Interplay of Zeb2a, Id2a and Batf3 regulates microglia and dendritic cell development in the zebrafish brain Free

The immune cell compartments in the central nervous system (CNS) are complex and involve a diverse set of professional effector cells (Croese et al., 2021). In addition to parenchymal microglia, the brain-resident macrophages, other myeloid cells, including non-parenchymal macrophages and dendritic cells (DCs), have been reported to actively participate in crucial immunological functions, CNS plasticity and neurogenesis (Ziv et al., 2006).

Microglia are the main immune cells in the CNS and have been studied extensively under physiological and pathological conditions (Prinz et al., 2019). Like other tissue-resident macrophages, microglia manifest active phagocytosis to remove cellular debris and invaded pathogens in the CNS, both in early development and adult stages (Mazaheri et al., 2014; Peri and Nüsslein-Volhard, 2008). In addition to performing conventional phagocytosis, microglia have also been shown to be involved in neurogenesis by secreting neurotrophic factors, promoting neural circuit wiring and modulating synaptic pruning (Paolicelli et al., 2011; Squarzoni et al., 2014; Ueno et al., 2013; Wake et al., 2013). Intriguingly, several studies have indicated that microglia can also promote neuronal cell death through secreting reactive oxygen species or by unknown mechanisms (Li et al., 2012; Marín-Teva et al., 2004), but the physiological significance of this unusual activity remains unknown. As the key immune effectors in the CNS, microglia have two faces: beneficial and harmful. Under pathological conditions, microglia could undergo morphological and transcriptomic transformation and generate abnormal neuroinflammation by secreting proinflammatory cytokines, reactive oxygen species and peroxynitrites (Nimmerjahn et al., 2005; Prinz et al., 2019). Thus, dysregulation of microglia formation and function are thought to contribute to the development and progression of neurodegenerative diseases (Bohlen et al., 2019). Hence, a comprehensive understanding of the formation and function of microglia may provide potential strategies for developing therapeutic treatment of related neurodegenerative disorders.

DCs, including conventional dendritic cells (cDCs) and plasmacytoid dendritic cells (pDCs), are the key antigen presenting cells (APCs) for T cell activation. In mammals, DCs are rarely distributed in the CNS in steady state and are predominantly found in the meninges and choroid plexus (m/chDCs) (Anandasabapathy et al., 2011; Quintana et al., 2015). These brain-associated DCs are short-lived and continuously replenished by the bone narrow (BM)-derived pre-DCs (Anandasabapathy et al., 2011) or monocytes (Menezes et al., 2016). Using single-cell proteomic analysis, Mrdjen et al. have classified murine CNS-associated DCs into three subsets – cDC1s, cDC2s and pDCs – based on the expression of surface integrin CD11b (also known as Itgam) and CD24a (Mrdjen et al., 2018). The number of DCs and the ratio of DC subsets in the CNS changes during aging and neuroinflammation (Jordão et al., 2019). Importantly, the vessel-associated DCs in the CNS have shown to be the primary APCs to present antigens to mediate T cell activation, whereas the brain-associated macrophages and microglia fail to do so (Greter et al., 2005; Miller et al., 2007). More recent studies have shown that cDC2 are abundantly present in ischaemic brains, triggering the IL17 production of T cells to promote infiltration of neutrophils (Gelderblom et al., 2018), and depletion of the brain cDC1s leads to a more severe inflammatory response to brain ischaemia (Gallizioli et al., 2020). These observations indicate that the brain-associated DCs play an important role in neuroinflammatory regulation. Despite these efforts, the development and the precise functions of the brain-associated DCs remain elusive.

It is well-known that the formation of macrophages and dendritic cells are predominantly controlled by the genetic networks formed by multiple transcription regulators, including lineage-specific transcription factors PU.1 (Spi1) and IRF8 (Kierdorf et al., 2013; Xu et al., 2015) and other general transcription regulators, such as inhibitor of DNA binding 2 (ID2) (Bagadia et al., 2019; Hacker et al., 2003; Scott et al., 2016), basic leucine zipper transcription factor ATF-like 3 (BATF3) (Edelson et al., 2010; Hildner et al., 2008) and zinc finger E box binding homeobox 2 (ZEB2) (Bagadia et al., 2019; Scott et al., 2016, 2018; Wu et al., 2016). ID2, a member of the helix-loop-helix (HLH) transcription regulators, cannot bind DNA directly owing to the lack of a DNA-binding domain but rather regulates gene expression through dimerization with other basic HLH proteins to repress their DNA-binding activity (Benezra et al., 1990). Although ID2 was first recognized as a regulator of cell proliferation and differentiation (Garrell and Modolell, 1990), subsequent studies have shown that it plays a crucial role in the development and functional regulation of multiple immune cells, including natural killer (NK) cells (Li et al., 2021), CD8⁺ T cells (Cannarile et al., 2006) and DCs (Bagadia et al., 2019; Hacker et al., 2003; Scott et al., 2016). BATF3 belongs to the activator protein-1 superfamily, which forms heterodimers with JUN proteins to regulate gene expression (Murphy et al., 2013). In mice, Batf3 expression is restricted to immune cells (Murphy et al., 2013) and has been shown to be essential for the development of conventional cDCs and CD103 (Itgae)-expressing migratory DCs (Edelson et al., 2010; Hildner et al., 2008). ZEB2 was first recognized as a transcriptional inhibitor regulating epithelial-to-mesenchymal transition (Vandewalle, 2005). Later studies have revealed that ZEB2 is essential for the proper development and behaviour of macrophages (Scott et al., 2018), DCs (Bagadia et al., 2019; Scott et al., 2016; Wu et al., 2016), NK cells (van Helden et al., 2015) and T cells (Dominguez et al., 2015; Omilusik et al., 2015). Yet the roles of these transcription regulators in the development of microglia and brain-associated DCs remain largely unexplored.

Zebrafish have become a prominent vertebrate model for studying the development and function of haematopoietic cells (Geirsdottir et al., 2019; Li et al., 2012; Mazaheri et al., 2014; Peri and Nüsslein-Volhard, 2008). We have previously identified two phenotypically and functionally distinct microglial subpopulations, ccl34b.1⁺ microglia and ccl34b.1⁻ microglia, in adult zebrafish brain (Wu et al., 2020). Recent cross-organ single-cell transcriptome profiling has re-identified the ccl34b.1⁺ microglia and ccl34b.1⁻ subtypes as conventional microglia (ccl34b.1⁺) and brain-associated DCs (ccl34b.1⁻), respectively (Zhou et al., 2023). Remarkably, the ccl34b.1⁻ brain-associated DCs are further classified into three subpopulations, one pDC and two cDC subsets (cnn3a⁺ cDCs and ccl35⁺ cDCs). Hence, zebrafish serve as a useful model to study the development of brain-associated DCs and their ontogenic relationship with microglia.

In this report, we used zebrafish as a model system to explore the mechanisms controlling the development of microglia and DCs in the CNS. We found that Id2a and Batf3 are two key regulators essential for the development of brain-associated DCs, and Zeb2a acts as a core factor that promotes microglia proliferation and inhibits brain-associated DC specification.

A recent single cell RNA-sequencing (scRNA-seq) study has shown that, in addition to conventional microglia, the zebrafish brain also contains DCs, including pDCs and two distinct cDCs, ccl35⁺ cDCs and cnn3a⁺ cDCs (Zhou et al., 2023). To investigate the development of microglia (ccl34b.1⁺ cells) and brain-associated DCs (ccl34b.1⁻ cells) in the zebrafish brain, we re-analyzed the transcriptome datasets of the bulk RNA-seq microglia and DCs (Wu et al., 2020) and found that id2a, which encodes a member of the helix-loop-helix (HLH) transcription regulators (Massari and Murre, 2000; Norton, 2000), was highly enriched in the DCs (Fig. S1A). This finding was further confirmed by scRNA-seq analysis (Zhou et al., 2023) and real-time PCR analysis (Fig. 1A). Because the mammalian Id2 has been shown to regulate DC development (Bagadia et al., 2019; Hacker et al., 2003), we reasoned that id2a may play a crucial role in brain-associated DC development in zebrafish. To test this possibility, we outcrossed id2a-deficient mutants, which carry a 22-bp deletion in the first exon (Choi et al., 2017), with the TgBAC(ccl34b.1:eGFP);Tg(mpeg1.1:DsRedx) reporter line, in which microglia and DCs (both pDCs and cDCs) are marked by GFP⁺DsRedx⁺ and GFP⁻DsRedx⁺, respectively, to examine the impact of id2a-deficiency on the development of microglia and DCs. Anti-GFP and anti-DsRedx staining of the midbrain cross-section showed that, although the GFP⁺DsRedx⁺ cells remained largely intact (Fig. 1B; Fig. S1B-D), the GFP⁻DsRedx⁺ cells were completely absent in the id2a-deficient mutants from early stage to adulthood (Fig. 1B). To quantify the number of the above two cell types, we measured the density of the two cell populations in the midbrain of the mutants and siblings. The density of GFP⁺DsRedx⁺ cells (microglia) in the mutants were comparable with that in siblings or marginally reduced (Fig. 1C), whereas the DsRedx⁺ single positive cells (DCs) in the id2a-deficient mutants were dramatically reduced (Fig. 1C). RNAscope of csf1ra (microglia marker) and flt3 (DC marker), co-stained with Lcp1, visualized all the myeloid cells in the brains of mutants and siblings in the TgBAC(ccl34b.1:eGFP) background (Zhou et al., 2023), confirming that the GFP⁺DsRedx⁺ and DsRedx⁺ single positive cells indeed represented microglia and DCs, respectively. In the siblings in this study, the csf1ra signals were detected in the GFP⁺Lcp1⁺ cells, whereas the flt3 signals were found exclusively in the GFP⁻Lcp1⁺ population (Fig. S1E, top panels), confirming their microglia and DC identity, respectively. As expected, in the id2a mutants, the flt3⁺ GFP⁻Lcp1⁺ population was absent, whereas the csf1ra⁺ GFP⁺Lcp1⁺ cells were largely unaffected (Fig. S1E, bottom panels). These data indicate that id2a deficiency blocks the formation of DCs but has little effect on the development of microglia. To further confirm this conclusion, we performed fluorescence activated cell sorting (FACS) of the whole brains of 4-month-old TgBAC(ccl34b.1:eGFP);Tg(mpeg1.1:DsRedx);id2a mutants and siblings (Fig. 1D,E). Consistent with the above notion, the number of GFP⁻DsRedx⁺ cells (representing DCs) was dramatically reduced in id2a-deficient mutants (Fig. 1G), whereas GFP⁺DsRedx⁺ cells (representing microglia) remained relatively normal (Fig. 1F). The residual GFP⁻DsRedx⁺ cells detected by FACS in the id2a mutant brains were neither microglia nor DCs as they lacked both microglia and DC lineage markers (ccl34b.1, apoeb for microglia; ccl19a.1, siglec15l for DCs) (Fig. 1H) (Wu et al., 2020). Subsequent analysis revealed that these remaining GFP⁻DsRedx⁺ cells were B cells expressing high levels of ighd, ighm and cd79a (Fig. S1F), in line with the previous findings that the Tg(mpeg1.1:DsRedx) reporter line labels a small subpopulation of B cells (Ferrero et al., 2020). Taken together, these results demonstrate that id2a is essential for the formation of brain-associated DCs.

As id2a is known to express in various cell types, including neurons (Diotel et al., 2015), we wondered whether the lack of DCs in id2a mutants was autonomous or non-cell-autonomous. To address this issue, we generated a Tg(mpeg1.1:id2a) transgenic line, in which the expression of id2a is under the control of the myeloid-specific mpeg1.1 promoter, and outcrossed it with id2a mutant in the TgBAC(ccl34b.1:eGFP);Tg(mpeg1.1:DsRedx) background. Characterization of this triple transgenic mutant revealed a robust re-appearance of the GFP⁻DsRedx⁺ cells in the transgenic mutant brains (Fig. 2A,B, upper panel; Fig. S2A). These recovered GFP⁻DsRedx⁺ cells displayed a ramified morphology and were predominantly found in the white matter regions (Fig. 2A,C), where the ccl34b.1⁻ cells, which represent DCs, are shown to reside (Wu et al., 2020). To confirm that the recovered GFP⁻DsRedx⁺ cells in the triple transgenic mutants were indeed DCs, we sorted these cells from the brains of the triple transgenic mutant fish and examined the expression of DC-specific markers (ccl19a.1, siglec15l). qRT-PCR analysis showed robust expression of these DC markers in the GFP⁻DsRedx⁺ cells (Fig. 2D). Moreover, RNAscope of DC-specific marker flt3 on the brain sections of the TgBAC(ccl34b.1:eGFP);Tg(mpeg1.1:id2a);id2a triple transgenic mutants showed a co-localization of flt3 signals with the recovered GFP⁻Lcp1⁺ cells (Fig. S2B). Based on these observations, we conclude that id2a intrinsically regulates the formation of the brain-associated DCs.

To further explore the developmental regulation of different DC subsets in the brain, we re-analyzed the scRNA-seq data (Zhou et al., 2023) and found that batf3 was highly expressed in the cnn3a⁺ cDCs compared with other DC subsets and microglia (Fig. 3A). We speculated that batf3 might play a crucial role in cnn3a⁺ cDC development. To test this possibility, we generated a batf3-deficient mutant fish, which harboured a 286-bp deletion in the coding region of the baft3 gene, resulting in the removal of the bZIP domain (Fig. S3A). Transverse brain sections of the 1- and 4-month-old batf3 mutants in the TgBAC(ccl34b.1:eGFP);Tg(mpeg1.1:DsRedx) background revealed a normal number and distribution of the GFP⁺Lcp1⁺ cells (representing microglia) throughout the whole brain (Fig. 3B,C; Fig. S3B-D). However, the number of the GFP⁻Lcp1⁺ cells (representing DCs) were markedly reduced in the batf3 mutants (Fig. 3B,C; Fig. S3B-D). Interestingly, we found that the remaining GFP⁻Lcp1⁺ cells in the mutant brains displayed a rod-shaped morphology (Fig. 3B,C; Fig. S3B-D). To further characterize the nature of these rod-shaped cells, we sorted these cells from the batf3-deficient mutant brains in the TgBAC(ccl34b.1:eGFP)Tg(mpeg1.1:DsRedx) background and performed RT-PCR analysis. Results showed that these residual rod-shaped cells were devoid of cnn3a⁺ cDC markers (cnn3a, hsp90aa1.2, chl1a) (Fig. 3D), but they predominantly expressed the ccl35⁺ cDC markers (ccl35, nr4a2b, cd74a) (Fig. 3E) and pDC markers (ctsbb, p2rx3a) (Fig. 3F). These observations indicate that batf3-deficiency selectively blocks the formation of cnn3a⁺ cDCs but has no obvious effect on the development of ccl35⁺ cCDs and pDCs.

Previous studies in mice have shown that Batf3 and Id2 are critical regulators of DC development (Bagadia et al., 2019; Edelson et al., 2010; Hacker et al., 2003; Hildner et al., 2008; Scott et al., 2016) and Batf3 appears to act downstream of Id2 as its expression is induced by Id2 during cDC1 specification (Bagadia et al., 2019). Based on the phenotypes of the id2a- and batf3-deficient zebrafish mutants, we speculated that zebrafish batf3 might act downstream of id2a to regulate cnn3a⁺ cDC development. To test this hypothesis, we examined the expression of these two genes in the corresponding mutants. As we anticipated, we found that batf3 transcripts were markedly reduced in the id2a-deficient mpeg1.1⁺ cells in the brains (Fig. 3G), whereas id2a expression was largely unaffected in batf3 mutants (Fig. S3E), suggesting that batf3 is downstream of id2a. To further support this notion, we examined the expression of batf3 in the Tg(mpeg1.1:id2a) fish, in which id2a was ectopically overexpressed in macrophages/microglia and DCs (Fig. S3F). Consistent with our hypothesis, batf3 expression was markedly increased in the mpeg1.1⁺ cells in the id2a-overexpressing Tg(mpeg1.1:id2a) transgenic fish (Fig. S3F). Taken together, these results indicate that the expression of batf3 is regulated either directly or indirectly by id2a. To further explore whether batf3 is sufficient to promote cnn3a⁺ cDC formation, we generated a batf3 overexpression Tg(mpeg1.1:batf3) transgenic line and outcrossed it with the id2a mutants. Surprisingly, we found that overexpressing batf3 failed to restore the formation of the cnn3a⁺ cDCs in the id2a-deficient mutants (Fig. 3H; Fig. S3G). These results indicate that batf3 acts downstream of id2a and likely works together with other unknown factor(s) to promote the formation of cnn3a⁺ cDCs in the CNS.

Having revealed the crucial roles of id2a and baft3 in brain-associated DC development, we next turned our attention to the developmental regulation of microglia. From the RNA-seq datasets, we noticed that Zeb2a was highly enriched in the microglia but was barely detectable in brain-associated DCs (Fig. S4A) (Wu et al., 2020), which was further confirmed by qRT-PCR analysis (Fig. 4A). Given that mammalian Zeb2 has been shown to be involved in the development of various haematopoietic lineages (Bagadia et al., 2019; Scott et al., 2016; Wu et al., 2016), the restricted expression pattern of zeb2a suggested that zebrafish zeb2a might play an important role in microglia development. To test this hypothesis, we generated mutant zebrafish harbouring mutations in zeb2a and zeb2b, the two zebrafish orthologs of mammalian Zeb2 (Delalande et al., 2008). Both mutant fish carried a short indel that led to a premature stop codon, resulting in the loss of the functional N-terminal zinc fingers (Fig. S4B,C). To facilitate the characterization of the mutants, we outcrossed them with the microglia-specific reporter line Tg(ccl34b.1:eGFP) (Wu et al., 2020). Further analysis showed that zeb2b-deficient mutants developed normally and contained a normal number of GFP⁺Lcp1⁺ (microglia) and GFP⁻Lcp1⁺ (DCs) cells from early stages to adulthood (Fig. S4D,E), suggesting that zeb2b is dispensable for the development of both microglia and DCs. In contrast, in zeb2a-deficient mutants, microglia (as indicated by ccl34b.1⁺ cells) were markedly reduced from 3 days postfertilization (dpf) onwards (Fig. 4B,C), indicating that zeb2a plays a crucial role in the development of microglia. To further support this conclusion, we quantified the number of ccl34b.1⁺ cells at 14 dpf, before the death of the mutants at ∼20 dpf, and we found that zeb2a mutants indeed had significantly fewer ccl34b.1⁺ cells (Fig. 4D,E). To exclude the compensatory function of zeb2b, we generated the zeb2a/b double mutants, and results showed that the microglia phenotype as indicated by the reduction of ccl34b.1⁺ cells was comparable between the double mutants and the zeb2a single mutants (Fig. 4B,C), further supporting the notion that zeb2b has a minimum role in microglia development.

To further delineate the cellular basis underlying the reduction of microglia in zeb2a-deficient mutants, we performed an EdU incorporation assay to measure the proliferation rate of ccl34b.1⁺ cells at 13 dpf and found a 2-fold reduction of EdU⁺ cells among the ccl34b.1⁺ cells in the zeb2a mutants compared with that in the control siblings (Fig. 4G,H), indicating that loss of Zeb2a function impairs the proliferation of microglia. Taken together, these findings demonstrate that zeb2a is a key factor that promotes microglia expansion.

Previous studies in mice have reported that Zeb2 and Id2 play opposite roles in DC development by repressing the expression of each other (Cannarile et al., 2006; Geirsdottir, et al., 2019). We therefore wondered whether this bilaterally repressive interaction between zeb2a and id2a would be preserved in zebrafish. To address this issue, we created a heat-shock-induced overexpression model to examine the relationship between zeb2a and id2a. Briefly, wild-type (WT) fertilized embryos were injected with the hsp70:zeb2a-P2A-eGFP or hsp70:id2a-P2A-eGFP plasmids and heat-shocked at 1 dpf for 1 h to induce ectopic overexpression of zeb2a or id2a (Fig. 5A). The total RNA was then collected from the heat-shocked embryos at 2 dpf and subjected to qRT-PCR analysis. We found that zeb2a overexpression indeed led to a downregulation of id2a expression (Fig. 5B,C), whereas overexpressing id2a had no obvious effect on zeb2a (Fig. S5A,B). To further validate these findings in the cell-lineage overexpression context, we outcrossed the Tg(mpeg1.1:zeb2a) and Tg(mpeg1.1:id2a) transgenic lines with the TgBAC(ccl34b.1:eGFP);Tg(mpeg1.1:DsRedx) double reporter line, sorted out the DCs (GFP⁻DsRedx⁺) and microglia (GFP⁺DsRedx⁺) and measured the id2a and zeb2a mRNA level respectively by RT-PCR. Indeed, we observed a significant reduction of id2a expression in the zeb2a-overexpressing DCs (Fig. S5C,D) but no significant alteration of zeb2a mRNA was detected in the id2a-overexpressing microglia (Fig. S5E,F). From these observations, we conclude that Zeb2a acts as a transcriptional repressor to suppress the id2a expression, whereas Id2a does not appear to regulate zeb2a expression.

Given the fact that id2a is a crucial regulator for DC development (Fig. 1), we hypothesized that Zeb2a might constrain the development of brain-associated DCs. To test this hypothesis, we examined the formation of ccl34b.1⁻ cells (representing DCs) in zeb2a-deficient mutants at 14 dpf, when both microglia and DCs are detected in the brain in WT fish (Wu et al., 2020) and tested whether loss of Zeb2a function would lead to the expansion of DC pool in the brain. Indeed, results showed that the brain-associated DCs (as indicated by the ccl34b.1⁻ cells) were significantly increased in the mutant brains (Fig. 4D,F). The increase of the ccl34b.1⁻ cells in zeb2a-deficient fish was not due to the excessive proliferation of these cells, as their proliferation rate was reduced in the mutants (Fig. 4G,I). These results suggest that Zeb2a inhibits the commitment of myeloid progenitors toward the DC lineages. To further support this notion, we transiently overexpressed zeb2a in the pan-myeloid lineages using the mpeg1.1 promoter and asked whether it would alter the formation of microglia and DCs. Indeed, transverse sections of 1-month-old mpeg1.1:zeb2a-injected fish showed a modest, but noticeable increase of ccl34b.1⁺ cells, accompanied by a 2-fold reduction of ccl34b.1⁻ cells (Fig. 5D-F). Moreover, the proliferation rate of ccl34b.1⁺ and ccl34b.1⁻ cells remained largely unchanged upon zeb2a overexpression (Fig. S6A,B). Taken together, these results indicate that, in addition to regulating microglia proliferation, zeb2a also acts as a repressor inhibiting the specification of DCs in the zebrafish CNS.

In this study, we have demonstrated a transcriptional network determining the formation of microglia and brain-associated DCs in the zebrafish CNS. We have identified Id2a as a key transcription modulator governing the formation of brain-associated DCs and further shown that Batf3 acts downstream of Id2a and is required for the development of the cnn3a⁺ subset of DCs but dispensable for the formation of ccl35⁺ cDCs. Finally, we have revealed a dual but opposite function of Zeb2a in the regulation of microglia and DCs in the brain.

Our study has demonstrated that Id2a intrinsically regulates the development of all three subtypes of the brain-associated DCs as they are completely absent in id2a-deficient mutants. However, the underlying cellular basis remains unclear. Interestingly, we showed that forced expression of id2a with the mpeg1.1 promoter is sufficient to rescue, at least in part, the development of DCs in the id2a-deficient mutants. Based on our recent scRNA-seq analysis (Zhou et al., 2023), mpeg1.1 transcripts are detected in macrophages and DCs, as well as NK cells, suggesting that, in addition to expressing in all committed DC lineages, id2a may also be expressed in early macrophage/DC precursors (including macrophage/DC common precursors and DC precursors). It is therefore conceivable to speculate that Id2a may play a role either in the DC lineage fate determination from macrophage/DC common precursors or in the survival/differentiation of DC precursors. Based on the DC-specific phenotype of the id2a-deficient mutants, we favour the later possible mechanism: interfering with the survival of DC precursors and their subsequent differentiation into DC lineages. This speculation is supported by the previous findings in mice showing that Id2 is essential for the differentiation of DCs (Bagadia et al., 2019; Hacker et al., 2003; Scott et al., 2016) and NK cells (Li et al., 2021). Moreover, it was found that Id2-deficient mammalian effector T cells express a lower level of antiapoptotic gene Bcl2 and higher levels of proapoptotic genes Bcl2l11 and Ctla4 (Cannarile et al., 2006), supporting the idea that Id2 may directly control the survival of the DC precursors. A more comprehensive study using a conditional knockout or knockdown of id2a would be necessary to reveal the temporal requirement of Id2a in DC formation. Mechanistically, how Id2a regulates the formation of brain-associated DCs remains unclear. It is well-known in mammals that ID2 regulates gene expression through the dimerization with other basic HLH proteins, such as E proteins, thereby repressing their DNA-binding activity (Benezra et al., 1990). Indeed, ablation of E2A in Id2-deficient mice bypasses the impairment of bone marrow NK cells and secondary lymphoid tissue (Boos et al., 2007), illustrating that ID2 function is mediated, at least in part, through inhibiting E protein activity. We therefore speculate that an evolutionarily conserved mechanism would be employed in zebrafish. It would be interesting to test this hypothesis in a future study.

Our study has also shown that Batf3 acts downstream of Id2a and is specifically required for the formation of the cnn3a⁺ cDC subset. However, whether Batf3 and Id2a function within a single pathway remains unclear. In mammals, the development of CD8α⁺ DC subset has been found to require not only Id2 and Batf3 but also Nfil3 (Hacker et al., 2003; Hildner et al., 2008; Kashiwada et al., 2011). In addition, forced expression of Batf3 in the Nfil3-deficient bone marrow progenitors can rescue CD8α⁺ DC formation, suggesting that Batf3 acts downstream of Nfil3 (Kashiwada et al., 2011). As zebrafish Nfil3 has been reported to be involved in myeloid cell development (Progatzky et al., 2012), we speculate that Batf3, although its expression is regulated by Id2a, may also act downstream of the zebrafish Nfil3. Intriguingly, we found that forced expression of batf3 is unable to rescue the formation of cnn3a⁺ cDCs in the id2a mutants. This observation implies that additional factor(s), which presumably acts downstream of Id2a and in parallel with Batf3, may be required to promote the development of cnn3a⁺ cDCs. Alternatively, the mpeg1.1 promoter-directed batf3 expression could not reach the level required for the cnn3a⁺ cDC development or missed the time window essential for the cnn3a⁺ cDC development. Further study will be required to clarify this issue.

Finally, our study has discovered that Zeb2 has a dual role in microglia and brain-associated DC development: promotion of microglia proliferation and inhibition of DC specification. In line with our observations, conditional knockout of Zeb2 in mice causes a reduction of tissue-resident macrophages, including microglia in the brain, Kupffer cells in the liver and macrophages in the lung, owing to the impairment of their proliferation (Scott et al., 2018). Remarkably, murine Zeb2 is also found to promote fate commitment of pre-cDCs toward the cDC2 lineage through inhibiting Id2 expression (Scott et al., 2016). However, whether mammalian Zeb2 indeed functions as a determinant factor controlling the fate choice between tissue-resident macrophages and DCs remains unclear. The inhibitory effect on the specification of the brain-associated DCs also raises the possibility that in zebrafish microglia and brain-associated DCs may arise from a common myeloid progenitor. All these issues warrant further exploration.

All zebrafish lines used in this study were maintained at 28.5°C with a 14 h light/10 h dark cycle. After natural spawning, embryos were raised in 0.5× E2 medium containing methylene blue (egg water) until 4 dpf. To avoid pigmentation, embryos were transferred to 0.003% N-phenylthiourea (P7629, Sigma-Aldrich) in egg water at 10-24 hpf. The lines used in this study were: AB WT, id2a (Choi et al., 2017), batf3^sz2, zeb2a^sz3, zeb2b^sz4, Tg(mpeg1.1:id2a,myl7:mCherry)^sz200 [known as Tg(mpeg1.1:id2a)], Tg(mpeg1.1:batf3,myl7:mCherry)^hkz053Tg [known as Tg(mpeg1.1:batf3)], Tg(mpeg1:loxP-DsRedx-loxP-GFP) [known as Tg(mpeg1.1:DsRedx)] (He et al., 2018) and TgBAC(ccl34b.1:eGFP) (Wu et al., 2020). All animal experiments were performed under approval from the Hong Kong University of Science and Technology's Animal Studies Committee.

CRISPR/Cas9-mediated mutagenesis was conducted according to a previous protocol (Chang et al., 2013). The double-stranded DNA template for specific gRNA synthesis was PCR amplified from the pMD-19-T-gRNA scaffold vector and then purified with QIAquick PCR Purification Kit (28104, Qiagen). The MEGAshortscript Kit (AM1354, Invitrogen) and MEGAclear Transcription Clean-Up Kit (AM1908, Invitrogen) were used to transcribe and purify the gRNA. EnGene Spy Cas9 NLS (6 pmol) and gRNA (500 ng/ml) were co-injected into one-cell-stage AB embryos. The injected embryos were raised to adulthood for founder screening.

For the Tg(mpeg1.1:id2a) and Tg(mpeg1.1:batf3) transgenic lines, the coding sequences of either id2a or batf3 genes were cloned together with the mpeg1.1 promoter into a modified pBlueScript II SK (⁺) vector with the transposase recognition site (He et al., 2018). The vectors (30 ng/ml) and transposase mRNA (50 ng/ml) were co-injected into one-cell-stage AB embryos. The injected embryos were raised to adulthood for transgene screening.

For the heat-shock line, the hsp70:id2a-P2A-eGFP and hsp70:zeb2a-P2A-eGFP constructs were generated by inserting the coding sequences of either id2a or zeb2a genes, linked with P2A-eGFP, together with the hsp70 promoter into a modified pBlueScript II SK (⁺) vector with the transposase recognition site (Halloran et al., 2000). The constructs (30 ng/ml) and transposase mRNA (50 ng/ml) were then co-injected into one-cell-stage AB embryos. The injected embryos were raised to the desired stage for the heat-shock experiment.

For the mpeg1.1:zeb2a, the coding sequence of the zeb2a gene was cloned together with the mpeg1.1 promoter into a modified pBlueScript II SK (⁺) vector with the transposase recognition site (He et al., 2018).

Larval fish (7 dpf, 14 dpf) and juvenile fish (1-month old) were anaesthetized on ice and fixed in 4% paraformaldehyde (PFA) at 4°C for 20-24 h. Adult fish were anaesthetized on ice and their brains were freshly dissected and fixed in 4% PFA at 4°C for 20-24 h. After washing with PBST, the fixed samples were dehydrated with 30% sucrose in PBS at 4°C for 1 day, then embedded in coagulating solution (1.5% agar+5% sucrose), and finally cut into 30 µm-thick section using a CryoStar NX70 (Thermo Fisher Scientific). Immunohistochemistry was performed as previously described (Wendl et al., 2002). Images were captured by Leica SP8 confocal microscope. Embryonic fish were anaesthetized using 0.01% Tricaine (A5040; Sigma-Aldrich), mounted in 1% low melting agarose, and imaged using a Leica SP8 confocal microscope.

EdU incorporation was performed following the manufacturer's protocol using the Click-iT EdU Cell Proliferation Kit for Imaging, Alexa Fluor 647 dye (Invitrogen, C10340).

csf1ra and flt3 RNAscope single molecule fluorescence in situ hybridization (smFISH) was performed in 14 μm-thick mid-brain slices using the RNAscope Multiplex Fluorescent Reagent Kit [323100, Advanced Cell Diagnostics (ACD)]. Briefly, brains harvested from 1-month-old fish were cryosectioned transversally. Specific probes for csf1ra (1046331-C3, ACD) and flt3 (1208621-C1, ACD) were hybridized and their signals were amplified. Following RNAscope smFISH, immunostaining was performed for GFP and Lcp1, using the anti-GFP (Abcam, ab6658, 1:400) and anti-Lcp1 (Jin et al., 2009, 1:400) primary antibodies, respectively. The secondary antibodies were Alexa 488 anti-goat antibody (Invitrogen, A11055, 1:400) and Alexa 405 anti-rabbit antibody (Invitrogen, A48258, 1:400). Images of the brain sections were acquired using the 40× objective of a Zeiss Celldiscoverer7/LSM900 confocal microscope and were subjected to further processing using the ZEISS ZEN Lite software.

Adult TgBAC(ccl34b.1:eGFP);Tg(mpeg1.1:DsRedx) fish were anaesthetized on ice, and the brains of the anaesthetized fish were dissected and homogenized by needles and syringes in 1 ml 0.25% trypsin-EDTA (25200072, Thermo Fisher Scientific). The homogenized brain samples were then digested in 2 ml 0.25% trypsin-EDTA for 20-25 mins at 30°C and the reaction was terminated by 10 mM CaCl₂ and 10% foetal bovine serum. Dissociated cells were pelleted and washed with ice-cold PBS/1% bovine serum albumin. Finally, the cell suspension was passed through a Falcon 40-mm Cell Strainer (352340, BD Falcon) to filter out the cellular aggregation. Microglia (50/100 cells per sample) were sorted directly into lysis buffer (2% Triton X-100 solution) using a BD FACSAria III sorter (BD Biosciences).

The cDNA and library was generated following a previous Smart-seq2 protocol (Picelli et al., 2014). Cell lysate (100 cells per samples) was mixed with oligo(dT) primers and deoxynucleotide triphosphate (U1515, Promega) and incubated at 72°C for 3 min. The reverse transcription mix containing SuperScript II Reverse Transcriptase (18064014, Invitrogen) and a template-switching oligo was then added to the mixture for reverse transcription. Finally, the first-strand reaction products were amplified using KAPA HiFi HotStart ReadyMix (KK2602, KAPA) with 24 PCR cycles. The cDNA was purified using Ampure XP beads (A63881, Beckman Coulter), followed by quality control using the Agilent Fragment Analyzer System.

Total RNA of 2 dpf embryos (10-12 embryos, per sample) were extracted using RNeasy Mini Kit (74104; Qiagen). Purified RNA was reversely transcribed using SuperScript IV Reverse Transcriptase (18090010, Thermo Fisher Scientific). Real-time qPCR was performed using a LightCycler 480 Real-Time PCR System to quantify the transcripts of ccl19a.1, siglec15l, id2a, batf3, zeb2a and zeb2b.

The embryos at 1 dpf were collected into a 15-ml Falcon tube containing 10 ml of 0.5× E2 medium and submerged in a 37°C water bath for 1 h. Heat-shocked embryos were then changed back to 28.5°C and raised until the desired stage for further analysis.

All the statistical analyses were performed using GraphPad Prism version 8. Unpaired Student's t-tests were used to calculate the P-value for pairwise comparisons, with Welch's correction in cases of unequal variance. Two-tailed P-values are used for all t-tests. For multiple comparisons, significances are calculated using two-way ANOVA test. Data are represented as mean±s.d. For all statistical analyses, the P-values are displayed on the figures. P<0.05 is considered statistically significant.

We extend our gratitude to the Biosciences Central Research Facility, HKUST (Clear Water Bay) for allowing us access to the equipment required for this research. This study was also carried out in part in the Core Research Facilities, Southern University of Science and Technology, Shenzhen, China.

The peer review history is available online at https://journals.biologists.com/dev/lookup/doi/10.1242/dev.201829.reviewer-comments.pdf