Endothelial PERK restricts lymphoid regeneration by reducing DLL4-NOTCH3 signaling at the Pre-B niche

SUMMARY

Delayed immune recovery after hematopoietic stem cell (HSC) transplantation is associated with a poor clinical outcome. We study the role of unfolded protein response (ER stress) in hematopoietic regeneration within the bone marrow (BM) microenvironment. We reveal that BM endothelium PERK activation is a prominent feature of patients with leukemia and is a hallmark response in mice following ionizing irradiation. Ablating endothelial Perk boosts NOTCH ligand DLL4 expression and promotes DLL4-dependent early HSC and B progenitor regeneration. Single-cell analysis reveals that endothelial DLL4 activates NOTCH3 expressed by mesenchymal stroma cells, and that the PERK-DLL4 axis coordinates the regulation of lymphoid commitment. NOTCH3 is critical for the upregulation of IL7 following irradiation and the expansion of lymphoid progenitors. These findings not only unveil an ER stress-controlled vascular-stroma signaling mechanism in regenerative hematopoiesis but also highlight PERK blockade as a promising strategy to improve immune recovery after myeloablative transplantation.

Graphical Abstract

In brief

Zou et al. show that the adaptive ER stress response in bone marrow blood vessels restricts the post-transplant regeneration of immune progenitor cells by attenuating the expression of NOTCH ligand DLL4. At the single cell level, ablating endothelial PERK promotes immune recovery by enhancing DLL4-NOTCH3 signaling and IL7 cytokine production.

INTRODUCTION

Hematopoietic stem cells (HSCs) reside in a tightly controlled bone marrow (BM) niche that regulates and maintains hematopoietic homeostasis and regeneration.^1-3 Endothelial cells (ECs) and mesenchymal stroma cells (MSCs) (marked by LepR or Prx-1) provide critical pro-hematopoietic factors to support the HSC pool and IL7 for restricted progenitors including common lymphoid progenitors (CLPs).^4-8 Single-cell and spatial transcriptomics analyses reveal that CXCL12-abundant reticular cells (CAR cells), which overlap with the Prx-1⁺ or LepR⁺ MSCs, show peri-vascular localization.^5,7,9 In addition, NOTCH ligands expressed by endothelium are known to function as an angiocrine factor regulating NOTCH-dependent HSC function.¹⁰ Single cell transcriptome profiling revealed that Delta-like 4 (DLL4), a major NOTCH ligand expressed by BM ECs, displays dynamic regulation in response to 5-FU treatment,¹¹ suggesting that regulating NOTCH signaling may be a key step in post-myeloablative hematopoietic regeneration.

A favorable outcome after HSC transplantation is dependent on timely HSC regeneration and differentiation into multiple blood cell lineages.¹² In hematopoietic regeneration, the unfolded protein response (UPR), which is an adaptive response to endoplasmic reticulum (ER) stress, is critical for meeting the increased demand for protein synthesis required for rapid cellular proliferation. ER stress response also helps adaptation to conditions such as increased levels of reactive oxygen species (ROS) generated by ionizing radiation and chemotherapy.¹³ ROS may cause further damage to the macromolecules leading to protein misfolding and unfolding.^14-16 As part of UPR, activated PKR-Like ER Kinase (PERK; encoded by Eif2ak3) phosphorylates the alpha subunit of eukaryotic initiation factor-2 (eIF-2α), which in turn inhibits general translation initiation to restore ER homeostasis.¹⁷ Our recent work revealed that endothelial PERK plays an active role in the regulation of NOTCH ligand expression and leukemia-induced vascular niche remodeling.¹⁸ However, how the ER stress response in the BM stroma adapts to myeloablation and impacts hematopoietic stem cell and progenitor (HSPC) recovery remains unclear. To address the role of ER stress response in HSPC regeneration, we performed BM transplantation, imaging, and MSC functional analysis. We revealed that perturbation of endothelial Perk boosted DLL4 expression and promoted DLL4-dependent HSPC regeneration. Importantly, EC-expressed DLL4 and stroma-expressed NOTCH3 co-regulate lymphoid priming of regenerative HSC. Furthermore, we used single-cell analysis to define the cellular and molecular networks required for enhanced lymphoid recovery by targeting endothelial PERK-DLL4 axis, highlighting a previously unrecognized vascular-stroma signaling mechanism in regenerative hematopoiesis.

RESULTS

Ablating endothelial PERK or blocking PERK activation promotes early HSC and lymphoid regeneration

We investigated whether PERK is activated in the BM microenvironment during transplantation. Forty-eight hours after lethal irradiation, phosphor-PERK (p-PERK) was markedly increased in BM ECs but only slightly increased in MSCs (Figure 1A), consistent with reports that MSCs have a higher antioxidant capacity and are more radio-resistant than ECs.¹⁹ This selective activation prompted us to focus on endothelial PERK under irradiation-induced stress hematopoiesis. We generated mice that have PERK deleted in ECs by tamoxifen (VE-cadherin^ERT2-Cre/Perk^F/F or Perk^iΔEC mice). FACS and IHC analysis of BM CD31⁺ ECs of Perk^iΔEC mice revealed ~82% Perk deletion (Figure S1A) and marked suppression of PERK activation following irradiation (Figure 1B). Analysis of homeostatic peripheral blood cells and HSPCs revealed that Perk deletion had no effect on homeostatic hematopoiesis (Figures S1B-S1F). We then investigated stress hematopoiesis following transplantation and chemotherapy. We infused wild type (WT) BM cells (Ly5.1) into irradiated Perk^iΔEC mice (Ly5.2) or litter-mate controls (VE-cadherin^ERT2-Cre/PERK^F/+ or PERK^F/F). White blood cell (WBC) counts consistently elevated by 17%–42% over 1–4 months post-transplantation in Perk^iΔEC mice. The most pronounced effect was increase in donor-derived B lymphocytes in Perk^iΔEC mice compared to controls (Figure 1C). The absolute numbers of HSC, CLP, and B lineage progenitors (pre-proB, pro-B, and pre-B cells) all increased at 1 month in transplanted Perk^iΔEC mice (Figure 1D). By 4 months, there were no significant differences in the numbers of HSPCs in Perk^iΔEC mice compared to controls (Figure 1E). We then assessed hematopoietic recovery after 5-FU treatment. Perk^iΔEC mice initially displayed mitigation of WBC decline. WBC continued to recover faster on day ten and day twenty, displaying elevated B cell numbers and decreased myeloid output (Figure 1F). To explore the regenerative potential of targeting PERK, we treated mice transplanted with WT BM cells with a PERK inhibitor (PERKi) and performed hematopoietic analysis on day 21, a clinically relevant time point for assessing early lymphocyte recovery.²⁰ PERKi treatment increased circulating B cells (Figure 1G) and HSC/lymphoid progenitor regeneration modestly, likely due to the partial suppression of PERK activation (Figures 1H, 1I, andS1G). Blocking IRE1α/XBP1, the other major arm of ER stress activation^21,22 did not affect lymphoid recovery or progenitor regeneration (Figures S1H-S1J), indicating this is a PERK-dependent effect. Therefore, PERK blockade or genetically ablating endothelial PERK mitigated myelosuppression but enhanced lymphoid recovery in transplantation without impacting long term hematopoiesis.

Figure 1. Ablating endothelial PERK promotes early HSC and lymphoid progenitor regeneration.

(A) FACS analysis was caried out on MACSQuant Analyzer 16. Representative FACS profile and mean fluorescent intensity (MFI) of activated PERK (p-PERK) from three similar experiments in bone marrow endothelial cells (ECs) and mesenchymal stroma cells (MSCs). MSC is defined by Lin (CD11b, Gr1, B220, CD4, CD8, Ter119, and NK1.1)⁻ CD31⁻ while EC is defined by Lin⁻CD31⁺.

(B) Left image shows MFI of p-PERK in non-irradiated wild type (WT) (ctrl), irradiated WT (ctrl + IRR) and irradiated Perk^iΔEC mice (Perk^iΔEC + IRR) at 48 h following 850 cGy (Cesium-137) irradiation. Right image shows IHC staining of p-PERK in non-irradiated WT, irradiated WT, and irradiated Perk^iΔEC mice ECs. Dashed circles and arrows show outlined ECs stained by anti-pPERK. Scale bars, 100 μm. Control EC staining by anti-CD31 is shown in Figure S1A.

(C) Peripheral blood was analyzed for WBC counts, B cells, T cells, and granulocytes at 4, 8, 12, and 16 weeks after transplantation (n = 8–9/group; data pooled from three experiments).

(D) The frequencies of HSC (Lin⁻Sca-1⁺c-Kit⁺CD150⁺CD48⁻CD34⁻Flk2⁻), CLP (Lin⁻IL7R⁺Flt3⁺Sca-1^lowc-Kit⁺low) and B progenitors (pre-proB: Lin [CD11b, Gr1, CD3, NK1.1, Ter119)⁻B220⁺CD43⁺Flt3⁺CD19⁻; pro-B: Lin⁻B220⁺CD43⁺Flt3⁻CD19⁺; pre-B: Lin⁻B220⁺CD43⁻) were determined in control and Perk^iΔEC mice BM at 1 month after transplantation (n = 6–7/group; data pooled from 2 experiments).

(E) The frequencies of HSC, CLP, and B progenitors were determined in control or Perk^iΔEC mice at 4 months after transplantation (n = 8–9/group; data pooled from three experiments).

(F) WBC counts were determined on days 5, 10, and 20 after single dose 5-FU treatment. Frequencies and absolute numbers (not shown) of B cells, T cells, and granulocytes were determined on day 20. Absolute numbers showed similar changes.

(G) WT mice transplanted with WT BM cells received PERK inhibitor (PERKi; GSK2656157) (50 mg/kg) or control treatment twice daily for 2 weeks starting on day 3 after transplantation. Peripheral blood B, T and granulocytes were enumerated on day 21.

(H and I) BM HSC, CLP (H) and B progenitors (I) were also determined. MFI of p-PERK in (A) and (B) is presented as averages ±SEM (n = 4–5/each condition). Results (C–I) were pooled from three experiments and presented as averages ±SEM (n = 9–12/group). *p < 0.05, **p < 0.01, ***p < 0.001.

PERK-mediated HSC and lymphoid progenitor recovery is compromised by DLL4 ablation

DLL4 is the major NOTCH ligand expressed by BM ECs and is downregulated after chemotherapy.¹¹ However, the mechanism of this downregulation was unknown. We found that DLL4 level remained largely unchanged in ECs after irradiation. Intriguingly, its expression increased in PERK-deleted ECs by FACS analysis (Figures 2A and 2B). Consistently, blocking PERK suppresses activation of eIF-2α (p-EIF2α) in irradiated BM ECs in vitro and upregulates DLL4 (Figure S2A). Mechanistically, ablating PERK-mediated DLL4 upregulation after irradiation was not caused by the increased transcription of DLL4. Instead, we observed increased DLL4 protein stability by downregulating PERK (Figures S2B and S2C). We found that DLL4 in ECs is degraded mainly through the proteasomal pathway. We noted that while MG132 treatment and PERK downregulation had similar effects on DLL4 stability, combining PERK ablation and MG132 failed to further increase DLL4 protein levels, suggesting that ablating PERK likely increases DLL4 stability by perturbating its proteasomal degradation (Figures S2D and S2E). To assess whether DLL4 and PERK expression show similar changes in response to myeloablation in human disease, we evaluated BM specimens from patients with acute myeloid leukemia (AML) who underwent myeloablative chemotherapy. We observed disorganized vessels and decreased cytoplasmic/membrane staining of DLL4 in AML BM when compared to tissues from individuals without hematologic malignancies. In contrast, expression of p-PERK was enhanced in AML BM vessels and displayed a negative correlation with DLL4 (Figures S2F and S2G).

Figure 2. Endothelial DLL4 is regulated by PERK and is indispensable for post-irradiation HSC and B lymphoid progenitor recovery.

(A) FACS analysis was carried out on CytoFLEX. Representative FACS profile from 3 similar experiments showing DLL4-positive BM endothelial cells (gated on Lin⁻CD31/CD144⁺) at 72 h in non-irradiated WT mice (ctrl, −IRR), WT irradiated mice (ctrl, +IRR), Perk^iΔEC irradiated mice (Perk^iΔEC, +IRR), and DKO irradiated mice (DKO +IRR) following 850 cGy irradiation.

(B) MFI of DLL4 expression from (A) is presented as averages ±SEM (n = 4–6/group).

(C and D) WT mice (ctrl) and Perk^iΔEC mice transplanted with WT BM cells received isotype control or anti-DLL4 (15 mg/kg) twice per week for 2 weeks starting on day 8 after transplantation. Peripheral blood B, T and granulocytes (C) and BM B progenitors (D) were determined on day 21 (n = 5/group).

(E) IHC analysis of endothelial DLL4 expression in bone marrow aspirates of non-irradiated WT, irradiated WT control, Perk^iΔEC, and DKO mice. Dashed circles and arrows show outlined endothelial cells stained by anti-DLL4. Scale bars, 100 μm.

(F) Enumeration of WBC and FACS analysis of B cells, T cells, and granulocytes at 4, 8, 12, and 16 weeks after transplantation (n = 9–10/group; pooled from three experiments).

(G and H) Representative FACS profile and frequencies of HSC/CLP (G) and B progenitors (H) in control and DKO mice at 1 month after transplantation. Results in all images except (E) are presented as averages ±SEM. *p < 0.05 **p < 0.01, ***p < 0.001, ****p < 0.0001.

To distinguish whether enhanced HSC and lymphoid recovery in Perk^iΔEC mice is regulated by DLL4, we briefly blocked DLL4 by using anti-DLL4 antibody.^23,24 Because extensive DLL4 blockade is associated with vascular toxicity,²⁵ we treated Perk^iΔEC mice with 4 doses of anti-DLL4, starting one week after mice receiving BM transplantation, and analyzed HSPC regeneration on day 21. Blocking DLL4 reversed increased circulating B cells and B lymphoid progenitors observed in Perk^iΔEC mice (Figures 2C and 2D). To determine whether these effects are mediated by blocking endothelial-expressing DLL4 in the setting that permits long-term assessment after transplantation, we generated endothelial Dll4 knockout (VE-cadherin^ERT2-Cre/DLL4^F/F or Dll4^iΔEC) and Dll4/Perk double knockout (DKO) mice (VE-cadherin^ERT2-Cre/Perk^F/F/Dll4^F/F) and confirmed near-complete extinction of DLL4 in Dll4^iΔEC mice ECs (Figures 2E and S3A). We performed BM transplantation in DKO mice. B cells decreased while granulocytes increased in DKO recipients 1 month after receiving WT cells (Figure 2F). Unlike Perk^iΔEC recipients, which had increased HSCs and CLPs, DKO mice displayed slightly decreased HSC and CLP recovery (Figure 2G). B progenitors reduced markedly (Figure 2H) while myeloid progenitors showed no significant changes (Figure S3B). This observation suggests that DLL4 plays a role in post-irradiation HSPC regeneration.

Accordingly, we found that ablating EC Dll4 alone in Dll4^iΔEC mice is sufficient to impair HSPC homeostasis and regeneration. Consistent with other reports, we found that Dll4^iΔEC mice had increased homeostatic granulopoiesis but decreased B cells with decreased B progenitors, HSCs and CLPs in the BM (Figures S3C-S3F).¹¹ Niche factor analysis revealed that Il7 decreased modestly and Cxcl12 and Kitl remained unchanged (Figure S3G). After transplantation, Dll4^iΔEC recipient mice displayed similarly enhanced granulopoiesis but decreased B lymphopoiesis with marked suppression of CLPs and HSCs. B progenitors were variably depleted while GMPs expanded (Figures S3H-S3K). B progenitor reduction was accompanied with marked reduction of Il7 (Figure S3L). Notably, suppression of HSCs and B progenitors persisted 4 months after transplantation in Dll4^iΔEC mice (Figures S3M-S3O). Considered together, these findings suggest that endothelial DLL4 in post-irradiation BM is required for HSPC and B lymphoid regeneration. The data suggest that in this context Dll4 is epistatic to Perk, and its upregulation contributes to the enhanced regeneration of HSCs and B progenitors observed upon Perk ablation.

PERK-DLL4 regulates regenerative vascular homeostasis and integrity

DLL4 has an important role in angiogenesis by regulating endothelial cell growth and proliferation and by inhibiting angiogenic sprouting via activating NOTCH receptors on adjacent stalk cells, restricting them from becoming tip cells.^26,27 We investigated the impact of endothelial PERK-DLL4 axis on post-irradiation vascular regeneration. To mitigate the interference from transplanted BM cells, we performed non-lethal irradiation (550 cGy) to induce vascular damage and regeneration. It has been shown that the collapsed vascular network rejuvenates by 2 weeks after whole body irradiation.²⁸ Confocal analysis showed expansion of the vascular volume and increased Sca-1⁺ arterioles in the BM of Dll4^iΔEC mice, consistent with the role of DLL4 as a negative regulator of angiogenesis (Figures 3A and 3B). In addition, the BM vessel bed in Dll4^iΔEC mice displayed a distorted pattern characterized by reduced vessel lengths and disrupted structures, suggesting that DLL4 may restrict post-irradiation disorganized angiogenesis. In contrast, irradiated Perk^iΔEC mice, which expressed increased DLL4, showed the opposite phenotype, with a more organized BM vasculature (Figure 3A). While visual inspection of DKO marrow vasculature might exhibit less disruption compared with the Dll4^iΔEC mice, EC density quantification demonstrated an increase in DKO BM compared with control BM (Figures 3A and 3B), suggesting again that Dll4 functions downstream of Perk in this context. To investigate whether the PERK-DLL4 axis regulates vascular integrity, we monitored BM vessel leakage at 2 weeks after 550 cGy irradiation by intravital microscopy using both low- (LMW; 70 KD) and high molecular weight Dextran (HMW; 2,000 KD). DLL4-deficient vessels were more permeable to both LMW and HMW Dextran tracers, indicating compromised barrier function. In comparison, Perk ablation alone had no significant effect on vascular permeability while DKO mice showed increased leakiness compared to Perk^iΔEC mice, but decreased leakiness compared to Dll4^iΔEC (Figures 3C and 3D). Considering that PERK regulates protein translation and that PERK perturbation affects DLL4 stability, we speculate that PERK deficiency may enhance translation and stability of proteins that control intercellular junctions or actin cytoskeleton contraction, which can antagonize the deleterious effects caused by DLL4 deficiency. Nevertheless, these results establish DLL4 as a critical determinant of vascular integrity following irradiation. Further, elevated DLL4 expression upon PERK ablation restricts abnormal angiogenesis and stabilizes the BM vasculature, thereby mitigating irradiation-induced vascular niche disruption.

Figure 3. PERK-DLL4 regulates regenerative vascular homeostasis and integrity.

(A and B) Whole-mount immunostaining of BM vascular bed with Alexa Fluor 647-anti-CD31 (red), Alexa Fluor 647-anti-CD144 (red) and FITC-anti-Sca-1 (green) in non-irradiated WT (ctrl) mice, and in irradiated WT, Dll4^iΔEC, Perk^iΔEC, and DKO mice (A). Representative images were generated in Imaris based on the corresponding fluorescent signals. Scale bars, 200 μM.

(B) The percentage of total CD31⁺/CD144⁺ volume (EC total) and CD31⁺Sca-1⁺ arteriole volume (Sca-1⁺ EC) were determined in WT, Dll4^iΔEC, Perk^iΔEC, and DKO mice.

(C) Intravital 2-photon imaging of calvarium BM vessels was performed in three independent experiments to determine vessel leakiness after retro-orbital injection of 2000 kD FITC-Dextran (green) and 70 kD TRITC-Dextran (red) immediately before imaging. Three-dimensional images consisting of XY: 775 mm × 775 mm and Z: 140 mm (5 mm/step), were obtained every 30 s for 30 min. Images were drift-corrected, smoothed, and surfaces were generated on both FITC and TRITC signals. Representative images taken at 1 min and 30 min were shown. Images in yellow are combined images of FITC-Dextran and 70 kD TRITC-Dextran. Scale bars, 50 μM.

(D) The corresponding total volumes for each signal were calculated at each timepoint and normalized to the first timepoint. Results in bar graphs are presented as mean ± S.D (n = 5–6/group). **p < 0.01, ***p < 0.001.

PERK and DLL4 differentially regulate pathways supporting B progenitor and HSC regeneration

To elucidate how endothelial PERK and DLL4 differentially regulate HSC and B progenitor regeneration, we performed scRNA-seq of regenerating BM cells on day 11 after sublethal irradiation. Under this condition, we were able to study regenerative BM without interference from infused donor cells. We identified fourteen cell populations based on canonical markers (Figures 4A, S4A, and S4B). Consistent with the transplant model, Dll4^iΔEC mice displayed an expansion of myeloid progenitors (GMPs) and a reduction of B progenitors (preB/B), while DKO mice exhibited a profound loss of preB/B cells. In contrast, Perk^iΔEC mice showed a decrease in GMPs accompanied by an increase in B progenitors (Figure 4B). Pathway analysis in B progenitors revealed that defense response and humoral immune response were upregulated upon Perk extinction but downregulated with Dll4 ablation (Figure 4C). This was supported by the upregulation of genes implicated in lymphoid development and function in B progenitors of Perk^iΔEC mice (Figure 4D). In line with increased B progenitor production, Il7 and Il7r expression increased in Perk^iΔEC stroma cells and HSPC cells, respectively, but decreased in Dll4^iΔEC and DKO mice (Figure 4E).

Figure 4. ScRNA-seq reveals BM cellular and molecular network supporting B progenitor development.

(A) UMAP visualization and proportion of identified cell clusters from scRNA-seq analysis in 4 groups of mice: control (ctrl) (n = 4), Dll4^iΔEC (D-EC) (n = 4), Perk^iΔEC (P-EC) (n = 4), and DKO (n = 3).

(B) Proportions of GMP and B/pre-B cells in four groups of mice.

(C) Integrated dot plots for GSEA results in Gene Ontology-Biological Processes (GO-BP) pathway, illustrating significant changes identified when comparing Dll4^iΔEC mice vs. control (D-EC vs. Ctrl), Perk^iΔEC mice vs. control (P-EC vs. Ctrl), and DKO vs. Perk^iΔEC (DKO vs. PP-EC) for B/pre-B cells. Dot sizes demonstrated the normalized enrichment scores (NES) and color codes visualized significance where red indicates upregulated pathways and blue indicates down-regulated pathways.

(D) Differentially expressed genes for B/pre-B cells.

(E) Il7 gene expression in MSC cells and Il7r gene expression in HSPC cells. Results (D and E) were presented as averages ±SEM. Statistical significances were first determined by Kruskal-Wallis test across four conditions (adjusted p < 0.05), followed by post-hoc pairwise tests using MAST algorithm. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.

Similarly, in HSPCs, immune response and defensive response were upregulated by Perk ablation and downregulated with Dll4 ablation (Figure 5A). Analysis of the top differentially regulated genes in emerging HSPCs (Figure 5B) showed upregulation of the B-cell specific activating transcription factor (TF) Pax5 and HSC marker genes (Ly6a and Adgrl4)^29,30 in Perk^iΔEC mice, but downregulation of these genes by Dll4 ablation in DKO mice (Figure 5C). Genes regulating HSC quiescence (Cd53 and Emb)^31,32 and self-renewal or regeneration (Mecom and Stat3)^33-35 were also upregulated in Perk^iΔEC mice but downregulated or unchanged in Dll4^iΔEC and DKO mice. The downregulation of genes supporting HSC quiescence in Dll4^iΔEC mice prompted us to assess the role of Dll4 in HSC regeneration. We transplanted WT BM cells (Ly5.1) into lethally irradiated Dll4^iΔEC and control mice (Ly5.2) to reconstitute HSC pools (Figure 5D). Cell cycle analysis 1 month after transplantation showed that donor-derived HSCs in Dll4^iΔEC mice were cycling faster than those in control mice leading to a reduction of quiescent HSCs in the G₀ phase (Figure 5E). We isolated BM cells from the primary WT or Dll4^iΔEC recipients 4 months after transplant, normalized for HSC frequency with competitor BM cells (Ly5.2), and performed competitive secondary and tertiary transplantations (Figure 5D). The engraftment of donors pre-conditioned in Dll4^iΔEC mice significantly reduced in both secondary and tertiary Dll4^iΔEC recipients when compared with cells pre-conditioned in WT mice and engrafted in WT recipients (Figure 5F). Consistently, fewer HSPCs isolated from secondary competitive transplant Dll4^iΔEC recipient mice were in G₀ phase (Figure 5G). Collectively, these findings demonstrate that endothelial DLL4 is not only essential for HSC and B progenitor regeneration but also plays a key role in HSC quiescence maintenance which is essential for long-term hematopoietic regeneration.

Figure 5. PERK and DLL4 differentially regulate molecular pathways supporting HSC regeneration.

(A) Integrated dot plots for GSEA results in Gene Ontology-Biological Processes (GO-BP) pathway, illustrating significant changes identified when comparing Dll4^iΔEC mice vs. control (D-EC vs. Ctrl), Perk^iΔEC mice vs. control (P-EC vs. Ctrl), and DKO vs. Perk^iΔEC (DKO vs. PP-EC) for HSPC cells. Dot sizes demonstrated the NES and color codes visualized significance where red indicates upregulated pathways and blue indicates down-regulated pathways.

(B) Volcano plot shows DEGs of Perk^iΔEC (P-EC) vs. CTRL derived from MAST algorithm.

(C) Differentially expressed genes for HSPC cells. Bar plots visualized normalized cell-type specific gene expression as mean ± SEM. Statistical significances were first determined by Kruskal-Wallis test across four conditions, followed by post-hoc pairwise tests using MAST algorithm. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.

(D) Scheme of primary BM transplantation and competitive 2^nd and 3^rd transplantation.

(E) Marrow cells were stained with lineage antibodies (CD4, CD8, B220, Gr-1, CD11b, TER119, and NK1.1), c-kit, Sca-1, CD34, Flt3, CD150, pyronin Y (RNA dye), and Hoechst 33342 (DNA dye). The relative proportion of cells in G₀, G₁, and S/G₂/M phase of the cell cycle was analyzed on gated HSCs (Ly5.1) (n = 4–6/group).

(F) Proportion of control and Dll4^iΔEC primed donor-derived BM cells (Ly5.1) in 2^nd or 3^rd transplant recipients.

(G) Proportion of Lin− and LSK cells (Ly5.1) in G₀ phase in 2^nd transplant recipients. Results in bar graphs (E–G) are presented as mean ± SEM. **p < 0.01, ***p < 0.001.

PERK-DLL4 axis orchestrates the stroma network for HSC lymphoid priming

We further examined signaling pathways and HSC-promoting cytokines in the regenerative stroma. We distinguished four major cell clusters, including CARs, osteoblasts, sinusoidal ECs, and arterioles (Figure 6A). CARs displayed the highest expression levels of CXCL12 and SCF (KITL) and were separated into two subsets: the adipogenic lineage precursors (MALP)³⁶ or adipo-CAR cells and the osteogenic CAR or osteo-CAR (Figure S4B).⁹ Pathway analysis revealed an enhanced response to IFNβ and pyroptosis in Perk^iΔEC ECs (Figure 6B). We then examined HSC cytokine gene expression. Il7 expression increased 2-fold in adipo-CARs of Perk^iΔEC mice, its expression markedly decreased in Dll4^iΔEC and DKO mice (Figure 6C). Cxcl12 and Kitl displayed a milder but significant increase and decrease in Perk^iΔEC and DKO mice, respectively (Figure S5A). We queried the influences of these HSC supporting cytokines on various cellular compartments by CellChat analysis. Il7 from adipo- and osteo-CARs displayed strong communication with Il7r present on preB/B, T/NK, and MEP cells in Perk^iΔEC mice, but these interactions were much weaker in Dll4^iΔEC or DKO mice (Figure S5B). Cxcl12 from CARs and arterioles strongly interacted with Cxcr4 in preB/B cells in Perk^iΔEC mice while it showed the strongest communication with MEPs in DLL4-deficient mice. There were substantial interactions between Kitl (SCF) in arteriolar ECs and CARs with HSPCs and GMPs in control BM. These interactions were enhanced in Perk^iΔEC mice but repressed in Dll4^iΔEC and DKO mice (Figures S5C and S5D).

Figure 6. Signaling pathways and HSPC-promoting cytokines in the regenerative stroma network regulated by PERK-DLL4 axis.

(A) UMAP visualization and proportion of five stromal cell subsets from scRNA-seq analysis.

(B) Integrated dot plots for significant GO-BP pathway, illustrating significant changes identified when comparing Dll4^iΔEC mice vs. control (D-EC vs. Ctrl), Perk^iΔEC mice vs. control (P-EC vs. Ctrl), and DKO vs. Perk^iΔEC (DKO vs. PP-EC) for sinusoidal ECs. Dot sizes demonstrated the NES and color codes visualized significance where red indicates upregulated pathways and blue indicates down-regulated pathways.

(C) UMAP and bar plots of Il7 expression across four groups of mice. Bar plots visualized normalized cell-type specific gene expression as mean ± SEM.

(D) Integrated dot plots for significant GO-BP pathway, illustrating significant changes identified when comparing Dll4^iΔEC mice vs. control (D-EC vs. Ctrl), Perk^iΔEC mice vs. control (P-EC vs. Ctrl), and DKO vs. Perk^iΔEC (DKO vs. PP-EC) for adipo-CAR cells. Dot sizes demonstrate NES and color codes visualized significance where red indicates upregulated and blue down-regulated pathways.

(E) Interferon-Stimulated Gene (ISG) score defined by BP pathway (Ifi35, Ifi44, Ifit1, Ifi73, Il7, Irf7, Oas3, Bst2, Cxcl10, Cxcl12, Ddx60, Hsd17b1, and Stat1) in stroma ECs, CARs and osteoblasts across four groups of mice. Bar plots visualized quantified ISG scores as mean ± SEM.

(F) ISG signature scores in Adipo/Osteo-CARs and ECs across four groups of mice.

(G and H) Inferred transcription factor (TF) activities across four groups of mice for adipo-CARs (G) and sinusoidal ECs (H).

Statistical significances (E–H) were first determined by Kruskal-Wallis test across four conditions (adjusted p < 0.05), followed by post-hoc pairwise tests using Wilcoxon rank-sum test. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.

Like ECs, CARs of Perk^iΔEC mice exhibited an enhanced response to IFN (Figure 6D). NK/T cells expressed the highest levels of IFNγ and demonstrated strong communication with adipo-CARs in Perk^iΔEC mice, but not in DKO mice (Figures S5E and S5F). Among stroma subsets, CAR cells had the highest interferon-stimulated gene (ISG) score, calculated from the core enrichment genes of the Hallmark interferon pathway (Figure 6E). The ISG score increased across stroma compartments including adipo-CARs, osteo-CARs, and ECs in Perk^iΔEC mice but was suppressed in these subsets in DKO mice (Figure 6F). We applied SCENIC gene regulatory network analysis to investigate TF activities from the scRNA-seq data. IFN response TF activity, including Irf7, Irf1, and Stat1, was all upregulated in Perk^iΔEC adipo-CARs (Figure 6G) and sinusoidal ECs (Figure 6H). This upregulation was abolished in DKO mice, indicating that these are DLL4-dependent IFN responses promoted by PERK perturbation. In summary, scRNA-seq analysis revealed that the enhanced B lymphocyte recovery in Perk^iΔEC mice was associated with a strong Il7-Il7r signaling axis and heightened IFN response, both of which were markedly attenuated in the absence of DLL4.

NOTCH3 is the major NOTCH homolog in CARs, and Notch3^−/− mice exhibit impaired post-irradiation lymphoid priming and B progenitor regeneration

CellChat analysis of scRNA-seq data identified Notch3 as the prime NOTCH homolog expressed by CARs interacting with DLL4 and Jagged1 (JAG1) (Figures 7A and 7B). Importantly, osteo- and adipo-CARs display higher NOTCH pathway activity than other stroma subsets. NOTCH activity was enhanced in Perk-ablated stroma which had increased DLL4 but was suppressed in DKO stroma (Figure 7C). To determine whether NOTCH3 regulates hematopoiesis, we first examined Notch3^−/− mice under homeostasis and found that steady state hematopoiesis was essentially normal (Figures 7D-7F). We then performed reciprocal BM transplantation to determine if NOTCH3 regulates hematopoietic regeneration. While BM cells from Notch3^−/− mice transplanted into WT recipients had normal myeloid and lymphoid lineage development (Figures S6A-S6C), Notch3^−/− recipient mice receiving WT BM displayed reduced early B lymphoid recovery and reduced MPP4, CLP, and B progenitor regeneration (Figures 7G-7I), suggesting that NOTCH3 in BM stroma may regulate myeloablation-induced B progenitor recovery.

Figure 7. Notch3 deficiency in BM stroma impairs B progenitor regeneration.

(A) EC Notch ligand expression and CAR cell Notch receptor expression in the BM.

(B) Dot plot of inferred Notch ligand (sender: EC) – Notch receptor (receiver: EC, CAR-adipo, CAR-osteo, and fibroblast/osteoblast) interaction by CellChat analysis.

(C) Notch signaling activity score (Notch3, Notch2, Hes6, Hey1, Hey2, Lfng, Mfng, Maml1, and Tnc) in stroma subsets in control mice (top panel) and in stroma subsets of PERK^iΔEC mice and DKO mice (bottom panel). Bar plots (A and C) visualized quantified Notch activity scores as mean ± SEM. Statistical significances were first determined by Kruskal-Wallis test across four conditions (adjusted p < 0.05), followed by post-hoc pairwise tests using Wilcoxon rank-sum test. **p < 0.01, ***p < 0.001, ****p < 0.0001.

(D) Analysis of peripheral blood B cells, T cells, and granulocytes from homeostatic WT and Notch3^−/− mice. Absolute numbers were shown (N = 6/group).

(E) Frequencies of HSC, CLP, and MPP4 (Lin⁻Sca-1⁺c-Kit⁺CD150⁻CD48+Flk2⁺) were determined from WT control and Notch3^−/− mice BM (N = 6/group).

(F) Numbers of BM B progenitors were determined from WT control and Notch3^−/− mice (n = 6/group).

(G–I) Analysis of peripheral blood B cells, T cells, and granulocytes at 1 month after lethal irradiation and transplantation of WT donor BM cells in Notch3^+/− and Notch3^−/− recipient mice (G). Frequencies of HSC/CLP/MPP4 (H) and pre-proB/preB/proB (I) were determined (n = 4/group). Results from (D)–(I) are presented as averages ±SD. *p < 0.05, **p < 0.01.

(J) FACS analysis was caried out on MACSQuant Analyzer 16. Representative FACS profile from 3 similar experiments showing PDGFRα⁺/CD51⁺ (DP) MSCs (Lin⁻Ter119⁻CD31⁻PDGFRα⁺/CD51⁺) and quantification of DP MSCs in WT or Notch3^−/− mice BM (n = 3/group).

(K) FACS analysis was caried out on MACSQuant Analyzer 16. Representative FACS profile from 3 similar experiments showing intracellular IL7 expression and quantification of IL7 MFI in DP MSCs isolated from mouse BM eight days after 550 cGy irradiation (n = 3/group).

(L) A representative photograph showing LK (500/well) cells cultured with SCF and FLT3L (SF) alone or in SF + mesenspheres on day 4 (left) and quantification of expanded total CD45⁺ cells after 5 days of coculture by FACS. Scale bars, 50 μM.

(M) Quantification of expanded lymphoid-primed HSPCs (CLP+MPP4) after 5 days of coculture by FACS. Results from (J)–(M) are presented as averages ±SD (n = 3/group). **p < 0.01, ***p < 0.001.

To investigate the role of NOTCH3 in the HSC niche, we examined CXCL12 and IL7 production in HSC niche-enriched PDGFRα⁺ and CD51⁺ double-positive (DP) MSCs, which expressed the highest level CXCL12 among all stroma cells (Figure S6D).³⁷ There is a slight decrease of DP MSCs in Notch3^−/− mice (Figure 7J), but Notch3-deficiency has no effect on CXCL12 expression (Figure S6E). While WT DP MSCs showed upregulation of IL7 in response to irradiation or IFNβ stimulation (Figure S6F), Notch3-deficient DP MSCs displayed attenuated post-irradiation IL7 surge but unaltered response to IFNβ (Figures 7K and S6G). Finally, we assessed MSC niche function using a modified mesensphere coculture assay developed by Frenette’s group.³⁷ Coculture of WT mesenspheres with BM progenitors (Lin⁻c-kit⁺) in the presence of SCF and FLT3L (SF) led to 6.1-fold and 2.8-fold higher expansion of CD45⁺ cells and lymphoid HSPCs than LK cultured with SF alone. Expansion of lymphoid HSPCs by Notch3-deficient mesenspheres, however, reduced by 25% (Figures 7L and 7M). Together, these data indicate that NOTCH3 promotes post-irradiation IL7 upregulation and lymphoid priming.

We also investigated the functional significance of JAG1, another major NOTCH ligand, in post-irradiation hematopoietic regeneration by performing BM transplantation in endothelial Jag1 deficient mice (VE-cadherin^ERT2-Cre/Jag1^F/F or Jag1^iΔEC). Unlike Dll4^iΔEC mice, Jag1^iΔEC mice displayed no differences in early HSPC recovery (Figures S7A and S7B) when compared with control mice. Further, there is no significant defect in vascular regeneration in Jag1^iΔEC mice after irradiation (Figure S7C). Considered together, we concluded that, in the post-irradiated HSC niche, MSC-expressed NOTCH3 and EC-expressed DLL4 regulate lymphoid priming and B progenitor regeneration.

DISCUSSION

Enhancing lymphoid recovery is a critical unmet need in HSC transplantation.^38-42 A low absolute lymphocyte count within 3 months of allogeneic HSC transplantation is a major risk factor for worse outcome.^41,43 However, strategies to enhance lymphocyte regeneration are limited, largely due to gaps in our understanding of regenerative mechanisms under myeloablative stress. In this study, we revealed a previously underappreciated role of endothelial ER stress signaling in shaping hematopoietic regeneration. Specifically, activation of endothelial PERK restricts early HSCP regeneration in post-myeloablation transplantation. Our work demonstrated that ablating endothelial Perk in irradiated mice increases BM endothelial DLL4 and NOTCH signaling in the HSC niche. We revealed that BM stroma NOTCH3, the prime NOTCH receptor that pairs with EC DLL4, regulates irradiation-induced IL7 upregulation and promotes MSC-supported lymphoid progenitor expansion. Further, single cell analysis reveals that the PERK-DLL4 axis orchestrates the regenerating network by regulating HSC lymphoid commitment, CAR cell NOTCH activation, and HSC niche factor expression.

MSCs are essential niche cells for supporting HSC homeostasis.^44,45 Single cell and spatial analysis support a pivotal role for the perivascular niche to instruct lymphoid progenitor development. However, the mechanism that regulates the lymphoid program in the perivascular region in response to stress conditions such as myeloablation has not been elucidated. The close link between the vessels and the perivascular MSC suggests that direct cell-cell interactions are important for HSC niche function. NOTCH pathway is a fundamental cellular interaction mechanism that controls cell fate decisions via ligand-mediated nuclear translocation of NOTCH receptor. Several studies including ours have revealed a NOTCH-regulated mechanism for supporting lymphoid priming in the BM.^46-48 Here, we provided compelling evidence that endothelial DLL4 and CAR cell-expressed NOTCH3 are important regulators of lymphoid progenitor regeneration. NOTCH3 is essential for the development of perivascular MSCs in dental mesenchyme.^49,50 NOTCH3 also plays a critical role in pathological tissue remodeling.^51-53 In the synovium of rheumatoid arthritis, for example, NOTCH3 drives the perivascular and sublining fibroblast transcriptional gradients initiated from ligand-expressing ECs. Consistently, our work demonstrates that CAR/MSC-expressed NOTCH3 is the prime NOTCH receptor which engages with endothelial DLL4 to regulate post-transplant lymphoid priming and B progenitor regeneration.

In addition to its NOTCH-promoting function in CAR cells which is critical for B progenitor regeneration, we revealed that endothelial DLL4 also maintains HSC quiescence, corroborating our previous work showing increased HSC cycling by blocking DLL4.²⁴ These findings extended beyond previously reported homeostatic B cell reduction in mice with endothelial DLL4 deficiency.¹¹ Further, consistent with other reports that DLL4 deficiency induces hyperpermeability by increasing transcytosis, causing non-size-selective leakage, and/or dysregulating adherens junction assembly,^54,55 our findings suggest that DLL4 functions as the guardian of vessel integrity. These roles of DLL4 may explain why HSC and B progenitor recovery were more severely impaired in Dll4^iΔEC than in Notch3-deficient mice. ER stress response is critical for restoration of angiogenic homeostasis after irradiation.⁵⁶ Post-irradiation vasculature in PERK-deficient mice displayed a DLL4-dependent maintenance of organized architecture and vessel integrity, suggesting the interplay of PERK and DLL4 plays a key role in post-irradiation vessel regeneration. Upon PERK activation, phosphorylation of eIF-2α inhibits general translation initiation. However, others reported that DLL4 translation under hypoxia-induced stress response can be maintained through PERK-regulated activation of the internal ribosome entry site (IRES) in the 5′-UTR of DLL4 mRNA.⁵⁷ While DLL4 level was not significantly increased under the stress condition induced by myeloablation, we observed increased cell surface DLL4 level by PERK ablation. Further, in vitro studies suggest that PERK ablation increases DLL4 stability by interfering with DLL4 proteasomal degradation. Considering together, these findings suggest that PERK can regulate DLL4 expression through translational or post-translational modification depending on the type of ER stress insult. Further studies are needed to illuminate exactly how Perk perturbation alters the proteosome degradation machinery or controls DLL4 degradation through additional mechanism. Importantly, PERK activation and DLL4 suppression are prominent features of BM endothelium of patients with leukemia undergoing myeloablative treatment. Given the absence of an FDA-approved direct DLL4/NOTCH3-promoting therapeutic strategy, results from our animal studies suggest that pharmacological PERK blockade could potentially improve early lymphoid recovery in post-myeloablation transplantation.

IL7 is essential for B lymphoid lineage commitment and differentiation.⁵⁸ Increased expression of IL7 in Perk^iΔEC CAR cells can drive the accelerated B progenitor recovery. Like the upregulation of IL7 in irradiated thymic stroma and the induction of IL7 by IFN in keratinocytes,^59,60 irradiation and IFNβ increase MSC production of IL7. We found that irradiation-induced increase of IL7 is impaired by Notch3 deficiency, while Dll4 ablation in endothelium also led to IL7 reduction based on scRNA-seq analysis. Whether NOTCH directly regulates IL7 expression remains to be determined. Notably, we identified a strong interaction between T/NK and stroma cells and induction of CAR cell IFN activities by targeting Perk. Irf1 is preferentially induced by IFNγ, while Irf1 can induce the expression of ISGs and upregulate type I IFN.⁶¹ Increased IFNβ in Perk-deleted mice can in turn drives IL7 production in CAR cells. IFNγ released from T/NK cells is likely regulated by NOTCH, and thus Notch may indirectly regulate CAR IL7 release through IFNγ. Alternatively, IFNγ can be induced by BM cells sensing self-ligands (dsDNA and dsRNA) released upon cell death, as Perk extinction increases pyroptosis. IFN plays a context-dependent role in HSCs, with short-term exposure promoting HSC expansion.^62-64 Together, these findings suggest that Perk ablation elicits DLL4-dependent coordinated IFN response, NOTCH activation, and IL7 release from CAR cells, which synergistically promote HSPC and B progenitor regeneration.

In summary, we identify that vascular PERK-DLL4 axis and CAR-expressing NOTCH3 as central regulators of a specialized “pre-B niche” that governs lymphoid regeneration following transplantation. By linking endothelial ER stress, vascular-stromal NOTCH signaling, IFN response, and IL7 production, our findings not only uncover fundamental mechanisms of lymphoid regeneration but also highlight PERK inhibition as a promising therapeutic strategy to improve immune recovery after myeloablative transplantation.

Limitations of the study

This study has several limitations. Firstly, analyses of the PERK-DLL4 axis were based on an endothelial-specific Perk single knockout or a combined Perk and Dll4 knockout model (DKO), precluding the assessment of effects solely caused by the increased DLL4 associated with PERK ablation. Although our work focuses on the PERK-DLL4 axis in bone marrow endothelium, it should be noted that DLL4 can regulate angiogenesis and HSC quiescence independently of PERK; therefore, the phenotypes observed in DKO mice are compounded by these non-PERK-regulated effects. Secondly, while PERK ablation increases DLL4 stability, the precise molecular machinery governing PERK-controlled DLL4 turnover remains incompletely defined and requires further investigation. Finally, the tamoxifen-inducible VE-cadherin^ERT2-Cre model targets whole-body endothelial cells, limiting the ability to exclusively determine the impact of the bone marrow endothelial adaptive ER stress response on hematopoietic regeneration.

STAR★METHODS

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Mice and treatment

Animal studies were approved by the IACUC (IS00007266 & IS00009402). B6;129S1-Notch3tm1Grid/J mice⁷¹ (JAX#010547) were backcrossed to C57BL/6 mice and confirmed to be 100% C57BL/6 background by genome scan analysis. Previously described floxed mice (Perk^F/F,72 Jag1^F/F,73 Dll4^F/F and VE-cadherin^ERT2–Cre mice⁷⁴ were maintained on the C57BL6 background and crossed to generate inducible endothelial-specific knockout of Perk, Dll4, or Jag1 by five consecutive doses of tamoxifen. PERK and DLL4 expression were assessed by anti-PERK (BS-2469R; Bioss Antibodies Inc.) and anti-DLL4 (HMD4-1; Thermo Fisher) through FACS. For pharmacologic treatments, mice received GSK2656157 (Selleckchem) (50 mg/kg in 10% DMSO and 90% corn oil) through oral gavage twice daily starting on day 3 after transplantation for 12 days. BI09²² was administered daily by intraperitoneal injection (25 mg/kg) for 5 days, starting one day before irradiation. For DLL4 blockade, mice received anti-DLL4 (Genentech) through intraperitoneal injection (15 mg/kg) every three days starting on day 8 after transplantation for a total of 4 doses.

BM transplantation, irradiation, and 5-FU treatment

BM transplantation was performed as described previously.⁷⁵ Briefly, 2.0 × 10⁶ total BM cells from WT donor mice (Ly5.1) were transferred retro-orbitally to lethally irradiated (950 cGy Cesium-137 irradiator or 850 cGy X-ray irradiator; lethal for at least 50% of mice by day 30 (LD50/30) in our facility) WT recipients (Ly5.2). For competitive secondary and tertiary transplantation, BM cells from primary or secondary transplant recipients (Ly5.1) were assessed for HSC frequency, adjusted for the equivalent HSC numbers within 1.0 × 10⁶ competitive bone marrow cells (Ly5.2), and infused together with 1.0 × 10⁶ competitive cells into lethally irradiated secondary or tertiary recipients (Ly5.2). Non-lethal irradiation (550 cGy X-ray) was performed in assessing endothelial DLL4 expression, whole mount and intravital imaging, and scRNA-sequencing. Chemotherapy with 5-FU was performed by a single IP injection of 5-FU (150 mg/kg).

MSC culture and treatment

BM stroma cells were collected and enriched as described.³⁷ Total BM cells were flushed from two femurs and two tibias, resuspended in MSC medium (MesenCult Expansion Kit, Catalog # 05513, StemCell) after red cell lysis, and plated in 6-well plates and allowed to adhere. Non-adherent hematopoietic cells were removed daily for the first 3 days and then twice per week with MSC medium. MSCs were harvested for FACS analysis after reaching 75% confluence or further enriched by sorting. Mesensphere formation was carried out using a modified protocol.⁷⁶ Briefly, MSCs cultured in 2D were digested with 0.05% trypsin, neutralized with 10 volumes of 10% FBS and resuspended in MSC 3D culture medium at clonal density in ultra-low attachment 96-well plates (Corning #3474). The MSC 3D culture medium contained 15% chicken embryo extract, 0.1 mM β-mercaptoethanol, 1% non-essential amino acids (Sigma), 1% N2, 2% B27 supplements (Gibco), and 20 ng/mL each of fibroblast growth factor (FGF)-basic, insulin-like growth factor-1 (IGF-1), epidermal growth factor (EGF), platelet-derived growth factor (PDGF) and oncostatin M (OSM) (Peprotech) in DMEM/F12 (1:1)/human endothelial (Gibco) (1:2). MSC cultures were maintained at 37°C with 5% CO₂ in a water-jacketed incubator and left untouched for 1 week to allow mesensphere formation, with half medium changes performed weekly. For coculture assays, mouse BM Lin⁻c-kit⁺ (LK) cells were prepared using CD117 microbeads following anti–biotin magnetic microbeads (Miltenyi Biotec) after incubation with a cocktail of biotinylated antibodies (Gr-1, CD11b, CD4, CD8, NK1.1, B220, and TER119). Approximately 500 Lin⁻c-kit⁺ cells were cocultured with ~20 mesenspheres in medium (StemSpan; STEMCELL Technologies) supplemented with 50 ng/mL of SCF and 100 ng/mL of Flt3L (R&D Systems) for 5 days at 37°C. For analysis of IL7 in MSCs, cells were stimulated with or without IFNβ (1500 IU/mL, 4 h, 37°C) followed by analysis IL7 expression by FACS.

METHOD DETAILS

Analysis of peripheral and bone marrow hematopoiesis

BM reconstitution was analyzed at d21 and monthly thereafter by manual peripheral blood white blood cell counts (hemocytometer) and FACS immunophenotyping. Three independent groups of transplanted mice were analyzed separately, and results were pooled. For BM HSPC reconstitution analysis, BM nucleated cells collected from two femurs and two tibias were stained with a cocktail of biotinylated lineage antibodies followed by analysis of HSPC (Lin⁻Sca1⁺c-kit⁺). Antibodies were purchased from BD (San Jose, CA), eBioscience (San Diego, CA), and Biolegend (San Diego, CA) and included those against the following antigens: CD4 (RM4-5), CD8α (53-6.7), B220 (RA3-6B2), CD11b (M1/70), Gr-1 (RB6-8C5), TER119 (TER-119), c-KIT (2B8), Sca1 (D7), CD150 (9D1), CD34 (RAM34), CD16/CD32 (2.4G2), and Flt3 (A2F10). Analysis of HSPC (Lin⁻Sca1⁺c-kit⁺) was achieved by gating on lineage⁻ (CD4⁻CD8⁻B220⁻CD11b⁻Gr-1⁻TER119⁻) cells. Analysis of CAR and endothelial cells was achieved by gating on CD45⁻TER119⁻CD71⁻ cells. Antibodies used for analysis are CD31 (390), CD71 (RI7217), PERK (C33E10), p-PERK (T980) (16F8), DLL4 (HMD4-1), Cxcl12 (79018), and IL7 (PA5-79509). FACS analyses were performed on CytoFLEX (Case Western Reserve University), BD LSRII or MACSQuant Analyzer 16 (Houston Methodist Hospital) after washing cells with PBS. For p-EPRK and Cxcl12 analysis, BM cells were incubated with CD31, TER119, and CD45 for 20 min on ice and then fixed (BD Biosciences #557870) and permeabilized (BD Biosciences #554723) for 20 min at room temperature. Cells were then incubated with the p-PERK or Cxcl12 antibodies for 20–30 min. For IL7 analysis, cells were incubated with anti-IL7 antibody (1:100, Invitrogen #PA5-79509) at room temperature for 45 min after fixation/permeabilization followed by staining with FITC conjugated anti-rabbit antibody (Biolegend #406403) at room temperature for 20 min.

qRT-PCR analysis

Total RNA was isolated using Direct-zol RNA Miniprep Plus Kits (Zymo Research, R2070) according to the manufacturer’s instructions. cDNA was synthesized using iScript Reverse Transcription Supermix (BioRad, #1708841). Quantitative RT-PCR (qRT-PCR) reaction was performed using cDNA with SsoAdvanced Universal SYBR? Green Supermix (BioRad, 1725271) on a real-time PCR system (CFX Duet, BioRad).

Immunohistochemistry

Studies involving human marrow specimens were approved by the Institutional Research Board (IRB) of the Houston Methodist Research Institute. Expression of p-PERK and DLL4 was evaluated by immunohistochemistry (IHC). Tissues sections were incubated overnight at 4°C with the primary antibodies against p-PERK (1:200, Cell Signaling) DLL4 (21584-1-AP, 1:400, Proteintech). Slides were scanned using the MoticEasyScan Pro 6 (Motic) and acquired images were analyzed with the accompanying software. For mouse bone marrow IHC, femur and tibia were isolated and temporarily preserved in ice-cold RPMI-1640 with 5% FBS on ice. Thick filter papers were pre-wet with ice-cold RPMI-1640 with 5% FBS. BM was aspirated with 23G needle and then attached to the filter paper before fixation with 4% formalin overnight. The fixed BM was first transferred to 70% ethanol to wash off formalin and then embedded in 1% agarose (A9539, Sigma-Aldrich) and cooled down to solidify. The solidified BM/agarose cubes were then embedded in paraffin. IHC staining of PERK (1:200) and DLL4 (1:400) were performed on 4 μm-thick slides with EDTA antigen retrieval buffer (#00-4956-58, eBioscience IHC Antigen Retrieval Solution - High pH (10X), Invitrogen). Primary antibodies were incubated at 4°C overnight. Secondary antibody and DAB staining was performed with Ultra Streptavidin HRP Kit according to manufacturer’s protocol. Slides were then subjected to 10% hematoxylin (ab220365, Abcam) counterstaining, dehydration and sealed with coverslips. Photos were taken with Olympus Model BX43 Microscope Configured with DP74 Digital Camera System.

Multiphoton imaging and analysis for vessel leakiness

Intravital 2-photon imaging preparation was performed as previously described.^75,77 Mice were anesthetized with isoflurane and placed in a stereotactic holder. The skin above the calvarium was carefully removed and a well was created with dental acrylic around the calvarium to hold sterile saline for imaging. The mouse was positioned under the microscope to image the frontal bone near the junction of the interfrontal and coronal sutures. Vessel dye containing 1 mg each of 2000 kD FITC-Dextran or 70 kD TRITC-Dextran (Sigma-Aldrich) was injected retro-orbitally immediately before imaging. Images were acquired on a Leica SP5 microscope equipped with a Coherent Chameleon Ti/Sapphire laser tuned to 800 nm and a 4-channel-NDD detector while mice were under inhaled anesthesia (1%–2% isoflurane) on a warmed microscope stage (37°C). Three-dimensional images consisting of XY: 775 mm × 775 mm and Z: 140 mm (5 mm/step), were obtained every 30 s for 30 min. Images were analyzed with Imaris software (Oxford Instruments). Datasets were drift-corrected, smoothed, and surfaces were generated on both FITC and TRITC channels. The corresponding total fluorescent volumes for each tracer were calculated at each timepoint and normalized to the first timepoint.

Whole-mount immunostaining

To analyze the vascular network, 10 μg each of Alexa Fluor 647 anti-mouse CD31 and Alexa Fluor 647 anti-mouse CD144 (BioLegend) were administered intravenously 10 min before euthanasia, as described.^78,79 Sternal bone fragments were fixed in ice-cold 4% PFA for 3 h. Whole-mount fragments were directly imaged on a Leica SP5 inverted confocal imaging system. Fluorescent emissions were collected using internal detectors set to 494–538 nm (GFP), 553–617 nm (PE), and 644–710 nm (AF647). High resolution three-dimensional (xyz) scans were performed using a Leica 10x objective (N.A. 0.4), with XYZ voxel sizes of 1.5 μm × 1.5 μm × 5 μm. Images were analyzed using Imaris software (Bitplane, Inc, Belfast, UK). Volume data for each image was determined by generating surfaces in Imaris based on the corresponding fluorescent signals. The overall volume of the image was used to normalize the data and to determine percentages.

Protein half-life analysis

DLL4 protein half-life was assessed after 100 μM cycloheximide was added to HUVEC cells. HUVEC cells were authenticated by sing specific protein markers like CD31 and VE-cadherin via flow cytometry and were regularly screened for Mycoplasma. To study whether DLL4 is degraded through the lysosomal or proteasomal degradation pathway, 100 μM Chloroquine or 25 μM MG132 were included with cycloheximide in cultured HUVEC cells. Cells were lysed at times indicated in RIPA lysis buffer supplemented with protease/phosphatase inhibitor. Protein lysates were used for Western blot and DLL4 protein levels were quantified after normalized to actin.

QUANTIFICATION AND STATISTICAL ANALYSIS

scRNA-seq library construction

For scRNA-seq, mice were irradiated with a sublethal 550 cGy (RS 2000) sufficient to induce evident damage to the BM vasculature and myelosuppression.⁸⁰ Single-cell RNA sequencing libraries were prepared according to manufacturer’s instructions (10X Genomics, 3′ V3) using pooled mouse marrow cells (n = 4–5/genotype; female and male mixed) collected on day 11–12, a time point when the collapsed vascular network has recovered by whole mount immunostaining and leakiness assay in the wild type mice based on other reports and our own analysis.⁸⁰ The scRNA-seq libraries were generated by Chromium nest GEM Single Cell 3′ v3.1 kit (PN-1000263, 10x Genomics, Pleasanton, CA) according to the manufacturer’s protocol. Pre-made libraries were sequenced on a Novaseq 6000 (Illumina) instrument to obtain a minimum of 30,000 paired end reads per cell.

scRNA-seq data quality control and cell annotation

The FASTQ data were processed using CellRanger (version 7.1.0), and downstream analyses were conducted in R with Seurat package (version 5.0.1).⁶⁵ Specifically, doublets were eliminated using DoubletFinder,⁶⁶ and ambient RNA counts were corrected with DecontX.⁸¹ Low-quality cells, characterized by high mitochondrial content and low gene detection rates, were also filtered out. The quality control workflow resulted in 38,031 total cells, with each condition ranging from 8,062 to 10,366 cells. Gene counts were SCTransformed and identified top 3,000 highly variable genes (HVGs) to obtain principal components (PCs). All libraries were integrated by Harmony using top 30 PCs and then clustered by the Louvain method.⁸² Cell clusters were annotated as fourteen cell population based on known cellular markers. SingleR and reference mapping approaches were used to verify cell type annotations (using bone marrow single cell atlas) to ensure accurate annotation of HSPCs/cycling progenitors under irradiation conditions. Stromal cells were further subset (n = 2798) and re-clustered following the above-mentioned approach. The refined clustering identifies five stromal cell subtypes. Parameter titrations (data not shown) were applied using different numbers of HVGs, PCs, and clustering resolutions to ensure a robust phenotyping of stromal cell types.

scRNA-seq computational analysis

The integrated scRNA-seq data were used to identify differentially expressed genes (DEGs) using MAST algorithm,⁶⁷ followed by pathway analysis. MAST is a two-part, generalized linear modeling approach that can robustly detect DEGs and calculate fold changes of gene expression, even with varying numbers of single cells in different groups and for lowly expressed genes. MAST result reported log2 fold-change (L2FC) and Benjamini-Hochberg (BH) adjusted p-value (P_adj). For gene expression bar plots, gene expression data were first normalized to a unit of count per 10,000 (CP10K) and grouped by experimental conditions. Group-wise statistical comparisons were reported using the results from MAST algorithm. Then, all DEGs were ranked based on L2FC × Truncated (−LOG10(P_adj)) for Gene Set Enrichment Analysis (GSEA). Gene Ontology (GO) database was used for functional annotation in the ClusterProfiler package.⁶⁹ Gene signature activity scores (e.g., ISG, NOTCH activity) were quantified using the AddModuleScore function in Seurat at the single cell resolution. Cell-cell communication was analyzed using CellChat based on log-normalized counts.⁷⁰ CellChat inferred cell-type-wise communication probability for each experimental group and ligand-receptor (LR) pairs. The bubble and circle plots visualized the cell-cell communication strength for selected signaling pathway and LR pairs. Gene regulatory networks were analyzed using pySCENIC,⁶⁸ which computationally inferred the activities of transcription factors (TFs) and signature activities at single cell resolution from gene expression data. Wilcoxon rank-sum test was deployed to statistically compare the differentially activated TFs across experimental conditions.

Statistical analysis

Statistics of all paired experiments were analyzed by two-tailed Student’s t test or two-way ANOVA. Statistics of scRNA-seq were described in the figure legends.

Supplementary Material

Supplemental information can be found online at https://doi.org/10.1016/j.celrep.2026.117953.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
PERK Polyclonal Antibody	Bioss	RRID: AB_10855213
DLL4 (delta-like 4) Monoclonal Antibody (HMD4-1)	Thermo Fisher Scientific	RRID: AB_11043100
APC anti-mouse CD117 (c-Kit) Antibody	Biolegend	RRID: AB_313221
PE/Cyanine7 anti-mouse Ly-6A/E (Sca-1) Antibody	Biolegend	RRID: AB_493596
CD150 Monoclonal Antibody (9D1), FITC	Thermo Fisher Scientific	RRID: AB_2534267
FITC Rat anti-Mouse CD34	BD Biosciences	RRID: AB_395017
BD Pharmingen™ Purified Rat Anti-Mouse CD16/CD32 (Mouse BD Fc Block™)	BD Biosciences	RRID: AB_394657
PE anti-mouse CD135 Antibody	Biolegend	RRID: AB_1877217
Biotin anti-mouse CD4 Antibody	Biolegend	RRID: AB_312711
Biotin anti-mouse CD8a Antibody	Biolegend	RRID: AB_312743
Biotin anti-mouse/human CD45R/B220 Antibody	Biolegend	RRID: AB_312989
Biotin anti-mouse/human CD11b Antibody	Biolegend	RRID: AB_312787
Biotin anti-mouse NK-1.1 Antibody	Biolegend	RRID: AB_313391
Biotin anti-mouse Ly-6G/Ly-6C (Gr-1) Antibody	Biolegend	RRID: AB_313369
Biotin anti-mouse TER-119/Erythroid Cells Antibody	Biolegend	RRID: AB_313705
APC/Cyanine7 Streptavidin	Biolegend	RRID: AB_3754458
APC anti-mouse CD31 Antibody	Biolegend	RRID: AB_312905
CD71 (Transferrin Receptor) Monoclonal Antibody (R17217 (RI7 217.1.4)), APC	Thermo Fisher Scientific	RRID: AB_1834355
PERK (C33E10) Rabbit Monoclonal Antibody	Cell Signaling Technology	RRID: AB_2095847
Phospho-PERK (Thr980) (16F8) Rabbit mAb	Cell Signaling Technology	RRID: AB_2095853
Phospho-PERK (Thr982) Polyclonal Antibody	Thermo Fisher Scientific	RRID: AB_2815938
eIF2 alpha Antibody	Cell Signaling Technology	RRID: AB_2230924
Phospho-eIF2 alpha (Ser51) Antibody	Cell Signaling Technology	RRID: AB_330951
CXCL12 Monoclonal Antibody (79018), FITC	Thermo Fisher Scientific	RRID: AB_2608710
IL-7 Polyclonal Antibody	Thermo Fisher Scientific	RRID: AB_2746625
FITC Donkey anti-rabbit IgG (minimal x-reactivity) Antibody	Biolegend	RRID: AB_893531
DLL4 Polyclonal antibody	proteintech	RRID: AB_2878888
DLL4 Antibody	Cell Signaling Technology	RRID: AB_2092960
IL-7 Polyclonal Antibody	Thermo Fisher Scientific	RRID: AB_2746625
BD Pharmingen™ Alexa Fluor® 647 Rat Anti-Mouse CD144	BD Biosciences	RRID: AB_2737608
BD Pharmingen™ Alexa Fluor® 647 Rat Anti-Mouse CD31	BD Biosciences	RRID: AB_2738313
BD Pharmingen™ FITC Rat Anti-Mouse Ly-6A/E	BD Biosciences	RRID: AB_10898185
Pacific Blue™ anti-mouse CD48 Antibody	Biolegend	RRID: AB_756140
PE-cy7 B220	Biolegend	RRID: AB_313004
PE anti-mouse CD19 Antibody	Biolegend	RRID: AB_2629817
CD43 Monoclonal Antibody (eBioR2/60), FITC, eBioscience™ - 11-0431-82	Thermo Fisher Scientific	RRID: AB_465040
BD Pharmingen™ PE-Cy™7 Rat Anti-Mouse CD45R/B220	BD Biosciences	RRID: AB_394458
PE anti-mouse CD3 Antibody	Biolegend	RRID: AB_312663
Alexa Fluor® 700 anti-mouse CD45 Antibody	Biolegend	RRID: AB_493715
FITC anti-mouse Ly-6G/Ly-6C (Gr-1) Antibody	Biolegend	RRID: AB_313371
PE/Cyanine7 anti-mouse/human CD11b Antibody	Biolegend	RRID: AB_312799
FITC Donkey anti-rabbit IgG (minimal x-reactivity) Antibody	Biolegend	RRID: AB_893531
beta-Actin (8H10D10) Mouse Monoclonal Antibody	Cell Signaling Technology	RRID: AB_2242334
α-Tubulin (DM1A) Mouse mAb	Cell Signaling Technology	RRID: AB_1904178
Anti-rabbit IgG, HRP-linked Antibody	Cell Signaling Technology	RRID: AB_2099233
Anti-mouse IgG, HRP-linked Antibody	Cell Signaling Technology	RRID: AB_330924
BV605 Rat Anti-Mouse CD140a	BD Biosciences	RRID: AB_2740111
PE anti-mouse CD51 Antibody	Biolegend	RRID: AB_2129493
DLL4 blocking antibody	Genentech	YW152F
anti-Ragweed isotype control IgG1 antibody	Genentech	N/A
Biological samples
Human bone marrow FFPE slides	This paper	N/A
Chemicals, peptides, and recombinant proteins
Tamoxifen	Selleckchem	S1238
GSK2656157 PERK inhibitor	Selleckchem	S7033
B I09	Chih-chi Andrew Hu lab	N/A
BD Phosflow™ Fix Buffer I	BD Biosciences	557870
BD Perm/Wash™ Perm/Wash Buffer	BD Biosciences	554723
MesenCult™ Expansion Kit	StemCell	05513
Trypsin-EDTA (0.05%), phenol red	Gibco	25300120
Fetal Bovine Serum, qualified, heat inactivated, United States	Gibco	16140071
chicken embryo extract	MP medicine	092850145
B27 supplement (50×), serum free	Gibco	17504044
N-2 Supplement (100X)	Gibco	17502048
Mouse fibroblast growth factor (FGF)-basic	Gibco	450-33-50UG
Mouse IGF-I Recombinant Protein	Gibco	250-19-50UG
Mouse EGF Recombinant Protein	Gibco	315-09-100UG
Mouse PDGF-BB Recombinant Protein	Gibco	315-18-10UG
Human Oncostatin M (227 aa) Recombinant Protein	Gibco	300-10-10UG
DMEM/F-12	Gibco	11320033
Human Endothelial SFM	Gibco	11111044
CD117 MicroBeads, mouse	Miltenyi Biotec	130-091-224
Anti-Biotin MicroBeads	Miltenyi Biotec	130-097-046
Recombinant Mouse SCF Protein	R&D Systems	455-MC
Recombinant Mouse Flt-3 Ligand/FLT3L Protein	R&D Systems	427-FL
Recombinant Mouse IFN-beta Protein	R&D Systems	8234-MB
StemSpan™ SFEM	StemCell	09650
Fluorescein isothiocyanate–dextran (FITC–Dextran 2000)	Sigma-Aldrich	52471
AAT Bioquest TRITC–dextran conjugate (average MW = ~70K)	Sigma-Aldrich	AATB-21708
Penicillin-Streptomycin (5,000 U/mL)	Gibco	15070063
7-AAD Viability Staining Solution	Biolegend	420404
DAPI (4′,6-Diamidino-2-Phenylindole, Dilactate)	Biolegend	422801
Zombie Aqua™ Fixable Viability Kit	Biolegend	423102
Complete Endothelial Cell Medium	Cellbiologics	M1168PF
BD Horizon™ Brilliant Stain Buffer	BD Horizon	566349
Endothelial Cell Growth Medium (500 mL)	Sigma-Aldrich	211-500
DharmaFECT Duo Transfection Reagent	Dharmacon	T-2010-01
ACK Lysing Buffer	Gibco	A1049201
RIPA Lysis and Extraction Buffer	Thermo Scientific	89901
Laemmli SDS sample buffer, reducing (6X)	Thermo Scientific Chemicals	J61337.AD
Opti-MEM™ I Reduced Serum Medium, no phenol red	Gibco	11058021
DMEM, high glucose	Gibco	11965092
SuperSignal™ West Pico PLUS Chemiluminescent Substrate	Thermo Scientific	34577
Trans-Blot Turbo 5x Transfer Buffer	BioRad	10026938
Precision Plus Protein WesternC Blotting Standards	BioRad	1610399
EveryBlot Blocking Buffer	BioRad	12010020
StarBright Blue 700 Goat Anti-Mouse IgG	BioRad	12004159
10x Tris/Glycine/SDS	BioRad	1610732
10x Tris/Glycine Buffer for Western Blots and Native Gels	BioRad	1610734
Direct-zol RNA Miniprep Kits	Zymo Research	R2052
TRIzol™ Reagent	Invitrogen	15596018
SsoAdvanced Universal SYBR® Green Supermix	BioRad	1725272
iScript™ Reverse Transcription Supermix for RT-qPCR	BioRad	1708841
Cycloheximide solution	Sigma-Aldrich	C4859
(R)-MG132	Sigma-Aldrich	M8699
Chloroquine	Selleckchem	S6999
Critical commercial assays
Ultra Streptavidin HRP Kit (Multi-Species, DAB) Antibody	Biolegend	929501
MesenCult™ Expansion Kit (Mouse)	StemCell	05513
Deposited data
Raw and analyzed single cell RNA sequencing data	This paper	GEO: GSE308168
A Balanced Bone Marrow Reference Map of Hematopoietic Development (Human)	Zeng et al.	https://cellxgene.cziscience.com/datasets
Experimental models: Cell lines
Primary Umbilical Vein Endothelial Cells; Normal, Human (HUVEC)	ATCC	PCS-100-010
Mouse Primary Bone Marrow Derived Endothelial Cells (BMEC)	Cellbiologics	C57-6221
Experimental models: Organisms/strains
Mouse: B6;129S1-Notch3tm1Grid/J mice	The Jackson Laboratory	Stock #010547
Mouse: PERKflox/flox	Adoro lab
Mouse: JAG1flox/flox	Proweller lab
Mouse: DLL4flox/flox	Adam lab
Mouse: VE-cadherinERT2-Cre	Adam lab
Oligonucleotides
Kitl Forward 5′-GAATCTCCGAAGAGGCCAGAA-3′	This paper
Kitl Reverse 5′-GCTGCAACAGGGGGTAACAT
Cxcl12 Forward 5′-TGCATCAGTGACGGTAAACCA-3′
Cxcl12 Reverse 5′- TTCTTCAGCCGTGCAACAATC-3′
Il7 Forward 5′- TTCCTCCACTGATCCTTGTTCT-3′
Il7 Reverse 5′- AGCAGCTTCCTTTGTATCATCAC-3′
Actin Forward 5′- TCACACTGGCATCGTGATGGACTC-3′
Actin Reverse 5′- TCCTGCTTGCTGATCCACATCTGC-3′
Recombinant DNA
SignalSilence® PERK siRNA I	Cell Signaling Technology	9024
Software and algorithms
Imaris	Oxford Instruments	https://imaris.oxinst.com/
CellRanger 7.10	10x Genomics	https://www.10xgenomics.com/
R	The R Project for Statistical Computing	https://www.r-project.org/
Seurat package for R	Hao et al.65	https://satijalab.org/seurat/
DoubletFinder package for R	McGinnis et al.66	https://github.com/chris-mcginnis-ucsf/DoubletFinder
DecontX package for R	Yang et al.	https://www.bioconductor.org/
MAST algorithm package for R	Finak et al.67	https://rglab.github.io/MAST/
pySCENIC package for R	Van de Sande et al.68	https://github.com/aertslab/pySCENIC
clusterProfiler package for R	Wu et al.69	https://www.bioconductor.org/
Cellchat package for R	Jin et al.70	https://github.com/jinworks/CellChat
Samtools	Genome Research Limited	https://www.htslib.org/
FlowJo	Waters Biosciences	https://www.flowjo.com/
ImageJ/FIJI	NIH	https://imagej.nih.gov
Image Lab	BioRad	12012931
Other
Costar® 6-well Clear Flat Bottom Ultra-Low Attachment Multiple Well Plates, Individually Wrapped, Sterile	Life Sciences	3471
Corning® 96-well Clear Flat Bottom Ultra-Low Attachment Microplate, Individually Wrapped, with Lid, Sterile	Life Sciences	3474
MS Columns	Miltenyi Biotec	130-042-201
LS Columns	Miltenyi Biotec	130-042-401
4–15% Mini-PROTEAN™ TGX Stain-Free™ Protein Gels, 15 well, 15 μL	BioRad	4568086
10% Mini-PROTEAN® TGX Stain-Free™ Protein Gels, 15 well, 15 μL	BioRad	4568036
ImmunoBlot PVDF Membrane	BioRad	1620177
Extra Thick Blot Filter Paper	BioRad	1703965

Highlights.

• Endothelial PERK activation is a hall marker of bone marrow ablation

• Ablating endothelial PERK promotes DLL4 and post-myeloablation immune recovery

• EC DLL4 paired with MSC NOTCH3 promotes IL7 release and lymphoid regeneration