Like a lot of people in the community around cystic fibrosis, I’ve become a bit obsessed with understanding a relatively fundamental problem: put generally, how do proteins fold?, or more specifically, how does that folding cause or alleviate the symptoms of cystic fibrosis? Cystic fibrosis is, after all, an inherited condition characterized by a lack of functional CFTR protein. The body’s problems in CF are myriad, but they all result from one of more than a thousand possible mutations in this single protein. So my time, resources and obsession have been funneled far too frequently in the last months toward understanding what that protein does, how it comes together, and how a new class of medicines—so-called “small molecules”—work to help restore some of that protein’s functionality. Most of the people in the community around CF are excited about these molecules, because one of them—a so-called “potentiator”—was approved by the FDA in January of 2012 to treat a very small subset of the CF population who carry a mutation called “G551D.” It’s a good drug. It greatly slows and may actually cease the progressive lung disease that most often leads to death in people with CF, and it reduces a measure of what is called “sweat chloride” to below the diagnostic threshold for disease.

As more and more of these molecules affecting CFTR folding have been found, though, I’ve become more and more curious about what exactly they do. And by extension, I’ve become more and more curious about the basics around what CFTR protein is and what it does. So I thought I would write up what I’ve found so far, and to share the understanding I’ve come to about how pharmaceutical companies market the potential efficacy of the drug candidates they’re bringing to trial. The basic message I’ve come away with is this: it’s a long way from lab bench to bedside. But that doesn’t mean the path to clinical efficacy is impossible, even if we should treat corporate claims with some skepticism.

So what is CFTR?

Well, the short answer you hear from people is that it’s a protein that sits at the surface of epithelial cells and lets chloride ions out. That’s certainly a decent summary. The reason that you need to let chloride ions out is that they join sodium (at least in the respiratory tract) in the so-called “ASL” (airway surface layer), making a saltier solution in the fluid just outside the cell membrane and drawing more liquid to the area outside the cell. With epithelial cells, this liquid is really important, because those cells have little hairs on them, called cilia, which beat back and forth very quickly in a normal person’s body, moving the fluid and the mucus that is constantly forming in this extra-cellular area. If you don’t have enough water, the mucus gets too dense and the cilia can’t whip around. That leaves mucus sitting in place, rather than moving through the body to pick up and usher out any bad critters the body doesn’t like. And clogged-up, dense mucus is a really good place for those bad critters to multiply once they make it into the body.

You want CFTR to work so that you can have the right balance of fluids in and around the cell, which means you’ve also got mucus that trundles happily along on its bed of tiny hairs. In CF, it doesn’t work. So you get the bad results.

However, this story sells CFTR a bit short. It’s a complicated beastie, and if all it did was open a span in a cell membrane that lets chloride ions out, that would be boring. So here’s a bit more complicated picture:

When a protein is born, it comes about as a long series of amino acids held together with peptide bonds and forming peptide chains. You can think of it like a very, very long one of those paper chains people make for parties out of single strips of paper that are cut up and glued together in rings. First it comes out as a chain from the cell’s protein synthesis team of mRNA, tRNA, and rRNA. Then it has to fold into the shape. The folds happen because bits of the protein want to get closer to one another due to their chemical composition, scrunching the chain up into a variety of possible shapes, and the whole thing hangs together with little hooks that some people call “peptide backbones.” One of the most common mutations in CF, ∆F508, removes a crucial peptide backbone, and the entire assembly falls apart as the protein moves to the cell surface. More on this later, though. Let’s just say that the protein folds into a big functional shape, and its component shapes are those needed to perform the task that the protein accomplishes.

In CFTR, that task (letting chloride ions out of the cell) gets accomplished by five separate “domains” of the protein working in concert, all of which perform an important function. The first domain produced in protein synthesis is the MSD1, or Membrane Spanning Domain 1. Its function is to open a channel in the cell’s membrane that lets chloride ions escape the cell. However, it doesn’t just stay open all of the time. The MSD1 only opens up when there is ATP (adenosine triphosphate) caught up in the next domain on the chain. That domain, the next to be built in protein synthesis, is the NBD1, or Nucleotide Binding Domain 1. The NBD1 essentially catches molecules of ATP, which is a co-enzyme, and breaks it into ADP and inorganic phosphate. This releases energy and allows the protein to move ions out through the MSD1—if, and only if, there is an enzyme called PKA (protein kinase A) bound up in the next domain down the line: the R domain.

The R domain appears to be an oddity, and my sense of the literature is that it’s not clear how exactly it works. However it does, it seems to make the NBD1 and NBD2 better at gathering ATP, and unless there is PKA caught in the R domain, the MSD1 doesn’t open up the cell membrane to let chloride ions out. So it’s important, and seems to be linked to the activity of both NBD1 and NBD2. However, as the diagram above shows, it’s not linked directly to NBD2 in the chain’s sequence, meaning it must exert its influence there through a bit of protein that hooks all of the domains together. The next domain on the chain is another Membrane Spanning Domain, MSD2, and the one after that is another Nucleotide Binding Domain, NBD2. They appear to work much like the MSD1 and MSD2, though I’m fuzzier on this because they are far less frequently discussed. So with this and any other detail, I’d appreciate any input that I can get to improve this discussion.

So, that first diagram was—as you can see from the 3d model above—a bit of an oversimplification. This model, however, is more like it. Those tight green coils go into the cell membrane (MSD1 and MSD2, separated by shade) and are the channels on which chloride ions leave the cell. The pink nest-like bundles below are the butterfly nets for ATP that activate the green coils, and the big gray blob is the R domain, which gathers the PKA. As you can see, when it folds itself into shape, it has a number of segments that work to hold the shape of the protein and link the domains together. You can kind of imagine the interlinked actions of the whole by thinking of an impossibly elaborate draw-string bag.

This Duluth Pack bag is actually a good example, though it’s not quite elaborate enough. At the top of the bag, you have a circular channel that the draw string is threaded through. Now let’s say that in order to get your arms through the slack, the top had to be cinched up entirely. That gets closer to understanding the interlinked chain of events required for CFTR protein to function appropriately. But it’s still not quite elaborate enough.

Instead, let’s say you had two inter-dependent circular channels at the top of the bag. In order to open one side, the other has to be closed, though they can both be closed—sort of like a figure-8 at the top of the bag, but with only enough strapping running through it to open one side at a time. And you still need the top cinched in order to get your arms through with enough slack, but the top can’t close until the bag gets something in it. Now with our CFTR-bag, you have to get some PKA into the R domain side of the bag. It’s a good shape and size of opening to fit PKA into it, though, so let’s say some drifts into the bag. Well, now the bag automatically cinches shut on the R domain side only to open up on the other, which is its NBD side. This, luckily, opens the top of the bag to a shape that’s just right for fitting some ATP. And as chance would have it, some ATP just drifted in. The pressure of the ATP as it sits in the bag then cinches that side of the bag shut on top as well. This cinching lets out slack in the straps, which are just loose enough to let some chloride ions get through them.

Not the most elegant analogy, I know. But it gives a rough sense, to my mind, of how all of these things have to fit together not only by sequence, but also by physical necessity and shape. The protein has to get itself into a shape where all of these dependencies can be effectively linked, and that shape isn’t an abstraction of modeling—it’s an actual physical shape that the protein has to fold itself into in order to function. And when the protein doesn’t get into the exact right physical shape, you get defects, degraded protein and disease.

What goes wrong with CFTR in CF?

So defects and disease: they are the background noise of the place from which most people step into the world of CFTR proteins. As I mentioned earlier, there are thousands of mutations on this protein that can lead to the clinical diagnosis of cystic fibrosis. So on some level, it’s not particularly useful to think about cystic fibrosis as a single condition—rather it is a diagnosis that covers a spectrum of conditions that all result from mutations on the same protein. I would say that the thing uniting them all is that they all result in the protein’s degradation or its failure to perform the function that the body requires it to perform, but that’s not even inclusive enough. There are mutations that leave some function of the protein, too (these are so-called “residual function” mutations). We could say, as much of the CF community does, that there are “classes” of mutations which can be roughly grouped together. But as I hope will become clear later, this isn’t really as descriptive or helpful as it sounds. Every mutation messes up the protein in its own special way, and when we get into our discussion of small molecule medicines below, I think it will be clear that we have to be mindful of those individual difference.

The most common mutation in CF is ∆F508. We think it’s been so successful because, as heterozygotes, carriers of a mutation at ∆F508 have had a better survival rate with certain pandemic illnesses than non-carriers. So nature, in its blind and dumb way, has managed to select for that mutation in a limited way and in limited populations. But it really does mess up CFTR protein. What is missing from the CFTR made with this mutation is a single link in the chain—a little piece of phenylalanine—in the region of the protein that makes up NBD1. However, it’s an important piece, because it helps assemble the shape of the whole protein—its one of those hooks I mention above that hold the scrunched bits of chain together. And without it, the protein gets all floppy and falls apart on its way to the cell surface (with some help from another cellular mechanism, which I’ll describe below).

What would it normally hook together? Well, a 2008 study, from which I’ve take my 3d model above, indicated that this piece of phenylalanine would normally hook NBD1 in with MSD2 at one of the “cytoplasmic loops” holding the pores in that domain together. Remember the tight green curls? Those poke through the lipid layers of the cell membrane, and inside the cell, dangling through the cytoplasm that all the cell parts float inside, there is a little string of the protein that ties each curly tube together like the casing-ends on a sausage (the blue curly guy up above). This casing-ends loop should work a bit like a hook that hangs a chunk of the NBD1 from it, but when the phenylalanine isn’t there, the NBD1 can’t catch the loop to hang. See the way the F508 lines up with the F1068 and the F1074 in the model above? With ∆F508-CFTR, the protein is missing this hook. And the protein lacks a similar interface with MSD1 and NBD1 (at CL1) because of this same missing piece. So then the wobbly protein comes up to the cell’s quality control mechanism, the ERAD (or, Endoplasmic Reticulum Associated Degradation) system.

In the ERAD system, chemical feelers drift around the protein and they bind to common weak points that identify a protein as requiring degradation (i.e., cutting up and losing the parts). These weak points are basically odd shapes with distinct chemical signatures in the poorly formed protein that indicate that it won’t work as well as it should. Sometimes that might be a hydrophobic region, like we find in ∆F508-CFTR. When ∆F508-CFTR goes through the ERAD, one of its chemical feelers, which is called “RMA1" but let’s call it “sticky-tag” for clarity’s sake, sniffs out and attaches itself to the loop at the casing ends where we’d normally find phenylalanine, because it’s noticing that this is a region in the protein that is hydrophobic. Then, as the wobbly protein moves down the line, this “sticky-tag” has essentially marked the protein for further chemicals that—to put a fine point on it—scissor the protein to pieces. Or at least that’s one narrative we can give for how things go wrong (see Younger, et al., “Sequential quality-control checkpoints triage mis-folded cystic fibrosis transmembrane conductance regulator,” Cell 126.3, 2006).

Another way of understanding the degradation that happens with ∆F508 is that yet another kind of intra-cellular chemical, called keratin-8, builds itself up on a similar spot in the protein once it arrives in the cytoplasm, but the same essential process happens after that. One keratin-8 latches on, then a whole mess of its friends do the same, and this cuts the whole protein apart (“Disruption of cytokeratin-8 interaction with F508del-CFTR corrects its functional defect,” Hum. Mol. Genet. 21.3, 2012).

So if you wanted to fix the ∆F508 version of the CFTR protein, one thing to do would be to send a molecule into the cell that binds onto the CFTR protein while it is being assembled so that it can hold the shape of the protein together. This basically covers up the hydrophobic region of the protein that’s exposed when we don’t have our phenylalanine hook, so the ERAD system doesn’t see the defects. And that’s how pharmaceutical companies are looking to treat cystic fibrosis right now—by developing drugs that break down after ingesting a pill, distributing these little molecules throughout the body to the cells affected by these protein mutations.

Testing CFTR function

To this date, the only proven and effective drug that uses this molecular approach to treat the protein defect at the heart of cystic fibrosis is called ivacaftor, or Kalydeco. It would be too complicated to explain in detail the mutation that ivacaftor helps to alleviate and the way that it does that work. At least to do that here. So instead I’ll just say that the mutation it helps with is called G551D, and the place in the protein chain that it affects is (as you might have guessed from its number) not terribly far away from ∆F508 in the NBD1. That means that it introduces a problem with the way the protein deals with ATP, and results in an MSD1 that cannot open. Remember, the weird drawstring analogy: you have to move NBD1 to open MSD1, and because the protein doesn’t have the right movement in NBD1, it can’t open MSD1. So Kalydeco helps MSD1 open up by making ATP processing unnecessary, probably binding to MSD1 itself. There is a good paper that explains this in more detail (notice that these authors use “TMD” or “transmembrane domain” for what I’ve been calling MSD). But for simplicity’s sake, let’s just note that it doesn’t fix the protein, but does increase the probability that chloride ion channels in the MSD will be open when they need to be.

So how do we know when one of these molecules is working, and how do we know how well it is doing the job we want to it to do (i.e., rescuing enough function so that the protein can let some chloride ions out of the cell)? Well, the standard test for determining whether a cell’s CFTR are functioning is called a “patch clamp” test, though it’s a bit deceptive of me to call that “the test.” There are actually scads of variations. And there’s also something called an “Ussing chamber test,” which is the one used quite often by pharmaceutical companies to determine partial CFTR function. Here’s an easy to follow diagram for one example of a patch clamp test:

Now these are tests designed to measure the electrical function across the entirety of a cell, or even sometimes at a single protein channel. Why do we want to measure electrical function to see whether CFTR is working? Well, the trick is in the fact that the chloride it lets out of the cell is an ion, meaning it’s a charged particle. If the CFTR is working, that should mean that when you prod the cell with an electrode (and how, exactly, you do this depends on what you’re trying to learn), you see a reference current’s worth of ionic activity through the cell membrane. Current, you may remember from school or from your fuse box, is measured in amperes, and it’s essentially a measure of volume. You’re checking how much traffic there is, of a very specific kind—chloride ion traffic. And you get that traffic to flow, typically, by introducing a chemical “on-switch” called forskolin (forskolin stimulates cAMP activity—another complicated digression—but basically it’s the juice you need to get transmembrane ion transfers running).

Remember that I said above that pharmaceutical companies are more often using something called an “Ussing chamber?” Part of the reason for this is that Ussing chambers are very good for testing what they want to know—how are a whole mess of cells, spread out in a layer together, going to function when you add some of these small molecules to them? In these tests, you take a single layer of epithelial cells on a permeable “support” and fit it into a chamber that has a special solution in it. This layer of cells divides the chamber in half, so that if you measure the voltage on one side—by virtue of the fact that epithelial cells are actually kind of special and have what are basically “up” and a “down” sides (apical and basolateral, if you’re interested)—you can measure ion transport by seeing how much of a difference you see from the other side. In other words, in healthy cells, you have ions kicking out of the epithelial cell layer into the solution, changing the voltage of the apical side of the chamber. And you can see just how many ions by using another electrode to inject more current into the solution until the voltage difference between the two sides of the chamber equals zero.

In testing VX-770 in 2009, VX-770 being the corporate code-name for ivacaftor or Kalydeco, the scientists working for Vertex (the corporation who discovered and commercialized the molecule) used this test with a reference set of human epithelial cells and then with cells having a couple of different mutations that had been treated with VX-770. Here’s what they found:

See the “IT” (apologies—I can’t figure out the subscript for that T)? That’s ion transport. And the µA refers to a measure of current, microamperes, over a square centimeter. The numbers on the left are the volume of microamperes, and the numbers on the right indicate the % these volumes represent of what you see with reference “non-CF-HBE” or “non-cystic fibrosis-human bronchial epithelial cells.” Another way this gets written is “WT” or “wild-type.” The bits on the bottom tell us about the concentration of the molecule that has been applied. Now an interesting detail from the Van Goor study is that even in the absence of any molecular helpers like ivacaftor, they found that there was ~4% CFTR ion transport on cells with CF-causing mutations. In other words, and this is important so I’m going to put it in bold type and italics, even CF-causing mutations demonstrate some ion transport activity (about 4% of normal) with their CFTR proteins in the Ussing chamber. And this replicates studies from the 1990s that show that forskolin-induced ion transport happens in CF-mutant cells at around <5% of normal volumes. Hold onto this for later: there are studies from more recently suggesting something slightly—but importantly—different.

So how does this translate as a medicine that affects the whole body, and not in carefully doped cells lined up in a very careful row?

Measuring CFTR activity in living people

Well, one way of answering that would be to see what clinical effects you get when you give ivacaftor to people with cystic fibrosis. Here’s what that data looks like:

Huge, unambiguous swings up in lung function. And, as other data have shown, reduced pulmonary exacerbations, fewer IV antibiotics, and much much more. But what we’ve been talking about so far hasn’t been clinical manifestation of disease. We’ve been talking about something more basic—how well cells can get rid of chloride ions. There are actually two ways of working backward from the whole system of the body to the sort of measure of ion transport we saw in the Ussing chamber data. One is by looking at sweat chloride levels, and the other is by looking at the results of a similar test of electrical function to the Ussing chamber and patch clamp tests we were talking about above. Here, for reference’s sake, is what the study above saw in sweat chloride tests:

I’ll discuss this data in a bit, but it is somewhat confusing because it isn’t a directly comparable sort of test to the Ussing chamber test. It is relatively easy information to gather, though, so it’s something that a lot of researchers interested in figuring out how well small molecule drugs are working have examined. The test more directly analagous to the Ussing chamber and patch clamp tests is something called a “nasal potential difference” test.

The nasal potential difference test is annoyingly difficult to do, invasive, and only conducted at a few places around the U.S. It’s also the only good in vivo test of chloride electrical function that I know. What is it? Here’s how Johns Hopkins’s CF center describes the test:

[Nasal] potential difference can be measured by placing an electrode on the lining of the nose. After the electrode is positioned, the lining of the nose is bathed in a series of solutions that contain different salts. These solutions are designed to change the flow of ions across the epithelium in predictable ways, thus changing the potential difference in predictable ways. These solutions contain (1) a Ringer’s saline solution (a special salt solution used to obtain the baseline NPD), (2) amiloride which blocks sodium channels, (3) a chloride-free solution and (4) isoproterenol, which stimulates CFTR.

Early studies proving the viability of nasal potential difference tests as standards for both diagnosing CF and identifying lower CFTR activity (in individuals with other respiratory conditions, like asthma) were done by the same researchers now developing an inhaled gene therapy product in the UK (see this paper for their early studies). And it is one of the chief measures used in their research for gauging the efficacy of the gene therapy products being developed (here is an abstract explaining the results of their single-dose “run-up” trial). As a way of measuring the move from basic science to bedside, it is one that is relatively well-accepted. It is not what Vertex (the major developer of small molecules for CF right now) has wanted to measure for most of their studies. It’s actually kind of hard to find an example of Vertex’s scientists capturing NPD in their studies, but there is one. They looked at NPD for the Phase II study of Kalydeco. And here’s what they found (presented as published five years later):

The authors of the study this is from conclude that ivacaftor restores around 35% of CFTR activity, but that’s curious given the data that they offer in support of that claim. It looks much more like ~30% to me for the dose they commercialized, until you add in sweat chloride data (which we’ll discuss momentarily). How does that compare to clinical manifestations of CF? Well, there are some uneven and weird data out there, but the 2014 study above pulls everything that its authors could find together into a table, and here are some takeaways: carriers of a CF mutation on their CFTR protein who have some manifestations of mild clinical symptoms (chronic rhinosinusitis, like me) have CFTR activity that is around 75% of normal on this test; people with one or two mutations, but whose only symptom is something called “congenital bilateral absence of the vas deferens” (CBAVD) have CFTR activity that is in a range between 35 and 50% (this is actually broken out in the study); pancreatic sufficient CF patients, whose symptoms tend to be milder but still offer evidence of chronic disease, have a range of around 6-15% activity; pancreatic insufficient CF patients are <5%.

So given the range of CFTR activity indicated in the nasal potential difference test for a variety of conditions, it would seem to be the case that for a molecule to have clinically meaningful activity in its rescue of protein function, you probably want to aim for somewhere between the 15% of activity that you see in CF patients with pancreatic sufficiency and the 75% of activity that you see in carriers with this test (and we have to remember: comparison by the same measure is what should count here). Given the range of clinical presentations in the middle, it’s probably better to err on the high side of that spectrum before we can talk about things like “functional cures” as you hear some people doing with ivacaftor. While it seems that ivacaftor rescues enough protein function to render the disease invisible to sweat chloride testing, it only leads in vivo to an amount of electrical activity somewhere between that of pancreatic-sufficient-CF and CBAVD, which is more or less free of the progressive disease we see in CF. It’s in the fuzzy realm of CF-related disorders, which—while a hell of an improvement over CF—are still full of complications for one’s health.

Corporate claims about CFTR

But if you listen to Vertex, you find the claim that ivacaftor rescues somewhere between 35 and 40% of CFTR activity in vivo—right around that of CBAVD. And the study I’ve cited just above, which included the lead scientist for Vertex’s development of ivacaftor, Fred Van Goor, is where they root these claims, writing

This model demonstrated that in response to treatment with ivacaftor,
patients achieved a restoration of CFTR activity in the range of
35%–40% of that seen in control patients without CF.

So how do they get those larger numbers? What is this “model?” Well the high end is pretty easy to figure out: they base that on a much larger dose than the one actually given to patients (250mg, rather than 150mg). Recall the chart above: the trial subjects on the 250mg dose had NPD tests indicating 43% of normal CFTR activity. But the 35% is harder to figure out—their NPD data doesn’t justify the claim. However, when they include sweat chloride, they are able to make a different, problematic seeming chart.

This is a complicated chart, because it’s a scatterplot that isn’t labeled completely: you have to dig through the study’s references to see what they’re pointing out. But basically, what the study authors have tried to do is coordinate NPD data with sweat chloride data, because sweat chloride is seen as a measure of CFTR activity as well. However, what doesn’t become clear here, in part because the curve is obscured by the more linear readout of NPD data points, is that sweat chloride data has to be measured on a curve that swings up heavily as CFTR activity approaches zero. In other words, a relatively modest amount of CFTR activity can drop the sweat chloride measurement by quite a lot. And past that point, the numbers in the literature start to jump all over the place until you hit carrier and non-CF controls. So if you’re working between the two of these data points to come up with a more comprehensive way of coordinating CFTR activity and clinical manifestations of disease, you’ll necessarily make modest amounts of CFTR activity look more substantial as you try to close the gap between the more linear NPD data and the curved sweat chloride data. This is manipulating data to make things look better than they are, even if the place that this gets patients to is terrific and transforms their outlook. And because of the obvious clinical benefit of a drug like ivacaftor, it artificially deflates the level of CFTR activity that you want to see as a goal in new drug development.

This discussion gets us into a tricky and hard-to-navigate question: how much CFTR activity is enough to offer a real benefit to patients? And there are a couple of different ways into that question. I’ll answer it in part with some very new data (like yesterday: June 24) in a bit. But, as I’ve done above, you can also start from controls that approximate the place you’re trying to get. And that still leaves an important piece unsolved: how much of what you rescue in vitro can a given drug design manage to rescue in vivo? There are a lot of factors in this: how long does the molecule you’re developing stay active, how is it administered, how is it metabolized in the body? If Vertex’s ivacaftor is a good control for this, does that mean we can expect 60% in vitro to mean 36% in vivo for other drugs? 75% in vitro to mean 45% in vivo? Or can we expect other manufacturers to attain results closer to what they see in the lab? I think it’s too early to answer that question, though the glimpses we’ve seen from other companies developing molecules that perform a similar function to that of ivacaftor are interesting on this subject.

Ivacaftor is broken down relatively quickly from the body in the presence of the right enzymes (this is looking at a class of enzymes called “CYP”s, which are a good yardstick for measuring how long a drug will stay active in the body). This means, for it to work well, you need higher doses and you need them more often than is likely to be the case with more advanced drug candidates. With the novel potentiator compared against ivacaftor above, you see a great deal of sustained activity after introducing CYPs. There’s every reason to think that the major hurdle to achieving more parity between in vitro and in vivo CFTR activity is the length of time the molecule stays active, and not something more fundamental differentiating the two measurements. Then again, ivacaftor is not water soluble, meaning that its administration relies on other mechanisms to dissolve it and distribute it through your body. So it’s probably, A) not getting everywhere you want it to go on a regular basis, and B) more of a pain to take (and so not as well-administered) than future competitors will be.

All of this leads me to think that you can still use in vitro CFTR activity as a rough predictor for how in vivo CFTR activity is going to look, just not necessarily with ivacaftor. Its interactions with other chemicals in the body and the fact of its insolubility in water just make it a bad drug in living persons compared to its potential in vitro. So we can keep responses to molecules in vitro as a tool to help anticipate clinical efficacy. But what about the other half of the question? How much do you need to correct to meet that benchmark?

How much CFTR is enough? VX-809/VX-770 combos as a test case

On the heels of their success in commercializing ivacaftor (VX-770, you’ll recall), Vertex selected a candidate molecule that they found actually saved some ∆F508 CFTR protein from degradation by the ERAD which we discussed above. They called this molecule a “corrector” and published some in vitro results for the drug in 2011. We’ll look at those below. But another study was done by independent researchers in 2013 to assess the molecule’s method of action. What they did to figure this out was quite clever—they basically captured fragments of CFTR protein of various lengths to see what VX-809 would bind to, and then established the smallest possible fragment (i.e., the minimum length of the protein you have to produce along its chain) that responded to VX-809. And when they did this, they found two things: first, that VX-809 binds to the MSD1, and second, that it seems to do this at a junction of this part of the protein responsible for holding that domain together as it links into CL1.

That’s an interesting finding, because it means that we’re still not talking about a molecule that binds to the region of NBD1 that’s actually unstable because of the deletion of phenylalanine at position 508 (remember that zoomed in view of the squiggly bits connecting to one another above?). Instead, VX-809 seems to work by making a new link possible between another domain and the unstable NBD1. And that’s promising, because it means that there are potentially a lot of options for fixing the assembly of the protein and getting it through ERAD (where it gets tagged for degradation), through the proteasome (where various bits and pieces glom onto it and rip it apart when it isn’t built well), and to the cell surface (where it still needs help to open its channels and let ions out).

So VX-809 rescues some CFTR protein. How much? Well, the increase in functional CFTR protein is not that great. But when you add ivacaftor to it and “potentiate” the stabilized MSD1, you get a decent amount of activity. Here’s what Vertex’s charts say, as published in their 2011 discussion of VX-809 and VX-770 working in concert as a combo therapy:

Alone, they find that VX-809 rescues between 10 and 20% of CFTR activity, probably around 14% (I know—the chart doesn’t look like that; but the discussion of the data in the article says, “an increase in chloride transport from 3.4 ± 0.7% to 13.9 ± 2.3% of that measured in HBE isolated from four non-CF donor lungs,” and an earlier chart makes that more obvious).

However, when we look at another study from 2011 that measured nasal potential difference (NPD, you’ll recall) in patients who received a monotherapy of VX-809, you find no statistically significant improvement at all in activity. Here’s what that looked like:

You also saw some pretty nasty side-effects for patients who enrolled in the study: drops in FEV1 (a rough measure of lung function), chest tightness, and so on. However, as the chart from the Vertex study in 2011 shows, when you add ivacaftor to VX-809, you get a much greater increase in activity. The discussion of their results actually indicates an increase to about 25% of non-CF-HBE (even though the chart suggests a much higher number, and the table that they include with the study suggests a different number yet, also higher). Let’s just trust the discussion part, since it’s actually the lowest claim that Vertex’s lead scientist on this project, Van Goor, is making. So VX-809 alone restores 14% or so of CFTR activity in vitro and when you add a small amount of VX-770, ivacaftor, you get to ~ 25% CFTR activity.

Well, here is where we run into a big problem. We’ve already noted that ivacaftor doesn’t have a perfectly efficient translation from in vitro to in vivo measures of CFTR activity. And when we look at VX-809's NPD data, there appears to be almost no efficiency in affecting in vivo CFTR activity. So what does that ~25% CFTR activity in vitro mean?

Getting the bare minimum

Now, there have been no NPD tests for subjects on a combination study that I can find. So there are no tests telling us how much in vivo CFTR activity is restored using this set of tools. But we do now have some fairly good data on clinical outcomes. A phase 3 trial just completed that looked at how people who are homozygous for the ∆F508 mutation respond to VX-809/VX-770 in combination. It was huge news and I’m sure that if you’re reading this you know about it. But what it means for CF research is still pretty unsettled. I started writing this essay in part to try to estimate what its results might mean, from the perspective of a layperson who is just concerned about his child’s future. Here’s that data, such as it is:

You’ll note that the gains in FEV1 are relatively modest, but over this large of a group, they are certainly statistically significant. Basically, what they suggest is that the drug is doing something for people—while it might affect a number of people negatively (remember VX-809 alone is not particularly kind to lung function, and in the phase IIa from 2011, researchers saw some serious declines in lung function), it helps others quite a lot. This averages out to the good side of the mean baseline, and that’s a good thing. But this alone might not be persuasive to a regulatory body like the FDA. The stronger evidence for making a drug with this modest an increase in lung function available to patients comes from looking at the larger picture—more data points, which is particularly important given that it targets a protein to restore electrical activity, and not the airways specifically. It’s actually kind of perverse not to include NPD improvement as an endpoint when that’s the only well-accepted way to test the thing you’re actually changing in patients’ bodies. But here are the other endpoints that Vertex did gather in this phase 3:

I think these make a strong case for the combo. Increased BMI, statistically significant increases in the quality of life questionnaire, large percentages of patients who saw greater than 5% relative increases in their lung function, reduced exacerbations and lower frequency of IV antibiotics (that last point included elsewhere). But—and this is an important point—I think this is a barely approvable drug. It should be approvable; it has demonstrable clinical impacts on average, and if you look at some of the cancer drugs that get approved by the FDA, this does at least as well as those closer regulatory calls. But it is just over the edge of offering clinical benefit that’s demonstrable enough over a wide population to justify FDA approval.

And that leads me to my concluding questions: first, is this what 25% in vitro CFTR activity gets you; second, is there a better measure than electrical activity for estimating the clinical impact of a drug in vivo? To answer the former, I don’t think we know, and that’s galling. The 25% in vitro activity demonstrated with VX-809 in combination with VX-770 should absolutely be enough to show real clinical impacts. But it simply cannot be translating into anything like 25% in vivo. Firstly, VX-809 shows no significant increase in nasal potential difference on its own, and VX-770, ivacaftor, shows only about 60% efficiency in its translation to a living patient. The best case scenario for the combination would be duplicating that 60% efficiency, which would mean roughly 15% in vivo CFTR activity. But that’s optimistic, given the very poor effect on in vivo CFTR activity that we see with VX-809 monotherapy (on its own). I think a more realistic estimate of in vivo CFTR activity (until we see this data from an independent researcher or from Vertex) with the combination of VX-809 and VX-770 is ~10% +/- 2%. In other words, we can work backwards to estimate that the combination improves CFTR activity to around the levels we see in CF patients who have pancreatic sufficiency, but not higher. Certainly not outside of the range for progressive lung disease, and certainly not close to levels like we see in patients with CBAVD as the only symptom and a CF mutation or two.

Stanford C-Sweat tests and other measures of CFTR activity

So what other tools do we have to determining the amount of CFTR activity a given molecule or combination of molecules might be generating in patients? Ideally, you’d want a test that was less invasive than nasal potential difference, and which had the same sort of linear readout of function built in to it—i.e., something you don’t have to have an enormous curve in order to read. Some Stanford researchers managed to come up with a test that promises to be something like this. With some caveats, though: they know it doesn’t work well at the extreme low end, which is where all of the movement happens between CFTR activity and clinical outcomes.

Here’s how the researchers describe their hypothesis for this test:

Sweat glands also secrete fluid via a CFTR-dependent mechanism (hereafter termed ‘C-sweat’) when cholinergic pathways are blocked and β-adrenergic pathways are stimulated to elevate [cAMP]. C-sweating is completely absent in CF subjects, and remarkably, when normalized to methacholine (MCh)-stimulated sweating (hereafter termed ‘M-sweat’), it is half-normal, on average, in CF heterozygotes. This was the first clear demonstration of a gene-dosing effect in cystic fibrosis. It indicates the direct dependence of C-sweating on the level of functional CFTR in the sweat glands, and thus provides a near-linear readout of CFTR function.

Now, that “near-linear” part is surprisingly important. The levels of CFTR function (i.e., not activity—this is a significantly register we’re in, where activity doesn’t necessarily correlate with function) that the test was able to measure cannot find any CFTR function for either pancreatic sufficient or pancreatic insufficient CF patients, except for a single outlier. And when I say “find” I mean visually—the test involves visually assessing the number of chemical granules present in a slide. In any event, here are the data that they found in generating their controls:

Now, this is really intriguing stuff. And part of the reason it draws you in if you’re interested in CFTR is that it potentially demonstrates just how little function you need to produce meaningful clinical changes. If you can have patients with CF-related conditions, but without classic CF lung disease, who are only seeing <2% CFTR function according to this measurement, that’s very interesting. And according to this method of measurement, patients on Kalydeco had an average of ~4% CFTR function (see “A little CFTR goes a long way” in PLOS ONE, 2014). Which both puts them into near-disease-free territory, as well as indicating just how much farther this class of medicine could and should go.

But, like I said, these numbers may be way off. As the authors of this study detail, they were troubled by the lack of any visible CFTR activity in patients who had pancreatic sufficiency. They write that “it is not credible that pancreatic sufficient CF subjects have less than 0.01% CFTR function,” and so they are clear about the fact that at the low end of functionality—where you’re likely to see the meaningful changes for patients as their clinical outcomes climb out of those of classic CF—their scale doesn’t work. We may need to change the numbers at the low end significantly when better data are available from this test.

But like I said before—it’s suggestive. It indicates, for one thing, that electrical activity might not correlate exactly with function, and that function might have an incredibly steep curve in its correlation with clinical outcomes. Below, I’ve compiled what it looks like when you put some of these numbers next to one another.

Pancreatic insufficient CF: <5% npd electrical activity, 0% c-sweat;

Pancreatic sufficient CF: 6-15% npd electrical activity, 0% c-sweat;

G551D treated with Kalydeco: 30% electrical activity, 4% c-sweat;

CF-related conditions: 25-80% electrical activity, 1.5-80% c-sweat;

CBAVD: 35-50% electrical activity, 15% c-sweat;

Heterozygous carriers: 75-94% electrical activity, 23-71% c-sweat;

Non-CF: 100% electrical activity, ~100% c-sweat.