Histone deacetylase 3 controls lung alveolar macrophage development and homeostasis

Alveolar macrophages (AMs) derived from embryonic precursors seed the lung before birth and self-maintain locally throughout adulthood, but are regenerated by bone marrow (BM) under stress conditions. However, the regulation of AM development and maintenance remains poorly understood. Here, we show that histone deacetylase 3 (HDAC3) is a key epigenetic factor required for AM embryonic development, postnatal homeostasis, maturation, and regeneration from BM. Loss of HDAC3 in early embryonic development affects AM development starting at E14.5, while loss of HDAC3 after birth affects AM homeostasis and maturation. Single-cell RNA sequencing analyses reveal four distinct AM sub-clusters and a dysregulated cluster-specific pathway in the HDAC3-deficient AMs. Moreover, HDAC3-deficient AMs exhibit severe mitochondrial oxidative dysfunction and deteriorative cell death. Mechanistically, HDAC3 directly binds to Pparg enhancers, and HDAC3 deficiency impairs Pparg expression and its signaling pathway. Our findings identify HDAC3 as a key epigenetic regulator of lung AM development and homeostasis.

A lveolar macrophages (AMs), the resident macrophages in lung alveoli, are important for the maintenance of homeostasis in the airways and are involved in the development of a variety of pulmonary diseases 1 . Recent lineagetracing studies have demonstrated that tissue-resident macrophages (TRMs) originate from embryonic yolk sac (YS) erythromyeloid progenitors (EMPs) and can self-renew in most adult tissues at steady state without contribution from bone marrow (BM) hematopoietic stem cells (HSCs) [2][3][4] . YS EMP cells differentiate into premacrophages (pMacs) that colonize all embryonic organs as TRMs starting at embryonic day 9.5 (E9.5) 5 . The YS EMPs also enter the fetal liver (FL) starting from E12.5, where they expand and differentiate into pMacs and/or monocytes that travel to local tissues through the blood and eventually become tissue-specific TRMs 4 . Fetal monocytes begin to accumulate in the developing lung at E14.5 and then differentiate into immature AMs (preAMs) expressing F4/80 int CD11b int , which postnatally mature into AMs expressing CD11c hi Siglec-F hi CD11b lo, 4,6,7 . Like the majority of other TRMs, AMs self-maintain locally at steady state throughout adult life 4,8 . Under stressed conditions, AMs are either self-maintained by local proliferation or completely replaced by bone marrow-derived cells in a challenge-dependent manner 9,10 .
A few genes have been identified as requirements for AM development and homeostasis. Granulocyte macrophage colonystimulating factor (GM-CSF) was shown to be essential for the perinatal differentiation of fetal monocytes into preAMs and for the full maturation of AMs postnatally 7 . Transforming growth factor-β (TGF-β) is also crucial for the embryonic differentiation of fetal monocytes into preAMs, their postnatal maturation, as well as the homeostasis of adult AMs 11 . Both GM-CSF and TGFβ signaling pathways induce the expression of PPAR-γ, a signature transcription factor essential for the development of AMs 7,11 . Macrophage identity and homeostasis require the precise regulation of gene expression that is governed by different epigenetic mechanisms, such as DNA methylation, histone modification, and chromatin structure 12 . However, very little is known about the epigenetic factors that regulate AM development from EMPs and fetal monocytes, as well as its postnatal homeostasis.
Histone deacetylases (HDACs) are enzymes that regulate gene expression by modifying chromatin structure through the removal of acetyl groups from target histones or through direct deacetylation of non-histone proteins. HDACs exhibit limited substrate specificity and rely on transcription factors with specific DNA-binding and/or chromatin-binding activities in order to target specific genes 13 . Class I HDACs (HDAC1, 2,3,8) can assemble into multi-component co-repressor 13 or co-activator 14 complexes that regulate the transcription of a broad array of genes which fundamentally impact cellular physiology, organism development, and disease pathogenesis [15][16][17][18] . Several pan HDAC inhibitors (HDACi) have been approved by the FDA for antitumor therapy 19 . More recently, HDACi was suggested for use in the treatment of asthma and other inflammatory lung diseases 20,21 . HDAC3, belonging to the class I HDAC family, is required for the macrophage inflammatory gene expression program 22 and serves as an epigenetic brake in macrophage alternative activation 23 . However, it remains unknown whether HDAC3 is involved in the ontogeny and homeostasis of AMs.
In this study, we aim to investigate the role of HDAC3 in the embryonic development and postnatal maintenance of AMs. Using mice with TRM or AM lineage-specific or time-specific HDAC3 deletion and the bone marrow chimeric mouse model, we demonstrate that HDAC3 deficiency leads to severe impairment of AM ontogeny, maintenance, maturation, and regeneration. Single-cell RNA sequencing analyses reveal that AMs have four unique sub-populations and that a lack of HDAC3 gives rise to AM subset-specific pathway dysregulation. Moreover, loss of HDAC3 results in dysregulated mitochondrial oxidative phosphorylation and cell survival. Mechanistically, HDAC3 directly binds to Pparg gene enhancers and regulates PPAR-γ signaling during AM development. Our findings uncover HDAC3 as a key epigenetic factor in the regulation of lung AM embryonic development and maintenance after birth.

Results
HDAC3 is required for the embryonic development of AMs. Before birth, F4/80 hi CD11b int pMac-derived fetal macrophages 5 and F4/80 int CD11b hi FL monocytes sequentially colonize the developing lung around E12.5 and E14.5, respectively 8 . Fetal monocytes further differentiate into F4/80 int CD11b int preAMs, which become CD11c hi Siglec-F hi CD11b lo AMs during the first week of life. To investigate the role of HDAC3 in AMs, we first examined the expression pattern of HDAC3 in the lung fetal macrophages and preAMs during embryonic development, as well as AMs at a young age and adulthood, using qRT-PCR. As shown in Fig. 1a, HDAC3 was expressed in lung fetal macrophages at E14.5 and in preAMs at E16.5 and E18.5, but its expression was dramatically increased in AMs from the young (P14,~20-fold) and adult (~40-fold) mice. These results suggest that HDAC3 is potentially involved in AM embryonic development and maintenance after birth.
To determine whether HDAC3 regulates embryonic development of AMs, we generated Csf1r iCre Hdac3 fl/fl conditional knockout (cKO) mice, in which HDAC3 is deficient in the Csf1r-expressing myeloid cells, including TRMs and monocytes 24,25 . Both HDAC3 mRNA and protein were efficiently depleted (>80%) in preAMs from the HDAC3cKO newborns as determined by qRT-PCR and flow cytometry, respectively (Fig. 1b,  c). We next examined the fetal macrophages, preAMs, and fetal monocytes in the lung from Csf1r iCre Hdac3 fl/fl cKO mouse embryos and wild type (WT) littermates between E12.5 and right after birth (P0). There was no significant alteration in the frequencies of F4/80 hi CD11b lo fetal macrophages and F4/ 80 lo CD11b hi monocytes in the lung between cKO and WT embryos at E12.5 (Fig. 1d, e, gates shown in Supplementary  Fig. 1a). However, a modest but significant reduction in the frequency of fetal macrophages was observed in HDAC3cKO embryos at E14.5 and E16.5, followed by a gradual diminish afterwards. On the other hand, the frequency of F4/80 int CD11b int preAM cells that emerged at E16.5 in the fetal lung was dramatically decreased in HDAC3cKO compared to WT littermates starting at E16.5 until birth. In contrast, fetal monocytes showed compensatorily increased distribution in the absence of HDAC3 starting at E14.5. Immunofluorescence staining of lung tissue sections at E18.5 further confirmed that the number of lung preAMs was significantly reduced in the HDAC3cKO mice (Fig. 1f). Collectively, these results suggest that HDAC3 is indispensable for the development of YS pMac-derived lung macrophages and FL monocyte-derived preAMs during embryogenesis.
Next, we investigated whether HDAC3 deficiency blocks EMP development and differentiation into pMacs and FL monocytes in the YS and FL, respectively. Interestingly, within the CD45 + Lin -(CD45-expressing; lineage negative) population, the frequencies of CD117 hi F4/80 -EMPs, CD117 -F4/80 -pMacs, and CD117 -F4/ 80 + macrophages in the YS remained unaltered in HDAC3cKO embryos compared to the WT at E9.5, E10.5, and E12.5 (Fig. 1g, h, gates shown in Supplementary Fig. 1b), suggesting that HDAC3 is dispensable for EMPs and their further differentiation into pMacs and YS macrophages. Furthermore, we observed comparable frequencies of EMPs in the FL between the HDAC3cKO and WT embryos at E10.5 and E12.5, suggesting that the loss of HDAC3 also does not affect EMP seeding in the FL (Fig. 1g, h). In addition, the frequency of CD64 lo CD11b hi-Ly6c hi FL monocytes remained unaltered during the embryonic stage (E14.5-E18.5) in the HDAC3cKO mice (Fig. 1i, j, gates shown in Supplementary Fig. 1c). HDAC3 deletion in FL monocytes, which was confirmed by qRT-PCR and flow cytometry ( Supplementary Fig. 2a, b), does not disrupt their differentiation into TRMs in other organs, such as the brain, spleen, kidney, and pancreas ( Supplementary Fig. 3a, b, gates shown in Supplementary Fig. 1a). Overall, HDAC3 deficiency does not affect EMPs, their ability to differentiate into pMacs in the YS, or FL monocyte development, but it may play key roles in pMac-derived macrophages and FL monocyte-derived preAMs in the lung.
HDAC3 is essential for the maintenance of AMs after birth. We next examined the AMs from the Csf1r iCre Hdac3 fl/fl cKO adult mice. As expected, the frequency of lung AMs (CD11c hi Siglec-F hi )  Supplementary Fig. 1d). This reduction was further confirmed by immunofluorescence staining of frozen tissue sections (Fig. 2b). After birth, preAMs quickly downregulate their CD11b expression and become mature AMs 7 . As shown in Fig. 2c, the remaining AMs in the HDAC3-deficient mice were found to highly express CD11b. This suggests that maintenance of mature AMs in the adult mice is very likely dependent on HDAC3. However, given that Csf1r.Cre-induced gene deletion occurs in the embryonic stage, the AM defects observed in the Csf1r iCre Hdac3 fl/fl cKO adult mice may be due to embryonic developmental dysregulation rather than a maintenance abnormality after birth.
Regeneration of AMs from the BM depends on HDAC3. AMs self-maintain at steady state independently of circulating precursors 9 . However, under certain inflammatory conditions or BM transplantation following lethal whole-body irradiation, BMderived monocytes can repopulate the AM niche 9,26 . We next sought to determine whether BM-derived AMs are also dependent on HDAC3. To do this, we established a BM chimeric mouse model by co-transferring BM cells from Hdac3 fl/fl or Csf1r i-Cre Hdac3 fl/fl cKO donors (CD45.2 + ) with competitor BM cells from B6.SJL WT mice (CD45.1 + ) into the irradiated B6.SJL hetero (CD45.1 + CD45.2 + ) recipient mice (Fig. 3a). As expected, in the chimeric mice, lung AMs were radiosensitive and largely replaced by BM-derived cells from both WT donor and competitor mice (Fig. 3b). However, the BM from HDAC3cKO mice completely failed to reconstitute lung AMs compared to the BM from WT mice. These findings suggest that HDAC3 is not only important for embryonically-derived AMs, but also required for the regeneration of AMs from adult BM under stress conditions.
HDAC3 regulates AM subset-specific gene expression profile.
Recent single-cell studies have shown that heart and liver TRMs are heterogeneous populations at steady state 27,28 . To characterize whether lung AMs are also heterogeneous, we sorted CD11c hi-Siglec-F hi AMs from Hdac3 fl/fl WT adult mice and performed single-cell RNA-sequencing (scRNA-seq) using the 10× Genomics platform. More than 2000 cells were analyzed and Tdistributed stochastic neighbor embedding (t-SNE) dimensionality reduction analysis identified 4 major AM clusters (Fig. 4a). Each AM sub-cluster has a unique gene expression pattern ( Fig. 4b) with their signature genes (Fig. 4c). Of the AM clusters, cluster 1 was the most abundant and expressed macrophageassociated genes, such as Abcg1, Mrc1, and Mpeg1. Cluster 2 expressed genes such as Birc5, Top2a, Rrm2, and Stmn1, which are involved in the cell cycle and mitosis. Cluster 3 had relatively higher expression of antigen-presentation genes, such as H2-Eb1, H2-Ab1, H2-Aa, and Cd74. Cluster 4 was enriched in many genes associated with immune response, such as Tnf, Il1a, Cxcl1, Cxcl2, and Cxcl3. The cluster-specific gene expression pattern suggests that each AM cell subset may have a distinct role in AM maintenance and function. In order to determine whether HDAC3 is required for the maintenance of a specific lung AM cell subset, we next performed scRNA-seq on the sorted AM cells from Cd11c Cre Hdac3 fl/fl cKO mice and compared the AM clusters between HDAC3cKO and WT mice. Although the composition of clusters 1 and 2 remained almost unaltered in the KO, the frequency of cluster 3 was increased from 3.3% to 4.9%, while the frequency of cluster 4 was reduced from 4.0% to 2.4% in the absence of HDAC3 (Fig. 4d,   of AM subsets in a cluster-specific manner. In addition, we also observed certain cluster-specific transcriptomic changes between WT and cKO mice ( Supplementary Fig. 5). Ingenuity pathway analysis (IPA) revealed that each AM cluster in the cKO mice exhibited distinct dysregulated pathways compared to that in their WT counterpart (Fig. 4f). In cluster 1, the key disordered pathways included phagosome maturation, unfolded protein response, mitochondria dysfunction, and BAG2 signaling pathways, which are associated with macrophage phagocytosis, polarization, survival, and stress responses. In cluster 2, the pathways related to cell cycle control were less enriched in the cKO cells. In cluster 3, several immune-responsive pathways were upregulated in the cKO cells, including E1F2 signaling, Th1/Th2 pathways, and dendritic cell maturation. In cluster 4, the pathways involved in cytokine signaling (i.e., IL-6 and IL-10 signaling) and acute phase response signaling were less enriched in the cKO cells, whereas Toll-like receptor signaling, neuroinflammation signaling pathway, and communication between innate and adaptive immune cells, were strongly enriched in the cKO cells. These data reveal that HDAC3 deficiency results in cluster-specific pathway dysregulation in AMs at steady state, which may lead to distinct function disorders in each AM cluster.
Interestingly, when we analyzed transcriptomes of AM populations without clustering, we identified a total of 186 differentially expressed genes, including 59 upregulated genes and 127 downregulated genes in the HDAC3-deficient AMs compared to their WT counterparts ( Supplementary Fig. 6a). IPA of these differentially expressed genes showed that several canonical pathways related to inflammation and immune responses were significantly inhibited in the HDAC3cKO AMs ( Supplementary  Fig. 6b). Moreover, the signaling pathways related to macrophage survival and apoptosis were also remarkably impaired, including TREM1 signaling 29,30 , NF-kB pathway 31,32 , and STAT3 pathway 33,34 . qRT-PCR analysis further confirmed the downregulation of the genes associated with these three pathways, such as Flt1, Mapk3, Map3k1, Tlr2, Fcgr2b, Tgfb1, and Tgfbr1 ( Supplementary Fig. 6c). These data suggest that the inhibition of TREM1, NF-kB, and STAT3 signaling pathways might reduce cell survival of HDAC3cKO AMs, leading to the decreased total number of AM cells in the absence of HDAC3 as we observed at steady-state maintenance (Fig. 2d). On the other hand, scRNAseq also revealed that PPAR-γ was remarkably reduced in HDAC3-deficient AMs ( Supplementary Fig. 6a), suggesting that HDAC3 might control AM homeostasis through regulation of PPAR-γ.
HDAC3 controls mitochondria function and survival of pre-AMs. To further obtain mechanistic insights into the embryonic deficiency of AMs in the absence of HDAC3, we first performed bulk RNA-seq analysis of lung preAMs from cKO and WT at E18.5. Based on the principal component analysis (PCA) of their transcriptomes, the molecular signatures showed a clear difference in preAMs between cKO and WT (Fig. 5a). A total of 9074 differentially expressed genes (P < 0.05) were identified in the preAMs between HDAC3cKO and WT, including 4612 downregulated and 4462 upregulated genes (Fig. 5b, Supplementary Data 1). Gene Ontology (GO) analysis further identified the altered biological pathways in the HDAC3cKO preAMs, including multiple metabolic pathways, the electron transport chain, cell proliferation, and cell death (Fig. 5c, Supplementary Data 2). Most of the identified pathways are consistent with previous studies in other cell types with HDAC3 deletion 14,35,36 .
Furthermore, gene set enrichment analysis (GSEA) identified some gene sets that were either depleted (normalized enrichment score, NES < 0, false discovery rate, FDR < 0.05) ( Supplementary   Fig. 7a, b) or enriched in HDAC3cKO phenotype (NES > 0, FDR < 0.05) ( Supplementary Fig. 8a, b). Notably, the genes involved in oxidative phosphorylation were significantly downregulated in HDAC3-deficient preAMs (Fig. 5d). Many subunits of protein complexes I-V of the electron transport chain, the key component of the oxidative phosphorylation process, were diminished in the absence of HDAC3 (Fig. 5e, f). In addition, we observed a significant decrease in Mitotracker Orange fluorescence-labeled AMs from cKO, which reflects the reduced mitochondria membrane potential and indicates impaired mitochondria function in HDAC3-deficient lung preAMs (Fig. 5g). Moreover, GSEA demonstrated that the absence of HDAC3 led to a strong enrichment of genes involved in apoptosis signaling (Fig. 5h-j). Enhanced apoptosis of preAMs at E18.5 was further confirmed by flow cytometry (Fig. 5k). These results suggest that HDAC3 regulates preAM mitochondrial oxidative function and cell death, which may be related to defective embryonic AM development in HDAC3 cKO mice.
HDAC3 deficiency causes impaired PPAR-γ signaling in pre-AMs. We next sought to identify the direct downstream targets of HDAC3 for AM development. To achieve this, we performed HDAC3 ChIP-Seq analysis of lung AMs from the WT mice, which was further integrated with our bulk RNA-seq data. We found 2502 genes shared in both datasets that may be primary targets of HDAC3 transcriptional regulation (Fig. 6a). IPA analysis of these differentially expressed genes revealed that PPAR signaling was one of the canonical pathways potentially targeted by HDAC3 (Fig. 6b). Indeed, the repression of Pparg mRNA expression was correlated with the loss of HDAC3 in HDAC3deficient preAMs (Fig. 6c). ChIP-seq analysis also showed that HDAC3 can directly bind to Pparg enhancers that are located in the upstream (within −30 kilobases (kb)) or intronic (+49 kb) regions from the transcription start sites in lung AMs from the WT mice, but not in the HDAC3cKO mice (Fig. 6d, Supplementary Data 3). The binding of HDAC3 to the Pparg enhancers was further confirmed by ChIP-qPCR using MH-S cells, a mouse AM cell line (Fig. 6e). Moreover, the GSEA of bulk RNA-seq data revealed that the absence of HDAC3 led to a significantly weaker enrichment of genes involved in PPAR-γ signaling compared to WT (Fig. 6f, g). Lack of HDAC3 resulted in the downregulation of PPAR-γ target genes, including Ppargc1b, Acsl1, Cidec, Fabp4, Mgst3, and Nr1h3, which was further confirmed by qRT-PCR (Fig. 6h). Collectively, these results indicate that HDAC3 directly targets PPAR-γ in AMs, and that loss of HDAC3 fundamentally abrogates the PPAR-γ signaling pathway during AM development.
PPAR-γ regulates mitochondria function and survival of pre-AMs. A prior study has shown impaired development of fetal lung preAMs in Vav1-Cre-induced, PPAR-γ-deficient mice at E18.5 7 , indicating that prenatal PPAR-γ is required for AM ontogeny. Using Csf1r iCre Pparg fl/fl cKO mice, we further found a modest reduction of F4/80 hi CD11b int YS pMac-derived lung fetal macrophages at E14.5 and a drastic loss of F4/80 int CD11b int fetal liver monocyte-derived preAMs at E17.5 from Csf1r iCre Pparg fl/ fl cKO mice (Fig. 7a), which were consistent with the phenotype observed in HDAC3cKO mice (Fig. 1d, e). As expected, CD11c hi Siglec-F hi AMs were essentially eliminated in the Csf1r i-Cre Pparg fl/fl cKO adult mice (Fig. 7a), which was consistent with the previous results using Vav1 Cre -and Cd11c Cre Pparg fl/fl cKO mice 7 . Moreover, PPAR-γ gene expression was detected in the lung fetal macrophages at E14.5, which was gradually upregulated in preAMs at E16.5 and E18.5, and maintained at high levels in the AMs from young and adult mice (Fig. 7b). Interestingly,      PPAR-γ mRNA was not expressed in the TRMs of the kidney, liver, brain, and heart (Supplementary Fig. 9), suggesting that HDAC3-PPAR-γ axis is specifically required for the development of lung AMs, but not for TRMs in other tissues. We next revisited a recently-published microarray dataset 7 and analyzed gene expression profiles of lung preAMs from Pparg fl/fl WT and Cd11c Cre Pparg fl/fl cKO mice at day 2 after birth. As expected, the PPAR-γ signaling pathway in preAMs was severely impaired in the absence of PPAR-γ, which was consistent with our observations in Csf1r iCre Hdac3 fl/fl cKO mice (Fig. 7c). Interestingly, oxidative phosphorylation-related genes, which were severely diminished in HDAC3cKO mice, were robustly reduced in PPAR-γ-deficient mice (Fig. 7d-f). ChIP-qPCR analysis using MH-S cells further showed an enrichment of PPAR-γ binding sites in the promoter regions (within 3 kb of the transcription start sites) of oxidative phosphorylation-related genes, including Atp5l, Cox5a, Cox5b, Cox6a1, Ndufa4, Sdhb, Uqcrfs1, and Uqcrq, all of which were downregulated in the absence of PPAR-γ (Fig. 7e, f), suggesting a possible direct impact of PPAR-γ on mitochondria function-related genes in AMs (Fig. 7g). Moreover, the apoptosis-associated genes were highly enriched in PPAR-γdeficient mice compared to their WT counterparts, which was also consistent with our observations in HDAC3cKO mice (Fig. 7h-j). These data suggest that the PPAR-γ signaling pathway serves as a key pathway that contributes to the dysregulation of mitochondrial oxidative function and cell survival in the HDAC3-deficient AMs. Taken together, our study demonstrates that the HDAC3-PPAR-γ axis is indispensable for normal mitochondrial oxidative function and cell survival during AM development (Fig. 7k).

Discussion
The concept of TRMs has been refreshed in the past few years. Recent lineage-tracing studies have revealed that adult mouse TRMs are derived from embryonic YS EMP-derived macrophages and FL monocytes rather than from bone marrow-derived precursors, and they are self-maintained at steady state after F2r Smad7 . Although a few genes (e.g., Csf2, Pparg, and Tgfb) have been identified as requirements for AM development, little information is known about the epigenetic regulators that instruct the development and maintenance of AMs. HDAC3, belonging to the class I HDAC family, plays a critical role in many biological processes, including circadian rhythm 17 , intestinal homeostasis 38 , and tissue thermogenesis 14 . Here, we demonstrate that HDAC3 is required for embryonic development and postnatal maintenance of lung AMs at steady state. In addition, HDAC3 controls AM regeneration from BM progenitors after lethal whole-body irradiation. Our data further indicate that HDAC3 is dispensable for development of TRMs in the brain, spleen, kidney, and pancreas, suggesting that HDAC3 regulates TRM ontogeny in an organspecific manner.
As the resident macrophages in lung alveoli, AMs are long lived with proliferative self-renewal capability, and control the maintenance of airway homeostasis through their capacity of clearing surfactant and cell debris at steady state 1,39 . During respiratory infection or inflammation, AMs also function as antigen-presenting cells by transporting exogenous antigens to lung-draining lymph nodes and releasing a wide array of cytokines and chemokines to reprogram both innate and adaptive immune responses 40,41 . However, it remains unclear whether multifunctional roles of AMs are derived from AM heterogeneity at the individual cell level. Recent advances in single-cell technology have provided a powerful tool to study tissue and cell heterogeneity 28,[42][43][44] . To address if the AMs are heterogenous at steady state and if HDAC3 regulates AM heterogeneity and subset function, we performed scRNA-seq of mature AMs in the adult mice from WT and CD11cCre.HDAC3 KO mice. We found that the murine lung tissues contain four transcriptionallydistinct AM cell subsets. The most abundant cluster (cluster 1) accounted for over 85% of the total lung AMs and expressed high levels of genes related to lipid metabolism, including Abcg1 45 and Ltc4s 46 . Thus, this abundant AM cluster may be specialized for pulmonary surfactant lipid clearance. It has been shown that the adult AM network is maintained throughout life by a slow local proliferation of lung AMs rather than from BM precursors 8 . Interestingly, one of the minor AM clusters (cluster 2, 6%) was characterized by higher expression of cell cycle-related genes, and may represent proliferating AM cells that slowly renew the AM network, which has yet to be further tested. Another minor AM cluster (cluster 3, 3%) was enriched in many antigenpresentation-related genes and may represent the AM subpopulation that possesses antigen-presenting function 47 . Furthermore, the most minor AM cluster (cluster 4, 4%) highly expressed a variety of genes involved in immune responses and may serve as the major source of AMs for secreting pro-inflammatory cytokines and chemokines during inflammatory conditions. Together, our data suggest that individual AM subpopulations might be specialized for unique functions to regulate lung homeostasis and antipathogen immunity.
Interestingly, HDAC3 deficiency resulted in unproportionable changes in the number of AM sub-cluster cells, indicating a cluster-dependent effect of HDAC3 on AM homeostasis. Moreover, the transcriptomic changes and their associated pathway dysregulation induced by HDAC3 deletion in AMs also showed a cluster-specific pattern, i.e., mitochondria dysfunction in cluster 1, cell cycle control disorders in cluster 2, antigen-presenting dysfunction in cluster 3, and immune response dysregulation in cluster 4. Notably, the mitochondria dysfunction found in the abundant AM cluster (cluster 1) is consistent with our observations from bulk-RNA seq data showing impaired mitochondrial oxidative function and deteriorative cell death in the absence of HDAC3 during AM development. Given the importance of mitochondria function in immune metabolism and cell survival of macrophages 48 , we reason that the loss of cluster 1 caused by impaired mitochondria function largely contributes to the reduced number of total AM cells in the HDAC3cKO mice. Moreover, the disorders of cell cycle control and DNA damage response in the proliferating cluster (cluster 2) likely lead to less proliferative capacity of this cluster, which may also partially contribute to the reduced number of HDAC3-deficient AM cells. On the other hand, loss of HDAC3 significantly affected the antigen-presenting function of cluster 3 and the immuneresponsive function of cluster 4, suggesting that HDAC3 might crucially control the function of both AM clusters during immune responses. Further studies are needed to assess the role of HDAC3 in the functionality of individual AM subpopulations in the pathogenesis of lung diseases, such as asthma and cancer. Furthermore, it remains to be determined whether fetal immature preAMs have subpopulations and how HDAC3 regulates preAM subsets, if any, during embryonic development. However, given severe developmental defects on preAMs in the absence of HDAC3, we currently could not obtain enough cells from the KO littermates for scRNA-seq analysis. Advanced technology, e.g., mass cytometry, may serve as a promising and powerful tool to address these interesting questions in the future 49 .
inflammatory lung diseases 20,21 . However, due to common side effects of these pan HDAC inhibitors, developing drugs with high selectivity for individual HDAC has drawn great attention. On the other hand, different subtypes of HDACs appear to play disparate roles in immune defense and disease progression 52 . Understanding the underlying mechanism by which individual HDACs regulate immune cell development, maintenance, and function will help us to identify potential therapeutic treatments. In the present study, deletion of HDAC3 leads to repressed PPAR-γ expression in fetal immature AMs, as well as adult mature AMs, as confirmed by bulk RNA-seq and scRNA-seq analyses, respectively, suggesting the essential role of the HDAC3-PPAR-γ axis in the development and maintenance of AMs. In addition, our data imply that the HDAC3-PPAR-γ axis controls fetal immature AM development, at least partially through regulation of mitochondria function and cell survival. Interestingly, PPAR-γ binding sites are enriched near the genes associated with oxidative phosphorylation, suggesting that PPAR-γ might directly regulate mitochondrial oxidative function in AMs. On the other hand, given lack of PPAR-γ expression in TRMs from other organs, the HDAC3-PPAR-γ axis seems to be only required for TRM development in a lung-specific manner. It should be noted that a very recent single-cell study revealed an enrichment of PPAR-γ hi macrophages in human lung adenocarcinoma lesions that likely promotes an immunosuppressive microenvironment 53 . Thus, HDAC3i could be promising immunotherapeutic drugs to treat lung cancer patients by targeting the HDAC3-PPAR-γ axis in lung macrophage populations, which needs to be further investigated.
In conclusion, our data have provided evidence that HDAC3 serves as a key epigenetic regulator to control AM development at the embryonic stage, as well as in AM maintenance and regeneration after birth. Accumulated research has indicated that AMs are involved in the development of cancer, autoimmune, and inflammatory diseases in the airway. Thus, our findings may shed light on HDAC3 as a potential therapeutic target for AM-based intervention in cancer and autoimmune diseases.
Male and female mice from 6-18 weeks of age were used. All experiments included age-matched and sex-matched littermate controls. Embryonic development was estimated considering the day of vaginal plug formation as embryonic age of 0.5 days.
All mice were housed under specific pathogen-free conditions at temperatures of 20-26°C with 30-70% humidity and a 12-h light/12-h dark cycle at Henry Ford Health System. Experimental animal protocols were performed in accordance with the guidelines of the Institutional Animal Care and Use Committee.
Isolation of embryonic and adult cells. Pregnant females were sacrificed by CO 2 exposure. Embryos ranging from embryonic days E10.5-E18.5 were removed from the uterus and washed in 4°C phosphate-buffered saline ([PBS], Invitrogen, Carlsbad, CA). Embryos were exsanguinated through decapitation. Lungs, liver, brain, kidneys, spleen, and pancreas were collected, minced into tiny pieces, and incubated for 30 min in PBS containing 1 mg/ml collagenase D (Roche, Basel, Switzerland), 0.01% DNase I (Worthington Biochemical Corp., Lakewood, NJ), and 3% FBS (Hyclone, San Angelo, TX) at 37°C. Erythrocytes from all the samples were lysed for 3 min with 0.83% NH 4 Cl buffer. The cells from the adult tissues were isolated by the same method with some modifications. The adult lungs were harvested, minced, and incubated in PBS containing 1 mg/ml collagenase D (Roche), 0.01% DNase I (Worthington), and 3% FBS (Hyclone) at 37°C for 45 min. The erythrocytes were lysed as described above. All the cell suspensions were passed through a 70 μm cell strainer (BD Biosciences, San Jose, CA).
Flow cytometry and cell sorting. Single-cell suspensions were centrifuged at 450 × g for 7 min, resuspended in ice-cold staining buffer (1× PBS with 2% FBS), and placed in 96-well round-bottom plates. After incubation with purified anti-FcγRII/III antibody (clone 2.4G2) at 4°C for 15 min, cells were stained with a mixture of fluorescent surface antibodies at 4°C for 30 min. The stained samples were either directly analyzed by flow cytometry or fixed with 2% formalin at 4°C for 20 min prior to storage at 4°C overnight. The full list of antibodies used can be found in Supplementary Table 1. Flow cytometry was performed with FACSAria™ II or FACSCelesta™ flow cytometer (BD Biosciences). Data were acquired using BD FACSDiva software version 8.0.2 (BD Biosciences) and analyzed using FlowJo 10.5.3 (BD Biosciences).
Mitochondria function and cell apoptosis assays. The MitoTracker Orange (Invitrogen/Thermo Fisher Scientific) stock solution was prepared according to the manufacturer's instructions. Briefly, 50 μg of lyophilized MitoTracker Orange dye was dissolved in DMSO to a final concentration of 1 mM and stored at −20°C in the dark. Cells were incubated with 50 nM MitoTracker Orange in the prewarmed RPMI 1640 medium at 37°C with 5% CO 2 for 15 min , followed by two washes with PBS. The stained cells were analyzed by flow cytometry (Ex 554 nm, Em 576 nm).
Cell apoptosis was measured using the Annexin V Apoptosis Detection Kit (eFuor TM 450 or APC, eBioscience/Thermo Fisher Scientific). In brief, freshlyisolated cells were washed once with 1× binding buffer and resuspended in 100 μL of 1× Binding Buffer at 5 × 10 6 /ml. After adding 3 μL of fluorochrome-conjugated Annexin V, the cells were incubated at RT for 13 min. The stained cells were washed once with 1× binding buffer and analyzed by flow cytometry.
Tamoxifen treatment. Tamoxifen (Sigma Aldrich, St. Louis, MO) was dissolved in corn oil (Sigma Aldrich) containing 10% (vol/vol) ethanol (Thermo Fisher Scientific) and was intraperitoneally administered at 50 mg/kg of mouse weight for five consecutive days.
Bone marrow chimeras. B6.SJL Hetero (CD45.1 + CD45.2 + ) mice were lethally irradiated once with 950 rads. Donor BM cells were harvested from age-matched and sex-matched CD45.2 + Hdac3 fl/fl and Csf1r iCre Hdac3 fl/fl mice, as well as B6.SJL (CD45.1 + ) mice. The BM cells from Hdac3 fl/fl or Csf1r iCre Hdac3 fl/fl mice were mixed with the cells from B6.SJL mice (at a 4:1 ratio) and then co-transferred to the irradiated B6.SJL Hetero recipient mice (10 × 10 6 mixed cells/mouse). The recipient mice were sacrificed 10 wks after reconstitution. Immunofluorescence. A piece of lung or liver freshly isolated from adult mice or embryos at E18.5 was immediately frozen in OCT compound for cryostat sectioning (5 μm per section). The frozen tissue sections were fixed with acetone for 15 min at −20°C. After rinsing with PBS for 5 min, samples were blocked with anti-FcγRII/III antibody at RT for 15 min followed by staining at 4°C overnight with PE-labeled anti-Siglec-F (adult lung) or PE-labeled anti-F4/80 (fetal lung) antibody. The samples were then stained with 1 μg/ml DAPI at RT for 1 min and mounted under a coverslip with one drop of IMMU-MOUNT (Thermo Fisher Scientific). Specimens were visualized on an Olympus FSX100 fluorescence microscope (Olympus Scientific Solutions Americas Corp, Waltham, MA) at ×10 magnification. At least 5 independent images were collected from each specimen. Images were collected using FSX-BSW (version 03.02.12) software (Olympus). The cell density of AMs (Siglec-F + ) or preAMs/TRMs (F4/80 + ) was measured based on cell numbers counted in the fields using cellSens Dimensions Imaging Software version 1.15 (Olympus).

RNA extraction and
Bulk RNA sequencing. To achieve optimal library quality, cDNA synthesis and preamplification for total RNA of the sorted lung preAMs from WT or Csf1r i-Cre Hdac3 fl/fl cKO embryos at E18.5 were performed using the SMART-Seq v4 Ultra Low Input RNA Kit (Clontech, Mountain View, CA) following the manufacturer's instructions. Briefly, cDNA was pre-amplified by a thermal cycler (Applied Biosystem) using 12-14 cycles and sheered into 150-400 bp fragments using a Bioruptor Pico sonicator (Diagenode, Denville, NJ). The DNA ends were repaired using End-It™ DNA End-Repair Kit (Epicentre, Charlotte, SC). The end-repaired DNA fragments were ligated to adapters by T4 ligase (New England Bioloabs, Ipswich, MA) after a single A was added at 3′ by Klenow Fragment (3′ → 5′ Exo-) (New England Bioloabs). MinElute Reaction Cleanup Kit (Qiagen, Gaithersburg, MD) was used for DNA cleanup after each step. Illumina P5 and P7 primers with indicated adapters (Supplementary Table 3) were used for final library amplification. The 150-400 bp final product was purified on a 2% E-gel (Thermo Fisher Scientific). Sequencing was performed by the DNA sequencing core facility at University of Michigan using a 50 bp single end read setup on the Illumina Hiseq 4000 platform.
Single-cell RNA sequencing. Single-cell sequencing libraries were generated by an established protocol using the 10× Genomics Chromium Single Cell 3′ Reagent Kit (v2 Chemistry) and Chromium Single Cell Controller 59 . Briefly, 5000 FACS-sorted cells were loaded into each reaction for gel bead-in-emulsion (GEM) generation and cell barcoding. Reverse transcription of the GEM (GEM-RT) was performed using a Veriti™ 96-Well Fast Thermal Cycler, (Applied Biosystems) at 53°C for 45 min, 85°C for 5 min, and a 4°C hold. cDNA amplification was performed after GEM-RT cleanup with Dynabeads MyOne Silane (Thermo Fisher Scientific) using the same thermocycler (98°C for 3 min, 98°C for 15 s, 67°C for 20 s, 72°C for 1 min, followed by a 12 cycle repeat at 72°C for 1 min and a 4°C hold). Amplified cDNA was cleaned up with SPRIselect Reagent Kit (Beckman Coulter, Brea, CA) that was followed by a library construction procedure, including fragmentation, end repair, adapter ligation, and library amplification. An Agilent 2100 Bioanalyzer (Agilent, Santa Clara, CA) was used for library quality control. Libraries were sequenced on an Illumina HiSeq4000 using a paired-end flow cell: Read 1, 26 cycles; i7 index, 8 cycles; Read 2, 98 cycles.
ChIP sequencing. Cell pellets from freshly-isolated AMs (1.38 × 10 6 and 0.77 × 10 6 cells for WT and cKO, respectively) of adult Hdac3 fl/fl WT (n = 13) or Csf1r i-Cre Hdac3 fl/fl cKO mice (n = 26) were directly frozen at −80°C. The samples were then submitted to EpiGentek (Farmingdale, NY) for ChIP, library preparation, and sequencing. Briefly, the cells were fixed with 1% formaldehyde and chromatin was isolated using a ChromaFlash TM Chromatin Extraction Kit. Chromatin was sheared using the Episonic2000 Sonicatin System. ChIP was performed using an antibody against HDAC3 (Abcam, #ab7030, polyclonal, 2 µg/500 µl assay buffer) on the chromatin, and input DNA (without immunoprecipitation) was used as background. After adapter linking, the library was size selected (100-300 bp) and PCR amplified. Ten nanomolar of each sample library was provided for next generation sequencing on a HiSeq 4000.
scRNA-seq data processing. Paired-end read counts for all known murine genes underwent barcode and unique molecular identifier (UMI) deconvolution. They were subsequently aligned to the mm10 reference genome using the 10× Cell RangerTM [v2.1] pipeline 67,68 . A total of 2000-3000 cells were captured, with 53,000-73,000 mean reads per cell, and 1200-1300 median genes per cell. Datasets were subsequently analyzed using R-3.3.2 package Seurat's Canonical Correlation Analysis (CCA) workflow to characterize the differences between pooled samples 69 .
The following quality control metrics were employed. To identify potential doublets, cells expressing uncharacteristically high numbers of genes (>2500) were excluded (double the median number of genes). Low-quality cells were excluded based on a low number of genes detected (<300) or having high mitochondria genetic content (>5.0%). In addition, uninteresting sources of variation within the data were removed. Genes removed include sex-specific genes (Xist and all Y chromosome genes), ribosomal structural proteins as identified by gene ontology term GO:0003735 and the RPG: Ribosomal Protein Gene database 70 , non-coding rRNAs, Hbb, and genes expressed in <3 cells.
The global-scaling normalization method, LogNormalize, was employed to normalize gene expression measurements of each cell by the total expression, multiplying this by a factor of 10,000, followed by log-transformation. Highly variable genes in each data analysis were identified, and the intersecting top 500 genes in each dataset were used for clustering and downstream analyses. Datasets underwent scaling based on nUMI and percent mitochondria gene content. The number of CCs (1:15) used to cluster cells was determined by manual inspection of scree plots. Further pathway analysis was performed using IPA.