modENCODE as Part of a Drosophila Search for Cancer Therapeutics – Ross Cagan

Ross Cagan:
Okay, great. Thank you very much. I actually learned quite a bit about chromatin and transcription
these last couple days, so I appreciate that. And also because the modENCODE program is
having already in my lab, and is going to have in many labs, a big impact. And I appreciate
the opportunity to talk about how we’ve been able to take advantage of this. My lab is
a fly lab. I should also say before I go on that nothing I’m going to talk about today
is relevant to this, but I do want to disclose that I am a co-founder of a company, stockholder
and so on. So here’s the basic problem that we have,
and that is it’s a lack of success of drug discovery in clinical trials. And this has
really been hurting, for example, the cancer field, which has the lowest success rate of
drugs in clinical trials of any major disease, despite the fact that it has been the field
of most focus. For example — and I’ll talk about colorectal cancer in a bit — three
percent of drugs that go into clinical trials for colorectal cancer succeed. In fact, we
really don’t have much — many good drugs for colorectal and it’s already the second
leading killer — cancer killer — of Americans. So what’s the problem? Obviously, all these
drugs have succeeded with our standard mouse models, otherwise they wouldn’t have gone
into the clinics. And I’m going to discuss some of the issues about the complexity of
the models that we use, and I’ll show you some of the data and some of the things we
found in flies that suggest or indicate what some of the issues are. In a larger sense, these are some of the issues
I’m going to talk about. First of all, our tendency to simplify disease — and I think
my thing is dying here — our tendency to simplify disease — I’ll see how it goes.
That would be too easy [laughs] because this advances as well. I’ve got it all worked out.
So let’s see how far this goes, and then I’ll use the other one. So first of all, we have
a tendency to simplify disease, and while that makes working on things easier, it has
been a problem because cancer, and diabetes, and so on, they’re not simple diseases, and
it’s okay to embrace that complexity. I’m good, thanks. Second of all, there’s been an assumption
in the field, and I think not necessarily a correct one, that identifying the driver
of the disease is the same as identifying the best therapeutic target. This is a more
subtle point and one that we often miss as basic researchers because we really are focused
on identifying drivers of disease, and I think we’re going to have a panel discussion later,
where I hope this issue comes up, that there really are differences if you’re focused on
mechanism versus focused on therapeutics. And this is one of the reasons why there is
a difference. And I’m going also talk about the differences between single- and multi-targeting
drugs. Each has their advantages and disadvantages. I’m going to show you our efforts to go after
multi-targeting drugs, and I think it’s obvious where I stand on the cell based versus whole
animal screening against — again, each has its ups and downs. Our expertise is in whole
animal. So I’m going to talk about two stories today,
or at least two cancers, if I have time. And the first one I’m going to start with is something
called medullary thyroid carcinoma. This is — all you need to know is a couple things.
This is a very, from a genetic standpoint, a very simple cancer. It is due, in most patients,
to an activating mutation in the ret receptor tyrosine kinase. All right. So red is a classic
RTK. It binds gene GDNF and has a co-factor. It activates downstream pathways like Ras,
and Src, and PI 3-kinase, and so on. I’ll get back to those in a moment. Patients that have MTCs have either inherited
or spontaneous mutations in either the extracellular domain that drive dimerization in the absence
of ligands, or the intracellular domain that actually open up the activation loop and,
in fact, those receptors can be active as monomers. The result is a proliferation of
the parafollicular cells, or C cells of the thyroid, and basically these are the one — this
is the one tumor type — thyroid tumor type — that can be fatal. These tumors can metastasize
and go to distant sites in the liver, and so on, and if allowed to progress, as I said,
can be fatal. And when we started this project, there were no approved chemotherapies for
this cancer, and I’ll show you our efforts towards that. So how do you model thyroid cancer in a fly?
Well, flies don’t have a thyroid. To be fair, they have a very skinny neck, so we may have
missed it, but I don’t think so because they don’t have calcitonin. [laughter] So, instead, we took oncogenic Ret, the various
isoforms, and we targeted them to the eye, right? And just to cut to the chase, as you
can see, compared to a normal eye, the result is proliferation, and you could see little
tumor-like growths here. Compensatory apoptosis, which isn’t cell fate if you drive it during
development. Many of the aspects that you see in real tumors, we see here. Now this
is not a perfect model. As I said, it’s not even in the correct cell types, but as I’ll
also show you, it’s been a useful model. So we’ve done many things with this model.
Perhaps one of the more interesting things is that we’ve used that to develop a method
for screening flies in relatively high-throughput using a variety of compound libraries. And
what we do is we take liquid handlers that move food and drug into each well. We have
a modified fact sorter from Union Biometrica that spits 10 embryos into each well. We put
an oxygen-permeable lid on top. The flies hatch out, they eat the food, they eat the
drug, and it’s a simple, like a Phase 3 clinical trial in a dish. Okay? We ask very bottom
line; we’re not making any assumptions about what the best therapeutic target is. We just
ask for a drug that makes the tumors go away, and the fly lives through the experience.
That is efficacy with minimal toxicity. Also, bioavailability, compound stability, and so
on. And I have to say, this has worked out great.
And I’m just going to cut to the chase here because I’m actually going to a second, more
recent story that was just published a couple of weeks ago. So one drug that we hit was
particularly notable, and that’s a compound originally called ZD6474, which was developed
by AstraZeneca. In the presence of this compound and the oncogene, you could see that the eyes
almost completely normal. And what was striking about this in the fly is that it could cure
the fly tumors at a concentration that was about 30-fold lower than was required to harm
the fly. So if you were a human, you would say that’s a therapeutic index of 30, if concentration
in the food matches concentration in the animal. So I just want to cut to the chase here because
I have a lot to talk about. There is important work done by Massimo Santoro,
showing that ZD6474 also worked well in human cell lines, and then this was picked up by
Sam Wells, our close clinical collaborator, long-time collaborator, and brought to clinical
trials, where, last April of 2011, it was approved as the first chemotherapy for medullary
thyroid carcinoma. And so I just jumped across many years. All right? So, what is — and it has a commercial name
— Caprelsa. All right? You can buy it at your local drugstore. So how does Caprelsa
work and what did we learn from that to maybe make a better drug? And to go right to it,
Caprelsa does not, we think, work primarily released exclusively through targeting Ret
itself. All right? In fact, it’s a terrible Ret inhibitor. Caprelsa is what I would call
a low affinity kinase inhibitor. It hits many targets, but it seems to hit them in just
the right ratio that the fly can take that hit. We can take that hit. But the tumor has
difficulty selecting around it. And, in fact, the best data that Sam tells me to date is
that patients that have taken this drug for at least two years, only 20 percent have become
resistant to date to this drug. And I think there’s just too many avenues that you need
to get around, the tumor needs to get around to get away from this. So we’ve taken this to heart, and I’m going
to talk about how we use that idea to go after what I’m going to call rational polypharmacology,
whereas this would be, I guess, irrational polypharmacology. And to enter this story,
I need to tell you that we did a genetic screen against oncogenic Ret, and most people here
would understand how we would do this. This is a simple donit [spelled phonetically] modifier
screen. And when I say “we” I mean Renee Reed, a former student at the lab. She identified
140 genetic modifiers that are required for Ret to drive that tumor. And just to summarize,
the majority of them fell into three pathways: Ras, PI 3-kinase and Src. And I have to tell
you we’ve really beaten this to death. There’s a lot of signaling both within the tumor and
in the neighboring tumor. This is why you can’t model this well in a dish. But what
was interesting is another hit that we had, and that brings me to modENCODE, was a chromatin
remodeling protein, or at least associated protein, called Sin3a. This is homologous
to actually two proteins in humans: Sin3a and Sin3b. It has been — it’s well known
as a transcriptional repressor, okay, as an HDAC. But that’s actually not true. Here is a schematic of — and this is work
by Tirtha Doss [spelled phonetically]. Here’s a schematic of the complex that it’s in. This
is the best known complex as an HDAC. But interestingly, when we began this collaboration
with this work, I contacted Kevin White from modENCODE. We supplied them an antibody. They
did a ChIP-seq for us, and what we found was about 100 targets, so it wasn’t many hundreds,
that’s good. And the targets that it hit were interesting. Okay? He’ll go away. So here’s about half the targets.
We validated all these. What’s interesting about this is when we knock Sin3a down, if
you look here in the red, most of the targets — not all, but most — actually go down.
That is, it acts as a transcriptional activator. And I can tell you that 80 percent of the
targets for Sin3a, it acts as transcriptional activator; 20 percent, it’s a repressor. And
what’s more interesting to me is that virtually all of its targets — that it’s an activator
— are tumor suppressors. They’re a who’s-who of how to make a tumor. And virtually all
of the things that acts as an inhibitor, it targets [unintelligible] inhibitor, are oncogenes.
All right? It’s remarkable how they can activate gaps and inhibits gaps and so on in the same
pathways to really drive a pathway in one direction. And just to show you a few examples of many,
we hit Ras itself, Akt, actually, multiple things in the Ras’s pathway, Akt and PI 3-kinase,
Csk, the major negative regulator Src; a host of active remodeling proteins, regulators
of jun kinase. And what’s interesting about that is that I had assumed that Ret activated
these pathways through direct binding of partners that would activate Ras, and PI 3-kinase,
and Src. And what this told us is that a primary, if not the primary regulator of these pathways
actually rotates through transcription and then comes back to regulate the receptor.
In fact, we think Sin3a is a major regulator of Rtks in general, and if you’re going to
have a tumor, you need to get out away from Sin3a. And our data in eight different human
tumor lines shows that all eight of those, a variety of tumors, Sin3a is consistently
and strongly down-regulated in these tumors. All right? So we postulate that you need to
drive Sin3a down just to be able to escape this repression and actually have a full-blown
tumor. Now, how do we use this to get into drug discovery?
That takes me here, and a collaboration with our lab, especially Tirtha Doss, and Kevan
Shokat’s lab with his post-doc, Arvin Dar. And to get further with this to develop compounds
to take Caprelsa further, we set up a screen against oncogenic Ret model, not in the eye
anymore, but actually several places in the fly, causing tumors to develop in multiple
places in the fly and it actually kills the animal, and so Tirtha set this up, so that
half of the animals make it to pupation and half died. None of them make it to adult. Now, if I was a pharmaceutical company, the
simple thing to say is, “All right, let’s make a clean Ret inhibitor. Everything that’s
wrong in this fly is due to oncogenic Ret. We know this because we made the fly. So Kevan
and Arvin developed one called DLO6 that’s as close to a clean Ret inhibitor as I’m aware
of. It works beautiful in cell lines in a dish. However, in the animal, at the highest
dose we can give it before we kill everything, it basically does nothing. And we have a lot
of data, and I’ll show you more, that there’s no correlation between activity against Ret,
generally, in our screens, and utility as a drug. And that’s probably because Ret, like
many oncogenes, is important for cell viability, for organism viability. That’s why they drive
tumors, because they’re fundamental to the biology of the cell. Hitting them with drugs
will often cause tops [spelled phonetically]. Okay? All right. So let me cut to the chase here
and say that we screened through a library developed by these guys, a library that specialized
in hitting multiple targets. We had one hit, which we call AD1, for Arvin Dar 1, and you
could see it’s a pretty good hit, that more flies make it to pupation, and now we’re beginning
to get adults. Now interestingly, and here’s the — here’s the composition matter here
— interestingly, if you look at close analogs of AD1, AD1, and AD3, AD2 is highly toxic.
And yet the only difference chemically between 2 and 1 is just this trifluoromethyl group
on the terminal fennel [spelled phonetically] group here is lost, is missing an AD2. That
small change causes this to become highly toxic, despite the fact that this would look
the same in cells in a dish, and so on. The animal’s telling us it has marked differences.
AD3 has this extra methylene group at the center core here; that tiny difference causes
AD3 to be almost inactive. All right? So it’s very striking how the animal gives
you a very different result, and this, too, I think, is a key point. But also, remember
that I told you there were three pathways that matter: Ras, PI 3-kinase and Src. So
what do these drugs do to the rate-limiting enzymes in these pathways: Raf, Tor, and Src?
If you look at in vitro kinase data, we find that AD1 hits all of them. All right? It hits
Ret but notice they all hit Ret. There’s no correlation between activity and Ret, okay?
But AD1 hits Src, it hits BRAF, it hits mTor, so you’re good to go. It gets all three pathways
that are genetics that are needed. AD2, interestingly, it hits Src, it hits mTor, but it doesn’t
hit BRAF. Now why would not hitting a target actually
make you toxic? That was counterintuitive to us at the beginning, and just to cut to
the chase, we weren’t the first to show this. A number of laboratories, including ours,
have evidence for this, that there’s a feedback loop between Tor and BRAF. If you don’t — that’s
required to suppress this pathway. If you inhibit Tor, this feedback inhibition is relieved,
Ras signaling goes up throughout the animal, and that’s toxic. And we have a lot of data
showing that when you give this drug, Ras signaling goes up everywhere in the animal.
Okay? And the tumors actually get worse. Now, how do we know the toxicity is due to this?
Well, let’s look at AD1. Let’s remove one genomic copy of the downstream target erk.
There’s all your toxicity. Okay? So, from these sorts of experiments we realize
we could actually begin to walk through the kinome and identify better targets, and also
identify what we call anti-targets, those targets that need to be left alone, or the
drug becomes toxic. And so we play this game, for example, with AD1. If we remove a copy
of erk — now that’s a drug I’d love to have. Okay? And that tells you that we can get a
better balance between these two things. So we played through this game quite a bit. I’m
not going to show you all the data. And we went back to Kevan and Arvin and said, “Okay,
we need a drug. It needs to hit Src, it needs to hit BRAF, it needs to leave mTor alone,
but it still has to knock the pathway out, so you have to hit the next step down below
the feedback loop, which is S6 kinase. Can you build that drug?” And to my continually
stunned amazement, they can do this. They sent back two drugs after they did computer
modeling. They sent back two drugs, shall call AD1b — oh, and I should say if you remove
a copy of Tor here, AD1 becomes AD2. Also evidenced, the importance of leaving that
alone. Okay? And they sent two drugs back that have that profile, AD1b and AD1c. And
you can see these are drugs that we really are interested in, and they’ve been refined
to maximize therapeutic index. Let’s look at AD1B. It hits Src, it hits BRAF, it leaves
mTor alone, it hits S6 kinase, we’re good to go. All right? How does this work in a mammal? Is this just
a fly phenomenon? And the answer is no. So here’s the drug that was approved last year,
Caprelsa, and it works well in human cells in a dish. These are MEN2B cells, but you
can see AD1 and AD1b work much better — about 500-fold better in a dish. If you look at
a xenograft, and I’d love to take better models but oncogenic Ret put in a mouse really doesn’t
do much, so we looked at xenograft models. We actually force the tumors. We forced the
issue by growing the tumors for 46 days, full-blown tumors and then ask the drugs to actually
reverse it, which is different from previous studies. And in this more stringent case,
Caprelsa really struggles to deal with those but AD1b does well. And importantly, if we look at body weights
of the mouse, based on those body weights, at high doses, AD1b shows little difference
to vehicle whereas Caprelsa shows the expected toxicity. Okay? So — and this drug is now
being licensed, and we’re hoping that, with luck, this will be in clinical trials within
the year. So to summarize the section, to wrap this
up, we think that model organisms, using sophisticated network analysis like modENCODE, which we
were able to take advantage of because it was there, it was essentially off-the-shelf
information, combined with sophisticated modeling of medicinal chemistry, can be very useful
for identify both targets, things that need to be knocked out but in a complex way, not
single targets, but also what we call anti-targets, things that need to be left alone or they
can drive toxicity, and only in these whole animal approaches, in my opinion, can you
actually do this to generate better, more sophisticated, and hopefully, we’ll see, more
useful drugs. All right, you guys with me, because I’m going
to finish with one last little story. Okay. And that’s this. And this takes me to the flipside. So I showed
you what we think is the importance of developing polypharmacology, rational polypharmacology.
Let’s go to the other side and talk about models. And the importance of embracing complexity,
which I believe is exactly what modENCODE is about, or at least it’s stepping towards.
And I mentioned at the beginning colorectal cancer; we’ve been modeling breast, lung,
and colorectal, as well as thyroid cancers, and I’m going to show you this work by a really
terrific post-doc in lab, Erdem Bangi. And what he did is he went into the human sequencing
data — I’m sorry, am I standing in your way? He went into the human sequencing data and
asked, “What are the most common triple, quadruple, quintuple combinations of oncogenes and tumor
suppressors that you see in currently sequenced patients?” So we worked with the Vogelstein
group, and I can tell you that to date, the most common quadruple combination mutations
you see in patients is RAS P10 APC P53. All right? This is the second most common quadruple,
this is the third. Erdem built all of these, targeted the trans genes to the gut, using
conditional activation, turn it on only in the adult, okay? So no cheating in that. And
he also, in addition to the quadruples, he built all the subset triples, all the subset
doubles, all the singles. That’s 15 fly lines for each quadruple. And for those of you are
wondering why we stopped at four, there are no two currently sequenced tumors that share
five genes. Okay? The statistics fall off sharply. All right, so we learned quite a bit about
[inaudible] details. Erdem has done an impressive job. Am I feeding to them? All right. Could
it be my phone? Can you hold that? I actually did that once, it’s a true story, and my mother
called, and I handed it to somebody right there, and they chatted for a little while
with my mom. [laughter] [laughs] So at least my mom now believes me
when I actually do give talks. So Erdem took these four — I don’t know why that came out
— Erdem took these four hit models, and he’s done a very extensive analysis, and we’re
pushing towards a network analysis. I really don’t have time to get into this. But some
of the phenotypes you find are hyperproliferation, multilayering, and release of cells from the
tissue and distant migration to form secondary metastasis-like growth, I’ll say. Also, aspects
of senesence and apoptosis. And what’s really fascinating about this is that different aspects
of these emerge as we look at different combinations of oncogenes and tumor suppressants. All right?
And from a network analysis, this is really an interesting problem and one we’re trying
to grapple with. Just to show you what a tumor looks like,
so, and also to be clear that this four-hit fly has 10 transgenes in it, so this is very
difficult to build in a mouse; I know, because they’re trying. Here is part of the gut, so
it includes GFP. This is part of the gut here. Here’s the muscle wall around the gut. This
is the hind gut. This is what’s called the trachea, for those of you who aren’t familiar.
This carries oxygen in the fly. And what happens is, in addition to a polyp-like formation
and growth, and so on and so forth, is that periodically cells will actually pop out of
the tissue. They’ll push their way through the muscle wall. This one has extended a process.
It’s found its source of oxygen. It corkscrews around it, and the cell will walk right out
of the gut. And in many cases, just walk along the trachea off into distant sites, really
throughout the animal, as far as the head. Okay? Here’s another example, a bird’s eye view.
The gut is actually below the surface here. Here’s the muscle wall. Here’s the trachea,
and here’s the cell. It’s found its target; it’s now hopped up onto the trachea here,
and it’s off to the races. All right? It’ll go off to distant sites. In at least some
of our tumor models, we also — just for those of you who are interested — we do see the
equivalent of neoangiogenesis where it’s just neotracheogenesis of oxygen bearing trachea
back into the growing tumors, and so on. Okay? So many of the aspects that you see in human
tumors, we see here. Now here’s my final question: What’s the difference
between a one-hit and a four-hit colorectal model, all right? And what can it tell us
about why drugs have failed at such a high rate in clinical trials? There’s many differences,
and we’re really trying to wrap our heads around it through network analysis, RNA-seq
analysis, and so on. But let me just show you something — one thing that we’ve learned. First of all, if you just look at the cells,
here’s a one-hit model, equivalent to the fly — the mouse K-Ras model. So if you put
Ras into the fly gut, you will get cells coming out of the gut, you’ll get proliferation,
they will go to distant sites. But if you just look at the cells, they’re quite different.
They’re smaller and less robust than the four-hit mouse, okay? But here’s a more interesting,
I think, and a more practical difference. And that is that Erdem took 16 drugs, many
of them clinically relevant, many of them failed in clinical trials for colorectal cancer,
and he asked, “What is the sensitivity of a one-hit model versus a four-hit?” And that’s
shown here. Of the 16 drugs he tested, 13 of them worked fine in a one-hit fly to knock
those tumors down. All right? But zero of 16 worked in the four-hit mouse — in the
four-hit fly. And I think that’s really telling us exactly why these drugs have failed, because
the complexity of our models — our models just don’t have the sufficient complexity,
and remember, we’ve made all the two- and three-hit combinations and so on. We’ve mapped
out the resistance for multiple drugs, and using that information in biochemical analysis
and so on, we’ve now begun to identify combinations of drugs. For example, these combinations
— that is actually successful at knocking these tumors down. And Erdem has done a lot
of biochemical analysis. This is really a fascinating story on why particular combinations
of drugs will work on these, and why others won’t, and so on. And this is something we
can talk about in the discussion section. So let me finish up by telling you what I
told you — excuse me. What I told you is that we’ve taken a whole-animal approach to
try to build complex drugs. So we’re not focusing on making the drug cleaner and cleaner, we’re
actually going the other direction. The problem with polypharmacology is that you need to
know all the targets for medicinal chemistry reasons so you can modify these drugs without
messing with the active sites. And so we’re basically offering the pharmaceutical industry
the opportunity to develop them in a rational way so you can keep track of the activities
so you can modify them for PKPD and so on without messing with what’s important in the
drug. We’ve used modENCODE to really explore. I
called it the epigenetics; I’m not sure that’s quite right. But to explore the transcriptional
control of the factors that actually driving this tumor genesis, and that has been fundamentally
useful for our ability to identify the key pathways, and once we have those key pathways
in hand, that puts us in a position to work with medicinal chemists to attack those pathways
in a way that’s useful. And through the chemical genetics we’ve developed a method we call
“rational polypharmacology” that we hope will be useful. And, finally, I finished by talking about
the importance, especially with model organisms, of taking advantage of these model organisms.
They’re really the only ones that can embrace this complexity so readily, and so quickly,
and cheaply to develop complex models, to take advantage of the sequencing data that’s
going on, and not to focus on a single target, but to go ahead and embrace that complexity.
And I showed you reasons for that, and to cut to the chase, a four-hit model is nothing
like a one- or a two-hit model. And they’re nothing like them in the ways that matter,
like drug sensitivity. Okay? And this is my lab and the guys who have done
this work, and also my thanks to modENCODE, as well as NIH and ACS. So, thank you very
much. [applause] Manolis Kellis:
Hey, Ross, breathtaking presentation. I really, really enjoyed it. Ross Cagan:
Thank you. Manolis Kellis:
So, I have several questions here. So, how do you go — I mean, in a way, part of the
challenge is translating all that to humans. So, what are some of the challenges here?
First of all, the networks themselves might not be conserved. Secondly, you actually need
to build a model for the disease in Drosophila. So, my question is, from your own experience,
are these sort of limiting the number of diseases, the number of pathways for which this will
be, you know, this type of success story will be possible? And, now, conversely, if you
want to now take these approaches and apply them directly to mammals, for example, the
two rate-limiting steps seems to be the complexity of the networks that we can build, and that
complexity is now increasing with, you know, ENCODE, mouse ENCODE and so on, where you
can build these more complex networks. Of course, building the mice with two and three
and four, five hits can be much more difficult. But I’m just wondering if you see other limiting
steps that, you know, that go beyond this simple view. Ross Cagan:
Okay, so that’s a lot of questions, so let me see if I can remember and walk through
them. The first question, I believe, was conservation between flies and mammals, and also you had
a bigger — a larger question, which, if I can rephrase it, which is, essentially, when
are flies and model organisms good and when aren’t they? And I think that that’s where
you have to choose carefully, and we try to choose carefully, excuse me, in terms of what
we can model and what we can’t, because flies are not good for many diseases that we consider,
actually, and have turned down. In terms of — so there are some playgrounds
that you can play in with flies that work well. The kinome is one of them. There may
be an exception, but I can’t think of any. Every chemical kinase inhibitor that we have
tried in flies and also looked to validate has played out as hitting its target. And
I’m talking about functionally looking at, you know, phospho antibodies and so on, okay?
So I think that those work well. Also, Erdem Bangi, whose post-doc in my lab now was at
Novartis and they published a paper looking at a number of various pathway inhibitors,
notch pathway inhibitors, hedgehog, and so on, and shown that flies work well for those.
So I would not suggest that these are going to be successful for everything. For example,
those drugs that need to be metabolized — some will work, and actually my company has checked
one or two of those, and the cleavage for the ones we checked is fine. But I would guess
in most cases it’s not going to work because fly P450 is going to be different. So you have to be smart about what you model,
all right? Maybe, in retrospect, it was dumb of us to model thyroid cancer in a fly. It
just — I was too naive at the time, and it worked out well because it was such a simple
signaling pathway. And as you say, at the end of the day, you have to carefully validate
what you’re looking at in flies, so we’ve done — what I’m not showing you is the parallel
work that, for example, Erdem has been doing with flies and human cell lines. All right,
so we knock this out or we add this drug, we see this change, we better see the same
thing in human. And to a remarkable step, he has. I wouldn’t push the point that we
have a perfect gut model. Fly gut has many differences than human, but, again, I think
it’s a useful one. So it’s a little bit rambling, but to answer your question on that. Now, I would love the community to take up
the question of network analysis, which was, I think, was your last question. So what we’re
doing now, and in fact I just finished my part of it at 7:00 this morning, due today,
is we are working with network analysis people in the Eric Shock [spelled phonetically] group,
Jin Ju [spelled phonetically] and Wi Chang [spelled phonetically], and for example, with
colorectal cancer, we’re doing RNA-seq against all 15 combinations, subcombinations. Okay,
that’s our goal. To add — so we have drug sensitivity, although we need to expand that.
We have phenotypes. We want to get the RNA-seq, and we can plug all of that into network analysis
and then compare that to the analysis that has already been done for colorectal cancer
in humans. So this is where we’re going with that part of it. There will be, I predict,
similarities and differences, and what’ll be useful for us is that will point us in
directions that we can be useful. Okay? Male Speaker:
I have a question. Can these fly models be utilized for studying disease reoccurrence
and drug resistance? Ross Cagan:
Disease… Male Speaker:
Reoccurrence. Ross Cagan:
Reoccurrence. Male Speaker:
Yeah, and drug resistance. Ross Cagan:
And resistance. So the question was can flies be used to study disease reoccurrence and
resistance? Yes, and, I mean — yes. [laughs] It can, and in fact, had some conversations
with pharmaceutical companies when they’ve given inhibitors of certain pathways and resistance
emerges quickly and they want to know what those resistance pathways far. I have to say
I’m a little up and down about that because we can find pathways you can hit that could
potentially drive resistance, but that doesn’t mean they actually are driving resistance
in patients, because there are many ways around these initial pathways. So to answer your
question, yes, you can do it in principle but I expect, and what we’ve seen from the
sequencing data of patients, is that it has been either particular to the patient or there
have been mutations that were either so — they were obvious and you didn’t need flies, like
an active mutation would crop up in the receptor, or it was something surprising, some downstream
thing, that, again, flies, I don’t think would be helpful. So I would say where flies can be helpful
is this polypharmacology. Not so much predicting resistance, because there are many roads to
resistance, but developing a drug that’s so complicated for the tumor to get around that
it would have difficulty selecting around it. That, I think, would be more reasonable
as a utility for drugs — for flies. Male Speaker:
So, a lot of small molecules we work with tend to have multiple targets. So I was wondering
if you had tried anything — I know you made — tried to increase selectivity with your
design, but of course you could get unwanted targets and things as well, and seems like
flies would be a great system to sort that out with as well, if you had a suite of these
heterozygous kinase mutants that you could just toss these on. Have you tried anything
like that? Ross Cagan:
You mean with the low-specificity kinase inhibitors to try to find things — Male Speaker:
Yeah, or even the ones you think are high-specificity, toss them on as suite of — you know, you
tested them on the obvious candidates, which is great. Ross Cagan:
Oh, no, I’m sorry, I didn’t show this. I’m sorry, I wasn’t clear on that. When I showed
the in vitro kinase data, what I’m not showing is it’s the full in vitro gen panel. Okay? Male Speaker:
Okay. Ross Cagan:
And so we actually know quite a bit, and I’m not showing you all of its targets. Male Speaker:
So vitro, you test them first and then you go in in vivo. Ross Cagan:
And then we go in in vivo and validate the in vitro ChIP, because the in vitro ChIP from
in vitro gen is human. Male Speaker:
Yeah. Ross Cagan:
And then we go into the flies and we say, okay, it says it’s going to knock out Srcs,
so we look at the downstream target, and yes, it does, that kind of thing. So to answer
your question, you could do this with a low-specificity library, absolutely. And — yeah, let me just
leave it at that. You can do it with that. What I think is powerful is if you can then
work with a chemist — you consider that a starting material. So the drugs I showed you
were not high specificity. They were fairly high specificity, the AD1, and so on and so
forth. They’re fairly high specificity in that they don’t hit 20 targets. They maybe
hit six or seven. But what’s really powerful is if you can then, whatever drug you start
with, I don’t care how many targets it hits because in flies you can check them all, is
to be able to work with a chemist and flush some of those out. The chances you’ll find
a drug that’s going to get everything just the way you want it is possible but then… Male Speaker:
You know, your multigenic model of the disease causational response for the drugs — the
data you presented or the information you presented is presented essentially a validation
information. You have known genes which have — you have put together and [unintelligible].
I would assume it will be at least possible, it may not be easy, to include [unintelligible]
of genes which are distantly suspected to be related to the disease. Ross Cagan:
That’s right. If I can rephrase the question, we were focusing on obvious targets. What
about not-obvious targets? So what I probably didn’t make clear is when I said there were
three pathways that matter and so on, that was a little bit of cherry picking, but actually
not entirely. So we had done an unbiased genetics screen across the genome to ask for all targets
required — not all, but to the level we did it, because when we did it, not — we didn’t
have access to mutations in every gene, but we had access to mutations of thousands — bless
you. So we did an unbiased screen, and of course I’m presenting a simple talk where
I said [unintelligible]. What I’m not going through is that there are a number of other
pathways. For example, when signaling is key for here. Hedgehog signaling, we also hit
that as well. And I haven’t talked about those in part because, you know, we don’t have a
lot to say about those. So you’re absolutely right, and we do have that information in
hand — our functional screen, and you could decide, in principle, to match this initial
drug with a Wnt inhibitor, for example, and that would be reasonable as well. But we do have that information. We didn’t
pick those three just because they were obvious. We picked them because our genetics pushed
us in that direction. And, actually to finish that point, when you look at Sin3a and its
targets, it’s astonishing how it hues to those three pathways’ not 100 percent, but it’s
astonishing the high percentage that really hued to Ras, Src, PI 3-kinase, act and remodeling,
little bit of junk, hits one notched target sort as [unintelligible]. That’s kind of it.
So we are getting focused back over and over again, but I agree with you that — Male Speaker:
Yeah, so in principle your multigenic model could be used to sort of make a decision whether
a given gene, inclusion or exclusion, is it rival gene or just a passenger gene. Ross Cagan:
Absolutely. And — yeah, so that’s a great point. And on that point, we have a work we’ve
been doing with Francis Collins on the diabetes side. I didn’t talk about our diabetes work,
but we just feed flies a high sugar diet and they become diabetic. That’s all you need
to know, okay; I love this model. And we actually went through their GWAS studies, and in each
region, we said, “Okay, there is five genes that we’ve defined if we just make it arbitrary
interval at 100 KB. Which one’s the driver?” All right? So we tested the fly orthologs
for each of those, and in many cases we were actually — it would say, yes, probably that
guy. In some cases we were even able to say, “It’s this guy plus these two flanking guys,”
which you would miss in standard analysis because they’re — the statistics of the neighboring
genes are swamped by the peak of the one next to it and so on. Absolutely, I think flies are terrific for
that, and just to finish this point, in that thyroid cancer, we actually published a study
where we used the fly genetic screen to predict susceptibility loci to secondary tumors in
a related cancer syndrome, called multiple endocrine neoplasia type 2, which the major
thing — problem they have is this thyroid tumor. And we were actually able to identify
two that were at least markers and possibly the drivers of susceptibility to these secondary
tumors. So I absolutely agree with you. Flies and worms have great potential if they are
used that way to really explore various aspects of drivers, and susceptibility loci, and so
on. I think they have a lot of promise for that, for sure. So, Stu [spelled phonetically] standing up
means I need to sit down, yeah? Male Speaker:
Two things before break. We’ll have coffee outside and also upstairs, and I’d like to
ask people to be back at 10:45. I’d also like to thank our morning group of speakers for
a set of excellent talks. So please join me. [applause] [end of transcript]