Abstract

The vertebrate central nervous system is protected by the blood–brain barrier and meningeal membranes, which ensure immune privilege1. In the mammalian brain, microglia and barrier-associated or border-associated macrophages (BAMs) provide immune surveillance and scavenge wastes2, yet how evolution shaped immune-cell diversity and function is not understood. In zebrafish, a vascular-derived mural lymphatic endothelial cell (muLEC) lineage fulfils scavenger cell functions at central nervous system borders3,4,5. Here we identify the transcription factor odd-skipped related 2 (osr2) as a specific marker and regulator of muLEC differentiation and maintenance. osr2 controls the transition of muLECs from interconnected endothelial cells to individual scavenger cells in part by means of control of cadherin-6. muLECs are more transcriptionally similar to BAMs than to other mammalian meningeal cells and share several functions in tissue homeostasis. However, BAMs are absent from zebrafish and muLECs from mice and humans. Analysis of osr2, lymphatic endothelial cell (LEC) and BAM markers in diverse vertebrate species reveals muLECs as an ancient lineage and BAMs a recent mammalian specialization. muLECs and BAMs share functional analogies but are not homologous, providing an example of convergent evolution. This highlights the physiological importance of meningeal scavenger cells and the developmental plasticity of LECs in generating specialized cell types throughout evolution.

Main

The vertebrate blood–brain barrier and meningeal membranes serve as barriers that together control influx and efflux of cells, nutrients and metabolites at the central nervous system (CNS)1. As such, the CNS is immune privileged and patrolled by highly specialized immune cells2. Key innate immune cells in mammals are the parenchymal microglia and the border- (also known as barrier-) associated macrophages (BAMs)6,7. BAMs develop in the mammalian embryo from yolk sac-derived monocytic progenitors, are replenished throughout life by low level self-repopulation, and in disease settings by influx of circulating monocytes7. BAMs express a unique repertoire of scavenger receptors such as MRC1 (CD206), LYVE1, STAB1 and others that are involved in the phagocytosis of diverse ligands and tissue wastes7. They are localized to perivascular spaces at brain borders, the choroid plexus and the meninges, where they are thought to both clear wastes and regulate immune responses. BAMs have been shown to influence progression of Alzheimer’s and Parkinson’s disease phenotypes and are implicated in stroke and brain cancer8,9,10,11. Most of our understanding of immune-cell types in the CNS comes from studies in mammalian systems. Consequently, we know very little about the evolutionary diversification of CNS immunity at a cellular and functional level, in other parts of the phylogeny where immune privilege is intact.

The teleost fish Danio rerio (zebrafish) has a conserved blood–brain barrier and cell lineages of the neurovascular unit as well as a conserved innate immune system12,13. Whereas zebrafish macrophages, neutrophils and microglia have homologous roles to their mammalian counterparts13, BAMs have not been described. However, a new population of isolated muLECs (also known as brain LECs or fluorescent granular perithelial cells) was previously discovered within the CNS adjacent to the pial meningeal layer in an equivalent location to BAMs in mammals3,4,5. muLECs develop from the choroidal vascular plexus to form loop-like vascular structures on the dorsal surface of the zebrafish brain by 5–7 days postfertilization (dpf). Cells within these loop-like structures then disperse into individual cells as a sparse mural population covering the meningeal blood vascular network in juvenile and adult fish3. This developmental process is dependent on canonical drivers of lymphangiogenesis (vegfc,* flt4*,* ccbe1*, prox1a)3,4,5,14, glial cell function15 and independent of haematopoietic programs5. muLECs are a specialized lymphatic cell type with a distinctive mesenchymal morphology and established functions in scavenging macromolecular wastes into endocytic vesicles in a manner similar to scavenger endothelial cells (ECs)16,17. This scavenging function is dependent on Mrc1 (also known as CD206), Stab1, Stab2 and the endocytic adaptor protein Dab2 (refs. 16,18,19,20). So far, a lineage-specific marker and developmental program controlling muLEC differentiation has not been identified.

We examined the developmental transition of muLECs as they populate the zebrafish midbrain surface. At a gross morphological level, muLECs adjacent to the mesencephalic veins progressively disconnect from each other and lose their elongated EC appearance to become separated from one another and more rounded in morphology between 7 dpf and 30 dpf (Fig. 1a,b and Extended Data Fig. 1a). To identify the earliest transcriptional drivers of this change, we collated a single-cell RNA sequencing (scRNA-seq) time course for ECs, integrating datasets spanning key stages of muLEC development in the brain. This included 40 h postfertilization (hpf), 3 dpf, 4 dpf, 5 dpf and 7 dpf, encompassing sprouting of LECs from embryonic veins (40 hpf), emergence of muLECs at midbrain meninges (3–4 dpf) and the formation of muLEC loops (5–7 dpf) (Extended Data Fig. 1b–e). We annotated cell clusters on the basis of known marker genes (Extended Data Fig. 1f and Supplementary Data Table 1) and ECs (n = 15,898) were reclustered (Fig. 1c,d and Extended Data Fig. 1g–j). A putative muLEC cluster was identified as transcriptionally distinct from other EC clusters, including LECs. To investigate transcriptional differences between muLECs and LECs, we performed differentially expressed genes (DEG) analysis at 3 dpf and 7 dpf. Several genes marking LECs were strongly downregulated or absent in the muLEC cluster (for example, cadherin-6 (cdh6), tbx1)14, whereas others were reduced in expression (for example, prox1a). Conversely, genes associated with scavenger activity (for example, lyve1b, mrc1a) were upregulated (Fig. 1e–g and Supplementary Table 2). Notably, the transcription factor odd-skipped-related 2 (osr2), previously described for its role in the development of palate, synovial joint, kidney and bone formation in mammals21,22,23 and fin chondrogenesis in zebrafish24, was the most upregulated gene in the muLEC cluster at 7 dpf (Fig. 1e–g). osr2 expression initiated in ECs between 40 hpf and 3 dpf and was specifically restricted to the muLEC cluster at all timepoints, being the most cell type specific marker across the whole transcriptome (Fig. 1f,g and Extended Data Fig. 2a–c’). cdh6 and tbx1 were LEC-specific markers not expressed in muLECs (Fig. 1g, Supplementary Table 2 and Extended Data Fig. 2b,c”).

Fig. 1: osr2 is a specific marker of muLEC development.

a, Schematic of anatomical areas imaged. a′, EC nuclei (*Tg(fli1a:nGFP)y7, blue) and muLECs (Tg(−5.2lyve1b:DsRed)*nz101, orange) in 3-dpf, 7-dpf, 14-dpf and 30-dpf zebrafish right midbrains. b, muLEC length at mesencephalic vein scored as longest distance along any cellular axis from a′ (n = 12, n = 106, n = 10, n = 274, n = 123, n = 191 cells at stages indicated). c,d, UMAP showing EC scRNA-seq at different timepoints (n = 15,898 cells). Coloured by phenotype (c) and stage (d). e, Volcano plot of DEGs between muLECs and LECs at 3 dpf and 7 dpf (n = 220 (102 up, 118 down), and n = 211 (75 up, 136 down), respectively). Pink indicates DEGs (adjusted P < 0.05 and absolute log2[FC] > 1) and purple genes of interest. Two-sided Wilcox test, Bonferroni correction. f, UMAPs showing ECs by timepoint (40 hpf, 3d pf and 7 dpf), coloured by phenotype (upper) and osr2, mrc1a, lyve1b and prox1a expression (lower). Colour scale represents log-normalized expression. g, Dot plot showing marker gene expression in muLEC, VEC and LECs. Gene names and stages are indicated. Size of dots represents proportion of expressing cells, colour scale log-normalized expression. h–h′′, osr2 expression (Tg(osr2-gal4, UAS:mTagBFP2);Tg(UAS:RFP) knock-in, magenta) and lymphatics (Tg(−5.2lyve1b:DsRed)*nz101, cyan) at 7 dpf (h), and higher magnification of facial (h′) and trunk lymphatics (h′′). i–i′′, osr2 expression (magenta) and LECs and VECs (cyan) in a dorsal view of the 7-dpf brain (i) and magnification of muLEC loop adjacent to the mesencephalic vein (individual channels, i′ and **i′′*). Co-expression confirmed in more than ten animals. AEC, arterial EC; imVEC immature VEC; MV, mesencephalic vein; sEC, sparc+ EC. Images are confocal projections. Error bars represent s.e.m. Scale bars, 100 μm (a′,h,h′,h′′,i), 25 μm (i′′). Illustration in a created in BioRender. Usseglio gaudi, A. (2025) https://BioRender.com/w7lx2hd.

Source data

To validate osr2 as a definitive muLEC lineage marker, we generated a knock-in line. This identified expression in embryonic tissues previously reported to express osr2 (ref. 24) and expression colocalized with lyve1b in muLECs but not LECs at 7 dpf (Fig. 1h,i″ and Extended Data Fig. 2d). To further validate this, we took advantage of a published single-nuclei assay for transposase accessible chromatin sequencing (snATAC-seq) dataset14, and analysed differentially accessible chromatin peaks at the osr2 locus at 4 dpf. In muLECs, we discovered a putative enhancer in a distal region 45 kilobases (kb) (+45 osr2) downstream of the osr2 promoter. This region was cloned into a zebrafish enhancer detection vector and was sufficient to drive muLEC expression in a stable transgenic line (Extended Data Fig. 2e–g’). Taken together, these data confirm osr2 as a EC lineage-specific marker of muLECs.

To understand osr2 function in muLEC development, we generated a predicted loss-of-function mutant using CRISPR genome editing (Extended Data Fig. 3a). osr2 mutants and their wild-type siblings showed indistinguishable growth and morphology at all timepoints (Extended Data Fig. 3b). Mature muLECs are typically isolated, individual cells with limited cell-to-cell contact, unlike ECs in vessels. In mutants, muLEC number was unchanged (Extended Data Fig. 3c–f), but the characteristic transition to individual scavenger cells was abnormal (7 dpf, 14 dpf, 21 dpf and 30 dpf; Fig. 2a,a′ and Extended Data Fig. 3g–i’). muLEC length was increased in osr2 mutants compared with sibling controls at all 14 dpf, 21 dpf and 30 dpf, but unchanged at 7 dpf (Fig. 2b, Extended Data Fig. 3j). In addition, the percentage of muLECs engaged in cell-to-cell contact with other muLECs was reduced in osr2 mutants from 14 dpf up to 30 dpf (Fig. 2c). Together, these data show a role for osr2 in the morphological transition of muLECs from ECs to individual cells. In addition, we raised mutant and sibling animals to 1 year of age and assessed muLECs. Although mutant brains were morphologically normal, we observed a marked loss of muLECs (Fig. 2d,d′), identifying a requirement for osr2 in the maintenance of muLECs.

Fig. 2: osr2 regulates differentiation and maintenance of muLECs.

a,a′, muLECs (*Tg(−5.2lyve1b:DsRed)*nz101) of sibling (upper) and osr2 mutant (lower) at 30-dpf right midbrain (a) and mesencephalic vein (a′). b,c, Average muLEC length (b) and percentage of non-contacting muLECs (c) at mesencephalic vein in 7 dpf (n = 40, n = 23); 14 dpf (n = 14, n = 8); 21 dpf (n = 11, n = 5); 30 dpf (n = 10, n = 10) siblings (green) and mutants (blue), respectively. b, **P = 0.0023, *****P *< 0.0001. c, *****P *< 0.0001. d, muLECs in 1-year-old (yo) sibling (upper) and osr2 mutant (lower). d′, Brain width (upper) and total muLECs (lower) for d (n = 5 siblings, n = 6 mutants), ****P < 0.0001. e,e′, UMAP of scRNA-seq at 30 dpf (level 02, RNA_snn_res.0.2, n = 6,942) coloured by phenotype (e) and genotype (e′). f, Stacked bar plot showing genotype composition per cell type. g, UMAP coloured by mean mixing metric score. h,h′, UMAPs of sibling (h) and mutant osr2−/− (h′) ECs, coloured by phenotype (sibling n = 3,473 cells, mutant n = 3,543 cells). Arrows indicate RNA velocity. i,j, Stacked bar plot for DEGs (ECs, adjusted *P *< 0.05 and absolute log2[FC] > 0.5, i) and volcano plot for muLEC DEGs (n = 82 (55 up, 27 down, j). Pink, DEGs (adjusted *P *< 0.05 and absolute log2[FC] > 0.5). Purple, genes of interest. Two-sided Wilcox tests with Bonferroni correction. k, cdh6 expression (*Tg(−48cdh6:E1b:EGFP)*uom134, magenta) in muLECs (green) in siblings (top) and osr2 mutants (bottom). k′, Higher magnification. k′′, cdh6 mean intensity from k (n = 54 siblings, n = 30 mutants), *****P *< 0.0001. l, muLECs at mesencephalic vein (*Tg(−5.2lyve1b:DsRed)nz101, magenta) and blood vessels (*Tg(kdrl:EGFP)s843, green) in sibling (top) and cdh6 mutants (bottom). m,n, Average muLEC length (m) and percentage non-contacting muLECs (n) from l. 5 dpf n = 30, n = 13; 7 dpf n = 32, n = 11; 10 dpf n = 24, n = 13, for siblings and mutants, respectively. ***P < 0.0001. One-way ANOVA in b,c,m,n. Unpaired two-tailed t-test for d′, k′′. All images are confocal projections. Error bars represent s.e.m. Scale bars, 100 μm (a,d,k,l), 25 μm (a′,k′).

Source data

To further understand the role of osr2 in muLEC development, we performed scRNA-seq of ECs from siblings and osr2 mutants at 30 dpf (Extended Data Fig. 3k–m). We annotated cell clusters on the basis of known marker genes (Extended Data Fig. 3n and Supplementary Data Table 1) and ECs (n = 6,942) were reclustered (Fig. 2e,f and Extended Data Fig. 3o,p). On clustering, we observed that although mutant and sibling muLECs were both present, the mixing of cells in the muLEC cluster was heterogeneous (Fig. 2g and Extended Data Fig. 3q). Separate clustering of sibling and mutant datasets coupled with velocyto trajectory analysis revealed a notable defect in the differentiation trajectory of osr2 mutant compared with sibling control muLECs. Mutant muLEC transcriptomes were closely associated with LEC transcriptomes with notable mixing of the lineages in velocyto trajectory analysis, whereas wild-type muLECs had transitioned to a distinctly separate cell type (Fig. 2h,h′ and Extended Data Fig. 4a–c). Furthermore, analysis of Slingshot pseudotime differentiation trajectories confirmed a reduction in differentiation of mutant muLECs (Extended Data Fig. 4e,f). A DEG analysis comparing mutant and sibling muLECs identified 82 DEGs (27 downregulated, 55 upregulated) (Fig. 2i and Supplementary Data Table 3). Changes in gene expression included the dysregulation of angpt2a and its receptor tie1 (ref. 25) (involved in EC adhesion and lymphatic development) and the increased expression of some cell–cell adhesion molecules in muLECs (Fig. 2j and Extended Data Fig. 4d). This included expression of both cldn5b and cdh6, normally high in wild-type LECs, but lost in differentiated wild-type muLECs. Both genes were highly expressed in osr2 mutant muLECs that had failed to fully differentiate (Extended Data Fig. 4g). As this data is consistent with a defect in differentiation and muLECs are active in scavenging meningeal wastes, we assessed muLEC scavenging in osr2 mutants. We found that mutants show reduced scavenging capacity as early as 7 dpf, which is consistent with abnormal differentiation (Extended Data Fig. 4h–j). Overall, this suggests that osr2 mutant muLECs still form, but do not differentiate as well as their wild-type counterparts, leading to morphological and functional deficits in the lineage and a failure in maintenance over time.

cdh6 encodes a cadherin predicted to control cell–cell adhesion and expressed specifically in LECs in zebrafish, whereas cdh5 is only found in blood vessels in this model system14. Notably, the developmental transition to individual muLECs is concomitant with the progressive downregulation of cdh6 (Extended Data Fig. 2b,c”). Using a cdh6 enhancer reporter transgenic line, we confirmed that cdh6 expression labels all LECs but is downregulated during muLEC development (Extended Data Fig. 4k–l’ and Extended Data Fig. 2b,c”). Crossing this line onto the osr2 mutant, we confirmed an increase in cdh6 expression in muLECs lacking osr2 (Fig. 2k–k′′). To determine whether loss of cdh6 function might contribute to the morphological transition from EC to muLEC, we generated a CRISPR mutant for cdh6 (Methods). We found that mutants lacking cdh6 showed muLECs with more rounded and isolated morphologies at 5 dpf, 7 dpf and 10 dpf, whereas there was no gross change to lymphatic vessel morphology in these animals (Fig. 2l–n and Extended Data Fig. 4m). This is consistent with an earlier morphological transition from EC towards individual scavenger cells occurring in cdh6 mutants than in wild-type animals. Thus, downregulation of cdh6 and cell–cell adhesion is probably an important aspect of the osr2-regulated morphological changes observed during development of the lineage.

The mammalian CNS is devoid of LECs, with the presence of lymphatic vessels only seen at the dura in close proximity to the skull26. Although rare cells expressing some LEC markers have been observed in human histological analyses27, a large population resembling zebrafish muLECs has never been reported. To hunt for a mammalian muLEC population in an unbiased manner, we isolated cells from mouse meninges for scRNA-seq on the basis of their ability to scavenge Alexa Fluor 488-labelled acetylated-LDL (488acLDL). Previous work has shown that 488acLDL is taken up by zebrafish muLECs3 and this moiety would be expected to be taken up broadly by cells with scavenging ability. We finely dissected the leptomeninges and some associated parenchyma from the mouse cortex, dissociated cells, cultured for 1 h in 488acLDL and fluorescence-activated cell-sorted all cells that had taken up the fluorophore. These cells were then sequenced to profile meningeal cells with scavenging capability (Fig. 3a and Extended Data Fig. 5a,b). We assigned clusters based on expression of known marker genes (Extended Data Fig. 5c and Supplementary Data Table 1). Large populations of ECs and immune cells were identified as well as smaller populations of glia and oligodendrocytes (Fig. 3a). We captured parenchymal microglia and two distinct macrophage populations, perivascular macrophages and BAMs on the basis of published profiles28. Despite isolating *n *= 16,262 cells capable of scavenging 488acLDL, we did not identify any cells expressing LEC markers or Osr2 that would be homologous to zebrafish muLECs. We further interrogated the publicly available Allen Brain Cancer Cell Atlas dataset29, which captured more than 4 million cells, including approximately 230,000 immune and vascular cells, and found no evidence of muLECs in the mouse brain (Extended Data Fig. 5d). To explore scavenging cell populations in human leptomeninges, we took advantage of two previously published human scRNA-seq datasets that included leptomeningeal cells30,31 (Fig. 3b and Extended Data Fig. 5e–i). We assigned clusters on the basis of expression of known marker genes and we investigated the expression of OSR2, LEC and muLEC marker genes across cell clusters (Extended Data Fig. 5j–l). We found that human meninges were also devoid of muLECs and, thus, muLECs appear to be absent from mammals.

Fig. 3: muLECs are absent from mammals but similar to BAMs.

a, UMAP of scRNA-seq of dissected mouse cortex scavenging cells (n = 16,262 cells) coloured by phenotype (right). b, UMAP of selection of cells from GSE2453111 (ref. 30) (23 weeks postconception human leptomeningeal cells). Level 02 (n = 10,986 cells) coloured by phenotype. c–e, Heatmap of spearman correlation coefficients (left) and Jaccard similarity (right) between indicated species: mouse versus zebrafish (c); human versus zebrafish (d) and human versus mouse (e). Sorted by Jaccard similarity (high to low). Colour represents positive correlation coefficient and Jaccard similarity, respectively. Those that did not pass significance testing are denoted with an x. f, Dot plot showing expression for lymphatic, scavenging, proteolytic and haematopoietic marker genes for zebrafish, mouse and human (data from refs. 30,31). Size of dots represents proportion of expressing cells, colour represents average log-normalized expression. g, Meningeal blood vessels (Tg(kdrl:EGFP)s843, green) in control (upper) and ccbe1 MO (lower) injected animals. muLECs (Tg(−5.2lyve1b:DsRed)nz101, magenta) shown in the inset. h, Variance of vascular density in control and ccbe1 MO. Two-sided Levene’s test ***P = 0.0005 (n = 7 control, n = 13 ccbe1 MO). i, Neutrophils (*Tg(lyz:BFP)zf217Tg, green) in control (upper), *flt4um203 mutant (middle) and ccbe1 MO (lower) 4-mpf brains. muLECs inset. j, Neutrophil numbers in control, flt4 mutant and ccbe1 MO (n = 22 control, n = 19 flt4 mutant, n = 9 ccbe1 MO). k, Microglia and/or macrophages (*Tg(mpeg1.1:TagBFP)bcz53Tg, green) in control (upper), *flt4um203 mutant (middle) and ccbe1 MO (lower) 4-mpf brains. muLECs in inset. l, Microglia and/or macrophage numbers in control, flt4 mutant and ccbe1 MO (n = 8 control, n = 6 flt4 mutant and n = 10 ccbe1 MO brains). Unpaired two-tailed t-test **P = 0.0033* for* control versus flt4 mutants, P = 0.0017 for control versus ccbe1 MO in j. *P = 0.0251 for control versus flt4 mutants, **P = 0.0092 for control versus ccbe1 MO in l. NK cells, natural killer cells; MG, microglia; Prolif. BAM, proliferative BAM; PVM, perivascular macrophage; VSMC, vascular smooth EC; wpc, weeks postconception. Scale bars, 100 μm. All images confocal projections. Error bars represent s.e.m. Illustrations in a and b created in BioRender. Usseglio gaudi, A. (2025) https://BioRender.com/w7lx2hd.

Source data

To more deeply compare scavenging cells across species and to identify transcriptionally similar cell types in an unbiased manner, we compared our mouse and human datasets with the muLEC transcriptome from zebrafish. This approach used two independent methods (a correlation coefficient and Jaccard similarity score) correlating gene expression as well as the most specific markers of cell types across species. muLECs showed the highest correlation coefficient and Jaccard similarity score with the mouse and human BAM clusters (Fig. 3c–e, Extended Data Fig. 6a–e’ and Supplementary Data Table 4). Detailed analysis of shared gene expression between zebrafish muLECs and mouse BAMs revealed expression of scavenger receptors (for example, homologues of STAB1, MRC1,* LYVE1*) and factors involved in protein turnover (homologues of CTSD, CTSH, CTSZ, CTSLA, LGMN and LAMTOR4) (Fig. 3f). muLEC specific genes were lymphatic marker genes (for example, homologues of FLT4, PROX1) and BAM-specific genes were associated with the complement system (C1QA,* C1QC*), known BAM-specific markers (homologues of MS4A7) and well-known haematopoietic marker genes32 (Fig. 3f, Extended Data Fig. 6f,g and Supplementary Data Table 4). Expression of immunomodulatory genes by BAMs but not muLECs suggests specialized roles for BAMs. The development of muLECs at the zebrafish meninges has been demonstrated to be controlled by glial Vegfc production and meningeal fibroblast-produced Ccbe1 (ref. 15). We saw expression of these lymphangiogenic factors in the zebrafish but not mammalian meninges, potentially explaining the species specificity of the lineage (Extended Data Fig. 6h–k’). These data together indicate that muLECs and BAMs are distinct cell types that share transcriptional similarities attributable in large part to genes involved in scavenging of extracellular wastes.

In zebrafish, muLECs can clear tissue wastes and change morphology on injury33 reminiscent of macrophages, but muLECs are generated by lymphangiogenesis and do not develop through haematopoiesis5. To better understand the nature of muLECs and their activation on injury, we generated a scRNA-seq dataset that compared injured versus uninjured larval heads and brains at 24 h postinjury in 7 dpf larvae (Methods and Extended Data Fig. 7a–c). The injury assay led to local accumulation of macrophages, changes in muLEC morphology as previously reported34 and broadly increased lysosomal activity at the wound (Extended Data Fig. 7d,e). The dataset captured a large number of cell types and we correlated gene expression between muLECs and all other cell types (Extended Data Fig. 7f). Apart from high similarity to LECs (consistent with previous work3,5), the second highest correlation score was with a cluster of macrophages or microglia (Extended Data Fig. 7g). Analysis of individual marker genes revealed that muLECs and macrophages and/or microglia both expressed a selection of cathepsins (ctsd,* ctsh*,* ctsz*) involved in phagocytic functions35,36,37 (Extended Data Fig. 7h,i and Supplementary Data Table 5), but only muLECs expressed the scavenger receptors typical of mammalian BAMs (stab2,* lyve1b*, mrc1a)7. Zebrafish macrophages and/or microglia but not muLECs showed expression of myeloid lineage markers (mpeg1.1, csf1ra,* spi1a*,* mfap4*) (Extended Data Fig. 7j) and we found no evidence of BAMs present in the zebrafish brain in our own or published datasets38 (Extended Data Fig. 7k,l). Next, we examined the differences seen on wounding. We noticed that the muLEC cluster showed heterogeneous mixing of control and wounded cells and quantification of mixing metric suggested muLECs undergo a substantial transcriptional change on wounding (Extended Data Fig. 8a–d). DEG analysis revealed that muLECs had the highest number of DEGs in response to injury of any cell type (Extended Data Fig. 8e). Gene ontology (GO) terms associated with upregulated genes in muLECs following injury related to endocytosis, proteolysis and metabolic regulation (Extended Data Fig. 8f,g). Upregulated were scavenging cell markers (stab1,* stab2*,* ctsla*, ctsh, lgmn,* lamtor4*) and genes related to proteolysis, peptidase and lysosomal activity (Extended Data Fig. 8h). Taken altogether, these data indicated that muLECs should be considered ‘macrophage-like’ on the basis of anatomical, transcriptional and functional analogy. As a functional test, we delivered clodronate liposomes to 6 dpf zebrafish embryos by angiogram. Then 24 h postinjection, both macrophages and muLECs were significantly depleted (Extended Data Fig. 8i,j). In addition, as muLECs change morphology on wounding, we tested whether this change is dependent on normal differentiation of the lineage. We examined osr2 mutants following wounding and found that mutant muLECs did not undergo the expected morphological transition (Extended Data Fig. 8k–l). Altogether, muLECs differentiate into macrophage-like cells that are activated on injury and are transcriptionally reminiscent of mammalian BAMs, but the zebrafish brain lacks a true BAM population.

As well as scavenging wastes, other physiological functions attributed to BAMs include signalling to endothelium and modulating local immune responses (for a review, see refs. 7,28,39). muLECs can produce Vegfs and modulate meningeal angiogenesis early in development3, but their influence over mature vasculature and immune-cell populations has not been studied, whereas some immune-cell types (such as microglia) share related functions18,40. We found that injection of optimized doses of a morpholino oligomer (MO) depleting Ccbe1 (ref. [41](https://www.nature.com/art