I'm going to be, I'm Charlie Marmer, and I'll be the sole presenter for this session. I'm the chairman of the Department of Psychiatry at NYU, Grossman School of Medicine, and I direct, among my responsibilities in my research lab, I direct the Center for Precision Medicine in Alcohol Use Disorder and PTSD. The goal of our center is to advance evidence-based diagnostic criteria using laboratory tools that go beyond patient self-report and family interviews, and to adjust treatment specifically to each patient, that is to say, to go beyond treatment guidelines, which are designed for average patients to actually personalize treatment in the way treatment is done for cancer and heart disease. And I'm going to be describing some genomic tools, neurocognitive tools, neurocircuitry tools, and clinical features that can help us personalize both diagnosis and treatment, and eventually get to the point where in psychiatry we can subtype our psychiatric disorders so that the patient that you're consulting with when you go back to your practice can be given treatment that is specific to them, rather than running through an algorithm which may take two, three, four trial and error attempts before we get the correct treatment. So that is the goal of my center, and I'm delighted to present my work to you. I'm going to present six studies today. Three of them are complete, and three of them are in progress, but they're designed as an ensemble to illustrate the principles of how in the future we will use precision medicine tools borrowed mostly from oncology and cardiology to advance the personalized treatment of psychiatric illness. It's very challenging, and we'll talk about why it's been so difficult in our field. You're, of course, aware of many of the challenges and barriers, but we'll talk about some of them. So these are my disclosures. I'm very heavily supported by NIH, DOD, the U.S. Army, and multiple foundations in my research. None of my disclosures represent specific conflicts for the presentations I'll be giving today. I like to playfully say I have no serious conflicts in this presentation other than the usual unresolved Oedipal conflicts. So what are the challenges in psychiatry? These are self-evident, I think, to all of you as experienced clinicians and interested in clinical research, but let me go through them one at a time to say what really the tremendous challenges we face in our field and how we're going to try to address them. So compared to other fields of medicine, psychiatry is the only major field of medicine that is not based in terms of diagnostics and therapeutic decisions, is not based in laboratory medicine. And one of the disappointments I've had in my lifetime, but I appreciate the challenges, is that this was equally true when I was in residency training and remains true for our residents now. We make diagnoses by interviewing our patients. We use some structured tools, which are helpful. We interview, we get collateral history, but fundamentally, we don't use many objective observable features, particularly we don't use much in the way of blood tests or imaging tests to confirm our diagnoses, which is interesting because we use a lot of blood tests to rule out other problems. So we know if our patient's having a thyroid problem or an infectious disease that's causing their symptoms, but we don't have laboratory-based tests, for the most part, to confirm or disconfirm the diagnosis of unipolar and bipolar depression, PTSD, schizophrenia, and other disorders. And that's despite hundreds of millions, if not billions of dollars of research on the molecular and neurocircuitry basis of psychiatric illness, we still don't have things that are scalable and useful in everyday clinical practice. And as someone who is very busy administrating a large department of psychiatry and has a very large research lab, I still do 10 hours a week of clinical work, and I'm constantly frustrated that I can't use any of the studies that I'm doing in my research lab and in the field to actually help me diagnose and plan individualized treatment with my patients. And so that is the goal. We're going to try to address them. We need to get past subjective complaints. Now what are the problems with subjective complaints? In psychiatry, much more so than in medicine, because of the obvious problems of stigma, denial, and other problems, our patients' subjective complaints are often unreliable. And in particular, in the fields that I'm interested in, alcohol use disorder and PTSD, there is systematic under-reporting of the difficulties the patients have. I don't know if you've had this experience, but I've sat across from one of my patients and asked them, are you having any difficulties with alcohol at this time? And the answer is, well, no, Dr. Marmer, since I've been in care with you, I'm managing very well. And then I get a call the next day from the family that they found multiple bottles of vodkas in the patient's bedroom. And so the patient is not a habitual liar. The patient is deeply ashamed of their alcohol use. I sit down with war fighters and tell me that they're not having nightmares, flashbacks, and startle reactions. And I get a call a week later from their spouse that they can't sleep at night because the patient's trying to strangle them during a combat-related traumatic nightmare. So our patients, particularly in sensitive areas such as alcohol, stress, mood disorders, and particularly in psychotic disorders, they're unreliable reporters. As a matter of fact, the organ of the illness, insight and judgment are a part of every mental status examination. And psychiatric illness in every form interferes with insight and judgment. And yet we rely heavily on the insight and judgment of our patients to make diagnoses. That's one problem. Another problem is most psychiatric care is not provided in specialty care by us. Most psychiatric care in America, even in highly developed countries, is provided in primary care. Seventy percent of all patients with stress, anxiety, and depression, alcohol, and drug abuse are treated in primary care, not in specialty care. And primary care doctors don't have the time or training to do the kind of very sophisticated interviewing that we do to try to sort through all these problems of denial and over and under reporting. So we need tools, high-throughput screening tools, for primary care. In the developed world, almost all of psychiatry, in the developing world, I have colleagues working in sub-Saharan and East Africa, almost all of the psychiatric care is provided in primary care, and they're tremendously disadvantaged. Another problem, which is obvious to all of us, is that psychiatric disorders are highly heterogeneous. Now, that's true of medical disorders as well, but in some ways we're like practitioners in the 18th or 19th century trying to deal with fever of unknown origin. There could be 50 causes. There are 50 causes of depression. There are at least six, maybe more, clear clinical subtypes of depression on clinical grounds, but we don't have laboratory confirmation for them. So we need to be able to partition the heterogeneity of psychiatric disorders. We haven't been able to do what's been done so successfully in other fields of medicine, which is to subtype our disorders for purposes of risk assessment, determining the course. If someone presents to my office and to my consulting practice following a sexual assault, I don't have really good objective tools to know whether they're likely to recover in the first few months with brief intervention, or whether they may be disabled for years with their psychiatric difficulty. And so we don't have good ways to determine the course, and worst of all, I think, or most challenging is we don't know which treatment for which person at which point in their illness in an objective, reproducible, scalable form. So we're practicing psychiatry the way oncologists practiced oncology when I was a medical student. If a woman presented with breast cancer or a man presented with prostatic cancer at that time when I was a medical student, they were given some combination of radiation surgery and chemotherapy. The order wasn't clear, necessarily, and it wasn't clear when you gave a combination of treatments, and think about how oncology is practiced now. If a woman presents with breast cancer now, the cancer is biopsied, objectively identified, single cell genomics are done, the genomic markers are identified, and the patient is provided with monoclonal antibodies that are specific to their cancer. And those monoclonal antibodies may have been developed for lung cancer or prosthetic cancer or some other form of cancer. It's not important. What's important is that's the right treatment for that person with the right configuration of their molecular pathways that have been disrupted by the cancer. Now what do you think is the fundamental advantage that cancer oncologists have that we don't have? We're very sophisticated in psychiatry, as you'll see from some of my presentations. We understand as much about molecular pathways of neuropsychiatric illness in principle as oncologists know about their molecular pathways. What's the problem that we have in our field? We are the only field of medicine that doesn't do what? We don't biopsy the organ of our illness. Now if you're interested, email me, because on a single occasion we actually did that. A patient was depressed for 20 years, had failed to respond to all known psychotherapies for depression, psychoanalytic, cognitive behavioral, interpersonal, all known pharmacotherapies for depression, SSRIs, SNRIs, tricyclics, monamine oxidase inhibitors, anticonvulsants, and every other drug that's ever been used for the treatment of depression over 20 years, and failed to respond to ECT. Think about the damage to that person's life over 20 years of being chronically ill with depression and having been subjected to all those treatments, each of which has their own very significant side effects, and each of which affects molecular pathways of every major signaling system in the brain. So in this case there was a concern the patient might have had an autoimmune encephalitis that had been underdiagnosed. We felt it was ethical in this case to biopsy the patient's brain. The patient went to neurosurgery. They had a biopsy, a needle biopsy, to a silent area of the prefrontal cortex, and that sample was sent to neuropathology. It came back with massive invasion of inflammatory scales. Patient did not have a systemic inflammatory disease. Their peripheral markers, HSCRP and SED rate were elevated, but not dramatically. So this was an unexpected finding. The patient was treated with a new generation anti-inflammatory drug. I think it was infliximab or one of those drugs, and it crosses the blood-brain barrier, and for the first time in 20 years the patient had relief from their depression. So imagine 20 years of trial and error, failed treatments, and the only guidance for treatment came after doing a brain biopsy. You can ask the question, why don't we biopsy the brains of our patients? It's probably a legacy of the bad public relations we had, and justifiably so, with lobotomy, but the fact of the matter is we need to be able to actually know what's going on in the brain to treat difficult cases in our practices. And I'll talk to you about, we think we have a way to do this now, a liquid biopsy of the brain, I'll talk about that later. Okay, what about practice guidelines? How many of you actually practice using practice guidelines? Show of hands. About a third. The rest of us, of course, think that we know much better than practice guidelines. We have a term for that in psychiatry, which is delusional thinking, but I'll leave that aside for the moment. The problem with practice guidelines is they're not developed to practice personalized medicine. They're developed to answer a simple question. On average, if a patient presents to your practice with a new onset major depressive disorder, where should we start? And if it doesn't work, where should we go next, and so on. And there's published guidelines for every psychiatric disorder, I've been part of the effort to try to do that for PTSD, it's been done for AUD and every other disorder, and they're quite limited in their utility. The patient that I described to you with 20 years of depression had been treated with practice guidelines, and they had failed for 20 years, because this patient didn't have a form of depression that fit into the standard guidelines. Okay. Guidelines are mostly trial and error. First and second line and third line treatments are average responses, not personalized responses. And another thing, what is the problem with clinical trials? I'm going to give you some startling data from one of our recently completed studies of a clinical trial. What's the fundamental problem with clinical trials in our field, would you say? Speak loud, because they're... Partially, yes. Other problems with them? What's the problem with placebo responses in our field? We get very robust placebo responses in clinical trials. The reason for that is someone who has toiled in the trenches in clinical trials, and suffered the slings and arrows of placebo responses. Matt Friedman and I did a 10-site study of sertraline for PTSD in veterans. Now, as you know, sertraline and paroxetine are the only two FDA-approved drugs for the treatment of post-traumatic stress disorder. We did a study, a 10-site study with male veterans, and we got no... Drug did not separate from placebo at all in veterans. Why? One, the veterans that were enrolled in the study had failed multiple previous treatments, so they were poor prognosis. Second, they were being treated in VA medical centers with very limited resources. When they came into our trial, they got a huge amount of attention from our research coordinators, and we had a massive... It's not that the drug didn't work. The drug gave a 50% response. There was a massive placebo response, because psychiatric patients who have been getting minimal treatment really do respond to the therapeutic alliance, to the human contact, to the caring and compassion and concern, which these young, enthusiastic research coordinators were very passionate about caring for the patients in the clinical trials. So, we get killed on placebo responses. I'm going to show you startling results from a study of alcohol use disorder. It was a gabapentin prodrug, in which drug did not separate from placebo at all. However, using advanced analytics, we were able to find a subgroup of patients in the drug group that did far better than placebo. So, another problem we have is we need to know how to find the likely responders within a heterogeneous group of outcomes. Okay. And finally, and this is more my problem probably than yours on an everyday basis, but it's our collective problem as a field. We are trying to develop biomarkers for psychiatric illness, and where we're looking is in peripheral blood, in urine and in CSF, saliva, and what's the problem with us? They're very far away from the brain, and they're very noisy. I did a systematic review of the literature on DNA and RNA markers in peripheral blood. They're exceptionally complicated. They come from multi-different organ systems. They primarily come from PBMCs, which do not reflect the pathways in the human brain. And even viruses and bacteria affect the findings we do when we study the genetics of peripheral blood. So if you want to develop a blood test for a psychiatric disorder, good luck, because it doesn't actually reflect what's going on in the brain, in the molecular pathways in our brain, and it's very noisy and difficult to replicate. We've had some luck, but I'll show you the complications. So until we can get at the actual pathways in the brain, so we have some ideas now how to do that. It's done in post-mortem studies, which are not too helpful. It's done in stem cell studies, which are interesting but very expensive. We need a reliable, inexpensive, scalable way to determine what's actually happening in the brain in our patients, and I'm going to suggest that we can now do that for the first time using neuronally-derived exosomes, and I'll talk a little bit about them later. So solution. We need biomarkers, but they have to reflect what's actually going on in the organ of our illness, the brain. We need to be able to subtype patients for risk, course, and treatment. We need to personalize treatment. Every one of our patients presents with a different configuration of clinical symptoms, a different pattern of alcohol use, a different history of trauma, different neurocognitive ability, and different molecular and circuit features. As one example, among patients with post-traumatic stress disorder, we've now shown that 30 to 40% of them have very significant cognitive difficulties. They're quite impaired in terms of attention, concentration, working memory, and other cognitive functions, and they don't do well in conventional treatments, such as PE. In fact, I'm going to present a study and show you one of the first precision medicine findings in PTSD, where we showed that PTSD patients, male veterans replicated in female civilians, very different population, male veterans with combat-related trauma, women with mostly sexual trauma. If they had significant neurocognitive impairment and some circuit features of that, they did not respond to prolonged exposure therapy virtually at all. That's an example of precision medicine. We need advanced computational modeling to try to determine what is the heterogeneity and who is and is not likely to respond to a treatment. The idea of precision medicine is very simple to describe and very, very difficult to do in our field. It reminds me of the investor Warren Buffet, who was asked, how have you been so successful? He said, well, I have a formula, and it's very easy to describe the formula. The problem is it's very difficult to actually make the decisions. This is the problem we have. It's a very easy idea. Our patients are highly heterogeneous, so if we take a patient with PTSD and we refer them for so-called evidence-based treatment with prolonged exposure and SSRI, the problem is they're highly heterogeneous, so on average, maybe at best 30 or 40 percent will respond. But if we actually knew what the problems were in terms of the clinical features, cognitive functioning, molecular pathways, and brain circuits, and could subtype the patient, then we could say this patient requires this treatment, and 80 percent of them would respond. And the patients with the other subtypes would not respond, and probably even more important, we would know not to give them those treatments. My depression case is an extreme example of that, right? The only thing the patient could respond to was an anti-inflammatory drug because the cause of their depression was massive neuroinflammation, and you can run through all of the other evidence-based treatments, but because of the heterogeneity of the presentation, the patient is not going to do well. Okay, so this is what's done in cancer, and it's done so successfully that patients that used to be in this category are now in the 80-plus response category with specific treatment for their form of cancer. We need to do that in psychiatry and including in AUD. Okay, so I'm going to, I hope, I've spent my whole life studying PTSD and TBI, and one of my residents who's cynical and hilarious said, Dr. Marmer, you spent your whole life studying PTSD and TBI. Why do you give us both disorders with your lectures? So I'm going to give you six studies, and I hope you won't be concussed afterwards because they're fairly complicated and they'll go fairly quickly. The first three studies, two of them in veteran populations and one in civilians and veterans, are completed studies. Study one is a study of an attempt to develop a blood test for screening for PTSD with all the problems. Second, we'll be using voice markers where we've had very good success. Third, we're looking at cognitive impairment in PTSD and how it affects treatment response. And then three studies, two of them funded by NIAAA and one by the DOD, where I'm going to show you what the next generation studies, oh, I'm sorry, I'm going to show you what the next generation studies will actually look like, let me go back here, will actually look like in very sophisticated precision medicine studies. So study one, this study is funded by the Department of Defense. It's a very unusual study. We're in the 13th year of a consortium. The study has received $100 million in Department of Defense funding over 13 years, and it's designed to answer a simple question. Can we find a blood test that could be used in primary care practice in active duty military and veteran populations to screen for PTSD? And I'm going to show you the first result in which we have candidate blood markers. Now, I just want to say by way of background, another problem in our field is we're a little bit like the person who's lost their keys in the dark and is looking for them under the lamppost. The light's there, but the keys aren't there, right? So that's one of our problems also. So we've been looking in PTSD and AUD, we've been looking at very simple markers. In PTSD, we look at the HP axis and look at cortisol and we look at adrenergic markers and if we're sophisticated, a few renin-angiotensin markers. But we don't do what's done in cancer, which is look at the entire human genome and really trying to find out which molecular pathways are actually perturbed. Now we've attempted to do that for the first time in PTSD in this consortium and let me show you what we found. So we use an approach called systems biology. Systems biology was developed in oncology and it takes the following approach, that we have to understand all of the elements in a molecular pathway from DNA to DNA expression, from DNA expression through transcripts through to proteins, metabolites, and other features and to try to understand how the system, which is complex and multi-genetic, works together to explain our findings. So we've done that in the following way in this study. So the attempt is to develop a blood test for screening for PTSD. We use the usual machine learning approaches. You have to have a discovery sample where you get candidate blood markers and then a completely independent test sample to try to see if they replicate. So we started with 83 PTSD cases and 83 controls. In this case, they were male PTSD cases and male healthy controls. They were all Iraq and Afghanistan veterans. They'd all seen moderate to heavy combat exposure. Some had childhood trauma. Some did not. Some had comorbid depression. Some did not. So we have a validation and test sample. And once we found our candidate markers, we recruited an additional 29 cases and 40 controls to see if they would actually separate cases from controls in a completely independent sample on which we had not developed the panel of blood biomarkers from the first sample. You have to be very careful about that because you're biased if you try to replicate within the same population. Okay. So where to start? Very difficult question. The problem is the human genome is very complex, as you know. There are 22,000 genes in the human genome that code for proteins, which we're interested in. And they actually, these 22,000 genes have very complicated expression patterns. So we are able to study 850,000 messenger RNA transcripts from these 22,000 genes and study then many hundreds of proteins and metabolites. And yet we have limited sample size, so we had to be parsimonious. We did something very interesting called wisdom of the crowds. We brought together seven major academic centers of people with expertise in molecular pathways, in this case PTSD. We could have done it for AUD. So this is six universities, Harvard, UCSF, NYU, and several other universities. And we also brought together the Army's most sophisticated molecular biology laboratory. So we included in our group Frank Doyle, who is presently the Dean of the School of Engineering at Harvard, and is one of the world's leading experts on how to do computational work in these pathways. Kerry Ressler, one of the leading experts in the neurobiology of stress. The metabolism group at UCSF, Owen Walkowicz, Cindy Mellon, and Victor Roos and others. And we included Lee Hood, who is the world's leading expert in proteomics and who actually developed PCR. So this is like an elite team of molecular biologists with the knowledge of stress. And we asked them to nominate, included Rachel Yehuda, who is the world's leader in the HP axis in stress, and went from Sinai. And we asked them to nominate some panels of candidate markers. And it was wisdom of the crowds. I don't know if you're aware of this, but when you take your minimum daily requirement of vitamins every day, do you know how those were determined? Interesting question. I looked into it. They were determined by the wisdom of the crowds. At the beginning of World War II, the Army wasn't sure what vitamin levels to include in rations for soldiers deployed to war. So they brought together experts in nutrition, and they sat around a table, and they nominated the levels. So that's how it was done. It wasn't empirically validated. Wisdom of the crowds. Wisdom of the crowds is not a bad method, actually, if you have the right, if the crowds are selected to have the right wisdom, right? It can be a little difficult sometimes. So we started with 50 panels, and they included 343 molecular features that were agreed upon across all of the known published studies in PTSD, and by the experts doing research on them. And we allowed them to be genes, epigenetic marks, transcripts, proteins, metabolites. Any feature in blood you want was on the table. And we ended up with a consensus. We started with 343. They included, actually, two physiological measures. Heart rate was one of them. Interestingly, they included 20 features, which you can get from routine clinical labs. Glucose, fasting glucose, lipids, other things that are available in CEDRATE, CRP, things that are available in the labs. And if you're interested, I won't be presenting this today, but we just published a study to see if we could use routine clinical labs that you can order on any of your patients to see if we could separate from PTSD and TBI. And we did pretty well, actually. So routine clinical labs using 46 labs that are available, that are taken when you have your annual physical. If you do a very sophisticated data mining of them, you can begin to separate psychiatric disorders. It's very interesting. Please email me if you'd like me to send you any of these papers. I'm happy to give you my slides, and I'm happy to send all of our publications. I'm easy to find. I'm the chair at NYU. So just email me, and I'll send you whatever you're interested that I talk about today. But clinical labs are important. We included eight endocrine markers, including cortisol and ACTH. We included 27 metabolites, 156 methylation probes. This was very interesting, that it turned out, based on the wisdom of the crowds, that the single largest candidate for candidate blood biomarkers for PTSD were epigenetic changes. Maybe that's not surprising, because trauma has a profound effect on the epigenome. I was surprised by that. We included 81 micro RNAs, which are small, non-coding RNAs, but they affect the expression of other genes, 42 proteins, four other small molecules, and then a couple of complicated nonlinear combinations. So these were the features. So the first thing we did was ask, if you have all these features, using advanced machine learning, using advanced machine learning, can you begin to eliminate the features without decreasing the accuracy of diagnosis? So the very first thing we did is we asked, how well do we classify PTSD cases from controls if we use all 343 features? And the answer was, a little hard for you to see here, but maybe I can, can you see this arrow? Yeah. So you can see that the classification rate, which is based on the area under the curve here, is about 0.8586, which is very good. In other words, we got a high level of accuracy of classifying PTSD cases from controls if we used all 343 features. But as you know, molecular pathways are very interrelated. The genes and the epigenetic marks and their transcripts and their proteins are part of pathways, and they're interdependent. So they're unlikely to be independent features. And the question is, could we reduce them and get a minimal number that are most predictive? And we used a procedure called recursive feature elimination, which simply means one at a time you drop the features and watch to see how the accuracy declines. And what's interesting is you can see that as you go from 343 features down to 77 features, the accuracy does not degrade at all, meaning that the 77 contain all the information that you had in the 343. But once you start to drop below that point, the accuracy starts to quickly drop off. So that led us to having 77 candidates. And the next thing we did is we did repeated machine learning to see which of those 77 candidates were most important. And this is a little bit complicated. This is a machine learning paradigm that basically runs, does a whole series of repeated tests on the sample to try to determine which of the 77 features, each of these bars represents one of the 77 features, is most important in classifying cases from controls. And you can see as you move from left to right across this figure that when you get to a certain point that there are a limited number of features that are highly important in the classification. The others are less important, right? So that's an approach we've taken. And we ended up by selecting the features in blood that were the most important, that each had a certain threshold of at least 37, 30 percent level feature important. So we went from 343 features to 77, and now we have 28 features which really carry, allow us to classify with almost the same accuracy as the original 343. And here's what they are. Here's the final panel. By the way, I'm just presenting this to you not so much because this is definitive. All has to be replicated in multiple samples. I'm presenting this to you as a way of thinking about precision psychiatry that in the future is going to lead us to accurate diagnostic tests and precision treatments. So the first thing we found is that heart rate's very important. It survived. Heart rate turned out to be one of the top features. Three metabolites. Interestingly enough, the metabolites, when was the last time you all thought seriously about the Krebs cycle? Tell me when. When was it? I love it. I myself must confess to the following. I made this comment at another seminar, so forgive me if you came to that one. But I remember a biochemistry test in second year medicine on the Krebs cycle. And I remember thinking, thank God, I'll never have to think about the Krebs cycle again. And now my life is completely centered on the Krebs cycle, which is some kind of punishment for my arrogance, of course. The other thing is, I have a very good friend here at UCSF who turns out to have done a fellowship with Hans Kreb. And he actually told me about Kreb and his life, and it's very interesting, and how he did this work. And there are many, many wonderful stories I could tell you about Hans Kreb, who was an incredible person. You remember that he also discovered the Krebs-Urea cycle, which is the most important cycle in all of biochemistry, because it accounts for protein metabolism. And how many of you are in teaching settings where you have residents working with you? Good. And how many of the residents complain about workload and work hours? Good. So I tell my residents the following story when they tell me that 80 hours is too many hours a week to work. I tell them about Hans Kreb, who was a PhD biochemist working in Germany before World War II. He was Jewish, and he was a German Jew, and he was a very gifted biochemist. And he decided he would go to medical school because he wanted to contribute to health, not just basic biochemistry. So he was in medical school right before World War II, and he was working on the Urea cycle. He had worked out part of the features of the Urea cycle, and he asked his supervisor, who was a member of the Nazi party, if he could have one day a week during his internship to continue his research on the Urea cycle. And his supervisor, who knew he was Jewish, said, if you're that smart that you have time to do research as an intern, I'm going to assign you two medical wards to cover instead of one. Now during that year of his internship covering those two wards, Hans Kreb completed the work on the Urea cycle, for which he won the Nobel Prize. That's a conversation we have about work-life balance. Of course, you understand I may not be able to go back to NYU because it's too dangerous for me, so I may have to stay here. But it's an apocryphal story. So Hans Kreb turned out to be an incredible person, very interesting, did the most important work in biochemistry ever. But it turns out that among the molecular features that we found that discriminate PTSD cases from controls, three of them are from the Kreb cycle, citrate, pyruvate, and lactate. And they are, as you remember, features of aerobic and anaerobic metabolism. And in PTSD, there's a shift from aerobic to anaerobic metabolism. And those features, citrate, pyruvate, and lactate goes up. Pyruvate goes down, whatever, those are the ratios. And they turn out to be very, very important. Four microRNAs were very important, one of which, miR95P, which is involved in neurogenesis. Two clinical labs, insulin and MPV. Eleven methylation probes are important. And seven proteins are important. And we were able to take that panel, and in a completely independent sample of cases and controls, we were able to classify those cases from controls with these 28 molecular features with overall accuracy of 81%. So this is the first, one of the first examples ever in psychiatry of the use of a blood test for high-throughput screening. Now, we're not saying that this is a definitive diagnostic test because it requires validation, but that it is useful for separating cases from controls. I'm going to move on to the second case. Can we use, are there other objective ways to differentiate cases from controls? We've been very interested in human speech. And we think that human speech carries, we think of all the digital markers that are now being studied in our field, and NIH, by the way, just put out on our request for applications for digital markers to advance psychiatric illness. We think of all of them, the most interesting is voice. How many of you, how many of you do telepsychiatry? Everybody. How many of you have done telephone psychiatry? Do you think there's any difference? Minimal, right? It turns out the data show that telephone psychotherapy is probably very close in terms of its effectiveness to video-based psychotherapy. Why would that be? What is it about the human voice? Let me ask you a simple question. You pick up the phone after you walk out of this lecture, and you call a family member. It's not a FaceTime call, it's a phone call. How many seconds does it take for you to know how they're feeling? Five seconds? The reason for that is that we are herd animals, and before the internet, we survived by vocal communication. So the human voice evolved, and mother-child relationships, and family relations, and group relationships, the human voice evolved to express emotion. So I think for that reason, telephone psychotherapy is probably very close to voice plus face. Speech is a primary form of human communication, it was evolutionary to develop emotional states. It's simple, non-invasive, low-cost, ubiquitous. In the age of smartphones, voice is ubiquitous, it can be used for high-throughput screening. We think of all the digital markers of emotional health, voice will turn out to be the most important. Let me show you some very interesting results. We studied 22 PTSD cases and 77 healthy combat-exposed controls. These were all Iraq and Afghanistan veterans. We did recordings of clinical interviews. We did a very simple thing, we asked them to describe their worst combat experience, or their worst war zone experience, we recorded it with high-fidelity recordings, and we asked them to talk about a neutral subject. We got 35 minutes worth of speech, we now find we can do it with five minutes of speech, and I was very lucky, I got a call out of the blue from Stanford Research Institute in Menlo Park, very close to here, from a group of bioengineers who said, Charlie, would you like to collaborate on voice markers for stress and other psychiatric disorders? I didn't know who they were, I found out subsequently that they were the group of engineers that developed Siri, and they also developed the code for Dragon Naturally Speaking, which is the best voice translation software. These are the top bioengineers in the world in the area of voice and voice quality, and they wanted to work, and they had DOD funding, we worked together. Here's what we did. We took the audio recordings from these PTSD cases and controls, we isolated the speech segments of the patients, the participants, and we ran them through the SRI platform, in which they had developed for some of these other applications, for example, for Siri. It turns out that these bioengineers, these are PhD biophysicists and bioengineers, they're able to fractionate the human voice spectrum into 40,000 unique biophysical elements. By the way, this has nothing to do with the content of speech. This is not natural language processing. This is just the biophysics of the waveform of speech. They used an advanced form of Fourier analyses, and they developed 40,000 unique features in the human voice spectrum, and we used them, and we used machine learning to isolate the ones that were most interesting, and we used a random forest classifier, and we were able to reduce the 40,526 features to 18 features, and here's the accuracy we got for classifying PTSD cases from controls without any knowledge of the content of speech, with no clinical features, only the wave spectral form of their voices, 89.1% accuracy in differentiating cases from controls. We're now replicating this in women and civilians, and if it holds up, we will have a voice test that you can load up and use when you're doing telepsychiatry and get a readout on the voice quality of your patients. These are the important features. The names are very technical, as you can imagine, as they're biophysical features, but it's interesting. When I looked at the features, I thought they would be features reflecting anxious arousal in the cases, but the PTSD cases turned out to have vocal quality that was more flat, more muffled and unclear. There was less tonal change, and so this appeared to be reflective of emotional numbness in the more chronic cases as opposed to anxious arousal. Study number three, and this is a precision medicine treatment study. This is a study that we conducted, and I was fortunate to be part of a collaboration with Amit Atkin and his group at Stanford in the Department of Psychiatry and some other collaborators. This is a large collaborative effort as well. We published this work in Science Translational Medicine, and we asked a very simple question. If we can understand the cognitive functioning of patients with PTSD using cognitive tests, simple cognitive tests like measures of attention, concentration, and working memory, and also use circuit-based cognitive tests from neuroimaging, can we find a subtype that responds better or worse to the so-called gold standard treatment for PTSD, prolonged exposure therapy? We really tried to take on the precision medicine question. PTSD cases are heterogeneous, as are all psychiatric disorders. Can we find a subgroup that's responsive and not responsive to prolonged exposure? By the way, my group has published in JAMA in two different articles repeatedly that only about 40 to 50 percent of patients with PTSD respond well to prolonged exposure therapy. Half do not. So what's the difference? We used neuroimaging from fMRI to interrogate major circuits in the fMRI circuitry. And just a little bit of circuit background for you. The way human brain circuits work is they're organized into networks. And to work efficiently, the networks need to meet two criteria. They need to be segregated well from each other so they're not overlapping, and they also need to communicate well with each other. So networks, we could talk about the fear network, we could talk about the reward network, we could talk about the executive function networks, they all work this way in the human brain. There are complex signaling pathways through the brain, and these networks need to be well differentiated from each other, but they also need to communicate and talk to each other. And we actually can derive indices of how well the networks are segregated and integrated to be high functioning. And here's what we found. We found in PTSD cases, compared to healthy controls, we found that healthy controls, shown in the open bar here, have high levels of network integration, but among the PTSD cases, there were two groups. Those with very poor network integration, very impaired, and those with essentially impact network integration in this circuit. And this was an executive function attention circuit. And the same was true, that was found in female civilians with sexual trauma, and replicated virtually exactly in male veterans with combat trauma. So what have we shown? We've shown that PTSD is highly heterogeneous with regard to neurocognitive functioning and neural circuitry functioning, and that there are at least two major groups. Those among PTSD patients, there are those with relatively intact cognitive function and highly impaired cognitive function. And the question is, do they respond the same or different to treatment? Here's the answer. The circuit prediction of outcome was very powerful, and what we found is, if you'll ignore for a moment the weightless control groups, which did not improve, among those who received PE, those who received PE and who were found to have intact cognitive functioning and intact circuit behavior, had a very robust response to prolonged exposure therapy. And those with impaired cognitive functioning did not separate from the weightless control group. So this is one of the first examples of an actual precision medicine finding in psychiatry where circuit features and clinical features, including you can do this, by the way, you can do this pretty well with your own patients if you do a bedside test of attention, concentration, and working memory. You can give them serial sevens, you can give them five objects to remember for five minutes, and if they're impaired on those, they're likely not to respond very well to prolonged exposure therapy. Incredibly interesting finding. Okay. Next. I'm going to switch to AUD. There have been some efforts in AUD to try to identify precision biomarkers or clinical features that will explain who will and will not respond to treatment. These have been done for naltrexone, for sertraline, for topiramate, and I would say overall the results have been unimpressive and not replicable. There are a few features, there are a few genes. For example, Hank Kranzler published in 2014 that a polymorphism of the GRK1 gene was associated with topiramate response. Actually it didn't replicate in subsequent populations, and for naltrexone there was a polymorphism of the OPRM1 opioid receptor gene, but again, we don't think that we can advance the field of psychiatry by looking at single genes for single transmitters, receptors, or transporters because psychiatric illness, as you know from our large GWAS studies of hundreds of thousands of subjects, psychiatric illness is highly polygenic. In psychiatry we're unlucky in many, many ways. One way we're unlucky is we don't have simple gene phenotype relationships. We don't have Mendelian inheritance. We have 50, 100, 200 more genes that are important for all of our psychiatric disorders. What's the other problem in psychiatry? The other problem is our readouts are very complex. Whatever is wrong with the genes, whatever variants we have, they express themselves in the form of complex readouts such as cognitive rigidity, such as emotion dysregulation, such as cognitive dysfunction. Those are ... I don't know if you know who Alec Widge is who's at the University of Minnesota. He's doing deep brain stimulation work in depression. Alec gave grand rounds at NYU last week and he was saying, you know, I started my work, I thought, look how well deep brain stimulation has worked for Parkinson's disease. Surely, we should be able to find the circuit and do this for depression. And he said, I found, after it was very humbling, there's a very simple problem. In Parkinson's disease, you can sink a wire into someone's brain. You can go to the basal ganglia. The patient can have a very, very rapid hand tremor. And when you hit the right circuit, the hand stops shaking. It's binary. It's 0, 1. Tremor, no tremor. You found the circuit. Try doing that in depression. The readout is so complicated. So we're left with polygenic diseases with very complex, hard to define symptomatic readouts. And that's the struggle we have. All right. So I want to talk for a moment now about how to find the needles in the haystack. If so many of our clinical trials are, if so many of our clinical trials, drugs do not separate from placebo, and yet we know that among those treated with drugs, a subgroup do very well separate from placebo, but we don't know who they are. So we have developed some very advanced analytic computational models at NYU. I'd like to share the idea with you now, because we're actually going to go back and reanalyze a very large number of failed psychiatric trials in which drug did not separate from placebo, and we think we can find the subgroups of patients for whom drug actually is superior to placebo. And let me give you, I'd like to give you one example, and it's from alcohol use disorder. This is an NIAAA-sponsored symposium today. I'm a PTSD researcher who got very interested in AUD because so many of my patients suffer from AUD, but I'm a little late to the AUD world, and I've spent 30 years in PTSD research, but I'm now doing a very deep dive into AUD and AUD with PTSD. So this study was a wonderful study. I don't know if you're familiar with it. It's a study, it's a randomized controlled trial of gabapentin-anacarbal extended release for alcohol use disorder. This is not a PTSD study. It's a straight AUD study of very heavy drinkers. The study was actually conducted by the intramural division at NIAAA by Dan Falk and Ray Litton. It's a very, very elegant study. I think it was 10 sites, total sample size was 343 or 346, and gabapentin-anacarbal is a prodrug, meaning it's metabolized to gabapentin. So this is a gabapentin study, and what did they find? After spending $10 million, they found the following. 338 men and women, 338 was the sample size, 24 weeks of treatment with gabapentin-anacarbal, 10 academic U.S. sites, study conducted between June 2015 and February 2017, very strict inclusion-exclusion criteria, and they used gabapentin extended release 600 milligrams twice a day, and the result was profoundly disappointing. Drug did not separate from placebo. A huge investment, that's what they found. The drug was tolerated, but it didn't separate. We said, based on some recent work we've been doing using advanced computational models, we think we can find the subgroup of patients for which drug actually is better than placebo. Now the problem is, how do you do that? It's very difficult to do, because in a perfect world, what you'd want to do is clone the patients, and for every patient you assign to gabapentin, you would assign their exact behavioral and genetic clone to the placebo. We don't do that. What do we do? We randomize people. Now we could have a whole lecture on what are the problems with randomized clinical trials, but I can tell you having done a very deep look at this issue now, randomized trials of less than 5,000 people, maybe even less than 10,000 people, which are routinely done in cardiology, when there is a clinical trial in cardiology to determine if a drug, drug A is better than drug B for atrial fibrillation, the sample size is 40,000. It's not 400, okay? And what's the problem with doing a sample size of 300 people to compare drug and placebo? The problem is randomization does not guarantee equivalent groups, because the clinical and biological features which determine whether a person will or will not respond to a psychiatric drug are very complex, and with small samples of a few hundred people, you're not going to get equivalent groups in terms of likely responders. So we said we think we can do better than that, and so we reanalyzed this group, our paper, which I'm happy to send to you, is called A Likely Responder Reanalysis, and our goal was to use very sophisticated causal modeling to do the equivalent of the twin study. That is to say, our strategy was as follows. We very carefully studied the gabapentin group, and we found that their outcomes varied from very bad to very good. We carefully studied the placebo group. Their outcomes responded from very bad to very good in terms of the decrease in numbers of heavy drinking days over the course of the trial. That was the primary endpoint. Delta heavy drinking days. And on average, when you compared them, they were equivalent. But there's no reason to believe that the same people that are responding to drug are responding to placebo. They may be very, very different, and we think they were. So we studied that. It's a very complex method, but we start by looking at those in the gabapentin group who were likely responders, who did well on gabapentin. We look at a vector of their features, their clinical features, and other data we have on them, their levels of anxiety and depression before the trial, their levels of heavy drinking days before the trial, their impulsive decision-making behavior. We looked at close to 300 features we had, and we were able to develop a set of features which was associated with a good prognosis in the gabapentin group. We then went through, using machine learning, were able to identify a group in the placebo group with the same features. That is to say, they had the same, had they been assigned to gabapentin, they would have responded well. So we matched them, essentially, on a large number of features which were associated with a good response in the drug group. And then we used a complicated set of machine learning approaches to try to test to see if we could identify a subgroup of the gabapentin group that outperformed placebo, and a subgroup of the placebo group that outperformed gabapentin. And in the interest of time, I would like to say that we were able to do that. And what we found was very interesting, if I can jump to this, we were able to identify a group of likely responders and unlikely responders to gabapentin. So now, although drug did not separate from placebo, we were able to identify a group by clinical features of people likely to respond to gabapentin, and another group unlikely to respond, and they separated well. Actually, this is a test of our ability to predict them, which we did. And what did we find? We found that among alcohol use disorder patients, those who had lower levels of anxiety and depression before they were randomized to drug or placebo, had higher levels of cognitive and motor impulsivity, and had higher levels of alcohol consumption before treatment, responded much better to gabapentin than they did to placebo. So even though the trial failed, and even though gabapentin overall did not separate from placebo, for the patient in front of you, if they presented with low levels of anxiety and depression, so-called internalizing symptoms, high levels of behavioral and cognitive discontrol, anger, impulsivity, violence, impulsive decision-making, and very heavy drinking, they did much better on gabapentin. Perhaps not surprising, because gabapentin is an anti-invulsive, and probably very helpful for those with high levels of emotional and cognitive dysregulation. So that's just a proof of concept for you. On the basis of our findings, NIAAA was sufficiently excited about it that they awarded me a large R01 to try to see if we could really deeply understand this finding. And it turns out, in their wisdom, they drew blood on every subject before and after treatment. They drew blood for DNA and RNA before and after, DNA before treatment and RNA before and after treatment, and that blood was sitting in a biorepository in Bethesda at NIAAA, no one has ever looked at it, and now that we have a method to determine who will and will not respond to gabapentin, they awarded me a large R01 to take those samples to do very sophisticated, wall-to-wall genomic analysis on them, and see if, in addition to the clinical features, low anxiety and depression, high cognitive and motor impulsivity, and heavy drinking, can we find genetic, epigenetic, and transcriptomic evidence to better explain who will and will not respond. So we're actually analyzing that data right now. This is the study, we have samples before and after treatment. This is the overall plan for the study. In the interest of time, I'll just say we're ahead of schedule. We have ascertained all the DNA and RNA markers from these cases. We're working with, I think, the country's most sophisticated genomic laboratory, Adriana Hege at NYU. I'm sort of just somebody who stumbles into good luck. I stumbled into SRI, the best bioengineers in the world. I got a call from Adriana Hege one day saying, Charlie, I understand you're interested in DNA and RNA markers for your precision psychiatry studies in AOD and PTS. Yes, Adriana, I am. I can help you. Well, what can you do? Well, I'm able, in my lab now, for under $300 to sequence the entire human genome on every one of your subjects. Right, you remember how much money and how much time it took to sequence the first human genome? So we're going to have 22 million molecular features on every subject in this study, and we're going to figure out which ones make a difference and don't make a difference to explain the outcome. Okay, I'm gonna conclude. I wanna leave a few minutes for discussion just to say very briefly that NIAAA awarded me a very generous P01 grant. This is a $6 million center grant to see if we could do a really deep dive and really do precision medicine on AOD and PTSD. And you know what all the issues are. You know, of course, because you're interested in AOD, some of the incredibly brilliant work that George Kube has done to define the stages of alcohol use, including the hedonic binge intoxication phase, the anhedonic withdrawal negative phase, and then the self-medication phase to manage the negative withdrawal. And this is a more detailed model of the molecular features that George Kube defined from his animal and human studies, and we'll be studying them all in this P01. We're particularly interested in the balance between glutamate and GABA and excitatory and inhibitory signaling, which we think is very important for alcohol use disorder and is maybe how drugs like topiramate and gabapentin actually work for alcohol use disorder. The anticonvulsants, one of the things they probably do is restore a better balance of excitatory and inhibitory signaling, which is very disrupted by chronic heavy alcohol use. So topiramate, again, I don't want to go into all the rationale why we studied topiramate. It's not FDA-approved for alcohol use disorder, although I think it should be. And here's the design of our study. We're randomizing 150 people, two-to-one ratio of topiramate to placebo, but what's really important, and what I really want to tell you, ignore all these details for a minute, what I really want to tell you is on each one of these subjects, we are getting very detailed clinical evaluation, in-depth neurocognitive evaluation by expert neuropsychologists. We're getting 17 tubes of blood on each subject before and after treatment to do every genetic, epigenetic, RNA transfer, methylation mark, every transcript, every microRNA, every protein, and every metabolite, and we also get structural neuroimaging, resting-state fMRI, task-based fMRI, and TMS-guided EEG on these subjects, and we're going to try and integrate all this information to advance precision medicine. This is the design. And the GRIK1 gene is one isolated candidate, but we'll look at that. We're not too optimistic about it. But this is, I'd like you, if you remember one slide, I'd like you to remember this slide. This is a map for how to conduct precision medicine in psychiatry. You do a clinical trial, but you embed in the clinical trial very deep clinical assessments, very expert neurocognitive assessments, every molecular pathway that's known to be relevant to the psychiatric disorder, and every brain circuit that's known to be related to the psychiatric disorder, and then you use advanced computational models to try to see who does and does not respond to treatment. And that's what we'll be doing, and we'll use that analytic approach that I showed you a moment ago that identified the responders versus the placebo responders in the GABA-Penton study. Finally, I'll just leave you with the latest study we've got. I was very fortunate to be awarded a $8 million DOD Center Grant to try to identify unique molecular and circuit subtypes of PTSD, alone and in combination with AUD. And in this study, this is showing you the five cohorts, we will have 3,000 PTSD cases and 12,000 controls, and we will do wall-to-wall clinical cognitive molecular and where we have imaging markers. And the goal of the study will be to hand off to us as clinicians, for the first time, reliable subtypes of PTSD that differ clinically, cognitively, and in terms of their molecular and circuit features, and we think will differ in terms of their treatment. As an example, we already know that those with cognitive impairment don't do well in prolonged exposure therapy. But the question is, which ones do well in search line? Why search line? Why not benlofaxine? Why not lamotrigine? Why not ketamine? Why not psilocybin? We need to really, our field, we're practicing infectious disease without any way to determine the sensitivities and specificities of the antibiotics we're using. And the initial trials of this, to look at markers of metabolism, as you know, have been quite disappointing. It hasn't actually moved the field forward very well. Slow and fast metabolizers, it's an interesting idea, but it hasn't actually improved clinical practice very well. We really want to advance clinical practice so that when you assess, when we assess, as I do every day, a new referral for PTSD or AUD, I can say, this is a type three PTSD patient, and I don't want to start with psychotherapy. I want to start with TMS for this patient. That's our goal. So I think in the interest of time, I should say we pre-specified four, we believe using the so-called wisdom of the crowds that there are four types of PTSD. PTSD with impaired cognition, PTSD with mild symptoms, PTSD with severe symptoms, and PTSD with prominent depression symptoms. There may be a fifth PTSD with prominent dissociation symptoms, and we're studying them all. I'd like to stop at that point because I want to leave a few minutes for Q&A, but my goal today, and thank you, you all look well, and it looks like you survived the onslaught of this lecture, and don't share my residents' view of my inflicting PTSD. My other favorite comment was from a UCSF medical student, by the way, when I gave what I thought was a very elegant lecture on insomnia. I used to teach the course of Introduction to Psychiatry for 167 UCSF medical students when I was at UCSF before I was recruited to NYU, and one of my students who I knew well, and I knew he was going into orthopedic surgery, and I knew he cast a jaundiced eye over psychiatry, said at the end of the lecture, he stood up and said, Dr. Marmer, I want to thank you for a very scholarly and I think useful lecture on insomnia. As a future surgeon, my patients will have insomnia, and I appreciate it. However, I need to tell you, you failed to include one of the most important weapons in the battle on insomnia. I said, well, I included behavioral approaches, drugs, lifestyle, sleep hygiene, and what did I left out? He said, your lectures. So I'm glad that everyone is self-alert and awake. So thank you so much for your attention, and please come to the mic if you have questions so we can all hear. Thank you, you're very kind. Questions? On any of the things I talked about, or any things that interest you in AUD, PTSD, that I may not have touched on? Yeah, thank you, very, very interesting lecture, especially as someone who went the other direction, long-term addiction psychiatrist who ended up helping one of my colleagues at the University of Maryland do a PTSD study, pre-gablin for AUD and comorbid PTSD. After the session, I hope you have time to address some directed questions to you. But my question is, if one is not a clinical trial without all the detailed data that you have, could you somewhat accomplish your post-matching by using something simpler like propensity score matching to try to evaluate the difference between the responders, non-responders, placebo, and active groups? So for active duty, yes, it's very, very important. The work that we're funded for by the Department of Defense is actually directed at the active component more than at veterans. And fundamentally, what the Army has asked us to do is to find objective ways to screen for risk and readiness for deployment through the studies of the molecular circuit and other markers we're studying, not for the purpose of disqualifying people from serving or even being redeployed sometimes, but rather the Army knows that most of the people that they assess will not disclose their PTSD symptoms while they're active duty. They may disclose them when they separate, but not while they're active duty. I don't know if you've ever worked with any Marines, but one of my friends in the Marines told me, said, Dr. Marmor, do you know what the Marines call psychiatrists? Said, I have no idea. Said, I thought something very disparaging. He said, no, we call them wizards because they have the power to make you disappear. If you tell them that you have nightmares, flashbacks, and startle reactions, you will disappear. You'll never be redeployed again. They're very fearful of it. So they won't disclose their symptoms for the most part. So we are working to develop high throughput screening in primary care practice to be used to make informed decisions about deployment readiness and to provide people with the care they need so they can be sufficiently resilient to be deployed or redeployed, or in some cases, to be deployed to less intense roles such as combat support and service support roles. So that's what we're doing. That's what I've been doing for 20 years with the NYPD as well. Thank you. Yes. Hi, thank you so much for the talk. Pick up the mic so everyone can hear you. My question was, among people with PTSD, are there any biomarkers or any differences that you've been able to notice for patients with PTSD who might turn to alcohol versus those that might not? PTSD that are? PTSD that turns to alcohol compared to those that do not. Yes. So we will have the answer to that question from my P01 when we analyze it. And we will have a very clear idea of the relative contributions of PTSD and AUD to the biomarkers. We do think the biomarkers will be different. We think the biomarkers in PTSD will be more related to adrenergic functioning, to HPA functioning, to renin-angiotensin functioning, and to the Krebs cycle markers. In AUD, they're likely to be a somewhat different set of markers, probably more related to reward, circuitry, and molecular pathways, and more related to imbalance between excitatory and inhibitory amino acids. That is our hypothesis, but we will have the data in the next two years. So we'll be happy to come back and continue the conversation. Yes. Oh, it's fine. I'm sorry I missed the first part of your talk. So I have a feeling you covered it thoroughly, but I see in your list of subtypes, you don't have the adrenergic subtype as a subtype, at least in this group. Whereas it seems there is some research suggesting, especially in alcoholics, that those with a high blood pressure may be preferentially responsive to prazosin. A study just came out this year already supporting that. I think one of the great difficulties in our field is that among our more complex patients, which is the ones we all care for in our, the simpler patients with the simpler disorders are actually mostly managed in primary care. So if someone has a simple drinking problem and they don't have an underlying PTSD or bipolar disorder or cocaine abuse, we often don't ever see them. So our patients come wrapped up in complex comorbidities and we think that we're going to need to define the subtypes based on those comorbidities. I can say to you, and I'm not sure if you were here when I presented my initial blood test findings on PTSD, we were able to demonstrate that the molecular features which discriminated PTSD from case controls did so after controlling for level of depression. So they were not driven primarily by depression. Yes. Dr. Marma, I just wanted to ask, one of the points that you made about randomization and versus the likelihood to respond to gabapentin in that study, that you analyzed the study again and saw that you were able to separate placebo from responders, from the gabapentin group based on the more nuanced characteristics of those patients. How would you, like what conclusions can we make of this going forward in terms of trials that randomization, as you mentioned, is not? Sorry? That we can draw from the idea that randomization by itself is not necessarily sufficient for a small group of patients to be able to separate placebo from a treatment arm. Yeah, it's difficult. We really need to do, the work is at an early stage. The problems are very complex. I think we are, I think, one thing I didn't have a chance to talk about today is we're on the edge of a revolution in neuropsychiatry. It's a quiet revolution that's not being spoke about. And it goes back to my concern that we don't biopsy the organ of our illness. There have been recent developments to show that every cell in the human body releases vesicles. They're not cells, they're vesicles. The vesicles carry a genomic cargo of RNA, proteins, and metabolites, and they're used mostly for cell-to-cell signaling in every organ. They're particularly important in the brain. And up till now, it's been very, and they're, they also, the vesicles that are released by brain cells, neurons, glial cells, astroglial cells, and others, actually cross the blood-brain barrier. And they are in peripheral blood, but they do represent the molecular pathways that are actually occurring in specific, not only in specific in the brain, but they differ and can be differentiated by which type of brain cell. And in the most recent studies by Karl and Aska, my colleague, they're even able to say whether these exosomes came from the hippocampus or the amygdala or the basal ganglia. So we are going to repeat all of the studies I presented today using the molecular features from these neuronally-derived exosomes, which we can now isolate in blood and have a window into what's actually happening in brain molecular pathways. If this works as well as we hope it will, this will revolutionize our field and we'll be able to answer much more precisely all of the questions we've been asking because the blood is just too noisy. So we look forward to coming back. Yes, please. Come to the mic, though. I just have a follow-up. Like, what time frame do I need to take after a clinical application? This is a wonderful question. I'm very careful about answering that question because I was asked that question about 10 years ago by one of my sponsors who funded my work on a blood test for PTSD. I said, I thought, well, maybe we could do it if it all goes well in five years. And the answer from the sponsor was, I hope it all goes well. Because if it doesn't go well, you will be sleeping with the fishes in the East River. So I'm now very leery on answering these questions, but I would say to be fair that over the next five to 10 years, we can revolutionize our field through some combination of studying neuronal exosomes, induced pluripotent stem cells to try to replicate the most important findings, post-mortem studies, and peripheral blood studies, and triangulating all of them. So I am much more optimistic now than when I was asked that question 10 years ago. But it is difficult. Our field is complicated. Our phenotypes are more difficult than other medical disorders. And the heterogeneity is profound. And life experience, and genes, and culture, and health equity all play a role. And we have to sort through it all. But I am very optimistic about it. And I think five to 10 years from now, we'll all be practicing a form of precision medicine. Hi, thank you very much for a very delightful lecture. That's my pleasure, thank you. My question is more about the phenotype, and particularly the concept of trauma that has become so widespread. Often trauma, and at a very primary level, is confused with PTSD. People self-diagnose themselves PTSD, their counselors and therapists often say that you have PTSD. Is there any scope of refinement in the definition of post-traumatic stress disorder? Already DSM-5 TR, or DSM-5 had done a good job of work on that. But then when you're going for biomarkers, and all those blood tests, should the effort not be made to make the definitions a bit more stringent? Yeah, it's a very complex question. It's a very astute question, so thank you. A few quick comments. In primary care practice in developed countries, like America, the most prevalent psychiatric illness is major depressive disorder. In primary care practice in developing countries, the most prevalent psychiatric diagnosis is not major depressive disorder, it's post-traumatic stress disorder. So in sub-Saharan Africa, in East Africa, in other developing regions of the world, in the few studies that have been done of careful assessment of psychiatric illness in primary care, the number one diagnosis is PTSD, and it's because of domestic violence, war and conflict, massive sexual violence and other gender-based violence towards women, and exposure to disaster, and exposure to death and dying through HIV-AIDS, pandemics, other infectious disease. So the burden of traumatic exposure in developing countries is enormous. Now, that doesn't mean that everyone who's exposed develop PTSD, and it's why, in the studies I presented to you the healthy controls we chose were all heavily exposed to trauma. So that in all the studies I presented to you on the blood biomarkers, the voice biomarkers, and the other studies, the healthy controls had similar levels of trauma exposure to the cases. So we do have ways. Now, as regards how should we diagnose PTSD, I was one of the people who contributed to the DSM diagnosis for PTSD earlier in my career. I think, in some ways, the ICD classification may be more robust than the DSM, and the reason I say that is the six symptoms of PTSD in the ICD classification are those that, when I did a historical review of the description of PTSD from the first case report that we could find till now, and by the way, the first published case report on PTSD is from 3,400 years ago from the kingdom of Assyria. It was a single case study of a war fighter who presented with nightmares and flashbacks after battle experiences, and the case was written by the physician to the court of the kingdom of Assyria. The single best description of PTSD, if you want a near-perfect definition, was actually by Shakespeare. It's in Henry IV, Part I, in which Lady Percy describes her husband, the king's, combat-related PTSD. It's a relatively short soliloquy. I recommend it to all of you. It is a perfect description of PTSD, and beyond that, a very deep description of how PTSD affects the partner of someone living with PTSD. It is incredibly brilliant, and if you want the real symptoms of PTSD, go to Shakespeare. Had I read it before I served on the DSM committee, I would have resigned from the committee and submitted the Henry IV quote. It's three o'clock, dear friends, thank you.

precision medicine

psychiatric disorders

Dr. Charlie Marmer

NYU Grossman School of Medicine

subjective self-reports

objective diagnostic tools

customized treatments

genomic tools

neurocognitive tools

neurocircuitry tools

PTSD

machine learning

biomarkers