Again, thanks to the organizers for this invitation. I think this is my disclosure slide. There we go. All right, so we're going to start here with the case as well. This is a 56-year-old Caucasian man who presents to you with painless jaundice, itching, and a 10-pound weight loss. By history, he said he had upper respiratory infection about four weeks ago and got a 10-day course of Augmentin. He's not gone anywhere. He's not been exposed to anyone else who's sick. He's got a pretty simple medical history, a little bit of hypertension, treated with hydrochlorothiazide. He reports no known drug allergies. He doesn't really smoke or drink excessively. And on physical exams, pretty unremarkable, no abdominal tenderness, no masses, but he has obvious sclerolictoris and no visible rash. What's of concern is his liver enzymes. His AST is 220, his ALT is 320, and his LFAS is 650 with a bilirubin of 8 and an elevated INR at 1.3. Looking for clues here, carefully reviewing his CBC, the differential is normal, his creatinine is normal. And so he's coming to you saying, what's caused my liver injury? And you as the clinician have to figure out what to do. So I think all of us from our medical school teaching and experience would say, well, we need to evaluate someone for painless jaundice. If I can move this slide forward. So imaging, right? Painless jaundice, you get an ultrasound. There's no pancreatic obiliary disease. And of course, viral hepatitis we know is a very common cause of an acute liver injury. We check for A, B, and C, it's all negative. And starting to think of less common things like autoimmune hepatitis, you check autoantibodies which are negative, but his gamma globulins are a little bit high. And so he looks okay, so he comes back to see you a week later, and lo and behold, labs are not getting any better, his bilirubin's up to 15. And so with the diagnostic uncertainty, you decide to do an outpatient liver biopsy, largely I think to rule out things that you maybe didn't suspect. And I don't know if we're ruling in with a liver biopsy, but we see marked cholestasis and some mild ductular proliferation. So this then brings up the question, how do you make a diagnosis of idiosyncratic drug induced liver injury? And how confident are you in that diagnosis when you're seeing that patient in front of you? Is there anything that we could have done to predict dilly risk or susceptibility in an individual patient? And then in the bigger picture, do we have the tools right now in 2019 to order tests before giving a drug that would predict and prevent adverse drug reactions? And then what are you going to tell this patient about his prognosis, is he going to be okay or do you need to worry about him? So these are the questions that come up. So we've already heard about the definition of precision medicine, the use of an individual's global information to guide decisions regarding prevention, diagnosis, and treatment of disease. And of course, our goal is to maximize efficacy and obviously individualize the ability for a patient to comply with the treatment plan while maximizing safety and minimizing adverse events. So we all know about clinical parameters, but you know, this symposium is really based upon where are we at with the omics, as we heard. And we've heard a lot about pharmacogenomics, largely because that's been the largest set of data that we have with all the different forms of liver disease. But as you know, transcriptomics, proteomics, and so on, and we just heard a very nice talk on microbiome. So pharmacogenomics is really what I'm going to focus on predominantly here. But I would say in general, we've been looking for a while, for about 20 years, and we're still looking for the use of these tests. And I think we need more reliable and predictive biomarkers for individual diseases. The low-hanging fruit, if you will, has already been obtained. And then how much incremental value is a pharmacogenomic test, either for diagnosis, prognosis, or treatment decisions? Is it clinically actionable if you order a test? Is it going to change what you do or what you tell your patient? And then it gets complicated when you start to get the interpretation of multiple gene panels, and how do you interpret the test and explain it to your patient, and so on. So if you look at the bigger picture, if you look at drug labels, there's over 1,000 drugs approved in the U.S. About 10 percent of them, when you look at the package inserts, actually have some component of pharmacogenomic data in them from the drug development. However, there's only about 10 instances where pharmacogenetic testing is recommended prior to therapy. So don't worry, you haven't fallen behind. Now let's talk a little bit about idiosyncratic drug-induced liver injury. You know, when we talk about this, this is the one that you can't predict, right? It's due to an individual patient's unique features, which we don't know yet, hence the research. It's really independent of the dose, or the route, or the duration of which you give the medication. And it's pretty uncommon with most drugs on the market, anywhere 1 in 1,000 to 1 in 1,000, so it's a very difficult area to study. That being said, if you look at a global population like was done in Iceland over a two-year period, the overall incidence in the adult population is somewhere between 10 to 20 per 100,000 patients. If you do the math and extrapolate to the U.S. population, that would be about 60,000 cases of clinically apparent drug-induced liver injury, and we know it affects children as well as elderly, and probably more so in the elderly because they get more medications. There's also probably a slight predominance of women over men, but again, that may be just an exposure effect. In the U.S., from the Dillon study, we know that antibiotics are the leading cause of drug-induced liver injury in adults, and then followed by herbal and dietary supplements. But really, what we all struggle with with studying this entity is making the diagnosis because it's a clinical diagnosis of exclusion, and as a hepatologist, like this case, it's going to be less than 1%. It should be on the lower part of your differential diagnosis, and then it gets even more complicated when you prospectively study this because you see there's a variety of phenotypes from the same drug. That being said, we will continue to work very much on this area because it continues to lead to drugs failing during development and being pulled from the market. So it's a clinical diagnosis, and you look for the temporal association between taking the drug and the onset of liver injury, which in most cases is less than six months. Then you need to follow the patient after the drug is stopped, but there are some drugs that are a little bit unusual, like Augmentin, that can present with jaundice three to four weeks after the drug has been stopped. You look for the lab profile when the patient presents. Is it hepatocellular, cholestatic? But again, we're seeing individual drugs may have a mixture of different biochemical profiles, and it gets very complicated in the United States because the average American is taking five to six prescription medications. Which one is the culprit? So we fall back frequently to liver biopsy, but we also have to have excluded all those other causes simultaneously while DILI is rising towards the top of your differential diagnosis. But at the end of the day, you really want to hang your hat on an objective confirmatory test to say that this drug caused this liver injury in my patient, and we don't have it. So we go by causality assessment, which is just probability of synthesizing all the clinical data. So in the DILIN network, what we use is this expert opinion score, where if we know for sure it was acute hep A, that's an unlikely case. If it's possible, but we still think something besides the drug caused it, we call that a possible case. And then if it's more likely than not that the drug caused the liver injury, we call it a probable, very likely, or definite, based upon the amount of testing you did, characterization of the case, and what you know about the hepatotoxicity profile of the drug. And we've shown that this expert opinion process, as subjective as it is, but at least it's ordinal and categorical, is more reliable than some of the currently available instruments that are out there like the RUCAM. So it's clinical diagnosis, and the severity is very important as well to assess an individual patient, and it gets complicated, and INH is a perfect example. We know that this is an intrinsically hepatotoxic drug, and one to 10% of people will get a mild transaminitis, but interestingly, if you continue dosing in those individuals with low-level AST and ALT elevations, they will develop tolerance, and you can continue to treat through that. So is that patient different than the one who then goes on with continued dosing to get jaundiced? Probably. There may be different mechanisms with a different level of liver injury, and then the patient who presents to you in full-blown acute liver failure with encephalopathy and coagulopathy who needs a liver transplant, do they have a combination of risk factors that got them to there as opposed to other patients? So our generic mechanism for DILI is fairly fuzzy right now because it may be different, and I think it is different for most drugs, but what we tend to think is that as the drug gets into the body, it needs to be metabolized to get out of the body, and we know that that occurs predominantly in the liver as well as the kidney and the gut. We have oxidative metabolism, and the ADME genes, which we've known about quite a bit, have variability amongst us, and perhaps variability in this oxidative metabolism may predispose you to that. Well, the support behind that is actually not very good with what we know so far. With that being said, if you take the idea that perhaps inadvertent metabolism to reactive metabolite or a neoantigen occurs in the liver, and then that sets off a series of intracellular events like oxidative stress, mitochondrial toxicity, inhibition of biliary transporters, that in the appropriately susceptible individual, that may then set off a reaction from the immune system to lead to a different form of liver injury, whether it's cholestasis, hepatitis, necrosis. And of course, it may be that a lot of people go that way, but they have protective, adaptive mechanisms that prevent them to developing the full-blown DILI event. And therefore, perhaps the immune system may be more important than getting to the original cause of the hepatotoxin. So genetics, this is a logical area to apply pharmacogenomics, and we have. The DILI network has been collecting cases since 2003, and there's several other networks around the world also doing this. In over an eight-year period, we were able to collect 783 Caucasian DILI patients. We had a replication cohort of 307, and we had a large number of population controls. And within this population of cases, there were over 200 suspect drugs. And when we did a GWAS looking at greater than a million SNPs across the human genome, we found no association, when you aggregated the cases together, by the liver injury pattern, age, presence of immunoallergic features, and so on and so forth. So common variants do not appear to be responsible for DILI as a general phenotype, is what we can say based upon that work. However, if it was a rarer variant and we didn't look for it with the chip that we had, that's still as possible. Or it could be that we were inadequately powered for an individual drug. And it turns out it's probably the latter from other data that's come out. So drugs in development where DNA was collected and were proven to be hepatotoxic, with as little as 40 cases of lamirococcib or 70 cases of zymelogatrin with treated controls, lo and behold, there are drug-specific HLA alleles that associate in the patients who got liver toxicity versus those that didn't. You can see odds ratios there of four to five. But the overall allele frequency of these variants is actually pretty common. And then with post-marketing experience with drugs like Lapatinib, Augmentin, several HLA alleles are associated with that. And then Flucoxycin, which is probably the strongest genetic association that we've seen, you can tell that there's some genetic susceptibility, but it's drug-specific. Now it turns out as we were patient and continued to accumulate more cases and repeated the GWAS recently, with 1,800 Caucasians and other African-American and Hispanic cases, we did find one gene that was present across all DILI cases, and that's PTPN22. This was just published. And this very interesting gene associated with other autoimmune phenomenon. Overall odds ratio was relatively low at 1.4, which is probably why we didn't pick it up on the first pass. But it's the first non-MHC-reported SNP associated with DILI risk, and it works across all drugs. And just to show you how we can then add genes together with the Augmentin story, you can see here in this table that if you look for the PTPN22 allele and these two HLA alleles that if in a dose-dependent manner, if you have all three, your odds ratio is 13 versus not having any of those genetic alleles. Now flucloxacillin is sort of the best story here, if you will, of susceptibility to drug-induced liver injury on a genetic basis, and this is in the HLA-B5701, and you can see it was very strongly associated here. Eighty-four percent of the cases had it in the initial discovery, and 87 percent in the validation cohort versus 15 percent with a significant odds ratio. But because it's such a frequent allele in a rare event, only one in 500 of the susceptible individuals will actually get DILI. Now how does this play out? How good is this testing for predicting drug-induced liver injury? Well, it's not very good at identifying who will get it, but it will tell you who won't get it. So in other words, it has a high negative predictive value. So if you don't have the allele and you have someone with jaundice who took flucloxacillin, you can use it to assess the patient, but you can't identify someone who's necessarily going to get the injury or not. And I would point out that actually the performance characteristics of these genetic tests are actually not that much different from what we use for diagnosing autoimmune hepatitis based upon autoantibodies. Now are there other tests beyond genetics? And we're working on it. Certainly liver-specific biomarkers that may be more sensitive and specific than ALT are being developed, like GLDH and MIR22, but they're not going to be DILI-specific. They're just going to be probably liver-specific, and we clearly need more validation there. We looked at patients who already had a DILI event and looked at their lymphocytes and exposed them ex vivo to the suspect drug, and lo and behold, that lymphocyte transfer information test did not work very well in our hands. There's an intriguing technology called the Metaheps, which was recently published where lymphocytes from their peripheral blood are cultured up into hepatocyte-like cells and exposed to the drug with very promising data, but we need to see validation of this. So going back to clinical practice, there's very few instances where we can use pharmacogenomics to prevent adverse reactions, and I just point out to you the two most famous ones, if you will, which is carbamazepine, which can cause a very severe, life-threatening Stevens-Johnson syndrome, and when this was studied in Asia, it turns out that the individuals who got the Stevens-Johnson syndrome had an overrepresentation of this particular HLA allele, and so it's actually required by the FDA, if you have an Asian patient, to actually test for that allele and not give the drug to someone who has that allele, and same thing for abacavir, which is an antiretroviral, a very strong association between HLA-B5701 and the development of this hypersensitivity reaction, and if you actually do the testing, you reduce the instance from 9% to 0%, so we do have two instances of preventing adverse reactions, but these aren't dilly. So what else can you do for prognosis? Histology we use a lot, and there are some useful things, so if you see granulomas and eosinophils, that's a good thing, those patients generally do better, whereas if you see significant necrosis or fibrosis, those patients do worse, but that's not surprising, I think. High's Law, you're all familiar with, is just looking at a combination of the ALT and the bilirubin in someone with suspected dilly. Patients who meet that criteria have about a 10% mortality. Within Dillon, we looked at High's Law versus the MELD score in 7% of patients who died, and lo and behold, the MELD score is actually better than High's Law at predicting patients who aren't going to do well, and if you get to the point of acute liver failure, things are pretty dire, and actually only about a 30% transplant-free survival. So we're working through proteomics and discovery to try to identify better prognostic markers. This was a recent paper suggesting that MCSFR and osteopontin may add value, but you can see we have to probably add these to other things that are part of the MELD score for them to be utilized, and then there's a recent interesting publication in Gastro from the Indiana group looking at very simple bedside parameters for patients who have dilly, and the Charlson Comorbidity Index here, you can see in patients with a MELD score of less than 19, that if you had an index of greater than 2, you're much more likely to die than if your index was less than 2, and same thing if your MELD score was elevated. So simple clinical things may come back to add to what we know. Now I would point out to you there are a very valuable resource for everyone to use, which is the Liver Talks database developed by the NIDDK and the National Library of Medicine, where there's basically an e-textbook of 1,200 prescription drugs and over 100 herbal and dietary supplements with what the phenotype of human liver injury is, and I think that's a very valuable resource for now in 2019 to try to come up with more confidence in your diagnosis when you're seeing patients. I also would point out to you there's about 100 drugs being approved every year. It's very difficult to keep up in this space, so I would refer you to the FDA website where they're constantly collating all of the different associations that have been identified, and just again to summarize, only 10 pre-therapy pharmacogenetic tests are currently recommended in the United States for efficacy, and only four to prevent adverse drug reactions. So we have a long ways to go, and there's several other databases here adding to this effort. So to summarize, where are we at with precision medicine and idiosyncratic drug-induced liver injury in 2019? DILI is an infrequent but important cause of liver injury, and the difficulty with establishing the diagnosis has in part inhibited the research, but now with networks and more awareness, I think we're starting to make some progress getting adequate numbers. I personally think that DILI, there's a component of genetic susceptibility, but I don't think that's going to tell the story. We know that already actually from having done GWAS in over 2,000 cases, but it's probably some of the variants, and it's probably going to end up being drug-specific genetic susceptibility above and beyond the PTPN22, and that in clinical practice, at least in the UK and other parts of Europe where flucloxacillin is available, HLA-B5701 is actually useful to confirm a diagnosis of someone who took flucloxacillin and developed jaundice, but it doesn't prevent it. And then in terms of outcomes, we have the MELD score, we have High's Law, we have this Charleston Comorbidity Index, which we may be able to use, but obviously we're looking for more precise individualized ability to predict prognosis and still working in that. So in summary, I think we need additional studies to improve our ability to use the principles of precision medicine in diagnosing, preventing, and prognosticating in our individual patients with DILI. Thank you.

clinician

complex case

liver injury

antibiotic treatment

idiosyncratic drug-induced liver injury

precision medicine