Melanoma 2003

SLIDES & TRANSCRIPTS
Monday, May 5, 2003

Commentary: Pro

Jeffrey M. Trent, Ph.D.

Slide 1:

Well, moving from the NIH to the non-profit research institute, TGen, has returned me not only to my home, Arizona, but also to a region of the country, according to Dave Alberts, where, especially if you back out the Native American or Hispanic population within Arizona, that has six times the national average in terms of its incidence rates for melanoma, as well as its mortality rates.

So, Nick Hayward and others will try to convince you that Australia is where you have to move to study melanoma, but I can assure you that, within the context of Arizona -- I think Vern will try to say even in the context of Michigan -- there is just a remarkable opportunity to study this because of the unfortunate incidence rates that were shared with us this morning, that this is a disease that is worth focusing our efforts on.

I applaud the NIH for continuing, and the National Cancer Institute for continuing, to put forth melanoma, despite its relatively lower incidence rates on the kind of programs like this, that are important for moving this forward.

I have been asked not to share about melanoma research programs that are being developed, of course, but I was asked to pontificate, and to do so in an entirely positive way, to be able to say that genes are deterministic, that they, in fact, explain all of melanoma and all of life, and that that was my charge.

I will mention I was interviewed a week ago, in case anybody heard that, with my twin brother, John, on National Public Radio, just as an example of exactly the opposite, that is, that genes are not deterministic.

My brother is a minister, for example, and I think that was the example that they gave for differences, that we aren't merely the sum of our genes and that the truth perhaps lies in the middle.

Slide 2:

So, what I will focus on today is something that I think, again, has been around for a long time, and it highlights the fact that the genome project has brought alongside it, in the merger of technology to medicine, something that has been recognized since the mathematical days of really just the early turn of the century, the statement that every real advance goes hand in hand with the invention of sharper tools and simpler methods.

Slide 3:

I will spend no time mentioning to you what you already know, that some of the tools for expression analysis, as Paul Meltzer mentioned, as well as Mike Bittner, have really led us to be able to begin to formulate questions and hypotheses.

Slide 4:

I won't spend any time on the protein side. We will have talks about this specifically coming forward. The end of this, ultimately, when we are able to collect protein information, fractionate it, is ultimately to generate the same sort of algorithmic datas, to try to look at cancer associated programs, to implement them into the clinical spectrum.

Slide 5:

You have heard about the technology development at the level of the position of individual bax. This is from, of course, the slide that you saw just a few moments ago from Boris Bastian, showing Dan Pinkel's excellent work on using bax as one of the templates for being able to develop whole genome expression analysis.

Slide 6:

This slide perhaps doesn't project very well, but one of the ways that Mike Bittner, Paul Meltzer and myself have been involved with investigators from Adjuvant Technologies, which will be presenting shortly, tried to do this at the recent AACR meeting -- so we will do it at a later point -- is to be able to show that we have been able to work with Allogen to develop gene level positional resolution.

That means that, instead of using bax as the template, using cDNAs or, in fact, oligos for genes, to get individual copy number changes of ones versus twos, which is one of the important ways we think of adding positional resolution to the type of studies that you saw a few moments ago in regards to using bax for that.

Slide 7:

All of these technologies are really being developed to try to somehow translate the information from these high throughput technologies into clinical studies, and I will focus really on three brief areas for how I think this is impacting us in a positive way.

Slide 8:

First, I will speak to one of the universal challenges, and just to make a comment or reference to that, and ultimately will discuss, then, some of the technical challenges and integration of genomics into clinical trials but, without question, access to tissue for high throughput studies is a huge problem and something that requires massive effort.

I have been fortunate to work under the direction of Andy Von Eschenbach as head of the tissue committee for the National Dialogue on Cancer.

I think you will hear, over the course of the next year, the development of a national approach to try to solve this.

Even though it seems like a problem that has been tackled over and over again, there is hope that there is, in fact, some specific focus that could move forward on trying to develop approaches to help provide access, not just within the research community and the academic community, but the corporate community as well.

Slide 9:

Technical challenges, I will discuss very briefly, relate to the computational analysis of data and integrating that, and I will focus on that and target validation.

Slide 10:

I would like to give just one example of the fact that computational biology will change how we really look at the data from these projects, and this is work from Paul Meltzer's lab, a really talented fellow from his group, Heather Conliff(?), who has just taken the type of expression information that you heard, but mapped it back against the now growing information about individual genes, so annotating each one of the genes.

This is looking at estrogen responsive genes in a so-called heat map but then, underneath it, being able to take the information about the specific pathways that these genes are involved in, overlay that information, that ontology information, and increasingly this will help, particularly those of us that, again, are still challenged in terms of our remembrance of biochemical pathways for matching the information about specific targets, I believe, to types of studies that you have seen in terms of the work that Mike Bittner and others have presented.

Increasingly, this is the world we have to deal in. We have to be able to think about computational models to address clinical questions, really incorporating and maximizing the information that we get from these various platforms.

Slide 11:

Really, I think, this can be borne out in a study that we have been fortunate to work on, again, with Steve Rosenberg and Franco Marincola initially, and Yuan Jiang and a number of others, in trying to develop the question of whether we could find genetic separators that would allow us to predict response, particularly for IL-2 based response in patients with melanoma.

Slide 12:

Of course, it has been since 1985 since the classic paper of Steve Rosenberg came out in the New England Journal of Medicine, mentioning that IL-2 based therapy could help patients with metastatic melanoma.

Slide 13:

I have always wanted to do this. Steve Rosenberg gave me these slides, so I could show them, to actually show an x-ray, just to show a patient response, which is why, of course, we do this.

This is Steve Rosenberg, left side, untreated, right side treated sample.

Slide 14:

This is a sample of interdermal metastasis, again, after IL-2 based therapy in this woman.

Those are remarkable. As all of you know that the issue is, it has been 15 years since that. The response rate is still eight to 10 percent in these patients, to get these types of dramatic results.

It has still been impossible, at least to this date, to accurately prospectively predict which patients would benefit and, of course, that is one of the areas we would like to see.

Slide 15:

So, I won't really go into this data, but Kristin Carr, an outstanding student who is now going to be working at the National Cancer Institute, is a fellow in pathology going forward with Lance Liotta.

This really illustrates the fact that what this really shows is that data from this kind of study, in trying to look at gene expression information to predict whether or not we could pull out responders from non-responders, in the upper left-hand side it shows the sort of standard analyses that we would perform for statistical analyses of these data doesn't really separate things. They are sort of lumped together and it is really sort of difficult to get predictions.

We really have to go to really some of the efforts that were mentioned by Mike Bittner in terms of multivariate discriminators and really much more complex ways. I will show you in a moment how costly this is, in terms of computing power, to really be able to pull apart and separate the responders which, in that lower right, are shown in green versus the non-responders in blue, and develop small sets of samples that can be useful for these kinds of separations.

Slide 16:

So, with that information, we hope to be able to integrate, as Dr. Elder showed you from his, and I took this out of his presentation, again, to just mention that we want to try to combine this information with the kind of information from immunohistochemistry that lets us go back and validate against samples.

Slide 17:

Clearly, tissue arrays play a role in that. We have been fortunate to work, again, with Paul Duray and Lance Liotta and a number of other people, to develop arrays for a progression.

I know that investigators in this room have taken advantage of these, as well, for going forward, and I think that kind of high throughput technology will be important.

Slide 18:

I will mention, just in the last technologic point, that high throughput technologies to match and functionally validate these gene candidates is critically important.

We have heard several times about RNA based knock downs of individual genes. We have been interested, in the case of melanoma, for looking at all the genes that we have identified as important in terms of their potential, particularly effects on motility and knocking them down sequentially, simultaneously, in a single experiment.

Slide 19:

We have worked with one of the experimenters in our group, Spira Moises, who is now at TGen and trying to develop high throughput ways for taking advantage of the concepts of this reverse transfection of siRNAs and to cell based systems.

Slide 20:

So, what one does here is to spot a series of siRNAs that were against a series of specific genes -- for example, all the genes that we have identified as important in motility changes and relation to melanomas, put them onto a given array, grow cells over them, and look at their integrated effect into the cells and then be actually able to do combinatoric analysis based on looking at this information.

Slide 21:

Here, for a different system, it just shows a comparison that was done at CIT here at the NIH in concert with investigators now at TGen as well, to develop some of the approaches for putting this into a high throughput context, looking at these individual spots, getting the information, putting it into a format that could be looked at in a high throughput flash, and then comparing that to flow cytometry, is something that we think will really be important going forward.

Slide 22:

Let me then just end by talking about the integration of genomics into clinical trials and the importance, again, of reducing dimensionality, but then showing you how we are introducing this into trials in two different ways.

Slide 23:

The first way, in terms of reducing dimensionality has been in concert with a really outstanding investigator at Arizona State University, who used to be at Motorola Life Sciences.

Let me just mention again, that in the kind of study that is required to be able to find genes that strongly separated the response to IL therapy versus those patients who did not respond, that required, as mentioned by Mike Bittner, super computing power.

You are dealing with billions of combinations when you begin to look at pathways of multiple genes in combination rather than, for example, just the millions of combinations that you would need to look at for looking at univariate or even bivariate analysis.

Once you get to multivariate analysis of these data, we have been very fortunate to work with groups like IBM Life Sciences and others to apply those types of technologies.

Ultimately, what you are trying to do is reduce the thousands of genes down into a small manageable number of genes, certainly under 50, perhaps under 10, that best separate what you want, response from not, transition of stage of disease from vertical to radial, etc., etc., get a very small number of type classifiers that you can really predict against, and then expand that to a large population of patients.

The way that we have been working on doing this is developing ties to groups that are developing nanobiology to do this.

Slide 24:

[No text is associated with this slide]

Slide 25:

This is an example of this type of electronic array that they have developed.

Slide 26:

Really, the whole goal here is to try to develop sample to answer technology.

This is an example of the work that they have done in terms of actually you place cells on one of these chips. It purifies, lyses the cells, does the PCR hybridization, the detection.

You are beginning to see in the upper left-hand corner the filling of this trough after mixing with the microfluidics that they have developed.

Then it allows you a sample detection to be able to look against some of these small sets of discriminating genes to be able to develop this technology.

Shown in the lower right-hand side, you can come up with approaches to try to then read out this information in a very short period of time.

The cost for these, as they begin to develop them, are so reasonable that we think this is one of the ways that we will increasingly see the incorporation of this information, that is, to be able to find genes that separate, for example, response from non-response, put them into a small setting, apply them to every single patient sample that we look at, and use that information in a prospective fashion.

Slide 27:

The last way that you will hear about, I hope, over the course of this is identifying and integrating genomics into the clinical trials and doing it with a sense of urgency, something that we have to do.

Slide 28:

I will just mention very briefly one study going on in Arizona. This is not our work. This is work of Bob Gilley, Victor Firubi and David Mount(?),

Slide 29:

to try to use mulitmeric ligands to touch patients with melanoma. Again, our contribution is our data from our expression analysis to help this.

Slide 30:

Essentially, what one does there is a molecular phenotyping approach with multimeric ligands that are developed against specific moieties that would be important, of course, for either imaging or for therapeutics potentially.

Slide 31:

It is a remarkable increase in your binding efficiency if you go to multimeric ligands for various receptors instead of just using a ligand to one specific receptor, and genomic mining is what it takes to find those that are related within a spatial distance that allows you to do this type of hetero multi-ligand recognition.

Slide 32:

Clearly, again, the issue here is being able to apply and identify targets based on expression information, but then use this type of imaging as a surrogate for the type of expression information that would require invasive sampling of a particular case, and we are very hopeful that this is something that the combined efforts will make a difference in.

Slide 33:

So, I started with this, I will end with this. I think there is great hope that, as we advance some of the technologies, that we will, in fact, be able to develop simpler methods for being able to impact this disease, and I thank you very much for your kind attention.

Slide 34:

[No text is associated with this slide]