ATDEPARTMENT OF HEALTH AND HUMAN SERVICES

FOOD AND DRUG ADMINISTRATION

CENTER FOR DRUG EVALUATION AND RESEARCH

ADVISORY COMMITTEE FOR PHARMACEUTICAL SCIENCE

CLINICAL PHARMACOLOGY SUBCOMMITTEE

Tuesday, November 18, 2003

8:30 a.m.

Advisors and Consultants Staff Conference Room

5630 Fishers Lane

Rockville, Maryland

PARTICIPANTS

Jurgen Venitz, M.D., Ph.D., Chair

Hilda F. Scharen, M.S., Executive Secretary

MEMBERS

David D'Argenio, Ph.D.

Marie Davidian, Ph.D.

Hartmut Derendorf, Ph.D.

David Flockhart, M.D., Ph.D.

William J. Jusko, Ph.D.

Gregory L. Kearns, Pharm. D., Ph.D.

Howard L. McCleod, Pharm.D.

Mary V. Relling, Pharm.D.

Wolfgang Sadee, Ph.D.

Lewis B. Sheiner, M.D.

Marc Swadener, Ed.D.

GUESTS AND GUEST SPEAKERS (NON-VOTING)

Acting Industry Representative:

Efraim Shek, Ph.D.

Guest Speakers:

Richard Hockett, M.D.

Pertti Neuvonen, M.D.

FDA

Shiew-Mei Huang, Ph.D.

Peter Lee, Ph.D.

Lawrence Lesko, Ph.D.

C O N T E N T S

Call to Order:

Jurgen Venitz, M.D., Ph.D. 4

Introduction of the Committee 4

Conflict of Interest Statement:

Hilda F. Scharen, M.S. 5

Open Public Hearing 8

Introduction:

Lawrence Lesko, Ph.D. 8

Drug Interactions

Introduction:

Shiew-Mei Huang, Ph.D. 8

Evaluation of CYP2B6-Based Interactions:

David Flockhart, M.D., Ph.D. 17

Evaluation of CYP2C8-Based Interactions:

Pertti Neuvonen, M.D. 40

Committee Discussion 63

Pharmacogenetics: Integration into

New Drug Development

Introduction:

Lawrence Lesko, Ph.D. 73

Academic Perspectives:

David Flockhart, M.D., Ph.D. 83

Industry Perspectives:

Richard Hockett, M.D. 100

Practitioner Perspectives:

Mary V. Relling, Pharm. D. 129

Committee Discussion 152

Concluding Remarks 170

P R O C E E D I N G S

Call to Order

DR. VENITZ: Good morning and welcome everyone to the second day of the Clinical Pharmacology Subcommittee Meeting. This is the continuation of yesterday's topic area. My name is Jurgen Venitz and I am the Chair. I would like to start by introducing all the members of the committee and invited guests around the table.

Introduction of the Committee

DR. D'ARGENIO: David D'Argenio from the University of Southern California.

DR. FLOCKHART: Dave Flockhart from Indiana University.

DR. SHEINER: Lewis Sheiner, University of California, San Francisco.

DR. SWADENER: Mark Swadener, Boulder, Colorado.

DR. JUSKO: William Jusko, University of Buffalo.

MS. SCHAREN: Hilda Scharen, FDA, Center for Drugs, Executive Secretary.

DR. KEARNS: Greg Kearns, University of Missouri.

DR. DERENDORF: Hartmut Derendorf, University of Florida.

DR. DAVIDIAN: Marie Davidian, North Carolina State University.

DR. SHEK: Efraim Shek, Abbott Laboratories.

DR. McCLEOD: Howard McCleod, Washington University.

DR. RELLING: Mary Relling, St. Jude Children's Research Hospital, Memphis.

DR. SADEE: Wolfgang Sadee, Ohio State University.

DR. LEE: Peter Lee, COPB, FDA.

DR. HUANG: Shiew-Mei Huang, Center for Drugs, Office of Clinical Pharmacology and Biopharmaceutics.

DR. LESKO: Larry Lesko from FDA, Office of Clinical Pharmacology and Biopharmaceutics.

DR. NEUVONEN: Pertti Neuvonen from the University of Helsinki, Finland.

DR. HOCKETT: Rick Hockett, Eli Lilly.

DR. VENITZ: Thank you, everyone. Let me turn over the microphone to Ms. Hilda Scharen. She is going to read the conflict-of-interest statement.

Conflict of Interest Statement

MS. SCHAREN: The following announcement addresses the issue of conflict of interest with respect to this meeting and is made a part of the record to preclude even the appearance of such at this meeting.

The topics of today's meeting are issues of particular matters of broad applicability. Unlike issues before a committee in which a particular product is discussed, issues of particular matters of broad applicability involve many industrial sponsors and academic institutions.

All special government employees have been screened for their financial interests as they may apply to the general topics at hand. Because they have reported interests in pharmaceutical companies, the Food and Drug Administration has granted general-matters waivers of broad applicability to the following SGEs which permits them to participate in today's discussion; Dr. David D'Argenio, Dr. Marie Davidian, Dr. Hartmut Derendorf, Dr. David Flockhart, Dr. William Jusko, Dr. Gregory Kearns, Dr. Howard McCleod, Dr. Mary Relling, Dr. Wolfgang Sadee, Dr. Jurgen Venitz.

A copy of the waiver statements may be obtained by submitting a written request to the agency's Freedom of Information Office, Room 12A30, of the Parklawn Building. Because general topics could involve so many firms and institutions, it is not prudent to recite all potential conflicts of interest but, because of the general nature of today's discussion, these potential conflicts are mitigated.

We would also like to note for the record that Dr. Efraim Shek is participating in today's meeting as an acting, non-voting, industry representative.

In the event that the discussions involve any other products or firms not already on the agenda for which FDA participants have a financial interest, the participants' involvement and their exclusion will be noted for the record.

With respect to all other participants, we ask, in the interest of fairness, that they address any current or previous financial involvement with any firm whose product they may wish to comment upon.

Thank you.

DR. VENITZ: Thank you, Hilda.

Two housekeeping issues before we get started. You may have noticed in the original agenda for the second, we had a topic on Pediatric Population PK Template. Due to time constraints, that topic had to be deferred to our next meeting or one of our next meetings.

Open Public Hearing

I have also been informed that we won't have any presenters at the open public hearing today so we might be able to get an early adjournment.

Having said that, I would like to ask Dr. Lesko to introduce the topics for today and give us our charge.

Introduction

DR. LESKO: Thank you, Jurgen. I am not going to do much with the first topic, cytochrome. I will Dr. Shiew-Mei Huang do that and then, after that, I will introduce the pharmacogenetic topic. So let me turn it over to Shiew-Mei.

DRUG INTERACTIONS

Introduction

DR. HUANG: Good morning.

[Slide.]

The first topic this morning, we will talk about CYP2B6 and CYP2C8 drug interactions.

[Slide.]

Recall, at the last April meeting of this committee I have discussed that the CDER Drug Interaction Working Group is revising the guidance, the In Vivo Drug Interactions Guidance, which was published in 1999. Because of the emerging technologies and tools, available, we have additional information which prompted us to update this guidance which is about three-years old.

We are going to use information that is obtained from various workshops cosponsored by the agency or the information that was published in the PhRMA Position Paper or from internal research from the reviewers about industry practices and literature data.

As I discussed last time, we would like to propose to include the information on classification of CYP3A inhibitors in this revised draft guidance which will be published for public comment again so that when we have drugs that are substrates of 3A, we will be able to prioritize our study and we will be able to label drugs that are strong or moderate inhibitors in the labeling to facilitate the priorities of the interaction or clinical significance of interactions in the drug label.

We also discussed that we are seeing increasing submissions that interactions are based on P-glycoprotein. Based on our discussion in April, the majority of the committee members agree that the digoxin is a good substrate for P-glycoprotein although it is also a substrate for other transporters such as organic anion-transporting peptide. Still, right now, it is probably the best substrate to study because the clinical significance of the interaction outcome.

In addition, in this '99 guidance, we will also include in vitro evaluation technologies discussing various substrates, inhibitors, inducers for key cytochrome P450 enzymes. I will discuss that a little bit more. In keeping with the impending publication of the Final Rule of Physician Labeling, we will also discuss case examples indicating certain drug interactions that may be put into the Highlights Section of the new Physician Labeling in addition to an additional section of drug interaction in the labeling.

[Slide.]

I just want to briefly discuss the current practices on cytochrome-P450-based interactions. In the in vitro evaluation, our reviewers have been recommending and industry has been consistently performing the evaluation of these key enzymes; cytochrome P450-1A2, 2C9, 2C19, 2D6 and 3A, both for reaction phenotyping, determining the metabolic pathway of the new molecular entities. In addition, these other enzymes, 2A6, 2B6, 2C8, 2E1 often are also evaluated.

For enzyme-modulating effects for inhibition; again those five key enzymes have been most consistently evaluated--if not, our reviewers would provide feedback--also, for induction, since 2D6 has not been shown to be induced. These are the four enzymes, plus some of the 2B6, 2C8 increasingly have been studied in this in vitro evaluation.

As far as in vivo or clinical human interaction studies, again, our reviewers have communicated and the sponsor has been conducting the studies to evaluate other drug effects on the new molecular entity and the drug's effect on others. They are often prioritized based on the in vitro evaluation of cytochrome P450.

For example, if the reaction phenotyping is indicating 3A as a major enzyme, there is usually a study involving a strong inhibitor of 3A. If this compound is shown to be inhibiting certain enzymes, then the effect on others with appropriate probe drugs are often conducting. Increasingly, we have seen both in vitro and in vivo evaluation of P-glycoprotein-based interactions using various substrates in vitro, with digoxin, or in vivo with digoxin, fexofenadine, as a substrate.

Depending on the drugs or previously known similar compounds, other pathways such as phase-II metabolizing enzymes or sudden peptide transport or if it is renally secreted, certain compounds that are inhibiting renal active secretion have also been evaluated in various submissions.

[Slide.]

So why do we want to discuss CYP2C8 today? The various cases of rhabdomyolosis involving gemfibrozil in statins; there are data to show that monotherapy of gemfibrozil and statins, on their own, they have shown some dose or concentration-related increase in the incidence of myopathy or rhabdomyolosis. So this could be a pharmacodynamic interaction. However, we are seeing reports in the pharmacokinetics of statins that have been changed because of coadministration of gemfibrozil--I show cases there--since gemfibrozil does not appear to interact with these statins, via CYP3A, even some of the statins with 3A substrates.

There is a possibility of other enzymes or transporters that are being affected by gemfibrozil such as CYP2C8, 2C9, UGT, glucuronosyltransferases or organic anion-transporting peptides.

[Slide.]

For example, just look at the sample of literature data. Many of these were published by Dr. Neuvonen and, later on, he will elaborate on each study results more in detail. You can see here the examples from statins such as fluvastatin, a 2C9 substrate here. It didn't show an interaction with gemfibrozil.

Rosuvastatin, as shown yesterday by one of our presenters, there is a two-fold increase. Simvastatin acid, lovastatin acid and cerivastatin, there are various degrees of increase in area under the curve when gemfibrozil was given together. These were in healthy volunteers. Another, rosiglitazone, a 2C8 substrate, repaglinide, also as 2C8 substrate also so a different degree of interaction. Here, with repaglinide, it is up to more than an eight-fold increase when gemfibrozil is given.

As a comparison, trimethoprim, which, in the literature is also shown to be affecting 2C8, has a relatively smaller effect on rosiglitazone.

[Slide.]

In our submissions, we have seen recently compounds such as Drug A which has been shown to be metabolized by CYP2C8. The major cytochrome P450s, 3A, 2C9, 2C19, 2D6, may not appear to affect this metabolism. So what do we do if we would like to know its interaction potential with this drug. Especially as we discussed yesterday, certain safety biomarkers such as QT prolongation have been increasing evaluated when drugs are submitted for approval.

If we need to evaluate QT prolongation, we either use supertherapeutic dose or we try to stress the system using enzyme-inhibitors to increase the exposure and try to anticipate the worst-case scenario. In that case, what can we do to increase the exposure to see what is the maximum exposure that will happen, assuming this is the case, what inhibitors are available for us to evaluate.

Or in another case, Drug B, which has been shown to inhibit CYP2C8 in vitro, what are the ideal or probe substrates of 2C8 that we can evaluate this drug's effect on other drugs? So this is about 2C8.

[Slide.]

So why are we interested in CYP2B6 which we are discussing today? There are recent studies on efavirens and bupropion which have shown that 2B6 is the key or the principal enzyme responsible for efavirens metabolism and one of the key pathways for bupropion. There are recent data on inducers of 2B6 such as some HIV protease inhibitors, dietary supplements such as St. John's wort.

Our submission with Drug C is sometimes metabolized by 2B6 in vitro. So, again, we would like to see the clinical significance of other drugs' effects on it, what kind of inhibitors are available there for us to evaluate their clinical significance.

[Slide.]

Today, we have invited two experts in the field; Dr. David Flockhart to talk about CYP2B6 and drug interactions. Dr. Flockhart and his colleagues at Indiana University have recently published research data on efavirenz metabolism and will give us a review in this field.

We also have Dr. Pertti Neuvonen from University of Helsinki. Dr. Neuvonen and his colleagues have published numerous articles characterizing strong inhibitors such as ketoconazole, itraconazole, on various probe substrates of 3A to estimate their extent of interaction. He has published a lot of grapefruit-juice-related interaction and, more recently, he has published various gemfibrozil and statin interaction data, and also in vitro evaluation of various substrates and inhibitors and inducers.

[Slide.]

The issues for them to discuss and for the committee to consider are what is the clinical significance of 2B6- and 2C8-based interactions and are there tools available, are there pro-inhibitors for the clinical evaluation of 2B6- or 2C8-based interaction, or do we have substrates that their interactions are mostly based on 2C8.

Some of the examples that may be shown later may have a lot of possible transporters involved and we would like to know whether there are good inhibitors and substrates that will be able to provide us useful information particular to these two enzymes. Also, maybe there are other areas that we need to focus on based on this particular evaluation.

[Slide.]

These enzymes are felt to be important from our working-group discussion and this is just to show you the big group of our Interaction Working Group members from our Office of Clinical Pharmacology and Biopharmaceutics, members from the Office of Pharmaceutical Science, members from--used to be from CBER, Center for Biologics, and also from Office of the Commissioner who wants to see what our current evaluation is and the labeling impact, whether these are consistent with the new proposed rule and how would this facilitate the healthcare providers and patients to use the labeling depending on how we will address the interaction issues in the label.

With that, I would like to introduce Dr. David Flockhart to discuss 2B6-related interactions.

Evaluation of CYP2B6-Based Interactions

DR. FLOCKHART: Thank you, Shiew-Mei. It is a great pleasure to be here this morning, particularly, I must say, on the same podium as Dr. Neuvonen whose work I have followed for a long time. We have actually published together and collaborated but we have never met until yesterday evening.

[Slide.]

I am going to talk about cytochrome P450 2B6. Those of you who know me will know that I really don't know much about this. But it is a subject of a great deal of interest in our Division of Clinical Pharmacology at Indiana and the work in 2B6 is led by Zeruesenay Desta. Dr. Desta has currently a series of projects aimed at defining probes and inhibitors of this important enzyme. So I am going to talk a little bit about our thinking about new ways of evaluating it.

[Slide.]

I am going to talk about some data on expression because that is historically important in terms of understanding why we have not spent a lot of time on this cytochrome up until now and then talk about some potential substrates, both in vitro substrates and substrates that might be used in the clinic, and then talk about inhibitors in the same kind of context, ones that might be used in test tubes and ones that might be used in people.

2B6 is a heavily inducible enzyme. I think that is one thing that is really obvious from its study at this point. So interest in inducers of it is as important, and there may be a large number, actually, of important, clinically important, interactions with this enzyme that result in low concentrations, particularly of HIV medications, that we, as yet, are--well, we are not unaware of but we, as yet, don't understand in terms of the mechanism.

[Slide.]

I think the main reason, as is the case with many isoforms, and this would have been the case in the past even for the two Cs, all of them, 2C9, 2C19 and 2C8, is that the early antibodies that we always talk about, and I am referring in particular to the classic paper published by Shimada and Guengerich ten or fifteen years ago now which first documented by Western Blot the amount of different P450s in the liver.

On that, the amount of 2B6 expression was very low. It actually isn't shown in their diagrams but was estimated in the test to be a minor component and less than 1 percent of the total P450. Therefore, and as recently as two or three years ago, in conversations with Grant Wilkinson at Vanderbilt, he was absolutely convinced that it played a tiny role in human drug metabolism.

[Slide.]

This was why Grant thought that. This is taken from a review article that Dr. Desta and I have put together, but if you look just at the detection percent on the left here, from a series of studies published in the late 1990s--well, I guess throughout the 1990s--there is a relatively small n in the studies. So these are all individual livers but, in a significant number, you can't even pick up the enzyme at all.

If you just recall, you will see the rough numbers here. So, 1 to 2 picamoles per milligram of protein, which is not a lot of P450, in the liver was detected except in one study, this one from Japan, in which 19 picamoles were picked up.

This has a lot to do with the specificity and sensitivity of the antibodies we were using at the time. But the number of studies indicates the potential interest in this isoform.

More recently, and you will notice that the dates of these references on the right are later--this is up until the present--in every liver tested, or most livers tested in these studies, you can actually pick up the enzyme. Not all, though. And this may relate to genetic polymorphisms that have been described but not terribly well characterized to this point.

But you will notice, on the last slide, I talked about 1 to 2 picamoles being present. The average, in these studies--I haven't gone through the somewhat disingenuous exercise of trying to average all these things, but you see that it is significantly higher, probably a lot higher, with the newer antibodies and there are some that are significantly higher.

Also, I think now people in the field would agree there is a consensus that we have a specific antibody. When you study using these antibodies, the variability--there is a huge variability in protein expression but also in RNA expression. The RNA expression data is currently confusing because Aaron Schutz and other people at Mary Relling's institution have shown quite nicely there are multiple splice variants of this enzyme that might contribute to variations in its activity. So we have yet to sort out really confidence assays for the RNA. But, suffice it to say, the amount is considerably more than we thought it was originally.

[Slide.]

So new mono and polyclonal antibodies of higher sensitivity and specificity have made it clear that there is a greater frequency of detection. I think, in all the livers we have ever tested now, the enzyme is there, and there is more of it and it looks rather less than 0.1 percent. It averages about 6 percent of the total liver with absolute maximum amounts that are really quite significant, presumably in livers that are turned on or people that are turned on for one reason or another, up to 25 to 44 percent.

[Slide.]

This is taken from the paper that Shiew-Mei referred to which is our in vitro study of efavirenz metabolism. A couple of points about this. Pharmacologists always have to put up diagrams, structures, but there is an important unusual group on this, this triethylene planar group, monoplanar, group, out here which is common to a number of substrates of cytochrome P450 2B6. I am going to show you data that basically show that this is the dominant route, this is the main route, by which efavirenz is metabolized in people.

A minor route here we have recently shown is mainly cytochrome P450 3A, but this, in people, is about one-hundredth this route. So this route is the dominant means of clearing this drug from the body. It is an 8-hydroxylation in the 8 position down here whereas the 3A-mediated metabolism is A7. It catalyzes the 7-hydroxylation right here. You count from this side, so this is 7. This is 8 down here.

There is also metabolism of this, of the metabolite, of the metabolite, although less quickly to the 8,14-dihydroxy.

[Slide.]

Let me just show you some of the data that supports this. These are data simply showing the clearance of efavirenz, itself, from an in vitro incubation. So this is the disappearance of the parent. They might have parent left, if you like, and you see the only one isoform under these conditions which is 1 micromolar efavirenz, approximately the concentration reached at steady state during normal dosing of about 600 milligrams a day. Only one isoform reduced it.

So these are the data that initially got us interested in it. This is fairly comprehensive. It does include 2C8, both 3A isoforms. There is a difference between 3A4 and 3A5 out here which has held up in subsequent studies. 3A5 seems to be a more efficient catalyst of efavirenz metabolism than 3A4.

[Slide.]

Our very first clinical data--this is the first time I have shown this--phase I and phase II here; phase I is in the absence of rifampin and phase II is after 10 days of rifampin treatment. You do see a decrease in bioavailability and an increase in the rate of metabolism of efavirenz in vivo.

This is something that you see with cytochrome P450 2B6 but you also see it, obviously, with cytochrome P450 3A, 2C9, 2C19, a number of other isoforms.

We are currently conducting a study of about 100 people in which we are trying to determine at what point in this curve it would be intelligent to conduct a phenotyping study; in other words, one that might allow us to do a single point determination to study a large number of people in order to get some sense of the clinical variability of this enzyme in vivo. That might be, and we don't know the answer to this yet, a urinary ratio of efficacy to 8-hydroxyefavirenz or it might be a serum ratio. But we don't have those data yet and I can't talk about it.

[Slide.]

This is a similar drug. This was published in January of this year. It is a drug with a number, DPC963 but you will notice a similar structure up here. I am simply putting up this complicated slide to make the point that 2B6 catalyzes metabolism of an efavirenz analogue as well.

You will note that this drug also is metabolized notably by 3A as well. But, again, the dominant route to metabolism is by 2B6. All this is saying, really, is that there are a number of related drugs that are metabolized by the same pathway.

[Slide.]

2B6 is also a low-affinity catalyst of S-mephenytoin metabolism to Nirvanol. This is from a paper published in 1996 that we became interested in. I say low affinity because mephenytoin, which we can't use without and IND anymore--it is off the market in the United States, unfortunately. It is a valuable probe drug, obviously, for cytochrome P450 2C19, but, in this study, the metabolism of mephenytoin not to is 4-hydroxy metabolite, which is 2C19-mediated reaction, but to Nirvanol which is the demethylation reaction of mephenytoin which was studied. These authors showed that only one isoform did this.

This was the only data in this paper, but we became interested in this idea because we had been interested in 2C9 team and actually had done a study which we published in 1992 showing that S-mephenytoin and omeprazole could be used as probes for that.

So, what we wanted to do at the time, was to take this large 200-person study, take their urine and see if we could actually do 2B6 phenotyping from the same urine. Unfortunately, that turned out not to be the case.

[Slide.]

This is largely because of work done by a very smart Korean post-doc, Jim Ko, who showed at the time, before we actually got into wasting these valuable urine samples, that this same reaction, the N-demethylation of mephenytoin at this concentration can be carried out by two isoforms, 2B6, but also 2C9.

He went on to show in subsequent studies that the high-affinity catalyst was 2C9 and not 2B6. So it remains unclear at the moment whether or not one can use mephenytoin as a probe for 2B6. Personally, I think it is rather compromised.

[Slide.]

I am going to skip this. This basically just shows the metabolism of mephenytoin.

[Slide.]

These are our data suggesting the R-mephenytoin might be a substrate probe for 2B6. Certainly, in vitro, possibly, in the future as one isoform that does it again. These data are rather thin in the sense that they only are recombinant enzyme data. We haven't done careful studies because, at the time--this is 1996--we didn't have any confident in vitro inhibitors of 2B6 that were specific and only recently have we been able to have those.

[Slide.]

Now, there are a number of inhibitors that are now published clear and obvious. Not all of them, unfortunately, are specific. It is very clear that both paroxetine and sertraline can inhibit this isoform using in particular bupropion as a probe, the hydroxylation of bupropion as a probe.

Antiretrovirals including nelfinavir and ritonavir are potent inhibitors. Both ticlopidine and clopidogrel have been shown by our group to inhibit 2C8. Clopidogrel is metabolized primarily by it. We and others have shown also that thioTEPA, the chemotherapy agent, is an inhibitor of cytochrome P450 2B6. We are pretty confident that that happens in vivo and we have some data to indicate that it is fairly specific in vitro.

[Slide.]

These are some of those data suggesting that this is a specific P450 inhibitor in vitro. This is just a percent of control activity with a series of cytochrome P450 isoforms with a series of different probes. This is 100 percent so everything should be here. But when you coincubate thioTEPA, I believe at 1 micromolar in this experiment, you see a decrease principally in 2B6 although there is a little inhibition of 1A2 as well.

When you look at this carefully, and you do a dose-response to thioTEPA, and these are data that we published, I think, three years ago now, you see that 2B6 is preferentially inhibited compared to the others. There are decreases in all of these but the potent inhibition with an IC50 or 5 micromolar which is well below, actually, the concentration that this drug reached in vivo, it is here.

So, because of these data, we believe that thioTEPA can be used as an in vitro inhibitor if the conditions are done right and this low concentration can be used as a specific in vitro inhibitor of this enzyme. That is an important tool to allow us to study it further.

It is the case, obviously, that the thioTEPA is a chemotherapeutic agent and you can't just give thioTEPA to normal volunteers. So it is not something that we are going to be able to use in vivo.

[Slide.]

This just shows the potency. We used S-mephenytoin metabolism at high concentrations, relatively high concentrations, to be inhibited by thioTEPA. These are just Dixon plots indicating that you see linear kinetics and potent inhibition.

[Slide.]

Cyclophosphamide has also been described to be metabolized by this enzyme and it was first described really carefully by a series of nice studies done by Irv Wainer and his group at Georgetown University and in Montreal when Irv was there. This is the structure of cyclophosphamide. Its metabolism to its principal active metabolite which is 4-hydroxycyclophosphamide is carried out primarily by 2B6--that is why it is in the red--but also by these isoforms. A number of groups including David Waxman's group and a number of others have contributed to these studies as has John Slattery's group at Seattle.

[Slide.]

So the difficult position we are in is we have these in vitro data. It is not really--it wasn't really clear, how much of this actually occurs via 2B6 in vivo. But we have noted a study from Holland by Huitema in 2000. What this is is a study of a sequential treatment in cancer patients with cyclophosphamide and thioTEPA.

In this situation here, what you are looking at is concentration of drug against time in two different sequences. So first, in this situation, cyclophosphamide is given prior to thioTEPA and you see the normal kinetics that you would expect of cyclophosphamide. I just want to point out that the concentration of the parent drug is notably higher than that of the metabolite which is in the squares below.

On the other hand, if you coadminister thioTEPA, you give it I.V. at the same time, you see a notable decrease in the red in the parent concentration and a notable increase in the metabolite concentration. So--I'm sorry; I got that the completely wrong way around. This is the parent here, which goes up, and the metabolite goes down. So this is an inhibition of cyclophosphamide metabolism, not an induction.

So we think, because of these data, that thioTEPA is acting to inhibit 2B6 in vivo and resulting in a change in cyclophosphamide pharmacokinetics.

[Slide.]

As I indicated a moment ago, there are a lot of inducers of this enzyme. In fact, many of the substrates of this enzyme, we have not yet found one that doesn't seem to auto-induce its own metabolism. Rifampin, hyperforin, phenobarb, ritonavir, phenytoin, carbamazepine, all induce. We are familiar with these as ligands for PXR and sometimes CAR.

The HMG-CoA reductase inhibitors, interestingly, have been shown in some situations to induce 2B6 metabolism as have nevirapine induces its own metabolism as does efavirenz. Clotrimazole has been shown in vitro to as well and there is recently a clinical study indicating that artemisinin induces the metabolism of bupropion.

[Slide.]

So, overall, 2B6 is a significant contributor to hepatic CYP expression. The number of substrates is growing and I anticipate that the number of submissions to the agency will grow although Shiew-Mei tells me that there are a lot more substrates coming over the FDA's desk that are 2C8 than there are 2B6.

Efavirenz and bupropion, we believe, are specific in vitro probes. I haven't spent much time talking about bupropion because we haven't studied it much ourselves. But there is a reason we haven't studied it and that is because we felt early on that it was pretty clear that its dominant route of metabolism is via 3A and not through 2B6. Although it does have a fairly specific 2B6 route of metabolism, the hydroxylation, most bupropion is via another route. So this compromises its utility as a probe.

ThioTEPA is a specific inhibitor of 2B6, we believe. There is clearly no evaluable specific inhibitor yet of 2B6 in vivo that we can use and we really need one, you know, to be able to prove for sure that a lot of reactions are occurring via this enzyme in vivo.

Lastly, we do think efavirenz is a potentially valuable in vivo probe for the activity of more currently evaluating that.

Thanks for your attention. I would be glad to take a couple of questions.

DR. VENITZ: Thank you, David.

Any specific questions for David? Shiew-Mei?

DR. HUANG: You listed ritonavir as a 2B6 inhibitor and later on as an inducer. This is similar to the situation with ritonavir with 3A.

DR. FLOCKHART: Yes.

DR. HUANG: Ritonavir is shown to self-induce. It is an inducer for 3A. It is an inhibitor of 3A. Although the 10-day or 14-day study, most of the studies with ritonavir have shown an inhibition effect.

DR. FLOCKHART: Yes.

DR. HUANG: That is the basis for ritonavir and nelfinavir.

DR. FLOCKHART: The nelfinavir with ritonavir is inhibition.

DR. HUANG: Right. I wonder if, for 2B6, do we know about what is the net effect?

DR. FLOCKHART: No. It is clearly in vitro. Several groups now--I think three groups have shown that it is an inducer in vitro. In our hands, it is a good inhibitor in vitro but we don't know in vivo. I guess, with ritonavir, you have to be a little careful about that to net. After about ten days or two weeks, it is clearly an inhibitor but there are periods in between when it would be a net inducer.

DR. SADEE: David, I assume that this is all hepatic activity.

DR. FLOCKHART: The data that I am presenting is hepatic activity. It is clearly present. I cut out a slide showing some of our work and the work from Mikael Akoban's group and Aaron Schutz' work indicating that it present in a lot of tissues. It is not just an hepatic enzyme.

DR. SADEE: Because that could also make a big difference in terms of inducibility. If there is also a lot of extrahepatic activity, the inducibility will--

DR. FLOCKHART: Absolutely. Notably, in my business, it is present in the breast. It is present in muscle. It is present in CNS. It is a very widely distributed isoform which may have all kinds of interesting implications. It is also an effective catalyst of a lot of endogenous things like testosterone, estradiol and so on.

DR. SADEE: It could be present in tumor as well.

DR. FLOCKHART: It is present in some tumors. That has been shown, active in tumors, RNA protein and activity.

DR. SADEE: Another question. The variability; is this caused by maybe pathophysiology? Is there anything known about particular states of liver disease?

DR. FLOCKHART: I would really be going out on the edge to suggest that, Wolfgang, at the moment. But it clearly a very inducible enzyme. You can turn it on very easily. It seems to be less inhibitable, at least in our hands. It is something that is very sensitive to the PXR and CAR-inducing mechanisms and maybe others.

DR. SADEE: One more comment, and it is semantics, basically. I always feel very uncomfortable about specific inhibitors.

DR. FLOCKHART: Yes; you never know until you have studied for infinity.

DR. SADEE: That is why it is so--just for official use, I would strongly recommend using "selective."

DR. FLOCKHART: Selective; okay. I have tried to use the term "relatively specific."

DR. LESKO: David, what do we know about the distribution of 2B6 activity in the population and what is the range of expression or activity, say, from low to high? Is it like 3A4, for example, or is it like something else?

DR. FLOCKHART: The problem is we don't really have a good probe at the moment in vivo, so I think we are conducting a study right now that should give us a handle on that using efavirenz. There are studies that have been done by Ed Lecluse and others indicating there is a fair amount of variability, ten- to twenty-fold variability, in bupropion hydroxylation in people.

But my problem with that is that some of that could be influenced by 3A activity in alternative routes. So, in this particular setting, 2B6 variability, I think we really don't have the data yet, Larry. I would be reluctant to extrapolate the in vitro variability in livers although, of course, that is about the variability you see in 3A.

DR. HUANG: We can elaborate on the discussion later on but I just want to follow on Wolfgang's discussion on nonselective inhibitors. Recently, we have been discussing drug interactions whereby you want to use two drugs to inhibit two major, equally major, pathways in order to create a worst-case scenario.

DR. FLOCKHART: A really bad thing, to create a really bad thing.

DR. HUANG: A really bad case. So it may not be a bad idea to use a nonspecific inhibitors where you could inhibit one major pathway and the other one that you are also concerned with.

Here, you have listed several that are not specific. For example, ritonavir has various pathways.

DR. FLOCKHART: Right.

DR. HUANG: But if you know this drug is not metabolized by all the other pathways, and we know that when it is a strong inhibitor--

DR. FLOCKHART: That is a very, very good point. So Bob Temple has, many times, and I am sure both you and Larry have made the point that, if you want to study the worst possible interaction with 3A, you have got to kill the thing with ketoconazole. So one could make the case, if something is metabolized, both by 3A and 3B6, that you could coadminister a drug that inhibits both, like Ticlid, like ticlopidine, which is a fairly effective inhibitor of both drugs.

Ketoconazole, actually, interestingly, at high concentrations, you have to use a fair amount of ketokonazole but it seems to kill 2B6 as well if you go high enough.

DR. HUANG: Just to clarify; for ticlodipine, is it the parent drug that is active for both, or is the metabolite.

DR. FLOCKHART: The parent drug is the inhibitor.

DR. HUANG: For both; okay.

DR. FLOCKHART: Yes.

DR. HUANG: Thanks.

DR. SADEE: I have one more question, David. If the variability of 2B6 is as high as it appears to be and the variability in 3A4, for instance, also, so you would have a substrate for both. Then, in one person, there would be a 3A4 substrate. In another person, it may be the 2B6 substrate and the other enzyme may play no role. So I am just wondering about labeling this or presenting the information that this is a substrate for both enzymes and, in reality, in individuals, there may be other--

DR. FLOCKHART: I guess that might be the case. I don't have data yet, Wolfgang. I think that generically I would agree with you. I think there may be people for whom there is very little 3A activity and 2B6 would be the dominant route. My bias, at the moment, and it is a bias based on not much data, but I will share the data, is that 2B6 is really dominantly the enzyme for efavirenz.

Even when you turn on with rifampin, you don't see a lot of 3A contribution. The basis for that is the big difference in affinity between the two isoforms for efavirenz and the fact in the urine of the patient that I showed you that was induced, we see the 2B6 hydroxylation, the 8-hydroxylation route, really turned on. There is a lot of that metabolite in the urine and very little of the 7-hydroxymetabolite although that is increased as well.

So I think, in that situation, when it is really turned on, there is more 3A. But it is still a dominantly a 2B6 drug.

DR. VENITZ: Any further questions? Thank you, David.

Our next speaker is Dr. Neuvonen. He is going to share with us his experiences with 2C8.

Evaluation of CYP2C8-Based Interaction

DR. NEUVONEN: Thank you, Mr. Chairman, dear colleagues and committee members.

[Slide.]

In my talk about CYP2C8 and drug interactions, I will review substrates, inhibitors and inducers of 2C8, some in vivo interaction studies and finally present some suggestions for in vitro and in vivo studies.

[Slide.]

CYP2C8 is highly expressed in the liver. The protein content of 2C8 is on the same level as that of 2C9 and clearly than that 2C19. There is lots of interindividual variation in the protein content of 2C8 and 2C8 seems not to be detectable in the intestine.

[Slide.]

Many drugs are substrates for 2C8. In vitro studies, paclitaxel, amodiaquine and torsemide have been used. 6-alpha-hydroxy paclitaxel is a 2C8-mediated reaction and amodiaquine is metabolized mainly by 2C8. Torsemide is metabolized both by 2C9 and 2C8 but, in some conditions, this can be used as a marker substrate.

In vivo studies, cerivastatin, repaglinide and rosiglitazone have been used as substrates. Also many other compounds are substrates for 2C8. For example, many of the substrates of CYP3A4 are also substrates for 2C8. But the relative contribution of different CYP enzymes may depend on the substrate concentration used, for example, in in vitro studies.

[Slide.]

This slide shows the relationship between amodiaquine metabolism and paclitaxel, 6-alpha-hydroxylase activity. As can be seen, amodiaquine clearance and formation of N-desethyl-amodiaquine correlate very well with the activity of paclitaxel 6-alpha-hydroxylase. This was a study where microsomes from ten human livers were used.

[Slide.]

Trimethoprim is a competitive of 2C8. It has a Ki value of about 32 micromolar and it is relatively selective up to 100 micromolar concentration.

[Slide.]

As can be seen here, the inhibition of other CYP enzymes is very little, up to the concentration of 100 micromolar.

[Slide.]

However, when higher concentrations are used, and here are shown 250 and 500 micromolar concentrations, trimethoprim inhibits, for example, 2D6, 3A4, 2C19, 2C9, 1A2 enzymes.

[Slide.]

Quercetin is a competitive and potent inhibitor of 2C8. It has a Ki value of about 2 micromolar but quercetin is also a potent inhibitor of 1A2. So it is a nonselective inhibitor of 2C8.

Glitazones are potent inhibitors of 2C8. Gemfibrozil is nonselective but it seems to work both in vitro and in vivo. There are also many other nonselective inhibitors; for example, many substrates of 3A4 seem to be inhibitors of 2C8.

[Slide.]

Here are shown Ki values of some glitazones. Rosiglitazone and pioglitazone are relatively selective for 2C8 whereas trogliazone inhibits more 2C9.

[Slide.]

Some of the so-called selective probe inhibitors used as a diagnostic inhibitors in in vitro studies are also inhibitors of 2C8 at the concentrations generally used. For example, ketoconazole at the concentration of 1 micromolar considerably inhibits activity of 2C8. Ketoconazole is a noncompetitive inhibitor with an apparent Ki value of 2.5 micromolar.

So data regarding ketoconazole--let's say that the inhibition data where it has been used as an inhibitor of CYP2A isoforms may include also inhibition of 2C9. Also DDC is a significant inhibitor of 2C8.

[Slide.]

2C8 is clearly inducible. In vitro rifampin is a more potent inducer of 2C8 than of 2C19 or 2C9 and rifampin is more potent as an inducer of 2C8 than, for example, phenobarbital or dexamethasone. In in vivo studies, rifampin clearly decreases, for example, the AUC of repaglinide which is a substrate of 2C8. It decreases the AUC roughly by 60 percent. Of course, here, maybe the induction of 3A4 contributes to the finding, but probably it is best to measure extent to induction of 2C8.

[Slide.]

In the following, I will present some in vivo interaction studies where gemfibrozil and some statins or oral antidiabetics have been administered. All these studies are randomized crossover studies in healthy volunteers where gemfibrozil or placebo or a comparator have been given for three to four days. Then, on Day 3, a single dose of either cerivastatin, simvastatin, lovastatin, repaglinide or rosiglitazone has been administered.

[Slide.]

Here are data on the effect of the gemfibrozil on cerivastatin. As can be seen here, gemfibrozil greatly increases the AUC of unchanged cerivastatin. The AUC was increased on average five or sixfold and in 110 healthy subjects the increase was tenfold. Also, the concentrations of cerivastatin, lactone or M1 metabolite, which is formed by CYP3A4, are greatly increased by gemfibrozil whereas the concentration of M23 metabolite is drastically decreased. This M23 metabolite is formed by CYP2C8.

I think that this pharmacokinetic interaction greatly contributes to this toxicity of the gemfibrozil/cerivastatin combination which has been previously found.

[Slide.]

Gemfibrozil inhibits cerivastatin metabolism also in vitro, the formation of 23 metabolite is clearly reduced by gemfibrozil.

[Slide.]

Here are shown the effect of gemfibrozil of the pharmacokinetics of simvastatin or simvastatin acid. Gemfibrozil increased AUC of simvastatin acid about two, threefold, whereas the AUC of the parent simvastatin was unchanged.

[Slide.]

Here is shown the role of CYP enzymes in simvastatin metabolism and it can be seen that simvastatin acid is metabolized by CYP3A4 but also partially by 2C8. It seems that gemfibrozil could inhibit this 2C8-mediated partway. Of course, there are also some alternative explanations for the finding.

[Slide.]

Gemfibrozil also increases the AUC of lovastatin acid whereas the AUC of parent lovastatin remains unchanged. Bezafibrate had no effect.

[Slide.]

Here are shown the effect of gemfibrozil, itraconazole and their combination on the concentrations of repaglinide and its M1 metabolite. Repaglinide is a short-acting oral hypoglycemic agent. Plasma concentrations of unchanged repaglinide were increased greatly by gemfibrozil whereas itraconazole had only a minor effect on plasma concentrations of repaglinide. The combination of gemfibrozil and itraconazole drastically increased plasma concentrations of repaglinide.

Gemfibrozil increased greatly the concentration in one metabolite which is formed mainly by CYP3A4 and, as expected, itraconazole greatly reduced it.

[Slide.]

Here are shown the effect of two CYP3A4 inhibitors and gemfibrozil on the AUC of repaglinide. The data regarding clarithromycin are derived from another study. Clarithromycin and itraconazole both increased the AUC roughly 40 percent. Gemfibrozil increased it on average eight-fold and the combination of the gemfibrozil and itraconazole about twenty-fold. Of course, the flat glucose-lowering effect was clearly increased along with these increased concentrations.

[Slide.]

Here are shown the effect of gemfibrozil of rosiglitazone. The AUC of rosiglitazone was about two, three-fold--increased two, three-fold, by gemfibrozil and both the Cmax and half-life were increased.

[Slide.]

In the final two figures, I will suggest some possibilities for in vitro and in vivo interaction studies, in vitro human liver microsomes or recombinant human 2C8 enzymes can be used. Paclitaxel and amodiaquine seem to be well suitable substrates. Torsemide is useful only with recombinant 2C8 because also 2C9 is metabolizing torsemide and forming just the same metabolites.

Trimethoprim, quercetin and pioglitazone or rosiglitazone can be used as inhibitors and rifampin is useful as an inducer.

[Slide.]

For in vivo studies, repaglinide can be used as a probe compound, probe substrate. Also, rosiglitazone is useful. Cerivastatin would be also useful but, of course, it may be difficult to get for in vivo studies. Amodiaquine is probably too toxic to be used in interaction studies.

Gemfibrozil can be used as an inhibitor. Of course, one should remember that it is nonselective. It inhibits, for example, 2C9 and also some transporters, at least OATP2. Trimethoprim is more selective but it is not very potent. By now, there have been only very few in vivo data about trimethoprim as an inhibitor of 2C8.

Pioglitazone and rosiglitazone could also be possible inhibitors. Rifampin is a useful inducer but, in conclusion, further studies are needed to find optimal probe substrates and probe inhibitors, particularly for in vivo interaction studies with 2C8.

Thank you.

DR. VENITZ: Thank you, Dr. Neuvonen. Any questions? Mary?

DR. RELLING: So some of those interactions that you described at the end with gemfibrozil were remarkably potent interactions. So do you suspect that there are other mechanisms involved besides just CYP2C8, with gemfibrozil, for example? That was a 1900 percent effect on AUC.

DR. NEUVONEN: I agree that there may be also other possibilities. For example, the role of OATP2 inhibition should be clarified in these interactions. But surprisingly all substrates of 2C8 we have studied by now together with gemfibrozil, there has been a significant interaction with gemfibrozil and those substrates.

DR. RELLING: That is in proportion to their relative KM's or Ki's roughly.

DR. NEUVONEN: Not very well. So it is not sure if it is a parent gemfibrozil or some of its metabolites, for example. Of course, we are not aware of the liver concentration of gemfibrozil.

DR. HUANG: I was just going to add to it. I guess many of these drugs that you study as a substrate, with gemfibrozil as a 2C8 substrate, the concentration of 2C8 may vary among these drugs and so KM may not be the only determining factor.

But I want to comment on do we know anything about gemfibrozil dose and the effect on some of the statins? Do we know any dose effect.

DR. NEUVONEN: We have used the standard dose, 1200 milligrams per day and we have not studied possible dose-effect correlation. So I have no answer to that at this time.

DR. HUANG: I was wondering, is there an interaction between itraconazole and gemfibrozil?

DR. NEUVONEN: Between itraconazole and--

DR. HUANG: Itraconazole and gemfibrozil.

DR. NEUVONEN: I am not aware of it. We have not studied it.

DR. HUANG: Okay, because the nineteen-fold increase was only when itraconazole was added.

DR. NEUVONEN: Actually, I would like to correct my previous answer. Of course, we measured the concentration of itraconazole too in these studies. If I remember correctly, it was, perhaps, that plasma concentrations of gemfibrozil were decreased.

DR. HUANG: Decreased.

DR. NEUVONEN: Yes. It has been reported in the publication. If I remember correctly; yes. We thought that it could be a displacement from protein binding or something like that, but we have no final--

DR. HUANG: I was just wondering, the higher effects of gemfibrozil on repaglinide when itraconazole was additionally added to the regimen, was it due to its effect of gemfibrozil or just added other mechanisms of interaction because itraconazole, itself, doesn't really affect repaglinide.

DR. NEUVONEN: In the case of repaglinide, I guess, or at least one explanation could be, that repaglinide is metabolized by 2C8 and 3A4. If both of these metabolic enzymes will be blocked, then this could explain more than the additive interaction observed with these compounds.

DR. VENITZ: Dr. Derendorf?

DR. DERENDORF: You mentioned quercitin as a potent inhibitor. I would assume that data comes from in vitro studies. When you give quercitin, you hardly find any in the blood. It gets converted to the conjugate. So is there any information on the quercitin conjugates or any in vivo interaction data.

DR. NEUVONEN: Actually, we have no experience of our own with quercitin. These data are based only on the literature. So I have not seen any in vivo studies with it.

DR. VENITZ: Larry?

DR. LESKO: What is the nature of the relationship between the in vitro data and the in vivo data on the substrates and inhibitors? In other words, is there a qualitative rank order that, if I have a sensitive substrate in vitro, I would see the same sensitivity in vivo in the comparative sense, or, conversely, if I had a weak inhibitor in vitro, would it serve as a weak inhibitor in an in vivo situation for the same substrate.

DR. NEUVONEN: You mean, basically, now with 2C8 enzyme?

DR. LESKO: With 2C8.

DR. NEUVONEN: Actually, our data with trimethoprim--well, based on our in vitro data, we calculated, if I remember correctly, that in vivo it should inhibit roughly 20/70 percent 2C8 activity. I think that the in vivo data, actually, we have in press in line with these findings. So trimethoprim increases the AUC of

of repaglinide but not as much as gemfibrozil.

DR. VENITZ: David?

DR. FLOCKHART: Two things. I think I would like to congratulate you for just doing the experiment with both itraconazole and gemfibrozil. As Mary points out, it is a big effect. But I think this is relevant to the kind of evolution of the guidances about drug interactions. We have all been talking about multiple drug interactions. I think many of us have been saying for many years that, while the real world is people are taking many, many different drugs, we have been studying one-on-one drug interactions.

So I would just like to emphasize the point that we need to move into a mode, and I know Larry is aware of this problem, of studying more multiple-drug interactions. There has been data, really, for twenty-five years indicating that, in the elderly, they really get into bad adverse drug reactions once they are over five or six medications, at least in the V.A. system. I think that is important.

I would also--since you are here and in this country, I would like to thank you again for coming and for the large contribution that your group has made to our understanding of these things over many years. Many, many times we, in the U.S., have talked around doing studies, thinking of doing things. Your group has actually been the one that has actually done it.

DR. NEUVONEN: Thank you. Actually, one point I would like to add is that we should not look too much at the mean increases but just to look at the interindividual variation in the extent of interaction because I guess that just those adverse effects are coming from those patients who are most sensitive and, therefore, the variability in the extent of interaction should be overreported.

For example, in the case of cerivastatin, there were, even in the material of ten homogeneous students, an increase of 10 in 1, so what is the variation in a typical population.

DR. KEARNS: I think you just answered the question I was going to ask. It is remarkable. Not only have you come a long way, but you have managed to at least read my mind a bit. But, my point is from a regulatory perspective. To me, and maybe this is just a very simple way of thinking about it, but it is the constitutive expression of the enzyme in a patient that will determine the extent of the interaction.

So, from a regulatory standpoint, when you are contemplating putting in labeling about an interaction and you may be basing that on mean data, how do you reconcile that with respect to a prudent warning. If it is a drug that has a huge therapeutic index, it makes no difference. But if it is a drug that is used to treat cancer or other narrow-therapeutic-index drugs, it is a big issue.

So, to my friends at FDA, how are you going to deal with that?

DR. LEE: May I answer that? In the last two advisory meetings, we actually proposed a method to look at the probability of an adverse event due to the drug-drug interaction. So we would look at the PK safety relationship and calculate, based on the distribution of PK change--and calculate what will be the probably of an adverse event.

So we would not only look at the mean value but also look at the patients who are on the extreme.

DR. SHEINER: That's the right thing to do except that now your data requirements go way up because you are now talking about estimating sort of tails of the distribution, not that they are not the most important. They are because we are concerned about 1 in 100, 1 in 1000, events. There are a series of things. But the amount of data you need to actually get a confident estimate of something like the tail area is really, really nasty. It is not just like twice as much.

Have I got that right?

DR. HUANG: Just to add to that, I think at the last advisory committee meeting we presented a case where we are estimating the percent population that may have QT prolongation more than 30 milliseconds due to drug interactions or due to renal disease because that particular example, the drug is both metabolized and renally excreted. So, actually, the assimilation also shown was a percentage of population which would result in QT prolongation more than 30 milliseconds where you have both renal failure, a certain creatinine-clearance range, and having ketoconazole. So we are approach that quantitative approach. We haven't done the multi drugs yet but we are doing two different conditions to estimate that. We have not applied widely but we are starting to.

DR. KEARNS: I understand that. That is laudable. But, again, and I hate to go back to the QT discussion because it is always painful for me, but ketoconazole is an IKR-channel inhibitor. Until you can factor in the intrinsic ability of that interacting substrate to have its own pharmacologic effect that may produce an adverse effect, then the kinetic piece is just part of it.

As Dr. Sheiner just mentioned, then the n goes up way big to factor out maybe the pharmacodynamic piece of it.

DR. HUANG: Yes; our reviewers take note of that and actually this was in the consideration when we look at the data on some of the inhibitors that we would recommend in order to increase the exposure of drugs that we are evaluating for QT prolongation.

DR. KEARNS: Still on the point that Greg raised, I don't know if you had something in mind as an alternative but, yes, in fact, averages are used, or mean values are used, along with some other considerations, I suppose, in making recommendations in the label. Is what you are asking related to the way this information is expressed in the label?

For example, is it leading to expression of ranges of let's say area-under-the-curve increases? How else can you do it, I guess, is what I am sort of trying to get to. What are some alternatives to the way it is done currently?

DR. KEARNS: I wish I knew. But what is troubling me sitting here as a pediatric pharmacology person is that, if we look at developmental expression, activity of the enzymes changes over time. There are not a lot of drug-interaction studies in children to see, at three months of age, if you look at the P450-based interaction and the extent of it, how far do you move it kinetically, compared to when the enzyme is fully expressed.

Again, it boils down to the therapeutic range so there is this clinical need for people to generalize and to put interactions on tables and charts or to memorize the important ones. I think, at some level, that is good. It is like a warning. But, at another level, if the pharmacist refuses to fill the prescription because there is a drug interaction in the label, then patients can be deprived of therapy where the interaction for a given person may not exist in a meaningful way.

So I don't know the answer, but it is a problem.

DR. HUANG: That is why at least one of the approaches that we are taking is to warn about the most significant interaction. That is why we are trying to put in the labeling that you are dealing with a drug with a strong inhibitor or, if this drug is given with a strong inhibitor what you should do.

Hopefully, this will be caught up in the computer system where you can search for only strong inhibitors and that is where you put maybe three flags instead of one to make a difference between all these interactions that will come up as a warning when patients are--I think that is the first step. At least that is what we are trying to do to minimize the trivial interactions and flag the important ones.

DR. SHEINER: Not to dwell too much on the technical side, but means are bad descriptors for distributions that are highly skewed. This is probably what you have got here. It depends on which way you look at it. If you look at AUC increase and it goes up twenty-fold, that is a huge skew to that side.

If you just flip it upside down and say you are looking at the amount of active enzyme or something like that, then that is going towards zero and that actually compresses the thing.

So there usually is some reasonable transformation, whether reciprocal or square root or whatever, that will allow you to get a more symmetrical distribution and then allow you to maybe make a little bit more confident statement about what fraction of people are beyond a certain limit. That is sort of a very simple type of a thing.

The other point is progress can be made here because these are really population issues. In other words, we study the population and if we can know what the distribution of various isoforms of the enzymes and so on are, and we can know what the distribution--perhaps this is a little bit tougher--of the sensitivity of individuals if it varies.

If it doesn't vary, if it is just a matter of this particular enzyme has this particular inhibition potential from that drug, then maybe we can get a lot from sort of these pooling data across multiple sources rather than having every manufacturer have to go out and get his panel of people and go and do the same thing over and over again.

DR. HUANG: Just to add another point. We are starting to--at least for extreme cases, we have started to put it in labeling; for example, Strattera, which was talked about yesterday, or last week, actually. In the labeling, we actually talk about CYP2D6 inhibitors effects on an extensive metabolizer versus a poor metabolizer.

So I guess, in the past, we just mentioned, it is a D26 substrate and with a 2D6 inhibitor, you may need to be aware of the adverse events and--we didn't say dose adjustment. But, in poor metabolizers, we do not expect to have an interaction. So I think this needs to be taken into consideration. We have started to put this information on the labeling so, at least in the extreme cases, where we know that a poor metabolizer, you don't expect an interaction. That we are putting in.

The subjects with intermediate metabolized activity, then you may see variable interaction, extent of interaction. I think we are starting to see this and I think this may be discussed more in the later session. But at least we try to address one aspect.

DR. LEE: Just to follow up Dr. Sheiner's suggestion. Are you suggesting that if we see a sort of increase of AUC or PK due to an inhibitor we can verify the distribution of the increase to a population PK type of analysis using the pooled data?

DR. SHEINER: No; I wasn't saying that. I am not exactly sure how you could verify anything. I was just saying that when you think about how you describe--let's say even in the label, how you describe what you are likely to run into. I am saying if you have a very skewed distribution, the mean is not a good descriptor of what is going on.

It is sort of like we saw yesterday with the QT interval. We can't get at the individual parts of the heart and their conduction and their repolarization but the mean there is really insensitive to the fact that you have got heterogeneity which is what the issue there is.

It is the same thing here. You have got heterogeneity in the population as to how much enzyme they have got. So X amount of drug will be a lot of problem for somebody but not much for someone else. You want to find some way of A, estimating what is important, and B, expressing it in such a way that people can understand it. All I am saying is taking the average may not be what you want to do.

DR. VENITZ: I think you have got a lot of general comments back on drug-drug interaction. Let me get back to what you guys what us to talk about which is 2B6 and 2C8.

Committee Discussion

DR. VENITZ: The question put in front of the committee is what our recommendations would be, as to committee support, given the state of the art in our knowledge on 2B6 and 2C8. I think you are primarily interested in in vitro substrates, in vitro inhibitors and in vivo substrates and in vivo inhibitors.

What is the committee's feedback or response to that question?

David, do you want to summarize?

DR. FLOCKHART: Just for 2B6, I think we have a decent couple of substrates in vitro. We have got efavirenz and we have bupropion. I think in some settings, mephenytoin is a reasonable substrate probe as well. As inhibitors, the only specific in vitro one, selective in vitro one--excuse me--is thioTEPA that I am aware of. I don't think we have specific inducers and I don't think we have validated in vivo probes.

DR. VENITZ: That was my conclusion, too, listening to David. Any additional comments on 2B6? I am looking at Dr. Neuvonen. Maybe you want to summarize what your recommendations would be with respect to 2C8 in vitro inhibitors, in vitro substrates, in vivo inhibitors and in vivo substrates.

DR. NEUVONEN: In vivo assay substrate, I would recommend repaglinide because it seems to the most sensitive of those compounds which are easily available. Of course, rosiglitazone can also be used, but it may be not so sensitive a marker.

As inhibitors, I would like to use gemfibrozil even with great reservations regarding its mechanism of action because it seems to be so potent. But trimethoprim is more selective and actually I have no data regarding pioglitazone and rosiglitazone. They may be in the future more useful but actually further data are needed.

DR. VENITZ: Any additional comments by anyone on the committee?

DR. SHEINER: How do you usually sort of probe for these things? We have seen some exquisite experiments in which you have done area under the curve and things like that which you can't argue with that. But, in a typical situation where you are trying to--I am thinking again about gathering information on populations; what do you do to decide whether somebody has or has not got a given enzyme or some drug does or doesn't inhibit another one in a sort of a survey sense.

You can't do intensive PK studies, crossover studies, in that many people. So what are the techniques you try to use to decide what these distributions are?

DR. FLOCKHART: I think there are techniques, but they haven't been used a huge amount, Lew. There are a number of not necessarily recent, but there are a number of studies over the years where people have looked in large databases to look for well-known interactions. I am thinking of things like interactions between ACE inhibitors and potassium, those kinds of things, the things that are fairly well documented, and looking in large populations to see how real they really are.

DR. SHEINER: How would you know?

DR. FLOCKHART: If you have the mechanism biologically understood, you can go into a large database like the Reagan Strafe Institute database at Indiana and look at the number of people who actually coprescribe those two things who actually get hyperkalemia.

That kind of activity is valuable, I think. We haven't done enough of it. But increasingly, as we move towards being able to use databases like that more--for two reasons. One is there are more of them. Two is the data in them is becoming more reliable. Three, I guess, is they are becoming more accessible. So I think those kinds of estimates are things that are not something that we talk about or use widely, certainly in medical practice, at the moment but it is the kind of data that really ought to be integrated into a doctor's thinking about coprescribing drugs.

DR. VENITZ: Shiew-Mei?

DR. HUANG: Just a clarifying question. Dr. Neuvonen, you mentioned, during your talk, that there are quite a few CYP3A substrates, that they are also CYP2C8 inhibitors. Are you talking about some of the 2C8 inhibitors or just some of the substrates that we have not evaluated as inhibitors?

DR. NEUVONEN: If I remember correctly, there was a study published in British Journal of Clinical Pharmacology some two or three years ago where they showed that many of the typical substrates of 3A4 were inhibitors of 2C8 so that when they are used in vitro, concentrations which were roughly five times the KM volumes, regarding the 3A4 enzyme, these compounds caused nearly total inhibitor of 2C8. I guess it was a paper by Ung et al. I can't remember exactly.

DR. HUANG: Thanks.

DR. SADEE: I just have a general question. When preclinical data are being submitted, are all these P450s covered in the preclinical data that are submitted to the FDA or is it mandatory now? What is the status?

DR. HUANG: I guess you meant nonclinical human microsomal data.

DR. SADEE: Right.

DR. HUANG: For reaction phenotyping, for metabolic pathway, in addition to the five critical enzymes, 1A2, 2C9, 2C19, 2D6 and 3A, most of the time, for reaction phenotyping, we also see 2A6, 2B6, 2C8 and 2E1 data. For inhibitors, the five are the ones that we most consistently see. Sometimes, we also see 2B6 and 2C8.

For induction, it is 3A is the majority that we look at. In addition, some of the 2C9 and 2C19. Increasingly we are seeing 2B6 and 2C8 in addition to 1A2.

DR. SADEE: So there is no guideline as to what preferably would have to be presented?

DR. HUANG: In the past, we have stressed those five that I mentioned earlier because it constitutes 90 percent of the metabolism of most drugs as metabolized by CYP enzyme. But, increasingly, the tools are available as we discussed today when we have more specific probes and we have inhibitors in vitro available. We are going to include those in our guidance on what substrates, conditions, were studies so that the study will be valid to be able to be evaluated.

However, in vivo, based on today's discussion, we are probably not ready to make a strong recommendation until we have a better idea. I guess some of the substrates, we might be able to recommend, and some of the inhibitors, especially in light of possibly inhibiting multiple pathways. So, ever if they are nonspecific enzymes, they might be able to be useful in certain conditions.

DR. SHEINER: I have got to get back to Greg's question. How is that going to translate in labeling? What are you going to say when you find that there is a possibility that lots of different drugs taken together could make a big difference in the metabolism of something else.

DR. FLOCKHART: I don't think you are going to do that. It is going to be guided. So, for example, at the moment, pick a drug, Versed, midazolam. We have in the label that you see a big change with ketoconazole, erythromycin, clarithromycin. That is totally appropriate. It is the main metabolic route.

But what we are seeing here, really, is that increasingly companies, for good reason, are coming up with drugs that avoid one isoform for genetic reasons and for drug-interaction reasons. That is to everyone's benefit, probably, because they have alternative routes when one is cut down. But I think the next level of sophistication here is really to be able to say, okay, I know this drug is a 3A, 2D6 drug and what happens if I put in ritonavir, which kills both enzymes. That is the logical sequel to Bob Temple's saying the worst interaction would be keto. Well, for that drug, the worse interaction may be ritonavir, something that kills both.

DR. HUANG: I just want to add that looking at these interactions, some of them are multiple interactions, some of them are specific to drug interaction. The utility is at least twofold. One is to help us in designing our study and to evaluate the safety database. For example, as shown yesterday, we look at the most stressed system where the exposure would increase because of multiple--right now, we are talking about one at a time--multiple drug interaction.

So what kind of exposure do we need to evaluate? So that is what these interactions can provide us, and the other one is the labeling that has been discussed where we have different degrees of labeling depending on the severity of interactions. Sometimes, we contraindicate or sometimes we modify the dose or dosing interval to accommodate a certain drug interaction.

We have not given specific instructions when multiple drugs are given together. Right now, it is still individual drugs.

DR. VENITZ: Any final comments on the metabolic drug interactions? Mary?

DR. RELLING: Just that, based on what Lew is saying, the most important thing is to carefully describe what has been done to determine which enzymes are involved in the disposition of the drug. We can't predict five years from now what potent 3A inhibitors or PGP inhibitors or 2C8 inhibitors may come on the market that we don't know about and we have to trust pharmacists and physicians to keep educating themselves, to keep providing public sources of what those inhibitors and inducers are. But you can't expect the manufacturer to list all the drug interacting agents at the time the drug is approved. But you can expect them to carefully list what has been tested and what hasn't and give a guesstimate of KMs or affinities so somebody can come up with--a knowledgeable person can come up with recommendations of how to avoid or modify drugs.

DR. SHEINER: I really like that. So the dossier, so to speak, is about your drug.

DR. RELLING: Yes.

DR. SHEINER: And not about all the other ones.

DR. RELLING: You are responsible for your drug.

DR. SHEINER: Right.

DR. NEUVONEN: I would like to add to the previous, that when studying the contribution of different CYP enzymes in vitro, I hope that the substrate concentration used is as close to that in vivo as possible because the contribution of different enzymes may be quite different at different concentrations. I think there have been some artificial data previously based on those kinds of errors.

DR. VENITZ: Final words on drug interactions? Thank you.

We are moving to our next topic and our last topic for today, pharmacogenetics. I am going to ask Larry to give us the introduction.

Pharmacogenetics: Integration into

New Drug Development

DR. LESKO: Thank you.

[Slide.]

We are in the home stretch talking about a related topic but still somewhat different. I want to introduce the topic of pharmacogenetics and integration into new drug development. This is actually the first public advisory committee in which these issues, I think, have been discussed in a general way, although we have had other meetings that have discussed specific pharmacogenetic issues.

This is really the beginning of a discussion on this topic. I anticipate we will have many more of them within this committee and, perhaps, some others. So today is really a starting point to open up the discussion of where we ought to be going with pharmacogenetics as it matures in the context of drug development.

I think of drug development as not only what a sponsor does during the research phase in getting an NDA put together but drug development also includes the regulatory decision stage as well so a lot of what we are talking about encompasses that entire scope.

[Slide.]

I mentioned yesterday that pharmacogenomics is one of the key areas in the FDA's new strategic plan that came out in August. As part of that strategic plan, there are some target goals for the development of guidances related to this topic for the purpose of advancing pharmacogenomics in drug development and its use in public health.

We had a workshop last week on the first of these guidances that was released on November 1. It was called Genomic Data Submissions. This DIA workshop was intended to gather public comment on this draft guidance and also to raise issues related to the integration of this information in drug development and how it might be submitted to the FDA in one of various pathways depending on the criteria that define it.

Dr. McClellan opened up the conference, and this quote is taken from his presentation which reflects the strategic plan and the interest that he has as well as our Center Director and that is we need to speed up the use of genomics to help make our medicines safer and more effective.

Part of speeding that up is to provide guidance to the industry, particularly in an area that is evolving where there is a lot of uncertainty as to how the FDA views this data and how it is going to use it. So this was the first of several guidances which are targeted for the genomics area. Two more are targeted for 2004. One of them is a general pharmacogenomics guidance which will touch upon the issues I will introduce today.

[Slide.]

Pharmacogenomics, or pharmacogenetics, is a broad area so I want to try to narrow the discussion a little bit and thus I will define pharmacogenomics as a tool, a tool to segment phenotypes based on genotypes. Pharmacogenomics, in and of itself, doesn't necessarily cause bad things to happen or good things, but it is a way of finding out information about patients. What we do with that information is, of course, what we want to discuss.

The focus is on interindividual variability in pharmacokinetics. We can also talk about pharmacodynamics but not for today. The problem is basically one dose given to many genomes results in different degrees of variability and different degrees of exposure; that is, the patients.

For the purposes of today, let's define phenotype as an exposure metric--for example, area under the curve--or pharmacokinetic parameters such as intrinsic clearance, and let's define genotype as some inherited variation in drug-metabolizing enzymes.

[Slide.]

The problem is interindividual variability. This is a major obstacle for effective therapeutics, as we all know. This variability predisposes people to risk. We give the same dose to many patients. We have some that react fine, some that have adverse events and some that don't react at all. So there is a wide spectrum of patients. Part of that is thought to be related to the genetic characteristics that affect the metabolic activity.

It has become quite common in clinical pharmacology to conduct studies routinely during drug development to focus on the so-called intrinsic and extrinsic factors that affect PK. These include the well-known ones of demographics such as age, gender, ethnicity and race, the diseases, hepatic and renal, and, as we just discussed, the whole spectrum of drug interactions.

What we do with the information is look at the potential need for dose adjustments based on changes in exposure, usually, sometimes changes in exposure and response. Then, based on that change in exposure under the special-population situation, we recommend adjusted doses that we think will provide exposure that is considered safe and effective.

Where we have come to is that genotypes have become known to influence exposure and these influences are as large, if not greater than, the factors that we routinely consider in the clinical pharmacology area of drug development. I am talking about the factors that relate to the alleles of the common enzymes that have polymorphic aspects of the drug metabolism.

[Slide.]

This is not necessarily new. Everyone here is familiar with the well-known polymorphisms and drug metabolism. We discussed TPMT extensively in our first two meetings and 2D6 is well known, responsible for a high percentage of the drugs in the marketplace and 2C9, less drugs, but some significant drugs with a high incidence of adverse events such as a warfarin.

So the evidence is growing. There is more and more information appearing in the literature on the importance of genetic factors, both retrospective analysis and prospective studies. While all this is not new, what has changed in the landscape recently is the potential that we have to deal with the variability.

[Slide.]

Tests for the cytochrome P450 genotypes have become more widely available, potentially, in the future, FDA approved, and, if available, and if sensitive and specific enough, these tests can be used as an adjunct tool, not much different than blood levels of drugs for individualizing doses of drugs that are substrates for these enzymes. The value of this type of information is that, unlike therapeutic drug monitoring, this can be done in advance of giving the drug as opposed to after administration of the drug.

Likewise, the evidence of clinical utility of these tests is increasing both in the published literature. Oftentimes, years back, it was retrospective but, more recently, in prospective literature. This is not equivocal evidence, necessarily, and there is a lot of debate about what level of evidence underpins the clinical utility. This is another area that is still evolving.

[Slide.]

Related to regulations, we have labeling regulations that talk about evidence that is necessary to support the safe and effective use of the drug. This includes dosing adjustments in selected subgroups of the larger population. In any case, that labeling should describe this evidence and identify tests or actions that are needed for the selection and monitoring of patients who need the drug.

This, if we interpret it in the context of pharmacogenetics, would also lead one to conclude that a genetic test, if suitably validated analytically and clinically, would be a valuable adjunct for label information.

[Slide.]

So the problem that we have to solve, not today but in the next coming year, let's say, is I think we need a systematic way of thinking about pharmacogenomics in drug development; for example, a type of decision tree. When are pharmacogenomic studies important based on some prior in vitro studies, let's say, of drug metabolism? What phases of development might this information be efficiently and effectively gathered?

What types of studies ought to be designed and conducted? How should these results be interpreted and, probably most importantly, at the end of the day, how do we put these results in the label and translate it for the benefit of practitioners and patients?

[Slide.]

One example of a possible strategy, just to start somewhere; let's say we had in vitro data that indicated a pathway of drug metabolism was the major pathway for clearance of the drug and that pathway has known polymorphisms. One might think about determining the differences in pharmacokinetics in the important genotypes in phase I healthy volunteers and then, taking that pharmacokinetic information and assessing its significance in terms of differences using some exposure-response relationships involving biomarkers or clinical endpoints.

That may be where things stop. Maybe there is some significance, but one might think about including complete or partial DNA collection in phase II trials and/or phase III trials in patients. One could design this collection as a prospective sparse-sample strategy with formal population PK analysis looking at genotype as a covariate as we have done before with other covariates in the area of, for example, age or race or ethnicity.

One can also look at retrospective analysis of genotype associations with clinical and safety endpoints and then, from this data, collectively conclude that this is or isn't an important variable in the drug-concentration response relationship. There may be other ways to gather this information but that is the purposes of opening up this discussion. But this is one starting point.

Lastly, labeling products with the information; conceptually, it seems like it would be similar to other special populations defined by other factors.

[Slide.]

Then, finally, there are the questions that we want to put on the table for the committee. The way we planned this session is basically to begin to hear what the issues are and, thus, we have asked the presenters to look at this issue from three different perspectives.

Dr. Flockhart will look at it from his experience in academic research, significantly in the area of 2D6 and some other areas. We have asked Dr. Hockett to come from his experience with developing atomoxetine and what the issues were in that program in terms of what we know about that. And then, thirdly, we asked Dr. Relling to present a clinical view as a clinician--a new drug came on the market that is a substrate for one of these enzymes; in the future, what would you like to know about it.

With those three perspectives, then, we hope we get the issues on the table for discussion and the two questions that we have here, are the approaches presented to study the influence of pharmacogenetics on exposure response sufficient and appropriate. It may actually be a premature question because we don't really have a lot of approaches and it is okay with me if we end up just discussing the issues that might lead us to answer that question in the future.

I think the second question is important; are there criteria or approaches that the agency should consider recommending to sponsors. Again, this may be premature but I think, overall, if we have a good discussion on the issues surrounding the question and the problem we are trying to solve, I think it would be very beneficial to our thinking and, perhaps, we can come back to these questions at a later time for more specific recommendations.

DR. VENITZ: Thank you, Larry.

Let me ask David to come back and take the podium and give us the academician's perspective.

Academic Perspectives

DR. FLOCKHART: I am going to talk about two things, really.

[Slide.]

One is a large picture of how we might approach this process and the second thing is Shiew-Mei asked me specifically to talk about--this is pretty funny--2D6 while I was here. What I heard on the phone was 2B6. So I spent a lot of tie developing my 2B6 presentation before and I didn't realize she also wanted me to talk about 2D6.

So, can you clean up your accent a little bit.

The other thing that Larry has not talked about and I think he does deserve a fair amount of credit for, and the Office, in general, does, and that is for what I think is a real kind of series of acts of leadership that led to the labeling changes for the TPMT enzymes. That is something we have known about for a long time, but the recent Committee on Pediatric Oncology basic approval of what this committee would have recommended, I think, is a real step forward.

Now I think we have to approach other things and so 2D6 came up logically as a next subject. I like to think about big decisions like this in diagrams and some of you are aware of this, pyramids and other things. I have tried to be a little bit more organized this time and presented this way of making decisions as a target, a circular target.

The idea here is that you go from the middle out towards the wider world of healthcare professionals prescribing and patients being treated. You start in the middle with a valid genetic test which is really the basis after you have decided that there is a real distinction, of course, that that test can make.

And then I think you could argue that we may even need a guidance on this. I think there are a lot of things about a genetic test that we assume but which are not written down in code and there is a fair amount of confusion about. Howard, among others, has educated many of us about how many snips in the human genome are wrong and how many we haven't picked up.

I think the characteristics of a genetic test and the series of hoops such as genetic tests might have to jump through from a regulatory point of view are important things that might be the subject of a guidance.

Outside that, once you have that, there is obviously the correlation between that and phenotype. Larry just really alluded to this series of discussions. How do you do that? There are lots of ways of doing it. You can do it retrospectively. You can do it prospectively in a very highly expensive and organized way or you can do it using random sampling. There are lots of efficient ways to do this. But which are the ones we trust and which are the ones we think we should seal with you like the imprimatur of the FDA in terms of a good way for a company to do a test like this. That might also be a subject for a guidance.

Then, beyond a simple correlation of genotype and phenotype, there is the real world, the real dirty world, of drug interactions, diseases, races, genders and really large clinical trials. The genetic tests that we come up with must be robust enough to survive in that environment. I think one might come up with recommendations for how to do that as well.

I am not sure this last one needs to be here. This is economic assessment. But it is something that is in people's mind all the time. It has been done for TPMT. It has been done recently, several times, including I just saw an article this morning, yet another article, about 2C19 and Helicobacter pylori, Greg, demonstrating its economic effectiveness.

But I think this is important to the people who are doing the testing. It is important to healthcare professionals and it is certainly important, I think, to pharmaceutical companies, what is the value of these tests in the larger picture. That is also potentially a subject at least for discussion.

[Slide.]

A way of thinking about this is--and this is an old diagram that I have just reorganized a little bit. If you think about the population treated with a drug--and here I have just got the Y axis, really. This could be a unidimensional. But this is a schematic representation of a population treated, an average drug, where about a third of people don't have a response. So this is no response and this is a response on the upper side.

What we are really doing here is coming up with a genetic variant that would divide these people up one way or another.

[Slide.]

In an absolutely ideal situation, you would have this, an ideal parameter separation where the relative risk between the two things is huge. Unfortunately, there may be situations where this is the case. I am thinking potentially of hemochromatosis and a number of other situations like that.

[Slide.]

But, in fact, in reality, in my experience, anyway, there is hardly ever--maybe I should never say never, but this hardly ever happens and you are nearly always dealing with a messy situation like this. So it becomes important to have a parameter that makes this distinction, that separates these two things.

I think, myself, this is probably a disease-specific parameter. I say a disease rather than a drug or a population because, for many, many diseases, there are separators already. I work in breast cancer. You can predict a person's response to therapy for breast cancer with a large number of things; the stage of the tumor, the grade of the tumor, the number of lymph nodes, the age of the woman. We routinely put this into regular clinical decision making in terms of what we are going to do with women who have breast cancer.

A genetic test that is going to improve on that has to survive in that decision-making matrix. It has to be something that will improve it. I think it is not enough to say it would just survive. It has got to improve it.

So what do we do here? I am really just putting this up for a matter of discussion It is one thing to call it just statistically significant. The clinicians amongst us would say you need to do more than that. It has got to be clinically as well as statistically significant. But we spend a huge amount of our time just testing for the p-value and really not thinking enough about more clinically relevant statistics like the relative risk or, in fact, the absolute risk between these two things.

As clinicians, certainly as someone who teaches clinical pharmacology, I try and encourage our residents and interns and medical students to think in terms of absolute risk because it is a more valuable thing in many contexts and, indeed, to think about the number needed to test or the number needed to treat. So the number of patients you would need to treat, to come up with a significant outcome, or, in this case, the number of patients you would need to test in order to come up with someone who really had a significant difference on one side or the other; what is the parameter we should use?

I am not standing here saying we should use one or the other. I am saying we should have an intelligent and informed discussion about how we do this. I, personally, am biased towards thinking this is a disease-specific thing and that, in breast cancer, I could give you the relative risk caused by four lymph nodes. I could give you the relative risk brought about by a woman being aged--having a stage 3 tumor.

I know those numbers. Therefore, if I had an equivalent change caused by a genetic test, I would think that might be something valuable. Something that was less than that would not be as useful.

[Slide.]

I am going to change tracks completely. That ends my general statements because Shiew-Mei asked me to talk about 2B6--I mean, 2D6. The specific question that Shiew-Mei asked me to address was the question of distinction between the extremes. So, I guess, in some ways, it is related to the same thing.

[Slide.]

Just to summarize very quickly about 2D6, we know it is absent in 7 percent of Caucasians. Fascinatingly and interestingly, it is hyperactive in 30 percent of East Africans including Ethiopian and Saudi Arabians and a number of people in Spain. It ketolyses the primary metabolism of a large number of drugs which is why we are talking about it, really, and is potently inhibited by a large number of equally interesting drugs.

[Slide.]

This is frozen? This slide didn't come out in the handout? It is a big figure.

[Slide.]

Just to make some points about this. These are old data from the Swedish group. Debrisoquine is a probe for 2D6 activity. This is the number of subjects. We can clearly distinguish these people because they are two logs different from the mean over here. So poor metabolizers are, in general, a completely separate phenotypic group. There is a cutoff here. There is also a cutoff up here and, for the very fast people, these are actually, I am increasingly coming to believe, very distinguishable as well.

We had someone recently who destroyed codeine at a rate, really, that was almost 100 times someone in the middle here. So there are unusual people at the extremes out here but it is not really, if we are honest about it--like, there is nobody in here. This is something like a thousand subjects. So, inevitably, if you increase this to a million subjects, there would be people in here who it is hard to distinguish.

If you are talking about 2D6, this is 7 percent of the population. But this is much more of the population. This is well over 30, 40 percent of the population. There are people who are intermediate metabolizers of one kind or another. So the difficult question for a company is what do you do about these people. Do you make any kind of dosing recommendation at all or do you just leave that there.

Now, the case for making any recommendation would be that there would be, if there is a difference in pharmacokinetics that is real in this group, and secondly that there is a large number of people in that group. What I am going to say is two things. I am going to say that the answer to this is really sometimes it is worth it but not always.

[Slide.]

This is from Michael Eichelbaum's data in a paper published with Esmeier and a number of others in 1997. It is a very bad slide, I'm afraid, but it basically shows that this is ultrarapid metabolizers and poor metabolizers by genotype here. You can see that there are a group of people who you genotypically predict to be in the middle but, nearly always, they overlap with these people over here. So, for this given genotype here, which is a star-1-star-1 genotype--this is the old nomenclature; I'm sorry--it overlaps over here whereas this also overlaps.

There are a group of people, the star-10s here, who are intermediate. But, certainly, when this was published, six or seven years ago, there weren't clear ways of distinguishing this group.

Since this was published, and I am missing the allele slide that I had, we have really relatively ethnic-specific alleles, the star-10 allele among Asians--I say relatively, because it is not absolutely. You can pick up star-10 in Caucasians and you can pick up star-10 in Africans, but it is a relatively Asian allele.

Star-17 is an African allele. Andrea Guideker and Greg Kearns' group has shown the importance of star-29 in African-Americans as well. So it is possible that it is able to define people--it is possible now to define people more who are in this group and we can discuss that a little bit.

[Slide.]

Now here is the difficulty. These are also data from the Swedish group. So this is the number of functional alleles against nortriptyline concentration. You are simply looking at concentration on a normal, not a log, scale against time. So this is the number of functional alleles. A poor metabolizer would have a rate and a half-life like this.

But you notice that, if one allele is deficient--so if this would be a star-4 heterozygote, for example, someone who had one knocked-out allele, one completely dead, nonfunctional, completely inactive half of the DNA and the other is perfectly active, and that person has a very slightly different pharmacokinetic profile from this person, but a very notably different pharmacokinetic profile from someone who has two alleles knocked out.

This is true for a number of drugs but not all. So this is a situation where, if you change from two active alleles to one, you see a significant change. This is a substrate-specific thing, I believe, and there are substrates where, if you go from two to one, you don't see much change.

[Slide.]

But we, and others, have modeled these kinds of data. So, if you look at the number of functional alleles at a low dose, 25 milligrams, you see people come into the therapeutic range and, at a middling dose, you see people exceed the therapeutic range and, at the 75 milligrams TID dose, you see people go way above the therapeutic range and people who have two or three functional alleles fall nicely in the therapeutic range.

[Slide.]

Now, these kinds of data have been used by the Europeans to come up with dosage guidelines. This is just a diagram from the omega document on dosing nortriptyline. So this is doses of nortriptyline recommended for different 2D6 phenotypes and genotypes in Europe. So this, again, is the same debrisoquine diagram that I showed you, number of subjects, rate of metabolism in the inverse. The poor metabolizers are over here.

The genetic variants are indicated in these cartoon forms. The X is a knocked-out allele, so that would be here and here, and the multiple-copy alleles are over here. And the doses predicted from the model by the European group--I am trying not to designate any particular person because there were so many people involved in doing this--were a 500-milligram dose, 100 to 150-milligram dose, or 10 to 20-milligram dose. So this is a ten-fold difference, a fifty-fold difference, from one end to the other of nortriptyline dose according to the phenotype and genotype.

Obviously, what these people have done here is they have made a recommendation in the middle, even though I showed you a moment ago that there is not a huge difference between the pharmacokinetics of nortriptyline in a heterozygote compared with someone over here. But they have gone ahead and done it anyway because this variation is so large.

So the important question, I think, for us, is are there substrates where we should do a similar thing.

[Slide.]

I am just putting these questions out. So two recommendations. These are really both recommendations for discussion. In the long-term, over the next several meetings, we should define and make clear a disease-specific parameter that is a target for useful pharmacokinetic tests and, secondly, for these three isoforms, at least, and I would recommend that these be the first addressed, we recommend a genotype and phenotypic test that defines this. We, at least, can get into this discussion.

Personally, I am not here yet. I haven't got this really clear in my mind and I am not sure how we would recommend doing this but it is an important thing that is worth discussing.

So I will stop there and I think I might sit down as well. If there are any points-of-information questions that people have, I would be glad to deal with that.

DR. VENITZ: Are there any information questions for David before we get into our discussion?

DR. SADEE: With the heterozygotes, it is not clear why they would be, necessarily, closer to the homozygous null carriers.

DR. FLOCKHART: You are right. It is not clear. It is an observation.

DR. SADEE: So, most likely, the ones that one finds to have this, the other allele has something wrong with it, too, that may be less well expressed.

DR. FLOCKHART: You mean, it is not a knock-out? The other allele is--

DR. SADEE: No; one is a null allele and the other one would be less well expressed in some fashion.

DR. FLOCKHART: Conceivably an interaction because of the absence of one allele, you mean.

DR. SADEE: Yes; or the ones that you find have relatively poor metabolism. It is just there is another genotype that affects this that we don't know about.

DR. FLOCKHART: What you are talking about is, in this situation, where one allele is dead.

DR. SADEE: Right.

DR. FLOCKHART: And this situation is where both alleles are dead.

DR. SADEE: Yes.

DR. FLOCKHART: So what--

DR. SADEE: Then the gene, the allele that is not dead, is somehow impaired and that may be a polymorphism that is not described.

DR. FLOCKHART: Oh; I see what you meant. I'm sorry. We might be missing one here in this particular setting. That is possible there; yes.

DR. HOCKETT: If you get more than a few patients, that can't be the explanation because there aren't that many alleles that decrease function a little bit that would give you that picture. So it is going to depend how many patients went into the formation of this graph.

DR. FLOCKHART: And how many alleles, actually, because we have done so much on 2D6. I mean, we are still beating up new alleles. We have 43, 44 new alleles. Really, if all of them were tested here, and I don't know that they were, but the vast majority--these are people who know what they are doing, I think, in general, Sweden--the vast majority would have been tested here so it is possible that the average--that if this were one patient, which it is, that that could be the case. But if this were a population average, and I think you could plot a population average like this, it would be hard to explain it that way.

DR. VENITZ: Any other questions? Thank you, David.

Then our next speaker is Dr. Hockett. He is going to give us the industry perspective discussing a recently approved drug.

Industry Perspective

DR. HOCKETT: Good morning.

[Slide.]

I appreciate the chance to address the committee with an industry perspective but I caution you, there is no way I can give you an overall industry perspective so you have to take this in light of what this would be consideration of one person at Eli Lilly.

[Slide.]

Like David, I am going to start off with a few general comment. I am actually going to get on my soapbox for a couple of slides. I think there have been a couple of difficulties for the field in pharmacogenomics and I will go through those.

The title is a case study of Strattera. I will talk a little bit about Strattera because that is the most recent example of where a genetic test has been put in the label, at least a mention of one. Then I will talk, again, about some more generalized thing about pharmacogenomics and how we think they are going to apply and what I think, from my perspective, would be nice to see as far as CYP2D6 if it would have been required in the Strattera label.

The first problem that I think the industry has had and, thankfully, it is getting less and less as we progress, is illustrated on this slide where there are far too many definitions of what we are talking about; pharmacogenomic, pharmacogenetic, applied genetics, applied genomics. I reminds me a little of the "po-tay-to"/"po-tah-to" argument as to how you pronounce this.

I have seen the slide of David's that now says that pharmacogenomic and pharmacogenetic are actually just a two-snip change of the same terminology. I didn't steal that from him but he has done that. In fact, you can see several groups. EMEA, which is the FDA equivalent in Europe, has got a very broad definition. The PWG, which is a loose consortium of pharmaceutical companies and biotech groups called the Pharmacogenomic Working Group, actually has split the definition. Why they have chosen, and I am with this group, to split hairs is still unknown to me. It is not very helpful. Even at Lilly, we have subdivided this. In pharmacogenomics, we have a little bit narrower view. It really means we want to understand the genetic influences of how people respond to drugs.

None of those are right or wrong, but you can say it leads to confusion in the field.

[Slide.]

The second problem that we have had is pharmacogenomics has been hyped, I think, an overamount in the field. We are not going to have a choice. We are actually going to have to do this. The field, pharmaceutical companies, will be dragged, kicking and screaming if we don't help lead the way.

I illustrate this from this U.S. News and World Report that actually fell on my doorstep in January of this year where the cover of this said, "This drug is for you." There have been several magazines that do this. Interestingly, if you open up this and look at the article, the gist of this was that we are all going to run around with our human genetic sequence on a card about the size of a credit card. That will allow physicians to figure out which diseases you are going to get, which drugs he can give you to prevent those diseases you are susceptible to and, if you get a disease that wasn't predicted, what drug.

Boy, that may happen. But it isn't going to happen anytime soon. There are multiple problems, not the least of which, how much is it going to cost me to sequence a single person. The first time we did it was several billion. We are probably a log-fold or two less than that now, but, even if it was a million dollars, how many of us are actually going to have the sequence done.

Second off, even if I could sequence everybody in this room, I don't know how to interpret all the variation yet. There is not enough data for me to understand disease susceptibility versus drugs.

In fact, I have put a collection of my favorite hyped sayings for pharmacogenomics here. I am not going to go through those, but some are rather interesting such as, "Applying pharmacogenomics to drug development will cut cycle times to 1.5 to 2 years." I don't see that ever happening. I just think that is not going to be true and I think we are deluding ourselves.

[Slide.]

What this combination has done is what I would like to illustrate on this slide. This is, in applying new technologies, you have this gentleman with the telescope and let's equate that with pharmacogenomics. He hasn't got his eye on the prize. He is looking in the wrong place. I think this has deflected what we should be talking about in pharmacogenomics, when he has missed the comet over here in the sky.

This comet, I think, for pharmacogenomics, is developing new genetic biomarkers that will allow us to predict how people are going to respond to drugs, not we are going to change cycle times, not that I am going to be able to predict everything. But, in certain instances, we are going to develop specific biomarkers that are going to help us do it.

[Slide.]

Now I say this because at Lilly I sat down with my colleagues and we developed a list of how we are to apply genetics to drug development. We really apply that in three areas; in the discovery arena, in preclinical toxicology, where we give these drugs to animals and try to make sure that they don't destroy a whole bunch of organs when they then go into humans, and then in the clinical side.

You can see there are lots of different things but, in reality, we have two key activities and two key activities only. The first one of these is to identify and understand targets. We want to use genetics to try to figure out where there is the next available drug target for an unmet medical need. Then the second one is to develop human biomarkers where I can actually predict, then, who should be on a particular drug, either for a positive reason--they are going to have efficacy--or a negative reason--to avoid toxicity or adverse events. That is what we are going to talk about.

[Slide.]

What I have listed here, then, are the broad categories where genetics is going to be applied in medicine currently. We have two big areas called disease-susceptibility biomarkers and drug-activity biomarkers.

Now, the disease-susceptibility ones are those that would predict you are going to come down with a genetic disease. You are familiar with several of these, especially under the single disease genes of Mendelian inheritance. This is where I think I would differ with David when he said the absolute distance between a genetic event and a response never happens. It actually does in the Mendelian inheritance like sickle-cell anemia.

If you get two copies of the disease, you have the disease and, if you don't, you don't. But, other than that, he and I agree precisely. However, in complex diseases, that is much less the way it is. If you take Alzhemier's disease and Apo4, it has got a fairly large relative risk but it does not separate the population at all, and we will come back to that.

Then the drug-activity biomarkers which some would call the true pharmacogenomic biomarkers. This is where I think, as a drug company, we need to spend all our time. I have put the one in green that we are talking about today those things that happen when you have defects or variants in metabolic enzymes and that leads to changes in PK profiles and can lead sometimes to profound toxicities.

[Slide.]

I have put a list of things on here where we, as a drug company, would choose to include genetics in drug development. Contrary to some prevailing opinions in at least the lay press, we don't like to give drugs to people who are going to respond badly. It is not very cost effective for us to have adverse events and severe toxicities. So we are very much in favor of trying to identify those individuals and keeping them off our drugs.

We may get into the discussion are we willing to subdivide our market, et cetera. That is actually an entirely different topic. But you can see we are planning to apply this very early in discovery and all through clinical development; phase I studies of a particular type, mainly in the PK variety, Phase II and III if we can use to figure out who is going to respond either positively or negatively to our drugs.

[Slide.]

For Strattera, it is primarily metabolized by CYP2D6. You can see there are profound differences in the plasma clearance, a ten-fold difference if you are poor metabolizer. In fact, the AUC has got a ten-fold difference, ten-fold higher in this case, if you are a poor metabolizer and the half-life is significantly extended.

Obviously, we were interested and concerned about this. Did this lead to safety concerns or just did it have tolerability or efficacy issues. That, obviously, the interplay between those things, would have profound implications for the label. So, if it developed a very severe toxicity, it may become a label requirement. If it is simply a tolerance issue, it might not be. And you will see that is, indeed, what happened.

[Slide.]

When you look at the clearance of Strattera, and this is the number of patients, and the plasma clearance here, this very much looks like the metabolizer status that Dave showed on one slide and I am going to show in just a minute where you have got the poor metabolizers down here in black. You have got the extensive metabolizers here or the wild-type variants and then the ultrametabolizers here. It looks very much the same for Strattera as it does for any kind of drug.

[Slide.]

I am going to show you just one slide of data. It came from a single study. It is the best data that I think illustrates the point. We did some initial clinical pharmacology studies to look at what the maximum dose was. We looked at some CYP2D6 genotypes obtained under double-blind conditions. Therefore, the clinicians are now going to start patients on a dose of the drug not knowing what their genotype is. Then are then going to adjust the dose based on toxicity, tolerability and efficacy.

In the end, then, we are going to compare EMs to PMs and see where they ended up and where there are large changes in the ultimate dose they were given for efficacy, toxicity and tolerability and were there any differences between EMs and PMs.

[Slide.]

So that is what happened. It is illustrated on this slide where you can see the extensive metabolizers are in green, the poor metabolizers are in purple. The bottom is weeks of therapy and the Y axis is the mean dose in milligrams per kilogram per day. You can see the comparison between EMs and PMs is essentially there is no difference.

So, without understanding EM to PM differences in prescribing these drugs, they actually ended up on the same dose which means we haven't got a profound toxicity problem with PMs in Strattera.

[Slide.]

To summarize several different kinds of studies on this slide, there were some adverse-event discontinuations in all studies. In fact, poor metabolizers had a slightly higher level than extensive metabolizers except they were based on insomnia and irritability not on profound toxicity. So, in the end, what we really had was a tolerability question and not a safety question.

There was a slight hint of efficacy increase in PMs especially on an ADHDH response scale compared to EMs, but we didn't have enough patients in there to make that terribly profound and, obviously, there weren't enough patients to affect the label.

[Slide.]

So, in negotiations with the FDA, CYP2D6 was put in the label. In fact, it occurs seven times in the Strattera label in the Pharmacokinetics Section, Adverse Events Sections, Drug-Drug Interaction Sections and the Laboratory Testing Section. But it is not a requirement because there is no profound safety issue dealing with CYP2D6 in Strattera.

Here is one of the verbatim quotes. Actually, this has been mentioned already today where it talks about the incidence of poor metabolizers, et cetera, as well as having to pay attention to the alternate drugs that may induce a poor metabolizer status.

[Slide.]

Obviously, as far as Lilly was concerned, that is almost a non-event in pharmacogenomics. It is in our label. We were happy to have it mentioned in the label. For medical reasons, we don't mind people testing. But it didn't make any sense to require it because there wasn't a toxicity issue and we agreed and we came to terms.

Obviously, that doesn't give you a whole lot to talk about and so I am going to expand this a little bit in how do you define PM status and how actually, if you do have one that is required, would you put it in the label.

For some of the P450s, it is actually pretty easy because there are a couple of alleles. Dave has already alluded to CYP2D6. It is more problematic. There are actually 44 alleles defined, as he already said. This is a typo. Actually, there are 21 alleles that have been defined that have absent activity. The vast majority of those are at such low frequency that they probably shouldn't be routinely ordered and we will come back to that in just a second.

There are at least two that are classified as decreased or intermediate alleles, star-10, star-29, star-17 and then a duplication exists in this where you can have more than two copies of the gene, et cetera.

[Slide.]

All of that leads to some problems. In fact, as Dave has already alluded--and he and I didn't talk about our presentations beforehand and they ended up being remarkably similar--you have vast differences in ethnic groups.

Here are Caucasians that have a 5 to 7 percent incidence of poor metabolizers. You have Asians where the poor metabolizers are actually less than 1 percent. But then they have a significant number of intermediate metabolizers. In fact, there can even be differences among Asian groups.

And then there are a bunch of ethnic groups that we don't have any data, or at least there is no published data, on what this means. What this is going to come down is you are probably going to have different recommendations based on different ethnic groups and different alleles that need to be ordered if you are talking about a different ethnic population.

[Slide.]

Just like Dave, here is my requisite phenotype/genotype slide. It is rather complicated but I think there are about three or four important points to make on this slide. On the bottom, here, is the metabolic ratio which, in this case, is a dextromethorphan/dextrorphan ratio. On the Y axis is the genotype under the star allele nomenclature. Here, there are three, or the amplified status, two, one and no functional alleles.

Then you have got the designations here of where the ratio is in relation to the genotype. The first important point, as Dave has already pointed out, there seems to be relatively good separation of poor metabolizers. I have seen at least a dozen or eighteen different studies that show the same kind of thing. It is relatively easy and there is a decent phenotype/genotype correlation for poor metabolizers.

For the rest of these, there actually isn't, in my mind, a very good separation, in fact, if you have got one allele. There is a huge overlap if you have got two functional alleles, at least for dextromethorphan. Obviously, this kind of decision has to be data driven and, if you get additional data, especially with different drugs that separate them, that is true.

Even if you look in places where they have only one functional star-10 allele, which, in this case, is this star-4, star-10, where they have three patients here, those people are clearly not over here in the intermediate. They are well within the extensive-metabolizer status for this drug.

So I would agree with Dave. It is not very clear. Then, if you look under the ultrametabolizers up here, I think, for this case, they have such an overlap that it is not useful to distinguish between those two.

[Slide.]

Obviously, then, you have a decision; are you going to require a phenotype or a genotype. I have put just a few things up here. It is not exhaustive. There are some advantages to going to either side. From a drug-development perspective, I would prefer a genotype. The reason for that is I can measure it at once, as Dave said, before I give any drug and I can actually measure a bunch of alleles at a time and get more than one drug, or more than one metabolic status for one particular enzyme out of this.

In fact, we are developing a chip at Lilly where I am going to be able to look at 120 different genes all at one time for a relatively inexpensive cost and, obviously, then, we are going to prefer to do that kind of thing.

[Slide.]

I have just a couple more slides and then I will end because I think I am getting close. There are, then, a bunch of other considerations that come into play when you decide to do a genetic test. I can't do justice to this topic. It is probably worth an hour's presentation in and of itself.

But I am going to touch on two of them, and those are the first two on the list here. The ethical, legal and social implications of this can be rather profound. It really has to do with the population's reticence at doing genetic testing and their fear that something bad is going to happen to them, like insurance is revoked or they are going to be labeled in some way if they do a genetic test.

For this reason, I am in favor of having something like a metabolic enzyme be one of the first tests that are propagated here because it doesn't have the disease-association status that some other things such as complex disease would have and will potentially be swallowed by our public much easier than a different kind of test.

[Slide.]

Obviously, that is a profound thing that we have to get over. We also have to make sure that we educate them properly and try to get rid of the hype for what this can do and talk more about what it actually is going to be practical to do on that kind of scheme.

[Slide.]

Then the last one is the utility of the information and biomarker. Once again, these are very similar to the slides that Dave already showed. This is the best case; if I have got a genotypic variation and a response, I get absolute discrimination between the two. I agree, that almost never happens.

We will be lucky if we can get them that have this kind of separation. This would probably be acceptable. If they are like this, I don't think they are even going to be instituted or accepted. The example we have, I think, the best example we have for where it is a poor separation is the Apo-E4 variant that causes at least 50 percent of Alzheimer's disease in the Caucasian population. But it has such poor separation between those that are going to get Alzheimer's disease and those that are not, that I don't know what to tell the patient if they come down with an E4; you are at slight increased risk to get Alzheimer's but certainly not guaranteed. That kind of test, obviously, is not going to be very widely accepted.

[Slide.]

The last slide I have is CYP2D6 recommendations. I think the PM genotype predicts the PM phenotype in roughly 99 percent of cases. That is at least shown very well in two very large studies that have been published. Since there are 21 alleles that actually cause a null phenotype, you would think that is very challenging but, if you look at these very large studies, they do it with only about five or six. Those are the most frequent ones that are found in these populations. Here is the listing of these more frequent alleles that pick up about 99 percent of this.

I think, to avoid confusion, the FDA should specify that you can do both phenotype and genotype as acceptable methods for defining this PM status, but I think this should include a recommendation for what is minimal genotyping, a minimal number of alleles that would be acceptable for that kind of genotyping, similar to the list that is supported in the literature.

I don't think the genotypic designations of UM, IM and EM have--they have distinguishable phenotypes on a population basis but not on an individual patient basis and, therefore, I am not actually in favor of indicating them by current data. The important point here is current data. If we generate specific instances where you have a separation between the two, obviously, it has to be a data-driven decision.

Then, recapitulating what has happened with the Strattera label, genotyping for these mutants is warranted only when a compound's margin of safety is exceeded in poor metabolizers and, if it is, then I fully am in support of it actually being a requirement in the label.

[Slide.]

The last one of these things I just have is we are all in favor of getting the right targets, the right drugs, into the right patients. Within our education program, though, we have to make sure we convey what we think is the appropriate time line for this. Unfortunately, my guess for this appropriate time line is after I am done, actually, practicing in this field. I think it is going to be dozens of years before we get to that ubiquitous type.

With that, I will stop, take general questions, if you like, but I think there is going to be a discussion in the end.

DR. VENITZ: Thank you, Dr. Hockett.

Any specific questions, informational questions, about his presentation?

DR. FLOCKHART: Just one question, Rick, within an excellent presentation. But the graph you showed of Strattera, the population-distribution graph, you had shaded the UMs, the ultra-rapid metabolizers.

DR. HOCKETT: Yes.

DR. FLOCKHART: How had you defined them? There was a big overlap, but how did you define them?

DR. HOCKETT: Whether they had three or more copies of a functional allele which had been the star-2 allele. That is how they were defined. I have seen descriptions of thirteen copies, but we never saw anybody over four copies. It was always three or four.

DR. FLOCKHART: That is one family in Sweden.

DR. HOCKETT: Yes; right. Exactly. Generally, I don't see that number. It is usually just three or four, it appears.

DR. KEARNS: Rick, when you did your presentation, you had a slide that suggested when Lilly might include pharmacogenetics in--did you find, in the PK data for Strattera, that having 2D6 genotype was useful in examining your PK data?

DR. HOCKETT: Yes. It clearly helped distinguish who was--we had a very high correlation between the genotype/phenotype. So when there was a poor metabolizer by PK, it came up poor metabolizer by genotype as well.

DR. KEARNS: So do you think it is reasonable and, I guess, where I am going here is in pediatric studies, in particular, where the numbers of subjects in a PK study may be smaller then in an adult phase I or phase II, that, when there is a drug that is metabolized by a polymorphically expressed enzyme, having that genotype data, assuming it does correlate with phenotype, can be useful in separating out, perhaps, is there an age effect on the disposition of the drug.

DR. HOCKETT: Let me answer that, or respond to that, in two ways. We are developing a program at Lilly where we are going to be doing metabolic-enzyme and transporter testing out of every phase I patient who comes through our clinical trials. That will include pediatrics because we think it will help us understand the PK.

The only hesitation I would have is in definition of what useful means. I think there is going to be a scientific useful and a drug-development useful. Generally, we don't have enough patients with enough PK outliers to know precisely what is going on and say, with absolute certainty, that it is due to a particular genotype.

We don't necessarily need that to understand if we have to worry about it in phase II and phase III. So if you allow me that distinction between absolutely scientifically proving and then figuring out what we have to follow in phase II and phase III, then I would agree that it will be useful for us to understand and have the genotypes on these individuals.

DR. SHEINER: I am not questioning the conclusions you drew about the drug you spoke about where you found that this difference in metabolism didn't reflect in the difference in outcome, but I do want to discourage the use of the design wherein you conclude that some genetic difference or anything else is not important because you find that physicians ultimately don't wind up adjusting doses differently in the two groups.

The medical profession has a long and glorious history, not only of not noticing what harms they do but actively promoting harmful therapies. So I don't think that is a sensitive way to design a study although I am sympathetic with the notion of saying, what are the practical consequences as opposed to the sort of theoretical ones. But I think we can probably come up with a better design to try to see whether something actually makes a difference than that one.

DR. HOCKETT: Point taken.

DR. HUANG: Either you are going to do a prospective study or retrospective genotyping if you are going to have another 2D6 drug with what we know about, the more alleles that we know are null alleles, which the assay may not be available before. What would be your minimum alleles that you would like to test in order to conclude that the genotype may not have an effect on your adverse events.

DR. HOCKETT: Do you mean in 2D6? I would put it at six or seven, which is what we typically measure, although, by the middle of next year, I think it will be a moot point. The chip we are going to build is going to test 40 or 42 alleles for 2D6 and we won't have to worry about that.

DR. HUANG: Even those that are available, because I just wanted--because not all chips have all the alleles. What are the essential ones on based on, in your opinion, expert opinion?

DR. HOCKETT: Which is available? The ones that were listed--

DR. HUANG: It doesn't matter, available or not. What are the key ones?

DR. HOCKETT: The key ones are the ones that are most frequent that you are going to see. So, in 2D6, it will be three, four, five, six, nine and you might add a couple of others like 16 or 15. But it is really a frequency question. So we typically do about six or seven. Those are the ones that I would continue to really look at.

DR. HUANG: Because you cited two large studies. I assume they are mostly a Caucasian population.

DR. HOCKETT: Yes. The poor metabolizer status in things like Asians and African-Americans, we will add one or two alleles depending on those although, for CYP2D6, right now, I don't think an intermediate status is--and the poor metabolizers for Asians are basically the same alleles as Caucasians but just less frequent.

So that is why I say it is six or seven. We would add a 17 or a 21 for African Americans or Japanese and then those are the ones that we frequently look at. We are still running about a 99 percent genotype/phenotype correlation in everything we have seen.

DR. FLOCKHART: Could I just amplify that a little bit because this is an important point. We routinely add, on the basis of ethnicity--we don't have your chip yet, so we conserve our resources by looking for star-10 in Asians and star-17 and 29 in Africans.

There is a problem with that in thinking about this because we can't really separate those phenotypes yet. So a star-17/star-17 homozygote strictly is not distinguishable in most studies of most drugs from an extensive metabolizer.

Key in this decision-making algorithm is whether there is a gene-dose effect. If there is a really clear intermediate group, and I think there probably are drugs where that is the case. Tamoxifin is one of them where you do see that. But, in general, I think that is going to be a really hard thing. Whether we actually recommend it, I think, depends on whether there is a phenotypic difference.

So I think, in general, I would agree completely with Rick. You need relatively small, five or six, to do it. But I would also agree with Rick in that the point is going to be moot in a couple of years when we will have lots of ways of doing it.

DR. VENITZ: Wolfgang?

DR. SADEE: I do come back to the issue of the heterozygous, even the patients with two "normal" alleles. The spread is so large that it is, in some cases, convenient to say they are poor metabolizers and intermediate metabolizers. But half of the intermediate metabolizers are very close to the poor metabolizers and the ratios that you can see here in the genotype and phenotype plots are such that one would worry about a fairly large percentage of the patients having a very slow metabolism even though they are not classified as poor metabolizers.

So, again, it would appear that there are yet unrecognized polymorphisms probably in the promoter regions and other regions that contribute to this or whatever else factors contribute to that. So, in some cases, it may be useful to just say, here is the group of poor metabolizers and those are going to be the only ones who are at risk. But that may be few cases because the intermediate metabolizers may have such a poor metabolism that they are also at risk.

So how do we deal with that?

DR. HOCKETT: The one thing that is not contained in the genotype/phenotype graph I showed you was a reproducibility among a series of individuals. So you are dealing with a single determination here. I would have bet that there is a fair amount of variability within the group that you can swap them from one position to another within their own distribution graph.

Therefore, I don't know how to interpret those that are close to poor metabolizer status as to whether or not, if you measured them repeatedly, they are always in that position. I would bet they are not, but Dave might be able to answer that question because I have never seen that kind of data.

DR. SADEE: I think that is a key question because, if they are just in the same position, then this is some intrinsic factor. If that is extremely variable, then all bets are off.

DR. HOCKETT: As long as the variability doesn't flip them over to the poor metabolizer on the other side of the ratio.

DR. VENITZ: Larry?

DR. LESKO: Rick, I wanted to ask about the early study in the clin-pharm area. Was this study done by enrolling X number of subjects and then retrospectively looking at their genotype to figure out the difference in pharmacokinetics or was it prospectively enrolled to get suitable numbers in each of the genotypes that you were interested in.

DR. HOCKETT: No. Every study that was done with Strattera for genotyping was done retrospectively. We did collect some things prospectively, but we kept them double-blinded to try to answer the question in a different way. I should say that there were a couple of late phase III trials where they separated the individuals based on poor metabolizer status, but the early stuff was all done retrospectively.

DR. LESKO: Do you think that is the most efficient way to do it?

DR. HOCKETT: No, but that was our first foray into one of these drugs that was going to be necessary. I think we have learned a fair amount. I think we would change our approach slightly.

DR. LESKO: What do you think would be more efficient?

DR. HOCKETT: This is going to open up another can of worms. I think, prospectively, it would be, especially if we find that there is a toxicity that we have to identify or deal with with poor metabolizers. Then we have to gear up to make sure we get an FDA-approved test when our drug is released, is the most efficient, because, for us to be able to sell a drug that requires a test, at least my understanding is you are going to require at least a fair amount of work going down the road to an FDA-approvable assay for that to happen. That has, then, got to be done in parallel. Otherwise, I can't sell my drug.

DR. LESKO: I was sort of coming from another standpoint, the increased cost of screening people to get suitable numbers of genotypes versus just sort of increasing the enrollment in a study and hoping that the breakout occurs--

DR. HOCKETT: Yes; it is far most cost-effective to screen people even if it is $300 or $400 than to enroll them. The average cost in most clinical trials is about, what, $10,000 a patient to carry them through a clinical trial. So if I can screen a bunch to keep that number down, I am much better off.

DR. VENITZ: Thank you, again.

Our last presentation for this meeting is Dr. Relling. She is going to give us the practitioner's perspective for pharmacogenetic testing.

Practitioner Perspectives

DR. RELLING: Good morning.

[Slide.]

I think it has been implicit in what we have all been saying that obviously there are some drugs where the therapeutic range is so wide we don't need to know anything about how to prescribe them and we are willing to give a very high population dose to everybody in order to achieve a high probability of efficacy and a low probability of toxicity, and that it is for drugs with narrow therapeutic ranges.

Of course, anticancer drugs definitely fall in this range where the dose that one needs to achieve a reasonable probability of efficacy is so close to the dose that achieves serious toxicity that anything that we can do to help us to individual doses in any given patient is something that we would try to have.

[Slide.]

So let's go ahead and make the assumption that getting the right dose of the drug for the disease being treated is important. Of course, sometimes, that can be true, but there may be other approaches to titrating the dosage besides doing something like genetic testing.

So, in cases where that might be problematic is, of course, the probability of response and the adverse effects should be related in some way to drug exposure and titrating dose may not be optimal. Either the disease would be too serious to risk a period of undertreatment, and I think the this we just heard about, for example, ADHD, might be a disease where it is not so serious if the patient goes a few weeks with a suboptimally controlled disease whereas there are other diseases where spending even a few weeks at suboptimal control could compromise overall long-term outcome, that the adverse effects are so serious that it is not ethical to risk then and that you are really bound to do whatever you can to adjust the dose as accurately as possible from Day 1 or that the response or the adverse effects are delayed to too difficult to monitor.

Too difficult to monitor, for example, might be something extremely expensive or extremely invasive, Swann-Ganz catheters or some implantable device that just wouldn't be reasonable for following patients long-term or that, really, there is nothing that you can monitor while you are seeing the patient week after week or month after month to give you a clue as to what might be going on with long-term adverse effects.

Again, we have, in cancer, got examples of that that comes from our association between the cumulative incidence of a very late adverse effect, the development of irradiation-induced brain tumors whose onset didn't occur until five years after the start of radiotherapy, so that was over six years after the start of treatment for acute lymphoblastic leukemia that was related to a single genetic polymorphism and a single gene, this TPMP or thiopurine methyltransferase gene.

So, obviously, there is nothing that would could monitor during this period of therapy when patients were receiving their thiopurine daily for two-and-a-half to three years that would give us any clue that the patient would ultimately develop a life-threatening secondary brain tumor. So that is an example of a late effect that we need something earlier to monitor to figure out how to adjust doses.

[Slide.]

In the diseases that we treat at St. Jude, the most common pediatric tumor is acute lymphoblastic leukemia. I think some of the phenotypes that we monitor in this disease are illustrative of how we have to go about monitoring therapy. So ALL is treated with, as I said, two-and-a-half to three years of almost daily chemotherapy with anywhere from five to eight drugs almost all of which cause myelosuppression and those patients are monitored weekly for their blood counts. Myelosuppression is something that we can monitor and sometimes make dose adjustments in therapy to prevent that myelosuppression, at least in the following week or ten days.

Vincristine-induced peripheral neuropathy is another example that has a relatively short onset adverse effect. It is possible to adjust the doses of vincristine to try and avoid that adverse effect as patients are being treated. As we start going out, the onset starts getting longer and the relationship to therapy more complicated. So the use of glucocorticoids like prednisone and dexamethasone have been associated with the development of avascular necrosis but exactly when it happens, what its onset is, what the best way to prevent it is, is not clear so that now we are left, when a patient has symptoms or MRI imaging indicating vascular necrosis, we cut the dose or we stop the dose but we have no idea if it is right thing to do in terms of long-term overall outcome of that disease and cure.

Methotrexate neurotoxicity can be quite delayed. Sterility, long-term obesity might be five, six, ten, fifteen years after the start of therapy. Ultimately, whether the patient is cured is a decision that can't be made until you are at least five years from the diagnosis of the disease and the development of secondary tumors is also one that is three, four, five, six, seven years out.

So monitoring therapy during the period of treatment isn't feasible in this case and having anything to help us adjust doses prospectively would be worthwhile.

[Slide.]

Also, to make the point we are all making, that we recognize this has to be made in the context of other factors that we know affect drug pharmacokinetics and pharmacodynamics. So, as there are some drugs for which renal function might really be the most important determinant of exposure and it is likely that there are not strong polymorphisms, for example, in drug-metabolizing enzymes that could be important but whatever the environmental or nongenetic influence on drug disposition, it does have to all interact with the patient's constitutive genetic state.

[Slide.]

What I have been struck with is the conversations we have just been having is we are focusing where the light is shining. We are focusing on the polymorphisms that we already know are important, like CYP2D6 and TPMT. But I guess I am a strong believer that I do think we will discover additional genetic polymorphisms in the next ten, twenty, thirty years that we currently have no idea are important, so that to make decisions about drug development based on phase I studies doesn't seem to me to be an option. There has to be DNA collection throughout all phases of drug therapy.

I have been told that Dr. Sheiner is someone that likes us to think in a sort of organized way about decision-making so I am trying to use this as a little bit of a platform for what do I want to know, how sure do I need to be and what am I willing to assume as a clinician who wants to have prescribing information for pharmacogenetics.

[Slide.]

I want to know whether specific genetic polymorphisms influence the probability of response or adverse effects. Whereas there can be twin studies or family studies that indicate a genetic component in drug response, I think we are talking about wanted to identify individual genetic defects that may be problematic. So we are talking about specific proposal polymorphisms.

[Slide.]

And we want to have some idea of how the polymorphisms affect drug response, by interfering with protein products involved in absorption, distribution, metabolism, excretion or the response or pharmacodynamics to the drugs. That is because of the point we made earlier that, in order to have an idea of how to put this in the context of drug interactions and diseases, we have to have an idea of what the underlying mechanism is involved.

So if it is a genetic polymorphism and a drug-metabolizing enzyme, then I should have heightened sensitivity to the administration of any other drugs that are substrates for those same enzymes and that providing this information in the context of all that information, the nongenetic information, is important.

[Slide.]

Also, to give me a little bit of information in the labeling about what doses or routes of the drugs were tested when pharmacogenetic information was collected so that, in situations where doses are relatively low or exposures are long, a 24-hour infusion instead of a two-minute I.V. push, the effect of the drug saturating an enzyme or a protein product could be quite different.

So let me understand a little bit about how the studies were done. And the same would be true in terms of predicting how relevant polymorphisms and hepatic metabolism would be helpful to know if there is oral or prolonged exposure versus very short acute exposures.

[Slide.]

What am I willing to assume? We have kind of been talking about this all morning. The in vitro data and preclinical data can be helpful so even if the clinical information isn't strongly supportive of an effect, having the basic information about what enzymes are involved in the metabolism or the handling of a medical is helpful if only for doing things like predicting three and four drug interactions. As we heard about this morning, three and four drugs is a whole lot different than just two drugs interacting, to help the prescribers use the information that we know about the effects of polymorphisms from other drugs on the drug of interest.

Again, by using basic principles of pharmacology, the clinician may be able to make a more sophisticated decision about how to use the medication by providing that information.

[Slide.]

This is what I think was mentioned earlier also, this European group has tried to get together and come up with some dosage recommendations that would be reasonable to put into place now for some drugs that are substrates for 2C9, 2C19 and CYP2D6. They have come up with recommended starting doses for a number of drugs in poor metabolizers and extensive metabolizers, and, in once case, where there were sufficient data, in the ultrarapid metabolizes.

Having this kind of information, again, although the clinician would have to be careful, by knowing about how the medication is handled, how the drug us dosed relative to the concentrations that are likely to saturate these protein products, you might willing to state that an ultrarapid metabolizer receiving another drug in this class might be deserving of a higher dose even though there might not be clinical data specifically testing that drug at those higher doses in those genotypes.

[Slide.]

What do I want to know? I do want to know the frequency of the specific genotypes in at least the three largest ethnic racial groups, understanding that Hispanics are, in many cases, a larger ethnic group but that they are going to be somewhere in between these three groups in terms of allele frequencies, in general.

You basically want to know the frequency of the common homozygous genotype, heterozygotes and those that are homozygous variant or defective.

[Slide.]

Giving allele frequencies is another possibility that I think most clinicians are not really comfortable going through Hardy-Weinberg calculations. So I think clinicians are going to be more comfortable with knowing the frequency of the genotypes rather than allele frequencies.

[Slide.]

We have talked about the difference between phenotype and genotype. While it is true that phenotype is the bottom line, phenotype can be influenced by concurrent drugs, by diet, by disease. This information could be important to put in the label as long as it is clear to the clinician that that is the truth whereas, of course, the patient's germ-line DNA is the patient's germ-line DNA with the possible exception of stem-cell-transplant survivors whose blood DNA is not going to be their germ-line DNA.

It has the advantage that it must only be studied once, although, again, with the caveat that the technology could improve so that genotype might need to be repeated in the future as technology improves. It has already been mentioned that genotype is probably more susceptible to false negatives than phenotype is just by virtue of the fact that probably no genotyping test is going to capture all inactivating alleles or mutations.

[Slide.]

Some concepts about genotyping tests that I think we have to educate ourselves about, that there are multiple types of variant and wild-type alleles for every gene. We have already heard about those. We have already talked about false negatives, and that the number of false negatives really depends on the proportion of the inactivating variants that a genetic test is going to account for.

I think, in my mind, this is going to be the responsibility of the person providing the test results to indication what variants they test for and, given current data, what proportion of inactivating variants their test covers, and that putting that in a label is probably not feasible because that is a piece of biology that is going to change rapidly over time. So I don't think we should hold manufacturers of individual drugs to that standard.

[Slide.]

That patients can be heterozygotes. Clinicians are going to get back results that will indicate more than one mutation in some cases. Again, the better the interpretation of the test, the less information has to go in the label and the less we have to worry about clinicians being able to understand this. I do think, again, this is going to be the responsibility of the people providing genetic tests to say, here is what the raw genetic results are. We are willing to interpret the haplotype likelihoods this way and so there is a 95 percent chance that this result means that the patient is a heterozygote and there is 1 in 100 chance it means that this patient is homozygous-deficient.

[Slide.]

Again, some knowledge of genetics and molecular biology will be helpful as pharmacogenetics gets incorporated into labels. We have heard people debating about the role of assessing heterozygotes but I do think, in most of these cases, is it going to be a reasonable assumption that heterozygote phenotypes are usually in between the two homozygous genotypes and that, although there may not be strong clinical data for that particular drug indicating a different dose is indicated in heterozygotes versus homozygotes of one genotype or the other, given a patient has other concurrent drugs, given a patient might have other altered routes of metabolism or excretion, it is reasonable for the clinician to make some assumptions about heterozygotes and so provide the clinician with that information.

We have already talked a lot about gene duplications and how a gene-duplicated allele along with a heterozygote variant allele could confound interpretation. Again, I would put more of that responsibility on the provider of the test result and not that kind of detailed information being requested in the label, necessarily.

[Slide.]

Again, the more information the clinician has about how they understand how these different mechanisms of genetic variants might affect the expression of a protein product will be better if the clinician understands that a gene deletion obviously means the gene can't be expressed at all. There is no controversy, that an early stop codon means there absolutely can't be any protein, that gene duplication means there might be more active protein and that things like conserved amino-acid substitutions or promoter polymorphisms are likely to have a less significant effect, that will be helpful but, again, interpretation of the genetic results should take care of most of these relatively complicated decisions.

[Slide.]

As I am writing all this down, I am thinking, is this too much to expect of clinicians? It may be, but I do think that there are plenty of examples where we expect a high degree of sophistication in clinicians in being able to prescribe drugs. Now, with the availability of ematinib, the 922 translocation in peripheral blood or bone marrow really needs to be followed to see how it is progressing within a patient.

That can be assessed several different ways, by cytogenetic tests, by FISH, by RT-PCR, and there may be a lot of clinicians who don't understand the subtleties between the way that those tests work. But that doesn't mean that we don't expect them to have some idea of how to follow diseases in these patients.

There are many drugs for which G6PD deficiency is either a warning or a contraindication and most clinicians don't understand how those tests are done. They don't know whether they are phenotype or they are genotype, that we are expecting them to try to get them, to try to utilize them, to try to prevent adverse effects for patients prescribed some of those drugs in some cases.

I noticed in the Hepatic Dosing FDA Guidelines, the Child-Pugh score is used repeatedly to describe how to interpret the liver dysfunction in patients. I would wager to guess that there are many clinicians using drugs for which the Child-Pugh score is described in the label but they don't understand exactly how to calculate that or what those numbers mean, and we can go on and on.

So I do think it is a lot to expect of clinicians but I don't think that that means it shouldn't be done.

[Slide.]

What else do I want to know? I don't want to know a lot of the details about phenotype but at least tell me whether it is a blood test or a urine test, give me a little idea of the direction of the phenotype, so that could be AUC, that could be enzyme activity, and how, at least, directionally, it relates to the genotype and give me some idea of what interferes with the phenotyping test so I know whether it is reasonable to try on the patient.

For genotyping, we have already mentioned at least an idea of the number of inactivating variants, their approximate frequencies and it would be helpful--again, this could be provided by the person providing the genotyping test to understand what proportion of inactivating variants their genotyping test accounts for in at least the major racial ethnic groups.

[Slide.]

Also, that negative results can be very helpful, so just understanding that a drug has been tested to see whether it is a substrate for different genetically regulated polymorphism gene products and knowing that it is negative may be helpful and that that information should be included where possible.

[Slide.]

How sure do I need to be? I think it is helpful to just provide examples of real data and I guess I would prefer that we leave the option somewhat open as to exactly what kinds of data are presented. Knowing the average or standard deviation or the median plus-or-minus the confidence interval for the dose in three genotypes, homozygote, wild-type, heterozygote and homozygous variant at some specific doses.

For example, given here are some doses. That can be helpful. Understanding the frequency of a serious toxicity like QT widening along with confidence intervals in patients of different genotypes. Given a dose, a fixed dose, what proportion of patients displayed evidence of response or what proportion of patients displayed evidence of toxicity?

[Slide.]

The literature is filled with these kinds of examples that I think would be helpful in the labeling.

[Slide.]

This is an example of the frequency of the median and confidence intervals for severity of mucositis in patients who are homozygous CC, heterozygote or homozygous TT for an enzyme involved in folate metabolism who are given methotrexate as transplant preparative regime.

[Slide.]

This is an example of the warfarin milligram-per-day dose in patients who were titrated to achieve a target INR. One can see the degree of overlap among the genotpyes, see that there is overlap but that there will be differences in the median and range of doses tolerated by patients in those various genotypes.

This is the proportion of patients cured based on their 2C19 genotype in the wild type versus heterozygote versus homozygous variant genotypes treated with a standard dose of omeprazole.

[Slide.]

So our favorite gene polymorphism TPMT; this shows the difference in enzyme activity of frequency distribution and the mean tolerated weekly dose of 6 mercaptopurine in the 1 percent of patients who are homozygote mutant, the 10 percent who are variant heterozygote and the 90 percent who are homozygous wild-type, the same polymorphism, the cumulative incidence of requiring a dosage decrease based on myelosuppression in the homozygous variant, heterozygote and wild-type patients along with confidence intervals for that cumulative incidence.

I think any of that kind of information is information that clinicians can interpret if they want to understand how to best prescribe medications in their patients.

[Slide.]

In terms of the labeling sections that may be relevant for clinicians, I think that we have heard about most of these today, that the Clinical Pharmacology Section is very important to provide general background information, a little bit of information about what doses of drug were used, what concentrations were used in in vitro studies, a little bit about how the studies were done, where relevant to put information in the Warnings, Precautions and Adverse Reactions and Overdosage Section, and to provide some information on dosage and administration, especially given that what is right there right now includes information on dosage adjustments, given degrees of renal dysfunction and hepatic dysfunction which often have far less ability to discriminate doses that have been true for many pharmacogenetic polymorphisms that have been associated with different doses and adding information on what has been observed in different genotypes for dose of a drug, I think, in that section is important.

[Slide.]

The other principle that I think has been illustrated already in a couple of labels that have been approved by the FDA, having just cross references among sections I think is a good idea. So, if there is something about genetic polymorphisms in clinical pharmacology, it can state, "Please see the Adverse Reaction Section for additional information on dosing of these drugs."

[Slide.]

A couple of miscellaneous items in terms of terminology that I think should be considered, that it should be allowed to use colloquial terms where it is relevant and people may be familiar, so extensive and poor metabolizers, fast and slow acetylators, that is fine to use if they are already out there in the medical literature. To try to avoid the word "mutant" if possible. Most people don't like to hear themselves referred to as mutants-- although I don't mind at all, and I am homozygous variant for all kinds of things--that the terms variant and defective are more neutral and probably descriptive.

Avoid the word "normal" if possible and use wild-type or describe what the effect is on the phenotype, high activity, normal expression. All of these star HUGO nomenclature designations that those of us in the field throw around are not going to be very interesting to most clinicians.

If they can be easily mapped to the wild-type, common or variant-defective allele in the label or at least in the genetic test, I think that will be helpful to prescribers, but we are going to have to deal with the fact that this is confusing, that there may be several HUGO designations for a wild-type allele and lots for defective or variant alleles.

[Slide.]

I apologize because I have still got a couple of typos in here. In terms of a decision-tree, it is really nothing other than what we have been talking about. If the ability to titrate the dose intraindividually is apparent without compromising the patient, that, just based on response, then I don't think we really have to look very much further on how to prescribe the drug intelligently.

But, if not, and if the drug is complicated by late effects or invasive monitoring or, as I mentioned, very serious diseases where under or overtreatment is not an option, then are there other simple lab tests that can be used or, like Bill Evans used to say, "If you can use shoe size, you use shoe size." You use what works.

If that is not an option, and pharmacogenetic tests are available and an option, then, yes; they should be used and I don't really think we have to decide on phenotype versus genotype. I think both kinds of information should be provided to prescribers.

So I will stop there and be happy to take questions.

DR. VENITZ: Thank you, Mary.

Any questions for Mary? Let's get started on the discussion,

Committee Discussion

DR. VENITZ: You all have the questions that Larry and the FDA are asking us; are the approaches presented to study the influence of pharmacogenetics on exposure response sufficient and appropriate and a follow up question, are there any other criteria or approaches that FDA should consider recommending to sponsors?

So I will open the floor for general discussion as well as any questions that you might have for Mary's presentation.

DR. SHEINER: Mary, can you just flash up the one slide again? Is that possible, or is it gone? Has it disappeared? I think it was maybe the first or second one.

That's it. I just wanted to say, Mary, that I knew that anti-cancer drugs were dangerous but I didn't know that you could get more than 100 percent toxicity at a high dose.

DR. McCLEOD: It is more than one toxicity.

DR. SHEINER: I wanted to say I am pleased that you used the three questions that I have asked, but that was actually more in the line of, if you are going to do an investigation, sort of a learning study or a confirming study, because when you get to decisions things get a bit more complicated and you need utility functions and stuff like that, sort of like that, sort of that how certain you need to be becomes what is it worth to you. So life gets a lot more complicated.

But I did want to say I really like the way you sort of put it all together there. The problem is you had an awful lot of, "What do I want to know?" We have got to do some kind of distillation. Maybe some people can handle it and, as you say, the expectation is the people in the field taking care of people will have to be able to respond to these things, but we have got to distill it down. That was a lot of, "What do you want to know?"

And you went so far as saying, "And I want to see some real data." I know that is you and I know that is us, but it is a big demand and, you make the label huge that way, you may find that you get an unintended consequence which is nobody pays attention to it, which is already a problem with labels. They go on and on.

The other point I just wanted to make about your last slide when you said, if you can titrate, then maybe you just should titrate. One of the things we shouldn't lose sight of is, even though therapeutic drug monitoring is not as good as effect, if the issue is a pharmacokinetic change, and if you are worried about this drug is inducing and that drug is blocking, and so on, in the end it comes down to what is the drug level. It may be the easiest thing to do is just to find out and not have to worry about all those details.

DR. RELLING: Yes. I put drug levels as phenotype.

DR. SHEK: Again, looking at this decision tree and looking at the adjusted dose, with regard to practicality, I would assume, for injectables it might be easy but how practical is it with the dosage that industry is putting that you can adjust the dose. Do we have also to look at more flexibility there which might have its own economic impact where you have more variability in the dosage, strains that a supplier, a manufacturer, will come up with to enable you to do it.

Although maybe drugs, they wouldn't have the toxicity, but I believe that side effects might have an impact on compliance and might have an impact on efficacy, people are not complying because of the side effects. I don't know how much flexibility is there, whether the industry has to respond and come up with more dosage flexibilities.

DR. RELLING: As somebody in pediatrics, we deal with this all the time. I mean, we just have to come up with different dosages based on formulations. But there are examples where I am sure there is pressure on the industry to come up with more formulations.

Again, I don't think we can let the fact that different doses may be required in different patients be the reason not to have individualized doses. We have got to figure out a way to do it.

DR. HUANG: Going back to your decision tree, and your question whether we can titrate to the response and, if so, then you adjust dose accordingly. All the examples that you have shown, which one do you think the clinician will not answer, "I can adjust the dose?" For example, some of the warfarin and this whole list of tricyclics where the physician was saying, "No; I cannot adjust and I am going to go to the left," or the majority would go to the right, where they say, "We could adjust according to the response."

DR. RELLING: I will let somebody else handle tricyclics. My impression is that under-treatment of psychiatric disorders is a major problem, the fact that there is this assumption that there is a huge proportion of the population that just intrinsically don't respond and nobody knows why and it is only a trial-and-error period of six to eight weeks. I think that causes unbelievable morbidity in this country right now.

There might be a lot clinicians who say they can do that, but having better information about how to come up with a good starting dose, I would think would be critical in that area. Almost every anti-cancer drug is a drug that can't be titrated based on response accurately or reliably. I don't know--let everybody else put in their favorite compounds. I guess there are others where it is not problematic. Insulin is one where you can titrate to response.

DR. McCLEOD: I think the warfarin example is a good example of why you can titrate to response but it is not good enough. The cohort data that David Veenstra and others have published identify that the people with the homozygous variant genotype were able to be titrated to a good INR. It took an average of 94 days and we all know, in the area, that is the first 70 days that are most critical for preventing clot post arthritic--or hip replacement or in the case of atrial fibrillation.

So it can be done, just not in a timely enough manner to prevent some events. How many of those events is arguable. In other situations, it is not as big a deal. If you have a mild rheumatoid arthritis and you want to get the methotrexate dose right, you have a few weeks to get it wrong. It is inconvenient and patients don't like it, but it is not life-threatening or associated with high morbidity.

So I think maybe that decision tree needs to go how soon you need to get it right because, if you need to get it right quickly, then it may be that a lab test will be more appropriate and can be done, as Rick mentioned, before you ever give the drug as opposed to having to wait and respond.

DR. SADEE: I think, looking at all the data, there is a fundamental problem in that we have a few polymorphisms in the P450s that are clear. They abrogate the function of a protein and that is useful in a fraction of the patients. But then there is additional variation that is really very, very large. So you cannot say we cannot titrate the dose on the basis of genotypic information because it may only take care of a very small fraction of the problem.

Maybe it is useful to just think about the fact that the cytochromes that are highly polymorphic are a very unusual example in that it hardly ever happens in any other gene that nonmutations, mutations such as abrogate the function altogether of a protein, accumulate to such high levels, let's say 30, 40 percent of allele frequencies in some cases. So that is a very unusual situation.

If you do a genomewide study and those studies have been published now, then polymorphisms in promoter regions, polymorphisms affect the stability of mRNA processing, splicing, et cetera, are probably five times as prevalent or maybe even ten times as prevalent as those that affect protein function. That is where most of the research has been going.

So I agree with Mary's statement that anything we do should make sure that the polymorphisms that we put into any labeling are seen as just maybe the ones that we know right now, that there is room for additional polymorphisms that can be 100 KB upstream of a gene, nobody has ever looked at it and they are extremely important, could affect the expression tenfold, easily.

So these polymorphisms may appear over the next few years. So whatever we do needs to be predicated by the sense that we actually only know a very small portion. Lew, you said we have to distill it down, and that is correct. But we can only distill down if we know from where we are distilling down.

And I think we are still, even in the cases of 2D6, quite a way from knowing all the important variations that occur, not even to talk about epistasis, compart heterozygosity, haplotype information, you name it. All those are complicating factors that you definitely want to touch after having distilled down, but you have to know it ahead of time. There is no good method to determine whether two functionally important polymorphisms are on the same allele or on the opposite allele except for maybe the methods you are using now, and they work pretty well. But nobody is using it.

So there is a lot of uncertainty and that is the difficulty of what we are dealing with, to distill down from an entirely incomplete piece of information to something that then is supposed to educate us how to use dosages. That will be rather difficult.

DR. VENITZ: Let me add something and maybe reiterate something that Lew had mentioned early on and this goes back to my favorite utility function implicit in both of your presentations. If I look at Strattera, the reason why you ultimately didn't care about the phenotype is because you were worried about insomnia. The reason why Mary cares about it is because her toxicities are life-threatening, at least potentially.

If you had to pick the perfect pharmacogenetic test or the perfect scenario where it might be useful, you want to pick something where the stakes are very high. Either the stakes may be the consequences of toxicity or the consequences of lack of efficacy. That is why I think oncology is a perfect area for that because the stakes are very high.

A lot of other diseases or indications you may find, yes, there are relevant genotypic differences that are reflected and genotypic differences that you can measure in terms of exposure of response, but the consequences, clinically speaking, are insignificant. Those are the ones where there is very little at stake and it is very difficult to convince practitioners that are already having a tough time translating all the nice research that we are doing into practice. It is very difficult for us to convince them to actually change anything.

So, in terms of strategic planning on the FDA side, I would focus on the scenarios where there is lots at stake as opposed to picking the ones we know a lot about but clinically the relevance is limited at best.

DR. SHEINER: There is also an interesting signal-noise issue. It is right that we focus on the poster children, the big effects and so on, sort of to get people's consciousness up about what is going on. But I remember, and maybe I am showing my age too much here--I remember 30 years ago digoxin. This was the classic drug to adjust for renal function.

Yet, when we looked at a huge number of patients receiving digoxin, they were mostly old and their creatinines were around 2 because they were old and their kidneys were not working as well as young people. But the number of people with renal disease in a random population in a hospital ward was rather small and, if you just did the sort of standard statistical test and asked, did renal function help when you put it into the regression. Very little information about how you ought to dose that drug in practice was conferred by knowing the creatinine.

Now, that is not true. The person with a creatinine of 10, obviously, you learned a lot. But they were very, very rare so you couldn't get it to show up. Now, does that mean we should be sort of segregating out the outliers and saying, "But that is who we really care about," or does it mean we really want to talk about average behavior.

These are all issues that don't really come up as scientists because you are trying to push the knowledge forward but do come up very much in a regulatory agency. How much do you hold people's feet to the fire? How much do you put in the label? How much do you prevent things from happening. It is very tough to answer because, again, you need population data. You need somebody is going to be realistic about the way they evaluate it rather than somebody who has got a flag to wave or an ax to grind.

It is just starting here. You found a couple--I think it was brilliant of Larry to limit this discussion to metabolic enzymes of a certain type. But this is opening a Pandora's box of thousands of possible genetic variants and their implications for pharmacodynamics, pharmacokinetics, lord knows what else and just the thought of how you were going to deal with in some way in which you do pay attention to the important ones and not to the unimportant ones is really almost daunting.

DR. SADEE: I think focussing on the ones where it really makes a difference, the dosing, and you mentioned tricyclics, but also the treatment of psychosis, of first-case psychosis, is a real problem because, if it is not treated properly, it may cause damage for the rest of the life of that particular patient.

And yet you do not know whether a drug is effective until maybe six or eight weeks. At least that is the conventional wisdom. There may be better techniques. So if you underdose because there is a high metabolizer, for instance, you wouldn't know about this and these patients would be damaged for life. So I think that is another situation where it is extremely critical to get the dosage right.

With respect to all these multiple polymorphisms and unknown factors that we are talking about, I think we must be aware of also the increasing knowledge about epigenetic changes and accommodate of the modeling which is exploding into our face. There may be absolutely no polymorphism and it still may be epigenetic; that is to say, there is a stable genetic change in the gene that you may not see by the normal genotyping where the gene is silenced or where the comatin is remodeled.

That appears to be malleable even though it was thought to be once a gene is silenced, that will be for life but it can be reversed. So these are epigenetic changes we do not even touch upon and they may be also huge in their effect.

DR. LESKO: Just to elaborate a little bit on "the stakes are high issue." The current let's call it "model" in drug development is to look at covariates that affect pharmacokinetics early on and then react to that in one form or another in terms of drug dosing.

I guess I am trying to get to maybe a better understanding of why the issue of stakes are high would be any different in a genetic or genotype-defined population than the stakes are high for any drug in which we study routinely hepatic disease, renal disease, and so on.

I understand it is only interpretation but how does it differ as a cofactor that might become something that is a routine factor to study in drug development with the decision about what to do about it later on. It almost sounds like, "I don't want to study this cofactor unless the stakes are really high."

But it is part of understanding the basic informational content of the clinical pharmacology of the drug, so I sort of want to pursue that thinking a little bit.

DR. VENITZ: Fundamentally, I don't think it makes any difference. What I was referring to is how you can translate that into actually changing the practice. If you pick the low-hanging fruits, you have a better chance of convincing people that this is actually important. Otherwise, we are going to swamp labels with pharmacogenetic information that, in reality, is not going to be used.

You heard what Lew was saying about distilling information. I am talking about information may not even be relevant so distilling it to the point that it doesn't even appear on the label.

DR. SHEINER: I think it is very different. The difference is that how many drug-eliminating organs are there. There is the kidney, the lungs and the liver and that is about it. So there are only a few things you need to look at. We lump all hepatic diseases together. Maybe we shouldn't, but we do.

So it was doable. We are now entering a realm where the number of possible things you could have to look at just keeps on multiplying. Not only does that produce terrible problems in false positives and the ability to extract from 100-people's worth of data when you have got a thousand covariates which one makes a difference. We have got to be much more intelligent about this.

A drug that is excreted unchanged, you have got to look at the kidneys. Basically, it is how well do they work. It doesn't matter what disease has caused them to not work so well. It is a doable containable problem. This is not. So that is the difference. There is no conceptual difference but the difference is we are in a very different universe. We are in a thousand-dimensional universe. And everything changes.

DR. LESKO: So what do we do about it. It gets to another question I was thinking about and it is that when studies like this would be conducted, they might be conducted, for example, in a phase I healthy volunteer population. Typically, that information, whether it is drug interactions or anything else is extrapolated to many other populations for the purposes of adjusting doses without a lot of consideration of issues other than the differences in exposure.

So I guess what I was wondering is, as we have gone through, actually two days and maybe, Greg, you touched upon this is if you had a genotypic difference demonstrated in a test population which would be a phase I healthy-volunteer population, as you extrapolate that knowledge to other populations, where would it become more important or less important?

For example, in the elderly where you have maybe in an extensive metabolizing group slower metabolism so the differences become closer, genotype doesn't make much difference. In young kids, maybe the development process doesn't make much difference.

Does genotype interact with other covariates that are out there in that little circle that Mary showed? What do we know about those sorts of issues?

DR. RELLING: I think that the Strattera example is interesting. I would like to know a lot more information. Where you see this incredible bimodal distribution, in some estimate of I think it was a parent oral clearance, I don't know where those doses that were tested were relative to the doses that were actually used in the chronic dosing over weeks that you showed us where it ended up that there was no difference in the delivered dose.

I don't know what was titrated week by week in order to decide whether to go up or down on the dose. But, yes; presumably those other nongenetic factors--it may just include simple things like what dose you are working at, which is the other way of saying what AUC are you working at, which is kind of what Larry just said. If you are very, very old and everybody has lousy clearance, you may wipe out the importance of a polymorphism. If you are very, very young and everybody has beautiful clearance, you may wipe out the effect of a polymorphism.

So that is why I am afraid, even though I know that that was a lot of slides of information that I want, I think to really use the information smartly, you are going to need to have a fair amount of information and you are going to need to assume a pretty high level of functioning about understanding of pharmacology and pharmacokinetics to use the information optimally.

Anything you do to make it real simple so it looks like the package inserts we have now is going to wipe out so much of the complexity that really helps clarify the information that it will be misleading. I have thought about why does the label have to be manageable? Nobody reads it anyway. It is all on the Web. Why not make it huge. Why not make it fully referenced, fully graphicized? Put a lot of information there. Make it the world's best review article on the drug.

Now that is all electronic, what does it matter how big it is? And then put in everything that affects it including drugs and age and renal function and liver function and put in tons of information.

DR. VENITZ: Any other comments or recommendations? Larry, do you want to wrap things up?

Concluding Remarks

DR. LESKO: I think we are getting near the end and getting pretty tired. I think we have been overwhelmed by information from the last two days and it has been extremely valuable to us to get the comments and input that we have.

As usual, we have to distill a lot of what we heard over the last two days and try to take each of the four, five different projects we brought to this committee and move them forward to the next level.

I guess I will just close by expressing my thanks and appreciation to the committee for their input into the topics. I would express thanks to the guest presenters that we had. They added a lot to the meeting. Appreciate that. And thanks to all of the FDA presenters that were able to put on the presentations during the course of the last two days.

So, as always, it has been a very good experience and a learning experience. Thank you.

DR. VENITZ: Let me add my thanks to the invited guests for coming that far, to the committee members for freeing their time and for the FDA staff for organizing it. Let's adjourn the meeting. Have a safe trip home.

[Whereupon, at 12:00 p.m., the meeting was adjourned.]

- - -