Melanoma 2003

SLIDES & TRANSCRIPTS
Monday, May 5, 2003

Profiling Melanomas: Technology and Approaches

Michael Bittner, Ph.D.

Slide 1:

This is certainly a topic that is much on our mind. Many of the people who went out to join the translational genomic research institute are very interested in, how do you run things all the way from basic science into applied clinical use.

One small part of that is what you can learn from profiling. The questions we like to think about are, what happens in the wiring? What genes are coordinated and how?

What processes are acting? We would like to know what is going on, so we can try to make some sense out of the diverse things that we see happening.

Then, there is this whole question of, if you could see into this really well, would you be able to put together some notion of a diagram of how things are connected, and are there switching elements, major genes that have a wide regulatory influence.

Slide 2:

We are going to go in two parts. One, what is the state of the art now? Mainly, it is a study of similarity and differences.

Slide 3:

Transcription is a result of a bipartite system. The genes themselves specify a lattice on which various factors and effectors bind. They serve as points of assembly for even larger complexes.

The cell, on the other hand, provides a context that drives what is going on at those lattices, what is going to happen next with the gene.

The problem for all of this is that the product of transcription ultimately alter the regulatory context itself. So, this is a closed and cyclic loop which offers some problems for interpretation.

Slide 4:

So, we like to think of genes as little thermometers, a read out device.

Slide 5:

You have promoter sites which interact with the cell. You have a particular cell and that gene, when placed in a cell of a given context, produces a typical level of transcript.

Slide 6:

Different genes having promoters that are essentially the same, will respond in similar patterns to the cellular context when in the same cell.

Slide 7:

Genes, therefore, sample the cellular context.

Slide 8:

Different contexts produce different results. So, you have got a readout that is very deep. There are a lot of genes. If you follow what they are doing, you can say a lot about what there is going on in the cell.

Slide 9:

The way this is typically inferred is, when you are thinking about the patterns of promoter response, you just hold the samples constant and you ask, are there patterns of gene activity -- so, these are the genes along this side -- which repeat in different samples.

Those genes which have consistent patterns are linked in some way. They share information about the process, driving transcription.

It is easy to get caught up in looking at these things. These are MHC responsive elements in melanoma, and these are proteolipid genes and lipid synthesizing genes, and so, you need something to constrain your curiosity and keep you from going too far afield.

Slide 10:

You can also look at it in the opposite direction. So, the cell's context can be assessed by sampling many different genes.

Slide 11:

In this case, we hold the genes constant and we group the cells by the similarity of what has happened with the genes, and these two different kinds of domains show that there are basically at least one consistent domain of cellular context, and many other ones that are inconsistent in this set of samples.

Slide 12:

So, given all of those relationships, what can you do?

Really, the problem now is, how do you correctly form a question that you want to ask, so that you can say something about melanoma.

It is all pretty simple. You have got these profiles of samples.

You do some sort of analysis to clump the samples either by clinical type -- you can know something about them ahead of time -- or you can clump them by molecular type, what I just showed.

Then you can use either linear methods, all the sons and daughters of F&T testing, rank testing and that sort of thing, and you can use non-linear methods, genes at work, coupled two-way clustering, strong feature analysis, to build up a ranked list of genes which have the power to differentiate those two classes.

You have got this gene list. You can use that to study what is going on with the samples. Can you build classifiers for the sample so that you know, based on a particular gene's action, what kind of sample that you are looking at, the GIST story Paul just told you?

You can identify process components, what is going on in this cell. Some particular phenotypes have readily identifiable molecular consequences, and you can try to find explanatory candidates. Given the most powerfully discriminating genes, do they drive these processes?

Slide 13:

So, a now published study that was done by Yuan Jiang and Ashani Weeraratna looked at what can you find out about that first clumping of cells that I told you about.

One of the main differences was WNT5A was a very discriminatory gene between the two of them. You look at the process.

The process profile looked like, when WNT was up, invasive sorts of characteristics -- proteases, extracellular matrix production -- were up, and in the end, they went all the way down to biochemistry and could show that, typically, here is the leading edge of an invasive cell. The red staining is high WINT-A expression.

They could show that this molecule looks, for the all the world, like something that drives invasion.

Slide 14:

Another way you can go is look at these patterns, now you don't know ahead of time what is going on.You look at the patterns of molecular subtypes.

As you accumulate more and more samples, you become familiar with these and you can say, ah, among these NCI samples, given to us by Dr. Rosenberg's lab, Franco Maricola and Nina Wang, this looks odd. In fact, it looks like something we have seen in other studies.

What is the difference between these two? How good is that discrimination? Well, the discrimination is huge. These patterns run for 600 or 700 genes. The SWOG samples are easily pulled apart from the NIH samples.

Slide 15:

The largest clinical difference between those was the survival, average survival, and the most discriminating gene was a pro-apoptotic gene.

This little section here shows high staining in dav for that gene and, as you can see, the nuclei are tending to disappear in here.

This is interesting because, as was said before, melanoma tends to start out having edged out apoptosis as a control for it.

We think that, in these patients, this gene provides a route to reactive apoptosis in the presence of those mini-IAP and other barriers to apoptosis.

Slide 16:

So, you would really like to do more than just cataloging these things. You would like to have some way to link these in a causal way, the directness problem that has been mentioned before.

Slide 17:

What kind of genes could you see where they would change the regulatory context sufficiently that you could actually see the consequences of their expression in the absence of a time dependent series?

Another problem with melanoma is that it is hard for us to get a nice longitudinal study. We look forward to start using the arleto model mouse and look at this.

You might expect that the genes that dictate the level of transcription factors or effectors, or that perhaps change their actions or influence the accessibility of the target genes to their action.

Slide 18:

Do we think those are running biologically? Well, one of the oldest and strongest thoughts about it was Waddington's conjecture of canalization.

He says, look, when development happens, you can have huge sons and daughters, small sons and daughters.

However, the process is always driven in a way that produces the desired end result in terms of the number of features and the workableness of the embryo.

We see in recA and lexA in E. coli and p53 in mammalian cells controls that can be brought up when the cell is in mortal danger, that act for repair, to drive repair and such, which you might expect that one of these big effectors would be induced in extremes.

Computationally, it makes good sense. At the beginning of last century, they were thinking about what would you have to do to have effective computation when you have got lots of input.

That feeds back into Waddington's conjecture. It is easy to get paralyzed by lots of input. We run up against it ourselves.

Kaufmann, who has looked at what you might need to do to do effective biological computation, for many years now finds canalizing functions in binary simulations that look like they might apply to biology. These were later recognized to be a set of post-production functions.

Slide 19:

So, if you had a regulatory context switch, you would expect it to be invoked in some context. It's state wouldn't be readily correlated with other genes.

When operating, you might expect to see the states of many genes making that state predictable in a multi-gene predictor.

Slide 20:

So, let's think that we have three cellular contexts here, and gene three, we are going to posit, is our big contextual switcher.

In the first two contexts, the behavior of the other genes are not in any way -- you can't write a simple logical rule that predicts the behavior of gene three. In the third one, you possibly could, because now you have something on there.

Slide 21:

When we actually look in data -- now you have to use super computers and all to start to find out, well, how many genes that don't intrinsically predict at the level of correlation do predict when put into multivariate things -- when this gene is off, here is a logical truth table.

If gene one is in state -1, gene two in -1 and gene three in -1, what is the state of the target gene. The important thing is that you see many possible ways of having the gene be zero, but only one set of gene predictions, all of which predict exactly the same thing, work when the gene is on.

We have thousands of these tables for the melanomas we are looking at and they all work this way. Over 300 independent genes seem to be affected. So, we think this is an important context gene.

How would you look at it?

Slide 22:

One way is, this gene is on in many of Dr. Rosenberg's patients. So, if you look at the total survival of all the patients he has given us to look at and then stratifying it based on the staining of that gene, you can see, in these Kaplan-Meyer curves that this gene correlates quite well with longevity.

Slide 23:

In fact, you can look all the way back to the early days of this cancer. Dr. Duray, of Dr. Rosenberg's group, a pathologist, has had a look at radial growth phase cells that are going over into actual overt melanoma.

There is no staining for this gene in the RGPs, but in the overt melanoma, they actually stain quite heavily.

This is a marker for the transition of these cells. It is an anti-proliferative, which would be a reason for it to be co-determined with life expectancy.

Slide 24:

In many cases, it allows you to look at survivors way out there.

So, here is a patient who has survived for almost 20 years with stage four melanoma, who comes in every year or so to have pieces of melanoma removed.

If you look throughout the history of that patient, you see that constantly this gene is on in the affected areas, the melanomas.

Slide 25:

So, we know the transcriptional activity reveals information about how cells are wired up and what kind of states they are in.

We know that you can identify differences in these states, and you can make testable hypotheses out of this, and validation is very important. You need to be able to go back and say something about what you have speculated on.

The study of these can show you genes with wide regulatory influence, and integrating mechanics of biological regulation with the signal processing and computational processes that have grown up in the last 100 years as people have thought about how might you compute effectively, is one way to delve into biological systems.

This certainly is no panacea. It gives you very focused observations but, in some cases, we think that these may be - WNT5A is one of those things that transmits through a GPCR and is, therefore, very drug-able. These other genes may be modulate-able. They may ameliorate the effect of cancer. Thanks.

Slide 26:

[No text is associated with this slide]