Abstract

   The tissue environment governs vascular remodeling, a key determinant
   of collateral circulation (CC) in ischemic disease, yet the mechanisms
   driving CC in adults remain unclear. Plasma profiling from patients
   with peripheral artery disease (PAD) and ischemic murine muscle
   revealed dysregulated lipid metabolism, including elevated APOA1
   binding protein (AIBP), with levels positively correlating with PAD
   severity. Myeloid cells enriched at CC sites increased AIBP expression
   postischemia. Genetic deletion of AIBP expanded CXCR4⁺ capillary
   endothelial cells (CECs) with stemlike and proliferative properties
   that remodeled into functional collaterals, a process blocked by CXCR4
   inhibition. Mechanistically, AIBP bound the endocytic receptor LRP2 to
   promote endothelial uptake of high-density lipoprotein (HDL)–associated
   miR-223, a repressor of CXCR4. Disruption of this AIBP–LRP2–HDL–miR-223
   axis restored CXCR4 and rescued CC growth. These findings define a
   two-phase mechanism in which stemlike CECs first expand and then
   transition to arterial fates, establishing a therapeutic strategy for
   revascularization in ischemic vascular disease.
     __________________________________________________________________

   Targeting the AIBP–LRP2–miR-223 axis reprograms endothelial cells to
   augment collateral circulation in ischemic disease.

INTRODUCTION

   Ischemic vascular disease, such as peripheral artery disease (PAD), is
   a prevalent and debilitating condition that causes tissue ischemia in
   the lower extremities due to vascular narrowing or occlusion ([58]1).
   Although stenting is a common treatment approach, its long-term
   effectiveness is often limited by in-stent restenosis and diffuse
   vascular disease, underscoring the need for alternative strategies to
   improve blood flow ([59]2). Collateral circulation (CC), defined by the
   formation of anastomotic vessels that bypass occluded arteries, serves
   as a critical compensatory mechanism to restore tissue perfusion
   ([60]3–[61]5). The development of robust CC is associated with better
   clinical outcomes ([62]6), making it a promising therapeutic approach.
   However, the mechanisms underlying collateral vessel formation remain
   poorly understood, and no treatments are currently available to enhance
   CC.

   Traditional models of CC have primarily focused on arteriogenesis, the
   enlargement of preexisting collateral arteries ([63]7), and
   arterialization, the transition of capillary endothelial cells (CECs)
   into arterial endothelial cells (AECs) ([64]8). Although these
   processes are well documented, the capacity of CECs to form direct
   artery-to-artery connections has been recognized in a prior study
   ([65]8), yet the underlying mechanisms and physiological relevance of
   this phenomenon remain largely unknown. Further investigation into this
   alternative route of vascular remodeling is needed to fully elucidate
   its contribution to CC and potential for therapeutic targeting.

   Emerging evidence suggests that chemokine receptor type 4 (CXCR4)
   signaling plays a pivotal role in arterial development and CC formation
   ([66]9). Studies have demonstrated that CXCR4 regulates arteriogenesis
   in both developmental and pathological contexts ([67]10). For example,
   CXCR4 is critical for the reassembly of CXCR4^+ AECs during neonatal
   heart regeneration ([68]11) and guides the migration of endothelial
   cells (ECs) from venous to arterial vasculature during zebrafish
   development ([69]5, [70]12). Despite these advances, the molecular
   mechanisms driving CC formation in the adult lower extremities remain
   largely uncharted.

   We hypothesize that the extracellular microenvironment, shaped by
   tissue ischemia, plays an integral role in orchestrating CC. Ischemic
   injury and subsequent immune cell infiltration trigger cellular
   responses that reshape the secretome—a complex repertoire of proteins
   and signaling molecules—which can either facilitate or impede vascular
   remodeling ([71]13). To identify key regulators within the secretome,
   we performed unbiased proteomic analyses of plasma from patients with
   PAD and healthy controls, as well as muscle tissue from a murine model
   of femoral artery ligation (FAL), an established model to induce CC in
   the hindlimbs ([72]14). These analyses highlighted lipid metabolism as
   a critical pathway, with APOA1 binding protein (AIBP) (aka NAXE)
   emerging as a central player linking lipid metabolism to vascular
   remodeling.

   Our findings uncover a previously unappreciated role for AIBP, derived
   from local myeloid cells, in regulating proximal CC formation. Contrary
   to reported models that emphasize CXCR4^+ AEC–mediated reassembly
   ([73]11), we show that AIBP depletion promotes the expansion of
   procollateral CXCR4^+ CECs exhibiting a stemness signature. A subset of
   these stemlike CECs commits to an arterial fate and contributes to
   functional CC. This identified mechanism not only challenges existing
   paradigms but also underscores the therapeutic potential of targeting
   the AIBP pathway to enhance CC and improve clinical outcomes in PAD and
   other ischemic conditions.

RESULTS

Unbiased profiling of secretory proteins highlights alterations in lipid
metabolism

   We hypothesized that the tissue environment, shaped by secretory
   components, orchestrates vascular remodeling in ischemic disease. To
   identify secreted proteins altered in PAD, we performed a proteomic
   analysis of patient plasma ([74]Fig. 1A). We detected a total of 1337
   proteins in both healthy individuals and patients with PAD (table S1).
   Among these, 664 (fold change ≥ 1.2) were increased and 329 (fold
   change ≤ 0.83) were reduced in PAD plasma versus normal plasma
   ([75]Fig. 1B). Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway
   analysis of the up-regulated proteins revealed enrichment in
   cholesterol metabolism ([76]Fig. 1C). Reanalysis of RNA sequencing
   (RNA-seq) data from a prior study ([77]15) yielded consistent findings
   (fig. S1, A and B). Among the proteins identified within the 10 most
   significantly enriched pathways linked to lipid metabolism (table S2),
   AIBP, a key regulator of vascular function ([78]16–[79]18), was
   markedly up-regulated in the plasma of patients with PAD ([80]Fig. 1D).
   This plasma increase was mirrored by higher AIBP levels in ischemic
   muscle tissue from patients with PAD ([81]Fig. 1, E and F). In
   contrast, other lipid metabolism–related proteins such as FABP5 and
   APOC4 showed no corresponding change in ischemic tissue (fig. S2, L and
   M). The concordance between plasma and tissue AIBP levels points to a
   potential direct role in ischemic tissue, prompting us to focus on AIBP
   in subsequent investigations. Subsequent analysis of AIBP protein
   expression in an independent cohort of patients with PAD (basic
   characteristics in table S3) confirmed a significant increase compared
   to healthy controls ([82]Fig. 1G). Notably, AIBP levels exhibited a
   negative correlation with the ankle-brachial index (ABI), where higher
   ABI values (<1.4) typically indicate better outcome of PAD ([83]Fig.
   1H). Consistent with this, AIBP levels were reduced in the patients
   with PAD with longer painless walking distance (>50 m) ([84]Fig. 1I).

Fig. 1. Unbiased proteomic analysis identifies lipid metabolism and AIBP
association with PAD.

   [85]Fig. 1.
   [86]Open in a new tab

   (A) Experimental design of proteomic profiling of plasma derived from
   healthy individuals and patients with PAD. (B) Volcano plot showing
   significantly up-regulated (red) and down-regulated (blue) proteins in
   PAD versus healthy controls. The representative lipid related proteins
   are indicated. (C) KEGG pathway enrichment analysis of up-regulated
   proteins (≥1.5-fold change). (D) Heatmap displaying significantly
   up-regulated proteins (≥1.5-fold) associated with vascular function in
   patients with PAD (n = 3 healthy; n = 3 PAD). (E and F) Representative
   Western blot analysis (E) and quantification (F) of AIBP protein in
   less ischemic versus ischemic muscles from patients with PAD (n = 8).
   (G) ELISA-based quantification of plasma AIBP levels in healthy
   individuals (n = 20) and patients with PAD (n = 48). (H) Pearson
   correlation between plasma AIBP levels and ABI, with correlation
   coefficient (R) calculated using GraphPad Prism 10. The R value
   reflects the strength and direction of the relationship. (I) ELISA
   analysis of plasma AIBP levels in patients with PAD stratified by
   pain-free walking distance. (J) Plasma AIBP levels in patients with PAD
   with collateral category 1 (n = 20) versus category 2 (n = 17),
   assessed by ELISA. (K) Time-course analysis of plasma AIBP levels in B6
   mice following FAL (n = 3 per time point; each sample pooled from
   plasma of three mice). Means ± SEM; *P < 0.05; **P < 0.01; n.s., not
   significant.

   We further evaluated collateral vessel formation in patients with PAD
   using computed tomography angiography (CTA) (fig. S1C), as previously
   reported ([87]19) (table S4). As expected, AIBP levels were lower in
   patients with more collaterals (category 2) compared with those with
   fewer collaterals (category 1) ([88]Fig. 1J). To investigate whether
   AIBP levels differ by sex, we performed enzyme-linked immunosorbent
   assay (ELISA) and found similarly elevated AIBP levels in both male and
   female patients with PAD compared with their respective controls (fig.
   S1, D and E). The negative correlation between AIBP and ABI (fig. S1, F
   and G), as well as the association between lower AIBP and improved
   walking distance (fig. S1, H and I), was consistently observed in both
   sexes.

   To determine whether these pathways are associated with CC, we
   subjected C57BL/6J (B6) mice to FAL. The adductor muscles, which harbor
   abundant collateral vessels, were harvested from the ischemic and
   control limbs for label-free proteomic analysis (fig. S2A). Our data
   revealed 849 up-regulated proteins (fold change > 1.5) in FAL muscles
   compared to controls (table S5), with 140 of the 1730 proteins being
   uniquely detected in the FAL group (fig. S2B). Among the total
   differentially expressed proteins, 16.6% were classified as secretory
   proteins according to curated human secretome database (fig. S2C, top).

   In support of our hypothesis that extracellular environment dictates
   CC, we found a profound increase in secreted proteins in the ischemic
   adductors versus the control adductors (50.2%; fig. S2C, bottom).
   Consistent with findings in humans, KEGG pathway analysis of these
   up-regulated proteins revealed enrichment in lipid metabolism pathways,
   including cholesterol metabolism and peroxisome proliferator–activated
   receptor (PPAR) signaling (fig. S2D). GO term analysis further
   identified enrichment in secretion-related categories such as
   extracellular region/component and secretory granule (fig. S2E).
   Twenty-eight proteins within the lipid metabolism pathway exhibited
   >1.2-fold increase (fig. S2F). Notably, plasma AIBP levels were also
   elevated in mice following FAL ([89]Fig. 1K). This ELISA-based method
   was validated using AIBP knockout mice, in which AIBP was undetectable
   (fig. S2G). This up-regulation of AIBP occurs predominantly at the
   translational or posttranscriptional level rather than the
   transcriptional level, as evidenced by increased protein levels
   detected by Western blot without corresponding changes in mRNA levels
   observed by quantitative polymerase chain reaction (qPCR) analysis
   (fig. S2, H to K). In agreement with the results from human PAD tissue
   analysis, FABP5 and APOC4 expression remained unchanged in the ischemic
   hindlimb tissues of mice (fig. S2, N and O). Together, our data from
   both patients with PAD and FAL mouse model indicate that the lipid
   metabolism pathway, in particular AIBP expression, is significantly
   increased in ischemic tissue, with AIBP levels mirroring the disease
   severity.

Extracellular AIBP is derived from leukocytes post-FAL and its up-regulation
by hypoxia

   A prior study assessed AIBP secretion using cell surface AIBP as a
   surrogate marker ([90]20). To identify the tissue source of
   extracellular AIBP, we isolated ischemic adductor muscles post-FAL and
   prepared a single-cell suspension for flow cytometry analysis. We
   surveyed eight cell types within the ischemic tissues, including
   fibro-adipogenic progenitors (FAPs), ECs, vascular smooth muscle cells
   (VSMCs), muscle stem cells (MUSCs), pericytes, skeletal muscle cells,
   macrophages, neutrophils, and monocytes. Compared to corresponding cell
   types in unoperated controls, a substantial elevation of cell surface
   AIBP was observed in monocytes (1.5-fold increase) and neutrophils
   (5.6-fold increase) following FAL ([91]Fig. 2, A to D). Given that
   hypoxia is a key feature of ischemic injury, we examined its effect on
   AIBP expression in THP-1, a human monocyte cell line. Western blot
   analysis revealed a time-dependent increase in AIBP expression under
   hypoxic conditions ([92]Fig. 2E), which was corroborated by flow
   cytometry ([93]Fig. 2, F and G) and ELISA of culture media ([94]Fig.
   2H). The sequential up-regulation of intracellular, cell surface, and
   secreted AIBP suggests an expected route of AIBP extracellular
   transport. These findings were replicated in murine J774A.1 monocyte
   cell line (fig. S2, P and Q), further supporting a myeloid origin for
   AIBP under ischemic conditions.

Fig. 2. AIBP originates from leukocytes following FAL, and its expression is
up-regulated by hypoxia.

   [95]Fig. 2.
   [96]Open in a new tab

   (A) Representative flow cytometry analysis of AIBP expression in
   various cell types isolated from adductors on day 7 post-FAL, including
   Sca-1^+CD31^−CD45^− FAPs, CD31^+CD45^− ECs, SMA^+CD31^−CD45^− VSMCs,
   VCAM1^+CD31^−CD45^− MUSCs, Desmin^+CD31^−CD45^− cells (skeletal
   muscles, SKM), and NG2^+CD146^+ pericytes. (B) Representative flow
   cytometry analysis of AIBP expression in F4/80^+CD31^−CD45^+
   macrophages, Ly6G^+CD11b^+CD31^−CD45^+ cells (neutrophils), and
   Ly6c^+CD11b^+CD31^−CD45^+ cells (monocytes) isolated from mice
   adductors on day 7 post-FAL. (C) Quantification of AIBP^+ cells across
   the populations shown in (A) and (B). Data represent n = 3 independent
   experiments, with each sample consisting of adductors pooled from five
   mice. (D) Representative flow cytometry analysis of AIBP expression in
   Apoa1bp^−/− monocytes (left) and AIBP fluorescence minus one (FMO)
   control in wild-type monocytes (right). Cell viability was assessed
   using the Zombie Aqua Fixable Viability Kit. n = 3, with each adductors
   pooled from five mice. (E) Immunoblotting analysis (top) and
   quantification (bottom) of AIBP expression in THP-1 following hypoxia
   (1% O[2]) treatment for the indicated durations. n = 7 repeats. h,
   hours. (F and G) Representative flow cytometry analysis (F) and
   quantification (G) of AIBP expression in THP1 cells exposed to hypoxia.
   n = 6 independent experiments. IgG, immunoglobulin G; FITC, fluorescein
   isothiocyanate; a.u., arbitrary units. (H) ELISA of AIBP concentration
   in the culture medium of THP1 cells exposed to hypoxia. n = 3
   independent experiments. Means ± SEM; *P < 0.05; **P < 0.01;
   ****P < 0.0001; n.s., not significant.

   To assess the functional relevance of extracellular AIBP in CC, B6 mice
   were administered intravenous recombinant AIBP or control bovine serum
   albumin (BSA) following FAL. We performed micro–computed tomography
   (micro-CT) analysis to quantify the hindlimb arterial network in the
   ligated adductors (fig. S3A). We found that AIBP treatment markedly
   reduced arterial interconnections, as evidenced by decreased vascular
   volume and impaired growth of collateral vessels (fig. S3, B and C).
   Immunofluorescence staining of α–smooth muscle actin (α-SMA) further
   validated these findings (fig. S3, D and E).

   Collateral artery growth requires EC proliferation. To analyze this, we
   performed Ki67 immunostaining on the ischemic adductors. Compared to
   BSA, AIBP delivery suppressed the endothelial proliferation in the
   ischemic muscle (fig. S3, F and G). Laser Doppler imaging revealed ~70%
   recovery of arterial flow in the BSA-treated mice by day 21, whereas
   flow recovery was significantly impaired in the AIBP-injected mice
   (fig. S3, H and I). Consequently, mice receiving AIBP exhibited greater
   gastrocnemius necrosis (fig. S3, J and K) and worse mobility (fig. S3,
   L and M), and this was analyzed using (i) Tarlov score that assesses
   motor function and (ii) modified ischemia score that evaluates the
   severity of ischemia (table S6) ([97]21). These findings indicate that
   elevated extracellular AIBP limits collateral growth and exacerbates
   vascular and functional deficits following ischemic injury.

Identification of AIBP-regulated stemlike CXCR4^+ CEC cluster following FAL

   To elucidate how AIBP depletion promotes CC, we performed single-cell
   RNA sequencing (scRNA-seq) on unsorted single-cell suspensions from
   Apoa1bp^−/− and control adductors at day 7 post-FAL (fig. S4A), an
   approach that minimized tissue processing–induced transcriptome change.
   After excluding low-quality cells on the basis of mitochondrial gene
   content (fig. S4, B and C), we retained 12,391 control and 8359
   Apoa1bp^−/− cells. ECs constituted ~3.6% of total cells, consistent
   with normal muscle vasculature. T-distributed Stochastic Neighbor
   Embedding (t-SNE) analysis also identified fibroblasts and immune cells
   (fig. S4D).

   Differentially expressed gene (DEG) analysis [Log[10](Fc) > 0.25]
   revealed that Apoa1bp deficiency up-regulated Icam1, Icam2, and Vcam1
   (fig. S4, E and F), suggesting an enhanced immune response ([98]22).
   Apoa1bp^−/− ECs also showed increased expression of arterial markers
   ([99]23) Gja4, Gja5, and Gjc1 (fig. S4, G and H). NOTCH1 is a
   well-established regulator of arterial specification. Although Notch1
   expression was reduced in Apoa1bp^−/− ECs, expression of its canonical
   downstream targets—Hes1, Hey1, and Hey2—remained largely unchanged
   (fig. S4, I and J), implying a NOTCH1 signaling–independent mechanism.

   We next focused on EC subpopulations. T-SNE analysis identified four
   distinct EC clusters ([100]Fig. 3A), each defined by unique marker
   genes ([101]Fig. 3B). Two clusters were CECs (CEC1 and CEC2),
   distinguished by Plvap ([102]22) in CEC1 and Cxcl10/Cd52 ([103]24)
   CEC2, with the latter also expressing proliferation markers Ki67 and
   Pcna. Cluster 3 comprised AECs bearing arterial markers Sema7a and
   Efnb2 ([104]24), whereas cluster 4 included venous ECs (VECs) enriched
   in Nr2f2 (CoupTFII) (fig. S4K).

Fig. 3. scRNA-seq analysis reveals the crucial role of AIBP in collateral
vessel growth.

   [105]Fig. 3.
   [106]Open in a new tab

   (A) Unbiased clustering of 944 ECs isolated from adductor muscles of
   control and AIBP knockout mice using scRNA-seq and t-SNE dimensionality
   reduction. Four transcriptionally distinct EC subpopulations were
   identified: AECs, VECs, CECs, and LECs. (B) Heatmap displaying the top
   50 marker genes for each EC cluster. (C) t-SNE visualization of
   individual ECs. Cyan depicts control ECs, and yellow marks show
   Apoa1bp^−/− ECs. (D) Quantification of the CEC2 subpopulation in
   control versus Apoa1bp^−/− ischemic adductors. (E) Violin plots
   depicting Cxcr4 mRNA expression across EC clusters. (F)
   Stemness-associated gene list in CEC2 based on our ChatGPT analysis.
   (G) Upstream regulator analysis in CEC2 using Ingenuity Pathway
   Analysis. FDR, false discovery rate. (H) Box plot showing stemness and
   proliferation scores across EC clusters. (I) Stemness analysis of the
   CEC2 cluster performed using CytoTRACE. (J) Bubble plot depicting
   enriched metabolic pathways in CEC2. (K) Pseudotime trajectory analysis
   of AEC, CEC1, CEC2, LEC, and VEC clusters to infer lineage
   relationships. (L) CellChat analysis indicating strong communication
   between macrophages/neutrophils and the CEC2 cluster. (M) Probability
   of cell-cell interactions between macrophages/neutrophils and EC
   clusters (AECs, CEC1, CEC2, and VECs) predicted by CellChat. (N)
   Ligand-receptor interaction analysis between macrophages/neutrophils
   and specific EC subsets. Means ± SEM; ****P < 0.001.

   Apoa1bp deficiency doubled the proportion of CEC2 ([107]Fig. 3, C and
   D), which expressed the stemness marker Cxcr4 ([108]Fig. 3E). Cell
   identity analysis using ChatGPT-based algorithms ([109]25) identified
   14 stemness-associated genes enriched in CEC2 ([110]Fig. 3F). Upstream
   regulator analysis highlighted MYC as the top driver ([111]Fig. 3G),
   consistent with its known role in driving stem cell proliferation and
   metabolic reprogramming. CEC2 also exhibited a proliferative signature
   ([112]Fig. 3H). Cell fate assessment revealed that both CECs and VECs
   exhibited more stemlike characteristics compared to AECs or lymphatic
   endothelial cells (LECs) (fig. S4L). Intriguingly, mechanistic target
   of rapamycin (mTOR) signaling was suppressed as rapamycin and Torin1
   emerged among the top regulators. Given mTOR’s role in restraining
   autophagy and promoting anabolic metabolism, its inhibition may
   facilitate autophagy-driven repair and support a metabolically
   adaptive, proregenerative EC phenotype in the ischemic settings
   ([113]26–[114]28). In addition, p53 also emerged as an upstream
   regulator, consistent with its role in stress-induced angiogenesis
   ([115]29) and mTOR adaptation ([116]30).

   AIBP knockout significantly elevated the stemness score ([117]Fig. 3I)
   and cell cycle score (fig. S4M). In line with this, flow cytometry
   analysis revealed a significant increase in c-Kit^+CXCR4^+ ECs in
   AIBP-deficient mice post-FAL (fig. S4N). Genes related to lipid
   metabolism were also up-regulated in the CEC2 cluster ([118]Fig. 3J),
   paralleling our proteomic findings. Pseudotime analysis suggested a
   differentiation trajectory from CEC2 to CEC1 and ultimately to AECs,
   with a minor contribution to VECs ([119]Fig. 3K and fig. S4O),
   suggesting that these stemlike capillary cells transition toward
   arterial fates during CC. Along this trajectory, lipid metabolism genes
   remained dynamically regulated (fig. S4P), implying that CEC2 exhibits
   proliferative, proregenerative, and metabolic adaptations to form
   functional CC.

   Cell-cell cross-talk plays a critical role in CC. CellChat analysis
   revealed preferential cross-talk between immune cells—particularly
   macrophages and neutrophils—and the CEC2 subpopulation, surpassing
   their interactions with other endothelial subtypes ([120]Fig. 3, L and
   M). In line with a prior study ([121]11), our receptor-ligand analysis
   confirmed the interaction between macrophage-derived CXCL12 and CXCR4
   in AECs. Furthermore, we identified migration inhibitory factor (MIF)
   as the key ligand from macrophages binding to CXCR4 in CEC2 ([122]Fig.
   3N), consistent with MIF’s role in regulating endothelial progenitor
   cell function and neovascularization ([123]31). These findings are
   consistent with the inflammatory signature associated with CEC2 as
   lipopolysaccharide was identified as one of the upstream regulators
   ([124]Fig. 3G). Overall, these data suggest that AIBP-regulated
   stemlike CECs play a pivotal role in driving CC through mechanisms that
   involve immune cross-talk, metabolic rewiring, and mTOR suppression.

Apoa1bp depletion increases formation of small-sized collateral vessels in
the ischemic hindlimbs

   We created Apoa1bp knockout mice to investigate the functional role of
   AIBP ablation in CC (fig. S5, A and B). To determine whether AIBP
   governs formation of preexisting collateral vessels, we performed
   micro-CT analysis to quantify arterial number and diameter ([125]Fig.
   4, A and B) and conducted SMA immunostaining in the adductor muscles
   (fig. S5, C and D) of unoperated control and Apoa1bp^−/− adult mice.
   Both groups exhibited comparable collateral vessel profiles, consistent
   with the quiescent nature of vasculature in healthy adults, which
   limits the detectability of active collateral formation. In contrast,
   vascular remodeling is more prominent in the early postnatal period
   ([126]32). We therefore expanded our investigation to postnatal day 6
   (P6) mice at the well-established anatomical locations. Specifically,
   we analyzed interconnected vessels between the proximal caudal femoral
   artery (PCFA) and the saphenous artery (SA) in the adductor muscles and
   still found a similar number of collateral vessels (fig. S5, E and F).
   These findings indicate that AIBP loss does not affect collateral
   vessel development under baseline, nonischemic conditions in either
   adult or early postnatal mice.

Fig. 4. AIBP deficiency impairs hindlimb collateral vessel formation.

   [127]Fig. 4.
   [128]Open in a new tab

   (A and B) Representative micro-CT images (A) and quantification (B) of
   vessels <100 μm in diameter within the region of interest (ROI) from
   nonischemic limbs of Apoa1bp^−/− and control littermates. 3D rendering
   images are shown. n = 3 mice per group. Scale bar, 1 mm. (C and D)
   Representative micro-CT images (C) and quantification (D) of vessel
   volumes across various size ranges within the ROI from ischemic limbs
   on day 21 post-FAL. Yellow arrowheads indicate collateral arteries.
   n = 5 mice per group. Scale bar, 0.5 mm. (E and F) Representative
   confocal images (E) and quantification (F) of CD31^+ (green) and SMA^+
   (red) arteries in adductors from Apoa1bp^−/− and control mice on day 21
   post-FAL. n ≥ 5 mice per group. Scale bars, 20 μm. DAPI,
   4′,6-diamidino-2-phenylindole. (G and H) Representative confocal images
   (G) and quantification (H) of CD31^+ (red) and Ki67^+ (green) cells in
   ischemic adductors. Arrowheads indicate proliferating ECs
   (CD31^+Ki67^+). n ≥ 5 mice per group. Scale bars, 20 μm. (I and J)
   Tarlov scores (I) and modified ischemic scores (J) of Apoa1bp^−/− mice
   and control littermates at the indicated time points following FAL.
   Scoring details are provided in table S6. n = 5 at each time point.
   Data are shown as means ± SEM. *P < 0.05; **P < 0.01; ***P < 0.001;
   ****P < 0.0001; n.s., not significant.

   However, by day 21 after FAL, AIBP knockout muscle demonstrated a
   significant increase in the volume of small-diameter collateral
   arteries (lumen: 20 to 100 μm), as evidenced by micro-CT imaging
   ([129]Fig. 4, C and D), whereas larger vessels (>100 μm) remained
   unchanged between the two groups. This increase in small collaterals
   was accompanied by greater arteriole density (SMA^+) and enhanced EC
   proliferation (Ki67^+) in the ischemic adductors of Apoa1bp^−/− mice
   ([130]Fig. 4, E to H). Correspondingly, AIBP knockout mice manifested
   clinical improvements, as indicated by improved modified ischemia
   scores and enhanced Tarlov scores ([131]Fig. 4, I and J). Although
   preexisting collaterals in the adductor muscles can undergo enlargement
   and remodeling to support flow recovery, we applied a canonical
   flow–elicited arteriogenesis model ([132]33) (fig. S5G) and observed no
   enlargement in the diameter and the volume of collaterals (fig. S5, H
   to J) and no increase in blood perfusion (fig. S5, K and L) as well as
   Tarlov and modified ischemia scores in Apoa1bp^−/− mice (fig. S5, M and
   N).

   These effects of Apoa1bp deficiency on collateral formation were also
   reproduced in female mice (fig. S6, A and B), which showed increased
   collateral number but not areas, as indicated by immunostaining of
   adductor cross sections (fig. S6, C and D). To confirm that
   extracellular AIBP regulates collateral vessel formation, we delivered
   intravenous recombinant AIBP to the Apoa1bp^−/− mice. This treatment
   abolished the enhanced perfusion recovery in Apoa1bp^−/− mice (fig. S7,
   A and B), reduced arteriole density (fig. S7, C and D), and suppressed
   EC proliferation (fig. S7, E and F). These results collectively suggest
   that extracellular AIBP limits collateral artery formation.

Loss of AIBP primers the expansion and arterial transition of stemlike CEC2

   To analyze the tissue distribution of CEC2 post-FAL, we performed CXCR4
   immunostaining in combination with detection of endomucin (EMCN)
   ([133]34), a capillary marker protein. We found that CXCR4 is
   predominantly enriched in CECs at day 7 postinjury, with a progressive
   decline by day 21 (fig. S8, A and B). Costaining for CXCR4 and SMA
   revealed a gradual increase in CXCR4^+ AECs over time, accompanied by
   rising SMA^+ arteriole density. Apoa1bp deficiency led to a marked
   increase in CXCR4^+ CECs at day 7, a difference that diminished by day
   21 (fig. S8, A and B). In contrast, Apoa1bp^−/− mice exhibited
   increased CXCR4^+ AECs at day 21, with no significant difference at day
   7 (fig. S8, C and D).

   We further investigated the EC subpopulations using flow cytometry
   analysis. Our flow cytometry gating is shown in fig. S8 (E and F). Stem
   cell antigen-1 (Sca-1), a marker of arterioles ([134]34), enabled
   distinction between CECs and AECs. On day 7, there were more
   CXCR4^+EMCN^+CD31^+ CECs and fewer CXCR4^+Sca-1^+CD31^+ arteriolar ECs,
   a pattern that reversed by day 21 in B6 mice (fig. S8G), indicating a
   dynamic shift in EC subpopulations over time. Accordingly, AIBP
   knockout increased the number of CXCR4^+ CECs at day 7 and arteriolar
   ECs at day 21 (fig. S8, H to J). Collectively, our data support a model
   where the stemlike CXCR4^+ CECs proliferate and convert to mature AECs,
   a process repressed by AIBP.

   To define the vascular identity of CXCR4^+ ECs in situ, we performed
   whole-mount immunostaining of the entire thigh muscle. Confocal imaging
   revealed that CXCR4 expression was primarily localized to the capillary
   network on day 7 post-FAL, with a transition toward mature arteries by
   day 21 in wild-type B6 mice (fig. S8K). As expected, Apoa1bp^−/− mice
   exhibited a significant up-regulation of CECs on day 7 and an increase
   in CXCR4^+ arterioles by day 21 post-FAL (fig. S8, L and M).
   Furthermore, we observed an increased number of proliferating CXCR4^+
   CECs in AIBP knockout mice (fig. S8, N and O). In sum, these data
   suggest that AIBP absence augments CC formation following ischemic
   injury.

   To determine whether the stemlike CECs augments CC, CXCR4^+ CECs were
   isolated from the ischemic hindlimbs of donor mice 7 days post-FAL and
   subsequently transplanted intravenously into recipient mice immediately
   following FAL. Assessment by micro-CT and immunofluorescence revealed a
   significantly increased density of collateral vessels in recipients of
   CXCR4^+ CECs relative to those receiving CXCR4^− CECs (fig. S9, A to
   D). Furthermore, laser Doppler imaging demonstrated accelerated
   restoration of hindlimb blood flow‌ in the CXCR4^+ CEC cohort (fig. S9,
   E and F). Concomitant with these improvements, both Tarlov locomotor
   scores and modified ischemia scores exhibited significant enhancement
   (fig. S9, G and H). Collectively, these findings demonstrate‌ that
   CXCR4^+ CECs promote neovascularization and functional recovery
   following ischemic injury.

CXCR4 signaling mediates the Apoa1bp deficiency effect on collateral
formation

   CXCR4 drives artery reassembly by promoting EC migration ([135]11). To
   determine the effect of AIBP on CXCR4 signaling in vivo, we delivered
   AMD3100, a specific inhibitor of CXCR4 ([136]35), into control and
   Apoa1bp^−/− mice. The resulting phenotypes mirrored those observed in
   mice receiving recombinant AIBP. Specifically, AMD3100 administration
   abolished the enhanced perfusion recovery seen in the Apoa1bp^−/− mice
   following FAL (fig. S10, A and B). Furthermore, AMD3100 treatment
   impaired collateral formation in Apoa1bp^−/− mice, as evidenced by the
   lack of CC with vessel diameter less than 100 μm (fig. S10, C and D), a
   reduction of SMA^+ arteries (fig. S10, E and F), and a decrease in EC
   proliferation (fig. S10, G and H). Mice treated with AMD3100 also
   demonstrated weakened physical activities compared to either control or
   AIBP null mice receiving saline (fig. S10, I and J).

   In vitro, AIBP and high-density lipoprotein (HDL) in combination, but
   neither alone, inhibited endothelial proliferation and migration (fig.
   S10, K to N). To further evaluate CXCR4 dependency, we knocked down
   CXCR4 in human umbilical cord endothelial cells (HUVECs) using CXCR4
   small interfering RNA (siRNA) (fig. S10O). Cotreatment with AMD3100 and
   CXCR4 siRNA did not produce additional inhibition of EC proliferation
   compared with either alone (fig. S10P; treatments 4 and 7 versus 8),
   suggesting that AMD3100 and siRNA converge on the same CXCR4-mediated
   pathway. Together, these results suggest that AIBP regulates
   endothelial behavior and ischemia-induced CC through CXCR4-dependent
   signaling.

AIBP down-regulates CXCR4 via enhancing endothelial uptake of HDL-associated
miR-223

   Among known regulators of CXCR4 expression ([137]36, [138]37), miR-223
   is notable for its association with HDL ([139]38). We found that
   miR-223 predominantly resides on HDL rather than non-HDL components
   (fig. S11A). In line with this, miR-223 abundance positively correlates
   with HDL levels (fig. S11B). Because our previous works established
   that AIBP enhances HDL binding to ECs ([140]16, [141]17), we
   hypothesized that AIBP facilitated cellular miR-223 delivery through
   HDL.

   To test this, we used both HUVECs and human microvascular endothelial
   cells (HMVECs), which align with the CEC subpopulation identified in
   our scRNA-seq analysis. Cells were incubated with Cy3-labeled HDL and
   Fam-labeled miR-223 in the presence or absence of AIBP. As expected,
   AIBP increased HDL uptake ([142]Fig. 5, A and B, and fig. S11, C and D)
   and significantly elevated intracellular miR-223 levels at 6 and 16
   hours ([143]Fig. 5C). In contrast, free miR-223 alone failed to enter
   ECs, even in the presence of AIBP (fig. S11, E and F), indicating that
   HDL association is essential for intracellular miR-223 delivery.
   Notably, AIBP selectively increased mature miR-223 without elevating
   pre–miR-223 ([144]Fig. 5D), and transcriptional inhibition with
   actinomycin D had no effect on mature miR-223 levels ([145]Fig. 5D and
   fig. S11G), suggesting that the rise stems from uptake rather than de
   novo synthesis. Furthermore, RNase treatment of HDL reduced the miR-223
   content ([146]Fig. 5E and fig. S11H), reinforcing the role of HDL as
   the delivery vehicle.

Fig. 5. HDL–miR-223 mediates AIBP effects on ECs.

   [147]Fig. 5.
   [148]Open in a new tab

   (A and B) Representative confocal images (A) and quantification (B) of
   HDL uptake in HUVECs. n = 6 independent repeats. Each data point
   represents the average fluorescence area for each coverslip, calculated
   by summing the total fluorescence from all cells in randomly selected
   fields on a coverslip and dividing by the total cell number. Scale bar,
   10 μm. (C) Quantification of miR-223 abundance. HUVECs were treated
   with AIBP, HDL, or their combination for the indicated durations. n = 3
   independent repeats. (D and E) Quantification of mature miR-223 and
   precursor miR-223 (pre–miR-223) levels following coincubation with AIBP
   and HDL, in the presence or absence of actinomycin D (ACTD; 5 μg/ml)
   and RNase (20 μg/ml) for 2 hours (E). n = 3 independent experiments.
   (F) qPCR analysis of CXCR4 mRNA levels in HUVECs transfected with
   either miR-negative control (NC) or miR-223 (left) and anti–miR-NC or
   anti–miR-223 (right). n = 3 independent experiments. (G and H) Western
   blot analysis of CXCR4 (G) and quantification (H). HUVECs were
   transfected with miR-223 for the indicated times. n = 3 independent
   repeats. FC, fold change. (I) Immunoblotting of CXCR4 (top) and
   quantification (bottom). HUVECs were transfected with control or
   miR-223 or coincubated with HDL and AIBP followed by control or
   anti–miR-223 transfection. n = 3 independent experiments. (J) qPCR
   analysis of miR-223 levels. CXCR4^+CD31^+CD45^− ECs were sorted from
   ischemic control or Apoa1bp^−/− adductors on day 7 after FAL. The
   sorted cells were lysed using the miRNeasy Micro Kit to extract total
   RNA. n = 3 independent repeats. Each sample pooled from four mice. Data
   are expressed as means ± SEM. *P < 0.05; **P < 0.01; ***P < 0.001;
   ****P < 0.0001; n.s., not significant; n.d., not determined.

   To determine whether miR-223 regulates CXCR4 expression in ECs, we
   performed gain-of-function and loss-of-function studies. Overexpression
   of miR-223 down-regulated CXCR4, whereas its inhibition up-regulated
   CXCR4 ([149]Fig. 5F). A brief 6-hour transfection of miR-223 alone
   significantly decreased CXCR4 protein levels ([150]Fig. 5, G and H).
   Similarly, incubating HUVECs with AIBP and HDL for 16 hours reduced
   CXCR4 expression (lanes 1 versus 2 in [151]Fig. 5I), and this effect
   was reversed by anti–miR-223 (lanes 3 versus 4 in [152]Fig. 5I). In
   vivo, CXCR4^+ ECs from AIBP knockout adductors exhibited reduced
   miR-223 levels ([153]Fig. 5J). The data validate endothelial CXCR4 as
   the direct target of miR-223 both in vitro and in vivo.

   To investigate the uptake mechanism, we blocked clathrin-dependent
   endocytosis with chlorpromazine (CPZ) ([154]39). CPZ lowered
   AIBP-mediated uptake of HDL and miR-223, as evidenced by their reduced
   fluorescence intensity (fig. S12, A and B), decreased intracellular
   mature miR-223 levels, and restored CXCR4 mRNA and protein level in
   HUVECs (lanes 3 versus 5 in fig. S12, C and D). In addition to
   clathrin, caveolae also mediates endocytosis of cell surface proteins,
   which can be disrupted by methyl-β-cyclodextrin (MβCD) ([155]40). In
   contrast to CPZ, MβCD treatment had no effect on miR-223 or CXCR4
   levels in HUVECs treated with AIBP and HDL (lanes 3 versus 7 in fig.
   S12, C and D). CPZ alone recapitulated MβCD and CPZ coadministration
   effect (lanes 5 versus 8 in fig. S12, C and D). These data suggest that
   clathrin-mediated endocytosis is critical for AIBP-driven HDL–miR-223
   internalization.

   Because CXCR4 localizes to lipid rafts ([156]41, [157]42), we examined
   whether AIBP-mediated raft disruption ([158]16, [159]43–[160]45) lowers
   CXCR4 levels. Treatment with AIBP and HDL for 16 hours reduced CXCR4
   levels in lipid rafts (lanes 1 versus 4 in fig. S13, A and B), but this
   effect was reversed when HDL was pretreated with RNase (lanes 4 versus
   5), suggesting that miR-223 underlies CXCR4 reduction. In contrast,
   shorter (4 hours) AIBP-HDL incubation also disrupted rafts, but
   eliminating miR-223 did not rescue CXCR4 levels (lanes 2 versus 3 in
   fig. S13, C and D). We further assessed lipid raft levels using
   fluorescently labeled CTB, which detects core raft lipid-GM1
   gangliosides. Our data show that miR-223 levels did not change lipid
   raft abundance (fig. S13, E and F), when compared with MβCD treatment
   (fig. S13G). These data suggest that AIBP-driven lipid raft disruption
   and miR-223–mediated CXCR4 suppression are distinct processes, with the
   latter depending on clathrin-mediated uptake of HDL-bound miR-223.

Lipoprotein receptor–related protein-2 is the endothelial receptor that
mediates AIBP function in CC

   To elucidate the mechanism by which AIBP regulates CC, we used TurboID
   proximity labeling to identify AIBP-interacting proteins (fig. S14, A
   and B). Flow cytometry analysis confirmed that AIBP was significantly
   enriched on the EC surface after transfection with AIBP-TurboID (fig.
   S14, C and D). Following AIBP-TurboID overexpression and subsequent
   optimization (fig. S14, E and F), biotinylated proteins were isolated
   and analyzed by mass spectrometry. A total of 684 membrane-associated
   proteins were detected, and KEGG pathway analysis revealed enrichment
   in pathways related to endocytosis and fatty acid metabolism (fig.
   S14G). We chose to focus on lipoprotein receptor–related protein-2
   (LRP2) (fig. S14H), given its established role in lipoprotein
   metabolism ([161]46).

   We validated the binding of AIBP to LRP2 using coimmunoprecipitation
   (Co-IP) ([162]Fig. 6A). To further confirm this interaction in situ, we
   performed proximity ligation assay (PLA), which detects protein-protein
   interactions via DNA probe-conjugated antibodies, producing fluorescent
   puncta when in close proximity. As expected, HMVECs incubated with
   recombinant AIBP exhibited significantly more PLA puncta compared to
   BSA-treated cells ([163]Fig. 6, B and C). Consistently,
   immunofluorescence staining showed colocalization of AIBP with LRP2 on
   the endothelial plasma membrane following recombinant AIBP treatment
   ([164]Fig. 6D). As anticipated, LRP2 deficiency attenuated
   AIBP-mediated disruption of lipid rafts (fig. S14, I to K, lanes 9
   versus 10). Together, these data identify LRP2 as the endothelial
   receptor that binds AIBP.

Fig. 6. LRP2 mediates AIBP binding and function.

   [165]Fig. 6.
   [166]Open in a new tab

   (A) Co-IP of AIBP-LRP2 interaction in HMVECs treated with recombinant
   AIBP. Lysates were immunoprecipitated with anti-LRP2 antibody and
   analyzed by Western blot as indicated. (B) Representative PLA confocal
   images showing AIBP-LRP2 binding. Scale bar, 10 μm. (C) Quantification
   of PLA puncta from (B); n = 3 mice. (D) Confocal images of AIBP (red),
   LRP2 (green), and CD31 (white) in HMVECs. n = 3 mice. Scale bar, 10 μm.
   (E) qPCR of Lrp2 mRNA in CD31^+CD45^− ECs from adductors of mice
   transduced with AAV9-shCtrl or AAV9-shLrp2. (F and G) Confocal images
   (F) and quantification (G) of LRP2 (green) and CD31 (red) in adductors
   following EC-specific AAV9-shCtrl or AAV9-shLrp2. n = 5 mice. Scale
   bar, 20 μm. (H and I) Micro-CT images (H) and vessel volume
   quantification (I) in ischemic limbs on day 21 post-FAL. Yellow
   arrowheads: collateral arteries. n = 5. Scale bar, 0.5 mm. (J and K)
   Flow cytometry (J) and quantification (K) of CXCR4^+EMCN^+ ECs in
   ischemic adductors. n = 3 samples (five mice pooled per sample). (L and
   M) qPCR of miR-223 (L) and Cxcr4 (M) in CD31^+CD45^− ECs from ischemic
   adductors (day 7 post-FAL). n = 8 to 12 mice per group. (N) Model: AIBP
   binds endothelial LRP2, triggering clathrin-mediated endocytosis and
   uptake of HDL–miR-223, leading to CXCR4 suppression. Means ± SEM;
   *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001.

LRP2 mediates the AIBP-facilitated HDL–miR-223–dependent down-regulation of
CXCR4

   Next, we assessed whether LRP2 mediates the effects of AIBP on
   HDL–miR-223 uptake. We knocked down LRP2 in HMVECs (fig. S15A) and
   assessed HDL-Cy3 uptake in the presence of AIBP. Our data showed that
   LRP2 knockdown significantly reduced cellular HDL uptake (fig. S15, B
   and C) and intracellular miR-223 levels (fig. S15, D and E). Notably,
   LRP2 deficiency rescued CXCR4 protein expression following incubation
   with AIBP and HDL (lanes 4 versus 5 in fig. S15, F and G), indicating
   that LRP2 is required for AIBP-HDL–induced CXCR4 suppression.

   To examine whether LRP2 mediates AIBP’s effects on CC, we generated an
   endothelial-specific adeno-associated virus (AAV)9–short hairpin RNA
   (shRNA) construct targeting endothelial LRP2 (AAV9-shLrp2), in which
   the Cdh5 promoter drives shLrp2 expression within an miR30 backbone
   ([167]47). Following intravenous delivery into B6 mice, AAV9-shLrp2,
   compared to control AAV9-shRNA, elicited EC-specific Lrp2 knockdown, as
   evidenced by qPCR analysis of Lrp2 in ECs isolated by
   fluorescence-activated cell sorting (FACS) ([168]Fig. 6E) and reduced
   vascular LRP2 protein in ischemic adductors ([169]Fig. 6, F and G), as
   confirmed by confocal microscopy analysis.

   Micro-CT analysis showed that endothelial LRP2 knockdown abrogated
   AIBP-mediated suppression of CC formation (lumen < 100 μm) ([170]Fig.
   6, H and I). Consistent with this, Lrp2 knockdown restored the density
   of SMA^+ arterioles (fig. S16, A and B) and number of Ki67^+ ECs in the
   ischemic adductors following recombinant AIBP delivery (fig. S16, C and
   D). Accordingly, mice receiving AAV9-shLrp2 exhibited improved blood
   perfusion recovery following FAL (fig. S16, E and F).

   Flow cytometry demonstrated that AAV9-shLrp2 attenuated the
   AIBP-induced reduction of CXCR4^+ CECs (CXCR4^+EMCN^+CD31^+CD45^−) at
   day 7 post-FAL, resulting in increased CXCR4^+ CEC abundance ([171]Fig.
   6, J and K), a finding corroborated by immunostaining (fig. S16, G and
   H). By contrast, no difference in the number of CXCR4^+ AECs
   (CXCR4^+Sca-1^+CD31^+CD45^− cells) was detected (fig. S16, I and J).
   Last, the AIBP-induced increase in miR-223 and reduction in CXCR4
   levels were abolished by LRP2 knockdown but remained intact in control
   mice ([172]Fig. 6, L and M). Collectively, these data suggest that AIBP
   binds to LRP2 to promote the uptake of HDL-associated miR-223, leading
   to the down-regulation of CXCR4 in CECs ([173]Fig. 6N).

Neutralization of extracellular AIBP by MM02 promotes collateral vessel
formation

   The above data led us to investigate the impact of neutralizing
   extracellular AIBP on collateral formation. To this end, we generated a
   panel of murine anti-AIBP monoclonal antibodies (mAbs). The antibodies
   #2, #5, and #6 selectively recognized endogenous AIBP in wild-type but
   not AIBP-deficient muscle tissue (fig. S17A). Among these, MM02
   uniquely disrupted AIBP binding to HUVECs (fig. S17B) and most
   effectively reversed AIBP-mediated suppression of HUVEC proliferation
   (fig. S17C), HDL uptake, and miR-223 delivery (fig. S17, D to F).
   Furthermore, preincubation with MM02 abolished the AIBP-mediated
   reduction of CXCR4 mRNA and protein in HMVECs compared to the control
   antibody (fig. S17, G to I).

   These findings established a functional role for extracellular AIBP in
   limiting CC formation. We next evaluated the therapeutic effect of MM02
   on CC. Administration of MM02, but not a control antibody,
   significantly increased collateral artery volume by day 21 following
   FAL ([174]Fig. 7, A and B). Immunohistochemical analyses revealed a
   higher density of SMA^+ arterioles in MM02-treated mice ([175]Fig. 7, C
   and D), accompanied by a marked increase in Ki67^+ ECs by day 7
   post-FAL ([176]Fig. 7, E and F). This proliferative response was
   paralleled by increased CXCR4^+ CECs in MM02-treated adductors at day 7
   (fig. S18, A and B) and a subsequent rise in CXCR4^+ AECs observed by
   day 21 (fig. S18, C and D), when compared with control antibody-treated
   mice.

Fig. 7. Systemic delivery of AIBP mAb MM02 enhances collateral vessel
formation post-FAL.

   [177]Fig. 7.
   [178]Open in a new tab

   (A) Representative Micro-CT images showing collateral arteries (yellow
   arrowheads) in B6 mice treated with control or MM02. Scale bar, 1 mm.
   (B) Quantification of total vascular volume (≤100 μm in diameter) in
   the ROI. n = 5 mice. (C and D) Confocal images (C) and quantification
   (D) of CD31^+SMA^+ vessels in ischemic adductors on day 21 post-FAL.
   n = 11 to 13 mice. Scale bar, 50 μm. (E and F) Confocal images (E) and
   quantification (F) of Ki67^+CD31^+ proliferating ECs at day 7 post-FAL
   after control or MM02 treatment. Arrowheads indicate double-positive
   cells. n = 5 to 8 mice. Scale bar, 20 μm. (G) qPCR of miR-223 in
   CXCR4^+CD31^+CD45^− ECs from adductors on day 7 post-FAL. n = 3 (each
   pooled from four mice). (H and I) Representative laser Doppler images
   (H) and perfusion ratios between injured and control limbs (I) on day
   21 post-FAL. n = 7 mice. (J and K) Tarlov (J) and ischemic scores (K)
   at indicated time points post-FAL. n = 7 mice. (L and M) Representative
   hematoxylin and eosin (H&E) images (L) and necrotic area quantification
   (M) in ischemic adductors on day 21 post-FAL. n = 6 to 10 mice.
   Means ± SEM; *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001; n.s.,
   not significant.

   Furthermore, analysis of CXCR4^+ ECs isolated on day 7 revealed a
   significant reduction in miR-223 levels in the MM02-treated adductors
   ([179]Fig. 7G). MM02 treatment accelerated perfusion recovery
   ([180]Fig. 7, H and I), alleviated tissue ischemia and impaired
   mobility ([181]Fig. 7, J and K), and mitigated necrosis ([182]Fig. 7, L
   and M). Consistent with its extracellular mechanism of action, MM02
   treatment significantly reduced plasma AIBP levels on day 7 and day 14
   post-FAL (fig. S18E). These results suggest that neutralizing
   extracellular AIBP promotes collateral vessel development and
   functional recovery following ischemic injury.

   Previous studies have established that capillary remodeling mediates CC
   formation via fluid shear stress-dependent mechanisms, including the
   up-regulation of shear-responsive genes. In AIBP-deficient mice
   subjected to FAL, we observed a marked up-regulation of the
   shear-responsive gene Klf2 at postoperative day 14 (fig. S19, A and B),
   implicating shear stress as a predominant driver of collateral vessel
   maturation. Our above work demonstrated that Apoa1bp deficiency
   promotes expansion of the CXCR4^+ stemlike CEC subpopulation in the
   early stage of ischemia. To assess whether the enhanced collateral
   formation observed in AIBP-deficient mice at later stages depends on
   sustained shear stress, we established a model of complete iliac artery
   ligation (IAL) (fig. S19C), which eliminates distal hindlimb perfusion
   and thereby abolishes flow-mediated cues. Notably, although AIBP
   knockout mice exhibited an expanded CXCR4^+ CEC population on
   postoperative day 7 (fig. S19, D and E), neither AIBP-deficient nor
   control mice formed mature collateral arteries by day 21 (fig. S19, F
   and G). Together, these findings suggest that persistent fluid shear
   stress is essential for collateral artery maturation in AIBP-deficient
   mice during the later phase of ischemia.

DISCUSSION

   In this study, we revealed the lipid metabolism pathway as a critical
   regulator of CC and identified previously unknown procollateral CEC
   type typical of stemness. All previously reported EC subtypes that
   contribute to arteriogenesis exhibit stemlike features, including both
   CECs and VECs. In contrast, LECs—primarily derived from VECs—show low
   stemness, comparable to AECs. Through complementary loss-of-function
   and gain-of-function approaches, we demonstrate that AIBP—a lipid
   metabolism protein up-regulated in the skeletal muscle of patients with
   PAD—acts as a negative regulator of CC formation. In the ischemic
   hindlimb, AIBP depletion enhances CC by promoting the expansion of
   CXCR4^+ CECs rather than opening preexisting collaterals.

   Our results support a two-phase model of CC formation (fig. S20): Phase
   1—sprouting phase: an initial stage during which CXCR4^+ stemlike CEC2
   subtype sprouts and establish interarterial angiogenic connections.
   This phase is enhanced by Apoa1bp deficiency. Phase 2—remodeling phase:
   a subsequent stage characterized by the progressive transition of
   interarterial capillaries into functional collateral arteries,
   involving the up-regulation of flow-responsive genes. During this
   process, the CEC2 subpopulation is converted to CEC1 and then to AECs.
   This trajectory is mechanistically distinct from the reassembly of
   fate-committed AECs previously proposed in the literature ([183]11).
   Notably, AIBP’s function is preferentially confined to the first phase
   as AIBP deficiency does not enhance CC in the classical arteriogenesis
   model. Therefore, the procollateralization effect of AIBP deficiency
   manifests specifically in the FAL model ([184]17), which uniquely
   couples angiogenesis with arteriogenesis, but is absent in models of
   classical arteriogenesis that rely preferentially on flow-induced
   remodeling ([185]33). Our observations align with reports supporting a
   role for CECs in arterial regeneration during ischemic tissue repair
   ([186]48).

   Although VEGF gene therapy has shown limited efficacy in clinical
   trials and is no longer actively pursued as a therapeutic strategy, our
   study identifies an alternative and potentially more targeted approach.
   By uncovering the AIBP–miR-223–CXCR4 regulatory axis that restricts
   capillary-to-artery transition, we propose a mechanism-driven strategy
   that may offer greater precision in enhancing collateral growth,
   particularly in patients with impaired arteriogenesis. We show that
   AIBP suppresses the sprouting phase of CC by facilitating HDL-mediated
   delivery of miR-223 via LRP2, which down-regulates CXCR4 expression and
   restricts CEC2 expansion. From an evolutionary perspective, AIBP-driven
   suppression of CC may serve to limit excessive vascular remodeling,
   which could otherwise predispose tissues to aberrant vessel formation
   and instability ([187]49–[188]52). Neutralization of AIBP using a mAb
   enhanced collateral network formation and accelerated perfusion
   recovery. Given that the antibody targets extracellular AIBP, it is
   expected to exert its effects without disrupting AIBP’s intracellular
   functions ([189]53, [190]54).

   In this study, we identified neutrophils and monocytes as the
   predominant sources of AIBP during the early phase of ischemic injury,
   and F4/80^+CD11b^+ macrophages also express AIBP, albeit at lower
   levels compared to neutrophils and monocytes. The temporally distinct
   expression patterns among myeloid subsets highlight the dynamic
   regulation of AIBP in response to ischemic stress and underscore its
   potential multifaceted role across different stages of the immune
   response.

   Our study reveals a moonlighting role for HDL as an miR-223 carrier in
   CC, distinct from its canonical role in lipid transport. This
   noncanonical function is amplified by myeloid cell–derived AIBP
   localized at CC sites ([191]55, [192]56), reinforcing the emerging view
   that myeloid-derived factors are critical orchestrators of
   extracellular signaling networks during CC. These myeloid cells likely
   coordinate CC establishment by modulating the extracellular
   environment. Although traditionally viewed as “good cholesterol”
   ([193]57, [194]58) emerging evidence suggests that HDL exhibits
   functional heterogeneity and may exert context-dependent effects on
   vascular biology. Our findings indicate that the HDL–miR-223 axis,
   particularly in the presence of elevated AIBP, may impair collateral
   vessel formation by facilitating miR-223 uptake into ECs. These results
   highlight the complex and potentially dualistic role of HDL in vascular
   remodeling and underscore the need for further investigation into HDL
   subtype-specific functions.

   In vascular development, arterialization is a distinct process that
   relies on NOTCH1-dependent cell cycle inhibition ([195]59), enabling
   the transition of proliferative ECs into a quiescent arterial fate.
   During the later stages of arterial growth and remodeling, CECs and
   VECs serve as a cellular reservoir ([196]22), contributing additional
   ECs to the expanding arterial network. We observed a similar transition
   in CC, where proliferative CEC2 give rise to less proliferative CEC1,
   consistent with the requirement for cell cycle inhibition during
   capillary arterialization. Our prior work showed that AIBP up-regulates
   the NOTCH1 pathway in the developing retinal vasculature ([197]17). In
   this study, AIBP depletion led to reduced Notch1 expression in ECs of
   ischemic tissue, although expression of canonical downstream targets
   (Hes1 and Hey1/2) remained unchanged, suggesting a context-dependent
   modulation of NOTCH signaling. Of note, bioinformatic analysis
   nominated transforming growth factor–β1 (TGF-β1) as a top upstream
   regulator of CEC2 gene expression. TGF-β1 may enhance NOTCH activity by
   up-regulating Ephrin-B2, NOTCH4, and HEY family members or by
   remodeling the extracellular matrix to facilitate arterial
   specification ([198]60–[199]62). Collectively, our findings define a
   two-stage model of collateral vessel formation that relies on the
   dynamic expansion and fate transition of CXCR4^+ CECs. Our studies
   reveal extracellular AIBP as a key repressor of this process and
   highlight its potential as a therapeutic target to augment
   revascularization in ischemic disease.

Limitation

   Although APOC4 and FABP5 were also identified in our initial screening,
   their functional roles in CC were not examined in depth; future work
   using different ischemic stages, genetic models, and assessments of
   cross-organ communication, expanded cohorts, extended time points, and
   genetic models is needed to clarify their potential contributions.
   Although MM02 demonstrates therapeutic potential, its clinical
   translation may be constrained by the requirement for repeated antibody
   administration. To overcome this, future efforts will explore Fc
   engineering, sustained-release platforms, and site-specific delivery
   strategies to enhance pharmacokinetics and patient adherence. In terms
   of outcome assessment, although the modified ischemia score offers
   clinical practicality, it remains semiquantitative and may not fully
   differentiate true tissue regeneration from superficial improvements
   due to necrotic tissue retraction or desiccation. Our data support a
   potential fate transition of stemlike CECs into AECs; however, this
   conclusion remains inferential in the absence of direct lineage-tracing
   evidence. Future studies incorporating genetic fate-mapping strategies
   specific to these endothelial subpopulations are necessary to
   rigorously define this trajectory. Although direct artery-to-artery
   collateral formation has been observed in preclinical models, current
   clinical imaging modalities lack the spatial resolution and molecular
   specificity to confirm this mechanism in patients. Emerging
   technologies—including single-cell transcriptomics, spatially resolved
   profiling, and AI-assisted imaging—may enable precise characterization
   of collateral vessel formation in human tissues. These innovations are
   poised to bridge the translational gap between mechanistic discovery
   and clinical application.

MATERIALS AND METHODS

Mass spectrometric analysis of the TurboID proteins

   Peptides were loaded onto a capillary and analyzed using the timsTOF
   Pro (Bruker Daltonics) mass spectrometer. The electrospray voltage was
   set to 1.60 kV. Precursors and fragments were detected with a scan
   range of 100 to 1700 m/z (mass/charge ratio) in the time-of-flight
   (TOF) detector. The timsTOF Pro operated in PASEF mode, selecting
   precursors with charge states of 0 to 5 for fragmentation, with 10
   PASEF-MS/MS scans acquired per cycle. Dynamic exclusion was set at 30 s
   ([200]63).

Bioinformatics analysis

   Functional enrichment. KEGG pathway enrichment analysis was performed
   using the KEGG database. Fisher’s exact test was applied to analyze
   KEGG pathway enrichment significance for differentially expressed
   proteins (using identified proteins as the background), with a
   significance threshold of P < 0.05 ([201]64).

Bulk RNA-seq analysis

   We analyzed the DEGs in the public bulk RNA-seq dataset ([202]15) using
   functional enrichment analysis as above described in the Bioinformatics
   analysis section.

Single-cell RNA-seq

Sample preparation

   We performed FAL surgery on four Apoa1bp^−/− mice and four control
   littermates. Ischemic adductor muscles were collected and stored in
   tissue storage solution (Miltenyi, 130-100-008) for subsequent RNA
   isolation and library preparation.

Quality control and library preparation

   scRNA-seq was conducted by GENE DENOVO (China). Ischemic adductor
   muscles were digested, and the resulting single-cell suspension was
   loaded into the 10X Genomics GemCode Single-cell instrument to generate
   Gel Bead-In-Emulsion (GEMs). cDNA libraries were prepared using
   Chromium Next GEM Single Cell 3′ Reagent Kits v3.1. Upon GEM
   dissolution, barcoded cDNAs were reverse transcribed from
   polyadenylated mRNA. Silane magnetic beads were used to purify the
   cDNA, which was then amplified for library construction. The final
   library was sequenced using Illumina paired-end sequencing, with the
   16–base pair (bp) 10x Barcode and 10-bp UMI encoded in Read 1 and cDNA
   fragments sequenced in Read 2. Sample index sequences were incorporated
   into the i7 index read.

EC cluster identification and marker gene analysis

   Raw BCL files were converted to FASTQ files, aligned, and quantified
   using 10X Genomics Cell Ranger software (v.3.1.0). Gene matrices from
   individual samples were imported into Scanpy (v.1.9.1) for downstream
   analysis. The integrated expression matrix was scaled, and dimensional
   reduction was performed using principal components analysis (PCA). The
   number of principal components used was determined by examining their
   contribution to the total variance, using a screen test to identify the
   most influential components. Seurat was used for graph-based
   clustering, with distances between cells calculated on the basis of
   identified principal components. A shared nearest-neighbor (SNN) graph
   was constructed on the basis of Euclidean distances in PCA space,
   refining edge weights using the Jaccard distance. Clusters were
   identified using the Louvain method to maximize modularity. Gene
   expression within each cluster was compared to all other cells using
   the Wilcoxon rank sum test. Up-regulated genes were defined by a
   1.28-fold increase, expression in more than 25% of cluster cells, and P
   value < 0.05. Genes were considered functionally related if involved in
   similar biological processes. Heatmaps were generated from
   cluster-averaged gene expression, row-wise z-score scaled, and
   visualized using the heatmaply R package (v.0.15.2).

Cell cycle pseudotime analysis

   Cell pseudotime trajectories were analyzed using Monocle 3, reducing
   data to one dimension. The cell fate trajectory was visualized in a
   treelike structure, including branches and tips, with parameters
   sigma = 0.001, lambda = NULL, param.gamma = 10, and tol = 0.001. The
   resulting cell fate trajectory was visualized in reduced dimensional
   space and represented as a treelike structure, including tips and
   branches, to reflect the progression of cell states ([203]65).

Pathway and transcription factor activity analysis

   Gene set variation analysis (GSVA) was used to convert the gene-by-cell
   matrix into a gene set–by–cell matrix. Transcription factor activity
   was analyzed using the SCENIC package (v.0.1.5), applying the 20,000
   motif database for RcisTarget and GRNboost analysis. A normalized
   expression matrix of 14,917 filtered genes served as the input.

Western blot analysis

   Western blotting was performed as previously described ([204]66). In
   brief, the adductor muscles were homogenized in protein lysis buffer
   (Sangon Biotech, catalog no. C500028-0010, China) on ice. HUVECs and
   HMVECs were lysed in Beyotime lysis buffer (catalog no. P0013B, China)
   with protease inhibitors. Protein concentration was measured using a
   BCA Assay Kit (Beyotime, catalog no. P0012, China). SDS–polyacrylamide
   gel electrophoresis (PAGE) and Western blotting were carried out
   according to standard protocols. PVDF (polyvinylidene difluoride)
   membranes were blocked in 5% skim milk in TBST [1X tris-buffered saline
   (TBS) with 0.05% Tween 20] and incubated with primary antibodies
   against AIBP (1:1000, homemade), CXCR4 (1:1000, Abcam, catalog no.
   124824), glyceraldehyde-3-phosphate dehydrogenase (GAPDH) (1:10,000,
   Abcam, catalog no.181602), CAV-1 (1:5000, Abcam), FLOT1 (1:200,
   sc-74566, Santa Cruz Bio), FLOT2 (1:1000, PTM-5369, PTM BIO), LRP2
   (1:1000, A24690, ABclonal), and anti-rabbit (1:10,000, Abcam, catalog
   no. 6721) or anti-mouse (1:10,000, Abcam, catalog no.6789) secondary
   HRP (horseradish peroxidase) antibodies. Signals were detected using
   the SuperSignal West Pico PLUS Chemiluminescent Substrate (Thermo
   Fisher Scientific).

AIBP binding assay

   The AIBP binding assay was performed as previously elaborated
   ([205]17). Recombinant human AIBP was first conjugated to
   Cy3–N-hydroxysuccinimide ester according to the manufacturer’s
   instructions (APEXBIO A8107, USA). The Cy3-conjugated AIBP was
   incubated with HUVECs on ice for 4 hours. The cells were subsequently
   washed twice with ice-cold phosphate-buffered saline (PBS), fixed with
   4% paraformaldehyde (PFA) for 10 min, and then washed twice with
   ice-cold PBS. The fluorescence intensity was measured using a
   microplate reader (Biotek ELx800), and data were expressed as relative
   intensity per well.

HDL isolation

   HDL and HDL3 were isolated as previously described ([206]67). Fresh
   plasma from healthy donors was obtained from Xiangya Hospital of
   Central South University. Plasma was adjusted to a density of 1.063
   kg/liter using KBr and subjected to ultracentrifugation at 35,000 rpm
   for 12 hours at 4°C. Crude HDL was isolated and further purified by
   adjusting to densities of 1.125 and 1.21 kg/liter with KBr, followed by
   ultracentrifugation. The HDL3 fraction was collected and dialyzed
   against PBS. Purity was confirmed with Sudan black staining and agarose
   gel electrophoresis. The purity of HDL and HDL3 was verified with
   Coomassie blue staining and agarose gel electrophoresis ([207]67).

Caveolar fraction isolation

   Caveolar fractions were isolated through a detergent-free,
   discontinuous gradient ultracentrifugation method as described before
   ([208]68). In brief, HUVECs were washed twice with ice-cold PBS and
   cells were scraped from the two D150 plates in 0.5 M sodium carbonate
   buffer (pH 11.0) containing protease inhibitor cocktail, homogenized,
   and sonicated on ice. Following protein quantification, protein
   extracts were mixed with MBS buffer [25 mM MES and 0.15 M NaCl (pH
   6.5)] with 45% sucrose (w/v) in the ultracentrifugation tubes, and
   overlaid with 35 and 5% sucrose. After ultracentrifugation at 35,000
   rpm for 20 hours at 4°C in an SW-41 rotor (Beckman), a discontinuous
   sucrose density gradient of 5 to 45% was generated. Eleven 1-ml
   fractions were collected from the top to the bottom. An equal amount of
   each faction was analyzed by SDS-PAGE using CXCR4 (Abcam, catalog
   no.124824) and Cav-1 (Abcam) antibodies.

TurboID mediated biotinylation

   Turbo-ID assay was performed as previously described ([209]69). A
   pLenti-AIBP-TurboID-HA plasmid was prepared for lentiviral packaging.
   HUVECs in the logarithmic growth phase were cultured in 10-cm dishes.
   When the cell density reached 70 to 80%, the viral concentrate carrying
   the AIBP-TurboID-HA gene was added. After 24 hours of viral infection,
   the medium was replaced with a fresh medium. Over the next 7 days, the
   medium was changed every 2 days using a G418-containing medium (300 to
   500 ng/ml) for selection, and cells were monitored daily. Cell lysates
   were collected for immunoblotting analysis to confirm successful
   transformation. Cells stably expressing AIBP-TurboID-HA were designated
   AIBP-HUVECs.

   For biotin labeling, AIBP-HUVECs were seeded into 6-well plates. Once
   they reached approximately 80% confluence, the medium was removed, and
   cells were washed twice with PBS and incubated with biotin-labeling
   buffer containing 500 μM biotin and 200 μM adenosine triphosphate
   (ATP). The medium was then replaced with complete medium containing the
   same biotin-labeling buffer. Cells were collected at the following time
   points: 0 and 10 min and 1, 6, and 12 hours. Protein lysates were
   analyzed via Western blot using a streptavidin-peroxidase antibody.

   For further experiments, AIBP-HUVECs were cultured in T75 flasks and
   divided into two groups: group I: cells cultured without additional
   treatment and group II: cells cultured in Dulbecco’s modified Eagle’s
   medium supplemented with 500 μM biotin and 200 μM ATP for 12 hours.
   Cells were collected, washed with cold PBS, and lysed in cold lysis
   buffer [8 M urea, 150 mM NaCl, 20 mM tris-HCl (pH 7.4), and 1% protease
   inhibitor cocktail] using cell scrapers. Lysates were sonicated and
   cleared by centrifugation (15,000g, 10 min, 4°C).

   For biotin pull-down, streptavidin magnetic beads were prewashed with
   radioimmunoprecipitation assay (RIPA) buffer (1 ml, 2 min) and the
   prepared protein lysates were incubated with the beads overnight at
   4°C. Beads were washed twice with RIPA buffer (1 ml, 2 min), once with
   1 M KCl (1 ml, 2 min), once with 0.1 M Na[2]CO[3] (1 ml, 10 s), once
   with 2 M urea in 10 mM tris-HCl (pH 8.0) (1 ml, 10 s), and twice more
   with RIPA buffer. Beads were not allowed to sit in Na[2]CO[3] or urea
   for extended periods to prevent aggregation. After the final wash, the
   beads were transferred to fresh tubes, washed twice with PBS, and
   stored at −20°C. Beads were then sent to OE Biotech Co. Ltd. (Shanghai,
   China) for mass spectrometry analysis.

Coimmunoprecipitation

   Co-IP assay was done as previously described ([210]70). Briefly, HUVECs
   were incubated with AIBP (200 ng/ml) or BSA for 2 hours on ice. Cells
   were then lysed in the lysis buffer containing 1% Triton X-100, 1%
   NP-40, 0.5% sodium deoxycholate, 150 mM NaCl, 10 mM Tris (pH 7.4), 1 mM
   EDTA, 50 mM NaF, and a protease inhibitor cocktail. The LRP2 antibody
   was cross-linked using an IP kit (Thermo Fisher Scientific, Cat: 88805)
   following the manufacturer’s protocol. The prepared protein lysates
   were incubated with the LRP2 antibody-bound protein A/G resin overnight
   at 4°C. After three washes with lysis buffer, the bead pellets were
   incubated with sample loading buffer and prepared for Western blot
   analysis to detect bound AIBP.

FAL model

   The surgery of FAL was performed in 9-month-old control or AIBP
   knockout mice as previously reported ([211]14) with minor changes.
   Briefly, surgical procedures were performed in mice under anesthesia
   and sterile conditions. Mice were injected with analgesic buprenorphine
   SR (0.15 mg/kg). Anesthesia was induced with 3% isoflurane and
   maintained with 1% isoflurane. Unilateral hindlimb ischemia was
   established by double ligation of the left femoral artery at both the
   proximal (near the inguinal ligament) and distal (just above the
   bifurcation of the saphenous and popliteal arteries) sites. The artery
   was transected between the two ligatures to ensure complete occlusion.

   All animal procedures adhered to the guidelines of National Institutes
   of Health for the care and use of laboratory animals and received
   approval from the Institutional Animal Care and Use Committee at
   Xiangya Hospital, Central South China University.

AAV9-ShLRP2cdh5 construction and administration

   rAAV9 were constructed by the vendor General Biology, and
   AAV9-Cdh5-Mir30-m-shRNA (Ctrl) and AAV9-Cdh5-Mir30-m-shRNA (LRP2) were
   used ([212]71). This design enables endothelium-specific LRP2
   knockdown. AAV9-ShLRP2cdh5 [1.0 × 10^11 vector genomes (vg)] or
   ShCtrl-AAV9cdh5 (1.0 × 10^11 vg) was injected via the tail vein. Two
   weeks later, ECs of the recipient mice were FACS sorted, and qPCR was
   performed to validate the Lrp2 knockdown efficiency.

Micro-CT angiography

   Micro-CT analysis of the hindlimb vasculature was performed as
   described previously with modifications ([213]72). Briefly, mice were
   injected with heparin (2000 U/kg), 5 min before euthanasia. After
   confirming cessation of cardiac activity, mice underwent cardiac
   perfusion with vasodilation solution [Dulbecco’s phosphate-buffered
   saline (DPBS) containing 100 μM adenosine, 10 μM sodium nitroprusside,
   and heparin (20 U/ml)] to remove blood from the circulatory system.
   Mice were subsequently perfused with 4% PFA followed by an injection of
   0.8-ml Microfil (MV-122, Flow Tech, Carver, MA) in the descending
   aorta. The specimens were stored in 4% PFA and kept at 4°C overnight to
   allow polymerization. Samples were scanned using a Bruker Micro-CT
   SkyScan 1176 system (Kontich, Belgium). Scan settings are as follows:
   voxel size, 6.534165 μm; medium resolution, 58 kV; 431 μA; 0.5-mm Al
   filter; and integration time, 1000 ms. Density measurements were
   calibrated to the manufacturer’s calcium hydroxyapatite (phantom).
   Analysis was performed using the manufacturer’s evaluation software.
   Reconstruction was carried out by NRecon (version 1.7.4.2). 3D images
   were obtained from contoured 2D images by methods on the basis of
   distance transformation of the grayscale original images (CT vox;
   version 3.3.0). 3D and 2D analyses were performed using the software CT
   Analyzer (version 1.18.8.0). 3D structural parameters, including vessel
   volume and vessel numbers, were analyzed.

Laser Doppler measurement of blood flow

   Blood flow was measured by scanning both rear paws with an LDI analyzer
   (Moor Infrared Laser Doppler Imager Instrument, Wilmington, Delaware)
   at designated time points: before and after the surgical procedure
   (days −1, 0, 3, 7, 14, and 21). During the procedure, animals were kept
   under 1% isoflurane anesthesia on a heating pad. The level of perfusion
   in both the ischemic and nonischemic hindlimbs was quantified using the
   mean pixel value within the region of interest, and the relative
   changes in hindlimb blood flow were expressed as the ratio of the left
   (ischemic) versus right (normal) laser Doppler–detected blood perfusion
   ([214]73).

Functional scoring

   Mice were examined preoperatively, immediately postsurgery following
   recovery from anesthesia, and at 1 and 2 weeks postoperation.
   Functional grading was performed according to the Tarlov scale and a
   modified ischemic scale designed to detect milder ischemia. Details of
   the functional and ischemic scoring criteria were elaborated in table
   S6 ([215]21).

Statistical analysis

   Statistical analyses for [216]Figs. 1 to [217]7 were performed using
   GraphPad Prism 10, and results are presented as means ± SEM.
   Significance was determined using appropriate tests on the basis of
   data structure: unpaired two-tailed Student’s t tests with Welch’s
   correction ([218]Figs. 1, E, H, and K; [219]3I; [220]4, B, F, and H;
   [221]5F; [222]6, C, E, and G; and [223]7, D, F, G, and M), one-way
   analysis of variance (ANOVA) followed by Tukey’s ([224]Figs. 1G and
   [225]6, I and K to M), Dunnett’s ([226]Fig. 2, E and G to J), or
   Sidak’s post hoc tests ([227]Fig. 5, B, H, and I), and two-way ANOVA
   followed by Tukey’s ([228]Fig. 2C), Bonferroni’s ([229]Fig. 4, D, I,
   and J), or Sidak’s post hoc tests ([230]Fig. 7, I to K). Nonparametric
   tests with Dunn’s post hoc analysis were used for multiple comparisons
   ([231]Fig. 1I), and the two-stage step-up method of Benjamini, Krieger,
   and Yekutieli was used to control the false discovery rate in [232]Fig.
   5 (C and J). Statistical significance is denoted as *P < 0.05,
   **P < 0.01, ***P < 0.001, ****P < 0.0001, with “n.s.” indicating no
   significance and “nd” indicating not determined.

   Other detailed experimental protocols and the source of reagents used
   in this study are available in the Supplementary Materials.

Use of artificial intelligence (ChatGPT)

   We used ChatGPT (GPT-4, Open AI) for the capillary stemness score
   analysis. We used the following prompts in our analysis: “You are an
   expert in vascular biology and single-cell analysis. Score ‘capillary
   endothelial cell stemness’ for each gene I provide. Use a stemness
   score (1–10), with 10 being the highest and 1 the lowest.”

Acknowledgments