Tip:
Highlight text to annotate it
X
>> Good morning everyone.
Thank you for joining us today for the NCI CBIIT Speaker Series,
a knowledge sharing forum featuring both internal and external speakers
on topics of interest to the biomedical informatics and research communities.
My name is Tony Kerlavage.
I'm the head of the Informatics Program here at NCI CBIIT.
I'll remind you that today's presentation will be made available on the wiki
for the speaker series as a screencast with voice-over and also posted
on the Speaker Series YouTube playlist.
If you Google NCI Speaker Series, you'll find that Wiki page.
Information about future speakers is available via Twitter and on our blog,
so please check out those sites for the latest information.
You can just Google NCI blog and our Twitter handle is @NCI_NCIP.
Today, I'm quite happy to welcome Dr. Cheryl Marks who is Associate Director
of the NCI Division of Cancer Biology.
Dr. Marks received her Ph.D. in Biochemical Genetics
from the George Washington University and did post-doctoral training
in the National Institute of Neurological Disorders
and Stroke Intramural Program,
and she joined the NCI as a program director in 1987.
The title of her presentation today is the "Oncology Models Forum,
An Interactive Platform for Collaborative Translational Research."
And with, that I'll turn the floor over to Cheryl.
>> I hope this microphone is working.
Everyone can hear fine?
OK. Cool. I think it's important to point out right at the outset that in fact,
this is a virtual presentation about what is in fact a virtual program.
We have not launched the Oncology Models Forum but I would like to describe it
to you today and use this time to open a discussion and suggestions
that might help us in fact to formulate this program effectively.
We started about 15 years ago,
a program in our division called "The Mouse Models of Human Cancers Consortium."
It's a research program and its main goal right at the outset was
to generate a whole new suite of models for translational research ultimately.
For 15 years, the program has gone on, the program ends this August and they've done their job.
We now have, along with other laboratories have adopted the techniques
to do this have generated many more models.
So, the cancer research community has many, many, many models to choose from.
That's also the challenge.
After 15 years of supporting research on cancer models,
we thought it was time for us to begin to serve the needs
of the greater community by having an open forum where people who need
to use models could come and find out a great deal more about what is the art
and science, a lot of its art of cancer modeling as well
as what are the resources available, with whom might they collaborate
and a whole host of other educational sorts of opportunities.
But when we thought about how we might formulate the new program,
this Oncology Model Forum, the Mouse Models Consortium people were really very
helpful in pointing out where some of the opportunities are
and they lie as you will see in integrative use of Mouse Models
and other animal models in cancer research.
To do that well requires an abase, not just as information
but of tools to be able to analyze the data that are brought
in from animal models and mingle with appropriately with information
from clinical trials, from epidemiology and from prevention studies.
So, the opportunity exists to put together communities
who in fact don't really interact as much as they could.
Although, we have example and I'll show you a few of them where the juxtaposition of expertise
in different fields is really making a difference in this area of integrative use of cancer and tumor models.
So, when we were thinking about how we might do this, we were looking for a platform.
One of the first things that we thought about was, "OK. Fine, we can do a website on steroids."
But a website on steroids which is great at deploying information is not the sort
of entity that allows you to encourage collaborations across communities.
And because the whole field of cancer research of every kind, as well
as animal modeling, is a worldwide activity,
we really need something that is facile
to pull people together from a variety of places.
So, about the time we were thinking about this, the Institute began to--
a pilot project with a platform group at Purdue University.
The platform is called HUBzero.
Some of you may have heard of it, some of you may not.
So, this is HUBzero. It is an open source platform.
It was built -- and I really like this part --
I love projects that are made with other people's money.
NSF paid for the development of this and it hosts a lot of different organizations who use it very effectively.
I spent a lot of time on this site looking to see how other people are using it in fact for their programs.
So, what is HUBzero? It's an open source software platform.
It has software. It has middleware.
It has all kinds of approaches that have been tried by other communities
to in fact do team science in a virtual setting.
And this is really helpful to the sorts of things that we want to do.
So, this is always a question: why not just a website on steroids?
Well, one of the nice things about this and I'll get to this a little later
on is that in fact people can do experiments together in the space in HUBzero.
And then, as they accumulate particularly sort of the history of the project,
exactly what they did is they went along to demonstrate that you went
from point A to point B, et cetera.
Then, you can assign a digital object identifier to it
and it becomes a publication which you can cite
and I'll come back to that in just a moment.
So, the NCI has set up NCIP Hub and if you haven't visited NCIP Hub,
I suggest that you do so because if you move around that particular entity
and you see what some of the capabilities are, then you too might be attracted
to it for your program to think, "Huh, this might facilitate some of the things
that we really want to do as well."
At any rate, because NCIP has already established a hub,
we expect to put the Oncology Models Forum under the umbrella of NCIP Hub.
Now that won't constrain what we need to do because the hub is very versatile
and we hope to attract to manage this forum
the kinds of people who know about community building
and are comfortable using information and information systems
and tools that is then built by others rather than insisting that they have to build their own.
So, what are our goals for this particular forum?
We've tried to be in what we will advertise for this forum, very basic.
We don't really want to tell the cancer research translational community what they need.
They know what they need, and so those who will apply will be encouraged to go
out to the community, find out what the needs are
and then develop their application to meet those needs in a variety of ways,
and I'll go through a little bit of this as we go along.
But, the major goals are in fact to deploy
and refresh the knowledge about animal modeling.
This is an enormous task.
We're not talking about a modest amount of information.
We are talking again about worldwide connections
that enable the best of cancer modeling.
Another thing that we've learned from the Mouse Consortium over 15 years is
that face-to-face meetings occasionally are really, really valuable.
You get a great deal more interaction than you do in a web setting.
And so the forum applicants are going to be required,
or the successful awardees are going to be required,
to convene an annual meeting right here in lovely Shady Grove building.
We can then have workshops elsewhere, perhaps in our cancer centers,
perhaps abroad and then we will have regular webinars as well to deploy information.
And we hope to use the HUBzero capabilities,
this interaction collaboration platform, as a way to support collaborations
across the many communities who could benefit
from knowledge about using animal models.
So, how are we going to support the goals of the forum?
Well, this is of necessity quite brief
because these funding announcements are under development.
I see my colleagues from the Division
of Extramural Activities who are really working with us to try to make sure
that what we put out is, makes sense to the community
but we will have a funding announcement for a U24. U24 will guide the forum.
But, all the work of the forum in other words,
many of the projects that will enhance how we use mouse models translationally
will be supported by two other funding announcements. One is collaborative R01 projects.
These are sort of like little starter P01s if you will.
They're a chance for a clinician or a radiologist or a team of people
from very different expertise coming together,
thinking about a problem that an animal model might help to solve
or animal models or some other way of tackling questions
about how do we use animals appropriately, translationally.
And those collaborative teams would bring a variety of expertise to the table.
We also expect to have smaller projects.
R01s. Some of the goals of the program are in fact
to make the model, make a given model better.
So, maybe that would be supported by a regular R01
or a competing revision to an ongoing R01.
But, all of these projects will be designed to help us
in the Institute do this kind of research and connection
to the translational types of work and the connection the clinic
or to epidemiology much more satisfactorily and consistently.
Probably some of you have heard over the last couple of years about this issue of reproducibility.
It is a challenge in science and I'll get back
to why I think the HUBzero platform will really help us begin
to accomplish some of those goals as well.
So then, because many of the needs
for integrative cross-species research are information technology,
bioinformatics, there are projects that could come in to serve those needs
in response to the ITCR FOAs that already exist.
And in fact, there's several of the projects that are now funded. For example,
a U24, a network analysis that will be very helpful to us because those kinds
of tools are really critical to the integrative use
of animal models and human data.
We also expect to be able to connect to other NCI
or NIH networks and consortia.
We can also encourage people and are encouraging people to apply from overseas
because I said, mouse modeling as well
as cancer research is done in a global sense.
So, the two related funding announcements to the U24,
we're trying to encourage cross-disciplinary teams and they would--
they may be focused on a variety of things such as,
what are the real technical aspects of trying
to use mouse models in human integratively?
There may be differing expertise that's required
to address an unmet translational need.
Some of the things have to do with, how do you use mouse models really well
to do combinations of therapies or to try combinations in a different schedule?
So, these are hard to do as clinical projects but by marrying the needs
of the clinical research community with people who know how to use animal models
to ask those questions might help us in the future to design some
of these alternative types of clinical trials.
We also think it's very important to point
out that there are many other disease states that are precursors
or are raising the risk of developing cancer -- pancreatitis, asthma,
some of the other diseases that are in fact not funded in the NCI
but the expertise exists in other Institutes.
We will be encouraging people to reach out to people who study normal biology,
who study other disease biology that's related to cancer and merge their expertise with cancer.
And then last but not least, again I mentioned that we want to make sure
that there are people who can help us work with in the forum.
We're going to require that the awardees attend the annual meeting as well
as at least one of the workshops and to help us formulate the webinars.
So, the other one is smaller projects.
R01s and competing revisions, we think they will have a narrow risk scope,
although they maybe multi-TI R01s.
They may be focusing rather than a large question that focusing on some narrow technical topic.
And again, they may be rather than doing something that's global,
they may be looking for something that is much more discreet in terms of improving translational use.
And if possible, just like the length collaborative R01s, if they are--
if it's possible, they might use the HUBzero platform or the NCIP Hub platform
in order to foster collaboration.
So, those are in general what we're talking about.
So, the ITCR FOAs, I just said that there's--
this is just the tip of the iceberg in terms of the kinds
of bioinformatics tools that we need to do this kind of integrative biology.
And much of this can be prompted by people taking existing bioinformatics tools
and simple repurposing them so that they are much more relevant to a cross-species comparison.
We need tools to help us capture the workflow.
One of the things when you talk about reliability or reproducibility has to do with transcriptional errors.
If there were a better way of capturing data from the animal experiments and being able then to deploy it
in using the HUBzero platform then people would have a better idea of what kind
of data are captured and what goes in to ensuring reliability and robust use of animal models.
So, it isn't a matter in many cases of going out and reinventing.
There are lots of opportunities to put things together.
And one of the nice things about the ITCR is in fact, they too will have annual meetings
and we hope to get an off a lot more cross-talk between that program and this Oncology Models Forum.
So, I've talked here a little bit about cross-species research
but if the collaborative platform is actually there, in other words, when we use HUBzero, what kinds
of projects would we be thinking about supporting?
And I'm not going to give you a science lesson at this point but I've picked
out a few projects that I think show what sorts of things we're thinking about,
and the community is thinking about, in terms of using that HUBzero platform.
So, one area of research that goes to the heart of reproducibility has to do with how you validate a model.
This term was coined years ago by the Mouse Consortium when we were first asked,
you know, it's a mouse, it's developing mouse cancer, how do this relate to human cancer?
And so, we put together a team of pathologists, it is now a worldwide network,
we can call upon 450 pathologists. These are not medical pathologists --
These are medical pathologists, not mostly veterinary pathologists
that they've learned how you examine a mouse tumor and how you relate it to a human disease.
And so, this particular report that I have up here is just one of 14
or 15 that the Mouse Consortium produced over 15 years.
This is the latest one that has to do with prostate models.
Well, what you can see here is that there are stages, for example, of cancer
that can be related to the human disease.
And over the course of many years, we've developed not just the best ways
to fix the tissue and stain it and all of those technical procedures.
We've also worked with the NCI to develop the ontology that allow us
to cross-compare to human and the vocabularies that enable us
to capture whatever the concept is in mouse and then cross-compare it to human.
That's taken a lot of effort.
And for that, we thank the NCI for its support because without having the people
who do the ontology development and work with us to capture this sort
of information, we wouldn't know in many cases,
we wouldn't have a standard way of calling this.
I can only tell you that this particular, at this particular meeting,
we didn't just look at genetically engineered models,
even though that has been primarily the focus of the Mouse Consortium
because the people in the Mouse Consortium don't just use genetically engineered models.
They use every tool in the shed to inform the design of this GEM models.
And so, this particular meeting included people who came
with patient-derived xenograft specimens as well as dog and rat.
So, this paper represents a compendium of analysis,
cross species analysis of mouse prostate tumors.
So, although it may not look like a lot of data,
this little, as our pathologist-in-chief Robert Cardiff likes to say,
the postage stamp pathology that you see in a publication.
Behind these pictures is a whole large collection of whole slide images. They're very, very big images.
And if you're a person who's developing a new model
or you have your patient-derived xenograft then you're able
to get good whole slide images, maybe you want to cross-compare
with the mouse images or the products of these particular meeting.
So, what do you do? Do you download all of those images onto your own server?
That's a significant problem.
There are firewall issues. There's all sorts of issues.
There's also the problem that these have been annotated with a variety of tools
and you may not have those tools.
So, HUBzero, that platform, can help us accumulate all of this data
and this is just a small, small subset of what exists worldwide and allow people
to come to the site and maybe even on their iPads, look at the images
and cross-compare theirs without having
to download absolutely everything onto their own servers.
Then, of course they forget to take it off the server
and people get irritated and yada yada.
But, this is as I said only one of 15 papers that's been published
in the last 15 years and increasingly, this can serve as a tool to explain
to people, what do we mean by a validated model?
This is the beginning point for validation.
This isn't just the only thing that people do.
But, the point is that this is, this constitutes enormous amounts of images
and we need a facile way to share those images worldwide.
So, I want to go on to a paper that really illustrates how you make a model
and validate it and then use it translationally.
This is the laboratory of Cory Abate-Shen at the Columbia Cancer Center.
And what this illustrates is that you start with human data.
These are two clinical studies that were done and they showed
that if you had a combination of high PI3-kinase and Ras present,
then you didn't live as long as people who have a much lower overexpression
or a much lower expression of those two pathways.
And the same is true of this study although, they were asking a different question.
This was prostate cancer specific survival and this had to do with a biochemical recurrence.
But the point is that the data here suggested to Dr. Abate-Shen
and her colleagues said if they wanted to have a model to study metastatic prostate cancer,
then perhaps they should engineer it a certain way
so that it would have the same genotype as these patients.
And so, they did that and you can see up here that in fact,
there's a strain of mice that only develops prostate intraepithelial neoplasia.
We don't call it PIN in mice. We call it mouse PIN, to make sure people don't get confused.
This particular animal corresponds to the patients
who would only have elevated PI3-kinase but this corresponds to patients who have both.
And I show you over here, just a survival curve, which is reminiscent of the human survival curves as well.
So, again, one of the things that you do with a mouse model to see
if you can learn from it for your clinical trials
or for your clinical investigations as well as to validate it
for how well it resembles human cancer and therefore,
you have a comfort level using it then you do some
of the same things that you do to patients.
You can see up here in the top that they use MRI
to investigate what's going on in the mouse patients.
These are again whole slide images and you wouldn't just be looking
at the whole slide image for one of the mice.
You'd be doing a whole slide image for all of them in the cohort.
And then, you have high power path images of the same pathology and then,
commonly just as we do with patients, you stain the tissues for various markers
and some of it, some of these are very standard things and some of them are far more elegant.
And again, are large images to capture.
So, for each of the animals you're going to have a suite of data that's really quite big.
Moving along, so these are just very standard or classic types of things that people do in mice.
The next thing is to look at in the standard analysis what are the networks
and pathways that are perturbed in these animals.
And so, not to belabor the way in which this is done
because these could be captured as a process on the HUBzero platform
and everyone could see how you took the data and how you analyzed it
and how you made your cross-comparison rather than having to your--
you're trying to write it out as a description to go in your paper.
At any rate what you, what this shows is that in fact,
the mouse models do recapitulate very much of the signatures of the human.
Over here are indolent tumors and malignant tumors and then,
when they did the models just themselves,
they came up with this signature of indolent cancer.
I can tell you that although, I'm not showing the paper,
that particular signature was a candidate biomarker signature.
It was tested on a retrospective set of human sample.
And now, it is in a clinical trial with human specimens in a prospective trial
because that particular signature appears to be able to look at biopsies
of human patients and tell whether they're likely to progress or not.
And then again, we have the sorts of things which mouse modelers tend to do
which is to then use the tissues for discovery.
What's actually, what are, which genes are really involved
and how can I document which of those genes is actually involved?
So, I'm going to go back to this again because I spoke to Dr. Abate-Shen
and her colleague, Andrea Califano, who is a bioinformatician,
he does computational biology and they have been working on this biomarker signature.
And they have a new paper that's about to go into a major, major, journal and they sent 200 pages of appendix material.
Well, the journal was not pleased because 200 pages of supplemental material
because they wanted to document the process where they took the animal data
and they used it to inform human and vice versa.
They wanted to document their process. They wanted people to have access
to the process whereby they did these 200 pages.
So, the journal said, "Oh. No way. We're not going to let you do that.
You're going to have to pick out salient pictures here and there and we'll get some brief description."
HUBzero platform. You can actually work together
in that platform then you can document the process you went
through to examine the data to put it together, what tools you used,
the tools would be there for someone else to see.
And so, from a standpoint of being able to go back at some point in time
and say, "How did they do this?"
And also, "What did we learn about the validity of these kinds of experiment in mouse and human?"
It's a forensic sort of process.
But, the point is when we get to this issue of reliability and reproducibility,
we think this platform, if used by people
to document their collaborations, could be a very valuable tool for this kind of translational research.
Another one that uses an immense amount of data,
this paper came out I guess last year, it's a group from Stanford
and they're trying to find new drugs for small cell lung cancer.
For anyone who knows small cell lung cancer is not a really good suite of drugs that can be used with it.
So, they had a paper probably three years earlier in which they described how all bioinformatics tools
and so forth that they would use to do drug repositioning.
And again, I don't want to go into excruciating detail here
but they evaluated the therapeutic potential of some of the FDA-approved drugs
that were out there already by examining huge numbers of datasets,
of transcriptomics basically, where these agents were used in cells of all different kinds.
So, when I spoke to the guys form Stanford I said, "Gee. That must involve a lot of data. How do you make that work?"
And they said, "Well, you have to download all of it onto Stanford servers,
they have a firewall, there are problems with that, when you get it all there,
if you want to work with somebody else, they actually have to come to Stanford
and sit in our offices and work together."
So, I pointed them to the HUBzero platform and a few days later,
I got a response back saying "Huh. This might make life a lot easier for us."
The bottom line was that their process actually came up with the molecules
which are what as you saw, tricyclic antidepressants, who knew.
And they then tried them out in a variety of animal model.
So again, you have to capture a huge amount of information
about the animals themselves and how you made the models or how you develop them
in this particular setting, then you have cohorts of animals about which you have
to capture a lot of information.
And typical of people who do this kind of science using models, they use three different strategies.
Panel B is an animal that is a GEM model whose tumor was removed and a cell line was made out of it.
This is a cell line from human and this is what's called a patient-derived xenograft.
So, they used three modeling approaches.
And I won't go through the rest of the data in this paper but the point is
that now, they have a clinical trial in which they are taking one
of these antidepressants and they're trying it in small cell lung cancer patient.
And by the way for anyone who cares about this,
there actually is a pig in a cage in Iowa that developed small cell lung cancer
and we'd dearly love to try this particular approach on that particular animal.
At the moment, we don't have a way to support that sort of thing but we think that this funding announcement,
they in fact allow those folks to collaborate with mouse modelers and with clinical trialists to see
if in fact a swine model might be an improved model for small cell.
I don't know how we're going to afford to have troops of pigs but anyway,
the last topic is what is called a co-clinical trial.
Now, co-clinical trials very briefly,
a co-clinical trial is a trial in which you take a suite of models, not just one model but a suite of them
because of the heterogeneity of human cancer.
You take a suite of models that are appropriate for that particular subset
of human cancer, and you run a trial in parallel with patients who have progressed on the standard of care.
They are trying new combinations and the point is that you can run those trials
at the same time and as you begin to see biological change in the mice,
if it's important to tell the clinicians about then you can tell them,
"We're starting to see our little mouse patients with this genotype are failing.
So, you might want to start watching your patients a little bit more closely."
At any rate, this particular co-clinical trial was done as you can see,
with a host of people, as usual a large mob, and they're not all exactly in the same place.
And so, they need ways to communicate much more rapidly but the point was
that they wanted to see if they had patients who were on a clinical trial who had KRAS mutant lung cancers,
could they determine from this co-clinical trial if adding another drug in would help?
So, this is just a little bit of the data, these are--
up at the top there in A is a typical waterfall plot with the mice.
You can see that responders and the non-responders and you can also see
that it matters that the genotype matters. These are engineered mice.
And then, they added in the additional drug, this MEK--
ME-kinase inhibitor and you get some response.
But, the point is that these data were being generated while the drug trial was
underway and they were able to point out to the people doing the clinical trial
that patients who have this particular hallmark, KRAS and LKB1, were not as likely
to respond to this combination.
This is just a little bit more data they used, FDG-PET on both the mice
and the humans and they were able to come up with a signature if you will,
in the mice that helped to inform what they were seeing in the patients.
And this is simply continuing to generate data that shows
which of the combinations or which of these combinations in which patients it might be effective
and these are just the usual survival curves.
And so, what they found out was that if you had a, if you were a mouse
or a human with either a KRAS mutation or a KRAS and p53 mutation,
then you were likely, much more likely to respond to this combination.
And if you had KRAS and these LKB1 mutations, LKB1 is important in metabolism, it's an interesting finding.
And then, putting the FDG-PET together between mouse and human,
they were able to come with some biomarkers
that they believe would help the clinicians figure out who was on a trial and maybe failing it.
And the challenge here was that the trial that they were in running in parallel
with didn't have the resources available to go in and retake--
to take the patient's specimens and analyze them for some of these biomarkers.
Again, here's another opportunity. If you were to run one of these,
you might be able to show that if there were better ways to capture the molecular information
from the patient's specimen then one might be able to do a better job tailoring therapies.
So, this is actually one of the, there were several clinical trials of this type
that were going on at the same time as the co-clinical trial.
And if you read the paper, you find that in fact the two together,
these two drugs together in fact that does give a better outcome
for the patients, although, as you may remember from the mouse data,
obviously patients and mice progress on the combination after a while.
They do live longer, they do better but they do eventually progress.
And one of the cool things you can do is take those animals that have been through single or multiple therapies
and you can investigate why they are becoming resistant.
Again, looking for molecular hallmarks of resistance that then one might be able to investigate in patients.
So, to get back to the hub, this is the way we currently see it
and we see that we're going to have people who are going to be working in the hub.
We'll be doing, we hope-- We'll be on the collaborative R01s.
We'll be doing collaborative projects.
Some of the people on the R01s, well, yeah,
they'll probably be part of the forum.
They don't have to be but we're hoping to induce them to be.
The ITCR projects, some of them will fit
within the Oncology Models Forum, some of them won't.
And these are just some of the things that in fact we imagine that the R01s
and our connections to other entities within NIH and the NCI will allow us
to develop focus groups to serve a variety of communities.
We expect to be able to link much more strongly to international folks.
We will be working with cross communities to develop standards and protocols
that are reasonable for people to use, as well as having meetings
and accumulating resources and linkages to those resources.
But one last item, our educational projects.
We were thinking about this because we have already a number of videos
and things like that that we can deploy.
I don't-- I'm no expert in 508 compliance but it seems to us that perhaps
if we worked with communities of people, develop--
who develop educational materials,
that perhaps we can start developing materials that would be suitable for everyone.
>> All right.
Thanks very much, Cheryl, for this very interesting presentation.
[ Applause ]