move on to discuss that same question with ES cells.

CHAIR URBA: Go ahead and address that.

DR. CHEN: Okay. I don't know bone marrow transplantation, but I know that, for rodent, it's feasible that you'll be able to follow life span. You can follow for a year. That's not difficult. But for large animals, that will become difficult. I don't exactly know what the panel suggests.

DR. CHIEN: One issue there I think that is a point here is that we're not asking for an endogenous cell in the recipient to kind of go rogue, and become cancer. We're putting cells in there where we know their doubling rate, where we know these cells can form tumors. As I said before, that's the gold standard to know that you've got an ES cell, right? All that data exists.

We make teratomas all the time. All the IPS, you know, all those papers that came out saying the skin cell became an ES cell, one of the key criteria was is that they formed teratomas. That data is out there, okay, and it's weeks. We do it routinely in my lab. I mean, you can get, and that's where the data and the rodents will match what you would get in the larger animals as long as you give a larger cell load.

I want to make one other point about the large animal, and this, again, will belie my cardiological background. But, you know, there are going to have to be devices and other things to get the cells where you're going to get. They're not going to magically end up in your favorite organ system. Someone's going to have to devise devices to get them there.

This gets around. We haven't gotten to the imaging. I don't know if we're going there next. And those devices are going to have to be optimized for large animals. Let me put it this way, if we tried to optimize anything we do in cardiology in a cath lab in a mouse, we wouldn't get too far. Okay? There isn't one thing there that may be -- you know, the mouse has been very good to me, so I hate to dis it. But I don't think it's going to get us there for optimization of delivery strategies for cells. And so that is another thing that you can put in there.

So I just think there's no -- you can run, but you can't hide. You've got to get the safety in a large animal.

CHAIR URBA: Dr. Taylor, you had a comment about duration?

DR. TAYLOR: I do have a comment about duration, and it actually, directly, is relevant to what Ken said.

Ken, we've discussed delivery and the fact that you need large animals for years now, and that's the actual gold standard now if you're going to use a catheter, that you have to use something on the size of a human to be able to deliver those cells.

I guess I will go back to preclinical studies should be clinically relevant, and there are also data out there from gene therapy studies, and those are probably the most directly relevant, in my mind, in this regard with regard to tumorigenicity, and with regard to duration of monitoring.

The clinical monitoring for the original gene therapy trials was very long. It was, I believe, on the order of ten years, and that was actually somewhat problematic for the sponsors, but identified some problems that emerged that have turned out to be clinically relevant.

So I would argue that we go back to what the preclinical requirements are for gene therapy studies, and learn from those, and say, have they accurately predicted where we need to go, and if so, maybe just with, you know, minimal effort, change the word gene to cell, and think about that.

DR. SALOMON: I mean, I would have said something similar in that we've kind of already solved this for gene therapy, and I think the FDA should probably use a similar guidance. So at the time, we took different classes of vectors, and evaluated their risk, and pegged the long-term followup to that. And I suggested something similar could be done here, as well.

So we looked at, for example, retroviral, and other integrating lentiviral systems, and said, that should be lifetime followup. Whereas, an adeno-associated virus, for example, that exists -- some argument there, but exists primarily as an episome, could be very short term, like a year or so.

The other principle that's important here is that nobody suggested that sponsors withhold moving forward into clinical trials while you followed large animals for a year, or two years, or in rhesus macaque, it could be a 25 year model if you wanted to follow them for life, but rather that there were stipulations on the sponsors that they had to follow their patients, and/or their large animal models, in parallel to moving forward at the appropriate time of the trial.

DR. SNYDER: The only difference, as Ken brought out before, and I think the gene therapy field is very informative for us, is that those are still endogenous cells, whereas we're talking about implanting exogenous cells that often can be marked, and traced, and tracked, and characterized ex vivo before putting them in. So gene therapy worked on endogenous cells is the point that Ken made before.

DR. SALOMON: Just as a followup, Evan, then we came back and did the exact same thing with xenotransplantation.

DR. TAYLOR: And I would argue that it's all related to, as I said earlier, the original product, and how well characterized it is prior to going in. You've either solved it with non-ES cells, if it's a completely benign product that we can reproducibly create that's differentiated, or we've begun to address it with gene therapy and xeno if it's not.

DR. GERSON: I would just comment that I'm not that impressed with extended long-term followup in the animal settings. I am very impressed with the need for extended long-term followup in clinical settings. So I'm just -- the rodent, life span of the rodent is probably sort of okay if we have to. But I think some reasonable bound, I suspect less than a year for any of these animal studies, is probably quite adequate, but not predictive in the human.

CHAIR URBA: Dr. Chen, do you have other points you want us to consider before I give Dr. Goldman the last word on this question?

DR. CHEN: I think the last point is the dose.

CHAIR URBA: Does anybody want to comment? Have we heard enough about whether we're looking for percentages, or numbers of specifics?

DR. WOO: So this morning we heard five percent, which may be equivalent to 100,000. And then when the question was raised whether it's the percent or the absolute number that's important, the answer is maybe it's the number, the absolute number of cells, which that makes sense. But then they also published reports in the literature that it is not really 100,000 cells. As little as 1,000 cells will give you tumors in skin of the mouse that has been previously transplanted with fetal human tissue. Whether a human embryonic stem cell has been injected into it, this is a paper that was shared with me by Dr. Gunter over here.

So these numbers are falling all over the map, and we've got to get a much better handle in terms of how many cells are actually tolerable in an appropriate animal model for toxicity. Until we can demonstrate that, I think to talk about, well, maybe 95 percent or 99 percent is good enough, I think is not scientific, and clinically irrelevant.

DR. FRIEDLANDER: You know, I think coming back to this issue of numbers of cells versus percentage, I'm not disagreeing, these are very important, critical issues, but I think it's very model and disease appropriate. So if you're squirting, you know, cells into a concentrated area in a localized area of the brain or the eye, that's very different than administering them systemically where they pass through lungs, livers, et cetera. So I just think these are considerations we need to think about.

DR. WOO: Excuse me. Could I follow this up?

Then perhaps for that particular disease, the toxicity ought to be determined by the same route of administration, same route of administration. This is the same as in any other therapeutics, including the drugs, or monoclonal antibodies, or gene vectors.

DR. TAYLOR: I completely agree, Savio. I think earlier I said, does route matter more than site? We don't really know, and I think we need to answer that. But again, it comes back to what the product is and how you've defined -- I mean, if the product release criteria are, we have a 99.99 percent pure product, that's going to be very different than if the release criteria are, we have 20 percent undifferentiated cells, and 80 percent terminally differentiated cells. And we really have to tie this closely, I think, to the release criteria of the given product. Otherwise, it's pointless, and not going to be predictive in any way.

DR. SNYDER: Just one point this brings up, and I actually don't know the answer to this, and maybe this is not even the right place to discuss this.

Would we all feel more comfortable if the cells had a suicide gene in them that would cover a lot of this messiness? So I don't know the answer, but if we knew that what we were going to do was somehow reversible, or we could do a mid course correction, I think a lot of these -- I know I personally would breathe a lot easier about moving ahead. So that, probably, is a topic we need to discuss.

DR. GERSON: I'm going to go back to the comment I made earlier about teratomas. So I've got two comments about dose.

I think the dose of undifferentiated ES cells in the clinical setting ought to be no more than the absolute number of undifferentiated embryonic stem cells that did not form a teratoma in the appropriate animal model, and not scaled by animal to human. So in my mind, that number is sort of close to zero.

In terms of the cell dose of the differentiated cell product used for therapeutic settings in the phase one plus studies, I think one just has to be careful in the dose escalation strategy and be, for instance, much more conservative than we have with the adult cell dose escalation strategies that have been undertaken, because we really are in unchartered water.

DR. FRIEDLANDER: I was just going to come back to your point, Dr. Taylor, about, you know, a small percentage of a very large number is still enough cells to cause problems. So that's my only issue.

DR. MCDONALD: Yes. I just wanted to emphasize, I mean, the question is so complex. Get back to the dosing issue. It's related to where, how, when, and in what format. I mean, it's almost irrelevant to talk about numbers, with the exception of agreeing that there be zero undifferentiated ES cells in any human form of transplantation. But, you know, teratomas are much more likely to form when cells are put in groups than if they're not. You know, dosing is related to number of sites, number of cells, how they survive, how many more times they'll divide. I think it's impossible, almost, to give some answer in between other than an absolute answer. Like I see no reason why, under any circumstances, we would administer an undifferentiated embryonic stem cell for any human treatment.

DR. TAYLOR: I would come back to what I tried to say earlier, and obviously didn't say very well. We know that there are certain sites where tumorigenicity is more likely. And the preclinical data suggests that, and I think it's important to develop assays where we have to test a given cell product in those environments. And if there's tumorigenicity, then that product is not releaseable.

So we can define what those are, because it really doesn't matter if you're trying to treat the brain if you don't know all the cells go to the brain, or all the cells stay in the brain, and the same product will create a tumor in the pancreas, say, because that's one of the vulnerable sites for that given cell type.

So I think we need to create some assays that are predictive of tumorigenicity for undifferentiated cells, require that the human product be tested in rodent animal models who have those assays, and base release criteria on whether or not there's tumorigenicity, and tumorigenicity is not an acceptable outcome, period.

DR. BAUER: I just wanted to make a point about the earlier discussion about comparisons with gene therapy, and I think the idea of tying long-term followup to risk was a very nice paradigm that came out of there. But I think, with embryonic stem cells, for us to be able to decide what the risk is of those, perhaps rare cells, and how that interacts with long-term follow-up to classify at this point is a little bit of a challenge, just because of the biological complexity. So I just wanted to raise that as a point.

DR. SALOMON: Can I respond to that?

CHAIR URBA: Go ahead. And then we're going to have to finish up with a couple of things. Otherwise, we won't get to Questions 2 and 3.

DR. SALOMON: You know, I thought about that, too. I wrote to some notes to myself about, as an overall suggestion, to stratify by risk. I think in a guidance document, that's still going to be important. I don't think it's going to be for you, for stem cells, quite as easy as it was for us to say, this is an integrating vector, it's high risk; this is a episomal vector, it's not. So that point's well taken. I thought I said that.

But I do think that you can stratify risk. And I won't go on and on, but just examples would be the cell type. I mean, I think that there's one thing making a neuronal stem cell, and another thing making an insulin progenitor, and another thing making a myoblast.

The type of manipulations that are required, you know, you can really show a successful terminal differentiation of a cell that's very safe, and I think that you could call a lower risk. And you could begin to stratify risks here, whereas someone else might come along with an early process human embryonic stem cell, and I think the risk would clearly be higher.

So I'll stop there just in the interest of time. But that's the concept, I think, in a guidance document, would be to demand sponsors to profile the risk, and then evaluate them, and the stringency of what you demand based on that.

CHAIR URBA: So we'll run the table. Dr. Chien, Weir, Gunter, and then back to Dr. Goldman, and then onto Question 2.

DR. CHIEN: We've got a hot topic here.

One thing, and this would be suggestion, obviously. Nobody knows the answer to these things. But it seems to me, if I was doing these experiments, and let's say I was trying to move something forward, was excited about it, but was concerned about the safety, I think the safety issues here are actually twofold, because no one is going to propose to use human ES cells directly into a patient. No one is going to do that. They're all going to be progeny of human ES cells somewhere down the line at various stages. So there's two components.

One is, what are the residual

ES-like cells, or ES cells that are in the preparation that could become a teratoma? And for that, you have to have larger amounts of the starting material, because you want to start to see if any of them are in there. And that should almost be standardized.

Okay, I like the standardization issue. Because there's also going to be lot-to-lot variation within the same company, or whatever. You use the same protocol. This is probably the most complex biological therapeutic humanly imaginable. The chances that one lot on one week is going to be the same as the lot next week, and you have to have an assay to make sure that the thing that you did last week was the same thing you're doing this week, because it could go off, or something else could be slightly different, the culture conditions. So I think it's an internal control, as well. So for that component.

And the other one is, is the cell that represents the predominant cell in the population, the one that you're hoping has a therapeutic effect, can it go rogue? Okay? And for that, you need a different set of assays, and you're looking for a different kind of tumor, or whatever, a different kind of outcome. It could be an ectopic differentiated phenotype. But I think, again, that would be a different kind of scenario.

But I think the standardization for teratoma and teratocarcinoma for residually S-cells, which most of my comments were addressed to, should be standardized, and it should be nailed. And I'm almost sure that, if somebody looked carefully, they could find the prostate-specific antigen PSA for teratoma, or something like it. I mean, I think it's out there, somebody could start a company on it. It's a free idea. Go for it.

I think that's what we need. We need, like, hERG screening for this, and I bet you somebody can come up with it.

DR. WEIR: Well, I think so much has already been said, but I just want to make the point about generalized versus specific approaches, because we're talking about generalized approaches, how many cells can form a teratoma, what kind of tissue is teratomagenic, et cetera.

But I just want to make the point of how important I think it is for each specific disease to be really nailed. So if you're going to do diabetes, you've got to figure out all the animal models that would be pertinent to your goal in diabetes, and assess the tumor risk in those models. And if you're going to do Parkinson's, you know, that's a whole other set of things.

And every company that has a product, you've got, because the cell lines are different, you've got go to through the same thing for every different product. So I just want to emphasize the need for generalized knowledge, but also these very specific approaches for each disease.

CHAIR URBA: Dr. Gunter?

DR. GUNTER: Well, I think my point's already been made, actually, but I'll just echo the need to have a standardized assay for assessing tumorigenicity. And if we had that, we would likely understand sensitivity of the assay, hopefully, also, hopefully, robust us in other parameters. But then we'd feel much more comfortable, if we knew the sensitivity of the assay, about how we dosed. So that's important.

Another point I'd like to make is that it's clear, from my reading of the literature, there's a lot of interactions of ES cells with other cell types, and so the relative proportion of undifferentiated ES cells to the differentiated cells should not vary from what's anticipated to be used in the clinic. That could have significant effects.

Finally, in the background information, there was a nice paper in Cytotherapy by Lawrence that I thought, of all the papers I read for this meeting, it had the best potential assay for measuring tumorigenicity. So I want to thank the FDA for providing it to us, and suggest we look at it further.

Thank you.

CHAIR URBA: We're starting to hear some of the same things, but, Dr. Goldman, I'm not sure it lends itself to a nice summary of what we recommend, but you may have the floor for a few last words.

DR. GOLDMAN: That's a dubious distinction. So before any closing words, I wanted to respond to Dr. Chien and Dr. Chen's comments before, which I think cut to the heart of what constitutes tumorigenicity, or at least what our concerns should be with regards to tumorigenicity. Most of the comments of the last eight or ten folks have been directed at ES-derived tumorigenicity.

I would like to get back to the point that the differentiated, or semi differentiated derivatives of ES can, of course, generate their own lineage-specified tumors, and I actually worry about that more as a long term issue, getting back to Dr. Chien's point of how long, both preclinically in animal models, and, for that matter, patients, should be followed, and then think back to Melissa Carpenter's points before with regards to the relatively high incidence, at least I would take it, of aneuploidy within these populations.

So in some cases aneuploid cells, of course, are potentially tumorigenic in and of themselves, depending upon what deletions or rearrangements are in place, but I look at the greater problem as being, as these cells then differentiate, depending upon the site of deletion or rearrangement, we're going to run into differentiation-specific or lineage-specific abnormalities, both in terms of differentiation in competence, but also in terms of tumorigenicity. There are some deletions that will specifically lend themselves to a generation of gliomas, or neurofibromas, these are well studied in the literature, that will not will not lend themselves to tumorigenicity of more upstream phenotypes.

And of course, that, I think raises the question that, with human ES derivatives, which may take very long periods of time after transplantation to achieve final mature-differentiated state, you may see the late appearance of tumors that are specific to organ, that are specific to given site or phenotype, which may, in and of itself, indicate the need for very long term follow-up, both in animal models and people.

I can tell you from some of our own experiments with ES and tissue-derived glioprogenitors we have to often wait 13, 14, 15 months before seeing the terminal maturation of some of the cells implanted.

Now, we're typically using very early passage ES that we've established as karyotypically normal up front. We're more typically tissue derived, but I can imagine that if we're using more highly passaged lines where aneuploid is an issue, by the time we're out again 12, 13, 14, 15 months we might very well start to see the generation of tumor specific to the lineage that is being generated at that point in time.

So I would advocate, just on that basis, very long term followup. That's not to say that that need interfere with progression to early clinical trials. Again, it's a question of cost benefit, risk benefit as always, but I think we have to be very aware of the possibility not only of teratoma generation, or teratocarcinoma generation from ES, but also of organ- and cell-type specific tumors that may be restricted and much later in appearance.

As far as summarizing this discussion, it does seem that there is consensus, can I say that, on the -- not only the lack of enthusiasm for implanting known, undifferentiated ES under any circumstance, but perhaps a consensus that all efforts should be made to exclude undifferentiated ES to the points of current detection capabilities from any implanted cell populations. I think that's, if nothing else, perhaps the most critical point derived from this hour.

That the differentiation competence of these cells, their derivatives, have to be assessed in disease-appropriate models and that the nature of those models will determine essentially what our definitions become as to whether or not not only efficacy but safety has been achieved with these grafts. And then as essentially the risk benefit ratios are established for each disease target for each implanted phenotype that we perhaps draw a distinction between the risks inherent in inappropriate differentiation versus from the risks associated with tumorigenesis perhaps giving greater weight to the latter.

CHAIR URBA: Thank you. Yes.

DR. SCHNEIDER: I'd just like to make a comment. We were talking about cell numbers and assessing risk. I've heard numbers like 10⁵ and 10⁶ cells. But for some indications we're going to be using log orders more cells therapeutic.

For example, for islet transplantation, you need what, about half a million islets or a million islets to keep you glycemic, and if each islet has about 2,000 cells and it's mixed alpha and beta, we're talking about two billion cells, and so I recognize that the islets won't be put into a mass and they'll be separated from each other hopefully in some way. But I think that that should be considered as well, for that indication as well as for others.

CHAIR URBA: Okay. The program calls for a break at 4:00, but since we took such a long time on Question 1, we'll go along to Question 2 and we'll understand why people may have to sneak out for a few minutes or grab a cup of coffee.

So Question 2 has to do with the characterization of the human embryonic stem cell-derived preparations. We've been asked two questions.

Please which product characteristics might be predictive of adverse events such as ectopic or inappropriate differentiation, including tumorigenesis or other undesired outcomes.

And please include in your discussion the specificity and sensitivity of specific assays used to distinguish undifferentiated, appropriately differentiated, and inappropriately differentiated derivatives within a heterogeneous cell preparation.

We've laid some ground work for this in the previous two hours. Dr. Firpo, will you start us off?

DR. FIRPO: Okay. I think that a lot of the discussion on Question 1 was really more relevant for Question 2, so I'll try and be relatively brief.

In terms of the characterization of the cells themselves, the cell preparations themselves, I think that there are tools available that can help at least predict some of the issues with inappropriate differentiation, tumorigenesis both from the perspective of teratoma and from the perspective of tumors of specific cell types. So I'm just going to run down the list and maybe give a couple of examples without trying to be exhaustive.

First of all, I think that one of the initial characterizations of the cells has to be cell identity, and it is true that the commercial entities that will be working on commercializing and implementing clinical trials with these cells are going to be using their own proprietary cell lines. So this becomes important especially when there are multiple cell lines being used in the trial or in the company, but it will become even more important as we have patient-specific lines that become available. It becomes quite an important issue.

And then a question of how is the best way to do this has been -- actually it's still being argued in the embryonic stem cell community, there are very cheap commercial ways of doing this using STR, and snIP analysis used to be the standard, but I think that STR analysis is actually becoming the standard for this.

And the question becomes identity of the cells, but is also one of the ways that you can look in increasingly more sensitive ways at aneuploidies and genetic abnormalities in the cells that are at too fine a level to be detected by karyotype, which I'll talk about later.

The second issue is the viability of the cells, both before and after transfer, one, because this affects the dose, but, two, also because I think we need to keep an eye on what cells we're transferring, keeping in mind that the cells themselves may not need to survive very long.

I don't really think tumors are the only issue we need to discuss here today. Just want to make that clear. The cells don't necessarily need to survive very long to have a positive effect, a therapeutic effect in fact, or a negative result in that they could actually, for example, induce increased immunogenicity of the other cells, the desired cells, or they could actually reduce or increase the viability or proliferative capacity of the cells that we actually want to engraft.

The next issues is detecting contaminating cells, and for this there are several different tools available to us from other fields and some of them are increasingly or decreasingly sensitive or predictive for our purposes, and maybe a combination approach is the best way to go. One is that we need to look at the differentiated population for contaminating cells that may help or hurt engraftment. For example, endothelial cells that may increase engraftment, but conversely also increase tumorigenicity of any undifferentiated cells or earlier precursor cells that might yield tumors, teratomas, or differentiated tumors.

There may be inert cells that give us a false idea of the dose of the actual cells we're interested in transplanting, and then even more worrisome, contaminating cells that will give an adverse effect independently of the desired cells, the functioning cells, for teratomas, or, for example, cells that may create cytokines that change the function or the differentiation patter of the cells that we're interested in.

And as we look at how this is actually going to happen, we can think about the fact that the terminally differentiated cell product, for example beta cells or mature cardiomyocytes, are probably not the only cells that we're going to transplant. In fact, in those cases it might be more effective to transplant a precursor population, in which case we could transplant fewer cells and maybe fewer contaminating cells along with them.

The cells could become functional in conjunction with vascularization rather than having this sort of cell drop off that you would have if you put mature cells in. But also the earlier you go probably the more likely you're going to be infusing along with the precursor cell population undifferentiated cells that could give rise to teratoma or more cell-specific tumors.

So for this we have several tools. We have cell surface markers, many of which have been developed for adult cell or other cell product transplantation, for example CD34, for hematopoietic precursor cells. We can use reporter genes, although that is relatively sensitive. It gives us a way of monitoring the cells, but induces the problem of insertional mutagenesis and potentially causing problems down the road with tumorigenicity of the transplant.

We have PCR analysis, which is extremely sensitive, and then we also have in vitro and in vivo assays, which are going to be functional assays, increasingly important particularly when we're transplanting precursors that may not have function that we can assess as a transplanted unit. Now, of course PCR assays are extremely sensitive and reporter assays are more sensitive. Cell surface assays by immunostaining or by flow cytometry are somewhat less sensitive.

But each of these, if we were to look at specific markers indicative of undifferentiated human ES cells for which there are many cell surface markers available, would be quite valuable and certainly would be, if they are accurate, much more sensitive, lot more sensitive than teratoma assays in terms of being able to predict the ultimate long term tumorigenicity in people.

For example, we have SSEA4. We were talking about that this morning. But for that I think we need to correlate in realistic model systems what the expression of the cell surface marker is with the tumorigenicity of the cells. And for that I want to go back to hematopoietic stem cells.

The CD34 expression is often used as a way of sorting hematopoietic stem cells. And, in fact, in the literature you'll see people say hematopoietic stem cells as equivalent to CD34 positive cells. But we know from the breast cancer literature that it's absolutely not 100 percent accurate. And, also, we know from animal models that most of the CD34 cells are not stem cells and that that population is actually quite heterogeneous. So I think we need to have functional assays like teratoma assays correlated well with the cell characterization.

And this was also brought up already, but I think we also need to understand specific interactions between the contaminating cells and the cells we're interested in. And just as an example I think that there are many other reasons to do this, but that vascular cells or cells expressing specific cytokines that stimulate proliferation would be more likely to induce tumorigenicity in a very small dose of contaminating cells, where in a regular teratoma model you might actually need a higher dose of those cells.

And then finally the quality of the cells themselves, and I think that this was fairly extensively just recently, the karyotype of the cells, and for this there is really no consensus in the embryonic stem cell community. Most of these questions are actually being address by the embryonic stem cell banks themselves is how do we have to characterize these cells to really demonstrate that they're good quality cells, and I think the gold standard has been teratomas.

But when we get to the point of discussing what's the best way to analyze the genetic status of the cells, all consensus leaves the room. Banding karyotype is very sensitive in the sense that you can look at many, many metaphases on the same slide and see a small proportion of aneuploid cells that might be predictive of giving a specific tumor type. However, it's not very sensitive in the sense that smaller mutations are not easily seen.

Spectrokaryotyping has sort of the converse problem. It's not very sensitive in the sense that you can't cease, for example, inversions and small duplications or deletions on the chromosome. But you can actually see very small translocations that wouldn't be visible by g-banding.

And then the third method is CGH, which is raid-based analysis, where it's extremely sensitive in terms of very small deletions or insertions, but is very insensitive because you need to have about 25 percent of the cells with that specific mutation to be able to detect that population above the noise. And as the technology improves, that sensitivity will also improve along with it.

So I think that there are tools available to address the quality of the cells themselves. Of course we need to also characterize the cells that we're interested in looking at and what's the population of those. But I think that those kinds of studies also have to be correlated with in vitro and in vivo analysis so that we can come to some standards where we can say that this marker is, indeed, predictive of specific outcome, either positive or negative.

CHAIR URBA: Other comments? Dr. Taylor.

DR. TAYLOR: I think it's been pretty clear this morning that one of the issues on the table is not whether or not there will be heterogeneity in the final cell release criteria, but what's a tolerable degree of heterogeneity. And there are two different components in my mind to that.

What is the degree of differentiation required for the composition of the product, and what's the percentage of partially differentiated cells that we can tolerate?

I think it's probably impossible to set a given number for either of those in this room because ever preparation's going to differ. But I think it's important that we demand that the sponsor show that whatever number their product has correlates with safety and the benefit of that given product outweighs the risk.

I think we also have to take into account whether the site or route of administration changes this, and whether or not the host immune status changes it. Again, the patient population may matter dramatically. And then I also think, and along those lines, virtually every speaker this morning said that cells go where we put them and elsewhere, and that virtually every cell becomes what we want and something else.

And so I think it's important that we recognize there's an intricate relationship between delivery, between ultimate location and between the percentage of and degree of differentiation that's going to be required.

And then finally I think this probably an appropriate place to discuss what we heard earlier about suicide vectors and whether or not we would all feel more comfortable if there were a given way to overcome the potential limitations.

I'm pretty sure I don't want to go on record saying that we have to create a suicide option for these cells. On the other hand, I think we certainly need to go on record saying that we have to create a safety profile beyond a reasonable doubt, and that we need to explore a number of options in that regard.

CHAIR URBA: Dr. McDonald?

DR. MCDONALD: Yes. I was really going to make a comment on the last topic, so maybe we've already moved on.

CHAIR URBA: Stan?

DR. GERSON: I would -- If a sponsor can come up with a really good rationale for how the safety vector impacts, well, terrific. Otherwise, I wouldn't start with that premise just because of the risks of insertional mutagenesis for no other good reason.

I was impressed that there's in the context of the proliferation of the ES cells, the culture, the culture conditions, the length of culture, the number of passages, the predifferentiation strategy, a lot of things happen and I would encourage sponsors to be very cautious in their approaches to that so they didn't end up with a cell product with discernable karyotypic, genotypic differences which created new genetic unknowns in the product because it's going to be very difficult for anyone, including the Agency, to sort of assess risk benefit of that product if there's new abnormal genetic characteristics.

So I would encourage care in that, and in my mind that includes caution in willy nilly passage number tolerance. So I would somehow, limiting that would seem to me like a good idea because, otherwise, I think, to me, that's the most problematic element of the term inappropriately differentiated. I would also be worried about inappropriate genotypic variance.

The other element of inappropriate differentiation is we heard relates as much to the location and the disease model and the characteristic and the micro environment, and we recognize this is going to be a heterogeneous cell product, and the impact of the heterogeneity on inappropriate differentiation I think is actually very difficult to describe without specifics.

So I think in each application you're going to need to have a differentiation of how the sponsor will deal with the heterogeneity and the inappropriately differentiated component of the cell product. I don't think one can proscribe that. To me it's just having an assessment around that risk.

CHAIR URBA: Dr. Gunter?

DR. GUNTER: So I'll be brief here.

I think that regarding heterogeneity of the product, that will occur. And I think basically we should tolerate any degree of heterogeneity. But, there's an important but here, that needs to be consistent from batch to batch. And also, based on what we've already gone over, I don't think that we should tolerate evidence of undifferentiated embryonic stem cells in the product.

Regarding suicide vectors, it seems to me they may potentially cause more problems than they're going to solve. I'm not in the field, but I haven't seen data here that they actually work. So maybe that's something that we need to explore before we recommend it.

And I just want to mention a comment. This is really maybe off the wall. But inappropriate differentiation kind of scares me. I'm not sure we can really define what is inappropriate. Do we understand enough about the physiology of the disease or how the cells work to presuppose what phenotype they're supposed to have. You know, hopefully our preclinical models are good enough to tell us if the differentiated state of the cells is going to cause adverse effects.

DR. SNYDER: I just have a few comments, and I guess it reiterates what was just said, is that heterogeneity obviously is not just undifferentiated cells intermixed with differentiated cells, but it's also multiple lineages. And while Mary brought up that it could be bad. We, also, there's a vast amount that we don't understand where they actually could be good and some of the efficacy of the cells could be based on that heterogeneity.

Just one example that's been emerging in some of the work we've done is that the emergency of neural crest-derived neurons, in other words the autonomic nervous system, is linked directly to the emergency of vascular endothelial cells and you block the differentiation of vascular endothelial cells, which is mesoderm, and you completely lose your autonomic nervous system emerging spontaneously, and vice versa.

So the two of them creating the niche could be the basis of their efficacy and we just have to be very cognizant of that, that efficacy may be based on some degree of heterogeneity.

In terms of how to screen, it kind of gets back to what Ken was talking about in our earlier discussion about tumorigenicity. We need some kind of easy profile where we can fingerprint these cells so that we know that batch-to-batch we have at least a tolerable amount of heterogeneity.

And then finally in terms of suicide mechanisms, genes are just one of them, but some of it may turn out to be fortuitous, some of may just depend on being clever. For example, in the brain tumor field, one of the mechanisms for killing brain tumors is based on a pro-drug therapy -- a pro-drug strategy in which case a cell that re-enters the cell cycle self eliminates. And using the stem cell against brain tumors that's based on that strategy may also provide a built-in suicide mechanism. In other words, if the stem cell also re-enters, the cell cycle it self-eliminates.

So it's always dicey to put in any kind of genetic manipulation and I agree that a suicide gene just, while it might make us feel more comfortable, also introduces a complexity and even a risk. Again, as Doris said, we should somehow be sensitive to the fact or somehow think of clever ways that we can do mid course corrections through monitoring, and maybe we'll get to that in the last topic, so that if something goes wrong we can correct it or neutralize the effect.

CHAIR URBA: Dr. McDonald?

DR. MCDONALD: Yes. Just to comment on the heterogeneity issue.

I mean I agree. I think there's multiple lines of evidence that cross lineage. Treatments are necessary. So that mixing mesoderm and ectoderm or endoderm are necessary for certain effects that have been seen in some of our animal models. But the key answer there is always having the same heterogeneity in your product within some reasonable error.

I just wanted to make a comment on the tumorigenesis in general. I think another topic that has been touched upon, but maybe we should explore a little bit more, is perhaps how are we going to monitor and how we are going to assess for the indirect tumorigenesis effects of a cell product on the endogenous cells tumorigenesis.

So in our field in the central nervous system there's been some pretty elegant work that's been done showing that the neuroprogenitors being made and migrating, that there is a percentage of those cells, actually a surprisingly large percentage of those cells that have abnormal karyotypes that are normally eliminated.

But what if any of our cell-based products that we're using alter that response, not just to the cells that we're transplanting, but to the endogenous cells? And I think that's another thing that we need to have sensitivity to, both in terms of monitoring as well as our animal models.

CHAIR URBA: Dr. Woo?

DR. WOO: About this heterogeneity issue, I think you were talking about getting rid of cells that are contaminating in nature and they're really not needed for efficacy that's one thing. But if we're talking about we need accessory cells to make the therapy actually work and to think that we're going to start out with one embryonic stem cell and differentiate it in multiple passages to come to a certain phenotype, and then to demand at the end of it the product will have a certain percentage of cell type #1 and cell type #2 and cell type #3 to make the therapy work, then I think the burden is going to be on the sponsor to illustrate that that percentage can be reproduced from batch to batch and time after time, and I think that would be an insurmountable burden.

So if we need multiple cell types to make the therapy work, we need to develop two products and mix them together so that they will come to a certain proportion. Otherwise, it's not workable.

CHAIR URBA: Dr. Firpo?

DR. FIRPO: I just want to bring up one thing that I didn't mention before, but I think is probably worth looking at as a strategy. If we're talking about suicide genes, that it might make sense that if we're concerned about what we're putting in to actually do something like a purge where you can use immunological means, or even cytotoxic means that maybe not possible in vivo, to the cells before we transplant them. There's some precedence for this in bone marrow transplantation. But also that without having suicide genes, there may be immunological ways of killing tumor cells also in vivo, which I don't think has been explored at all.

DR. WOO: Just one comment on the suicide gene, and that is the suicide gene itself certainly brings in its own set of risks and so on. But also a suicide gene, if it is not human in nature, could potentially be immunogenic and that would help to get the cells eliminated. So I'm not so sure about suicide gene is the right approach to make sure that we can recall the therapy and we need to develop something better.

CHAIR URBA: Any other comments? Dr. Bauer, are you happy with the information so far?

DR. BAUER: Yes, I think that was a nice discussion of the points. Thank you very much.

CHAIR URBA: Any comments from our guest speakers today? You've been very quiet. No? Okay. Did you want to say something?

DR. ISACSON: If I may?

CHAIR URBA: You may. Use the microphone though.

DR. ISACSON: Clarification. So actually to the extent that I don't know, one can certainly purify as I showed dopamine neurons to purity, and certainly I believe others will be able to do so with any neuronal cell type.

Now, if requirements are that you have a reproducible mixture of cells, that's also very doable using flow cytometric techniques. So I don't think it's an, quote/unquote, unsurmountable barrier. And, in fact, if life cell therapies as Dr. Snyder imagines would require more cell types, I think that those requirements can be overcome technically.

DR. GOLDMAN: Let me follow up on that.

On some level I think it's actually axiomatic that what we're trying to do is generate the most differentiated phenotypes that are compatible with viable transplantation and integration. So when it becomes a question, as per Evan's point, of perhaps needing to have a number of ally phenotypes present to allow their mutual survival, I think the imperative then becomes having the separation technology so well developed as to achieve each of those individual lineages and then mix them back, so that at least we have some control then upon what we're doing.

DR. SNYDER: Though I would point out that, you know, as you know in the neural stem cell field, once you get into the neural lineage, there is kind of a division of labor in proper ratios for example, between the right ratio of oligodendrocytes to astrocytes, often the right number of neurons to nonneuronal cells, so it may be a challenge in other fields.

I agree with Ole that it's probably less of a burden for us in the neural field because we seem to be able to get a lot of the right ratios with the cross talking cells that we need.

DR. DINSMORE: If I can speak directly to the issue of purifying cells? You often purify away the therapeutic benefit of cells by purifying them. Take tologous myoblasts, if you purify them to heterogeneity in a FAC's machine, they don't work. You need some less manipulated cells than that. In some cases neurons have been successfully purified, but it's the rare case that 100 percent purified cells work when you subject them to things such as FAC sorting or the like.

DR. GOLDMAN: I would argue that point. I think the question is the markers upon which they are being sorted and the criteria that are being used for defining that population. And so it gets back to simply the definitional point of purifying the most differentiated population that is still compatible with the successful and viable transplantation and integration of that pool.

DR. TAYLOR: I just want to reiterate though what Jonathan just -- the point that he made that may not have been appreciated, which is homogeneous cell populations may not be the therapeutic tool. When we talk about CD34 positive cells, when we talk about virtually any cell population that's being given clinically now, it is not a homogeneous population. Preclinically, we and a number of other groups have shown that if you purify, purify, purify, you lose potency.

And so cells talk to each other and we don't -- we're young enough in this field that we don't fully appreciate how they talk to each other. And so I think we have to make room and release criteria for the heterogeneity that may be the biologically relevant molecules here.

DR. WEIR: Just a brief comment about diabetes. That's one area where we know if we had pure beta cells they'd work. We wouldn't need anything immature. Just a straight thing.

(Laughter.)

DR. FIRPO: Okay. So just to summarize. The discussion mostly revolved around heterogeneity and also sort of failsafe mechanisms in case too much heterogeneity leads to problems down the line. So we talked about heterogeneity is not desired, but actually may be required for engraftment and functionality of the cells, although it may be possible to isolate specific populations and add them back together.

It's not clear if that is actually going to lead to a result. In fact, in some populations the specific populations that are required may not be known or isolatable.

With respect to suicide genes, I think maybe I'm not - maybe I'm speaking too much. There is a consensus this would be nice, but that because of insertional mutagenesis or other immunological problems that it would cause, that this not really the desired way of getting rid of the cells. There may be some other means, but there didn't seem to be a consensus that this should actually be required.

And then aside from that, there was a pretty good consensus I think that the heterogeneity could potentially be tolerated as long as it was shown to not be dangerous and that it was consistent from batch to batch.

And then, finally, there was some discussion of giving some parameters as to the passage number of the cells, I think the cultured cells, in terms of batch-to-batch consistency.

CHAIR URBA: Okay. Thank you very much.

Now we'll move on to Question number 3, which has to do with patient monitoring.

Taking into account the capabilities of existing analytical tools and non-invasive monitoring technologies, please discuss features of early phase clinical trial design that will facilitate monitoring of patient safety following administration of human embryonic stem cell-derived cellular products.

Please comment on other trial design features, such as cell dosing, that can help to increase the probability of obtaining a measurable therapeutic benefit while ensuring maximum safety.

What I'm reading isn't exactly the way the slide has it, but I think it's all contained therein. So, Dr. Weir?

DR. WEIR: Okay. I'll make a few comments to get the ball rolling.

Obviously so much has to be done before the cells would go into people and we've had a nice discussion about that. But I wanted to talk first about the areas of the potential abnormal growth of the cells, the possible teratoma formation, and the possible malignancies and how we would -- what we have to do to monitor those.

And, first of all, the abnormal growth is of interest and that goes down to dosage issues. And, for example, with diabetes, if you had a situation where you got the wrong dose and it generated too many beta cells, you'd hope they would auto-regulate and just shut off themselves, but they might not. So you might have a situation where you'd have to go retrieve beta cells to prevent people from getting hypoglycemic. And I can imagine that in any other of these diseases that we are talking about that you could have a situation where you had too many dopaminergic neurons or too many whatevers and you'd have problems with that. So some you could monitor and some you couldn't.

Now, as far as the teratoma formation, if they were benign teratomas, they could be really problematic from just a space occupying point of view. For example, particularly in the brain or in the spinal cord, you just have a growing, benign teratoma and that could cause a heck of a lot of problems. On the other hand, it's conceivable that you could have some in a more convenient place, like in a subcutaneous site, where you might just be able to take it out. But it's a major concern, and, hence, all of our issues about putting in as safe a product as possible.

Now, malignancy is something that will just always be a specter that worries us. One, of course, the teratoma with its potential for malignancy. But there may be other things that stem cells do that would allow the appearance of malignancy by tampering -- tinkering with the immune system. So you really have a problem with a malignancy that could be anywhere in the body and it's sort of like somebody having a genetic predisposition to malignancy and how do you follow that so you could have every scan or MRI in the book and we're just going to have to make some decisions as to how we deal with that.

Now, Ken and other people have brought up the issue of cell markers, alpha feeder protein and, to be as yet discovered, blood tests that could tell you that there's a malignancy or a teratoma to be dealt with.

So I think we're going to need very specific approaches. We're going to need some general approaches for watching for malignancies that we want to catch earlier rather than later. But we're going to need some very specific things for each individual disease.

For example, spinal cord interventions are going to need, you know, probably MRIs of the spinal cord area. Things into the brain are going to need separate approaches than things for diabetes. So I think there are those variations.

And we've already talked about how long do we have to monitor people for, and the answer is forever. We don't know if things are going to develop 20 years from now and it's going to be rather complicated to work out the schedules as to what you monitor and how often. But, nonetheless, that's going to be our task.

So that's as much as I was going to say about the sort of tumors and abnormal growth. But, again, we're putting in these cells and we don't know what they're going to cause in terms of trouble. So I think we need to be rather broad in our surveillance.

Something as simple as human logic changes, weird autoimmune, anemias, or blood dyscrasias, or clotting disorders, I mean just standard things that you know I suppose you could dream up ways in which embryonic stem cells could cause problems. And I think particularly the immune system, I mean there are so many things that affect the immune system. So could you trigger autoimmune diseases, could you start getting glomerulonephritis?

You know, a whole spectrum of autoimmune disease, simple allergy-like reactions, things like we mentioned graft versus host, so I think that, again, I just want to emphasize the sort of broad monitoring that I think we have to do for these patients who are transplanted.

And I think maybe that's what I wanted to say as introduction and maybe we could fire away.

CHAIR URBA: How about other design features? Are some of the trials potentially risky enough that you should start in a patient who has cancer and Parkinson's disease? I mean could you select a higher risk patient population? I have lots of patients with macular degeneration who have cancer, too, that you could do things you might not do in somebody who's -- and then a second question would be, I think everyone's familiar with the study where antibodies were given to normal volunteers all at once, and one part of trial design is you treat patient number 1, how long do you have to wait before you treat number 2? Do you think these cells require any special attention to that versus what we would normally do?

You can answer that.

DR. WEIR: Yes.

DR. MCDONALD: Yes. I'll address one of the questions, which is -- really, both of those up there.

In our field, in spinal cord injury, we've already had to go through this, and it's interesting. Like, for example, a typical safety approach in spinal cord injury would be that, well, we're going to do the first therapeutic in the thoracic area as opposed to the cervical area because, God forbid if something goes wrong, you could afford to lose a couple of levels in the thoracic area. But you lose a couple of levels in the cervical area, now you have interference with respiratory function, this, that and the other.

But the converse is that now you've created a very insensitive test for problems. That is, a problem, a major problem could occur in the thoracic cord and we not even notice it clinically. Now this is compounded by the fact that basically you lose the ability to do MRI in most of these patients for any real sensitive tests.

So in modern days the majority of patients now are getting acute surgical stabilization. And a number of the trial designs have been, whether it be material based or cell based, is to let's go in when they're already going in for the operation. Let's not put a patient at an additional risk.

But the trouble with that is that now they have hardware in. And although this hardware is titanium, it completely obliterates most of the techniques that we have with any degree of sensitivity. There's a number of things you can do to help get rid of that distortion in part. But basically the high resolutions that we heard about today basically fall off the table.

So some of our most sensitive noninvasive imaging approaches now aren't applicable. So we've created a situation where it's a bit of a band-aid. We feel that it's safer to go in. Really, in essence, it's not any safer, but we're creating a trial design that's incredibly insensitive to detecting problems, or, therefore, also benefits.

So if you regain functions at two levels in the mid thoracic cord, we can't detect it with our gold standard tests, whereas in the cervical cord we can. And the approaches in both the companies that I've been involved in from an advisory standpoint and we had to go through the FDA in this regard, the choice was that it was an escalation approach. That is, we would escalate in terms of level of the spinal cord treated.

Beginning in the thoracic zone, we waited, what was it, I think eight weeks or 12 weeks before we went into the cervical spinal cord, and I think that was a reasonable amount of time, but I think that the major decisions about why to go in the thoracic cord in reality don't make much sense, and, two, really work against us in terms of both of those questions.

And I think this is just an example of things that have already been through just recently where I don't think we're adequately thinking about this because we're I think so worried about the potential negative that I think we really need to be careful about designing those, and those were trials that worked. But we're designing safety, phase one/phase two trials, that are insensitive both to therapeutic effects and problematic issues.

DR. WOO: I'd like to discuss a little bit about the trial design incorporating this duration of may therapeutic effects, maybe short term toxicity and long term toxicity.

Because of the phase one trial design that we're really doing those escalation to measure side effect, which is the first step in clinical trials. Well, first of all, I would say that lifetime monitoring a patient after treatment is maybe desirable as Dr. Weir mentioned. It cannot be utilized as a criteria for the dose escalation studies, and if that were the case, then the trial can never be completed. First patient is treated, life long monitoring before the second patient, and that's the end of the trial. So we have to be pragmatic about it and to have some sort of short term measurements in terms of toxicity and so allow the dose escalation to proceed and complete the trial if there is toxicity and follow up with efficacy studies.

In terms of the potential for long term toxicity, at this point no one knows. So the possibility is always there. You can never eliminate an unknown and so that would have to be taken care of in informed consent. And I don't know any other way out of this.

DR. GERSON: Could I just address the issue of monitoring cells in vivo?

Since these are allogneic cells, the fingerprint seems to be should be available, and I think study design including informed consent should encourage the request for biopsy of abnormal collection of cells and validation that the collection of cells is from the implantation using available genetic markers. I think that's sort of an appropriate premise important for the field, et cetera. But I think that rather than being implicit, it would be explicit as an expectation here.

I do not think that it's wise in this setting to force only the most extremist of patients to be considered for early phase. We don't have enough evidence of severity of expected toxicity here to I think justify that. I think we have to have the expectation that we'll be doing something which is safe and ultimately efficacious or in the patient population at risk, and I do think that very careful assessment of toxicity fairly frequently is required.

I also think that there should be imaging both locally at the site of the installation and systemically at some interval given the possibility unknown that there could be significant amount of migration. We haven't heard or discussed anything about the possibility of invading, if you will, the germ line, but that at least in the back of our minds still has to be considered.

CHAIR URBA: Chien?

DR. CHIEN: Yes. I was going come back to this is one where I think before I was advocating one size fits all. We should have a standardized, up front, early stage, everyone doesn't want to see undifferentiated ES cells in any prep or cell that could go rogue. So that's seems to be fairly straightforward.

In this case when you're looking at safety issues in the context of a patient that has gotten cells in a specific location, I think this is where there has to be kind of case specific rules. And so, for example, the things, and again just using my own thinking here, if you're going to put cells in say -- I'm not sure why you would do it, but let's just say in an intra-articular location. You know, you're going to regenerate cartilage. That's going to be probably fairly different than if you're going to put it into your heart for example. And one of the questions is, where do the cells go when they're delivered that way?

I think one of the things that we have not talked a lot about here is delivery, and how you deliver it is going to make a big difference on where the cells go. And it gets back to this imaging issue that we talked about earlier, and I am not convinced that one has to have elegant molecular imaging to be able to at any time survey the cell non-invasively and know where it's at and what it's doing.

I think that's too high a hurdle. I don't think it's even feasible. The technology isn't there and it's certainly not at a single cell level or resolution and you almost need that. So if you can't get that, then I'm not sure what you're getting for all that.

But you can optimize the delivery to make sure you get the cells as much as possible where you want them to go and where they don't. And you should also know where they're going where they might be clear in the liver, they might be clear in the lung depending upon what the target organ is. And then that organ should be surveyed more carefully than any other average organ and there're ways to do that.

So I would do that on a case by case basis. You could have a generic thing, but I think it should be more specialized for the indication.

The other thing is I do think there are unique clinical circumstances where cells can be put in and cells can be taken out. So let me use my favor organ system, the heart.

Obviously, transplantation, organs are transplanted. There could be unique circumstances where you could do things in a human setting, and, of course, it needs to benefit that patient also. So this is the other tricky thing.

And it needs to be thought about very carefully, but I think there are cases where you could justifying delivering cells to a patient and then taking the organ out and seeing whether or not the cells actually got to where you wanted and did they become what you wanted, and then you're looking at the organ because you put another organ in. And then maybe later on that patient will be a candidate for the real McCoy, but I think you'd find that patients might participate in that.

So I think that this is where I think thinking about unique clinical niches could be useful. So I wouldn't just automatically dismiss it and say you couldn't actually get what you wanted.

CHAIR URBA: Dr. Chamberlain?

DR. WEIR: Actually, just to build on that, Ole Isacson brought that important study of Parkinson's disease after many years that I guess was done on people who died, and for some of these bad diseases that we might be trying to treat, maybe there should be some sort of plans that when people do die, that we can get some of that tissue and study it properly.

CHAIR URBA: Dr. Chamberlain?

DR. CHAMBERLAIN: Well, I was just going to actually make the same comment that Dr. Gerson made a minute ago. That when you're monitoring patients perhaps for the rest of their life for tumors, at least 25 percent of these patients are going to get a cancer anyway. So you really have to have a way to distinguish whether it was the ES cell or not. And so there should I think be a strict requirement that every batch of cells that's prepared by a company have a DNA profile attached to it so you can go back and determine.

The other thing is that one of the things I worry about other than the obvious unique feature of embryonic stem cells having the ability to form teratomas is that I think once you get beyond that and we establish criteria for how to screen the cells or purify them properly and then monitor for tumors, beyond that I think we need to be a little bit careful that we don't establish so many standards beyond what are normally required for conventional stem cell transplants using adult stem cells because those have been done for many years very safely. And, as I say, other than the potential tumor issue, that perhaps some of the existing levels of scrutiny for adult stem cells should simply be applied to the embryonic stem cells as well.

CHAIR URBA: Dr. Chappell?

DR. CHAPPELL: I would like to comment on some of the design aspects that are required for these types of studies and contrast them to perhaps the most common and oldest dose escalation trial which is in cancer.

For cytotoxic agents in cancer the goal is to get as high a dose as possible, where possible is defined in terms of safety. Here I think from what I understand the discussion the goal is to get as high a dose as possible and necessary, where possible again refers to safety, but the necessity relates to efficacy.

Unlike cytotoxic agents after a certain number of cells, if they work, you don't want to escalate regardless of the safety issues from what I understand. And that means that we're in the realm of what's often called phase one and two, efficacy safety trials done simultaneously, which have a lot of good characteristics anyway, but they can be somewhat complex, and people expect more out of them than they can give sometimes.

And going back to my earlier point of asking for an incidence of efficacy and safety if they're binary or some level, some continuous variable, I ask you to consider a situation where you're trying to show that the incidence of some event toxicity is below 20 percent, which is pretty high I think, certainly for teratoma it's very high. So what kind of sample size would you need?

And suppose you have the most favorable possible outcome? No patients in your study get that toxicity and you use 95 percent one-sided confidence interval, which is pretty loose. Then 13 patients even with zero toxicities, the confidence interval would overlap 20 percent. You'd need 14 or more patients. And 20 percent's high. For five percent you would need approximately 200 patients per single dose and I could just see the industry representatives pulses rising.

And so that urges, first of all, as always, large sample size. But secondly are surrogate variables. I don't know what might be a surrogate variable for a teratoma. If you don't have that, then you're going to need very large sample sizes to determine them.

And, similarly, for efficacy, you need some kind of surrogate variable for efficacy and/or very large sample sizes. And there are a lot of phase one/two designs out there from cancer and neurology to which this bears resemblance, not coincidentally which we can borrow and build on.

The second issue is timing, and, again, there's a difference with cytotoxics. Some of the toxicities I've seen listed in the background documents are quick. But some of them, like teratomas, seem to be slower, and I asked around during lunch about how long you might expect, and there's no maximum, but the minimum of my answers tended to be six months to a year. And, again, you go into the area of impracticality if you have say five or ten doses. It takes six months to accrue patients, which would be nice. It takes six months follow up and so you have a year for each dose, plus the planning and analysis, and you're talking about ten years for a trial.

There are designs which are called staggered entry. But by staggered I don't mean sequential, one dose after another, but where the doses overlap. That is, before you're finished with the first dose, you start the second dose and keep tracking the first dose. Now that can make people nervous because if suddenly at the end of say six months, for the first dose you see a tremendous number of toxicities, then you've placed dose, cohort which you've already started accruing at risk, and recognize that risk, but the alternatives are that you completely ignore what happens after a month or whatever the usual followup.

So I urge you to stress long followup until you know what you're doing so well that you've determined that the relevant outcomes are ascertainable with a short followup, and I'm confident that that will take a long time.

So to summarize the last part, I urge you to take advantage of recent results in phase one trial designs, of which there are a lot, to have fairly long followup. And, actually, in cancer, those have been used. Some have gone through INDs to the FDA, so it has passed and they've been supported by the NIH. So progress is being made.

CHAIR URBA: Dr. Gunter?

DR. GUNTER: So I wanted to comment on the long term followup of patients in these studies and I think that everyone that's going to be doing these clinical trials wants to follow good clinical practice. But I must point out that following patients for their lifetime is quite an undertaking for any clinical trial and I have to actually question whether it's feasible to do it and follow GCP.

Now I will say that one of the best guidance documents I've seen from the FDA was one that discussed long term followup for adverse events following gene therapy, and it was a nice risk-based document that based the followup on the type of vector used and other risk factors.

I hate to assign work to the FDA at these meeting, but it might be possible to come up with some kind of similar document that prescribes long term followup based on the risk of the cellular product, which presumably could be based on something like the percent of undifferentiated ES cells in the product. But I have to say the retrovirus document is a 15 year followup just so we all note that for the record in the worst case scenario.

A couple of other things I wanted to mention. Just thinking out of the box in these kind of clinical trials, it might be useful thinking about ways that the cells could be given to allow them to be retrieved if necessary. And I'm not very creative, but I thought of a couple of things.

One is encapsulation. Another in a patient population with a chronic disease might be given the cells subcutaneously first, followed by an excisional biopsy, and then getting a second dose to the same patient in the target area.

I thought the talk this morning was really interesting on newer imaging technologies and I hope we can talk a little bit about them. But I want to point out that some of -- well, maybe it's just a question. Are we going to run into problems using experimental imaging technologies in conjunction with experimental therapies? Is that going to cause problems of interpretation?

And then Dr. Chappell already hit the dose staggering issue that I wanted to cover, so thank you for that.

CHAIR URBA: Dr. Salomon?

DR. SALOMON: I was just thinking this was one opportunity that we could be a little concrete because the reality here is that how can you move any kind of field forward, where there were risk averse or not risk averse, if you're not concrete about what you need to do to do a trial. So I'd like to just suggest something and everybody can tear it up.

I say you start with an animal model and you come up with the best judgment for the dose. Everybody can discuss what they like. The animal model, we already had lots of comments on it. But bottom line is the sponsor starts with the animal model or animal models and does their best judgment dose and that gets accepted.

Then you start the clinical trial with half that. Again, I'm just trying to be concrete here. I think there would be exceptions to that. For example, if you were talking about, Ken, acute heart attack, you've got on chance to save the guy's life. Well, you know, you might not start with a half dose in that situation. So, again, you'd have to risk stratify, which is a point I made earlier.

Then how many patients do you do on a cohort? You know, three patients per dose cohort works pretty well for gene therapy and cell therapy trials. I would suggest that we don't necessarily have to suggest inventing a new wheel for that one.

Again, how long do you wait then after you do that cohort of patients? I'd suggest the following:

1) Take twice the time in the best animal model they have, or the average of the best set of animal models that the sponsor has and it's accepted by the FDA. Take twice the time demonstrated in that animal model or models to get a stable outcome based on the most significant measure of the outcome.

Treating spinal cord injury and the outcome's walking, then, you know, follow it up, the first cohort, twice as long as it takes the best animal model to start walking, that kind of thing. And again, there would be exceptions.  

And this is my last point. It goes back to risk stratification. If you have a really high risk, high yield situation, one could apply that and say, okay, fine, that's good. But if you have a very low risk situation, a child for example, and one might modify that and go longer followup before you onto the next cohort.

CHAIR URBA: Dr. Kwak?

DR. KWAK: I sense the discussion is winding down. So I wanted to come back to an earlier point made by Dr. Gerson initially.

With the interest of perhaps emphasizing its importance for inclusion in a guidance document and that's the idea that in clinical design that we incorporate some kind of multiple sampling or encourage multiple sampling so the question of persistence can be monitored as a secondary end point. I think this is going to be important both from a scientific standpoint as well as, not just safety, but obviously it will impact on the potential benefit of the maneuver.

And, secondly, I think that's an important question to answer individually in individual clinical protocols because the immunosuppression regimen will likely differ between the different protocols, and so the same answer may not be obtained in different protocols.

CHAIR URBA: Dr. Taylor:

DR. TAYLOR: Although I'm not sure how this can be rigorously regulated at this point, I think I would be remiss if I didn't comment on the fact that there are preclinical data in adult stem cell therapies and in adult even in clinical data as well as in published preclinical data that sex makes a difference. And so I would encourage the sponsors to think about included sex-based -- be clear about including appropriate numbers of males and females in safety data so that adverse events don't emerge in large clinical trials.

CHAIR URBA: Dr. Snyder?

DR. SNYDER: So just to get back to a point that I think is very interesting that Kurt was raising.

While I agree that realtime imaging of prelabeled cells is probably an unrealistic hurdle right now, I think we should encourage that as a very admirable, worthwhile future goal for future development, not only for tracking where the cells migrate, but also it could be an early indicator of when tumorigenesis is imminent, maybe before there's a clinical symptomatology or even a blood test or a profile.

I think the other thing we need to discuss, particularly when we're talking about patients, it was raised before by Dan, is what immunosuppressive regimes are we going to recommend? Because when we actually go into patients, we're going to have to deal with whether or not the sponsor has an adequate immunosuppressive regime, if any at all, and how we'll deal with that.

And then one final topic we probably should discuss. It was in some of the preliminary stuff you sent out to us is how do we regard migration? We tend to think of putting in the cells and migrating away from the target area as a bad thing, but for certain kinds of indications migration may actually be a good thing.

If we're talking about ALS for example, being able to migrate up and down along the neuraxis may actually be a desirable thing, certain kinds of spinal cord injury. Targeting metastatic cancer migration, particular in the brain, migration might be a good thing. So I think before we leave, we probably should address at least some of those issues.

DR. MCDONALD: I just wanted to make a comment about what I thought was very important, the surrogate markers, specifically that we're talking about what can we do in specific trial design for phase 1, 2 trials that is doable that gets better at efficacy and problems that perhaps, of course, we wouldn't do in phase 3, 4 trials. And I think that specifically relates to this issue of tumorigenicity, and it relates to my earlier comment about the lack of effectiveness of the most relevant and widely available imaging technique MRI for detecting problems.

You know I think some of the more recent data on high resolution PET, okay, this would become a standard in phase 1, 2 trials in order to detect a surrogate marker that will become a uniform standard both in space and time for any product. That is, you want to come up with this standard that will be uniform, independent of the product, where the product's put, how long after you put the product in do you analyze.

And so there are a number of cell surface receptors systems that are expressed in proliferating cells, highly proliferative cells, and/or tumorigenic cells that high resolution PET imaging could get at. And that is something that would be feasible, although there's not many sites to do it, that would be feasible in a phase 1, 2 trial and you could really detect, at least as a uniform surrogate marker, as potential indicator, I think that's something that could be implemented in the near future.

CHAIR URBA: Dr. Tomford?

DR. TOMFORD: As a clinician, I know that the human response can be very variable regardless of how you try to control it. So I'd just like to make the point, I think it's very, very important to make sure that what goes into the patient is very well characterized so that that way you can at least provide some control to what comes out. It's like the garbage in/garbage out situation. Unless you know precisely what's going in, that doesn't mean it has to be standardized to mean it can be heterogeneic and maybe that's an advantage, but unless you know precisely what is going in, and that involves standard tests, then I think it's difficult to tell what's going to come out of the human.

DR. CHAPPELL: I'd like to respond to Dr. Salomon's invitation. Not to tear it apart, but at least to comment on the three using -- to stick with the inertia and use the same three per cohort design and suggest that it broadened, that design possibilities be broadened for two reasons.

First of all, efficacy obviously plays a role. I've not heard anybody deny that in setting a dose. And nobody that I know of has a three-person, three-subject cohort efficacy setting. So you need more information than that.

And, secondly, even for pure toxicity studies, three-patient cohorts, when they've been examined carefully, the effects many times has been disastrous in that it may look good, but then it moves uncertainty about toxicity to later in the trial, phase 2 where toxicity is not examined as much, or even phase 3 where you have just a catastrophic disaster, it stopped early, people wondered how it could get that far.

If you have three patients and toxicity as high as 50 percent, then you still have a very good chance of not observing any. It's just one-eighth, which is somewhat similar in chance to succumbing to Russian roulette, but for each cohort, and even worse these are patients who are already sick, and sometimes toxicities are confused with consequences of the illness. And so it can be a bit muddled if you use sample sizes that small.

But the advantage is that since we're combining two phases of clinical research, you have a bigger budget, literally and also in terms of patients, to have a larger sample size.

CHAIR URBA: How many patients per dose does your trial have in it?

DR. CHAPPELL: Well, it depends what you're -- so you'll never get a straight answer unless you ask a quantitative question.

(Laughter.)

CHAIR URBA: That's not quantitative about how many. How many?

DR. CHAPPELL: Okay. To answer what question? So typically the studies I've been --

CHAIR URBA: The phase 1, we've been focusing on toxicity.

DR. CHAPPELL: Between 50 and 100. So if 50 and 100 for a phase 1, 2 trial that gives you usable information on toxicity and efficacy to design a very early type of phase 3 trial or randomized phase 2 trial. So between 50 and 100 is typical.

But if you're demanding very, very strong evidence for no teratomas, that number would have to be increased.

DR. SALOMON: I want to clarify one thing. I was talking about the size of cohort in a dose escalation study in a phase 1, maybe /2. Actually it's a phase 1, 2 that I recommended because the timing was associated with an outcome. I wasn't talking about three patients in the study. I think the whole thing would be the typical thing that we do for cell therapy, instead of having many of the same issues, and that'd be 12 to 20.

CHAIR URBA: But I think he's arguing correctly?

DR. CHAPPELL: I understood that.

CHAIR URBA: Yes, right.

DR. CHAPPELL: Sorry.

CHAIR URBA: Yes. And I understood your response and your response was that we get away with doing it that way. But when we actually put three patients on dose X and there's no toxicity, and then go to Y and there's no toxicity, our confidence level that there really is truly no toxicity with that drug at that dose level is very broad because it's only three patients.

DR. CHAPPELL: Yes, too broad.

CHAIR URBA: Too broad for teratomas is what you're saying?

DR. CHAPPELL: From what I've heard here. If you would be comfortable with the potential 50 percent teratoma rate when you don't see any.

CHAIR URBA: Anybody want to respond to that?

DR. SALOMON: Just point out the obvious. That if you want to do it, if the suggestion, that is to be concrete, and advise the FDA to do a dose escalation study with 20 patients in a cohort, that that would not be practical.

DR. CHAPPELL: Between three and 20.

(Laughter.)

DR. CHAPPELL: Of course it also depends on how many cohorts. If you have 20 cohorts, then, of course, we'll have to be lower. And it depends on the disease.

Right now I'm conducting a phase 1 trial in radiation oncology for radiation treatment of prostate cancer and we have 50 in three cohorts and they're accruing just fine it being prostate cancer. So, of course, it depends on the disease.

DR. TAYLOR: I would also say it depends on the duration of time you wait between cohorts.

DR. WEIR: I wonder if with in thinking about some of the more serious disease with the worse prognosis whether there'd be some advantage in leading off with those clinical trials, in which case you might learn a great deal about teratoma formation, which would then enable you to change how you do things for something that are less threatening diseases.

DR. WEIR: Just to continue on the teratoma. It seems to me that the expectation is that, with these provisions, we would not be observing teratomas, and, therefore, it would be a cumulative experience assessment by the FDA and that they would modulate their instruction based on observed teratoma events rather than having anyone take on the responsibility, if you will, of proving that these cell types are not producing teratomas.

Secondly, it seems to me, as we discussed earlier really throughout the day, that based on disease, cell site of infusion, severity of illness, et cetera, that the cohort and dose escalation range would take care of itself and just have to be considered. But I don't see how it could be proscribed here.

CHAIR URBA: Your model with the cumulative reporting would assume that all stem cells are equal, and that may be the best that we can do and still be reasonable, but you're likely going to have 12 patients from one company and 15 from another and they're going to be using a different set of cells. And that'll be useful information, but it's not the same as what he's describing.

DR. GERSON: Correct. On the other hand, if I could take the experience of