Mucus sialylation determines intestinal host-commensal homeostasis.

Intestinal mucus forms the first line of defense against bacterial invasion while providing nutrition to support microbial symbiosis. How the host controls mucus barrier integrity and commensalism is unclear. We show that terminal sialylation of glycans on intestinal mucus by ST6GALNAC1 (ST6), the dominant sialyltransferase specifically expressed in goblet cells and induced by microbial pathogen-associated molecular patterns, is essential for mucus integrity and protecting against excessive bacterial proteolytic degradation. Glycoproteomic profiling and biochemical analysis of ST6 mutations identified in patients show that decreased sialylation causes defective mucus proteins and congenital inflammatory bowel disease (IBD). Mice harboring a patient ST6 mutation have compromised mucus barriers, dysbiosis, and susceptibility to intestinal inflammation. Based on our understanding of the ST6 regulatory network, we show that treatment with sialylated mucin or a Foxo3 inhibitor can ameliorate IBD.

Regulation of T helper cell differentiation by the interplay between histone modification and chromatin interaction.

The development and function of the immune system are controlled by temporospatial gene expression programs, which are regulated by cis-regulatory elements, chromatin structure, and trans-acting factors. In this study, we cataloged the dynamic histone modifications and chromatin interactions at regulatory regions during T helper (Th) cell differentiation. Our data revealed that the H3K4me1 landscape established by MLL4 in naive CD4+ T cells is critical for restructuring the regulatory interaction network and orchestrating gene expression during the early phase of Th differentiation. GATA3 plays a crucial role in further configuring H3K4me1 modification and the chromatin interaction network during Th2 differentiation. Furthermore, we demonstrated that HSS3-anchored chromatin loops function to restrict the activity of the Th2 locus control region (LCR), thus coordinating the expression of Th2 cytokines. Our results provide insights into the mechanisms of how the interplay between histone modifications, chromatin looping, and trans-acting factors contributes to the differentiation of Th cells.

Restraint of IFN-γ expression through a distal silencer CNS-28 for tissue homeostasis.

Interferon-γ (IFN-γ) is a key cytokine in response to viral or intracellular bacterial infection in mammals. While a number of enhancers are described to promote IFN-γ responses, to the best of our knowledge, no silencers for the Ifng gene have been identified. By examining H3K4me1 histone modification in naive CD4+ T cells within Ifng locus, we identified a silencer (CNS-28) that restrains Ifng expression. Mechanistically, CNS-28 maintains Ifng silence by diminishing enhancer-promoter interactions within Ifng locus in a GATA3-dependent but T-bet-independent manner. Functionally, CNS-28 restrains Ifng transcription in NK cells, CD4+ cells, and CD8+ T cells during both innate and adaptive immune responses. Moreover, CNS-28 deficiency resulted in repressed type 2 responses due to elevated IFN-γ expression, shifting Th1 and Th2 paradigm. Thus, CNS-28 activity ensures immune cell quiescence by cooperating with other regulatory cis elements within the Ifng gene locus to minimize autoimmunity.

MLL4 prepares the enhancer landscape for Foxp3 induction via chromatin looping.

MLL4 is an essential subunit of the histone H3 Lys4 (H3K4)-methylation complexes. We found that MLL4 deficiency compromised the development of regulatory T cells (Treg cells) and resulted in a substantial decrease in monomethylated H3K4 (H3K4me1) and chromatin interaction at putative gene enhancers, a considerable portion of which were not direct targets of MLL4 but were enhancers that interacted with MLL4-bound sites. The decrease in H3K4me1 and chromatin interaction at the enhancers not bound by MLL4 correlated with MLL4 binding at distant interacting regions. Deletion of an upstream MLL4-binding site diminished the abundance of H3K4me1 at the regulatory elements of the gene encoding the transcription factor Foxp3 that were looped to the MLL4-binding site and compromised both the thymic differentiation and the inducible differentiation of Treg cells. We found that MLL4 catalyzed methylation of H3K4 at distant unbound enhancers via chromatin looping, which identifies a previously unknown mechanism for regulating the T cell enhancer landscape and affecting Treg cell differentiation.

Tcf1 and Lef1 transcription factors establish CD8(+) T cell identity through intrinsic HDAC activity.

The CD4(+) and CD8(+) T cell dichotomy is essential for effective cellular immunity. How individual T cell identity is established remains poorly understood. Here we show that the high-mobility group (HMG) transcription factors Tcf1 and Lef1 are essential for repressing CD4(+) lineage-associated genes including Cd4, Foxp3 and Rorc in CD8(+) T cells. Tcf1- and Lef1-deficient CD8(+) T cells exhibit histone hyperacetylation, which can be ascribed to intrinsic histone deacetylase (HDAC) activity in Tcf1 and Lef1. Mutation of five conserved amino acids in the Tcf1 HDAC domain diminishes HDAC activity and the ability to suppress CD4(+) lineage genes in CD8(+) T cells. These findings reveal that sequence-specific transcription factors can utilize intrinsic HDAC activity to guard cell identity by repressing lineage-inappropriate genes.

Regulation of pluripotency and self- renewal of ESCs through epigenetic-threshold modulation and mRNA pruning.

Embryonic stem cell (ESC) pluripotency requires bivalent epigenetic modifications of key developmental genes regulated by various transcription factors and chromatin-modifying enzymes. How these factors coordinate with one another to maintain the bivalent chromatin state so that ESCs can undergo rapid self-renewal while retaining pluripotency is poorly understood. We report that Utf1, a target of Oct4 and Sox2, is a bivalent chromatin component that buffers poised states of bivalent genes. By limiting PRC2 loading and histone 3 lysine-27 trimethylation, Utf1 sets proper activation thresholds for bivalent genes. It also promotes nuclear tagging of messenger RNAs (mRNAs) transcribed from insufficiently silenced bivalent genes for cytoplasmic degradation through mRNA decapping. These opposing functions of Utf1 promote coordinated differentiation. The mRNA degradation function also ensures rapid cell proliferation by blocking the Myc-Arf feedback control. Thus, Utf1 couples the core pluripotency factors with Myc and PRC2 networks to promote the pluripotency and proliferation of ESCs.

NAD activates olfactory receptor 1386 to regulate type I interferon responses in Plasmodium yoelii YM infection.

Olfactory receptors (Olfr) are G protein-coupled receptors that are normally expressed on olfactory sensory neurons to detect volatile chemicals or odorants. Interestingly, many Olfrs are also expressed in diverse tissues and function in cell-cell recognition, migration, and proliferation as well as immune responses and disease processes. Here, we showed that many Olfr genes were expressed in the mouse spleen, linked to Plasmodium yoelii genetic loci significantly, and/or had genome-wide patterns of LOD scores (GPLSs) similar to those of host Toll-like receptor genes. Expression of specific Olfr genes such as Olfr1386 in HEK293T cells significantly increased luciferase signals driven by IFN-ß and NF-κB promoters, with elevated levels of phosphorylated TBK1, IRF3, P38, and JNK. Mice without Olfr1386 were generated using the CRISPR/Cas9 method, and the Olfr1386-/- mice showed significantly lower IFN-α/ß levels and longer survival than wild-type (WT) littermates after infection with P. yoelii YM parasites. Inhibition of G protein signaling and P38 activity could affect cyclic AMP-responsive element promoter-driven luciferase signals and IFN-ß mRNA levels in HEK293T cells expressing the Olfr1386 gene, respectively. Screening of malaria parasite metabolites identified nicotinamide adenine dinucleotide (NAD) as a potential ligand for Olfr1386, and NAD could stimulate IFN-ß responses and phosphorylation of TBK1 and STAT1/2 in RAW264.7 cells. Additionally, parasite RNA (pRNA) could significantly increase Olfr1386 mRNA levels. This study links multiple Olfrs to host immune response pathways, identifies a candidate ligand for Olfr1386, and demonstrates the important roles of Olfr1386 in regulating type I interferon (IFN-I) responses during malaria parasite infections.

Correction: Magnaporthe oryzae fimbrin organizes actin networks in the hyphal tip during polar growth and pathogenesis.

[This corrects the article DOI: 10.1371/journal.ppat.1008437.].

WASF3 disrupts mitochondrial respiration and may mediate exercise intolerance in myalgic encephalomyelitis/chronic fatigue syndrome.

Myalgic encephalomyelitis/chronic fatigue syndrome (ME/CFS) is characterized by various disabling symptoms including exercise intolerance and is diagnosed in the absence of a specific cause, making its clinical management challenging. A better understanding of the molecular mechanism underlying this apparent bioenergetic deficiency state may reveal insights for developing targeted treatment strategies. We report that overexpression of Wiskott-Aldrich Syndrome Protein Family Member 3 (WASF3), here identified in a 38-y-old woman suffering from long-standing fatigue and exercise intolerance, can disrupt mitochondrial respiratory supercomplex formation and is associated with endoplasmic reticulum (ER) stress. Increased expression of WASF3 in transgenic mice markedly decreased their treadmill running capacity with concomitantly impaired respiratory supercomplex assembly and reduced complex IV levels in skeletal muscle mitochondria. WASF3 induction by ER stress using endotoxin, well known to be associated with fatigue in humans, also decreased skeletal muscle complex IV levels in mice, while decreasing WASF3 levels by pharmacologic inhibition of ER stress improved mitochondrial function in the cells of the patient with chronic fatigue. Expanding on our findings, skeletal muscle biopsy samples obtained from a cohort of patients with ME/CFS showed increased WASF3 protein levels and aberrant ER stress activation. In addition to revealing a potential mechanism for the bioenergetic deficiency in ME/CFS, our study may also provide insights into other disorders associated with fatigue such as rheumatic diseases and long COVID.

RGS4 controls airway hyperresponsiveness through GAP-independent mechanisms.

Regulators of G protein signaling (RGS) proteins constrain G protein-coupled receptor (GPCR)-mediated and other responses throughout the body primarily, but not exclusively, through their GTPase-activating protein activity. Asthma is a highly prevalent condition characterized by airway hyper-responsiveness (AHR) to environmental stimuli resulting in part from amplified GPCR-mediated airway smooth muscle contraction. Rgs2 or Rgs5 gene deletion in mice enhances AHR and airway smooth muscle contraction, whereas RGS4 KO mice unexpectedly have decreased AHR because of increased production of the bronchodilator prostaglandin E2 (PGE2) by lung epithelial cells. Here, we found that knockin mice harboring Rgs4 alleles encoding a point mutation (N128A) that sharply curtails RGS4 GTPase-activating protein activity had increased AHR, reduced airway PGE2 levels, and augmented GPCR-induced bronchoconstriction compared with either RGS4 KO mice or WT controls. RGS4 interacted with the p85α subunit of PI3K and inhibited PI3K-dependent PGE2 secretion elicited by transforming growth factor beta in airway epithelial cells. Together, these findings suggest that RGS4 affects asthma severity in part by regulating the airway inflammatory milieu in a G protein-independent manner.

Aberrant expression and localization of the RAP1 shelterin protein contribute to age-related phenotypes.

Short telomeres induce a DNA damage response (DDR) that evokes apoptosis and senescence in human cells. An extant question is the contribution of telomere dysfunction-induced DDR to the phenotypes observed in aging and telomere biology disorders. One candidate is RAP1, a telomere-associated protein that also controls transcription at extratelomeric regions. To distinguish these roles, we generated a knockin mouse carrying a mutated Rap1, which was incapable of binding telomeres and did not result in eroded telomeres or a DDR. Primary Rap1 knockin embryonic fibroblasts showed decreased RAP1 expression and re-localization away from telomeres, with an increased cytosolic distribution akin to that observed in human fibroblasts undergoing telomere erosion. Rap1 knockin mice were viable, but exhibited transcriptomic alterations, proinflammatory cytokine/chemokine signaling, reduced lifespan, and decreased healthspan with increased body weight/fasting blood glucose levels, spontaneous tumor incidence, and behavioral deficits. Taken together, our data present mechanisms distinct from telomere-induced DDR that underlie age-related phenotypes.

Loss of mitochondrial Ca2+ uptake protein 3 impairs skeletal muscle calcium handling and exercise capacity.

Mitochondrial calcium concentration ([Ca2+ ]m ) plays an essential role in bioenergetics, and loss of [Ca2+ ]m homeostasis can trigger diseases and cell death in numerous cell types. Ca2+ uptake into mitochondria occurs via the mitochondrial Ca2+ uniporter (MCU), which is regulated by three mitochondrial Ca2+ uptake (MICU) proteins localized in the intermembrane space, MICU1, 2, and 3. We generated a mouse model of systemic MICU3 ablation and examined its physiological role in skeletal muscle. We found that loss of MICU3 led to impaired exercise capacity. When the muscles were directly stimulated there was a decrease in time to fatigue. MICU3 ablation significantly increased the maximal force of the KO muscle and altered fibre type composition with an increase in the ratio of type IIb (low oxidative capacity) to type IIa (high oxidative capacity) fibres. Furthermore, MICU3-KO mitochondria have reduced uptake of Ca2+ and increased phosphorylation of pyruvate dehydrogenase, indicating that KO animals contain less Ca2+ in their mitochondria. Skeletal muscle from MICU3-KO mice exhibited lower net oxidation of NADH during electrically stimulated muscle contraction compared with wild-type. These data demonstrate that MICU3 plays a role in skeletal muscle physiology by setting the proper threshold for mitochondrial Ca2+ uptake, which is important for matching energy demand and supply in muscle. KEY POINTS: Mitochondrial calcium uptake is an important regulator of bioenergetics and cell death and is regulated by the mitochondrial calcium uniporter (MCU) and three calcium sensitive regulatory proteins (MICU1, 2 and 3). Loss of MICU3 leads to impaired exercise capacity and decreased time to skeletal muscle fatigue. Skeletal muscle from MICU3-KO mice exhibits a net oxidation of NADH during electrically stimulated muscle contractions, suggesting that MICU3 plays a role in skeletal muscle physiology by matching energy demand and supply.

Fecal Microbiota Transplantation for Sleep Disturbance in Post-acute COVID-19 Syndrome.

BACKGROUND & AIMS: Post-acute COVID-19 syndrome (PACS) is associated with sleep disturbance, but treatment options are limited. The etiology of PACS may be secondary to alterations in the gut microbiome. Here, we report the efficacy of fecal microbiota transplantation (FMT) in alleviating post-COVID insomnia symptoms in a nonrandomized, open-label prospective interventional study. METHODS: Between September 22, 2022, and May 22, 2023, we recruited 60 PACS patients with insomnia defined as Insomnia Severity Index (ISI) ≥8 and assigned them to the FMT group (FMT at weeks 0, 2, 4, and 8; n = 30) or the control group (n = 30). The primary outcome was clinical remission defined by an ISI of <8 at 12 weeks. Secondary outcomes included changes in the Pittsburgh Sleep Quality Index, Generalized Anxiety Disorder-7 scale, Epworth Sleepiness Scale, Multidimensional Fatigue Inventory, blood cortisol and melatonin, and gut microbiome analysis on metagenomic sequencing. RESULTS: At week 12, more patients in the FMT than the control group had insomnia remission (37.9% vs 10.0%; P = .018). The FMT group showed a decrease in ISI score (P < .0001), Pittsburgh Sleep Quality Index (P < .0001), Generalized Anxiety Disorder-7 scale (P = .0019), Epworth Sleepiness Scale (P = .0057), and blood cortisol concentration (P = .035) from baseline to week 12, but there was no significant change in the control group. There was enrichment of bacteria such as Gemmiger formicilis and depletion of microbial pathways producing menaquinol derivatives after FMT. The gut microbiome profile resembled that of the donor in FMT responders but not in nonresponders at week 12. There was no serious adverse event. CONCLUSIONS: This pilot study showed that FMT could be effective and safe in alleviating post-COVID insomnia, and further clinical trials are warranted. CLINICALTRIALS: gov, Number: NCT05556733.

Predicting regional somatic mutation rates using DNA motifs.

How the locus-specificity of epigenetic modifications is regulated remains an unanswered question. A contributing mechanism is that epigenetic enzymes are recruited to specific loci by DNA binding factors recognizing particular sequence motifs (referred to as epi-motifs). Using these motifs to predict biological outputs depending on local epigenetic state such as somatic mutation rates would confirm their functionality. Here, we used DNA motifs including known TF motifs and epi-motifs as a surrogate of epigenetic signals to predict somatic mutation rates in 13 cancers at an average 23kbp resolution. We implemented an interpretable neural network model, called contextual regression, to successfully learn the universal relationship between mutations and DNA motifs, and uncovered motifs that are most impactful on the regional mutation rates such as TP53 and epi-motifs associated with H3K9me3. Furthermore, we identified genomic regions with significantly higher mutation rates than the expected values in each individual tumor and demonstrated that such cancer-related regions can accurately predict cancer types. Interestingly, we found that the same mutation signatures often have different contributions to cancer-related and cancer-independent regions, and we also identified the motifs with the most contribution to each mutation signature.

Prediction of survival and analysis of prognostic factors for patients with AFP negative hepatocellular carcinoma: a population-based study.

PURPOSE: Hepatocellular carcinoma (HCC) has a poor prognosis, and alpha-fetoprotein (AFP) is widely used to evaluate HCC. However, the proportion of AFP-negative individuals cannot be disregarded. This study aimed to establish a nomogram of risk factors affecting the prognosis of patients with AFP-negative HCC and to evaluate its diagnostic efficiency. PATIENTS AND METHODS: Data from patients with AFP-negative initial diagnosis of HCC (ANHC) between 2004 and 2015 were collected from the Surveillance, Epidemiology, and End Results database for model establishment and validation. We randomly divided overall cohort into the training or validation cohort (7:3). Univariate and multivariate Cox regression analysis were used to identify the risk factors. We constructed nomograms with overall survival (OS) and cancer-specific survival (CSS) as clinical endpoint events and constructed survival analysis by using Kaplan-Meier curve. Also, we conducted internal validation with Receiver Operating Characteristic (ROC) analysis and Decision curve analysis (DCA) to validate the clinical value of the model. RESULTS: This study included 1811 patients (1409 men; 64.7% were Caucasian; the average age was 64 years; 60.7% were married). In the multivariate analysis, the independent risk factors affecting prognosis were age, ethnicity, year of diagnosis, tumor size, tumor grade, surgery, chemotherapy, and radiotherapy. The nomogram-based model related C-indexes were 0.762 (95% confidence interval (CI): 0.752-0.772) and 0.752 (95% CI: 0.740-0.769) for predicting OS, and 0.785 (95% CI: 0.774-0.795) and 0.779 (95% CI: 0.762-0.795) for predicting CSS. The nomogram model showed that the predicted death was consistent with the actual value. The ROC analysis and DCA showed that the nomogram had good clinical value compared with TNM staging. CONCLUSION: The age(HR:1.012, 95% CI: 1.006-1.018, P-value < 0.001), ethnicity(African-American: HR:0.946, 95% CI: 0.783-1.212, P-value: 0.66; Others: HR:0.737, 95% CI: 0.613-0.887, P-value: 0.001), tumor diameter(HR:1.006, 95% CI: 1.004-1.008, P-value < 0.001), year of diagnosis (HR:0.852, 95% CI: 0.729-0.997, P-value: 0.046), tumor grade(Grade 2: HR:1.124, 95% CI: 0.953-1.326, P-value: 0.164; Grade 3: HR:1.984, 95% CI: 1.574-2.501, P-value < 0.001; Grade 4: HR:2.119, 95% CI: 1.115-4.027, P-value: 0.022), surgery(Liver Resection: HR:0.193, 95% CI: 0.160-0.234, P-value < 0.001; Liver Transplant: HR:0.102, 95% CI: 0.072-0.145, P-value < 0.001), chemotherapy(HR:0.561, 95% CI: 0.471-0.668, P-value < 0.001), and radiotherapy(HR:0.641, 95% CI: 0.463-0.887, P-value:0.007) were independent prognostic factors for patients with ANHC. We developed a nomogram model for predicting the OS and CSS of patients with ANHC, with a good predictive performance.

Causal relationship between circulating glutamine levels and idiopathic pulmonary fibrosis: a two-sample mendelian randomization study.

BACKGROUND: Idiopathic pulmonary fibrosis (IPF) is a progressive and debilitating respiratory disease with a median survival of less than 5 years. In recent years, glutamine has been reported to be involved in the regulation of collagen deposition and cell proliferation in fibroblasts, thereby influencing the progression of IPF. However, the relationships between glutamine and the incidence, progression, and treatment response of IPF remain unclear. Our study aimed to investigate the relationship between circulating glutamine levels and IPF, as well as its potential as a therapeutic target. METHODS: We performed a comprehensive Mendelian Randomization (MR) analysis using the most recent genome-wide association study summary-level data. A total of 32 single nucleotide polymorphisms significantly correlated to glutamine levels were identified as instrumental variables. Eight MR analysis methods, including inverse variance weighted, MR-Egger, weighted median, weighted mode, constrained maximum likelihood, contamination mixture, robust adjusted profile score, and debiased inverse-variance weighted method, were used to assess the relationship between glutamine levels with IPF. RESULTS: The inverse variance weighted analysis revealed a significant inverse correlation between glutamine levels and IPF risk (Odds Ratio = 0.750; 95% Confidence Interval : 0.592-0.951; P = 0.017). Sensitivity analyses, including MR-Egger regression and MR-PRESSO global test, confirmed the robustness of our findings, with no evidence of horizontal pleiotropy or heterogeneity. CONCLUSION: Our study provides novel evidence for a causal relationship between lower circulating glutamine levels and increased risk of IPF. This finding may contribute to the early identification of high-risk individuals for IPF, disease monitoring, and development of targeted therapeutic strategies.

CBFB cooperates with p53 to maintain TAp73 expression and suppress breast cancer.

The CBFB gene is frequently mutated in several types of solid tumors. Emerging evidence suggests that CBFB is a tumor suppressor in breast cancer. However, our understanding of the tumor suppressive function of CBFB remains incomplete. Here, we analyze genetic interactions between mutations of CBFB and other highly mutated genes in human breast cancer datasets and find that CBFB and TP53 mutations are mutually exclusive, suggesting a functional association between CBFB and p53. Integrated genomic studies reveal that TAp73 is a common transcriptional target of CBFB and p53. CBFB cooperates with p53 to maintain TAp73 expression, as either CBFB or p53 loss leads to TAp73 depletion. TAp73 re-expression abrogates the tumorigenic effect of CBFB deletion. Although TAp73 loss alone is insufficient for tumorigenesis, it enhances the tumorigenic effect of NOTCH3 overexpression, a downstream event of CBFB loss. Immunohistochemistry shows that p73 loss is coupled with higher proliferation in xenografts. Moreover, TAp73 loss-of-expression is a frequent event in human breast cancer tumors and cell lines. Together, our results significantly advance our understanding of the tumor suppressive functions of CBFB and reveal a mechanism underlying the communication between the two tumor suppressors CBFB and p53.

TFH cells depend on Tcf1-intrinsic HDAC activity to suppress CTLA4 and guard B-cell help function.

Precise regulation of coinhibitory receptors is essential for maintaining immune tolerance without interfering with protective immunity, yet the mechanism underlying such a balanced act remains poorly understood. In response to protein immunization, T follicular helper (TFH) cells lacking Tcf1 and Lef1 transcription factors were phenotypically normal but failed to promote germinal center formation and antibody production. Transcriptomic profiling revealed that Tcf1/Lef1-deficient TFH cells aberrantly up-regulated CTLA4 and LAG3 expression, and treatment with anti-CTLA4 alone or combined with anti-LAG3 substantially rectified B-cell help defects by Tcf1/Lef1-deficient TFH cells. Mechanistically, Tcf1 and Lef1 restrain chromatin accessibility at the Ctla4 and Lag3 loci. Groucho/Tle corepressors, which are known to cooperate with Tcf/Lef factors, were essential for TFH cell expansion but dispensable for repressing coinhibitory receptors. In contrast, mutating key amino acids in histone deacetylase (HDAC) domain in Tcf1 resulted in CTLA4 derepression in TFH cells. These findings demonstrate that Tcf1-instrinsic HDAC activity is necessary for preventing excessive CTLA4 induction in protein immunization-elicited TFH cells and hence guarding their B-cell help function.

Development and Validation of an Interpretable Conformal Predictor to Predict Sepsis Mortality Risk: Retrospective Cohort Study.

BACKGROUND: Early and reliable identification of patients with sepsis who are at high risk of mortality is important to improve clinical outcomes. However, 3 major barriers to artificial intelligence (AI) models, including the lack of interpretability, the difficulty in generalizability, and the risk of automation bias, hinder the widespread adoption of AI models for use in clinical practice. OBJECTIVE: This study aimed to develop and validate (internally and externally) a conformal predictor of sepsis mortality risk in patients who are critically ill, leveraging AI-assisted prediction modeling. The proposed approach enables explaining the model output and assessing its confidence level. METHODS: We retrospectively extracted data on adult patients with sepsis from a database collected in a teaching hospital at Beth Israel Deaconess Medical Center for model training and internal validation. A large multicenter critical care database from the Philips eICU Research Institute was used for external validation. A total of 103 clinical features were extracted from the first day after admission. We developed an AI model using gradient-boosting machines to predict the mortality risk of sepsis and used Mondrian conformal prediction to estimate the prediction uncertainty. The Shapley additive explanation method was used to explain the model. RESULTS: A total of 16,746 (80%) patients from Beth Israel Deaconess Medical Center were used to train the model. When tested on the internal validation population of 4187 (20%) patients, the model achieved an area under the receiver operating characteristic curve of 0.858 (95% CI 0.845-0.871), which was reduced to 0.800 (95% CI 0.789-0.811) when externally validated on 10,362 patients from the Philips eICU database. At a specified confidence level of 90% for the internal validation cohort the percentage of error predictions (n=438) out of all predictions (n=4187) was 10.5%, with 1229 (29.4%) predictions requiring clinician review. In contrast, the AI model without conformal prediction made 1449 (34.6%) errors. When externally validated, more predictions (n=4004, 38.6%) were flagged for clinician review due to interdatabase heterogeneity. Nevertheless, the model still produced significantly lower error rates compared to the point predictions by AI (n=1221, 11.8% vs n=4540, 43.8%). The most important predictors identified in this predictive model were Acute Physiology Score III, age, urine output, vasopressors, and pulmonary infection. Clinically relevant risk factors contributing to a single patient were also examined to show how the risk arose. CONCLUSIONS: By combining model explanation and conformal prediction, AI-based systems can be better translated into medical practice for clinical decision-making.

Enhancing Patient Selection in Sepsis Clinical Trials Design Through an AI Enrichment Strategy: Algorithm Development and Validation.

BACKGROUND: Sepsis is a heterogeneous syndrome, and enrollment of more homogeneous patients is essential to improve the efficiency of clinical trials. Artificial intelligence (AI) has facilitated the identification of homogeneous subgroups, but how to estimate the uncertainty of the model outputs when applying AI to clinical decision-making remains unknown. OBJECTIVE: We aimed to design an AI-based model for purposeful patient enrollment, ensuring that a patient with sepsis recruited into a trial would still be persistently ill by the time the proposed therapy could impact patient outcome. We also expected that the model could provide interpretable factors and estimate the uncertainty of the model outputs at a customized confidence level. METHODS: In this retrospective study, 9135 patients with sepsis requiring vasopressor treatment within 24 hours after sepsis onset were enrolled from Beth Israel Deaconess Medical Center. This cohort was used for model development, and 10-fold cross-validation with 50 repeats was used for internal validation. In total, 3743 patients with sepsis from the eICU Collaborative Research Database were used as the external validation cohort. All included patients with sepsis were stratified based on disease progression trajectories: rapid death, recovery, and persistent ill. A total of 148 variables were selected for predicting the 3 trajectories. Four machine learning algorithms with 3 different setups were used. We estimated the uncertainty of the model outputs using conformal prediction (CP). The Shapley Additive Explanations method was used to explain the model. RESULTS: The multiclass gradient boosting machine was identified as the best-performing model with good discrimination and calibration performance in both validation cohorts. The mean area under the receiver operating characteristic curve with SD was 0.906 (0.018) for rapid death, 0.843 (0.008) for recovery, and 0.807 (0.010) for persistent ill in the internal validation cohort. In the external validation cohort, the mean area under the receiver operating characteristic curve (SD) was 0.878 (0.003) for rapid death, 0.764 (0.008) for recovery, and 0.696 (0.007) for persistent ill. The maximum norepinephrine equivalence, total urine output, Acute Physiology Score III, mean systolic blood pressure, and the coefficient of variation of oxygen saturation contributed the most. Compared to the model without CP, using the model with CP at a mixed confidence approach reduced overall prediction errors by 27.6% (n=62) and 30.7% (n=412) in the internal and external validation cohorts, respectively, as well as enabled the identification of more potentially persistent ill patients. CONCLUSIONS: The implementation of our model has the potential to reduce heterogeneity and enroll more homogeneous patients in sepsis clinical trials. The use of CP for estimating the uncertainty of the model outputs allows for a more comprehensive understanding of the model's reliability and assists in making informed decisions based on the predicted outcomes.