Intracerebral Hemorrhage Induces Monocyte TNF Signaling in Patients That Is Suppressed by BAF312

ABSTRACT

Objective

Intracerebral hemorrhage (ICH) causes high morbidity and mortality, with neurotoxic inflammation driven by infiltrating monocytes. This study is an in‐depth longitudinal examination of the immune response during the first week of ICH in the presence and absence of the immunomodulatory drug BAF312 (Siponimod).

Methods

We performed single‐cell RNA sequencing on peripheral blood samples taken 1, 3, and 7 days post‐ICH in patients treated with BAF312 (5 patients) or placebo (2 patients). The correlation of gene expression with time, drug treatment, measures of ICH severity, and clinical outcome was examined. Results were validated by comparison with previous studies and measurement of plasma cytokine concentration.

Results

A broad peripheral inflammatory response involving both lymphoid and myeloid cells peaked at day 3 post ICH. BAF312 markedly reduced peripheral blood T and B lymphocyte numbers by day 3. BAF312 also impacted the myeloid response, suppressing TNF signaling in classical and nonclassical monocytes. Multiple cytokine signaling pathways were decreased, though BAF312 did not impact plasma cytokine or chemokine concentrations. Surprisingly, increased monocyte TNF signaling was associated with better 90‐day clinical outcome.

Interpretation

These findings demonstrate broad peripheral leukocyte activation after ICH and suggest that during ICH, BAF312 suppresses both lymphoid and myeloid responses. The positive association of monocyte TNF signaling with better outcome points to the positive role of monocytes during the subacute stage of ICH and supports a complex role of monocytes in this disease. Larger studies will determine the extent to which these findings apply beyond the small cohort examined here.

Abbreviations

BBB

blood–brain barrier

CBA

cytometric bead array

GSEA

gene set enrichment analysis

ICH

intracerebral hemorrhage

LFC

log‐fold change

mRS

modified Rankin score

NES

Normalized enrichment score

NIHSS

National Institutes of Health Stroke Scale

S1P

sphingosine‐1‐phosphate

S1PR

sphingosine‐1‐phosphate receptor

1. Introduction

Intracerebral hemorrhage (ICH) is the second most common form of stroke, affecting 2 million people annually worldwide [1, 2]. It is the deadliest type of stroke, with a 30‐day mortality rate of 40% [3, 4, 5]. The influx of blood into the brain parenchyma triggers inflammatory cell death, which leads to the recruitment of blood monocytes, which differentiate into macrophages that contribute to inflammatory damage in the acute and early subacute stages of ICH [6, 7]. After several days, macrophages primarily contribute to reparative processes including phagocytosis of cell debris [6]. T cells worsen outcome, while regulatory T cells limit inflammation and neurological deficits [8, 9, 10, 11]. An effective immunomodulatory therapeutic must limit inflammation‐dependent damage without impairing the reparative roles of immune cells.

S1P is a bioactive lipid that signals through five receptors (S1PR1‐S1PR5). BAF312 binds S1PR1 and S1PR5, downregulating S1PR1 expression and acting as a functional S1PR1 antagonist [12]. Developed as an improved alternative to the S1P analog FTY720 (fingolimod), BAF312 has more selective binding and a reduced elimination half‐life [12, 13, 14]. BAF312 and FTY720 sequester T and B cells in secondary lymphoid organs and are approved for treatment of multiple sclerosis [12, 15, 16, 17, 18, 19]. They also act on nonlymphocyte populations [20]. BAF312 and FTY720 suppress astrocyte activation, while FTY720 reduces microglial production of TNF, IL‐1β, and IL‐6, while increasing production of BDNF and GDNF [21, 22, 23, 24].

BAF312 and FTY720 can reduce inflammation and improve outcome after ICH and subarachnoid hemorrhage in rodents, and FTY720 has shown efficacy in ICH patients [25, 26, 27, 28, 29, 30, 31, 32]. FTY720 and BAF312 treatments are associated with decreased brain edema, fewer brain‐infiltrating T cells, less white matter loss, and improved functional outcomes. The effects are small and are not seen in all studies [33], but there is a compelling need to identify a therapeutic for ICH [34].

Additionally, clinical trial samples can provide insight into the biology of the disease and the impacts of a therapeutic even when the therapeutic does not improve patient outcome [32, 35, 36]. Single‐cell RNA sequencing enables examination of all cell subsets from precious patient samples. RNA sequencing has produced valuable insights into the biology of ICH [35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45], but most extant studies examine animal models and/or utilize bulk RNA sequencing of whole peripheral blood.

The present study utilized single‐cell RNA‐seq in a longitudinal analysis of peripheral blood from 8 ICH patients treated with BAF312 or placebo. In placebo‐treated patients, inflammatory responses peaked at day 3, mirroring the time‐course previously reported in the hematoma [35]. Analysis of 3 published datasets [37, 38, 39] examining bulk PBMC from a total of 190 humans and 10 pigs revealed that the genes upregulated after ICH in bulk PBMC are primarily expressed by monocytes. In BAF312‐treated patients, monocytes exhibited reduced cytokine responsiveness by day 3 of treatment. Contrary to expectation, patients whose monocytes exhibited increased TNF signaling had better long‐term outcomes. This suggests that the beneficial functions of active monocytes may outweigh the negative effects of TNF [46, 47, 48, 49].

2. Methods

2.1. Study Design

Patient enrollment was performed at 3 approved US study sites, according to the parent Stage 2 clinical trial to study the Efficacy, Safety, and Tolerability of BAF312 in patients with ICH, performed by Novartis Pharmaceuticals (https://clinicaltrials.gov/study/NCT03338998). The trial was randomized, placebo‐controlled, and subject‐ and investigator‐blinded. Eligible subjects were 18–85 years of age with an acute supratentorial ICH 10–60 mL within 24 h of enrollment.

BAF312 was administered daily by intravenous up‐titration on days 1–7 and then by 10 mg oral tablets on days 8–14. Participants were followed to day 90 for neurological (mRS and NIHSS) and safety conditions during 3 clinic visits. Recruitment was put on hold due to the COVID‐19 pandemic. After 7 months, an interim analysis was reviewed by the Data Monitoring Committee and the trial was terminated due to lack of potential clinical efficacy. Before the hold, peripheral blood from 8 patients was collected in ACD tubes (BD 364606) at approximately 1, 3, and 7 days post ICH and shipped to our lab overnight in NanoCool refrigerated shippers (Path‐Tec, Midland, GA), consistent with the methods previously described to preserve transcriptional profiles [50].

2.2. Sex as a Biological Variable

Previous studies have found increased type I interferon responses in females during sterile neuroinflammation including ICH [51]. This substudy used every sample available, but 7 of the 8 patients assessed were female, precluding meaningful quantification of sex effects.

2.3. Collection of Patient Clinical Data

CT scans were performed at approximately days 1, 8, and 15 post ICH. ICH volume, perihematomal edema volume, and relative perihematomal edema volume were determined by CT. Functional outcome testing was performed by trained, blinded investigators at each study site.

2.4. Peripheral Blood Sample Processing

PBMC were isolated by Ficoll gradient centrifugation as previously described [35] and then counted and cryopreserved. All 3 timepoints for a given patient were thawed and analyzed by 10X on the same day to limit longitudinal batch effects. Live singlet cells were sorted on a BD FACSAria. Samples from patients 3, 4, 6, 7, and 8 were stained with BioLegend TotalSeq‐A or TotalSeq‐B antibodies to allow for multiplexing. Samples from patients 3 and 4 were thawed and analyzed together, as were patients 7 and 8. The cells were then counted and analyzed on the 10X Genomics platform. Cells from 1 patient had low viability after thawing and were not analyzed by RNA sequencing, though the plasma from this patient was included in cytokine measurements. In total, 7 patients were analyzed by single‐cell RNA‐sequencing (2 placebo, 5 BAF312). One BAF312 patient experienced bradycardia after 2 days and discontinued treatment. This patient is excluded from all analyses except the correlation of gene expression with clinical characteristics and outcome (Figure 5, Figure S4A).

FIGURE 5.

Increased TNF response was associated with better functional outcome. (A–E) GSEA of monocytes, comparing patients with: Day 1 hematoma volume greater than 30 mL versus less than 30 mL (A); Day 1 perihematomal edema greater than 30 mL versus less than 30 mL (B); Day 1 relative perihematomal edema greater than 1 versus less than 1 (C); Day 1 NIHSS of greater than 9 versus 1–9 (D); Day 90 mRS of 0–3 versus 4–6 (E). (F) Differential expression of select inflammatory genes in patients with Day 90 mRS of 0–3 versus 4–6. (G) GSEA as in panel E, but after downsampling and fitting to a linear model that normalizes for ICH score. (H) GSEA of hematoma monocytes/macrophages isolated from patients 4–7 days post ICH, comparing mRS of 0–3 versus 4–6 at 1 year post ICH, from Askenase, et al. 2021.

2.5. Flow Cytometry

Flow cytometry was performed on a BD FACSAria while sorting for total live singlets. Cells were stained with CD45 PE‐Cy7 (BioLegend 304016), CD14 PE (BioLegend 301806), CD56 FITC (BD 562794), CD3 APC‐eFluor 780 (eBioscience 47–0036‐42), CD4 violetFluor 450 (BD 560345), CD8 V500 (BD 560774), and LIVE/DEAD Fixable Red Dead Cell Stain (Thermo L34972). Data were analyzed using FlowJo v10.10.0 (BD).

2.6. Single Cell RNA‐Sequencing Data Processing

Fastq files were aligned to the GRCh38‐2020‐A genome (10X genomics) using Cell Ranger count v7.1.0 (quality metrics in Table S1). Aligned reads were processed in R v4.3.2 with Seurat v4.3.0. Counts from all patients and timepoints were normalized with SCTransform and integrated using reciprocal PCA. Multiplexed samples were demultiplexed with HTODemux. Cells with fewer than 500 RNA features, more than 6000 RNA features, or more than 25% mitochondrial genes were excluded. After QC, 103,301 cells were analyzed from the 21 samples (3 timepoints × 7 patients). Cluster annotation was performed manually, using SingleR v2.2.0 [52] to map datasets from Mulder et al. [53] (monocytes and DCs) and Hao et al. [54] (all populations) onto our dataset. Clusters were pooled into 26 populations and doublets and dead cells were removed (Figure S1A,B). These populations were pooled into 7 lineages (proportions per lineage per patient in Figure S1C). To limit the impact of variable cell recovery, each population was downsampled proportionally: CD16⁺ monocytes, NK, CD8⁺ T, and B cells to 400 cells, CD4⁺ T cells to 800 cells, and classical monocytes to 1600 cells (Figure S1D).

TABLE 1.

Patient baseline characteristics.

Age (years)	Sex	ICH Vol. (mL)	ICH location	GCS	NIHSS (Day 1)	mRS (Premorbid)	ICH score	Treatment
69	F	14.49	Basal ganglia	13	18	0	0	Placebo a
46	F	7.28	Basal ganglia	14	14	0	0	BAF312 b
48	F	21.37	Basal ganglia	14	16	0	0	BAF312
73	F	45.90	Basal ganglia	11	19	0	2	BAF312
70	F	34.32	Occipital lobe	15	3	0	1	Placebo
69	F	24.48	Basal ganglia	12	10	0	1	Placebo
81	F	48.47	Occipital lobe	15	8	1	2	BAF312
64	M	26.63	Occipital lobe	15	4	0	0	BAF312

Abbreviations: GCS, Glascow Coma Scale; mRS, Modified Rankin Score; NIHSS, National Institutes of Health Stroke Scale.

^a Samples from this patient were not analyzed by RNA sequencing.

^b BAF312 treatment in this patient discontinued within 48 h due to bradycardia.

Differential gene expression used FindMarkers and module scoring used AddModuleScore. TNF signaling was scored using genes of the MSigDB HALLMARK_TNFA_SIGNALING_VIA_NFKB geneset. ICH gene sets from published studies were defined as follows: Bai 2024—genes up‐ or downregulated ≥ 2‐fold in bulk peripheral blood in ICH versus hypertension; Durocher 2019—genes up‐ or downregulated ≥ 1.6‐fold in bulk peripheral blood in ICH versus matched vascular risk factors; Walsh 2019—top 100 DEGs in bulk peripheral blood of pigs 6 h post‐ICH versus pre‐ICH.

For association of gene expression with disease severity, good outcome was defined as mRS < 4. Other clinical metrics were dichotomized at thresholds that split the cohort roughly in half while avoiding splitting similar values.

For some analyses, cells were pseudobulked by lineage within each sample. This further corrected for cell number differences [55] and enabled ICH score to be included as a covariate in the linear model for differential expression (edgeR v3.38.4 glmQLFTest). Ingenuity Pathway Analysis Upstream Regulators analysis used a Benjamini‐Hochberg adjusted p‐value cutoff of 0.05, bias‐corrected activation z‐scores are reported. Geneset enrichment analysis (GSEA) was performed with MSigDB Hallmark Gene Sets (fgsea v1.28.0); single‐cell DEG lists were ranked by logFC and pseudobulked DEG lists by t‐statistic (limma v3.58.1 eBayes). Geneset variation analysis (GSVA) was performed with MSigDB Hallmark Gene Sets (GSVA v1.52.3); linear model statistics were calculated with lm (stats v4.4.1).

2.7. Cytometric Bead Array

Plasma cytokine and chemokine concentrations were determined via Cytometric Bead Array (CBA) (BD Biosciences) per manufacturer's protocol. CBA Flex Sets used: TNF (558273), IL‐1α (560153), IL‐1β (558279), IL‐6 (558276), CCL2 (558287), CXCL10 (558280), CD40L (560305), IFNγ (558269), IL‐10 (558274), and IL‐4 (558272). Values below the lower limit of detection (LLOD) were set to LLOD/2.

2.8. Statistics

Single cell differential expression: Seurat FindMarkers (Wilcoxon Rank sum test, LFC threshold of 0.3, genes expressed in > 10% of cells, Bonferroni correction). Pseudobulked differential expression: edgeR negative binomial GLM (glmQLFit and glmQLTest, Benjamini‐Hochburg correction). GSEA: fgsea v1.22.0 fgsea (Benjamini‐Hochberg adjusted p‐values < 0.5 considered significant). All differentially expressed genes and GSEA pathways are available in supplemental data. IPA: Ingenuity Pathways Analysis v9.0 Upstream Regulators Analysis (DEG adj. p‐values < 0.05; IPA pathways with BH‐corrected p‐values < 0.05 and containing > 3 DEGs reported). Cytokine concentration statistics: rstatix v0.7.0 wilcoxon_test (Holm's‐corrected p‐values < 0.05 considered significant). Box plot hinges are first and third quartiles, whiskers extend to the most extreme value within 1.5 × IQR.

2.9. Study Approvals

The Human Investigations Committee (IRB) of Yale University approved this study. Patients or their designated surrogates provided written informed consent according to the Declaration of Helsinki.

3. Results

3.1. Patient Baseline Characteristics

Peripheral blood was collected from 8 patients at approximately 1, 3, and 7 days post ICH (Figure 1A). Patient age ranged from 46 to 81 years (mean 65 ± 12 years) (Table 1). Seven of 8 patients were female. Baseline ICH volume ranged from 7.3 to 48.5 mL (mean 27.9 ± 14.4 mL). Four hemorrhages were in the basal ganglia and 3 in the occipital lobe. Baseline NIHSS ranged from 3 to 19 (mean 11.5 ± 6.2). Additional demographics, clinical characteristics, and outcome data are in Table S2.

FIGURE 1.

Single cell RNA sequencing analysis pipeline and population identification. (A) Schematic of the clinical trial showing days of treatment with BAF312 or placebo, blood draws, functional tests, and CT scans. (B) UMAP of all 103,301 cells analyzed from the 21 integrated samples after QC filtering. (C) Feature plots showing expression of canonical markers of the lineages analyzed further in the study.

3.2. Cell Population Subclustering and Pooling Into Lineages

Patient PBMC were analyzed by single cell RNA‐sequencing as described in Methods. Monocytes, CD4⁺ T cells, CD8⁺ T cells, MAIT cells, NK cells, B cells, dendritic cells, HSPCs, and platelets were identified (Figure 1B). Naïve, memory, and activated cells were distinguishable (Figure 1B and Figure S1B). For subsequent analyses, cell subsets were pooled and focus was on the major lineages: classical monocytes, CD16⁺ monocytes, NK cells, CD4⁺ T cells, CD8⁺ T cells, and B cells. Lineage identity is demonstrated by canonical lineage marker expression (Figure 1C). Proportions of major lineages per patient and timepoint are in Figure S1C.

3.3. Peripheral Inflammatory Response Peaked 3 Days Post ICH

In healthy individuals, PBMC are 10%–20% monocytes and 70%–90% lymphocytes [56]. By scRNA‐seq, day 1 PBMC in our cohort were approximately 50% classical monocytes, 10% CD16⁺ monocytes, 15% NK cells, 15% T cells, and 3% B cells (Figure 2A). Flow cytometry confirmed these proportions (Figure S2A). Lineage ratios remained stable over days 1–7, with no significant changes over time (Figure 2A, Figure S2A).

FIGURE 2.

Peripheral inflammatory response peaks 3 days post ICH. (A) Percentages of adaptive (left) and innate (right) immune populations over time after ICH in placebo‐treated patients. (B) Time course of select inflammatory gene expression in monocytes from placebo‐treated patients after ICH. (C) GSEA with MSigDB Hallmark Pathway genesets of monocytes of placebo‐treated patients, Day 3 or Day 7 versus Day 1. (D) GSEA of different cell lineages from placebo‐treated patients, comparing Day 3 versus Day 1 and Day 7 versus Day 1, showing only results for MSigDB Hallmark Pathways TNFA_SIGNALING_VIA_NFKB geneset. (E) Module scoring of the TNFA_SIGNALING_VIA_NFKB geneset at days 1, 3, and 7 post ICH. (F) Enrichment at day 1 in our cohort of gene modules modulated by ICH in previous studies. Left panel indicates expression, in our cohort, of genes upregulated in the previous study. Right panel indicates expression, in our cohort, of genes downregulated in the previous study.

Monocytes were quickly activated, upregulating inflammatory genes from day 1 to day 3: FOSB (LFC 0.35), JUN (LFC 0.85), NFKBIA (LFC 0.77), IRF1 (LFC 0.35), and CXCL8 (LFC 1.1) (Figure 2B, Table S3). Expression of these genes peaked at day 3 and declined from day 3 to day 7: FOSB (LFC −0.63), JUN (LFC −1.1), NFKBIA (LFC −1.5), IRF1 (LFC −2.3), and CXCL8 (LFC −2.8). GSEA revealed upregulation of multiple inflammatory pathways from day 1 to day 3, including responses to TNF (NES 2.5), IFNα (NES 2.3), and IFNγ (NES 2.1) (Figure 2C, Table S4). The inflammatory response peaked around day 3, as shown by the change from day 1 to day 7: TNF Signaling via NFκB (NES −2.9), Inflammatory Response (NES −2.3), IFNγ Response (NES not signif.). The IFNα Response remained elevated at day 7 (NES 1.9). TNF signaling was modulated in all lineages and we focused our downstream analysis on TNF. TNF signaling increased from day 1 to day 3 in classical monocytes (NES 2.5), CD4⁺ T cells (NES 2.0), CD8⁺ T cells (NES 2.0), B cells (NES 2.4), and NK cells (NES 1.9) (Figure 2D, Table S4). For most populations it peaked at day 3 (day 1 to day 7: classical monocytes (NES −2.9), CD16⁺ monocytes (NES −2.7), CD4⁺ T cells (NES 1.7), CD8⁺ T cells (NES not signif.), and B cells (NES not signif.)). NK cells were an exception: TNF signaling was more enriched at day 7 versus day 1 (NES 2.2) than day 3 versus day 1 (NES 1.9). Module scoring for the MSigDB HALLMARK_TNFA_SIGNALING_VIA_NFKB geneset revealed this pathway was already highly expressed at day 1 post ICH (Figure 2E). TNF signaling was higher in classical and nonclassical monocytes than in other lineages and was not restricted to a subset of monocytes (Figure 2E).

Module scoring was performed to examine expression in our dataset of DEGs identified in prior studies: within 48 h post‐ICH in humans [37], 4–120 h post‐ICH in humans [38], or 6 h post‐ICH in swine [39]. Enrichment of these genesets in our samples was similar at all timepoints; day 1 is shown. Genes upregulated in bulk PBMC in those studies were enriched in classical monocytes (module score 0.07 to 0.3), but not in lymphocytes (module score −0.1 to −0.05) in our cohort (Figure 2F). Genes downregulated in human PBMC [37, 38] were enriched in T cells (module score 0.003 to 0.25) and NK cells (module score 0.14 to 0.17), but not in classical monocytes (module score −0.14 to −0.05) in our cohort (Figure 2F).

3.4. Peripheral Inflammatory Response Decreased by Day 3 in BAF312‐Treated Patients

BAF312‐treated patients demonstrated significant lymphopenia by day 3, the first posttreatment timepoint (Figure 3A). Decreases were seen from day 1 to day 3 in proportions of peripheral CD4⁺ T cells (24% to 1.8%), CD8⁺ T cells (13% to 2.7%), and B cells (6.6% to 2.9%), while the proportion of monocytes increased (24% to 61%). Flow cytometry confirmed these changes (Figure S2B). Analysis of lymphocyte subpopulations revealed low day 3/day 1 ratios of CD4⁺ T_Nv (0.024), CD4⁺ T_CM (0.074), CD4⁺ T_EM (0.14), CD8⁺ T_Nv (0.049), CD8⁺ T_CM (0.18), and B_Nv (0.27) cells in BAF312‐treated patients (Figure 3B). Some lymphocyte populations were not yet reduced at day 3: CD8⁺ T_EM (0.91), B_Int (1.15), and B_Mem (0.94), but by day 7 these had decreased (day 7/day 1 ratios: CD8⁺ T_EM (0.45), B_Int (0.43), and B_Mem (0.18)). Regulatory T cells were halved (day 3/day 1 ratio: 0.47, day 7/day 1 ratio: 0.54).

FIGURE 3.

Peripheral inflammatory response decreases by day 3 of ICH in patients treated with BAF312. (A) Percentages of adaptive (left) and innate (right) immune populations over time after ICH in BAF312‐treated patients. (B) Ratios of lymphocyte subpopulations in BAF312‐treated patients at Day 3 or Day 7 versus Day 1. (C) Time course of select inflammatory gene expression in monocytes from BAF312‐treated patients after ICH. (D) GSEA with MSigDB Hallmark Pathway genesets of monocytes of BAF312‐treated patients, Day 3 or Day 7 versus Day 1. (E) GSEA of different cell lineages from BAF312‐treated patients, comparing Day 3 versus Day 1 and Day 7 versus Day 1, showing only results for MSigDB Hallmark Pathways TNFA_SIGNALING_VIA_NFKB pathway. (F) Module scoring of the TNFA‐SIGNALING_VIA_NFKB geneset at days 1, 3, and 7 post ICH.

In contrast to placebo‐treated patients, inflammatory gene expression in monocytes from BAF312‐treated patients was downregulated from day 1 to day 3: FOSB (LFC −0.70), JUN (LFC −0.74), NFKBIA (LFC −1.01), IRF1 (LFC −0.32), and CXCL8 (LFC −1.91) (Figure 3C, Table S5). Expression of these genes stayed relatively constant from day 3 to day 7: FOSB (LFC 0.087), JUN (LFC not signif.), NFKBIA (LFC not signif.), and CXCL8 (LFC not signif.). IRF1 was an exception, increasing from day 3 to day 7 (LFC 0.40). GSEA showed decreases in TNF Signaling via NFκB (NES −3.1) from day 1 to day 3 and in IL‐2 STAT5 Signaling (NES −1.6) and Inflammatory Response (NES −2.2) from day 1 to day 7 post ICH (Figure 3D and Table S6). The TNF Signaling via NFκB geneset also decreased from day 1 to day 3 in CD16⁺ monocytes (NES −2.9), CD8⁺ T cells (NES −2.1), and NK cells (NES −2.7) in BAF312‐treated patients (Figure 3E, Table S6). Module scoring indicated that TNF signaling decreased uniformly within each lineage, suggesting that BAF312 did not target a specific subset (Figure 3F).

3.5. BAF312 Decreased Inflammatory Response Compared to Placebo

Compared with placebo, monocytes from patients treated with BAF312 had lower expression of NKFBIA (LFC −1.8), FOSB (LFC −0.86), JUN (LFC −1.4), and CXCL8 (LFC −4.2) at day 3 post ICH (Figure 4A, Table S7). Effects were smaller at day 7: NKFBIA (LFC −0.37), FOSB (LFC not signif.), JUN (LFC −0.26), and CXCL8 (LFC −1.4). GSEA showed an association of BAF312 treatment with decreased monocyte TNF (NES −3.2), IFNα (NES −1.8), IFNγ (NES −2.1), hypoxia (NES −2.1), and general inflammatory (NES −2.3) response pathways at day 3 (Figure 4B, Table S8). GSEA found no impact of BAF312 on these pathways at day 7. The suppressive effect on TNF Signaling via NFκB at day 3 was also seen in CD16⁺ monocytes (NES −2.8), CD4⁺ T cells (NES −2.3), CD8⁺ T cells (NES −1.9), and B cells (NES −2.0) (Figure 4C, Table S8). BAF312 did not impact TNF signaling in NK cells at day 3. At day 7, BAF312 decreased TNF Signaling in NK cells (NES −1.9), CD4⁺ T cells (NES −2.2), and B cells (NES −1.7).

FIGURE 4.

BAF312 decreased inflammatory response compared to placebo. (A) Differential expression of select inflammatory genes in BAF312‐treated versus placebo‐treated patients. (B) GSEA of monocytes from BAF312‐treated versus placebo‐treated patients. (C) GSEA of different cell lineages in BAF312‐treated versus placebo‐treated patients, showing only results for MSigDB Hallmark Pathways TNFA_SIGNALING_VIA_NFKB pathway. (D) Enrichment of cytokine signaling pathways in BAF312‐treated versus placebo‐treated patients, measured by Ingenuity Pathways Analysis Upstream Regulators Analysis. Numbers on bars indicate the number of genes overlapping with the pathway. (E) Concentration of cytokines in patient plasma, measured by Cytometric Bead Array. Dashed lines indicate lower limit of detection.

ICH score was similar between the placebo‐treated (mean 1.0) and BAF312‐treated (mean 0.8) patients (Table 1). To nonetheless address the potential confounding impact of differences in initial stroke severity, transcript counts were pseudobulked and ICH score was included in the differential expression model. Statistical significance of individual DEGs was lost after pseudobulking, but the trend of BAF312 reducing inflammatory gene expression was maintained (Figure S3A–C).

Ingenuity Pathway Analysis (IPA) Upstream Regulators Analysis identified cytokines that could be responsible for the decreased inflammatory response in BAF312‐treated patients. BAF312 treatment was associated with decreased expression of signaling pathways for IFNγ (z‐score −8.2), TNF (z‐score −7.0), IL‐1β (z‐score −6.7), and IL‐6 (z‐score −5.1) at day 3 post‐ICH (Figure 4D, Figure S3D). This impact persisted at a lower magnitude at day 7: IFNγ (z‐score −2.7), TNF (z‐score −2.5), IL‐1β (z‐score −2.5), and IL‐6 (z‐score −4.3). Plasma cytokine and chemokine concentrations (IFNγ, TNF, IL‐1β, CCL2, CXCL10, CD40L, IL‐1α, IL‐6, IL‐4, and IL‐10) were assessed by cytometric bead array (BD Biosciences) and revealed no significant differences between placebo‐ and BAF312‐treated patients (Figure 4E, Figure S3E).

3.6. Increased TNF Response After ICH Was Associated With Better Functional Outcome

Patients were stratified by clinical measures of ICH severity and outcome, and GSEA was performed to identify transcriptional pathways that differed between groups. Blood draws for this substudy were not taken at all timepoints for which clinical measures were examined, but these clinical data enabled analysis of associations between peripheral immune responses at days 1, 3, and 7 with multiple clinical measures taken at various timepoints (Tables S2, S9, and S10). Higher day 1 hematoma volume (> 30 mL) was associated with higher TNF signaling at day 3 (NES 2.2) (Figure 5A). Greater day 1 perihematomal edema volume (> 30 mL) was also associated with higher TNF signaling at day 3 (NES 2.3) (Figure 5B). Higher relative perihematomal edema (> 1) at day 1 was associated with increased TNF signaling for the entire first week post ICH (day 1 NES 2.3, day 3 NES 2.6, day 7 NES 2.2) (Figure 5C). Surprisingly, higher day 1 NIH Stroke Scale (NIHSS) score (> 9) correlated with less TNF signaling at day 1 (NES −1.9) and day 3 (NES −1.9) (Figure 5D).

Patients with better day 90 functional outcome (mRS < 4) had monocytes with greater TNF signaling throughout the first week (day 1 NES 2.4, day 3 NES 2.6, day 7 NES 2.2) (Figure 5E). These patients also had less oxidative phosphorylation in the first week of ICH (day 1 NES −2.2, day 3 NES −1.9, day 7 NES −1.7) (Figure 5E). At the gene level, lower day 90 mRS was associated with higher day 1 and day 3 expression of a subset of the inflammatory genes that are upregulated during ICH, notably NFKB1 (day 3 LFC 1.1), NFKBIA (day 3 LFC 1.3), CXCL8 (day 3 LFC 3.0), and VEGFA (day 3 LFC 1.2) (Figures 2B and 5F). Full results for all cell types and clinical measures are in Tables S9 and S10. After pseudobulking and including ICH score in the model, the association of lower day 90 mRS with increased TNF signaling and decreased oxidative phosphorylation was maintained (Figure 5G). To visualize this association on a per‐patient basis, Gene Set Variation Analysis (GSVA) was performed using the Hallmark TNF signaling geneset. By GSVA, there is a statistically significant association between day 90 mRS and TNF signaling at day 1 but not at days 3 or 7 post ICH (Figure S4A). This association is intriguing, but must be interpreted with caution, as it is based on analysis of only 3–4 patients per treatment condition.

To validate these results, we reanalyzed a previously published bulk RNA‐seq dataset of monocytes from the peripheral blood and hematoma of patients in the MISTIE III trial [35]. We found that in the MISTIE III trial, good outcome (mRS < 4) was associated with lower expression of oxidative phosphorylation (NES −2.2) and TNF signaling (NES −2.3) pathways in peripheral blood monocytes (Figure S4B). Good outcome in the MISTIE III trial was also associated with increased TNF signaling (NES 3.2) and decreased oxidative phosphorylation (NES −2.1) in hematoma‐derived monocytes/macrophages (Figure 5H), consistent with our findings in peripheral blood monocytes.

4. Discussion

Longitudinal sample collection in this study enabled transcriptome‐wide characterization of the peripheral immune response in the first week after ICH. Patients from both the placebo and BAF312 arm had moderate lymphopenia at day 1, which has been reported in ICH and is associated with larger hemorrhage volume and decreased survival [57, 58]. This lymphopenia was markedly increased at days 3 and 7 in BAF312‐treated patients, confirming a robust drug effect within this subacute window. A prior report described an increase in peripheral blood monocytes at days 3–5 post ICH [59]. We did not see statistically significant changes in monocyte numbers after ICH, possibly due to limited sample size. The small size of this study's cohort limits the generalizability of its findings, but we discuss here possible interpretations of its findings, contingent on validation by larger studies.

The contribution of TNF to pathological neuroinflammation during ICH is well documented [6, 35, 46, 60]. The MSigDB Hallmark TNF signaling via NFκB geneset was highly expressed post ICH in all cell lineages examined in this study and was used as our primary readout of inflammation. Monocytes expressed this geneset at high levels by day 1, consistent with TNF production in the rat brain as early as 2 h post ICH [46]. Expression of type I inflammation‐associated molecules in the hematoma and CSF are highest in the first 1–4 days after ICH and decrease by day 7 [6, 35, 60]; our findings suggest that the peripheral blood responses resemble those in the hematoma. The response of most major blood leukocyte lineages peaked at day 3, in line with the reported shift to a less inflammatory and more reparative response after day 3 [6, 35].

Module scoring revealed that genes upregulated during ICH in previous studies [37, 38, 39] were highly expressed in our cohort. Those studies used bulk‐RNA‐Seq to analyze unsorted PBMC or whole blood; we found that the upregulated signatures were predominantly expressed by monocytes, with minimal expression in other lineages. This suggests that the monocyte response dominates the acute peripheral blood transcriptional response during ICH. Genes downregulated in bulk PBMC after ICH were enriched in T cells and NK cells in our cohort, suggesting a decrease in numbers or activation levels of these populations relative to monocytes.

BAF312 was associated with profound suppression of the peripheral immune response. Naïve and central memory T and B cell populations decreased first, but by day 7 the effector subsets were reduced as well. This complements the finding that effector T cells and transitional B cells are resistant to reduction from circulation by BAF312 after 6–12 months in multiple sclerosis [12, 15].

The study focused on monocytes because of their established importance in the first week of ICH, the relative paucity of data on BAF312's effects on monocytes, and because lymphopenia limited analysis of lymphocytes. BAF312 was associated with inhibited inflammatory gene expression in monocytes at day 3 through day 7, by which point monocytes in placebo‐treated patients had also decreased their inflammatory gene expression. BAF312 therefore appears to have accelerated the resolution of inflammation. Ingenuity Pathways Analysis revealed that BAF312 also suppressed pathway genes for IFNγ, IL‐1β, CD40L, and other inflammatory mediators. Plasma cytokine concentrations did not differ between groups, however, consistent with previous findings that FTY720 does not decrease serum inflammatory cytokine concentration in humans or during murine MCAO [32, 61]. BAF312 may therefore limit the ability of monocytes to respond to cytokines.

Acute‐stage monocyte gene expression was strongly associated with long‐term functional outcome. The number of pathways identified, their normalized enrichment scores, and their statistical significance were greater for correlations of gene expression with day 90 mRS than with hematoma volume, perihematomal edema, or day 1 NIHSS. Patients with good day 90 mRS had, surprisingly, higher TNF signaling throughout the first week after ICH. In support of this finding, good outcome in the MISTIE III trial was associated with higher TNF signaling by hematoma monocytes/macrophages. This seems at odds with reports that blockade or genetic ablation of TNF improves functional outcome after ICH [46, 47, 48, 49]. It is possible that the beneficial effects of infiltrating macrophages in breaking down the hematoma and resolving disease in the subacute stage of ICH may outweigh the direct neurotoxicity of TNF [6]. Increased microglial activation has been associated with improved day 90 mRS [62], and TNF has neuroprotective effects during ischemic stroke that may extend to ICH [63, 64]. Larger studies will be needed to resolve the pro‐inflammatory versus reparative roles of TNF‐responsive macrophages and the relationship between TNF and perihematomal edema.

5. Limitations

This study has several limitations. Samples came from a clinical trial that was terminated early, which reduced the number of samples, limited statistical power, and precluded analysis of sex effects. Samples were shipped overnight before processing, and gene expression or plasma cytokine levels might have changed during transit. Confirmation of the RNA‐sequencing results was limited to flow cytometry, plasma cytokine concentrations, and comparisons with published datasets. The impact of BAF312 on TNF signaling and the role of TNF in ICH were not directly confirmed by experiment.

Author Contributions

J.H.D.: designing research studies, conducting experiments, acquiring data, analyzing data, writing the manuscript; S.D.‐P.: analyzing data, editing the manuscript; K.N.S., J.H.D., C.J.M., P.G.M.W., N.P.: acquiring data; A.H., H.E.H., K.B.W.: acquiring data, editing the manuscript; L.H.S.: designing research studies, analyzing data, writing the manuscript.

Funding

This work was supported by the National Institutes of Health, 1S10OD030363‐01A1, P30 CA016359, T32HL007950. American Heart Association, 19EIA34770133, 830877.

Conflicts of Interest

J.H.D.: Spouse employed by Alexion Pharmaceuticals; spouse owns stock in AstraZeneca.

Supporting information

Figure S1: Clustering and downsampling. (A) UMAP including doublets and dead cells that were removed before downstream analysis. (B) Expression dotplot of select genes that help identify the indicated subclusters of cells. (C) Proportions of cells found in select major cell lineages, in each patient, at each timepoint. Proportions do not add to 100% because minor lineages are not represented. (D) Cell numbers collected per lineage, per sample. Dashed lines indicate the number of cells to which each cell type was downsampled.

Figure S2: Population frequency changes after ICH by flow cytometry. (A, B) Cell proportions in placebo‐treated patients (A) and BAF312‐treated patients (B), measured by flow cytometry. Adaptive immune populations are shown on the left and innate immune populations are shown on the right.

Figure S3: BAF312 decreased inflammatory response compared to placebo. (A–D) Samples were pseudobulked by cell type and fit to a linear model in which gene expression is normalized for ICH score. (A) Differential expression of select inflammatory genes in monocytes from BAF312‐treated versus placebo‐treated patients. (B) GSEA of monocytes from BAF312‐treated versus placebo‐treated patients. (C) GSEA of different cell lineages in BAF312‐treated versus placebo‐treated patients, showing only results for MSigDB Hallmark Pathways TNFA_SIGNALING_VIA_NFKB pathway. (D) Enrichment of cytokine signaling pathways in BAF312‐treated versus placebo‐treated patients, measured by Ingenuity Pathways Analysis Upstream Regulators Analysis. (E) Concentration of cytokines in patient plasma, measured by Cytometric Bead Array. Dashed lines indicate lower limit of detection.

Figure S4: Association of outcome with immune response when measured by GSVA, as well as in MISTIE III study. (A) GSVA analysis of monocytes from placebo‐treated (black circles), BAF312‐treated (maroon squares), and BAF312‐treatment‐interrupted (yellow diamonds) patients, plotted against day 90 mRS. Patient number is indicated below symbols. Coefficient of determination (R ²) and p‐value of linear regression shown. (B) GSEA of peripheral blood monocytes isolated from patients 4–7 days post ICH, comparing mRS of 0–3 versus 4–6 at 1 year post ICH, from Askenase, et al. [35].

Table S1: Sequencing metrics. Sequencing quality metrics as reported by Cell Ranger v7.1.0.

Table S2: Patient demographics and clinical characteristics. Patient demographics, medical history, central reading data, and clinical outcome measurements.

Table S3: DEG of Placebo‐treated patients—comparisons by timepoint. Differential gene expression, performed using the FindMarkers() function as described in Methods. DEGs were calculated for each of the 6 main lineages identified. Three comparisons were made for each population in placebo‐treated patients: Day 3 versus Day 1, Day 7 versus Day 3, and Day 7 versus Day 1.

Table S4: GSEA of Placebo‐treated patients—comparisons by timepoint. GSEA was performed using the fgsea() function of the fgsea package in R. Input genes were the DEGs shown in Table S3, ranked by log₂(fold‐change).

Table S5: DEG of BAF312‐treated patients—comparisons by timepoint. Differential gene expression, performed using the FindMarkers() function as described in Methods. DEGs were calculated for each of the 6 main lineages identified. Three comparisons were made for each population in BAF312‐treated patients: Day 3 versus Day 1, Day 7 versus Day 3, and Day 7 versus Day 1.

Table S6: GSEA of BAF312‐treated patients—comparisons by timepoint. GSEA was performed using the fgsea() function of the fgsea package in R. Input genes were the DEGs shown in Table S5, ranked by log₂(fold‐change).

Table S7: DEG of placebo‐treated versus BAF312‐treated patients. Differential gene expression, performed using the FindMarkers() function as described in Methods. DEGs were calculated for each of the 6 main lineages identified. Three comparisons were made for each population: placebo‐treated versus BAF312‐treated at Day 1, Day 3, and Day 7.

Table S8: GSEA of placebo‐treated versus BAF312‐treated patients. GSEA was performed using the fgsea() function of the fgsea package in R. Input genes were the DEGs shown in Table S7, ranked by log₂(fold‐change).

Table S9: DEG of pooled patients by clinical characteristics. Differential gene expression, performed using the FindMarkers() function as described in Methods. DEGs were calculated for each of the 6 main lineages identified. Patients were bifurcated into groups based on clinical measures of ICH severity and outcome and DEGs were identified that distinguish these groups at Day 1, Day 3, and Day 7.

Table S10: GSEA of pooled patients by clinical characteristics. GSEA was performed using the fgsea() function of the fgsea package in R. Input genes were the DEGs shown in Table S9, ranked by log₂(fold‐change).

Table S11: Supporting Data Values. Raw data values are included for all plots displaying nontranscriptome data.

DeLong J. H., Díaz‐Pérez S., Sheth K. N., et al., “Intracerebral Hemorrhage Induces Monocyte TNF Signaling in Patients That Is Suppressed by BAF312,” Annals of Clinical and Translational Neurology (2026): 1–12, 10.1002/acn3.70485.

Funding Statement

This work was funded by National Institutes of Health grants 1S10OD030363‐01A1, P30 CA016359, and T32HL007950; American Heart Association grants 19EIA34770133 and 830877.

Data Availability Statement

Values for all data points in graphs are reported in Table S11—Supporting Data Values. Raw and processed RNA‐sequencing data are available in the PubMed Gene Expression Omnibus (GEO) database under the accession number GSE313043. R scripts are available at https://github.com/SansingLab/DeLong_BAF312_suppresses_monocytes_during_ICH.