Editor’s note: These remarks have been very lightly edited. Dr. Frederico Goodsaid is associate director for operations in genomics, Office of Clinical Pharmacology, Office of Translational Science, Center for Drug Evaluation and Research, reviewed FDA expectations.
March 2008 | “In 2005, we came out with the pharmacogenomic guidance and in that guidance, a pathway for data sharing with the agency was defined, which at that time was called voluntary genomic data submission, a VGDS. And that pathway for submission of data to the agency allowed the sharing of information before it was to be included within INDs or NDAs. That pathway allowed the number of ways in which we could, in fact, interact with different companies when they wanted to share this type of data. And the pathway defined the voluntary data submission meeting and we’ve changed that G to an X to more accurately include the fact that we are dealing with a whole bunch of different kinds of exploratory biomarkers, not just genomic biomarkers.
In a voluntary exploratory data submission meeting, the data is received from a company and it’s tracked and archived by the FDA and the Interdisciplinary Pharmacogenomic Review Group goes over that data and has a face-to-face meeting with the sponsor that has submitted that information. That face-to-face meeting includes a review of the data that was received. And in that face-to-face meeting we have a chance to first of all learn from the sponsors about the different kinds of platforms that are being used to generate the data and platforms that are being used to analyze the data and to interpret it. And it also allows us to show the sponsors the capability that we have in doing this kind of analysis.
Here we’re talking about voluntary data submission of exploratory biomarkers. When we deal with a regulatory submission with pharmacogenomic data being used to support very specific label plates, when you get to that level, you’re no longer talking about voluntary data submission, you’re talking about really knowing how to analyze genomic data and to interpret genomic data. In a regulatory world, you may have different reasons, different ways in which you would look at genomic data, or biomarker data, in general. You could look at the biomarker data, try to explain the pathway to get some information about the pathway associated with—and either a safety or an efficacy claim. You are worried about the specific list that you’re working with that defines that biomarker, that list of genes, that list of proteins. You’re also worried about the biology, in general, that you’re learning with those biomarkers.
In the specific case of the voluntary submissions, however, it is mostly about the biology and you’ll see over the next few minutes why that makes pathway tools so important in this type of work. Voluntary data submission meetings are useful to a sponsor because it gives them experience with biomarker data submissions, because they allow the sponsor to discuss the exploratory biomarker data with the FDA and to introduce those exploratory biomarkers to the review division that will be receiving the IND or an NDA later on. It also provides formal guidance on how these biomarkers are to be applied in drug development. It’s a shared experience in the use of genomics and other platforms. It’s also a way, as I said earlier, to have the sponsor be aware of how ready we are at the FDA to evaluate this type of data.
There have been several examples of the voluntary exploratory data submission meetings. We’ve had cases where we’ve been dealing with candidate gene approaches compared to whole genome scans to identify efficacy biomarkers. We’ve dealt with gene expression profiles in peripheral blood, gene expression patterns, genomic biomarkers to predict responders and non-responders, the use of registry to identify biomarkers, toxicogenomic approaches, and [other data]. The impact at the external level has been well beyond our ability to interact with the sponsor because this type of information tends to generate other questions and tends to generate other needs and other possibilities.
We’ve come up with a number of CRADA through the industry. We have one CRADA on the qualification on the [inaudible] biomarkers. We have another one with a collaborative research and development agreement on trying to look at a specific case study for drug test co-development. And finally we have a collaborative research and development agreement that has been signed off on the improvements and the application of biological pathway tools. We’ve also come up with new guidance. One for guiding principles to make these meetings as joint meetings [with] your peers or your peer counterparts. And another guidance which goes over the technical issues associated with the generation and interpretation of genomic data and submission to the agency.
We’ve worked with a number of consortia. The Predictive Safety Testing Consortium, which has submitted a number of biomarker sets for qualification to the agency, and the Microarray Quality Control Consortium, which is an ongoing massive effort to try to understand the sources of variability in microarrays, data generation, and analogy. And the draft companion guidance for the pharmacogenomic guidance really stems from the fact that we identify through those voluntary exploratory data submissions, a number of issues that we needed to address, if you were going to make sure that pharmacogenomic data could be submitted and interpreted in a way in which we could get consistent results throughout the agency.
We started off with a white paper that was discussed at a workshop that was held here in DC on best practices and development of standards for the submission of genomic data to the FDA. And we followed that up with the issuance of the draft guidance in August of last year and, of course, we’ve had quite intensive feedback on that draft guidance and we should be wrapping up the final text for that guidance, hopefully, over the next six months or so.
The Table of Contents of this guidance really reflects very nicely what the experience for the voluntary exploratory data submission has taught us, the different steps, some of them rather straightforward, some of them still a bit contentious as to what you should do when you’re generating microarray data, for example. In this case, the Section 2 of Gene Expression Data for Microarray, when we were writing it and it ranged from writing something about which there wasn’t a whole lot of controversy in labeling reactions and for microarrays, those are fairly straightforward issues to talk about.
When it came to RNA isolation, well, you know there’s also quite a good consensus on how that’s supposed to be done. When came to the analysis that yielded your differentially expressed gene lists, that’s still a pretty open and very contentious area of discussion. And finally, when it came to the interpretation, the biological interpretation of that data, of list of differentially expressing, we are dealing with a pretty open question. Not necessarily introducing [the] violent responses you might get when you’re discussing the analysis to generate the differentially expressed gene list, but nevertheless, one which can be a bit unclear when people set out to interpret those differentially expressed genes lists or a list of SNPs, for example.
So as far as the biological interpretation of lists of differentially expressed genes, we are worried about what are the biological and disease functions or pathways that are associated with significantly or represented genes on the list. And if you think about what happens when you do the analysis, there is room for different lists that will come out as you try out different kinds of analysis procedures on the microarray data. And those different lists of genes can then also be interpreted in different ways as you go into your biological interpretations.
There are questions about how many pathways are affected? What type of pathways are affected? What is the preferred mechanism of action and toxicity of the gene expression changes? What is the tissue specificity of the pathways and the gene function in relation to biological processes? How does the magnitude and/or pattern of those perturbations compare to the treatments of reference compounds? And what about the known pharmacological or toxicological properties of those compounds? So to be able to use these tools to try to gain biological pathway information about those lists of either differentially expressed genes or SNPs is something that is not trivial. It does not have a single answer—it does not have a single tool that will give you those answers. At this point, what we have is the fact that a combination of tools, [the] very hard work of looking over different sources of information can be used to address a particular question of interest that you want to define.
A variety of analytical platforms are available and they’re either free on the web or they may be purchased through a commercially available product. As far as the FDA is concerned, we’d like to use as many of them as possible. We have a tool for analyzing the array data called Array Track and we have all those platforms—I think they were ornaments on a Christmas tree because we really want to make sure we cover as many of them as possible. Understanding that ultimately we need to come up with a consensus interpretation that is based on where all these databases lead you.
Sometimes that consensus can be hindered by many different factors. In some instances you may have an absence of information of the compound of interest in the reference database. If it’s not there, you’re not going to see it. And that could be a big problem. Sometimes there’s lack of annotation for a particular pathway. Those pathways are not there. It can be either an instance of genes may replace specific pathways in one system, but they may not be represented in the same pathways in another pathway [system]. It could be that the genes may not have been evaluated in a particular platform. Either way, the information differs and you have to make sure ultimately that a reviewer here at the agency, whether you’re dealing with voluntary data submissions or regulatory INDs or NDAs, that a reviewer is able to have a good biological sense in the interpretation of all these sources of information.
It’s also important to distinguish whether the information has been limited to findings supported by experimental evidence, or whether you’re talking about some sort of review paper that started to have some speculation about this finding. The bottom-line is, regardless of where you get the information to come up with the biological interpretation, you have to make sure that you understand what you’re working with and that you’re able to link all these sources of information together to make sense of them. We recommend that people just sit down and take a look at PubMed after they’ve consulted all these different databases and they can come up with a hypothesis that can be accurately supported by the literature in this area.
The information submitted should include, but not necessarily be limited to, the type of database used for annotation, including the version and vendor name; the methods and approaches, statistical tests used to identify over-represented pathways in each database; the references used to justify any user defined annotation; and summary by the sponsor of the interpretation of the pathway annotation results. It should be possible for us here at the agency, just like we can try to reconstruct the lists of differentially expressed gene, for example, as you start with the raw data from the microarrays, we ought to be able to reconstruct the biological interpretation by looking at the databases that provided the information for the sponsor.
Here’s an illustration which I think will help [you] understand how important this can be. We have a case study proposed where you have compounds A and B that are structurally related to each other. Compound A binds to a target receptor; compound B does not. A sponsor submits a differential gene expression data for studies with both compounds in a non-clinical animal model. So now, we go ahead and we look at the genes in common for compounds A and B, at a P of less than 0.01 and a full change of more than 2.0. Under those conditions, we have six genes that are shared by those compounds.
Then we go the next step and we take a look at the biological interpretation of each of those lists, using Ingenuity’s [tool], and what we find is that there are different groupings for the genes that are contained in those lists. For the compound A genes, we have a category for lipid metabolism, molecular transports, small molecule biochemistry, carbohydrate metabolism, and cancer. For compound B, some are shared, such as the lipid metabolism genes. Others seem to be somewhat different. Self-cycle genes come up here, cell [inaudible] genes come up here.
Either way, the point that, in this case, becomes important is what happens in the middle. These are two compounds that are very similar, but one of them binds the target, one of them does not. And what is common about them is what might tell us something in terms of off target effects that we could be worried about at the agency. And in this case, there’s only six genes that are shared by these two compounds and they have to do with some perturbation gene. If you go ahead and expand the [criteria] for the list of genes and now you look for a P value of less than 0.05 and a fold-change of more than 1.8, then you can come up with a larger number of genes that are shared. Those 32 genes in the middle can then be used to go through the same kind of biological pathway analysis. And if you do that, you find out that you can generate a type of map that you could get from Ingenuity, which in this case clearly lays out the PPAR alpha, a PPAR gamma gene, together with other genes that are associated with this. So in common, these two compounds have the ability to induce this particular gene.
You can then characterize those commonly held pathways by looking at the canonical pathways in this tool and again, at the top of the list are the fatty acid synthesis pathways. You can also look at different functions and this again brings you back to the lipid metabolism that we saw earlier. So we have a number of ways to characterize those biological pathways and this is just one tool, one example, but one that really allows a reviewer here, whether it is a medical officer or one that specializes in genomics or biomarkers as we do in our genomics group, to actually make some biological sense of what is being observed. Keep in mind, again, this is only one of several sources of information that will be used to get to a conclusion.
But we think that a pathway tool like this can be developed further, and that’s why we drew up a collaborative research and development agreement with Ingenuity. The role of this agreement is multiple. On the one hand to leverage the Ingenuity pathway analysis system for the VGDS and we’ve already done that. We have access to the tool and we’ve integrated within ArrayTrack, as I said earlier, as one of several Christmas tree icons that can be invoked from within the tool to be able to analyze the list of differentially expressed genes that you get out of ArrayTrack.
The agreement that we have with Ingenuity is intended to jointly develop a pathway-based solution for the regulatory review of biomarkers, pharmacogenomic, and toxicogenomic data, and to facilitate for a reviewer how this analysis would be done. Now this is only the last step in the process of analysis, so we have to do a lot more work if we want to really make the work that a reviewer would do with this type of data easier. But this is an important step because it is at the tail end of the reviewing duties that need to be done. The idea is to expand the content of the tool to make sure that it covers the drug chemical and biomarker universe that we’re interested in.
We are also interested in three areas that will go beyond where the obvious is today. Expanding the content is an important area that I just mentioned, to come up with biomarkers, toxicology, ontology, and content that make sense, that really cover the kind of biomarkers and the kind of drugs that we are actually working with. As I said earlier, if a biomarker or a drug or a compound are not in that particular database, we can do anything we want. We just don’t see them. And that’s not going to help much with the review function.
Now assuming that we have the right database to work from, the next step would be to come up with different ways of looking at those databases—species centric, tissue centric, drug centric, tox centric, to be able to have the reviewer give a report from each of these different points of view and that way to be able to cover the kinds of questions that will come up as that review result is actually looked at by another group of reviewers. In the future, hopefully, we come up with some sort of fingerprinting tool. We don’t know how far we’ll be able to get to that goal.
But to summarize then, pathway tools are being used by the FDA in interpretation of development data and other kinds of biomarker data. They’re used with the biological interpretation as well as with the review of this data. And the integration of these tools, through what we hope will be a genomic [inaudible] as I said earlier, we need to work on the earlier stages to make them as easy as possible for reviewers that may not necessarily be experts in the analysis of genomic and other biomarker data. That will, of course, increase their value as review tools.”
This article appeared in Bio-IT World Magazine.
Subscriptions are free for qualifying individuals. Apply Today.