A csf1rb mutation uncouples two waves of microglia development in zebrafish

Microglia are tissue-resident macrophages that play key immune and housekeeping roles in the central nervous system (CNS) (Prinz et al., 2019; Sierra et al., 2019). During development, microglia support neurogenesis by releasing trophic factors (Tong and Vidyadaran, 2016), efficiently engulfing apoptotic neurons (Peri and Nusslein-Volhard, 2008) and pruning supernumerary synapses (Paolicelli et al., 2011). In the adult brain, microglia actively protrude branches to monitor the CNS microenvironment and interact with other cell types in order to maintain homeostasis (Davalos et al., 2005; Nimmerjahn et al., 2005). The indispensable role of microglia in fostering CNS homeostasis becomes evident in human genetic conditions that cause microglia deficits or dysfunctions (Li and Barres, 2018), which can result in severe pathologies such as Nasu-Hakola disease (Paloneva et al., 2002) and adult onset leukoencephalopathy with spheroids (Rademakers et al., 2011). Moreover, microglia are regarded as key mediators of the severe and prolonged inflammatory response triggered by CNS damage, which represents a major therapeutic hurdle in neurodegenerative disorders (Colonna and Butovsky, 2017).

During embryogenesis, microglia arise from yolk sac-derived primitive macrophages, which seed the developing neuroepithelium before the onset of neurogenesis (Alliot et al., 1999; Boche et al., 2012; Cuadros et al., 1993; Herbomel et al., 1999). Lineage-tracing studies performed in the mouse model have shown that these early microglia are maintained throughout life (Ginhoux et al., 2010), although later hematopoietic waves might partially contribute to the adult microglia pool (De et al., 2018). Similar to the mouse, microglia ontogeny in zebrafish initiates from amoeboid-shaped primitive macrophages, which colonize the neural tissue starting at 35 h post-fertilization (hpf) and then differentiate into branched microglia at around 60 hpf (Herbomel et al., 2001). Unlike their mammalian counterparts however, embryo-derived microglia are not maintained in zebrafish, and the adult microglial network is established through a second wave of progenitors that seed the brain parenchyma later during development, fully replacing the initial population by the end of the juvenile stage (Ferrero et al., 2018; Xu et al., 2015). Cell transplantation and fate mapping experiments identified embryonic hematopoietic stem cells (HSCs) arising from the hemogenic endothelium in the dorsal aorta (DA), as the source of adult microglia in this model (Ferrero et al., 2018). Collectively, these findings have opened new avenues of research regarding possible functional differences between the two zebrafish microglial waves, as well as between mouse and zebrafish adult microglia, owing to their distinct cellular origins. Although zebrafish genetic models deficient for each microglia population would facilitate such comparative studies, little is known regarding the genetic regulation of adult microglia ontogeny, and so far no viable mutant resulting in the specific loss of adult microglia has been reported.

The tyrosine kinase colony-stimulating factor 1 receptor (Csf1r), also known as M-CSFR, is a fundamental regulator of mononuclear phagocyte homeostasis in vertebrates (Stanley and Chitu, 2014). It is predominantly expressed in macrophages and their precursors (regardless of their developmental origin), and exhibits pleiotropic effects including cell proliferation, differentiation and survival. Accordingly, CSF1R deficiency in human, mouse and rat leads to a dramatic reduction in tissue macrophage development, including microglia (Erblich et al., 2011; Oosterhof et al., 2018; Pridans et al., 2018; Rojo et al., 2019). Once established, microglia also rely on CSF1R signaling for their maintenance in the brain parenchyma and can be efficiently depleted in the mouse brain through pharmacological blockade of CSF1R (Elmore et al., 2014; Squarzoni et al., 2014). In humans, deficiencies in CSF1R signaling have been associated with neurodegenerative disorders, further highlighting the central role of CSF1R in microglia homeostasis (Oosterhof et al., 2019; Rademakers et al., 2011). In vivo, two non-homologous cytokines serve as ligands for CSF1R: Csf1 and Interleukin 34 (Il34) (Lin et al., 2008; Stanley and Heard, 1977). Both show distinct spatial and cellular distribution in the brain parenchyma (Cahoy et al., 2008; Zeisel et al., 2015) and elicit both overlapping and non-redundant biological responses in regional microglia (Easley-Neal et al., 2019; Greter et al., 2012; Kana et al., 2019; Wang et al., 2012).

As a result of a teleost-specific whole genome duplication, zebrafish possess two paralogs of the csf1r gene: csf1ra and csf1rb (Braasch et al., 2006). Fish deficient in both genes (csf1r^DM) lack microglia from the embryonic to the adult stages (Oosterhof et al., 2018), thus mimicking the Csf1r^−/− mouse phenotype (Dai et al., 2002; Ginhoux et al., 2010). In contrast, individual mutants exhibit a less severe microglial phenotype, characterized by a transient loss of microglia in csf1ra^−/− zebrafish embryos and a moderate reduction of adult microglia in both csf1ra^−/− and csf1rb^−/− single mutants (Oosterhof et al., 2018). Based on these phenotypes, it was suggested that both paralogs exhibit redundant functions. This prompted us to revisit the precise contribution of each paralog to microglia ontogeny, in light of the newly established model of microglia ontogeny in zebrafish in which the two distinct primitive and definitive microglia populations temporally overlap. We previously demonstrated that the kinase insert domain receptor like (kdrl):Cre model offers a powerful tool to discriminate primitive macrophage-derived embryonic microglia from HSC-derived adult microglia in vivo (Ferrero et al., 2018), based on the hemogenic nature of their precursors. Exploiting this approach, we uncovered non-overlapping functions for csf1ra and csf1rb and identified csf1rb as a unique regulator of adult microglia development. In addition, we also demonstrated a specific contribution for the csf1rb paralogue to HSC-derived myelopoiesis, consistent with the specific HSC origin of the adult microglial population.

Previous studies have documented the restricted expression of csf1ra in neural crest-derived cells, early macrophages and microglia during embryonic development (Caetano-Lopes et al., 2020; Herbomel et al., 2001; Parichy et al., 2000) (Fig. 1A-E), as well as its role in primitive myelopoiesis (Herbomel et al., 2001). However, although csf1rb has been previously linked to microglia biology (Mazzolini et al., 2019; Oosterhof et al., 2018), its expression has not been assessed in the context of developmental hematopoiesis and little is known about its specific functions. Using whole-mount in situ hybridization (WISH), we found that csf1rb exhibits an expression profile distinct from that of csf1ra during embryogenesis, with no expression in either neural crest or primitive macrophages. Rather, csf1rb transcripts are first detected at around 30 hpf in the otic vesicle (OV), as well as in a small number of hematopoietic cells in the posterior blood island (PBI) (Fig. 1F). The latter is consistent with expression in erythro-myeloid progenitors (EMPs), which are transiently found in the developing embryo. At 36 hpf, expression of csf1rb increases in the OV, and appears in cells located along the DA, a site involved in the onset of HSC formation (Fig. 1G). Over subsequent stages, expression in the OV disappears but expands in the DA and, at 48 hpf onwards, csf1rb expression is observed in the caudal hematopoietic tissue (CHT), a region of definitive hematopoiesis (Fig. 1H). At 72 hpf, the developing thymus also contains csf1rb-expressing cells, reminiscent of HSC-derived lymphoid progenitor immigrants (Fig. 1I,J). Notably, although csf1ra and csf1rb show distinct spatial expression patterns during development, both transcripts overlap in microglia in the brain and retina starting at 72 hpf (Fig. 1E,J).

Because expression of csf1rb was mainly found in sites of definitive hematopoiesis, we performed WISH in runx1 mutant embryos, which lack HSCs. Although expression in the OV and in the PBI at 30 hpf is normal, we observed a strong reduction of csf1rb transcripts in the DA, CHT and thymus of homozygous embryos, thus identifying csf1rb-expressing cells in these anatomical locations as HSC-dependent (Fig. 1K-O). As expected, microglial expression of csf1rb was not affected in runx1^null embryos, consistent with their ontogenic relationship with primitive macrophages, which are runx1-independent (Ferrero et al., 2018) (Fig. 1O). Collectively, these results indicate that csf1ra and csf1rb have non-overlapping distribution during early development, except for microglia.

We next assessed transcript expression of csf1r paralogs in mononuclear phagocytes isolated from adult tissues. As a source for these studies, we used Tg(mhc2dab:GFP; cd45:DsRed) double transgenic fish, as we previously demonstrated that the mhc2dab:GFP; cd45:DsRed transgene combination enabled the isolation by fluorescence-activated cell sorting (FACS) of pure populations of mononuclear phagocytes (Wittamer et al., 2011), including resident microglia (Ferrero et al., 2018) (Fig. S1). As shown in Fig. 2A, csf1ra was highly expressed in adult microglia, as well as in mononuclear phagocytes isolated from whole kidney marrow (WKM), spleen and skin. In contrast to the ubiquitous expression pattern of csf1ra, we found high levels of csf1rb transcripts in microglial cells, very little expression in WKM macrophages and no expression in skin and spleen macrophages. Analysis of a publicly available WKM single-cell sequencing dataset (Lareau et al., 2017; Tang et al., 2017) also revealed major differences in the expression profiles of the two paralogs across the adult hematopoietic tree. Mirroring the patterns observed during embryogenesis, csf1ra expression was restricted to macrophages, whereas csf1rb transcripts were predominantly found in blood progenitors (Fig. 2B-D). In addition, csf1rb was also expressed in macrophages, as well as a few cells scattered within the neutrophil cluster (Fig. 2B-D). However, as visualized in the scaled-heatmap in Fig. 2E, the relative expression of csf1rb within blood progenitors was at least three times higher than in macrophages. Notably, when compared with progenitors, csf1rb expression appears to be negligible in neutrophils (Fig. 2E). Together with our WISH analyses, these results suggest that csf1rb expression labels blood progenitors through life and identified microglia as a unique population of mononuclear phagocytes to display csf1rb expression outside the WKM.

To study csf1r function in vivo, we used two zebrafish mutant lines with no functional csf1ra or csf1rb gene. The zebrafish panther line carries a point mutation in csf1ra, replacing a valine by a methionine in position 614 (Parichy et al., 2000). This change induces an impaired functioning of the kinase activity of the receptor, resulting in the disruption of internal cell signaling. This model has been previously used to demonstrate the contribution of csf1ra to microglia development (Herbomel et al., 2001; Oosterhof et al., 2018) and constitutes a valuable tool for our investigations. The zebrafish line sa1503 harbors a splice site mutation in the csf1rb gene, leading to the inclusion of 86 nucleotides from intron 11 and a premature stop codon (Fig. S2A). This nonsense mutation results into the synthesis of a truncated protein that lacks the receptor kinase domain and is expected to be non-functional. The absence of normal csf1rb transcripts possibly resulting from the presence of alternative splicing sites was confirmed through RT-PCR and sequencing analyses, thus validating that the mutation fully interferes with splicing (Fig. S2B,C). Homozygous csf1rb^sa1503 fish exhibit normal external morphology and behavior and, like the csf1ra^−/− mutant, survive to adulthood.

Previous studies indicated that csf1ra is not required for early myelopoiesis, as primitive macrophages develop normally in csf1ra^−/− embryos (Herbomel et al., 2001). Extending these analyses, we found no effect on the number of mfap4⁺ primitive macrophages in csf1rb mutants, as determined using WISH (Fig. 3A,B). To study whether both paralogs were simultaneously required for primitive myelopoiesis, we intercrossed the two single mutant lines and derived csf1ra/b double mutant embryos (hereafter referred to as csf1r^DM). As shown in Fig. 3A,B, the complete loss of csf1r had no consequence on primitive macrophage ontogeny, as the number of mfap4⁺ cells in double mutants was similar to that of wild-type and single homozygous mutant embryos. This is consistent with our recent findings using the macrophage mpeg1:EGFP reporter line (Kuil et al., 2020). We next investigated the requirement of the different paralogs for the establishment of embryonic microglia, which differentiate in the brain parenchyma from primitive macrophages starting at 60 hpf (Ferrero et al., 2018; Herbomel et al., 2001). As a readout for microglia differentiation, we used WISH to analyze the expression of apoeb, a microglia signature gene. Quantification of apoeb⁺ cells present in the optic tectum at 72 hpf showed that the number of microglia was dramatically decreased in csf1ra-deficient embryos (0.8±0.4 cells; mean±s.e.m.) when compared with wild-type (20.8±1.4 cells), unaffected in csf1rb-depleted embryos (19.1±1.3 cells) and similarly strongly reduced in csf1r^DM embryos (0.8±0.5 cells) (Fig. 3C,D). These results indicate that independently, csf1ra, and not csf1rb, is important for establishing the first wave of microglia during zebrafish embryogenesis.

Because it was previously reported that csf1ra^−/− embryos exhibit a partial recovery of microglia cells at 6 days post fertilization (dpf) (Herbomel et al., 2001), we next examined the status of microglia in the mutants later during development using WISH. Although the numbers of apoeb⁺ cells were stable from 3 to 6 dpf in wild-type and csf1rb^−/− embryos (∼20 cells per optic tectum), we observed a gradual increase in the number of microglia (from 0.7±0.4 cells at 3 dpf to 9.7±1.1 cells at 6 dpf) in embryos carrying the csf1ra^−/− mutation (Fig. 3E,F). At 6 dpf, microglia cell numbers in csf1ra^−/− embryos accounted for ∼50% of total microglia cells found in sibling controls. Interestingly, at the same developmental stage, repopulation of the brain parenchyma by microglia was not observed in double mutant embryos, which remained devoid of apoeb-expressing cells. This observation indicated that recovery of microglia in csf1ra-deficient embryos is mediated by csf1rb, which suggested there may be a compensatory role for csf1rb in microglia development in the absence of csf1ra. However, when we FACS-sorted mpeg1:EGFP⁺ cells from the heads of 6 dpf wild-type and csf1ra^−/− embryos, we found no significant difference in expression of csf1rb transcripts between both genotypes (Fig. S3). Taken together, these data suggest that the partial recovery of embryonic microglia in csf1ra^−/− embryos is csf1rb-dependent but does not require a compensatory increase in csf1rb mRNA.

We investigated the source of the repopulating microglial cells in csf1ra-deficient embryos. Indeed, microglia recovery in these embryos could result either from a delay of differentiation of primitive macrophages or from the early and atypical contribution of HSCs, the precursors of adult microglia. We discriminated between these two possibilities by crossing the csf1ra mutant line to Tg(kdrl:Cre; bactin2:loxP-Stop-loxP-DsRed^express (also known as ßactin:Switch-DsRed); mpeg1:EGFP) triple transgenics (Fig. 3G). As we previously showed, primitive macrophage-derived embryonic microglia are GFP⁺, DsRed⁻ in this setup (Fig. 3H), whereas mononuclear phagocytes originating from EMPs or HSCs are GFP⁺, DsRed⁺, owing to the hemogenic nature of their precursors (Ferrero et al., 2018). Confocal microscopy analysis of live embryos revealed that GFP⁺ microglia present at 6 dpf in csf1ra-deficient embryos did not express the DsRed transgene, thus demonstrating their lineage relationship with primitive macrophages (Fig. 3I). These findings indicate that recovered microglial cells in the csf1ra mutant share the same cellular origin as their wild-type counterparts and point to a delay of primitive macrophage differentiation as the cause of the observed phenotype.

Like the single homozygous mutants, csf1r^DM are viable and fertile, allowing for investigations into the role of Csf1r signaling in the establishment of definitive microglia. As a way to discriminate between embryonic and adult microglia in our analyses, we relied again on mutant fish carrying the kdrl:Cre; ßactin:Switch-DsRed; mpeg1:EGFP triple transgene (Fig. 4A) and performed confocal analyses of brain sections immunostained for GFP and DsRed. In line with previous findings, the density of GFP⁺ microglia cells in the brain parenchyma was decreased ∼60% in single csf1ra^−/− and csf1rb^−/− mutant fish compared with their wild-type siblings (Fig. 4B,C,D,N) and showed a dramatic reduction (90%) in adult animals lacking both paralogs (Fig. 4E,N). However, analysis of DsRed transgene expression to assess their primitive or definitive identity revealed striking microglial phenotypes (Fig. 4F-O). In wild-type fish, all mpeg1⁺ microglia were DsRed⁺, as expected from their known HSC origin (Fig. 4F,J,O). Similarly, GFP⁺ microglial cells from csf1ra^−/− fish also co-expressed DsRed, indicating that adult microglia ontogeny still occurs in the absence of csf1ra (Fig. 4G,K,O). In contrast, the majority of the remaining GFP⁺ cells in csf1rb^−/− animals were found to be DsRed⁻, thus excluding them as microglia derived from the adult wave (Fig. 4H,L,O). Based on the lack of DsRed expression, these mpeg1:EGFP⁺ cells likely represent residual primitive microglia. This is further supported by observations that in csf1r^DM animals, which lack primitive microglia, the very few cells present in the brain parenchyma all expressed DsRed (Fig. 4I,M,O). Collectively, these data indicate that csf1rb, and not csf1ra, is essential for establishing the definitive wave of microglia in zebrafish.

To characterize the developmental dynamics leading to the observed phenotype, we examined the brains of Tg(kdrl:Cre; ßactin:Switch-DsRed; mpeg1:EGFP) wild-type and csf1ra or csf1rb mutant larvae at 21, 28, 35 and 50 dpf. As we previously reported, this time window encompasses the progressive replacement of GFP⁺ DsRed⁻ primitive microglia by definitive GFP⁺ DsRed⁺ microglia in the brain parenchyma (Ferrero et al., 2018). These kinetic analyses revealed distinct phenotypes among the mutants. Consistent with our previous observations, in wild-type fish the percentage of adult DsRed⁺ microglia steadily increased over time (Fig. 4P). In contrast, the brain of csf1rb mutants remained largely devoid of DsRed⁺ cells at all time points, suggesting that microglial progenitors fail to colonize the CNS in the absence of csf1rb (Fig. 4P). Surprisingly, in csf1ra^null animals we observed a shift in the emergence of adult microglia. At 21 dpf, when GFP⁺ DsRed⁻ primitive microglia are still predominant in wild-type brains, the majority of microglia in csf1ra^−/− fish already express DsRed (Fig. 4P). Based on these observations, we hypothesized that primitive microglia detected in the csf1ra^−/− brain at 6 dpf fail to maintain through the juvenile stage. Interestingly, considering the overall density of mpeg1⁺ microglia across time, irrespective of the origin, we observed that in wild-type individuals the density of microglia increased from 1.7±0.07 cells/mm³ to 2.5±0.2 cells/mm³ between 21 and 50 dpf, mirroring the progressive expansion of the DsRed⁺ cells. In csf1rb mutants, microglia similarly expanded from 1.2±0.1 cells/mm³ to 2.2±0.3 cells/mm³ between 21 and 35 dpf (Fig. 4Q). Given that these cells are from embryonic origin, such findings suggest that a partial compensation takes place in the brain of csf1rb^−/− fish in the absence of DsRed⁺ adult microglia. However, the potential of primitive microglia to compensate for the lack of the adult wave appears to be limited, as cell density dropped to 1.8±0.07 cells/mm³ at 50 dpf (Fig. 4Q) and remained lower than in wild-type fish throughout adulthood (Fig. 4N). The curve of microglia density across time followed a different trend in csf1ra^−/− fish, in which DsRed⁺ cells successfully established in the brain by 21 dpf, but did not undergo the steady expansion that we observed in wild-type. This result suggests that, although dispensable for their ontogeny, csf1ra is likely required for maintaining adult microglia after they colonize the juvenile brain. Overall, we concluded that csf1ra and csf1rb are required for the maintenance and specification of embryonic and adult microglia in zebrafish, respectively, and that individual loss of function of either paralog results in reduced microglia densities in the adult.

We sought to dissect the mechanisms linking csf1rb to adult microglia development. Based on the expression profile of csf1rb in hematopoietic progenitors and the developmental relationship between adult microglia and embryonic HSCs, we hypothesized that csf1rb regulates definitive hematopoiesis in zebrafish. Using WISH, we did not detect any significant alteration in the expression of runx1 in the DA of csf1rb-deficient embryos, indicating normal specification of the hemogenic endothelium (Fig. 5A,B). In addition, at 3 and 6 dpf, c-myb (also known as myb) expression in the CHT and the pronephros, which specifically labels HSCs and progenitors, was not changed in csf1rb^−/− embryos (Fig. 5C-F). This demonstrates that neither the emergence of HSCs nor the maintenance of progenitors during embryonic development requires csf1rb.

We evaluated a possible requirement for csf1rb during HSC differentiation. In the zebrafish embryo, T lymphopoiesis starts at around 50 hpf, with HSC-derived thymocyte precursors migrating to the developing thymus (Hess and Boehm, 2012; Murayama et al., 2006). We found that T cell development was not affected in the absence of csf1rb, as expression of the early T cell marker rag1 was detected in mutant embryos at levels similar to that seen in wild-type (Fig. 5G,H). Next, we assessed the myeloid potential of csf1rb-deficient HSCs. As the different waves of myelopoiesis temporally overlap during embryonic development, we used triple transgenic kdrl:Cre; ßactin:Switch-DsRed; mpeg1:EGFP embryos to discriminate in the CHT between newly born definitive macrophages (GFP⁺ DsRed⁺) and primitive macrophages (GFP⁺ DsRed⁻) having colonized the site from the periphery (Fig. 5I). In wild-type embryos, we found that for the first 48 h of development, all GFP⁺ macrophages present in the CHT (40 cells on average) are derived from primitive hematopoiesis, as indicated by their lack of DsRed expression (Fig. 5J,K). The first definitive macrophages, identified as DsRed⁺ GFP⁺ cells, are detected in the CHT at around 48 hpf (∼3 cells on average) (Fig. 5J,K). This population then slowly increases over time, and at 6 dpf, the CHT contains on average 113 double-positive cells per embryo. By that stage, definitive macrophages in the CHT outnumber primitive macrophages, accounting for up to 80% of the total GFP⁺ population. Having delineated the kinetics of differentiation of definitive macrophages in the CHT, we next performed similar quantification of macrophage numbers in single mutants. In csf1ra-deficient embryos, we found that both the kinetics of appearance and total number of DsRed⁺ GFP⁺ double-positive cells were similar to the wild type, thus indicating that the developmental program of HSC-derived macrophages was not affected (Fig. 5K-M,O). In contrast, csf1rb^−/− embryos exhibited significantly decreased numbers of definitive macrophages at each time point (4 versus 19 at 72 hpf and 19 versus 113 at 6 dpf) (Fig. 5K-O). Collectively, these findings demonstrate that Csf1rb activity specifically supports the embryonic development of HSC-derived macrophages. Of note, genetic complementation tests performed using an independently derived mutant allele of csf1rb fully validated the observed phenotype as a consequence of csf1rb perturbation (Fig. S4). Interestingly, analysis of CHT GFP⁺ primitive macrophages during the 48 hpf to 6 dpf time-window also revealed major phenotypic differences among the mutants. Consistent with previous observations (Herbomel et al., 2001) and our own results suggesting a role for csf1ra in controlling early macrophage invasion in embryonic tissues, we found that the CHT of csf1ra mutants became colonized by 30% fewer primitive macrophages (Fig. 5J). By comparison, the numbers of primitive macrophages present in the CHT of csf1rb mutant embryos were similar to that of the wild type. These findings suggest that in primitive macrophages csf1rb is dispensable for cell migration. Collectively, these results demonstrate that csf1rb, and not csf1ra, is required for definitive myelopoiesis in the zebrafish embryo.

To evaluate whether Csf1rb functions are required for life, we examined WKM cell suspensions from mpeg1:EGFP transgenics by flow cytometry. These analyses revealed that the relative percentage of mpeg1:EGFP^high cells, which identify macrophages in adult fish (Ferrero et al., 2020), was reduced 60% in csf1rb^−/− animals, whereas csf1ra^−/− fish exhibited an intermediate phenotype (30% decrease) when compared with wild type (Fig. 6A,B). Importantly, the presence of macrophages within the fluorescence-negative cell fraction of each mutant line, possibly resulting from the regulation of mpeg1 transgene expression by csf1r, was excluded by expression analyses (Fig. S5). Intriguingly, the mpeg1:EGFP^low population was almost completely absent in csf1rb mutant fish, in contrast to csf1ra^−/− animals that had percentages of these cells similar to those of their wild-type siblings (Fig. 6B). This suggests that the WKM of csf1rb^−/− animals is devoid of mpeg1-expressing B lymphocytes (Ferrero et al., 2020). Consistent with these observations, flow cytometric analysis of the entire WKM revealed the existence of a distinct light-scatter profile in csf1rb mutant fish, characterized by a 40% decrease of both the lymphoid and myeloid scatter fractions and a concomitant two-fold increase of progenitors (Fig. 6C,D). As the myeloid fraction mostly contains mature neutrophils, this suggests a differentiation block at the myeloid progenitor stage, a phenotype compatible with the expression profile of csf1rb in the WKM. Collectively, these findings indicate that the absence of csf1rb results in functional deficiencies in myelopoiesis in the adult, and a concomitant lack of mpeg1-expressing B lymphocytes.

Tissue macrophages constitute a highly heterogeneous compartment, based on their origin and the niche they inhabit (Bennett and Bennett, 2019; Guilliams et al., 2020). CSF1R signaling is a common pathway regulating the development of most macrophages in vertebrates, as demonstrated by their dramatic loss (including microglia) in Csf1r-deficient mice and csf1r-deficient zebrafish (Chitu and Stanley, 2017; Dai et al., 2002; Kuil et al., 2020; Oosterhof et al., 2018). However, although CSF1R is represented only once in mammalian genomes, zebrafish possess two copies of the gene, and their relative contribution to myelopoiesis has remained unknown. Focusing on the specific context of microglia ontogeny, in this study we have thus investigated the functions of each paralog during macrophage development. Although our gene expression analyses demonstrate that csf1rb, like csf1ra, is expressed within the hematopoietic compartment, they also reveal a divergence in their expression profiles, with csf1ra expressed in all tissue macrophages (regardless of their primitive or definitive origin) and csf1rb restricted to microglia and definitive blood progenitors. Given that, in the mouse, Csf1r is expressed throughout the path of macrophage differentiation (from hematopoietic progenitors to mature cells) (Hawley et al., 2018; Sasmono et al., 2003), such complementary patterns suggest that subfunctionalization, a process in which the two gene copies partition the ancestral function (Force et al., 1999), may have contributed to the evolution of this family in zebrafish. Accordingly, csf1ra signaling is required for the establishment of primitive macrophage-derived embryonic microglia (Herbomel et al., 2001; Oosterhof et al., 2018), whereas csf1rb controls the ontogeny of definitive macrophages, including adult microglia. Our work thus demonstrates that csf1ra and csf1rb are jointly required to fulfil the roles of mammalian CSF1R in myelopoiesis.

Interestingly, despite these functional divergences, we also provide evidence that csf1rb is able to compensate, at least partially, for the absence of csf1ra. For example, whereas csf1rb loss has no effect on microglia development in embryos with a functional csf1ra paralog, csf1rb signaling is responsible for the partial recovery of microglia observed in csf1ra^−/− larvae, as indicated by the absence of recovery in csf1r^DM mutants. As microglia repopulating csf1ra^−/− larvae entirely derived from primitive macrophages and did not exhibit a compensatory overexpression of csf1rb, it thus appears that in this setting the basal expression level of csf1rb on embryonic microglia is sufficient for driving the recovery process. Our findings also provide new insights into the functions and complex interplay between csf1r paralogs and the three Csf1r ligands identified in zebrafish: Il34, Csf1a and Csf1b. Similar to the mouse (Greter et al., 2012; Wang et al., 2012), Il34 acting through Csf1ra is thought to control the migration of primitive macrophages to the embryonic neuroepithelium in zebrafish (Kuil et al., 2019; Wu et al., 2018). Accordingly, il34-deficient embryos phenocopy both the microglial loss at 3 dpf and the partial recovery at 5 dpf observed in csf1ra mutants (Herbomel et al., 2001; Kuil et al., 2019). Given that csf1r^DM larvae are completely devoid of microglia at 6 dpf, this suggests that csf1rb-mediated microglia replenishment in csf1ra^−/− larvae is independent of Il34 signaling. Further investigation in zebrafish mutants combining Csf1 ligand and receptor knockout would be instrumental to understand whether either Csf1a or Csf1b signaling through Csf1rb represents an alternative pathway to drive microglia migration into the brain when the Il34-Csf1ra interaction is abrogated.

A major finding of our study is the demonstration that the second microglial wave in zebrafish is completely abolished in absence of csf1rb, thus uncovering a selective role for csf1rb in the establishment of HSC-derived adult microglia. Indeed, although microglia cells are present in each single mutant (albeit at comparably reduced cell density), lineage tracing of definitive microglia development revealed the microglial population present in the brain of csf1rb^−/− adult fish remains of primitive origin. This is in sharp contrast to csf1ra-deficient and wild-type fish, in which the HSC-derived adult population fully replaces the primitive microglial pool. Interestingly, however, although adult microglia develop normally in csf1ra-deficient juvenile fish, their numbers drop towards the adult stage. This suggests that, although being dispensable for the ontogeny of the definitive microglial wave, csf1ra likely contributes to microglia maintenance within the adult brain parenchyma. The opposite microglial phenotypes observed in csf1ra and csf1rb mutants also shed light on potential dynamics between the two microglial waves during development. Through time-course analyses, we found that the incomplete recovery of primitive microglia in csf1ra juveniles is compensated by an earlier establishment of the definitive wave compared with the wild type. Conversely, the lack of definitive microglia in csf1rb mutants results in the primitive pool being retained in the adult. By analogy with the current view established in the mouse model (Guilliams et al., 2020), it is conceivable that competition for the juvenile brain niche regulates the exchange between the two microglial waves, in a scenario in which efficient seeding of the brain by definitive microglia would require the regression of the primitive wave. Another plausible hypothesis is that the adult wave may actively participate to the removal of the primitive population, therefore explaining the maintenance of embryonic microglia in csf1rb-deficient adult animals. Future studies, making use of new and more sophisticated tools, will be required to address these complex questions. Nevertheless, as neither definitive microglia in csf1ra mutants nor primitive microglia in csf1rb mutants achieved the cellular density seen in wild-type adults, other intrinsic or extrinsic factors aside from niche availability are likely to affect microglia homeostasis in the adult brain.

Our investigations into the molecular mechanisms underlying the microglial phenotype of csf1rb mutant fish revealed that HSCs give rise to myeloid cells in a csf1rb-dependent fashion. Indeed, we found that csf1rb-deficient fish display a broad deficit in definitive myelopoiesis, as supported by the decrease of HSC-derived macrophages during embryonic development and in the adult hematopoietic niche. These findings are consistent with the runx1-dependent selective expression of csf1rb on blood progenitor cells throughout life. In addition, csf1rb is dispensable for HSC emergence in the aorta-gonad-mesonephros and, unlike csf1ra, does not appear to control cell migration to the different niches. On the whole, this suggests that the depletion of adult microglia and definitive macrophages in the csf1rb mutant results from a deficit of differentiation at the level of HSC-derived myeloid progenitors. These data add to previous findings in zebrafish (Yu et al., 2017) and mouse (Azzoni et al., 2018) showing that distinct molecular mechanisms regulate the emergence of subsequent macrophage waves. Overall, zebrafish csf1ra and csf1rb mutants may thus provide insightful models for the functional dissection of each microglial population and to better understand microglia development from an evolutionary perspective.

Finally, a surprising finding of our study is that fish lacking the csf1rb paralog also lack a population of mpeg1⁺ B cells in the WKM. As we previously showed that these cells account for the majority of IgM-expressing B lymphocytes in zebrafish (Ferrero et al., 2020), these observations suggest that B lymphopoiesis is globally impaired in csf1rb mutant animals. This is interesting because, although no such phenotype has been reported in adult CSF1R-deficient mice so far, expression of CSF1R was recently identified on a subset of embryonic myeloid-primed B-cell progenitors in the fetal liver, and its loss associated to defective fetal B-cell differentiation in vivo (Zriwil et al., 2016). Although the precise contribution of csf1rb to zebrafish B lymphopoiesis remains to be investigated, our work thus provides further support for a role of CSF1R beyond myelopoiesis in vertebrates.

Zebrafish were maintained under standard conditions, according to Federation of European Laboratory Animal Science Associations guidelines (Alestrom et al., 2019). All experimental procedures were approved by the ethical committee for animal welfare (CEBEA) from the Université Libre de Bruxelles. The following lines were used: Tg(mpeg1:EGFP)^gl22 (Ellett et al., 2011); Tg(kdrl:Cre)^s89 (Bertrand et al., 2010); Tg(actb2:loxP-STOP-loxP-DsRed^express)^sd5 (Bertrand et al., 2010); Tg(ptprc:DsRed)^sd5 (here referred to as cd45:DsRed) and Tg(mhc2dab:GFP_LT)^sd67 (Wittamer et al., 2011); panther^j4e1 (Parichy et al., 2000) (here called csf1ra^−/−); and runx1^w84x (Sood et al., 2010) (here called runx1^−/−). Csf1rb^sa1503 mutants (here called csf1rb^−/−), generated via ethyl-nitrosurea mutagenesis, were obtained from the Sanger Institute Zebrafish Mutation Project. Heterozygous adults from the F3 generation carrying the sa1503 allele were backcrossed to wild-type AB* fish at least four times before being incrossed to obtain homozygous mutants. The following primers were used to identify the point mutation in intron 11 by PCR on genomic DNA: sa1503-F (5′-CTCTCTCTGTGGCAACTCTATGGATG-3′); sa1503-R (5′-CGCTCCTGCTCCAAGAACCTG-3′). The independently derived csf1rb^re01 mutant line, which carries a 4 bp deletion in exon 3 (Oosterhof et al., 2018), was used for the complementation test. Unless specified, the term ‘adult’ fish refers to animals aged between 3 and 6 months.

To extract RNA from the CHT region, the tails of ∼100 embryos were dissected and rapidly dissolved in TRIzol reagent (Qiagen). RNA was isolated using the RNAeasy Mini Kit (Qiagen) and cDNAs were synthesized using the Super Script II reverse transcriptase (Thermo Fisher Scientific). The following primers were used to amplify the region of the csf1rb coding sequence encompassing the mutation site from embryonic cDNA: Forward (5′-CGCTGGAGACGATTGTGACTTCTATAC-3′), Reverse (5′-GACCAGGCCAATAGCAGTTGCTTG-3′).

Embryos were collected at desired stages of development, anaesthetized in E3 medium containing 0.1 mg/ml tricaine (Sigma-Aldrich) and digested with Liberase TM (Roche) in PBS for 1 h at 33°C. Kidney marrow cells were prepared through mechanic resuspension, as previously described (Stachura and Traver, 2011). Other adult organs (spleen, skin and brain) were triturated and treated with Liberase TM at 33°C for 30-60 min. Cells were then filtered through 40-μm nylon mesh and washed with 2% fetal bovine serum in Hank's Balanced Salt Solution by centrifugation (290 g). Just before flow cytometry analysis, Sytox Red (Invitrogen) solution was added to the samples at a final concentration of 5 nM to exclude non-viable cells. Flow cytometry and cell sorting was performed on a FACS ARIA II (Becton Dickinson). Analyses were performed using the FlowJo software (Treestar).

qPCR reactions were performed on a Bio-Rad CFX96^TM real time system (Bio-Rad), according to the manufacturer's instructions. Biological triplicates were compared for each subset. Relative amount of each transcript was quantified via the ΔCt method, using elongation-Factor-1-alpha (eef1a1l1; ENSDARG00000020850) expression for normalization. Primers used are reported in Table S1.

Probes for apoeb (ENSDARG00000040295), runx1 (ENSDARG00000087646), c-myb (ENSDARG00000053666), mfap4 (ENSDARG00000038681), csf1ra (ENSDARG00000102986) and csf1rb (ENSDARG00000053624) were synthesized in vitro. For the csf1rb WISH, we combined two probes hybridizing different parts of the transcript in order to increase signal strength. The following primers were used for the generation of the csf1rb probes from 4 dpf larvae cDNA: Fw1, 5′-ATCATTGCAGTGCTGACCTGTATG; Rv1, 5′-GGTGAGCTCCAGGTGAAGTTGTAG; Fw2, 5′-ATGGCCAACCAATCCATTTCTGAG; Rv2, 5′-AGTAAGCATTCCTTGCGGGATGTT. Zebrafish embryos at the desired stages were fixed in 4% paraformaldehyde (PFA). WISH was performed using standard methods (Thisse and Thisse, 2008).

The sample size was chosen based on previous experience in the laboratory, for each experiment to yield high power to detect specific effects. No statistical methods were used to predetermine sample size. Homozygous mutant animals used in this study were obtained by heterozygous mating. No fish were excluded. After WISH analyses, individual embryos/larvae were imaged and genotyped to identify homozygous mutants, using previously established protocols (Dobrzycki et al., 2018). For analyses performed on adults, genotyping was performed on tail biopsies collected from individual euthanized fish, in parallel to brain dissection. Randomly selected samples for each genotype were then immunostained in one batch, assessed phenotypically in a blind manner and grouped based on their genotype.

Embryos and larvae (up to 10 dpf) were fixed in 4% PFA, and then directly subjected to whole-mount antibody staining. For juvenile and adult fish, the brains were dissected, fixed in 4% PFA (at pH 8.5 for juveniles to preserve the endogenous fluorescence), and subjected to either tissue clearing (Susaki et al., 2015) and imaging (15- to 50-dpf fish), or incubated overnight in 30% sucrose, cryosectioned (30 µm-thick transversal) followed by antibody staining (adult fish). The following antibodies were used: chicken anti-GFP polyclonal antibody (1:500, Abcam, ab13970), rabbit anti-DsRed polyclonal antibody (1:500, Takara, 632496), goat Alexa Fluor 488-conjugated anti-chicken IgG antibody (1:500, Abcam, ab150169) and donkey Alexa Fluor 594-conjugated anti-rabbit IgG (1:500, Abcam, ab150076). Imaging was performed on a Zeiss LSM 780 inverted microscope, using a Plan Apochromat 20× objective for adult sections and a LD LCI Plan Apochromat 25× water-immersion objective for whole-mount embryos and tissue-cleared brains. Images of entire adult brain sections were obtained by combining 15 tiles using Zen software (Zeiss), for a total area of 1.80 mm².

Adult microglia were quantified on 10 transversal sections to obtain a homogeneous representation of the rostro-caudal axis of each brain. The image of each slide was obtained by stitching 12 to 15 neighboring tiles on the x-plane via Zen software (Zeiss). The mpeg1-positive cells on each slide were manually counted using the Zen software and divided by the area of the slide to obtain the cell density per µm². To analyze microglia development at juvenile stages on whole-mount brains, four to six contiguous fields encompassing the zebrafish optic tectum and cerebellum regions were acquired, at a variable depth depending on the volume of the brain. The total amount of microglia in each field was divided by the volume to obtain the density of cells per mm³. Images of the whole CHT of larvae were obtained by combining three to five consecutive fields of acquisition (depending on the stage) from the caudal to the rostral extremities. Mpeg1-postive cells were subsequently counted in each field.

Adult homozygous csf1rb^sa1503 carrying the kdrl:Cre; ßactin:Switch-DsRed; mpeg1:EGFP triple transgene were intercrossed to homozygous csf1rb^re01 fish, as well as to wild-type animals in order to demonstrate the recessive nature of the csf1rb^sa1503 mutation. The progeny was sorted for triple transgene expression at 2 dpf, then fixed at 6 dpf for phenotype assessment.

Statistical analyses were performed using the Prism Software (GraphPad), using a Kruskal–Wallis test followed by post hoc Dunn's multiple comparisons. A P-value smaller than 0.05 was considered as significant. In all experiments, n indicates the number of individuals (biological replicates) included in the analyses and the graphs represent either mean±s.e.m. (columns) or individual values (dots). The precise number of individuals analyzed in each experiment is specified in the figure legends.

We thank Mireia Rovira, member of the Wittamer lab, for critical discussion and comments on the manuscript, and Tjakko van Ham for sharing the csf1rb^re01 mutant line. We are also grateful to Marianne Caron for technical assistance, to Christine Dubois for support with flow cytometry and to Vivianne de Maertelaer for her help in conducting the statistical analyses.