Hong Zheng

Neutrophils, the most abundant leukocytes in human blood, have multifaceted roles in health and disease. Here we assembled a single-cell atlas of human neutrophils (SCAHN) by integrating transcriptome data from 1 million human neutrophils from 781 samples and 43 independent datasets. Using SCAHN, we identified multiple neutrophil subsets with distinct functions across tissues and robust changes in gene expression and subset proportions in infection, cancer, immune suppression and sex differences at single-cell resolution. Cross-species reference mapping of mouse and nonhuman primate neutrophils onto SCAHN revealed both conserved and species-specific features. Using 17,023 bulk blood transcriptome profiles from 40 datasets, we found that neutrophil-specific signatures predicted mortality in acutely ill patients at presentation, were associated with cancer diagnosis and survival, and correlated with metabolic dysfunction and long-term all-cause mortality. Our study highlights how neutrophil biology can improve immune risk stratification and develop effective diagnostic and therapeutic strategies.

The hyperinflammatory syndromes in critically ill patients, including trauma, sepsis, and acute lung injury, are characterized by dysregulated neutrophil responses that contribute to tissue damage and poor outcomes. Using murine models of cytokine storm induced by trauma and lung injury, we identified transforming growth factor β (TGFβ) as a central regulator of immune checkpoint in neutrophils. TGFβ signaling modulates neutrophil activation and upregulates the expression of programmed death-ligand 1 (PDL1). Disruption of TGFβ signaling during hyperinflammation restores the migratory capacity of neutrophils but leads to excessive activation, severe pulmonary tissue damage, and increased susceptibility to spontaneous bacterial infection in the lung. Mechanistically, PDL1 expression alters neutrophil behavior within lung capillaries, promoting intravascular clustering and restricting tissue infiltration. Targeted deletion of PDL1 in neutrophils reverses hyperinflammation-induced clustering, restores effective trafficking to infectious foci, and enhances host-protective immune function while limiting pathological neutrophil hyperactivation. These findings define a TGFβ-PDL1 regulatory axis that restrains the pathogenicity of neutrophils during hyperinflammation, revealing a checkpoint mechanism that balances host defense and tissue integrity.

Background: Post Acute Sequelae of COVID-19 (PASC), also referred to as Long COVID, is an infection-associated chronic syndrome with heterogenous symptom profiles that occurs in a subset of people following SARS-CoV-2 infection. Despite proposed viral persistence mechanisms, no therapeutic benefit was observed in two randomized placebo-controlled trials of nirmatrelvir/ritonavir (NMV/r) in adults with Long COVID, including the Selective Trial of Paxlovid for PASC (STOP-PASC) and PAX LC. This systems immunology analysis aimed to characterize immune profiles of participants during clinical trial intervention, identify biomarkers associated with patient-reported outcomes, and investigate potential mechanisms underlying Long COVID. Methods: We performed comprehensive immunological profiling of 152 STOP-PASC trial participants using plasma proteomics (Olink® Explore HT 5400 panel), autoantigen arrays, viral serology, and microclot assays at baseline, day 15, and week 10. We assessed associations between immune features and patient-reported outcomes. We also conducted meta-analysis of nine independent Long COVID proteomics cohorts (n=590 total samples) to identify conserved inflammatory signatures. Results: NMV/r treatment at day 15 compared with baseline induced transient changes in plasma proteins that normalized by week 10, primarily impacting myeloid cell/monocyte, lysosome, and complement activation pathways. Cardiovascular symptoms were negatively associated with SARS-CoV-2 antibody levels at baseline. No widespread differences in autoantibody profiles, Epstein-Barr virus (EBV) reactivation, or microclotting were observed between STOP-PASC Long COVID participants, pre-pandemic controls, and individuals without Long COVID. Meta-analysis of publicly available Olink® data from Long COVID cohorts identified a conserved 60-protein Long COVID Signature (LCS) score revealing multi-compartment immune activation involving monocyte, neutrophil, and T/NK cell modules. Conclusions: These findings advance our understanding of Long COVID immunology and may help direct future proteomic biomarker endpoints for Long COVID clinical trials.

Critical care syndromes such as sepsis, acute respiratory distress syndrome (ARDS) and trauma continue to have unacceptably high morbidity and mortality, with progress limited by the inherent heterogeneity within syndromic illnesses. Although numerous immune endotypes have been proposed for sepsis and critical care, the similarities and differences between these endotypes remain unclear, hindering clinical translation. The SUBSPACE consortium is an international consortium that aims to advance precision medicine in critical care through the sharing of transcriptomic data. Here, evaluating the overlap of existing immune endotypes in sepsis across >7,074 samples from 37 independent cohorts, we developed cell-type-specific gene expression signatures to quantify dysregulation within immune compartments. Myeloid and lymphoid dysregulation were associated with disease severity and mortality across all cohorts. Importantly, this dysregulation was also observed in patients with ARDS, trauma and burns, suggesting a conserved mechanism across various critical illness syndromes. Moreover, analysis of randomized controlled trial data revealed that myeloid and lymphoid dysregulation are associated with differential mortality in patients treated with anakinra in the SAVE-MORE trial (n = 452) and corticosteroids in the VICTAS (n = 89) and VANISH (n = 117) trials, underscoring their prognostic and therapeutic implications. In conclusion, our proposed immunology-based framework for quantifying cellular compartment dysregulation offers a potentially valuable tool for understanding immune dysregulation in critical illness with prognostic and therapeutic significance.

Viral pandemics and epidemics pose a significant global threat, with emerging and re-emerging viruses responsible for four pandemics in the 21st century alone. While macaques have been utilized as a model for understanding viral disease in a controlled setting, it remains unclear how conserved the antiviral responses to diverse viruses are between macaques and humans. To address this critical knowledge gap, we conducted a comprehensive cross-species analysis of transcriptomic data from over 6000 blood samples from macaques and humans infected with one of 31 viruses, including Lassa, Ebola, Marburg, Zika, and dengue. Our findings demonstrate that irrespective of primate or viral species, there are conserved antiviral responses which are consistent regardless of infection phase (acute, chronic, or latent) and viral genome type (DNA or RNA viruses). Moreover, by leveraging longitudinal data from experimental challenges, we identified virus-specific response dynamics such as host responses to Coronaviridae and Orthomyxoviridae infections peaking 1-3 days earlier than responses to Filoviridae and Arenaviridae viral infections. Additionally, through comparative analysis of immune responses across viruses, we identified a unique enrichment of lymphoid cellular response modules in macaque Flaviviridae infection that persists in human responses to dengue. Our results underscore macaque studies as a powerful tool for gaining new insights into viral pathogenesis and immune responses that translate to humans, which can inform viral therapeutic development and enable pandemic preparedness. One sentence summary: Using longitudinal macaque viral challenge studies, we identified shared and virus-specific responses to infection that replicate in human viral disease - thereby demonstrating the utility of macaque models of viral infection to understand antiviral biology and for pandemic preparedness.

The human leukocyte antigen (HLA) locus plays a central role in adaptive immune function and has significant clinical implications for tissue transplant compatibility and allelic disease associations. Studies using bulk-cell RNA sequencing have demonstrated that HLA transcription may be regulated in an allele-specific manner and single-cell RNA sequencing (scRNA-seq) has the potential to better characterize these expression patterns. However, quantification of allele-specific expression (ASE) for HLA loci requires sample-specific reference genotyping due to extensive polymorphism. While genotype prediction from bulk RNA sequencing is well described, the feasibility of predicting HLA genotypes directly from single-cell data is unknown. Here we evaluate and expand upon several computational HLA genotyping tools by comparing predictions from human single-cell data to gold-standard, molecular genotyping. The highest 2-field accuracy averaged across all loci was 76% by arcasHLA and increased to 86% using a composite model of multiple genotyping tools. We also developed a highly accurate model (AUC 0.93) for predicting HLA-DRB345 copy number in order to improve genotyping accuracy of the HLA-DRB locus. Genotyping accuracy improved with read depth and was reproducible at repeat sampling. Using a metanalytic approach, we also show that HLA genotypes from PHLAT and OptiType can generate ASE ratios that are highly correlated (R ² = 0.8 and 0.94, respectively) with those derived from gold-standard genotyping. Highlights: - Benchmarking of HLA genotype prediction accuracy from 5’- and 3’-based single-cell RNA-seq data compared to molecular HLA genotyping - Quantification of transcript coverage by 5’- and 3’-based single cell sequencing methods - Accurate prediction of complex HLA-DRB345 copy numbers using supervised learning - Balancing accuracy and performance through composite HLA genotyping - Meta-analytic approach to summarizing allele-specific expression of HLA genotypes at single-cell level

RNA sequencing has emerged as a promising approach in cancer prognosis as sequencing data becomes more easily and affordably accessible. However, it remains challenging to build good predictive models especially when the sample size is limited and the number of features is high, which is a common situation in biomedical settings. To address these limitations, we propose a meta-learning framework based on neural networks for survival analysis and evaluate it in a genomic cancer research setting. We demonstrate that, compared to regular transfer-learning, meta-learning is a significantly more effective paradigm to leverage high-dimensional data that is relevant but not directly related to the problem of interest. Specifically, meta-learning explicitly constructs a model, from abundant data of relevant tasks, to learn a new task with few samples effectively. For the application of predicting cancer survival outcome, we also show that the meta-learning framework with a few samples is able to achieve competitive performance with learning from scratch with a significantly larger number of samples. Finally, we demonstrate that the meta-learning model implicitly prioritizes genes based on their contribution to survival prediction and allows us to identify important pathways in cancer.

SARS-CoV-2 pandemic, the fourth pandemic of the decade, has underscored gaps in global pandemic preparedness and the need for generalizable tests to avert overwhelming healthcare systems worldwide, irrespective of a virus. We integrated 4,780 blood transcriptome profiles from patients infected with one of 16 viruses across 34 independent cohorts from 18 countries, and 71 scRNA-seq profiles of 264,224 immune cells across three independent cohorts. We found a myeloid cell-dominated conserved host response associated with severity. It showed increased hematopoiesis, myelopoiesis, and myeloid-derived suppressor cells with increased severity. We identified four gene modules that delineate distinct trajectories associated with mild and severe outcomes, and show the interferon response was decoupled from protective host response during severe viral infection. These modules distinguished non-severe from severe viral infection with clinically useful accuracy. Together, our findings provide insights into immune response dynamics during viral infection, and identify factors that may influence patient outcomes.

Background: As missing values are frequently present in genomic data, practical methods to handle missing data are necessary for downstream analyses that require complete data sets. State-of-the-art imputation techniques, including methods based on singular value decomposition and K-nearest neighbors, can be computationally expensive for large data sets and it is difficult to modify these algorithms to handle certain cases not missing at random. Results: In this work, we use a deep-learning framework based on the variational auto-encoder (VAE) for genomic missing value imputation and demonstrate its effectiveness in transcriptome and methylome data analysis. We show that in the vast majority of our testing scenarios, VAE achieves similar or better performances than the most widely used imputation standards, while having a computational advantage at evaluation time. When dealing with data missing not at random (e.g., few values are missing), we develop simple yet effective methodologies to leverage the prior knowledge about missing data. Furthermore, we investigate the effect of varying latent space regularization strength in VAE on the imputation performances and, in this context, show why VAE has a better imputation capacity compared to a regular deterministic auto-encoder. Conclusions: We describe a deep learning imputation framework for transcriptome and methylome data using a VAE and show that it can be a preferable alternative to traditional methods for data imputation, especially in the setting of large-scale data and certain missing-not-at-random scenarios.