And hopefully what I'm going to do, and my goal is here, is to fill in, to give some, you know, great insight into some of the amazing basic and translational science work that's going on in this field. And what I'm going to do here is additionally try to wrap that together, put it in some clinical context. It'll be complementary to if you saw Dr. Vanessa Nomolini's presentation on immunophenotyping yesterday. I hope you had a chance to catch that. How do we wrap this all together? There's lots of scary terms we've been talking about. Biomarkers, phenotypes, endotypes, MDSCs. It's kind of scary. Let's see if we can find our way down the yellow brick road here. So my disclosures are I am named on a U.S. patent that's held by a company called IFDX for whole blood ELLISBOT technique for diagnostic approaches to clinical immune endotyping. I have no equity position in the company at all. But the biggest disclosure I think is that this review is just really the tip of the iceberg, both in content and contributors. In 15 minutes I can't delve completely into this topic. So I'm hoping I'm going to show you some selected examples of some of the rigorous and robust and most interesting examples of this and hopefully kind of put all this together. So we talk a lot about the promise of precision medicine, but what is that? So the powers that be, the NIH and the FDA, it's an innovative approach that takes into account individual differences in patients' genes, environments, and lifestyles, and it's about matching the right drugs or treatments to the right people based on genetic or molecular understandings of their disease. And so, you know, what's the rationale for this? Well, I think we're starting to realize that, you know, the traditional gold standard of evidence-based medicine, randomized controlled trials, the results are based on the average treatment effect across the population and do not apply to all patients equally. And so what precision medicine ideally will do, will target the correct treatment to the appropriate patient at the optimal time. And so let's kind of put that in context. So let's think about as clinical care, you know, critical care practitioners, what's the most, you know, practice-changing study of the last five years? And so let's talk about the recovery trial and the implementation that really changed practice for a lot of us over the last five years in just incredibly rapid fashion. So the recovery trial testing use of dexamethasone in hospitalized patients with COVID-19 requiring oxygen, which I understand is a subgroup analysis, but it is consistent with the current NIH COVID-19 treatment guidelines of how we should be treating patients, reduced mortality for one in every 34 patients. Now a reasonable number needed to treat is thought to be somewhere between five to 10. Clearly we can do better. So how does that apply to sepsis? Well, I think we've kind of been through this and we've seen this. If you look at the sepsis wars of the last, you know, 90s and 2000s, we've been trying to target some of this amazing mechanistic work that has been done over the last 30 years is just littered with immunotherapy casualties. You pick your favorite, APC, anti-TNF, steroids, anti-IL-1. I mean, none of this has been shown when translated to randomized clinical trials to have a meaningful effect on outcomes after sepsis. So why are there no immunotherapy silver bullets in sepsis? We know so much. There's so much incredible basic translational work being done. What is the answer? Why can't we translate this? Well, maybe none of the drugs are effective, but they work in preclinical and early phase one, two trials. The mechanisms seem clear. Why aren't they working? Maybe preclinical animal models don't replicate the complexity of human disease is extremely controversial in the research world right now, but clearly research men are inbred clones and humans are not, right? And drugs effective in preclinical animal studies may not be, may be effective in only some human subpopulations of this very broad disease, which we have with sepsis. And clinical trial outcomes currently may not detect or reflect the complexity of disease and trying to focus on short-term mortality, you know, either inpatient with the first seven days, you know, it makes it nearly impossible to reach these treatment goals because maybe we need to be looking at organ dysfunction and longer term survival. So the future of sepsis therapeutic trials is going to be prospectively identifying the patients who both require intervention and are likely to benefit from a specific intervention. So the goals that are going to have to be, we need to be able to identify sepsis patients at risk of developing severe complications, persistent organ failure, long-term death. And among these at risk, identify individual patients, maybe by some kind of biomarker or phenotype or endotype, who would benefit from a specific targeted therapy. So what are the challenges to this? Well, the interventions have to be rapid. This isn't cancer. We can't genotype tumors and act on it maybe in the next four weeks. You know, we're on a time frame of minutes to hours, you know, at best, maybe a day. So the efficacy of current prognostic biomarkers, and I'll come back to that in a second, what prognostic biomarker means in sepsis and how it can guide interventions is right now remains completely unproven. And importantly, biomarkers in therapies must use diagnostics that are developed under CLIA oversight and FDA cleared. You can't just pull stuff out of the lab and say we're going to start using this on patients to guide trials. So is there a role for biomarkers in phenotypes and sepsis interventional trials? So it's really important to understand the prognostic and predictive biomarkers. So prognostic biomarkers are what you see most of in the literature, right? They're identified from observational data, used to identify patients more likely to have a particular, usually adverse outcome. I kind of like to say these are a dime a dozen, we see them all the time, whether it's a physiologic parameters like Apache or ISS in trauma, maybe something specific like an IL-6 concentration. But the reality is that these prognosticate a patient's clinical outcome. A predictive marker is very different. It predicts a patient's response to a targeted therapy. And in sepsis, there are very few, arguably no true predictive biomarkers, at least that we can utilize right now. So sepsis biomarkers. What is a biomarker? And I'm not a person who likes to get lost in definitions and kind of minutiae of words, but unfortunately in this realm, definitions are important. So what is a biomarker? It's a characteristic that is objectively measured and evaluated as an indicator of normal biological processes or pharmacological responses to a therapeutic intervention. So what are some commonly used available biomarkers for diagnosis and prognostication in sepsis? Many of you may use these. These are some of the most common ones. CRP is not very specific. I don't find it to be very helpful. Most of us use lactate as a surrogate marker for organ perfusion in patients who are in more shock. Maybe the poster child of biomarkers is glactamate or beta-D-glucan, very specific and actually might change your management right there at the bedside. Procalcitonin remains pretty controversial, I would say. Many people may use it for shortening duration of antibiotic therapy or suggesting when not to start antimicrobials, but the reality of it is, click, that overall evidence that single readily available clinical biomarkers, if they significantly affect care and improve outcomes in sepsis, the evidence for that is low. Now there's other ones we are not using clinically yet. There are more uncommonly user-available biomarkers for diagnostic prognostication in sepsis, receptin, neutrophil parameters, CD64 expression or cell-free DNA or microRNAs. The reality is most of these are not currently available or practical to guide clinical care. Now they might have some increase in sensitivity, specificity for diagnosis and outcome prognostication, but it's not clear whether or not these are going to be valid across heterogeneous sepsis populations. And the actual impact on clinical care that you're going to have at the bedside for a treatment target or patient selection, again, remains unproven. So the practicality, efficacy, and cost-effectiveness of individual biomarkers to significantly modify clinical care and improve outcomes in sepsis remains questionable. So what are sepsis phenotypes? I'm sure you're starting to read about this now. And do they matter? Well, I think most people would agree that the major barrier to progress in sepsis is the heterogeneity of the disease. It's a very broad definition of a clinical syndrome that ignores significant heterogeneity of the disease at the level of the individual patient that you're taking care of at the bedside. And so the arguable strategy is, well, can we identify disease phenotypes to prognosticate outcomes and guide clinical care? And so, well, we talked about, you know, I don't want to get lost in definitions, but I had a very frustrating time trying to actually pin down consensus definitions for the terms I'm using in this talk, so let's talk about it a little bit. So, you know, when you don't know where to go or you're looking for consensus definitions, where do you go? Let's start with Wikipedia. Phenotype, a single or combination of observable characteristics or disease traits without any implication of a mechanism. No citation needed. So how does that apply to within sepsis? Again, this was even more difficult to pin down a definition on. So what do we do when we're really confused in the modern age? Well, we ask Dr. AI. So what does Dr. AI say? Well, the term sepsis phenotype refers to distinct subgroups of sepsis patients characterized by specific clinical, biological, and or molecular features and are identified through the analysis of patient data, including clinical variables, lab results, and genomic data, and are associated with different outcomes and responses to treatment. Okay, so hopefully we have some kind of generalized understanding about what we're talking about when we're talking about phenotypes. So what about sepsis phenotypes? And I think the most robust example to give is the work by Dr. Seymour, Yenay, Angus, and the Pitt Group regarding their alpha through delta phenotypes. And so what they did is they looked and identified four phenotypes using 24 EMR clinical variables, clinically available variables, with 27 clinically available serum biomarkers that carry a subrepresentation of inflammation and organ dysfunction. Again, these are clinically available information. And what they found is after, you know, analyzing these and validating them in multiple, across multiple cohorts and trials, which I'll talk about in a second, you know, they found that there were within group similarities in circulating biomarkers and organ dysfunction between these groups of patients with different mortality risk across the phenotype. And the examples would be is that the alpha phenotype was the most common with the lowest level organ dysfunction and vasopressor use. The delta phenotype was the severe phenotype with a high degree of shock and high mortality and liver dysfunction. So each phenotype has a different pattern of organ, a different clinical pattern and presentation. I think the really interesting thing about this paper, about how we're going to use this, is that they were able to, you know, show some evidence that phenotypes exhibit different treatment effects when applied across prior clinical trials. So they did some fancy Monte Carlo simulation to estimate the differential estimated treacle effect on previous trial populations. And just as two examples, so in the access trial, toll-like receptor antagonists that, you know, basically found on the far left, no signal at all of any effect. But actually, when you look at the higher severity phenotypes, that there was evidence of not insignificant amount of increased harm. Or if you look at early goal-directed therapy in the process trial, they showed actually, even though there was no signal, maybe even harm across the entire population, but within the alpha phenotype, that there might be evidence of significantly more benefit. But in the severe phenotype, more severe phenotype combinations, that there is evidence for significantly increased risk of harm. So what does this mean? Well, phenotypic enrichment might help enhance treatment effect in future sepsis clinical randomized controlled and adaptive trials. Now we can even get a little bit deeper. People start talking about subphenotypes. This is work from one of my colleagues at the University of Washington, Pavan Bhattaraju, looking at organ-specific subphenotypes. These are kind of phenotypes within phenotypes. And so he's been able to show that there are two subphenotypes in sepsis-associated AKI. When you look at, he did a latent class analysis across a discovery and validation core, similar to the Seymour work. And using combination of clinical variables and some more research-oriented biomarkers under the law of underlying biology. But he was able to show that in these two phenotypes, the SP2, the subphenotype 2 population of AKI, that there's increased risk of 7-day renal non-recovery in these patients and 28-day mortality. But importantly, when you look at and incorporate this into previous trials within the VAST trial, the SP2 phenotype showed, even though it was a higher mortality cohort, that there is decreased mortality in the patients who utilized vasopressin as opposed to norepinephrine. And very recently, he's applied this to the Clover's trial. And again, even though the SP2 cohort has significantly higher mortality, that the restrictive fluid strategy within these patients significantly was associated with reduction of mortality. So subphenotype enrichment may help enhance treatment effect in future, again, clinical trials in organ-specific treatments and outcomes. So the summary of sepsis phenotypes, the strengths, well, they can broadly address hopefully large heterogeneity, at least start to address the difficult-to-define sepsis populations, which is the world we live in. They're practical, based on common and readily available clinical parameters, and they have the potential to significantly leverage EMR data because of this. There's an enhanced ability to link patient correlations of their characteristics to outcomes, and it may help enrich enrollment populations for future trials, as I mentioned before. What are the weaknesses? There's still significant heterogeneity within this phenotypic homogeneity. They're still there. And importantly, there's not necessarily any relationship or insight into the underlying biological disease process of the individual patient. So whether or not this is able to guide therapy for an individual patient remains unclear. So what are sepsis endotypes, and do they matter? Again, dive into the literature to try to come up with a consensus definition of this. Good luck. So again, we go back to Dr. A.I. And what do they tell us? An endotype is a subtype of a condition which is defined by a distinct functional or pathological mechanism. Okay. Sounds reasonable. So what about this in sepsis? And this is most commonly, but not exclusively, in the context of genomics in sepsis. I think the early work and kind of early development in this field, we really need to give much credit and recognition to the late Dr. Hector Wong in pediatric sepsis endotypes looking at whole genome analyses across a large amount of gene expressions, narrow it down to 100 gene signature, which was able to subclass three classes of pediatric sepsis. And the subclass A was the highest mortality. But importantly, there's some evidence and linkage to underlying biological processes here that we can see, including defects in adaptive immunity and aberrations in glucocorticoid and zinc pathways, which might lead us to a direction of how could we mechanistically target these individual patients. And I think in adult sepsis, I think the best example is probably the MARS endotypes. Again, gene-wide expression analysis across multiple validation cohorts, and they stratified it to four endotypes based on 140 gene signature. Again, differences in the different endotypes that not only had a difference in mortality, but also had some insight into the actual underlying biology, with the most highest mortality phenotype showing extreme repression of innate and adaptive immune pathways, whereas the higher surviving cohort showed more evidence of expression of adaptive immune functioning and signaling. So a competent immune system. So, what I would argue is that effective precision medicine approaches will require the recognition of disease endotypes. And again, this goes down to, I don't want to get lost in semantics, but this is important to understand the difference between what is a phenotype and what is an endotype. A phenotype is what you see with your eyes, if you want to think about it that way. It's a group of patients defined by clinical signs, symptoms, or common laboratory valuables at your fingertips. And this is what most commonly drives clinical management at the bedside. This is what you do as a clinician. You actually see this with your eyes. And it does not necessarily relate to, unfortunately, or give insight into the mechanisms of what's going on in that individual patient. So an endotype is what lies beneath. It's a group of patients defined by the underlying biologic mechanisms. And they may look the same to your eyes. And they reflect different host biology between individuals with the same clinical, quote unquote, disease. These two patients are septic. They look the same. And importantly, this is probably what dictates the response of an individual to a mechanistically targeted therapy, i.e., the battlefield of immunotherapy casualties that I showed you at the beginning of the talk. We've done some work on this in severe injury and trauma, looking at both the, not only the level of individual immune proteomic markers, but also the trajectory over time to show these different mechanisms, you know, three different subgroups of these types of mechanisms in severely injured trauma patients. So what tools do we have right now to determine endotypes? It sounds like a good concept. We have a lot of fancy, very accurate tools to use. Large scale proteomics, metabolomics, flow cytometry, we're approaching the capability to mark up to 40 to 50 markers on individual cells nowadays, so very deep cellular immunophenotyping. I mentioned genomics and transcriptomics. There's also functional assays such as microfluidics. These are really powerful tools. And do these tools allow us to get even deeper mechanistic insight into sepsis? Absolutely. Phil has shown us that you can look at actually the, at a single cell level at an individual patient and determine individual defects in mechanisms, including patients, two sepsis patients having immunosuppressive transcriptomic profile in their early immune and immature lineage of MDSCs, but also severe lymphopenia and immunosuppressive transcriptome across multiple T cell subsets. So really stratifying these patients at the individual cell level. This sounds promising. It sounds kind of exciting. At least I think it sounds kind of exciting. But is it practical? What's the turnaround time? And the reality is that most of these standard research analyses that we do, I mean, they're at best days, probably weeks before we can get some information from that, and that's not really going to work in sepsis. So we need clinically relevant technical platforms to move this work forward. And so we've started to take those first steps towards real time point of care diagnostic genomics. We did that, and we looked at and attempted to validate a trauma metric that predicts complicated outcome, severe or persistent organ dysfunction, and or early death that was initially validated in the trauma glue grant, or that was initially designed in the trauma glue grant project back in the thousands. And we used a clinically available platform, the Nanostring device, which is 510K cleared for clinical diagnostic use and can utilize any open set of gene sets and algorithms that you have. And importantly, we were able to develop an 18 to 24-hour laboratory drive protocol to execute this study from patients that, blood that was drawn from serially injured patients within 24 hours. And we were able to validate in a point of care type fashion that this could be highly prognostic for the development of persistent organ dysfunction and or death of a trauma as compared to something, you know, readily clinically available things like an injury severity score, an Apache score, or even something like IL-6 as a potential biomarker. So what about genomic diagnostics in sepsis? This is a bigger market share, obviously, so industry has definitely entered the game. And the interesting thing is that the gene sequences and sets are in the public domain, but the device technology and algorithms can be proprietary. I think that's why industry has kind of gotten on the bandwagon here. And there has been, you know, really what's moved this forward is federally funded industry and academic collaborations to accelerate advancements in technology. An example of that was work that we did at the University of Florida in collaboration with a company called Inflamatics, funded by the BARDA mechanism, where we were able to validate in real time a transcriptomic severity metric that predicted it was able to differentiate both bacterial and viral sepsis as well as prognosticate 28-day outcomes. And importantly, that the timeframe for diagnostic genomic results is rapidly condensing because of point of care genomic technology. We're approaching the point, actually, where we're going from days in the next year there will be devices that can do this at the bedside within 15 to 30 minutes. Are there other practical and available methodologies to assist in determining the endotype of the individual shock patient? The Spies Consortium, which is Phil and Link Moldau at the University of Florida, Richard Hotchkiss at Wash U and Chip Caldwell at the University of Cincinnati, myself, Ken Remy and Monty Mazur at Case Western, and then Vlad Milinovich and Tom Griffiths who are basic scientists at Iowa and Minnesota. And we've been looking at other alternatives to try to, you know, clinically relevant ways to endotype an individual shock patient. One of the focuses has been use of ELISPOT. This technology has been around for, you know, not a short amount of time. It's actually really, in essence, a single cell ELISA. And what we've developed a way to take a whole blood sample, plate it, stimulate it either with CD3 or CD28 or LPS, and then measure interferon gamma and TNF alpha production at a single cell level for an assessment of both the cellular response to sepsis as well as cellular function by the amount of interferon gamma or TNF that's being produced. And importantly, this facilitates rapid testing ex vivo of responses to treatments. And you know, we've been able to show that this could be used as a functional readout for immunosuppressive endotype. During the pandemic, we kind of tried to turn lemons into lemonade and started looking at this in COVID and bacterial sepsis, looking at it together, and we were able to show using ELISPOT immunophenotyping that COVID patients and sepsis patients have evidence of severe and progressive adaptive immunosuppression, which when you ex vivo appear to be rescued by L7. So can we use techniques like this at the level of the individual patient to see if in the mechanistic therapy can actually work in that patient because we understand their endotype. All right. So that's our summary of sepsis endotypes. They give us a deeper understanding of the biology of sepsis at the level of the individual patient, their enhanced diagnostic and prognostic potential. They may be able to enhance enrollment in clinical trials mechanistically, targeted therapies importantly, and truly potentiate a precision medicine approach to sepsis. The weaknesses, you know, there are still currently utilized research methods that aren't widely available or cost effective for clinical use right now. Some might argue the science is not quite ready for prime time. And we do need to take caution on how small we're slicing this biologic pie versus whether or not it can be clinically practical in the context. So the final takeaway, if anybody who's really been kind of intellectually stimulated by this talk, this is a great, I think, kind of review article by a bunch of the experts in the field kind of just describing where we're at with sepsis subclasses. I think the ideal case in sepsis subclassification is integration across multiple data types. It's going to take all these approaches and in a combination and probably a lot of high end computation to be able to incorporate this data that is suggestive of a plausible biologic mechanism, is prognostic for clinically relevant outcomes, and importantly predictive of a response of an individual to treatment. Currently available biomarkers are likely inadequate to meaningfully impact sepsis care right now. Phenotypic enrichment may help enhance treatment effect in future clinical trials. And the ability to determine the endotype of individual patients, in my opinion, is likely necessary to successfully implement novel mechanistic based therapies and a true precision approach, medicine approach, to sepsis. Thank you very much and I guess we'll open up to questions.

precision medicine

sepsis treatment

biomarkers

phenotypes

endotypes

genomic tools