Slfn4-mediated Stat3 signaling promotes suppressive bone marrow monocytes in a murine second-hit sepsis model

Abstract

Background

Sepsis survivors who develop recurrent or secondary infections often exhibit prolonged immunosuppression and impaired pathogen clearance, yet the underlying mechanisms remain poorly defined and targeted therapies are limited.

Methods

We established a murine second-hit sepsis model to examine immune remodeling during the immunosuppressive phase after an initial inflammatory insult. Bone marrow immune remodeling was characterized by flow cytometry and single-cell RNA sequencing. Purified myeloid subsets were subjected to functional assays, transcriptomic analyses, and molecular studies to define how Slfn4 regulates the suppressive program of monocytic myeloid-derived suppressor cell-like (M-MDSC-like) cells.

Results

We found that Slfn4 marks an immunosuppressive M-MDSC-like subset in the bone marrow during second-hit sepsis, and our data support a role for the Slfn4–Stat3 axis in maintaining its suppressive phenotype. Genetic silencing of Slfn4 or pharmacologic inhibition of Stat3 was associated with reduced M-MDSC-like cell abundance, partially restored T-cell function, and improved survival. Mechanistically, our data supports a model in which SLFN4 enhances Stat3 activation in association with post-transcriptional repression of Socs3, potentially through interaction with an AU-rich element within the Socs3 3′-UTR. In addition, sildenafil, a PDE5 inhibitor, decreased M-MDSC-like cell abundance and enhanced bacterial clearance in vivo.

Conclusions

These findings implicate the Slfn4–Stat3 axis in bone marrow-associated immunosuppression in this murine second-hit sepsis model and support further investigation of this pathway in the late immunosuppressive phase of sepsis.

Supplementary Information

The online version contains supplementary material available at 10.1186/s10020-026-01515-3.

Introduction

Despite advances in critical care, sepsis remains a pervasive therapeutic challenge, characterized by a dysregulated immune response that extends well beyond the acute inflammatory phase (Kox et al. 2025). A subset of sepsis survivors may enter a stage of persistent inflammation, immunosuppression and catabolism syndrome (PICS) (Poll et al. 2021). Because this stage of sepsis lacks universally accepted diagnostic criteria, no standardized animal model fully captures its complexity (Yang et al. 2014). This gap has hindered efforts to define the mechanisms underlying post-sepsis immune dysfunction. Recently, experimental paradigms have increasingly turned to a “second-hit" strategy, in which a sub-lethal inflammatory stimulus is superimposed once the first-hit cytokine storm receded (Nascimento et al. 2021; Bauhofer et al. 2006). Although such models do not recapitulate the full heterogeneity of clinical secondary infections, they provide a useful framework for studying the transition from early inflammation to sustained immune suppression.

Among the cellular mediators implicated in sepsis-associated immunosuppression, myeloid-derived suppressor cells (MDSCs) are considered important contributors to immune paralysis (Hegde et al. 2021). MDSCs are generally classified into polymorphonuclear MDSCs (PMN-MDSCs), characterized in humans as CD45⁺CD11b⁺CD15⁺CD14⁻CD66b⁺LOX-1⁺, and monocytic MDSCs (M-MDSCs), defined as CD45⁺CD11b + CD14⁺HLA-DR⁻. In mice, the corresponding populations are commonly described as CD45⁺CD11b⁺Ly6G⁺Ly6C⁻ and CD45⁺CD11b⁺Ly6C⁺Ly6G⁻, respectively (Veglia et al. 2021). Previous studies have shown that expansion of MDSCs reflects the immune status of critically ill patients. For instance, a CD15^int M-MDSC subpopulation with high PD-L1 expression has been reported to predict mortality (Schrijver et al. 2024). Functionally, these cells suppress T-cell responses through multiple mechanisms, including L-arginine depletion, production of reactive oxygen and nitrogen species, and release of PD-L1⁺ extracellular vesicles (Fortin et al. 2024; Wang et al. 2024). However, identifying suppressive myeloid populations in inflammatory settings remains challenging, and markers that both predict function and define the regulatory program of these cells are still lacking.

The Schlafen (Slfn) family comprises evolutionarily conserved genes involved in cell proliferation, differentiation, and immune regulation. Slfn expression is predominantly regulated by upstream signals such as type I interferons (Jo and Pommier 2022). Recently, Slfn4 has been characterized as a marker for MDSC in gastric metaplasia and a participant in myeloid differentiation (Ding et al. 2022). In our single-cell RNA sequencing analysis of bone marrow cells (BMCs) from septic mice, Slfn4 emerged as one of the most prominently upregulated transcripts in a monocyte-like suppressive population. This finding raised an important question: does Slfn4 merely mark these cells, or does it actively drive their expansion and immunosuppressive function? Addressing this question is essential for establishing Slfn4's potential as a therapeutic target for sepsis.

Here, we used a murine second-hit sepsis model in combination with single-cell RNA sequencing and functional assays to define bone marrow myeloid populations associated with prolonged immune suppression. We hypothesized that a Slfn4-enriched monocytic population contributes functionally to this suppressive state rather than serving only as a marker, potentially through Stat3-related signaling. We therefore examined the role of Slfn4 in the suppressive phenotype of these cells, explored its relationship with Stat3 signaling, and evaluated whether pharmacological modulation of this pathway could attenuate suppressive myeloid dysfunction in vivo.

Materials and methods

Animals

Male C57BL/6 J mice (6–8 weeks old) were purchased from Shanghai Jiesijie Experimental Animal Co., Ltd. All mice were housed in a specific pathogen-free (SPF) facility under standard conditions (12-h light/dark cycle, 22 ± 2 °C, 50–60% humidity) and acclimatized for at least one week before experimentation. All animal procedures were approved by the Institutional Animal Care and Use Committee of Naval Medical University.

Second-hit sepsis model

To establish the second-hit sepsis model, mice were assigned to four groups: Control, LPS-only, E. coli, and second-hit. Mice in the Control group received PBS pretreatment followed by PBS challenge. Mice in the LPS-only group received LPS treatment. Mice in the E. coli group received PBS pretreatment followed by live E. coli challenge. Mice in the second-hit group received LPS pretreatment followed by live E. coli challenge.

For LPS priming, mice received an intraperitoneal injection of lipopolysaccharide (LPS; Escherichia coli O111:B4, Sigma-Aldrich) at 10 mg/kg. Forty-eight hours later, mice were challenged intraperitoneally with live E. coli. The 48-h interval between LPS priming and bacterial challenge was selected to model a relatively hyporesponsive post-inflammatory state, as supported by transient cytokine elevation after LPS that returned toward baseline by 48 h. Mice were monitored for survival, body weight, rectal temperature, and clinical status at the indicated time points after each intervention. Clinical status was assessed using the Murine Sepsis Score (MSS) (Mai et al. 2018). Mice reaching a predefined humane endpoint (e.g., total score ≥ 12 or a score of 4 in any single parameter) were immediately euthanized, and euthanasia was recorded in place of spontaneous death as an experimental endpoint.

Bacterial culture and inoculum preparation

Escherichia coli (ATCC 25922) was cultured in Luria–Bertani (LB) broth at 37 °C with shaking to logarithmic growth phase. Bacteria were collected by centrifugation, washed, and resuspended in sterile PBS for intraperitoneal injection. Bacterial concentration was estimated by optical density (OD₆₀₀) and adjusted to the desired concentration in PBS. The final inoculum for the main experiments was approximately 5 × 10⁷ CFU (corresponding to approximately 0.06 OD₆₀₀) per mouse, on the basis of pilot dose-escalation studies.

For bacterial burden analysis, blood and tissue samples were collected aseptically, serially diluted in sterile PBS, and plated on LB plates. Colony-forming units (CFUs) were counted after overnight incubation at 37 °C.

Sample collection

At the indicated time points, mice were anesthetized and euthanized for sample collection. Peripheral blood (PB) was collected for serum preparation. Serum was obtained by centrifugation after clotting and stored at − 80 °C until analysis. BMCs were harvested by flushing femurs and tibiae with PBS or culture medium using a 26-gauge needle. Spleens (SP) were mechanically dissociated into single-cell suspensions. Red blood cells were removed using ACK lysis buffer. Cell suspensions from PB, BM and SP were passed through a 70-μm cell strainer before downstream analyses.

Flow cytometry and cell sorting

Cells were stained in CSB (BioLegend) after viability assessment (Zombie Aqua, BioLegend 423,102) and Fc block (anti-CD16/32, Bio X Cell BE0307, 10 min, 4 °C). Surface markers included CD45, CD11b, Ly6G, Ly6C, B220, CD4, and CD8 (Supplementary Table 1). For myeloid subset analysis, phenotypically defined M-MDSC-like cells were gated as CD45⁺CD11b⁺Ly6G⁻Ly6C^hi cells, whereas PMN-MDSCs were gated as CD45⁺CD11b⁺Ly6G⁺Ly6C^lo cells. For sorting experiments, the indicated populations in BM were purified using a BD FACSAria III. Post-sort purity was routinely > 95% assessed by flow cytometric re-analysis. Data were acquired on a CytoFLEX LX (Beckman Coulter) and analyzed using FlowJo software.

Histopathology

Lung and kidney were fixed in 4% paraformaldehyde, paraffin-embedded, and sectioned (4 μm) for H&E staining. Histological analysis was evaluated in three random fields per sample by a blinded pathologist. Lung injury was scored 0–20 based on alveolar edema, inflammation, hemorrhage, hyaline membrane, and structural disruption, with each parameter graded from 0 to 4. Kidney injury was scored 0–5 based on the percentage of damaged tubules.

Cytokine measurement

Cytokine levels of IL-6, IL-10, TNF-α, and IL-1β in serum were measured using ELISA kits according to the manufacturer’s instructions. Samples were stored at − 80 °C and measured in duplicate; concentrations were interpolated from standard curves.

Electroporation for Slfn4 knockdown

For knockdown experiments, CD45⁺CD11b⁺Ly6G⁻Ly6C^hi BM-derived M-MDSC-like cells were sorted from second-hit mice as previously described. Purified cells were transfected immediately with Slfn4 siRNA or non-targeting control siRNA using the 4D-Nucleofector system (Lonza). For each reaction, 1 × 10⁶ cells were resuspended in 100 μL of P3 Nucleofector Solution and electroporated using program FF-120. siRNA was used at a final concentration of 100 nM (10 pmol per reaction). Cells were then transferred to prewarmed complete medium for recovery, and after 2–3 h were subjected to subsequent analyses.

In vitro induction of M-MDSC-like cells and Slfn4 overexpression

BMCs were isolated from naive mice and cultured in complete RPMI-1640 medium containing 10% fetal bovine serum and 1% penicillin/streptomycin at 37 °C in 5% CO₂. To generate M-MDSC-like cells in vitro, freshly isolated total BMCs were cultured for 72 h in the presence of GM-CSF (20 ng/mL) plus IL-6 (20 ng/mL). For cytokine comparison experiments, BMCs were cultured with GM-CSF alone, IL-6 alone, or GM-CSF plus IL-6, as indicated.

After 72 h induction, cultured cells were analyzed by flow cytometry, and CD45⁺CD11b⁺Ly6G⁻Ly6C^hi cells were sorted for downstream analyses. Sorted cells were then electroporated with a Slfn4 expression plasmid or empty vector using the 4D-Nucleofector system (Lonza). For each reaction, 5 × 10⁵ cells were resuspended in 100 μL of P3 Nucleofector Solution and electroporated using program FF-120 with 5 μg plasmid DNA per reaction. Cells were then immediately transferred to prewarmed complete medium containing GM-CSF (20 ng/mL) and IL-6 (20 ng/mL) for recovery. After 24 h, cells were used for downstream functional analyses.

RT-qPCR

Total RNA was extracted with TRIzol (Thermo Fisher) and reverse-transcribed (1 μg RNA, PrimeScript RT Kit, Takara). RT-qPCR used QuantiNova SYBR Green (Qiagen) on QuantStudio 5 (Applied Biosystems). Primers are listed in Supplementary Table 1. Expression was normalized to GAPDH and calculated by 2^−ΔΔCT method.

Western blot

Cells were lysed in RIPA buffer with protease/phosphatase inhibitors (Thermo Fisher). Proteins (20–30 μg) were separated by 10% SDS-PAGE, transferred to nitrocellulose, and blocked with 5% BSA. Membranes were incubated with primary antibodies (Supplementary Table 1) overnight at 4 °C, followed by HRP-conjugated secondary antibodies. SLFN4 was detected with anti-Flag. Bands were visualized by ECL and quantified with ImageJ.

Apoptosis detection

Apoptosis was assessed using the Annexin V-FITC/PI Apoptosis Detection Kit (BD Biosciences, 556,547). Briefly, sorted M-MDSC-like cells were washed twice with cold PBS and resuspended in binding buffer. Annexin V-FITC and PI were added according to the manufacturer’s instructions. After incubation for 20 min in the dark, cells were analyzed by flow cytometry within 1 h.

T-cell suppression assay

CFSE-labeled whole splenocytes from naive mice were used as responder cells and co-cultured with sorted M-MDSC-like cells at a ratio of 4:1. The co-culture system also contains anti-CD28 (2 μg/mL) and anti-CD3 (2 μg/mL) for T cell proliferation. After 48 h, cells were stained for CD4 and CD8. T-cell proliferation was quantified by CFSE dilution, and suppression was calculated by comparing division indices between cultures with and without M-MDSC-like cells.

Single-cell RNA sequencing

Single-cell suspensions were loaded onto the NovelCyto System (NovelBio) using limited dilution (Lai et al. 2022). Cells were paired with barcode-beads in microwells, lysed, and poly-adenylated RNA hybridized to beads. After reverse transcription, cDNA was tagged with UMIs and cell barcodes. Libraries were prepared by second-strand synthesis, WTA adaptor ligation, and 18-cycle PCR amplification. Sequencing libraries were enriched for 3′-end transcripts, quantified (Bioanalyzer 2200, Qubit HS DNA), and sequenced on DNBSEQ-T7 (MGI, 150 bp PE).

scRNA-seq data processing

Data were analyzed by NovelBio using the NovelBrain Cloud Platform (https://singlecell.novelbrain.com/login). Fastp (v0.24.0) was used to remove adaptor sequences and low-quality reads. Clean reads were aligned and quantified with STARsolo (v2.7.11b) against the mouse mm10 reference genome (Ensembl v100). To correct for sequencing depth variation, down-sampling was performed on the basis of mapped barcoded reads per cell across all samples, and an aggregated gene expression matrix was generated for downstream analysis.

Cells with low transcript complexity or poor quality were excluded before downstream analysis. Dimensionality reduction, clustering, and UMAP visualization were performed on the filtered matrix, and cell clusters were annotated on the basis of canonical marker genes. Doublets were estimated using Scrublet. Monocyte-macrophage subsets were further analyzed for pseudotime trajectory using Monocle 3 and for differentiation potential using CytoTRACE2. Cell–cell communication analysis was performed using CellChat. Single-cell pathway activity was evaluated using ssGSEA and AUCell with curated gene sets from MSigDB. Differentially expressed genes were subjected to GO, KEGG, and GSEA analyses using standard pipelines.

RNA immunoprecipitation (RIP)

HEK293T cells were cultured in DMEM with 10% FBS and transfected using jetPEI. For RIP and stability assays, cells were co-transfected with Flag-SLFN4 and Myc-SOCS3 (500 ng each). For reporter assays, HEK293T cells were co-transfected with pMIR-Socs3 reporter constructs, pRL-TK (20 ng), and Flag-SLFN4.

RIP was performed using the Magna RIP™ RNA-Binding Protein Immunoprecipitation Kit (MBL International, Cat# 17–700). Briefly, HEK293T cells were formaldehyde-crosslinked and lysed. Lysates were incubated with anti-Flag M2 magnetic beads (Sigma). RNA–protein complexes were immunoprecipitated, and RNA was extracted using TRIzol. Reverse transcription was performed using the PrimeScript RT Kit (Takara), followed by qPCR to detect Socs3 mRNA enrichment.

RNA stability assay

At 24 h post-transfection, cells were treated with 5 μg/mL actinomycin D (ActD) for 0, 15, 30 and 60 min. Total RNA was extracted using TRIzol, reverse-transcribed, and Socs3 levels quantified by qPCR (normalized to β-actin). Remaining mRNA percentage was calculated relative to 0 min.

Luciferase assay

Mouse Slfn4 was cloned into pCMV-Flag-C; the full-length Socs3 3'-UTR, 5'-UTR or CDS was inserted into pMIR-REPORT (Thermo Fisher) downstream of firefly luciferase. pRL-TK (Promega) served as a transfection control. Cells were lysed 24 h post-transfection. Firefly luciferase activity was measured using the Dual-Luciferase Assay System (Promega) and normalized to Renilla activity.

In vivo treatment

For Stattic treatment in the second-hit model, mice were intraperitoneally injected with Stattic (5 mg/kg, dissolved in 1% DMSO) or an equal volume of vehicle at 30 min before live E. coli challenge. For sildenafil treatment, sildenafil was prepared in 0.5% carboxymethylcellulose sodium (CMC-Na) aqueous solution and administered by oral gavage at 25, 50, or 100 mg/kg, as indicated. Mice received a single dose of sildenafil or vehicle (0.5% CMC-Na) 30 min before live E. coli challenge.

Computational modeling of protein-nucleic acid interactions

The sequences of SLFN4 protein and Socs3 mRNA were retrieved from UniProt (www.uniprot.org) and the NCBI database. Protein and nucleic acid structures were predicted using AlphaFold 3 (Abramson et al. 2024) and the 3DRNA/DNA method (Zhang et al. 2022). Then, we chose the higher-scoring predicted structures for further molecular docking, with three target nucleic acid molecules and the protein simulated using the HDOCK server (Yan et al. 2017). The top-ranked binding poses for each complex, based on docking score and confidence score, were selected. The docking results revealed a binding model where the nucleic acid wrapped around a positively charged groove of the protein.

Molecular dynamics (MD) simulations of the protein-nucleic acid complexes were conducted using Gromacs 2023.3 (Spoel et al. 2005). Systems were parameterized using the AMBER ff14SB_parmbsc1 force field, placed in a TIP3P water box, and ionized. After minimizing the energy of the system, all systems were equilibrated with NVT and NPT for 298.15 K and 1 bar. A production run of 100 ns was performed with a 2-fs timestep.

Trajectory analysis included root mean square deviation (RMSD), root mean square fluctuation (RMSF), radius of gyration (Rg), solvent accessible surface area (SASA), and inter-molecular hydrogen bonds. The binding free energy and per-residue energy (ΔG_bind) decomposition were calculated using the MM/PBSA method to identify key interacting residues. A negative ΔG_bind value for a residue indicates a favorable contribution to binding, whereas a positive value suggests an unfavorable effect. All analyses and visualizations were performed using standard Gromacs tools, PyMOL (Cerutti et al. 2009), and Origin.

Statistical analysis

All data are presented as mean ± SD. Statistical significance was assessed using an unpaired Student’s t-test or one-way ANOVA with Tukey’s post hoc test, as appropriate. Survival curves were analyzed using the Kaplan–Meier method and compared with the log-rank test. A two-sided P value < 0.05 was considered statistically significant. All analyses were performed using GraphPad Prism 10.0. Sample sizes were determined on the basis of pilot experiments and prior experience with the model. Exact sample sizes for each experiment are provided in the figure legends. No a priori power calculation was performed.

Results

Establishment of a second-hit sepsis model after LPS priming

To establish a murine second-hit sepsis model, mice were primed with LPS (10 mg/kg, i.p.) and challenged 48 h later with live E. coli (Fig. 1A). Mice were assigned to four groups: Control, LPS-only, E. coli, and second-hit. The LPS-only group confirmed the model conditions: all mice survived, and serum cytokines (IL-1β, IL-6, TNFα, IL-10) remained elevated at 12 h but normalized by 48 h (Fig. S1A). These data indicate that systemic cytokine elevation after LPS had largely subsided by 48 h, consistent with a relatively hyporesponsive post-inflammatory state under our experimental conditions.

Fig. 1.

Second-hit sepsis is associated with immunosuppression and expansion of bone marrow CD45⁺CD11b⁺Ly6G⁻Ly6C^hi cells. A Schematic of experimental design. Mice received intraperitoneal LPS (10 mg/kg) followed 48 h later by intraperitoneal E. coli (5 × 10⁷ CFU). Control, LPS-only, and E. coli-only groups were included. B Survival curves. The E. coli group showed 0% survival within 24–48 h, whereas the second-hit group exhibited delayed mortality; Control and LPS-only mice showed 100% survival (log-rank test, ***P < 0.001, n = 10 per group). C Serum cytokine levels measured 12 h after bacterial challenge. IL-1β, IL-6, TNF-α, and IL-10 were elevated in the E. coli group compared with Control (***P < 0.001), whereas IL-1β, IL-6, and TNF-α were reduced and IL-10 was increased in the second-hit group relative to the E. coli group (*P < 0.05, **P < 0.01, n = 5–6). D Flow-cytometric quantification of myeloid subsets in bone marrow. CD45⁺CD11b⁺Ly6G⁺Ly6C^lo cells were decreased (****P < 0.0001), whereas CD45⁺CD11b⁺Ly6G⁻Ly6C^hi cells expanded to nearly 10% of total BM cells in the second-hit group (***P < 0.001 vs Control and E. coli, n = 5–8). E Flow-cytometric analysis of splenic lymphocyte subsets. CD4⁺ and CD8⁺ T-cell populations were reduced after second-hit (*P < 0.05, ***P < 0.001 vs Control or E. coli, n = 5–8). F UMAP visualization of annotated BMC clusters from Control, E. coli, and second-hit groups. G Bar plot showing the proportion of each annotated cell cluster across groups, highlighting expansion of monocyte-macrophage cells in the second-hit group. H, I Total number and strength of ligand-receptor interactions inferred by CellChat across major BM cell subsets. J Relative contribution of each group to the total number of CellChat-inferred interactions

Second-hit sepsis induces severe immunosuppression and prolonged mortality

E. coli challenge alone caused rapid mortality, with all mice dying within 48 h, whereas the second-hit group exhibited delayed and prolonged mortality, with deaths extending beyond 48 h (Fig. 1B; log-rank test, P < 0.0001). Systemic cytokine profiling at 12 h showed marked increases in IL-1β, IL-6, TNF-α, and IL-10 in the E. coli group compared with Control (all P < 0.001). In contrast, LPS priming before E. coli challenge resulted in lower IL-1β, IL-6, and TNF-α, but higher IL-10, compared with E. coli alone (all P < 0.05; Fig. 1C), suggesting a shift from an acute hyperinflammatory response toward immune hypo-responsiveness. Bacterial burden in blood and spleen is shown in Fig. S1B. In second-hit mice, viable E. coli counts peaked at 24–48 h post-inoculation in both PB and SP, indicating impaired bacterial control. Histological analysis of lung and kidney tissues is shown in Fig. S1C. The E. coli group exhibited severe lung injury but only mild kidney injury, whereas the second‑hit group showed severe lung injury and marked kidney injury (Fig. S1D, E).

Together, these findings indicate that E. coli challenge alone was associated with acute inflammatory injury and rapid death, whereas the second-hit condition showed impaired bacterial control together with features consistent with immune hypo-responsiveness.

Second-hit sepsis is associated with preferential accumulationof CD45⁺CD11b⁺Ly6G⁻Ly6C^hi cells

To assess immune alterations after the two-hit challenge, we performed flow cytometric analysis of leukocyte subsets in PB, SP and BM. In the BM, the proportion of CD45⁺CD11b⁺Ly6G⁺Ly6C^lo cells was significantly reduced (P < 0.001), whereas CD45⁺CD11b⁺Ly6G⁻Ly6C^hi cells expanded to nearly 10% of total BM cells in the second-hit group (P < 0.001, Fig. 1D), consistent with accumulation of a monocytic myeloid population. A similar increase in CD45⁺CD11b⁺Ly6G⁻Ly6C^hi cells was observed in PB (P < 0.05, Fig. S1F). In parallel, CD4⁺ and CD8⁺ T-cell populations were reduced in SP and BM (Fig. 1E and Fig. S1G). Minimal change in the splenic myeloid subsets (Fig. S1H) and PB T-cell frequencies (Fig. S1I) in second-hit compared with E. coli.

Single-cell RNA sequencing unravels functional monocytic myeloid cells in bone marrow

To further characterize bone marrow immune heterogeneity in second-hit sepsis, we performed scRNA-seq on BMCs from Control, E. coli, and second-hit mice at 12 h after challenge. After quality filtering, 27,308 high-quality cells were retained for downstream analysis (Fig. S2A-C). UMAP analysis identified 16 transcriptionally distinct clusters, which were annotated using canonical marker genes (Fig. S2D-F). These clusters corresponded to major immune subsets, including granulocytes, monocyte-macrophages, hematopoietic/progenitor cells, dendritic cells, T cells, B cells, and cycling cells (Fig. 1F and Fig. S2G). Comparison across groups revealed marked shifts in cellular composition, particularly an expansion of monocyte-macrophage populations in the second-hit group (Fig. 1G and Fig. S3A). To assess differentiation states, we next performed CytoTRACE2 analysis (Fig. S3B, C). Notably, Slfn4 was the gene most strongly negatively correlated with stemness, suggesting an association with a terminally differentiated state.

We next performed CellChat analysis to examine intercellular communication within the BM microenvironment. Monocyte-macrophage populations exhibited the most extensive ligand-receptor crosstalk among all major cell subsets (Fig. 1H, I), monocyte-macrophage-associated interactions were increased in the second-hit group (Fig. 1J). Further analysis of this compartment identified several putative immunoregulatory ligand-receptor pairs, including Sirpb1a-CD47, CD48-CD244, and Fn1-(Itga4 + Itgb1) (Fig. S3D).

Identification of a distinct Slfn4⁺ monocytic subset

To further resolve the monocytic compartment, we re-clustered monocyte-macrophage cells and identified 11 subclusters (M01-M11) (Fig. 2A). Analysis of cluster-defining genes and pathway enrichment supported a relatively suppressive profile for M01 and a more inflammatory profile for M06/M11 (Supplementary Table 2). In particular, the M01 subcluster was derived almost exclusively from the second-hit group, indicating strong enrichment under second-hit conditions (Fig. 2B). The top marker genes for each subcluster are shown in Fig. S4A. M01 was characterized by prominent expression of Slfn4 together with Il1rn, Rsad2, Cd274, Ifi204, Gbp2, Acod1, Cxcl5, Ifit2, Ifit3, Saa3, and Nos2. Slfn4 expression was highly concentrated in this subcluster (Fig. 2C, D), and additional marker-gene expression patterns are shown in Fig. S4B. These data identify a distinct Slfn4⁺ monocytic subcluster with a transcriptional profile consistent with immunoregulatory activity.

Fig. 2.

Identification of a terminally differentiated Slfn4⁺ monocytic subset enriched in bone marrow after second-hit sepsis. A UMAP visualization of 11 monocyte-macrophage subclusters (M01–M11), color-coded by cluster identity (n = 9,847 cells pooled from all three groups). B Relative abundance of each monocyte-macrophage subcluster across groups. M01 was almost exclusively derived from the second-hit group (599 of 618 total cells). C Violin plots showing Slfn4 expression across experimental groups. D Feature plot showing that Slfn4 expression was largely restricted to the M01 subcluster. E, F Monocle 3 pseudotime trajectory revealing two major terminal fates, with fate 1 predominantly populated by M01 cells. G Slfn4 expression along pseudotime, showing progressive upregulation toward the terminal phase of fate 1. H T-cell suppression assay. BM CD45⁺CD11b⁺Ly6G⁻Ly6C^hi cells were sorted by flow cytometry and co-cultured with CFSE-labeled splenocytes at a 1:4 ratio for 48 h. Cells from second-hit mice suppressed proliferation of CD4⁺ (*P < 0.05, vs E. coli), and CD8⁺ T cells (***P < 0.001, vs E. coli, n = 3–4). I, J qPCR analysis of sorted BM CD45⁺CD11b⁺Ly6G⁻Ly6C.^hi cells showing increased expression of (I) Slfn4 and (J) Arg1, Cd274, and Nos2 in the second-hit group (*P < 0.05, **P < 0.01, ***P < 0.001 vs Control or E. coli, n = 3–4). K Flow cytometric analysis of CD274 expression on sorted cells. CD274 levels were elevated in the E. coli group and further significantly increased in the second-hit group (**P < 0.01, ****P < 0.0001)

The Slfn4⁺ monocytic subset exhibits features consistent with an M-MDSC-like phenotype

To investigate the differentiation trajectory of this subset, we performed pseudotime analysis using Monocle 3. The resulting trajectory revealed two major terminal fates, with fate 1 predominantly populated by M01 cells (Fig. 2E, F). Consistent with this pattern, Slfn4 expression increased progressively along pseudotime and peaked at the terminal phase of fate 1 (Fig. 2G). Gene-module analysis identified three major modules, among which module 2 was specifically associated with fate 1 (Fig. S4C). GO and KEGG analyses of module 2 genes showed enrichment for pathways related to type II interferon response, regulation of IL-6 production, TNF signaling, cytokine-cytokine receptor interaction, and Jak-Stat signaling (Fig. S4D, E). Cells from the second-hit group were preferentially distributed at the terminal end of the pseudotime trajectory, consistent with fate 1 commitment (Fig. S4F).

To functionally assess the suppressive potential of this population, we sorted BM CD45⁺CD11b⁺Ly6G⁻Ly6C^hi cells and co-cultured them with CFSE-labeled splenocytes. These cells from second-hit mice suppressed proliferation of both CD4⁺ and CD8⁺ T cells, with a more pronounced effect on CD8⁺ T cells (Fig. 2H). We then quantified expressions of Slfn4 and several genes associated with MDSC suppressive programs, including Arg1, Cd274 and Nos2. All four transcripts were significantly upregulated in the second-hit group (Fig. 2I, J). The flow cytometry further verified higher CD274 protein expression on sorted cells in the second-hit group (Fig. 2K).

Collectively, these findings identify a second-hit-enriched Slfn4⁺ monocytic subset with a terminal differentiation signature, suppressive activity in co-culture, and transcriptional features consistent with an M-MDSC-like phenotype.

Enrichment of type II interferon and Jak-Stat3 signaling pathways in the Slfn4⁺ M-MDSC-like subset

To further define transcriptional programs associated with the second-hit condition, we compared monocyte-macrophage cells from second-hit and E. coli groups. Differential expression analysis identified Slfn4 among the top upregulated genes in second-hit mice (Fig. S5A). GSEA further highlighted enrichment of Jak-Stat signaling and interferon-γ-related pathways (Fig. S5B, C). Because interferon signaling is a known regulator of SLFN-family genes and Stat3 is central to MDSC biology, we next quantified these pathway activities at the single-cell level. ssGSEA and AUCell analyses showed that both the type II interferon response (Fig. S5D-F) and IL6-JAK-STAT3 signaling signatures (Fig. S5G-I) were most strongly enriched in the M01 subcluster.

Based on these observations, we hypothesized that Slfn4 may functionally contribute to the immunosuppressive program of this subset, potentially in connection with Stat3 signaling.

Knockdown of Slfn4 relieves immunosuppressive function of M-MDSC-like cells in second-hit sepsis

To evaluate the functional role of Slfn4, BM CD45⁺CD11b⁺Ly6G⁻Ly6C^hi cells from second-hit mice were sorted and transfected with Slfn4 siRNA (Si-Slfn4) or negative control siRNA (Si-NC) (Fig. S6A). After 2–3 h of cell recovery, qPCR confirmed effective knockdown of Slfn4 (Fig. 3A). In co-culture assays, Slfn4 knockdown attenuated the ability of these cells to suppress CD4⁺ and CD8⁺ T-cell proliferation (Fig. 3B). Annexin V/PI staining further showed increased late apoptosis after Slfn4 knockdown (Fig. S6B), suggesting that Slfn4 supports survival of this monocytic suppressor population. Consistent with this, western blotting revealed reduced STAT3 phosphorylation in Si-Slfn4-treated cells (Fig. 3C). p-JAK was also decreased, whereas SOCS3 protein was increased; in contrast, p-AKT and p-ERK were unchanged, supporting relative selectivity for the Jak-Stat3 pathway (Fig. S6C). In parallel, expression of the suppressive effector genes Cd274, Arg1, and Nos2 was significantly reduced following Slfn4 knockdown (Fig. 3D).

Fig. 3.

Slfn4 enhances the suppressive activity of bone marrow monocytic cells through Stat3 activation. A qPCR analysis of Slfn4 expression after siRNA transfection in sorted BM CD45⁺CD11b⁺Ly6G⁻Ly6C^hi cells of second-hit mice. Slfn4 knockdown reduced Slfn4 mRNA expression by approximately 70% (*P < 0.05 vs si-NC, n = 4). B CFSE-based T-cell suppression assay. Slfn4-silenced BM CD45⁺CD11b⁺Ly6G⁻Ly6C^hi cells showed reduced suppression of CD4⁺ and CD8⁺ T-cell proliferation, whereas Si-Slfn4 alone had no direct effect on T-cell proliferation (***P < 0.001, ****P < 0.0001 vs E. coli, n = 3–4). C Western blot analysis of phosphorylated STAT3 (Tyr705) and total STAT3 in sorted and transfected BM CD45⁺CD11b⁺Ly6G⁻Ly6C^hi cells. The p-STAT3/STAT3 ratio was decreased after Slfn4 knockdown (*P < 0.05). D qPCR analysis showing reduced expression of the immunosuppressive genes Cd274, Arg1, and Nos2 after Slfn4 knockdown (*P < 0.05, **P < 0.01 vs Si-NC, n = 3). E–H Gain-of-function experiments in in vitro-generated BM-derived M-MDSC-like cells. Naive BMCs were first cultured with GM-CSF and IL-6 for 72 h, after which CD45⁺CD11b⁺Ly6G⁻Ly6C^hi cells were sorted, electroporated with Slfn4 expression plasmid or vector control, and recovered for 24 h before analysis. E qPCR analysis showing increased Slfn4 expression in M-MDSC-like cells after Slfn4-OE (**P < 0.01 vs vector, n = 4). F CFSE-based T-cell suppression assay. Slfn4-OE cells markedly inhibited CD4⁺ and CD8⁺ T-cell proliferation (**P < 0.01, ***P < 0.001 vs E. coli, n = 4). G Western blot analysis showing an increased p-STAT3/STAT3 ratio in Slfn4-overexpressing cells (*P < 0.05, n = 3). H qPCR analysis showing increased expression of Cd274, Arg1, and Nos2 in Slfn4-overexpressing cells (*P < 0.05, **P < 0.01, n = 3). Data are presented as mean ± SD. Statistical analyses are described in the Methods

Slfn4 promotes a Stat3-associated suppressive phenotypein vitro

For gain-of-function studies, we first established an in vitro protocol to generate M-MDSC-like cells from naive BMCs (Fig. S6D). Total BMCs were cultured under the indicated cytokine conditions for 72 h and then analyzed by flow cytometry. Whereas GM-CSF alone induced mainly macrophage-like F4/80⁺ cells and IL-6 alone yielded cells with relatively low Ly6C expression, the combination of GM-CSF and IL-6 generated a myeloid population with higher Ly6C expression (Fig. S6E). Based on this optimization, BMCs cultured with GM-CSF plus IL-6 for 72 h were used for subsequent overexpression experiments.

The induced CD45⁺CD11b⁺Ly6G⁻Ly6C^hi cells were sorted and then electroporated with a Slfn4 expression plasmid or empty vector, followed by 24 h recovery in GM-CSF- and IL-6-containing medium (Fig. S6D). RT-qPCR confirmed increased Slfn4 expression (Fig. 3E). In co-culture assays, the sorted Slfn4-OE cells showed enhanced suppression of CD4⁺ and CD8⁺ T cell proliferation (Fig. 3F). Moreover, Slfn4 overexpression reduced late apoptosis (Fig. S6F), increased STAT3 phosphorylation (Fig. 3G and Fig. S6G), and upregulated Cd274, Arg1, and Nos2 expression (Fig. 3H). Together, these results support a role for Slfn4 in maintaining the survival and immunosuppressive phenotype of M-MDSC-like cells, in association with Stat3 activation.

Inhibition of Stat3 attenuates sepsis-induced myeloid dysfunction and improves survival

Given the association between Slfn4 and Stat3 activation, we next examined whether pharmacologic Stat3 inhibition could counteract the suppressive monocytic program in vitro and in vivo. In the in vitro M-MDSC differentiation system, treatment with the Stat3 inhibitor Stattic did not alter the Slfn4 mRNA expression (Fig. 4A), while partially reduced the CD8⁺ T-cell suppressive activity conferred by Slfn4 overexpression (Fig. 4B). Stattic also reduced p-STAT3/STAT3 levels (Fig. 4C) and increased late apoptosis (Fig. S6H), phenocopying the effects of Slfn4 knockdown.

Fig. 4.

Pharmacologic STAT3 inhibition during the immunosuppressive phase reverses myeloid dysfunction and improves survival. A In an in vitro M-MDSC-like cell differentiation model, treatment with Stattic (5 μM) did not affect Slfn4 mRNA expression. B CFSE-based T-cell suppression assay showing that Stattic restored CD8⁺ T-cell proliferation previously suppressed by Slfn4-overexpressing M-MDSC-like cells (*P < 0.01, n = 3). C Western blot analysis of SOCS3, phosphorylated STAT3, and total STAT3 in Slfn4-overexpressing monocytic cells treated with or without Stattic (5 μM). SOCS3 protein remained low in Slfn4-overexpressing cells irrespective of Stattic treatment (**P < 0.01 vs vector, n = 3). D Kaplan–Meier survival analysis after Stattic intervention in vivo. Stattic significantly prolonged survival compared with vehicle treatment (P = 0.045 by log-rank test, n = 10). E Flow cytometric analysis of BM CD45⁺CD11b⁺Ly6G⁻Ly6C.^hi cells at 12–24 h after E. coli challenge. Stattic reduced the proportion of this population from 3.2% to 0.3% (*P < 0.05 vs vehicle, n = 6). F Flow cytometric analysis showing that Stattic increased the proportion of splenic CD8⁺ T cells (***P < 0.001 vs vehicle, n = 6)

In vivo, Stattic administered 30 min before the secondary E. coli challenge significantly improved survival in second-hit mice (Fig. 4D). Consistent with this benefit, Stattic-treated mice showed a reduced proportion of BM CD45⁺CD11b⁺Ly6G⁻Ly6C^hi cells (Fig. 4E) together with an increased splenic CD8⁺ T-cell population (Fig. 4F), suggesting partial reversal of the immunosuppressive phenotype.

Slfn4 post-transcriptionally regulates Socs3viadirect mRNA binding

Our findings placed Stat3 as a candidate downstream pathway of Slfn4. During these analyses, we noted that Socs3 mRNA was elevated after E. coli challenge and was further increased in the second-hit group (Fig. S4B), whereas Slfn4 overexpression reduced SOCS3 protein abundance (Fig. S6G). In addition, SOCS3 protein remained suppressed in Slfn4-OE cells even after Stattic treatment (Fig. 4C). Because SOCS3 is a key negative regulator of Stat3 signaling, this discordance between transcript abundance and protein expression implied post-transcriptional regulation by Slfn4.

To test this possibility, we co-transfected HEK293T cells with Flag-SLFN4 and Myc-SOCS3 expression plasmids. Immunoblotting confirmed robust SLFN4 expression and a marked reduction in Myc-SOCS3 protein, whereas Socs3 mRNA abundance remained unchanged (Fig. 5A), indicating post-transcriptional suppression. RIP assays further showed strong enrichment of Socs3 mRNA in Flag-SLFN4 immunoprecipitate (Fig. 5B), supporting direct association between SLFN4 protein and the Socs3 transcript. However, Actinomycin D chase experiments showed no difference in Socs3 mRNA decay kinetics between groups (Fig. 5C), suggesting that Slfn4 does not regulate Socs3 through altered mRNA stability.

Fig. 5.

SLFN4 associates with Socs3 mRNA and suppresses SOCS3 protein expression. A HEK293T cells were co-transfected with Myc-Socs3 and either Flag-Slfn4 or empty vector. SOCS3 protein and Socs3 mRNA levels were analyzed by immunoblotting and RT-qPCR, respectively. SOCS3 protein showed an apparent decrease in the presence of SLFN4, whereas Socs3 mRNA levels were not significantly affected (ns, n = 3). B RIP was performed in co-transfected HEK293T cells using an anti-Flag antibody, followed by RT-qPCR analysis of co-transfected Socs3 mRNA. Socs3 mRNA was significantly enriched in Flag-SLFN4 immunoprecipitates compared with control (*P < 0.05, n = 3). C To assess mRNA stability, transfected HEK293T cells were treated with actinomycin D (ActD), and Socs3 mRNA levels were measured at the indicated time points. SLFN4 did not significantly alter Socs3 mRNA decay (ns, n = 3). D AlphaFold2-predicted structure of full-length mouse SLFN4, colored according to pLDDT confidence values. The putative RNA-interacting surface is indicated. E Schematic of Socs3 mRNA showing three conserved sequence elements (SLS1, SLS2, and ARE). The AU-rich element (ARE) was selected for docking analysis. F Structural docking model of SLFN4 with the Socs3 ARE RNA fragment, showing a putative interaction with a positively charged surface region of SLFN4. G HEK293T cells were co-transfected with a luciferase reporter containing the Socs3 3′UTR together with Flag-Slfn4 or empty vector. Reporter activity was significantly reduced in the presence of SLFN4 (*P < 0.05, n = 3). Data are presented as mean ± SD

Slfn4 targets the 3'UTR of Socs3 and induce translational repression

To gain structural insight into the interaction between SLFN4 protein and Socs3 mRNA, we performed computational modeling and molecular dynamics simulations. AlphaFold2 modeling generated a high-confidence full-length mouse SLFN4 (UniProt Q3UHC5) protein structure (Fig. 5D). The conserved structural elements within the Socs3 transcript were identified through cross-species sequence alignment. Three evolutionarily conserved elements were retained for structural analysis: Stem-Loop-1 (SLS1, –22 to –8nt), Stem-Loop-2 (SLS2, + 8 to + 28nt) and the AU-rich element (ARE, + 1180 to + 1205nt). Their secondary/tertiary folds were generated with 3dRNA and AlphaFold3 (Fig. 5E; Fig. S7A). Among the tested RNA fragments, the 3′-UTR AU-rich element (ARE) showed the highest predicted binding affinity for SLFN4 (Fig. S7B and Supplementary file 1). Molecular dynamics simulations supported a stable interaction between SLFN4 and ARE, with electrostatic contacts contributed primarily by positively charged residues on SLFN4 (Fig. 5F and Fig. S7C-G). Detailed MD process and results were all presented in the Supplement File.

To experimentally validate the binding region, we performed luciferase reporter assays. Co-transfection of Flag-SLFN4 significantly reduced luciferase activity from the Socs3 3′-UTR reporter, but not from reporters containing the 5′-UTR or coding sequence (Fig. 5G), indicating that SLFN4 targets the Socs3 3′-UTR and represses expression at the translational level rather than by promoting mRNA decay.

Related SLFN-family-positive suppressive monocytic signatures in human sepsis and a murine CLP Model

To assess whether this transcriptional program extends beyond our second-hit model, we re-analyzed published scRNA-seq datasets from human COVID-19-associated sepsis (GSE157344) and murine CLP sepsis (GSE207651). In human mononuclear phagocytes, we identified a SLFN12⁺ monocyte subset with high CD274 expression (Fig. S8A, B), and GO analysis suggested enrichment of interferon- and immune-regulatory pathways (Fig. S8C). In the murine CLP dataset, Slfn4⁺ monocytes increased at 48 h after surgery (Fig. S8D) and co-expressed multiple genes associated with suppressive myeloid states, including S100a8, S100a9, Lcn2, Cd177, and Anxa1 (Fig. S8E). GO and KEGG analyses further supported enrichment of interferon-related, T-cell regulatory, and Jak-Stat-associated pathways (Fig. S8F, G). These independent datasets support the presence of transcriptionally related SLFN-family-positive monocytic populations with immunoregulatory features across distinct sepsis settings. Given species differences within the SLFN family, these cross-dataset analyses should be interpreted as transcriptional parallels rather than direct orthologous equivalence.

Sildenafil attenuates suppressive myeloid features and improves survival in vivo

To explore the therapeutic relevance of this pathway, we treated mice with sildenafil, a phosphodiesterase-5 (PDE5) inhibitor previously reported to modulate suppressive myeloid cells (Ding et al. 2023). The sildenafil was administered 30 min before the secondary E. coli challenge (Fig. 6A). Results suggested that Sildenafil at 50 mg/kg significantly improved survival in the second-hit model (Fig. 6B), whereas 25 mg/kg and 100 mg/kg did not (Fig. S8H). Sildenafil treatment also reduced bacterial burden in PB (Fig. 6C, Fig. S8I). Flow cytometric analysis showed that sildenafil decreased the proportion of BM CD45⁺CD11b⁺Ly6G⁻Ly6C^hi cells (Fig. 6D) and increased CD4⁺ and CD8⁺ T-cell populations in the SP (Fig. 6E). To determine whether this intervention affected the suppressive phenotype of these cells, BM CD45⁺CD11b⁺Ly6G⁻Ly6C^hi cells were sorted from treated mice and tested in co-culture assays. Cells from sildenafil-treated mice exhibited weaker suppression of CD8⁺ T-cell proliferation than those from vehicle-treated mice (Fig. 6F). In addition, Slfn4, Arg1, Cd274, and Nos2 mRNA levels were reduced in these sorted cells after sildenafil treatment (Fig. 6G, H). Sildenafil also reduced the CD274 protein on surface of sorted cells (Fig. 6I).

Fig. 6.

Sildenafil disrupts the Slfn4⁺ M-MDSC axis, restores T-cell immunity and improves survival. A Schematic illustration of the sildenafil treatment schedule. B Kaplan–Meier survival analysis showed that sildenafil treatment (50 mg/kg) significantly improved 7-day survival compared with vehicle (P = 0.036, log-rank test, n = 10). C Blood bacterial load in surviving mice was measured at 48 h after infection. Sildenafil treatment reduced bacterial burden compared with vehicle-treated controls (**P < 0.01, n = 5). D Flow cytometric analysis of BMCs showed that the proportion of CD45⁺CD11b⁺Ly6G⁻Ly6C^hi cells was significantly decreased in sildenafil-treated mice compared with vehicle-treated mice (*P < 0.05, n = 5–6). E Flow cytometric analysis of spleen cells showed increased proportions of CD4⁺ T cells and CD8⁺ T cells in sildenafil-treated mice compared with vehicle-treated controls (*P < 0.05, ****P < 0.0001, n = 4). F CFSE-based T-cell suppression assays were performed using CD45⁺CD11b⁺Ly6G⁻Ly6C^hi cells isolated from sildenafil-treated or vehicle-treated mice. Cells from sildenafil-treated mice showed reduced inhibitory effects on CD8⁺ T-cell proliferation (**P < 0.01, n = 4). G Slfn4 mRNA levels were measured by RT-qPCR in BM CD45⁺CD11b⁺Ly6G⁻Ly6C^hi cells isolated 12 h after E. coli infection. Sildenafil treatment reduced Slfn4 expression compared with vehicle treatment (P < 0.05, n = 6). H RT-qPCR analysis of sorted cells showed that the expression of Arg1, Cd274, and Nos2 were reduced after sildenafil treatment compared with vehicle-treated controls (*P < 0.05, **P < 0.01, n = 4). I Flow cytometric analysis showed that CD274 protein level on sorted cells were reduced after sildenafil treatment compared with vehicle-treated controls (*P < 0.05, n = 4). Data are presented as mean ± SD from independent experiments

Together, these findings show that sildenafil treatment is associated with reduced accumulation and suppressive activity of this monocytic population in vivo.

Discussion

In this study, we identified a Slfn4-enriched monocyte-like population that emerged in the bone marrow during second-hit sepsis. These cells exhibited phenotypic and functional features consistent with a suppressive myeloid state. Functional perturbation experiments further supported a role for Slfn4 in promoting the immunosuppressive phenotype of these cells, at least in part through enhanced Stat3 signaling. In vivo, pharmacological perturbation of this pathway attenuated suppressive myeloid accumulation and was associated with improved immune parameters. Together, these findings implicate the Slfn4–Stat3 axis as an important regulator of bone marrow-associated immune dysregulation in this model. Rather than establishing a complete or exclusive mechanism, our data support Slfn4 as a functionally relevant component of the suppressive myeloid program that develops after a secondary septic challenge.

The prolonged sepsis is increasingly recognized as a dynamic process that progresses from initial hyper-inflammation to a state of immune hypo-responsiveness and existing pathogen burden (Jamieson et al. 2013; Wang et al. 2016). Patients in this phase are particularly susceptible to recurrent nosocomial infections, opportunistic pathogens, and poor long-term outcomes (Chadda and Puthucheary 2024). Recent single-cell atlases of human and murine sepsis have revealed broad immunosuppressive landscapes (Reyes et al. 2020; Kwok et al. 2023), yet experimental systems that reliably model this transition remain limited. To address this challenge, second-hit models have been widely used to reproduce selected features of post-sepsis immune dysfunction, typically by combining an initial inflammatory insult with a delayed secondary microbial challenge (Bauhofer et al. 2006; Semple et al. 2019; Guan et al. 2013; Wang et al. 2022). Based on previous explorations, we developed a sequential second-hit protocol: primary LPS injection to permit development of endotoxin hypo-responsiveness, and a secondary challenge to capture aspects of delayed antibacterial immune dysfunction. Notably, tolerance-associated myeloid remodeling is unlikely to be uniformly detrimental. In our model, LPS preconditioning appeared to blunt the early inflammatory surge after secondary bacterial challenge, which may confer short-term protection from overwhelming acute inflammation. At the same time, second-hit mice showed impaired bacterial control and prolonged mortality, consistent with a maladaptive persistence of immune hypo-responsiveness once effective antimicrobial immunity is required. These observations highlight the stage-dependent nature of post-inflammatory myeloid remodeling rather than implying that tolerance-associated programs are uniformly protective or harmful.

While tracking the immune profile after the second-hit, we observed a marked accumulation of CD45⁺CD11b⁺Ly6G⁻Ly6C^hi myeloid cells within BM, together with the emergence of a Slfn4-enriched monocyte-like cluster in the single-cell dataset. Pseudotime analysis positioned this cluster toward a terminal-state trajectory, and co-culture assays with whole splenocytes demonstrated a strong capacity to suppress T-cell proliferation, accompanied by elevated expression of suppressive effector genes. These findings highlight the heterogeneity of monocyte responses across infection settings. Whereas E. coli infection alone preferentially induced monocyte clusters with more inflammatory transcriptional features, the second-hit condition favored emergence of a Slfn4-associated suppressive state.

Considering that the phenotypic boundaries between M-MDSCs and inflammatory monocytes remain incompletely resolved in the bone marrow, our study complements rather than contradicts prior sequencing studies that have highlighted suppressive PMN-MDSC and M-MDSC populations in the circulation and spleen of septic patients (Reyes et al. 2020; Kwok et al. 2023; Xiao et al. 2025). For example, Kwok et al. (Kwok et al. 2023) identified a mature CD66b⁺ neutrophil population in whole blood with T-cell suppressive properties, underscoring the heterogeneity of suppressive myeloid responses across tissues. Given the central role of bone marrow in emergency myelopoiesis, our data highlight this compartment as an important site in which suppressive monocyte-like cells may emerge or be conditioned during prolonged sepsis.

Members of the Schlafen family have been implicated in diverse cellular processes, including proliferation, differentiation, and immune regulation (Jo and Pommier 2022). Although the molecular functions of individual SLFN proteins are still being defined, prior work from Merchant and colleagues established Slfn4 as a myeloid differentiation factor associated with spasmolytic metaplasia and later identified SLFN12L/Slfn4⁺ suppressive myeloid cells in Helicobacter-induced gastric metaplasia in both mice and humans (Ding et al. 2016, 2020, 2022, 2023). These studies provided an important framework for considering Slfn4 as more than a lineage marker. In our second-hit sepsis model, we similarly identified a Slfn4⁺ monocyte-like population with strong suppressive effects on CD4⁺ and CD8⁺ T-cell proliferation. Moreover, genetic knockdown of Slfn4 partially reduced the suppressive phenotype, promoted apoptosis, and attenuated Stat3 phosphorylation, whereas Slfn4 overexpression enhanced expression of key suppressive mediators such as Arg1, Cd274, and Nos2. Taken together, these data support a functional contribution of Slfn4 to the maintenance of a suppressive myeloid program in the bone marrow, rather than serving only as a passive biomarker.

Mechanistically, our data support a model in which Slfn4 amplifies Stat3 signaling, potentially through post-transcriptional regulation of Socs3. RIP-qPCR, RNA stability, and luciferase assays were consistent with a direct effect of SLFN4 on the Socs3 transcript, and in silico docking further identified an AU-rich element within the 3′-UTR as a plausible binding interface. Given the established role of sustained Jak/Stat3 signaling in promoting MDSC expansion and suppressive differentiation (Su et al. 2018; Condamine and Gabrilovich 2011), and the function of Socs3 as a key negative regulator downstream of IL-6/IL-10 signaling (Yuan et al. 2023), this mechanism may help explain how Slfn4 supports a persistent suppressive myeloid state in late sepsis. The proposed mechanism is summarized in Fig. 7.

Fig. 7.

Proposed mechanism by which Slfn4⁺ M-MDSC-like cells promote immunosuppression during second-hit sepsis

These findings offer a potential mechanistic explanation for how Slfn4 may reinforce Stat3 activation and thereby stabilize the suppressive phenotype. However, this model should still be interpreted with appropriate caution. Although the computational and functional data converge on a plausible SLFN4–Socs3 regulatory interaction, direct biochemical validation of RNA binding specificity and translational repression will be required to establish the precise molecular mechanism.

Our pharmacological data further suggest that this pathway may be modifiable in vivo. Stat3 inhibition with Stattic reduced suppressive myeloid accumulation and improved immune parameters in second-hit sepsis, in line with previous work showing that Stat3-targeted approaches can limit MDSC expansion (Wang et al. 2024). In addition, sildenafil was included as an exploratory intervention based on prior studies linking PDE5 inhibition to MDSC modulation (Porwal and Das 2024; Lasser et al. 2024). Recent clinical studies have highlighted the potential relevance of MDSC-associated immune dysfunction in sepsis, both as biomarkers of immune dysfunction and as possible targets for immunomodulatory intervention (Meisel et al. 2009; Uhel et al. 2017). Our findings support the concept that targeting the Slfn4-associated suppressive program may have therapeutic relevance, particularly during the sustained immunosuppressive phase after the initial insult. However, these interventions should be considered proof-of-concept rather than directly translatable treatment strategies, as the function of Slfn4⁺ suppressive monocytes is likely context- and stage-dependent, and therapeutic benefit may critically depend on intervention timing.

Several limitations should also be acknowledged. First, our LPS + E. coli second-hit model captures selected features of prolonged sepsis but does not fully reproduce the clinical heterogeneity of human disease. Second, although the bone marrow population identified here showed transcriptional and functional features consistent with M-MDSC-like cells, further phenotypic resolution will be needed to distinguish these cells more clearly from inflammatory monocytes. Third, while multiple assays support a role for Slfn4 in Socs3/Stat3 regulation, the proposed SLFN4–Socs3 interaction remains incompletely resolved at the biochemical level. Finally, only male mice were used in this study, and potential sex-dependent differences in suppressive myeloid remodeling should be addressed in future work.

In summary, using a second-hit sepsis model, we identify a Slfn4-associated suppressive monocyte-like program in the bone marrow and support its functional link to Stat3 activation and impaired immune responsiveness. These findings provide a mechanistic framework for understanding bone marrow-derived immune dysregulation during prolonged sepsis and support further investigation of the Slfn4–Stat3 axis as a candidate pathway in sepsis-associated immunosuppression.

Authors’ contributions

Study concept and design: X.M., X.H., H.Y.; drafting of the manuscript: X.M., H.W.; implement of the study: X.M., H.W., H.W., Y.H., Y.Y.; analysis and interpretation of data: J.C., L.P.; revision of the manuscript: X.H., H.Y.. All authors have read and approved the final manuscript.

Funding

This work was supported by the National Natural Science Foundation of China [grant number 82270004] and the Tengfei Project [grant number TF2024XSJJ10]. All authors declare no conflict of interest.

Data availability

The data that support this study are available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

All animal experiments were approved by the Institutional Animal Care and Use Committee of Naval Medical University and conducted in accordance with institutional guidelines.

Competing interests

The authors declare no competing interests.