IL-7-dependent and -independent lineages of IL-7R-dependent human T cells.

Infants with biallelic IL7R loss-of-function variants have severe combined immune deficiency (SCID) characterized by the absence of autologous T lymphocytes, but normal counts of circulating B and NK cells (T-B+NK+ SCID). We report 6 adults (aged 22 to 59 years) from 4 kindreds and 3 ancestries (Colombian, Israeli Arab, Japanese) carrying homozygous IL7 loss-of-function variants resulting in combined immunodeficiency (CID). Deep immunophenotyping revealed relatively normal counts and/or proportions of myeloid, B, NK, and innate lymphoid cells. By contrast, the patients had profound T cell lymphopenia, with low proportions of innate-like adaptive mucosal-associated invariant T and invariant NK T cells. They also had low blood counts of T cell receptor (TCR) excision circles, recent thymic emigrant T cells and naive CD4+ T cells, and low overall TCR repertoire diversity, collectively indicating impaired thymic output. The proportions of effector memory CD4+ and CD8+ T cells were high, indicating IL-7-independent homeostatic T cell proliferation in the periphery. Intriguingly, the proportions of other T cell subsets, including TCRγδ+ T cells and some TCRαß+ T cell subsets (including Th1, Tfh, and Treg) were little affected. Peripheral CD4+ T cells displayed poor proliferation, but normal cytokine production upon stimulation with mitogens in vitro. Thus, inherited IL-7 deficiency impairs T cell development less severely and in a more subset-specific manner than IL-7R deficiency. These findings suggest that another IL-7R-binding cytokine, possibly thymic stromal lymphopoietin, governs an IL-7-independent pathway of human T cell development.

Human-augmented large language model-driven selection of glutathione peroxidase 4 as a candidate blood transcriptional biomarker for circulating erythroid cells.

The identification of optimal candidate genes from large-scale blood transcriptomic data is crucial for developing targeted assays to monitor immune responses. Here, we introduce a novel, optimized large language model (LLM)-based approach for prioritizing candidate biomarkers from blood transcriptional modules. Focusing on module M14.51 from the BloodGen3 repertoire, we implemented a multi-step LLM-driven workflow. Initial high-throughput screening used GPT-4, Claude 3, and Claude 3.5 Sonnet to score and rank the module's constituent genes across six criteria. Top candidates then underwent high-resolution scoring using Consensus GPT, with concurrent manual fact-checking and, when needed, iterative refinement of the scores based on user feedback. Qualitative assessment of literature-based narratives and analysis of reference transcriptome data further refined the selection process. This novel multi-tiered approach consistently identified Glutathione Peroxidase 4 (GPX4) as the top candidate gene for module M14.51. GPX4's role in oxidative stress regulation, its potential as a future drug target, and its expression pattern across diverse cell types supported its selection. The incorporation of reference transcriptome data further validated GPX4 as the most suitable candidate for this module. This study presents an advanced LLM-driven workflow with a novel optimized scoring strategy for candidate gene prioritization, incorporating human-in-the-loop augmentation. The approach identified GPX4 as a key gene in the erythroid cell-associated module M14.51, suggesting its potential utility for biomarker discovery and targeted assay development. By combining AI-driven literature analysis with iterative human expert validation, this method leverages the strengths of both artificial and human intelligence, potentially contributing to the development of biologically relevant and clinically informative targeted assays. Further validation studies are needed to confirm the broader applicability of this human-augmented AI approach.

Autoinflammation in patients with leukocytic CBL loss of heterozygosity is caused by constitutive ERK-mediated monocyte activation.

Patients heterozygous for germline CBL loss-of-function (LOF) variants can develop myeloid malignancy, autoinflammation, or both, if some or all of their leukocytes become homozygous for these variants through somatic loss of heterozygosity (LOH) via uniparental isodisomy. We observed an upregulation of the inflammatory gene expression signature in whole blood from these patients, mimicking monogenic inborn errors underlying autoinflammation. Remarkably, these patients had constitutively activated monocytes that secreted 10 to 100 times more inflammatory cytokines than those of healthy individuals and CBL LOF heterozygotes without LOH. CBL-LOH hematopoietic stem and progenitor cells (HSPCs) outgrew the other cells, accounting for the persistence of peripheral monocytes homozygous for the CBL LOF variant. ERK pathway activation was required for the excessive production of cytokines by both resting and stimulated CBL-LOF monocytes, as shown in monocytic cell lines. Finally, we found that about 1 in 10,000 individuals in the UK Biobank were heterozygous for CBL LOF variants and that these carriers were at high risk of hematological and inflammatory conditions.

Your Blood is Out for Delivery: Considerations of Shipping Time and Temperature on Degradation of RNA from Stabilized Whole Blood.

Remote research studies are an invaluable tool for reaching populations in geographical regions with limited access to large medical centers or universities. To expand the remote study toolkit, we have previously developed homeRNA, which allows for at-home self-collection and stabilization of blood and demonstrated the feasibility of using homeRNA in high temperature climates. Here, we expand upon this work through a systematic study exploring the effects of high temperature on RNA integrity through in-lab and field experiments. Compared to the frozen controls (overall mean RIN of 8.2, n = 8), samples kept at 37°C for 2, 4, and 8 days had mean RINs of 7.6, 5.9, and 5.2 (n = 3), respectively, indicating that typical shipping conditions (~2 days) yield samples suitable for downstream RNA sequencing. Shorter time intervals (6 hours) resulted in minimal RNA degradation (median RIN of 6.4, n = 3) even at higher temperatures (50°C) compared to the frozen control (mean RIN of 7.8, n = 3). Additionally, we shipped homeRNA-stabilized blood from a single donor to 14 different states and back during the summer with continuous temperature probes (7.1 median RIN, n = 42). Samples from all locations were analyzed with 3' mRNA-seq to assess differences in gene counts, with the transcriptomic data suggesting that there was no preferential degradation of transcripts as a result of different shipping times, temperatures, and regions. Overall, our data support that homeRNA can be used in elevated temperature conditions, enabling decentralized sample collection for telemedicine, global health, and clinical research.

Tuberculosis in otherwise healthy adults with inherited TNF deficiency.

Severe defects in human IFNγ immunity predispose individuals to both Bacillus Calmette-Guérin disease and tuberculosis, whereas milder defects predispose only to tuberculosis1. Here we report two adults with recurrent pulmonary tuberculosis who are homozygous for a private loss-of-function TNF variant. Neither has any other clinical phenotype and both mount normal clinical and biological inflammatory responses. Their leukocytes, including monocytes and monocyte-derived macrophages (MDMs) do not produce TNF, even after stimulation with IFNγ. Blood leukocyte subset development is normal in these patients. However, an impairment in the respiratory burst was observed in granulocyte-macrophage colony-stimulating factor (GM-CSF)-matured MDMs and alveolar macrophage-like (AML) cells2 from both patients with TNF deficiency, TNF- or TNFR1-deficient induced pluripotent stem (iPS)-cell-derived GM-CSF-matured macrophages, and healthy control MDMs and AML cells differentiated with TNF blockers in vitro, and in lung macrophages treated with TNF blockers ex vivo. The stimulation of TNF-deficient iPS-cell-derived macrophages with TNF rescued the respiratory burst. These findings contrast with those for patients with inherited complete deficiency of the respiratory burst across all phagocytes, who are prone to multiple infections, including both Bacillus Calmette-Guérin disease and tuberculosis3. Human TNF is required for respiratory-burst-dependent immunity to Mycobacterium tuberculosis in macrophages but is surprisingly redundant otherwise, including for inflammation and immunity to weakly virulent mycobacteria and many other infectious agents.

SARS-CoV-2 brainstem encephalitis in human inherited DBR1 deficiency.

Inherited deficiency of the RNA lariat-debranching enzyme 1 (DBR1) is a rare etiology of brainstem viral encephalitis. The cellular basis of disease and the range of viral predisposition are unclear. We report inherited DBR1 deficiency in a 14-year-old boy who suffered from isolated SARS-CoV-2 brainstem encephalitis. The patient is homozygous for a previously reported hypomorphic and pathogenic DBR1 variant (I120T). Consistently, DBR1 I120T/I120T fibroblasts from affected individuals from this and another unrelated kindred have similarly low levels of DBR1 protein and high levels of RNA lariats. DBR1 I120T/I120T human pluripotent stem cell (hPSC)-derived hindbrain neurons are highly susceptible to SARS-CoV-2 infection. Exogenous WT DBR1 expression in DBR1 I120T/I120T fibroblasts and hindbrain neurons rescued the RNA lariat accumulation phenotype. Moreover, expression of exogenous RNA lariats, mimicking DBR1 deficiency, increased the susceptibility of WT hindbrain neurons to SARS-CoV-2 infection. Inborn errors of DBR1 impair hindbrain neuron-intrinsic antiviral immunity, predisposing to viral infections of the brainstem, including that by SARS-CoV-2.

Human TMEFF1 is a restriction factor for herpes simplex virus in the brain.

Most cases of herpes simplex virus 1 (HSV-1) encephalitis (HSE) remain unexplained1,2. Here, we report on two unrelated people who had HSE as children and are homozygous for rare deleterious variants of TMEFF1, which encodes a cell membrane protein that is preferentially expressed by brain cortical neurons. TMEFF1 interacts with the cell-surface HSV-1 receptor NECTIN-1, impairing HSV-1 glycoprotein D- and NECTIN-1-mediated fusion of the virus and the cell membrane, blocking viral entry. Genetic TMEFF1 deficiency allows HSV-1 to rapidly enter cortical neurons that are either patient specific or derived from CRISPR-Cas9-engineered human pluripotent stem cells, thereby enhancing HSV-1 translocation to the nucleus and subsequent replication. This cellular phenotype can be rescued by pretreatment with type I interferon (IFN) or the expression of exogenous wild-type TMEFF1. Moreover, ectopic expression of full-length TMEFF1 or its amino-terminal extracellular domain, but not its carboxy-terminal intracellular domain, impairs HSV-1 entry into NECTIN-1-expressing cells other than neurons, increasing their resistance to HSV-1 infection. Human TMEFF1 is therefore a host restriction factor for HSV-1 entry into cortical neurons. Its constitutively high abundance in cortical neurons protects these cells from HSV-1 infection, whereas inherited TMEFF1 deficiency renders them susceptible to this virus and can therefore underlie HSE.

An interactive web application for exploring systemic lupus erythematosus blood transcriptomic diversity.

In the field of complex autoimmune diseases such as systemic lupus erythematosus (SLE), systems immunology approaches have proven invaluable in translational research settings. Large-scale datasets of transcriptome profiling have been collected and made available to the research community in public repositories, but remain poorly accessible and usable by mainstream researchers. Enabling tools and technologies facilitating investigators' interaction with large-scale datasets such as user-friendly web applications could promote data reuse and foster knowledge discovery. Microarray blood transcriptomic data from the LUPUCE cohort (publicly available on Gene Expression Omnibus, GSE49454), which comprised 157 samples from 62 adult SLE patients, were analyzed with the third-generation (BloodGen3) module repertoire framework, which comprises modules and module aggregates. These well-characterized samples corresponded to different levels of disease activity, different types of flares (including biopsy-proven lupus nephritis), different auto-antibody profiles and different levels of interferon signatures. A web application was deployed to present the aggregate-level, module-level and gene-level analysis results from LUPUCE dataset. Users can explore the similarities and heterogeneity of SLE samples, navigate through different levels of analysis, test hypotheses and generate custom fingerprint grids and heatmaps, which may be used in reports or manuscripts. This resource is available via this link: https://immunology-research.shinyapps.io/LUPUCE/. This web application can be employed as a stand-alone resource to explore changes in blood transcript profiles in SLE, and their relation to clinical and immunological parameters, to generate new research hypotheses.

FLT3L governs the development of partially overlapping hematopoietic lineages in humans and mice.

FMS-related tyrosine kinase 3 ligand (FLT3L), encoded by FLT3LG, is a hematopoietic factor essential for the development of natural killer (NK) cells, B cells, and dendritic cells (DCs) in mice. We describe three humans homozygous for a loss-of-function FLT3LG variant with a history of various recurrent infections, including severe cutaneous warts. The patients' bone marrow (BM) was hypoplastic, with low levels of hematopoietic progenitors, particularly myeloid and B cell precursors. Counts of B cells, monocytes, and DCs were low in the patients' blood, whereas the other blood subsets, including NK cells, were affected only moderately, if at all. The patients had normal counts of Langerhans cells (LCs) and dermal macrophages in the skin but lacked dermal DCs. Thus, FLT3L is required for B cell and DC development in mice and humans. However, unlike its murine counterpart, human FLT3L is required for the development of monocytes but not NK cells.

In search of a function for human type III interferons: insights from inherited and acquired deficits.

The essential and redundant functions of human type I and II interferons (IFNs) have been delineated over the last three decades by studies of patients with inborn errors of immunity or their autoimmune phenocopies, but much less is known about type III IFNs. Patients with cells that do not respond to type III IFNs due to inherited IL10RB deficiency display no overt viral disease, and their inflammatory disease phenotypes can be explained by defective signaling via other interleukine10RB-dependent pathways. Moreover, patients with inherited deficiencies of interferon-stimulated gene factor 3 (ISGF-3) (STAT1, STAT2, IRF9) present viral diseases also seen in patients with inherited deficiencies of the type I IFN receptor (IFNAR1/2). Finally, patients with autoantibodies neutralizing type III IFNs have no obvious predisposition to viral disease. Current findings thus suggest that type III IFNs are largely redundant in humans. The essential functions of human type III IFNs, particularly in antiviral defenses, remain to be discovered.

Human inherited PD-L1 deficiency is clinically and immunologically less severe than PD-1 deficiency.

We previously reported two siblings with inherited PD-1 deficiency who died from autoimmune pneumonitis at 3 and 11 years of age after developing other autoimmune manifestations, including type 1 diabetes (T1D). We report here two siblings, aged 10 and 11 years, with neonatal-onset T1D (diagnosed at the ages of 1 day and 7 wk), who are homozygous for a splice-site variant of CD274 (encoding PD-L1). This variant results in the exclusive expression of an alternative, loss-of-function PD-L1 protein isoform in overexpression experiments and in the patients' primary leukocytes. Surprisingly, cytometric immunophenotyping and single-cell RNA sequencing analysis on blood leukocytes showed largely normal development and transcriptional profiles across lymphoid and myeloid subsets in the PD-L1-deficient siblings, contrasting with the extensive dysregulation of both lymphoid and myeloid leukocyte compartments in PD-1 deficiency. Our findings suggest that PD-1 and PD-L1 are essential for preventing early-onset T1D but that, unlike PD-1 deficiency, PD-L1 deficiency does not lead to fatal autoimmunity with extensive leukocytic dysregulation.

A data browsing application for accessing gene and module-level blood transcriptome profiles of healthy pregnant women from high- and low-resource settings.

Transcriptome profiling data, generated via RNA sequencing, are commonly deposited in public repositories. However, these data may not be easily accessible or usable by many researchers. To enhance data reuse, we present well-annotated, partially analyzed data via a user-friendly web application. This project involved transcriptome profiling of blood samples from 15 healthy pregnant women in a low-resource setting, taken at 6 consecutive time points beginning from the first trimester. Additional blood transcriptome profiles were retrieved from the National Center for Biotechnology Information (NCBI) Gene Expression Omnibus (GEO) public repository, representing a cohort of healthy pregnant women from a high-resource setting. We analyzed these datasets using the fixed BloodGen3 module repertoire. We deployed a web application, accessible at https://thejacksonlaboratory.shinyapps.io/BloodGen3_Pregnancy/which displays the module-level analysis results from both original and public pregnancy blood transcriptome datasets. Users can create custom fingerprint grid and heatmap representations via various navigation options, useful for reports and manuscript preparation. The web application serves as a standalone resource for exploring blood transcript abundance changes during pregnancy. Alternatively, users can integrate it with similar applications developed for earlier publications to analyze transcript abundance changes of a given BloodGen3 signature across a range of disease cohorts. Database URL: https://thejacksonlaboratory.shinyapps.io/BloodGen3_Pregnancy/.

The immunopathological landscape of human pre-TCRα deficiency: From rare to common variants.

We describe humans with rare biallelic loss-of-function PTCRA variants impairing pre-α T cell receptor (pre-TCRα) expression. Low circulating naive αß T cell counts at birth persisted over time, with normal memory αß and high γδ T cell counts. Their TCRα repertoire was biased, which suggests that noncanonical thymic differentiation pathways can rescue αß T cell development. Only a minority of these individuals were sick, with infection, lymphoproliferation, and/or autoimmunity. We also report that 1 in 4000 individuals from the Middle East and South Asia are homozygous for a common hypomorphic PTCRA variant. They had normal circulating naive αß T cell counts but high γδ T cell counts. Although residual pre-TCRα expression drove the differentiation of more αß T cells, autoimmune conditions were more frequent in these patients compared with the general population.

Design of a targeted blood transcriptional panel for monitoring immunological changes accompanying pregnancy.

Background: Immunomodulatory processes exert steering functions throughout pregnancy. Detecting diversions from this physiologic immune clock may help identify pregnant women at risk for pregnancy-associated complications. We present results from a data-driven selection process to develop a targeted panel of mRNAs that may prove effective in detecting pregnancies diverting from the norm. Methods: Based on a de novo dataset from a resource-constrained setting and a dataset from a resource-rich area readily available in the public domain, whole blood gene expression profiles of uneventful pregnancies were captured at multiple time points during pregnancy. BloodGen3, a fixed blood transcriptional module repertoire, was employed to analyze and visualize gene expression patterns in the two datasets. Differentially expressed genes were identified by comparing their abundance to non-pregnant postpartum controls. The selection process for a targeted gene panel considered (i) transcript abundance in whole blood; (ii) degree of correlation with the BloodGen3 module; and (iii) pregnancy biology. Results: We identified 176 transcripts that were complemented with eight housekeeping genes. Changes in transcript abundance were seen in the early stages of pregnancy and similar patterns were observed in both datasets. Functional gene annotation suggested significant changes in the lymphoid, prostaglandin and inflammation-associated compartments, when compared to the postpartum controls. Conclusion: The gene panel presented here holds promise for the development of predictive, targeted, transcriptional profiling assays. Such assays might become useful for monitoring of pregnant women, specifically to detect potential adverse events early. Prospective validation of this targeted assay, in-depth investigation of functional annotations of differentially expressed genes, and assessment of common pregnancy-associated complications with the aim to identify these early in pregnancy to improve pregnancy outcomes are the next steps.

Recombinant IFN-γ1b Treatment in a Patient with Inherited IFN-γ Deficiency.

PURPOSE: Inborn errors of IFN-γ immunity underlie Mendelian susceptibility to mycobacterial disease (MSMD). Twenty-two genes with products involved in the production of, or response to, IFN-γ and variants of which underlie MSMD have been identified. However, pathogenic variants of IFNG encoding a defective IFN-γ have been described in only two siblings, who both underwent hematopoietic stem cell transplantation (HCST). METHODS: We characterized a new patient with MSMD by genetic, immunological, and clinical means. Therapeutic decisions were taken on the basis of these findings. RESULTS: The patient was born to consanguineous Turkish parents and developed bacillus Calmette-Guérin (BCG) disease following vaccination at birth. Whole-exome sequencing revealed a homozygous private IFNG variant (c.224 T > C, p.F75S). Upon overexpression in recipient cells or constitutive expression in the patient's cells, the mutant IFN-γ was produced within the cells but was not correctly folded or secreted. The patient was treated for 6 months with two or three antimycobacterial drugs only and then for 30 months with subcutaneous recombinant IFN-γ1b plus two antimycobacterial drugs. Treatment with IFN-γ1b finally normalized all biological parameters. The patient presented no recurrence of mycobacterial disease or other related infectious diseases. The treatment was well tolerated, without the production of detectable autoantibodies against IFN-γ. CONCLUSION: We describe a patient with a new form of autosomal recessive IFN-γ deficiency, with intracellular, but not extracellular IFN-γ. IFN-γ1b treatment appears to have been beneficial in this patient, with no recurrence of mycobacterial infection over a period of more than 30 months. This targeted treatment provides an alternative to HCST in patients with complete IFN-γ deficiency or at least an option to better control mycobacterial infection prior to HCST.

Human inherited CCR2 deficiency underlies progressive polycystic lung disease.

We describe a human lung disease caused by autosomal recessive, complete deficiency of the monocyte chemokine receptor C-C motif chemokine receptor 2 (CCR2). Nine children from five independent kindreds have pulmonary alveolar proteinosis (PAP), progressive polycystic lung disease, and recurrent infections, including bacillus Calmette Guérin (BCG) disease. The CCR2 variants are homozygous in six patients and compound heterozygous in three, and all are loss-of-expression and loss-of-function. They abolish CCR2-agonist chemokine C-C motif ligand 2 (CCL-2)-stimulated Ca2+ signaling in and migration of monocytic cells. All patients have high blood CCL-2 levels, providing a diagnostic test for screening children with unexplained lung or mycobacterial disease. Blood myeloid and lymphoid subsets and interferon (IFN)-γ- and granulocyte-macrophage colony-stimulating factor (GM-CSF)-mediated immunity are unaffected. CCR2-deficient monocytes and alveolar macrophage-like cells have normal gene expression profiles and functions. By contrast, alveolar macrophage counts are about half. Human complete CCR2 deficiency is a genetic etiology of PAP, polycystic lung disease, and recurrent infections caused by impaired CCL2-dependent monocyte migration to the lungs and infected tissues.

Genome-wide detection of human intronic AG-gain variants located between splicing branchpoints and canonical splice acceptor sites.

Human genetic variants that introduce an AG into the intronic region between the branchpoint (BP) and the canonical splice acceptor site (ACC) of protein-coding genes can disrupt pre-mRNA splicing. Using our genome-wide BP database, we delineated the BP-ACC segments of all human introns and found extreme depletion of AG/YAG in the [BP+8, ACC-4] high-risk region. We developed AGAIN as a genome-wide computational approach to systematically and precisely pinpoint intronic AG-gain variants within the BP-ACC regions. AGAIN identified 350 AG-gain variants from the Human Gene Mutation Database, all of which alter splicing and cause disease. Among them, 74% created new acceptor sites, whereas 31% resulted in complete exon skipping. AGAIN also predicts the protein-level products resulting from these two consequences. We performed AGAIN on our exome/genomes database of patients with severe infectious diseases but without known genetic etiology and identified a private homozygous intronic AG-gain variant in the antimycobacterial gene SPPL2A in a patient with mycobacterial disease. AGAIN also predicts a retention of six intronic nucleotides that encode an in-frame stop codon, turning AG-gain into stop-gain. This allele was then confirmed experimentally to lead to loss of function by disrupting splicing. We further showed that AG-gain variants inside the high-risk region led to misspliced products, while those outside the region did not, by two case studies in genes STAT1 and IRF7. We finally evaluated AGAIN on our 14 paired exome-RNAseq samples and found that 82% of AG-gain variants in high-risk regions showed evidence of missplicing. AGAIN is publicly available from https://hgidsoft.rockefeller.edu/AGAIN and https://github.com/casanova-lab/AGAIN.

Autoantibodies against type I IFNs in humans with alternative NF-κB pathway deficiency.

Patients with autoimmune polyendocrinopathy syndrome type 1 (APS-1) caused by autosomal recessive AIRE deficiency produce autoantibodies that neutralize type I interferons (IFNs)1,2, conferring a predisposition to life-threatening COVID-19 pneumonia3. Here we report that patients with autosomal recessive NIK or RELB deficiency, or a specific type of autosomal-dominant NF-κB2 deficiency, also have neutralizing autoantibodies against type I IFNs and are at higher risk of getting life-threatening COVID-19 pneumonia. In patients with autosomal-dominant NF-κB2 deficiency, these autoantibodies are found only in individuals who are heterozygous for variants associated with both transcription (p52 activity) loss of function (LOF) due to impaired p100 processing to generate p52, and regulatory (IκBδ activity) gain of function (GOF) due to the accumulation of unprocessed p100, therefore increasing the inhibitory activity of IκBδ (hereafter, p52LOF/IκBδGOF). By contrast, neutralizing autoantibodies against type I IFNs are not found in individuals who are heterozygous for NFKB2 variants causing haploinsufficiency of p100 and p52 (hereafter, p52LOF/IκBδLOF) or gain-of-function of p52 (hereafter, p52GOF/IκBδLOF). In contrast to patients with APS-1, patients with disorders of NIK, RELB or NF-κB2 have very few tissue-specific autoantibodies. However, their thymuses have an abnormal structure, with few AIRE-expressing medullary thymic epithelial cells. Human inborn errors of the alternative NF-κB pathway impair the development of AIRE-expressing medullary thymic epithelial cells, thereby underlying the production of autoantibodies against type I IFNs and predisposition to viral diseases.

Harnessing large language models (LLMs) for candidate gene prioritization and selection.

BACKGROUND: Feature selection is a critical step for translating advances afforded by systems-scale molecular profiling into actionable clinical insights. While data-driven methods are commonly utilized for selecting candidate genes, knowledge-driven methods must contend with the challenge of efficiently sifting through extensive volumes of biomedical information. This work aimed to assess the utility of large language models (LLMs) for knowledge-driven gene prioritization and selection. METHODS: In this proof of concept, we focused on 11 blood transcriptional modules associated with an Erythroid cells signature. We evaluated four leading LLMs across multiple tasks. Next, we established a workflow leveraging LLMs. The steps consisted of: (1) Selecting one of the 11 modules; (2) Identifying functional convergences among constituent genes using the LLMs; (3) Scoring candidate genes across six criteria capturing the gene's biological and clinical relevance; (4) Prioritizing candidate genes and summarizing justifications; (5) Fact-checking justifications and identifying supporting references; (6) Selecting a top candidate gene based on validated scoring justifications; and (7) Factoring in transcriptome profiling data to finalize the selection of the top candidate gene. RESULTS: Of the four LLMs evaluated, OpenAI's GPT-4 and Anthropic's Claude demonstrated the best performance and were chosen for the implementation of the candidate gene prioritization and selection workflow. This workflow was run in parallel for each of the 11 erythroid cell modules by participants in a data mining workshop. Module M9.2 served as an illustrative use case. The 30 candidate genes forming this module were assessed, and the top five scoring genes were identified as BCL2L1, ALAS2, SLC4A1, CA1, and FECH. Researchers carefully fact-checked the summarized scoring justifications, after which the LLMs were prompted to select a top candidate based on this information. GPT-4 initially chose BCL2L1, while Claude selected ALAS2. When transcriptional profiling data from three reference datasets were provided for additional context, GPT-4 revised its initial choice to ALAS2, whereas Claude reaffirmed its original selection for this module. CONCLUSIONS: Taken together, our findings highlight the ability of LLMs to prioritize candidate genes with minimal human intervention. This suggests the potential of this technology to boost productivity, especially for tasks that require leveraging extensive biomedical knowledge.