Cancer update: precision oncology comes into its own

The first post in this series attempted to introduce the vast field of cancer, with an eye to describing what I expected would be represented in the new developments at ASCO this year.  In many respects, those expectations were realized and expanded upon.

The second post looked at combination therapies and how they're changing the cancer treatment landscape.  Now, the use of multiple drugs in combination isn't new to cancer research, so why did I devote a whole post to it?  Because the way we can use these approaches is changing in subtle but powerful ways.  In the past, we might focus on a small number of patients at research institutions whose cancer biology got additional attention.  In other words, we could look at a few 'example' patients and hope to learn information we could generalize to all cancer patients.  Of course, cancer doesn't work that way.  Each patient's cancer is different, and we've known that extrapolating out from one patient to many is a sub-optimal strategy.  It's just the only strategy we had available to us.

Though we do see trends of certain pathways in certain cancer types, there's only so much you can do with an approach of "treat liver cancer patients like [THIS] and lung patients like [THAT]."  We've been moving in the direction of figuring out the underlying biology of which combinations of therapies might work together well for a while now, but now there are a number of specific genetic abnormalities we can look at to tell us when a combination therapy will be required to ensure a therapy is effective.  This trend will likely continue in coming years, with more biomarkers helping to define combinations of drugs that work, where only one drug would never have had any impact.

The third post looked at why certain patients fail current standard therapies.  If we can know ahead of time which patients will benefit from each treatment we can ensure patients get to the right treatment sooner.  The nature of cancer treatment is that even if a therapy is effective at eliminating most of a patient's cancer, the cells that weren't hit come back stronger.  This is similar to how we use antibiotics against bacteria.  If you give away penicillin like candy you'll soon see resistant strains of bacteria that don't care about your antibiotics treatment.  The same happens with cancer cells and standard therapies.

It's important to remember that this process isn't directed, it's selected.  A directed process would aim to achieve a specific goal: become resistant to penicillin/chemotherapy.  A selected process is just a subset of all possible variations, one that is capable of overcoming the selecting event.  If cancer cells were using a directed approach, the mutated cells would suddenly be immune to the cancer treatment, but would be otherwise identical to the cells our treatment killed.  Instead, treatment selects among all the cells those that are already resistant to therapy.  These tend to be cancer cells with the most mutations.  With cancer, more mutations usually means more aggressive.

Getting patients effective treatments earlier in their fight against cancer is vital to overcoming this disease.  How do we do this?  Once again, the answer comes down to precision medicine.  We're looking for specific biomarkers that tell us whether a therapy will work or not.  This is slightly different from the post that focused on combination therapies, which asked whether new treatments would be effective.  This is about making sure current therapies work when we give them.

One of the scary things about getting diagnosed with is the idea of getting chemotherapy that ultimately doesn't do anything, or that only extends your life by a couple of months.  Just the process of getting chemo is a pain.  Your doctor doesn't hand you a pill to take.  You have to sit in a chair with an IV on your arm for a couple of hours getting an infusion.  Then you go home and the real 'fun' begins.  Chemo kills all the rapidly-dividing cells of your intestinal tract, so you don't just get sick you get violently ill for days.  Of course your hair all falls out.  Meanwhile your whole immune system is practically wiped out, so you're likely to get sick (and could die) from just about anything.  And remember you're already sick from the medication itself!  Is it worth it to go through the hell of chemotherapy for three months in order to extend your life by three months?  Probably not (unless you're just holding on to see your daughter graduate from college).  Is it worth it to go through that same chemo in order to cure your cancer and go on with your life for the next 30 years cancer-free?  Almost certainly.  These genetic markers pave the way to answering the all-important question patients, doctors, and insurers really want to get at: am I looking at short- or long-term treatment efficacy here?  Will it be worth it?

In this post, I'd like to round things off with a short discussion of something I identified in the first post: namely treatments that specifically target a genetic abnormality.  This is different from the biomarker approach, because often multiple biomarkers will tell us whether a certain therapy or combination of therapies will/will not work.  If my treatment relies on DNA repair, there are perhaps a dozen different genes involved in that process, and therefore a dozen possible mutations I could look for to tell me whether the treatment will work.  The probability these treatments would have been 'effective' without looking at the relevant biomarkers is fairly high.  If something is likely to work for 35% of cancer patients, we'll see that in the data during clinical trials.  We want to know which patients fall into that 35%, but where that information is not available we can probably get there through trial and error.  If, instead, a treatment targets a very specific gene mutation you'll never see it unless you're specifically looking for the biomarker.  During ASCO the frequently-mentioned example was larotrectinib, which targets a mutation present in only about 1% of all cancers.  The only way to know whether to use that drug is if you already know the one biomarker it works on.

The difference here is that with one approach the biomarker makes our treatment paradigm more effective more often, while with the other approach the biomarker is a vital part of the process of discovering and treating the patient.  During one session of ASCO, this topic was discussed at length.  One of the presenters said he'd seen a number of drugs targeting specific genes, similar to larotrectinib, that were eventually dropped by drug companies because it's too difficult to find the patients who have the right gene mutation that their drug targets.  The issue is that not enough patients are getting testing that can detect a large number of gene mutations (like NGS), and they can't afford to test 100 patients for every one they go on to actually treat.

Thus, we're left with a chicken-and-egg problem, where there aren't enough of these kind of genetically-targeted therapies to make the NGS testing worth doing at scale, so patients and payers don't buy them.  Meanwhile, in order to develop this kind of targeted therapy, we need many more patients to get NGS testing.  We're left with far fewer (and less specific, less effective, more toxic) therapies than we might otherwise have.

If we pull together all these threads, we see a way forward for targeted therapy.  As the number and usefulness of biomarkers proliferates, the justification for ordering a multiplex assay like NGS (instead of singe-gene tests) gets stronger.  If all you're looking at is two genes, NGS is more expensive and takes longer than just ordering the tests for the two genes you're interested in.  But at around five gene targets it starts to make more sense to order NGS over a bunch of one-off targets.

Multiplex assays have been around for a while.  Early in the 2000's we saw so-called SNP chips (pronounced 'snip chip') capable of analyzing thousands of single nucleotide cancer-specific differences.  They got cheaper over time, so cost wasn't the main issue.  The problem with SNP chips was they could only detect changes in a single base pair of DNA.  Therefore they were of limited usefulness, and never really supplanted single-gene tests in oncology.  Next-gen sequencing is a more recent tool that is becoming more well-known as it's becoming broadly available.  With NGS, you get actual sequence information about a whole gene.  So for example, larotrectinib only works on patients with a gene fusion (one gene stitched to another).  Until recently you could only get that kind of information by looking for it in one-off tests; you don't get it with a SNP chip.  NGS can look at dozens of gene fusions while also looking for hundreds of other targets.

One of the most common tests I see is from Foundation Medicine (FM).  They're one of the earliest and largest companies doing NGS testing.  Their first test became commercially available in 2012.  At only seven years old, that's practically brand new in research terms.  The first drug approved based almost entirely on NGS testing came a mere 6 years later.  This is a testament to how fast cancer research is at capitalizing on new discoveries.

NGS as a tool promises to dramatically change the landscape of cancer as we figure out how to work it into standard clinical practice.  Things we could previously only do at academic research institutions we now have a path to get to community clinics.  As we get to the tipping point with insurance/payers where it makes sense to pay for the tests, testing will have knock-on effects elsewhere.  Patients will be able to skip a therapy that wouldn't have worked anyway.  Or we'll learn the patient needs a second drug to make the first one work.  Other therapies that didn't make sense to develop in the old days will blossom as the patients who could benefit are identified.  These would be targeted therapies with high specificity, high success, and few off-target effects.

All these improvements will require institutional changes at the foundation of cancer treatment.  Today, we look at cancer in a very tissue specific way.  The most important feature we focus on is the organ the cancer cells originally came from, whether that's breast, liver, lung, prostate or something else.  That's not going away any time soon, but we'll need to layer on another specialty on top of that framework in the near future: molecular oncologist.

Groups of these specialists will review patient cases, analyze pathways that are activated/inhibited, and use this information to advise what will be most effective against each person's cancer.  This is already being done in some academic research centers.  As testing becomes more common, molecular specialists will need to roll down to smaller community centers as well.  There was some discussion at ASCO about how to start making this happen.

These are all mild, near-future speculations.  Many of them are in the process of being implemented today.  They are exciting in their potential to dramatically improve treatment outcomes and ensure patients don't have to undergo difficult but futile therapies.  I'd like to finish off with some more long-term speculations.

Let's assume many targeted therapies and biomarkers are validated and become broadly available.  Certainly many current clinical trials are focused on exactly this type of approach.  Now imagine ten years from now you get cancer, and your doctor sends out for NGS testing right away.  You go talk to your doctor to discuss the results.  He tells you the standard information you'd get today, "You have stage 4 pancreatic cancer.  This is a very aggressive type of cancer and we need to treat it quickly before it spreads and becomes fatal."

In addition, he also tells you some genetic information, "We looked at the genes that are driving your cancer, and compared it to the therapies available to you.  It looks like immunotherapy isn't going to do you any good, but we can give you a PARP inhibitor along with an ATR inhibitor, and that should be very effective given your tumor's genetic profile.  You also have an NTRK fusion and mutations in four other genes we looked at.  We have drugs for three of these, including NTRK."

Now, you might think you'd have a long discussion about which of these therapies to try first, and I suspect at first that's what we'll do.  But what if instead we gave you multiple treatments at once?  Not just the combination therapies I already discussed, but treatments against all the different genes driving your cancer cells.  If we know your cancer is 80% likely to respond to a treatment, and we give you three or four treatments at once the likelihood of the cancer responding goes up dramatically.

Remember that the way cancer cells become resistant to a treatment is through selective pressure from the treatment itself.  If a resistance mutation is present in one in a million cells, a single-mechanism approach will kill off 999,999 cells, leaving only the handful of resistant cells behind.  Once those few cells grow back in a month or so all your cancer cells are resistant and the drug doesn't work anymore.  However, if you treat a patient with two drugs at once, the chance that the same cell will have the resistance mutation for both pathways is much lower.  So low its likely no cells have both mutations, meaning all the cancer cells die.  The chances of a 'cure' through this approach go up with a third or fourth drug given at the same time.  This is similar to using combination pesticides on crops.

Are we ready to try this kind of approach?  Not even close.  The first problem is finding any mutations we can do something with for patients who get NGS testing.  There are many mutations in any one person's cancer; but often we don't have a treatment that targets them, and there are many mutations we're not testing for because we don't know about them.  Those problems should go away over time, simply by continuing the same strategies we're pursuing today.

The other problem is that an approach where each patient gets a different blend of treatments isn't something our current clinical trials paradigm is designed to work with.  The FDA has been very understanding in approving drugs like larotrectinib, where the patient population is so low you can't do phase 2 comparison trials with standard of care.  Larotrectinib was approved based only on treating patients with the mutation, not on comparison with other treatments.  Now, those patients responded at an unheard-of 75% rate.  (For context, many therapies are approved with response rates of less than 30%.)  The hope for precision medicine is that this kind of high response rate becomes the norm, not a fluke.  However, if we start treating patients with multiple uncommon drugs, we won't know what kind of drug interactions we might see.

Normally we run a phase 1 dose-escalation safety trial for any new combination therapy.  But in a world where oncologists cobble together a therapy based on the genetic report, the selection of drugs are specially tailored to each patient.  There are potentially thousands of different drug combinations, and there's no way to run a clinical trial analyzing safety for every combination.  Eventually, the FDA and other regulatory agencies are going to have to figure out new ways to assess drug safety when treatments become so specific they're only given to one patient.

Is sequencing and precision medicine the magic bullet that will finally end cancer? I don't know.  I haven't even talked about other exciting therapies like CAR-T cells (and CAR-NK cells!).  This avenue of research is advancing quickly - especially in leukemia.  Interestingly, they were also using sequencing to help target their therapies and ensure they delivered the right chimeric antigen receptor to the patient.  Even if we don't realize all the promises of precision medicine, it's clearly going to play a major role in cancer care in coming years.  I suspect we're already past the tipping point, watching as the whole field shifts in an accelerating way.

Comments

Popular posts from this blog

Reverse Engineering Life