Cancer Of The Lymph Nodes

female speaker: dr. david solit. i have to start witha personal note. he's part of my family. and so i've always beenexcited about david and who he is as a person,and i'm excited about his work as well. he's my nephew. but more importantlythan that, he's the director at the kraviscenter for molecular oncology

at the memorial sloan ketteringcancer center in new york city. he's both a researcherand clinician, and works on allareas of cancer. and today he's going to talkabout an aspect of his research that i find fascinatingwhen i understand it. and so thank you so much david,for coming and talking to us. david b. solit: thanks forthe invitation to present. i am out here this weekendriding in a cancer fundraiser called cycle for survival.

so since i was in town, iguess i had the opportunity to create thisgoogle youtube stream and speak to you guystoday about what we've been doing at sloan kettering. so really what i want totalk to you about today is how we're trying to usea big data computational approach to identifynew drug targets, and how we're actuallystudying what we call extraordinary responders.

these are patients whoreally beat the odds. these are people who have acancer that most patients die of, and oftentimes very quickly. but these are thoserare individuals who overcame those odds andare alive many years later. and what we try to do is analyzetumors from those patients so that we can learnlessons from that and hopefully identifynew drug targets that we can apply in thebroader population of patients.

so i've been at sloankettering now it's 18 years. and i run both the research lab. i also see patients asa medical oncologist. and really what the focusof my lab's efforts is is to really try toidentify these drug targets. and so we try to acceleratethis target discovery. and it's a pretty simpleparadigm that we follow. it's really up here onthis first slide here. we try to define targets.

and i'm going to spenda few minutes talking about what these targetsare and how we find them. we then, in thelaboratory, try to identify a drug that selectivelywill inhibit that altered target that'smaking the cancer cell grow or something downstreamfrom that drug target that'll kill the cancer cell. and then what is amajor challenge-- and i'll talk to you aboutthat for the second part

of the talk-- ishow do we identify the individual patients whohave an alteration in their drug target so that we can givethem a particular drug? and that's reallywhat we try to do. so what are these targets? there's really two main waysto identify these targets. one's called a genotypeto phenotype approach. and then i'll mention to you theopposite, which is a phenotype to genotype approach.

so the genotype to phenotypeapproach-- or g to p-- is where we take a largegroup of tumor samples and then we retrospectively domolecular profiling on them. and this is usually what'scalled next-gen sequencing at this point. there are threebillion base pairs of dna within the human genome. and what we tryto do is sequence that genome in theseindividual tumors

and then use astatistical approach to find things that are commonlyaltered in multiple patients with the same cancer. so if you see the samemutation again and again in a particularcancer type, that gives you a statistical cluethat that mutation maybe is important for thegrowth of that cancer. and so this was really the paperthat really changed my career. i was what was called apostdoctoral fellow working

in the laboratoryof someone else. i hadn't gottenmy own laboratory when this paper cameout in a journal called "nature," which isone of the big biology journals in the field. and it was a very simpleexperiment that they did. what they did is they took1,000 cancer samples-- some tumors, some celllines that were tumors that they now hadgrowing on plastic--

and they sequenced acommon pathway that's activated in cancer calledthe map kinase pathway. and then what theywere looking for was genes that hadn't been knownto be mutated that were mutated in these tumors commonly. and so when theysequenced that pathway, they found thisgene called braf. and this was interestingbecause at the time, nobody had studied braf.

there's actually threedifferent types of raf proteins. there's araf, there'sbraf, there's craf. craf is also calledraf1, and it's the one that everyone had beenbiologically focusing on. and it turned out that peoplewere really missing the boat. and this non-biased approach,by just sequencing 1,000 tumors, identified that braf mutationswere found in about 80 of those, so in about8% of all those tumors. and that was about50% of melanomas,

but also scatteredthroughout other tumor types. what was notable aboutthese braf mutations is that about 95% of themwere the exact same change in the dna. so it was a substitutionof one amino nucleotide for another nucleotide,which causes a single base pair aminoacid change in this protein. and when you do that, youactually activate that protein constitutively.

so that's why thesepeople get cancer. they have an abnormal proteinthat's always turned on. and when you have asituation like that, if you can develop an inhibitorof that activated enzyme, you oftentimes havea new cancer drug. and that's what we testedin the lab after that. so what we did is we took apanel of cancer cell lines. we looked to see which ones hadbraf mutations, which ones had a mutation upstream calledras that had already

been discovered afew years prior, and then whichones had activation of this, what's calledmap kinase pathway, but due to upstream activation. and what we foundin the laboratory was that these tumorsthat had braf mutations were selectivelydependant upon braf. because if you have a mutation,and the cancer cell is not dependant upon that mutation,then developing a drug

that targets thatmutation's probably not going to have too muchof a dramatic effect. if you actually lookat the chronology here, this experimentwas done in 2002. we did this biologicexperiment in 2006. and then it tookabout til 2010 til we had a good drug that wecould actually test in patients. and this is just some earlydata from the phase one trial. that's the first in human studyof a drug that selectively

inhibits raf. and this is a patient withvery advanced melanoma. all of these dark dots hereon this pet scan are cancer. and you can see 15days later, after being on this selectiveinhibitor of braf, this patient's had a completeresponse on our pet scan. and this is a cat scanshowing regression of those tumors inthis patient's lung. and ultimately ina randomized study,

we were able toshow that this drug, if you give it topatients with melanoma who have this braf mutation, thatthose patients live longer. and actually we were ableto show in the laboratory that if you give thissame drug to cells that lack the braf mutation,those cells actually grow faster. so it really becomescritical that for each individual patient,before we treat them,

we know whether theyhave the braf mutation or whether they don'thave the braf mutation. because if we give themthe braf inhibitor, they're likely to benefit ifthey have a braf mutation. if we give themthe braf inhibitor and they're brafwild type tumor, their tumor mayactually grow faster and they may die quicker. so really what the challenge hasbecome somewhat in our cancer

population is howdo we figure out who's got what mutationswithin their tumor? because again,these mutations can be scattered over thisentire human genome, which makes up about three billiondifferent base pairs. so that's one way toapproach this problem. that way is really astatistical approach. it takes large numbers. it takes a large consortium.

and there's an effort calledthe tumor cancer genome atlas that's trying to dothat across the country. and that's one way to do it. and we actually came upwith the opposite idea. can we actually take patientswho were on cancer drugs and they responded incrediblywell to those cancer drugs-- most patients didn't, or maybenobody else didn't-- and can we find out what was the mutationthat made them respond? because if we couldfigure that out,

we already have a gooddrug because it worked in that individual already. and if we could just findother patients just like them, we have a drug readyfor those individuals. so could we expandout the number of patients who would benefit? and the real inspirationfor this work was this clinical trial thatwe ran in sloan kettering a few years ago.

so this was a trialwe ran back in 2008. we did this trial inpatients with bladder cancer. and like every otherbladder cancer trial we've done over thepast 20 to 30 years, this was a negative study. so we enrolled 45 patientson to this clinical trial. only two of thosepatients responded. and based upon that result,statistically one would say ok, this drug which wewere testing here

called everolimus-- it's madeby a company called novartis-- that this drug doesn't work inpatients with bladder cancer. so let's maybe moveon to the next idea. but despite the fact thatin the entire population it wasn't impressive,this patient here-- and i'm showingher cat scans-- really jumped out at us. and so just to give youher history briefly, her history actuallybegins off the slide here.

in the summer of2009, she presented with a bladder cancerthat had already spread to her lymph nodes. so it was already whatwe call metastatic. and what we try to do inthese patients is we give them chemotherapy, we try toshrink the tumor down, we try to cuteverything out, and then in a percentage of thosepatients we cure the cancer. but unfortunately,by the fall of 2009,

her cancer had come back, it hadspread back to the lymph nodes. and you can see here in january2010, she had this cat scan. and really this area here thatthe red arrows are pointing at, that's her tumor. and that spread outthrough her abdomen. and actually in this situation,a patient with bladder cancer who's had a priorline of chemotherapy, we have no standard of care. and the average survivalof these patients

is only nine months. and so we reallyencourage these patients to go on clinical trials. and for no particularreason other than it was an open clinical trial atsloan kettering at the time, this woman enrolled onthe everolimus study. and despite the fact that thisdrug really didn't work well in anybody else,by april 2010 she had achieved a nearcomplete response.

by july 2010, acomplete response. what i mean by that is thatall the tumors had completely resolved. if you look at july 2011,still in a complete response. '12, '13, '14. she was actually just back inour clinic a few weeks ago. and she's now at herfive year anniversary being on this compound withoutany evidence of disease, despite the factthat this again,

in the overall population,was a negative study. in fact, every otherpatient on this study has died of their disease. she's the only one who's alive. and not only is she alive, hercancer's completely gone away. so again, what makes her unique? so at the time, back in2008, 2009, the technology that we had available reallyallowed us to look at one gene at a time for mutations.

and that was pretty limiting. we could have a hypothesisbased upon the biology around everolimus thatmaybe this gene or that gene was important for this response. so we did that type of approach. we looked at a few genesand a candidate approach. couldn't find what really wasunique about this patient. but what happened along the wayis that technology improved. and this was both what wecall sequencing technology

and then also just thecomputational power, our ability to analyze thisdata expanded to the point where we could nowdo whole genomes. and so actually, the firstwhole genome sequence of a normal cellwas not completed until about the year 2000. the first cancer genomes wereonly about five years ago. and those firstcancer genomes were costing on an average hundredsof millions of dollars

to complete in their entirety. the technology eventuallyprogressed to the point where, for just a small amountof money at the time, for this pairedgenome for sequencing both their tumor and theirnormal cost about $20,000. that's already downto about $2,000 today, just a few years later. but again, by being able tosequence her whole genome and being able toanalyze all that data,

we're able to figure out everysingle place in her tumor genome where there'sa mutation and it's different thanher normal genome. and if you do thatanalysis, you come up with 17,000 mutationsthat are in her tumor that are not in her normal. and you say well, howare you going to sort through 17,000 data points? but it's a realtestament to how much

we've learned about thebiology of these genes over the past severaldecades that we're able to very quicklybioinformatically map every single one ofthese 17,000 mutations to what we call thereference genome. we can ask, whichof these mutations are present in a codingregion of a gene? which of thesemutations would be predicted, basedupon an algorithm,

to change the protein function? and which of these genes thatchange a protein function when they have a mutation wouldbe predicted to be associated with everolimus response? and when you dothat, you can take all 17,000 of these mutationsand actually boil it down to just two-- one called tse1and the other called nf2. and the reason thesewere so interesting is that when you mutatedboth tse1 and tse2,

you activated aprotein torc-1, which is the direct targetof everolimus. so if you look at thispatient in retrospect, she was the perfectpatient to get this drug. we simply didn't know itgoing into the clinical trial, because we didn'tknow what genes were mutated in her tumor. actually, at about thetime-- i'll just go back-- that we were gettingthese results back,

there was actuallya clinical trial published that this samedrug everolimus works in patients who have a conditioncalled tuberous sclerosis. this is an inherited diseasewhere patients get cancer at a very young age dueto germ line mutations in the same gene, tse1. and so these are people whoare born with already one defective copy of this gene. if they lose the othercopy in a normal cell,

they then get tumors. and because normally you haveto lose both copies of a gene, if it's a tumorsuppressor, to get a tumor, these people are verysusceptible to getting cancer and they get cancerat a very young age. so we've now taken thisextraordinary responder paradigm and we'veapplied it to other cases. and i just want to mentionone other quickly here today. this is a second patient whowas a young woman in her 40s.

she presented with amass in her abdomen. it was a localized tumor, sothe surgeons opened her up. they went ahead andcut out this tumor. and if you look at thistumor under the microscope, this patient has what'scalled a small cell cancer of the urothelial tract. and this type of tumor has aparticularly grim prognosis. so if you look at our seriesof this type of cancer at sloan ketteringover the past 10 years,

essentially all patientswith this disease die of their disease. and so given that really grimprognosis, the doctors who were treating her wentahead and gave her some chemotherapy,the [inaudible], even though she didn'thave any cancer. and we call thisadjuvant therapy. it's the type ofchemotherapy you give try to prevent thecancer from coming back.

we do this very commonly inbreast cancer and in colon cancer. but unfortunately, despitegetting this chemotherapy, just a few months later,her cancer came back. it had now metastasizedto the kidney. she now had metastases,and so spread to the lymph nodes in the abdomen. and so what they went aheadand did is they actually went, took her back to surgery.

they cut all thiscancer out again. usually that doesn't work,but they were really desperate and had really no other options. and unfortunately, just afew months later, the cancer had come back againin the abdomen and had now spread to the bone. and so this was reallynot a good story. this is one that probablywas not going to end well. and so what we recommendedis that we just

don't have good standardtreatments for you. we referred her fora clinical trial, and they referred her towhat's called a phase one clinical trial. this is really the first[? in-man ?] human study where we're not even sure what'sthe side effect of the drugs, what's the right dose to give. and so she ultimately again,for no particular reason other than it was anopen study at the time,

was enrolled on the study of achemotherapy called irinotecan, and a second drugcalled azd7762, which is an inhibitor ofa kinase-- or an enzyme, you could say--within the cell that's important for dnarepair and checkpoint to allow for dna repair. and the hypothesiswith this trial was that the cancer cells wouldbe more susceptible to dna damage from the chemotherapy.

this kinase inhibitor wouldprevent this dna repair from happening in thecancer cells selectively, and that would thus enhance theeffects of the chemotherapy. and it was a good idea. there was some evidencein the laboratory that this could work. but unfortunately, theysaw a lot of side effects with this combination. and they really weren'tseeing a lot of activities.

so midway through treatmenton this clinical trial, the company astrazenecaactually discontinued the entire program. so they stopped thedevelopment of this azd7762. and that was despite the factthat this patient had actually achieved a complete response. so all her cancer had gone away. but she was reallythe only one where we saw that type of response.

and so given the fact that theyhad discontinued the program, she had to stop taking the drug. they continued the chemotherapyfor a few more months, eventually had to stopthat to give her a break. they were expecting the cancerwould come right back once they stopped the treatment. and that was aboutfour years ago. her cancer's nevercome back and she's now probably cured of her cancer.

so again, this is a womanwho was cured of a cancer that everybody diesof on a drug that was so disappointingin the clinic that the company discontinued itduring the first clinical trial that was being done. so we went ahead and wesequenced this woman's entire tumor genome. and this is calleda circos plot. it's a way for us justto represent graphically

the complexity ofher tumor genome. we just take the chromosomes,we line them up in a circle. these line show translocations,so a breakage of one chromosome where it's recombinedinto another. this inner circle shows copynumber alterations, so gains or losses of particular genes. and then this woman actuallyhad 19,000 somatic mutations. these are alterationsthat are in the tumor that are not present in the normal.

so again, how do we sortthrough all this information? just to give you somehints of how we do that. one thing we do is we canactually align these genes to the homologues of thesegenes in other species. so we can take particular genesthat are mutated informatically and align them to thesame gene in a mouse, or the same gene ina drosophila fly, or the same gene in a yeast,or the same gene in a bacteria. and when you have areaswhere the gene is identical

in a human, in a mouse,in a dog, in a drosophila, in a yeast, in a bacteria,that's a part of the gene that you simply can't mutatewithout altering its function. so we can prioritizethat mutations in that part ofthe genome are more likely to be functional ordisruptive of the protein's function. and so when you go ahead anddo that, again we go back and we take the 19,000genes, we match them

up to the coding regions. we actually narrow it downto just a couple hundred doing that. and then we take thosecouple hundred candidates, we align them evolutionarily. and when we do that, there wasa particular gene called rad50 where there was asingle base pair substitution in that gene whereit was in a part of the protein called the d loop.

and that part is essentiallythe same in humans, in mice, in drosophila, inyeast, in bacteria. and so that really allowed us toprioritize this particular gene mutation for further functionalvalidation versus for example, this atr mutation,which was also mutated in thispatient's tumor but was not in a known functionalarea of the protein and was not in an area ofevolutionary conservation. we can also take thesedifferent genes that

are mutated and based uponcrystal structures of proteins that have been made, we canactually model off the crystal structure where themutation is in the protein and what this specificamino acid substitution is potentially going to do to theinteraction of that protein with other proteins. and so this is justshowing in this patient who had this unbelievablecurative response, that this is a protein complex of actuallythree different proteins called

the mre complex. and this mutation inrad50 would actually be predicted to be on theinterface between rad50 and another protein calledmre11, would actually disrupt binding ofthese two proteins, and would actuallydisrupt the function. so based upon again, thisevolutionary conservation and this crystalstructure modeling, we're able to takethese 19,000 mutations

and really focus onthis one mutation that seems most interesting. but even there, it's reallya computer prediction. it's a guess. and we still need togo back to the lab to do functionalstudies to prove that that hypothesismakes a lot of sense. so what we did in this patient,to make a long story short, is we took advantage of thisevolutionary conservation

within this proteinto actually do a cross species validation ofthis mutation in a yeast model. so because the yeastprotein for this pathway is identical inhumans and in yeast, we can knock thismutation into yeast and see if it disruptsthe protein function and whether itwould maybe explain this woman'sincredible sensitivity to the chemotherapy.

and so that's reallythis experiment here. and what we're showingin this experiment is if you knock in thisrad50 mutation into yeast- and actually yeast justhave one copy of the genome, these are haploid yeast--you get some sensitization to the chemotherapy here. but if you actually knock inthis rad50 mutation and also knock out thecheck one pathway-- and so mec1 is thehomologue of atr,

which activatesupstream check one-- you get what's calleda synthetically lethal interaction. so you see a killingof these cells when you have both the rad50mutation and knockout check one when you give thechemotherapy that you don't see when you give thechemotherapy to cells that don't have thisparticular mutation profile. so again, like the firstpatient i talked to you about,

this patient wasthe perfect patient for this combination of drugs. we simply didn't know itgoing into the clinical trial. but one could now hypothesizethat if we knew the mutation status of all ofthese patients going into these clinical trials, wecould do these clinical trials more effectively, hopefullyget them done quicker, and hopefully see betterresults going forward. so this is really the paradigmof this extraordinary responder

initiative that's now gonenationwide through the national cancer institute. we take a clinical trialwhich was disappointing, but for which there weresome patients who seemed to benefit from the treatment. we take tumors fromthose patients. we do some sort ofgenetic analysis to figure out potentially whatmade those patients unique. it's not showing upon the slide here,

but we always go backto the laboratory to do functionalvalidation of that biology. and then what we needis we need an assay, we need a test that willallow us to perspectively now look for those particularmutations in everybody. so we know who haswhich mutations so that we can do an iterativeclinical trial where we just enroll patientswith that mutation onto this drug to ask howmany of them are going to have

similar type of responses. so how are we goingto find the patients? what is the assaythat we're going to do to actually do this analysis? and so this is what we'redoing at sloan kettering. we run an assay calledmsk-impact-- impact standing for integrated mutationprofiling of actionable cancer targets. and so this is anassay of 341 genes.

and i took out the slide. and so what we knowabout cancer is that genes are notnecessarily distributed in terms of theimportant mutations all the way throughout the genome. that there's certain partsof the genome, certain cancer genes, that are more likelyto be important than others. and so what we can doto try to save costs, to save money so that wecan sequence more patients

and get more from our sequencingresource, what we can do is we can take dna from multiplepatients, we can chop it up, we can then attach a dna barcodeto each of those individual dna molecules, and then mix allthose dna molecules together in the same reaction. and then we can usethose barcode dnas in a very powerfulcomputer at the end to actually figure outwhich dna molecule came from which patient.

and so by mixing all thesein the same reaction, that allows us to save money. and i hope you'vegot a plug. [laughs] female speaker: it's there. i don't know why it's not. keep talking. david b. solit: ok. the other way wesave money is we can actually pull downthe parts of the genome

that we're most interested in. so we develop dates. we develop dna dates thatwill go and hybridize with the parts of the genomewhere the cancer mutations are most likely to be found. and that again allows us tosave money on the sequencing resource in the future. and so we'vedeveloped this assay. it's a custom assay.

yeah? audience: question. [inaudible] david b. solit: so this iscalled next-gen sequencing, and it's really adigital platform. so we sequence each individualdna molecule separately. and what we try to do is tryto sequence each of these dna molecules 500 or 1,000 times. and the reason thatwe need to do that

is because the normal dna ismixed in with the tumor dna. we can't usually geta pure population of just cancer cells. and so if the tumor that we cutout from the patient and then smushed up andextract the dna out of it, if that's 70% normalcells and 30% tumor cells, we need to getbetter sensitivity for finding the mutations. we want to havevery deep coverage.

and that means sequencingthat same part of the genome again and again andagain to make sure we don't miss the mutation. yeah. female speaker: [inaudible]. david b. solit: so we'vecreated this initiative at sloan kettering. and really the initiativeis pretty simple. what we're tryingto do here is we're

trying to sequenceevery single patient with advanced metastaticcancer at sloan kettering using this assay. and so we see about45,000 unique patients with cancer at sloankettering every single year. about 75% of thesepatients are cured. about a quarter of themdevelop recurrent disease and are thus at highest riskof dying of their cancer and thus the most inneed for novel therapies.

and so we've got about10,000 to 12,000 patients like that at sloankettering every year. and so we are goingto attempt, and we've begun this effort justover the past year, use this assay to sequence everysingle one of those patients. we actually sequenceboth the tumor and we also sequencethe normal dna. and by sequencing the normaldna in all these genes, there's actually the potentialto identify a mutation that's

in the normal cellsthat was actually the reason that the patient gotthe cancer in the first place. so genes like brca1or brca2, these are risk factor genes thatmake it more likely that you're going to get cancer. what's the majorhurdle to doing this? previously up to thisyear, it was technology. but now it's reallymoney, like many things. so it's very expensiveto do the sequencing.

and that's why i wasfocusing on how can we bring the cost of thesequencing down by mixing dna from multiple patientstogether and just sequencing parts of the genome? and it's bothexpensive and it's not standard to do this typeof testing in most patients with cancer. so we can only billinsurance companies to do this type of testingin six adult tumors.

so patients with lungcancer, colorectal, melanoma, gist tumors,thyroid, and patients with diffuse gliomas,which are brain cancers. so people with breast cancer,folks with prostate cancer, endometrial cancer, ovariancancer, bladder cancer, these are not tumors that we canget any sort of insurance or medicare reimbursementfor doing the sequencing. so to be able to do thissequencing, we need to go out and we need to raise funds.

and i'm actuallyhere this weekend riding with cycle forsurvival to raise funds to do this type of sequencing. and so our plan is totry to do this as a what we call a researchnon billable test, try to sequence everyone,and then use this information to try to drive patients tothe most appropriate treatment. we've also made a decisionat sloan kettering that we're going to makethe data that we generate

using this assay availableto everyone at the center in real time. so what we've done is we'vecreated this web-based portal and database called thecbioportal for cancer genomics. and there's botha private version of this portal and a publicversion that you can access. but this is really almost cancerbioinformatics for dummies. it really is away for biologists who don't have acomputational degree, maybe

knowledge of how to doa unix command prompt or other training, to beable to browse this data. so if you're a clinician andyou're not an informatician, or you're a biologistand just don't have a bioinformaticsdegree, this type of portal lets you figure outvery quickly where mutation's found in particularcancer types, what's the co-mutationpattern, how often are those mutations found.

and this data's now availableto everyone at sloan kettering. and this is the summarypage of that portal. and i just blew it up hereso that you can see it. it lists how many of eachcancer that we've done. 441 patients with breastcancer, 393 with lung cancer, 222 with colorectal,all down the line. and we're not justdoing common cancers. we're also doingvery rare cancers. we've done one patient forexample with a wilms' tumor.

this is a very rarepediatric cancer. we've done a couple patientswho have fallopian tube cancers, another very rare cancer. and so we're doing both commoncancers and rare tumor types. there's a patient centeredview for these tumors. and you can see thisis a patient who had a colorectal cancer whichhad one of these braf mutations that we now have a drugagainst that i mentioned to you a little bit earlier.

but the patient also has anumber of other mutations that are picked up by the assay. and so this typeof sequencing may allow us to understandmaybe why some patients with braf reputations whoget the braf inhibitor do better than others. so some patients mayhave a co-mutation that makes their tumorless dependent upon braf and thus less sensitiveto that inhibitor.

we've also createda database which is out there to actually educatethe clinicians as to what this information means. so many of the oncologists whotreat these patients really just never got theeducation in cancer genetics to even understand what we'retalking about when we talk about all these gene mutations. and so there's 20,000 plusgenes in the human genome. i don't know offthe top of my head

the four digit code forevery single one of those and what that gene does. and so we almostneed a computer tool to be able to, whenthis data comes back, be able to give us data. this gene has thisfunction and when you have this particularmutation in the gene, that's a functional mutationand causes sensitivity to a particular drug.

so this database is somethingthat's a work in progress. but something that actuallydoesn't exist publicly is something we'reactually hoping maybe we can get together withother major cancer centers and do a wikipedia type ofmodel where everybody puts in their specificknowledge expertise and eventually youthen crowd source this sort of importantdata for all 20,000 plus genes in the genome.

but currently, wehave this data just for the genes thatwe have in our assay. and even there,it's only partial. so it's a real work inprogress, and a real problem. because when this data comesback, most oncologists again, they don't know what theseparticular mutations mean and what they do. so again, this is--yeah, go ahead. so if we actually go backone slide here before this

comes up. audience: [inaudible] david b. solit: yeah. ok, so can wecompare our database to 1000 genomes or thesanger database and others? and cbioportal has become apublic database in some ways. so all the tumor cancer genomeatlas that's publicly available is now searchablein the cbioportal. other publisheddatabases are now

brought into the cbioportal. and so this hasactually become, i think, the primarytool for most people who are not cancerbioinformaticians to really sort through the data. but you know, the problemis is that not everyone even makes their data publiclyavailable in a way that we can bring itin to these databases. so just to give youone example, there's

a company that does germ linetesting for brca called myriad. and they've probably testedclose to 200,000 plus patients with breast and ovariancancer for these germ line alterations. but they consider thatdata to be proprietary. they don't release it, andso we don't have that data. so it really isgoing to be important that anything public getsinto these type of databases. and it's not something thata cancer center necessarily

was in the best position to do. i mean, that'snot our expertise. and so we've created thisinitiative at sloan kettering. but i'll tell you,we'd love to reach out to people who have betterexperience putting together these type of databases tomake these tools even better. so i think thatremains a problem. there's a lot ofproprietary data out there, but i think there's enoughin the public domain

that if you really consolidatedit into one source, it would be incredibly valuable. so again, that's how we'regoing to find the mutations. and now i just want totalk for a few minutes about how are we going todo the clinical trials that are actually going to take thisgenetic information to really test these new drugs? and so there'sreally two main ways we can do these clinical trials.

one is called an umbrellaprotocol or a master protocol. and these protocolsare usually centered around a particular cancer type. so this would be a masterprotocol for patients with lung cancer, as an example. and so what we do here iswe sequence all the patients with lung cancer, maybeat multiple institutions. and then depending uponwhat mutation they have, we put them on differentdrugs based upon that.

and many of the doctorslike this approach, because as i mentionedearlier, it's very hard to find the fundsto sequence the patients. and so if you have aclinical trial like this, you can ask the drug companiesmaybe that make these drugs-- the egfr inhibitor, the brafinhibitor, the mac inhibitor-- to maybe pay for the sequencing. so this is one waythat we try to do this. and there are somemaster protocols open

for patients with lung cancerand some other tumor types. there's a few problemswith this approach though. one is that oftentimesit's not the same drug company that makes each drug. and so drug company one mightmake the egfr inhibitor, drug company two maymake the braf inhibitor, and so on and so forth. so sometimes you can spenda year or two negotiating with different drugcompanies to get

the different drugs into thisone giganto clinical trial. and by the time you'veactually secured the drugs and opened the trial, maybeone or two of these drugs have already failedin another context and they're not really the bestin class drug that's available. so that's one downside. the other downsidei'm going to focus on is that oftentimes theseare small clinical trials. and maybe you're sequencingin this clinical trial

200 patients. and if you've got amutation in the population that you're studyingthat's only 1% or 2% or 3% of the population,you're not going to get enough patientson to this clinical trial with that particularmutation to really answer the primary question, whichis if you have this mutation, do you respond to this specificinhibitor of that drug? so that's one approach.

but if you could findthe right patient , you can sometimes seethese incredible effects. this is a patientwith lung cancer. and actually in lungcancer, these braf mutations i mentioned to you earlier onlymake up about 1% to maybe 2% of the population. but if you can find that rarepatient who has a braf mutation and give them thesame drug that works in melanoma-- in thiscase, [inaudible],

which is a selectiveinhibitor of raf-- you can see these typeof incredible effects. this is a patient who's beenon this drug for over two years with a disease that mostpatients progress very quickly and die within a year or two. so again, if you canmatch the right patient on to the right drug by somemechanism, this can work. but for these rare mutations,this type of master protocol design is really inefficient.

so what we've tried to dois create a different type of trial design that works formutations where the mutation is found in a lot ofdifferent cancers, but is not commonin any of them. and so an exampleof that is a gene called erbb2, which is mutatedin pretty much every different but usually about1% or 2% or 3%. so there's a lotof these mutations. if you add them up,there's probably

10,000 patients inthis country a year who have a erbb2 mutant tumor. but in most cancers, likebreast cancer, it's only 2% so if you wantedto just test for it and then enroll abreast cancer trial, you'd have to test100 patients to find two who have the mutation. and that's a veryinefficient way to do that. so what we've decidedto do is we've

created this new type of trialthat we call a basket trial. and what a baskettrial is it's a trial that's focused around a specificgene or mutation and not around a particular tumor. so we can take any patient. it doesn't matter whatkind of cancer you have, as long as you have themutation we're interested in. and for this study of adrug called neratinib, we're looking for patientswith a erbb2 mutation.

and it doesn't matter if youhave a bladder cancer or colon cancer, an endometrial,gastric, ovarian, breast. it doesn't matter. if you have a erbb2 mutation,you can go on to the study. we always have anotherarm for these studies as well, because maybethere's some other rare cancer type that we didn't evenknow had erbb2 mutations. but we find one, we wantto give the opportunity to test whether that patient'sgoing to respond to the drug.

and we've seen some incredibleresponses with this approach. so this is a patient who hasone of these erbb2 mutations. and we generally don't testfor these mutations in breast cancer, which is what she has. and this is a woman who'dbeen through many lines of chemotherapy,none of which were working because they weren'ttargeting the right mutation. and you can seefrom her pet scan, she had this incrediblyhigh burden of disease.

all of these are actuallymetastasis from her cancer. this is her brain,this is her bladder. some of this is herheart, but most of these are all mutations. and actually see lessuptake in the heart because so much of the pettracer here-- the glucose that we inject, theradioactive glucose tracer-- is going up in the tumor. and so she was started onthis drug on this basket study

in july of 2004. within two months, she'shad a complete response. but the key issue here isnot that it's surprising that this erbb2 inhibitor,which is what neratinib is, would work in thiserbb2 mutant patient. the real problem is thatalmost all patients who have this mutation-- this v77lmissense mutation in erbb2-- are unaware that theyhave this mutation because we don't testfor this mutation.

so since nobody's testing forit, we can't find the patients. we can't put them onthe trial and we can't prove that the drug works. so it's really a chickenand the egg problem. how do you prove the drugworks without knowing who has the mutation? how can you actually justifydoing the testing unless you know that the drug works? because otherwise,the insurance company

won't pay for the testing. and so to try to break thischicken and the egg cycle, we've just made thiscommitment that we are going to test everyone usinginstitutional funds until we can essentially enroll enoughpatients onto these trials to prove that thesedrugs work or don't work. it's a little more complicatedthan i'm making it out to be. there are multiple differenterbb2 mutations in cancer. they may all respond a littlebit differently to the drug.

and so what we do with thebasket study, for example, is we can take thisbreast cancer arm. we can initially enroll all thepatients with erbb2 mutations. since we've alreadyseen a positive result on the overall arm, wecan then do sub-baskets where we enroll acohort of patients that have this s310 mutation,a cohort of patients with an l755 mutation, a cohortwith v77l on down the line. and again, we alwayshave another arm

for these raremutations to give access to the drug to these patients. and the question is, can weeventually get to a point where we've got a rareenough mutation disease combination where almosteverybody's responding that the fda would simply allowus to change standard of care at that point? so typically, the fda hasrequired us, for cancer drugs, to do randomized studies.

where one half of thepatients or so get the drug, the other half get a placeboor an older standard drug that doesn't work very well. and the questionis, can we avoid those type of randomized studiesby very carefully selecting the patients so that the benefitis so high on a small study that it compels the communityto just change standard of care? because you could imagine if wewere seeing this type of result in 80% or 90% percentof the patients

with this mutationin breast cancer, a patient with that exactmutation in breast cancer doesn't want to go on a placebo. and it might sound ridiculous toeven consider that possibility. but the [inaudible] study thati told you about earlier was that exact situation. and despite that,we actually had to do a placebo controlledstudy by the fda. in that case, there were enoughpatients to do that trial.

i don't even think there'senough patients in the country to do a placebo controlled studyof just v77l erbb2 mutations with neratinib. so hopefully, the regulatorswill accept this and change the way we do things. question? so i guess thequestion was are we potentially missing importantmutations or other alterations by just sequencing the 410genes that our current assay is

looking at? and the answer isabsolutely correct. but my initial answer thatquestion would be this is an example of notletting the perfect be the enemy of the good. so since we can't dowhole genome sequencing on every singlepatient, that shouldn't prevent us to do 410 geneson every single patient. so we're doing what we'refeasibly able to do right now.

but the expectation isthat this sequencing is going to becomecheaper over time, and the computationalpower to do the analysis gets cheaper andquicker over time. and everybody i'msure in this room knows of moore's law,which is the rate at which computational powerbecomes more powerful and less expensive. the rate at which sequencingis becoming more powerful

is actually muchdramatically faster than what you see with moore'slaw with semiconductors. so we're seeing thisincredibly fast drop in the cost of sequencing. and there's companieslike illumina and others that aremaking these machines that are driving this improvement. and it's probablya few years away from a point where we willbe getting that whole genome

sequence on every patient. we just feasiblycan't do it now. but there areprobably some patients who are going out and havingtheir whole genome sequenced. there might be a fewpatients of means that can convince somebody to do that. my lab does it in thisretrospective approach for these rare,extraordinary responders. but each one of those not onlycosts the lab tens of thousands

of dollars to do,but it takes actually a team of informaticiansweeks to do the analysis. so we need the costof the sequencing to come down andthen what we need is better algorithmicapproaches for the analysis so that you can take thesebillions of data points and be able to analyzethem much more quickly. and one way to dothat is to have this central databaseof all known variant

changes that you'reable to reference. and so the factthat we've only done a small number ofwhole genomes so far make it difficult for usto know whether something found in a particulargenome is important. because the more youdo, the more often you see these patternsthat allow you to know that something is real or not. but it's a great question.

so what are the advantagesof this basket approach? i'm a clinician, but i'm also acancer biologist, a geneticist. what this allowsus to do is really test that definedbiologic hypothesis. does a erbb2 inhibitor work inpatients with a erbb2 mutation? how often does it work? if it works in somepatients and not others, we can collect tumorsfrom all those patients, we can maybe do wholegenome sequencing

in that whole cohort,and figure out what are the co-mutations thatdetermine that heterogeneity within even that small class. we're also doing onthis trial what's called a co-clinical trial,where we enroll patients not just with knownactivating mutations but also patients who have novelmutations that have never been biologically characterized. and what we do is whilethose patients are

on the clinicaltrial, we biologically characterize thosemutations in the laboratory. and we may find thatfor some of them, they're notfunctional mutations. so if i go back to thisslide of erbb2 mutations, these hot spots have now allbeen biologically validated and are real. as we get more andmore sequencing data, some of these rare mutationswill become hot spots.

but some of these are just whatwe call passenger mutations-- the random changes in thedna due to the toxin that induced the dna damage. and they're notdriving the cancer. and so we don't havethat knowledge for all of these mutations so far. we just have it forthe most common. but as patients comeon to the study, we study individual mutationsthat have not already

been validated. what are the challengeswith the approach? the primary criticism weget from drug companies, clinicians,regulators, is that we fail to identify patientswho could respond but they don't have thebio-marker we're testing. so in the case of erbb2,you don't necessarily have to have a erbb2 mutationto benefit from neratinib. we actually know that's true.

maybe you can have an activatedreceptor from a ligand that's binding. so while that's true, it'snot the point of the study, and we can always study thosethings in later studies. i call it the sad fact,and that's in academia, it's sometimes difficult toget the lung doctors to work with the colon doctors to getthem to work with the bladder doctors. they're actually not used todoing clinical trials that way.

so the way they train, theway they practice over decades is you just dobreast cancer trials, and they're separate thandoing lung cancer trials, and they're separatethan doing colon trials. and the concept of doing atrial where any type of cancer can be in that trialis something that's foreign to the community. but i think that's startingto change with some of the success we've seen.

and again, i'm justgoing to mention that the primary hurdle reallyremains that identifying the patients is a challenge. the vast majority of patientswith cancer in this country don't get any testing. there's only a few institutionsdoing this right now. the scale, that number is goingto keep going up over time. but most patients in thecommunity don't get this. there are some companies thatare now specifically opening up

to do this type of testing. one's calledfoundation medicine. they charge $5,800 for thetest, and generally insurance doesn't cover. so for most patients,it's not really feasible to getthis test performed. i just have two more slides. i just want to point outthat the testing's not just for selection of what we calltargeted therapies, which

are drugs that inhibit aspecific mutant protein. this was one of thefirst patients of mine that i see in the clinicwhich i did this testing for. and this was a patientwith prostate cancer who did very poorly withall the standard treatments. and so we went ahead andwe biopsied his cancer after it had become metastaticand had spread to other sites. and then we did thetesting, and what was notable about histumor is that he actually

had 29 mutationsidentified by the assay. and that's surprisingin prostate cancer, because most prostate cancersonly have a few mutations. so prostate cancer's knownfor a very low mutation rate. but this guy had a veryhigh mutation rate tumor. and he actually had twogenes that were mutated. the red bars are a little off. one's called mlh1, theother's called msh2. and these were bothtruncating mutations.

and what's interestingabout these genes is that when you getmutations in these genes in the normal cells, youhave a condition called lynch syndrome, whichleads to colorectal cancers, endometrial cancers. not typically prostatecancers, but you get these hypermutated tumors. and what we've learned is thatif you test immunotherapies-- these are a new class of drugsthat harness the immune system

to attack the cancer-- thesetype of hypermutated tumors are more likely torespond than tumors that have fewer mutations. and that's probablydue to the fact that these hypermutatedtumors are expressing proteins that are seen by theimmune system as being foreign. they're not self, they'renot the normal proteins for that individual. they've got a slight changeand so they're seen as foreign

and it's easier to stimulate theimmune system to attack them. so we actuallyhave a basket study where you can haveany type of cancer. if you have one of thesetype of hypermutated tumors, this basket studydesign provides access to this type of immunotherapy. in particular for this trial,a drug called medi4736. this is a drug that binds toa protein called pd1, which leads to a checkpointeffect and a stimulation

of the immune systemto fight the cancer. and i can say that thispatient here is actually seeing early benefitwith this anti pd1 therapy with thishypermutated prostate cancer. these type ofhypermutated tumors were not something we sawwhen we did a retrospective analysis of early tumors. but notably, we've foundalready about five or six of these patients outof the first 150 or so

of the prostate cancers thatwe've done perspectively and it's possible thatthese hypermutated tumors are overrepresented in thoseadvanced metastatic patients because thesehypermutated tumors are less likely to respondthe standard treatment and more likely to recur andprogress to metastatic disease. so again, figuringout in the end what is driving an individualpatient's tumor, we're hoping to use thatinformation to really guide

the treatment of patients. just one more point andthen i'll finish up. this is this companythat i mentioned that's been doingthis type of testing. this was a report that was sentto me from a patient at johns hopkins asking if they couldbe on this erbb2 basket study. and they had aerbb2 mutation, they had an s310f erbb2 mutation. but there was thisnote on the report that

was what's called subclonal. and so what that meansis that it's likely at least you caninfer informatically that this mutation is not likelypresent in every single cancer cell. and so to just show you fromour data what that means, here's a patientwith breast cancer who has an l755s mutation. the allele frequencyhere is the number

of reads in the tumorthat have the mutation versus the normal reads. and we never see this as 100%,because typically you only mutate one copy of the gene. and also there's thisstromal contamination of the normal cells,which are obviously normal for the mutation. but what's notable, it's 58%here, 20 something percent for the other two mutations.

this is probably amutant where there's two copies of themutation for erbb2 and one copy forthese other two genes. if i look at thisother patient, this is a patient withbladder cancer where can see this p53 mutation'sin 57% of the reads, the tse1 is in 61% of thereads, but the erbb2 mutation's only in 3% of the reads. and the way you canstatistically get that

is if the erbb2 mutation isjust in some of the cancer cells and not in all the cancer cells. and one could imagine thatif the mutation that we're targeting is not inevery single cancer cell, when we give the drug, the cellsthat don't have the mutation are probably going to quicklygrow out and become resistant. and this type of what wecall clonality analysis is very difficult to do when youhave limited amounts of data. so if we only have asmall list of mutations

because we're onlysequencing a few genes, it's very difficult to do this. but if we do whole exomeor whole genome sequencing, we can actually definethis clonality much better. it still takes a huge amountof computational resource to do this type of analysis. but as this technologygets better, this is the type ofdata we're going to be getting for every patient.

and it's not goingto simply be do you have the mutation as abinary variable, yes or no, but is that mutation what wecall an early truncal mutation? is it present in all cellsor is present in just some? this is an interestingpatient where we have four tumors that we sequenced. and this is how thecbioportal shows that. and you can see forthese four genes, these four mutations werefound in all four tumors.

but you could see the erbb2mutation was just in tumor one. this pik3ca mutation wasjust in two and four. we actually have drugsagainst both of these targets. the egfr mutation wasonly in tumor three. the p10 only in tumor two. so these are alldruggable mutations, but they're not inevery single tumor cell. and so this is an additionallevel of complexity that is going tomake this difficult.

and is really going torequire to eventually use combinations of these inhibitorsto really see maximal effect. so i'm just goingto leave it there and just acknowledgea number of people. i really want to acknowledgegopa iyer, a very talented-- he was initially a postdoc fellow,now junior faculty member who was in my lab andreally started with me this extraordinaryresponder initiative. the center i run, which is thekravis center for molecular

oncology at sloankettering, i've got three associate directors. one, mike berger, who developedthis impact assay and really deserves a lot ofcredit for that. another, agnes viale, who runsour operational sequencing core. and then barry taylor,who we actually recruited back fromucsf just recently to head what we callcomputational oncology.

and this is really a newfield that didn't exist before the last couple years. and so what'scomputational oncology? these are people who usecomputational tools to answer clinical and biologic questions. so we now have a wholegroup of scientists that that's what they'regoing to be identified as, as computational biologistsor computational oncologists. and then just noteseveral other people.

marc ladanyi, who runs ourclinical sequencing lab with maria arcila. dave hyman, who runs a lot ofthese basket studies with me, and jose baselga, who'sour physician in chief, and has really provideda lot of leadership to get this sequencingeffort up and running. so thanks againfor the invitation. happy to answer any questions,either now or formally later. so thank you.

[applause] audience: you showed a slidewhere one person had you said, 20 some mutationsin the prostate. david b. solit: yes. this is-- audience: and then you had aprevious slide further back where you did a single tumor. and you had 19,000somatic mutations. audience: and fromwhat i read actually,

that's more the commoncase versus the cancer that you have veryfew mutations. the common case is thatyou have tens of thousands of mutations in a single tumor. david b. solit: so letme just clarify this as an issue maybei wasn't clear on, which is really differencesin the denominator. so this assay thati'm showing you here, this was an assay ofjust 341 cancer genes.

so this was our capturebased approach where we just pull down parts of the genome. and when we do that, thescale's obviously different because we're only sequencinga small part of the genome. we're sequencing thepart of the genome with all the knowncancer genes in it, but it's a very smallfraction of the total genome. and so this is a lot ofmutations for sequencing just that small part.

audience: ok. david b. solit:the 19,000 referred to sequencing the entire genome. audience: right. david b. solit: andwithin that 19,000, that patient probably hadabout 140 coding mutations. and if you actuallyjust ran impact, you probably only would havefound about eight or nine mutations.

so it's really a differenceof what fraction of the genome are we sequencing? those numbers are going tobe very different, yeah. audience: so maybe the premiseof my question is wrong. but my question was ifyou've got so many-- you kind of boiled it downto a much smaller number-- but if you've got 19,000 somaticmutations, how many, if you were to look at so-callednormal cells in a normal person, a typical assay, howmany mutations would you

find in normal cells? david b. solit: sothe normal cells have no mutations bydefinition, because we're using that as the referencefor each individual patient. so there is the normalcoding sequence, which is really created by anamalgamation of many people. and then there are variationsfrom individual to individual. so because there's thesepolymorphisms, or differences in the genomebetween individuals,

we actually don't want touse this reference sequence for the actual calling ofa mutation that's somatic. we want to actually comparethis individual patient's normal to this tumorand actually then digitally subtract it. if you don't dothat, you'll end up with thousands ofadditional polymorphisms between any individualand actually the reference sequence.

now when we sequence the normal,we don't sequence one cell. we sequence what is dnaextracted out of thousands or tens of thousands of cells. and so if you actually sequencedjust one individual cell, you took hundreds ofthem, you might even see some mutationsthat are slightly different from thiscell to that cell. but we don't pick itup with this approach because we're averaging thatout and then that detail's lost.

audience: but it's radicallydifferent in a cancer tumor, right? david b. solit:but you know, yeah. audience: so a cancertumor, radically more. david b. solit:dramatically higher, yeah. audience: so howdo you explain that on a fundamental biological whenyou're looking at first causes? you would think,for example, or i would think-- justkind of a lay person--

that if you imaginethat cancer is caused by some initialsomatic mutation, why does that single mutationnot dominate and survive throughout the cancer? so why do you havetens of thousands? this is a real mystery. and it seems like a realproblem to this approach. because if you've gottens of thousands-- and some places i've readthey say hundreds of thousands

of mutations in a single tumor-- so-- audience: then how? david b. solit: so we actuallyhave a lot of sense for this now with the tumor cancergenome atlas that's been done. there's very wellcharacterized whole exome data. so not a lot ofwhole genome data, but whole exome data, showingthat, for example in melanomas, it's not uncommon to have3,000 or so exomic mutations.

and that's the tumor that hasthe highest mutation rate. there are many pediatriccancers-- lymphomas and leukemias-- that have anexceedingly low mutation rate. and some of it has to do withthe pathogenesis of the cancer. so what was thetoxin that caused it? so ultraviolet light is what'scausing these melanomas. and actually manyof the mutations are ones that are caused byultraviolet light induced changes.

whereas in lungcancers, which is also another hypermutated tumor,the mutation spectrum is a different typeof amino base pair change than what you see in aultraviolet induced melanoma. so many of these are passengers. how many are actuallydrivers, which make the difference betweena driver and a passenger is a driver really ishaving a functional effect in driving thecancer pathogenesis,

we really don'tknow in large part. the presumption is thevast majority of them are passengers, they'renot doing anything. but why are they there? they're therebecause there's just a stochastic randomness towhere these mutations occur in some ways. and unless you hit onsomething that actually is presenting asurvival advantage,

it doesn't really matter. so these mutations areaccumulating in a stem cell, and what happens is youget an early hit which maybe makes a cellproliferate a little bit more quickly than normal. so that cell starts toreplace the average population of the normal cells. but then in that, you geta lot of random mutations that are not doing anything.

but then eventually inone of those cells that's now built up more ofthose random changes you get a second hit. and then it accumulates. and then eventuallyyou can get a mutation in a gene like mlh1 and msh2,and these genes actually are important forthe dna replication. so once you get one of thesedna repair deficient mutations, there's just going tobe a huge number of now

additional mutationsthat are caused because these thefidelity of dna repair or dna reproduction is very low. and so that's part of it. i can't answer yourquestion completely. audience: it soundslike eventually this is going to go towardsif you have cancer, you'll get a screeningupfront and figure out what genomicalterations you have

and you can matchthem with a drug. do you think therate limiting factor on that will be technologyor insurance or physician adoption? which one of those isthe biggest hurdle? david b. solit: so until thispast year, it was technology. so even if you hadunlimited amounts of money, you simply couldn't do this. it would take a team ofdozens months or years

to do an individual genome. and that's just not feasible. but now my lab can doa whole genome sequence on any individualfor less than $10,000 if we spend x amount oftime to do the analysis. so it's getting quicker andit'll be turnkey more and more. so as the informatic tools,the algorithms get better, they're more automated. we actually are limitedby computational power

you guys could laugh. we have 1.5 petabytes ofmemory and that's all we got. and we've got a clusterwith this many processors. and our thing issome of the analysis is waiting to geton to the cluster. we just don't have enoughcomputational power to be doing more than we can. and we spent $2.5million last year alone on computationalresources,

and that's a lot of moneyfor a nonprofit hospital to be paying. so right now, technologyis still limiting. but that clearly isgoing to go away. the other problemis doctor adoption. i think even if thetechnology becomes completely turnkey and inexpensive, andeven if the insurance companies start paying for somedegree-- and they're not going to pay for wholegenome right away.

what they're going to do ispay for some number of genes and then as the costgoes down, they'll pay for more and more genes. and then maybeeventually it's cheaper to just do the wholegenome in everyone. but the otherdownside of this is doing this type ofsequencing somewhat opens up a pandora's box. and that's because we all haveprobably 10 to 20, if not more,

polymorphisms in ournormal cells that put us at risk for something. and it's notsomething that we all want to worry aboutfor our entire life. so just to give you aquick scenario here, we run our genetic test herefor a patient with cancer and the patient's cured. one of the genes in ourassay is called park2. and this is a gene that'simportant for cancer.

it's a tumor suppressor. it's a gene that was shown to bea tumor suppressor in a paper i published with a colleagueof mine, tim chan, just a few years ago. but the reason that thegene is called park2 is because it's foundin the germ line, so in the normal cells,of patients who have early onset parkinson's disease. and so imagine the scenarioof a patient coming in,

they get sequenced, we use thesequencing data to tell them how best to treat their cancer. and then the oncologistsays, oh by the way, you have a germ linemutation of park2 and you have an 80% chance ofgetting parkinson's disease by the age of 50, and we havenothing we can do about that. so did you want toknow that information? and as we find more and moreof those susceptibility genes, ones that maybe are lesspenetrant, meaning they're

less of a phenotypeor less dominant, we're all going tohave some of these. and is this going to createso much anxiety in all of us that some peopleare going to have a hard time dealing with that? so there is really a hugeamount of worry and concern in the community aswe're doing this testing to make sure that whenwe look at the germ line data in particular, that we'revery careful in how we counsel

the patients as towhat this means, and that we don't return a lotof information that is this is a change in your germ line ina gene that could be important but we don't know what it means. and then patients areworrying about it. so that's going to slowphysician adoption of this a little bit. but eventually, i thinkall of us in the field think eventually 10 yearsdown the line, 15 years down

the line, you're just goingto come in and in a day, here's your tumor sequence. and then we're going to matchyou up to certain therapies based upon that. but getting thereis going to require some significanttechnologic improvements. definitely bettercomputational algorithms for the analysis,better databases of what all these mutationsmean, not just somatically

but also in the germ linealterations, and then also better biologicunderstanding of what they mean so wecan counsel people that they don't worry aboutthings that are not important. and then probably evennew laws, because if you tell that patient that theyhave that park2 mutation, maybe they can't get lifeinsurance, they can't get health insurance. their family may havethe same mutation

and maybe can't get healthinsurance or life insurance. so this test is maybenot just affecting you, it's affecting yourfamily members. so there's a little waysto go until we sort out all those issues. audience: when youidentified the mutations and the responders,do you also sequence the genomes of thepeople that don't respond to confirm that theydon't have that mutation?

david b. solit: so for thepaper we published in "science" on that everolimus trial--i didn't show it here-- we unfortunatelydidn't have tumor for every patientfrom the study. but we did go back andsequence a host of patients from the study. and what we found thateven the patients who had minor responsesto everolimus, they also had a tse1 mutation.

but what we found is thatthe co-mutation pattern for those patientswas different. so this patient who hadthe extraordinary response had an nf2 mutationwith the tse1. and both of those,we think, cooperate to make that patientexquisitely dependant upon the target of that drug. other patients had a tse1mutation with other mutations that we think diminishaddiction to that target,

but maybe a combinationwould be helpful. i think the learning experiencefrom that was going back, we realized we didn'thave tumor from everyone. so i think now we're trying todo a much better job of making sure that when werun these studies, we collect these tumors andhave the material available for these type ofretrospective studies. but actually some of theextraordinary responders initiatives are notgoing to even be needed.

because if we have themutation data perspectively, it just becomes veryobvious sometimes why somebody responded. well, they had theright mutation. and so the key is again gettingthe tissue for study later on. and it's a challenge. i'm actually minimizing howbig of a challenge it is. it's not necessarily just goodenough to have a piece of tumor from five years agofrom the patient.

if they've had 10therapies along the way, the tumor may have changedsignificantly during that time and it's beyond what we're ableto do for every clinical trial, to get a new piece of tumorfrom every patient right before they go on the therapy. because that'sinvasive, it's costly. and so we work with whatwe have the best we can, and it's not always perfect. well, thank you.