I'd like to thank the organizers for the opportunity to talk today about one of my favorite topics, the subclassification of pediatric ARDS. I have no relevant financial disclosures. First, I'd like to invite you to take a moment to search your feelings with regards to pediatric ARDS, because somewhere deep down, we all know that all pediatric ARDS is not alike, in that there are different types of ARDS, and if we could only better classify and identify those differences, we could better care for our patients. The problem is that this is really hard. On the one hand, every day we come across patients that meet PARDS criteria and will get better no matter what we do, and we don't worry about those patients. At the same time, we can have a patient next door that was very similar to the one with mild rapidly resolving ARDS, and with the same injury, and this patient will have a prolonged and severe ARDS course. We might have a third patient that has ARDS a few weeks after bone marrow transplantation. We might manage each of these patients differently, but we don't have any clinical trial data that supports treating them differently. What's more, we don't really have a good way to classify the biological differences between these groups, so we can begin to develop targeted therapies in pediatric ARDS. We can talk about all sorts of subclassification schemes, but for now, I want to establish the basics using this figure from the new PALIC report just released. Anytime we divide a group by a classifier, we are creating subtypes. This can be by demographics, age, laboratory values, anything. Pediatric ARDS is a subtype of ARDS. A subphenotype is a subtype that can be discerned by something that we observe. The classification of ARDS into mild, moderate, and severe is phenotypic subclassification. Endotypes subclassify based on some component that is thought to indicate a biological difference within a group that may not be observable. For example, we might use serum angiopoietin 2 levels to classify ARDS endotypes as to those with and without endothelial injury. One logical way to subcategorize pediatric ARDS is whether it's caused by a direct lung injury like aspiration or viral infection, or if it's caused indirectly by something like sepsis or trauma. Clinical trials of surfactant replacement therapy use this approach to target surfactant inactivation in direct lung injury. We can also classify PARDS by latency of onset. By definition, PARDS must develop within seven days of injury, but one can imagine that the path of biology is different with delayed versus immediate onset. Etiology is also important. Although we might manage them similarly on the ventilator, single organ failure from respiratory viral infection is likely to be biologically different than ARDS from trauma or after sepsis. Over the course of childhood, the number of airway branches will increase from the mid-teens to the low-20s, and gas surface exchange area will increase from 5 to 100 square meters. The immune system also matures over this time, so it's likely that ARDS in the neonatal pediatric and adolescent population is biologically different. A large percentage of our PICU population has underlying comorbidity, and immunocompromised PARDS has not only clinical significance, but as I'll show you, epigenetic changes that support its consideration as a separate cohort in ARDS. While the degree of oxygenation impairment was the first method for subclassification and tracks closely with outcome, data that Anupa Hall will share shows that worsened ventilator independently predicts poor outcome, and we should probably be considering both when subclassifying based on gas exchange. I think chest X-ray or other imaging is kind of a gray zone between subphenotype and endotype since you usually have a pretty good idea of what your chest X-ray will look like based on your exam. But regardless, I placed it with the endotypic classification, and I wanted to highlight a pre-COVID paper in adults that tested whether we could use chest X-ray to guide ventilator management. In this study of 400 patients, half were randomized to standard management and half to a larger tidal volume and lower PEEP with the protein strategy. In intention to treat analysis, there was no difference, but after exclusion of 40 incorrectly classified patients, there was benefit for a personalized ventilator management strategy. We all incorporate chest X-ray data into ventilator management decisions, but the point here is that ARDS patients with different radiologic patterns have a different response to different ventilator strategies. Serum biomarkers can be used in a variety of ways to either assess injury to specific departments, as in the case of angiopointin-2s affecting protein D or RAGE, or provide a barometer on overall inflammatory status with a host of inflammatory cytokines. As you're probably aware, Carolyn Kalfi and Michael Monthea showed that serum biomarkers can be used in conjunction with clinical data to classify adult ARDS into two phenotypes. However, as they point out in the original paper and proved in a subsequent one, these aren't really endotypes because these same two groups can be identified using clinical variables alone. Retrospective analysis of ARMA, alveoli, and FACT trials showed that the hyperinflammatory group did better with a high PEEP and restrictive fluid regimen. Mary Dahmer and Heidi Fleury recently showed that the same two phenotypes in pediatric ARDS, indicating that there are a lot of the same processes occurring in adult and pediatric patients. Lastly, we can use highly dimensional and granular technologies to dive down deep into the molecular biology of parts. The processing of these samples, acquisition of data, and analysis might be more complex and difficult, but hold the potential to uncover more fundamental biology and develop more targeted therapeutics based on subtype. However, the degrees of freedom in these analyses is greater than the number of ARDS patients in the country per year, and without proper constraints, you risk emerging from a rabbit hole with as many parts endotypes as you have ARDS patients. Nonetheless, the importance of this approach is shown on the next slide. So, how are we going to use all this information to better care for our patients? I think this data really drives home the subphenotype versus endotype story. We analyzed peripheral blood transcriptomic data on day one of ARDS, and then we classified the relationship to the hyperinflammatory or hypoinflammatory phenotype 1 and 2 that were described by biomarkers. On the left, the CATS1 subgroup is older. The CATS1 and 2 subgroups are more likely to be immunocompromised and have longer duration mechanical ventilation, and the CATS1 subgroup had worsened oxygenation. This is the transcriptomic classification. On the right, we classified the patients into phenotype 1 or 2 based on the serum biomarkers and the clinical variables. While we recaptured the classification of subjects as CATS1, 2, or 3, there's only a weak association of subgroup class with peripheral blood transcriptome. Genes set in enrichment analysis of phenotype 1 and 2 were not related to inflammatory signaling, but rather apoptosis and cell survival. This is similar to analysis of the MARES cohort in 2019. The take-home from this is that we can't surmise the entire immune landscape by sampling the peripheral blood. We have to consider the other compartments as well. This is exactly what we did when we analyzed the transcriptome of the nasal epithelial cells in pediatric ARDS. We found that the transcriptome of these cells can be broken down into four different patterns with some overlap between each. We then analyzed these transcriptomes longitudinally. We realized that there were two different injury patterns. Subgroup B, which was characterized by inflammation and epithelial cell dysfunction, and subgroup D, which was characterized by a loss of epithelial stem cell function. Both subgroups underwent similar pattern of repair regeneration, which is subgroup A, and returned to homeostasis in subgroup C, which is the most similar to control. Subjects that had an initial pattern of B or D did not have worse initial PARDS, but did have longer PARDS duration than those that started with pattern A or C. One of the most straightforward ways to identify genes in pediatric ARDS is to perform genome-wide association studies. Angiotensin-converting enzyme 2 was the first gene to be found in association with ARDS, and as we saw with COVID-19, it's important in both the respiratory epithelial cell and endothelial cell function. This recently constructed table outlines all the genes with polymorphisms that are associated with ARDS or pediatric ARDS. While you can argue about how some of these genes were classified, by and large, they regulate functions that we would consider important in ARDS pathogenesis. While the idea of performing such assays to direct clinical care has been widely criticized as impractical and has some serious ethical and practical implications, such pressures are being pursued in fields like pharmacogenomics when it's important to know how patients will respond to a certain drug. It's not inconceivable that we would develop a limited panel of SNPs for genes related to immune response or cell function that could help us to subclassify and target therapies in pediatric ARDS. Epigenetics has been shown important in a host of complex multifactorial diseases like lupus, and we generated data indicating that there might be a subgroup of patients with an epigenetic predisposition to severe lung injury. Using the same nasal brushings I referenced before, we performed bisulfite sequencing to characterize the methylation patterns of respiratory epithelial cells. ARDS patients showed two patterns, and over time, those patterns were stable. Controlled subjects all had a pattern like subgroup 1. And when you compared the clinical characteristics of the two subgroups, all the immunocompromised subjects were in subgroup 2, and the genes that were hypomethylated were largely involved in inflammation and chemotaxis. When we performed multi-omic analysis using transcriptomethalone, what we observed were coordinating patterns of hypomethylation and gene transcription that were not only coordinated, but also suggested activation of some of the downstream programs not evident in analyzing each dataset individually, such as the activation of cell survival pathways in subjects with methyl subgroup 1 and transcriptomic subgroup B, suggesting that perhaps stem cells survive in this environment at the expense of function. So how are we supposed to balance precise characterization of biological heterogeneity with what is practical and potentially clinically useful in defining powered subgroups? First, we need to incorporate robust sample collection and preservation protocols in our studies. There is a balance between what is practical for sites to do with regards to sample collection and preservation, but the last five years have seen an explosion in the robust sample collection and processing and preservation tools, as people like plant biologists and deep-sea marine researchers want to incorporate single-cell molecular technologies into their research. My lab works on developing techniques feasible in the PICU or clinical labs so that we can use these technologies to link more traditional subgroup classification strategies with cell and molecular techniques that will lead to targeted therapies. As we all know, the cat is often out of the bag by the time someone arrives in the pediatric ICU. We need to understand what is happening in the lead-up to acute respiratory failure and what happens after it resolves. We also need to take a systems biology approach to pediatric ARDS. This review talks about the similarity of peripheral blood gene expression profiles in different critical illnesses and the overlap of subtypes that are identified in these illnesses. Before we lament how few patients we have in which to test our therapies in, let's consider how to better design our trials to enrich for patients likely to have a poor outcome with the pathobiology that we want to target. We're obviously going to need team science and collaboration moving forward, but if we can reduce biological heterogeneity and enrich for patients likely to have a poor outcome, we might be able to conduct meaningful clinical trials with a quarter of the number of patients we are enrolling in current ARDS trials. That being said, let's remember that we need to be selective in what we target and how we do so. We need to apply a high degree of rigor in developing personalized medicine approaches in ARDS. With each passing year, we have new tools that more specifically and potently target cell types, molecules, or pathways. We need to first prove that the animal or cell models in which we test these therapies recapitulate human disease before assuming that targeting something we see in the omic data in animals will give us the same result in people. We need to balance what we want to do with what is possible in a consortium. And so, moving forward, I look forward to speaking and collaborating with any of you. I encourage you to reach out if you have questions or want to discuss this further, develop collaborations. I would like to thank all my funding agencies as well as the members of my lab and collaborators which have made this work possible.

pediatric ARDS

subclassifications

targeted therapies

etiology

personalized medicine approaches