Mar 21, 2011
Jacquelyn K. Beals, Ph.D.
March 21, 2011 — Editor’s note: The first report of the sequencing of a complete cancer genome appeared in Nature in November 2008. Researchers at Washington University in St. Louis, Missouri, identified 8 previously unrecognized mutations in tumor cells from a patient with acute myelogenous leukemia (AML) that apparently contributed to the cancer’s proliferation. By June 2010, the American Association for Cancer Research noted the publication of 70 whole cancer genomes or exomes that have the potential to shape therapeutic choices.
In an interview with Medscape Medical News, Elaine R. Mardis, PhD, co-director of The Genome Center and associate professor of genetics and molecular microbiology at Washington University School of Medicine, talks about her recent presentation on whole-genome sequencing in breast cancer at the The Future of Genomic Medicine (FGM) IV conference, held March 3 and 4 in La Jolla, California, and sponsored by Scripps Translational Science Institute.
Medscape: You’ve presented at the FGM meetings for the past 2 years. Was your topic this year closely related to the material you talked about last year?
Dr. Mardis: The presentation that I gave was really a follow-on to a slide in last year’s talk, in which I had primarily focused on our work in AML. I also introduced the latest study that we’ve been working on in breast cancer — sequencing tumors that have been accrued in a clinical trial. The clinical trial is essentially randomizing women with estrogen-receptor (ER)-positive disease to 1 of 3 commercially available aromatase inhibitors. Pretreatment, we have a biopsy taken from the breast tumor; that’s the source of the DNA for our whole-DNA sequencing approach. The women then go on to 4 months of treatment with 1 of the 3 aromatase inhibitors; at the end of that 4-month treatment program, they go through surgery to remove what remains of the breast tumor.
The postsurgical workup of the tumor includes an immunohistochemistry assay that looks at a specific protein — Ki67. Ki67, through immunohistochemistry, is a measure of the proliferation or lack thereof, depending on the amount of Ki67 protein detected by the antibody reagent.
Using that Ki67 index, we stratify the patients into responses — that is, patients who have a low proliferation index, so their tumor is being shut down by the removal of estrogen (the action of the aromatase inhibitor) vs unresponsive patients (or resistant, depending on what you want to call them), who have a very high proliferation index. Their cells are still rapidly growing and dividing, apparently unaffected by the aromatase inhibitor.
In our whole-genome study, we’ve taken approximately equal numbers of responsive and resistant patients, a total of 50, and we’ve sequenced the whole genome for all 50 cases.
This includes a second genome, which is a blood normal, to help us identify which of the tumor-specific mutations are in fact somatic, and which of the mutations we identify in the tumor are unique to the tumor tissue.
Ultimately, we’d like to have a genomic predictor of response to aromatase so we can fashion a new clinical trial in which we use this genomic predictor. That would involve mutation assays directed specifically at the genes — and maybe specific areas of genome amplification — plus an algorithm that would take the results of that assay and get a probability of response to the aromatase inhibitor.
Women going on to our second-generation trial could be selected for response to aromatase inhibitors before going on to that trial. Another possible arm of the trial that comes from the genome sequencing is that, in this genomic assay that we’re looking at, we would identify genes for which targeted therapies are now available.
A great example of that is a gene called PIK3CA, a well-known kinase that’s highly mutated in breast cancer. Two or 3 phase 1 clinical trials are going on right now for agents targeted at those mutations in the gene. One arm of the trial, for example, would use an aromatase inhibitor only; the second arm could use a combination of an aromatase inhibitor plus PIC3CA-targeted agents in the patients who have these mutations in their genomes. Of course, that would stratify them into that arm of the trial selectively.
Medscape: Your study divided breast cancer patients into 2 approximately equal groups: one that was sensitive to an aromatase inhibitor and one that was not sensitive. In the general population, how common are tumors that are aromatase-sensitive?
Dr. Mardis: It’s my understanding that about 70% to 75% of all breast cancers are ER-positive. You would never, of course, use an aromatase inhibitor on an ER-negative tumor, such as the triple-negative or basal subtype. So of that 75% that qualify based on ER status, probably somewhere in the neighborhood of 48% will respond to the drug with tumor shrinkage. A much smaller proportion, maybe 10% of that 48%, would respond with what’s called a “complete pathologic response.” Their tumor would essentially not be there after the neoadjuvant therapy period.
Medscape: That’s still a significant number of women, isn’t it?
Dr. Mardis: Absolutely! And I think the flip side of looking at it is the fact that, like most drugs, aromatase inhibitors are not without side effects. For women who won’t respond, there are 2 factors at play. First, why would you put somebody on a drug if they’re not going to respond, and risk having possible unwanted side effects of the drug? Second, why would you spend 4 months treating them with an agent that isn’t going to work when you can be treating them with something else that’s much more effective? So part of the screening that I’m talking about would be directed at women who will respond to aromatase inhibitors, and of course that’s of interest. For women who come out negative or with a low probability of response, we would have to come up with some sort of regimen to deal with those negative responders.
The clinical trial that we’re using for these samples accrued more than 370 women. We’ve only studied 50, so we have plenty more to go back to. Once we think that we have a good assay and a good predicting algorithm, we plan to go back in a blinded fashion to additional samples from the trial, run the assay, run the predictor, and predict the response vs the resistant groups. Then we have the Ki67 scores for all of those women. We would be blinded to them at the first pass, but later we would have the ability to see how sensitive our predictor actually is and how often it’s right. That might require a couple of iterations to add in additional genes — or even reconsider the whole thing! I like to think we’re not that bad, but we’ll probably need a couple tries to see what’s effective before designing a clinical trial around it.
Medscape: Your research career has focused on sequencing, and breast cancer is one of many things that you’ve looked at. Did your presentation on AML last year also deal with cancer-genome sequencing?
Dr. Mardis: The first genome that we sequenced was an AML genome of a patient [Nature. 2008;456:66-72]. It has been our focus for whole-genome sequencing. We started with leukemias, primarily because they tend to be fairly quiet, genomically speaking. In particular, the M1-subtype AMLs that we focused on are highly diploid genomes. They’re commonly very frustrating for hematologists/oncologists because they comprise the majority of patients.
About 61% of AML patients have what are called intermediate cytogenetics. When cytogeneticists look at their chromosomes under a microscope, which is the standard of clinical care, they look highly diploid. There are no clues there as to whether the patients will be successful on the commonly used therapy. The goal is to knock out the immune system, called induction therapy, and then allow the patient to rebuild the immune system.
The converse of that is that some 20% of patients actually have a known translocation. For example, in M3-subtype AML, you have a t15:17 [translocation between chromosomes 15 and 17] that’s very well understood. It fuses 2 genes together [producing PML-RAR alpha fusion protein]. Patients go through induction, they’re consolidated with all-trans retinoic acid (ATRA), and almost every patient responds successfully to that treatment — they fall into what is called a favorable risk category.
The absolute opposite of that is adverse risk. Under the cytogeneticist’s microscope, these patients have very highly rearranged and deleted chromosomes. There’s a lot going on in those genomes and they have absolutely the worst possible outcome. So you can stratify the 20% of either one of those pretty nicely. It’s that middle 61% that we’ve been looking at since about 2002 — since 2006 with next-generation whole-genome sequencing — to try to find prognostic indicators so you can test for 1 or 2 genes and say: “This patient, based on all the patients that we’ve studied, won’t do well and will need more aggressive therapy going in.”
Late last year, we published a major finding on a gene of that sort [N Engl J Med. 2010;363:2424-2433]. DNMT3A, on Kaplan–Meier analysis, absolutely predicts patients will have a poor outcome if they have that mutation, normal cytogenetics, and no other features considered. That’s sort of the first big prognostic indicator.
As you can imagine, there are multiple investigators of AML with hundreds of samples in the bank who are now looking very, very hard at DNMT3A. Actually, a paper just came out in Nature Genetics [published online March 13, 2011] from a Chinese group looking specifically at M5-subtype patients and DNMT3A itself. Basically, they’re finding exactly the same thing we did, although we looked across most of the FAB subtypes [French-American-British classification system for AML]. We didn’t just focus on 1 subtype.
Medscape: Is there any difference between the approaches used for cancer-genome sequencing and for regular whole-genome sequencing? Or is it just the same technology but applied to genetic material from malignant cells?
Dr. Mardis: Basically, to understand what’s unique to the cancer, you have to have the comparative normal from that patient. So you’re really sequencing 2 genomes when you do the study. Other than that, the approaches are exactly the same. Of course, where life gets complicated is in the bioinformatics analysis!
Medscape: Cancer cells are rapidly dividing and not exactly “under control,” so do you find more variability between the genomes in various parts of a tumor than you find between healthy tissues of a person’s body?
Dr. Mardis: I think we have ways of looking at that, but I don’t know that we’ve been incredibly systematic about it. To be fair, one reason that AML is a little bit easier to study in that regard is that it’s a liquid tumor. We can look at very high coverage, go into those specific mutations once we’ve done the whole-genome sequencing, and study them again but at much greater depth.
We normally sequence about 30-fold coverage — oversampling as I see it — when we do the whole genome. When we do very high depth, we often look at 1000-fold coverage. You can then do a statistical analysis that tells you which mutations in the cell population are the oldest, which ones have been around the longest, and which ones are fairly new. This is one way of looking at the putative drivers, because the big looming question in cancer genomics is, of course: “What starts the tumor?” But it can also tell you, using a different kind of principal-component analysis comparing tumor and normal, how many subclones are in a population. That heterogeneity is really the question you’re asking.
We commonly think of solid tumors as being more heterogeneous than liquid tumors; we’re a little surprised at the heterogeneity of liquid tumors as well, but it’s not as broad. One thing we’re going to do, as a check of the clinical trial that I described earlier in breast cancer, is when those women are biopsied pretreatment, we actually take 2 biopsy cores from their tumor. Part of that is a little bit of “insurance,” because you want to make sure you actually core the tumor. That turns out to be technically a little difficult, and sometimes the surgeon will miss it, so you need to double your chances. But in some cases, we have 2 good-quality biopsies with lots of tumor cellularity. So one thing we plan to do, now that the sequencing is done and we’re in the analysis phase, is to go back and reexamine that second biopsy to see how similar or different it is from the first one we studied. I think that’s an interesting approach.
The other thing I talked about, which is still underway, is that we also have that resected tumor mass to look at [from the end of the 4-month program]. So, after the 4-month aromatase inhibitor, the tumor is removed. For those women who still had a discernable tumor mass, we’re planning to do whole-genome sequencing on those and evaluate what influence the aromatase inhibitor drug had on tumor cell heterogeneity. I think that’ll be very interesting, particularly in the context of responsive vs nonresponsive patients.
We predict that oftentimes women with breast cancer, because of the heterogeneity of the disease, may present with largely ER-positive disease. But there may be components — that may or may not even show up on immunohistochemistry — that aren’t incredibly sensitive, that are more basal-like, or more ER-negative.
In some cases, tumor cells may even be progesterone-receptor (PR)-positive but not ER-positive. One of the things we really want to get at by sequencing these resection samples is what changes.
One possibility is that maybe the aromatase inhibitor works largely by shutting down the majority of tumor cells that are ER-positive by removing estrogen from the body. But that could allow a subclone of cells, which aren’t responsive because they’re not expressing estrogen receptors, to actually flourish. There may be competition for survival, and by knocking down that predominant cell population, you may encourage a minor cell population. Part of what leads to that speculation, quite frankly, is that you have these women I described earlier, the 10% or so, who show a complete pathologic response. Maybe their tumors were highly homogeneous, while the rest have a heterogeneous tumor mass, so most of it disappears, but not all of it. That might be the factor differentiating complete pathologic response from partial response.
Medscape: One use of the cancer-genome information you’re getting is for selecting cancer therapies. Beyond the therapeutic choice, which is clearly important, will cancer-genome sequencing help us understand or prevent or halt the development of cancers?
Dr. Mardis: In the long run, yes. What we really want to understand by looking at these patterns of response is: Are there combined therapies that would work? For example, if a tumor mass is analyzed with immunohistochemistry, which is very commonly done (ER, PR, and HER2 evaluations are done for every breast cancer that’s removed), you may get clues, even from those gross-scale imaging approaches, as to whether the cancer has a complex nature. In a case like I was just describing, you may want to use an aromatase inhibitor to knock down the ER-positive component of the disease, followed by ACT [doxorubicin (Adriamycin), cyclophosphamide, followed by docetaxel (Taxotere)], a commonly used standard of clinical care for women with basal subtype or triple-negative disease. We don’t know those answers yet, but it’s fairly clear from some studies our collaborator Matt Ellis has done.
Matt’s an expert at using RNA expression for subtyping breast cancers (luminal, basal, HER2-positive, and a few others). He tends to look at the RNA expression pattern for the pretreatment biopsy vs the RNA expression pattern for the resection sample. It’s not what I’d call common, but we have a lot of good examples in which the pretreatment biopsy shows luminal A or luminal B ER-positive disease and the resection sample may show basal or HER2-positive disease. That speaks to the heterogeneity becoming less in the resection sample, to the point where the apparent subtype has even changed! I think we need to tease out some of the nuances there. For example, when tumors show a mixed population of cells on the risk predictor that does subtyping using RNA expression, is that a little bit less confident of a call? Some sort of background is being contributed by that minor tumor cell subpopulation that’s not luminal B, or vice versa. There’s more to do on that and it’s going to require a lot of careful laser-capture microdissection, sequencing, and classification comparing the pretreatment and posttreatment biopsy samples. There’s a lot more to do!
Medscape: In the panel discussion at the end of that session, what questions or themes did people want to pursue at more length?
Dr. Mardis: The person who really stole the show, and rightfully so, is the guy from T-Gen [Translational Genomics Research Institute], Dan Von Hoff [whose talk was entitled The Oncologists’ 6th Vital Sign — A Context of Vulnerability]. The 5 vital signs are the 4 that you think about, like blood pressure, heart rate, etc.; for cancer patients, you add a fifth, which is for them to gauge their level of pain on a scale of 1 to 10. Van Hoff went through all of this, which is great, and he knows his business. Then he said the sixth vital sign that he wanted to talk about is whole-genome sequencing for diagnosis.
Not to steal his thunder, but he gives a beautiful presentation because he’s lived it. What he’s doing is taking end-stage cancer patients for whom all treatment options have been exhausted — this is, of course, the group that you can do this on — and sequencing their tumor genomes. In some cases, but not all, he is able to use a targeted therapy that will address their disease as it currently exists. Dan is a very practical guy, so he starts all this description by talking about the 6 vital signs, then he quickly puts up what seemed like an inordinately large number of slides that list all the patients they tried treating this way over the past year that may have been initially successful but ultimately were killed by their disease. It’s fairly sobering, because you realize that these are people and these guys are trying very hard to do what they can to help these patients survive a little bit longer. But in many cases, it’s just not enough.
He then talks about the successes they’ve had, and those are remarkable! I think the talk gets people excited because it’s the most tangible connection to where all this is going. You can do it on end-stage patients now, today, but wouldn’t we rather be doing it on newly diagnosed patients for aiding their treatment in the near term? Would the 20% survival numbers that came out last week rise dramatically if we were able to do that? The extension of that is what we’re talking about here — to extend that approach to a clinical trial mechanism. Most clinical trials take a retrospective look and focus on the specifics that they think are associated with a specific body site. I think we’re proving over and over that this is a false notion. We find genes that are mutated in a lot of cancers, no matter where they occur in the body. A lot of the same mutated genes keep showing up; we just need more data to prove it. But at the end of the day, people get most excited and — as I understand it, this group at Scripps had a lot of physicians — they get it immediately. They understand how doing this in the future will affect their ability to treat patients.