Reclassifying cancer, guided
  by genomics

January 2001
Cover Story

Karen Titus

Cancer genomics researchers tend to be a nervously buoyant lot when they talk about their work. Little wonder, really.

For more than a decade, the large-scale application of molecular biology and molecular tools in cancer research has been accompanied by breathless pledges and giddy expectations of a medical revolution. Pharmacogenetics! Labs on a chip! Bye-bye, cancer!

"There are so many big promises being made right now," says Ken Buetow, PhD, chief of the Laboratory of Population Genetics at the National Cancer Institute. "It’s like one big promissory note." Certainly he’s had his fill of announcements from scientists who, as he puts it, "suggest that starting tomorrow, the entire world will be different."

Wouldn’t you be a little nervous too?

And yet. And yet. Oncology no doubt will be different, and Dr. Buetow, enthusiasm barely in check, is among the many who are eager to sound the trumpet.

After years of effort, Dr. Buetow and legions of other cancer researchers are enjoying an exponential increase in knowledge about tumors and its potential application in clinical settings. "We’re seeing some dividends from the investment we’ve been making in the raw infrastructure of genomics," he says. "We’re generating new reagents and beginning to apply them, and we can actually reclassify cancers in ways we couldn’t before."

More quickly than anyone realizes, say Dr. Buetow and others, researchers are looking beyond morphology toward genetics, identifying so-called diseases within a disease.

Efforts to subclassify human breast tumors are on the leading edge of such work.

"One of the very interesting things we’ve found is that estrogen receptor-negative tumors are at least two distinct types of tumors: those that have a basal epithelial cell-type profile, and those that highly express Erb-B2," says Charles M. Perou, PhD, assistant professor of genetics at the University of North Carolina. "Actually, some of our newer data suggest there might be even more types of ER-negative tumor, at least one of which looks to be of a luminal cell origin."

Moreover, the different subclasses of tumor appear to be associated with different clinical outcomes, says Stefanie S. Jeffrey, MD, chief of breast surgery, Stanford University School of Medicine. New findings based on tumors collected and analyzed at Stanford and by Norwegian collaborators "suggest five different classes of breast tumors that augment our usual pathological tumor classifications," she says. "They have very different survival curves, and the treatment for these separate classes of tumors would likely be extremely different."

With clinicians now identifying directed therapy for Erb-B2-positive breast tumors, "We may want to do the same thing for the Erb-B2-negative basal-like tumors as well," Dr. Perou suggests.

Much of the work on breast cancer comes out of Stanford, spearheaded by researchers in labs overseen by David Botstein, PhD, and Patrick O. Brown, MD, PhD. Though the researchers are also targeting other types of cancer (including prostate, lung, liver, renal, ovarian, and brain), breast tumors are an appealing quarry because of their great heterogeneity. "We reasoned that because there were many different morphologic subtypes, by gene expression we should see many different subtypes as well," says Dr. Perou, who until recently was a postdoctoral fellow in Dr. Botstein’s lab. "And on top of that, even within a single morphologically described subtype-in particular, infiltrating ductal carcinomas-there’s a great deal of clinical variability."

The approach at Stanford involves using genome-wide expression patterns to search out groups of genes that are either co-regulated or highly expressed in specific subpopulations of tumors. Such recurrent gene expression patterns can then be used to identify discrete groups of tumors; these tumor-defining sets of genes, in turn, could be of biologic and/or clinical importance.

The researchers use complementary DNA (cDNA) microarrays, developed by Dr. Brown, to determine whether genes are expressed, and at what level. They then turn to a hierarchical clustering gene expression analysis software tool, which examines and compares the gene expression pattern of each of the thousands of genes contained on the arrays. Genes with similar expression patterns are grouped together; the software displays this information in colored tables. "By looking at these colored tables, we can see sets of genes that are always expressed together, or always not expressed," Dr. Perou explains. These sets typically are highly expressed in some samples, and absent in others. "In both cases, that’s likely to be of biologic importance," he says.

In a paper published in Nature (Perou CM, et al. 2000;406:747-752), the Stanford researchers reported on their analysis of gene expression patterns in grossly dissected normal or malignant human breast tissues from 42 individuals. In addition to identifying the two ER-negative subtypes, they report the so-called molecular portraits provided by the gene expression patterns appear to represent each individual tumor itself, and not the particular tumor sample.

Initial findings based on 42 individuals don’t carry the weight of findings from hundreds of patients, of course, and the researchers involved in this work stress—repeatedly—that it needs to be confirmed and extended. Yet their enthusiasm is hard to ignore.

"This will totally change how we diagnose and treat breast cancer," says Dr. Jeffrey, who oversees the clinical database for the Stanford breast tumor research and assists with data analysis, and whose lab performs breast tumor RNA isolation and microarray hybridization.

The current means by which physicians determine whether a tumor is likely to shed into the circulation and metastasize are limited. As a result, adjuvant or hormone therapy is given to most patients whose breast tumors are greater than 1 cm, though "the great majority of women who get this therapy do not need it, necessarily," Dr. Jeffrey says. "Their tumor does not have that capacity to spread, or the chemotherapy may not directly affect their tumor."

Genome-wide analysis of tumors, on the other hand, should enable physicians to peg characteristics that predict the clinical behavior of the tumor, or to identify potential targets for novel, tumor-specific therapies.

What will it take to reach that point?

Plenty.

From a sheerly logistical standpoint, researchers have to obtain a sufficient number of tumors—containing high-quality RNA and accompanied by good clinical databases—to power studies large enough to identify significant subsets. "But it’s not always easy to do that," says Douglas Ross, MD, PhD, chief scientific officer of Applied Genomics and formerly a fellow in the Botstein-Brown lab. "Ideally, we would like to identify clinical correlates in cohorts going back 10, 20, or even 30 years. The best tissue for cDNA arrays is usually fresh or fresh frozen and stored a couple of years or less."

To circumvent that challenge, Dr. Ross and his colleagues at Applied Genomics (a spin-off of the Huntsville, Ala.-based firm Research Genetics), in collaboration with Stanford, are developing antibody reagents to the protein products of the genes in the novel subcategories identified by the cDNA microarrays. The antibodies are then used to immunostain tissue arrays containing hundreds of different tissue samples. This permits researchers to assess one antibody on as many as a thousand different tumors in a high-throughput fashion. With this approach, researchers can use archived paraffin blocks to do large retrospective studies, rather than rely on fresh frozen tissue.

Another issue, Dr. Jeffrey notes, is that most of the work to date has been done on large tumors, which tend to be late stage. Whether the researchers’ findings will hold up for earlier-stage tumors remains to be seen. And inherent in that challenge is another one: Earlier-stage tumors tend to be smaller and thus may lack genetic material in amounts sufficient to perform gene expression analysis. And, as Dr. Jeffrey points out, "Patient care trumps research."

In Dr. Jeffrey’s lab, researchers are refining RNA amplification techniques that will permit analyses to be done on even the smallest pieces of tissue. "We want to be able to use core biopsies rather than chunks of tumor," she says.

Matt van de Rijn, MD, PhD, assistant professor of pathology at Stanford, points to yet another problem: Most array techniques analyze the mRNA expression in an entire tumor mass. "You grind the tumor up, and there’s all kinds of nontumor cells in there: vessels, lymphocytes, stromal cells, histiocytes, benign breast cell ducts—you name it. So you’re measuring the global level of messenger RNA for all these different combined tissues."

Laser capture microdissection holds out the promise of providing purified tumor cells, though it’s a technique Dr. van de Rijn admits to struggling with in his lab. "Many people are looking at this approach, some with more success than others," says Dr. van de Rijn, who is assisting with the numerous cancer studies coming out of the Botstein-Brown and other Stanford labs, including his own. "But I think eventually the problems will be resolved. And then, just to fantasize a bit, you could compare, from one and the same patient, benign breast tissue, benign ductal cells, ductal hyperplasia cells, in situ carcinoma, and invasive carcinoma."

Researchers must also play a bit of a waiting game. "We have to get these correlates between gene expression and patient outcome before we can convince anyone that what we’re seeing is real," Dr. Perou says. But given the prospective nature of most of these studies, "You almost have to wait a few years to get the needed clinical followup data."

And let’s not forget the need for targeted therapies, assuming the above challenges are met. Says Dr. Ross: "The ability to subcategorize cancer is useful to pathologists only if it predicts something about the patient’s clinical course or response to therapy. It has to be useful to the clinician or patient; otherwise no one’s going to order it, right?"

Right. He and his colleagues at Applied Genomics are betting that their targeted reagents will enable pharmaceutical companies to distinguish which tumors will react to their own specific compounds or reagents.

"Let’s say a company has a drug candidate that seems to be effective for 40 percent of breast tumors," Dr. Ross explains. The goal would be to offer reagents to help predict which 40 percent those are.

Another hope is that the tissue array studies will yield serum-based markers for tumor monitoring, Dr. Ross says, though that hope, not surprisingly, is tempered by its own challenges.

If the subtypes being identified are new and assist disease management, they will then need to be scored in the clinical setting. Most observers suggest current research techniques are too unwieldy for nonresearch labs. "We don’t need to be looking at 48,000 genes in an everyday setting," Dr. Jeffrey says wryly.

Instead, clinical tests could appear in any number of guises.

"Will they be gene expression-based tests?" Dr. Perou asks. "Maybe. It might be that we end up doing a microarray on every tumor sample that comes in. Or it might be that we have to do a gene expression analysis on, say, 20 or 30 genes for a given tumor subtype, and those 20 or 30 genes will allow us to make the prognosis." In the case of the ER-negative tumors, he says, the subtypes were differentiated by 20 to 100 genes.

Microarrays are one possible tool, Dr. Ross agrees. "Then there are companies such as ours that are saying, ’Well, maybe, but we also believe that if the number of genes is at least relatively limited—10, 20, 50—that a panel of antibodies may be just as useful."

"It’s all going to shake out as we go forward," Dr. Ross says. "And as much as we are focusing on antibodies, I don’t know the answer."

Dr. Jeffrey offers two possible candidates: additional immunostains to complement the current lineup of ER, PR, HER2/neu, and Ki-67, and quantitative RT-PCR (Taq-Man assay), which measures RNA for specific products, but on a limited basis. "We’d be looking at tens or hundreds of genes rather than thousands," she says.

According to Dr. van de Rijn, plenty of people "are talking about having a limited number of arrays and doing quantitative RT-PCR on a group of genes. You would just isolate messenger RNA from one tumor, put it either on a chip or a multiwell plate or whatever, and run quantitative analyses for a bunch of genes. I could see that happening," says Dr. van de Rijn, who is also assisting with the Applied Genomics collaboration to develop antisera, testing gene reactivity on the tissue arrays.

A complete swap of old methods for new is unlikely, he says. "A lot of people, when they think about this, start to assume this will replace surgical pathology. But actually, I think current surgical pathology, taken together with immunohistochemistry, is doing a very good job of creating useful tumor categories," he says.

"Of course, it can always get better."

It’s already gotten better with diffuselarge B-cell lymphoma.

A group of researchers at the NCI are using cDNA microarrays to determine gene expression in normal lymphocyte development, says NCI senior investigator Louis M. Staudt, MD, PhD. Using specially designed "lymphochips," Dr. Staudt and his colleagues—including a contingent from Stanford and the University of Nebraska Medical Center—report finding two molecularly distinct forms of DLBCL. One type expressed genes characteristic of germinal center B cells, and the other expressed genes normally induced during in vitro activation of peripheral blood B cells.

Strikingly, the two types appear to have distinctly different survival rates.

While most patients with DLBCL respond initially to therapy, only about 40 percent survive beyond five years, Dr. Staudt notes. "It’s been a big clinical puzzle why there’s always a subgroup that is cured by multiagent chemotherapy and always a subgroup that is not. The actual percentage of patients who are cured by chemotherapy has not budged in the last 20 years."

"When we looked at the gene expression profiling," he continues, "we found we could divide it nicely into two prognostic groups." Patients whose tumors have a profile resembling the germinal center B cell have a 75 percent chance of being alive at five years, while those whose tumors resembled an activated peripheral blood B cell (called activated B-like lymphomas) did poorly—less than a quarter of them were alive at five years. The researchers reported on their work in Nature last year (Alizadeh AA, et al. 2000;403:503-511).

The information provided by these so-called cell-type signatures is more detailed than that provided by traditional immunostaining methods, Dr. Staudt maintains. "Diffuse lymphoma has for many years included two different diseases that nonetheless look morphologically the same under the light microscope. In the most recent formulation of the WHO diagnostic criteria, all cases are essentially lumped together as diffuse large cell lymphoma. But when you look at the genes that are expressed, there are two very clear groups expressing over a thousand different genes between the two."

Based on this work, Dr. Staudt and his colleagues are expanding their studies with the Lymphoma-Leukemia Molecular Profiling Project, or LLMPP. A large group of pathologists and clinicians from eight institutions worldwide are using lymphochips—which contain genes that are preferentially expressed in lymphoid cells and genes with known or suspected roles in immunology- or cancer-related processes—to look for additional lymphoma subgroups.

"The most interesting question right now is why we don’t have complete predictive power," says Dr. Staudt. While the initial DLBCL work does "a pretty good job segregating patients into prognostic groups, nonetheless, even in the favorable group there were patients who died early on." The goal is to understand the precise differences between the patients with germinal center B-like DLBCL. "It’s probably a question of which individual genes and which individual pathways within the cell are active," he says.

The large number of cases expected to be contributed by the LLMPP group should generate statistically significant data. Beyond looking at diffuse lymphoma, "We want to look at all the lymphoid malignancies," Dr. Staudt continues. This would include follicular lymphoma, mantle cell lymphoma, and the large number of less-common types of lymphoma and leukemia. "We expect the same ’disease-within-a-disease’ concept will be recapitulated," he says. If it is, the plan is to then test whether the different subgroups within a given diagnostic classification have different clinical outcomes.

Wing C. Chan, MD, a professor of pathology at the University of Nebraska Medical Center and co-chair of the LLMPP, suggests the project will identify new molecular targets for therapy as well as useful diagnostic markers. "If we can understand the abnormal genetic pathways in different types of lymphoma—and I think there are a limited number of those pathways—then we should be able to find new therapeutic targets," he says. "That’s what’s exciting about this work."

Even when a fair amount is known about the cause of a certain cancer, molecular studies can be useful. Take liver cancer, for example.

Relying on several decades of epidemiological studies, physicians know that hepatitis B virus and aflatoxin exposure are important environmental risk factors for the disease. Nevertheless, the majority of people who have lifelong, chronic HBV infection or exposure to aflatoxin B1 do not develop liver cancer. Using reagents generated by the NCI’s Cancer Genome Anatomy Project and Genetic Annotation Initiative, Dr. Buetow and colleagues have studied genes involved in detoxification of aflatoxin B1.

Much to the researchers’ initial frustration, even in a simple exposure like aflatoxin B1, which is a single carcinogenic compound, some 15 to 20 different genes could be involved in its immediate detoxification.

Much to their eventual satisfaction, however, "We found there were differences, and some of the variant forms of the genes we identified showed significant differences in individuals who developed liver cancer versus individuals who didn’t," Dr. Buetow says. When they looked at HBV and genetic constitution, "We saw that individuals who had a hepatitis B infection and the at-risk genotype for the aflatoxin detoxification locus were 75 times more likely to develop liver cancer than those who didn’t have those joint units," Dr. Buetow notes.

Based on these and other cancer genomics studies, "There’s going to be a whole series of diagnostics tests for answering the question, ’Are you susceptible to developing disease?’" Dr. Buetow predicts.

The potential payoff of the studies could indeed be enormous, and the researchers involved in these efforts don’t mince words when they consider what lies ahead.

"This is part of the future of diagnostic pathology," says Dr. Staudt.

"This is revolutionary," says Stanford’s Dr. van de Rijn. "Its impact will be in the same realm as the development of PCR and monoclonal antibodies. This is not something that will go away in a couple of years."

But, characteristically, they then cover their pronouncements with a few choice caveats.

"It’s unbelievably exciting," says Stanford’s Dr. Jeffrey, before adding, "But what I’m afraid of is people calling up and saying, ’Well, I have a tumor, can you tell me what it is?’"

They’re not ready for that. Yet.

Karen Titus is CAP TODAY contributing editor and co-managing editor.