What Keeps Cancer Drug Developers Awake at Night

The range of comments, as well as the general tone, of a recent clinical cancer meeting prompted us to ask developers of cancer drugs what they saw as the major issues and hurdles in designing and testing new compounds. Their comments suggest that empirical methodologies, albeit informed by understanding of the biology around a target, will dominate clinical thinking in the near term. Clinical setbacks over the past several years, a lack of validated biomarkers, and an awareness that preclinical models are of limited utility in establishing dose and identifying likely responders, have reinforced that perception. Developers' principal concerns therefore often focus on things that are within their control; in particular, how to resource and manage a program for the long haul.

Efficacy in patients remains the only true validation of a drug and its target.

By Mark L. Ratner

• Discussions with developers of oncology drugs suggest that empirical methodologies, albeit informed by understanding of the biology around a target, will dominate clinical thinking in the near term.

• Clinical setbacks over the past several years, a lack of validated biomarkers, and an awareness that preclinical models are of limited utility in establishing dose and identifying likely responders, have reinforced that perception.

• Developers' principal concerns therefore often focus on things that are within their control; in particular, how to resource and manage a program for the long haul.

At the AACR-NCI-EORTC International Conference on Molecular Targets and Cancer Therapeutics held in Boston in November 2003, Alex Matter, MD, PhD, the leader of the team that developed imatinib (Gleevec) at Novartis AG , laid out many of the hurdles facing developers of cancer drugs. Indeed, the recent history of the first drugs to target the epidermal growth factor receptor (EGFR)—AstraZeneca PLC 's approved drug, gefitinib (Iressa) and Genentech Inc. 's late-stage candidate erlotinib (Tarceva), both of which failed to demonstrate efficacy in large, randomized late-stage trials at considerable cost to their sponsors (see Exhibit 1)—suggests that industry still has much to learn about choosing the optimal molecular targets for therapeutic intervention and finding the appropriate clinical setting in which drugs against those targets will be efficacious. (See "The Fits and Starts of Targeted Cancer Drug Development," START-UP, December 2003 (Also see "The Fits and Starts of Targeted Cancer Drug Development" - Scrip, 1 Dec, 2003.) and "Testing Drugs against New Targets: Like Playing Blind Man's Bluff?" IN VIVO, October 2002 (Also see "Testing Drugs Against New Targets: Like Playing Blind Man's Bluff?" - In Vivo, 1 Oct, 2002.).)

Topping his list of technology issues, Matter cited developers' still-limited ability to determine the relevance of a target, which requires having viable biomarkers that allow clinicians to look at pharmacodynamic endpoints in early clinical trials, to establish dosage. Such biomarkers haven't yet made their way from the bench to the bedside: it's still early days. Nor have screening tools and structural information about molecular targets provided many good leads—crystal structures for targets often don't exist, and even with that information in hand, it may not be representative of what happens in vivo (such was the case with Gleevec, he noted). Nor are xenograft models predictive of clinical experience in many cases.

But while technology issues certainly are formidable, they are also part of the normal evolution of drug development. In fact, the principal concerns that may keep Matter awake at night lie elsewhere and center on things that are within his control: the most significant were building a team, resourcing it adequately, keeping it in place over the long haul of a development program, and championing the program within an organization, he explained. With Gleevec, for example, the initial internal marketing assessment was that peak sales would reach maybe $30 million. "We had to prove that the drug had potential value beyond CML [chronic myelogenous leukemia]," he said, "and convince management to trust us. If R&D doesn't show a commitment," he went on, "management will easily sense that. They'll set hurdles and present difficult choices," as a way of testing how much faith and commitment a team has to a project.

More generally, Matter suggested, despite 20 years' experience with genetically well-established targets, only a handful of drugs have made it to market—a record that in a subsequent conversation he called "terrible." Along with the need for better biomarkers and other preclinical tools, Matter attributed the failure to poor interplay among academics, clinicians, regulators, and industry—despite good science—and also to a failure of creativity and tendencies to follow the crowd.

The range of Matter's comments, as well as the general tone of the AACR meeting, which for the second consecutive year focused largely on the trials and tribulations of EGFR drugs, prompted us to ask other developers of cancer drugs what they saw as the major issues and hurdles in designing and testing new compounds. During the AACR meeting we also sat down with Julian Adams, PhD, CSO of Infinity Pharmaceuticals Inc. , who shared the podium with Matter and others during a roundtable discussion of the issues around target selection. Adams led the development of the proteasome inhibitor bortezomib (Velcade), first at Proscript Inc. and eventually at Millennium Pharmaceuticals Inc. , which acquired Velcade when it bought LeukoSite Inc. (which itself had acquired Proscript a few months earlier). [See Deal], [See Deal] We chose Adams because of Velcade's rapid development—mirroring that of Gleevec—once it was determined to have anti-cancer effects. (See "Millennium Makes Its Case as a Drug Developer," IN VIVO, March 2003 (Also see "Millennium Makes Its Case as a Drug Developer" - In Vivo, 1 Mar, 2003.).) Later we also talked with David Schenkein, MD, VP, oncology clinical development at Millennium.

We also sought the views of Kapil Dhingra, MD, VP and head, global oncology drug development, and David Heimbrook, PhD, VP, oncology research and global therapy head, discovery oncology, both at Roche , which has a core oncology franchise and a balanced portfolio of conventional and next-generation cytotoxics, antibody drugs (via its close relationship with Genentech), and pipeline candidates against several molecular and enzymatic targets.

Finally, inspired by Matter's comments about the importance of team building, we put the "What keeps you up at night?" theme to biochemist Geoffrey Yarranton, PhD, and his long-time colleague and collaborator Dan Shochat, PhD. Yarranton spent 16 years at Celltech Group PLC , where he led a joint venture between Celltech and American Cyanamid (now Wyeth ) that developed the first approved antibody-drug conjugate, calicheamicin (Mylotarg). He then joined Coulter Pharmaceuticals Inc., prior to its merger into Corixa Corp. [See Deal], where he led the team that refiled the BLA for the now-approved lymphoma drug tositumomab (Bexxar)—like Mylotarg, an antibody/radioisotope conjugate, and an entirely different class of drug than small-molecule kinase inhibitors such as Iressa or Gleevec, with very different issues of mechanism, formulation, and delivery. Yarranton formed Celscia Inc., which merged with KaloBios Inc. this month, with Shochat, who for his part invented Mylotarg and established much of the process for the scale-up and manufacture of Bexxar. (See "KaloBios Inc.," START-UP, December 2003 (Also see "KaloBios Inc." - Scrip, 1 Dec, 2003.).)

Echoing Matter, and reflecting the difficult development path for immunoconjugates, Yarranton and Shochat also emphasized the importance of keeping a team together—and therefore access to the history of a drug as it goes into and through clinical testing.

More than anything else, what came out of those discussions was the strong sense that empirical methodologies, albeit informed by understanding of the biology around a target, will dominate clinical thinking in the near term. The failures of Iressa and Tarceva when used with combination chemotherapy in lung cancer, if unexpected, are also informative. Indeed, a critical message delivered by Matter, as well as by many other speakers at the AACR meeting, was that cancer drug development is now in the midst of a trend going away from a rational, mechanism-based approach to a modified empirical one—sometimes referred to as informed, or guided, empiricism. (See "AstraZeneca's Nuanced View of Cancer Drug Development," IN VIVO, January 2004 [A#2004800007].)

Put another way: clinicians increasingly are using information about specific targets to guide their interrogation of the biology, but are at the same time more mindful that efficacy in patients remains the only true validation of a drug and its target. As Herman Eisen, MD, of the Massachusetts Institute of Technology told START-UP in 1999: "Tumor immunology brings together everything we don't know about the immune system."

Even Velcade has suffered setbacks in this regard: in early January, Millennium announced the failure of a Phase II trial of Velcade in combination with irinotecan (Camptosar) in Camptosar-refractory colorectal cancer patients. The good news: unlike with Iressa and Tarceva, Millennium only intended to enroll 175 patients. The message: first run trials with depth (namely, test a drug in many settings), and only later expand into larger trials of greater breadth.

David Schenkein, MD, VP, oncology clinical development, at Millennium Pharmaceuticals Inc. and Julian Adams, PhD, CSO of Infinity Pharmaceuticals Inc.

David, You came to Millennium two years ago from Tufts-New England Medical Center, where you remain an associate professor. Was the move into industry driven by the belief that you would be better able to follow through on translational research?

David Schenkein, MD: For me, one of the most significant issues in cancer drug development is having the right people and collaborations in place. That collective team needs to be nimble, and one of the things that prompted me to leave my position as director of a cancer center to come to Millennium was the ability to build a team that could look at issues both internally and in the external cancer environment—to make those kind of rapid-pace decisions that allow us to stay ahead of the disease and develop new agents.

One of the things we continue to work on is the open-door, close-communication cross-talk between our clinical and preclinical folks. The concept of a two-way street linking the bench and the bedside – doing translational research, in other words—is even more important when you develop novel drugs. Velcade was a good example of that. Exciting preclinical data gave us an enormous leg up in understanding dose and schedule and what to look for. The clinical team obviously ran with that, and we moved rapidly into a broad development program. Velcade showed striking activity against virtually all tumors tested. It allowed us to develop and design a broad preclinical program, but importantly, it also provided a rationale for the combination studies we're doing now.

Velcade was designed to hit a single target, the transcription factor Nf-kB. Do you think it's better to have two highly specific drugs, one against one target and a second against another, to engineer multi-targeting into a single molecule, or to hope for some favorable off-target effects?

Schenkein: I'm not sure that the answer in oncology will be highly targeted compounds that only affect an individual pathway—the Gleevec model. It may be that molecules that hit one target but affect multiple pathways, like Velcade, ultimately give you more flexibility. You need to ask the questions molecule by molecule, indication by indication.

How do you optimize clinical trial design based on assumptions about mechanism?

Julian Adams, PhD: If you want to do good drug development and establish a good, durable efficacy response, to the extent that you understand the mechanisms you're addressing, you're better off doing a rationally designed initial experiment based on mechanism, as opposed to many one-off, probing experiments, as was the case with development of matrix metalloproteinases, for example. That wasted lots of years and lots of patients, and you still don't have an answer. You have a lot of failed trials, but no definitive answer that there is no role for matrix metalloproteinases. Arguably, the science was not adequately developed to find a role for those agents.

What was the mechanistic rationale behind Velcade?

Adams: Prior to the drug's association with cancer, which was only made in 1996, we were looking at RA and other inflammatory diseases where Nf-kB was certainly known to play a role, in TNF production and so forth. We'd published on this, but we had a narrow therapeutic index. When the association with cancer was shown in three back-to-back papers in Science in 1996, we galvanized our thinking, focusing on tumor types where Nf-kB was hot: prostate, myeloma, lymphoma.

As it turned out, myeloma was a smoking gun. So we enriched the population of myeloma in Phase I immunological studies and saw some responses. That may have been a lucky break, but we were prepared to be lucky because we were looking for a signature of response based on a hypothesis.

Schenkein: We all look for clues of activity or safety, or PK [pharmacokinetics] that push you in one direction or another. We went so quickly into myeloma with Velcade because we began to see early signals of activity in myeloma, even though there were small numbers of patients in the Phase I trial. You hope for striking activity in a refractory patient population, and if you see it, the observation reshapes the development program.

How is that working with your pipeline drug 518, which targets FLT3 in leukemia?

Schenkein: FLT3 is an exciting target, and not just because there are three or four companies working on it, but because there's a lot of biology known there. The potential market may be larger than just leukemia. We know that many of the FLT3 inhibitors hit other targets, as well, which may be important in other tumors. It'll be a continuing proof of principle, similar to Gleevec, to understand how important some of the pathways may be.

We're early on with 518 in the sense that we're trying to get a first read on some safety and efficacy signals from our first human clinical trial. There's little question that acute leukemia will be a different hurdle than chronic leukemia was for Gleevec, in the sense that acute myeloid leukemia [AML] is a more serious proliferative disease than chronic leukemia, and more than one pathway may be affected. Unlike with Gleevec, we may need to run combination trials in addition to single-agent data to see effectiveness.

Julian, I heard you say the other day on the podium that you were agnostic about some kinds of targets, especially antiangiogenic approaches. True? What target classes do intrigue you?

Adams: We have seen one antiangiogenic-positive study—with VEGF. We have dozens of failures. So I'm not sure whether the failures are failures of pharmacology, or poor target.

I think IGF-1 is a good target. I am also intrigued by regulating stress responses in a tumor cell.

It's important to take advantage of the fact that there's a major system--and it's only one system--for protein turnover that predominates in cells, and that the heat shock protein folding chaperone, HSP90, which is the predominant way cells fold, maintains the tertiary structure of protein folding. HSP90 has the elements of a good potential target.

But inhibiting HSP90 in tumors may not be sufficient as a therapy. Ultimately, a drug like geldanamycin or Velcade still may be best used in combination, no matter what its intrinsic activity.

What lessons have you learned from other companies' experiences?

Adams: I think the most important aspect of drug development is knowing, in great detail, the pharmacology of your drug. But to know you've hit the target, and then to be able to measure that to be able to choose a dose, means also having markers of pharmacodynamics, and then other surrogate biomarkers that say that dose and schedule will produce this response—not necessarily a one-to-one response, but a relationship. Without that relationship, it's difficult to develop the drug.

The other thing is to be able to stratify patients, and for that we're truly in the early days. We have to understand how to enrich the patient population that has the target there, and know that. Again, it means doing early, hypothesis-driven trials.

Schenkein: There have been high-profile successes and failures. The events highlight the need, as industry takes molecules forward, for understanding how these pathways interact with more conventional therapies, and really understand dose, schedule, and sequence before we do clinical trials involving thousands of subjects.

Given the number of compounds now in the clinic that were brought forward without markers, and presumably without the understanding you just said industry must have, do you expect to see more failures in the short term?

Schenkein: I do. We just saw it with several high-profile Phase III trials in lung cancer with novel agents and chemotherapy. Remember, trials take years to get going. As with trials of Iressa and Tarceva, we are probably still waiting for earlier assumptions to play out.

I would not be surprised if we saw more, similar disappointments, because of the attention we have to pay to understanding the pharmacodynamic markers that will tell us if the compound is being used in the right clinical setting—before we go to late-stage clinical trials. I'm all for Phase III trials, but you really need to be ready before you go into those.

Should a more rational approach lead to a more rational outcome?

Schenkein: We have tools available to us, now. When we were combining three instead of two cytotoxics in lung cancer, that was, rightly, an empirical clinical trial. On the other hand, when combining novel pathway inhibitors and conventional drugs, we do need to learn as much as we can about the interaction of those pathways up front. You should try to understand as much as you can before doing large trials.

The other issue is to target the patient population you think will respond, instead of the more traditional "Treat all comers" approach. If some of the novel compounds in development have striking activity in a small patient population, you'll miss that effect if you enroll everybody: Herceptin would not have had a positive outcome if Genentech had enrolled all women with breast cancer, for example. Nor would Gleevec have shown activity in most leukemias.

All the preclinical experiments in the world aren't going to predict what will happen in the clinic. There's tremendous tumor heterogeneity inherent within patients with malignancies. The more we learn before we go into those trials, the better we'll sleep at night.

Was Iressa's problem a lack of measurement tools up front, prior to clinical development?

Adams: Yes. We have the tools and the technology to interrogate whether knocking down a gene actually sensitizes the cell. RNAi will help enormously where you don't have a drug a priori. But then again, once you get the drug, you have to study it rigorously; for example, you may be way under- or overdosing if you base your work on what happens in a mouse.

So we've seen some progress and some hiccups. Would you expect to see the same rapid clinical development for other agents that you've seen with Velcade?

Schenkein: We use the same set of criteria for our pipeline compounds that we did with Velcade. We have put in place the same commitment to understanding the preclinical rationale, having appropriate pharmacodynamic markers. For targeted therapies, in particular, we want to know early on that we are hitting our target. And once we know we're hitting a receptor tyrosine kinase, for example, I also want to know—in Phase I–whether our drug is turning that kinase off, independent of the maximum tolerated dose [MTD].

One of the challenges of oncology drug development of targeted therapies is that the old dogma of finding a dose-response relationship by dosing until you hit MTD may no longer be appropriate. When we talk about Phase I studies combining novel agents, we're talking about biologic endpoints, not toxicity endpoints.

At the AACR, Alex Matter pointed out that after 20 years of drug development around gene-associated targets, there's only a handful of drugs. Should that be viewed with disappointment?

Adams: I think there have been a lot of wasted efforts. There have been notable examples of drug development done very well, and there have been examples of development done very poorly. Knowing a gene is overexpressed doesn't necessarily give you an edge in development. You also need to know whether it is promoting the disease, or not.

How much stock do you place in preclinical models?

Schenkein: Preclinical studies are a reasonable guide that gives us comfort. They're important to do, but obviously the true testing has to be done in humans. Xenografts only go so far: I don't look for them to do more than give us an underpinning for moving into the clinic. It's the clinical trials that allow us to move from step one to step two to step three.

Are brute force screening approaches, such as reprofiling existing compounds using some sort of novel assay or technology platform, a good substitute for rational drug development?

Adams: If every company did reprofiling, I'd say this is a failure of creativity. That there are some companies doing it [e.g., CombinatoRx Inc. ] is a good thing; we are not smart enough to understand everything about systems biology.

David Heimbrook, PhD, VP, oncology research and global therapy head, discovery oncology, and Kapil Dhingra, MD, VP and head, global oncology drug development, at Roche.

Cancer drug developers lament that xenograft models often don't correlate with the eventual clinical experience. What's your view?

David Heimbrook, PhD: There are limitations to the interpretation of preclinical models—what happens in a cell or in a mouse—compared to what will ultimately happen in a human.

That said, we aggressively follow the application of diagnostics and biomarkers from the earliest stages of our drug development efforts all the way through to the end. This is a place where Roche, as the number one player in oncology therapy and diagnostics, has an advantage. Identifying the right biomarkers and pharmacodynamic markers, first in animals and then in patients, allows you to have confidence that you're hitting your target enzyme hard. That's one of the things that allows you to interpret clinical trials, to show whether you're achieving the doses you need to achieve to hit your target. Then you can do the efficacy studies that follow the secondary biomarkers to see whether you're actually having a benefit once you do suppress the target. We address those important questions early on in in vivo studies, and translate and carry through into the clinic.

Do you have a predilection for developing multi-target or single-target drugs, and if the former, how can you ascertain the contribution inhibiting each target makes to improving outcomes?

Heimbrook: That's a gray area. Even if a compound is directed with absolute specificity against a single molecular target, the downstream effects of inhibiting that target are pleiotropic. You can see this, for example, with HDAC inhibitors—you hit one enzyme, and a lot of things happen in the cell. So even with highly specific targeted drugs, you'll have a "shrapnel" effect. With small molecules, obviously, there's always the possibility that you're hitting things other than the target that you think you're really directed against, because specificity is rarely absolute. With kinase inhibitors, in particular, the off-target effect can be significant.

The best you can do is to come up with a hypothesis linking the target and the biological effect, which is driven by expression in the tumor, and then direct your drug discovery and development efforts against that target. If the drug happens to interact with other proteins, with other enzymes, you determine whether that is likely to impact safety or likely to impact efficacy, and you decide from there whether to proceed with that particular compound, or whether you need one that's more or less selective.

Do you have biomarkers to ensure this?

Heimbrook: From a preclinical perspective, we'll generally have exploratory markers to use, to see how they track with efficacy. At the early translational stage, going from bench to bed, you're only looking at hypotheses, and knowing which outcomes will track to the clinic is difficult to predict. You really won't know, when you first go into the clinic, whether that biomarker will contribute value or not.

Many of your pipeline compounds for treating solid tumors, including epithilone D [partnered with Kosan Biosciences Inc. ] and topoisomerase inhibitors [partnered with Ipsen ] target enzymes. How important are measurements of enzyme levels as biomarkers?

Heimbrook: It's often useful to measure inhibition of the enzyme in the target tissue, ideally in the cancer, so you know you have the right dose.

Obviously, exposure in an animal can be difficult to translate into humans. Having pharmacodynamic markers—measurements of target suppression—therefore becomes an important readout in early clinical development, to know you have the right dose and are shutting down the pathway before you see toxicity. And whatever the clinical dose you're using going forward, you want to know that you have adequate exposure to hit the desired target. Otherwise, you're doing a very different experiment than the one you think you are.

You can measure kinase inhibition using downstream markers, looking at phosphorylated substrates of the kinase you want to inhibit. By doing this, you can see if you're inhibiting the target at this dose, and therefore interpret the clinical experiment to see whether inhibiting EGFR, for example, offers a clinical benefit.

What are the key clinical challenges facing Roche, in your opinion?

Kapil Dhingra, MD: Finagling the pathway of development to be able to optimally position a drug is probably the biggest challenge for us. That's because of competition in the field: there are at least 400 open INDs for oncology drugs at any one time, and it's difficult to position a drug with so many players, all trying to get their drugs to the market in multiple, and often the same, tumor types.

Given that dynamic, do you have a preference for being first in class with a drug, or a fast follower? Can you address this in terms of your oral fluoropyrimidine, Xeloda, which competes with Bristol-Myers Squibb Co. 's UFP?

Dhingra: First, we look first to see what proprietary position we have with respect to a target and a drug. There are times when we are in fast-follower mode, as with Tarceva, after Iressa. There are times when we're first in class, as with Herceptin or Avastin.

On the other hand, we were not the first with an oral fluoropyrimidine. However, we had the conviction that our drug was better and was tumor-targeted. The data from our clinical program proved that our conviction was well placed. So we are not hung up on whether we're first or second in class.

What about balancing your portfolio?

Dhingra: We discuss that, but we don't approach the decision-making for any individual drug based solely on portfolio-balancing considerations. Of course some things will be of greater interest, while others will be of lesser interest to us. But fundamentally, we all accept our ignorance of cancer biology sufficiently that we are willing to take into our portfolio anything that acts on a good target or is a promising modality for treatment.

Like everyone else, we are interested in pathways where we can show a difference between the tumor and normal tissue, whether in the form of DNA mutation or upregulation at the level of RNA or protein expression. That's what gives us the confidence to work on a target.

Our portfolio speaks for itself in that we have a mixture of traditional cytotoxics and next-generation cytotoxic drugs such as Xeloda, the topoisomerase I inhibitor diflomotecan, and epthilone D. We also have drugs that target various oncogenes and growth factors, as well as novel monoclonal antibodies such as those targeting the MUC-1 protein [through an alliance with Antisoma PLC ] and drugs that are important for supportive care, such as hematopoietic growth factors. Cancer is a sufficiently heterogeneous disease that it would be a mistake for any company to simply focus on one pathway or type of target.

You've mentioned your antibody drugs, via Genentech. Would you look for a small molecule against a target for which you already had a large molecule?

Dhingra: If we think a small-molecule inhibitor can deliver the same efficacy and is more convenient for the patient, we would. For example, in angiogenesis, we could potentially look at a small molecule even though we have Avastin in our portfolio. But we've also learned that, for some targets at least, a monoclonal antibody may be more efficacious than a small molecule. This is an interesting, emerging observation that we don't fully understand. It's not necessarily that we'd seek a small-molecule replacement for a large protein. And like everyone else, we've encountered challenges discovering small-molecule inhibitors against some targets.

Is there an inverse correlation between a drug's specificity and market size?

Dhingra: It's reasonable to assume that we won't see a single agent that suddenly cures lung cancer or breast cancer. But that doesn't mean targeted therapies are relegated to niche markets. I wouldn't consider drugs like Herceptin or MabThera as targeting a niche market. MabThera, for example, was launched originally for follicular lymphoma, which is considered a small indication. But then we found it worked in other lymphomas, and now its sales are more than $1 billion.

How do you decide whether to pursue a single-agent or a combination approach to treatment?

Dhingra: Whether a drug will work as a single agent will depend on when in the natural history of a cancer you plan to use it. Presumably, early in the process of carcinogenesis fewer genetic lesions exist. As time goes on, the number increases. So it is conceivable that if a cancer is detected early enough, one drug or a simple combination may cure it. But for most common tumors, it's fair to say that for the short- to mid-term, we will be using combinations.

I think the experience with small-molecule EGF inhibitors such as Iressa and Tarceva is a good example of how tumors are often a lot smarter than we are. The development program was fast because everyone had a high degree of confidence that EGFR is expressed at high levels in a large proportion of lung cancers and both drugs had shown significant single-agent activity in patients with advanced lung cancer. The expectation was that by combining these drugs with chemotherapy, you'd see a survival benefit in the first-line therapy setting. Why didn't we see a survival benefit? We're all speculating on the reasons. So, even when we develop targeted therapies, the selection of patients is not always so simple as looking at expression of the target.

I think it's likely that in some cases, the field will have similar experiences going forward. But people are also becoming a lot smarter about how to develop these types of drugs. There are two routes to take: one is to discover biomarkers early in the preclinical phase, so that before you get to an extensive clinical program you have potentially validated biomarkers that might predict efficacy. Another way, as we've done with Avastin, is to invest more in Phase II randomized trials, which can provide an initial estimate of the activity of the drug in a 100-150-patient trial before proceeding with a large development program.

Geoffrey Yarranton, PhD, CEO of KaloBios Inc., and Dan Shochat, PhD, also of KaloBios.

You and Dan formed a company last year, Celsia (recently merged with KaloBios), with the express purpose of reuniting an established drug development team. What are the biggest teachings you carry with you from your prior experiences with Mylotarg, a first-in-class cancer drug, and Bexxar?

Geoffrey Yarranton, PhD: The way we approached Celsia, at least on the biologicals side, was to emphasize finding antibodies that can access a target in vivo. We know they can access a target in a test tube and often in a mouse, but the real issue is whether antibodies can get to a target—their distribution is very different from a small molecule's.

World-beating technology won't help if you choose the wrong target, and we believe we have a pretty good track record of picking the right targets and the right indications in which antibodies will work.

We picked the target for Mylotarg because AML, while not a big market, is a disease of the bone marrow and the blood system: When you inject an antibody and measure the volume of distribution, it's mainly in the blood compartment. That's important because antibodies are big molecules that don't get out of the blood stream and into tissue easily. With Mylotarg, 80% of the injected dose goes to the tumor, so you can dose up the antibody until it saturates the leukemic cells. With solid tumors, by contrast, less than 1% of the injected dose hits the tumor, and the rest had to be cleared. We knew that at the beginning of the program and chose the approach as the best example of a targeted antibody, because it really was targeted.

We could also take out bone marrow and treat it with the conjugate so we could see if the conjugate kills primary tumors, as opposed to tumor cell lines. Because the bone marrow is accessible, exposing the primary tumor to the antibody conjugate was similar to injecting a patient with the antibody. In fact, the response rate we saw in in vitro systems was very similar to the response rate we see in patients. That validated our approach.

That's fine for leukemia. But it's also a special case because it's a disease of the blood. How do you go after less accessible tumors using that approach?

Yarranton: You're quite right, it's an in vivo delivery issue. The key is to make sure the liver is more resistant to the drug conjugate than is the tumor, because that's where the conjugate is taken up and processed, which activates the drug portion of the conjugate. If you dump a lot of conjugate into one of the clearance organs you're going to see a lot more toxicity—dose-limiting toxicity.

It comes back to selection of targets and understanding what happens when a specific antibody binds to its target. What we learned about anti-CD20, the antibody in Bexxar, is that the antibody has the ability to trigger apoptosis. If you then add ADCC (an antibody's effector function) and then a radioisotope, you have a combination chemotherapy in one molecule. It should be possible to make antibody-drug conjugates that act like combination chemotherapy, but with greater tumor selectivity, and therefore a wider therapeutic window.

It's wrong to think that you only want the antigen to be a receptor for targeting. You need to be able to target with molecules that have additional anti-tumor functions—triggering apoptosis, or blocking or triggering a receptor. There are enough targets out there to select one where antibody binding to the target creates an anti-tumor effect.

Not only are there lots of targets, and you can create thousands of antibodies against a target; sometimes the trick is also figuring out how to get the right antibody to the right epitope on the molecule: we're finding that you have to select antibodies against a range of epitopes and then select the best antibody with the best activity. Even within CD20, where there's a very small extracellular domain, about 45 amino acids, there's a range of antibodies with different properties. Some trigger apoptosis effectively, some don't.

Dan Shochat, PhD: You have to pick the disease and the right conjugation platform, and you need more than one platform; namely, several types of cytotoxic agents—radioisotopes or several types of drugs.

Do you want to develop conjugates, now?

Shochat: Not necessarily. You start with antibodies that have biological effect on their own. If that is enough, you stop with that. If not, you have to choose the right agent to attach—an isotope, a drug, or some other agent, perhaps a peptide with biological activity. You want to widen the therapeutic window by adding a "payload" that already possesses a differential between its effect on tumor cells and on normal cells. And that depends on the pathway the antibody modulates.

So is using a naked antibody a reasonable approach to killing a solid tumor?

Shochat: Take Erbitux [the anti-EGFR antibody from ImClone Systems Inc. /Bristol-Myers Squibb]. I don't know if it'll work, but the approach is absolutely legitimate: use an antibody with biological activity, then give chemotherapy. If you can increase the sensitivity of the tumor to the drug with an antibody, you may be successful. I'm not sure we've achieved that yet with Herceptin or Erbitux: they improve chemo, but we have to aspire to a bigger improvement in the future.

Yarranton: I think we should take heart from the fact that these drugs work in combination with chemotherapy. But the chemo is still given with systemic toxicities. It'd be even more effective to combine the antibody with a drug and get a targeted combination effect.

Did you come to Coulter specifically to work on the Bexxar BLA?

Yarranton: I joined in 1998, and the first BLA, to which FDA responded with a Non-Approvable Letter, was filed in 1999. I came into the program after that, to manage the team and help refile the BLA.

The initial BLA contained a number of firsts. It was the first radiolabeled therapeutic application FDA received; it was the first fully electronic filing for the group we were working with, and when you're dealing with a radiolabeled antibody, you may think it's trivial, but it takes a number of different skills within the FDA to review it: radiodosimetry, biologics and chemistry—you don't often see all of these in product applications. It was complicated, and Coulter was a young company.

As a small company, you'd prefer to not have everything you do be a first. It was the first—and only—facility that worked at the 55-curie level. There were too many firsts, and when you're blasting out those things you're also developing the review process.

With the hindsight of that experience, would you do it differently?

Yarranton: We licensed Bexxar in 1998 to SmithKline Beecham, and they found also that it was a complicated filing. SB, and then GSK, was a great partner. There were a lot of delays but they stuck by the program, they believed in the product. This gave us comfort that even one of the best companies saw Bexxar as a complicated, yet useful product.

You need to have a continuous dialog with the FDA. I think the development process was a lot more complicated than we would have liked. Dan inherited a hollow-fiber process, which is not a great system to scale up. He got what he got. This wasn't something the company started, it was in-licensed and we had to go from there.

We're keen on getting programs as early as possible. A lot of smaller companies tend to try to develop drugs as quickly as possible, for obvious reasons, but if you have to go back and re-do things, it ends up costing more money and it takes longer. Or if you go to out-license, a partner may say we'll have to go back and re-do Phase I, and drop the value of the compound accordingly.

Another consequence is a misalignment of process development and clinical development, so when you have a Phase II program you have it not just from a clinical perspective, but also from a process perspective. I think our experience in this gives us a competitive advantage.

Remember, Coulter Pharmaceutical was formed initially with Bexxar as the lead program, but the antibody had already been in the clinic, which was the attraction. So we were also stuck with the product's history—the cell line, the initial process for producing the protein. We had to make a number of major switches in the program after trials were under way, and the FDA would ask us for comparability studies, which slowed us down. And when you come to file the BLA, you have to put all that history in, and people will go back and look at it and often the people associated with the program at the company and the FDA have changed and there are new discussions around the history.

Were you able to keep the Bexxar team together through approval?

Shochat: Most of the development work was done with the same team. The team that developed the radiolabeling is still there. I'm still consulting. But certainly some people left when Corixa bought Coulter.

Speaking more philosophically, you need to keep the history of development, from research to commercialization, alive in a company. The people that do the early work are not needed as the drug moves up the ladder, and that part of the team is disbanded. If you don't have people to carry the history, later on the company doesn't know how and why a drug was developed as it was. We used to give talks at Coulter on the history of development of Bexxar. People found it interesting, sure, but it also kept the original torch alive.