Hello and welcome. My name is Bob Dixon and I'm speaking on the topic of latest in microbiome research and plotting microbiome research in the critical care environment. By way of backgrounds and credentials, I am a faculty member at the University of Michigan in the Division of Pulmonary and Critical Care Medicine, also in the Department of Microbiology and Immunology. I am a practicing intensivist and I care for patients in our medical ICU. I'm also a Deputy Director of the Max Harry Weill Institute for Critical Care Research and Innovation. I'm tasked with talking about the important topic of where the microbiome field is headed as regards to critical illness. In order to do that responsibly, I think we first need to talk about where we are, then we can talk about where the field is headed. So it's in the distant future, over the horizon, 10 years or more from now, in the not-too-distant future, the next 5 to 10 years, and what I anticipate is coming in the next few years. So from the top, where are we? What can we say with confidence about the microbiome's role in critical illness in 2024? First off, it's really not new, so it's an old familiar friend to critical care research. In the 1800s, Eli Metchnikoff speculated that the 2 to 3 pounds of bacteria that reside in the lower GI tract are probably important, both in health and sickness, and when they escape from their compartmentalization in the lower GI tract, they probably wreak havoc on the immune response and organ failure. He wasn't wrong. In the 1950s, Jacob Hein and his colleagues in Boston did important work that showed even in sterile model shock, like hemorrhagic shock with resuscitation, modulating the gut microbiome has consequences on endorhine dysfunction and mortality. He showed that in models of hemorrhage and resuscitation, if you pre-treat the gut with antibiotics, you can improve outcomes for what otherwise would be a lethal model. In the 1980s, there was quite a bit of enthusiasm for the gut microbiome. They didn't call it at the time, but John Marshall famously said, the guts need motor multi-organ failure, and of course, that launched decades of study into selective decontamination of the digestive tract, which is commonly presented as a pneumonia prophylaxis strategy, but it's really a microbiome-targeted investigation or intervention. It's selective for a region. It is quite deliberately designed to protect the anaerobic commensals in the lower gut and suppress the outgrowth of gram-negative rods. It's probably the most thoroughly studied intervention in all of critical care medicine, and JAMA meta-analysis from just over a year ago showed a 99-plus percent probability that it has a mortality advantage. So there's a sense in which the microbiome is not only not new, it's the most thoroughly studied and proven biologic entity in the ICU. The gut microbiome, the myocardic microbiome in general, is a major source of biologic heterogeneity in our patients. This is a buzzword and a really important topic in critical care research in our current era. We know that these diseases, like sepsis and ARDS and multi-organ failure, are really not one thing. They're actually heterogeneous entities, clinical entities, that reflect heterogeneous underlying biology. And unfortunately, most of the phenotyping and sub-phenotyping that's been done has overlooked what I think is the largest source of biologic heterogeneity in our patients. To give you a sense of how heterogeneous the microbiome is, this is data from our group and some collaborations with Bob Woods here at the University of Michigan. If we take rectal swabs acquired from critically ill patients in our ICU and we measure two things. In panel A, there's the density of bacteria, so basically how many bacterial gene copies we find per unit per milliliter of specimen. You see tremendous variation, 10 to the 5th, 10 to the 6th, order of magnitude variation across patients in bacterial density in the lower cut. And we can show you in this reference, it's not all attributable to contamination or variation in sampling. There really isn't this much difference across our patients. It's correlated with comorbidities, it's predicted clinical outcomes, and it's importantly influenced by our clinical interventions like antibiotics. On the right, panel B, what you see is not just how much is there, but who's there. So if we compare 120 patients, the identity of the bacteria we find in the lower gut, every dot you see is an intersection of two patients. There's tremendous variation. You have some patients that have more than 20, 25 shared bacterial families, and you have some that have literally no economic overlap at all. So one way to think about this is any two patients in your ICU are 99.9% identical in the host genome, but they may have literally no genomic overlap when it comes to the microbiome. So if you're looking for biologic heterogeneity, it's hard to find something that's more impressive than the microbiome. I think this point is so important that I've had ties and scarves made, I gave it out to my lab. I'm not wearing it today. The microbiome matters both in the gut and the lungs. I suspect most of the discussion during this important deep dive will be related to the gut microbiome, but the lung has its own microbiome. I'm a pulmonologist. I like to spend my time thinking above the diaphragm. Of course, the microbiome work drags me below. So part of my job is to drag the field above the diaphragm and think about respiratory microbiota. We've come a long way in the last 20 years. When I was a med student, our summary understanding of the lung microbiome was that the normal lung is free from bacteria, and of course, that's changed profoundly with the advent of sequencing-based ways of characterizing bacterial communities in respiratory specimens. The lung microbiome field is booming. There's now more than 500 publications per year, and it is a topic for another day. But suffice to say, it does matter clinically and biologically among critically ill patients, especially those with respiratory failure. Microbiome acts on the host by way of multiple mechanisms. Asking the question, how does the microbiome mediate end-organ dysfunction or function in the ICU is simplistic. That would be like asking, how does a kidney contribute to homeostasis or disruption of homeostasis in critically ill patients? More than one way. The way I think of it in terms of how the gut contributes to lung injury or just take the lung as emblematic of other organs, end-organs that fail among our critically ill patients. Perhaps the most familiar way is colonization resistance against potential pathogens. So this is old news. We know that if you blast the microbiome with raw spectrum antibiotics, the patient is at increased risk for C. discs and enterococcus and other nosocomial or secondary opportunistic pathogens that can exploit that lung's niche and overtake the gut, disseminate, and cause secondary infections. So there's nothing new there. We also know that in critically ill patients with shock due to sepsis or other forms of hypoperfusion, the gut wall gets porous and leaky. And we find evidence of direct translocation of gut bacteria from the gut to end-organs like the lungs. This was speculated for decades. Jacob Klein in the 1950s speculated it. In the 80s, it was thought to be debunked based on culture-dependent studies that found minimal culture-identified evidence of translocation. But of course, in the contemporary sequencing-based era, we know that that's a pretty insensitive way of assessing whether gut bacteria are present in a specimen. My lab, coming up on a decade ago, showed a nature of microbiology that if you look at humans and animal models of sepsis and acute respiratory distress syndrome, we find that the gut microbiome is disorderly. We find evidence of gut bacteria translocating to end-organs like the lungs. And the degree of translocation is correlated with disease severity and end-organ inflammation and injury. Subsequent to that, other groups, like Carolyn Kalfi's group at UCSF, have looked at respiratory specimens from critically ill patients. They found that respiratory microbiota are altered in patients that go on to develop organ failure, ARDS in this case, in a very specific way. So specifically, if you look at all the bacteria that discriminate patients that do and don't develop ARDS in this patient of trauma, in this total trauma patient, they're most differentiated by the presence of gut bacteria, specifically Enterobacteriaceae, in the respiratory tract of ARDS. In a collaboration with Louis Voss from the Netherlands, our group looked at many BAL specimens from mechanically ventilated patients, similarly found bacteria in the respiratory tract differentiating patients with and without ARDS, and remarkably found the exact same taxonomic correlates. So the same OTU, the same Enterobacteriaceae taxonomic group was associated with ARDS and adverse clinical outcomes in both cohorts. So again, indirect evidence that the translocation of gut bacteria to the respiratory tract is an index of disease severity and predictive of bad outcomes. Some unpublished work I can share. So at our Weill Institute, we're spoiled by having a large animal lab that lets us ask questions that you couldn't feasibly do either in animals or ethically do in human subjects. We have a model of sepsis. This is an E. coli tylenephritis model, and the animals develop clinical, physiological, biochemical evidence of septic shock. And one thing we can do with these large animal models is actually measure what goes into the lungs and what comes out. So just to review basic anatomy, we can sample the pulmonary artery, which of course takes the entire cardiac output and puts it through the lungs, and we can sample the carotid artery, which is on the arterial side, the left side of the heart, and ask, is there a step down or a step up in a biologic measurement? In this case, bacterial DNA. So we can ask if anything is being consumed or generated by the lungs. And when we do that with bacterial DNA, we find that quite consistently, there is more bacterial DNA going into the lungs on the pulmonary artery side than there is coming out on the arterial side. So evidence that the lungs indeed filtering, probably other end organs as well, filtering bacterial DNA and bacterial products that are translocated from the gut to the blood. But this leaves two other key mechanisms by which the gut microbiome forms end organ function and dysfunction in critical illness. The first is immunologic. We know, arguably, the gut microbiome is the largest immune organ in the body, and it plays a crucial role, not just in immune development, but immune calibration dynamically over the course of life. And finally, as an endocrine organ, or as a metabolic organ, we know that, again, you have two to three pounds of bacteria in your lower gut. Taking in the carbon that we feed it, breaking it down, turning it into small molecules of short-chain fatty acids that get into the bloodstream and have important end organ effects in terms of homeostasis. So we think the immunologic and metabolic function of the gut microbiome needs to be considered in understanding how it contributes to health, homeostasis, and disruption in critical illness. The microbiome is predictive of clinical outcomes, and this is now a really consistent and robust signature across studies. To give you an example, our collaboration with Louis Boss, I mentioned before, if we look at lung microbiota in mechanically ventilated patients, we find that the burden of bacterial DNA, so the more bacterial DNA we find in as many VAL specimens, the worse patients do in terms of ventilator free days, being alive and off the ventilator at 28 days. This is true even if we exclude or control for patients with any clinical suspicion for pneumonia. It's at some sub-pneumonia concentration of bacteria in the respiratory tract we find important associations with alopecia or inflammation, and prediction, prognostic indices that tell us who's going to be alive and off the ventilator in four weeks. This is true of the gut as well, so this is an important paper, I think, from the University of Chicago group that found that among critically ill patients with COVID, not just gut bacteria but gut bacterial metabolites, they derived a signature of essentially what we expect to find as metabolites generated by a robust, healthy, largely anaerobic lower gut microbiome, and found that you can derive an index from this that did a decent job discriminating survivors and non-survivors, even when controlling for other factors, like for diabetes and illness. So we think that it's not just the presence of the bugs, it's their function, it's what they're making, and what those metabolites are doing locally and systemically in those. We know the microbiome contributes to previously described sub-phenotypes. As I mentioned, a common refrain in critical care research in our era is, let's break these messy syndromes down into biologic sub-phenotypes. One thing our group has done is explored previously described sub-phenotypes of critical illness and asked, do they differ in their gut or respiratory microbiota? Here's one example that we published earlier this year, last year in the Blue Journal. So this is a paper from Siva Bhavani and Matt Chirpett, done several years ago, and they're at Chicago. They're looking at sub-phenotypes of sepsis patients using only temperature trajectories. So if you look at thousands of patients with sepsis, and you just follow their fever curve, so what happens to their temperature curve over their ICU stay, they consistently, across cohorts, sort themselves out into four phenotypes. There's the hyperthermic slow resolvers, the ones that come in hot and take a long time to defer best. The hyperthermic fast resolvers, the ones that come in hot and defer best quickly. The normothermic and the hypothermic, and what they've shown over and over is that these are consistent sub-phenotypes that differ clinically, demographically, and prognostically. Typically, the hypothermic patients do worse. They also differ in their cytokines and maturity, and that's a lot in play. But we don't know what's driving this variation. What is the actual explanation for why some septic patients fall into one sub-phenotype or the other? We have hypothesized in our group, and this is work led by Cale Bongers, one of my trainees, that maybe the gut microbiome has something to do with it. As I mentioned, it's a really important and under-considered source of variation in immune response and metabolism, and what is temperature other than a sort of integrated index of your immune response and your metabolism, your metabolic state. So, the first thing we did was we looked at our own population of septic patients at the University of Michigan and asked, do they cluster the same way? And they did. We found the same four sub-phenotypes of temperature trajectories, the hot stay hot, hot defer best, normothermic, and hypothermic. And they looked similar demographically to the previously described sub-phenotypes. We have rectal swabs on these patients, so we asked, do these sub-phenotypes, these temperature trajectories, differ in their own gut microbiota sampled towards the beginning of their ICU stay? And they do, quite consistently. It was a relatively small number of taxa that explained most of the variation across these sub-phenotypes, and they were quite consistent. It was the formicides phylum, so the dominant gram-positive phylum that we find in many gut microbiota. And specifically, I'll call your attention to everything in right here is Lachnosporaceae. So, Lachnosporaceae is a really important bacterial family. It's a very potent producer of short-chain fatty acids. It is depleted by anaerobic antibiotics, and it explained most of the temperature variation in humans. So, we asked if we could model this in mice, and we could. So, this first panel A is showing that if we model sepsis in blackthip mice, these are conventional STF mice that you just get from Charles Rivera Jackson. If you give them endotoxin, they get hypothermic. That's what we're seeing here, conventional LPS, and they normalize after about three days. If you use germ-free mice, so mice that have no microbiome at all, have never been exposed to a living viable microorganism, they have a very blunted temperature response if you give them the same septic insults. The same dose of endotoxin, and they barely budge their temperature, and they normalize quickly. If you look at the variation across arms, so here in panel B, we're talking about conventional mice who have a microbiome, but we get them from different shipments and different vendors, so they're quite similar from each other. You see that there's tremendous variation. Some mice drop their temperature dramatically by up to 10 degrees Celsius, and some barely drop their temperature at all, whereas germ-free mice are quite uniform, suggesting that in this experimental model, the microbiome is the main source of the biological variation that explains differences in temperature. And as with our human cohort, there was a relatively small number of bugs that explained all the variation in temperature, and specifically, it was the same ones, so that Lachnos graciae again, so this anaerobic gram-positive family of Firmicutes, potent short-chain fatty acid producer, was the most discriminating taxonomic group by temperature trajectory. And what we found is that just the relative abundance of this one bacterial family alone, Lachnos graciae, explained more than a third of variation in temperature response following the model. We could also put it back into the mice, in the germ-free mice, and change its temperature, so it's not just an epiphenomenon. It's actually participating in this physiologic variation. We know by this point the microbiome is not just an artifact and a confounding result of the fact that patients get antibiotics that distort the microbiome. We know it's involved in disease pathogenesis. This is through a series of animal models, an important paper from, coming up on a decade ago, from the Abuse for Signals group in the Netherlands, showed that if you deplete the lower gut microbiome in experimental mice, then they lose protection from pneumococcal pneumonia, both in terms of mortality and bacterial clearance. And this has been a very consistent and robustly replicated finding in other models. So from our group, we've used a Klebsiella pneumonia model and found that depleting the gut with Tiptazo, which has potent anti-anaerobic activity, makes them less able to clear bacteria and still into the respiratory tract. And even sterile models like Hyperoxia. So we can turn a non-lethal model of hyperoxia exposure into a lethal model by pre-treating the animals with Tiptazo, with Zosyn. So again, if you experimentally remove the anaerobes from the lower gut, animals are more susceptible to organ failure and mortality in models of critical illness. And finally, one thing we know for sure is that, as I've already alluded to, microbiome and respiratory microbiome are altered by our clinical interventions. So we may talk about modulating the microbiome therapeutically over the horizon. Guess what? We're already doing it. We just don't really understand what we're doing. So here's one data slide. There's plenty like this. But our group has shown that if you look at gut microbiota of patients when they arrive to the ICU, you can actually tell if they got cefepime or Tiptazo in the emergency department based on anaerobic coverage. So just a single slug of the personal Tiptazo Bactam lowers the concentration of anaerobic bacteria and lower gut specimens by 100-fold or 1,000-fold. Similarly, other exposures you might not think about also change microbiota. So Shanna Ashley is a postdoc that worked with me, and she published several years ago that oxygen, inhaled oxygen, both changes respiratory and lower gut microbiota. It changes the identity of subsequent pathogens that are cultured out from humans. And our animal models show that it actually participates in pathogenesis by way of using antibiotic pretreatment for germ-free animals. So that is a world study of where I think the field is. Now I'm going to move on to where we're headed. Starting with over the horizon, in the distant future, what can we expect in 10 to 20 years of microbiome research? I think the big concept here is that we aspire to treat the microbiome as a therapeutic target. We obviously have many, many therapies directed at the host. We're living the dream in terms of precision medicine, in that we can very selectively and very precisely modulate specific immune pathways by way of biologic agents. We're nowhere near that in our ability to modulate the microbiome. We have very clunky, coarse interventions like antibiotics and gut decontamination. We have very ineffective interventions like probiotics. And I think there's a lot of work that needs to be done so that we can coherently and rationally and deliberately modulate the microbiome's therapeutic effect. To give you some inspiration and make you think that this is actually worth pursuing, we can take some inspiration from the field of oncology. So there have been three papers in science all telling essentially the same story, which is that if you have patients with melanoma who are getting checkpoint inhibitor therapy, so anti-PD-1 immunotherapy, so mobilize the host's immune response to kill the cancer. If you compare responders and non-responders in their gut microbiota, they are different from each other in consistent ways. And then, follow-up study, you can take the gut microbiota from the responders, administer it to the non-responders, and change their responsiveness to immune therapy. So this is suggesting that the variation in immune response that we want to mobilize to treat the cancer is actually downstream of an upstream source of immunization, which is the gut microbiome. So this is the dream. This is what we want to do. Actually, therapeutically modulate the microbiome to make patients respond better to the therapies that we're trying to do to fix host dysfunction. But we're miles from that when it comes to modulating the microbiome, specifically in the ICU. What would this look like in terms of how could we tackle this problem of therapeutically targeting the microbiome? I'm going to go in reverse order of feasibility, namely starting with the hardest, which is to say engineering an ecosystem. This is awfully challenging. What they're doing in the melanoma studies is fecal microbiota transplant. So they're taking a slurry of fecal contents or fecal contents from volunteers and administering them colonoscopically to volunteers. That is occasionally done in the ICU with increasing frequency. To my knowledge, it's reserved for patients with refractory and life-threatening febrile colitis. That's a special exception. And obviously, there's a lot to be learned from that. But I'm talking more broadly about could we modulate the microbiome to treat non-infectious or other infectious causes other than C. diff. This is quite tricky. The work that's been done to date, I think, is quite clarifying. So the biggest, I think, best-designed probiotic study to date was in JAMA a few years ago. And the authors wanted to prevent ventilator-associated pneumonia. And they randomized patients to receive a probiotic. So a very reasonable probiotic administration back to bacillus remnosus. They accomplished the study well. They achieved separation in terms of exposure. And it was a complete flat effect. No effect on VAP or any important secondary endpoints. I think that there are a lot of lessons to be learned from this. I think it's hopeless. And I think it's probably pointless to try to do more simple probiotic studies like this. For one, patients were not phenotyped ahead of time. We don't know what their microbiota were. So this would be like just giving empiric immune-modulating therapy to patients without knowing their own underlying immune test response. Giving a single species empiric probiotic capsules, I think, is pretty hopeless, especially among critically ill patients, for a few reasons. Even in healthy folks, people outside of the ICU taking probiotics, they don't really work the way you want them to. It's not as simple as displacing the bad bugs with the good bugs. Often, probiotics don't take. They don't see. You can't detect them in the lower gut because they've got to outcompete the existing bugs. They have to survive and endure the acidic stomach, the alkaline duodenum. And there's plenty of competition of well-adapted bacteria downstream. In the ICU, you have the added challenge of every patient getting antibiotics. In this cohort, lactobacillus is a relatively wimpy bug. When it comes to antimicrobial therapy, penicillin kills it. And more than 80% of patients were receiving antibiotics that had activity against lactobacillus. So this feels quite hopeless in the ICU. Good luck getting your patients off of broad-spectrum antibiotics. This feels like trying to reseed a forest while the forest fire is still raging. So long as our patients are getting carpet-bombed with systemic antibiotics, I think just administering single strains of probiotics empirically are unlikely to be terribly effective. So I have made pessimists when it comes to that. Additionally, we should do this very cautiously. This is not a completely benign intervention. This study had a non-trivial number of patients that developed infections with lactobacillus, including some serious events that were bloodstream infections and prolonging the ICU stay for these patients. So I think that trying to engineer an ecosystem by way of viable organisms is a heavy lift, and I think there may be better ways forward. I may be wrong about this, but I'd be surprised if any simple probiotics replacement studies are accepted in the next decade. A little less hard, though, would be repeating bacterial products instead of the bacteria themselves. So we think, remember back to our model of how the microbiome is important, end-organ function and dysfunction. A really important mediator of all of this is that top line, chemical dissemination of microbiome-derived metabolites. And we think biologically and clinically, this may be a very promising way to move forward. It's easier to give bacterial products to a patient than to give viable bacteria that take root and repopulate the ecosystem that's been depleted by probiotics. I'll give you some unpublished data from my lab. Mark Adame is a graduate student working with me, looking at questions like this. What are the microbiome-derived mediators of end-organ dysfunction? He started with the observation that we previously showed you, which is that in modeling sepsis, so exposing lactic mice to endotoxins to make them physiologically septic, the gut microbiome explains a lot of variation in their temperature response. So that same bacterial family, Lachnospiraceae, explains more than a third of variation in temperature loss after endotoxin exposure. What we found is that if you measure short-chain fatty acids, so prominent gut microbiome-derived metabolites in the blood of these animals, that explains that variation. So it's like a very similar curve. What you're seeing is that the amount of acetate in the blood, much of which we have reason to think comes from the gut microbiome, explains variation in temperature loss after sepsis. We think it's going from the gut to the blood, even though host varieties can make acetate as well, for a number of reasons, but one of them is this. If we measure acetate levels in the blood in unexposed animals and animals that are made septic with endotoxin, we see a surge, and we see that slip in the cecum. So in the lower gut, you see all your acetate records are gone. Indirect evidence that the gut is essentially serving as a reservoir for this important small molecule that, in stressed states like sepsis, get into the blood, come into the end organs, and perform important functions there. We know that if you give acetate, you can recapitulate the findings that you see in the correlation level. Crucially, our clinically used antibiotics devastate acetate production in the gut. So here's the concentration you expect in control mice. Here's what happens if you give a mouse three days of peptazo, so commonly used zosyn. It essentially makes acetate undetectable in the lower gut, and cefepime does not have the same effect. So I worry that every time we just empirically treat someone with broad-spectrum anti-anaerobic antibiotics, we're doing this to this essentially important endocrine organ that is important in the context of health. So the question we're asking is, can we give it back? Can we give acetate, butyrate, proprionate, and other short-chain fatty acids back? On this, I'm partnering with Kathleen Springer, who is also a deputy director at the Weill Institute, with me. And Kathleen has been spearheading this effort to rescue animals by giving them those short-chain fatty acids back. So we pre-treat these animals with anaerobic depleting antibiotics, dropping enteric concentrations of the short-chain fatty acids, and Kathleen puts together a cocktail, hopefully at physiologic concentrations of acetate, proprionate, and butyrate, and then administers it via enema at the time of the insult to see if we can rescue them and restore both fecal as well as blood concentrations of short-chain fatty acids and attain the immunologic and endocrine function benefits that we anticipate we're going to get. This is what it looks like. So we know that we can muck up the microbiome with antibiotics, unsurprisingly. If we take – these are wild pigs, not wild pigs, but they're not STF pigs. If we take these pigs and we give them several days of antibiotics, they do what they're expected to do. So if we give anaerobic depleting antibiotics, like enteric metronidazole and mecamycin, proteobacteria take over and the protective anaerobes go away. If we give them anaerobic enriching, so basically selective decontamination of the beta-subtract, if we give those empirically, the proteobacteria, the gram-negatives, stay low, and the bacteroides take over. How about the metabolites? How about those important short-chain fatty acids the gut microbiome makes? They act in the same direction. So when we measure blood concentrations of acetate, propionate, and butyrate in antibiotic-treated animals, they go in the direction to expect. So if you give the anaerobic depleting antibiotics, whole blood short-chain fatty acids go down. If you give the other ones, they go up. And over the time point of an injury, we found that if you administer this enema, the short-chain fatty acid enema, you can restore blood concentrations of short-chain fatty acids. So this is the same data I just showed you. This is whole blood short-chain fatty acids. Before we give them the antibiotics, giving them anaerobic-depleting antibiotics drops it, and then giving them the short-chain fatty acid enema restores them to physiologic or superphysiologic concentrations. So we know that they're, in principle, can be done. And exciting, this is preliminary data, but it looks like it does have a consequence on the host immune response. So this is in partnership with Scott Denstead. We take monocytes, peripheral monocytes, we stimulate them with endotoxin, and we measure how much we're here on gamma, just a way of saying how twitchy and reactive your native immune response is. We see that if we deplete the gut microbiome of anaerobes, these monocytes get twitchier. They get more hyperinflammatory, which we think is relevant to sepsis pathophysiology. And then three, six hours after we initiate this butyrate, acetate, proprionate enema, it's restored back to its former homeostatic function. So we are optimistic. I'm more optimistic about a metabolite rescue approach than I am about reconstituting the microbiome by way of probiotics or microbiotic transplant. And finally, the easiest thing would just be to minimize dysbiosis in the first place. And I'm going to talk quite a bit more about this in my very last section, but my prediction in the next 10 years is that we start thinking of lower gut anaerobes as an opportunity for minimizing e-epigenesis. In the same way that many of the strides in critical care research have been minimizing harms from our supportive care, lung protective ventilation, continuous weakening trials, how can we minimize the damage we're doing with these necessary interventions? I think thinking hard about whether patients truly need anti-anaerobic coverage that depletes the gut and protective bacteria, I think it's going to be a major push in the next decade. But we also have to think diagnostically. So long as we can't characterize the microbiome in real time, you could argue we're in the dark. We're not going to blindly modulate something, at least not effectively, without knowing what's there in need of modulation. Honestly, this has been unthinkable until relatively recently because the way we characterize the microbiome has largely been done using short-rate sequencing. So Illumina-style shotgun sequencing, with or without PCR amplification of the 16S gene. It is time intensive, it takes a lot of time at the bench, typically run things in batch, bioinformatically challenging. It's just never been feasible that we would characterize the microbiome at the bedside. And I think that is changing and will change in the next decade, namely because of this. So Oxford Nanophores, the British company that makes the MinION, you may be familiar with it. It sequences very differently from Illumina products. Basically, individual strands of nucleic acid, DNA or RNA, go one at a time through this core. It measures the difference in average across that core, recurring across that core, generates a squiggle, and then there's a base color that can turn that squiggle into A's and G's or P's and T's. So it reads the genetic code, kind of like reading EKG for DNA. This is what the squiggle looks like. It's very noisy. Machine learning determines genetic code. And then there are increasingly fast tools that can quickly turn the genetic code into taxonomic classification. In 2017, we showed that you could use this approach for pathogen detection. So we found that to give me a BAL specimen from a critically ill patient, within a few hours, we could be sequencing it in real time, identify the pathogen a day ahead of when the clinical microlabs grows out the pathogen by way of culture. Others have taken this further. This adjustment of radius groups, an important work showing that this can actually be scaled up and done, again, for pathogen detection, not for microbiome characterization. And I think it's impressive and convincing that by their protocol, they can get this down to about six hours from the time you give them to a respiratory specimen when they tell you what bugs are in there, with really outstanding sensitivity and lower specificity. And I think that's a really fundamental problem the field has to face, is that we're going to find a lot of bugs through these approaches that we don't often reconcile into our conventional model of pathogen, non-pathogen. So I think there's no reason that we can't use this technology to characterize the microbiome in real time. It also may be bioinformatically faster and more lightweight than we think, because there's actually a lot of rich biologic information embedded in the squiggle itself. So not waiting for that time-intensive step of going from the raw squiggle to the base call to the genomic alignment. Some collaborators here at the University of Michigan, Josh Welch, who you may know, have shown that you can just use machine learning to interrogate that squiggle and identify whether something's human or microbial, and if it's microbial, what is it taxonomically? So you can think of this just like Shazam on your phone, can listen to sound waves and quickly figure out what the song an artist is. This would be like quickly looking at a specimen and quickly figuring out what exactly it comes up for, identifying it in real time. And most of the work that's been done on this has been on respiratory specimens, which are harder to work with. Either they're very low biomass to begin with, like BAL fluid, or the ratio of host to bug is 99.9 to 0.1. For lower depth specimens, I think, in a way, this is actually easier to work. So I think it's not unthinkable that before long, we can be characterizing the microbiome in real time at the bedside. In the not too distant future, what are we doing in the next five to 10 years? For this, I think the key story is longitudinal sampling and deep orthogonal phenotyping of critically ill patients. So a lot of the important work that's been done to date has been single center, by and large, cross-sectional, by and large, descriptive, and just describing how is the microbiome altered, how does it correlate with disease severity and the predictive outcomes. But I think we're at the point where we need to sample patients at multiple times during their ICU stay. The microbiome changes, it's very dynamic, and it changes by the day, unlike the host genome. We also need to integrate it with other cutting-edge, multi-omic measurements of the host. And I think I'm really excited to be part of the APS consortium. So for those who don't know, APS is, you can think of it as the successor to the PETL network, which is itself the successor to the ARDS network. It is going to be a large, multi-center, perspective, observational cohort study of critically ill patients with ARDS, pneumonia, and sepsis. There'll be six centers, clinical centers, also about 22 hospitals. 4,000 critically ill patients longitudinally sampled at multiple time points, both during their inpatient stay and for many of these patients at three months, six months, and 12 months following discharge. I think the recovery of the microbiome is a really important and unknown aspect of long-term functional recovery. And suffice to say, there will be comprehensive, high-dimensional biologic phenotyping, as well as clinical and psychosocial phenotyping of these patients. So this will be an incredible opportunity to understand how the microbiome fits into other facets of critical illness pathophysiology. And finally, my last section will be on what do I expect to happen in the near future? And I hinted towards this earlier. I think we are going to change how we think about the harm of antibiotics. I think we all know that antibiotics are not a free lunch that come with associated adverse effects, either drug reactions or promoting resistance. But I think evolving work is going to, emerging work, is going to show that we have to think much more concretely about the direct harm that comes to individual patients from antibiotics. For this, I'll share just a couple studies. So one of my trainees, Rishi Chandra, who's at the University of Michigan, was motivated by something that I described earlier, which is that in multiple animal models, depletion of gut anaerobes worsens susceptibility to pneumonia and other causes of critical illness. It's a very consistent finding. If you blow up the lower gut microbiome in animals, they do worse with infectious and non-infectious models of critical illness. So Rishi took advantage of the fact that among our critically ill patients, we have variation of whether they get anti-anaerobic antibiotics or not. So piperacillin, tabubactam, or Cetaphe, in terms of examples of anti-anaerobic or non-anaerobic antibiotics. Initially, in this paper in ERJ, Rishi looked at over 3,000 of cancer-ventilated patients, to a first approximation, all of them were getting antibiotics in the first 72 hours, but two-thirds had anti-anaerobic coverage, and one-third lacked anti-anaerobic coverage. And this is largely medical ICU patients that have no firm indication for anti-anaerobic antibiotics. So these are not intestinal perforation, these are not deep head and neck tissue infections. These are patients with largely pneumonia, USF, and bloodstream infection, things that the guidelines do not recommend anti-anaerobic coverage. And he found a dramatic difference in outcomes. Both in VAP, infection-free survival, and all-cause survival. And all of the deaths were not attributed to secondary infections, so it's clearly more interesting and important than that. But he found a roughly five to six percent difference, absolute difference, in mortality in patients who got anti-anaerobic antibiotics early in their ICU stay. And you could push back, you could say, okay, well, that's a single center, probably confounded, even though we did everything we could to control for severity of illness, co-exposures, but it's one center in Michigan, let's not generalize too much. Luckily, a group in Amsterdam, again, used for Cigna's group, looked at 16,000 patients coming to the emergency department, used the same criteria we did for differentiating who did and did not get anti-anaerobic antibiotics, and found the same separation. They found a difference in mortality, more harm in the patients who got the anaerobic, anaerobic-depleting antibiotics, like Tiftezo. But you could say, okay, it's no longer a single center, but maybe they were subject to the same confounding. Maybe it was just confounding by indication. There was something different. There was something that motivated physicians to give the anti-anaerobic antibiotic patients to the patients who did, and maybe that's what drove the difference. So we were fortunate in that we could take advantage of a 15-month drug shortage. So this is new data that will be published around the time of this webinar, this symposium. Rishi looked at a 15-month shortage in Tiptezo, piperacillin-tezobactam, that occurred and as you can see from this figure, it dramatically changed our prescribing practices. It went from about a quarter of all hospitalized patients getting Tiptezo to nearly none, and then after the shortage arose, it flipped. So we have this natural experiment where essentially the only predictor that determined whether you got sepapemersosin or Tiptezo was the calority. Did you fall within the drug shortage? And this lets us actually do a very powerful approach called instrumental variable analysis. So we can actually treat it like a natural experiment and treat it as essentially a randomization to determine if exposure to anaerobic defeating antibiotics like zosyn were associated with harm. And lo and behold, we found the same thing, a 5% absolute difference in mortality, and it was borne by the zosyn-treated patients. So early anaerobic antibiotics associated with harm increased mortality, robust to every sensitivity test that we can do. So I think that this is leading us to conclude that the microbiome, like we say, is an organ. It has its own function and dysfunction, and we are already modulating it. This is not over the horizon. This is now with an incomplete understanding of biological and clinical consequences. So I would love the field, especially the substance research field, to think a little more nuanced, be a little more nuanced in their thinking. It's not just about tying to antibiotics that matters. It's which antibiotic matters. And I specifically want us to reckon, to have a reckoning with most of our substance patients do not need anti-anaerobic antibiotics, and I worry that we're causing direct harm to them. So my prediction is that in the next five years, we start coming to terms with this, and perhaps we treat lower gut anaerobes as something that's worth saving in the same way that we use lung protective ventilation to protect the uninjured lungs of mechanically ventilated patients. That is it. That's my prediction for the next decade in microbiome research. And I, of course, appreciate the invitation and opportunity to speak to you all, and I look forward to the discussion.

microbiome research

critical care environment

University of Michigan

Pulmonary and Critical Care Medicine

microbiome's role in critical illness

advancements in microbiome research methods

microbiome modulation

antibiotic use in critical care settings

critical care environments

historical significance

microbiome variability

antibiotics impact

anaerobic bacteria

real-time microbiome characterization

longitudinal sampling