stan lipkowitz:i want to thank both suburban and the genome institute for inviting me, and thank you forthat introduction. actually, in the work that i'm going to talk about today, i'll probablyspeak -- i'm really not going to talk about any of the work we do in the lab, but reallydo an overview of how genomics and genetics are -- have and are currently impacting thetreatment and diagnosis of breast cancer, and how this is likely to be expanded in thenear future. so to begin this lecture, what i'd like todo is, for the purposes of my talk, define what i mean by genomic medicine. and simplywhat i mean is the use of molecular genotype, which usually means the sequence of the dna,the sequence of the genes, and the molecular
phenotype, which is the expression eithermeasured by messenger rna or by protein expression, to predict disease incidence, outcome, and/ordictate treatment. now in cancer biology, there are two genomes, and it's importantto keep this in mind. one is the tumor genome, which is referred to as the somatic genome,and the other is the patient genome, or the germ line. and last month you heard from larrybrody about germline mutations in genes like brca1 and brca2, that predispose -- that predisposepatients to getting breast cancer. today, most of my talk is going to focus on usingthe tumor genome to dictate decisions about management of patients, although i will touchon germline mutations as well later on in the talk.
now in the past, treatment was based on clinicalfeatures of breast cancer, and this is true for pretty much all tumors, that featuressuch as size, pathologic grade, the spread to the regional lymph nodes. but also in thepast, it involved expression or genetic abnormalities, so genomic features of the tumor, but of onlya few genes in the tumor. now this is not a new story, and in fact, the -- really thestory begins with this publication, which i would say is the first therapy based ontumor phenotype, its measurement, or its affecting the estrogen receptor, which of course isstill used today in the management of breast cancers and really important feature of themanagement of breast cancer. but this paper by george beatson was publishedin the lancet in 1896 about a novel treatment
for patients with inoperable breast cancer.so this is the only case history i'll describe today, but the patient he saw was a youngwoman, 33, premenopausal, who, several months earlier, had presented with a very large locallyadvanced tumor. it was 11 by 8 centimeters, there was clear skin involvement, and sheunderwent a mastectomy. three months later when he saw her she had diffuse chest wallinvolvement, she had ulceration of her skin, involvement of lymph nodes, and apparent metastaticdisease in her thyroid. so this patient at this point was unresectable, and at the time,in the 1890s, there was no other treatment for this patient. but what he did is her removedher ovaries a month later, and over the next couple of months, she developed a completeremission in her breast cancer, and actually
survived for four years before dying frombreast cancer. so why did -- why did beatson do this? well,what he was known, at the time, from both animal studies that he and others had done,was that the ovaries somehow controlled the growth and differentiation of the breast.what was also -- had been observed, but never really acted on, was that patients who hadbreast cancer and then went through menopause sometimes had spontaneous remissions of theirbreast cancer. so his reasoning was, "well, gee, maybe inducing menopause will cause aremission in this patient," and of course we now know that that's absolutely correct. so why does this work? this works becauseof the tumor phenotype. breast cancers, in
about two-thirds of all cases, express a proteinknown as estrogen receptor, which sits in the cytosol of cells as in inactive proteinuntil estrogen comes along and binds to the receptor, and that creates an active complex,which one of its primary functions is to go into the nucleus and act as a transcriptionfactor, where it regulates the expression of other genes, and these genes drive thegrowth and proliferation of cells. so what beatson did, of course, was removethe estrogen, so it turned the system off, and of course the therapies we use today aredesigned at either removing the sources of estrogen in patients or interfering directlywith the function of the estrogen receptor. but -- so this is molecular medicine in avery real sense. it's targeting the tumor
phenotype. now, beatson, of course, didn'tknow any of this and it took about 70 or 80 years to work out these pathways in the timeafter he did the experiment. but it works because the molecular phenotype and manipulationof that phenotype. the first therapy based on tumor genotypeis also not a new story. this is a therapy that began about -- in the 1980s, when dennisslamon and his group identified amplification of a gene known as her-2/neu. so this is justone piece of data from their paper, but it also shows what is old-style genomics andlow throughput genomics. these are southern blots. these are blots where dna is run ona -- run on a gel and transferred to a filter paper, and then a probe is used that identifiesa specific target, in this case, the gene
known as her-2/neu. now in every cell in anormal individual, there should be two copies of any normal gene, and so all of these shouldlook identical from normal cells. and you can immediately see that these are 70 or sotumors, and there's tremendous variation in the intensity of the signal, meaning the copynumber of genes in these tumors varies, which of course is not normal. and you can see,for example, in tumor number 18 there's a tremendously increased amount of signal. thissignal in lane 3 represents a normal deployed cell. so what you can see is there's tremendousvariation, and in about a third of the cases, he found what was defined as genomic amplification,increased copy numbers of the genes. he also found another striking feature. when he lookedat the prognosis of patients, the patients
with gene amplification, especially high levelsof amplification, did worse than those patients who had normal copy number of her2, and thishas been confirmed in, i think, more than 100 studies since -- in the time since that. now what's her2? her2 is a protein, it's shownin red here. it sits in the membrane of cells; it's a growth factor receptor. and when it'sactivated, it becomes an active tyrosine kinase, it phosphorylates other proteins. and as i'llshow you in a minute, that activates pathways important to the growth and survival of cells.it's a member of the epidermal growth factor receptor family, which has four members: egfreceptor her3 and her4, and a whole host of ligands which bind to these proteins. her2works as a partner with the other members
of the family -- so shown in red you can seeit'll dimerize with either egf receptor her3 or her4 -- and in response to ligands, itturns on pathways that are important for growth or survival of cells, and it leads to a wholebunch of outcomes that make sense when you think about this as a bad prognostic feature.it causes proliferation, it protects cells from cell death, causes invasion, migration,it affects angiogenesis. so all of these features make sense that it's, in fact, a bad prognosticfeature. more importantly, because of this, it's alsobecome a very important target. in the roughly 15 to 20 percent of patients who have amplificationof this protein, we can interfere with the function of this protein, either with antibodieslike trastuzumab or pertuzumab, which bind
to the outside of the cell and block the functionof the protein here, or some more molecular inhibitors, such as lapatinib, which willbind to the protein and inhibit the kinase activity. so in a very real sense, genomicmedicine that's been going on now for quite some number of years. but both of these areexamples of looking at one gene at a time in the patient's tumor. and that's really not where we are today.we still use clinical features, we still use the individual genes, estrogen, and progesteronereceptor, and her-2/neu amplification as measures of prognosis and also measures of what wouldbe the most effective treatments for the patient. but we've also started to incorporate moreglobal genomic measures of tumor prognosis
and predictive measures of therapies for thetreatment of cancers. and i'm going to walk about two today in a little bit of detail:the recurrent score and gene expression microarrays. so let's start with the recurrent score. thiswas developed to stratify the risk of relapse and the need for chemotherapy, in early stagepatients who were hormone receptor positive, node negative, and could be treated with hormonalagents such as tamoxifen. so why did -- why were people interested in doing this in thisgroup? well, we know this is a group of patients who will by and large do well, but some ofthem relapse. and yet over the course of the, say, the '90s and the early part of this century,we started using more and more chemotherapy so that we were treating a larger fractionof these patients who otherwise do well with
chemotherapy, in spite of the fact that onlya few of them really needed treatment. so the question was, could you stratify the riskof relapse and identify the patients who are more likely to benefit from chemotherapy withsome sort of test based on their genome? what they did is they developed a 21-geneset assay by starting -- by culling the literature and microarray experiments, and i'll definewhat a microarray is later. and then they designed a quantitative reverse transcriptionpcr assay from formalin fixed paraffin-embedded tumor tissue. so first, as an aside, i wantto explain what i mean by quantitative reverse transcription pcr, because i think you'llhear about this more and more in the lectures and probably, as well, in the literature.so this is an assay that starts with rna,
in this case, from a tumor sample on a slide,and uses an enzyme known as reverse transcriptase to turn that rna into a copy of the rna knownas cdna, or -- it's a dna copy of the rna. in the next step of this assay, you combineda complementary piece of dna shown in green to the dna that you've just copied, and thatprobe has two molecules attached to it. one is a fluorescent molecule shown as a reporterin r, and the second is a molecular that quenches the fluorescent. so this probe is actuallysilent at this point. but then you do polymerase chain reaction,where you copy the dna into multiple copies. and what happens in the course of that reactionis the dna in this green probe gets degraded and that frees up the reporter molecule fromthe quenching molecule and it becomes fluorescent,
and this can be measured in a fluorometer.as you repeat rounds of pcr, you get more and more freed reporter, and the amount ofthis reporter is proportional to how much starting material you have, and so you canmeasure this actually continuously as you're doing these assays. so another term you'lloften see is real-time pcr, but that's synonymous with quantitative, and you have quantitativereverse transcription pcr because you're starting with rna. so this is a very effective wayof measuring the amount of a protein. so what did they do? they measured, from theliterature search and array experiments they did, they found 21 genes that were usefulin this assay, and these are just the list of the genes. some of them are related toestrogen receptor, the estrogen receptor and
progesterone receptor, and some other targets.these are actually food prognostic features. her2 and a related gene that's amplified withher2f often, grb7 [spelled phonetically]. a number of genes known to predict high proliferationrates in tumors, genes that are known to be involved in invasions, some other genes whichcan't easily be categorized in these categories, and then a number of reference genes justas internal normalizations. and what they do is they perform this assay on all of thesegenes, normalize them to the standards, and then they have an equation that accounts forall of these genes and results in a score. some of them are -- give you negative values,so these are good prognostic features and they lower the score. these are bad prognosticfeatures and they raise the score, as are
these and these. and so what they wind up with is a linear-- not a linear, but a continuous predictor of recurrence. so this is a study where -- oneof the features of the study that was so powerful is they designed this to use qpcr in formalinfixed paraffin-embedded tissue, which means they can immediately go back to large studiesthat had been done over the last 20 or 30 years, take samples from that, and apply thistest to those samples, and validate whether their predictor is a true measure of recurrence.and so this is just the recurrence rate. this is their recurrence score, and you can seein blue that the higher the recurrence score, the higher the likelihood of recurrences,and this is from a randomized study that was
done by nsabp in about -- in the early '90s. now what they did is they then sort of groupedit, they just bend this data into three groups, what they'll call a low-risk group, an intermediate,and a high-risk group, and when you do that what you see is it predicts prognosis. sothis is, again, going back to a study that had looked at patients treated with tamoxifen-- actually, the study asked, does tamoxifen decrease the rate of recurrence. and thatstudy, which is a multi-center study, showed, in fact, that tamoxifen was beneficial inearly stage, node negative, hormone receptor positive patients. so when they looked atthe patients who had received tamoxifen, what they found is those patients that they calledlow-risk, with a low recurrence score, less
than 17, less than or equal to 17, they hada very good outcome. there were patients, though, who had a score above 31 that hada worse outcome, a higher likelihood of relapse, and a group in between. and these are the numbers down here. so abouthalf of the patients in this group that we would normally would say, these are good,this is a good prognostic group, half of them, in fact, have a very low risk at 10 yearsof recurring. only 6 or 7 percent, if treated with hormonal therapy. on the other hand,about a quarter of them have a fairly high risk; 1 in 3 of those patients will recur.so this data becomes very useful for saying, "okay, these patients may not need more therapy,"and we'll come back to that in a minute, "whereas
these patients do." now this allows us to stratify risk beyondwhat would normally be done in the clinic. so again, just to show you two parameters,it's well known that small tumors tend to do better than large tumors when you lookas a clinical staging of patients. but when you look -- and that's borne out by this test,so if you look at the patients who do well, have a low risk, it's higher in the smallertumors than it is in the larger tumors. but what's clear is there are small tumors whohave a high recurrence score, and there are large tumors with a low recurrence score.and that's also been known, that while in general size predicts the likelihood of recurrence,it's not a perfect predictor. so this allows
us to stratify risk further than the clinicalparameter. as another example, grade is, again, a well-used parameter that predicts recurrence,and high-grade tumors are more likely to recur than low-grade tumors. but again, within eachgrade of the tumor, there -- most of the low-risk tumors, in fact, have a low recurrence score,but some of them have a higher recurrence score. and similarly, most of the -- or theplurality of the high -- of the poorly differentiated tumors have a high recurrence score, but stillthere are a significant number that have a low recurrence score. so this allows us tolook at patients and start to stratify their risk further than the clinical parameters. on the other hand, if we look at her-2/neuamplification, the single gene that was -- has
been established now for more than 30 years,what we find here is that virtually all of the patients have a high recurrence score.so in fact, if you have her2 amplification, you probably don't need to do a recurrencescore because you already know that it's going to be high. and this is out of the same setof data, so you can see again, about half of the patients overall have a low risk, aquarter have an intermediate risk, and a quarter have a high risk, but when you look at her2,they're all biased towards the high risk. so in some cases, a single gene may give youas much information as the recurrence score, but in the other cases, we learn a lot aboutwho may need more therapy. now this has also been applied to -- althoughit was developed for node negative tumors,
which are the patients who we clearly areover treating, it's also true for node positive tumors, so this is just showing a study wherethey looked at recurrence score in node negative and node positive tumors. and what you cansee again, in all cases, node negative, one to three lymph nodes, or four positive lymphnodes, that the patients -- there's a continuous relationship between recurrence score andthe likelihood of relapse. what you can also see, though, is that at every score, the nodepositive patients are at a higher risk of recurring, so that when you stratify the patientsagain into these bins of low, intermediate, or high-risk -- this is from the data i justshowed you -- these are the node negative patients. again, low-risk do very well; onlya few percent of them recur, whereas there
are patients where about a quarter of thepatients recur in the high-risk group. you can similarly stratify the node positive patients,although again, notice that all the numbers are lower. so this is a group that has a worseoutcome overall, but even within that group, you can start to stratify better and worsepatients. so this is good; this says we can say, "thispatient needs more than therapy." in some patients, for example, a patient in this group,probably additional therapy is going -- the toxicity will outweigh the benefit. but isthere direct data that says that this test actually can predict the outcomes of chemotherapy?and again, there are. so again, taking advantage of the fact that they could go back to studiesthat had been collected prospectively, and
this was a study designed to test whetherchemotherapy was beneficial in node negative, hormone receptor positive patients beyondtamoxifen. and this is the result of that study, which you can see the blue bar is chemotherapyand tamoxifen, and the yellow bar -- the yellow line is the results of tamoxifen alone. andoverall, there was a significant advantage to adding chemotherapy to this group, butyou can see that overall, the group does pretty well and -- well, the statistical significantdifference, it's not a huge difference. but when you stratify the patients by recurrencescore, you can -- and remember, about half the patients will have a low risk -- you cansee that in the low-risk patients, they do very well and there's no additional benefitto adding chemotherapy to these patients.
in contrast, when you look at the high-riskpatients, what you can see is that a much higher percentage of those patients will relapse;somewhere between 30 and 40 percent of those patients relapse by 10 years, and there'sa tremendous advantage to giving those patients chemotherapy. so most of this difference isdriven by the effects in these patients. there's also an intermediate-risk group, where, atleast from this study, you'd say there's no advantage to giving chemotherapy to this groupas well. now -- so it's clear that you, from this data, that this group doesn't need chemotherapy,this group benefits. i will say that this group is a gray zoneat the current time, and the reason for that is twofold. first of all, remember this isa continuous variable. so if this is a continuous
variable, it means that patients at this end,with a 30, really aren't going to be appreciably different from this group that begins at 31,or a patient at the low end of this group with a score of 18 really isn't going to bethat different from a patient with 17 in this group. so where to draw the line in this groupbecomes problematic. and the second problem with this study is it was based on an olderstudy where the chemotherapy certainly wouldn't be considered state-of-the-art or optimalchemotherapy for a patient today, so that it's not clear that this group really doesn'tbenefit or where to draw the line. and there are ongoing studies asking just that question:if you take a group in this intermediate group of patients in this intermediate category,and randomize them to what would be state-of-the-art
chemotherapy, do they benefit from chemotherapyor not? now a second type of test -- so that's a testbased on pcr, and it's based on -- and can take advantage of formalin fixed paraffin-embeddedtissue. and that's something that's quite beneficial, and i think the penetrance ofthat test into the clinic is, in part, because most of our patients, their samples go intoformalin and get fixed, and so -- but we can still measure that. but another type of testis using a cdna microarray. and the test that's been approved in this country is known asthe mammaprint. similar to the last test, it was developed to predict risk of relapsein early stage patients, although in this case, they were both hormone receptor positiveand negative, and node positive and negative.
and what this relies on is 70 genes from acdna microarray, which started out by looking at, essentially, the entire genome, 25,000genes. before i go into this test, let's talk a little bit about what a cdna microarrayis, though. so microarray technology is a very powerfulay of querying the entire set of genes very rapidly. what you can do is you can, usingtechniques that are very similar to the techniques used to make microchips in a computer, youcan print a known sequence of dna, an oligonucleotide, onto a chip. so that's just shown here asnine spots on this chip. and so these oligonucleotides could, for example, be an oligonucleotidethat would detect each of the genes in your genome. and these chips, the densities ofthese chips at the current time can hold,
i think, hundreds of thousands, if not millions,of spots. so you can essentially query the entire genome, which is approximately 25 or30,000 genes, in one experiment. what you then do is make a probe; in this case, forcdna microarray, you would isolate rna from your sample of interest, in this case, a breasttumor, label it with a fluorescent probe, and hybridize it to this chip. and if themessage exists in the pool of messenger rnas from the sample, it will hybridize and givea signal, and that signal intensity will be proportional to the amount of copies of thegene -- of the rna that are expressed. and then this can be read in a chip reader, whichthen gives a result that, in the raw data, looks like that, which isn't particularlyinterpretable, but that data can then be digitized.
and this is an example from the original mammaprintpaper. so these are 70 genes across the bottom. it's about 300 patients, so each row representsa patient, each column represents a gene. and the way the data is presented is the intensityof the signal compared to a normalization control, shown here in the middle as black,is either lower or higher. so low is green and high expression is red. so very simply,it's, i guess, a molecular rorschach test. if you look at this, these are the patientsthat did well, and these are the patients that had early relapse. and just looking atthis, while it's not important to look at any one spot, you can see that there's a patternthat the patients fall into. so for example, the good patients have low expression of thesegenes and generally high expression of these
genes. and the patients who do poorly, it'sa little more heterogeneous, but again, higher expressions of these genes and lower expressionof these genes. but you can take any patient, then, and simply ask, if we do this microarrayand quantify the data, does the patient cluster with this group statistically or with thisgroup? and when they -- and when they did that, whatthey found is that these signatures predict the outcomes. so the patients with a goodsignature do relatively well, and this is both relapse and overall survival, and thepatients with a poor signature do poorly. and since this test was designed to look atboth er-positive and negative and node positive and negative patients, in fact, the data holdup for both -- in this case, i'm showing node
negative patients or node positive patients,but again, the signature predicts the likelihood of relapse of these patients. so this is,again, a very powerful test. it allows us to stratify patients. we know that some patientsdo well, we know that some patients do poorly; this allows us to start to identify whichpatients are more likely to be in those groups, and therefore, target therapy to the patientsshown in red who are more likely to need it. now this is actually, i think, used more ineurope where it was developed. it was developed in the netherlands by a group that's headedby renã© bernards, and one of the problems with this, although i think that's likelyto be overcome in the near future, is that this, in general, needs fresh frozen tissueto do these sorts of microarrays. in the current
state of the art, that requires that people,at the very beginning, in the operating room, say, "okay, this sample has to go not intoformalin, but into a non-fixative solution." in europe, they're actually managing to dothat in studies that are ongoing. i think in this country it can be done and it is donein many cases and studies, but i think within the near future, my understand is they'llbe able to do these assays out of formalin fixed tissue as well, so that may not be alimiting feature. now how do these tests compare to one another?and this was a study published in the new england journal a few years ago, where a groupcompared the recurrence score to this 70-egne profile, and to two other tests which used,in this case, more than 400 genes. and again,
all of these tests were developed in slightlydifferent -- for slightly different uses. this was developed, as i said, for node negative,hormone receptor positive patients. this was developed, really, for all-come or early stagepatients. this test looked retrospectively at a group of patients. this actually wasn'tdeveloped for cancer at all. it was looking at what happens to fibroblast when you inducea wound, and the response in fibroblast to wounding. but it was recognized that manyof the signatures you saw in wound response looked like signatures you see in cancer cells. the point is that all of these tests identifybetter patients and worse patients in terms of outcome, and when they were compared, itturned out if you took any patient, if it
was predicted to be more likely -- a higherlikelihood of relapse in this -- by this test, all of the tests seemed to behave similarly.they all predicted outcome, more or less -- more or less equivalently. interestingly, verylittle overlap in the genes that were used. there are 25,000 genes in the genome, andyet, there's almost no similarity in the genes that were used in these tests. what was similar,though, were the processes. they all looked at invasion, they all looked at proliferation,they all looked at things that prevent cell death. so while it's not -- it wasn't importantwhat gene was chosen, they all seemed to focus on the same processes. so i'm going to -- so this where we are today.we're already using genomic measures of gene
expression in tumors to decide who shouldand shouldn't get chemotherapy, who needs further treatment and who doesn't need furthertreatment. but where are we going in the near future? so again, treatment will be basedon clinical features of the tumor, and it continues to be based on that. we continueto use estrogen receptor and her2, but i think the measures of multiple gene expression aregoing to become more powerful as we go forward. they're going to allow us to stratify risks.they're also going to start to inform pharmacogenomics, and of course i'm going to talk a little bitabout whole genome sequencing, because i think that's really, ultimately, where a lot ofthis is headed. so this is the -- one of the profiles thatused 400 genes, and it's -- it's a study done
by sorlie et al., from a group that took anumber of tumors where the outcomes were known and did a microarray analysis. they focusedon 400 genes, which allowed them to stratify the patients into several groups as seen herein the color code. and what they found was, first of all, so the luminal patients shownhere in blue, orange, and light blue, are all the hormone receptor positive patients.so they clustered together. so the assay asks, are there genes which associate tumor sampleswith one another? and it doesn't put in the data that these patients were er positiveor her2 positive. but in asking are there genes which can identify subgroups, in fact,what fell out were the -- that er positive patients cluster together. the her2-amplifiedpatients clustered together, as shown in,
i guess, this purple color. and then in thered at the very end, patients we call triple-negative breast cancers that don't express hormonereceptors and don't express her2 -- or have her2 amplification cluster together. so this non-supervised clustering identifiedthe subsets we already knew were important. but we gained more information -- and thathas prognostic significance as the survival curves show. this is probability of survivalin the patients, and you can see that the different groups have different survivals.but there's an important distinction here, or important thing that comes out of thisstudy. when we look at er-positive patients, all of these patients in these first threegroups are er positive. and yet it's known
again that not all er-positive patients respondto hormones, not all er-positive patients do well. and you can see that the group thathe's calling "luminal a" in this study, which are the patients that generally would havehigh levels of hormone receptors expression, they do very well in terms of outcome. butthe patients in these other two groups of hormone receptor positive patents don't doso well. so the power of this is it allows us to, again,look -- and to look at the patients who have hormone receptor, and say, "well, not allof them are equivalent. some of them are going to do well, some of them aren't going to dowell." and really, what we need to be focusing our attention on is why don't these patientsrespond to hormones and do as well as these
patients? again, similarly, just to make apoint, the triple negative patients are known to do poorly, and again, this identified agroup of patients that are predominantly the triple negative. he called them basal characteristics-- we'll come back to that in a second -- and they, in fact, have the worst outcome in thisgraph. this data was generated before the introduction of herceptin, so the purple linehere are the her2 positive patients before the instruction of herceptin. so they do verypoorly in this group. currently, this group would be somewhere up in this range, withthe introduction of herceptin. now what do they mean by luminal and basal?so again, this is a definition where the array -- the gene expression of luminal cells looklike the lumen of the breast duct. so if you
looked at the cell that line the ducts ofthe breast, they would have a gene expression profile similar to these tumors. if you -- thebasal cells are the cells that line the outside of a duct in the breast, and they would havea gene expression profile similar to these tumors. doesn't -- to be careful, it doesn'tmean that these tumors arose from luminal cells. it means they look like luminal cells.the question of where these cells come from and what the source of any of these cancercells are is still an open question. but they look, at the end of the day, like the cellsin the lumen of the breast, and these look like the ones in the basal area of the breast. so focusing on this group for a second, weknow this is a group that does poorly. we
don't have targeted therapies for them asyet. they're triple negative; they can't use hormones because they don't have hormone receptors,we can't use her2 because they don't have her2 amplification. so the question is, canwe use array data to start to get better information about these patients? and this is a recentlypublished paper from jennifer pietenpol's group at vanderbilt, and the answer is yes.so this is that group of triple negative patients, and you might expect that a group definedby the lack of markers is going to be a heterogeneous group. and in fact, it is. she could identifyseven subsets. so again, this is just microarray data, andjust without worrying about -- so looking at this data, these are the genes, each rowis a gene, and going across each column is
a patient. and it's not important what thegenes are, but you can see that they cluster into groups. so for example, the second grouphas high expression of these genes, third group has high expression of these genes.and using what looks to be about 1,000 genes or so, you can identify patients -- that thesepatients fall into one of seven different subsets. now this is all, again, pre-clinicaldata. but as this data was analyzed, it became apparent that some pathways were more importantin one or another of these, and so what they did is they took cell lines -- again, thisis preclinical -- that represented different subsets of these patients, and treated themwith different drugs, and in fact, there's differing response to standard chemotherapeuticagents or targeted agents in the different
subsets of triple negative of cancers. so this is all early in development. whatit means is, it's certainly in terms of hypothesis generating, is that if a triple negative patientfalls into one of these different categories, maybe the therapies that we should be usingin these patients should be different. now of course that's a question that has to betested clinically, but it begins to allow us to ask questions about, are there waysto treat these patients more effectively, and again, not -- it individualizes treatmenteven further than just saying they're triple negative. it's saying they're triple negativein one of these seven categories. i'm sure this is not likely to be the end of the story,and that there are probably going to be other
characterizations of these sorts of patients.but again, it's the power of using multiple genes to dissect the molecular phenotype ofthe tumor. now what i want to do now is talk about anothertopic where i think we're going to impact very profoundly in the foreseeable futurewhich is pharmacogenomics. and again, as a definition, it's using genetic information,either the sequence of the genes or the expression of those genes, the genotype or phenotype,to predict efficacy or toxicity. and again, i just want to remind you that in a tumor,there are two genomes: there's the tumor genome and the patient genome. so let's focus firston the tumor genome and pharmacogenomics in terms of that.
so here, it's the presence of a therapeutictarget predicts the treatment benefit. well, we already know this, but this is, in fact,pharmacogenomics, although it wasn't necessarily thought of in those terms. if a patient hasthe estrogen receptor, we use hormonal agents. if they have her2 amplification, we use her2-targetedtherapies. so in many respects, the expression of specific genes in the tumors predicts efficacy.and in fact, the absence of these markers predict the lack of efficacy in those tumors,of these agents. so this is something that's already being used in the clinic. as a second example, i'm going to go backto something that larry brody talked about last month when he discussed the brca1 and2 mutations. so this is a case where these
aren't necessarily the targets, but they potentiallypredict what therapeutic intervention may be beneficial. so if you remember his talk,the brca1 and 2 genes, a large part of their role in a cell is to help repair dna damageof a specific type, and those are double-stranded dna breaks, caused by things like ionizingradiation, or other genotoxic agents. and this is the predominant mechanism those breaksare repaired by in a normal cell, and it depends on having a normal copy of brca1 or 2 around.there are other ways to repair this dna damage, it's just they are less efficient and morelikely to cause errors. but in a patient who has one defective copyof brca1, this pathway still works, because all you need is a copy of brca1 or brca2 towork. but in the tumor, you've loss both copies.
so in the tumor cells from a patient thathas brca mutation, they have no functional homologues recombination because this pathwayhas lost two -- or one or two of the critical components. and so this pathway becomes muchmore important to the repair of this dna. and larry, at the end of his talk, showedthe results of one study, but the idea is, if you take, then, a tumor that's dependenton this and interfere with this pathway, the tumor cells should be very susceptible toany kind of dna damage, either spontaneous or induced, whereas normal cells, which stillhave a functioning copy of brca1 or 2, should actually be relatively spared. and, in fact, that's the idea between -- behinda new generation of drugs known as parp inhibitors.
parp is one of the enzymes involved in thesealternative repair pathways. if you inhibit this pathway, the tumor cell will be moresensitive and will die. and, in fact, in early phase studies where these have been used inbrca1 or 2 mutant breast and ovarian cancer patients as a single agent has very high responserates, and so this is actually very promising. and then if you think about coupling thiswith dna damaging agents, it's likely that this is a therapy that will be effective.so why is this pharmacogenomics? it's because the mutations in the tumor are predictingwhat's likely to be effective therapy in these patients. now turning to patient pharmacogenomics, andthis is a different topic, and again, it's
a very broad topic. i know people like dougfigg give a whole lecture on this alone. but the idea is that not only does the genotypeof the tumor matter, or the phenotype of the tumor, the genotype of all of us matter interms of a response to drugs. and in this case, the presence of genotypic or phenotypicmarkers in any individual patient can predict, again, the drug toxicity or efficacy. nowremember, these are normal -- all of these are the normal genome, so all of us in thisroom, if we were to sequence our dna, would have differences in many different genes,typically or most commonly single nucleotide changes in those genes. but before we talkabout that, let me just talk about how this can be used -- is already being used in theclinic.
so when we think -- we don't always thinkof it in terms of this, but in a patient who's being treated with hormonal agents for aner positive breast cancer, we think about whether the patient is pre or postmenopausal,because the sources of estrogen are different in a pre and postmenopausal patient. in apremenopausal patient, the predominant source of estrogen is the ovarian -- hypothalamicpituitary ovarian access, whereas in a postmenopausal patient, the estrogens that are produced areproduced by converting adrenal androgens through an enzyme known as aromatase, into estrogen.so this is pharmacogenomics. in a premenopausal woman, her phenotype is such that we can targetovarian estrogens. this is what beatson did over 100 years ago. he removed the ovaries,and so we can continue to do that, either
with oophorectomy, or we can do use drugsthat turn off the ovary. on the other hand, this would do no good,of course, in a postmenopausal woman, and conversely, in a postmenopausal woman, weuse inhibitors of this enzyme, aromatase, to block the production of estrogens by theconversion of adrenal androgens. and again, just as aromatase inhibitors won't work forpremenopausal women, targeting the ovary doesn't work for postmenopausal women. and finally,if we target directly the estrogen receptor, that can be done in either pre or postmenopausalwomen. but again, this is looking at the patient's phenotype. this has nothing to do with thetumor itself to decide what is the best treatment for this patient.
the other more classic way of thinking ofpharmacogenomics is metabolic enzymes that may affect it, for example, the cytochromep450 enzymes, and again, this is what i was referring to. most commonly, these are singlepoint changes in the bases. they're not mutations so much as they're a variation between anytwo people in the population. and these can be measured, so i'm going to come back toa little bit of technology. just as we did microarrays to look at expression of genes,you can imagine printing a microarray, but now these spots don't represent probes forindividual genes. they could be probes for very -- single nucleotide polymorphisms inthe same gene. so all of these nine spots are probes for the same gene, but shown downbelow, the gene has sequence variations from
individual to individual. again, these are normal people, these arenormal genes. but these sequence variations can affect the activity of metabolic enzymes.and again, using these arrays, you can measure, again, hundreds of thousands, if not millions,of these in one setting. so if you knew a particular cytochrome that was important,you could just sequence that. but as we go forward, what can be done as an array thatwill allow us to look at the many polymorphisms, and it's conceivable in the future that thiswill be something done on most patients, because it's not just cancer drugs, but drugs foralmost any disease we treat will be affected by metabolic enzymes. and you can imaginehaving a profile and saying, "in this patient,
we need to be careful with this drug." so the best example i know of this, thereare no good examples of single nucleotide polymorphisms in breast cancer as yet. butthere's a drug called 6-mercaptopurine used predominantly in pediatric tumors, and thereare good metabolizers and bad metabolizers of that drug, and if you carry a polymorphismthat makes you a bad metabolizer, you're much more likely to have toxicity and the dosingof the drug is affected. and there are a number of drugs where there's true for at this point.so this is something that will affect, again, our decision about what drugs are the bestdrug for that patient, will there be undue toxicity? in some cases, drugs needs to beactivated by the metabolic enzymes, and is
this a patient where the drug will or won'tbe activated. and so i'd like to end, then, with what issomething that, again, is being done already, but not really in a clinical setting, andthat's whole genome sequencing. and this schematic from greg feero's recent new england journalpaper that summarizes molecular techniques in genomic medicine, it talks about the differencebetween sequencing, what i guess is traditional sequencing, and the next-generation sequencing.so this is just a schematic, but in traditional sequencing, you clone a piece of dna and thenyou sequence it, and this can be done in a reasonable throughput fashion so that youcan sequence maybe 100 copies of the gene at a -- 100 different pieces of dna at a time,and you get about 100,000 bases of sequence.
and it would take, probably, if you had areally good person doing this, probably a week or two to do this, and it would take30,000 experiments like this to sequence the genome once. so this is, you know, not a practicalway to sequence the genome. but the current sequencers don't rely on theold methods; they rely on either solid phase or fluid phase methods that sequence not 100fragments at a time, but millions of fragments. they do relatively short reads on those fragments,but you generate about 30 gigabases -- i'm sorry, 100 gigabases of data. so if you thinkabout it, the human genome is three billion bases. this is covering the human genome 30times over in one experiment. and practically speaking, i think this data can be acquiredprobably in a week or less, from the time
you have a dna sample to the time you havethe data at the end of the day. so this would allow you to acquire the sequence on many,many, many genes on the entire genome of the tumor or the patient. what you've probably seen in the literatureis the $1,000 genome. the cost of this is rapidly falling to the point where it willcost about $1,000 to do this. we're not at a point where we can use this data, so theanalogy larry brody actually gave me, it's a little bit like a ct scan. ct scan's uselesswithout a radiologist to read the ct scan and tell you what you're looking at. so thisdata can be generated very quickly. the problem right now is now you have 100 gigabytes ofdata, or a huge amount of data. it's interpreting
that data, really, that takes a long time. but just to show you some of the things thathave been done with it, and again, this is all experimental and not clinical. this isfrom a paper published by bert vogelstein's group in science a few years ago, where theytook 11 breast cancer -- and actually, 11 colon cancer -- cell lines and sequenced the-- all of the coding sequence in that, so the sequence that turns into messenger rna,not the entire genome. what -- this is a graphical representation. they're chromosome 1 throughchromosome x, so all of the human chromosomes are arrayed this way, and the little dotsand peaks represent mutations seen -- this is just the breast cancer sample -- in thesamples. and everywhere you see a dot, there's
-- that means there was a mutation in thatgene, and if there's a little purple hill it means there was more than one -- more thanone sample had a mutation. and what immediately falls out are severalthings. first of all, there are two really high peaks in this. this is p53, so out ofthe 11 samples, something like seven or eight of them had mutations in p53. so that's nothingthat we didn't already know; p53 is one of the most frequently mutated genes in the humangenome. this is another gene known as pi 3-kinase. it's a kinase that's involved in lipid metabolism,but it's also very important in signaling in cells, especially towards survival pathways.and again, that's a peak that came out from this data.
but there's another thing that comes out ofthis data. this is 11 samples; this would be the equivalent of sequencing 11 patients,and you can immediately see there's something like 10 or 20 mutations, on average, per patient.so there are a lot of mutations, and one of the difficulties, then, in deconvoluting thisdata, is, well, which of these are really meaningful, and which of these are noise?so what -- for vogelstein, which are drivers of the tumor phenotype and which are passengers,are the terms he's applied to this. but we can hone in on the pi 3-kinase examplewith a little more detail. so out of the 11 samples, this is that peak. this is showingboth the breast and the colon; the blue is the breast. so out of 11 samples, half ofthem, five of them, had mutations in this
particular gene, in this protein, and it wouldturn the protein on, and that would actually drive the proliferation and survival of thecells. but this schematic is the pi 3-kinase pathway, and so we really have to think notin terms of genes, also, but in terms of pathways and in terms of more system-type analysesof these tumors. because what the rest of the little circles, and blue and red for colon,breast and colon, respectively, show is that this pathway is targeted in more than -- morethan just hitting this particular protein, so that you see that their mutations throughoutthis pathway and most of them, in that effect, is to turn this pathway on. so a lot of thelittle hills you saw still target this pathway. so actually, if you-- if instead of showingindividual genes you said, well, which of
these hills represent this pathway, that wouldhave been a very tall peak. now why is this useful information? again,not ready for routine use in the clinic, but it turns out that there are data that supportthat activation of this pathway make a patient resistant to hormonal therapies or to her2-targetedtherapies. so that means that you might predict these patients won't do as well with herceptin,for example, or with -- or tamoxifen or an aromatase inhibitor, and importantly, thatthese patients may benefit from combinations of therapy that block this pathway and thentarget the more traditional pathways. so this is very useful information. the problem withthe genome data right now is culling the data down to meaningful clinical information, andthat's really, probably, the long part now;
acquiring the data is very rapid. so with that, i'll end with this summary slide.the past was really looking at tumor characteristics and one or a few genes. we currently are alreadylooking at genomic -- using genomic medicine. we're looking at arrays that look at anywherefrom 20 to 70 genes in breast cancer. but the future really is going to be to expressionof hundreds of genes at a time, and also sequencing and snp arrays to decide what are the bestchoices of drugs for a given patient, for a given tumor, who needs treatment, who doesn'tneed treatment. i'll stop there and take questions. [applause] go ahead.
male speaker:what happens to the tumor cell that survives chemotherapy [inaudible]? stan lipkowitz:you know, so chemotherapy is mutagenic, and so people have looked at that. clearly, there-- you know, i can't give you a specific example, but clearly, if you look at microarray dataof resistant tumors, it's different. if you look at genomic data, there are acquired mutations.there have been data -- i don't remember if the patient was treated or not, but for example,people have done whole genome or exome sequencing on a lobular cancer, both from the originalcancer and from metastasis of that cancer. clearly, there are mutations that are thereat the get-go that also show up in the metastatic
disease, but there are new mutations in themetastatic disease, and you would imagine that therapies -- whether it's going to dotwo things. therapy is genotoxic; most of the chemotherapy we do causes genetic damage.radiation is genotoxic. but also, we're going to select -- it's just like bacterial resistance-- we're going to select for preexisting clones that may not have been detectable that havea mutation. so it's hard to know if the therapy also did it, but it probably causes that.go ahead. female speaker:yeah, excellent talk. i really enjoyed it. i have a question: i recently came acrosssomeone mentioning that p10 mutations are beginning to show up in a lot of breast cancerpatients, and somehow, i didn't think that
that was associated with it. if you couldcomment on that. stan lipkowitz:so if you look at the original p10 patients, it's not as -- female speaker:will you repeat the question? stan lipkowitz:okay. so the question was, what about p10 mutations in breast cancer. and i'm tryingto remember if they showed up in this slide. yeah, so two things. first of all, if you'lllook at the regional paper identifying p10. p10 is a tumor suppressor gene; it's a negativeregulator of this same pathway. p10 is down here. it turns off the pi 3-kinase pathway.it's a phosphatase that removes the phosphate
that pi 3-kinase puts on. these are lipidphosporylation events, not protein. and if you look at the original cloning paper, itactually was found as a progression feature in both glioblastoma, but breast cancer. sothe original identification of p10 said it would occur in breast cancer, but not thatcommonly. and in bert vogelstein's sequence, two out of the 11 samples had p10 mutations.so i think that's certainly likely to be true. these are inactivating mutations, but allof these data say, well, inhibiting this pathway in one shape or form might be a good idea.obviously, the genes that are inactivated are hard to think about as targeting becausetheir activity is lost. but i think that is going to be something -- it's not going tobe a common mutation, but really, the theme
here is not necessarily that p10 is common,but the pi 3-kinase pathway mutations are probably more common than we know. in thisset of tumor cell lines, a very biased set, it was half of the cells that had a pi 3-kinasemutation. go ahead. male speaker:what is the mortality rate of breast cancer at this time, as courtesy to 1960? what isthe mortality rate of breast cancer? stan lipkowitz:so, two things. so first of all, the overall mortality rate of breast cancer. so the wayi look at that is -- so the question is, what's the mortality rate of breast cancer now comparedto 1960? i have to confess, i'm not really the epidemiologist in the audience, but ingeneral, it's lower now than it was then.
it you look, there has -- there was a relativelysteady mortality of breast cancer, and one way to look at it very simply is there areabout 200,000 cases of breast cancer every year, and about 40,000 deaths. so the mortalitydue to breast cancer is about 20 percent of all-comers. and that, of course, varies basedon what your stage and presentation is, and so forth, and your molecular features. andthat was actually very constant -- i don't think t increased over the '60s or '70s, andthe primary driver of that -- of the rate was that if you detected it early and didsurgery, those patients did well. but it didn't change all that much. starting at about 1990, however, the ratehas been decreasing a few percent, so i think
that the rate of mortality now is probably2 or 3 percent lower than it was in 1990. the -- so it's not a dramatic effect, butit's clearly going down. and the decrease has been attributed to two things: one isincreased use of screening, and the other half is the increased use of advanced chemotherapyin more patients. so it's thought to be both the -- so both of them targeting early stagepatients. to my knowledge, the mortality for metastaticbreast cancer hasn't changed. essentially, except for anecdotal data, we don't cure patientswho have metastatic breast cancer. so all of the decrease in mortality can be attributedto finding it earlier, and also treating the early stages more aggressively to deal withmicrometastatic disease that's spread.
male speaker:and mammograms? how about mammograms? don't they -- stan lipkowitz:yeah, so about half of that decrease is attributed to screening due to mammograms and -- i guessgiven the date of onset, it's really not mri or more advanced screening techniques, butscreening in general has thought to decrease the incidence. the estimate, again, theseare epidemiologic estimates. it was a new england journal article -- don berry was,i think, one of the main authors -- that estimated that that decrease was primarily 50 percentdue to screening and about 50 percent due to more aggressive therapy. the other thingthat's made a dramatic difference in the incidence
of breast cancer in the last few years, andi think larry brody showed this slide, was when the women's health studies showed thathormone replacement therapy not only didn't decrease heart disease, but increased breastcancer. people very rapidly stopped using hormone replacement therapy, and within twoor three years, there was a fall in the incidence of breast cancer, which also will drive afall in the mortality. but really, since 1990, there's been a steady, slow by steady, decline. male speaker:to what extent can this level of scrutiny used to look at the non-cancerous breast tissue,in an effort to predict what factors generate [inaudible] to begin with?
stan lipkowitz:so the question is, looking at, i guess -- or the breast prior to the malignancy, or -- ithink there's a lot of work on that. it's clear from work of people like mina bissellthat the tumor really arises, it's an organ, and it arises in an environment, and the environmenthas a major impact. and she's, for example, shown data that you could take tumor cells,and if you put them into one environment, they form a tumor, and if you put them intoanother environment, the tumor is suppressed. so it's clear that the environment has animpact. it's clear that the genotype has an impact,so if you carry a brca1 or 2 mutation, your chances of developing breast cancer over thecourse of your life are manifold higher. the
average -- the number that's quoted is roughly1 in 9 women will develop breast cancer; if you carry a brca mutation, the penetranceis anywhere from 40 to 80 percent, so much higher risk. so that's telling you the genotypeof the non-malignant cells matters. in terms of what other features matter, weknow that there are environmental factors that can increase your risk of breast cancer.there are people who are looking at this. i don't know of any data that i could citeto you that says, okay, this is something that's going to predict someone's risk, butthat's something people would like to do. can you, ideally using a non-invasive method,look at a person's tissue in the breast before they ever develop cancer and predict theirrisk? people are doing that. they're doing
ductal lavage, people are doing biopsies onhigher-risk patients to see if they can identify markers. i don't think anything's ready foruse, aside from the normal risk factors, brca mutation, family history, but those are beinglooked at. male speaker:clearly, [inaudible] in the presence of non-malignant disease, either prophylactic or lasting surgery,and that would provide a population of patients who are, almost theoretically, non-malignant. stan lipkowitz:right, and i think people are doing that. so people are looking at the breast -- it's-- the question is, do you have enough natural history in any given group of biopsies, doyou have samples from patients? so if you've
had -- the better would be breast reductionsurgery where you get samples, because then the patient still has a breast, and if theydevelop breast cancer, you can say, these people had breast reduction and didn't developbreast cancer. and that's being done. i don't know of any data that says there is somethingwe could really use for that today, but those are ongoing studies that i'm sure that arebeing done in a lot of place. go ahead. male speaker:have gene expression microassays been done on men with breast cancer? stan lipkowitz:good question. so my -- my knee-jerk would be i'm sure it's been done. i don't know -- ihaven't seen any data that specifically look
at that. we treat men with breast cancer likewe treat women because we don't really have a lot of data to guide us. there are cleardata that things like tamoxifen are beneficial. there's a lot of controversy, what about aromatase?isn't a guy just like a postmenopausal woman, and the answer is, well, we don't know, sowe don't know what to do. so that's an interesting question. i haven't seen a publication ofthat sort. i can't imagine it's not been done. obviously, the cases are rare. i should saythat men with breast cancer, depending on how you look at it, it's one -- it about 1percent of all breast cancer, so since breast cancer is so prevalent, it's not -- it's uncommon,but it's not incredibly rare. female speaker:i'll just make this the last question, please?
stan lipkowitz:go ahead. male speaker:following up on the previous question with the tumor microenvironment, can you commenton how the stroma, which is immediately invasive to the tumor, how that can be sort of markedupon as an expression [unintelligible]? do you understand how the stroma is affectingthe tumor? stan lipkowitz:so a lot of -- there's a lot of research going on on just that, and it's -- i'm sorry, howdoes the stroma impact the tumor. so first of all, let me just make a point. most ofthe expression analysis that's done is done on the tumor, which means tumor cell and stroma,because tumor's an organ. and that was a decision
made, for example, the group at stanford whoreally started and blazed the way on this, joe gray's group, they made that decisionreally out of practicality. it was easier just to grind up the whole tumor than it isto try to microdissect the tumor from the microarray. and their feeling was you wouldget information from this. and in fact, you do, because when you look through the data,the best example that comes to immediate mind, in lymphoma, there is prognostic informationthat comes from looking at the whole tumor, but some of the prognostic information isan immune signature of the response to the tumor and not the tumor cells themselves.so -- and that's true in breast, as well. so you're getting information.
but people are microdissecting, they're lookingat stroma versus normal. there are clearly data that suggest the stroma is differentin tumors. there are data that go back a long time, probably 10 or 15 years. in prostatecancer, if you take fibroblasts from patients with prostate cancer versus patients without,they support the growth of prostate cancer cells very differently. the normal don't andthe tumor fibroblasts do. so people are looking at that because that becomes another target.and the intriguing part of that is, the normal cells, their genomes are more stable. so ifyou start to target the normal cells, maybe you won't have as much problem with developmentof resistance that you do in the tumors. but that's an active area of research. they'rea lot of people looking at the stroma.
female speaker:okay, and with that, i would like to thank dr. lipkowitz --
