Drug target 을 어떻게 찾지? 어떤 target 이 좋을까? 라는 상당한 고부가가치의 질문에 ‘바로 이거야’ 라고 선뜻 답할 수 있다면, 내가 여기서 안 이러고 있지. $$$$
어째든간 이래저래 insight 를 길러보야 할 일이다. 본문에도 언급되어 있지만 empirical 한 작업이고, 무수한 trial-and-error 없이는 나올 수 없으니깐, 오늘도 피펫질이나…
계속적으로 자신의 경험을 바탕으로 모델을 만들어가는 일이 중요하다.
First, a quibble. Diseases: his high bound seems about right, but the reality is that incidences can go down to essentially 1 case. Particularly in our world of high throughput sequencing, in which Mendelian diseases can be identified with tiny numbers of patients or in which we can parse cancers down to specific combinations of putative causative genetic alterations. Now at some point rare diseases are just too rare for commercial success, but I’ve had commercial-minded folks at three different companies dodge that question, so I don’t know where that is. Even if a genetic condition is ultra-rare, finding them is a golden opportunity, as these are natural genetic experiments. PCSK9 is the poster child for this approach, but I expect there will be many more.
Let’s look at drug targets from the perspective of making drugs against them. First we need to decide what is a drug, and that even became quite a mess. Therapeutics either in use or being considered by reputable groups include energy (X-rays, UV, sound), engineered human cells, engineered bacterial cells, mixtures of natural bacteria, bacteriophages, various flavors of gene therapy, proteins, nucleic acids, small organic molecules and even inorganic ions and gases. This essay will mostly focus on small molecules with perhaps some touching on some of those others.
In any disease, there are a set of biological/biochemical phenomena which represent a deviation from the biological norm and we wish to correct. A goal of modern biology is to parse these down to lots of subphenomena, chains of events which constitute the disease. Sometimes these may even be simple, but it is possible to make complex the simple.
For example, there are a growing number of single-gene Mendelian disorders. Some of these can be treated with simple dietary interventions (e.g. phenylketonuria — and by simple I mean in concept; it is well known that actually following a PKU diet requires constant vigilance in the style of Alastor Moody). Genzyme made a mint selling an enzyme replacement therapy for Gaucher’s disease. These direct assaults on the disease are appealing, but perhaps not always possible An alternative would be to somehow use a drug to ameliorate the downstream side effects of the imbalance or prevent the imbalance.
Not a metabolic error, but an example would be sickle cell anemia, which is caused by a defective beta hemoglobin. There’s growing hope that this can be cured with gene therapy approaches, but some existing treatments actually work by stimulating the body to use a different form of hemoglobin, the fetal hemoglobin used prior to being born.
As another example, I’ve spent a sizable amount of my professional career thinking about attacking cancer via understanding what goes on in a cancer cell. What are the critical genetic circuits that make a cancer cell a cancer cell? I also spent time at Infinity on a program, which sadly failed, targeting the tissue around a cancer cell, the stroma. There is a great deal of evidence that the tissue around a tumor can interfere with treatment. There’s also the whole approach of targeting angiogenesis, which hasn’t proven to be a panacea for cancer but remains an important topic.
Actually, more than a little time at Millennium was spent trying to understand how much of bortezomib’s therapeutic effect in myeloma is due to effects on the myeloma cells themselves and how much is from altering their deranged interactions with other members of the bone marrow community. For example, myeloma is sometimes diagnosed due to a bone fracture because myeloma cells can alter the balance in bone between the bone-building osteoblasts and the bone-destroying osteoclasts. The latter are my favorite bit of Greek root onomatopoeia, from clastos “to break”. Can’t you hear a pot shattering with that word? Anyway, the usual balance between these is responsible for the great strength of bone and the ability to heal, but myeloma cells drive excess clastos.
Now there’s a huge resurgence of effort in the immune angle on cancer. How can we stimulate and/or train immune cells to go after tumors? Conversely, when immune cells are co-opted by tumors, which can also happen, how do we alter that? Trying to stray back to my point, we can tackle tumors by going after the tumor, going after the cells near the tumor or targeting the immune response to cancer. Similar ideas hold for many other disease areas: you can target the diseased tissue or some other part of the body. How should we target diseases? With apologies to Malcolm X, “By any means necessary”.
Thinking about this reminded me of my deficiencies in higher level physiology. When my parents throw medical questions at me one of my standard joke defenses is that I didn’t study much at levels higher than a cell. Genes beget RNAs that often beget proteins which change cells and biochemistry, but too often I stop there (doesn’t help that I work on bacteria, in which there isn’t a lot above that and what exists isn’t well understood). Cells are in tissues, tissues in organs, organs in systems and ultimately everything comes together to function as a human body. Thinking about drugs ultimately requires thinking about all those levels.
Okay, so what’s a target? Well, most targets fall in a few categories: lipid, DNA, RNA and protein.
Lipids haven’t been a hyperproductive target, but there are ionophore antibiotics you can buy over-the-counter in the U.S., so we shouldn’t discount them altogether. Some anesthetics may work by simply altering membrane properties, or at least that was a hypothesis that has existed in the past. But it is difficult to get specificity against lipids and membranes this way, so not an active area.
Some drugs just damage DNA directly. Primarily anticancer drugs — nasty alkylating agents and such. Again, hard to get specificity.
RNA isn’t the target of many drugs, but perhaps that will change in the future. Certain antibiotics, particularly the aminoglycosides (e.g. streptomycin) do specifically target RNA. Serial bioentrepeneur Michael Gilman’s newest company is trying to target RNA as well, though his effort as well as antisense and siRNA approaches are typically aimed at changing proteins. The biology of interest is happening at the protein level, as opposed to aminoglycosides that really are targeting functional RNA (ribosomal RNA).
So then proteins. They are the bulk of targets and probably will be for the forseeable future. And then the big problem is druggability, which I wrote on a while back. To summarize, we can identify human proteins which are relevant to disease but many of these are difficult-to-impossible to reach with drugs. That’s why approaches like drugging RNA and siRNA are appealing or others which are emerging: to drug the undruggable.
To give an idea of the problem, we’ve known for something like 30 years now that a huge number of solid tumors, and also some hematologic ones, either have mutations in KRAS or NRAS (or rarely HRAS) or are signalling through these proteins. They are small GTPases with many protein interactions; the extremely slow GTPase activity serves as a molecular timer and some interactions make the timer run faster or slower, whereas others are the downstream acceptors of the timing signal. But KRAS has essentially defied drug discovery, because it is an intracellular protein with few crags on its surface. My employer is one of the groups (something that wasn’t disclosed when I wrote the original piece) trying to drug KRAS despite these challenges. Of the flood of possible disease targets that have emerged in the last few decades, KRAS is more then norm than the exception.
If you want to hit an extracellular target, then antibodies are an option, unless your target is behind the blood-brain barrier. For intracellular targets, small molecules rule — with those exceptions. For a small molecule to work, the general rule is it must bind specifically near the active site of the protein. Or it needs to find a nice pocket somewhere that allosterically alters the protein, perhaps changing the shape or locking in one specific structural pose.
The end result of this is that most small molecule drugs target a very small number of categories of proteins. Enzymes, but not all enzymes. Cell surface receptors, particularly the G-protein coupled receptor (GPCR) type, but again not all receptor types. Transporters and ion channels in the membrane. Nuclear hormone receptors, which are receptors but not cell-surface ones. There’s all sorts of exceptions (which I can’t think of) and I think there are a few small molecule drugs with unknown or controversial mechanisms (lithium isn’t settled yet, is it?).
Then there is the problem of specificity. If I give a drug to a patient targeting GPCR X, how many other GPCRs will I hit? What side effects will that cause? Or perhaps GPCR X mediates the pathology of interest but also some perfectly normal physiology elsewhere? The drug may hit some totally unrelated protein as well, particularly after the liver and its enzymes have given it a going over. Perhaps the drug accumulates in some strange way that enables it to hit a low-affinity target in an undesirable location. Or pH or other local conditions alter the drug’s action in some interesting way.
Plus, you can’t get a small molecule to do just anything. With enzymes, your molecule is almost certainly going to be an inhibitor. So if you want to stimulate an enzyme, that enzyme probably won’t be a direct target — you’ll probably use “the enemy of my friend is my enemy” thinking to find something whose inhibition leads to stimulation of the enzyme. Channels and receptors are more complex things, sometimes you can find agonists which activate them and other times you can find antagonists which truly block them or inverse agonists that bias them away from being active.
As noted, there are many efforts to break out of the frustration of druggability, but all are early. Our approach is to reprogram some nifty natural product chemistry. Three natural product drugs which break the usual rules of druggabilty by binding to flat molecular surfaces are rapamycin, FK506 and cyclosporin. Each does this by “cheating”; rather than tackling a target alone they bind an abundant intracellular protein with one end of the molecule and then that hybrid molecule binds to the flat surface on the target with amazing specificity.
Another approach with some conceptual similarity are PROTACs, which is a strategy by which small molecules are used to bring together a target protein and a cellular enzyme that gives proteins a kiss-of-death. Apply a drug of this type and the targeted protein should disappear from the cell in a dose-dependent manner.
Vertex’s first cystic fibrosis drug pulls off a minor miracle: a drug which actually improves the function of a flawed protein. I was one of the fools who thought this approach was doomed to failure when Vertex first announced the effort; it is great to be proven wrong. This idea of “chemical chaperones” appears to be gaining some momentum, but how many disease targets could it apply to?
There’s probably a bunch out there I haven’t seen or lost track of. Cell penetrating peptides and ways of getting antibodies into cells have been discussed, but I don’t know how they stand. Local unicorn Moderna wants to express functional mRNAs as drugs. As with any nucleic acid scheme, they will have to deal with problems of delivery (liver or direct injection are great, anywhere else appears to be a challenge) and toxicity (both Moderna and Alnylam have had recent news around toxicity for their RNA-based drugs).
So suppose you have a novel target and you’ve found a candidate drug against it. You’re far from done. The problem is we really don’t understand the biology at anywhere near the level you’d need to predict outcomes. Until you’ve drugged patients, you don’t have a real chance of understanding the biology of drugging patients, and even then you don’t know every possible drug/diet/environment/genetic interaction that might show up in a few patients and wreck havoc. Laboratory approaches to validating targets give insight, whether they are cell culture models or RNAi experiments or overexpression experiments or what have you, but they aren’t the same as drugging a target. Nor are those natural genetic models I praised near the beginning of this piece; partial inhibition of a target chemically with defined temporal endpoints is still different than a complete genetic knockout which has been lived with since conception.
How so? Well, for one thing cells are resilient. Particularly cancer cells, but it applies everywhere. There may be some less efficient or backup pathway that can be activated. Or some sort of rescue-and-repair pathway. Or perhaps your choice of blocking the pathway doesn’t pan out because too many signals enter downstream of your target. Perhaps the dynamics of the target’s activity play a complex role. Biology is complex and our ignorance runs deep.
Another, touched on above, are undesired effects of your drug. Hitting related targets. Hitting the same target in cells and tissues you don’t wish to target. Hitting completely unrelated targets. There could also be some previously uncharted pathway that depends on your target and now isn’t happy either. Or you’ve inhibited an enzyme, and now the precursor it works on backs up in the cell, triggering an undesired response. To give one notorious example similar to this, creating a fat-like substance (Olestra) for making potato chips which can’t be absorbed in the gut means you’ve introduced a very good lubricant to the lower portion of the gut. ‘Nuff said.
Of course, even getting to targets can be a problem. Look at the wasteland which is Alzheimer’s disease approaches; it’s easy to conclude that little of what we think we know about the biology of this awful disease is of any value. You can moderate your risk on a specific target using a cell-based screen, but now you’ve traded one kind of risk (target) for another type (model). Again, for something like Alzheimer’s there may simply not be any trusted models for a phenotypic screen.
I have hardly scratched the surface of what can go wrong in drug development program. For example, animal models can give grossly misleading results. There’s plenty more; since I don’t specifically focus on these I let the people who do keep the catalog of what can go awry in their sector. Sometimes it seems remarkable that any drug has ever been successfully launched and made a difference, but of course many have. Remembering that — and all the patients who still have pressing needs for new drugs — is sometimes critical to keeping the spirit to continue plunging ever forward.