in EGF receptor. About 20, 25% a recently discovered translocation involving the ELK gene. And both of these types of lung cancers now have very, very dramatic effective therapies that target either EGFR or ALK with very dramatic clinical responses.

Similarly in melanoma, we've heard about B-Raf mutations. About 65% of melanomas have B-Raf mutations. And these respond very nicely to be B-Raf inhibitors. Again, you see here a PET scan showing dramatic responses.

So they're great successes that we can point to. On the other hand there's a tremendous amount of serendipity to how some of these things have evolved. This is how we got involved in the lung cancer studies in the first place, through a Boston Globe article, referring to one woman. There had been trials of thousands of patients with lung cancer, treated with EGF receptor inhibitors, most of whom had absolutely no benefit. But, there was this woman living in Boston, Kate, who had a dramatic response and did very, very well for over 10 years now. The question was why.

We ended up sequencing her tumor, finding EGF receptive mutations, and understanding that a subset of patients with what appeared to be similar cancers, have genetic abnormalities. You can't do this over and over again. And what I'm going to try and talk about is really the art of trying to convert from serendipitous observations, to trying to put some order in terms of finding these new types of biomarkers and applying them.

And I want to touch quickly on two initial platforms that we've been developing at MGH. And then focus on primarily on the third.

So the first is, how do we match the right drug with the right patient. And how do we do that pre-clinically so we don't end up treating thousands of patients with drugs that don't work. The second is, how do we apply this pre-clinical information that's really wonderful in the lab, but how do we apply that to the clinic. And what are some of the challenges. And finally, I want to talk about circulating tumor cells and ways obviously a really monitoring patients during therapy.

So the first is really a platform that was started by my colleague Jeff Settlemen, who's now at Genentech and now by Cyril Benes. And the idea was that if you could model the entire genetic heterogeneity of human cancers in the lab, could you actually find these drug susceptibility pathways.

So we've collected 1,000 cell lines-- and this is now a platform that's shared by MGH in the Sanger Center-- 1,000 different cancer cell lines representing as many different cancers as we can. All of which are annotated. And all of which are robotically treated with different drugs. Every drug that comes from pharma, biotech, the sigma catalog, your best friend's lab, anything you could possibly find. And then the idea is to try to find the sensitivity patterns. The scary thing is that, if you do this with EGF receptor mutations, you find sensitivity patterns and lung cancer within about three days.

If you do that with the B-Raf inhibitors that don't work in melanoma, you see no response. If you do it with drugs, B-Raf inhibitors that do work in melanoma, you see dramatic responses.

So the things that took us such a long time to figure out are now totally obvious, using these kinds of models. The big step forward now, is to apply those to the unknown. To drugs whose targets and not understood, to combinations of drugs, to try to find things pre-clinically on a scale that we haven't done before. But the big lesson here is that if you look over the last 50 years, everything has been modeled in one, two, three, maybe dozens of cancer cell lines. Individually, a cancer cell line means nothing. In aggregate, in the thousands, you then start to see patterns that can recapitulate what you see clinically.

The next challenge is, how do you bring this to the clinic. And the usual way in which we do clinical trials is that we treat all patients with a cancer with a particular drug. And then we realize that some of them responded, we try to get their tumor, we usually get about 10% of the tumors, and then we try to analyze these. And that's what has worked so far. But you can't do that for a living. Not only does it not work, not only is it slow and expensive, but if you don't enrich for a rare genetic variants, you will never have them in your clinical trial to analyze anyway.

And in fact, most of the lung cancer trials had two or three patients with EGF receptor mutations. And most of these trials were failures because they had not pre-screened for these patients.

So we actually went ahead and set up a collaboration with our pathology department to genotype every tumor that's biopsied or resected for the known, actionable mutations. Those for which there are drugs today, those for which there are likely to be drugs tomorrow, or the next day. This is basically a platform that just illustrates how you can find a melanoma mutation.

But the key here was not to do this in the lab. Not to do this with fancy technologies, but to do that within the pathology lab itself. So at this point, extra section is cut on every two tumor that's biopsied, and the analysis is done on the standard pathological specimen, in order to make this as easy as possible. And to try to get a turnaround time of about seven working days from the time a patient biopsy is obtained to the time you have this information.

This is the kind of information that you get. So we've now looked at over a thousand lung cancer samples. And you can see very clear patterns. Many of them have K-ras mutations. We've heard some about potential drugs that may target this mutation. At this point, it's still an area of investigation. You see a whole other pattern of lung cancer is quite distinct from these that had EGF receptor septic mutations, and others that have these ALK k translocations.

So what it ends up doing, is that as a patient comes in with a new diagnosis of lung cancer, you can very quickly identify those that have K-ras mutations, EGFR mutations, ALK translocations, or unknown mutations. These will get chemotherapy. And these will be treated upfront with targeted therapies.

And this is the proof of the pudding, this is a study that was recently published led by Eunice Kwak, which showed that we ended up genotyping 1,000 lung cancers to get about 100 that had this specific genotype, the AML for all translocation. There's a drug available, this one is from Pfizer, which targets ALK. And again, you see the waterfall plot that David had described. Over 90% of patients had benefitted from this drug, if they were pre-typed for EML4-ALK translocations.

Had this drug been tested in all patients with lung cancer, it would have failed, because less than 5% of patients with lung cancer have this kind of mutation. At this point, this is a new way of doing clinical trials. The drug trial is going up for FDA approval now, based on an accelerated pathway. And again, validating this whole idea of initial genotyping of tumors.

Now, how do you genotype these tumors. The challenge that we face in the clinical world, is that we've become so good at diagnosing cancer with tiny little needles, that in fact, we rarely have enough initial material in tumors to get a biopsy and to be able to do these kinds of genetic tests.

And that takes me to my last topic, which is the idea of using circulating tumors in the blood to analyze molecular characteristics of cancer. Now, this is very beautifully shown here, the tumor is green, the cells are green, you can see them in the blood. But in fact, we don't really know how the cells go from the primary tumor into the blood. We don't know how many of them that actually are in the blood. The best analysis is about one tumor cell in a billion blood cells.

We don't know how many different types of cells. How many of these a dead. How many of these on the test metastatic precursors. What do they look like. And we don't know how they leave the bloodstream to cause metastasis. Other than that, we're doing really well.

This is a collaboration with my MIT-trained colleague, Mehmet Toner, who runs a biomems and engineering lab at MGH. And he has developed a number of microfluidic device is to try to capture incredibly rare cells. This is our latest CTC chip, it's generation two. And basically, it's a microfluidic platform. You can see about two to four mLs of blood going through this device.

And to avoid the laminar flow of fluid through a microfluidic device, there are irregularities that you can see here, what we call herringbones so chevrons, the cause turbulent flow of the blood through this device. The walls of the device are coated with antibodies to epithelial markers, so we can capture circulating tumor cells as they flow through this device. This is what they look like. You can see a relatively small, normal white blood cell, which stains here in red, the nucleus is in purple. And you can see in this case, a prostate circulating tumor cells stains in green.

The pictures are actually quite nice, we have them all on our wall if you want to come visit. But you can actually see these. You can stain them for different markers of tumor. This is prostate-specific membrane antigen. This is a blood marker to distinguish a contaminating white cell. And we can start seeing these in clusters, as well.

In fact the appearance of clusters is a little scary. We've all learned it cells squeeze through capillaries in now lungs. How in the world do we get these clusters. But 5% of patients so we've analyzed have these relatively large clusters of tumor cells. You can see them here in green fluorescence and here by light microscopy. Which suggests again, that tumor emboli may make some contribution to the spread of cancer in these advanced stages.

Now, what can you do with these cells. Well, you can certainly count them. In here, what you can see in blue, is the volume of a tumor on an x-ray, lung cancer being treated with chemotherapy. In red, you can see the count of circulating tumor cells dropping precipitously with successful treatment.

You can also start to ask molecular questions. So this is a lung cancer patient with an EGFR mutation. And you can see again, in green, the x-ray diameter shrunk as the patient was given an EGFR receptor inhibitor. It stayed down for about a year, and then the tumor regrew, which is unfortunately very, very common.

The number of circulating tumor cells is shown in red. That dropped as well. Stayed down and came up as well. But the important fact is that at every step of the way, we could harvest the cells, isolate DNA or RNA, and analyze the genotype of the tumor. And in this example what we can see, is that the activating, sensitizing drug mutation, shown in green, was present at diagnosis, and present throughout the course of this patient's treatment.

What we can see is the emergence of drug resistance. And the analogy has been made to HIV. It's exactly correct here. We can see a drug resistance allele present at the time of diagnosis, before the patient was treated. But at that time, it was less than one in 100 alleles. And as we get successful treatment, and we get selection within the tumor, the resistance allele becomes prevalent. We see this over and over again. Rare alleles being selected. And this gives us the opportunity of monitoring cancers as they evolve. Identifying the dominant mechanisms of resistance and hopefully treating that.

This is a separate case which we had an ALK translocation. We can stain for ALK. And again, you can see here, within two weeks of treatment, the number of cells in the blood decreases. And again, there are more and more questions that we can ask using these kinds of cells.

And I want to end with basically a picture of heterogeneity of CTCs. The key is, can we identify what are the markers that are present within CTCs. How do they differ from the original tumor. Are there subsets of genes that could of a particular interests within the circulating tumor cells.

And to do this, we've actually collaborated with a single molecule, next generation sequencing companiy, Helicos, which can really identify by next generation sequencing, a digitcal gene expression profile for normal blood or blood captured on a mock chip, versus patients who have CTCs in the peripheral blood. And the question then, is to identify all of these red genes, and see what's expressed in CTCs, which are not present in the primary tumor, and how do individual CTCs differ from each other.

So in ending, I want to leave you with a kind of a vision for where we would like to be in cancer treatment in the near future and the distant future. And the idea in trying to bring all this together is that, there are a number of steps that we have to bring forward in the treatment of patients with cancer. Clearly, the first thing is, and has always been, the biopsy of the patient, the biopsy of the tumor. Increasingly with small needle biopsies, but you have to have enough of the tumor to know what kind of cancer it is. You then enter a diagnostic phase, which is very different than it used to be. Certainly, you start with a microscope, but very quickly you end up with sequencing data. And then you have to put all of these mutations within the context of a pathway. As you've heard from Eric Lander, some of these mutations are relatively rare, but they fall within pathways that may be druggable.

And finally, when you get to the druggable stage, there's a lot of information that needs to inform that selection of the drug. Partly pre-clinical screening for which drug works, as well as matching of drugs with particular mutations in particular pathways. And then the part which is becoming the most challenging now, is once you've identified a treatment, you can't wait three, four, five months to get a CAT scan and say wow, does this work or does it not work. Most patients who enter a clinical trial only live long enough to enroll in two trials, because by the time you find out if the treatment works or doesn't work, the disease has progressed. So early monitoring by repeat biopsies to look at the tumor and see how it's changed. By PET scanning to look at metabolic profiles, or using circulating tumor cells is becoming really, really important in terms of monitoring for early responses and adjusting treatment as the tumor regrows.

And again, the lesson from infectious disease is that the better we are at treating cancer, the more we can trigger cancer, the more the cancer adapts, changes. And the more we have to know what a cancer looks like in real time. Now again, in the spirit of the Koch Institute, this is the team that has worked on the circulating tumor cell project with the great support of Stand Up To Cancer. And it involves bioengineers, many of them trained at MIT. It Involves clinicians, many of them trained at MIT. It involves molecular biologists, almost all of whom were trained at MIT, creating these kinds of large and multidisciplinary teams. which at least I feel is going to be essential to making progress in the war on cancer. Thank you.

YAFFE: I'd like to thank all the speakers for their wonderful talks. This is now going to move to a discussion session. It's going to be led by Corbin Meacham. She's a graduate student in Mike Hemann's lab. And she will read questions that have been submitted by the audience or were submitted on the web page, for our panel to discuss.

MEACHAM: Great. So the first question is a two part question. First, what do you see as the necessary steps to move towards a broad implementation of individualized therapy? And what do you foresee as being the most critical contributions from biotech and pharma in the short term?

HABER: Thanks very much, David. Let me start with the second half, which I think is easier. I think we've had a lot of very productive interactions with biotech and pharma. I think the question has always been, is there broad appreciation for the fact that the era of the wonder drug is gone. And that the market and the applications are going to be small. And that even if fewer patients actually benefit from a drug, if they benefit enough, and take it for a long enough period of time, the pharmaceutical industry remember remains viable.

Our impression is it everybody pretty much agrees with that but it still hurts. And there is still varying degrees of appreciation for trying to develop drugs and market them in that way. And to me, at least one of the telltale signs is that when you talk with drug companies, the diagnostic budgets for new diagnostics, and biomarkers, is usually a tiny, tiny fraction of the pharmaceutical budget.

And again, as you can see, if you have to diagnose 1,000 patients to find 100 with whom a drug works, that's a very significant reordering of priorities.

HEMANN: I would say in response to the initial question, I think that you need to, in terms of broad implementation of targeted therapeutics, have essentially, actionable prognoses. You need to have a set of indications where directly, you can apply a specific targeted therapeutic meaningfully. You have to have a model of that scenario and some way to know if this happens, then this is the appropriate therapy. And I think up until now, that's really only happened when people have gone looking for how we can start extracting towards the wider constellation of mutations that we know are present in cancer. I think that's the real challenge in the end.

LIVINGSTON: Perhaps just two things to add to that. I think success in a linear set of clinical trials that to get to a point where as Mike pointed out-- patient with tumor x, genotype y-- has a greater than, very significant number chance of responding in a way to prolongs life to some therapeutic that make sense based on that genotype. That's an important step.

And the second thought is that, it's one thing to do an experiment at MGH, or Dana Farber, or Beth Israel, or Columbia, wherever-- a great academic medical center where there are resources and very smart people. It's another thing to actually have that become practiced care in a community, where the vast majority of cancers treated in the United States. And that's a huge step. Even now, there are centers that do not analyze. Or if they even take samples and try to send them off to get a genotype of the EGF receptor known in non-small cell lung cancer, can they obtain that information in any kind of speed, whatsoever. And if they can, it's usually at abnormal cost.

So the communities of the world, where a lot of medicine is practiced, need to become attuned to the benefits, and the actual necessities of practice of these methods, before they'll ever get to be widely understood and practiced.

YAFFE: I would just say drugs in the future will almost invariably be paired with diagnostic tests, so that every person, every drug, will have a test that has to be done in order to determine whether that'll be useful drug or not. I think as it's been pointed out, the expertise of drug companies lies more in the area of developing the therapeutics then in developing the diagnostic tests. And so it becomes a challenge to try and develop two simultaneous technologies that can work well together.

In addition, I think we're going to have to change the reimbursement policies. And I think this is what David's alluding to. So that physicians to begin with, automatically obtain the set of diagnostic tests that will target particular therapies. We also have to find a way to reimburse at a reasonable rate providers who can provide these types of tests.

HEMANN: I think they those lines, the good news is the therapies of the future may not be new therapies. They may be existing therapies that are used significantly better. And so, there may be incredible value in the portfolios that drug companies already have and compounds they already have. If we can specify better use of those agents.

LIVINGSTON: Just one other thought comes to mind. So far, the examples that people have talked about, where there is a targeted therapeutic effect of a positive nature worthy of report, have been clinical settings in which there is either a single mutant oncogene, or a pair of mutant cancer genes. But rarely more.

One has a sense by looking at those wiring diagrams-- which of course have grown geometrically more complex in the 10 years since the original review that was referred to before-- and the results of the cancer genome anatomy project, and other kinds of global genome analyses, suggests that the combinatorial affects multiple disorder genes-- some disordered on a non genetic basis-- are going to be hurdles that have to be jumped in order to see therapeutic effects.

And we're nowhere near, in our basic science appreciation of a cancer cell, at the point we can, with facility, point to which combination is responsible for Mrs. Jones' neoplastic cell behavior, and the lethality of that tumor. There are neoplastic entities called benign tumors, but they're not lethal. What's the difference between lethality and nonlethality. We tend to say metastasis-- that may not be all.

MEACHAM: This is a related question and it was addressed with some specific examples during the talks. But can you guys comment more broadly on the success of targeted therapies that have been used in the clinic?

HABER: So I would say that the examples that we have, the ones in lung cancer, and melanoma, and breast cancer are incredibly dramatic. And they're really a proof of principle that when you have a right, it really works. The two big challenges of course, what about all the other cancers for which we don't have that kind of fit. And then the fact that these responses don't last that long. So your average EGFR response lasts about a year. The average B-Raf response in melanoma lasts about six months.

So the whole question of resistance mechanisms, how to target those, has become incredibly important. And that's not something again, for which most drug companies have started developing the next generation of agents, which becomes a major limitation.

YAFFE: Maybe, and this alluded to previously, it'll be the staggered use of therapies, much like is done in HIV. Where you target the subset, a large fraction of cells that are likely to respond to a particular drug, followed by targeting those cells that you think may emerge as being resistant. And then back to the original strategy.

So I think we're really going to have to rethink both the drugs we use, and the way in which we administer them.

HEMANN: I think what's helpful about the targeted therapeutics is, as you heard from both of these guys, the mechanisms of resistance to targeted therapeutics are also somewhat targeted as well. So it's unlike emerging resistance to the front line genotoxic chemotherapy, where any number of things could happen. Here, we have a situation where there are known mechanisms of resistance and there's a possibility to specifically target that resistance mechanism, or the resistance allele that occurs. So even if a targeted therapy fails, at least it homes in more on a targeted mechanism of resistance that may be more directly actionable.

LIVINGSTON: And there may be examples-- In fact there are increasing number of examples-- where the best way to learn what a therapeutic effect is all about, is from the patient, her or his self, in groups. There are examples where drug x, which is supposed to attack two targets, one target, has a dramatic-- far in excess-- of anything that has been seen before. Or the lack of effect is far greater than one would have predicted. And by studying those patients' issues, In a more informed way, new insights will come.

So the patients have a lot to teach us by taking the results of clinical trials-- as Daniel and his colleagues did when they found a woman, who among 1,000 or more, responded dramatically to Gefitinib and studied her in great detail. Well, the people who don't respond also, as in the case we talked about before, may be just as interesting. And so, once again, I think deep, ever deeper, mechanism-driven investigation of a patient's fresh tumor may turn out to be quite important.

HABER: There's one of the points that I think is worth mentioning, particularly in these economic times when we wonder about the efficacy of some of these targeted drugs. When Herceptin was first used against HER2 breast cancer, it was used in the metastatic setting. And if you look at their improvement, it was really one of these curves where you can barely stick a finger between the two curves. And you could debate the benefit for quite some time.

Recently, we wanted to do a circulating tumor cell study in patients with metastatic or HER2 positive breast cancer. And we couldn't get enough patients to complete the study. When we called our breast cancer oncologists, they told us well, we don't see these patients anymore. Because what's happened over the last five years is that Herceptin is now used in the early setting when a breast cancer diagnosis is first made. Before it's actually spread, it's given in the adjuvant setting. And that's where you actually see cures.

So increasingly, what you're going to see with targeted therapies is initial application in the setting of metastatic disease, and responses that feel good, but don't last a long time. But if they get moved to the earlier setting, I think that's what we're going to start moving the curve, in terms of cures and cancer. At least for those that have these particular genotypes that we can study.

MEACHAM: So this question was addressed during your talk, but maybe everyone would be interested in commenting on it. So we now have really a wealth of information from genomics and proteomics and bioinformatics. And so, when do you think that these things will actually become clinically useful for the treatment and prevention of cancer?

HABER: I think that they already have. But I think that it would be very foolish for us to say that we know everything that we need to know. So I think the universe of what we know, the universe of genomics and proteomics analysis needs to expand considerably before we can start applying them. But already, we can see in a few examples how in specific instances this can be applied. I think it then comes back to David's point about how do you apply these.

And ironically, as you walk from the Broad Institute to the clinics, where patients are being treated, you take a couple steps down in terms of technology. So the irony is that if you have a test that can be done as a immunohistochemistry test, it can be done cheaply. It can be reimbursed. And you can have the answer within a few hours. If you have to do next-gen sequencing to identify your results, then that will take a long time. So I think in addition to the universe of information that we have, we have to kind of drive it down to a relatively simple and actionable type of assay that can be used in the clinic.

LIVINGSTON: You argue that the ultimate advantage for a cancer investigator interested in therapeutics, even interested in disease mechanisms, would be to, with great facility, be able to isolate a relatively small number of fresh primary tumor cells from a patient's tumor. Keep them alive. Be able to study them biologically, in vivo. That is, while they're alive. As opposed to, as elegant as it is, we analyze genotypes and phenotypes today on dead cells.

But to be able to analyze living cells, and be able to do dynamic experiments where we actually follow, in time, which we can't do now, the dynamics of signal transduction in a pathway that's under attack. That's an enviable goal that may be worth our government investing in.

Just a small pitch.

HEMANN: I think invariably that data has to lead to modeling approaches. That there's an essential step by which you can test outside of the patient. What that alteration or set of alterations means ultimately in therapeutic outcome. And the extent to which we can rapidly do that and recapitulate the actual response of tumors in vivo, I think it has tremendous value. But I think you'd like to be able to know what's going to happen with any defined alterations.

YAFFE: Cheap, multiplex devices. A device that would give you a readout for not just one gene or one protein, but 10 genes, or 10 proteins, or 20 proteins, or the activity states of five different pathways. In a way that you could use at the bedside. And that you could employ serially, to follow what's happening in a patient's tumor during the course of treatment are going to be essential, in order to make optimal use of the information that we're gaining.

LIVINGSTON: Here's a point. I don't know if this respected, esteemed colleague is out there, but I know from talking with my colleagues at MIT, that one of you has the ability to create a sensor the size of a rice seed and introduce it into a body cavity. And follow n number of analytes kind of close to real time. Well, we monitor the heart that way. Daniel and I did that when we were house officers. And I was a house officer considerably before he was. He did it with fancier equipment than I did. But we were doing that in the 1960s. Today, we don't monitor the living behaviors of a tumor anyway except radiographically and with our hands and by talking to a patient, almost ever. Except when we do a PET scan for example.

Well, can you imagine what one could learn simply by sampling biochemical behavior of a tumor through a sensing or series of sensing devices that could give us 24/7 readouts on a tumor before, during, and after confrontation with the appropriate targeted agent? That, potentially could give us insights of great value. Like for example, does targeted agent x always work by a cell autonomous pharmacologic mechanism, or are there ever cell-cell communications that are required for a tumor to regress, for example.

HEMANN: To some extent, the reason why that analysis is probably not done is that there's not a clear sense of what we would do with the information that comes out. Would it change patient outcome, would it change the way that they're treated. So it requires I think, a step forward and the ability to look at network data, large data sets. Do the appropriate modeling to again, leave it to a point where it makes sense. Where we can deduce something from that kind of information.

YAFFE: We should also keep in mind that signaling pathways in tumor cells aren't static events. And tumors are very clever. When you target one pathway, they evolve ways to work around that pathway. And so it may be that the correct approach is to use the drugs we have, but to use them in very clever ways.

To intentionally force the tumor to signal down a pathway that it doesn't ordinarily use in order to survive. And then target that pathway. So I think there is potential to drive tumors to function in the way we want so that we can use the drugs we already have in a better way.

MEACHAM: Okay, so this is a more general question. We've had quite a bit of interest in cancer stem cells and there was interest in wondering if you could talk about what cancer stem cells are, and how they might impact therapy.

HABER: Oh dear.

The cancer stem cell model is incredibly appealing because it connects normal organ development with tumor genesis. And it's incredibly powerful because if it's correct, then you could be theoretically wasting your time treating 99% percent of tumors cells that would only divide a few more times, instead of attacking the stem cell. I think there's some evidence in some cancers that there is a very rigid cancer stem cell model. Definitely leukemia has come to mind, some brain tumors. But overall, I think the interpretation cancer stem cell model is becoming looser and more flexible.

And the understanding is that the hierarchy may actually not be so rigid. And the arrows may go in different directions. And some of the most seminal work comes from Bob Weinberg here, pointing out that what's called epithelial-mesenchymal transitional EMT, is a change in the cell fate of a particular cancer cell. And that arrows can go in different directions. The second most important observation was recently comes from Sean Morrison's lab suggesting that in fact, if you immunosuppress a mouse enough, almost all cells in the melanoma can act as stem cells.

So I think we're in a funny state right now where there's definitely some sense that there's some hierarchy in cancers. And well-differentiated blood or epithelial cancers may have more of that hierarchy then less differentiated cancers. But I think the idea that we're wasting our time treating the dividing cells is actually losing ground. And the idea of a fluid dynamic between the more proliferative and more quiescent cells within the cancer is, I think, gaining ground.

HEMANN: I think from an operational perspective in the context of therapy, cancer stem cell is taken to mean perhaps, a subset of cells that would be less responsive to therapy, in addition to those that can relapse or repopulate a tumor. And it's really unclear whether that in fact as represent a really functionally distinct populations of cells or a set of cells that have found themselves in a particular location. Which is particularly good. It is embedded with growth factors and nutrients that allow it to protect itself from therapy or an anatomical location that is somehow privileged. And so I think that just simply speaks to the heterogeneity of tumor populations and finding strategies to actually treat that heterogeneity.

LIVINGSTON: I think that having been said, it's still an open question as to whether cells that are tumorigenic in some sort of assay-- right now the only assay we have is this in the immunologically incompetent mouse-- represent a special threat to the host. Whether they represent a special threat for tumor growth is one thing, whether they represent a special threat with respect to clinical outcome i.e. death. And that hasn't been explored. And probably would be a fruitful place to look.

MEACHAM: I think we have time for one more question. And this is one of those where the answer's probably going to be both, when I give you two options. But, which do you think will have a greater impact on improving patient outcomes-- advances in early detection of cancer, or improvements in treatment and therapy?

HABER: I think there's a tremendous movement towards arguing that we're not doing well fighting cancer, so we should prevented from developing in the first place. And that's really really nice, if we know how to do that. And I think the problem is that nice as that might seem, the scientific leads in chemo prevention, cancer prevention, and even early detection, really not that encouraging right now.

Early detection is nice, but we're learning that small cancers spread early. It's not so much the size, as the biological nature of it. Chemo prevention took a big hit with VIOXX and hasn't really recovered very well, in terms of preventive approaches.

So I would argue that we have to do what we know how to do. We can measure, which is treat cancer when it's in there. And then apply those kinds of insights and those kinds of treatments to the earlier treatment of cancer with the goal at chemo prevention, once we actually know what we're doing.

HABER: I'm with Daniel on that. I mean, the single biggest carcinogen is aging. And it always has been. It's not that there aren't defined, chemical, environmental carcinogens-- there are. But aging is the primary carcinogen and so it's really hard to imagine with our current understanding, how one would go about the business of attempting to interfere with that process, and at the same time interfere with the outcome.

That said, aging research is fascinating these days. Really fascinating. Beginning with work that came from here and my colleagues across the river, in other places. And so, as one begins to understand the real molecular links between aging, and the development of the neoplastic process, there may be insights to future reductions of cancer risk. Even among aging populations that might be useful. So basic science in this area is absolutely is actually imperative in my opinion.

YAFFE: Maybe the critical part of early detection is identifying the population at large, those who are most at risk. So that we leverage our ability to use diagnostic tools appropriately. A terrific bad example for instance is PSA, a prostate-specific antigen, which as I suspect, many of the audience are aware at one time was used as a biomarker for prostate cancer for screening. It is no longer used because the rate at which we detect tumors, which are likely to be clinically insignificant, is just way too hot.

So part of the best approach to either prevention or early detection is knowing exactly which cohort of patients to screen. It may be that when we all have our own personalized genome on a chip that we wear around our neck, we'll have a good idea of exactly who should be screened for what, at what point in their lives.

LIVINGSTON: I think it is fair to say, and important in strong defense of the people who do study cancer prevention-- which I think is still a primary science-- is that there are behaviors in people that clearly predispose them to the development of neoplasms at younger ages than they otherwise might have developed it. And that kind of research, an effort to reduce self-induced elevated risk is very, very important. And I think could be more widely practiced in the country than it is, frankly.

JACKS: So why don't we bring this session to a close, and thank Mike Yaffe, and Corbin, and the rest of the panelists for their fine contributions.

I just want to say a few words. It falls to me, as it often does in these settings, to give a few concluding remarks. I must say, I'm jealous to the folks who got to sit in the panel seats because they got those Janet Jackson microphones, which I didn't get to have.

Very stylish on you, especially Mike. But this was a wonderful day. It started with a discussion of history and even pre-history, talking about the wonderful accomplishments that have been made in cancer research here and elsewhere. Which have given us a very complete-- not totally complete-- but a very complete understanding about how normal cells become cancer cells. We've learned a lot of the secrets that cancer cells use. There are still some secrets that we need to uncover. But, we've learned a great deal. And we've also begun to see how that information can be translated into new ways to treat and control the disease. And we've seen lots of examples today of the new technologies and the new approaches, that are being used and that will be required for us to do this better.

And I take away from the series of discussions and presentations a great sense of optimism-- that we're actually getting close. That we can sort of see over the horizon and have a sense for what it's going to take. That doesn't mean we're going to cure all cancers, it doesn't mean we're going to do this next week. But we have a sense of how we're going to get there. And that we will be successful in controlling this disease in the not too distant future.

So I want to thank everyone who contributed to today's symposium. The speakers, the session chairs, and I want to give a particular shout out to the young discussion moderators, who did a terrific job leading those sessions as well. So if we can give everybody a hand.

I want to thank David Mindell and the MIT 150 Steering Committee that asked us to put together this symposium as part of the MIT 150 celebratory exercises. It was great honor to do so. I want to thank Robert Urban, who did a great deal of work in organizing the program. Also Susan Hockfield, for her contributions this morning, which I thought were excellent. And Susan was also a huge supporter of the development of the Koch institute at MIT. The idea of bringing engineers and cancer biologists together under one roof. And the power of that for MIT, for cancer research, and really for the world.

And also I'd like to thank David Koch for his generous contributions to allow us to do that. He provided substantial funds to underwrite our efforts to get this initiative going. And we're very excited about what we'll be able to do in the new Koch Institute, and how we think-- and we think increasingly confidently-- will change the course of cancer.

We were hopeful that this session would highlight the progress in cancer research. And I think you saw today just how far we've come. We have a ways to go and I want to encourage the young people in the audience to take on this challenge. We have a lot of our trainees, and young people, and even some undergraduates in the audience.

You've seen the power of new knowledge and the power of new approaches. And I encourage you to use your skills and your talents to help us finally succeed in this most challenging problem.

MIT has a 150 year history of education, and also invention, and innovation. It's in the DNA of MIT to solve problems, to use science and technology to overcome some of the great challenges that face society. And MIT, for many years, has used these approaches to deal with the cancer problem. And I would say collectively, we've done so very successfully. And I'm very optimistic and very encouraged about what the next generation of cancer researchers here at MIT will do to overcome this most challenging and most important problem.

And so with that, I thank you for participating today. I encourage you to stay in touch with us in the Koch Institute. To come and visit the new building, to come and visit our public gallery, and learn a bit more about what we do. And for those of you who are interested and intrigued by the kinds of things you heard today, I also encourage you to come back to this room on June 10 for the 2011 Koch Institute symposium, which will focus on metabolism and cancer. Thanks very much.