The Impact of Therapy on Glioma Evolution

February 09, 2024

Information

Yale Cancer Center Grand Rounds | February 9, 2024

Presented by: Dr. Roel Verhaak

ID11288

To CiteDCA Citation Guide

00:00I think we can start to get in.
00:02It's wonderful to be here this
00:05morning and to introduce somebody
00:07who is absolutely exceptional and
00:09who is newly recruited to our
00:12department in neurosurgery and also
00:14to Smilo in the Cancer Center.
00:16Dr. Rol Vierhok is Professor in
00:18the Department of Neurosurgery
00:20at the Yale School of Medicine.
00:22Following graduation with a PhD
00:24in Medicine from their Aerosmiths
00:26University Medical Center in Rotterdam,
00:28the Netherlands Role joined the
00:31Broad Institute Dana Farber Cancer
00:33Institute as a postdoctoral associate,
00:36supported by a fellowship from the Dutch
00:39Cancer Society during the time at the Broad,
00:41he was part of the team
00:42analyzing data from the TCGA.
00:44He led the Identification and
00:47Characterization of Gene Expression
00:49subtypes and glioblastoma work
00:51that resulted in a Seminole Cancer
00:54Cell 2010 publication will move
00:56to MD Anderson Cancer Center in
00:592010 to start his own laboratory.
01:02Since then,
01:03the Veerhawk lab has studied tumor
01:05evolution and mechanisms of therapy
01:07resistance in low and high grade gliomas.
01:10The group was foundational
01:12in establishing the Glioma
01:13Longitudinal Analysis Consortium,
01:15which has established a resource of
01:18molecular profiles over time on a
01:20large cohort of patients with glioma.
01:22They identified and described genetic
01:24scars and cellular phenotypes
01:26associated with glioma progression
01:28and disease recurrence.
01:30Extra chromosomal DNA amplifications were
01:32discovered as critical drivers and are
01:35now a major part of the team's research.
01:37After being affiliated with
01:39the Jackson Jackson Laboratory
01:40for Genomic Medicine in 2016.
01:42I can tell you our department leadership
01:45fought very hard to recruit him
01:48here to Yale and he joined us in the
01:51Department of Neurosurgery in 2023.
01:52Roll is a recipient of the
01:55AAAS Watchal Award,
01:56the Agilent the Early Career Professor
01:59Award and the Peter Stack Memorial Award.
02:02He's also Co founder of Boundless Bio.
02:05I can tell you in the short time that I've
02:06had the privilege of working with him,
02:08he's truly exceptional and I'm really
02:10excited for this talk and for all
02:13of the work that we have to come.
02:15So without further ado,
02:17thank you so much Doctor Veerhark.
02:20Thanks Jennifer.
02:21That very, very kind introduction.
02:23And so I joined the Department
02:25of Neurosurgery in April of last
02:29year after some discussions with
02:30Doctor Grinnell and others.
02:31And to be honest, it wasn't that hard.
02:33I was pretty convinced very quickly
02:35that this was going to be a great
02:37place to continue our research.
02:39As we were thinking about
02:40becoming more translational,
02:42it was we felt that we that being
02:43in a clinical environment would we
02:45greatly benefit our work and what
02:48what grader clinical environment
02:50and Yale School of Medicine and
02:52department of Neurosurgery.
02:53So I'm a Co founder of a biotech that
02:56won't be discussing that work today.
02:59I am also a consultant for neurotrials.
03:03So gliomas are the most common
03:07molecular tumor type in an in adult
03:11patients and the most devastating ones.
03:13They're characterized by an infiltrative
03:15growth into the environment,
03:16into the parencoa,
03:17and this makes these tumors
03:19exceptionally hard to treat because
03:20she can't go in with a knife and
03:22cut out the entire the entire
03:24structure for obvious reasons.
03:27Nowadays we recognize traditionally
03:29we would classify gliomas in
03:32adult patients by histopathology.
03:34Fortunately, we've gotten away from
03:36that as molecular markers do much more
03:39precise job in doing such classification.
03:41Nowadays we recognize gliomas based
03:44on two critical molecular markers.
03:461st, we set, we identified the presence
03:49of absence of a mutation in IDH one
03:52or IDH 2 isocitrate dehydrogenate.
03:54And for those cases that carry an
03:56IDH 1 mutation or an IDH 2 mutation,
03:58we further separate them by the presence
04:00of our absence of 1P19Q code deletion.
04:02So from some arm loss of 1P and 19 Q 19 Q.
04:06Predominantly the cases that
04:08have this code deletion,
04:10we call them codels are a majority
04:13isopathologically all it goes non code
04:15L So those cases that are IDH mutated
04:18but don't have that code deletion
04:20are in majority astrocytomas and the
04:22IDH wild type cases are mostly the
04:24glioblastomas and the patients survival
04:27patterns are according meaning that
04:29those cases that have no IDH mutation
04:32do particularly poorly clinically.
04:35That's not to say that any of these
04:38tuber types are better to have
04:39quote UN quote this as far as you
04:41can ever better have a tumor.
04:43Patients that are that carry the IDH
04:46mutant non CODEL tumors are typically
04:49diagnosed between 35 and 44 years of age,
04:52so very young in life.
04:54The codel patient typically
04:56is around 45 years of age.
04:58So again relatively early in life.
05:00So those patients might have much
05:02better outcomes but they'll mostly
05:04the majority will still succumb to
05:06disease even prior to the median
05:07age of diagnosis for IDH wild
05:09type tumors which is around 60.
05:11So these are all bad tumors.
05:15Here's the motivation for classifying
05:19gliomas by these two molecular markers.
05:22And in part it's of course
05:23it's clinically it's,
05:24it's the survival outcomes as I
05:26showed you on the previous slide.
05:27But here in this paper
05:30from the TCGA from 2015,
05:32we demonstrated that not only behave
05:34these tumors differently and respond,
05:37they they respond different to treatments,
05:39but they're really biologically
05:41different entities as reflected in
05:43the sets of molecular alterations
05:46that are commonly detected.
05:48For example,
05:49in the codel group we find that they
05:52are nearly universally carrying
05:54mutations in the Turk promoter,
05:57as well as relatively spurious mutations
06:00in genes like Notch One and NCIC.
06:04The non codels,
06:05even though they are IDH mutated,
06:07rarely contain Turk promoter mutations
06:11but are universally mutated in P53 and
06:1475% carry hairy alterations in ATRX.
06:17So very similar tumor types but
06:19molecularly quite different.
06:21And then finally IDH wild type tumors
06:24again 80% are turb promoter mutated and
06:28they are then a majority containing
06:31mutations in genes like e.g., FRCDK,
06:33N2AP10 and so on and so forth.
06:35So biological not just responding
06:37differently to treatment,
06:38not just different at in terms
06:40of at when they present in life,
06:42but biologically also quite different.
06:47Now after we were able to refine the
06:51classification of gliomas in adult patients,
06:54a major next a major challenge
06:55continues to be how these tumors
06:57respond to treatment and the lack of
06:58new treatments coming into the clinic.
07:03Now tumors initiate from a cell
07:05of origin and over time as these
07:07cells respond to challenges
07:08in the tumor microenvironment,
07:10for example presence of oxygen,
07:12lack of nutrients and so forth,
07:14you'll find that intratumoral
07:16heterogeneity starts to develop.
07:18And this is a consequence of
07:20these evolutionary processes and
07:21clonal selection where some cells
07:23are better able to deal with
07:25these limitations than others.
07:27Therefore they become they they
07:30they show clonal outgrowth.
07:31So at the time of diagnosis we're
07:33dealing with an with an heterogeneous
07:36tumor with different sets of tumor
07:38cells marked by specific and mutations
07:39and other kinds of gene alterations.
07:43Now critically this process doesn't
07:45end a diagnosis of course we impose
07:48treatments onto these tumors.
07:49You know surgery initially debulking
07:52surgery combined with radiation and
07:55chemotherapy which for gliomas in
07:57majority is stemozolamide, stemozolamide.
08:01And of course these treatments
08:04continue to impose these bottlenecks
08:06onto the tumor and those cells
08:09best able to deal with radiation,
08:11best able to deal with chemotherapy
08:13are the ones that are going to fuel
08:15the outgrowth and the tumor recurrence.
08:18So we felt that this would be,
08:19this would be an important process
08:22to study so that we could try and
08:24make these treatments more effective
08:26and potentially identify targets
08:28for new treatment development.
08:32So with that in mind,
08:33we started the Glioma Longitudinal
08:35Analysis or Glass Consortium in 2015.
08:39The glass consortium has set out to
08:42developed on the tail ends of the
08:44TCGA and it set out to develop a
08:47comprehensive molecular reference data
08:48set from pairs of tumors obtained and
08:52diagnosis and then after treatment,
08:53so the first tumor recurrence.
08:55But in reality we have been collecting
08:57tumors along the whole trajectory.
08:59So we have cases now for glass where we
09:02have 6 recurrences consecutively and
09:04we've molecularly characterized them.
09:06In other words, we've sequenced them.
09:08And then critically for glass,
09:10we really try to curate and obtain
09:14clinical annotation for all cases in
09:17the cohort because the value of a
09:20resource like this is significantly
09:22amplified if we know which tumors
09:24got treated in between time plans.
09:26Now why that we needed to do
09:28a consortium for this,
09:30it's because of things like patient
09:32mobility and the way tumor banks work.
09:34If you go to your average tumor bank,
09:36surely you'll find some tumors for which
09:39there's multiple time point specimens.
09:41But even for MD Anderson where I
09:43used to where we used to be one of
09:46the largest centers in the country,
09:48we were limited to a few dozen
09:50cases where we would have these
09:52multi time point specimens.
09:53And then as you're dealing with
09:55attrition due to tissue quality,
09:56at the end of the day,
09:58you really need an international
10:00collaboration to get to the large enough
10:03numbers to do any kind of robust analysis.
10:07So we started the consortium also
10:09still being enthusiastic of how well
10:11the TCJ collaboration went for us
10:13and now have developed a consortium
10:15that involves over 140 people
10:17spread across the globe essentially
10:19in 14 different countries.
10:21This is an older picture actually.
10:23If you would take a picture now,
10:24it would fill the room.
10:27So an important purpose of Glass
10:31is not just to create this,
10:33this data set but also to share it
10:36broadly just like the TCA so that
10:38not just we can do interesting
10:40analysis with them but of course
10:42that the whole community can do so.
10:43So in 2022 we released our latest
10:46public version of the glass data
10:48resource which is a cohort of 300
10:50/ 300 patients for whom we have
10:53collected multi timepoint DNA
10:55sequencing and or RNA sequencing.
10:58Now these 300 patients and I can tell
11:00you in the meantime now we're 2024,
11:02we've nearly doubled the cohort size
11:04and we'll be releasing that soon.
11:06So we continue to actively expand
11:08this cohort
11:11Right now in our code if you
11:12go to the URL shown here,
11:14you can find variants in clinical
11:16annotation for over 300 cases of
11:19which in majority are from IDH wild
11:22type tumors followed by the non
11:24Codells Finally the Codells as you
11:26can see we have a relative under
11:28representation of Codells here and
11:30that's likely due to the longer
11:31time to recurrence or Codell tumors
11:33compared to IDs wild type tumors.
11:35The shorter the time to recurrence,
11:38the higher the likelihood that two tumor
11:40specimens will end up in the same tumor bank.
11:42In addition to that RE resection is not
11:44standard for any of these and so that's
11:47another factor that comes into play here.
11:50You can maybe appreciate that the
11:52median age of diagnosis in these
11:54three groups is relatively young.
11:57And that is because in order to end up for
11:59data to end up in our in our resource,
12:02the patient has to have had two surgical
12:06procedures in order to obtain specimens,
12:08meaning that the patient has had to
12:10be in relatively good shape to be
12:12able to undergo those procedures.
12:13So we see a bit of a bias in
12:15median aid to diagnosis as well as
12:18in survival patterns shown here.
12:20That is what it is.
12:21We can't really address that.
12:23We try to address it by expanding
12:25the resources large as we can so
12:27that we kept capture as many patient
12:29groups as we can.
12:30Finally, I'm going to point out that
12:32our annotation in my opinion is really great.
12:34So we know for all patients whether
12:36they or nearly all patients,
12:38whether they have received tamizolamides,
12:40yes or no and whether they have
12:41received radio radiation therapy,
12:42yes or no.
12:43And we've got many more clinical variables.
12:46I just chose to highlight these
12:47on this on this particular slide.
12:50So I, in my opinion,
12:51it's really becoming a phenomenal resource.
12:55Now what can you do with a
12:56resource like this?
12:57I think you can do many things.
12:58But we initially started,
13:00we initially focused on
13:012 important questions.
13:02First,
13:03what is the impact of temozolomite on
13:05tumor evolution and on these gliomas?
13:08And 2nd,
13:08what is the impact of radiation?
13:13So treatment with temozolomite.
13:15Temozolomite is a DE alkylating agent.
13:18The repair process of the DE
13:20alkylation shows up in can show up
13:23as mutations and nucleotide changes.
13:26Nucleotide changes can conveniently
13:28be detected using sequencing,
13:30and that means that in a subset of tumors
13:32that are treated with temozolomide,
13:34A hypermutation phenotype will develop.
13:37So these are tumors where cells
13:38have been able to overcome the
13:40damaging effect of temozolomide,
13:42and they do so by repairing the
13:44damage caused by temozolomide,
13:45and the damage is then showing
13:47up as hypermutation.
13:48Very high mutational burdens
13:52across our cohort and this is slightly
13:54older version of our data set.
13:56But across our cohort we then see
13:58that when we compare mutational
14:00burden of the initial tumor and the
14:02post TMZ treated recurrent tumor,
14:04so this is only TMZ treated cases,
14:08we see very high mutational burdens in
14:10these recurrences and this is a log scale.
14:11So we chose a cutoff of 10
14:14mutations per megabase here.
14:17So across the three subtypes of glioma,
14:20we see that a subset recurs as hypermutated.
14:24The relative frequencies differ by subtypes,
14:27ID 12 type tumors 15 to 16%.
14:30For the IDH mutant tumors,
14:32we see that the rate of hypermutation
14:35development is much higher.
14:37We think this is due to the fact that
14:39IDH mutated tumors take longer to recur.
14:42Therefore, there's a more of an
14:44opportunity for hypermutation to develop.
14:50Hypermutation has been associated
14:52with relatively poor outcomes.
14:54What we found when we EPL evaluated
14:57the presence of hypermutation
14:58in TMZ treated tumors and then
15:00compared to time to progression,
15:02that it's actually similar
15:04between non hypermutated.
15:05So tumors that did not become hypermutated
15:08versus those that did become hypermutated.
15:10So time to progression doesn't really
15:13depend on the development of hypermutation,
15:15but once a tumor has become hypermutated,
15:18so after that recurrence the hypermutators
15:20do worse than the non hypermutators.
15:26That's not to say that we shouldn't be
15:29treating these patients with tenozolamide.
15:31Clinical trials such as the CAD non study
15:34from the ERTC have clearly demonstrated
15:36that tenozolamide has significant
15:38benefits across the patient population.
15:41So even though sometimes people will argue,
15:43well temozolomite causes hypermutation,
15:45hypermutation is bad.
15:47As a patient group,
15:49temozolomite is clearly beneficial.
15:56Emma and Kevin in our lab then chose
16:00to study similar questions but then
16:02for response to radiation therapy which
16:04causes a different type of DNA damage.
16:07It causes single single strand breaks
16:09as well as double strand breaks.
16:14And Long story short,
16:17they discovered that when comparing
16:19cases not treated with radiation to
16:22those that are treated with radiation,
16:24that treated cases develop a
16:27relatively high number of small 2
16:30to 20 base pair deletions across
16:32their scattered across their genome.
16:35So it's a bit of a similar
16:37phenomenon to hypermutation,
16:38but instead of single nucleotide changes,
16:41we found that radiation drives small
16:44deletions and in the treated cases we
16:46see significantly more small deletions
16:48arise compared to the untreated cases.
16:53Now radiation and temozolomide are
16:55often used in combination its standard
16:57of care for IDH wild type tumors.
17:00So are we observing this increase in
17:02small deletions because some of these
17:05tumors will are developing hypermutation?
17:11The answer is yes and no.
17:13Meaning that when we split up our cohort
17:16in those cases that are hypermutated
17:18as well as radiation treated,
17:20we find that hypermutation
17:22independent of radiation actually.
17:24So these are tumors that are hypermutated
17:26and have not been treated with radiation
17:28that they also show an increase
17:32in the number of small deletions.
17:34But importantly,
17:35those that are not have been mutated
17:37and have been radiated also show
17:39that small deletion increase.
17:41So hypermutation and radiation
17:43are independent factors driving
17:45the increase in small deletions.
17:48And in that sense small deletions,
17:50the small deletion increase
17:52burden increase is comparable to
17:54hypermutation for actemozolamide.
17:56And in our paper we actually found
17:59that this is true across cancers,
18:00not just gliomus,
18:04Gemma and Kevin and also evaluated
18:06anuploidies, in other words,
18:08broad losses and gains.
18:09So while small deletions will
18:11arise from double strand breaks
18:13that are subsequently repaired,
18:16anuploidies typically are a
18:17result of cell cycle of errors.
18:19During the cell cycle,
18:21for example MIS segregation,
18:25we compared gains, broad gains and
18:28broad losses and we did that between
18:31irradiated cases and unirradiated
18:33cases and found no difference in gains.
18:37But we found a significantly higher
18:39number of whole chromosome arm losses in
18:43irradiated versus non irradiated tumors.
18:45Similar to the small deletion increase,
18:50now homocygous deletion of CDK
18:51into A which is of course a cell
18:55cycle regulator has previously been
18:57associated with tumor progression
18:59especially in Ida mutant tumors.
19:03We compared not just homozygous
19:05deletion but also hemisygous deletion
19:07of CDK and to a first in untreated
19:10initial tumors in glass this is
19:12codels and non codels and this is
19:14focusing on the IDH mutant tumors,
19:16Codels versus non codels first
19:18and we find that non codels have
19:21a higher rate of CDK and to a
19:23homozygous as well as semisygous loss,
19:26but this is particular particularly
19:28pronounced in recurrences.
19:30So at recurrence non codal IDH
19:32mutant tumors significantly show
19:34a significant increase in the
19:37number of CDK N to a homozygous
19:38as well as hemisygous deletions.
19:41And this is then particularly true
19:44amongst irradiated tumors suggesting
19:46that there's a relationship between
19:48irradiation and CDK N to a loss.
19:53And again the presence or the when
19:56a CDK N to a loss either hemisygous
19:59or homozygous has been acquired,
20:02we see that that correlates associates
20:04with worse outcomes to treatment.
20:10We then evaluated the association between
20:13these broad aneuploidies and CDK into a loss.
20:17Here we grouped a bunch of cases.
20:19Actually this is not class data,
20:21this is span cancer data.
20:22We grouped those cases by non irradiated.
20:26Palliatively irradiated.
20:27So lower doses and accuratively irradiated
20:30tumors and find that the increase in
20:34the number of chromosome losses is
20:36actually only found in tumors that
20:38have homozygous deletion of CDK into A.
20:40So while we previously showed
20:43that irradiation appears to
20:45drive broad chromosome losses,
20:48what we're actually seeing is that
20:51irradiation associates with CDK into a loss,
20:53and it's really the CDK into a loss
20:56that then associates with aneuploidy
20:59because we're seeing a significant
21:00increase in the number of chromosome
21:02losses in unirradiated cases when
21:04homozygous loss of CDK into A is present.
21:08So irradiation itself does not
21:10appear to drive aneuploidy.
21:11It appears to drive CDK into a loss,
21:13which then drives to aneuploidy.
21:17And as with the hypermutation example,
21:21we're finding no significant
21:22difference in surgical interval in
21:25irradiated cases associating with the
21:28number of acquired small deletions.
21:30But post recurrence,
21:31those tumors that acquire the most,
21:33the highest number of new small
21:35deletions are the ones that stop
21:37responding to further therapy
21:39that have very poor outcomes.
21:41So acquired small deletions are a marker
21:43for further tumor response if you will.
21:52So this is the model that
21:53seems to arise from our data,
21:54which is perhaps not super surprising.
21:57These tumors as I started with
22:00originate from a cell of origin
22:02that starts to expand more
22:03quickly than the cells around it.
22:05And upon bottlenecks in
22:07the tumor microenvironment,
22:08subclones will further arise that
22:10then at the time of the diagnosis
22:12can be detected through sequencing.
22:14Then we impose this therapeutic
22:16barrier for surgery and then a chemo
22:19and radio and those cancer cells that
22:20are able to repair the DNA damage.
22:22So the ones that have the hypermutation
22:25phenotype or the ones that have acquired
22:27large numbers of small deletions,
22:29those are the ones that are
22:30able to repair the A damage and
22:32can subsequently be detected,
22:34have expanded sufficiently,
22:35have become large enough subclones that
22:38we can detect them through sequencing.
22:40So then it's not surprising that
22:42in recurrence these tumors that
22:44have these genomic scars,
22:45these signatures are the ones
22:46that have poor outcomes and stop
22:48responding to treatment because
22:50you're looking at cells that have
22:52already been able to have already
22:54shown that they don't care about
22:55further DNA damage or they don't
22:57care about chemo radiotherapy.
23:01Now when you summarize these
23:03numbers across our glass cohort,
23:07we see that amongst IDs wild type
23:09tumors that have been treated with
23:11temozolomide and or radiation that 15%
23:14develops the hypermutation phenotype.
23:17And of those that are that do not
23:19develop the hypermutation phenotype,
23:20another 16% requires large numbers
23:23of small deletion which leaves a
23:26relatively large group in which no
23:31genomic scars can be detected,
23:33A recurrence and they may have
23:35intrinsic mechanisms to deal with
23:38the toxic effects of therapy.
23:41When we look at IDH mutant
23:43tumors where the picture is more
23:45diverse because not all patients
23:47will receive the same therapies,
23:49we find that amongst those that have
23:50been treated with temozolomide,
23:5242% acquires hypermutation,
23:5335% of non hypermutators acquires
23:56the small deletion phenotype and
23:59again a subset shows neither.
24:04Now as Jen mentioned in the intro and
24:07previous work we have looked at gene
24:09expression patterns and this is focusing
24:12on GBM so IDH small type tumors.
24:14And we found that when we evaluate gene
24:17expression patterns we can and identify 3
24:20gene expression subtypes of glioblastoma,
24:22IDH well type glioblastoma which we labeled
24:24mesenchymal per neural and classical.
24:28When we evaluate A subtype classification
24:31in glass we see that you know we
24:34see the the relative distribution
24:35is of these three subtypes is
24:37what we typically would expect.
24:39A number of cases are classical
24:42mesenchymal or per neural at recurrence.
24:44We do appear to see a minor
24:47shift towards mesenchomal tumor.
24:48So we see a high number higher number
24:51of mesenchomal tumors and recurrent.
24:52So tumor progression appears to
24:56and some tumors coincide with
24:58mesenchomal transformation.
25:00But perhaps more importantly,
25:01we find that these subtypes are
25:04these subtypes identifications.
25:06Classifications are quite flexible
25:07because nearly half of our cases actually
25:10change subtypes between the initial
25:12time point and a recurrent time point.
25:17Now, like many tumor types,
25:18glioblastoma has been extensively
25:20studied using single cell sequencing
25:23and single nucleus sequencing.
25:25Our collaborator Mario Suva has let let
25:27the field in this respect and published
25:30a very influential paper in 2019,
25:32the Neftal ET al.
25:33Study which they found four
25:36predominant cell states of
25:38glioblastoma which they labeled
25:40oligo progenitor cell like neuro,
25:42progenital cell like astrocyte
25:45like mesenchymol like.
25:47You may notice that these terms are
25:49actually reminiscent of the subtypes
25:51that we had previously identified
25:53and that's maybe not surprising.
25:55If a tumor contains a majority
25:58mesenchymal like cells,
25:59the subtype signature will be mesenchymal.
26:01We've shown this using combined bulk
26:03RNA C and single cell RNA C data sets.
26:08Oh, and another important thing to remark
26:11here is that all GBMS each of the cell
26:14states can be detected in all GBM's.
26:16It's just a shift in numbers which
26:20perhaps explains the plasticity
26:22of expression subtypes over time.
26:24Now my lab, Kevin and Kevin Johnson
26:27and Kevin Johnson and Kevin Anderson
26:28have worked together to do single
26:30cell sequencing of gliomas as well
26:33with the purpose of identifying
26:35pan glioma cell states.
26:37Mario's Neftal at all cell states
26:39are focused on IDH well type tumors.
26:41Our effort initially focused on
26:43identifying cell states that could
26:46be identified across different
26:47types of glioma and we found those
26:50we labeled them stem like cells,
26:51proliferating stem like cell and
26:53differentiated stem like cell.
26:54And of course with single cell sequencing
26:56you can also identify non malignant
26:58cell states such as elutical dendrocytes,
27:00parasites,
27:00myeloid cells, cells,
27:01cells that are typically residing in
27:03the microenvironment of these gliomas.
27:08Now you can take, you can infer
27:11signature gene signatures from these
27:13single cell States and you can then
27:16use computational methods to project
27:18those signatures on bulk RNA C datasets
27:20such as the ones we have in glass.
27:23So you can use single cell signatures to
27:26deconvolute bulk expression profiles.
27:31So Fred Verne, former post doc in the lab,
27:33now a faculty member at the Jackson
27:35laboratory, took this approach,
27:38used the single cell inferred gene
27:41signatures from the Kevins and projected
27:44those onto the glass RNA C data sets.
27:48This is showing on the left IDH.
27:49Well the summary of IDH wild type tumors.
27:52On the right the IDH mutant
27:55tumors primary and recurrences.
27:57So when we aggregate all the data,
27:59first starting with IDH wild types,
28:02when we aggregate the presence of the
28:05single cell signatures across the cohort
28:08and compare initial to recurrent tumors,
28:11we do not find major shifts in our
28:14Panvama cell state cell state presence.
28:18Actually the major difference
28:20we found when comparing initial
28:23tumors to recurrent tumors is the
28:26relative fraction of oligodenrocytes
28:27in the tumor microenvironment.
28:32And maybe this is not too
28:35surprising if you look at the the
28:39invasive margins of these tumors,
28:41this is at the time of recurrence.
28:43This is where most of the tumor cells will
28:46have come from because this is the area
28:48of the tumor that's difficult to resect.
28:50So perhaps at recurrence you can
28:52imagine that at recurrence more
28:54of that margin is cut out.
28:56Therefore, we might be able to see more
28:59cells from that micro environment,
29:01in this case particularly oligodenvrosites.
29:06What what did seem more surprising
29:09to us is then again when Fred used
29:12computational methods to not just
29:14count enumerate the types of cells in
29:18primary to recurrent tumors but also
29:20looked at the actual gene expression
29:22profile of those cells that we found
29:25that a significant increase in
29:27neuronal signaling pathways amongst
29:29the malignant cell population.
29:32Of course you'll find neuronal
29:33signaling and oligodendrocytes but
29:35what we were finding is that also
29:37the malignant cells activate neuronal
29:39signaling pathways as recurrence.
29:42So we're seeing an increase in
29:45oligodendrocytes in the microenvironment
29:46but that appears to be converging
29:48with increased levels of neuronal
29:51signaling by the tumor cells.
29:54And when we use a public data set
29:57consisting of multi biopsy single cell
30:00RNA sequencing from glioblastoma patients,
30:03we could again confirm that the
30:06malignant single cells expressed
30:08higher levels of neuronal pathways
30:12when they were when the biopsies
30:14were obtained from the margins of
30:15the tumor relative to the core of the
30:18tumor confirming what Fred had found
30:20in our bulk analysis from glass.
30:25I'm actually going to skip this one.
30:27We decided this to then take
30:29this one step further in a large
30:31collaboration that involved Mario Suva,
30:33Itai, T Rush, Antonio Yavarone,
30:36Anna Lazarella as well as many
30:39postdoc and junior leads in
30:41in in these respective labs.
30:43Collaborating with MD Anderson,
30:44Duke and a number a number
30:46of other institutions
30:50we acquired. We generated longitudinal
30:52single nucleus RNA seed data for a large
30:55number of IDH wild type glioblastomas,
30:57again in the context of annotation for
31:00different types of therapy and we also
31:02were able to generate exomorhol genome
31:04sequencing on the majority of our core.
31:09So previously Mario and colleagues
31:14identified these four cell states that I
31:16mentioned earlier, NPCOPCACMS like cells.
31:21When we analyzed over 500,000 cells from
31:24this cohort and again to derive cell states
31:28as well as transcriptional meta programs,
31:31we find these same 4 cell States and the gene
31:34express's check features that come from them.
31:37Again, here is OPC, AC,
31:40mesenchymal, and NPC.
31:41But of course we found many more
31:44because of the much larger cohort
31:46as well as because in his Mario's
31:48initial study he had only untreated,
31:50he included only untreated tumors.
31:51And now we're looking at
31:54both primary and recurrences.
31:55So our large data set enabled
31:57us to find the number of new
31:59glioblastoma related cell programs.
32:06As we had also observed
32:09in our glass analysis,
32:10the relative number of malignant
32:13cells decreased at recurrence.
32:14So recurrent GBMS become less
32:17pure or more incorporate more
32:20tumor microenvironment cells.
32:21So we see a decrease in the number
32:24of proportion of malignant cells
32:25and an increase and we confirmed
32:27the increase in the number of
32:29oligodendrocytes that's because of
32:30the greater resolution of the single
32:32cell of the new single nucleus data.
32:33We also see a significant increase
32:35in the number of astrocytes in
32:37the number of neuronal cells
32:42converging with the result from glass
32:45that most tumors or many tumors
32:47change tumor subtype at recurrence.
32:50We find large shifts in cell states
32:54between primary and recurrent tumors
32:57and the one that maybe is interesting is
33:00hypoxia and I'll get back to that later.
33:03So a subset of this smaller color compared
33:07to glass acquired this small deletion
33:09phenotype that I mentioned earlier.
33:10In fact 10 of 46 tumors where we had
33:14converging DNA sequencing and single
33:17nucleus sequencing data acquired this
33:19small deletion phenotype as shown here.
33:22And what we're seeing is that when a small
33:25deletion phenotype has been acquired,
33:27tumors will increase.
33:29We find that more tumor cells show signs
33:32of hypoxia are responding to hypoxia.
33:37So radiation either drives or
33:42shows its most significant effects
33:44in cells in regions of hypoxia.
33:50When we then went back to our glass
33:52data sets, we could confirm that
33:54indeed tumors that acquire lots of
33:56small deletions are also showing
33:57an increase in hypoxic signaling,
33:59hypoxia cell state signaling compared to
34:01tumors that do not acquire small deletion.
34:07This is potentially relevant because
34:09hypoxia is a phenomenon you can
34:11detect through imaging and of course
34:14these results built upon large
34:16and large amount of literature
34:18demonstrating the convergence of
34:20radiation response with hypoxia.
34:26We then also focused,
34:28we then also performed A comparable
34:30analysis but looking at IDH mutant tumor.
34:32So we generated single nucleus
34:34RNA CC data on IDH mutants,
34:38a cohort of 35 cases and this is
34:40work led by Kevin Johnson who is a
34:42research scientist in our lab again
34:44with the same collaborator team.
34:51Mario Nita's labs have previously
34:54found that in IDH mutant tumors they
34:58they found less consistent cell cell
35:01state and gene expression programs.
35:02But they found that all that most tumors
35:06could be projected along an axis of stem
35:10like to stem like Sigma from stem like
35:14states to a more differentiated state.
35:17Because IDH mutant tumors in general are
35:20either astrocytoma or oligodenra glioma,
35:23They showed that astrocyte
35:28IDH mutant gliomas differentiate
35:29from a stem like cell to a astrocyte
35:33like cell phenotype whereas
35:35oligodenrocytes go from a stem like
35:37state to a more oligodenrocyte like
35:40state as potentially expected.
35:41So the non codels go to the left
35:44and the codels go to the right.
35:48In our paper from 2022 with
35:50Fred as first author,
35:51we had noticed that amongst IDH mutants
35:56those that show signs of treatment
35:59response either through hypermutation
36:01or through acquired sydicand to a loss,
36:04we saw an increase in the proportion
36:07of proliferating stem like cells
36:09which would be in the in the trunk
36:12of the axis shown on the left.
36:15So we took those results and we
36:17took those into consideration as
36:19we started to analyze these data.
36:21So first we Kevin generated these U
36:24maps that you can see in many papers.
36:26We had generated data from large
36:28numbers of nuclei,
36:29I would say quite unprecedented to show
36:32that you can infer your typical sets
36:35of cell state programs as shown here.
36:38So we have now generated a very
36:40large number of IDH smooth and single
36:42nucleus data and through that we can
36:45create a definition of cell States and
36:47associated metaprograms of IDH mutant tumors.
36:50And the metaprograms we arrived at
36:54actually are quite reminiscent of those
36:56that are shown in IDH wild type tumors.
37:03Interestingly, when we looked at the this,
37:05when we projected the 35 cases that we
37:09had analyzed that we had sequenced,
37:11we projected them on that same
37:13inferred Y axis as Mario and Itai had
37:16previously used in their analysis.
37:18We see that tumors tend to shift.
37:20So there's the circles here
37:21are the initial tumors and the
37:23triangles are the recurrent tumors.
37:26We see that nearly all tumors
37:28shift into an upward direction.
37:31So the relative amount of stem like or stem,
37:34the relative amount of stemness
37:36in these tumors almost universally
37:39increases upon recurrence.
37:41So tumor seems to de differentiate
37:47as a part of their tumor progression.
37:53This is also shown here.
37:54These are different some of the different
37:57meta programs that we had arrived at
37:59looking at different grades amongst
38:01Codells and non Codells and we see
38:05that a differentiated differentiation
38:07cell state such as the AC like
38:10state decreases upon with grade.
38:12And that's true for both Codells and non
38:16Codells whereas undifferentiated and
38:18number of cycling cells increases with grade.
38:23And when we actually pair up the tumor,
38:27so not split them by grade but
38:29actually look at paired samples
38:31again we confirm that the amount of
38:34undifferentiated cells increases,
38:36the amount of cycling cells increases,
38:38cells that show signs of stress increases
38:41in proportion and finally mesenchymal
38:43like cells increase in proportion.
38:49Now what is in my view most interesting
38:53about these observations is when we now
38:56separate our you know relatively modest
38:58cohort but still into cases that have signs
39:02of therapy induced genetic alterations.
39:05But those are the ones that also
39:07are also those are the ones that are
39:09driving these significant changes.
39:11So tumors that do not reflect therapy induced
39:15genetic alterations such as hypermutation,
39:17citic anti a loss or loss of new
39:20aneuploidies, we find no significant
39:22difference in cell states.
39:24It's only those tumors that show a
39:27treatment induced alteration that
39:28are the ones that also show changes
39:30in their gene expression programs.
39:36So to then re annotate the
39:39flow charts I showed earlier,
39:42we're seeing that subsets of IDH
39:44mutant as well as IDH wild type
39:47tumors acquire genetic alterations
39:48in response to treatment and we're
39:51now finding using our bulk and
39:53single nucleus our expression
39:55gene expression datasets that
39:57this coincides with increased cell
39:59cycle activity and proliferation.
40:02And D differentiation programs
40:04and IDH mutant glomus but with
40:08neuronal and mesenchomal signaling
40:10activity and IDH wild type tumors.
40:13So it leads me to summarize at the end here.
40:17IDH wild type gliomas so far seem to
40:20undergo tumor cell extrinsic changes
40:22which sets them apart from IDH mutant
40:25glomas which appear to a majority
40:27undergo tumor cell intrinsic transitions,
40:30which I think is a peculiar
40:32but interesting difference
40:35As we think about developing new
40:38therapies for these patients,
40:39this is something to take into consideration.
40:43And finally amongst the IDH mutant clairomas,
40:47the changes we are observing are mostly
40:49observed when in those tumors that have
40:52been treated and we find convergence
40:55between newly acquired genetic
40:57alterations with cell state transitions.
40:59So that leads to the question,
41:02are these tumors changing because
41:04of the treatment or are the
41:07oncologists treating the tumors that
41:09are more likely to change or both?
41:12That's something for a next analysis.
41:15With that, I come to the end,
41:17I'd like to thank all the people
41:19in the lab that worked very hard
41:21for these results and of course
41:22our funding our funders.
41:24Thank you very much.
41:29Thank you. Jen had to run to the OR,
41:31so I will handle the questions.
41:33Do we have any questions
41:34from the room or online?
41:38You mentioned immunotherapy,
41:40so are are there protocols now that
41:42are using some of these markers to
41:44determine who should get immunotherapy
41:45and which ones in this disease.
41:46So regrettably all the results so
41:48far I've shown that checkpoint
41:50inhibition does relative does little
41:53for these patients and that's likely
41:56because of the very immunosuppressive
41:57microenvironment in these tumors.
41:59There's very few active T cells.
42:00So you can treat them at checkpoint
42:02inhibition but without T cells that's
42:04going to not really result in any benefit.
42:07So moving forward the way to get
42:09immunotherapies to work in these patients
42:12would be to figure out how can we get
42:15T cells into the tumor and only then
42:17immunotherapy is is likely to have a chance.
42:19Got it. OK.
42:20Any questions in the room? OK.
42:21I will walk the microphone around.
42:23I'll go the front row here first.
42:28Hello. Oh thanks. A beautiful talk and
42:30I think you know just to it's something
42:33that we are all hoping to be able to
42:37replicate in different tumor types.
42:40What a what a great example of a
42:42a treasure trove of information.
42:45My question is about epigenetic regulation
42:47and I saw one slide with EZH 2
42:50your thoughts or if you've looked
42:52at sort of wrapping of chromatin
42:55epigenetic regulation specifically
42:58after radiation, if that's changed,
43:00if we can explore that with some of our,
43:02for example, ECH,
43:03two or other regulators there,
43:05inhibitors there,
43:08that's a great question.
43:11So just from a data perspective,
43:13we have been able to collect
43:15the NMS elation profiles,
43:17other members of the glass
43:19consortium have looked at those.
43:21What we see in the IDH wild type tumors,
43:23we don't see many changes from
43:25a Dena methylation perspective.
43:28These tumors have lots of things going on,
43:30but it doesn't really seem to change
43:32in directly their DNA methylation
43:34profile and the IDH mutants.
43:36We see that the subset of tumors goes
43:38from a relatively high amount of genome
43:41Y DNA methylation to a decreased amount.
43:44So and those are the ones that
43:46also are also the ones that change
43:48that acquire genetic alterations
43:49that change cell state programs,
43:51those also seem to demethylate or
43:55show demethylation genome wide.
43:57Now whether that has implications
43:59for treatment with ECH 2 inhibitors
44:01would be a bit of a stretch.
44:03I know those are being considered for
44:07the H3 wild type pediatric GBMS for example,
44:12but right now I don't have information on
44:14whether that will work for for adults.
44:17Great, we can go to Doctor
44:18crop and then doctor Contessa
44:19in the chat has a question.
44:21So we'll get him queued up
44:22to ask it verbally.
44:25Ian, very nice talk.
44:26And this question actually
44:27is a little bit similar
44:28to I think what Joe's getting at.
44:31So you've shown that in the the subset
44:34of the temozolomide treated patients
44:36developed this hypermutated phenotype
44:38and that's associated that leads to
44:40poor outcomes in those patients.
44:43It would seem that if you could
44:44potentially if you could identify
44:45those patients up front who were
44:47going to go down that path with
44:49treatment with temozolemide that
44:50you may decide it may be in their
44:52overall better outcome to avoid using
44:54temozolemide in those patients.
44:55So if you looked at baseline molecular,
44:59molecular genomic characteristics
45:01of the patients who go on to develop
45:03hypermutator phenotype to be able to
45:05if you could predict those up front,
45:07yeah. So for ID, it's wild type
45:08of course we have a great marker.
45:09It's MGMT methylation, right.
45:11So for that, I would say that's
45:14already most largely addressed.
45:15For IDH mutants,
45:16we have not looked at this very much yet.
45:19There's been another publication from
45:21a group in China that has established
45:23a large number of serial cases.
45:26They have suggested that low level changes
45:30in chromosome 8 would be predictive of
45:34risk of developing hypermutation and
45:36they link that functionally to MIC.
45:39I think that data is interesting.
45:41I think it could use some further
45:43validation now as we are expanding
45:45and working on our glass effort,
45:46a major change relative to our latest
45:50release and one we are working on right
45:52now is that we've accumulated a large
45:54amount of whole genome sequencing data.
45:55And I'm excited about that because
45:57with whole genome sequencing data,
45:58you can do things with mutational
46:00signatures and mutational signatures
46:02would reflect for example,
46:03potentially DNA damage repair processes
46:05that are ongoing in these tumors.
46:08So I'm hopeful that we can identify tumors
46:10that have DNA damage repair processes
46:12going on and that that would then be
46:15repredictive of response to demosolomide.
46:17That's all speculation.
46:18So hopefully in a year from now or so,
46:21we will have more definitive answers.
46:23Thanks.
46:23Great.
46:24Doctor Contessa,
46:24I'm told we don't have access
46:26to allow him to talk.
46:27Can I can you hear me a miracle?
46:31Yeah, this is OK go ahead.
46:32Oh, great role. That was fantastic.
46:35Fantastic talk, very exciting.
46:39So yeah, I just wanted to drill
46:40down a little bit on the the
46:42radiation induced mutations because
46:44there is this question, right.
46:46Is it that you're select that
46:49after radiation it's a selective
46:51pressure and you're winding up
46:53you know finding those those
46:56mutations that have gone on and been
46:59propagated in in different clones.
47:02And you know I think that
47:03the main question is,
47:04so if you're sequencing from a tumor
47:07and considering the stochastic
47:09nature of radiation are do you
47:11think you're going to be able to
47:13find those recurrent you know,
47:15small deletions and isn't that
47:17probably more consistent with you
47:19have a resistant clone which might
47:21be you know have adna repair defect
47:23which enables radiation resistance and
47:25so then you wind up having that you
47:29know radiation resistant clone moving on.
47:32And I and I think that's similar to
47:33what you would see with CDK and 2A,
47:35but I won't be too long.
47:36And I guess my main question is,
47:37so can you know you have these
47:39two different possibilities,
47:40Could you use a single cell analysis to
47:42analysis to try to differentiate between,
47:45right.
47:46Is it the radiation that's
47:48the cause or just the,
47:50you know that it's the
47:51the selective pressure?
47:52Yeah. Thanks.
47:53Thanks very much and and great question.
47:55So if we take hybrid mutation
47:58following tamizolomide as an example,
48:00because of the specific mutational
48:02signatures of mutations acquired
48:04after mint temozolomide,
48:06we're pretty sure that temozolomide is
48:09actually causing these these changes.
48:11And I think there's a lot of
48:14similarities between the small
48:15deletions acquired by after irradiation
48:18to the temozolomite example.
48:23One reason for saying that is that
48:24we have taken cell line models and
48:27irradiated them and then passes them for
48:2925 times or so or have made sure they
48:31went through a full cell cycle 25 * /
48:34a period of let's say 3 or so months.
48:37One of our MDP disease students
48:39has done that in the lab.
48:40And she said she showed that
48:42after about 3 months,
48:43you see a significant increase in the
48:45number of small deletions and tumors
48:47with or in cell lines with radiation
48:49versus those that have not been irradiated.
48:52And actually,
48:52and she actually spoke with your student
48:55after her exciting talk just two weeks ago.
48:57So that to me suggests a
49:00pretty strong causal link.
49:01Also, the types of small deletions,
49:04we've now made some progress
49:05in analyzing them.
49:06They carry a specific signature or they are
49:08associated with this specific signature,
49:10which again to me suggested
49:12there's a direct causal link rather
49:14than radiation causing clonal
49:16outgrowth of a particular clone.
49:21Yeah, I think that's what I wanted.
49:23Yeah, thanks. We should connect
49:25because I I have some some
49:27more comments and discussion.
49:29Great. I would love to. OK, I I just
49:31unmuted. Doctor Robinson, do you want
49:33to ask your question?
49:36Yeah, phenomenal talk.
49:36I was going to ask,
49:38I think you already answered this
49:39about the MGMT methylate if there's
49:40a difference in patterns resistance.
49:42But the other question I was going
49:44to ask is you know has there
49:46been efforts to kind of pursue
49:47synthetic lethal screens of some
49:48of these identified pathways,
49:50So CD and K things like that,
49:53That's a, it's a great suggestion.
49:56My speculation is that probably
49:57somebody has done that.
49:58We're not doing those in the lab right now.
50:01I guess you know challenges of
50:03course exist even though tell us this
50:06exists of course with getting any
50:08kind of molecules into the brain,
50:10you know if you have a target
50:12most many of our clinical trial
50:13failures that we've seen so far
50:15are actually related to blood brain
50:17barrier and and things like that.
50:19So I think your your your
50:20idea of course is very good.
50:24It'll take a little bit longer
50:26before we can actually see
50:29drugs and treatments materialize from that.
50:32And one follow up question,
50:33one thing that's always really been
50:35perplexing to me is that with EGFR 3 variants
50:37that if you put those in a Petri dish,
50:40those get selected out.
50:41So they're actually disadvantageous
50:42in a Petri dish.
50:43But obviously we see them in
50:45in real human tumors.
50:46Do you have any any kind of
50:48sense or any insights as to
50:49why those are advantageous in
50:51the real human environment,
50:52But they're not in a Petri dish,
50:54so it's hard to answer that directly.
50:58Maybe it has to do with the types of
51:01ligands that exist in the micro environment
51:04versus those that exist in a Petri dish.
51:06The the spin I would gift on this
51:08is that we find that all V3 variants
51:11exist on extra chromosomal DNAS.
51:14So EGFR is is amplified when the V3 is
51:18present and these amplifications typically
51:20reside on extra chromosomal DNAS.
51:23And there's a lot of, you know,
51:27that does a lot of things to these cells,
51:31including potentially putting a higher
51:33burden on the cells to produce all
51:36the DNA needed for the high numbers
51:38of copies that typically exist when
51:40there's extra chromosomal DNA.
51:41So that could be one reason.
51:43And in general,
51:44I think EC DNA is very potent in
51:48many ways and that probably has
51:50to do with why these are selected
51:52out more so than the V3 itself.