Introduction

The vertebrate brain uses varied strategies to clear extracellular material such as extracellular matrix and aggregation-prone proteins^1–3. Engulfment by specialized cells^4–7 – including macrophages such as microglia^8–16 – permits degradation and clearance of extracellular proteins and other macromolecules in the brain. Disruptions in clearance can lead to dysregulated accumulation of extracellular material in the brain interstitial space and cerebrospinal fluid (CSF), with detrimental consequences during aging and in neurodegenerative diseases^17–30. Thus, clearance mechanisms could represent key strategies for maintaining brain homeostasis, with promising therapeutic potential to counter brain aging and neurodegenerative diseases^31–37. Studying clearance strategies in varied species could reveal mechanisms not present in mammals and help determine how they could be used to counter brain aging.

An interesting model system in which to study clearance strategies is the African turquoise killifish Nothobranchius furzeri – the shortest-lived vertebrate that can be bred in captivity³⁸. The killifish has a median lifespan of only six months, 5 times shorter than the mouse³⁸, enabling scalable studies of processes that are disrupted with age^39–59. The killifish also recapitulates many conserved hallmarks of brain aging, including neurodegeneration^60–62, cognitive decline^63–65, and behavioral deficits⁶⁶. However, waste clearance in the brain has not been examined in this species. A short-lived vertebrate model in which to study clearance would enable longitudinal studies as well as the identification of interventions that boost clearance in the aged brain.

Here, we present a genetic model for studying clearance of a secreted fluorescent protein in the brain of the killifish. Using this genetic model, we show that a population of macrophages in the killifish brain is responsible for engulfment of diverse extracellular molecules. Intriguingly, we find that many killifish brain macrophages exhibit similarities to mammalian border-associated macrophages (BAMs) and monocyte-derived macrophages (MDMs), which are rare cell types in the adult mammalian brain. We also find that in our model, the engulfment capacity of killifish brain macrophages declines with age, which could contribute to clearance dysregulation and abnormal accumulation of extracellular material in brain aging.

A model for engulfment by vertebrate brain myeloid cells

To model clearance of extracellular proteins in the brain, we engineered a killifish strain using a CRISPR knockin method developed for killifish⁶⁷. In this strain, a secreted form of the red fluorescent protein oScarlet is expressed from one allele of the ELAVL3 locus, a vertebrate-conserved gene predominantly expressed in neurons⁶⁸ (Fig. 1a). We chose oScarlet for visualizing clearance processes due the relatively high stability of mScarlet-derived red fluorescent proteins^69,70. Importantly, to allow the secretion of oScarlet into the interstitial space, we added a signal peptide (SP, derived from the killifish HSPA5 protein⁷¹) to the N-terminal region of oScarlet. We also added a P2A self-cleaving peptide to prevent fusion of oScarlet to the ELAVL3 protein⁷² (Fig. 1a). We maintained the killifish line at the heterozygous state and refer to it as SP-oScarlet.

Figure 1.

We confirmed insertion of the SP-oScarlet knockin construct at the ELAVL3 locus by PCR and long-read sequencing of the entire locus, which confirmed appropriate junctions with the ELAVL3 gene and the presence of the SP-oScarlet insertion (Fig. 1 – Fig. Supplement 1a). We noted the presence of two tandem copies of the SP-oScarlet insertion, although the presence of a stop codon at the C-terminus of oScarlet likely leads to the production of a single oScarlet protein (Fig. 1 – Fig. Supplement 1a).

We verified that ELAVL3 and oScarlet transcripts were expressed in the same cells of SP-oScarlet fish, as expected, using in situ hybridization chain reaction (HCR)⁷³ on adult brain sections (Fig. 1 – Fig. Supplement 1b-c). By immunofluorescence staining for oScarlet, we also observed bright, punctate oScarlet signal in adult SP-oScarlet brains (Fig. 1b, Fig. 1 – Fig. Supplement 1d). Using HCR on brain sections, we verified the presence of cells with bright oScarlet protein signal (oScarlet^HIGH) that did not express oScarlet transcripts in the brains of SP-oScarlet fish (Fig. 1c, Fig. 1 – Fig. Supplement 1d). This suggests that the oScarlet protein is efficiently secreted from the neurons that express the oScarlet transcript, resulting in oScarlet protein that is likely engulfed by other cell types in the brain. Thus, the bright oScarlet puncta in the brains of SP-oScarlet fish can be found within cells that likely engulf, and do not produce, this protein.

To identify the cells that engulf oScarlet, we developed a flow cytometry strategy to distinguish cells with bright oScarlet protein signal (oScarlet^HIGH) from other cells in the adult brain (oScarlet^LOW) (Fig. 1d-e, Fig. 1 – Fig. Supplement 1e). We isolated oScarlet^HIGH cells from adult SP-oScarlet brains (Fig. 1e), and we verified that these cells were not present in the brains of wildtype killifish, even at older ages where more autofluorescent cells might be expected (Fig. 1 – Fig. Supplement 1f).

To unbiasedly characterize oScarlet^HIGH cells, we performed single-cell RNA sequencing on enriched oScarlet^HIGH cells FACS-isolated from the brains of young adult (49 days) and old (134 days) male SP-oScarlet killifish (n = 4 fish per age group, pooled), as well as oScarlet^LOW cells isolated from the young adult fish. We used transcription and translation inhibitors to minimize gene expression changes due to dissociation, as done in studies of mammalian brain macrophages^74–76. We performed Uniform Manifold Approximation and Projection (UMAP)⁷⁷ analysis of the single-cell RNA sequencing data from young adult animals (for analysis of the old dataset, see Fig. 4). This analysis revealed that most oScarlet^HIGH cells were transcriptionally distinct from their oScarlet^LOW counterparts (Fig. 1f).

By manually annotating cell types based on matching top UMAP cluster marker genes (Fig. 1 – Fig. Supplement 2a-b) to cell type markers known from literature in mammals, zebrafish, and killifish^68,78–80, we identified the vast majority (99.6%) of oScarlet^HIGH cells as myeloid immune cells (Fig. 1g).

oScarlet^LOW cells included neurons, oligodendroglia, radial glia and progenitors, pericytes, and T cells (Fig. 1g). We noted that endothelial cells were not well represented in our dataset, possibly due to our use of dissociation and FACS in our workflow. Only a very small percentage (0.5%) of oScarlet^LOW cells were myeloid cells, indicating that most myeloid cells likely engulfed oScarlet (Fig. 1g).

We could confirm by HCR on brain sections in situ that oScarlet^HIGH cells expressed a homolog of the conserved myeloid marker CSF1R^81,82 (Fig. 1h). We also verified that oScarlet^HIGH cells largely did not express ELAVL3, the driver for oScarlet expression (Fig. 1i).

Collectively, these data show that most oScarlet^HIGH cells are myeloid cells that have likely engulfed secreted oScarlet protein in the brains of SP-oScarlet fish. Thus, we have generated a genetic model for chronic engulfment by brain myeloid cells in a short-lived vertebrate.

The killifish brain includes a population of macrophages with transcriptional similarities to mammalian monocyte-derived macrophages

What type of myeloid cells engulf oScarlet in the killifish brain? Analysis of our single-cell RNA sequencing data indicated that oScarlet^HIGH cells are likely macrophages. Indeed, in addition to CSF1R, oScarlet^HIGH cells were enriched for APOEB (a homolog of human APOE) as well as antigen presentation-associated (e.g., HLA-DPB1, CD74), complement-associated (e.g., C1QC, C1QA), and scavenger receptor-encoding (e.g., MRC1, MARCO) genes (Fig. 2a). Moreover, by analyzing gene signatures from our single cell RNA sequencing dataset, we found that oScarlet^HIGH cells were enriched for signatures characteristic of macrophage functions, such as endocytosis-related and lysosomal genes (Fig. 2b). We also observed punctate oScarlet protein signal in proximity to late endosomes and lysosomes (marked by the protein marker LAMP1) by immunofluorescence staining (Fig. 2 – Fig. Supplement 1a), consistent with endocytosis and attempted degradation of extracellular material by macrophages. mScarlet, from which oScarlet is derived⁷⁰, is known to be relatively stable at the acidic lysosomal pH⁶⁹, which could explain the persistence of oScarlet signal in cells that engulf the protein.

Figure 2.

In the mammalian brain, most macrophages fall into one of three main classes^75,83: parenchymal microglia (derived from the embryonic yolk sac), and two classes of macrophages found at interfaces between the brain and periphery (blood vessels, meninges, and choroid plexus) – border-associated macrophages (BAMs) (also derived from the embryonic yolk sac), and monocyte-derived macrophages (MDMs) (of peripheral, hematopoietic origin)^84–93. BAMs and MDMs can strongly engulf substrates from the CSF¹⁰, with BAMs noted for their exceptional endocytic capacity⁷⁵.

Using principal component analysis (PCA) with marker genes, mouse brain macrophages from a dataset from Barr et al.⁷⁵ could be clearly transcriptionally separated into microglia, BAMs, and MDMs (Fig. 2c, Fig. 2 – Fig. Supplement 1b). In contrast, using principal component analysis with killifish homologs of the same marker genes, oScarlet^HIGH cells from SP-oScarlet killifish could not be transcriptionally separated into these classes (Fig. 2c, Fig. 2 – Fig. Supplement 1b). Consistent with this finding, wildtype killifish forebrain macrophages (i.e., cells from an APOEB+ IBA1+ LCP1+ cluster, originally annotated as “Microglia”) from an independent dataset from Ayana et al.⁶⁸ could not be transcriptionally separated into microglia, BAMs, and MDMs in the same analysis (Fig. 2c, Fig. 2 – Fig. Supplement 1b). Thus, we did not find evidence for the existence of transcriptionally distinct microglia, BAMs, and MDMs in killifish brains, possibly suggesting a more plastic population of brain macrophages in the killifish.

Killifish homologs of many of the strongest markers of mouse microglia, BAMs, and MDMs were not detected in killifish brain macrophages, while some killifish brain macrophage markers (e.g., CLDN5) were not detected in mouse brain macrophages (Fig. 2a, Fig. 2d, Fig. 2 – Fig. Supplement 2a, Fig. 2 – Fig. Supplement 3a, Fig. 2 – Fig. Supplement 4a).

Nonetheless, the expression profile of oScarlet^HIGH cells overlapped most with that of mouse MDMs (Fig. 2d, Fig. 2 – Fig. Supplement 2a, Fig. 2 – Fig. Supplement 3a). Similar expression of key MDM markers was observed in many wildtype macrophages in the Ayana et al. dataset (Fig. 2d, Fig. 2 – Fig. Supplement 2a, Fig. 2 – Fig. Supplement 4a). Hence, the adult killifish brain includes a population of macrophages with some transcriptional similarities to mammalian monocyte-derived macrophages (consistent with studies in zebrafish^80,94,95).

Macrophages in the killifish brain are endocytic and present at brain borders

To functionally test whether oScarlet^HIGH cells have in vivo engulfment capabilities like mammalian BAMs and MDMs, we injected the macromolecule dextran (a 70 kDa polysaccharide, conjugated to a fluorophore) into the brains of young adult to middle-aged (63-91 days) female SP-oScarlet animals (Fig. 3a). We aimed our injection for the brain ventricle to maximize dextran distribution via the CSF⁹⁶. In mice, similar approaches have been shown to result in engulfment of fluorescently-tagged substrates by BAMs and MDMs, and less so by microglia^10,97,98.

Figure 3.

We performed HCR and immunofluorescence staining on brain sections of fish one day post-injection. Cells with bright oScarlet protein signal were enriched for dextran engulfment, as well as for expression of APOEB, as predicted by our single-cell RNA sequencing data (p = 0.031 for each, paired Wilcoxon test) (Fig. 3a, Fig. 3 – Fig. Supplement 1a). We verified that the colocalization seen between dextran and oScarlet signal (Fig. 3a) was not a result of bleed-through or autofluorescence (Fig. 3 – Fig. Supplement 1b-c) and likely resulted from these substrates being localized to the same endosomes and lysosomes. Collectively, these data show that killifish brain macrophages can engulf diverse extracellular substrates (protein, polysaccharide) (see Fig. 5 for additional, ex vivo engulfment assays).

As noted above, in mammals, engulfment of substrates injected into the brain ventricles is a distinguishing functional characteristic of BAMs and MDMs, consistent with their localization at brain borders exposed to CSF flow^10,11. To determine the spatial localization of killifish brain macrophages, we perfused young adult to middle-aged (74-101 days) male and female wildtype killifish with sulfo-NHS-biotin to label vasculature⁹⁹ and performed HCR on brain sections for APOEB. We observed cells with high APOEB expression adjacent to (and on the parenchymal-facing side of, not within) vasculature in wildtype killifish brains, both near the brain’s external borders and within the brain interior (Fig. 3b). While we also observed some cells with high APOEB expression within the parenchyma, away from any detectable vasculature or other brain border, the distribution of cells with high APOEB expression was enriched toward sulfo-NHS-biotin-labeled vasculature in some brain regions (Fig. 3b, see Source Data Table 5 for additional metrics of colocalization with vasculature).

Thus, we have identified a population of macrophages in the killifish brain with several characteristics of mammalian BAMs and MDMs – expressing several markers associated with brain macrophages including complement components and scavenger receptors, exhibiting endocytic functions, and being at least in part localized near brain borders.

Killifish brain macrophages change transcriptionally with age in our model

We leveraged the killifish’s short lifespan to characterize brain macrophages with aging. We first used single-cell RNA sequencing to characterize oScarlet^HIGH cells from old SP-oScarlet killifish (same experiment as described in Fig. 1d). By performing UMAP analysis of oScarlet^HIGH cells from young adult and old SP-oScarlet brains, we found evidence of transcriptional differences between young adult and old oScarlet^HIGH cells (Fig. 4a). We found that oScarlet^HIGH cells from old brains were also predominantly macrophages, expressing markers (CSF1R, APOEB, HLA-DPB1, C1QC, MRC1) at similar levels to young adult oScarlet^HIGH cells (Fig. 4b).

Figure 4.

What transcriptional pathways in these killifish brain macrophages are impacted during aging? Gene ontology (GO) analysis showed an upregulation of translation-related pathways with age in oScarlet^HIGH cells (Fig. 4c-d). GO analysis also identified a modest downregulation in vacuolar acidification with age (Fig. 4c-d). Analysis of specific pathways relevant to engulfment and endolysosomal functions showed that these were also modestly decreased during aging (Fig. 4 – Fig. Supplement 1a). Inflammation-related genes were not strongly upregulated with age in oScarlet^HIGH cells (Fig. 4 – Fig. Supplement 1b), unlike in brain macrophage populations described in mammals^{75,79,100,101}.

Could some of the transcriptional responses observed with age be due to the lifelong, chronic engulfment of oScarlet, particularly given the relative resistance of oScarlet to rapid degradation? To address this possibility, we performed an independent single-cell RNA sequencing experiment of oScarlet^HIGH cells FACS-sorted from the brains of old (130 days) SP-oScarlet male animals (n = 3 fish, pooled), as well all Live cells FACS-sorted from the brains of old (122 days) wildtype male brains (n = 3 fish, pooled) (Fig. 4e). We subset each population to macrophages (i.e., cells from APOEB+ IBA1+ LCP1+ clusters).

We found that neither old oScarlet^HIGH nor old wildtype macrophages could be transcriptionally separated into microglia, BAMs, and MDMs (Fig. 4f, Fig. 4 – Fig. Supplement 1c). When comparing old oScarlet^HIGH macrophages to old wildtype macrophages, we found that these populations expressed marker genes (CSF1R, APOEB, HLA-DPB1, C1QC, MRC1) at similar levels (Fig. 4g-h).

GO analysis showed that translation-related pathways were enriched in old oScarlet^HIGH macrophages compared to wildtype macrophages (Fig. 4i, Fig. 4 – Fig. Supplement 1d).

Some inflammation- and pathogen-response-related pathways were mildly depleted in old oScarlet^HIGH macrophages compared to old wildtype macrophages (Fig. 4 – Fig. Supplement 1d). The engulfment-related pathways examined showed no consistent directional pattern when comparing old oScarlet^HIGH to old wildtype macrophages (Fig. 4 – Fig. Supplement 1e). These results suggest that the translation-related changes observed with age in oScarlet^HIGH cells may reflect a response to lifelong oScarlet exposure, rather than to aging per se. In contrast, changes in engulfment-related pathways are unlikely to be solely driven by lifelong, chronic oScarlet exposure.

Overall, our single-cell RNA sequencing results indicate that killifish brain macrophages in SP-oScarlet animals experience notable transcriptional changes with age, with translation-related genes increasing in expression with age (possibly as part of a response to chronic oScarlet engulfment) and some engulfment-relevant (vacuolar acidification) genes decreasing in expression with age.

Engulfment capacity of killifish brain macrophages declines with age in our model

We next functionally characterized killifish brain macrophages with age, using oScarlet^HIGH status as a proxy for macrophage identity – as supported by our single-cell RNA sequencing data – given the lack of antibodies and reporter lines in killifish to otherwise identify these cells. By performing flow cytometry on the brains of young adult (41-63 days) and old (148-167 days) male and female SP-oScarlet killifish (n = 6 fish per age group and per sex), we noted a decline with age in the oScarlet fluorescence intensity of oScarlet^HIGH cells in both males and females (p = 0.002 for each sex, unpaired Wilcoxon test) (Fig. 5a, Fig. 5 – Fig. Supplement 1a). These observations are consistent with the possibility that aging is accompanied by a potential decline in engulfment by brain macrophages, though the decline in oScarlet brightness in old oScarlet^HIGH cells could also be due to changes in oScarlet expression, secretion, or degradation, especially given that ELAVL3 gene expression has been shown to decrease with age in the killifish brain (Fig. 5 – Fig. Supplement 1b)⁵⁵.

Figure 5.

To assess more directly if the engulfment capacity of brain macrophages changes with age, we developed an acute ex vivo assay in the context of SP-oScarlet animals. In this assay, macromolecule substrates such as ovalbumin (a 45 kDa protein) and dextran (a 70 kDa polysaccharide) (conjugated to far-red and green fluorophores, respectively) were mixed ex vivo with brain cells during the 30 minutes of tissue dissociation⁷⁵, followed by flow cytometry to quantify fluorescent substrate engulfment by cells (Fig. 5b). Using this assay, we verified that oScarlet^HIGH cells from young adult SP-oScarlet brains were enriched for their ability to rapidly engulf ovalbumin compared to oScarlet^LOW cells (p = 0.031 for each sex, paired Wilcoxon test) (Fig. 5c, see Fig. 5 – Fig. Supplement 1c for additional metrics of engulfment).

We used this ex vivo engulfment assay to examine ovalbumin engulfment differences with age among oScarlet^HIGH cells in both males and females (n = 6 fish per age and per sex). Crucially, we showed that oScarlet^HIGH cells in old brains showed a decreased capacity to engulf ovalbumin compared their counterpart cells in young adult brains, in both males (p = 0.002, unpaired Wilcoxon test) and females (p = 0.009, unpaired Wilcoxon test) (Fig. 5d). Fewer old oScarlet^HIGH cells engulfed ovalbumin, and those that did showed reduced ovalbumin brightness compared to young adult oScarlet^HIGH cells that engulfed ovalbumin (Fig. 5 – Fig. Supplement 1d, see Fig. 5 – Fig. Supplement 2a-c for alternative gating schemes).

In a separate experiment, we also performed this ex vivo engulfment assay using dextran as a substrate (n = 4 male fish per age) and verified that oScarlet^HIGH cells from young adult (56 days) male SP-oScarlet brains also showed a trending, non-significant increased capacity to rapidly engulf dextran compared to oScarlet^LOW cells (p = 0.125, unpaired Wilcoxon test) (Fig. 5e, see Fig. 5 – Fig. Supplement 1e for additional metrics of engulfment). oScarlet^HIGH cells in old (156 days) male brains also showed a trending, non-significant decreased capacity to engulf dextran compared their counterpart cells in young adult brains (p = 0.057, unpaired Wilcoxon test) (Fig. 5f). Fewer old oScarlet^HIGH cells engulfed dextran, and those that did showed reduced dextran brightness compared to young adult oScarlet^HIGH cells that engulfed dextran (Fig. 5 – Fig. Supplement 1f, see Fig. 5 – Fig. Supplement 2d-e for alternative gating schemes).

Collectively, these data suggest that brain macrophages in killifish exhibit a functional decline in engulfment capacity with age.

Discussion

In this study, we generated a genetically modified killifish line to investigate engulfment of a secreted protein in the vertebrate brain, providing a model to study clearance. Using the secreted red fluorescent protein oScarlet as a model extracellular protein (here secreted principally from neurons), we found that a macrophage population in the adult killifish brain exhibits a high engulfment capacity. Our in vivo genetic model could be used to study secretion and engulfment in the vertebrate brain, help elucidate functional roles of brain macrophages, and identify interventions that boost engulfment in the aged or diseased brain. We note that some limitations of this genetic model include that oScarlet is expressed constitutively and is relatively resistant to degradation, precluding acute examination of oScarlet engulfment and potentially even affecting the state of cells that engulf oScarlet. We therefore developed additional methods in this study to assess acute engulfment with other substrates both in vivo and ex vivo in the killifish brain. Together, these tools in killifish complement recent approaches to study engulfment in mouse brains^10,75, and they provide a resource to understand the functions of brain macrophages in a short-lived vertebrate.

We found that many killifish brain macrophages display similarities to border-associated macrophages (BAMs) and monocyte-derived macrophages (MDMs), known engulfment-capable macrophage populations in mammalian central nervous systems^{10,75,76,89,102–115}. These similarities include: expression of marker genes such as scavenger receptors, localization to brain border regions, and functional enrichment for engulfment capacity for several substrates (the protein oScarlet in vivo, the polysaccharide dextran in vivo and the protein ovalbumin and polysaccharide dextran ex vivo). Our findings suggest that the adult killifish brain – in contrast to the mouse brain¹⁰² – represents a system where the need for waste clearance may be addressed by relatively abundant cells with characteristics of BAMs and MDMs, parallel to findings from adult zebrafish brains^80,94,95. This potential evolutionary difference is intriguing and could be due to differences in myelination^116–118, neurogenesis^119–121, or demands for pathogen response^122,123 or immune regulation¹²⁴ between fish and mammals or between aquatic and terrestrial environments. These observations also suggest there may be more plasticity among fish brain macrophages than previously anticipated. Understanding how macrophages other than microglia perform brain clearance in species where they are abundant could provide new avenues to promote clearance in mammals.

We showed that in our model, killifish brain macrophages decline in engulfment capacity with age. This is in line with recent mammalian studies highlighting that BAMs and MDMs, despite being rare populations in the mammalian brain, play key functional roles but experience alterations with age and under pathological conditions^{10,11,75,97,98,100,125–167}. Thus, BAMs and MDMs could represent promising avenues for therapeutic intervention in age-associated and other brain diseases. In mammalian BAMs, decline in engulfment with age has been linked to decreased expression of clathrin-mediated endocytosis receptors⁷⁵. While similar transcriptional changes with age were not observed in our single-cell RNA sequencing datasets, key endocytic machinery in the killifish brain may be regulated at the level of translation (which has been shown to decrease globally with age in the killifish brain⁵⁴), or at a post-translational level (e.g., decreased efficiency of proton pumping to acidify lysosomes¹⁶⁸), resulting in an evolutionarily conserved decline in scavenging capacity with age.

Overall, our findings highlight the underappreciated role that engulfment by macrophages, particularly those at brain borders, could play in critical waste clearance processes in the brain. This study also underscores the potential of the killifish as a powerful vertebrate model to investigate questions relevant to aging, disease, and evolution of the complex and specialized cell types of the vertebrate brain.

Methods

Killifish husbandry

The short-lived Nothobranchius furzeri GRZ strain³⁸ was used as the background killifish strain in this study. Fish were maintained using an Aquaneering recirculating aquatic system at a temperature of ∼27°C, pH of 7-7.5, conductivity of 1200 μS with a 12 hour/12 hour light/dark cycle with the light on between ∼7 a.m. and ∼7 p.m. For the first 1 month of life, fish were fed live brine shrimp (Brine Shrimp Direct, BSEP6LB). Brine shrimp feeding solution was generated by mixing 50 mL of brine eggs hatched over 48 hours at ∼27°C in a cone with 10 L reverse osmosis (RO) water and 1.5 cups salt (approximately 350 mL; Instant Ocean; Amazon, B000256EUS)). Fish were fed brine shrimp twice daily at ∼9 a.m. and ∼3 p.m. on weekdays (0.5-2 mL per feeding, with volume per feeding approximately scaling with each week of development). Fish were fed once daily on weekends and holidays. After 1 month, fish were fed food pellets (Reed Mariculture, Otohime C1) twice daily at ∼9 a.m. and ∼3 p.m. (and once daily on weekends and holidays) with ∼18 mg per feeding. Tests for pathogens were conducted every three months (Charles River Labs). Fish were housed in Stanford ChEM-H/Neuro Vivarium and Research Animal Facility (RAF), according to IACUC protocol 13645.

For breeding and collecting killifish embryos, plastic weigh boats (e.g., Heathrow Scientific, HS1420BB) filled up to halfway with sand were dropped in breeding tanks containing 1 male and 1-10 females (all 1-4 months old, 1.5-2.5 month old breeding females used when available; 2.8 L breeding tank for up to 3 females, 6.8 L breeding tank for up to 6 females; 9.8 L breeding tank for up to 10 females) to initiate mating, and approximately 3-6 hours later the embryos were collected by pouring the contents of the sand tray through a sieve.

Collected embryos were placed in 5-15 mL embryo solution liquid (embryo solution was made by adding 2 Ringer tablets (EMD Millipore, 1.15525.0001) and 100 μL methylene blue (Kordon, 37344) to 1 L Milli-Q water, for a final concentration of 0.01% methylene blue) in 35 or 60 mm diameter Petri dishes (Greiner, 627102 and 628161 respectively; dish size used dependent on number of embryos, with a maximum of ∼40 embryos in a 35 mm and ∼150 embryos in a 60 mm Petri dish) for 2 weeks at 28°C in a dark incubator (e.g., Thermo 50125590H). Dead embryos were removed each day, with the embryo solution exchanged on the first day and on an as-needed basis thereafter. Embryos were generally treated with Povidone Iodine Prep Solution U.S.P. (Dynarex; Amazon, B00859URLK) (added to the Petri dish with embryo solution at ∼1:1000 dilution, followed by 2-3 washes with embryo solution) immediately after collection.

At 14-16 days post fertilization (dpf), embryos were placed on autoclaved coconut fiber (Fluker’s; Amazon, B01HO27S4M) in 35 or 60 mm diameter Petri dishes (with a maximum of ∼40 embryos in a 35 mm and ∼150 embryos in a 60 mm Petri dish) and maintained at 27-30°C for an additional 2 weeks before hatching by placing in ∼5 mL cold (4°C) humic acid solution (Sigma-Aldrich, 53680, 1 g/L in RO water) in a 35 mm Petri dish left overnight at room temperature, followed by transfer to the aquatic system. In certain cases (noted in Supplementary Table 1), some embryos entered diapause, even at 28°C (this did not appear to be linked to maternal age or any other obviously noticeable parameter), and had to be induced out of diapause using a thermocycler program of 6 hours at 28°C and 6 hours at 37°C repeated 6 times before being allowed to continue to develop at 28°C and transferred to coconut fiber followed by hatching in cold humic acid.

Newly hatched killifish (fry) were housed on the system at a density of up to 20 fry per 0.8 L tank, with successive reductions in density as the fish grew older until reaching a density of 1 male fish per 1.8-2.8 L tank, or 1-10 female fish per 1.8-9.8 L tank (2.8 L tank for up to 4 females, 6.8 L breeding tank for up to 6 females; 9.8 L breeding tank for up to 10 females), beyond 1 month of age (sexual maturity).

Unless otherwise noted, original data in this study from all figures were collected from animals raised in Stanford ChEM-H/Neuro Vivarium after a general system bleaching (January 2025) and treated as described above. Data from Fig. 1b, Fig. 1c, Fig. 1h (images), Fig. 1 – Fig. Supplement 1c, Fig. 1 – Fig. Supplement 1d (4 of 10 fish), and Fig. 3b were obtained in another facility at Stanford (RAF), with specific husbandry variations noted in Supplementary Table 1 (see “Notes”). Data from Fig. 1 – Fig. Supplement 1a, Fig. 1 – Fig. Supplement 1f, Fig. 4f-i, and Fig. 4 – Fig. Supplement 1c-e were collected in ChEM-H/Neuro Vivarium before system bleaching (with bryozoans and mycobacteria prevalent), with specific husbandry variations noted in Supplementary Table 1 (see “Notes”). A list of animals used in each experiment/figure panel, with information on age (with hatch and sacrifice dates), notes on husbandry conditions, sex, and genotype, can be found in Supplementary Table 1. Importantly, the key phenotypes observed in this study were robust to variations in husbandry conditions.

Generation of SP-oScarlet killifish

To generate a killifish knockin strain expressing a secreted form of the red fluorescent protein oScarlet in the nervous system, we used our CRISPR-based knockin method⁶⁷. Overall, a construct encoding a signal peptide (SP) from killifish HSPA5 (signal peptide predicted using SignalP 5.0⁷¹) fused to the N-terminus of the red fluorescent protein oScarlet⁷⁰ was inserted by homology-directed repair just 5’ of the endogenous stop codon of the ELAVL3 (LOC107378364) locus in the killifish GRZ wildtype strain. Across multiple vertebrate species, ELAVL3 encodes an RNA-binding protein expressed strongly in neurons, as well as some glia such as oligodendrocytes^68,78, and has successfully be used in killifish as a driver for pan-neuronal expression⁶⁷.

The signal peptide allows the secretion of oScarlet from ELAVL3-expressing cells. To prevent unwanted fusion of oScarlet to ELAVL3, the construct also encodes a P2A self-cleaving peptide⁷² at the C-terminal end of ELAVL3 (full sequence of construct, including 200 bp homology arms at either end, in Supplementary Table 2; annotated SnapGene map file in Supplement 1).

To generate the killifish strain expressing this secreted form of oScarlet, which we refer to as SP-oScarlet, we followed the step-by-step CRISPR knockin protocol described by Bedbrook et al., 2023⁶⁷. For each round of injection attempted, a sand tray was dropped in a harem breeding tank (6.8-9.8L tank) containing 1 GRZ wildtype male and 5-10 GRZ wildtype females (all 1-4 months old, with 1.5-2.5 months old breeders used when available) to initiate mating, and approximately 3-4 hours later the embryos were collected by pouring the contents of the sand tray through a sieve.

Injection needles were generated from borosilicate glass capillaries (Sutter, BF100-58-10) pulled on P-97 pipette pullers (Sutter Instruments). A ramp test was performed and a Heat value of [RAMP value + 30] was used. Other settings used were Pull = 20-30, Velocity = 19-30, Time = 200. These parameters were optimized on each specific pipette puller used, with the resulting needles tested in pilot injections.

In the 30 minutes before collection, 1 μL of a crRNA targeting the 3’ end of the coding sequence of ELAVL3 (at 100 μM in Nuclease-Free Duplex Buffer (IDT, 11-01-03-01); sequence in Supplementary Table 2) and 1 μL of tracrRNA (at 100 μM in Nuclease-Free Duplex Buffer; IDT, 1072532/3/4) were added to 31.3 μL Nuclease-Free Duplex Buffer and annealed at 95°C for 5 minutes to generate the guide RNA (gRNA) solution. After this annealing, 10 μL of the resultant gRNA solution was added to 0.5 μL rCas9 protein (10 μg/μL; IDT, 1081059), and 5.5 μL PBS (Corning, 21-040-CV) and this mixture was heated at 37°C for 10 minutes to generate the ribonucleoprotein complex. After this heating, 8 μL of the ribonucleoprotein complex was added to 1 μL dsDNA homology-directed repair (HDR) template (150 ng/μL; IDT custom-synthesized, sequence in Supplementary Table 2, annotated SnapGene map file in Supplement 1), 1 μL HDR enhancer solution (IDT 10007921; 10 μM), and 0.33 μL phenol red (e.g., Sigma-Aldrich P0290-100ML) to generate the injection solution.

For injection, embryos at the 1 cell stage were aligned in rows on an injection pad (1% agarose in embryo solution, solidified over a custom-3D-printed mold with rows) with the single cell facing upwards placed under a dissecting microscope (Leica, MZ6). Embryos were individually injected with injection solution dispensed through the aforementioned needles (loaded using Microloader (Calibre Scientific, EPE-5242956003) tips; per-injection volume estimated to be in the nanoliter range). Needles were unsealed by brushing against sides of embryos. Injections were performed with a MPPI-2 Pressure Injector (ASI) with Air Out = Continuous, Pressure ∼ 10 psi. This procedure allowed for a constant, slow stream of injection solution flowing out of the needle (visible due to the use of phenol red); the needle was inserted into the single cell of an embryo and injection solution allowed to flow out until the single cell was visibly red (∼2-3 seconds).

After injection, embryos were placed in embryo solution in 35 or 60 mm diameter Petri dishes for 2 weeks at 28°C and dead embryos were removed daily. At 9-11 dpf, we identified F0 fish expressing the red fluorescent protein oScarlet in the nervous system (by visual inspection of embryos under a fluorescent dissecting microscope). At 14-16 dpf, embryos were placed on coconut fiber in 35 or 60 mm diameter Petri dishes and maintained at 27-30°C for an additional 2 weeks before hatching by placing in cold humic acid solution left at room temperature overnight (see above), followed by transfer to the aquatic system.

To generate a stable SP-oScarlet line, adult F0 fish (as determined by the visual screening of embryos described above) were crossed with GRZ wildtype fish. F1 fish and some subsequent generations were genotyped to identify SP-oScarlet progeny (see “Genotyping,” below). We outcrossed F1 SP-oScarlet fish with wildtype GRZ fish for >5 generations. Both

GRZ male x SP-oScarlet female and SP-oScarlet male x GRZ female crosses were used. In one cohort, we generated homozygous SP-oScarlet fish by crossing heterozygous parents and noted the homozygotes had relatively poor survival at the fry stage (<10% surviving to one month of age). Therefore, we maintained the line at the heterozygous state and we used heterozygotes for all experiments.

Genotyping

In most generations, SP-oScarlet progeny of SP-oScarlet x GRZ wildtype crosses were identified by visual inspection of embryos. Nonetheless, in the F1 generation and some subsequent generations, we used PCR genotyping to confirm the presence of an insertion at the ELAVL3 locus, using a three-primer PCR strategy with e.g., primers pFL1, pRL1, and pmCF (sequences in Supplementary Table 2). All primers used were synthesized by IDT.

Genomic DNA was extracted from a small (<1 mg) amount of tail tissue cut from a live fish, either from a 1 day old fry or an adult fish, anesthetized either on a dish placed over ice (for fry) or in 300 mg/L Tricaine (dissolved in system water; e.g., Pentair, TRS1) buffered with sodium bicarbonate to pH ∼7 (for adults).

The tail tissue was lysed in 30 μL (for fry) or 100 μL (for adults) 2% (volume/volume) Proteinase K (e.g., Invitrogen, 25-530-049) in DirectPCR lysis solution (Viagen Biotech, 102-T) in a PCR tube (e.g., USA Scientific, 1402-4708) using a thermocycler program of 55°C for 2 hours (lysis) and 80°C for 10 minutes (Proteinase K inactivation).

Next, 2 μL (for fry) or 3 μL (for adults) of the resulting lysate (unpurified) was added to 10 μL 2x GoTaq Green Master Mix (Promega, M7123), 6 μL Milli-Q water, and 0.5 μL each of primers pFL1, pRL1, and pmCF (each at 10 μM in Milli-Q water) following the manufacturer’s instructions for PCR with a total reaction volume of 19-20 μL and a thermocyler program of 98°C (pre-heating); 98°C for 30 seconds; 30 cycles of 98°C for 30 seconds, 55°C for 1 minute, and 72°C for 90 seconds; and finally 72°C for 5 minutes.

Then, 8 μL PCR product was resolved on a 1% agarose gel to verify the size of amplified bands. Samples with only one band of size 971 bp (pFL1, pRL1 band) were identified as wildtype, while samples with two bands (824 bp; pmCF, pRL1 band from knockin allele and 971 bp; pFL1, pRL1 band from wildtype allele) were identified as heterozygous SP-oScarlet (the >1.5 kb pFL1, pRL1 band from the knockin allele would not be expected to be amplified efficiently under these PCR conditions).

Separately, to validate the sequence of the genomic insertion in a SP-oScarlet heterozygous fish, we performed PCR with primers pFL1 and pRL1 outside the homology arms of the knockin construct. For this, genomic DNA was extracted from the sidebody muscle of heterozygous SP-oScarlet killifish (78 days old male). The fish was sacrificed in ice water and <25 mg of sidebody muscle was cut out with a scalpel (Integra, 4-122), chopped, and placed in 200 μL of DNEasy Blood & Tissue kit (Qiagen, 69504/6) Buffer AL in a 1.5-mL tube (e.g., Eppendorf, 0030108418). The DNEasy Blood & Tissue kit manufacturer’s protocol was then followed to extract DNA, which was eluted in 200 μL of Buffer AE.

DNA amount and quality was assessed by NanoDrop and 1 μL (<200 ng) of DNA was used as input in a PCR with 1 μL LongAmp polymerase (New England Biolabs, M0323S), 5 µL 5x LongAmp Taq Reaction Buffer (New England Biolabs, M0323S), 3.75 μL 2 mM DNTPs (e.g., Sigma-Aldrich, D7295-.2ML), 12.25 μL Milli-Q water, and 1 μL each of primers pFL1 and pRL1 (each at 10 μM in Milli-Q water) following the manufacturer’s instructions for PCR with a total reaction volume of 25 μL and a thermocyler program of 94°C (pre-heating); 94°C for 30 seconds, 30 cycles of 94°C for 30 seconds, 58°C for 1 minute, 65°C for 3 minutes; and finally 65°C for 10 minutes. Less than 8 μL PCR product was resolved on a 1% agarose gel to verify the size of amplified bands. The remaining PCR product was purified with an NEB Monarch^® DNA cleanup kit (New England Biolabs, T1030L) following the manufacturer’s protocol exactly and the eluted DNA sent to Primordium Labs for long-read (Oxford Nanopore) sequencing (“Linear/PCR” submission option).

Analysis of the long-read sequencing of the entire locus in SnapGene (by aligning to the expected insertion sequence as listed in Supplementary Table 2) revealed that there were two copies of the SP-oScarlet construct inserted. In silico translation with Expasy (https://web.expasy.org/translate/; 3’5’ Frame 3, oScarlet protein in frame) showed that there was a stop codon after the first copy, compatible with the presence of only one copy of oScarlet protein (Fig. 1 – Fig. Supplement 1a; full sequence of actual insertion in Supplementary Table 2).

Preparation of killifish brain sections

To prepare killifish brain sections, we first sacrificed fish in ice water (in most experiments, unperfused fish; in experiments in Fig. 3b, fish perfused with sulfo-NHS-biotin, see “Intracardial perfusion,” below). Whole fish heads were cut off and incubated overnight at 4°C in the dark with rocking in 1-5 mL 4% paraformaldehyde (PFA; e.g., Electron Microscopy Sciences, 15710 diluted 1:4 in phosphate-buffered saline (PBS) (e.g., Corning, 21-040-CV)) in PBS (volume used dependent on fish head size) in a 15-mL conical tube (e.g., Falcon, 352196). The following day, 3 washes with PBS (same volume as 4% PFA used in the preceding step) were performed for at least 1 hour each at 4°C in the dark with rocking. After the final wash, fish brains were dissected out of the heads in cold PBS using forceps (e.g., Fine Science Tools, 11252-20), starting with removal of the lower jaw, then the upper part of the head and eyes, and finally freeing the brain from the skull. The brains were then sunk in 0.5-1 mL 30% (weight/volume) sucrose (Sigma-Aldrich, S0389-1KG) in PBS overnight at 4°C in the dark. Once the brains had sunken, they were embedded in Optimal Cutting Temperature (OCT) compound (Fisher HealthCare, 4585). OCT blocks containing the brains were frozen over dry ice and stored at −80°C until sectioning.

Sagittal sections with a thickness of 50 µm were generated using a cryostat (Leica, CM3050 S) and MX35 Ultra blade (Epredia, 3053835) and added to Superfrost Plus slides (Fisherbrand, 1255015) that were stored at −20°C until further processing. Sections with visible and undamaged forebrain, midbrain and hindbrain tissue were included, with approximately position-matched sections (based on visual inspection of brain tissue gross morphology) used where appropriate for comparisons. RNase-free reagents and technique (namely, spraying tools and surfaces with 70% ethanol and RNaseZap (Invitrogen, AM9780, AM9782)) were used for all experiments in which hybridization chain reaction (HCR) was later performed on the brain sections.

Hybridization chain reaction

To perform hybridization chain reaction (HCR)⁷³ on killifish brain sections, we first placed slides with 50 µm-thick killifish brain sections were placed in RNase-free PBS at room temperature for at least 20 minutes to wash off excess OCT (preceded by a heating step at 37°C for ∼10 minutes in some experiments to minimize the occurrence of sections detaching from the slides).

After washing off OCT, slides were then permeabilized in 1% Triton (e.g., Sigma-Aldrich T8787-50ML) in 100% methanol (e.g., Sigma-Aldrich 34860-100ML-R) for at least 1 hour at −20°C, then washed twice in RNase-free 2x saline sodium citrate (SSC; diluted from 20x SSC (ThermoFisher, AM9763) in RNase-free water) with 0.1% Triton (250 µL per slide) for 15 minutes per wash at room temperature. HCR hybridization buffer (Molecular Instruments, “HCR Probe Hybridization Buffer,” for tissue sections; 250 µL per slide; preheated to 37°C) was applied to the slides, which were incubated at 37°C for 30 minutes with a coverslip on top. After the 30 minutes incubation, this hybridization buffer was removed and replaced with hybridization buffer containing the appropriate probe sets diluted 1:100 from stock. Once the hybridization buffer with probes sets had been applied, the slides were then incubated at 37°C overnight with a coverslip on top.

Sequences for all HCR probe sets used in this study are listed in Supplementary Table 2. These sequences were determined by a Python script that used the most 5’ or 3’ (as noted in Supplementary Table 2) 1012 bp of the longest killifish mRNA sequence for the gene of interest (or the entire mRNA sequence, in the case of shorter genes) to design up to 22 pairs of probes 22 bp in length each, with a 2 bp spacer between probes. Probes sets were ordered in tubes as lyophilized 50 nM oPools (IDT custom-synthesized) and resuspended in 50 µL of TE8.0 (Invitrogen, AM9849). After a 5 minutes wait, the tubes were vortexed and 20 µL of the resuspension solution was added to 20 µL of TE8.0. This diluted resuspension solution was vortexed, spun down, and spit into 5 µL aliquots which were used as the probe set stocks.

After the overnight incubation with probes sets diluted in hybridization buffer, HCR wash buffer (Molecular Instruments, “HCR Probe Wash Buffer,” for tissue sections; 250 µL per slide; preheated to 37°C) was applied to was applied to the slides, which were incubated at 37°C for two washes of 30 minutes each with a coverslip on top. The slides were then washed twice in RNase-free 2xSSC with 0.1% Triton (250 µL per slide) for 15 minutes per wash at room temperature. Next, HCR amplification buffer (Molecular Instruments, “HCR Amplification Buffer,” for tissue sections; 100 µL per slide; equilibrated to room temperature) was applied to the slides, which were incubated at room temperature for at least 30 minutes with a coverslip on top.

During this incubation, amplification hairpins (Molecular Instruments; 2 µL per hairpin per slide) were prepared. Hairpins were first snap-heated at 95°C for 5 min, followed by 30 minutes of cooling at 4°C. Hairpins used for each probe in this study are listed in Supplementary Table 2 (see “Notes”). Hairpin solution (100 total µL per slide) was prepared by adding hairpins (2 µL per hairpin per slide) to amplification buffer. After applying hairpin solution, slides were incubated at room temperature overnight in the dark.

After overnight incubation with hairpin solution, slides were washed twice in RNase-free 2xSSC with 0.1% Triton (250 µL per slide) for 5 minutes per wash at room temperature in the dark, then washed once in RNase-free 2xSSC with 0.1% Triton (250 µL per slide) for 5 minutes at room temperature in the dark, and then washed twice in RNase-free 2xSSC with 0.1% Triton (250 µL per slide) for 5 minutes per wash at room temperature in the dark before mounting in ProLong Gold with 4’,6-diamidino-2-phenylindole (DAPI; Thermo Fisher Scientific, P36931) under a coverslip. Mounted slides were dried overnight at room temperature in the dark and stored at 4°C in the dark until imaging.

For experiments in which HCR and immunofluorescence staining were performed sequentially on the same slides, after overnight incubation with hairpin solution, slides were washed five times in RNase-free 2xSSC with 0.1% Triton (250 µL per slide) for 5 minutes per wash at room temperature in the dark, before proceeding to blocking for immunofluorescence staining.

Immunofluorescence staining

To perform immunofluorescence staining on killifish brain sections, we first heated slides with 50 µm-thick killifish sagittal brain sections for 10 minutes at 37°C, then placed the slides in PBS at room temperature for at least 20 minutes to wash off excess OCT. Slides were then permeabilized in 0.5% Triton in methanol for at least 1 hour at −20°C, then washed twice in PBS for 5 minutes per wash at room temperature.

Next, for blocking, circles were drawn around sections with a hydrophobic pen (Edding 751) blocking buffer (PBS with 5% normal donkey serum (NDS; Fisher Sci., SP-072-VX10) and 1% bovine serum albumin (BSA; from 10% BSA stock; Miltenyi Biotec, 130-091-376)) was added to the areas within the circles for a 30 minutes incubation at room temperature. After blocking, the blocking buffer was removed and primary antibodies were added for overnight incubation at 4°C in the dark. All primary antibodies used in this study are listed in Supplementary Table 3 along with the dilution (in blocking buffer) at which they were used.

After incubation with primary antibodies, slides were washed 6 times with PBS with 0.1% Triton for 10 minutes each at room temperature in the dark, and then washed 3 times with PBS for 5 minutes each at room temperature in the dark, before the PBS was dried off and secondary antibodies and DAPI (Fisher Scientific, 62248) (all diluted 1:500 in blocking buffer) were added for 2 hours incubation at room temperature in the dark. All secondary antibodies used in this study are listed in Supplementary Table 3.

After incubation with secondary antibodies, slides were washed 6 times with PBS with 0.1% Triton for 10 minutes each at room temperature in the dark, and then washed 3 times with PBS for 5 minutes each at room temperature in the dark before mounting in ProLong Gold (ThermoFisher, P36930) under a coverslip. Mounted slides were dried overnight at room temperature in the dark and stored at 4°C in the dark until imaging.

Tissue clearing

To visualize oScarlet protein in SP-oScarlet brains at a whole-brain level, we performed tissue clearing using a modified Adipo-Clear protocol¹⁷⁰. A dissected young adult SP-oScarlet brain in PBS (processed as described in “Preparation of killifish brain sections,” above, up to dissection, then placed in room temperature PBS) was subjected to the Adipo-Clear protocol with ethyl cinnamate (Sigma-Aldrich, 112372) was used in place of dibenzyl ether. An anti-RFP antibody used to visualize oScarlet (details in Supplementary Table 3).

Imaging

Unless otherwise noted, all images were taken with a Zeiss LSM900 confocal microscope with Zen3.0 Blue Edition software. Zeiss Immersol Oil 518F (Zeiss 4Y00-R0DY-1007-3VF3) was used with the 40x and 63x magnification objectives. Imaging parameters, including the magnification objective used for each image, are listed in Supplementary Table 4.

The images in Fig. 1d, Fig. 1e, and Fig. 2 – Fig. Supplement 1a were taken with a Miltenyi Biotec Ultramicroscope Blaze, an EVOS M5000 tissue culture microscope, and a Zeiss Elyra7 structured illumination microscope (at Stanford’s Cell Sciences Imaging Facility (RRID:SCR_017787)), respectively.

Image processing and quantification

Image analysis scripts are listed in Supplementary Table 4. Confocal images taken were z-stacks with 10-20 slices 0.5 μm apart. Maximum intensity projections of 10 slices per image were generated in Fiji/ImageJ using the macros listed in Supplementary Table 4 and saved as TIF files. The max projection TIF files were loaded into QuPath (v0.6.0)¹⁷¹ for automated cell detection and, where needed to detect RNA transcripts, subcellular spot detection using Groovy scripts as listed in Supplementary Table 4.

Unless otherwise noted, all images shown in this manuscript were processed (maximum intensity projection, brightness/contrast adjustment, cropping) with Fiji/ImageJ. The image in Fig. 1d was processed in Imaris (specialized software for 3D image processing) and further processed in Fiji/ImageJ. The image in Fig. 2 – Fig. Supplement 1a was pre-processed with Zen Blue Software (for structured illumination mathematical correction) on the microscope used for image acquisition (at Stanford’s Cell Sciences Imaging Facility, RRID:SCR_017787) prior to processing in Fiji/ImageJ.

Since APOEB HCR signal was strong and individual transcript spots were often difficult to distinguish, mean APOEB RNA intensity per cell in 10x magnification images was used as a readout of APOEB expression. For all other HCR probes used, individual transcript spots were counted in 40-63x magnfiication images. QuPath measurements were exported as .csv files and analyzed in R (v4.5.1).

In quantifications of oScarlet immunofluorescence staining, oScarlet^HIGH cells were defined as those with both a mean fluorescence intensity in the oScarlet channel of >2.5 standard deviations higher than the mean of all cells in the image and maximum fluorescence intensity in the oScarlet channel of >25,000, due to the punctate nature of bright oScarlet protein signal.

In quantifications of dextran-Oregon Green 488 native fluorescence, dextran-positive cells were defined as those with both a mean fluorescence intensity in the dextran channel of >2.5 standard deviations higher than the mean of all cells in the image and maximum fluorescence intensity in the dextran channel of >25,000, due to the punctate nature of dextran signal.

In quantifications of APOEB HCR signal, cells with high APOEB RNA expression were defined as those with a mean fluorescence intensity in the APOEB RNA channel of >2.5 standard deviations higher than the mean of all cells in the image, since APOEB RNA signal was largely cell body-filling.

In quantifications of colocalization with vasculature, vasculature was defined as cells with a mean fluorescence intensity in the sulfo-NHS-biotin channel of >1.5 standard deviations higher than the mean of all cells in the image, since sulfo-NHS-biotin signal was largely cell body-filling. Cells colocalizing with vasculature were defined as those with a centroid <5 μm from the centroid of any vasculature cell (inclusive of vasculature cells themselves).

For representative sampling of all brain regions for quantification, we imaged one region of interest from each of the forebrain, midbrain, and hindbrain across 1-3 sagittal sections per animal, avoiding areas of visually obvious tissue damage.

In Fig. 3a and Fig. 3 – Fig. Supplement 1a, four images (one image per animal for each of four animals; listed in Supplementary Table 4) were excluded from quantification because they contained regions with a visually obvious very high density of cells with high APOEB RNA expression, which could potentially reflect a response to injury (see also “Injection of dextran into killifish brains,” below).

For quantifications in Fig. 3b, brain external borders were generally avoided when selecting regions of interest, because sulfo-NHS-biotin signal was much brighter at these external borders as compared to other areas in the brain, which we could bias quantification. As such, our analysis mostly examined APOEB+ cells in the brain interior.

Injection of dextran into killifish brains

To test the endocytic properties of cells in the killifish brain, we injected dextran-Oregon Green 488 (70 kDa) (Thermo Fisher Scientific, D7173), aiming for the ventricle to maximize distribution of dextran through the cerebrospinal fluid. Only females were used for these experiments since adult males have thick muscle above the skull that makes it difficult to visually locate the brain.

The brain injection procedure bears similarities to cerebroventricular microinjection procedures used for adult zebrafish^96,172, though specific adaptations were made for killifish. Brain injection needles (made identically to embryo microinjection needles) were filled with a mixture of 90% dextran (1 mg/mL dissolved in PBS) and 10% phenol red (for visualization), for a final dextran concentration of 900 μg/mL, using Microloader tips.

After loading, the needles were unsealed by breaking their points with fine forceps. Prior to injection, fish were individually anesthetized in 300 mg/L Tricaine (dissolved in system water) buffered with sodium bicarbonate to pH ∼7. Successful anesthesia was confirmed by the fish being non-responsive to tail pinching, at which point the fish was placed in between pins on a Tricaine-filled dish underneath the injection microscope.

Injections were performed with a Pneumatic Picopump (World Precision Instruments, PV380) with settings Vacuum = 0 in Hg (vac setting), Hold = 0 psi, Eject = 5-10 psi (Vent = hold), Range = 10 s, Period = 50, Duration = timed, with the needle stabilized and positioned using a DC-3K right apparatus (World Precision Instruments, DC3001R, 00-45-201-0000). We used a micrometer (Fisherbrand, 12-561-SM1) to estimate an approximate ejection volume of <0.5 μL (based 5 test ejections of 10% phenol red in PBS into a mineral oil (e.g., Sigma-Aldrich, M5904-5ML) droplet with similar needles, unsealing procedure, and approximately equal pressure settings as used for brain injections; although injecting through the skull further unseals the needle). We used the minimum Eject pressure necessary to eject liquid for this procedure, to minimize damage to the brain.

To aim for the ventricle, needles were inserted at approximately a 45° angle through the skin and skull at the most medial point of the forebrain/midbrain border (where the two lobes of the telencephalon and two lobes of the optic tectum meet). This point was approximately aligned along the anterior-posterior axis with the most posterior part of the eyes.

No brain coordinates were used to estimate the depth the needle needed to reach before ejection (by pressing foot pedal); the depth of needle insertion was approximately optimized by practice injections, with immediate spread of phenol red used as a readout of successful injection. We note that engulfment of dextran by brain macrophages was observed in some cases where immediate spread of phenol red upon ejection was not observed, possibly due to clogging in the needle limiting the volume of liquid ejected.

Fish were then placed in recovery tanks with fresh system water (not on aquatic system) and sacrificed 1 day (∼20-26 hours) after dextran injection.

We excluded one animal from the quantifications in Fig. 3a and Fig. 3 – Fig. Supplement 1a based on visually obvious lack of dextran signal throughout the brain (Fig. 3 – Fig. Supplement 1b). Of the 38 remaining images from six remaining animals, we additionally excluded four images (one each from four different animals) that contained regions of interest with visually apparent very high densities of APOEB+ cells, which could reflect a response to injury. These four excluded images are listed in Supplementary Table 4 (row 14, “Notes”).

Intracardial perfusion

To visualize vasculature, we perfused wildtype killifish with sulfo-NHS-biotin (Thermo Scientific, 21335). The perfusion procedure was similar to that described by Costa et al., 2026⁵⁵. Perfusion was performed with a syringe pump (KD Scientific, Legato 200 Series, 788200 syringe pump) with 30-gauge hubless needles with a point style 4 bevel (Hamilton Syringe, custom needle, 22030-01; 30 gauge, Hubless Needle, No Hub, 30 gauge, 1.5 inches length, point style 4 [12°]) attached using ∼0.25 meters of BD Intramedic PE Tubing (BD, 427400) to a 30 gauge metal hub blunt-end Luer needle (Hamilton Syringe, custom needle, 7748-16; 30 gauge, Metal Hub Needle, Point Style: 3; Needle Length: 0.375 inches) on a 20-mL disposable syringe with Luer Lock tip (‘Sterile Syringe Only with Luer Lock Tip’, Amazon, B08FJCSLFC).

Adult killifish were anesthetized in 300 mg/L Tricaine (dissolved in system water; e.g., Pentair, TRS1) buffered with sodium bicarbonate to pH ∼7, with the underbelly cut so as to expose the heart (full details of cutting described by Costa et al., 2026⁵⁵), with the perfusion needle inserted into the bulbus arteriosus. The perfusion flow rate was 3.5 mL/minute. First, 6 mL of 50 mM Tris pH 8.0 (filled in the disposable syringe; Thermo Fisher Scientific, 15-568-025) was perfused, followed by 4 mL of 0.5 mg/mL sulfo-NHS-biotin in PBS, and then 2 mL of 50 mM Tris pH 8.0.

After perfusion, fish were sacrificed in ice water and their carcasses incubated overnight at 4°C in the dark with rocking in >5 mL 4% paraformaldehyde in PBS (e.g., diluted from 16% paraformaldehyde in PBS (Electron Microscopy Sciences, 15714-S)) in a 15-mL conical tube (e.g., Falcon, 352196). Cuts were made in the underbelly to expose the internal organs and allow for complete fixation. The following day, 3 washes with PBS (same volume as 4% paraformaldehyde used) were performed for at least 1 hour each at 4°C in the dark with rocking. After this, the PBS was exchanged with 100% methanol (Sigma-Aldrich, 34860-4L-R) and the fish were stored long-term (>1 year) at −20°C. Before sectioning, heads were cut off and rehydrated with a reverse gradient of 2 mL each of (1) 75% methanol and 25% PBS; (2) 50% methanol and 50% PBS; (3) 25% methanol and 75% PBS; and finally (4) 100% PBS, sequentially for 15-45 minutes each on ice. Then, an additional wash in 100% PBS was performed for 15 minutes, and then brains were dissected for preparation of killifish brain sections.

Dissociation of killifish brains for single-cell RNA sequencing

We conducted two single-cell RNA sequencing experiments: one in October 2024 (Fig. 4e) and one in August 2025 (Fig. 1d). In the October 2024 experiment, we sequenced FACS-sorted oScarlet^HIGH cells from old male SP-oScarlet brains (n = 3 fish, pooled) as well as FACS-sorted Live cells from old male wildtype brains (n = 3 fish, pooled). In the August 2025 experiment, we sequenced FACS-sorted oScarlet^HIGH and oScarlet^LOW cells from young male SP-oScarlet brains (n = 4 fish, pooled) as well as FACS-sorted oScarlet^HIGH from old male wildtype brains (n = 4 fish, pooled).

For single-cell RNA sequencing experiments, fish were always sacrificed in the morning on a weekday, having not received feeding since the previous afternoon. Males were used for these experiments due to their larger size, maximizing cell yield for single-cell RNA sequencing.

For dissociation, we used protocols similar to those described by Mariën et al., 2024¹⁷³. Fish (not perfused) were sacrificed in ice water and brains were dissected in ice-cold PBS using forceps, starting with removal of the lower jaw, then the upper part of the head and eyes, and finally freeing the brain from the skull. After dissection, brains were cut into pieces approximately the size of a half telencephalon. Per sample (each sample consisted of a pool of 3-4 brains matched in age, genotype, and sex), all tissue pieces were transferred into a 1.5-mL microcentrifuge tube containing 800 µL of ice-cold dissection medium (DMEM/F12 with HEPES (Thermo Fisher Scientific, 11330032) with actinomycin D (transcription inhibitor; Sigma-Aldrich, A1410-2MG) added at 1:200, triptolide (transcription inhibitor; Sigma-Aldrich, T3652-1MG) added at 1:1000, anisomycin (translation inhibitor; Sigma-Aldrich, A9789-5MG) added at 1:368.5 and put on ice until all brains were dissected (up to 2 hours).

Once all the dissected tissues were ready in dissection medium on ice, fresh papain solution was prepared. Papain solution consisted of 50 µL of papain (suspension, from papaya latex; Worthington, LS003126), 20 µL DNase I solution (10 mg/mL; e.g., STEMCELL Technologies, 07900; dissolved in Milli-Q water), and 40 µL of L-cysteine solution (12 mg/mL) per 1 mL of DMEM/F12, with actinomycin D added at 1:200, triptolide at 1:1000, anisomycin at 1:368.5. This solution was sterilized through a 0.2-0.22-µm-pore-size filter (Sigma-Aldrich, WHA9914-2502 or Fisher Scientific, SLGP033RS) and vortexed to mix. L-cysteine solution was prepared by adding 120 mg of L-cysteine hydrochloride (e.g., Sigma-Aldrich, C6852-25G) to 10 mL of filtered, autoclaved Milli-Q water, and stored for up to 2 weeks at 4°C or 1 year at −20°C.

The dissection medium was removed by pipetting and 1 mL fresh papain solution was added to each sample. To start digesting the brain tissue, tubes with the samples were placed in an incubator for ∼6 minutes at 37°C. After the 6 minutes, the tubes were removed from the incubator and the brain tissue was dissociated mechanically by pipetting up and down ∼13 times with a 1-mL pipette tip (e.g., Fisher Scientific 12-111-164) cut at 0.5 cm. Four additional rounds of incubation and mechanical dissociation were performed. In the second round, mechanical dissociation was performed with a 1-mL pipette tip cut at 0.2 cm, and in subsequent rounds, mechanical dissociation was performed with an uncut 1-mL pipette tip. Papain digest was limited to 5 rounds (30 minutes) total to avoid pre-apoptotic cells. After this digest, cells were transferred into 5-mL FACS tubes with strainer caps (Corning, 352235), using sufficient pressure to make sure that the cells passed through the strainer. Then 3 mL of washing solution for single-cell suspension was passed through the strainer to clean the strainer and stop the enzymatic reaction. Washing solution for single-cell suspension was prepared by adding 975 µL of glucose (e.g., Sigma, G7021-1KG) 15% (weight/volume) in Milli-Q water, 250 µL of HEPES 1 M (e.g., Sigma-Aldrich, H3375-100G; dissolved in Milli-Q water), and 2.5 mL of FBS (fetal bovine serum, e.g., Millipore Sigma, F2442-500ML) to 46.9 mL DPBS (Dulbecco’s phosphate-buffered saline; Thermo Scientific, 14190094), stored for up to 2 months at 4°C, and sterilized with a 0.2-0.22-µm-pore-size filter. After addition of washing solution, contents of the FACS tubes were poured into 50 mL conical tubes (e.g., Thermo Fisher Scientific, 14-432-22) for subsequent centrifugation.

The conical tubes containing the cell suspensions were centrifuged at 500g for 10 minutes at 4°C, and the supernatants were carefully decanted and discarded. Debris was removed using the debris-removal solution, according to the manufacturer’s protocol. Cells were resuspended carefully in each tube with 15.5 mL DPBS. Then 4.5 mL of cold debris-removal solution (Miltenyi Biotec, 130-109-398), was added and gently mixed by inverting. This solution was gently overlaid (drop by drop) with 20 mL of cold DPBS (creating two phases), before centrifugation at 3000g for 10 minutes at 4°C, with full acceleration and full brake. After centrifugation, three phases formed; the top two phases were aspirated and discarded. If the phases were hard to distinguish, a conservative approach was taken and the top 20 mL was aspirated and discarded. Then each tube was filled up with cold DPBS (20 mL), closed, mixed by inverting three times with no vortexing and centrifuged at 1000g for 10 minutes at 4°C. The supernatant was aspirated and discarded, and the pellet was resuspended in 500 μL PBS to generate the single-cell suspension.

FACS for single-cell RNA sequencing

Single-cell suspensions were moved to 1.2-mL FACS library tubes (e.g., VWR, 83009-678), to which 5 μL Zombie NIR™ live/dead stain (BioLegend, 423105) diluted 1:10 in PBS was added. These tubes were incubated 15 minutes at room temperature in the dark, and then centrifuged at 300g for 5 minutes at 4°C. For each tube, the supernatant was removed to 100 μL, and the pellet was resuspended in 400 μL of PBS with 0.5% BSA (diluted from 35% BSA; Sigma-Aldrich A7979-50ML).

The contents of each tube were strained through a 5-mL FACS tube with strainer cap, vortexed to minimize cell clumping, and then run on a Sony MA900 cell sorter at about 1500-2000 events per second, with FSC threshold = 6%, FSC gain = 12, BSC gain = 35%, and all fluorophores gain = 40%. Compensation was not used due to the presence of only two fluorophores (oScarlet, same spectra as mScarlet; and Zombie NIR™) whose spectra were determined to be sufficiently non-overlapping per Thermo SpectraViewer. For each sample, 100,000 events were recorded and .fcs files were exported for analysis in FlowJo (v10.0.0). The flow cytometry plot in Fig. 1e was on Live cells as identified from the gating strategy shown in Fig. 1 – Fig. Supplement 1e. oScarlet^HIGH cells were defined as those having a FL3-A::mCherry-A (uncompensated) intensity of approximately >800 (October 2024 experiment) or >1000 (August 2025 experiment). oScarlet^LOW cells were defined as those having a FL3-A::mCherry-A (uncompensated) intensity of approximately <500.

Each sample from which oScarlet^HIGH cells were collected was run in its entirety on Normal mode (maximizes yield) to collect as many oScarlet^HIGH cells as possible (with yields ranging from approximately 35,000-65,000 oScarlet^HIGH cells according to Sony MA900 event count). For the Live cells collected from wildtype brains in the October 2024 experiment and the oScarlet^LOW cells collected from young adult SP-oScarlet fish in the August 2025 experiment, these were also run in Normal mode, but approximately 240,000 and 167,000 cells were collected respectively (according to Sony MA900 event count), which did not represent the entire sample, to avoid overloading the 10x chip in single-cell RNA sequencing. Cells were collected in 500 µL of RNase-free PBS with 1% BSA and 0.1% glucose in 1.5 mL DNA LoBind tubes (Eppendorf, 0030108418).

Single-cell RNA sequencing

After cell collection, tubes containing the cells were centrifuged at 300g for 5 minutes at 4°C. Supernatants were removed to 43.2 μL, and Master Mix was prepared and added to each sample exactly following the instructions in the 10x Genomics Chromium Next GEM Single Cell 3’ Reagent Kits v3.1 User Guide. Cells were not counted, in order to load all cells collected in each tube. Once Master Mix was added, the contents of each tube were pipetted up and down (to resuspend the cells fully) and loaded in their entirety onto the 10x chip.

After this, the instructions in the 10x Genomics Chromium Next GEM Single Cell 3’ Reagent Kits v3.1 User Guide were followed exactly with 11-12 cycles used in the cDNA amplification PCR and 13-14 cycles used in the Sample Index PCR (October 2024 experiment, 12 and 13 cycles, respectively; August 2025 experiment; 11 and 14 cycles, respectively).

Libraries were sequenced by Novogene on a NovaSeqX Plus sequencer (150 bp paired-end reads) to an approximate depth of 215-333 million reads per sample.

Identifying homologous genes across species

Killifish gene names were converted to their equivalents in human, mouse, or other species using the table presented in Supplementary Table 5, which was generated as described in the next paragraph. Killifish can have multiple paralogs corresponding to a single mammalian gene, due to whole-genome duplication in teleost fish^174,175. In such cases, for visualization purposes, one killifish paralog was selected (specified in figure legend). In cases where one killifish paralog had a much higher expression level than the other paralogs, the highest-expressed killifish paralog was selected.

Over half of the killifish NCBI gene names (11,715 out of 22,194) consist of locus numbers. Therefore, we generated new gene symbols for all the killifish protein coding genes, based on NCBI gene descriptions for killifish and human, and orthologs between killifish and other animals. For ortholog identifications, we ran sequence comparison using BLAST (ncbi-blast-2.5.0+) between killifish and 36 other animals, including multiple tetrapod, and fish species. We then generated a prioritized scheme for gene symbol assignment based on the following rules. We kept the gene symbols assigned by NCBI annotation. If there were no gene symbol assigned by NCBI, we matched the NCBI gene description between killifish and human and assigned the human gene symbols where the gene descriptions matched between the two species. For rest of the genes, we generated a consensus gene symbol based on ‘majority rules’ using ortholog symbols from 36 other animals. In case of ties, priority was given to the tetrapod symbols. We were able to assign proper gene identities to 93% of the locus numbers and improve gene symbols for 96.4% of all protein coding genes.

In the text, figures, and legends of this manuscript, all killifish gene names are capitalized and presented in italics (e.g., ELAVL3).

Analysis of single-cell RNA sequencing data

Demultiplexed reads provided by Novogene were aligned to the African turquoise killifish genome (Nfu_20140520) using Cell Ranger. A custom genome including the SP-oScarlet knockin insertion (as determined from sequencing as in Fig. 1 – Fig. Supplement 1a) as a separate chromosome was built according to the instructions provided by 10x Genomics (https://www.10xgenomics.com/support/software/cell-ranger/latest/tutorials/cr-tutorial-mr, “Add a custom gene to the FASTA and GTF”) for alignment of reads from SP-oScarlet fish, although we note that the SP-oScarlet transcript was difficult to detect even in ELAVL3+ cells. This is potentially explained by the SP-oScarlet transcript’s position several kilobases upstream of the 3’ end of ELAVL3, as ELAVL3 has a long untranslated region. The oScarlet transcript could be detected by HCR in ELAVL3+ cells in situ (Fig. 1 – Fig. Supplement 1c).

After alignment, raw_feature_bc_matrix outputs were downloaded and a Seurat¹⁷⁶ (v5.3.0) object was created (separately for each sample) using the Read10x and CreateSeuratObject functions (min.cells = 3, min.features = 200). For quality control, each of these Seurat objects was individually filtered to include cells with nCount_RNA > 1000, nFeature_RNA > 500, and <10% mitochondrial reads. No doublet removal was performed. After testing various doublet removal methods in preliminary analyses and ensuring that our key conclusions were robust to doublet removal method, we reasoned that due to the lack of reliable killifish cell type markers (particularly for myeloid cells), it would be challenging to assess the accuracy of singlet vs. doublet classifications. Quality control metrics per individual Seurat object are listed in Supplementary Table 6. Seurat objects were merged as appropriate. After each merge, JoinLayers was run on the merged object to allow Seurat to identify cluster markers and differentially expressed genes.

For all objects, merged or unmerged, clusters were defined by running NormalizeData (normalization.method = “LogNormalize”, scale.factor = 10000), FindVariableFeatures (selection.method = “vst”, nfeatures = 2000), ScaleData, RunPCA, RunUMAP (dims = 1:30), FindNeighbors (dims = 1:30) and finally FindClusters (resolution = 0.1-0.2). Cell types were defined by running FindMarkers for low-resolution clusters and manually matching markers to cell types based on literature, most notably Ayana et al., 2024⁶⁸ (killifish-specific) and Tabula Muris⁷⁸. Full lists of low-resolution cluster marker genes among all young adult cells from the August 2025 experiment can be found in Supplementary Table 7.

For subsetting to macrophages, APOEB+ IBA1+ LCP1+ clusters were selected based on expression from VlnPlot. These markers were used by Ayana et al. to define “Microglia” in their wildtype killifish forebrain single-cell RNA sequencing datasets⁶⁸, although our analysis shows that APOEB+ IBA1+ LCP1+ killifish brain cells have an expression profile more similar to that of mouse monocyte-derived macrophages rather than mouse microglia.

Young adult wildtype killifish forebrain macrophages were identified by downloading the dataset SRR24058857 (from Ayana et al., 2024⁶⁸) from NCBI Gene Expression Omnibus/Sequence Read Archive, aligning it to the Nfu_20140520 genome using Cell Ranger, and creating a Seurat object and performing quality control exactly as described above for this study’s original datasets, and then subsetting to the (one) APOEB+ IBA1+ LCP1+ cluster.

To compare killifish and mouse brain macrophages, we used a Seurat object corresponding to all young adult mouse brain immune cells analyzed by Barr et al., 2025 (see Barr et al., 2025⁷⁵ for details of single-cell RNA sequencing, genome alignment, quality control, annotation of cell types, etc.). This object, which includes cells collected at multiple circadian timepoints, was first subset to the circadian timepoint ZT12 (Time = ZT12), which corresponds to the beginning of the active phase (roughly equivalent to the circadian timepoint at which our killifish single-cell data was collected). The object was then subset to microglia, BAMs, and MDMs (Broad_Cell_Annotation %in% c(“Microglia”, “PBM”, “MDM”).

Then, we determined markers of mouse microglia, BAMs, and MDMs using FindAllMarkers (test.use = “wilcox”, min.pct = 0.1, logfc.threshold = 0.5). All markers of microglia, BAMs, or MDMs, including genes that were markers of multiple cell types (or all killifish homologs of these genes) were used for principal component analysis (PCA). This PCA was performed starting with a Seurat object for each of the following cell populations:

1. Young adult mouse brain macrophages (subset from all young adult mouse brain immune cells; Barr et al., 2025⁷⁵; Broad_Cell_Annotation %in% c(“Microglia”, “PBM”, “MDM”, “IFN-M”))

2. Young adult oScarlet^HIGH cells (this study)

3. Young adult wildtype killifish forebrain macrophages (Seurat object generated as described above; Ayana et al., 2024⁶⁸)

and subsetting to only the marker genes as described above, then re-running NormalizeData (normalization.method = “LogNormalize”, scale.factor = 10000), FindVariableFeatures (selection.method = “vst”, nfeatures = 2000), ScaleData, and finally RunPCA.

The strongest markers of mouse microglia, BAMs, and MDMs as depicted in the heatmaps in Fig. 2 – Fig. Supplement 3a and Fig. 2 – Fig. Supplement 4a were defined using using FindAllMarkers (test.use = “wilcox”, min.pct = 0.5, logfc.threshold = 3).

DimPlot was used to highlight cells by orig.ident (i.e., sample of origin) or cell type (reduction = “umap”) and produce PCA plots (reduction = “pca”). FeaturePlot and VlnPlot were used to highlight expression of specific genes. VlnPlot was also used to highlight sum log-normalized expression of genes in specific Gene Ontology terms. DoHeatmap was used to highlight expression of top killifish cell type marker genes as ranked by area under curve according to Presto¹⁷⁷ (no thresholds applied; cell types defined manually as described above) or to highlight expression of killifish homologs of the strongest mouse microglia, BAM, and MDM markers. Full lists of cell type marker genes among all young adult cells from the August 2025 experiment as determined by Presto can be found in Supplementary Table 8.

Differentially expressed genes between samples were determined using FindMarkers (test.use = “wilcox”, min.pct = 0.1, logfc.threshold = 0.5), and Gene Ontology terms upregulated among these genes were determined using enrichR (“GO_Biological_Process_2025”) with a background of all human genes with killifish homologs represented among the rownames of the combined Seurat object (containing both samples) and plotted using ggplot2. Differentially expressed genes were always calculated at the level of killifish genes, then converted to human gene names for Gene Ontology analysis (enrichR disregards duplicates). The sizes of the circles in Fig. 4d and Fig. 4 – Fig. Supplement 1d correspond to the numbers of unique human genes with any killifish homologs among the differentially expressed genes for the comparison made. As brains from n = 3-4 individual male fish were pooled to achieve sufficient numbers of oScarlet^HIGH cells, statistics were calculated at the per-cell level.

Full lists of differentially expressed genes per comparison made can be found in Supplementary Tables 9 and 12. Full lists of Gene Ontology terms upregulated per condition can be found in Supplementary Tables 10-11 and 13-14.

Dissociation of killifish brains for ex vivo engulfment experiments

To assess acute endocytic capacity of killifish brain macrophages, we performed ex vivo engulfment of ovalbumin-Alexa Fluor^® 647 (a 45 kDa protein) (Fisher Scientific, O34784) as well as dextran-Oregon Green 488 (a 70 kDa polysaccharide) (Thermo Fisher Scientific, D7173). For ex vivo experiments, fish were always sacrificed in the morning on a weekday, having not received feeding since the previous day.

Dissociation protocols largely followed those described by Mariën et al., 2024¹⁷³. Fish were sacrificed in ice water and brains were dissected in ice-cold PBS using forceps, starting with removal of the lower jaw, then the upper part of the head and eyes, and finally freeing the brain from the skull. After dissection, brains were cut into pieces approximately the size of a half telencephalon. Per sample (each sample consisted of 1 brain), all tissue pieces were transferred into a 1.5-mL microcentrifuge tube containing 500 µL of ice-cold dissection medium (DMEM/F12 + HEPES) and put on ice.

Once all the dissected tissues were ready in dissection medium on ice, fresh papain solution was prepared. Papain solution consisted of 50 µL of papain (suspension, from papaya latex), 20 µL DNase I solution (10 mg/mL; dissolved in Milli-Q water), and 40 µL of L-cysteine solution (12 mg/mL) per 1 mL of DMEM/F12, with the appropriate substrates added at 25 µg/mL each; see details of individual experiments below, “Ex vivo engulfment of ovalbumin” and “Ex vivo engulfment of dextran”). This solution was sterilized through a 0.2-0.22-µm-pore-size filter and vortexed to mix. L-cysteine solution was prepared by adding 120 mg of L-cysteine hydrochloride to 10 mL of filtered, autoclaved Milli-Q water, and stored for up to 2 weeks at 4°C or 1 year at −20°C.

The dissection medium was removed by pipetting and 0.5 mL fresh papain solution was added to each sample. To start digesting the brain tissue, tubes with the samples were placed in an incubator for 6 minutes at 37°C. Tubes were covered with foil for all incubations, due to the use of light-sensitive fluorescent substrates. After the 6 minutes, the tubes were removed from the incubator and the brain tissue was dissociated mechanically by pipetting up and down 13 times with a 1-mL pipette tip cut at 0.5 cm. Four additional rounds of incubation and mechanical dissociation were performed. In the second round, mechanical dissociation was performed with a 1-mL pipette tip cut at 0.2 cm, and in subsequent rounds, mechanical dissociation was performed with an uncut 1-mL pipette tip. Papain digest was limited to 5 rounds (30 min) total to avoid pre-apoptotic cells. After this digest, cells were transferred into FACS tubes with strainer caps, using sufficient pressure to make sure that the cells passed through the strainer. Then 2 mL of washing solution for single-cell suspension was passed through the strainer to clean the strainer and stop the enzymatic reaction. Washing solution for single-cell suspension was prepared by adding 975 µL of glucose 15% in water, 250 µL of HEPES 1 M (dissolved in Milli-Q water), and 2.5 mL of FBS to 46.9 mL DPBS, stored for up to 2 months at 4°C, and sterilized with a 0.2-0.22-µm-pore-size filter. After addition of washing solution, contents of the FACS tubes were transferred to 15 mL conical tubes for subsequent centrifugation. It is easier to see the pellet in conical tubes than in FACS tubes; while a wide variety of tubes are likely suitable for these centrifugation steps, we note that choice of tube can affect yield.

The conical tubes containing the cell suspensions were centrifuged at 500g for 10 minutes at 4°C, and the supernatants were carefully decanted and discarded. Debris was removed using the debris-removal solution, according to the manufacturer’s protocol. Cells were resuspended carefully in each tube with 1550 µL DPBS. Then 450 µL of cold debris-removal solution was added and gently mixed by inverting. This solution was gently overlaid (drop by drop) with 2 mL of cold DPBS (creating two phases), before centrifugation at 3000g for 10 minutes at 4°C, with full acceleration and full brake. After centrifugation, three phases formed; the top two phases were aspirated and discarded. If the phases were hard to distinguish, a conservative approach was taken and the top 2 mL was aspirated and discarded. Then each tube was filled up with cold DPBS (2 mL), closed, mixed by inverting three times with no vortexing and centrifuged at 1000g for 10 minutes at 4°C. The supernatant was aspirated and discarded, and the pellet was resuspended in 500 μL PBS with Zombie Aqua™ (BioLegend, 423101) diluted 1:1000 to generate the single-cell suspension.

Flow cytometry and data analysis for ex vivo engulfment experiments

To evaluate ex vivo engulfment by flow cytometry, 500 μL of single-cell suspension per dissociated brain were moved from 15 mL conical tubes to 1.2-mL FACS library tubes for incubation with Zombie Aqua™ live/dead dye. The suspensions were incubated 15 minutes at room temperature in the dark, and then centrifuged at 300g for 5 minutes at 4°C. For each tube, the supernatant was removed down to 100 μL, and the pellet was resuspended in 400 μL of added PBS with 0.5% BSA.

The contents of each tube were then strained through a 5-mL FACS tube with strainer cap, vortexed to minimize cell clumping, and then run on a Sony MA900 cell sorter at about 1500-2000 events per second (for some more dilute samples, 1000 events per second), with FSC threshold = 6%, FSC gain = 12, BSC gain = 35%, and all fluorophores gain = 40%. Single-color compensation controls were used (see details for individual experiments below). For each sample, 250,000 events were recorded and .fcs files were exported for analysis in FlowJo (v10.0.0). The gating strategy used to identify Live cells was similar to that shown in Fig. 1 – Fig. Supplement 1e, except with Zombie Aqua™ as the live/dead dye rather than Zombie NIR™. Scale data from Live cells were exported as .csv files that were analyzed in R (v4.5.1) for quantification.

Development of an ex vivo engulfment assay for ovalbumin

Data for ovalbumin engulfment were acquired over three experiments. In all experiments, ovalbumin-AlexaFluor 647 [ovalbumin] was used at 25 µg/mL (diluted from a 1 mg/mL ovalbumin stock dissolved in PBS) and the fluorophores used on the Sony MA900 sorter were FL3::mCherry (for oScarlet), FL7::BrilliantViolet510 (for Zombie Aqua™ live/dead dye), and FL10::Alexa Fluor^® 647 (for ovalbumin).

Importantly, in optimizing the ex vivo ovalbumin engulfment assay, we sorted ovalbumin+ cells and verified by confocal microscopy that ovalbumin was localized inside cells rather than binding to the surface of cells.

In the first experiment, four male 41 days old (young adult) and four male 160 days old (old) heterozygous SP-oScarlet killifish were used, as well as single-color compensation controls. The compensation matrix used is included in Supplement 2. Ovalbumin was added to a portion the papain solution after syringe filtration. In this experiment, ovalbumin+ cells were defined as having ovalbumin (FL10-A::Alexa Fluor^® 647-A, compensated) intensity of >1000.

In the second experiment, four female 57 days old (young adult) and four female 161 days old (old) heterozygous SP-oScarlet killifish were used, as well as single-color compensation controls (independent compensation controls were used for this experiment). The compensation matrix used is included in Supplement 3. Ovalbumin was added to a portion the papain solution after syringe filtration. In this experiment, ovalbumin+ cells were defined as having ovalbumin (FL10-A::Alexa Fluor^® 647-A, compensated) intensity of >1000.

In the third experiment, two male 63 days old (young adult), and two male 148 days old (old), two female 63 days old (young adult), and two female 167 days old (old) heterozygous

SP-oScarlet killifish were used. No new compensation samples were used in this experiment, as the sex-respective compensation matrices from the previous experiments were re-used. Unlike in the previous experiments, ovalbumin was added to the papain solution before syringe filtration. In this experiment, ovalbumin intensity values were generally lower than observed in the previous two experiments and ovalbumin+ cells were defined as having ovalbumin (FL10-A::Alexa Fluor^® 647-A, compensated) intensity of >500.

Our conclusions were not affected by whether ovalbumin was added to the papain solution before or after filtration. However, there was likely some ovalbumin loss when ovalbumin was added before filtration in the third experiment, and the effective ovalbumin concentration in the ex vivo engulfment assay was likely lower than the 25 µg/mL reported.

Data on brightness of oScarlet^HIGH cells in young adult and old fish were from these ovalbumin engulfment experiments. oScarlet^HIGH and oScarlet^LOW cells were defined as having oScarlet (FL3-A::mCherry-A, compensated) intensities of >1000 (>500 in Fig. 5 – Fig. Supplement 2b) and <500, respectively. Values presented in Fig. 5c-d, Fig. 5 – Fig. Supplement 1c-d, and Fig. 5 – Fig. Supplement 2b-c were normalized to the average value among for all control samples (always presented as the leftmost category on the x-axis) of the same respective sex and experiment day.

Ex vivo engulfment assay for dextran

Data for dextran engulfment were acquired in one experiment. Dextran-Oregon Green 488 (70 kDa) [dextran] was used at 25 µg/mL (diluted from a 1 mg/mL dextran stock dissolved in PBS) and the fluorophores used on the Sony MA900 sorter were FL1::Alexa Fluor^® 488 (for dextran), FL3::mCherry (for oScarlet), and FL7::BrilliantViolet510 (for Zombie Aqua™ live/dead dye).

Four male 56 days old (young adult) and four male 156 days old (old) heterozygous SP-oScarlet killifish were used, as well as single-color compensation controls run the previous day. The compensation matrix used is included in Supplement 4. On the day the eight experimental fish were run, dextran was added to the papain solution before syringe filtration.

In this experiment, dextran+ cells were defined as having dextran (FL1-A::Alexa Fluor^® 488-A, compensated) intensity of >800. oScarlet^HIGH and oScarlet^LOW cells were defined as having oScarlet (FL3-A::mCherry-A, compensated) intensities of >1000 (>500 in Fig. 5 – Fig. Supplement 2d) and <500, respectively. Values presented in Fig. 5e-f, Fig. 5 – Fig. Supplement 1e-f, and Fig. 5 – Fig. Supplement 2d-e were normalized to the average value among for all control samples (always presented as the leftmost category on the x-axis) of the same respective sex and experiment day.

Statistics

No statistical methods were used to determine sample sizes. Sample sizes were determined based on previous experiments with similar types of assays in mice and killifish. No randomization or blinding methods were used, but for dissociation-based experiments, each processing step was carried out in an alternating-by-condition order (e.g., young adult, then old, then young adult, etc.) to minimize the effects of order on the results. All image quantification was automated with scripts in QuPath (v6.0.0). All statistical tests were performed in R (v4.5.1) and plots made with ggplot2 or Seurat. Unless otherwise noted in the figure legends, Wilcoxon rank-sum tests were used to determine statistical significance (paired for matched comparisons; unpaired for all other comparisons). Numbers of experimental replicates, numbers of animals per condition, and whether statistical tests were performed at the per-animal or per-cell level are indicated in individual figure legends.

Figure supplements

Figure 1 – Figure Supplement 1

Figure 1 – Figure Supplement 2

Figure 2 – Figure Supplement 1

Figure 2 – Figure Supplement 2

Figure 2 – Figure Supplement 3

Figure 2 – Figure Supplement 4

Figure 3 – Figure Supplement 1

Figure 4 – Figure Supplement 1

Figure 5 – Figure Supplement 1

Figure 5 – Figure Supplement 2

Data availability

Sequencing reads are publicly available at NCBI GEO (GEO accession: GSE329891). Seurat objects generated in this study are available at Zenodo (record: 19672958). Other raw data (images, etc.) will be made available upon request. Code is available at Github (https://github.com/RNagvekar/killifish_brain_macrophages).

Acknowledgements

We thank Jason W. Miklas, Nimrod Rappoport, and Daniel J. Richard for critical reading of the manuscript; Natalie Schmahl and Sarah R. Boyle for help with killifish colony management; Jacob Chung, Rishad Khondker, Maddie Housh, Jason Yang, and Ian H. Guldner for help with killifish husbandry; Justin You, Aleksandra S. Tsenter, Merve Heinzer-Avar, Daniel Heinzer-Avar, and Julliana Ramirez-Matias for advice on brain sectioning and imaging analysis; Giulia Notarangelo, Andy P. Tsai, and Jaeyoon Lee for advice on flow cytometry; Olivia Y. Zhou for advice on design of single-cell RNA sequencing experiments, Nimrod Rappoport and Daniel J. Richard for advice on single-cell RNA sequencing data analysis; Alec J. Walker and Constanze Depp for advice on interpretation of single-cell RNA sequencing data; Ryann M. Fame and Blake J. Laham for advice on brain injections; Katrin I. Andreasson, Kyle G. Daniels, Andrew Z. Fire, Daniel F. Jarosz, and Judith Frydman, Charu Ramakrishnan, Lu Zhou, Gil Vantomme, Katharina Papsdorf, Eric D. Sun and all of the Brunet lab for helpful discussions and advice; and Nimrod Rappoport and Daniel J. Richard for code-checking. This project was supported, in part, by Award Number 1S10OD034400-01A1 from the National Center for Research Resources (NCRR). Its contents are solely the responsibility of the authors and do not necessarily represent the official views of the NCRR or the National Institutes of Health. Funding for this study was provided by the Simons Foundation (to A.B., B.S., T.W.C., and P.M.), the Knight Initiative for Brain Resilience (to A.B., T.W.C., and R.D.N.), the Glenn Foundation for Medical Research (to A.B.), the NOMIS Foundation (to A.B. and T.W.C.), the National Institute on Aging K99 Pathway to Independence Award (to R.D.N. and C.N.B.), the Simons Foundation Fellows-to-Faculty Award (to J.B.J.), and the Stanford Knight-Hennessy Scholars (to R.N.).

Additional information

Author contributions

This study was conceived by R.N., R.D.N., C.N.B., and A.B. R.N. designed the SP-oScarlet killifish line, and R.D.N. generated this line with help from R.N.. C.N.B. designed probes for some hybridization chain reaction experiments and checked code. R.N. designed and performed all experiments and analyses unless noted. A.N.P. performed structured illumination microscopy for LAMP1 imaging and helped with brain dissociations and FACS for ex vivo engulfment assays and single-cell RNA sequencing. P.R.K. (under the guidance of B.W.) performed cell type evolution analysis that aided interpretation of single-cell RNA sequencing data. H.J.B. (under the guidance of B.S.) helped design ex vivo engulfment assays and interpret single-cell RNA sequencing data, including by sharing unpublished mouse single-cell RNA sequencing datasets. A.M.M.J. (under the guidance of K.R.H.) performed brain clearing and 3D imaging. S.V.P. helped with brain injections. E.K.C. (under the guidance of T.W.C.) and J.C. performed sulfo-NHS-biotin perfusion for vasculature colocalization analysis. J.C. also helped with sample processing for single-cell RNA sequencing. F.B. helped with brain dissociations and flow cytometry for ex vivo engulfment assays and checked code. L.A.S. (under the guidance of S.R.Q.) and P.N.N. helped with 10x Genomics single-cell RNA sequencing. J.B.J. (under the guidance of P.M.) helped develop the brain injection protocol. E.K.C. (under the guidance of T.W.C.) and R.B. helped develop the FACS sorting strategy. R.B. also helped with brain sectioning and immunofluorescence staining optimization. P.P.S. produced the gene names conversion table used to identify homologs across species. The manuscript was written by R.N. and A.B., and all authors provided comments.

Funding

HHS | NIH | National Institute on Aging (NIA) (K99AG078499)

• Ravi D Nath

• Claire Nicole Bedbrook

HHS | NIH | National Institute on Aging (NIA) (K99AG076879)

• Ravi D Nath

• Claire Nicole Bedbrook

Simons Foundation (SF) (AN-NC-AB-Research-00811219)

• Anne Brunet

Glenn Foundation for Medical Research (GFMR) (SPO123100)

• Anne Brunet

NOMIS Stiftung (NOMIS Foundation) (Agreement-AB-26.2023)

• Anne Brunet

Knight Initiative for Brain Resilience

• Ravi D Nath

• Anne Brunet

• Tony Wyss-Coray

Knight-Hennessy Scholars

• Rahul Nagvekar

National Center for Research Resources (1S10OD034400-01A1)

• Angela N Pogson

Additional files

Supplementary Tables. Supplementary Table 1. Animals used in this study. Supplementary Table 2. Nucleic acid sequences used in this study. Supplementary Table 3. Antibodies used in this study. Supplementary Table 4. Images taken in this study. Supplementary Table 5. Gene names conversion table used to identify homologs across species in this study. Supplementary Table 6. Quality control metrics for Seurat objects generated from raw reads of killifish single-cell RNA sequencing data in this study. Supplementary Table 7. Cluster marker genes based on all young adult cells from August 2025 single-cell RNA sequencing experiment. Supplementary Table 8. Cell type marker genes based on all young adult cells from August 2025 single-cell RNA sequencing experiment. Supplementary Table 9. Differentially expressed genes between young adult and old oScarlet^HIGH cells from August 2025 single-cell RNA sequencing experiment. avg_log2FC > 0 indicates enrichment in old oScarlet^HIGH cells. avg_log2FC < 0 indicates enrichment in young adult oScarlet^HIGH cells. pct. 1 corresponds to the percentage of old oScarlet^HIGH cells with detectable expression of the gene. pct. 2 corresponds to the percentage of young adult oScarlet^HIGH cells with detectable expression of the gene. Supplementary Table 10. Gene Ontology: Biological Process (2025) terms upregulated in young adult oScarlet^HIGH cells from August 2025 single-cell RNA sequencing experiment. Supplementary Table 11. Gene Ontology: Biological Process (2025) terms upregulated in old oScarlet^HIGH cells from August 2025 single-cell RNA sequencing experiment. Supplementary Table 12. Differentially expressed genes between old oScarlet^HIGH and old wildtype macrophages from October 2024 single-cell RNA sequencing experiment. avg_log2FC > 0 indicates enrichment in old oScarlet^HIGH macrophages. avg_log2FC < 0 indicates enrichment in old wildtype macrophages. pct. 1 corresponds to the percentage of old oScarlet^HIGH macrophages with detectable expression of the gene. pct. 2 corresponds to the percentage of old wildtype macrophages with detectable expression of the gene. Supplementary Table 13. Gene Ontology: Biological Process (2025) terms upregulated in old oScarlet^HIGH macrophages from October 2024 single-cell RNA sequencing experiment. Supplementary Table 14. Gene Ontology: Biological Process (2025) terms upregulated in old wildtype macrophages from October 2024 single-cell RNA sequencing experiment.

Source Data Tables. Source Data Table 1. Corresponding to Fig. 3a. Source Data Table 2. Corresponding to Fig. 3 – Supplementary Fig. 1a. Source Data Table 3. Corresponding to Fig. 3 – Supplementary Fig. 1b. Source Data Table 4. Supporting information for analysis in Fig. 3b. Source Data Table 5. Supporting information for analysis in Fig. 3b. Source Data Table 6. Corresponding to Fig. 3b. Source Data Table 7. Corresponding to Fig. 5a, Fig. 5c-d, and Fig. 5 – Fig. Supplement 1c-d. Source Data Table 8. Corresponding to Fig. 5 – Fig. Supplement 2b. Source Data Table 9. Corresponding to Fig. 5 – Fig. Supplement 2c. For this analysis, the threshold for oScarlet^HIGH was set to 0. Source Data Table 10. Corresponding to Fig. 5e-f and Fig. 5 – Fig. Supplement 1e-f. Source Data Table 11. Corresponding to Fig. 5 – Fig. Supplement 2d. Source Data Table 12. Corresponding to Fig. 5 – Fig. Supplement 2e. For this analysis, the threshold for oScarlet^HIGH was set to 0.

Additional Supplements. Supplement 1. SnapGene map file of expected SP-oScarlet insertion. Supplement 2. Compensation matrix from first ex vivo ovalbumin engulfment experiment. Supplement 3. Compensation matrix from second ex vivo ovalbumin engulfment experiment. Supplement 4. Compensation matrix from ex vivo dextran engulfment experiment.