Towards Predictive Biomarkers in Early-Stage Her2 Positive Breast Cancer

March 04, 2022

Information

March 3, 2022

Yale Pathology Grand Rounds

Christina L. Curtis, PhD

ID7503

To CiteDCA Citation Guide

00:00OK, let's begin.
00:02Welcome everybody to
00:04pathology grand rounds today.
00:06I'm excited to introduce our speaker,
00:08Christina Curtis from Stanford.
00:10Christina comes has quite
00:12a long training history.
00:14Beginning in Heidel, Heidelberg,
00:16Germany for a Masters degree and
00:18then did her doctorate at USC and in
00:21computational and Molecular Biology program.
00:23Did a postdoc back over
00:25across the ocean in Cambridge,
00:26then came back to USC to begin
00:28as an assistant professor,
00:30but shortly thereafter.
00:31Joined Stanford as an assistant
00:33professor there and is now at the
00:36associate professor level at Stanford,
00:37and as you can see.
00:39We in pathology.
00:41Most of the people are in pathology,
00:43grand rounds or some way related to that.
00:45I've had a long appreciation for the
00:47importance of spatial information,
00:48but many of our genomics colleagues
00:51just grounded all up and I think
00:53that one of the reasons I was
00:55excited to invite Christina.
00:56She's one of those people that
00:58not only is an expert in genomics,
01:00as you can see,
01:01she's the director of the breast
01:02Cancer Translational unit and the Co
01:04director of the Molecular Tumor Board.
01:05But she's also very conscious,
01:07conscious of spatial information.
01:09And so the mixing of genomic information
01:12and spatial information is not an easy task,
01:15but I think Christina Curtis is
01:18one of the world leaders on this,
01:19even though she's still fairly
01:21junior and so rather than go through
01:23all the awards and fellowships
01:24and leadership positions she had,
01:26I'm going to let her speak for herself.
01:28And she's going to tell us about toward
01:29predictive markers and early stage.
01:31Her two positive breast cancer, Christina.
01:34Great, thanks so much for the
01:35kind introduction, David and I,
01:37I am delighted to share this with you.
01:39I hope next time. To be in person,
01:42as I'm sure we all do but but yeah,
01:45really delighted to share
01:46this sort of recent work,
01:48and I think often I have to say
01:50I've drawn a lot of inspiration
01:52from from you David in in this,
01:54as you were obviously a
01:55pioneer early in the field,
01:56and so we've been sort of waiting
01:58for these technologies to make
02:00it to a place where where they
02:02can be utilized by the masses,
02:04and so that that's what I'll speak to today.
02:05I'll just state these are my disclosures.
02:08The only point that is relevant to
02:10the discussion today is that I am a.
02:12Scientific advisor,
02:12banana string and I will discuss the
02:16mastering DSP technology and work
02:17that I had done Prior to joining.
02:20So as we all know,
02:22I'm a major objective of our current
02:26times is to affect precision oncology,
02:29and there's many pieces to this puzzle
02:33that range from really improving
02:35our understanding of prognostication
02:37through biomarker discovery,
02:38as well as predicting response to therapy,
02:41improving patient stratification,
02:42and ultimately,
02:43this goes on and actually can inform
02:46the drug development pipeline.
02:48And so there's a number of key goals.
02:50Here,
02:50mainly what I will focus on as a key
02:54area of interest from my own group is
02:56on patient stratification and really
02:59identifying aggressive subgroups of
03:01disease and tailoring our therapeutic
03:04approaches for these subgroups.
03:07Another key objective and and
03:08I will touch on this as well,
03:10is really on being able to
03:12predict response to therapy.
03:13And of course this is not only
03:15our new targeted and immunotherapy
03:17therapeutic agents,
03:18but also chemotherapeutic.
03:20Back bones that really remain the
03:23mainstay of many treatment regimes,
03:25but have been hard.
03:27A hard nut to crack with respect
03:29to prediction of response and
03:30ultimately a lot of the work in my
03:32own lab and I I won't dwell on this.
03:35I'll focus more on sort of the
03:36applications of these approaches
03:38is has been to use systems biology
03:40techniques and and I would say
03:41that one of the potential powers of
03:43this type of approach is that we're
03:45not only interested in developing
03:47predictive models or classifiers,
03:50but actually unraveling the biology
03:51so that we can.
03:53And develop mechanistic insights
03:56into into disease,
03:58and perhaps inform the next wave of
04:01therapeutic approaches and so really,
04:03what I'll talk about today is sort
04:06of a few pieces from my own labs work
04:09that have led from really omic technologies,
04:12but that are now moving
04:14towards clinical translation.
04:15Of course, across a very long road,
04:18and so the first story that
04:20I'll talk about really is about
04:22leveraging spatial approaches.
04:24In situ proteomic profiling to
04:25predict response in this case
04:27to her two targeted agents and
04:29her two positive breast cancer,
04:31and I'll describe really how
04:32we've gone about this.
04:34Actually,
04:34starting with dissociative and
04:36bulk technologies that that let
04:39us down some some harder paths and
04:41moving forward to use new technologies
04:44that are really quite emergent.
04:45So that's the first story that I'll
04:47share with you and really will be
04:49the bulk of my discussion today.
04:51Why I won't speak to some of the approaches
04:53that we've developed to for example,
04:56product, chemotherapy benefit and and
04:57these are really based on epigenomic
05:00biomarkers that have emerged from large
05:02transcriptional profiling efforts
05:03but also coupled with in vitro data.
05:06But this is of course another area of
05:08interest and I think that as we think
05:10about personalizing therapy again,
05:11we must be cognizant about how
05:14we do this for standard of care
05:17chemotherapeutic agents and really
05:19deescalating whenever possible.
05:21And of course. Escalating when necessary.
05:24I will try to close and and really
05:26speak to some of our other efforts
05:29that have been leveraging genomic
05:31biomarkers to guide therapy selection
05:33and high risk of relapse breast cancer
05:36and I'll just touch on some of the
05:39work that was foundational for this
05:41and our ongoing trials in this area.
05:44Right,
05:44so I think it goes without saying that
05:47really her two positive breast cancer
05:49is an archetype for precision medicine.
05:52This is of course one of our first
05:54exemplars where we had a targeted
05:56therapy for this copy number,
05:57amplified subgroup of disease,
05:59and we know that trustees map has
06:02been tremendously effective and
06:04has really changed.
06:05The landscape and outcomes
06:07for these patients.
06:08It's still the case,
06:09however,
06:10that despite the effectiveness of this
06:13agent that a subset of patients recur,
06:15and this has really led then.
06:16Down a path of developing a
06:19number of additional FDA approved
06:22agents including purchase Mab,
06:24TDM,
06:25one as well as small molecule
06:27inhibitors such as Neurontin and Pat
06:29nib to overcome this resistance,
06:32and so there's been numerous
06:33efforts in this area.
06:34We now have a wealth of FDA
06:36approved drugs and of course,
06:38in tandem to this.
06:39There have been considerable efforts to
06:41start to understand the mechanisms of
06:44resistance convergence on PR3 kinase.
06:46Pathway involvement of P-10 and
06:48so forth and and two dissect the
06:51contribution of these pathways to
06:53resistance and her two positive
06:55breast cancer.
06:55But really sort of coming back to this.
06:58It's still the case that while
06:59we need to escalate therapy for
07:01a subset of patients,
07:02there may be a subset of patients who
07:04actually do not require chemotherapy,
07:07and who could be spared these
07:09agents and so really,
07:10this sort of highlights the very
07:12critical need at this point in
07:14time to develop predictive mile
07:15markers to tailor therapy.
07:17And this is both for escalation but also
07:21deescalation and so just to highlight,
07:23you know how important this is.
07:25I I thought I would just demonstrate some
07:28of the pivotal trials in this space.
07:31Of course, NSA, BP,
07:33B31,
07:33amongst others that demonstrated
07:35the benefit of trustees Mab with
07:38respect to disease free survival.
07:39But since this time there's
07:41been numerous studies that have
07:43sought to further escalate therapy.
07:44These include the affinity trial of
07:47adjuvant trustees map in combination
07:49with pertuzumab as well as the
07:52Katherine trial which compared
07:53T DM one versus trustees map.
07:56And so really critical studies in
07:59the field have been highlighted
08:01very recently at ASCO and in tandem.
08:05There have been efforts to deescalate
08:07therapy and and one example of this is
08:10the a peachy trial which examined adjutant.
08:12Paclitaxel plus trustees map but
08:14with omission of chemotherapy,
08:16mainly anthracyclines.
08:16So this is a huge area and there's
08:19a lot happening in this space to
08:21really personalize therapy and
08:22in part enabled by by the many
08:25therapeutic options we do have.
08:26So I don't expect you to
08:27read this slide over here,
08:28but I want to say that this is really
08:31a place where there's been just
08:33tremendous efforts spanning multiple
08:35neoadjuvant trials in the early stage.
08:38Her two positive setting looking at both
08:41single and or dual agent approaches,
08:43and of course a key goal and the
08:45correlative science that has been
08:47done in tandem has really focused on.
08:49Can we develop predictive biomarkers?
08:53In the neoadjuvant setting,
08:54and so I'll just say that there's been a
08:57huge amount of sequencing of these cohorts.
08:59Calgb for 601 Pamela includes some of
09:01the most in depth data where there's
09:04been both EXO mandor targeted sequencing.
09:07There's been expression profiling some
09:08using arrays, some using RNA seek,
09:11really as discovery efforts to
09:13identify these biomarkers, and then,
09:15of course,
09:16a big component of embedded in that
09:18work has been the use and intrinsic
09:21subtyping or pan 50 based subtyping.
09:23To ask whether there is enrichment for
09:25the intrinsic subgroups and some of the
09:28associations that have been identified,
09:29there are an enrichment amongst
09:31responders in the her 2E or her two
09:34enriched group and then on treatment.
09:36As some of these trials have
09:39included a non treatment biopsy,
09:41there's been associations demonstrated
09:43between us which from her two enriched,
09:46for example,
09:47to normal like and so.
09:49This is clearly an important area where
09:51there's been numerous correlative studies.
09:53And then the other area that I
09:55that I also want to point out is
09:57that of course many have turned to
10:00assessment of tumor infiltrating
10:02lymphocytes or thiles in attempts
10:04to predict response to her two
10:06targeted therapy both at baseline
10:08and in a subset of trials on therapy.
10:11And so,
10:12just to speak a little bit more to that,
10:15this is one of the more recent studies
10:17that actually examined both the Pamela
10:20trial as well as the validation trial.
10:22And really, the goal was to ask,
10:24could we,
10:25for example,
10:26predict response to therapy and
10:28so just to highlight a few pieces
10:31of data from this study?
10:33And really one of the.
10:37If not further efforts to predict
10:40probably needs, but not at the moment.
10:43Sorry I hear some noise.
10:45Please please go on mute Susanna.
10:49Right so here really what they had
10:51done in this study was to assess
10:54tills at baseline, and at day 15.
10:57So this was our running biopsy
10:59prior to administration.
11:00Actually, of any chemotherapy it's
11:01not administered in this trial,
11:03and what you can look at is the change
11:05in tumor infiltrating lymphocytes,
11:07with orange showing an increase
11:09blew a decrease and stable,
11:12and in tandem in this study,
11:13they also actually examined
11:15tumor cellularity,
11:16and you can see some exemplars here.
11:18And so this really led to the development.
11:20Of an approach called cell till which
11:23attempts to combine the estimates
11:24based on tills with celularity you
11:26can see the area under the curve
11:29here about .7 in the panelist study.
11:31Actually it's higher for cellularity
11:33and so the goal was to sort of
11:36develop a combined classifier here,
11:38and this has really been.
11:39You know,
11:39one of the great successes in trying
11:41to predict response to her two targeted
11:43therapy all be it with variable
11:45responses across different cohorts.
11:47And so I'll just show that when
11:48this group went on to corroborate.
11:50These findings in the LPT trial.
11:52You can see that the performance was
11:56substantially inferior with an AUC
11:58under .7 using cell till here and
12:00so this shows you one of the recent
12:02attempts and of course this group
12:05probably needs no introduction to the
12:07opportunities and challenges around
12:09scoring tumor infiltrating lymphocytes
12:11and really standardizing these assays.
12:13But I wanted to present this to set
12:15the stage sort of for where the fields
12:16at and to say that you know really,
12:18despite many many attempts.
12:20To develop predictive biomarkers from
12:22the genomic from the transcriptomic,
12:25we still do not have a validated
12:27predictive biomarker and sell tools
12:29have emerged in the forefront.
12:30But there's more work to be done to
12:33really ask how consistent this is.
12:35Also across different agents and
12:37so this really sets the stage for
12:39where we began.
12:40Our journey in this field,
12:41which was a collaboration with Sarah
12:44Hurvitz and Dennis Slamon at UCLA
12:47on the trio USB 07 clinical trial.
12:49Now this trial looked at
12:51neoadjuvant trustees.
12:52Vanderlip at mid in early stage
12:53her two positive breast cancer.
12:55It was an investigator initiated
12:57trial and here you can see the sort
13:00of sample size is 130 patients.
13:02They were assigned either to our one
13:05trustees Mobileone ARM 2 lapatinib or
13:08the combination and what was you know,
13:10really intriguing to me about this
13:12trial was that core biopsies were
13:14collected not only at baseline but
13:16actually at run in after a single
13:18cycle of targeted therapy alone.
13:20Prior to the administration of chemotherapy.
13:22Now you can see the pathologic
13:24complete response rates,
13:25which was the primary endpoint here of 47%.
13:28The PATNAM was inferior here as it
13:30has been in other trials and the
13:33combination was modestly improved but
13:35not statistically significantly different,
13:37and so really this is this trial
13:40is distinct in the collection of
13:41a non treatment core biopsy prior
13:43to administration of chemo and at
13:45the time I felt that this would
13:47really afford us some unique
13:49insights into biomarkers of
13:51response after targeted therapy.
13:53Alone without having to deconvolve
13:55the effects of chemotherapy and so
13:57in initial work that was led by
13:59Sarah Hurvitz and Jennifer Castle,
14:01Gin a former fellow in my lab.
14:03Now, faculty at Stanford,
14:05we had embarked on an initiative,
14:07and this was actually embedded in the B
14:1007 trial design to leverage these pre
14:13on treatment and surgical samples to
14:16conduct bulk RNA expression profiling.
14:18And this was done using
14:21actually dual color microarrays.
14:23Which have now been largely
14:25supplanted by by RDC,
14:27but nonetheless you.
14:28You can see here the sort of try the
14:31design for this after a single cycle of
14:34of either of these agents we collected
14:37actually fresh frozen material and
14:38this was from a total of 89 patients.
14:40The fresh frozen material was then
14:42sent for RNA profiling and so of
14:44course the key question here is
14:45based on these expression profiles,
14:47could we predict pathologic
14:50complete response and?
14:52You can see that we took actually
14:54quite a deep dive into understanding
14:57the baseline clinical covariates and
14:59tumor features for these individuals.
15:02What you can see are the different
15:04measurements that were made.
15:05These include the her two fish ratio,
15:08which was.
15:10Obviously performed as part of
15:11the enrollment crunch cereal.
15:12We then went on to calculate
15:15the immune score.
15:16Her two IHC was also performed.
15:18Tills were scored and based on
15:20the RNA expression profiling,
15:22we were able to infer both
15:23intrinsic subtype M50 as well as
15:25the integrative subtypes which
15:27my group defined some years ago,
15:29and I'll speak to that more later
15:31and and then you have hormone
15:32receptor status and of course,
15:34Pathologic complete response.
15:35So really quite a rich data set with
15:38with numerous molecular correlate's.
15:39I'll cut to the chase and say that
15:42despite having all of these in hand,
15:44really none of them were
15:47robustly predictive of PCR.
15:48And that was true both at
15:50baseline as well as on therapy.
15:53But what we did learn from this
15:54study is that we were able to
15:56really start to deconvolve some of
15:58the contributions and something
15:59and illuminate some of the real
16:01challenges that come up with
16:04bulk admic sequencing data,
16:06and so many of the patterns that we
16:08saw are in line with what might.
16:10One might expect So what this plot
16:12shows here is actually the changes
16:15in gene expression during the
16:17short term on treatment biopsy.
16:19So after just a single cycle you
16:22can see the normalized enrichment
16:24scores for a variety of gene sets,
16:26and amongst those that are downregulated
16:29we have decreases in her two signaling
16:32proliferation and so forth.
16:34In in contrast we see increased enrichment
16:37for stromal and immune signatures.
16:40Just started after the short
16:42term therapy and there's a
16:43variety of signatures included.
16:44Many of them are of course correlated,
16:46and that's actually something that's
16:48really important to examine as we
16:50seek to parse these signatures,
16:51and so this is just showing really the
16:54Pearson correlation coefficient matrix
16:55based on the gene set enrichment,
16:58and hopefully what you can take away
17:00from this is that there's a huge
17:02degree of correlation for many of
17:04these pathways with one another,
17:05and this just sort of mirrors
17:07what we see over here. So indeed,
17:08in this short time course we are.
17:11Observing a number of patterns
17:12that we might expect to see.
17:16But there are, you know,
17:17we can also use these data to start to
17:19deconvolve what happens on therapy,
17:20and so, like many others,
17:22we also performed intrinsic subtyping.
17:25You can see that there's a preponderance
17:26of the her two enriched subgroup.
17:28This comprises roughly 53%
17:30of patients pretreatment.
17:32There's also a substantial normal,
17:33like composition here at treat.
17:36You know, prior to therapy, but on therapy,
17:39we do see this pretty dramatic switching,
17:41where a number of the her 2E
17:43cases actually become normalized.
17:44There is some other switching
17:46going on amongst.
17:47Luminous, but that's more modest,
17:48so we can assess these,
17:50and in these patterns are in
17:51line where others have reported
17:53and Pamela and similar trials.
17:55We can also use the bulk data to attempt
17:58to deconvolve the immune composition,
18:01but this is an exceedingly hard task,
18:04really, uh,
18:04you know there are many algorithms for this.
18:07I show one example,
18:08the cyber sort approach,
18:10which uses a reference matrix,
18:11and what I hope you take away from this
18:13is that you know apparent from these
18:15plots there's actually relatively.
18:17Modest changes from pre to on treatment
18:19or even at the time of definitive surgery.
18:22After completion of both
18:25targeted and chemotherapy.
18:26So this is a hard nut to crack.
18:29And actually the admixture
18:31in these populations.
18:33Really, we believe,
18:34hindered our ability to discover
18:36biomarkers in this context,
18:38and so I'll leave off there
18:40and say that this in tandem.
18:41You know,
18:42having sort of early reads into this
18:43and what others had been observing,
18:44which was similar.
18:46Let us too.
18:47Then take a new approach and so
18:50in collaboration with ministering
18:52while they were developing the
18:54digital spatial profiling platform,
18:56we really embarked on.
18:57This study was led by Katherine
18:59McNamara and MD,
19:00PhD student,
19:02who will match very soon and and
19:04was just a stellar lead and really
19:06taking on these new data types.
19:07And So what we did was to
19:09go back to this cohort.
19:10Now we turn to the formalin,
19:12fixed paraffin embedded tissue
19:14and we selected data from a.
19:16You know it was a total of 100 and.
19:18There were 122 samples that
19:20derived from 57 patients,
19:22and these were the ones that we
19:23felt we had adequate material
19:25left for for this assay,
19:26and so there were a total of
19:2820 in the discovery cohort,
19:2929 in the validation,
19:31again sampled at baseline,
19:33pretreatment at run in and at surgery.
19:36Right,
19:37so for those of you that are
19:39not familiar with this assay,
19:41we were focused on the Multiplex
19:44proteomic piece.
19:45There are indeed also technologies
19:48that allow one to profile the
19:50transcriptome that we're still very
19:52much in development at this time.
19:53The basic premise of this assay
19:55is that we have an indexing
19:57oligo nucleotide that is attached
19:59to this UV linker,
20:00and So what we can essentially
20:02do is to sustain our slide or FP
20:04slide with probes or antibodies of.
20:06Interest these oligos are UV,
20:09photo,
20:09cleavable and so that we can
20:11image the slides.
20:12Go in and select regions of interest or
20:14our allies and cleave off these oligos,
20:17aspirate them, and especially dispense them,
20:20and then they can be read off digitally,
20:22either using the Nanostring
20:23encounter or indeed via sequencing.
20:25And this process is repeated for a number
20:29of antibodies up to approximately 40 Plex,
20:31and so to give you a flavor for what we did.
20:34These were all four Micron
20:35sections that we took from this.
20:37You know almost this clinical trial for which
20:39we were using the sort of residual material,
20:42and we actually arrayed the pre on and
20:45surgical sample onto the same slide
20:47to mitigate batch effects and went
20:49in then and essentially use one of
20:51the the features of DSP which is to
20:54select based on phenotypic markers,
20:56in this case pants ID keratin to
20:59enrich for tumor cells as well as CD
21:0245 to illuminate immune cells.
21:05And of course this is coupled with the.
21:07Does DNA marker and so we can go
21:09in and really then take the tumor,
21:11enrich mask,
21:11or indeed take the inverted mask and enrich
21:15for the surrounding microenvironment?
21:16And we profiled a 43 Plex marker panel.
21:21This was really the panel that was in
21:23development at Nanostring at the time.
21:25We were fortunate that included a
21:28number of her two pathway members,
21:31AKT Phospho, AKT, and so forth.
21:33We actually had to add her to on.
21:34We could not convince them at
21:36the time to add.
21:38IAR, which was you know,
21:39disappointing in many respects,
21:41but I think many of the immune markers
21:44are really well represented here,
21:45and so this is the panel that we had now,
21:48just to give you a flavor,
21:49you know our eyes select selection.
21:51We could spend a lot of time talking about.
21:52There are huge study design
21:54considerations for how we do this.
21:56You know,
21:57in this trial cohort we essentially
21:59our goal was to select an average
22:02of four or ROI's part issue.
22:04This shows you an example where we
22:06selected 6 and in a different case where.
22:08You can see three of the four
22:10in this non PCR case,
22:12so you know we are essentially
22:14picking similar regions,
22:15but trying to be representative here and
22:17there are many ways to go about doing
22:19this now just to convince ourselves
22:21that the technology was working because
22:23we were amongst the first to view this,
22:26we went in and compared the
22:29normalized DSP KY 67 levels with the
22:31immunohistochemistry CHI 67 that we had
22:34for the same subset of samples and you
22:36can see that the correlation is .62.
22:38Reasonably good here.
22:40We did a similar analysis for her
22:43to where we had compared with,
22:45you know,
22:46I see based staining and again
22:48we're seeing trends that we would
22:50expect that convinced us of.
22:52You know that we were at least reading
22:54out some of this appropriately.
22:56So to give you a sense for the data.
22:59Really,
22:59what you can see here are just
23:01two examples
23:02of case 69.
23:03A pathologic complete response and case 58.
23:06I'm showing you individual regions.
23:09That we went in and profiled,
23:10and you can see the her two levels are
23:13indicated for these different regions above.
23:15And so this is just to show how
23:17we could visualize the data.
23:19But of course,
23:20what we're really interested in are the
23:22quantifications that we derive for this.
23:24So coming back to the sort of
23:27baseline characteristics that
23:28I had shared with you before,
23:30we were now very interested in looking at
23:32two key markers starting with her two.
23:35Of course as well as CD 45.
23:37So these are the protein markers of
23:39interest you can see on the bottom
23:41that we've stratified by intrinsic
23:42subtype as well as ER status obviously
23:44critical to account for here,
23:46and Patsy R,
23:46and so each of these dots represents
23:48a not at the log 2 normalized digital
23:51spatial profiling protein levels.
23:52You can see that there is reasonably.
23:54You know,
23:55good clustering for some of these in
23:57terms of the levels of expression
23:59within a sample, but some cases also exhibit,
24:01you know, pretty considerable variability,
24:04and so this is if we want to look at
24:07each of the regions individually.
24:08And that's of course something
24:10we can do with these data.
24:11So this gives us a sense for
24:15the heterogeneity present.
24:16In these markers,
24:17and neither were predictive at baseline,
24:19neither of these protein
24:21markers were predictive of PCR.
24:23We can also, of course,
24:25look at her two heterogeneity
24:27with far greater granularity,
24:29and so one of the ways that you
24:32could envision doing this is,
24:33as we've represented here,
24:34taking the PCR cases and non
24:36PCR cases on the bottom.
24:37This is again for each each individual
24:41region or summarizing the her two
24:44protein levels and you can see.
24:46How they look pretreatment
24:49versus on treatment.
24:50Whenever available,
24:51you can see that there's quite a bit
24:53more variability in the on treatment,
24:55her two levels,
24:55and actually we quantified this.
24:57You can look at the mean square error
25:01within patients versus between patients
25:03and see that they're far more comparable.
25:05Pretreatment versus on treatment.
25:07I mean,
25:08we really have a quite dramatic degree
25:10of change in her two heterogeneity,
25:14and so this this tells us
25:16something of course about.
25:17You know both that were
25:19likely hitting the target,
25:20but what that you know?
25:21How do we interpret the functional
25:24importance of this heterogeneity
25:25so you know the beauty of the
25:27multiplexing here is that it's not
25:29just her two that's of interest,
25:30but we can actually do these
25:32kinds of analysis for all of the
25:34markers that we've profiled.
25:35And so this is just stratifying
25:37by pretreatment on treatment,
25:39looking at all two all tumor markers.
25:42Not surprisingly,
25:43the greatest change in her two heterogeneity.
25:47Is in heterogeneous for her too,
25:49but this is followed by Phospho S6.
25:51If we instead look at the immune markers,
25:53we see that the change is
25:56overall relatively less compared
25:57to the tumor markers,
25:59but amongst the top markers we do
26:01see differences in our CD3 and CD8.
26:04And I should emphasize right now that all
26:06of the data that I'm showing you at this
26:08point is based on the pan cytokeratin
26:10enriched regions and focusing in on those.
26:12So similarly we can ask, well, you know,
26:14how do these markers change or differ?
26:16Or is there any association between patients
26:18that achieve a PCR versus those that don't?
26:21And this is now beginning to compare
26:23these on treatment values and and you
26:25can see that there are indeed pretty
26:27dramatic differences between responders
26:29and non responders.
26:30So this is all very well.
26:32I want to come back to say that
26:33of course you know we were.
26:34Being fairly pragmatic in
26:35our initial approach here,
26:36asking well can we.
26:37What can we learn from these
26:39samples at baseline?
26:40It would be ideal if we had strong
26:42predictors of response at baseline,
26:44so this is just showing you the CD
26:4745 pretreatment levels in a vial
26:49implant for the PCR cases versus
26:51non PCR as well As for CD 56 and
26:53what you can hopefully appreciate
26:55based on these violins are that
26:58there's really very no association
27:02in in the pretreatment markers,
27:04but once we start to look.
27:05On treatment,
27:06we're actually starting to see
27:07that there are significant
27:09differences between these markers
27:11in the PCR versus non PCR cases.
27:13So that gave us some,
27:14you know,
27:14sort of encouragement just in from
27:16a univariate analysis perspective.
27:18But what really got us excited?
27:22Was when we started to look at
27:24these markers in concert and So
27:26what I'm showing you here in this.
27:31It plot is that we're looking at the
27:33significance or the negative log 10
27:35FDR adjusted P value from the change
27:38from run into baseline again in the pan,
27:40CK enriched regions and so now you
27:43can look at the PCR cases and see that
27:45a whole host of markers are lower,
27:48quite dramatically lower at running.
27:50These include her two but also
27:52phospho 6 phospho Akt Chi 67,
27:54phospho, Erk and so forth.
27:57So really the whole ham pathway.
28:01Is showing a decrease?
28:03On treatment in patients that respond,
28:06there is a concomitant increase
28:07in a number of immune markers,
28:09including CD45, CD, eight others,
28:12really quite dramatic opposition here,
28:15and in contrast,
28:16we did not observe these patterns
28:18in the non PCR cases,
28:20so you can see that a handful of
28:22these markers are indeed achieved.
28:24Significance based on the FDR and there is a,
28:27you know, sort of a modest log 2 fold change,
28:30but it is much attenuated relative
28:33to the PCR cases,
28:34so this was the 1st.
28:35Sign that we had some signal and
28:37these data and was really quite encouraging.
28:41Now I wanted to come back to the
28:43RNA seek data RNA data that I
28:45showed you in the very beginning,
28:47and to say that we took the match samples.
28:49So just subsetting the larger
28:51RNA fresh frozen cohort where
28:53we failed to observe an association
28:55and asking what if we take these
28:57markers and just look at the
28:58RNA level and what you can see.
29:00Is that really we see no signal
29:03here now this of course could
29:06be attributed to many factors.
29:07It could be due to admixture in the bulk RNA.
29:11It could be due to the fact that
29:13we've actually enriched for a pan
29:15set of keratin tumor enriched,
29:16you know, population using DSP.
29:19And thirdly, it could be due to the fact
29:21that you know protein is more proximal.
29:23Readout of these signaling pathway
29:24changes that we want to observe.
29:25So multiple factors here, of course,
29:27the ideal comparator would be to do DSP RNA.
29:31Unfortunately,
29:32when we did this at the time,
29:34the RNA probes, it was a 96 Plex panel.
29:36We did do it the signal to noise was.
29:39Incredibly poor and we really
29:41couldn't use those data,
29:42so it is an experiment that that
29:43I'm sort of curious about to get
29:45back to which of these factors is
29:47driving our observations and I'll
29:49speak a little bit more to that more,
29:52but but you know,
29:53I will place some bets on the fact
29:56that we are enriching for tumor,
29:59and we are reading this out of the
30:01protein level so encouraged by our
30:03the previous slide and showing that
30:05there was an association between
30:07multiple markers in the PCR.
30:09Cases we then went on to take the
30:12logical next step which was to
30:14ask could we develop a classifier
30:16in our discovery cohort?
30:19Our very small discovery cohort and
30:22ask whether we could potentially
30:24predict response and so this is an L2
30:27regularized regression model because
30:28we were really trying to understand
30:30what these data could tell us.
30:32What we have done is to look at the
30:35DSP markers combined ingredients.
30:37So this is an and.
30:39Yep,
30:40pre and on treatment and we're
30:42comparing this with the sort of classic
30:44markers that we have in the field today.
30:46Which is estrogen receptor status,
30:49which we know is associated with PCR in the
30:51her two positive setting as well as Pam.
30:5350 You can see that this is this
30:54purple line with amine OC of .5
30:56so not telling us a whole bunch.
30:58We also try to combine these and
31:00ask whether this would improve our
31:02prediction and answer is no but but
31:04we were reasonably encouraged by
31:07this AUC of .733. Obviously this is.
31:10In, you know, in the discovery cohort alone,
31:14using cross validation and so you know,
31:17encouraged by this the next question was,
31:19well, OK,
31:20what markers are actually informative here,
31:22and so looking at the marker coefficients
31:24with this within this L2 model,
31:26what we noticed was that really CD
31:2945 and adjacent to this the next
31:31one up was Vista were amongst the
31:34the largest marker coefficient,
31:36so that gave us some clues and
31:38this was in the on treatment.
31:40Right,
31:40shown here in pink in the
31:43on treatment biopsy.
31:44So with this in hand we then you know,
31:46got a bit bolder and said, well, OK.
31:48What if we just look at CD 45 DSP alone?
31:52I mean, this would obviously be a
31:53simpler way to approach this problem,
31:55and the answer was that
31:57in our discovery cohort,
31:58so this is using cross validation.
32:00We got a very very high AUC,
32:03almost too good to be true of .9 and
32:05so then we actually took this into our,
32:09you know, withheld validation set where
32:11we assess this in the AC was .75.
32:14So not too bad.
32:15Still encouraging enough and and this
32:16LED us to then kind of come back to
32:18what I mentioned at the beginning,
32:19which was what we had on treatment pills.
32:22We had scored these for the entire cohort.
32:24We hadn't seen an association at
32:26large with that and we had reported
32:28that along with the trial.
32:30But you know what are these trends look like?
32:32So so here you can see the on
32:35treatment tell score broken down
32:37by non PCR versus PCR and you know
32:41we really see a far more striking
32:43separation of PCR versus non PCR cases.
32:45Saying CD 45 DSP they are correlated
32:47but not as well as one would like.
32:50And that begs a number of questions
32:52and and of course it would be really
32:55interesting to ask you know why is that
32:57the case in in this PO7 clinical trial?
32:59And why have others seen more of an
33:01association? For example in Pamela?
33:02But as I showed you before,
33:04not this doesn't always validate,
33:05and I think you know it raises the question.
33:07There's of course a lot of inter and intra
33:10observer variability in scoring chills,
33:12and so perhaps that's a contributing factor.
33:16But so with this information in hand
33:18and encouraged by the fact that CD
33:2145 DSP protein alone on treatment
33:23seem to be predictive of a PCR,
33:27we then went on to really try to reduce
33:30this approach to a more simplistic strategy.
33:33And So what we did was to then
33:35come back and perform CD.
33:3745 I mean a histochemistry on
33:40the cohort we had in hand,
33:42and we gathered as many additional
33:44cases from this clinical trial which
33:46was near expended at this point.
33:48To use that as a validation set,
33:50and so importantly,
33:52we built into this an effort to enrich
33:56with expert pathology guidance for
33:58tumor for tumor and rich regions,
34:00and this was really trying to mimic
34:02what we had done with the pants
34:05cytokeratin enrichment using DSP.
34:07We then use Q path to really
34:09automate this process and develop
34:10a digital pathology workflow.
34:12So how does this look?
34:13Well, so now this is taking, you know,
34:16a comparison of the match cases where we had.
34:19And then CD 45 immunohistochemistry
34:21shown in orange versus those
34:24cases where we had CD 45 DSP.
34:26You can see that they're
34:27largely in agreement.
34:28The IHC does slightly better,
34:31but really these are very,
34:34you know, reasonable AUC's to observe
34:35and just to put this in context,
34:37then what we're what we really want
34:39to be able to do is ask, well,
34:40what is the positive predictive value?
34:42The chance that a tumor will
34:45have a pathologic complete
34:47response to TCLTCH or TCHL?
34:49So the three different arms in the study
34:51and we were able to then, you know,
34:54use this to sort of set a cut point and
34:56the positive predictive value for a CD.
35:0045% positivity greater than 20% is .82 here,
35:03and this is in this combined cohort.
35:05If we do this only in the validation set.
35:08Where we have fewer cases,
35:10it drops to about .71.
35:12But Needless to say,
35:13this was really encouraging and
35:14begs the question, you know,
35:15sort of just to come back to this,
35:16but we could go from a Multiplex assay
35:18and reduce this down to a single
35:20marker that could be run in any lab.
35:22We know that CD 45 is incredibly
35:24robust and raises the possibility that
35:27perhaps this kind of approach could
35:30be used to guide therapy D escalation.
35:33Now I'll come back to this.
35:34This will require further validation
35:36in additional trial cohorts, namely.
35:39Ideally those that did not administer
35:42chemotherapy.
35:42Afterwards,
35:43and we are actively now validating
35:45us in a in a retrospectively in
35:48one of those trial cohorts,
35:50and if that pans out then of course then
35:53next steps would be a prospective effort.
35:55So moving you know back,
35:58I just want to spend a few more minutes
36:00saying that we of course had much more
36:02data in this cohort that we could mine.
36:04We are very keen to really leverage
36:07what started out as a discovery effort,
36:10but LED us to this new biomarker.
36:11I mean,
36:12this really was a piloting of the technology.
36:14That uncovered some pretty
36:15interesting biology here,
36:16but sort of coming back to the data
36:19I showed you before that the non
36:21PCR cases really didn't show very
36:23dramatic changes at the pretreatment
36:26versus run in time point.
36:28However,
36:28if we take those same patients and
36:30now look at them at at the surgical
36:33versus pretreatment time points,
36:34so after they've completed the full course,
36:36what we see is actually that by
36:39then they do indeed exhibit a
36:42a reduction in her two.
36:44We'll see that there is a reduction in
36:46CHI 67 these other downstream markers,
36:48however, are less downregulated,
36:50and this potentially suggests
36:52compensatory signaling in the non PCR
36:55cases that's active at the time of surgery.
36:57I will point out as well that we do
36:59see shifts by the time of surgery
37:01in immune markers, including CD 56.
37:03And of course you know this raises
37:07the question.
37:08Why that might be the most enriches
37:10this because we're seeing an
37:12effective natural killer cells
37:13in identifying and killing these
37:15chemotherapy stressed tumor cells.
37:17You know that's that's possible.
37:20And so really a lot to unpack here,
37:22but I think it's quite
37:23interesting that now we do
37:24start to see these changes now.
37:26On top of this, I haven't really
37:28talked at all beyond what we can
37:30do with the tumor in rich regions.
37:32But as I mentioned in the beginning,
37:33you know a benefit of the DSP
37:36approach and the phenotypic
37:37selection strategy that we deployed.
37:40There are many others that one could
37:42envision is that we can now select the
37:46surrounding tumor microenvironment area,
37:47and so we wanted to explore that.
37:50You know a bit here,
37:52and So what I'm showing you is
37:54now looking at the enrichment of
37:56just immune genes in either the
37:58surrounding microenvironment that's
37:59shown over here on the right.
38:02So that's an enrichment towards the TME.
38:05The TME area versus in the tumor over here,
38:08and so this is in the first instance
38:10looking at the pretreatment time point
38:12where you can see is that there's really
38:14a number of immune suppressive marks
38:16evident in the tumor in rich region.
38:18We see evidence for T cell exclusion.
38:20Based on you know, CD3 CD,
38:22four other markers.
38:23I do want to be cautious here
38:25and why we sort of, you know,
38:27interpret this a little bit
38:28carefully is that you know,
38:30I think there's a lot of
38:31open questions around,
38:32just the relative.
38:33So we say stickiness affinity of these
38:35antibodies in tumor versus immune
38:38populations that we don't understand well,
38:40and that will need to be parsed further.
38:43And I say that in part,
38:44just noting that we see very, you know,
38:46high enrichment of B7H4 over here.
38:48So we could take that with a grain of salt,
38:51but of course what we can do is
38:52assume that those are going to be
38:54equivalent overtime and now ask well
38:56what happens in terms of these marks
38:58from pretreatment to on treatment,
39:00and you can see that you know really
39:03in the tumor enriched region,
39:05things stay large,
39:05largely the same at this first time point,
39:07by the time we get to the post
39:09treatment time point,
39:10we actually do see that you know many
39:14of these markers have now shifted.
39:15We see far less evidence of, you know,
39:17sort of this T cell exclusion.
39:19Now they've.
39:19Appeared to infiltrate and and I'll
39:21point out that this Last Post treatment,
39:23timepoint were of course only
39:25looking at the non PCR cases.
39:27We're not looking at the responders who
39:29presumably had no tumor cells present.
39:32Right,
39:32so this this gives us some clues as
39:34to also potentially the timing of
39:37changes of these immune markers,
39:39and you know raises the question
39:41as to whether further efforts to
39:43profile at multiple time points
39:45on therapy might actually inform
39:47you know the timing of these,
39:49I mean infiltrates and maybe have
39:52relevance for contemplating the timing
39:54of immunotherapy in these populations.
39:55And of course that will
39:57require further study as well.
39:59So I guess you know just to
40:00really wrap up this part.
40:02I'll say that Multiplex proteomic profiling,
40:05coupled with pan Cytokeratin and
40:06Richemond can reveal dynamic changes
40:08in the tumor microenvironment
40:10during her two targeted therapy.
40:12These data are data really under score the
40:14value of having a non treatment biopsy.
40:16That's obviously can be difficult to achieve,
40:19but this was instrumental.
40:21And really where we saw the most predictive.
40:25A potential of any of these biomarkers.
40:29We see that CD 45,
40:31either expression or cell counts as measured
40:34by HC predicted PCR in an independent
40:37set and this really did outperform
40:40other candidate biomarkers such as ER
40:42or her to enrich status and so we do.
40:46We think that these findings have
40:48implications for tailoring therapy.
40:49Of course, far more work is needed,
40:51but really the dream would be to
40:53be able to scare to spare patients
40:56who safely can omit chemotherapy.
41:00And to make that determination
41:02early during their treatment course
41:03and so this will require further
41:06validation which is ongoing.
41:07But I'll just say that you know,
41:08I think there's many other open
41:10questions that this work informs.
41:12Of course,
41:12you know it will be interesting to see
41:15how predictive CD 45 is in other cohorts,
41:17as well As for other anti
41:19her two targeted agents.
41:20I'm still very interested to better
41:22understand the comparison with
41:24cell till that hasn't really been
41:26head-to-head in larger cohorts.
41:27I think this raises the question
41:28about what is the optimal timing.
41:30We had this essentially one cycle of
41:32targeted therapy two week window because
41:34that's when patients were biopsied,
41:36and it's convenient.
41:38There may be better windows,
41:40but of course understanding this
41:42timing is going to be critical for
41:45really optimizing our interventions.
41:46And then there's many other
41:49questions about whether CD 45,
41:51you know will correlate with long term
41:53outcomes and whether its prognostic
41:56and or predictive in other subgroups
41:58of breast cancer and other cancers.
42:01So I'll say that you know there are
42:03many other efforts to contemplate
42:05D escalation strategies.
42:07One of these is the ADAPT trial,
42:11which is looking at neoadjuvant
42:13pertuzumab plus trustees.
42:14Mab with or without paclitaxel.
42:16Another trial is for gain,
42:19which is looking at and FDG PET based
42:23biomarker of Pathologic complete
42:25response and of course these are
42:28you know ongoing and reporting out.
42:30And really, just highlight,
42:32I think the the real efforts of
42:35the Community to try to identify
42:37these biomarkers and two to optimize
42:40for our patients.
42:41Now I will say that of course
42:43I've I've mentioned this in the
42:45context of deescalation.
42:45Really the flip side of that coin is that
42:49ultimately we're very likely to need risk.
42:52Adapted novel trial designs to tailor
42:55therapy and deliver new drugs to
42:58high risk patients as needed and so.
43:00We've been contemplating that a little
43:02bit more in the ER positive her two
43:03negative setting and I'll just say that,
43:05you know,
43:05sort of building on from this work
43:07we've now embarked on a number of
43:09other efforts to really chart not only
43:12the tumor immune microenvironment,
43:13but to characterize tumor
43:15evolution through therapy.
43:16And that's been on a really
43:18long standing interest in
43:19my lab. We are using a
43:21variety of tools to do this,
43:22not only spatial proteomics,
43:24but also transcriptomics,
43:26which you know affords us maybe
43:29a less biased approach, and.
43:30Enables discovery efforts and and for many
43:32of these cohorts that we're working on,
43:35we've previously performed end up
43:37sequencing mainly at the bulk DNA level,
43:40but in some cases when possible.
43:42We're also doing this, you know,
43:44at the single cell level,
43:45to try to tease apart this biology,
43:47and so one of the areas that
43:50we're particularly interested in,
43:51and that I'll just summarize
43:53briefly is in understanding the
43:56determinants of breast cancer relapse
43:58and so to highlight this problem.
44:01You know,
44:01I think we all appreciate that
44:03prognosis has improved dramatically for
44:04early stage breast cancer patients,
44:06in part due to new therapeutic
44:08strategies and screening and,
44:10and we certainly know this is the
44:11case for her two positive disease,
44:13but but for many other subgroups.
44:15And yet at the same time more
44:17than 20% of patients will recur
44:19with Mets at distant sites,
44:22and this remains largely incurable.
44:24There was a very powerful meta
44:27analysis performed by panel
44:29published several years ago.
44:31Which demonstrated that there's a
44:32subset of women with early stage ER,
44:35positive breast cancer who have a
44:36persistent risk of recurrence and death
44:382 decades after their initial diagnosis,
44:41and these include women
44:43with node negative disease.
44:45As you can see over here,
44:48and so really illuminating what's
44:50been observed in clinical practice.
44:51Now,
44:52a key challenge of this has been
44:54that it's evident that our classic
44:57characteristics of nodal status,
44:59size, grade are insufficient.
45:01To predict recurrence,
45:02and really that the progress in the
45:04space has been impeded by the lack
45:07of cohorts with long term clinical
45:09follow-up and molecular data.
45:11And so this is really a segue to
45:15just a brief summary of of other
45:17work that we've been pursuing where
45:19several years ago now a decade ago.
45:21Actually,
45:22we sought to unpack the genomic
45:25landscape of breast cancer,
45:28really focusing on combining whole
45:30genome copy number based profiling.
45:33Quick transcriptomics and using
45:35unsupervised approaches we discovered
45:37that there are at least 10 molecularly
45:39distinct groups of disease.
45:41You can see these over here.
45:42This is the chromosome copy number
45:45in red shows amplifications,
45:46deletions in blue looking along
45:48the chromosome,
45:48and you'll recognize the Pam 50
45:50intrinsic subgroups and then our
45:52integrative subgroups on the outside.
45:54You can see that we really
45:56discover a number of groups.
45:58Not only do we recover, of course,
46:00the her two positive Group A great control,
46:02but we see other groups such
46:03as integrative cluster.
46:04One which has amplification of ARP,
46:056 KB,
46:06one on chromosome 17 Q integrative cluster,
46:096 amplification.
46:10Overexpression of FGFR,
46:11one on chromosome 8P12 and then this
46:15highly complex integrated cluster
46:17two with amplification of a cassette
46:19of chromatin regulators on 11 Q.
46:22So additional subgroups that are
46:24very much copy number defined
46:26and we were able to further show
46:28that these integrative
46:29subgroups really have
46:31distinct clinical outcomes.
46:32This is this cohort,
46:33obviously by virtue of the long
46:35follow-up predated the use of trustees,
46:37and Ave can see the integrative Cluster
46:385 or her two positive group here,
46:40but numerous other groups
46:42have very steep trajectories.
46:43So just to recap what what
46:45this really told us is that in
46:48addition to the her two positive
46:50subgroup or integrated Cluster 5.
46:52Many other subgroups are
46:53copying overdriven and might
46:54share similar characteristics,
46:56and so in recent years we've been
46:58able to go back and obtain the 20
47:01year clinical follow up for this
47:04metabolic cohort of over 2000 patients,
47:06and So what I'm showing you here is
47:08just a summary broken down by ear
47:10positive and ear negative patients.
47:12Each patient along the vertical.
47:14You can see the site of metastasis
47:16for these patients.
47:17You can see that they're vast,
47:18so there's a huge degree of organic tropism.
47:22But really,
47:23what's critical about these data
47:24with long term follow up is that
47:26now having this complete recurrence
47:28information allows us to study
47:29the rates and routes of distant
47:31relapse and their lethality,
47:32and so I won't dwell on these data.
47:35They're all publicly available,
47:37but what this led us to is to then really
47:40revisit these integrative subgroups
47:41and their association with relapse risk,
47:44and So what I'm showing you here is
47:47the probability of relapse ordered
47:49by increasing risk for individuals.
47:51And I'll walk through starting
47:53with her two positive.
47:53Again,
47:54before the use of trustees met Ian Black,
47:56you can see after surgery there risks
47:58over this 20 year interval in red
48:00after being disease free five years
48:02and in green disease free 10 years.
48:04And of course these trajectories
48:06are are very steep.
48:07We know her too is a bad actor
48:09prior to trustees moving.
48:10This has changed the game,
48:12but we were intrigued to see that just
48:14adjacent to the her two positive group
48:16where these four what we term high
48:18risk ER positive her two negative subgroups,
48:20integrative clusters 169 and two which
48:22happened to be defined by those.
48:24Hallmark copying appropriations.
48:25I just showed you,
48:27and so really.
48:28Hopefully you can appreciate that the
48:30risks of relapse are in excess of,
48:32you know,
48:32in some cases 55% and this persists.
48:35Five 1020 years after diagnosis,
48:37so we believe that this subset of
48:40patients may correspond to the late
48:42relapsing groups defined by PAN
48:44at all for which biomarkers have
48:46been lacking now adjacent to this.
48:48There are two subgroups of
48:49triple negative disease,
48:50so I see 10 is a classic baselight group.
48:53You can see the risk of relapse
48:55plateaus after five years.
48:56Integrative cluster for this,
48:57ER negative group actually has
48:59a increased risk of relapse,
49:00which better mirrors the ER positive
49:02groups and they have dramatically
49:04different immune landscapes and then
49:06over here we have our more typical risk.
49:08The majority of ER positive her
49:10two negative patients who really
49:12show show much more modest risk,
49:15so. Taking this information and comparing
49:17it to a sort of a typical risk group,
49:19we built the most powerful
49:21clinical models we could,
49:23incorporating all of the known covariates
49:25and what I'm comparing here is the
49:27clinical model plus immunohistochemistry
49:28versus integrative subtype information,
49:31and what I hope you can appreciate is that
49:33if we just look at immunohistochemistry
49:34data to separate these groups,
49:36we see that the risk is really
49:37homogenized the green line triangles
49:39are very similar across these groups,
49:40whereas when we incorporate
49:42intricate subtype,
49:42we see pretty dramatic separation and
49:44in this varies over time and it varies.
49:46In a subgroup,
49:47so we believe that this information
49:49informs the prediction of relapse risk.
49:51But critically, you know,
49:53really these groups have distinct drivers,
49:55and so I've already talked to you about
49:56what the integrative subtypes are.
49:58This is just showing you the
50:00landscape or amplification frequency
50:01for these different drivers.
50:02And really,
50:03there's many genes in these
50:05large copy number regions,
50:06so pinpointing the precise
50:07driver is a challenging task,
50:08but there's a number of candidates
50:10that emerge for each one and just to
50:13say that while individually these
50:15groups account for eight or five.
50:17Or you know, another 8% of the population.
50:19Together they account for 25% of all,
50:22ER positive, her two negative cancers,
50:24and the vast majority of distant relapses.
50:27And so we've been really intrigued
50:28to contemplate the fact that this
50:30may nominate new therapeutic targets
50:31in these high risk populations,
50:33thinking about honing in on their
50:35downstream targets, either the FGFR.
50:38Receptor itself, or indeed the ligands.
50:40And of course many downstream
50:42targets in the AKT mtor pathway.
50:44And so this actually motivated us to
50:47develop a window of opportunity trial
50:49to evaluate new therapeutic strategies
50:51in these high risk populations,
50:53and this is funded by the Department
50:56of Defense and and really this is
50:58a multicenter trial terpsichore,
51:00which we will biomarker stratify patients
51:03according to their integrative subtypes.
51:06Assign them into these individual groups.
51:08And conduct a window study where patients
51:11receive two weeks of targeted therapy.
51:13The readout of interest is a
51:16reduction in CHI 67 after therapy,
51:19and of course we're comparing the
51:21targeted agent alone or in combination.
51:22Sorry,
51:23the targeted agent in combination with
51:25ending therapy or endocrine therapy alone,
51:27and they are randomized,
51:28and so we're really excited about.
51:30This is a very ambitious trial,
51:31of course,
51:32to biomarker stratify in a very short window,
51:34and to do this in the early stage setting,
51:35but we believe that additionally,
51:37by collecting on treatment.
51:39And core biopsy surgical samples
51:41will actually be able to conduct
51:43similar studies to what I described
51:45before looking at the change in
51:47in these in these tissue samples
51:50in response to short term therapy.
51:52And so to enable this,
51:54we've really also set up a whole
51:56pipeline to do prospective biobanking.
51:58Both plasma tissue collection,
52:00but also the generation of organoids.
52:02And I'll say that that's been
52:04really ongoing work in
52:05my group to establish organized
52:07models that are representative of
52:09these high risk of relapse subgroups,
52:11because in fact they are vastly
52:13underrepresented by existing cell lines,
52:16and this is afforded us a real opportunity
52:18to have very high quality viable
52:20material for a number of assays and.
52:23And I hope to share with you
52:24some of that at another time,
52:26but also just to say that this is
52:28also really fueled a new center that
52:30we have for breast cancer metastasis.
52:32It's very much focused on delineating
52:34the evolutionary dynamics as well
52:35as micro environmental determinants
52:37of metastatic breast cancer,
52:38and it's really oriented around
52:40these integrative subgroups which
52:42we've defined seeking to define
52:44their definitive drivers as well
52:46as to leverage real-world data
52:48to evaluate these associations,
52:50but also to do a deep dive
52:52into the cellular topography.
52:54Of both the primary tumor and
52:56metastasis through therapy.
52:58And, of course,
52:59these are inexorably LinkedIn breast
53:01cancer and then a final piece of this
53:03and sort of give you the overwhelming
53:05schematic is actually now with these
53:07organoid models that we've established.
53:09We're actually conducting our very
53:11first crisper screens in 3D models
53:13to really pinpoint the definitive
53:15drivers of these subgroups,
53:16and and of course,
53:17we hope that this information
53:19will ultimately inform the next
53:21wave of clinical trials.
53:23It's all closed by thanking the many
53:26people in my lab who led this work.
53:28Fabulously.
53:28Talented scientists that it's really a
53:30privilege for me to work with every day.
53:33And you saw some of their
53:34pictures along the way.
53:35I just want to thank our collaborators at
53:37Sarah Hurvitz densely man and and Mike Press.
53:39And of course the initial work was also
53:41done in collaboration with Nanostring,
53:44and I'd be very happy to take any questions.
53:51Great thank you Christina.
53:53That was terrific.
53:55Vic lecture and we would all be
53:56clapping except that you can hear us.
53:58You can see is the logos these days,
54:01but I'm sure there must be questions.
54:04Anyone can either raise their
54:06hand or if they have questions
54:08just unmute and go ahead.
54:14Hi hi Christina, very nice work.
54:16This is iron crop.
54:19In the it's a very nice study when
54:22you have these the the neoadjuvant
54:24trial with the window of the her two
54:27therapy alone and you saw that you
54:29know this decrease in in her two
54:31signaling and increase in in immune
54:33markers in the responders or the people
54:35who eventually would be responders.
54:37Did you see any difference in that
54:40association when you looked at the
54:42type of her two therapy that they
54:44had given the potential for the
54:46putative differences met different
54:48mechanism action between.
54:49Antibodies and T keys.
54:51Yes we did and it was something near
54:54and dear to my heart to explore,
54:56but I have to say we were pretty
54:59underpowered because, you know,
55:00the whole trial whole trial was
55:02130 patients and then just for the
55:04subsets that we did DSP on that that
55:06that there was adequate material.
55:08The numbers in each of the arms were low.
55:10We kind of course grouped the
55:13trustees amebo only and the trustees
55:15met platinum are we see some hints
55:17of signal that would suggest that.
55:19You know, ABC mechanisms are are distinct,
55:22but we're really underpowered
55:24to fully tease that apart,
55:26and I think that it's ultimately just
55:28going to require a greater sample size,
55:30but it's a wonderful question,
55:31one that I ruminated on long and hard,
55:35and you know,
55:36I think I suspect that it's
55:38that that it will be different.
55:41We just didn't have the data.
55:42Yeah, thanks.
55:49Right, this is my year. Thank you.
55:51That was sharing the exciting
55:54part that was really nice.
55:57Difference in the pattern?
55:59You know signature or markers
56:01between the ACB 2 and three?
56:04If y'all look at those or with
56:06ACB 3 just you know worst thing
56:08you know the same type of markers
56:11persisting but much worse.
56:14Yeah, I mean we did.
56:16We did group them again
56:17because of power issues.
56:18You know just in terms of
56:20stratifying further on that.
56:21I agree it would be interesting
56:22we could go back.
56:23I think we're just we're really limited
56:25by the numbers here and you know.
56:27This was intended to be sort of a pilot
56:29effort to understand the technology,
56:31and we went back once we saw these
56:33results and gathered as many samples
56:35from this cohort as we could to go
56:37back and really bolster the numbers,
56:39but I think we're still just just fall short,
56:41and so you know,
56:42I think would be very
56:44interesting to to do this.
56:45Not not that there's so many
56:46of these additional cohorts
56:47that have on treatment,
56:48but there are others that are larger,
56:50so yeah,
56:51unfortunate.
56:54Thank you. Both very nice talk.
56:59So for the first part of Orlando strain
57:02studies when where you showed pretty
57:05good prediction power with on treatment
57:08parameter but not pre treatment.
57:10But I wonder whether you actually
57:12look like the the difference or ratio
57:15between some of the parameters and
57:16see whether there's a correlation.
57:18It might be better prediction power,
57:20that's right so we did so initially many
57:23of our models were actually combining
57:25sort of trying to get at this delta.
57:27The delta which I I thought was going to be,
57:29you know where it's at?
57:33Additional information that you're
57:34leveraging from having these
57:36pairs and it really turned out.
57:38That it seems that it's the
57:41that the on treatment value in
57:43and of itself held its own.
57:44It really held its own and we
57:46didn't get a huge benefit from that.
57:48Now there there may be many reasons
57:49why I mean, I'll say that you know,
57:52we were struck by the reduction in
57:54celularity in this cohort on treatment.
57:55Many of these patients,
57:56and of course we know that you know,
57:57there's cellularity estimates
57:59are not always concordant.
58:01We estimated this both molecularly,
58:03but obviously with expert pathology
58:05review and the sections were not
58:07identical for the different assays,
58:08of course.
58:09So there were a number of cases that
58:12you know had essentially 0 celularity
58:14upon path review on treatment more
58:17so than I would have expected we
58:18see in association with cellularity.
58:20But of course then what's interesting
58:22is you go in with DSP and you you start
58:24to illuminate this and we can see
58:26evident tumor markers there, right?
58:29So I mean just coming back to sort of
58:31some of the questions about the delta,
58:32I think.
58:35Yeah.
58:35You know some some questions emerge
58:37about like what's the heterogeneity
58:39in the pretreatment sample versus
58:41the on treatment?
58:43We're limited by the core biopsies.
58:44I think we see quite a bit of it and
58:46I I would say that we do detail some
58:48of this in the supplement and methods.
58:50There's a lot of comparisons one can
58:51do if you reduce the number of regions,
58:53which is one way to kind of come
58:55back to this question of, you know,
58:56are we measuring enough in terms
58:58of the delta?
58:59Because we had some cases for up
59:01to four ROI's and it looked to us,
59:03just, you know,
59:03sort of from a study design consideration
59:05that a minimum of two regions.
59:06Would be really helpful here,
59:09whether for improved that a ton is a you
59:11know another matter, not always feasible.
59:15But I think it's a.
59:16It's a great question that
59:17was our intuition as well,
59:19but for these various reasons
59:20we didn't really see a dramatic
59:22benefit of the delta.
59:27Maybe I'll take a chance to
59:28ask a question here as well.
59:29We're kind of running out of time,
59:31but maybe we have a couple more minutes.
59:33In particular one of those
59:34things we struggle with is how
59:36many fields of you are enough,
59:37and in your DSP study it's it
59:39looked like that there was.
59:41You were struggling with that same thing.
59:43Did you look at in in DSP?
59:45One of the blessings and curses of
59:47that technology is it's way more
59:49data than you can analyze years,
59:51and there's lots of other questions I
59:53have for you that I'll discuss later
59:55about other ways of analyzing that data.
59:57Did you look at averaging your
59:59fields of view versus looking at
01:00:01them individually and with that
01:00:03with the average or individual
01:00:05fields more informative?
01:00:07And did you look also at like how many
01:00:09you know comparing 2 versus 6 that
01:00:11looked like you had different numbers?
01:00:14Can you comment?
01:00:14That's right.
01:00:15Yeah, it's a hugely important question.
01:00:17I would say we were very.
01:00:19We spent a lot of time trying to parse
01:00:22this with this with the data that we had.
01:00:24I think that some of the analysis I've
01:00:27presented we did indeed take the average.
01:00:30There are ways to also,
01:00:32you know, wait that information,
01:00:34but in the sort of logistic regression
01:00:36models we were actually predicting,
01:00:38we did take an average across them.
01:00:41We subsequently, you know,
01:00:43we also investigated sort of.
01:00:44What if you had only?
01:00:45One versus 2 versus 4 and some had up
01:00:47to six and and as I sort of hinted at
01:00:50and actually analogous to what we've
01:00:52seen with genomic heterogeneity at
01:00:54minimum of two regions gets you a lot
01:00:56this ability to say anything about
01:00:58how heterogeneous these markers are,
01:01:00even between sort of two regions
01:01:03is hugely important,
01:01:05informative,
01:01:05and then there starts to be what appears
01:01:07to be a trail off in terms of the added
01:01:09benefit when you get up higher to six,
01:01:11you know,
01:01:11I think I wouldn't stake my life on
01:01:13whether choose the absolute Max we we
01:01:15try to collect more whenever possible
01:01:17because we're interested in the discovery.
01:01:19You know,
01:01:19and sort of questions around this.
01:01:20And I, I think also questions arise as to,
01:01:23you know how variable is this going to be
01:01:24for different subgroups of disease, right?
01:01:27So I don't think we've mastered that,
01:01:29but I felt at least encouraged that with two.
01:01:32We seem to be gaining information that that
01:01:35the heterogeneity itself was informative.
01:01:37Yeah,
01:01:38great thanks.
01:01:40Are there any other questions?
01:01:41Were already 5 minutes past,
01:01:42but if there's one more question,
01:01:44maybe we could take that if not.
01:01:48I don't see anyone raising their hands,
01:01:50so I'll assume that we're all
01:01:53happy with where we are and
01:01:55thank you very much for a very
01:01:57interesting lecture and well, thank.
01:01:59Thank you all for joining.
01:02:01Thank you.