cancer stem cells

hey john thanks for having me andthanks for the kind introduction really excited actually to tell this groupabout some of our work that sort of connecting encode like essays to reallyclinically entrenched problems and in particular in the area of cancer stemcells in glioblastoma. so glioblastoma is the most frequent human brain tumorand and despite resection of the tumor and radiotherapy, radiation, chemotherapyimmediate survival is very short and almost uniformly fatal so there is thisa lot of evidence if not all cells in glioblastoma are created equal butrather there is this subpopulation of very aggressive stem like cells thatglioblastoma that propagate the tumor so

by the time you resected this lesionhere there's actually a stem cells tumor stem cellssort of infiltrated throughout the brain right? knee seemed to be resistant orrefractory to therapy this is sort of a model of cancer stem cells i don't meanto say that all cancers have stem cells and it's a really controversial area butin glioblastoma there is evidence that not all cells are alike and that thereare these purple cells that can both self-renewal and and and propagatetumors of the purpose of this type of colon cancer stem cells or glioma stemcells. one can can model these cells and this is what we do actually take humantumors respected at mgh now

surgery and we can either expand themand stem like conditions in serum-free conditions as gliomaspheres these growspheres they look a lot like neural stem cells that you might be isolating fromes cells or something but these gliomaspheres they express cd133 which isa canonical marker of the stem like state and they're very aggressive so youcan put as few as 50 of these cells into a mouse brain and orthotopictransplantation and they will cause a tumor right. they are tumor initiating andpropagating cells you can take the same human tumor and you canexpand the cells in serum and these grow a cell line kind of the conventional cell linethat we all might work with

these cells though they don't expresscd133 you can put hundreds of thousands or millionsof these cells into mouse and they won't do anything. the mouse is just finethese cells want propagate tumors and they lacked stem like characteristics okyou can differentiate these cells so you can put these cells into here into theserum conditions will differentiate and they're no longer tumorigenic right sothey're sort of this directional thing that these are sort of primitive stateand they will differentiate you can't go back you take these cells and in serumthese non stem cells they won't go back to this initial state they basically won'tproliferate so this was very interesting to

us and let us tothis question of whether we could we could reprogram these differentiatedcells into tumor propagating glioomaspheres okay we basically set up thestrategy where we were going to look at the enhancer landscapes in thetranscription factors and then we would test the ability of these tfs toreprogram these cells into the into the gliomaspheres so whether we canidentify something like at tf code for so this is just you know very familiarnow to this group but this is basically i think pretty striking picture thatsays well if you take three different patients' tumors and you expand them inthe stem like conditions you can see

that the ... is very activein their these sort of an h3 k 27 acetylated peaks in the gypsy profiles thatare enhancers in the socks to locust but ifyou take the same tumors new expand them in the differentiated cell lineconditions you this is a very quiet locust right ...conversely if you look at some locust like being before which is so she woulddifferentiation you see precisely the opposite right now in these cases it'snot which patient's tumor you're looking at its whether you're isolating andexpanding a stem like population or a differentiated population that isunderlying at least this landscape but i

think you can see it's more strikingwhen you now each little row here different candidate enhancer or k 27 tosettle peak in the genome considered a large group of shared enhancer likeelements shared between the stem like cells announce themselves these havemotifs for some of the same cell cycle regulators rate these guys areproliferating but we're interested in here are these stem like cell specificpeaks you can see these guys are all active in the stem cells are not in thedifferentiated cells again motif analysis identifies each villagemotifs in socks motifs suggested sort of the neurodevelopmental program rightthe thing we can do is we can look at our

data we basically identified a set ofabout 20 different transcription factors that are predicted to bind to thesemotifs and that we posited might be you know explaining the different signaturein the cfcs i'm in which are expressed specifically in the cfcs so we went backand took the differentiated cells and we overexpressed each of thesetranscription factors one by one to ask whether they could reprogram it kind oflike the yamanaka experiment but we're looking for cancer stem cells sobasically if you take the the differentiated cells and you try toexpand them in sort of spheregenic conditions basically nothing happens andyou can see that here you get 0 sphere

forming cells are gonna start addingthese different transcription factors there are couple that kind of lit up andyou know you could sort of user if you if you were very optimistic you mightguess that paltry up to gives you a little bit of somethinggrowing in the plate no real cd 133 but we thought we might be moving in theright direction so we try to try and combinations of factors socks to you geta lot more sphere formation and so on ultimately by sort of iterating betweenadditions of factors and then looking at the enhancer maps that we were gonnagetting we were able to sort of come up with this cocktail and ultimatelydefined 40 tfs to socks to cell

to ... when you addedthem to the cell lines they would reprogram into spheregenic stem like cellsthat could propagate tumors and this is justshowing you a picture that you could take these reprogrammed cells you put themin a mouse they will cause a tumor and the tumor lookslike to some degree like glioblastoma ok so we call these induced gscsas you might hope you can see that thesecsc specific enhancer elements after you reprogram cells with the 4tfsthey get ... so you'rereally reprogramming also the chromatin

landscapes the chromatin landscape inthis case is a pretty good surrogate for the functional state of the cells anotheri think pretty critical data point here is that you only require transgeneexpression of the court he s so in some reprogramming again in summaryprogramming are different the differentiation experiments the factorsare required continuously to maintain a new states that raises the question ofwhether you really reprogram with you really achieve an epigeneticself-sustaining transition but here what happens is we can show that the theexact expression plasmids that we've expressed actually get shut off we cantell just by looking at the 3 utrs

and the ... actually come on you cantell just by looking at the rna expression patterns and you can alsotell just by looking at the loci and you can see that the enhancers in thepast we have to look at the enhancers in the socks to look as rightare coming on right so you sort of reprogram these guys in there nowself-sustaining part of how you know this now too is that you can now switchthese i gscs back into serum and the readily differentiate and they'lltotally shut off the core tfs so how might this be useful to understandwell for one thing we could go back and mapthe direct targets of these core tfs which

we will help you to stocks to sell twoof two by chip-seq there's a great deal of overlap the shore of binding each other'slow side and we think sustainingtheir expression i can tell you that all four of these guys are essential fortumor propagation the gscs but more than that when younow in further downstream targets of these core tfscan use the binding patterns and expression patterns and furtherdownstream targets i can tell you that essentially everything we've looked atand then try to knock down in gscs has turned out to be an essentialfactors this is sort of like you know

there's various proposals out therewhether you super and answers or specific tf signatures or whathere's an example where by knowing the core tfs driving the circuitryand examining their direct downstream targets you now have a set of regulatorsthat seem also to be essential for the tumor propagating state we wereparticularly interested in a school one which is a transcription factor thatit's a neurodevelopmental t effort activating which signalinggm's as well as the arc or to histone demethylase complex which actually threeor four different members of this complex are direct targets of the corecircuitry ok so in a sense knowing the

core circuitry now like like knowingthe core circuitry of ips cells told you a lot about the pluripotentregulatory circuitry we learned a lot about the gsc circuitry by using sort offamiliar encode type approaches combining them with with reprogrammingapproaches to define the circuitry i kinda told you this was some levelit's a very fundamental project i think that is clinically driven we're doingthis because we aren't want to understand the nature of these cellsthat are propagating tumors in humans so the question then is what's the statusof these core tfs in primary glioblastomas so here we wanted to weactually did a series of we did flow

cytometry as well as immunofluorescenceanalyses we labeled antibodies that recognize each of the core factors wealso use cd 133 antibody and we stained slides of primary glioblastomas frompatients and we can see there's a great deal of heterogenity as you mightexpect are notoriously heterogeneous tumor variable expressionof the core tfs but there clearly are a subpopulation of cells the pink oneshere that we know are quietly expressing all four of the core tfs in the tumorsand when we do what we do flow cytometry when we look at all themarkers these guys i can tell you are also cd 133 positive which is one of themarkers that enriches for tumor

propagation it's very likely that thereare surrogates within primary tumors the look very much like the stem cells thati just sort of functionally described in our model and such as shown here okin another way and that is by single-cell arnie seq a single cell ona sikh again of a primary tumors one can learn a little bit about the regulatoryprograms in the primary tumors and it's this is a short talk and try to focus onregulatory elements a little bit so i'm not gonna give you much on thesingle-cell rd seq so i want to tell you one interesting property we lookedat a few hundred cells from a few different glioblastomas we saw a lotof heterogeneity within each tumors that

variety of transcriptional program thesetumors at the rna level as you might expect from that last picture veryheterogeneous but what we want to take this opportunity to ask about the stemlike transcriptional programs and its presence in primary gbm so we defined astem like signature based on which is composed of genes that are expressed atlow levels in these differentiated cell lines that are highly expressed in thegsc this time like model that i talked about earlier in the top and then weasked what is the expression pattern these genes in single cells from a person'stumor. so this is just one tumor from a patient, and we looked at about 70 different cells.and i think you can see that there is a population

of cells at the right side of the slide herethat are highly expressing a lot of the genes in the signature, okay? so together with thelast picture where i said, "well, we've got the core tfs and we've got a protein levelanalysis," so we got stem-like cells that emulate our model in tumors. now i'm tellingyou that we've also identified cells with a transcriptional signature that looks a lotlike what our model looks like, okay? i'm telling you all this, but now i'm goingto tell you about a little caveat, okay? the problem was that if you take these cells,and you now ask about their cell cycle signature, it turns out they're not cycling. they'retotally dormant in class, okay? so we have this model of cscs with the right neurodevelopmentalprograms that seems to be emulating something

in a tumor. yet, our cscs are proliferatinglike mad, like probably most of the cultures that people work on in this room, becauseif a culture's not proliferating it's hard to work with. but the real in vivo gscs withthis developmental program seem to be out of cycle and pretty much quiescent and dormant,okay? so this presents a bit of a challenge to us as we think about modeling and targeting,okay? and we tell you now how we're trying to getat this population. i'm going to come to the population in a second, but i need to justsort of talk for a minute about therapeutics, okay, before i get back there. okay? so receptortyrosine kinasis. so rtk inhibitors, of course, have been a dramatic advance in cancer researchand brian druker and charles sawyers and others

have shown that this miraculous inhibitorof bc are able practically cures cml. this rtk inhibitor basically ablates, you know,this whole population in this whole tumor. so this led to a lot of excitement and nowrtks are being used in many settings. gmbs glioma seem like a perfect setting touse rtk inhibitors, right? look at the genetics of glioblastoma. you've got egfr amplification,a huge proportion -- you've got pdgfr amplification in others, altogether about 67 percent ofbrain tumors -- glioblastoma brain tumors -- harbor an amplification of an rtk. so,of course, this led to a lot of excitement to then go in the clinic and try rtk inhibitorsthat could target these genetic lesions and hopefully have an impact on glioblastoma.they haven't done anything in the clinic,

though, right? they've been -- they haven'tchanged the course of this disease whatsoever. so we wanted to understand this better, andyou'll understand where i'm going in just a minute. but we took one our stem cell models,one of these proliferating gsc models, and we took in particular one with the pdgfr amplification,okay? and we treated it with dasatinib, which is a really good pdgfr inhibitor. you cansee that dasatinib, this particular population is very sensitive to dasatinib. pretty lowconcentrations of this inhibitor kills the cells, okay? this is in contrast to a coupleother gliomalines [spelled phonetically]. if you take a glioma's stem cells from anegfr amplified one or a mic-amplified [spelled phonetically] line, they don't care aboutpdgfr so they're insensitive dasatinib, okay?

but look at this. even in the cells that aresupposedly totally dependent on pdgfr, there is this sort of window here where about fiveto 10 percent of the cells just won't die, and they remain viable. you culture them outfor a while, they actually start proliferating very slowly, okay? and they are quite insensitive.we call them persisters. i'll call them gsc-8 persisters or whatever, persister stem cells.they proliferate slowly, but they're pretty much impervious to very high concentrationsof dasatinib. so this cell within this genetically amplified pdgfr-dependent population, thesecells can acquire a different state that is independent of the rtk. so this is just showing you that we take thenaive cells and we can culture them with -- in

the presence of dasatinib and over time wekind of acquire this slow cycling population, okay? this is a little bit analogous to persistersthat have been defined in lung, melanoma, breast, and other settings. one interestingthing is it seems to be reversible, because we can remove the pressure of the drug andthese cells will go back to their initial state where they're proliferating very rapidlyand pdgfr-dependent, okay? it's kind of a reversible persister state. okay. this i think was pretty interesting. so wecan look at the regulatory elements of the different cell states, right? so this is allfrom the same gscs. here are the naive sort of proliferative gscs. and here are the persisters,right, that after they're going -- they're

cycling very slowly, they're pdgfr-independent,you can see enhancers that are common in various different patterns. and then there's thisbig group here that seems to be quite a bit stronger. and the persisters -- i think what'skind of striking about this is if you go back and look at the tumors, the tumors actuallylook much more like these persisters. so in a sense these rtk inhibited persister cellsthat are very slowly cycling, actually look a lot more like the primary tumors of thegsc -- the primary glioma tumors, okay, so that we might be doing a better job of emulatingthem. i want to go back and understand a littlebit, their circuitry and their regulation. on many levels, these persister cells lookmuch more like tumor cells, and they look

in particular like those tumor cells thatare quiescent, are non-cycling. and i'll just give you one example that we were kind ofexcited about that's, i think, relevant to this conference. this is kind of a complicatedgraph, but basically we're comparing the primary tumors -- different cells in the primary tumors.and at the top here are genes that are associated with the quiescent tumors, they're anti-correlatedwith cell cycle. here we're sort of comparing the naive gscsto the dormant gscs, and here are genes that are high in the quiescent, drug-tolerant onesthat we selected, okay? so things up here are high in the dormant non-cycling tumorcells and they're also high in the drug tolerant, non-cycling gsc model. what are the genes?well we were pretty struck that this is just

packed full of kdms, lysine demethylases,genes that encode these enzymes that remove the methyl marks off of histones. you mightnotice that in contrast the methyltransferases actually are down regulated in these drug-tolerantcells. we thought this was pretty interesting here that these kdms are being up-regulated.in fact, kdms have been -- particularly kdm-5 here, has been previously implicated in drug-tolerantcancer cells. let's see, i'll tell you just -- i don't havea slide here, but we spent a while looking at kdm-5 actually, and it turns out that kdm-5does not seem to be the critical player here. we can inhibit kdm-5, we can knock it down,we can actually pharmacologically inhibit it, but the cells don't care, the persistercells don't care. so they don't seem to be

dependent on gbm -- on kdm-5. this is theone that removes lysine for methylation [spelled phonetically], okay? but there's another set and there are thesekdm-6 and kdm-3 families that remove repressive marks that actually attract that seem to bea little more interesting. i'll tell you why. first of all, if you look at k27me3, repressivesort of polygon peaks, in the naive cells you see a pretty good signal, and then asthese cells transition into this drug-tolerant persister state, there's a global erasureof k27me3. you can actually see that -- let's see if i'm going to highlight it -- you cansee that here in the socstulocus [spelled phonetically], where you see a pretty goodsignal for k27me3 in the naive and then it

kind of goes away a little bit, in the persisters.but it's a very global effect. so you're erasing this repressive mark. you're doing the samething maybe equally dramatically to k9me3, another repressive mark, and i think you cansee a region here where you've got some pretty strong k9me3, and the persisters sort of haveerased that and it's very flat, okay? there's a kdm6, which is the one that removesk27me3. there's an inhibitor, gskj-4, it's not a perfect inhibitor, but it actually seemsto work pretty well because if you look at the naive cells compared to the persistercells, and then the persister cells treated with the drug, the k27me3 comes back up. sothe drug sort of blocks this demethylation effect, okay? and actually, what was reallystriking is that these persisters are really

in red here you can see that these long termpersisters are very sensitive to gskj-4, so they're very sensitive to an inhibition ofthe kdm6 at very low concentrations, suggesting that this erasure of these repressive marksmay be essential for these cells to transition into this drug-tolerant, quiescent state. so why do you need to do this? what's goingon? well, the persister cells have an altered developmental program, as you might suspect.so this is just looking at comparing naive to persister for k4 methylation and for k27acetylation and for k27me3. and we see a number of loci where there is k27me3 sort of silencinggenes in the naive cells, and then this gets erased in the persisters, and up comes k27acetyl, up comes k4me3. we looked in general

at the pathways. there are a couple thingsthat come on. one is very primitive neurodevelopmental genes that are associated with primitive neuralstem cells. the second is notch target genes. and in fact there's some very interestingliterature in the neural development field that suggests that in radioglio [spelled phonetically]development, and very primitive embryonic development of neural stem cells and theirtransition to transient amplifying cells, notch and rtks are sort of antagonistic andhaving a complex interaction. so the most primitive neural stem cells thatare quiescent have notch. the transient amplifying cells get the signal from egfr that causesthem to rev up and proliferate a little bit. this is happening in a very narrow windowof development, but we think this may be emulated

in glioblastoma. and, in fact, if you nowgo back and think about notch signaling, you can see that notch signaling is actually activatedin the persisters. you can just do western blots to stain for active form of notch. youcan see all the gene expression signatures associated with notch signaling. and the persistercells, but not the naive cells, are very sensitive to gamma secretase inhibitors, which targetnotch signaling. so this kind of leaves me here, as i'll wrapup, with this picture here. we think that the persister cells are sort of emulatinga quiescent neural stem cell state whereby these -- we're actually starting with thesenaive gscs that are proliferating and rtk-dependent, but through demethylation and perhaps a resettingof their epigenetic state, we get these persister

cells that are very slow-cycling, they up-regulatevery primitive stem-ness genes and they're notch-dependent. as i mentioned, this parallelsan event -- or sort of inversely parallels an event -- in normal development wherebyneural stem cells that are quiescent have not signaling and switch to an amplified statewith rtk signaling. we're very interested in how kdms are working,we think, to sort of help allow the transition between states or back to this primitive stateand allow particularly erasure of repressive marks or a repressive state can allow theright enhancers to come on and sustain this state. we're also very interested in whethernow, given the observation that the primary tumor cells look a lot more like this persisterstate and that they're very slow-cycling,

suggest that perhaps notch inhibitors or,better yet, inhibitors of some of the chromatin enzymes, might synergize with rtks and giveyou a better chance of targeting some of these glioma cells. so my final slide, i've shown you the enhancerlandscapes distinguish these tumor-propagating gscs from conventional cell lines that don'tpropagate tumors and don't have this functionality. i've shown you that we can identify core tfsthat are sufficient to reprogram gscs and these sort of define -- their targets defineessential drivers of the stem-like state. i've also told you that there's a large-scaleerasure of repressive marks seems to allow gscs to adopt a more quiescent, drug-tolerantstate that we think is emulating the primary

tumors, and how this may have implicationsfor glioma therapy that targets this other population, which we think is completely missedby existing therapies that -- all of them, because they all target proliferative cellstates. acknowledge mario suva is now a faculty atmgh, formerly a post-doc in the lab, led the reprogramming work. brian liau, gem sievershave been working on some of the persister cell story. the single-cell work was a collaborationwith itay tirosh and aviv regev at the broad institute and the klarman observatory, andwith mgh neuro-surgery, and chuck epstein and norm shoresh at the broad have been criticalfor a lot of the chromatin mapping work that we do. glad to take questions.

[applause] female speaker:[unintelligible] dependent pathways for these gmb cells. are you saying that now there isan epigenetic understanding or basis for oncogene addiction that they are able to target withsome of these drugs? bradley bernstein:let's see. so -- i mean, i think there's a point to this, that, you know, when cellsbecome dependent on oncogene for a particular state, you know, they will respond favorably.and i think that model holds very well for that eight percent of cells that are proliferatingrapidly in the tumor. the model doesn't hold so well for the other cells in this tumorthat are out of the proliferative state. cml,

cml oncogene, it works perfectly. in cml,99.9 percent of this chronic leukemia, 99.9 percent of the cells are in this proliferativestate, are driven by bcr-able, and so if you give them glivec [spelled phonetically] youpractically hear them. there's about .1 percent of the cells still in the bone marrow, youprobably need to give them glivec for a while until those sort of peter out, right? butit's a cure almost. glioblastoma is the opposite, right? that fraction of oncogene-addicted cells isthat eight percent that's proliferating, or there's 92 percent sitting in this dormantreservoir or something like that. so the model's great. it just depends on the tumor and alot of tumors have a very high fraction of

dormant cells in this other state -- otherepigenetic state that can transition and that poses a huge problem to using rtk inhibitorson their own. in the clinic you have to combine it. female speaker:thank you. male speaker:great talk. quick question in regards to the state of your persistent cells, you thinkthey're under-epigenetic instability, so the global reduction in k27me3, for instance -- bradley bernstein:[affirmative] male speaker:-- do you think that that's true of all cells,

or if you do ihc, you see that there's likestill a variability in some having high levels, some having low levels, and by the chip, itall looks down because the signal from all over is negated? bradley bernstein:could it be that you have a massive amount of heterogeneity? i don't think that -- imean, it's certainly possible technically. and in the tumors, i can't even say what theylook like in the tumors. but in the actual persister population that we're studying wherei show those things being erased, i think if there was a lot of epigenetic variabilityfrom cell-to-cell, there'd be some regions with middle levels and things like that. butin regions that are really repressed and deserts

where k27me3 is sitting, it shouldn't -- male speaker:yeah. bradley bernstein:-- be coming on and off, and it's gone. male speaker:yep. bradley bernstein:so the fact that those deserts also get totally erased, and there's no rationale that youthink some population of cells is expressed in this desert, tells me it's a global effecthitting all the cells in our model. again, i don't know what's going on in the tumor.it's very hard to work with single cells from the tumor in terms of understanding theirchromatin, but it's a good question and we

need to try to understand it. male speaker:good. thanks. [end of transcript]