SLIDES & TRANSCRIPTS
Wednesday, February 16, 2000

Can Treatment Be Tailored to Biologic Characteristics?
Peter Danenberg, PhD

Dr. ABBRUZZESE: I would like to invite Dr. Danenberg to come up now. I think one challenging technical issue, just while people are getting settled, is the use of core biopsies. We have done a lot of that ourselves, and there is going to have to be some significant effort put into quality control of the core biopsies. I can tell you that when you look at some of these, depending on the marker you want to look at, you may be biopsying necrotic material. A lot of attention, in terms of the technical aspects in working with the radiologists, trying to get the peripheral rim of viable tissue, is going to have to take place. That is a potential problem.

Dr. DANENBERG: The presentation is really a follow-up on Len Saltz' presentation. I don't have to give the introductory slide. I will save some time, but I just wanted to make a few other points that I think are important with this approach to improving cancer chemotherapy. First, I would like to discuss some technological issues. You may be wondering why we are measuring gene expression rather than protein expression. For measuring tumor response determinants to drugs you can measure either the relevant protein or the gene expression.

We chose gene expression in large part driven by technological considerations. To measure protein expressions precisely for a large number of proteins in small biopsies is still difficult. I know there is immunohistochemistry, but that is semiquantitative and subjective a lot of times. You may not have the requisite antibody available and, to this day, there is still no good antibody for DPD and others for TS. But if you want to study a new protein which you may not be able to do, gene expression is on the other hand driven by really dramatic technological advances in recent years, and it can be done for practically any gene for which the sequence is known. About 10 years ago, when I started this, we decided to do quantitative RTPCR, and the principle, of course, is that ultimately the protein level would reflect the gene expression level.

Of course, I know that is not always true, but I think in most cases it is true, and when we started this out we did this manually. We did a manual RTPCR. We got the PCR products. We ran them on a gel. We did serial dilutions, cut out the pieces from the gel with scissors, put them in liquid scintillation vials and counted them. It was very laborious, and we were getting discouraged, and we thought this project would never be done. Then a few years ago there was a real technological advance by the Perkin-Elmer Company and that was real-time RTPCR in a quantitative fashion.

This illustrates the principle of it. This utilizes two primers and a probe. This probe sits in the middle of the two PCR primers here in yellow. The probe has a quencher on it and a fluorescent probe and when they are sitting together in the DNA strand, the fluorescence is quenched, but when the strand starts getting extended by the tac polymerase this probe is cleaved, and the fluorescent molecules are released and the machine then picks it up.

For each PCR cycle you get an increase in fluorescence.

TOP

The data look like this. These bars here each represent a PCR cycle.

It is at this point above the threshold where the machine starts to detect the fluorescence. Each point here represents one PCR cycle. The further down the graph the signal starts getting detected, the less of the starting molecule there is. You can take the ratio between these threshold values and get a gene expression ratio or quantity of mRNA.

There are two samples here. This is beta-actin which is our internal reference gene, and these are two PS values. Here it is separated by 10 cycles or 2 to the 10th power,

TOP

and each PCR machine has 90 wells. You can do 90 PCR reactions in one run. You can probably do three runs in one day. That is almost 300 PCR reactions. This kind of technology makes the whole thing, this type of approach, feasible.

TOP

This is the slide that Len showed. This shows the low gene expression of TS and DPD can identify most of the responders in this group of cancer patients.

The points I want to reiterate are the following:

TOP

The first point is that these gene expressions can be used to predict survival because of the tight association between response and survival. All the patients with low TS and DPD values had considerably better survival than did those with one of those gene expressions being high.

TOP

You can use gene expression not just to predict response but also survival. The other point I wanted to make was that even though we had a small set of data, and I realize this is not a large number of points, these data are reproducible and have been done in other laboratories. About 2 years ago I was invited to Japan to make a presentation. I was invited by the Tiho(?) Company to make a presentation at some of their medical meetings, and I presented our TS and DPD data and recently one of the people who was there came through my lab and presented a seminar and, by an amazing coincidence, they started studying TS and DPD expression tumors, and here is their data of TS versus DPD. So they found the same thing with DPD, that high DPD expressors didn't respond to 5-FU.

Their data with TS on this scale wasn't quite as nice. Some of the what they called high TS were actually in the responding group. However, and this isn't apparent, their scale of TS isn't as high as ours. They didn't have the high TS expressions that we did, but nevertheless that is just a reiteration of the data.

TOP

They divide their data into quadrants, but those with low TS and DPD had an 86 percent response rate. Patients with high TS and high DPD had zero response rate and their so-called high TS's still had a response,

TOP

but they got almost the same survival curve that we did. The low TS and DPDs had a much better survival than the high TS and high DPDs. The data were reproducible in their hands, also.

I borrowed their slides, with permission, but they divided this so the potential quadrants of drug treatment, the low TS and low DPDs who got 5-FU therapy, the high DPDs might get 5-FU with a DPD inhibitor or alternately you might be able to give an antifolate based TS inhibitor like Tomadex, which should not be affected by DPD. In fact, we have shown that. We have some data about TS gene expression relative to a response with Tomudex, and indeed the high DPDs still do respond to Tomudex.

When I talk to these companies, sometimes I try to convince them that here is a potential market for their drugs. If they could identify all the colorectal patients with high DPD they would have a big market for their TS directed antifolates, but so far they haven't taken me up on that idea.

TOP

his quadrant we could treat with other drugs like CPT-11 and here, too, with the high TS. This is another one of the Japanese slides. One valuable thing they did was to measure DPD expression in different tissues. Here are primary tumors, liver mets and normal liver. There is much higher DPD expression than the tumors. This speaks to the importance of preparing the tissue properly.

In this case if you take the liver biopsy and you have normal liver in biopsy assay it is going to confound your DPD expression of tumor. So we get a false reading. In that case, it is very important that we try to make sure that you have mostly tumor tissue or entirely tumor tissue. It is not quite so important with TS because TS expressions generally in the normal tissue are similar to the tumor or a little bit lower. I have never seen one that is actually higher than the tumor.

TOP

This is another TS response slide from Germany, yet from another group.

TOP

They measured response in hepatic artery infusion of these drugs, another way of giving it and the high TSs, the progressing group all had high TSs which is kind of interesting. Just the progressors had high TS. There appears to be something biologically different about the tumors with TS values. That is the point I wanted to make, that these data are reproducible by other groups,

TOP

and the second point was that this also works in other tumors besides colorectal.

This is a study we did in gastric cancer published several years ago. Here we measured TS gene expression and ERCC 1 because these tumors were treated with a mixture of drugs, 5-FU and cisplatin, and we weren't sure how this was going to work out, whether TS would still be predicted with a drug combination involving 5-FU, but it was, and ERCC 1 was also predictive and ,just like the TS DPD graph, the responders here all have low values of either TS or ERCC 1. If you take this quadrant here you can pick out about 80 percent of the responders by TS and ERCC 1 values.

The rest of the time I will talk about our most recent work in esophageal cancer.

TOP

I saw these data in the literature on esophageal cancer. There were some small pilot studies out there where they showed that if esophageal tumors are treated neo-adjuvantly with chemoradiation, then a fairly high percentage of them were pathological complete responders, and those patients had a clear survival benefit whereas the others didn't.

I thought it would be an interesting study to see if I could identify those pathological complete responders beforehand just like we did with the colorectal with TS. These are the path CR patients and these are the known path CR patients. I was fortunate to get that study funded, and

TOP

these are some preliminary data we had with esophageal tumors from USC, but we divided them in the adeno and squamous groups.

It is interesting to note that the squamous tumors didn't have any high TS values. The TS was not diagnostic of response for the squamous tumors whereas the adenocarcinomas had the same range of TS that the colorectal cancers did up to about 20, and here TS was seen to be a diagnostic response and we had about the same response cutoff with the colorectal tumors, and we thought originally this was due to a difference in biology between squamous and adenocarcinoma, but most of these squamous, all of them actually were from Japan. A postdoc of mine brought them over for analysis. I am beginning to think that maybe this could be a racial characteristic of Japanese. Remember, I said in the previous slide that the TS was not quite as diagnostic in colorectal because there were no high TS expressors. It is the same phenomenon here, but of course, that needs to be investigated further.

Anyway, when we started the study, I proposed to get some fresh frozen specimens from a number of people who we are collaborating with, and I soon found out that that wasn't going to work very well because accrual was very slow, and I just wasn't getting enough specimens to do the study. That is when we developed this paraffin extraction procedure for RNA. It took us about a year to develop that so I could get it back in the archival samples and having done that I was able to approach a number of different investigators and basically ask for their specimens that they had stored up.

TOP

That is one of the problems with these kinds of studies. It was difficult to do.

I mean the people would agree to the study and they would say that it was very interesting, but the specimens -- it is not like they are located in one drawer somewhere. A lot of times they are in different community hospitals, and we have to go locate them and call the pathologists and persuade them to send the specimen over. We would send them a Fed Ex box and eventually we got a lot of the specimens we wanted but not all of them. So, it was difficult.

TOP

This is a group of patients from Munich actually. We actually had to travel there and get the specimens personally, and even then it was difficult. We had to hire some students to go in the basement and dig them out.

These are esophageal patients treated with 5-FU/cisplatin, and this is just a graph of survival versus response. The responders had a better survival than the non-responders, and so basically we didn't always have good response data. From now on most of our plots are in terms of survival, but it was possible to separate these patients into good and bad survival by their TS values, low TS and high TS. These are the Munich patients. There are 35 specimens here, statistically significant.

TOP

This is another group of patients that we got from North Carolina. They are Harfold's group and he treated these patients neo-adjuvantly with 5-FU/cisplatin radiation and we were able to pick out the pathological complete responders. They all had low TS compared to the others, statistically significant. We are trying to get some more of these.

These are all leftover specimens from studies. They had already cut up these blocks to a large extent. We had to take what was left over.

TOP

This is data that Harfold got by immunohistochemical staining. He tests a large number of genes. This is the only one that was actually correlated with response, glutathione-S-transferase, and so we did the same thing by gene expression

TOP

and got essentially the same results, slightly better separation by gene expression.

TOP

This is another set of patients from Munich treated with taxol 5-FU/cisplatin. We were interested in taxol because that is one of the drugs that is going to be used in esophageal cancer quite a bit, but here we tested beta-tubulin-3. There are other studies by Susan Horwitz that have shown that the tubulins were predictors of taxol response in breast cancer. We thought we would try it in esophagus, and low levels of tubulin-3 predicted for better survival here.

TOP

This is metablastin. This is supposed to be another taxol determinant. This time high levels of metablastin predicted for better survival and response. So, this is sort of the conclusion.

TOP

I think we have data that suggests that specific molecular markers may be useful in rational selection of chemotherapy and that is just my standard conclusion story,

TOP

but for this I just have one transparency that I want to show. This is a slightly better conclusion slide, but I think from our data so far it appears that it is possible, well, we know it is possible technically to measure candidate response determinants in pre-treatment biopsies from paraffin mainly. That is the main advance here.

We think it is possible to identify markers that predict tumor response or non-response to drugs. It is possible to identify single markers such as TS that predict response to multidrug therapy like 5-FU/cisplatin. The reason I bring that up is because people ask me a lot of times with all these drug mixtures, how can you hope to pick out tumor response determinants. Well, you can ask the question whether TS is a determinant for 5-FU oxali, for example, because that synergism may depend TS expression and effective inhibition of TS by 5-FU, and using two markers is much better than one, but they have to be independent markers, obviously. They cannot be co-regulated markers. Using TS alone we can pick out about half the responders. Using TS and DPD we can pick out 90 percent of the responders, and then I was asked to talk about the future a little bit.

Obviously these are preliminary data like Len said. We need larger studies where we nail down the response/non-response cutoffs for these determinants more precisely. We can do it more precisely now using the Tachman(?) and maybe even other technologies like array chips or something, but it is difficult. If you can identify these trials, sometimes it is difficult to get the specimens. For 5-FU we may have a trial in the Nordic Oncology Group. They said they would make available to us several hundred specimens treated with four, five, six hundred milligrams per ml of 5-FU. There we should be able to get much better TS or more complete TS data and of course the other thing, like Harold Johnson, is to implement the proper clinical trials that demonstrate actually that preselection of patients by their markers will have an ultimate benefit for cancer chemotherapy.

We predict that for selection by TS alone we should increase the response rate from about 20 percent to 40 percent. We should double the response rate for those patients with low TS and then of course, finally, once this is done you have to gain acceptance for the use of predictive markers for treatment of patients, but that is in the future after all this is done presumably.

Thank you.

(Applause.)

TOP