An NRIP1-Centric Monocyte Immunogenetic Bridge Between Heart Failure and Ischemic Stroke

Abstract

Background

Heart failure (HF) and ischemic stroke (IS) frequently co-occur and worsen each other. Circulating immune cells may act as a cross-organ inflammatory bridge, but the key immune lineage, genetically supported mediators, and druggable targets remain unclear.

Objective

To identify the immune-cell lineage, candidate gene, and tractable intervention target linking HF and IS.

Methods

Bidirectional Mendelian randomization (MR) of genome-wide association study summary statistics was used to assess causal links between HF and IS. Mouse single-cell transcriptomic datasets from HF and IS models were integrated to define a shared immune/stromal atlas and characterize Ly6C⁺ monocyte expansion. Ly6C⁺ monocyte marker genes were mapped to human orthologs and evaluated using peripheral-blood expression quantitative trait locus data for gene-level MR, followed by expression validation, colocalization, summary-data–based MR, and spatial transcriptomics. In HF single-cell data, Ly6C⁺ monocytes were stratified by Nrip1 expression for cell–cell communication, pathway enrichment, and pseudotime analyses. Candidate-drug prediction, druggability evaluation, molecular docking, molecular dynamics simulation, and in vitro validation were then performed.

Results

MR supported a bidirectional positive association between HF and IS. Both diseases showed expansion of inflammatory Ly6C⁺ monocytes at single-cell resolution. Among MR-implicated genes, NRIP1 was prioritized as a key gene associated with reduced IS risk, and genetic and spatial evidence supported its monocyte-mediated role in HF–IS coupling. Nrip1-stratified and pseudotime analyses suggested a homeostatic role through modulation of inflammatory thresholds and communication patterns. Drug prediction and in silico analyses indicated that retinoic acid–related small molecules may bind NRIP1 and exhibit preliminary druggability. In vitro, tretinoin alleviated ischemia-like HMC3 cell injury and inflammation by modulating monocyte NRIP1.

Conclusion

An NRIP1-centered monocyte network may represent a tractable immune bridge in HF–IS comorbidity, and tretinoin provides an entry point for NRIP1-targeted intervention.

Graphical Abstract

Introduction

Global Burden of Disease (GBD) studies have shown that ischemic heart disease and stroke have long ranked among the top two causes of death and disability-adjusted life years (DALYs) worldwide, with ischemic stroke (IS) accounting for approximately 80–90% of all stroke events and representing the major contributor to overall stroke burden.^1,2 Although heart failure (HF) is not the single most burdensome cardiovascular disease category in the GBD classification, it is characterized by a high prevalence, persistently elevated mortality and readmission rates, and a strong tendency to occur sequentially or coexist with stroke in the same patient, collectively forming a spectrum of heart-brain interactions marked by bidirectional dysfunction between the heart and the brain in clinical practice.^3,4 Patients with HF are at markedly increased risk of developing IS, whereas stroke-related arrhythmias, myocardial injury, and deterioration of cardiac function further accelerate HF progression as a form of neurogenic cardiac injury, thereby creating a vicious bidirectional cycle.^5–8 In recent years, the concept of a “heart-brain axis” has been proposed to capture these observations, emphasizing that beyond shared risk factors and hemodynamic disturbances, the heart and brain influence and amplify each other through multiple cross-organ pathways, including neuroendocrine activation, autonomic imbalance, systemic inflammation, and immune dysregulation.^9–11 Within the broad spectrum of cardiocerebrovascular disease, HF complicated by IS represents a common heart-brain comorbidity pattern with a particularly heavy prognostic burden, suggesting that the two conditions may share underlying pathological pathways and molecular targets. Rather than investigating each disease in isolation, starting from this comorbid pattern is more conducive to uncovering shared mechanisms underlying heart-brain comorbidity.

One of the central challenges in understanding heart-brain comorbidity is how to delineate, at the mechanistic level, the “cross-organ pathological pathways” that link myocardial injury to cerebral ischemia. Existing studies have largely focused on hemodynamic disturbances and neurogenic injury,⁵ whereas the concept of an “immune-inflammatory bridge” has gained increasing attention in recent years. Immune cells play a pivotal regulatory role in heart-brain comorbidity. Numerous clinical and experimental studies have demonstrated that circulating immune cells can transmit pathological signals between the myocardium and brain through cross-organ migration, secretion of inflammatory mediators, and cell-cell communication. On the one hand, after IS, activation of microglia and infiltration of monocytes drive focal neuroinflammation and disruption of the blood-brain barrier (BBB), while excessive sympathetic activation and cytokine storms can induce metabolic derangements and structural remodeling of remote myocardium.^6,9,12 On the other hand, in the context of the chronic systemic inflammatory state that characterizes HF, circulating monocytes and neutrophils remain persistently activated, infiltrate the myocardium, and participate in adverse remodeling, whereas cardiomyocyte-derived cytokines such as cardiotrophin-1 (CT-1) can cross the BBB, activate microglia and astrocytes, disrupt central immune homeostasis, and thereby increase susceptibility to IS and related brain injury.¹² Collectively, these findings suggest that peripheral immune cells, particularly monocytes and their derivative macrophages, may serve as a common cross-organ mediator of pathological progression between HF and IS. The immune–inflammatory bridge is not mediated by immune cells alone. Vascular wall structural cells, including endothelial cells and vascular smooth muscle cells, contribute to barrier integrity and vascular remodeling and regulate immune-cell recruitment and phenotypic transitions through adhesion molecule expression and chemokine secretion, thereby jointly shaping cross-organ pathological signaling in HF–IS comorbidity. Following IS, excessive activation of cardiac sympathetic nerves can upregulate adhesion molecules on endocardial and coronary endothelial cells, providing “anchoring-migration” sites for circulating monocytes, thereby promoting their infiltration into the myocardium and differentiation into pro-inflammatory macrophages. HF-associated myocardial injury can, in turn, influence the brain through cross-organ pathological signaling. Circulating inflammatory mediators and neural signals released during myocardial injury cooperatively regulate classical monocytes, enabling them to recognize adhesion molecules such as ICAM-1 and Itga4 on cerebrovascular endothelial cells and subsequently breach the BBB. Meanwhile, the inflammatory milieu within the brain, together with monocyte-derived cytokines, drives aberrant proliferation and phenotypic switching of cerebrovascular smooth muscle cells, which not only aggravates cerebrovascular stenosis but also, via neurohumoral regulation, exacerbates myocardial interstitial fibrosis, forming a vicious cycle of “myocardial remodeling-cerebrovascular pathology”.^13–15

Advances in genetic epidemiology and single-cell technologies have provided powerful tools for dissecting heart-brain comorbidity. Mendelian randomization (MR), which uses genetic variants as instrumental variables (IVs) to infer causal relationships, leverages the quasi-random allocation of genotypes at conception to mitigate confounding and reverse causation, thereby offering crucial support for elucidating the causal genetic basis of heart-brain comorbidity and linking genetic associations to cross-organ pathological mechanisms. The integrated application of single-cell and spatial transcriptomics allows high-resolution characterization of cross-organ cellular heterogeneity, molecular expression patterns, and intercellular communication networks, thereby precisely bridging genetic associations with cellular pathology and addressing a key knowledge gap between “genetic signals” and “cellular function”.¹⁶ Bidirectional MR analyses based on genome-wide association studies (GWAS) have preliminarily suggested a bidirectional positive causal relationship between HF and IS in terms of genetic susceptibility. However, these studies have largely remained at the level of disease phenotypes and have not clarified the underlying cellular and molecular mechanisms that implement such genetic links.¹⁷ In this context, the present study aimed to elucidate the cellular and molecular mechanisms linking HF and IS by establishing a multi-omics framework integrating “genetic validation, cellular dissection, molecular mining, and translational evaluation.” First, we reassessed the bidirectional causal relationship between HF and IS using large-scale GWAS summary statistics via two-sample bidirectional MR. We then integrated scRNA-seq datasets from myocardial tissue of a murine HFpEF model and brain tissue of a murine MCAO model to construct a shared immune/stromal lineage atlas, with particular attention to dynamic changes in monocytes and their subpopulations. Next, we incorporated peripheral-blood eQTL data and performed gene-level MR to screen key candidate genes potentially mediating HF–IS comorbidity, followed by genetic-consistency validation using SMR¹⁸ and colocalization analyses. In parallel, spatial transcriptomics was leveraged to define the spatial localization of prioritized genes within the IS tissue microenvironment. In the HF single-cell dataset, Ly6C⁺ monocytes were further stratified by expression of the prioritized gene, followed by cell–cell communication, pathway enrichment, and pseudotime analyses to delineate putative regulatory networks and dynamic features. Finally, guided by genetically supported candidate targets, we predicted candidate drugs and conducted a preliminary translational assessment using druggability metrics, and further performed in vitro experiments to evaluate the effects of perturbing the key monocyte gene on microglial injury and inflammatory responses, as well as the intervention effects of candidate drugs, thereby providing testable therapeutic hypotheses and potential treatment strategies for HF–IS comorbidity.

Methods

Bidirectional MR Analysis Between HF and IS

This study employed bidirectional two-sample MR to evaluate the causal relationship between HF and IS. GWAS summary statistics for HF (OpenGWAS ID: ebi-a-GCST009541; GWAS Catalog: GCST009541; 47,309 cases and 930,014 controls; total N = 977,323)¹⁹ and IS (OpenGWAS ID: ebi-a-GCST90018864; GWAS Catalog: GCST90018864; 11,929 cases and 472,192 controls; total N = 484,121)²⁰ were obtained from the IEU OpenGWAS database (https://gwas.mrcieu.ac.uk/). Case definitions for HF and IS followed those used in the original GWAS studies, as detailed in the corresponding primary publications. The GWAS summary statistics used here were adjusted using the covariate sets defined in the original GWAS analyses. Harmonization relied on the publicly released, fully adjusted summary estimates. All participants were of European ancestry to reduce bias from population stratification. All GWAS summary statistics used in this study were publicly available, de-identified, aggregated data. Therefore, no additional permissions or institutional review board approval were required for the present analyses, and data use complied with the terms of the respective resources.

HF and IS were alternately treated as the exposure and the outcome. Single-nucleotide polymorphisms (SNPs) strongly associated with the exposure (P < 5×10⁻⁸) were selected as IVs, and linkage disequilibrium (LD) clumping was applied (clump_kb = 10,000; clump_r2 = 0.001) to obtain independent IVs. Exposure and outcome summary statistics were harmonized to align effect alleles; ambiguous palindromic variants were removed, and variants unavailable in the outcome dataset were excluded. Instrument strength was assessed using the F-statistic (F = β²/SE²), and only non-weak IVs (F > 10) were retained. All IVs were selected to meet the three core assumptions of Mendelian randomization: (i) relevance (IVs are robustly associated with the exposure), (ii) independence (IVs are not associated with confounders), and (iii) exclusion restriction (IVs affect the outcome only through the exposure).

MR analyses were conducted using the “TwoSampleMR” R package. Causal effects of the exposure on the outcome were estimated using the inverse-variance weighted (IVW), MR-Egger, weighted median, simple mode, and weighted mode methods,²¹ and the results were presented in forest plots. Effect estimates are reported as odds ratios (ORs) with 95% confidence intervals. IVW was considered the primary analysis because it typically provides higher statistical efficiency and robust estimates when most IVs are valid and there is no evidence of directional horizontal pleiotropy.²² Heterogeneity was assessed using Cochran’s Q statistic; when P < 0.05 indicated heterogeneity, a random-effects IVW model was used instead of a fixed-effects IVW model.²² Directional horizontal pleiotropy was evaluated using the MR-Egger intercept test,²³ with P > 0.05 suggesting no evidence of directional pleiotropy. Leave-one-out sensitivity analyses were performed to assess robustness.²⁴ In addition, scatter plots were generated to visualize SNP–exposure and SNP–outcome associations, and funnel plots were used to inspect the symmetry of single-SNP causal estimates as an aid to detecting directional pleiotropy or heterogeneity patterns.

Acquisition and Dimensionality Reduction of scRNA-Seq Data

Single-cell transcriptomic data for HF and IS were obtained from the Gene Expression Omnibus (GEO; https://www.ncbi.nlm.nih.gov/geo/). The HF dataset was derived from myocardial scRNA-seq profiles of mice with heart failure with preserved ejection fraction (HFpEF) and matched controls in GSE295382, whereas the IS dataset was obtained from brain-tissue scRNA-seq profiles of mice subjected to middle cerebral artery occlusion (MCAO) and sham-operated controls in GSE174574. As no suitable publicly available human brain-tissue scRNA-seq dataset for IS was identified for the present analysis, mouse datasets were used for both HF and IS to avoid cross-species integration bias in single-cell comparisons.

Quality control and preprocessing were performed for both datasets using the “Seurat” R package. Low-quality cells were filtered based on the number of detected genes (nFeature_RNA) and the proportion of mitochondrial transcripts (percent.mt), retaining cells with nFeature_RNA between 200 and 5,000 and percent.mt < 20%. Expression matrices were log-normalized, and ScaleData was applied in downstream analyses. Highly variable genes were used for principal component analysis (PCA), and the elbow plot was used to select the major dimensions for subsequent analyses. To generate a comparable low-dimensional representation across myocardial- and brain-derived datasets and to mitigate source/batch-related effects on clustering and visualization, “dataset origin” was specified as the batch variable and Harmony was applied for batch-effect correction in PCA space. Graph-based nearest-neighbor construction and Louvain clustering were performed using the Harmony embeddings, followed by two-dimensional visualization with uniform manifold approximation and projection (UMAP).

Given that this study aimed to profile the shared cellular ecosystem of HFpEF and MCAO from a “heart–brain comorbidity” perspective, we first performed integration and batch-effect correction at the whole-dataset level to obtain a unified low-dimensional reference space for lineage alignment and robust annotation. On this basis, to construct an immune-inflammatory landscape that is directly comparable across diseases, we extracted immune/stromal lineages shared between the two datasets for HF–IS comparative analyses. Tissue-specific parenchymal cells (eg, glia/glia) were used only for within-disease descriptions and were not included in cross-model comparisons between HF and IS.

Extraction and Analysis of Monocyte Subtypes

In the integrated scRNA-seq dataset, major cellular lineages were annotated based on canonical marker genes, identifying T cells, B cells, monocytes, endothelial cells, and smooth muscle cells. Monocytes represented the most abundant immune population and were therefore selected as the primary immune cell population of interest. All monocytes were then extracted to generate a monocyte subset, within which data were re-normalized, highly variable genes were re-identified, and ScaleData and PCA were re-performed. Based on the elbow plot and JackStraw test, the top 25 principal components were selected for downstream analyses. Using Seurat, a graph-based neighbor graph was constructed and Louvain clustering was applied to define monocyte subclusters, which were visualized in two dimensions using UMAP. Differential expression analysis across monocyte subclusters was performed with FindAllMarkers (|log₂FC| > 0.58, adjusted P < 0.05) to identify subcluster-specific marker genes. Subclusters were functionally annotated according to marker-gene expression patterns and previously reported subtype markers, and annotation consistency and biological plausibility were assessed using UMAP feature plots.

Cell-Cell Communication Analysis

Given that Ly6C⁺ monocytes were a key subset commonly enriched in both conditions, we systematically characterized their intercellular communication networks. Based on Seurat-based cell annotations, Ly6C⁺ monocytes and other shared immune/stromal populations were extracted from the HF and IS datasets, and cell–cell communication networks were constructed using the “CellChat” R package. Using CellChat’s built-in mouse ligand–receptor database, signaling interactions between cell types were inferred, and the outgoing strength, incoming strength, and overall activity of each signaling pathway were quantified. Communication patterns and enriched ligand–receptor pathways were visualized using chord diagrams and bubble plots.

Gene-Level MR Analysis

Based on the org.Mm.egdb and org.Hs.egdb annotation databases, the “biomaRt” R package was used to map Ly6C⁺ monocyte marker genes identified in the mouse HF dataset to their human orthologs. Gene entries that could not be uniquely matched were removed. Peripheral-blood eQTL summary statistics were obtained from the eQTLGen Consortium (https://www.eqtlgen.org/, N = 31,684).²⁵ Genetically predicted gene expression was treated as the exposure and IS GWAS summary statistics were used as the outcome to perform gene-level two-sample MR. Instruments were extracted only for the candidate gene set derived from the Ly6C⁺ monocyte markers after mouse-to-human ortholog mapping. For each gene, SNPs associated with gene expression at genome-wide significance (P < 5 × 10⁻⁸) were selected as instruments. LD clumping was applied to obtain independent instruments using r² < 0.001 within a 10,000 kb window. Instrument strength was assessed using the F-statistic (F = β²/SE²), and only strong instruments (F > 10) were retained. The eQTL and IS GWAS summary statistics were harmonized to align effect alleles. Ambiguous palindromic variants were removed, and variants unavailable in the outcome dataset were excluded.

The primary MR estimates were obtained using the IVW method. MR-Egger, weighted median, simple mode, and weighted mode were applied as complementary estimators for robustness assessment. Cochran’s Q test was used to assess heterogeneity. Directional horizontal pleiotropy was evaluated using the MR-Egger intercept test. Leave-one-out analyses were performed to examine whether single variants disproportionately drove the results. Scatter plots, funnel plots, and forest plots were generated to visualize SNP-level consistency and potential bias patterns. For prioritized candidate genes, reverse MR analyses were conducted to reduce concerns about reverse causality. Multiple testing was addressed using false discovery rate (FDR) correction applied to IVW-derived P values. Genes that did not survive FDR adjustment were retained as nominally significant candidates and were further evaluated using orthogonal evidence, including colocalization and cell-type expression analyses. Effect estimates are reported as ORs with 95% confidence intervals and represent the change in IS risk per genetically predicted increase in gene expression.

Key-Gene Prioritization and Multilayer Validation: Expression, Spatial Localization, and Genetic Concordance

Differential Expression in IS Samples and Cell-Stratified Validation

For candidate genes prioritized by MR, IS differential-expression evidence was further integrated to conduct stratified cross-validation at the bulk, cell-type, and monocyte subcluster levels. First, heatmaps were used to display overall expression differences of candidate genes between the IS and control groups. Scatter plots were then used to show expression distributions across cell types. Violin plots were further used to compare candidate-gene expression between IS and control groups at the bulk, cell-type, and monocyte-subcluster levels, with statistical significance evaluated using the Wilcoxon rank-sum test. Subsequently, a UMAP embedding was used to visualize the spatial expression pattern of the key gene Nrip1 across monocyte subclusters in IS.

Spatial Transcriptomic Localization

Using a publicly available spatial transcriptomics dataset of mouse brain injury (GEO: GSE226208), we profiled the lesion-state–dependent spatial distribution of Nrip1 across injured brain tissue. This analysis was performed to delineate the functional localization of Nrip1 within the post-injury tissue microenvironment and to examine its relationship with lesion status and regional tissue context.

Colocalization Analysis

The “coloc” R package was used to perform colocalization analyses between cis-eQTL signals associated with the expression of key genes and IS GWAS signals.²⁶ Using SNPs filtered and harmonized based on the MR analyses, we performed Bayesian colocalization with default prior probabilities p1 = 1×10⁻⁴, p2 = 1×10⁻⁴, and p12 = 1×10⁻⁵, and then calculated the posterior probabilities for the five hypotheses (PPH0-PPH4), where PPH4 represents the scenario in which the same SNP jointly drives both gene expression and IS risk. Colocalization results were visualized using regional association plots and two-dimensional scatter plots of the test statistics. As an external consistency check, top cis-eQTL variants for NRIP1 in the Genotype-Tissue Expression (GTEx) Project v8 whole blood (https://gtexportal.org/)²⁷ were cross-referenced against eQTLGen results.

SMR Analysis

To further verify the central role of NRIP1 in cross-disease regulation between HF and IS, SMR was performed by integrating eQTL summary statistics from the eQTLGen Consortium with disease GWAS summary statistics, using HF and IS GWAS data as outcomes in separate analyses. SMR analyses were conducted using SMR v1.3.1 (smr-1.3.1-win.exe), with the 1000 Genomes Project European (1KG EUR) panel as the LD reference and a MAF threshold of ≥ 0.01. P values from the SMR output were adjusted for multiple testing using the FDR method implemented in R, and significant genes were identified in conjunction with the heterogeneity in dependent instruments (HEIDI) test to control false positives. Significant gene sets from HF-SMR and IS-SMR analyses were then obtained, and their intersection was determined using a Venn diagram to screen key genes jointly supporting heart–brain comorbidity.

Functional Characterization of Nrip1 in HF Monocytes

In the HF scRNA-seq dataset, Ly6C⁺ monocytes were divided into Nrip1-positive (Nrip1+) and Nrip1-negative (Nrip1−) subpopulations using the median Nrip1 expression as the threshold. Cell–cell communication analysis, ligand–receptor pathway enrichment, and KEGG/GO enrichment analyses were then performed to elucidate biological functions and potential molecular regulatory networks.

In addition, the “monocle3” R package was used to construct a single continuous trajectory on the existing HF and IS UMAP embeddings. An inflammatory-like monocyte cluster with high expression of Ly6c2, Ccr2, S100a8, and S100a9 was selected as the common root to order the two cell populations along pseudotime. Expression changes of early inflammatory marker genes, late macrophage-like marker genes, and Nrip1 along pseudotime were compared, with scatter plots and smoothed curves used to depict trends and to evaluate the association between Nrip1 and pseudotime as well as model fit. On this basis, mouse immune-related genes were extracted from the MSigDB C7 immunologic gene sets and combined with Nrip1 to form a candidate key-gene set. Normalized expression values were binarized into “on” and “off” states. Within the GeneSwitches framework, gene state-transition patterns along pseudotime were analyzed to identify “switch” regulatory genes at different stages of cell-state development. Dynamic changes of key genes during cell-state evolution were visualized using expression scatter plots and fitted curves.

Candidate Drug Prediction

To evaluate the druggability of NRIP1 and identify potential interventions, we queried the Drug–Gene Interaction Database (DGIdb, https://dgidb.org/)²⁸ using NRIP1 as the search term to obtain records of known or putative drug–gene interactions. Candidate drugs returned by DGIdb and their corresponding Interaction Scores were extracted and used as a reference measure of interaction-evidence strength.

Molecular Docking

To evaluate the structural plausibility and interaction mode between NRIP1 and candidate compounds, tretinoin and representative retinoic acid–related small molecules predicted by DGIdb were selected for protein–ligand docking analysis. The full-length human NRIP1 structural model was retrieved from the AlphaFold Protein Structure Database²⁹ The model was then prepared in AutoDock Tools by adding polar hydrogen atoms and assigning Gasteiger charges, and the processed structure was saved in PDBQT format as the receptor.³⁰ Because the full-length model contains extensive flexible regions, preliminary blind docking and pocket screening were first performed to identify the most plausible binding region, after which docking analysis was focused on a putative ligand-binding pocket within the structurally ordered core region. Three-dimensional structures of candidate small molecules were retrieved from PubChem and converted to PDBQT format using OpenBabel v2.4.1.³¹ Hydrogen atoms were added, charges were assigned, and ligand conformations were optimized in AutoDock Tools. Molecular docking was then performed in AutoDock Vina v1.2.3 using a semi-flexible docking protocol.^32,33 A cubic grid box centered on the predicted binding region was defined by specifying the grid center coordinates and box dimensions, and the default exhaustiveness parameter was applied. Multiple candidate binding poses were generated for each ligand, and the optimal binding mode was selected based on binding affinity and conformational plausibility. Noncovalent interactions, including hydrogen bonds, hydrophobic contacts, salt bridges, and π–π stacking, were analyzed using the Protein–Ligand Interaction Profiler (PLIP) platform,³⁴ and three-dimensional visualizations were generated in PyMOL v2.6.³⁵

Molecular Dynamics (MD) Simulation

To characterize the dynamic stability of ligand binding at the NRIP1 pocket, all-atom MD simulations of 100 ns were performed for the docked complexes using the GROMACS 2025.2 software package.^36,37 MD simulations were carried out on the pocket-containing local C-terminal structural region of NRIP1 extracted from the AlphaFold model, corresponding to the top-ranked docking region, rather than on the full-length 1158-aa protein, to reduce the influence of highly flexible peripheral regions. Each complex was placed in a periodic cubic simulation box and solvated with the TIP3P water model under the CHARMM36 force field, followed by the addition of Na⁺ and Cl⁻ ions to achieve a physiological salt concentration of 150 mM NaCl. After energy minimization, position-restrained NVT/NPT equilibrations (100 ps each) were conducted, and then a 100 ns production run was carried out with a time step of 2 fs. Trajectories were processed to remove periodic boundary artifacts and to center the system. To comprehensively evaluate the stability and conformational behavior of the protein-ligand complexes, several key structural descriptors were calculated from the simulation trajectories, including the root-mean-square deviation (RMSD) of Cα atoms, radius of gyration (Rg), root-mean-square fluctuation (RMSF) of residues, solvent-accessible surface area (SASA), and the number of hydrogen bonds between protein and ligand. In addition, free energy landscapes (FELs) were constructed to visualize the conformational stability of the complexes. FELs were generated using the first two principal components (PC1 and PC2) as reaction coordinates to depict the free energy distribution in conformational space, where deep blue basins represent thermodynamically favorable conformational states.

To further quantify binding stability and affinity, the last 20 ns of stable trajectories were analyzed using the “gmx_MMPBSA” script (which relies on MMPBSA.py v16.0),³⁸ and binding free energies were calculated with the molecular mechanics Poisson-Boltzmann surface area (MM/PBSA) method.³⁹ In this framework, the binding free energy (ΔG_bind) is decomposed into a gas-phase molecular mechanics component (ΔG_gas) and a solvation free energy component (ΔG_solv). ΔG_gas consists of van der Waals (ΔE_vdW) and electrostatic (ΔE_elec) interaction energies, whereas ΔG_solv is composed of polar (ΔG_polar) and nonpolar (ΔG_nonpolar) solvation energies. The overall relationship can be expressed as:. Lower ΔG_bind values indicate a more stable protein-ligand interaction and stronger binding affinity.

Absorption, Distribution, Metabolism, Excretion and Toxicity (ADMET) Prediction of Candidate Drugs

To systematically evaluate the druggability of the candidate drugs, ADMET properties were predicted using the pkCSM platform (http://biosig.unimelb.edu.au/pkcsm)⁴⁰. The SMILES strings of each compound were entered into the ADMET module of pkCSM, with a particular focus on gastrointestinal (GI) absorption and the BBB permeability parameter logBB. According to established thresholds, logBB > 0.3 was considered to indicate high BBB permeability, −1 < logBB ≤ 0.3 moderate permeability, and logBB ≤ −1 low permeability or poor penetration into the central nervous system,⁴¹ on the basis of which the candidate drugs were classified into high, moderate, and low BBB permeability groups. For toxicity assessment, predictions of hERG potassium channel inhibition (hERG I/II inhibition) and hepatotoxicity were extracted and recorded to comprehensively evaluate the potential risk of cardiotoxicity and hepatotoxicity for each compound.

In vitro Validation

Materials

Human CD14⁺ monocytes (Cat. No. RC-T3148) were purchased from Shanghai Yaji Biotechnology Co., Ltd. Human microglial HMC3 cells (Cat. No. CL-0620), RPMI-1640 Complete Medium (Cat. No. PM150110B), and HMC3 complete medium (Cat. No. CM-0620) were purchased from Procell Life Science and Technology Co., Ltd. (Wuhan, China). BeyoGold™ 24-well cell culture inserts (Cat. No. FTW070-48Ins) were purchased from Beyotime Biotechnology (Shanghai, China). Tretinoin (Cat. No. HY-14649) and Dimethyl sulfoxide (DMSO, Cat. No. HY-Y0320C) were purchased from MedChemExpress (MCE, USA). Human Monocyte Nucleofector Kit (Cat. No. VVPA-1007) was purchased from Vicbio Biotechnology Co., Ltd (China). The High-Capacity cDNA Reverse Transcription Kit (Cat. No. 4368813) was purchased from Thermo Fisher Scientific (USA). The CCK-8 assay kit and related stop solution were purchased from Nanjing Fengfeng Biomedical Technology Co., Ltd. (Nanjing, China). The BCA Protein Assay Kit (Cat. No. E-BC-K318-M), LDH assay kit, and ELISA kits were purchased from Elabscience Biotechnology Co., Ltd. (Wuhan, China). Anti-NRIP1 (Cat. No. Ag26585), anti-CCR2 (Cat. No. Ag33071), anti-β-actin (Cat. No. 20536-1-AP), HRP-conjugated goat anti-rabbit secondary antibody (Cat. No. SA00001-2), and HRP-conjugated goat anti-mouse secondary antibody (Cat. No. SA00001-1) were purchased from Proteintech (Wuhan, China). Transwell inserts with a pore size of 0.4 μm were used to establish the indirect co-culture system.

Establishment of the Transwell Co-Culture Model and Group Assignment

To evaluate the effects of monocyte NRIP1 alterations on ischemia-related microglial injury and inflammatory phenotypes, an indirect CD14⁺ monocytes/HMC3 Transwell co-culture model was established. HMC3 cells were seeded in the lower chamber, and CD14⁺ monocytes were seeded in the upper chamber of 0.4-μm Transwell inserts.

The cells were divided into five groups: Control, Model, siNRIP1, tretinoin, and siNRIP1 + tretinoin. The Control group consisted of normally cultured HMC3 cells co-cultured with siNC-transfected CD14⁺ monocytes. The remaining four groups consisted of OGD/R-treated HMC3 cells co-cultured with differently treated CD14⁺ monocytes. The siNRIP1 and negative control siRNA (siNC) were designed and synthesized by Sangon Biotech Co., Ltd. (Shanghai, China), and the sequences are listed in Table S1. CD14⁺ monocytes were transfected with siRNA by nucleofection using the Human Monocyte Nucleofector Kit according to the manufacturer’s instructions, and knockdown efficiency was verified by RT-qPCR and Western blot analysis. CD14⁺ monocytes in the upper chamber received the following treatments: Control and Model groups were transfected with scrambled siRNA (siNC) and treated with vehicle (DMSO); the siNRIP1 group was transfected with siNRIP1 and treated with vehicle (DMSO); the tretinoin group was transfected with siNC and pretreated with tretinoin; and the siNRIP1 + tretinoin group was transfected with siNRIP1 and pretreated with tretinoin. The working concentration of tretinoin was determined based on preliminary CCK-8 dose-finding experiments (Figure S1). Twenty-four hours after transfection, CD14⁺ monocytes were pretreated with tretinoin (1 μmol/L) for 2 h where appropriate, washed once with PBS to minimize direct carryover of tretinoin into the lower chamber, resuspended in co-culture medium, and then seeded into the upper chambers.

Except for the Control group, HMC3 cells in all other groups were subjected to oxygen-glucose deprivation/reoxygenation (OGD/R). When cell confluence reached 70%–80%, the original medium was replaced with glucose-free, serum-free medium, and the cells were exposed to hypoxic conditions (1% O₂, 5% CO₂, and 94% N₂) for 4 h. The medium was then replaced with reoxygenation medium, and the upper chambers containing the correspondingly treated CD14⁺ monocytes were placed above the lower chambers at the onset of reoxygenation for a further 24 h of co-culture. At the end of co-culture, cells from both chambers and culture supernatants were collected for subsequent analyses.

RT-qPCR Analysis of CD14⁺ Monocytes

At the end of co-culture, CD14⁺ monocytes from the upper chamber were collected, and total RNA was extracted using TRIzol reagent. After RNA concentration and purity were measured, equal amounts of RNA were reverse-transcribed into cDNA according to the manufacturer’s instructions. A SYBR Green-based real-time quantitative PCR system was used to detect the mRNA expression levels of NRIP1, IL1B, TNF, and CCR2, with GAPDH as the internal control. Primers were synthesized by Sangon Biotech Co., Ltd. (Shanghai, China), and the sequences are listed in Table S2. Melting-curve analysis was performed after amplification to verify specificity. Relative expression levels were calculated using the 2^-ΔΔCt method. At least three independent biological replicates were included for each group, and each sample was analyzed in triplicate.

Western Blot Analysis of CD14⁺ Monocytes

At the end of co-culture, CD14⁺ monocytes from the upper chamber were collected and lysed in RIPA buffer containing protease inhibitors. Protein concentrations were measured using the BCA assay. Equal amounts of protein were separated by SDS-PAGE and transferred onto PVDF membranes. After blocking at room temperature, the membranes were incubated overnight with primary antibodies against NRIP1, CCR2, and β-actin, followed by incubation with the corresponding HRP-conjugated secondary antibodies. Protein bands were visualized using enhanced chemiluminescence, and densitometric analysis was performed using ImageJ software, with β-actin as the loading control.

CCK-8 Assay for HMC3 Cell Viability

At the end of co-culture, the Transwell inserts were removed, and CCK-8 working solution was added to the lower-chamber HMC3 cell culture system and incubated in the dark for 2 h. A suitable volume of the reaction mixture was then transferred to a 96-well plate, and absorbance (A) was measured at 450 nm to assess HMC3 cell viability under different treatment conditions. Cell viability was calculated using GraphPad Prism 9.0 according to the following formula:

LDH Assay for HMC3 Cell Injury

At the end of co-culture, culture supernatants from lower-chamber HMC3 cells were collected, and lactate dehydrogenase (LDH) release was measured according to the manufacturer’s instructions to evaluate membrane damage and cell injury. Blank wells, low-control wells, and high-control wells were included: blank wells contained only reaction system and medium for background correction; low-control wells represented spontaneous LDH release from HMC3 cells under routine culture conditions; and high-control wells represented maximal LDH release after treatment with the kit-provided lysis solution. After the reaction, absorbance values were measured, and background values from blank wells were subtracted before calculating cytotoxicity using the following formula: , where A_{low control} represents absorbance from spontaneous-release wells and A_{high control} represents absorbance from maximal-release wells.

RT-qPCR Analysis of Lower-Chamber HMC3 Cells

At the end of co-culture, lower-chamber HMC3 cells were collected, and total RNA was extracted using TRIzol reagent and reverse-transcribed into cDNA. RT-qPCR was performed to detect the mRNA expression levels of IL1B, TNF, NOS2, and ARG1, using GAPDH as the internal control. PCR amplification conditions and data-processing procedures were the same as those described in RT-qPCR Analysis of CD14⁺ Monocytes. At least three independent biological replicates were included for each group, and each sample was analyzed in triplicate. Primer sequences are listed in Table S2.

ELISA for Inflammatory Cytokines in Lower-Chamber HMC3 Supernatants

At the end of co-culture, culture supernatants from lower-chamber HMC3 cells were collected and centrifuged to remove cells and debris. TNF-α and IL-1β levels were then measured using ELISA kits according to the manufacturers’ instructions.

Statistical Analysis and Reporting Guideline

Statistical analyses were performed using the “stats” and “car” packages in R software (version 4.3.0), and data visualization was carried out using the “ggplot2” package. All results are presented as mean ± standard error of the mean (SEM). Continuous variables were tested for normality using the Shapiro–Wilk test and for homogeneity of variance using Levene’s test. For data satisfying both assumptions, one-way ANOVA followed by Tukey’s HSD post hoc test was performed. For normally distributed data with unequal variances, Welch’s one-way ANOVA followed by the Games–Howell post hoc test was used. For non-normally distributed data, the Kruskal–Wallis test followed by Dunn’s test was applied. A P value < 0.05 was considered statistically significant.

The MR analyses in this study are reported in accordance with the STROBE-MR guideline, and the completed STROBE-MR checklist is provided as Table S4. The study protocol was not preregistered. An overview of the study design and analytical workflow is shown in Figure 1.

Figure 1.

Study workflow.

Results

Bidirectional Causal Association Between HF and IS

Bidirectional two-sample MR analyses were performed to investigate the genetic causal relationship between HF and IS, and the results indicated a robust bidirectional positive causal association between the two diseases. When HF was treated as the exposure and IS as the outcome, both the IVW and weighted median methods suggested a positive association, indicating that HF increases the risk of IS (IVW (multiplicative random effects): OR = 1.346, 95% CI: 1.035–1.749, P = 0.027; weighted median: OR = 1.252, 95% CI: 1.041–1.506, P = 0.017). Conversely, when IS was treated as the exposure and HF as the outcome, the IVW and weighted median methods similarly demonstrated a significant positive association, indicating that IS increases the risk of HF (IVW (multiplicative random effects): OR = 1.248, 95% CI: 1.033–1.507, P = 0.022; weighted median: OR = 1.143, 95% CI: 1.004–1.301, P = 0.044). Sensitivity analyses showed that the MR-Egger intercept tests did not detect horizontal pleiotropy in either direction (P > 0.05), whereas Cochran’s Q tests indicated the presence of heterogeneity (P < 0.05) and thus the multiplicative random-effects IVW model was applied for correction. Leave-one-out analyses confirmed the robustness of the causal estimates in both directions. Scatter plots provided an intuitive visualization of the bidirectional positive causal relationships, and funnel plots showed an approximately symmetric distribution, suggesting no evident directional pleiotropy or systematic bias. The bidirectional MR results are summarized in forest plots (Figure 2).

Figure 2.

Bidirectional two-sample MR analyses between HF and IS. (A–D) MR analyses with HF as the exposure and IS as the outcome. (E–H) MR analyses with IS as the exposure and HF as the outcome.

Transcriptomic Landscape and Monocyte Subset Heterogeneity in HF and IS

After quality control and preprocessing, both scRNA-seq datasets retained cells of sufficient quality for downstream analyses. Following batch-effect correction with Harmony, cells from different samples were well mixed in the low-dimensional space, and between-batch differences were markedly reduced. During integrated analysis, tissue-specific parenchymal cells such as cardiomyocytes and neurons were excluded, and only shared immune/stromal lineages were retained. UMAP-based clustering combined with canonical marker expression identified five major cell lineages: T cells, monocytes, endothelial cells, smooth muscle cells, and B cells. Further analysis showed that monocytes accounted for the highest proportion among these shared lineages and exhibited distinct distribution patterns across HF models, IS models, and their respective controls. Therefore, subsequent analyses focused on monocytes to further explore their potential roles and mechanisms in HF and IS (Figure 3).

Figure 3.

Initial filtering and processing of scRNA-seq data. (A) Violin plots showing quality control metrics for sequenced cells, including the number of detected genes (nFeature_RNA), RNA counts (nCount_RNA), percentage of mitochondrial genes (percent.mt), and percentage of hemoglobin genes (percent.HB). (B) PCA of the scRNA-seq data. (C) Scatter plot of the Harmony-corrected embedding after batch-effect adjustment. (D) UMAP visualization of the distribution of different cell types. (E) Cell-type proportions and cell counts.

After secondary clustering of monocytes, 25 subpopulations were identified. Based on the expression profiles of classical marker genes, these subclusters were categorized into two major phenotypic groups: Ly6C⁺ inflammatory monocytes and Ly6C⁻ relatively homeostatic monocytes. Compared with their respective control groups, both HF and IS samples showed a markedly higher proportion of Ly6C⁺ monocytes than Ly6C⁻ monocytes, suggesting that expansion of inflammatory monocytes may represent a shared immune feature of HF and IS (Figure 4).

Figure 4.

Analysis of monocyte subclusters. (A) PCA plot showing the distribution of monocytes. (B) Scatter plot of monocyte distribution after batch-effect correction with the Harmony algorithm. (C) Plot for principal component selection. (D) UMAP plot of 25 monocyte subclusters. (E) UMAP visualization of the spatial distribution of monocyte subclusters. (F) Proportions of monocyte subclusters in HF, IS, and control groups. (G) Expression profiles of marker genes across the 25 monocyte subclusters. (H) Expression profiles of marker genes in monocyte subclusters.

Cell-Cell Communication of Ly6C⁺ Monocytes in HF and IS

To investigate the microenvironmental interaction mechanisms underlying the expansion and functional changes of Ly6C⁺ monocytes, we constructed ligand-receptor communication networks between Ly6C⁺ monocytes and other cell populations in the HF and IS datasets. In HF samples, Ly6C⁺ monocytes exhibited overall weak communication activity, interacting only modestly with Ly6C⁻ monocytes, with the relevant signaling predominantly concentrated along the Ptprc–Mrc1 axis. This may be related to immune cell dysfunction, chronic inflammation, or immune tolerance in the HF context, which could suppress intercellular communication to some extent. In contrast, in IS samples, Ly6C⁺ monocytes established a much more active communication network with T cells, B cells, Ly6C⁻ monocytes, endothelial cells, and smooth muscle cells, with the most prominent interaction occurring between Ly6C⁺ monocytes and B cells. Pathway enrichment analysis indicated that these interactions were primarily enriched in the Ptn–Ncl signaling pathway, whereas the crosstalk specifically between Ly6C⁺ monocytes and B cells was predominantly enriched in the App–Cd74 pathway, which showed the highest signaling strength. Collectively, these results suggest that, in the context of acute ischemic injury, Ly6C⁺ monocytes not only expand in number but also engage in tight communication with lymphocytes and vascular-associated cells via specific ligand-receptor axes, potentially coordinating post-ischemic immune–inflammatory cascades and vascular remodeling (Figure 5).

Figure 5.

Cell-cell communication analysis in HF and IS. (A and B) Global cell-cell interaction network and Ly6C⁺ monocyte-centered interaction network in HF samples. (C) Enrichment of ligand-receptor signaling pathways involved in interactions between Ly6C⁺ monocytes and other cell types in HF samples. (D and E) Global cell-cell interaction network and Ly6C⁺ monocyte-centered interaction network in IS samples. (F) Enrichment of ligand-receptor signaling pathways involved in interactions between Ly6C⁺ monocytes and other cell types in IS samples.

Identification of Key Marker Genes and MR Analysis

Differential Expression-Based Screening of Key Marker Genes in HF-Associated Ly6C⁺ Monocytes

Differential expression analysis of Ly6C⁺ monocytes from HF samples identified 52 key marker genes, providing a candidate gene pool for subsequent investigation of cross-disease regulatory mechanisms between HF and IS.

MR Analysis Identifies Genes with Causal Effects on IS

To clarify the causal relationships between the above genes and IS, these genes were treated as exposures and their corresponding eQTLs were used as IVs to perform MR analyses with IS as the outcome. The analyses identified four genes (FBXL5, GUSB, NRIP1, and RNASE6) that showed nominally significant inverse causal associations with IS and were therefore designated as key candidate genes. MR-Egger regression and Cochran’s Q tests further confirmed that none of these four genes showed evident horizontal pleiotropy or heterogeneity (P > 0.05), supporting the reliability of the MR estimates. Forest plots were generated to visually summarize the negative causal effects of these genes on IS risk. Consistent with this, IVW results indicated that all four candidate genes were negatively associated with IS and linked to reduced IS risk (FBXL5: OR = 0.93873, 95% CI: 0.89524–0.98433, P = 0.009; GUSB: OR = 0.95566, 95% CI: 0.91381–0.99943, P = 0.047; NRIP1: OR = 0.91622, 95% CI: 0.85520–0.98159, P = 0.013; RNASE6: OR = 0.97467, 95% CI: 0.95562–0.99411, P = 0.011). However, none of the four genes remained significant after FDR correction; therefore, nominal significance is reported here and supported by the consistency of pleiotropy tests and sensitivity analyses.

Focusing on the core gene NRIP1, the MR scatter plot with linear fit clearly illustrated a negative association between NRIP1 expression and IS risk. Leave-one-out sensitivity analysis did not suggest that the association was driven by any single SNP, and the funnel plot appeared approximately symmetric, with no evidence of directional horizontal pleiotropy. Reverse MR provided no evidence supporting reverse causality (P > 0.05), further reinforcing the inverse causal association between NRIP1 and IS (Figure 6).

Figure 6.

Gene-level MR and spatial validation of candidate genes. (A) Scatter plot of MR results. (B) Forest plot of MR analyses assessing the associations between candidate key genes and IS. (C) Forest plot of reverse MR analysis evaluating the effect of genetic liability to IS on NRIP1 expression. (D) Funnel plot of SNPs associated with NRIP1 expression. (E) Leave-one-out sensitivity analysis for SNPs associated with NRIP1 expression. (F) Scatter plot showing the linear regression trend between NRIP1 expression and IS risk. (G) Spatial transcriptomic expression pattern of NRIP1.

Spatial Transcriptomic Characterization of Nrip1 Regional Distribution in Injured Brain Tissue

Spatial transcriptomic analysis revealed marked regional heterogeneity of Nrip1 expression in injured mouse brain tissue. Nrip1 was relatively highly expressed in the relatively intact (non-lesional) brain parenchyma, whereas its expression was reduced in the lesion core and peri-lesional (border) regions. This lesion-associated spatial pattern indicates that lower Nrip1 expression co-localizes with areas of more severe tissue disruption and microenvironmental perturbation, suggesting a potential role for Nrip1 in maintaining local microenvironmental homeostasis. Collectively, these findings provide tissue-level spatial evidence consistent with a potentially protective relevance of Nrip1 in ischemic brain injury (Figure 6).

Stratified Cross-Validation of MR-Associated Genes in IS

To further characterize the four MR-implicated candidate genes in IS, we examined their expression differences and cellular distribution patterns in the IS dataset. At the overall level, only Nrip1 and Rnase6 were significantly downregulated in the IS group compared with normal controls, whereas FBXL5 and GUSB were both significantly upregulated. Based on this, we next focused on Nrip1 and Rnase6, whose expression trends were consistent with the MR results. At the cell-type level, Nrip1 was markedly downregulated in IS monocytes, in line with its overall expression pattern, whereas Rnase6 was expressed in fewer than 10% of monocytes, suggesting a relatively limited cellular basis for acting as a key mediator within monocytes. At the monocyte subcluster level, Nrip1 showed no significant differential expression in Ly6C⁺ monocytes in IS, but displayed a downregulation pattern in Ly6C⁻ monocytes that was consistent with the overall monocyte trend, indicating that its expression difference may be primarily driven by the Ly6C⁻ subset.

Integrating MR effect sizes with these expression validations, Nrip1 not only showed a stronger negative association with IS risk than Rnase6, but also maintained a higher proportion of expressing cells in total monocytes and in Ly6C⁻ monocytes, providing a plausible biological basis for functional impact. Consequently, Nrip1 was prioritized as the main focus for subsequent analyses. Rnase6, by contrast, was not selected as a primary target at this stage due to its very low expression proportion in monocytes. The other two genes were not prioritized at this stage, not because they lack research value, but because their higher expression in the IS group is at odds with the protective effects suggested by MR analyses. This discrepancy needs to be clarified through preliminary studies such as human-mouse homology verification, assessment of technical bias, and exploration of potential dissociation between expression and function.

Notably, Nrip1 was initially identified as a differentially expressed gene in HF-associated Ly6C⁺ monocytes, and its genetically protective association with IS aligns with its overall downregulation in IS. However, Nrip1 did not differ significantly within the Ly6C⁺ subset in IS. This observation does not necessarily negate the contribution of Ly6C⁺ monocytes to the HF–IS immune bridge; instead, it suggests that Nrip1 may exert a “threshold/homeostasis-maintaining” mode of regulation, and/or that its functional carrier in modulating IS may reside in Ly6C⁻ monocytes and/or other cell types, which should be clarified in subsequent mechanistic analyses (Figure 7).

Figure 7.

Expression patterns of MR-identified candidate genes in IS. (A) Heatmap showing the expression levels of candidate genes in IS. (B) Bubble plot illustrating the expression of candidate genes across major cell types. (C) Violin plots comparing the expression of candidate genes between IS and control groups. (D) Expression levels of Nrip1 across different cell types in IS. (E) Expression levels of Nrip1 across monocyte subclusters in IS. (F) UMAP visualization of Nrip1 expression within monocyte subclusters in IS. **P < 0.01, and ***P < 0.001 for the indicated comparisons between the control and treatment groups.

Colocalization Analysis Validating the Specificity of the NRIP1 Genetic Association

Colocalization analysis provided evidence that the NRIP1 cis-eQTL signal and the IS GWAS association at the NRIP1 locus were consistent with a shared causal variant (PPH4 = 0.999). Within this region, rs9975024 represented the lead signal in both datasets and showed concordant association patterns in the bivariate statistic scatter plot (Figure 8). To assess cross-resource consistency, we compared top NRIP1 cis-eQTL variants in GTEx v8 whole blood with eQTLGen and found that multiple GTEx top signals corresponded to eQTLGen lead variants with consistent effect alleles and directions (Table S3), supporting the robustness of the NRIP1 regulatory signal in blood.

Figure 8.

Colocalization and SMR analyses. (A) Colocalization analysis of NRIP1 eQTLs with IS GWAS signals. (B) Overlap of SMR-identified genes between IS and HF. (C and D) SMR Manhattan plots for IS and HF. (E and F) Associations of NRIP1 expression with IS and HF risk.

SMR Analysis Confirms the Cross-Disease Genetic Regulatory Role of NRIP1

SMR analysis identified 69 genes that were simultaneously associated with the risk of both IS and HF, among which NRIP1 emerged as a shared core gene. Scatter plots showed that NRIP1 was significantly negatively associated with the risk of both IS and HF. Collectively, these findings not only corroborate the genetic causal association between NRIP1 and IS, but also indicate a close link between NRIP1 and HF susceptibility, providing direct genetic evidence supporting NRIP1 as a key cross-disease gene connecting HF and IS (Figure 8).

Potential Expression Patterns and Cellular Functions of the Key Gene

Given that Nrip1 was identified as a differentially expressed gene in Ly6C⁺ monocytes from HF samples, and that genetic, differential-expression, and spatial transcriptomic evidence collectively suggested its potential protective role in IS, we further explored its regulatory mechanisms in the HF scRNA-seq dataset at the levels of cell-cell communication and functional pathways. UMAP visualization showed that Nrip1 was overall expressed at low levels within the Ly6C⁺ monocyte cluster. Using CellChat, we constructed communication networks for Nrip1+ and Nrip1− Ly6C⁺ monocytes. The results indicated that both populations exhibited generally weak overall communication activity. However, Nrip1+ Ly6C⁺ monocytes displayed stronger homotypic interactions and showed detectable communication with B cells, whereas this feature was not observed in Nrip1− Ly6C⁺ monocytes. Ligand–receptor pathway enrichment further suggested that Nrip1-associated differences in monocyte communication showed a degree of interaction specificity. Interactions involving Nrip1− Ly6C⁺ monocytes with themselves and with Ly6C⁻ monocytes were enriched only in the Ptprc–Mrc1 axis. In contrast, homotypic interactions among Nrip1+ Ly6C⁺ monocytes were mainly enriched for App–Cd74, Ccl2–Ccr2, Ccl6–Ccr2, Icam1–ITGAM_ITGB2 and Ptprc–Mrc1 pathways, while their interactions with Ly6C⁻ monocytes showed selective enrichment for App–Cd74 and Ptprc–Mrc1. Notably, enrichment of the anti-inflammatory App–Cd74 pathway was observed exclusively in interactions between Nrip1+ Ly6C⁺ monocytes and B cells. This pattern suggests that Nrip1 expression is a prerequisite for Ly6C⁺ monocytes to establish anti-inflammatory communication with B cells, and that Nrip1 may help maintain immune homeostasis in HF and secondarily constrain IS-related pathology by modulating these pathways.

To clarify the underlying molecular mechanisms of these functional differences, we performed KEGG and GO enrichment analyses for Ly6C⁻ monocytes, Nrip1+ Ly6C⁺ monocytes and Nrip1− Ly6C⁺ monocytes. KEGG analysis showed that Nrip1− Ly6C⁺ monocytes were significantly enriched in immune response pathways (Leishmaniasis, human cytomegalovirus infection), metabolism and cellular function pathways (oxidative phosphorylation, phagosome), and neurodegeneration-related pathways (Alzheimer disease, Parkinson disease). In contrast, Nrip1+ Ly6C⁺ monocytes showed only weak enrichment in these pathways, and the significance of some pathways (eg, Leishmaniasis, diabetic cardiomyopathy) was notably reduced. GO enrichment analysis further indicated that Nrip1− Ly6C⁺ monocytes were enriched in biological process (BP) terms related to cytoplasmic translation and translation at synapse, cellular component (CC) terms such as cytosolic ribosome and ribosomal subunit, and molecular function (MF) terms including structural constituent of ribosome and mRNA binding. By contrast, Nrip1+ Ly6C⁺ monocytes were specifically enriched in BP terms such as lymphocyte differentiation and regulation of lymphocyte differentiation, CC terms such as vacuolar membrane and lysosomal membrane, and MF terms such as signaling receptor adaptor activity and protein heterodimerization. Taken together, these findings indicate that Nrip1 expression endows Ly6C⁺ monocytes with an anti-inflammatory, communication-oriented profile, while restraining excessive activation of pro-inflammatory and hypermetabolic pathways. This reshapes their target-cell communication patterns and favors a homeostasis-oriented pathway activity spectrum, providing direct evidence at both the cell-cell communication and molecular pathway levels for functional divergence of Ly6C⁺ monocytes in HF immune regulation and IS-related pathological processes (Figure 9).

Figure 9.

Nrip1-associated cell communication patterns and functional pathway characteristics. (A) UMAP plot showing Nrip1 expression across monocyte subclusters in HF. (B) Cell-cell communication networks of Nrip1+ and Nrip1− Ly6C⁺ monocytes. (C and D) Bubble plots of ligand-receptor signaling pathways involved in interactions between Nrip1− and Nrip1+ Ly6C⁺ monocytes and other cell types. (E and F) KEGG and GO enrichment analyses of metabolic pathways in Ly6C⁻ monocytes, Nrip1+ Ly6C⁺ monocytes and Nrip1− Ly6C⁺ monocytes.

Cell Trajectory and Pseudotime Analysis

To delineate the dynamic functional evolution of Ly6C⁺ monocytes in HF, we first constructed a single backbone trajectory based on Ly6C⁺ monocytes from HF samples. Cells were continuously distributed along pseudotime from an early highly inflammatory state to a later macrophage-like differentiated state. Early-stage marker genes Ccr2, Ly6c2 and S100a8/S100a9 were highly expressed at the beginning of the trajectory and then rapidly downregulated, whereas late-stage genes such as Adgre1, C1qa/C1qb and Lgals3 gradually increased toward the terminal phase, indicating that this trajectory captures the progression of Ly6C⁺ monocytes in HF from acute inflammatory activation toward a differentiated state. When Nrip1 was projected onto the same pseudotime axis, its expression in HF Ly6C⁺ monocytes remained globally low with only mild fluctuations, without the monotonic increase/decrease or switch-like transitions observed for the marker genes. GeneSwitches analysis further showed that, at the genome-wide level, scores of immune-related gene modules increased significantly along pseudotime, whereas Nrip1, due to its poor model fit, was not classified as a “switch gene” on the pseudotime axis. Together with the significant positive correlation between global immune gene expression and pseudotime, these findings suggest that the dynamic amplification of inflammatory cascades in HF Ly6C⁺ monocytes is mainly driven by other immune switch genes, while Nrip1 is more likely to act as a relatively stable background protective factor rather than a key temporal regulator of monocyte activation (Figure 10).

Figure 10.

Cell trajectory and pseudotime analysis. (A) Trajectory of Ly6C⁺ monocytes in HF, colored by Nrip1 expression. (B) Pseudotime expression trends of early and late marker genes. (C) Pseudotime analysis of Nrip1 in Ly6C⁺ monocytes from HF samples. (D) Histogram of binarized gene expression values. (E) Expression trajectories of key genes along the pseudotime axis. (F) Scatter plot showing the relationship between global gene expression levels and pseudotime.

We next constructed a similar pseudotime trajectory for IS monocytes. IS monocytes likewise exhibited a continuous distribution from an early highly inflammatory state to a later macrophage-like differentiated state. Ly6c2, Ccr2 and S100a8/S100a9 were enriched at the early end of pseudotime and declined rapidly, whereas Adgre1, C1qa/C1qb and Lgals3 progressively increased during the middle-to-late phases, supporting the biological plausibility of the trajectory. Along this temporal axis, Nrip1 again showed globally low expression with only modest local fluctuations, with a much narrower dynamic range than that of the marker genes, closely mirroring the pattern observed in HF. The concordant trajectories in HF and IS jointly support the notion that Nrip1 functions primarily as a “homeostatic susceptibility” factor in the setting of HF–IS comorbidity, rather than as a core temporal regulator driving the acute activation cascade of monocytes (Figure S2).

Prediction of NRIP1-Associated Candidate Drugs and Structure-Based Interaction Analysis

Based on NRIP1, we queried the DGIdb database and identified two drug entries, tretinoin and “retinoic acid agent” (Table 1), with tretinoin belonging to the retinoic acid agent class. To evaluate the binding potential of these candidates to NRIP1, we selected representative small molecules from the retinoic acid agent category and performed molecular docking against NRIP1. AutoDock Vina predicted binding energies of −7.8, −8.1, −8.5 and −7.1 kcal/mol for tretinoin, alitretinoin, Arotinoid acid and bexarotene, respectively, suggesting favorable predicted interaction potential. Among these, Arotinoid acid showed the lowest binding energy, suggesting the most favorable predicted affinity for NRIP1. PLIP and PyMOL visualization of the docking complexes suggested that tretinoin was predicted to interact with NRIP1 mainly through hydrogen bonds with ASP-1057 (2.5 Å) and SER-1058 (2.8 Å), whereas Arotinoid acid was predicted to form a hydrogen bond with GLN-1073 (2.5 Å).

Table 1.

Predicted NRIP1-Targeting Drugs and Interaction Scores from the DGIdb Database

Gene	Drug	Interaction Score
NRIP1	Tretinoin	1.68
NRIP1	Retinoic acid agent	0.86

Although molecular docking can predict binding poses and affinity, conventional (rigid or semi-flexible) docking is limited in its ability to capture conformational flexibility and solvent dynamics under near-physiological conditions. To more comprehensively assess the stability and interaction features of the complexes, we therefore performed all-atom MD simulations for the docked NRIP1–tretinoin and NRIP1–Arotinoid acid complexes. By monitoring RMSD, RMSF, Rg, SASA and hydrogen bond occupancy, we characterized the conformational evolution of each complex under quasi-physiological conditions. The results showed that protein backbone RMSD remained mostly < 0.5 nm and ligand RMSD < 0.2 nm in both systems, indicating that the ligands remained stably associated with the predicted binding region and that the complexes were relatively stable throughout the simulations. Rg values were mainly distributed between 1.75 and 1.85 nm with minimal temporal fluctuation, and SASA values remained within 115–130 nm², suggesting no major conformational expansion or collapse. Residue-level RMSF values were generally < 0.3 nm, with higher peaks restricted to loop or terminal regions, reflecting localized flexibility. The number of hydrogen bonds between ligand and protein fluctuated between 0 and 2, with one persistent hydrogen bond being maintained for most of the simulation time in both complexes, supporting stable hydrogen-bonding interactions. Notably, the NRIP1–Arotinoid acid complex transiently formed up to six hydrogen bonds around 90 ns, then returned to a stable range of 1–3 hydrogen bonds; during this period, both RMSD and Rg remained stable, indicating that this represented normal conformational fluctuation rather than loss of overall complex stability. To further characterize dynamic conformational stability, we constructed FEL plots using the first two principal components (PC1 and PC2) as reaction coordinates. Both NRIP1–tretinoin and NRIP1–Arotinoid acid complexes exhibited well-defined energy basins, indicating stable, low-energy and conformationally compact binding states (Figure 11).

Figure 11.

Molecular docking and MD simulations. (A) Molecular docking pose of NRIP1 with tretinoin. (B) Molecular docking pose of NRIP1 with arotinoid acid. (C–J) All-atom 100 ns MD simulations, including RMSD (C and D), Rg (E), SASA (F), residue-wise RMSF (G), intermolecular hydrogen bond counts (H), and FELs (I and J).

To quantitatively assess binding stability, we extracted the last 20 ns of the equilibrated MD trajectories and calculated the binding free energy (ΔG_bind) of each complex using the MM/PBSA method. Both NRIP1–tretinoin and NRIP1–Arotinoid acid complexes showed marked binding affinity, with ΔG_bind values of −30.39 kcal/mol and −47.49 kcal/mol, respectively (Table 2). Decomposition of the energy terms indicated that the binding stability of the NRIP1–tretinoin complex was mainly supported by the combined contribution of van der Waals interactions and polar solvation energy, whereas the strong binding affinity of the NRIP1–Arotinoid acid complex was predominantly driven by cooperative van der Waals and electrostatic interactions.

Table 2.

Average Binding Free Energies (kcal/Mol) Calculated by the MM/PBSA Method

Energy Contributions	NRIP1–Tretinoin	NRIP1–Arotinoid Acid
ΔEvdW	−26.27	−25.94
ΔEelec	22.12	−56.62
ΔGpolar	−19.50	38.62
ΔGnonpolar	−6.74	−3.54
ΔGgas	−4.15	−82.56
ΔGsolv	−26.24	35.08
ΔGbind	−30.39	−47.49

ADMET Evaluation of Candidate Drugs

To further evaluate the druggability of the NRIP1-targeting candidates, we used pkCSM to perform in silico ADMET prediction for four small molecules. The results showed that tretinoin, alitretinoin, Arotinoid acid and bexarotene all exhibited high GI absorption (> 90%), supporting the feasibility of oral administration. Tretinoin and alitretinoin were predicted to have high BBB permeability (logBB > 0.3), whereas Arotinoid acid and bexarotene showed moderate BBB permeability (logBB = 0.176 and 0.225, respectively), meeting the basic requirement for central nervous system targeting in IS. None of the four compounds was predicted to inhibit hERG channels, suggesting a low risk of arrhythmogenic cardiotoxicity and aligning with safety considerations in HF. With respect to hepatotoxicity, tretinoin, alitretinoin and Arotinoid acid all showed potential liver toxicity signals, whereas bexarotene did not exhibit obvious hepatotoxicity alerts. Taken together with the docking results, MD stability, and ADMET profiles, tretinoin and Arotinoid acid were considered representative candidates with strong NRIP1-binding capacity. By contrast, bexarotene showed a relative safety advantage and may serve as a reference scaffold for subsequent functional validation and lead optimization (Table 3).

Table 3.

pkCSM-Predicted ADMET Properties of the Candidate Drugs

Candidate Drug	GI Absorption (%)	logBB	BBB Class	hERG Inhibition	Hepatotoxicity
Tretinoin	93.488	0.311	High	No	Yes
Alitretinoin	93.488	0.311	High	No	Yes
Arotinoid acid	100.000	0.225	Medium	No	Yes
Bexarotene	100.000	0.176	Medium	No	No

In vitro Experiments Demonstrated That Tretinoin Attenuates Ischemic HMC3 Cell Injury and Inflammation by Modulating Monocyte NRIP1

Effects of siNRIP1 Transfection and Tretinoin Intervention on NRIP1-Related Changes in Monocytes

RT-qPCR and Western blot analyses were performed to verify the effectiveness of siNRIP1 transfection and tretinoin intervention in CD14⁺ monocytes. RT-qPCR results showed that, compared with the control group, NRIP1 mRNA expression was slightly decreased and IL1B mRNA expression was slightly increased in the model group, without statistical significance, whereas TNF and CCR2 mRNA expression levels were increased (P < 0.01 and P < 0.05, respectively), suggesting that the ischemia-related microenvironment induced inflammatory alterations in monocytes. Compared with the model group, NRIP1 mRNA expression was significantly decreased in the siNRIP1 group (P < 0.001), while IL1B, TNF, and CCR2 mRNA expression levels were significantly increased (P < 0.01 and P < 0.001), indicating that siNRIP1 transfection effectively downregulated NRIP1 in monocytes and promoted inflammatory changes. Compared with the model group, NRIP1 expression was increased in the tretinoin group (P < 0.01), whereas IL1B, TNF, and CCR2 expression levels were decreased (P < 0.01). Compared with the siNRIP1 group, NRIP1 expression was partially restored in the siNRIP1 + tretinoin group (P < 0.05), while IL1B, TNF, and CCR2 expression levels were reduced (P < 0.05), suggesting that tretinoin partially alleviated the inflammatory phenotype associated with NRIP1 knockdown.

Western blot results showed that, compared with the control group, NRIP1 protein levels were decreased and CCR2 protein levels were increased in the model group (P < 0.05, P < 0.01). Compared with the model group, the siNRIP1 group exhibited a further decrease in NRIP1 protein and a further increase in CCR2 protein (P < 0.05), whereas the tretinoin group showed increased NRIP1 protein expression and decreased CCR2 protein expression (P < 0.05). Compared with the siNRIP1 group, the siNRIP1 + tretinoin group showed partial restoration of NRIP1 protein expression and reduced CCR2 protein levels (P < 0.05, P < 0.001). These findings indicate that siNRIP1 transfection effectively downregulated NRIP1 expression in monocytes, whereas tretinoin intervention partially restored NRIP1 expression and ameliorated the alterations associated with NRIP1 knockdown.

Changes in HMC3 Cell Viability and Injury

CCK-8 and LDH assays were used to evaluate HMC3 cell viability and injury under co-culture conditions with differently treated CD14⁺ monocytes and to assess the effects of tretinoin. CCK-8 results showed that, compared with the Control group, HMC3 cell viability was significantly decreased in the Model group (P < 0.01), indicating that OGD/R markedly reduced HMC3 cell viability and that the model was successfully established. Compared with the model group, HMC3 cell viability was further reduced in the siNRIP1 group (P < 0.01), suggesting that NRIP1 knockdown in monocytes aggravated HMC3 cell injury and reduced cell viability. In contrast, HMC3 cell viability was increased in the tretinoin group relative to the model group (P < 0.05), indicating that pretreatment of monocytes with tretinoin partially improved HMC3 cell viability. Further comparison showed that HMC3 cell viability was partially restored in the siNRIP1 + tretinoin group compared with the siNRIP1 group (P < 0.05).

LDH results showed that, compared with the control group, the percentage of HMC3 cytotoxicity was significantly increased in the model group (P < 0.001), indicating that OGD/R aggravated HMC3 cell injury. Compared with the model group, HMC3 cytotoxicity was further increased in the siNRIP1 group (P < 0.001), suggesting that NRIP1 knockdown in monocytes further exacerbated ischemic HMC3 cell injury. Compared with the model group, the percentage of HMC3 cytotoxicity was reduced in the tretinoin group (P < 0.05), indicating that pretreatment of monocytes with tretinoin alleviated HMC3 cell injury. Further comparison showed that HMC3 cytotoxicity was decreased in the siNRIP1 + tretinoin group compared with the siNRIP1 group (P < 0.05). Collectively, these results indicate that NRIP1 knockdown in monocytes aggravates ischemic HMC3 cell injury, whereas tretinoin partially reverses this detrimental effect.

Changes in Inflammatory Gene Expression and Cytokine Secretion

RT-qPCR and ELISA were used to evaluate inflammatory changes in HMC3 cells under ischemia-like conditions and the effects of tretinoin intervention. RT-qPCR results showed that, compared with the control group, IL1B, TNF, and NOS2 mRNA expression levels were significantly increased in the model group (P < 0.05 and P < 0.01), whereas ARG1 expression was significantly decreased (P < 0.01), indicating that OGD/R induced inflammatory activation in HMC3 cells. Compared with the model group, IL1B, TNF, and NOS2 expression levels were further increased in the siNRIP1 group (P < 0.05 and P < 0.01), while ARG1 expression was further decreased (P < 0.01), indicating that NRIP1 knockdown in monocytes exacerbated the inflammatory response in ischemic HMC3 cells. Compared with the model group, IL1B, TNF, and NOS2 expression levels were reduced in the tretinoin group (P < 0.05 and P < 0.01), whereas ARG1 expression was increased (P < 0.01), suggesting that pretreatment of monocytes with tretinoin attenuated inflammatory activation in HMC3 cells. Further comparison showed that IL1B, TNF, and NOS2 expression levels were decreased in the siNRIP1 + tretinoin group compared with the siNRIP1 group (P < 0.05), while ARG1 expression was partially restored (P < 0.05), indicating that tretinoin still exerted a partial protective effect in the context of NRIP1 knockdown, although its effect was weaker than that observed in the tretinoin group.

ELISA results showed that, compared with the control group, TNF-α and IL-1β levels in HMC3 cell culture supernatants were significantly increased in the model group (P < 0.01 and P < 0.001). Compared with the model group, TNF-α and IL-1β levels were further increased in the siNRIP1 group (P < 0.05 and P < 0.01), whereas these inflammatory cytokines were decreased in the tretinoin group (P < 0.05 and P < 0.01). Compared with the siNRIP1 group, TNF-α and IL-1β levels were reduced in the siNRIP1 + tretinoin group (P < 0.05). Taken together, these findings suggest that NRIP1 knockdown in monocytes may aggravate the inflammatory response of ischemic HMC3 cells through paracrine mechanisms, whereas tretinoin intervention partially attenuates these changes (Figure 12).

Figure 12.

Effects of monocyte NRIP1 on ischemic HMC3 cell injury and inflammation and the intervention effect of tretinoin. (A) Schematic diagram of the Transwell co-culture model and treatment workflow. (B–E) RT-qPCR analysis of NRIP1, IL1B, TNF, and CCR2 mRNA expression in CD14⁺ monocytes. (F–H) Western blot analysis of NRIP1 and CCR2 protein expression in CD14⁺ monocytes. (I) CCK-8 assay of HMC3 cell viability. (J) LDH assay of HMC3 cytotoxicity. (K–N) RT-qPCR analysis of IL1B, TNF, NOS2, and ARG1 mRNA expression in HMC3 cells. (O and P) ELISA of TNF-α and IL-1β levels in HMC3 cell culture supernatants. Data are presented as mean ± SEM from three independent experiments. *P < 0.05, **P < 0.01, and ***P < 0.001 for the indicated comparisons (Model vs Control, siNRIP1 vs Model, tretinoin vs Model, and siNRIP1 + tretinoin vs siNRIP1).

Discussion

Previous clinical and epidemiological studies have convincingly demonstrated a high comorbidity between HF and IS. From the perspective of the “heart–brain axis,” these two conditions can mutually reinforce each other through multiple mechanisms, including hemodynamic disturbances, autonomic dysregulation, and systemic inflammation coupled with immune imbalance. However, most existing insights remain at a macroscopic pathophysiological level. Even MR studies, which have suggested a bidirectional causal relationship between HF and IS from a genetic standpoint, still fall short of explaining how such genetic links are instantiated as concrete cross-organ cellular and molecular alterations. Building on this gap, we integrated MR, scRNA-seq, spatial transcriptomics, colocalization, SMR, and preliminary in vitro validation to systematically dissect the immunogenetic basis of HF–IS comorbidity and to evaluate the translational relevance of the prioritized target and candidate drug. Overall, we obtained three main findings: (i) cross-disease single-cell integration revealed a synchronous expansion of inflammatory Ly6C⁺ monocytes in HFpEF myocardium and MCAO brain tissue, with a more active immune communication network under acute ischemic conditions, suggesting that these cells may serve as a key cellular hub for cross-organ inflammatory cascades; (ii) multi-omics integration identified NRIP1 as a cross-disease key gene with genetically protective effects and drug-development potential, implying that it may participate in HF- and IS-related immune homeostasis by modulating the inflammatory threshold and communication programs of monocytes; and (iii) candidate drug prediction and druggability assessment highlighted retinoic acid–related small molecules as potential agents targeting NRIP1 for intervention in HF–IS comorbidity, while preliminary in vitro experiments further supported that tretinoin may alleviate microglial injury and inflammatory activation by modulating monocyte NRIP1, thereby providing a testable translational entry point.

At the level of genetic causality, Zhang et al were the first to apply bidirectional multivariable MR to interrogate the causal relationship between HF and IS. They reported a bidirectional positive association in genetic susceptibility, suggesting that HF not only increases the risk of IS, but that stroke itself may, through secondary cardiac structural remodeling and functional impairment, in turn accelerate HF progression.¹⁷ Building on these findings, our study leveraged large-scale GWAS data from European-ancestry populations and used bidirectional two-sample MR to further confirm a robust bidirectional causal relationship between HF and IS. More importantly, we did not stop at genetic associations at the disease-phenotype level. Instead, we integrated multi-omics approaches to extend causal inference beyond “disease-disease” relationships to specific cell lineages and genetic loci, thereby constructing a more comprehensive evidence chain for heart-brain comorbidity across three dimensions, namely genetic liability, immune cell lineages, and candidate molecular targets.⁴²

A growing body of evidence suggests that circulating immune cells, together with the “immunoinflammatory bridge” formed through their interactions with vascular wall cells, may serve as a critical mediator of cross-organ pathological transmission between myocardial injury and cerebral ischemia. In this context, we focused on immune and stromal cell lineages and analyzed scRNA-seq datasets from myocardial tissue in a mouse HFpEF model and brain tissue in a mouse MCAO model. By applying unified quality-control criteria, Harmony-based batch-effect correction, and integrative clustering, we constructed a shared cellular lineage atlas across the two diseases and identified T cells, monocytes, endothelial cells, smooth muscle cells, and B cells as core cell types. Against this background, we observed a marked expansion of inflammatory Ly6C⁺ monocytes in both HF and IS models, indicating that Ly6C⁺ monocyte expansion may represent a shared immunological feature of the two conditions, which is highly consistent with previous reports. In HF, metabolism-related genes and the Wnt signaling pathway are aberrantly activated in circulating classical monocytes, and this phenotype correlates positively with the extent of myocardial fibrosis and HF functional class.⁴³ In IS models, IL-1β induced by cerebral ischemia drives epigenetic reprogramming of bone marrow hematopoietic stem cells, resulting in a long-lasting bias toward Ly6C⁺ monocyte production. These cells subsequently migrate across organs to the heart and differentiate into proinflammatory macrophages, promoting myocardial fibrosis via effector molecules such as MMP9, whereas blockade of IL-1β or the CCR2/CCR5 axis significantly attenuates cardiac dysfunction.^44,45 Meanwhile, excessive sympathetic activation after IS induces upregulation of adhesion molecules such as ICAM-1 in cardiac endothelial cells, providing “anchoring-migration” sites for circulating Ly6C⁺ monocytes and further amplifying brain-to-heart inflammatory coupling.¹⁴ Collectively, these lines of evidence support the notion that Ly6C⁺ monocytes and their macrophage derivatives not only contribute to the course of HF and IS individually, but also constitute a pivotal cellular hub mediating cross-organ propagation of inflammatory signals between the two diseases.

Further CellChat-based communication analysis showed that, in the HF model, ligand–receptor interactions between myocardial Ly6C⁺ monocytes and surrounding cells were relatively limited, mainly manifesting as a low-intensity interaction axis with Ly6C⁻ monocytes, suggesting a more “convergent” communication landscape in the setting of chronic remodeling. In contrast, in the IS model, Ly6C⁺ monocytes formed a comparatively more active ligand–receptor interaction network with T cells, B cells, and vascular-associated cells, and these interactions were enriched in multiple inflammatory and immunoregulatory pathways. Notably, monocyte–B cell signaling represented by pathways such as App–Cd74 was markedly strengthened, consistent with recent observations that B cells undergo phenotypic remodeling after IS and participate in inflammatory regulation and immune-cell recruitment.⁴⁶ This “low-intensity in HF versus high-intensity in IS” pattern is not contradictory to the chronic inflammatory state of HF. Rather, chronic remodeling often corresponds to lower-amplitude yet more persistent signal maintenance, tolerance, and homeostatic resetting, whereas acute ischemic injury is more likely to trigger a high-intensity, broad-spectrum amplification of intercellular communication networks. Compared with prior studies that separately profiled immune landscapes in myocardium or brain, our cross-disease single-cell integration further supports the plausibility that Ly6C⁺ monocytes function as a key immune hub in HF–IS comorbidity and participate in cross-organ transmission of inflammatory cascades, providing lineage-level evidence for the “immunoinflammatory bridge.”

At the molecular level, gene-level MR indicated that NRIP1 genetic variants were negatively associated with IS risk through eQTL effects. Colocalization analysis indicated that the NRIP1 eQTL signal and the IS GWAS signal shared the same causal variant, and SMR analysis further supported an association of NRIP1 with the genetic susceptibility to both HF and IS. Together with the expression levels and spatial distribution of Nrip1 in monocytes and brain tissue, these findings form a multidimensional evidence chain supporting NRIP1 as a cross-disease protective hub gene that links the genetic liabilities of HF and IS. Building on this evidence, we stratified Ly6C⁺ monocytes by Nrip1 expression and performed CellChat analysis using scRNA-seq data from the myocardium of HF mice. We found that Nrip1+ Ly6C⁺ monocytes exhibited stronger homotypic interactions and more prominent anti-inflammatory App–Cd74 signaling directed toward B cells than Nrip1− Ly6C⁺ monocytes. By contrast, Nrip1− Ly6C⁺ monocytes were significantly enriched for pathways related to infection responses, oxidative phosphorylation, and neurodegenerative diseases, displaying a proinflammatory lineage characterized by high metabolic activity, elevated translational programs, and heightened activation propensity. These features suggest that Nrip1 acts as a homeostatic gatekeeper along the continuum from ventricular remodeling to the post-ischemic inflammatory cascade by modulating the inflammatory response threshold and intercellular communication patterns of Ly6C⁺ monocytes. This finding extends the role of Nrip1 beyond its conventional identity as a regulator of myocardial metabolism to that of a regulator of immune-cell homeostasis, thereby positioning it to concurrently shape ventricular remodeling and the post-ischemic inflammatory cascade in the context of HF–IS comorbidity.

Notably, Nrip1 was initially identified as a differentially expressed gene in Ly6C⁺ monocytes in HF; however, within monocyte subsets in IS, its expression differences were mainly concentrated in the Ly6C⁻ monocyte subset, whereas no significant difference was observed in the Ly6C⁺ monocyte subset. This pattern clearly highlights the disease-dependent specificity of the cellular carriers through which Nrip1 exerts cross-disease regulation, suggesting that it may act via distinct monocyte subsets in HF and IS. Pseudotime trajectory analyses further showed that, along monocyte differentiation trajectories in both HF and IS, Nrip1 expression displayed only low-amplitude fluctuations over time and did not exhibit a typical “on-off” switch-like pattern, instead resembling a background factor involved in maintaining cellular homeostasis. Consistently, prior cross-disease studies have reported that key genes influencing both cardiovascular and neurodegenerative disorders tend to shape disease evolution by regulating cellular homeostasis rather than acute stress responses,⁴⁷ and the behavior of Nrip1 observed here closely aligns with this principle. NRIP1 (RIP140) is a canonical nuclear receptor coregulator that plays important roles in cardiovascular and metabolism-related diseases by modulating energy metabolism, mitochondrial functional homeostasis, and inflammatory transcriptional networks. Evidence indicates that NRIP1, as a downstream target of YY1, contributes to myocardial ischemia-reperfusion injury by promoting hypoxia-reoxygenation-induced cardiomyocyte damage and mitochondrial dysfunction.⁴⁸ It has also been shown to maintain mucosal immune homeostasis by suppressing retinoic acid-driven transcriptional programs for T-cell gut homing, whereas functional variants can lead to excessive immune activation and dysregulated inflammation.⁴⁹ Recent animal studies further demonstrate that deletion or deficiency of RIP140 in HF mouse models enhances myocardial fuel metabolic capacity, attenuates ventricular remodeling, and improves cardiac function, implying that RIP140 may function as a cell-type-dependent signaling node, primarily influencing remodeling through energy metabolism in cardiomyocytes while preferentially regulating inflammatory homeostasis and intercellular communication in immune cells.⁵⁰ Integrating our genetic and single-cell evidence, Nrip1 may modulate monocyte metabolic programs and inflammatory response thresholds, thereby shaping homeostatic maintenance and effector-lineage differentiation in both HF and IS and ultimately determining the magnitude of ventricular remodeling and the post-ischemic inflammatory cascade. Its downregulation in Ly6C⁻ monocytes may weaken repression of inflammatory gene networks and increase susceptibility to a proinflammatory state, which is consistent with the downregulation of Nrip1 and reduced expression in lesion regions observed in IS.

At the translational level, we predicted candidate drugs using the DGIdb database and performed multidimensional validation integrating molecular docking, MD simulations, ADMET evaluation, and in vitro experiments. Retinoic acid derivatives were supported as potential NRIP1-targeting agents, as they not only exhibited strong binding affinity to NRIP1 with favorable conformational stability, but also showed high gastrointestinal absorption, measurable BBB permeability, and a low predicted risk of arrhythmogenic cardiotoxicity, thereby aligning well with the safety and delivery requirements of cross-organ intervention in HF–IS comorbidity. Furthermore, in the CD14⁺ monocytes/HMC3 Transwell co-culture system, monocyte NRIP1 knockdown aggravated OGD/R-induced HMC3 cell injury and inflammatory activation, whereas tretinoin partially restored NRIP1 expression in CD14⁺ monocytes, reduced CCR2 and inflammatory mediator levels, and alleviated HMC3 cell injury and inflammatory responses. Notably, the neuroprotective effects of retinoic acid in central nervous system injuries, including IS, are supported by substantial experimental evidence. In MCAO models, pretreatment with retinoic acid alleviates neurological deficits, BBB disruption, and excessive glial activation by modulating the ubiquitin-proteasome system and suppressing NF-κB pathway activation.^51–53 In addition, all-trans retinoic acid has been reported to protect against acute IS by modulating neutrophil polarization, reducing neutrophil extracellular trap formation, and suppressing STAT1-related inflammatory signaling.⁵⁴ Together with our findings, these data suggest that retinoic acid may exert dual heart–brain protective effects, at least in part, through modulation of NRIP1-mediated immunoinflammatory responses, thereby providing a novel therapeutic avenue for HF–IS comorbidity. Nevertheless, some candidate compounds may still carry potential hepatotoxicity risks,⁵⁵ a challenge that could be addressed through structural optimization, targeted delivery systems, or combination treatment strategies. More broadly, drug development studies indicate that targets supported by human genetic evidence have substantially higher success rates in clinical development than those lacking such support.^56,57 Incorporating SMR and colocalization, in addition to MR, at the target-prioritization stage may further strengthen the credibility and translational priority of candidate targets. Therefore, by integrating MR, SMR, and colocalization during target prioritization, our study identified NRIP1 as a candidate target with robust genetic support, which may improve the likelihood of success in subsequent drug development efforts.

A major methodological strength of this study lies in the construction of a multilayered and mutually reinforcing evidence framework. First, by integrating MR, colocalization, and SMR analyses, we translated the association between HF and IS into a more credible candidate-gene level, thereby reducing, to some extent, the risk of false-positive findings arising from linkage disequilibrium and pleiotropy. Second, the joint use of cross-disease scRNA-seq and spatial transcriptomics enabled genetic signals to be further mapped onto specific immune cell lineages and their tissue microenvironments. Combined with cell-cell communication and trajectory analyses, this strategy characterized key cellular states from both spatial and dynamic perspectives, thereby providing biologically interpretable support for conventional GWAS and eQTL evidence. On this basis, we further introduced candidate-drug prediction and performed preliminary translational screening through molecular docking, MD simulation, and ADMET evaluation. More importantly, we additionally incorporated in vitro experiments, which provided preliminary experimental support for the biological plausibility and translational relevance of the prioritized target and candidate drug. Taken together, genetic causal-consistency evidence, cellular and spatial tissue-level evidence, and in vitro functional and translational-feasibility evidence jointly form a complementary multidimensional evidence framework, which helps more robustly define the key immune hub, potentially actionable target, and translational therapeutic leads in HF–IS comorbidity.

Several limitations of this study should be acknowledged. First, the GWAS datasets used for MR analyses were primarily derived from European-ancestry populations. Although this helps reduce bias from population stratification, the generalizability of the genetic architecture of HF, IS, and the regulatory effects of NRIP1 to other ancestries remains uncertain and requires validation in multi-ethnic populations and multicenter cohorts. Second, the gene-level MR and colocalization analyses relied mainly on peripheral-blood eQTL data, whereas immune regulation in the myocardium and brain exhibits marked cell-type specificity. The absence of single-cell eQTL datasets for specific subsets such as Ly6C⁺ monocytes may therefore limit the precision of cell-type–specific inference. Third, the cross-disease scRNA-seq analyses were based on mouse HFpEF and MCAO models. Murine Ly6C⁺ monocytes are not fully equivalent to human classical CD14⁺ monocytes, and experimental mouse models cannot fully recapitulate the clinical heterogeneity of human HF–IS comorbidity. Therefore, the monocyte-associated NRIP1 signature identified from mouse scRNA-seq data should be interpreted as mechanistic and hypothesis-generating evidence. Fourth, although FDR correction is commonly applied to control false positives in multiple testing, strict thresholds can be overly conservative in MR settings where only a small number of true causal associations are expected. Thus, nominally significant findings that do not survive FDR correction should be interpreted cautiously yet not dismissed outright, particularly when supported by additional evidence such as colocalization analyses and cell-type–specific expression patterns. In addition, the in vitro validation was based on a simplified CD14⁺ monocytes/HMC3 Transwell co-culture system and siRNA-mediated NRIP1 knockdown. Although this provides preliminary support for the proposed mechanism and drug intervention, it cannot fully recapitulate in vivo cross-organ immune trafficking, cellular heterogeneity, and microenvironmental complexity within the heart–brain axis, and thus warrants further validation using primary-cell systems and in vivo experiments.

Conclusion

Building on the bidirectional causal association between HF and IS, we integrated multi-omics evidence to identify inflammatory Ly6C⁺ monocyte expansion as a shared immune feature and to prioritize NRIP1 as a monocyte-associated cross-disease candidate gene with translational relevance. In vitro validation further provided ischemia-side functional support by showing that tretinoin partially attenuated microglial injury and inflammatory activation through modulation of monocyte NRIP1 under paracrine co-culture conditions (Figure 13).

Figure 13.

Mechanism diagram.

Funding Statement

This study was supported by the National Natural Science Foundation of China (82474352) and the Hunan Provincial Administration of Traditional Chinese Medicine (A2024011).

Key Highlights

1. Cross-disease scRNA-seq integration identified inflammatory Ly6C⁺ monocyte expansion as a shared immune feature in HF–IS comorbidity.

2. NRIP1 was prioritized as a monocyte-associated hub through integrated genetic, cellular, spatial, and translational evidence.

3. Tretinoin partially attenuated ischemia-like microglial injury and inflammation through modulation of monocyte NRIP1.

Data Sharing Statement

The GWAS summary statistics analyzed in this study are available from IEU OpenGWAS (HF: ebi-a-GCST009541; IS: ebi-a-GCST90018864). Peripheral-blood eQTL summary statistics were obtained from the eQTLGen Consortium (https://www.eqtlgen.org/). The scRNA-seq datasets are available in the Gene Expression Omnibus (GEO) under accessions GSE295382 (HFpEF myocardium) and GSE174574 (MCAO brain). The spatial transcriptomics dataset is available in GEO under accession GSE226208. The analysis code is available from the corresponding author upon reasonable request.

Ethics Approval and Consent to Participate

This study was conducted in accordance with the principles of the Declaration of Helsinki. The study used only lawfully obtained, publicly available, de-identified summary-level datasets from open databases, including IEU OpenGWAS, eQTLGen, and GEO. No individual participants were recruited, no intervention was performed, and no identifiable personal information was involved. According to the national policy applicable in China, namely the Measures for Ethical Review of Life Science and Medical Research Involving Humans (2023), research using lawfully obtained public data or anonymized information data may be exempt from ethical review under specified conditions. Therefore, no additional ethical approval or informed consent was required for the present study.

Author Contributions

Changsong Ding: Conceptualization, Methodology, Writing – review & editing, Supervision, Funding acquisition.

Hongyao Chen: Methodology, Formal analysis, Investigation, Visualization, Writing – original draft, Writing – review & editing.

Yao Xiao: Methodology, Formal analysis, Investigation, Visualization, Writing – original draft, Writing – review & editing.

Xiaoyan Yang: Data curation, Investigation, Writing – review & editing.

Ying Zhou: Data curation, Investigation, Writing – review & editing.

All authors have read and approved the final manuscript.

All authors gave final approval of the version to be published; have agreed on the journal to which the article has been submitted; and agree to be accountable for all aspects of the work.

Disclosure

The authors declare that they have no competing interests.