[No text associated with this slide]

TOP

So, I, too, changed the subject of my talk a little bit. As you have heard already, there has been some pulling of strings behind the scenes here at this conference, and Vern has been making little suggestions to speakers.

He suggested to me that I should make the case -- in fact, it was an assignment -- make the case that the histologic and also clinical assessment -- I added also the clinical assessment -- of melanoma is still relevant in a molecular world.

So, it is a different talk than the one I was thinking about preparing, and it makes me feel a little bit like the resident Luddite at this conference but, nevertheless, I think there are important points to be made.

These are to be made in regard to diagnostic testing, prognostic testing and predictive testing.

TOP

I think these terms are familiar to all of you.

The diagnosis, of course, is the assignment of a category that predicts behavior of a disease, to some extent, at least, and guides therapy, to some extent, at least.

Prognosis represents the identification of subsets within a diagnostic category that vary in behavior, again, in a way that defines therapy or guides therapy.
In traditional diagnosis, at least, a diagnosis and an estimate of prognosis are often necessary for management.

For example, to be told you have a melanoma, a patient, the next question is, what is the thickness of the melanoma, and there are very dramatic differences depending on that simple prognostic attribute, or in situ and metastatic melanomas are very different.

Then, prediction is the term now applied to the identification of cases that are likely to respond to particular modalities of therapy.

TOP

Now, starting with diagnostic testing, traditional histopathology uses what has been termed a hypothetic codeductive strategy.

This is a very rapid process whereby you see a pathologist or a clinician looking at a lesion on the skin as well, rapidly glances at it, and basically makes it, that is a melanoma, that is a basal cell carcinoma.

How do they do that? Well, at a glance, they recognize patterns or scenes that are very powerful in generating hypotheses, and very often that one glance or that one hypothesis is all that is needed to make a firm diagnosis.

If additional testing of that diagnosis is needed, it is done with a directed strategy of diagnostic tests using what I am calling low throughput markers to contrast with high throughput markers, some of which we have been hearing about today.

These, in fact, are mostly molecular markers at the protein RNA or DNA levels, at least now that we regard proteomics as a subset of genomics.
Not only in situ hybridization, immunohistochemistry, IHC, in situ hybridization, FISH, all of these are used in diagnostic practice, and even fusion transcript detection, and so on.

So, traditional histopathology combines an ancient strategy with the microscope with modern molecular tools that are selected tools.

I am going to contrast this, perhaps unfairly to some extent, with what I am calling high throughput molecular methods.

I am, again unfairly perhaps, characterizing them as emphasizing unselected markers and unsupervised statistical tests, eliminating the hypothesis steps above, and perhaps reducing the power, at least initially.

Here, we are talking about not only expression profiling but also CGH proteomics could be mentioned, has not yet been mentioned yet but should be.

High throughput sequencing has been mentioned, and we could talk about screening profusion transcripts. We could talk about screening for methylation status for multiple genes, and so on and so forth.

TOP

So, traditional diagnosis uses selected molecular methods. It is already molecular, as adjuncts to a differential diagnosis, based on an initial hypothesis generated by pattern recognition.

For example, here is a tumor in a stomach. Grossly, it could be a GIST, which we have heard about, a leiomyelosarcoma, more likely than leiomyoma. It could be a metastatic melanoma.

Here is histology from a tumor which could be any of the above as well. So, we can rapidly go from this scene to order a couple or three tests, such as a stain for smooth muscle actin, a stain for Melan-A for melanoma.

If it turns out to be c-kit positive, in this context, it is a GIST, and we really don't need to look at 44,000 genes to make that diagnosis.

TOP

Looking at that lesion, which is an advanced primary or metastatic disease, is really a relatively simpler case, but if we look at the Vogelgram which we have already seen defined, there are more complex problems based on the fact that the differences between these various steps are often more subtle than, say, the difference between a metastasis and a melanocyte.

The different classes of melanocytic tumors have vastly different properties, ranging, for example, in a nevus which proliferates for a limited amount of time, perhaps under the influence of a BRAF mutation, but then stops, perhaps under the influence of control factors like p16.

Perhaps when p16 is lost, inexorable proliferation occurs, but it is not until the lesions become not only invasive, but also tumorigenic that they have competence for metastasis. The molecular correlates of this are poorly understood at the present time.

TOP

Here are some pictures showing that, not only do we have this progression, but it is a compartmentalized progression, so that the remnants of the previous step are often present in a given lesion.

Here we have a compound or dermal nevus. Here we have a dysplastic nevus which is purely dysplastic, but here we have dysplasia developing at the sides of a nevus.

So, if we were -- again, we have a melanoma developing in what was histopathologically a dysplastic nevus radial growth phase, vertical growth phase developing, and metastatic melanoma.

Notice the clinical similarity between the tumorigenic phases of a primary and a metastasis. Yet, the meaning for the patient is radically different.

If we were to use an unsupervised method to diagnose this vertical growth phase, we have to be very careful that we were looking at the vertical growth phase and not at some other compartment of the lesion. So, traditional methods are going to be needed to separate these compartments.

TOP

We do have some markers that predict progression, but they are inadequate at present.

Here is what was promising at one time, beta 3 integrin expression in a melanoma with a vertical and a radial growth phase.

Here, we see strong expression of it in the membrane of vertical growth phase cells, negative expression and an associated nevus that happened to be present somewhere in here, and negative expression in the in situ compound. A metastasis from another case is also positive.

TOP

If we look at that in a larger number of cases, we see a phenomenon that applies to a number of these markers, and we have really no affected markers or combinations of markers for reliably discriminating these stages, although some of the steps can be discriminated.

I think Frank Meyskens already referred to the big quantitative leap between radial growth phase and vertical growth phase, but this marker does not distinguish between vertical growth phase and metastasis, nor does it distinguish between nevi and radial growth phase melanomas.

The radial growth phase is more like a nevus, and the vertical growth phase is more like a metastasis.

Not shown on this slide is the fact that benign Spitz nevi, already mentioned by Boris as problematical stimulants of melanoma, also expressed the beta 3 integrin.

TOP

So, if we summarize the limitations of our present methods, we can say that they need to be improved, of course.

They don't reliably distinguish stimulants, and we have many tumors that we can just say their potential is uncertain or they are atypical melanocytic neoplasms.

It is difficult to distinguish primaries from metastasis, and the utility of the marker depends on the diagnostic context.

For example, a melanoma marker could be significant if it is in a lesion that could be a melanoma, but not significant if it is in a lesion that is clearly a Spitz nevus.

The question remains open, will high throughput molecular testing ultimately supplant, or will it simply supplement current diagnostic testing.

Will we be adding perhaps a subset of the total genome to supplement the already existing process, or will it replace it as a stand-alone method.

TOP

I am going to turn to a very brief discussion of prognostic testing. What is the molecular basis of it.

Prognostic factors known at the moment include stage and attributes of the tumors themselves. Stage is a clinical finding.

The tumor attributes are presumably under genetic control, but the prognostic markers, we know, are often very complex and probably not associated with single genes.

TOP

Thickness, ulceration, read down the list, try to wave your hands and develop some idea of what the biological correlates of these might be. That is a tough enough job.

Try to associate a single gene with any of these, and it is impossible.

So, if we are going to find these markers useful and they guide therapy today, we will have to replace them with something that contains the information that they provide and, right now, that is not on the horizon.

TOP

What are the molecular markers for these complex traits? They are not presently understood. Again, will high throughput methods supplant or merely supplement these markers.

 

TOP

Perhaps most important to the theme of this meeting is predictive testing for melanoma. Predictive markers are markers that predict response to therapy.

There are none, I dare to say, presently available for melanoma. Somebody may contradict me. Predictive markers, in general, should be developed, and they should be developed for metastatic melanoma.

TOP

Why? Metastatic melanoma is the stage of progression that is responsible for almost all melanoma mortality.

We can recognize and distinguish metastatic melanomas from most other cancers with traditional methods.

The diagnosis of a metastasis versus a primary is primarily clinical, but melanoma versus others is primarily histologic.

TOP

We again use our traditional hypothetico-deductive pattern.

We can deduct that this is a metastatic tumor. We can hypothesize from some of these features seen here that it could be a melanoma. We can do confirming tests.

If we thought it might be a clear cell sarcoma, it can be a high tech molecular test. We can look for the fusion transcript to make the diagnosis.

Actually, in general, although this system works, it is this area of biology in which the high throughput methods have been most successful in identifying diagnostic categories of metastatic tumors. That is breast cancer versus melanoma versus colon cancer, and so on.

TOP

Really, we need to know more than just, is it a melanoma. A question is, are many or all examples of metastatic melanoma much the same, or are they mostly very different with respect to therapeutic response markers?

For example, you could argue that they are mostly pretty much the same and that BRAF is mutated in approximately 70 percent, and the ERK pathway is constitutively activated in most melanomas.

So, the final common pathway could be mostly the same in a majority of melanomas, we can easily detect phospho-ERK with a simple immunohistochemical test.

TOP

It is possible that inhibition targeted at any point across this pathway could be effective in treating melanomas, although toxicity may be higher based on the panoply of effects that many of these markers may have.

TOP

[No text associated with this slide]

TOP

If we turn our attention back to RAF again, if we have a BRAF inhibitor, does its efficacy depend on mutation status? I don't think we know that yet. I don't think we know if we have an effective inhibitor either.

If so, should all metastatic tumors be tested for BRAF status? We know that other tumors have BRAF mutations. If we have an effective marker, is there a case for unselected screening of essentially all tumors for all known inhibitor targets?

That may come in the future but, at the moment, we are looking at selected subsets.

TOP

For example, in a GI stromal tumor, which constitutes the majority of gastrointestinal mesenchymal tumors, it needs to be distinguished from smooth muscle and neural tumors in the GI tract.

It is actually defined and diagnosed immunohistochemically simply by the expression of c-kit, which represents an expression of a molecular phenomenon.

So, because of the therapeutic efficacy, it is important that pathologists need to have a high index of suspicion for these tumors, but based upon an index of suspicion, that is how the diagnosis is made, by an initial hypothesis followed by a simple molecular testing step.

TOP

What are the future directions? Well, the director's challenge and -- I won't read this to you, everybody is familiar with it -- suggested that these profiles will lay the groundwork for changing the basis of tumor classification from morphologic to molecular characteristics.

Do we need to throw out the baby with the bath water?

TOP

The case for high throughput tumor profiling would be that, despite the power of morphology, prognosis remains uncertain for most cancer patients.

Morphology and biological behavior depend upon the expression of thousands of genes acting together, and that knowledge of all of the molecules expressed in a tumor could be more informative than morphology alone, although morphology, that initial scene actually represents the final product of all of those molecules acting together in one fleeting image.

So, what will be the basis for future classification schemes? Will it be, in fact, unsupervised, comprehensive molecular profiling of tumors done as a black box, or will it be something that follows clinical and histologic screening evaluation followed by selective molecular profiling interpreted in the context of these findings.

TOP

At present, diagnosis, prognosis and therapeutic prediction remain multi-step processes, beginning with powerful screening steps followed by refined procedures which, again, are often at the molecular level today.

TOP

In the future, it is my assertion that this will not change radically. It is likely that histologic or other morphologic screening will continue to be useful as a screening step.

A point that I haven't really mentioned yet is to assure that the sample we are studying is tumor. It is a clinical and morphologic diagnosis often backed up by histology to be sure that we are looking at the right tumor and, not only the right tumor or the tumor itself, but also the correct compartment of the tumor.

We may need to purify our tumors with microdissection, possibly sticking a needle in the tumor aided by high resolution imaging, which is a step that could bypass traditional histology but still remains a traditional morphologic step.

The screening will be needed to select a diagnostic strategy appropriate for the differential diagnosis. That may be low throughput, medium throughput. As we suggest here, there may be subsets of the genome developed as black boxes. They might be chips, they might be something that we will reach off the shelf when the differential diagnosis is melanoma versus something versus GI tumors versus.

Other chips may be used in a somewhat unselected way against all metastatic tumors to see which of them express which therapeutic targets.
I would say that unsupervised molecular diagnosis is most effective in advanced tumors, perhaps because, although their genomics are complex, their morphology is more homogeneous.

I think that as we apply it to early disease, which hasn't been done yet, simulations and small differences between progression steps will confound the process to some extent, but I do feel optimistic that targeted testing along the lines of the hypothesis followed by the testing of the hypothesis will gradually improve the specificity of the diagnosis in these areas.

TOP

So, the question Vern proposed was, can histopathologists help molecular biologists develop targets based on a clearer understanding of the disease, rather than simply by brute strength analysis of thousands of genes, and I don't think that we are looking to say that the analysis of thousands of genes is inappropriate.

I think the answer is clearly yes.

I think the answer is that high throughput screening technologies need to be applied in the context of initial screening steps that are not only histologic but also clinical. Thank you very much.

TOP