Advancements in Predictive Toxicology
Will Play an Increasing and Accelerating Role in Drug Discovery and
Development:
An Interview with Dave Craford and
Jacques Retief of Affymetrix
The inability to accurately predict
toxicity early in drug development cost the pharmaceutical industry $8
billion in 2003, approximately one-third the cost of all drug failures.
Even when drugs successfully obtain FDA approval and reach the market,
they remain vulnerable to costly safety issues. A recent example is
Merck's withdrawal of the blockbuster drug Vioxx, which caused the
company's stock to plunge 25% in one day. Indeed, predictive toxicology
and toxicogenomics technologies are of growing interest to government
regulators, who have issued several reports recently calling for more
predictive toxicology and toxicogenomics approaches to be used in
assessing drug safety.
In the new CHA Advances Life Science
Report, Toxicogenomics and Predictive Toxicology: Market and
Business Outlook, Dave Craford, Vice President of Business
Development and Jacques Retief , Genomic Collaboration Scientist at
Affymetrix, discuss the use of gene expression arrays in predicting
toxicity.For more information about this report, please visit http://www.chadvisors.com/services/PredTox/overview.cfm.
DAVE:
With respect to predictive toxicology right now I think the majority of
big pharma and biotech are doing predictive toxicology, or thinking about
doing predictive toxicology, using gene expression technology to build
(and/or license) databases to identify signatures that could help them
farther up in their drug discovery process to better prioritize compounds
according to their potential toxicity. So the majority of the market is, I
think, either doing or planning to do this type of work. Within three
years, I assume everybody will be doing it and have some level of database
or database access and screening program that incorporates some type of
profiling.
I think there's a bunch of technical
issues to bringing the predictive toxicology to the main-stream drug
discovery process. From a commercial perspective, I think there's a strong
belief out there that these technologies will be beneficial to a drug
discovery and development process, but three years out, I think people
will have data that confirms that an investment in building a database and
screening using expression-based technologies will really help them better
prioritize compounds in their pipeline. Obviously, pharma companies have
different gates within their pipeline that help them identify the number
of compounds they move forward and they're trying to get the best
compounds along the fastest. I think in three years they will continue to
be making investments, there will be much more data to support the
hypothesis that predictive testing is useful to do, and they will be
beginning to reap the benefits of the work they have done.
JACQUES:
It really depends on how narrowly you define where you're doing your
toxicology, because a lot of drug companies are already using microarrays
in this context and very typically early on in the drug discovery process,
because microarray gives you a view of what's going on inside the cell and
that is very useful when you are discovering your targets. During this
process, you can get some indication of what your toxicology may be down
the road. So what's been happening is that the toxicology has been moving
more upstream in the drug development process and a lot of the companies
we talk to have started seeing the benefit of using microarrays earlier on
to see what off-target and adverse drug reactions there could be.
DAVE:
Jacques brings up a great point. I think when people think about
predictive toxicology, they're typically thinking about the toxic affects
of the compound onto the system, but I think what Jacques also could be
alluding to is the mechanistic toxicity from a systems perspective. Maybe
it's not the compound, it's just the wrong target, which is mechanistic
toxicity that you could find much earlier in the process by having a
better view of the biology than you could get with a full genome
expression array.
JACQUES:
That's exactly the point, and the more you know about the mechanism of the
toxicity or your off-target effects, the easier it becomes to translate
the information you get from one model system to another. For example,
very much of the early on work is done in in vitro systems. So if you have
an indication what your toxicity is going to be or what the potential
toxicity is going to be, then it helps you to focus your later studies and
the model systems that you would want to use later on to make sure that
you address those potential issues. Another way of putting it is that if
you understand the mechanism, you can look for that mechanism. If I take
it from, for example, a tissue culture, take it into a mouse, take it into
a rat, a dog or whatever other system you want to apply it in, and you can
look for these potential problems. And of course the advantage of doing
this is that if for some reason you find that your leading candidate has
got a potentially serious issue and for some reason it fails, it helps you
make an intelligent decision about other potential drug candidates in that
same family. So what we are talking about here is just a real huge gain in
information that assists us in making intelligent decisions in the whole
process and making these decisions earlier on.
There are a number of issues on the
technology side. I think the most obvious one is the fact that we get this
huge amount of information every time we run an array or an array
experiment. We get a tremendous amount of information and it is
challenging for the researchers to look through this information and turn
it into useful, biological answers. So what we have done in that context
is that we have fairly extensive annotations for our arrays to help people
move from just seeing the data to actually making a biological
interpretation. Because really, if you think about it, it's not about the
array, the array is just the way to connect to the biological processes.
And when I talk about biological processes, I refer to pathways and gene
anthologies and so on because you wouldn't want to make that kind of a
connection and say, okay, well, this is the process that is being affected
by my drug. So the way we've been addressing it is NetAffx website and
it's worked for us incredibly well because our platform and our system is
very open.
NetAffx is a database we have that
contains all the annotations to our array and its actually available to
the public. Anybody can just basically log in and look at all the
information that we have on the arrays and it's an enormous amount of
information. For example, for the rat array, which is the most poorly
annotated array that we have, I once did a calculation that if you had to
print out all the information on the rat array that we provide, it would
cover 45,000 pages worth of single-spaced 12-point text.
DAVE:
It's all about getting to the biology, and on our chips we have probes
that represent genes in different genomes. We have human, rat, dog, and
all of these are used in different types of tox studies, but a gene really
doesn't mean anything without a significant amount of annotation as to
what that gene is, is it a receptor, some type of transcription factor,
how does it interact with others in a biological system. What
toxicologists are looking at is the biological effect, and without NetAffx
and the data that Jacques is describing, one couldn't make the connection
between some data that comes off an array and a biological system.
JACQUES:
Without the annotations, for example, if you could see that the immune
response is being affected in, for example, a rat, then it makes it very
easy if you go to a mouse or into a dog that you can actually look now to
see if the immune response in that particular animal is getting affected
as a fairly simple example.
But because our data is freely available,
it has made it possible for academics and industrial researchers to
actually look at this and come up with other ways of displaying and
analyzing the data. For example, we do have fairly large data sets
available on the Web and we've found that academic researchers have used
these data sets to develop new algorithms for analyzing the data and that
really has moved the field forward substantially.
DAVE:
From a regulatory perspective, I think one of the obstacles that is being
resolved or at least addressed is the FDA's level of awareness and
education on how microarrays can be used to collect information on how
drug candidates are interacting in different model systems in man. It's a
complex technology and they've invested a lot of energy, I would say, over
the last couple years, both in collaborations, of which we're involved
with a few, and they've also put out a couple different guidance documents
that relate to the use of microarrays in preclinical and clinical
situations, which, again, are important to drive the field forward so that
people who are doing work with the technology understand how the FDA would
treat the data if it were submitted for a given purpose. PhRMA is very
active here as well and has, I think, a good collaborative relationship
with the FDA in the genomics area.
JACQUES:
Yes, that's really a very good point. And fairly recently there was a
predictive analysis workshop between PhRMA and the FDA. In this respect,
we find that transparency in our data really is helping because, again,
you can look at the algorithms and look at the data without being
concerned where it comes from.
DAVE:
There are two guidances that are in draft form that will impact will
impact data submitted on microarrays around toxicology. These guidances
would affect mechanistic or investigative tox applications. The CDER
issued a guidance last year that describes voluntary versus involuntary
biomarkers and what needs to be submitted versus not submitted, and that
will have ramifications. Then CDRH put out a draft guidance that was more
technically oriented on how you should do your study, and what is the FDA
going to look for to insure the data are valid. They put that out last
April and we're hoping that it will come to some level of conclusion, but
I think it will probably lag the CDER guidance in terms of availability.
For predictive toxicology, people are
using expression arrays. We have full genome expression arrays for human
and rat, which are probably the predominant organisms that people are
looking at, human in vitro, and then rat, both in vitro and in vivo
experiments. We also have dog. We also have people who run chimeric
primate samples onto the human arrays with a good degree of success, and
there have been a number of posters on that as well. I think we have
arrays available for the major organisms that people use for both model
systems and man.
Another development that we're working on
in the labs, that I think will be really important for predictive
toxicology, is that were adding a new format for our technology and
basically packaging the arrays into micro titre plates, which help people
industrialize the process, run greater numbers of samples for lower cost,
and actually get more reproducible results. So there will be a number of
organizations that use the technology in a micro titer plate package
format for these types of studies, because you can envision late-stage
screening, secondary/tertiary screening applications where they're looking
from a expression profiling point of view at the signatures of interest on
the compounds in cell-based screen.
JACQUES:
Yes, that's a real good point and I should also add that in any tox
screening generally you are screening a large number of compounds so,
typically, you would like to run a large number of arrays. So this kind of
high throughput system would help customers run enough arrays.
If we look at predictive toxicology and
tox responses in a slightly more general way, the new products that I
actually find very exciting are our genotyping products because one of the
issues that really has been a real challenge for the toxicology market are
things like idiosyncratic responses. With our new 100K genotyping arrays,
it starts becoming more viable to start looking at issues or situations
where individuals may have different drug responses.
DAVE:
We're not necessarily talking about predictive toxicology anymore, we're
talking about in a clinical trial setting, if you had an idiosyncratic
event using a full genome association study to identify potentially genes
that are responsible for that adverse event. And, as Jacques indicates,
one of the biggest costs in pharma is the idiosyncratic tox.
JACQUES:
Yes, this not your early classic predictive tox, but it becomes an issue
much later on when you start putting drugs in the clinic. One of the
examples that I'm thinking of is the release from the Mayo Clinic that is
looking at differential responses in hypertensive drugs in different
patient populations. So it's a real good example, actually, of how you can
use this technology right from its very, very early stages in the drug
development process all the way through right up to your geoclinical
point.
