Conquering Cancer: Paradigm Shifts - From Biology to Technology to Medical Applications

HYNES: Welcome back to the third session. I'm Richard Hynes from the Koch Institute. And I have the pleasure of chairing this session.

We have three truly excellent speakers for you this afternoon, two from MIT-- Susan Lindquist and Eric Lander-- and one from the West Coast, Lee Hood. I will introduce each of them as they come up to give their talk, rather than doing it all in one go at the beginning. And I'm not going to take any time to show slides and offer you my own opinions on things. We'll wait for the discussion. Maybe we'll do some of that then.

Just let me give you a quick overview of what this session is going to cover. We're going to first start off with Sue Lindquist, with an example of how one can take discoveries made in obscure organisms like fruit flies and yeast, take them all the way through from plants to implications for cancer and neurodegenerative disease. And then we're going to turn to the impact of genomics on medicine and cancer, with Eric Lander.

And finally we're going to end up with Lee Hood talking about personalized medicine and maybe a little bit of systems biology. I'm not sure exactly what he's going to say. But those are the two things for which he's best known, as well as inventing all the machines that Eric used to do the sequencing.

So as I said, I'm going to take more time. And I'd like for us to introduce Sue Lindquist, one of our own. Sue did her PhD at Harvard and then went to the University of Chicago, where she did a post-doc and stayed on for, I think, about 20 years on the faculty at the University of Chicago, where she, through all that period, did wonderful work. Some of which she's going to talk about the later extensions of in a minute.

10 years ago, she came to MIT, to the Whitehead, and was the director of the Whitehead for a period of time and has now gone back to her lab, which she loves and continues as a faculty member at MIT and at the Whitehead. And it's a great pleasure to introduce her to talk about heat-shock proteins and their implications. That wasn't the title she gave me, but it will do.

LINDQUIST: Thank you, Richard. It's a pleasure to be here today. Thanks for the invitation. I'm a fairly recent transplant to MIT. I came about 10 years ago, and I absolutely am crazy about it.

I'm going to tell you about some work that I started many years ago. In effect, this slide here is the only experiment that you'll see in this presentation that I did with my own hands. But basically, what this represents is a survival response, an illustration of the survival response, sometimes called the heat-shock response.

And what we have here is I grew a culture of yeast cells at 25 degrees. And I took two aliquots of the culture out. And one of them I put directly up at high temperatures, 50 degrees. And the other one I first gave about 30 minutes in an intermediate temperature and then put it up at high temperature. And so you can see that brief pre-treatment, just 30 minutes, provided incredible increase in the ability of the cells to survive that higher temperature insult.

And this is a very similar experiment, done by a graduate student in my lab, with Arabidopsis seedlings. Same basic idea, mild pre-treatment providing a tremendous amount of protection. You can do this with any organism on the planet. It's been done with fruit flies, it's been done with whole mice wrapped in electric blankets.

But basically, the same thing holds. A conditioning treatment allows a tremendous ability to survive a more severe treatment. And so what's going on in those cells while they're getting conditioned and learning to cope with the more severe conditions?

Before I go on to that, I should probably say that these mild conditioning pre-stress conditions will actually provide protection not just to heat but to all kinds of other stresses. They'll provide protection against oxidative stress, against heavy metal ions, against just osmotic stress, a huge array of different stresses.

So what magic is going on in those cells? Well, they're making a new set of proteins. So these are cells that were pulse labeled at 25 degrees, the normal growing temperature with a radioactive amino acid. And these cells over here were pulse labeled at that conditioning treatment.

And you can see that the pattern of protein synthesis is quite different. So they pull out all the stops. And they start to make new proteins that are going to help them to cope with the high temperatures and many other types of stress.

And it turns out that we were able to study this in a variety of different organisms. And we were working in different organisms, depending on what we wanted to do. But with yeast we could genetically manipulate these genes, way before we could do it in any higher organism.

And so we found something kind of interesting about hsp90 that turns out to have something interesting to do with cancer. This is my first little foray into the cancer field. Hsp90 we found is a very abundant protein normally. And it just gets up-regulated with heat. But normally the cells don't need anywhere near the amount that they're making. They can get away with much, much less.

But when they're under any condition of stress, they really need all they can make. And then we didn't know what was going on with this protein, what it was doing, until it was reported that-- by several different laboratories, kind of accidentally ran into it, they were working on cancer cells. And hsp90 was bound to the mutated oncogenic protein that was driving the cancer. And so we decided to take advantage of genetics in yeast and lower the hsp90 levels and see what happens when cells are able to use a lot of extra hsp90 or not.

So we expressed one of those cancer causing proteins, v-Src, in the yeast cell. And what we found was that in terms of just looking at Src accumulation, there was a little bit less Src accumulating. But pretty much the cells that had normal levels of hsp90 and cells that had much lower levels of hsp90 many were making pretty much the same amount of protein.

But when we looked at the activity of that oncogenic kinase, which is a tyrosine kinase, the activity was profoundly different. So the cells that had lower levels of hsp90, which could grow and multiply and do everything that a normal yeast cell can do, could not mature this oncogenic protein. And then we looked at the c-Src protein-- the normal, unmutated form of the protein-- and we found that well, we had to expose this 20 times longer, because of course it's a much less active kinase.

But when we did, there wasn't too much of a difference between the cells that had full bore, large amounts of hsp90 and the cells that didn't. So this protein, in other words, is not very needy of hsp90. This protein is very needy of hsp90.

And it turns out that this is not just true of yeast cells, not just an oddity of yeast cells. I had gotten the idea when I saw this that maybe hsp90 might be a really good therapeutic target in cancer, because you can in normal cells reduce the amount that you're using and they couldn't care less. But the ontogenic protein would not be able to be matured.

And I went to the intellectual property office at the University I was then working at, and they were quite rude. They laughed in my face. And so after a while, I kind of gave up on this story. But luckily other people did not.

And so here's an example of a experiment that was done by Luke Whitesell, a wonderful colleague of mine, looking at the ability of v-Src to transform and make these cells grow like crazy-- completely out of control-- or in the presence of an inhibitor of hsp90-- which causes them to revert to normal morphology-- and they're able to grow. And they do just fine, but they're normal cells.

And it's turned out that we understand very well now how this works. It's universal, these heat-shock proteins are conserved across all biological systems. And what happens is that the normal Src kinase folds back up on itself and keeps itself under control.

But the mutated form can't fold back on itself. And so it opens up, and it's splayed out. And it's actually very, very subject to degradation and aggregation and doesn't work very well. Except that hsp90 comes along.

It's a chaperone protein, that's a protein that helps other proteins to fold. It helps to get it folded, gets it inserted into the membrane. And then it goes away and leaves it free to be active. And we now know that this is actually true for many, many oncogenic kinases, in fact, the ones that were discussed earlier this morning as well.

And so it has become a fairly interesting new therapeutic strategy, just exactly what I had initially proposed a long time ago, is that you might be able to use inhibitors of this hsp90, protein because there's a large therapeutic window. Normal cells don't need very much. And these are just some of the pivotal licensing trials that are in progress right now. And this is an example of one of the earliest positive responses, seen in a patient with a breast cancer lung metastasis disappearing.

It's still early days. And we don't know how broadly this is going to be used. But the idea was that there might be a broad strategy here, to therapeutically target something basic about the biology of the cancer cells, rather than necessarily an individual target. I think they're both very, very useful.

So that's the way how one protein of this heat-shock response interfaces with cancer. What about the response as a whole? So I had, as I mentioned to you, I had kind of dropped this. And we were working on other aspects of the heat-shock protein functions, including hsp90s functions. But I had not really been enticed into working any further on cancer, until I came here.

And the Koch Cancer Institute is just simply so fantastic and so exciting and has so many wonderful people that I realized well, gee, I really would like to know how this ancient survival pathway interfaces with cancer cells. Are cancer cells using this survival pathway in more ways than just that one hsp90 protein is using it, to help mature and fold a cancer causing protein? And so because I was here, I was able to tackle this question.

And we took advantage of the fact that there's a master regulator of this whole response that's been conserved from yeast to man. And lots of different people have worked on it. And it turns out that it's a dispensable protein, except for stress responses. At normal conditions, you don't need it.

And so Ivor Benjamin had made a mouse that was deficient in this protein and could not make a stress response. And they could not survive a heart attack very well, or they could not survive a stroke very well. Otherwise, they were just perfectly fine. They lived a perfectly normal lifetime.

So we asked, what about their susceptibility to cancer? And the very first thing we did was to just use a rather simple chemical carcinogenesis protocol, where you paint the backs of the mice with a mutagen and a tumor promoter. And in the normal wild type mice, they're sprouting tumors all over the place here. And in the mice that are not capable of mounting that survival response, they're profoundly protected.

And this translates into a protection in terms of their lifetimes, as well. These are the mice have the knockout. And they're very, very-- they live a long time. These guys, the wild type, die.

And so that's a cancer that's driven by the Ras oncogene. How general is this? What about a cancer that's driven, as Jackie was talking about today, cancers driven by a very different type of mechanism, the loss of a tumor suppressor function?

So we went to Tyler Jacks. And we got his p53 mouse model. And here again, you can see that being deficient for the stress response provides profound protection. In other words, the cancer cells are using this survival pathway to help them survive in their really multi-faceted, deranged states.

So we looked at in a more detailed molecular mechanism how the cancer cells were using this stress response. And it turns out that the stress response was doing all sorts of things to help the cancer cells. It helped with cell proliferation.

It helped them to ignore apoptotic signals, the cell death signals. It changed all kinds of signal transduction pathways. It helped make the cells more growth factor independent.

And it changed their metabolism. As was mentioned earlier this morning, there's this Warburg effect. That really in part depends upon being able to mount the survival response. And when you think about it, such a profoundly multi-faceted survival response that affects so many aspects of the ability of all kinds of organisms to survive, it makes sense that it would be used in lots of different ways to provide a change in the biology of cancer cells.

So here's an example the Warburg effect. Cancers love sugars. And the difference between the heat-shock wild type cells, in terms of the uptake of sugars and the knockouts is really quite profound. That's just one example of data, I'm not going to dwell on this.

But of course, that's mice, right. So what about human cancers? And what about the maintenance of cancer, not just the initial induction of cancer?

And so one of the things we did was to go around to various labs at MIT and get lots of different cancer cell lines that had lots of different underlying genetic lesions from lots of different human tumors. And in fact, when we turned down the heat-shock response, they died. But what about renal tumors in a living human being?

This is a biopsy section of breast cancer. This is the normal tissue, and you can see it juxtaposed to the cancer tissue. And the brown staining is that master regulator of the stress response. So it's sitting in the cytoplasm and not doing anything much in normal cells.

When the cells in the cells that have been transformed in these tumors and are growing out of control, it gets expressed at a much higher level, and it goes into the nucleus, and it drives this program of survival. We at least can infer that this-- we know that the mice are using it. And the fact that the human tumors have so much HSF in the nucleus would suggest that they are being driven by this-- the cancer is helped and aided and abetted by the stress response.

That's not just specific to breast cancer. Here's prostate cancer. Here's lung cancer. And here's colon cancer.

Now it's not every cancer. It's only in about 70% of the cancers we've looked at. But it's in an awful lot of them.

And so we think this might be a strategy for therapeutically differentially targeting cancer cells in a very broad sense. And so we've set cells up with a response system that can report on whether or not they're making a heat-shock response, and whether or not they're inhibited for the production of other proteins as well. And I won't take you through all this. But I'll just say that we can very easily screen for both potentiators and inhibitors of cancer-- sorry, potentiators and inhibitors of HSF function.

And where did we do that? We did that at the Broad, where they have just as you will hear from Eric very soon, they have some really spectacular capabilities. And so we were able to screen, with the help of the Broad, 350,000 compounds for the ability to either inhibit the response or to potentiate the response. And the reason why we might also want to look for inducers of the response is that if you break the function-- it's an auto-regulated response. And so if you break the functionality of it, you'll wind up inducing a response as well.

Anyway, I'll just show you a couple of quick, very quick little vignettes of some things we're just starting to do now. And we found one compound-- I'm going to tell you just about two compounds that we followed. They're both natural products.

This one is from mahogany trees. And it actually had already been known to have anti-cancer properties. It was just not known why it might have anti-cancer properties.

And what's really cool about this compound is it's a very complex natural product. But total synthesis has been achieved, and that enables medicinal chemistry and SAR. And we've been doing that with John Porco's group at Boston University.

And this just shows you that cancer cells-- this is a transformed cell line. They're both transformed cell lines, but this one is a very highly malignant cell line, and this one is a very poorly malignant cell line. But basically, we seem to find-- we've tested a lot of lines-- the more deranged and metastatic and oncogenic the cell line is, the more it seems to depend upon the stress response.

And this is just the beginning. I showed you the glucose uptake before in the tissue culture cells. And this is-- a tumor is implanted in mice, treated with DMSO or tumors implanted in mice treated with this rocaglamide. And as you can see, it's interfering with that. We think it's interfering with a whole bunch of other stuff. We're just doing the survival trials on these mice now.

And finally, I just want to mention quickly another herbal compound that actually has been used as a wellness tonic in Ayurvedic medicine for 3,000 years. And you can buy it yourself in health food stores. But I don't recommend it, because the concentrations that are in different preparations of this vary by 100 fold in different health foods supplements.

Anyway, that compound, too, looks like it has some promise. This is an intracranial human glioma xenograft model, where we implanted glioma cells, which are a really highly, highly, malignant cell type into the brains of these mice. And what's cool about this compound is that people have been taking it because it helps them relax, and it makes them feel good. And that's because it gets into the brain.

And so in these mice that were not treated with Efferin, they die. But with Efferin, two different doses, it's slowing the advance of the cancer. And in some cases, it seems to be allowing the mice to live on without the cancer. This has been done a couple of times, but we need much larger numbers.

Anyway, I just want to say that I think that the environment here that's provided by the Koch and the Broad and the Whitehead and the biology department and this amazing town has just been a wonderful thing for me in terms of-- I'm not a cancer biologist. But it's enabled me to maybe try to make some kind of a contribution, in terms of understanding the generalized biology of cancer cells and how they might be using this stress response.

And I want to thank the people in my lab for all of the wonderful work they've done. These are the folks who did that initial work on v-Src that was initially rather ill treated when we first did it. And this is Mikko Taipale, who's working on the hsp90 now in the human cells and doing some really fabulous stuff, I didn't have time to show you his work.

Luke and Chengkai did all that work on the mice with the tumor models. Sandro Santagata did all of those sections and pathology and did the screen, set up the cells. Leslie, a natural products chemist and John Porco, a collaborator here at Boston University. And then all these amazing-- I can't begin to list all the people at the Koch Center who have helped and the Broad Institute who have helped. But I'm very grateful to you. Thank you

[APPLAUSE]

HYNES: Thank you Sue, for a wonderful talk and as you said, a great example of what collaboration can achieve and also a great example of taking basic science on flies and yeast and having it show its impact in cancer in not such a long time. That's really a delightful story.

So our next speaker, too, is an associate member of the Koch Institute, who lives across the street from the new Koch building Eric Lander, who is the director of the Broad Institute. Eric has two degrees in mathematics from Princeton and Oxford and no degree in biology that I know of. I was department head when we hired Eric. And he at that point was teaching a course in negotiation at the Harvard Business School. So he negotiated a good package. And--

[LAUGHTER]

--and ever since, he's negotiated from that point upwards. The Genome Center actually started in the Cancer Center, the old building. But the space wasn't big enough for Eric. So he moved off campus and set up the Whitehead MIT Genome Center, which played a big part in the sequencing of the human genome.

And once that was done, he looked around for something else to do and negotiated to set up the Broad, which has gone from strength to strength to exploit the genome. And he's going to tell us about some of that. Eric.

LANDER: Richard, thanks Richard. Well, 150 years of MIT. The Cancer Center represents one quarter of the history of MIT, about 36 years. It started in 1974. It's just fantastic to think that it's actually a whole quarter of MIT's history, or on the other hand, that it took a quarter of MIT's history to get a building worthy of the Cancer Center.

But it's happened. And I could not be more excited to look across from my window every day and see the Koch Institute, busily, busily functioning and bringing together science and engineering.

I haven't been around as long as that. But I have been around 25 years at MIT, so I've got one sixth of the history of MIT. And that's both exciting and mildly depressing, to think that I could represent some actual knowable fraction, meaningful fraction there, of that history. But it's been an amazing one sixth of MIT's history in the last 25 years.

I just want to talk a little bit about where things have been going over that 25-year period and think a little bit about where they're going to be going. So, yeah, we do have that. I want to talk about taking comprehensive views.

Mostly what I'm interested in is comprehensive views. Maybe it's because as a mathematician by origin, I like kind of stepping back and seeing the big picture. And it hasn't been possible to do that until relatively recently, really that last sixth of the history of MIT, has it been possible.

Now in cancer, much of the 20th century was devoted to arguing about what cancer was. The notion that cancer was a genetic disease was proposed as early as 1914 and then prominently forgotten. As late as 1970, you could find definitive reviews assuring us that the viral origin of the majority of malignant tumors is now documented beyond any reasonable doubt.

When people say things like that, you should reasonably doubt them. Because not too long thereafter, evidence began to accumulate that in fact, the majority of the causes of cancer was genetic, was rearrangements, mutations. That in fact, cancer was a disease of the genome, with Bishop and Varmus's discovery of viral oncogenes having cellular counterparts, with Bob Weinberg's incredibly important discovery of a first human proto oncogene.

And by 1986, one had a pretty good picture of some examples of point mutations, translocations, amplifications in the lesions, telling you that all the genetic mechanisms that could happen, do happen. And the question was, well, what's going on past there?

Renato Dulbecco wrote a very influential article in 1986 in Science Magazine, in which he said, look, we can either do this retail or wholesale. Either we're going the hand-to-hand combat, looking for a gene here or a gene there. Or we're going to step back and take the big picture. We're going to get the genome of, as he said, a selected species. He argued it should be the human.

And he and others around the mid 1980s, just around the time I had come down from Harvard Square to Kendall Square, were arguing for the idea that became the Human Genome Project. Well, that project got off the ground in 1990. It ran to about 2003. It produced first the draft sequence of the genome, around 2001 it was published. And a finished sequence of the genome completed around 2003 and gave us for the first time that big picture look.

Now I'll talk in a moment about what that's good for. But I have to say what has happened in the period since, roughly this last decade, is even more remarkable than the work that happened in the Human Genome Project in some ways.

The Human Genome Project, everybody agonized, should we do this project or not. It seemed so huge. How would we ever get the genome done? It was going to suck up all the money in biology.

And folks, look, somehow we're going to find ways to do it. Lee Hood, our next speaker, came up with ways to improve technologies to be able to do this. Many strategies were developed. And it turned out to be completely doable.

Boy it's good we went down that path. Because it's begun this virtuous cycle of improving and improving and improving technology. And my mind is blown by what's going on by this world of technology now. All sorts of new machines have come along, based on new approaches to sequence.

And to give you a sense of what the impact of these new ideas for sequencing has been, I'll just show you a graph I used to be very proud of. It showed our output of sequence from the year 1999 to 2006, when over at the Whitehead MIT Center for genome research and Broad Institute we went from 1 billion bases a year 70 billion bases a year. That felt really good.

And then as these new technologies came in, it looked like this. Two more years added to this graph looks like that. Let me add another year to that graph, looks like that. Let me complete it to the end of the year, like that.

About 125,000 billion bases of sequence last year. And we might pick up almost an order of magnitude this year. It is stunning. The cost of doing this has fallen by 100,000 fold already and will likely, before this technology levels out, fall by about a million fold. I don't know anything that's every decreased by a million fold.

It is stunning. It is really good we decided let's do the project. Let us figure out how to go down this path.

Now what has been the implication of this for cancer? Well, we're beginning to see that implication. We've seen some of it in the past decade. But for cancer, most of that impact, I think, is in the next five years or so, in terms of getting genomic information.

So what have we learned? Well, we've learned lots of things. I'm not going to talk about the many things we've learned about inherited diseases and human history and the evolution of the genome and what's in the genome. For these few minutes we have today, I just want to say what is it doing for us in cancer?

It's letting us understand the basis of disease in a completely unbiased fashion. By asking the cancer genome when it mutates, what's changed? And piling those data up for many, many patients and seeing.

What's the impact? Well, a few years after Renato Dulbecco wrote his article in 1986, the number of bona fide oncogenes in solid tumors-- I could get good numbers here for solid tumors-- was about a dozen. A decade later, around the time the Human Genome Project was completed, it was about 80.

Where are we now? Now, it's about 240. And lots of interesting surprises and things with therapeutic impact have come out.

Sequencing to look for mutations has pointed us to mutations in BRAFs in melanoma, in PIC3CA in colon cancer, EGFR mutations in lung cancer, IDH1 mutations in glioma. And each of these things has led either to therapeutic or diagnostic implications already, some of them already in the clinic, some soon to go into the clinic. It's led to discoveries of new types of oncogenes. Lineage specific transcription factors are turning up in multiple cancer types as important oncogenes.

There was an official dogma that translocations mattered in blood cancer but not in solid tumors. That's false. It turns out when you look, solid tumors have recurrent translocations. And then of course, combining genomic aberrations with functional studies, like RNAi, has let you intersect lists and find interesting and previously undescribed genes that play roles in cancer.

So this story, which I'm saying goes from about 2000 to 2010, began to get going. And around 2005, some of us were part of an NCI effort to create something that eventually became the Cancer Genome Atlas Project. The argument was kind of simple. It was, there was going to be a lot of information, and we should get it in a systematic fashion.

And just like the Human Genome Project, it provoked enormous controversy. The Human Genome Project folks said, oh, look, it's going to suck up all the money. Why don't we understand what we know. It's too expensive. Let's wait until it gets cheaper.

And of course, you know, the answer tends to be start down the road. Because unless you start, it doesn't get cheaper. Unless you start, you don't figure out how to do it.

And so let's start. And so a pilot project was organized. The pilot project ran for about three or four years. And now it's just beginning to get into its serious ramp-up phase.

Now, why were people concerned about this? Well, they were concerned because they said, you know, cancer is very complicated. It's hopelessly complicated. That was about half the objection. The other half of the objection was we already know all the genes in cancer.

And one wished you could get the two to argue it out with each other and have to only argue with one side at a time. But in any case, I come down squarely in the middle. Cancer is not hopelessly complicated. But we sure don't know everything, and we ought to be looking.

Of course, any one cancer is very complicated. All sorts of things go on in a cancer. There's lots of background noise.

But by looking at many cancers, many tumors of the same type, you can begin to pick out a common patterns. Genes are mutated 50% of the time, 20% of the time. With enough tumors, you can see that genes are mutated 3% of the time, and that's significantly above background, then it's telling you something.

So if you look at brain tumors, glioblastomas, you can find there are about 35 regions of the genome that show consistent amplification or deletion. And we can tell you a good story about 40% of them. The other 60%, we don't know what's going on.

That tells you it's not infinitely complicated, the level of amplifications and deletions. But we don't know everything. We can get pointers into what matters.

Now beyond looking at amplification and deletion, which you could do by some fairly easy tools, direct sequencing. Sequence the genome of a tumor. Sequence the genome of a normal. Compare them.

Any one individual differs from a reference sequence by about one letter in 1000. But an individual differs from a tumor by about one letter in a million. There are only about 30 or 40 or 50 coding changes in proteins. And 3,000 non-coding changes around the genome.

It's conceivable if you looked at enough of those, you could begin to sift them out. You could look at the copy number and the mutations and the rearrangements, et cetera.

I'll tell you a story about a paper that's coming out in the next couple of weeks from my colleagues Todd Golub and Gaddy Getz and Mike Chapman and others at the Broad and the Dana Farber about multiple myeloma. But I can tell you lots more stories like this. I'll just pick this is as a poster child example.

Multiple myeloma, it's about 20,000 new diagnoses a year. It's a malignancy of plasma cells, involving massive secretion of Ig. Well, we sequenced 50 genomes, or in some cases exomes-- the coding regions-- of multiple myeloma patients's tumors and normal pairs. Seven new pathways emerged from that.

Pathways, well, known pathways, KRAS, NRAS, p53. But 40 odd percent of the patients carry mutations affecting protein homeostasis. In retrospect, it makes sense. These things are protein factories, churning out protein. NF-kB pathway is clearly mutationally deranged in 25% of cases.

Two patients, only two, had mutations in the IRF4 transcription factor. But they had the identical mutation. And two more patients had mutations in a target of that transcription factor.

BRAF, one patient in our sample had a BRAF mutation. But there's a drug against BRAF. So we looked at a bunch more patients, and golly, about 4% of patients have BRAF mutations.

And I have no idea whether the BRAF drugs will work against them. But if you had somebody you loved with multiple myeloma and had that mutation, you'd sure want to know that.

The thrombin pathway is mutated in about 16% of patients. And that's an extracellular pathway. And what is that doing being oncogenic?

Maybe because thrombin comes back and acts on the par one receptor in a mytogenetic fashion. Maybe not, but you wouldn't have guessed that one. And et cetera, you get the point.

There's a lot of this going on, just at the Broad, right opposite the Koch, we've done about 1400 tumor normal pairs so far. Actually, this slide is a month old. We've done more than that, but I don't have the current number. This is roughly what's been going on there.

And this is just the start. As the costs of sequencing are falling down to $10,000, to $5,000, to I hope within another several years, $1,000, one will see tens of thousands, and frankly hundreds of thousands of cancer tumor normal pair sequence.

We already can see that there are important differences. They have very different mutation rates. Different types of cancers have different mutation rates, as you can see there. That's a log scale. They have different patterns of mutations, as you can see from the colorful pictures underneath that indicate the mutational patterns.

What do we have to do? What does the future look like? We need comprehensive views of cancer.

For patients, that's going to mean hundreds of thousands of tumor normal pairs. That's not a hard number to come by. In order to detect things that are running around at a couple of percent and are still significant, you need to look at a few thousand samples.

And there are 50 or 100 cancer types. Who knows, let's not argue about it. Let's begin to start and figure out sub-types and go deeper as we do that.

We need to do this for cell lines. There's a project that started here already to take 1,000 of the most important cancer cell lines and fully analyze them as well. And of course, animal models, all animal models, should be characterized mutationally in this same way.

But it isn't just sequencing. I've picked that to focus on right now. We need the same thing functionally.

We need to understand what are the essentials in cancer cells. Take RNAis against every gene in the genome. Knock out every gene in the genome in appropriate cellular models, in animal models, and be able to understand every essential gene. We need to be able to put back genes and ask which knockouts or additions can confer resistance on cells that have been insulted in various different ways, so we can build in advance. Look up tables of how a tumor will become resistant to drugs not 10 years later when the drug is in the clinic and patients now develop resistance in nine months, but right while you're developing the drug.

The real goal? Map this all back onto pathways. Map the mutations back onto the pathways, map the essentiality back onto the pathways. Map the potential sites of resistance back onto the pathways. Map the drugs available back onto the pathways.

And a physician of the future gets a report that says, aha, lesions here. Doesn't make sense to use the drug here, makes sense to use the drug there. And so on.

We need to think about not just the laboratory-- the structural information and the functional information-- but the clinical information. It's already time to start asking, when is this going to become the standard of care for cancer patients. It should be soon for anybody that I loved, that they could have this information. And when do we take all patients with cancer in the world and give them the opportunity to contribute their data freely to a common database, together with their clinical information and [INAUDIBLE] so that we can all be part of an experiment that we can learn from?

It has been a fantastic 150 years at MIT. It's been a fantastic 36 years of the Cancer Center. It's been a great 25 years that I've been here. But the next 25 and 36 and 150 years are going to be far more remarkable than that. And so I leave you with MIT plus 150, and I look forward to the discussion. Thanks.

[APPLAUSE]

HYNES: Thank you, Eric. I would like to remind you that we're going to have a panel discussion after the next speaker. So if you've got questions, and I'm sure those two talks must have stimulated many questions, write them on a card, and they will be collected. And we will try and address some of them in the panel discussion afterwards.

So a second wonderful talk, and now we're going to have a third one from Lee Hood, who's from the University of Washington and, in particular, from the Institute of Systems Biology, which he set up there. Lee is really actually our only M.D. on this group of speakers. He got his M.D. from Johns Hopkins and his other degrees from Caltech, where he spent many years and during that time, developed some of the earliest high throughput sequencing machine for both proteins and DNA, which initiated the revolution that Eric's just reviewed for you. And he moved from Caltech in the early '90s to Seattle, where he's been ever since.

His MIT connection is that in 2003, he won the Lemelson-MIT prize for innovation and invention. And I noticed in his long list of awards and things like that-- as the other two, I picked out one two only-- in 2008, Wired Magazine said he was one of the 15 people the president should listen to. So I hope you're getting his ear. But if you're not, you can tell Eric. Because Eric has the ear of the president. So Lee, good to have you.

HOOD: Thank you.

[APPLAUSE]

Well, I'm not sure he listened. But I'll tell you what I wanted to say to him anyway. What excites me about this symposia is the idea of paradigm changes. In fact, when I was a young assistant professor at Caltech in the '70s, I was really struck by reading a book called The Structure of Scientific Revolution, I think a really pioneering and transformational book that certainly changed my thinking about things.

I was fortunate enough to participate in a number of paradigm changes. And the first four that I participated in really set the framework for what I want to talk about, namely P4 medicine. And the essence of P4 medicine is how do you bring science to the patients, and how can you do it effectively?

So when I started at Caltech, as you heard in the 1970s, I was thinking about engineering and developing a series of instruments that essentially allowed us to translate biological information, DNA, later RNA, proteins, and the like. What that, interestingly enough, led to is what Eric talked about, the Human Genome Project. I got invited in '85 to the first official meeting on the Human Genome Project.

And I think the major conclusion we came to then was it was possible but difficult. And I must say, going out into the community and seeing 90% of the people firmly opposed for all the reasons that Eric enunciated-- and most of all, NIH was opposed to up until the very end-- was an interesting and fascinating experience.

Developing the automated DNA sequencer made me realize that, indeed, cross-disciplinary biology is going to be essential for the future. It is bringing in the engineers and physicists and the chemists and the computer scientists that really created the tools that allowed us to be able to move forward as we have. And I tried in the late '80s to set up such a department at Caltech, and the biologists weren't interested. Bill Gates let me do it at the University of Washington.

And that department, eventually, led to the Institute I set up now, the Institute for Systems Biology, whose mission was this systems science taking a systems approach to disease. And we'll talk just a little bit about that in a few moments.

But it was the sum total of all of these that really led to this thing I call P4 medicine-- predictive, personalized, preventive, and participatory. And what was interesting in thinking back over these paradigm changes, I think, were two important points. One is that each was really met with enormous skepticism. And the second was in every single case, we had to create new organizational structures, or I don't think they ever would have seen the light of day.

So what is P4 medicine all about? I would argue it's all about information. And my prediction is sometime in a 10 year future, individual patients will be surrounded by a virtual cloud of billions of data points. And we'll have the wherewithal to reduce that data dimensionality to simple hypotheses about health and disease. And a really important point about the data is it's going to be enormously heterogeneous. And indeed, I predict, that social data is going to be a very, very important aspect of this patient data mosaic.

So what then are the real foundations of P4 medicine, of systems biology? I would argue it is this view-- biology, medicine, is an informational science. The idea that there are two fundamental types of information, genome information and environmental signals.

The idea that those types of information are connected to the phenotype by biological networks and the execution of molecular machines. And of course the idea that there are many different levels of data that in a sense, if you're to really understand the system, have to be summed and integrated, so that you can explicate the various environmental modifications of each of these different levels of information, DNA to RNA to protein to networks and write up the hierarchy, ultimately to ecologies.

So by this token then, disease is a function of disease-perturbed networks, genetically or environmentally modified. It changes the envelope of information. And this altered dynamic envelope of changed information, both planes the pathophysiology of the disease, and gives us insights into diagnosis and therapy.

And indeed, I would like to give you just a very brief vignette of the disease that we've studied for more or less the last 10 years from a systems point of view. And that is prion neurodegeneration in mice. You can induce it at a particular time point. You can survey how the information in the brain changes at multiple time points across the full onset of the disease. You can subtract normal from diseased counterparts.

And if you do, you horrifyingly end up with a third of the mouse genes being changed, more or less. And you have to do signal to noise reduction procedures, especially that deal with the biological noise. That is, the phenotype is not one aspect that you'd be interested in, namely neurodegeneration, it's always many aspects of biology.

And you have to subtract out the unwanted aspects of biology. When we did this, we could get a 20-fold enrichment down to about 300 genes or so. And then we mapped them into four networks that were defined by histopathologic analysis, prion accumulation, glial activation, and then two aspects of neural degeneration. And for each, we were able to define networks that gave us fundamental insights.

And I would say two really important points came from looking at all of these networks. Number one is there are a lot of other networks that we never ever knew about. And the networks are disease-perturbed in a sequential manner, starting with prion replication and accumulation.

And number two, the dynamics of these networks explain virtually every aspect of the pathophysiology of the disease. And, of course, it led to interesting ideas to think about systems approach to disease.

I don't have enough to talk about the three different approaches that we've developed, one dealing with functional proteins that actually get secreted in the blood, as reflected by these disease-perturbed networks. A second organ-specific protein, I'll mention more about that. And a third we've done recently, and very fascinatingly in the brain, cell type specific proteins that get secreted in the blood that actually can let you interrogate how particular cell types are actually functioning.

So as you can see here, how do we go about identifying organ specific proteins that are secreted in the blood? And the procedure is not simple. You take deep transcriptome analyses of many different tissue types, organ types, in both human and mouse. And from those, you can deduce transcripts that are organ specific.

We can then actually test whether those organ specific transcripts make proteins that get secreted in the blood, using targeted proteomics mass spectrometry. And then finally, we have to be able to treat the blood. Because the range of proteins in the blood is of the order of 10 to the 11th or 10 to 12th, so we have to remove the top 20 or so proteins, so we can see down into the lesser dimensions of protein concentration.

And when we do that, we indeed can get significant numbers of organ specific proteins. Indeed, we have about 100 for the brain and 88 for the liver, the two organs we focused on. And these organ specific proteins, then, constitute fingerprints. And in a normal brain, the 100 proteins will display one set of concentration levels. In a disease-perturbed brain, you'll see those proteins whose cognate networks are disease-perturbed alter their concentration levels in a manner that's characteristic for each disease.