Hello, my name is Joe Arron. I’m a Senior Scientist in the Immunology Biomarker Discovery group at Genentech. And I’m going to talk to you about how we’ve exploited asthma heterogeneity to do a better job of developing drugs for asthma. This is a companion talk to Prescott Woodruff’s talk on asthma biology and heterogeneity, and I suggest you take a look at that one first so that my slides are a little bit less confusing. So, before I get started, I want to give a couple of disclaimers. First of all, while asthma heterogeneity is widely acknowledged, there is not yet a clearly defined drug development or regulatory path toward personalized healthcare for asthma patients. And there are no… currently, no therapeutic products approved for the treatment of asthma that specify the use of a companion diagnostic test on their label. Now, what I’m about to show you is our attempt to overcome these two disclaimers, but I do want to make it clear that none of this reflects anything that’s approved by any regulatory authorities yet. So, in drug development, it’s really more often a story of failures punctuated by occasional successes, rather than the other way around. And while there are many, many different reasons why a drug candidate might fail, I want to highlight four of them today. So, first of all, perhaps you’ve chosen the wrong target for your drug. You have a hypothesis about a particular target or pathway in a particular disease, but it turns out that that target is not a critical node in disease pathogenesis. A corollary to poor target selection is that perhaps going after that target is… it leads to unacceptable safety issues, where it’s unsafe to expose a patient to that drug. The second reason why a drug candidate might fail is that you’ve got the wrong molecule. So, what I mean by molecule is the drug candidate. So, this is something that you think is going to interact with your molecular target of choice. You may have insufficient affinity or avidity for the target, there may be off-target effects, where it binds to other molecules that you don’t want to inhibit or activate. You may have poor pharmacokinetics, meaning that the drug doesn’t distribute adequately to the tissue that you’re interested in. Or, perhaps, you’ve dosed it inadequately. A third reason why a drug candidate might fail is wrong outcomes. So, perhaps you’ve got a good target that’s relevant to your disease and you’ve got a good molecule that interacts with that target, but, in your clinical trials, if you’re not measuring a clinical outcome that is actually relevant to how that target manifests in the disease, or isn’t relevant in the particular population of patients that you’ve chosen to treat in this clinical trial, you may get it… end up with a negative result in your clinical trial. And the fourth reason that I really want to spend the most time on today is wrong patients. So, because asthma, as Prescott has already described, is a heterogeneous disease, if you’re treating everybody with… with a diagnosis of asthma and not accounting for this intrinsic heterogeneity of the disease, it may be the case that the patients who respond to your therapy are mixed in with the patients who don’t respond to the therapy. And if you can’t… have no way of distinguishing them, your aggregate effect that you see in the clinical trial will be blunted, and you may not actually see any effect at all. So, asthma is… has historically been treated empirically. So, I want to define asthma here, for our purposes, as a syndrome, rather than a specific disease entity, of loosely affiliated pathological processes. And this syndrome has a common clinical manifestation of reversible airway obstruction and hyper-reactivity. Now, the standard of care guidelines, which are issued by, in this case, the National Heart, Lung, and Blood Institute, suggest that when a patient presents with asthma, all the way here on the left side of the scale we have mild asthma, and we’re going to treat these patients’ symptoms by giving them a short-acting beta agonist, or SABA. And if that controls that patient’s symptoms, that’s all they’re going to get. However, if this fails to control their symptoms, you start to step up to a higher step on this diagram, where you start giving them an anti-inflammatory medication, such as low-dose inhaled corticosteroids. If low-dose inhaled steroids don’t work, what do you do? You step up. And you give them more inhaled steroids. If more inhaled steroids don’t work, what do you do? You give them more inhaled steroids. And if that isn’t leading to asthma control, you start putting on very high doses of inhaled steroids, and perhaps other medications as well. And ultimately, out here at the most severe end of the spectrum, we have patients that really require systemic corticosteroids. Now, systemic corticosteroids may be adequate to control these patients’ symptoms, but we really don’t want to keep patients chronically on systemic corticosteroids for many years, because they have a lot of adverse systemic effects. And so, in drug development, we’re really focused on these patients out here, which constitute what we would call the high-need, or severe, population of asthma patients. And while this is a relatively small fraction of all the patients with asthma — it’s maybe only about 10% of the patients — it actually constitutes about 2 million patients in the United States and Europe, so it really is a significant unmet medical need. So, historically, there have been a lot of different approaches to defining asthma heterogeneity. And in some ways what we’re doing with molecular asthma phenotyping stands on the shoulders of previous works that have used clinical or biological or other measures of asthma pathophysiology that appear to be heterogeneous within the population. So, in general, we see that there tend to be less inflammatory or more inflammatory subsets of asthma patients, which can be defined by things like the age of onset; the presence of atopy, or allergic responses to particular triggers; the triggers themselves; the clinical manifestations — so, some patients may have a lot of asthma symptoms and airway hyper-responsiveness, whereas others may be less symptomatic, but will occasionally have these really severe asthma attacks or exacerbations, which may require hospitalization; comorbidities, so, other medical conditions that are associated with asthma can associate with these subsets of asthma; and, recently, a lot of attention has been focused on the cellularity in the airway. So, if we’re looking at… at infiltrating cells such as eosinophils and neutrophils, in the airway, we see that a lot of asthma patients really don’t differ from healthy controls in terms of the cells that you see in the airway, whereas a subset of asthma have significant elevations of inflammatory cells, particularly eosinophils. Now, the problem with using this type of modality in clinical drug development is that it’s very hard to define these features objectively. And there’s a lot of overlap between these. There’s not really a hard line in between these two categories. And so using these features to select patients in a clinical trial could be very challenging. We really would like to have something a little bit more objective that’s related to our therapeutic target of interest. So, a simplified paradigm for looking at asthma heterogeneity might be to define it in terms of inflammation. So, as Prescott described to you earlier, in mild asthma it seems that everybody has some component of non-inflammatory bronchoconstriction, or airway smooth muscle hyper-responsiveness. A subset of these patients has elevated expression of type 2… T helper type 2, or Th2, cytokines. And these patients are characterized by increased levels of eosinophils in the airway. Now, when we go to severe asthma — this is the really unmet medical need that we’re trying to develop new therapies for — again, we see a subset that really doesn’t have a lot of inflammation but has a lot of bronchoconstriction. And then we see, again, a subset that has eosinophilic inflammation, a subset that maybe has more neutrophilic inflammation, and perhaps a subset that is an overlap of these. And this… the sizes of these circles in this Venn diagram are not really to scale or… or particularly well-defined. There’s a lot of debate about what constitutes a reasonable cutoff for these… these metrics. So, the key questions that we really set out to address are, number one, what are the molecular underpinnings of this heterogeneity? Number two, how can we use this information to develop novel targeted therapies? And ultimately, what we’re really trying to do is link therapeutic targets to the pathophysiological manifestations of the disease to outcomes that we can measure in a clinical trial. So, I don’t intend to go through this slide in great detail, other than to point out that airway inflammation is really complicated. And so there’s a lot of different mediators that contribute to T helper type 2 airway inflammation. The airway is a place where, basically, the body interacts with the environment. So, we… we inhale all sorts of different things on a constant basis, such as allergens, viruses, pollutants, other types of bugs. These interact with the stromal cells in the airway, leading to the production of epithelial “alarmins”, such as IL-25 and IL-33, or TSLP. These can contribute to either an adaptive or innate type of immunity. And this can ultimately lead to what we see pathologically in asthma, where we see increased mucus production, eosinophilic inflammation, subepithelial fibrosis, smooth muscle hyperplasia and hyper-reactivity, and angiogenesis. Now, there are a lot of different mediators in here and it’s kind of hard to decide, based on the way I’ve drawn this figure, which one of these would you target therapeutically. Now, I’m going to cut to the chase and highlight a particular mediator in here, interleukin-13 or IL-13, just to point out that IL-13 is actually involved in a lot of these different processes, at least in terms of preclinical models. And we know that IL-13 can actually be produced by a number of different cell types, including T cells, NKT cells, macrophages, eosinophils, basophils, mast cells, innate helper type 2 cells, and type 2 myeloid cells. And IL-13 can actually act on a number of different targets, leading to a lot of these different manifestations. So, one of the problems that we faced when trying to develop an inhibitor of IL-13 is data like this, where, in this study, investigators have actually measured IL-13 levels directly in induced sputum from asthma patients. And so, what you can appreciate here is that in normal healthy controls there’s not a lot of detectable IL-13. All these patients are just undetectable levels of IL-13. In mild asthmatics, about half the patients are undetectable and half the patients are detectable. When we go to moderate asthma patients — these are patients who have been put on inhaled steroids and are relatively well-controlled by their steroids, and so think back to Prescott’s talk where he showed that most of the moderate asthma patients that respond to steroids have a down-regulation of IL-13-induced gene expression — we don’t see a lot of patients with detectable IL-13. But now, when we go into the severe patients — these patients are defined by being refractory to inhaled steroids, so, even though these patients are on really high doses of inhaled steroids, they’re still quite symptomatic — again, we see a subset of about 30-40% of the patients who have detectable IL-13. So, if we wanted to design a clinical trial where we’re going to inhibit IL-13 specifically in asthma patients, we might guess that only these patients that actually have the target expressed in their airway have the potential to respond to an IL-13 inhibitor, whereas these patients that have undetectable levels of IL-13 are likely not to respond to the therapy. The challenges here are that IL-13… measuring IL-13 in the airways is actually really difficult. So, it requires either the production of an induced sputum sample or an invasive procedure such as bronchoscopy, which can really only be done in specialized pulmonary clinics and is difficult to implement on a broad scale. IL-13 is produced and consumed at the site of inflammation, so just a little bit of IL-13 is released from a cell, it immediately binds to a responder cell, which then responds. And so it’s difficult to detect the IL-13 itself. And as I mentioned, airway sampling is really technically quite complex and not logistically feasible in large, multicenter clinical studies of the type that we would need to do to get a drug approved. So, the key question here is, how can we identify these patients with elevated levels of airway IL-13 non-invasively? So, can we come up with a biomarker that doesn’t require direct sampling of the airway to reflect IL-13 activity? So, I’m gonna move into this… the part of my talk, now, where we’re going to talk about our strategies for discovering biomarkers for therapies that target type 2 or eosinophilic airway inflammation in asthma patients. So, first of all, I just want to give a couple of definitions. So, what do we mean by a biomarker? So, for the purposes of this talk, a biomarker… I… I think this definition from the NIH is very good. A biomarker is a characteristic that is objectively measured and evaluated as an indicator of normal biological processes, pathogenic processes, or pharmacological responses to a therapeutic intervention. And so one thing I want to point out here is it’s a characteristic. It doesn’t have to be a blood test. It doesn’t have to be a genetic polymorphism. It doesn’t have to be a gene expression level. It has to be a characteristic. So, it could be something like… as simple as gender or age. So, there are really four different types of biomarkers that we commonly use in… in drug development. And I’m gonna spend most of my time talking about a predictive biomarker, but I am going to touch on these other types as well. So, a predictive biomarker is a biomarker that would identify patients most likely to experience clinical benefit from a particular therapeutic intervention. And this would be a test that you do before you initiate treatment. So, you do the test and you decide, is this patient likely to respond or not to a particular intervention? A prognostic biomarker is something that would characterize the probability of future disease activity, and you may want to use this to stratify for clinical outcomes, for example, asthma exacerbations, and I’ll give you an example of that later in the talk. A pharmacodynamic biomarker is something that you measure before and after treatment, and what you’re looking for is evidence that you’ve actually engaged your therapeutic target with your drug. So, a pharmacodynamic marker will change in response to a therapeutic intervention, and we often use this for selecting the appropriate dose in patients. And finally, a surrogate biomarker might be something that is like a pharmacodynamic bio… is sort of like a combination of a pharmacodynamic and a [prognostic] biomarker, so, something that changes a little bit over the short-term, maybe an indi… indicator of a future clinical response. So, as Prescott very nicely described in the accompanying talk here, in a collaboration that we did with him, we were able to define two types of asthma patients by looking at gene expression signatures in bronchial epithelial cells collected from a bronchoscopy. So, if we take these three genes here, periostin, CLCA1, and serpinB2, these are all genes that are expressed in bronchial epithelium and could be induced, potentially, by interleukin-13. And so when we do an unsupervised clustering of 42 asthmatics and 28 healthy controls, just on the basis of expression of these three genes, we found something really interesting, that there were two really distinct clusters of patients. Over here, we’ve got a cluster of patients that has very high expression of all three genes, and has coordinate expression of all three genes. On the right side, we have patients that have lower expression of these three genes. Interestingly, the patients in this high cluster, here, comprised half of the asthma patients in the study, whereas, on the right side, these patients with low expression comprised the other half of the asthma patients, interspersed with healthy controls. And so on the basis of the expression of these three genes, these asthmatics over here are really indistinguishable from healthy controls, despite having very similar clinical manifestations of disease. We went on to show that in contemp… contemporaneously collected endobronchial biopsies the expression of type 2 cytokines is… is elevated in patients who had high expression of this gene signature. So, we see elevated levels of IL-13 and IL-5 in the asthmatics that were in cluster 1, which is the group that had high expression of those three genes. Furthermore, these patients with this high Th2 signature are the patients that have elevated levels of eosinophils in their airway. And so we think this is actually a really important finding, because this paradigm of eosinophilic or noneosinophilic asthma has been around for a fairly long time, but there wasn’t really a good molecular basis for this paradigm until we did this work. So, these Type 2 High and Type 2 Low asthmatics that we’ve defined by an epithelial gene expression signature signature are actually distinguished by the presence or absence of eosinophilic airway inflammation, the degree of subepithelial fibrosis, the composition of their mucus, and importantly, in this mild to moderate group of asthma patients, their response to inhaled steroid therapy. So, the Type 2 High patients responded very well to inhaled steroids in terms of their lung function improving, whereas the Type 2 Low patients really didn’t show much of a clinical benefit from inhaled steroid therapy. So, that’s all well and good, and I think this is a really important proof of concept that the response that’s typically seen to inhaled steroids in an unselected group of asthma patients may be actually driven by this subset of patients that has increased type 2 airway inflammation. However, the question that we really wanted to address is, well, what about these high-needs asthma patients that have really severe disease and who are not responsive to inhaled steroid therapy? So, as I told you a couple of slides ago, there’s a subset of patients that, even though they’re on a lot of anti-inflammatory medication, still has significant levels of airway eosinophilia and IL-13 expression. So, to cut a very long story very short, we needed to come up with a non-invasive way of identifying these patients that had this type of inflammation, because we can’t go and do bronchoscopy on thousands of patients in clinical trials. So, the desired characteristics of a biomarker of this pathway would be something that reflects airway inflammation; is mechanistically linked to our therapeutic target, in this case, IL-13; something that we can measure non-invasively, meaning we can measure it in a blood sample or measure it in exhaled air, but we don’t have to do an invasive procedure like a bronchoscopy; and things that can be measured using robust, reproducible assays. So, in this really simplified paradigm of airway inflammation, I’m showing you IL-13 and IL-5, which, as I showed you before, were both elevated in terms of their expression in the patients that had the high Th2 gene signature. Now, IL-13 in IL-5 can conspire to lead to eosinophilic airway inflammation. So, IL-5’s major role is to act as an obligate hematopoietic factor for eosinophils. So, IL-5 acts on hematopoietic progenitors in the bone marrow, leading to the production of more eosinophils. However, these eosinophils don’t really know what to do with themselves, because they’re not given a guidance cue. IL-13 acts directly on the structural cells in the airway to induce the expression of chemokines that bind to CCR3, which is a receptor that’s expressed on eosinophils, and so it recruits these eosinophils into the airway. IL-13 can also act on the cells in the airway to induce the expression of iNOS, which is a gene that encodes an enzyme that can convert arginine into citrulline and nitric oxide. Nitric oxide can actually be detected in exhaled breath as fractional exhaled nitric oxide, or FeNO. As Prescott described, IL-13 can also act on bronchial epithelial cells to induce the expression of periostin, which is secreted from the basolateral aspect of these epithelial cells, unlike another gene in the Th2 signature, CLCA1, which is expressed on the apical aspect of these bronchial epithelial cells. So, if you wanted to try to measure CLCA1 protein, you’d really have to sample the airway by either taking bronchoalveolar lavage or sputum, whereas in the case of periostin it’s being secreted down into the bronchial tissue, so there’s a possibility that it may actually be taken up by blood vessels and be detectable in the bloodstream. And so the three biomarkers that we really wanted to focus on in clinical development were eosinophils in the peripheral blood; exhaled nitric oxide, which can be easily measured by a device; and periostin, for which we needed to develop an assay to measure in the peripheral blood. So, we conducted another observational study in… this time, in severe asthmatics who were on very high doses of inhaled steroids yet still had highly symptomatic disease and compromised… compromised airway obstruction. And we simply compared patients that had elevated levels of eosinophils, either in the sputum or in their bronchial tissue, and we looked at these peripheral blood levels of serum periostin. So, this is eosinophils in the lung versus periostin in the blood. And what we see is that patients that have elevated eosinophils, either in the sputum or in the tissue, have significantly elevated levels of serum periostin. So, we believe that this corresponds to the Th2 High phenotype that was described in terms of the gene signature, and that, we believe, corresponds to patients that have elevated expression of IL-13 in their airways. And so we assessed periostin versus exhaled nitric oxide and blood eosinophils — and we threw in IgE for good measure — to see which was the best single predictor of airway eosinophilia. And this is a composite of sputum and tissue eosinophils. And so, what we find, shown in this receiver operating characteristic curve is that periostin was the single best predictor of airway eosinophils, as you can see in the red line. FeNO was a little bit… was also pretty good, but not quite as good as periostin. And peripheral blood eosinophils were a little bit behind that. So, now, I want to talk about what happens when we actually take these observations that we made from these cross-sectional observational studies into randomized placebo-controlled clinical trials of candidate therapeutics in asthma patients. And so this is one of the great privileges that I have, working at a pharmaceutical company, is that we actually get to do these very large-scale experiments in humans and see if our hypotheses that we’ve developed from all these observations actually play out in terms of clinical benefit for an unmet medical need. So, I’m gonna talk to you about two different clinical trials of two different therapeutics that target two different aspects of type 2 airway inflammation. So, I’m going to talk about lebrikizumab, which is a monoclonal antibody that inhibits IL-13, and I’m going to talk about all omalizumab, which is monoclonal antibody that inhibits IgE. And so in this very simplified schematic, you can see that IgE and IL-13 are both upstream and downstream of each other. Now, this obviously takes out a lot of detail about what’s going on, but in this… in this schema we have IgE… is combined to FC-epsilon receptors on mast cells. In the presence of allergens, it gets cross-linked. This mast cell can then degranulate and become activated and produce interleukin-13. IL-13 can in turn act on a B cell to induce isotype switching, so that this B cell produces IgE. And the cycle goes again. So, in this sort of crude sense, we wanted to see if these biomarkers that reflect IL-13 activity could predict clinical benefit from IL-13 or from IgE inhibitors. So, the two studies I’m going to talk to you about, the first one is called the MILLY study. This was a Phase 2 clinical study of lebrikizumab, which is an antibody against IL-13. There are 218 patients in this study, randomized 1:1, drug:placebo. These patients had moderate to severe asthma, so they were symptomatic despite being on inhaled steroids. They were treated for 24 weeks, and the primary outcome was the change in forced expiratory volume in one second, or FEV1, at 12 weeks. And so that’s a measure of airway obstruction, how much air can the patient blow out in one second. The other study I’m going to describe to you is called EXTRA. This was a Phase 3 three study of omalizumab in a much larger population of patients. Again, these had moderate to severe allergic asthma. They were symptomatic despite being on high-dose inhaled steroids. They were treated for almost a year, for 48 weeks, and the primary outcome here was the rate of severe asthma exacerbations. So, if we look just at these three biomarkers that I described to you, serum periostin, peripheral blood eosinophils, and FeNO, or exhaled nitric oxide, in either of these studies we see essentially the same thing. We noticed two things. One is that all three of these biomarkers are continuously distributed. So, here is the distribution for serum periostin, for blood eosinophils, and for FeNO. So, there’s no obvious bimodal distribution where we could say, ahh, this is where we need to put a cutoff in this study. The other thing that you’ll notice is that these three biomarkers, albeit weakly, are significantly intercorrelated with each other. So, a patient who has high serum periostin is gonna tend to have high levels of FeNO, and tend to have higher levels of blood eosinophils. So, we think they’re reflecting a similar process, despite the fact that these three measurements are really coming from three different sample types. So, the blood eosinophils is coming from whole blood, where we’ve counted… we’ve done a differential blood count, the serum periostin is based on the measurement of the protein in the soluble fraction of blood, and FeNO is actually measuring a metabolite in exhaled breath. And so, what we see, now, is if we… if… if we take all comers in this trial of anti-IL-13, or lebrikizumab, and we subdivide these patients based on their levels of periostin in the serum at baseline, what we see, here, is… this is plotting out the FEV1, or the… the difference in lung function over the course of the study. And in red we have the active arm, these are patients treated with lebrikizumab, in black we have patients treated with placebo. We see that patients that had serum periostin levels above the median level at baseline have a significant — both clinically and statistically significant — improvement in lung function over the course of the study, as compared to patients on placebo. Whereas patients that had low levels of serum periostin at baseline, really there’s no difference between the active and the placebo arms. We see very similar things if we subdivide patients according to the median levels of either FeNO or blood eosinophils at baseline. So, because, as I said, these biomarkers are continuously distributed across the population, we didn’t a priori have any way of deciding what would be an optimal cutoff, so we just used the median cutoff. And that actually turned out to be a pretty good cutoff. As you can see, patients below the median, so that’s half the patients in the study, really did not show any clinical benefit from lebrikizumab in terms of lung function as measured by FEV1. Whereas all of the clinical benefit observed was really in the patients… in the population of patients above the median. Now, if we go to the omalizumab study, this is looking at a different outcome measure, and what we’re plotting here is the rate of exacerbations per patient per year for each of four groups for either FeNO, periostin, or blood eosinophils. So, on the… so, you can look at any one of these three plots and it’s essentially the same story. What you see is that patients that were below the median level of the biomarker at baseline, in the placebo or omalizumab treatment arms, really have the same rate of exacerbations. Whereas patients above the median for, in this case, eosinophils, patients that received placebo have a much higher rate of asthma exacerbations in this study, whereas patients on omalizumab, again, have a lower rate. So really, the outlier in all three of these groups is the patients in the placebo arm that had high levels of the biomarker. So, getting back to something that I mentioned before about different types of biomarkers, we can show schematically, here, that these three biomarkers are actually both prognostic and predictive for asthma exacerbations in this clinical study. So, what we’re showing here is this dotted line, here, are patients that are diagnostic-positive, or who have high levels of the biomarker, that received placebo have the highest rate of exacerbations over time. Whereas the diagnostic-positive patients who received the drug actually have the lowest rate of exacerbations. Patients that are diagnostic-negative don’t have as many exacerbations in the placebo arm, but, because they don’t have a lot of activity of the targeted pathway, they don’t show much clinical benefit from the drug. So, these are, I think, a nice example of how you could use biomarkers to stratify both for the likelihood of a particular clinical outcome and for the likelihood that a particular subset of patients is more likely to benefit from an intervention. So, the… an… another way that biomarkers can be used is to look at pharmacodynamic effects. And so, what we’re plotting here is the pharmacodynamic effects of lebrikizumab — now, this is our anti-IL-13 — on serum periostin and FeNO levels. And what I’m showing you is the population distributions for these biomarkers in the placebo and the active arm at different times of treatment. So, at 0 is their baseline levels before they received any drug, 12 weeks, 24, and 32 weeks of treatment. And so what you can see here is that, at baseline, you know, the two groups have very similar distributions of periostin. And when we treat them, the placebo group has essentially… a very similar distribution, whereas we see a significant decrease in the periostin levels in the treated group at all of these time points. Same thing for FeNO, here. And what we’ve plotted here, on the… on this graph is the median level of serum periostin or FeNO that’s seen in a healthy non-asthmatic control population, and then the 25th and 75th percentiles. So, what you can see here, really, is that not only was there a significant decrease in the levels of periostin and FeNO on treatment, but we’ve actually normalized the distributions of these two biomarkers. So, in these patients who had asthma and who had an elevated distribution of these biomarkers at baseline, by inhibiting IL-13 we’ve reduced the distribution of these biomarkers to the same distribution that you see in a non-asthmatic population. So, we can infer from this that in asthma patients the excess levels of periostin and FeNO are due to interleukin-13. So, in terms of, how do we use these biomarkers and how do we go forward to develop these tests in conjunction with a novel therapeutic? Turns out to be really complicated from a clinical and regulatory perspective. And so complicated that we don’t even know exactly how this is going to play out at this time. But let me go back to something that I’ve talked about before. So, the characteristics that we really are looking for in a biomarker is that they accurately reflect airway inflammation, they’re mechanistically linked to the target, you can measure them non-invasively, and there are robust, reproducible assays for these biomarkers. So, if we look at the types of assay platforms that are available for these biomarkers, induced sputum is something which can be done in specialty clinics. However, there’s not a lot of consensus on exactly how sputum should be collected and processed. It turns out to be a very drawn-out procedure. It actually takes a patient… it may take a patient as much as half an hour or an hour to produce an adequate sample. And it may take several hours for a laboratory technician to process that sample. If we talk about exhaled nitric oxide, there’s actually an FD… FDA approved device which can measure nitri… exhaled nitric oxide, so that actually makes things a lot simpler if that’s a bio… if that’s a biomarker you wanted to use. Peripheral blood eosinophils, again, you would think this is something that’s fairly easy — you can just send your blood sample off to the lab and ask for an eosinophil count. The problem is there’s approximately ten different platforms that are in widespread use across the world, and each of these platforms counts the eosinophils in a slightly different way, and it’s very difficult to standardize those things. For serum periostin, I didn’t go into the details here, but this is a proprietary assay that we’ve developed in conjunction with Roche Diagnostics, and currently it’s not commercially available. But Roche Diagnostics is working really hard to come up with a very reliable, reproducible assay, so that when you measure serum periostin at your local clinical lab you will always get a reliable result from that. So, one thing which we didn’t talk about is, what is the intrapatient variability of these biomarkers? So, if you measure one of these biomarkers at a given point in time, what will happen if you just don’t do anything to the patient and measure that biomarker at some other point in time? We get some hint of this when we look at two samples taken a week apart from each other in the MILLY lebrikizumab study, before we gave a patient a dose of either drug or placebo. And what I’m plotting here is the coefficient of variation, or the CV, for periostin, FeNO, and blood eosinophils. And so what you can see here is that the CVs for FeNO and blood eosinophils is around 20%, whereas for periostin it’s around 5%. And so there could be a lot of reasons why this is. There are well-known diurnal variations in eosinophil counts. FeNO is probably reflecting episodic bouts of airway inflammation, and so it’s definitely going to change… fluctuate a lot over time, depending on the environmental exposures that a given patient has. So far, we’re fairly encouraged by the relatively low variability we see in periostin because we think it may be accumulating in the blood and reflect the aggregate amount of airway inflammation that has happened over the previous, perhaps, days or weeks. But a lot more work really needs to be done to understand how these biomarkers are going to perform in real life. So, to summarize what I’ve told you about biomarkers in asthma clinical studies, let’s go back to what we talked about on the first slide, which is four reasons why a drug candidate might fail. So, we talked about wrong target, wrong molecule, wrong outcomes, and wrong patients. And let’s take a look at this with respect to, for example, lebrikizumab, which is anti-IL-13 from the data that we have in a Phase 2 study. So, in terms of target, well, it appears that IL-13 is a key node in asthma pathogenesis, at least for some patients, because when we blocked it these patients had improvements in objective measures of their lung function. In terms of the molecule, I didn’t talk about this in detail, but it appears that the pharmacokinetics and adverse event profiles support continued clinical investigation, which means that the drug appears to be fairly stable in the blood and it appears… systemically… and it appears to be fairly well tolerated. In terms of exacerbations, we see significant effects on lung function over time — that’s the data I showed you. I didn’t show you data that it also appears to reduce the rate of asthma exacerbations. And, in a bronchial allergen… allergen challenge study, we also see some positive results. The most important thing that I really want to emphasize here, though, is patient selection. So, our proof-of-concept study that I showed you the data from was stratified to account for the possibility that there is heterogeneity in asthma. And importantly, we actually tested our hypothesis that patients who had high levels of these biomarkers would show increased clinical benefit relative to patients with low levels of these biomarkers. Periostin appears to be a robust, non-invasive biomarker of eosinophilic airway inflammation. And the predictive and pharmacodynamic effects that we see with respect to periostin levels do support our mechanistic hypothesis. Now, of course, this is all very preliminary data, and we currently have pivotal Phase 3 studies in progress. And so obviously these findings have to be substantiated in pivotal studies before the drug could be approved. But so far we’re very encouraged by what we’ve seen. So, this raises a lot of questions for the future. We talked about… I think Prescott talked in particular about the different things that IL-13 can do to airway pathophysiology. So, it can contribute to eosinophilia, mucus production, fibrosis, airway hyper-responsiveness. But we didn’t actually measure any of those things in our clinical trials. We measured how much air can a patient blow out in one second. So, I think we really need to understand, on a more mechanistic level, what does IL-13 inhibition do to these specific aspects of pathophysiology in vivo, in humans with asthma? To what extent does IL-13 affect other pathways? Now, there are other therapeutics that target such… things such as IL-5, IgE, IL-4 receptor out there, and it’s unclear what the degree of overlap between the biology targeted by each of these agents is. We also don’t know what it does to neutrophilic inflammation, which seems to be a feature of severe steroid-refractory asthma. Are these phenotypes stable over time? I showed you the stability of these biomarkers over one week, but what… what is the case over six months or a year or ten years in a given asthma patient? And then, finally, by addressing… by partially addressing an unmet medical need in severe asthma, where we said, well, here’s a subset of severe asthma that we can improve their lung function by targeting IL-13, we’ve also identified a subset of asthma that does not benefit from that intervention. So, what is the nature of these patients that have decreased airway inflammation? And how can we target it therapeutically? So, these are obviously items of great interest to us going forward in terms of the investigations we’re doing. And then finally, why is it that the severe inflammatory patients, even though they’re on high doses of steroids, still have inflammation? And why does blocking particular mediators of inflammation seem to improve their asthma, whereas blocking inflammation with steroids does not? Finally, I’d just like to acknowledge the virtual army of people who contributed to all the data I showed you here. And in particular, I really just want to call out the… our collaborators in academic groups, particularly at UCSF, the University of Leicester, Queen’s University Belfast, and the investigators who participated in the BOBCAT observational studies of severe asthma. And thank you for your attention.