A Cell Atlas of Microbe-Responsive Processes in the Zebrafish Intestine

Gut microbial products direct growth, differentiation and development in the animal host. Disruptions to host-microbe interactions have profound health consequences, that include onset of chronic inflammatory illnesses. However, we lack system-wide understanding of cell-specific responses to the microbiome. We profiled transcriptional activity in individual cells from the intestine, and associated tissue, of zebrafish larvae that we raised in the presence, or absence, of a microbiome. We uncovered extensive cellular heterogeneity in the conventional zebrafish intestinal epithelium, including previously undescribed cell types with known mammalian homologs. By comparing conventional to germ-free profiles, we mapped microbial impacts on transcriptional activity in each cell population. We revealed intricate degrees of cellular specificity in host responses to the microbiome, that included regulatory effects on patterning, metabolic and immune activity. For example, we showed that removal of microbes hindered transduction of vascular endothelial growth factor-dependent signals in the developing vasculature, resulting in impaired intestinal vascularization. Our work provides a high-resolution atlas of intestinal cellular composition in the developing fish gut and details the effects of the microbiome on each cell type.


INTRODUCTION
Research conducted with a variety of model organisms has revealed much about the importance of gut microbes for host health. Animals raised in sterile, germ-free environments, frequently exhibit defects in growth, immunity and metabolism (Bates et al., 2006;Hooper et al., 2001;Rawls et al., 2004;Reikvam et al., 2011). Of equal importance, sudden changes in composition or distribution of gut microbial communities are associated with severe and sometimes deadly illnesses, including gastrointestinal cancers and inflammatory bowel diseases (Belkaid and Hand, 2014;Zitvogel et al., 2015). Thus, it is critical to fully understand how the microbiota impacts development, growth, and cellular function of host organisms.
In recent years, zebrafish larvae have emerged as a valuable tool to characterize molecular and cellular regulators of host-microbe interactions (Brugman, 2016;Flores et al., 2020). Zebrafish develop within a protective chorion that shields them from environmental microbes up to forty-eight hours post fertilization. Once larvae exit the chorion, water-borne microbes colonize the gut lumen (Bates et al., 2006;Stephens et al., 2016;Wallace et al., 2005), where they influence development of host intestinal and extra-intestinal tissues (Bates et al., 2006;Cheesman et al., 2011;Kanther et al., 2011;Koch et al., 2018). From a technical perspective, zebrafish offer several advantages to pinpoint key events in developmental responses to the microbiota. Larvae are amenable to sophisticated genetic manipulations, including genetic modifications from the single cell stage (Grunwald and Eisen, 2002). Additionally, researchers have established simple protocols to generate large numbers of germ-free larvae, or larvae associated with defined microbial communities (Melancon et al., 2017;Pham et al., 2008), and the translucent epidermis is ideal for visualization of internal structures in fixed or live samples. Thus, zebrafish provide a convenient window to visualize microbial controls of vertebrate physiology.
Importantly, genetic regulation of intestinal function is highly similar between zebrafish and mammals. In both systems, orthologous signals, including those driven by Notch, BMP and Wnt pathways, direct development of absorptive and secretory cell lineages from cycling progenitors in the intestinal epithelium. While the zebrafish intestine possesses mucin-producing goblet cells and regulatory enteroendocrine cells, they appear to lack immune-modulatory Paneth and tuft cells. At present, our understanding of absorptive lineages is less complete. Similar to mammals, the fish intestinal epithelium includes regionally specialized enterocytes that harvest nutrients from the lumen Ng et al., 2005;Park et al., 2019;Wallace et al., 2005;Wang et al., 2010b). However, the extent of functional heterogeneity within enterocyte populations is unclear, and we do not know if the intestinal epithelium houses specialized absorptive cells such as antigen-capturing M cells, or recently described Best4/Otop2 cells (Parikh et al., 2019;Smillie et al., 2019). Likewise, despite experimental evidence for the existence of cycling progenitors (Crosnier et al., 2005;Rawls et al., 2004;Wallace et al., 2005), we lack expression markers that permit identification and manipulation of this essential cell type. Combined, these deficits have hampered our ability to harness the full potential of the zebrafish as a model of intestinal biology and host-microbe interactions.
From a microbial perspective, similarities between fish and mammals are also quite evident. For example, fish also rely on a complex network of germline-encoded innate defenses, and lymphocytebased adaptive defenses that prevent invasion of interstitial tissues by gut-resident microbes (Flores et al., 2020). Transcriptional studies showed that orthologous genes mediate microbe-dependent control of epithelial proliferation, nutrient metabolism, xenobiotic metabolism, and innate immunity (Hooper et al., 2001;Koch et al., 2018;Rawls et al., 2004;Reikvam et al., 2011). In vivo studies support a shared role for the microbiota in developmental processes including epithelial renewal, secretory cell differentiation, and gut motility (Cheesman et al., 2011;Troll et al., 2018;Wiles et al., 2016). Microbes also educate immune systems in fish and mammals, inducing control of mucosal inflammation and myeloid cell recruitment through Myd88-dependent TLR signals (Galindo-Villegas et al., 2012;Koch et al., 2018;Takeda and Akira, 2005). Like mammals, fish neutralize pathogenic bacteria via epithelial production of reactive oxygen species and antimicrobial peptides (Flores et al., 2010;Katzenback, 2015). Additionally, experimental evidence in fish revealed that epithelial alkaline phosphatase can detoxify LPS, a finding later corroborated in a mouse model (Bates et al., 2007;Goldberg et al., 2008). These findings demonstrate that zebrafish, alongside other models, can inform our understanding of microbial impacts on host development and disease.
While recent studies provided significant insights into microbe-dependent host processes, most of this work focused on how microbes impact whole organisms, the entire intestine, or intestinal regions.
Thus, a large knowledge gap exists in our understanding of cell type-specific processes that rely on microbial signals. A few studies addressed this disparity via FACS sorting of intestinal epithelial subpopulations (Arora et al., 2018), however this method requires foreknowledge of, and access to, cell typespecific antibodies or reporters, and assays are limited to one cell type per experiment.
To achieve unbiased cell type-specific analysis of microbe-dependent processes, and to overcome limited cell characterization of the fish intestine, we prepared a single cell transcriptional atlas of intestines, and associated tissue, from six-day post-fertilization zebrafish larvae raised in a conventional environment, or in the absence of a microbiome. In conventional fish, we identified nineteen distinct cell types in the gut and associated tissue, several of which were previously undescribed. We completed a high-resolution map of cellular responses to the microbiota that showed cell-specific microbial impacts on growth, patterning, and immunity in the host. In summary, our work resolves the cellular composition of the zebrafish intestine and defines cell-specific host transcriptional networks regulated by the microbiota.

A single cell atlas of the larval intestine and associated tissue
To trace the effects of commensal microbes on intestinal physiology, we prepared single-cell transcriptional profiles of gastrointestinal tracts, and associated tissue, from zebrafish larvae raised under conventional (CV) or germ-free (GF) conditions for six days after fertilization ( Figure 1A and Supplementary Figure 1). After filtering for dead cells and doublets, we determined gene expression profiles for 3,521 individual cells (1,939 GF; 1,582 CV). From binomial distribution analysis, we estimate that we captured cell types present at greater than 1% abundance, allowing us to identify most cell populations in the dissected samples.
To fill gaps in our understanding of intestinal cellular composition, we used a graph-based clustering approach to identify distinct cell types in the guts and associated tissues of CV larvae. We identified eighteen clusters of endodermal or mesodermal origin, and assigned identities based on expression of established fish or mammalian markers ( Figure 1B). Within the intestinal epithelium, we identified secretory cells marked by expression of genes required for production and release of peptide hormones  (Vanuytsel et al., 2013), our data raise the possibility that similar intercellular interactions influence growth in the developing zebrafish gut.
Finally, we found that vascular and myeloid cells were marked by expression of genes that control vasculogenesis, and immune cell activation and migration, respectively (Supplementary Figure 5).
In sum, we have unambiguously identified a panel of expression markers that distinguish all major intestinal lineages, including previously undescribed candidate progenitor cells.

Heterogeneity of absorptive lineages within the intestinal epithelium.
Like most animals, the zebrafish intestinal epithelium is dominated by regionally-specialized subtypes of absorptive cells required for acquisition and transport of material from the gut lumen Ng et al., 2005;Park et al., 2019;Wallace et al., 2005;Wang et al., 2010b). To uncover the heterogeneity of absorptive cell types within the zebrafish gut, we examined CV larvae for gene expression patterns that indicated regionally or functionally specialized absorptive cells.
86% of all CV intestinal epithelial cells were enterocytes that we stratified into seven groups based on distinct gene expression signatures ( Figure 2A). Individual clusters frequently had gene expression signatures consistent with localization to defined regions of the intestinal epithelium Wang et al., 2010b). For example, clusters one to three were functionally related and expressed genes involved in energy metabolism ( Figure 2B), as well as positional markers such as ada and fabp2 that indicate anterior localization ( Figure 2D). Cluster four was the dominant enterocyte type that accounted for 21% of all enterocytes, and also expressed markers consistent with anterior localization ( Figure 2D).
Functionally, cluster four cells were metabolically active, expressing genes required to process carbohydrates, lipoproteins and chitin ( Figure 2B). Cluster five cells were dedicated to ion transport and peptide catabolism ( Figure 2B), and positional marker expression patterns revealed reduced expression of anterior markers alongside upregulation of middle and posterior markers relative to ECs 1-4 ( Figure   2D), suggestive of middle to posterior intestinal localization. In contrast, cells from cluster seven were highly enriched for middle intestinal markers such as slc10a2 and tmigd1 ( Figure 2D), and expressed gene products involved in oxidative stress responses ( Figure 2B) indicating separate roles for putative distal EC clusters five and seven in ion absorption, or metabolite detoxification. Finally, we uncovered an enterocyte type -cluster six -that appeared predominantly anteriorly localized ( Figure 2D) and was highly enriched for expression of gene products that transduce interferon pathway signals ( Figure 2B), suggesting the zebrafish epithelium contains specialist enterocytes dedicated to interferon-based regulation of gut function.
In addition to conventional enterocytes, we found that the intestinal epithelium contained three additional absorptive cell types, two of which have not yet been described in fish. First, we manually searched our dataset for lysosome-rich enterocytes (LREs), which are protein-capturing enterocytes of the posterior middle intestine recently described in zebrafish (Park et al., 2019). We identified a small subset, originally clustered with EC7, that expressed LRE marker genes and was enriched for small molecule metabolism and transport ( Figure  In summary, the larval intestine houses a community of specialist cells that display a previously underappreciated degree of similarity to mammalian intestines, including mesenchymal sources for growthregulatory pathways; cycling cells capable of generating the intestinal epithelium; and regionallyspecialized absorptive cells that capture bulk material from the lumen.

Germ-free growth disrupts gene expression throughout the intestine and associated tissue.
Microbial factors direct development, differentiation, and function of host tissues. However, cellspecific microbe responses have been difficult to chart to date. To assess microbe-responsive genes at the cellular level, we initially confirmed microbial effects on expression of 24 previously identified microberesponsive genes (Rawls et al., 2004) and expanded these findings to show that many of the effects occur in subsets of host cells (Supplementary Figure 8). For example, expression of moesin a, involved in blood vessel development (Wang et al., 2010a), was specifically downregulated in vascular endothelial cells of GF fish. Next, we used our transcriptional data to resolve impacts of GF growth on all cell lineages described in this study. First, we mapped intestinal organization in GF larvae. Similar to CV counterparts,  Figure 9C). Additionally, extra-intestinal tissue of GF fish contained mesenchymal cells, acinar cells, hepatocytes, and myeloid cells dedicated to morphogenesis, metabolism, and immunity, among other processes (Supplementary Figure   9E and 9F). Thus, at a superficial level, intestines of CV and GF larvae appear similar in cellular anatomy, arrangement, and biological function.
However, examination of individual cell types, or biological pathways, uncovered pronounced effects of microbiome removal. First, we found that GF growth resulted in substantial shifts in cell type representation ( Figure 3C). Relative to CV larvae, GF larvae had elevated numbers of absorptive cells, fewer mesenchymal and progenitor-like cells, and a substantial depletion of cells that expressed vascular or myeloid markers. Among enterocytes, we detected a complete loss of enterocyte cluster six cells that express interferon pathway components ( Figure 3A and 3C). For example, ifit14 and ifit15, orthologous to human IFIT1 and IFIT2, were either lowly expressed or absent from GF epithelia ( Figure 3B and Supplementary Figure 10). Significant loss of CV marker genes was also detected in the progenitor-like cell compartment, where secretory cell differentiation factors dld, atoh1b, and myb were largely absent from her positive progenitor-like cells, instead appearing at low levels in Best4/Otop2 cells ( Figure 3B).
Our observation that GF intestines were defective in development of EC6, an immune-dedicated EC type, prompted us to assess immune response pathways in GF relative to CV intestines. For these studies, we examined the consequence of GF growth for expression of a representative set of microbial sensors, NF-kB pathway components, cytokines and chemokines. Upon integration of CV and GF datasets, we observed striking changes to germline-encoded immune response pathways. In CV larvae, we detected cell-restricted expression of key immune sensors and effectors ( Figure 3D). For example, myeloid cells expressed immune-regulatory cytokines such as cxcl8a, il1b, il17a/f3, tnfa, and tnfb, whereas the vasculature was characterized by enriched expression of microbial sensors such as tlr4ba and nod1, and the inflammation regulator ahr2. Similar to mammals, hepatocytes of CV larvae expressed the hamp antimicrobial peptide, while mesenchymal cells were characterized by enriched expression of NF-kB pathway members, the cxcl8b and cxcl12 chemokines, and the tgfb1a, tgfb2 and tgfb3 cytokines. Within the intestine, most enterocyte subtypes produced alpi2, a phosphatase required for detoxification of bacterial lipopolysaccharide, whereas endocrine cells expressed il22, an inflammatory cytokine that activates innate defenses in epithelial cells (Dudakov et al., 2015). Consistent with previous reports (Kanther et al., 2011), saa was expressed in mid-intestinal enterocyte population EC7, and we also observed enhanced nod2 peptidoglycan sensor expression in CV progenitor cells, similar to mice (Nigro et al., 2014). Removal of the microbiome significantly impacted organization of immune pathways in developing larvae. In particular, we noted a near complete loss of il17a/f3 and cxcl19 expression from myeloid tissue; il22 from endocrine and vascular endothelial cells; nod2 from progenitors; and greatly diminished expression of tgf isoforms in EC5, Best4/Otop2 cells, and mesenchymal cells. Combined, our data demonstrate a sophisticated partitioning of immune gene expression patterns across larval cell types, many of which indicate shared microbe-response pathways in zebrafish and mammals and show that growth of zebrafish larvae in the absence of a conventional microbiome disrupts immune processes throughout the host.

Cell-specific patterns of host responses to the microbiome.
Our analysis of microbe-regulated immune activity, and microbe-responsive marker gene expression patterns indicated cell-specific responses to a conventional microbiome. Consistent with a hypothesis that each cell type responds differently to microbial elimination, we found that GF growth caused substantial changes to host gene expression in enterocytes ( Figure 4A) but had minimal effect on acinar cells ( Figure   4B). Specifically, we identified plac8.1, whose mouse homologue is required to temper inflammation in response to microbes (Ledford et al., 2007), as the most downregulated gene in the collective enterocyte population ( Figure 4A). Comparatively, apoc1 and apoa2, lipoprotein components, were only moderately downregulated in acinar cells ( Figure 4B). Within the progenitor-like compartment, removal of the microbiota resulted in diminished expression of mitotic factors, and elevated expression of genes required for cell-cell adhesion ( Figure 4C, E), consistent with growth-promoting roles for the microbiome (Cheesman et al., 2011). Endocrine cells adjusted to removal of the microbiome with diminished expression of genes that control processing and transport of peptide hormones ( Figure 4C, F), and a significant drop in ins expression ( Figure 4F). Across the enterocyte subtypes, we detected gene expression shifts that indicate diminished lipid transport and metabolism, greatly enhanced peptide metabolism, and reduced innate immune activity ( Figure 4C, D), further supporting requirements of the microbiome to promote lipid metabolism (Sonnenburg and Bäckhed, 2016). We uncovered similarly pronounced effects of the microbiome on gene expression in extra-intestinal tissue. In particular, we noted suppression of immune activity in myeloid cells raised without a microbiome, shifts in developmental regulatory processes among mesenchymal cells, and metabolic adaptation in hepatocytes ( Figure 4G). Combined, our data show that growth of a developing larva in the absence of a microbiome changes the composition and transcriptional activity of all intestinal cells examined in this study.

The microbiome is essential for intestinal vascularization.
We discovered a pronounced loss of cells that expressed vasculature markers, and diminished expression of key vasculature developmental regulators in multiple cell types of GF larvae relative to CV controls ( Figure 3C, 4G, 5A). For example, larvae raised in GF conditions expressed lower amounts of proangiogenic transcription factors such as foxf1 and fli1a. Likewise, we detected significant drops in expression of Vascular Endothelial Growth Factor (VEGF)-type receptor tyrosine kinases, such as flt1 and the VEGFR2 ortholog, kdr, a prominent vasculogenesis inducer (Covassin et al., 2006) (Figure 5A).
Combined, these data raise the possibility that GF growth has detrimental consequences for formation of gut-associated vascular tissue. In larvae, intestinal vasculogenesis commences approximately three days after fertilization (Isogai et al., 2001) a time that matches microbial colonization of the lumen. At this stage, a population of angioblasts migrates ventrally from the posterior cardinal vein, establishing the supra-intestinal artery, and a vascular plexus that gradually resolves into a parallel series of vertical vessels and the sub-intestinal vein (Goi and Childs, 2016;Lenard et al., 2015;Nicenboim et al., 2015) ( Figure 5B).
The gut vasculature delivers nutrients from the intestine to the hepatic portal vein, supporting larval growth and development. To determine if the microbiome affects intestinal vasculogenesis, we used kdrl:mCherry larvae to visualize the vasculature of fish that we raised in the presence, or absence of a conventional microbiome for six days ( Figure 5B, C). We did not observe effects of the microbiome on formation or spacing of the supra-intestinal artery and the sub-intestinal vein, VEGF-independent developmental processes. In both groups, the artery and vein effectively delineated the dorsal and ventral margins of the intestine ( Figure 5B, C). In contrast, removal of the microbiome had significant effects on development of connecting vessels, a VEGF-dependent event. Consistent with this, we observed a near 50% reduction of intestinal kdrl:mCherry signal in GF larvae compared to CV counterparts ( Figure 5D).
Thus, we conclude that microbial factors are essential for proper development of the zebrafish intestinal vasculature.

DISCUSSION
Gut microbial factors are critical determinants of animal development (Sekirov et al., 2010).
Comparative studies with CV and GF larvae uncovered numerous microbial effects on the host, including impacts on proliferation, cell fate specification, and metabolism in the intestinal epithelium (Bates et al., 2006;Hooper et al., 2001;Rawls et al., 2004;Reikvam et al., 2011). Importantly, the molecular and genetic networks that determine intestinal development are highly similar between zebrafish and mammals (Davison et al., 2017;Lickwar et al., 2017). Thus, discoveries made with fish have the potential to reveal foundational aspects of host-microbe relationships. However, important knowledge gaps prevent us from maximizing the value of zebrafish-microbe interaction data. In particular, we know considerably less about the cellular composition of the zebrafish intestinal epithelium compared to mice and humans. For example, the zebrafish epithelium contains basal cycling cells that likely act as epithelial progenitors (Crosnier et al., 2005;Rawls et al., 2004;Wallace et al., 2005). However, we lack genetic markers that allow us to identify, or isolate, the progenitor population for experimental characterization. Likewise, it is unclear if the fish gut contains functionally heterogenous absorptive lineages, similar to the mammalian intestine. To bridge these deficits, and permit cell-by-cell definition of host responses to the microbiome, we prepared single-cell atlases of the larval intestine raised under conventional, or germ-free conditions.
We identified nineteen distinct cell types in the gut, and associated tissue, several of which were unknown in fish. Comparisons between conventional and germ-free fish allowed us to delineate impacts of the microbiome on growth, patterning, immune, and metabolic processes in each cell type. We believe these findings constitute a valuable resource that will support efforts to understand how microbes control vertebrate physiology.
Looking at the intestinal epithelium, we identified cycling cells that express classical intestinal stem cell markers, such as the Delta-like ligand dld, the Notch pathway components ascl1a and atoh1b, and several hes-related transcription factor family members that respond to Notch. Notably, expression of cell cycle markers, and dld declined in this population in GF larvae, consistent with diminished epithelial growth in fish raised in the absence of a microbiome (Bates et al., 2006;Rawls et al., 2004). Our data align with related work in vertebrates and invertebrates (Buchon et al., 2009;Cheesman et al., 2011;Rawls et al., 2004;Reikvam et al., 2011), and argue that we have uncovered intestinal stem cell markers for zebrafish larvae. In mammals, Lgr5 is a classical intestinal stem cell marker, with roles in modification of Wnt-dependent signaling (Haegebarth and Clevers, 2009). Wnt is important for microbe-dependent intestinal epithelial growth in zebrafish. However, the zebrafish genome does not appear to encode an lgr5 ortholog. Instead, the fish genome encodes the related lgr4 and lgr6 genes (Hirose et al., 2011). We did not observe enriched expression of lgr4 or lgr6 in the fish gut, raising the possibility that Wnt-Lgr signaling is not essential for specification and growth of fish intestinal stem cells. In this regard, zebrafish may be more akin to Drosophila, where Wnt activity has ancillary roles in midgut intestinal stem cell growth (Lin et al., 2008). In the future, it will be of interest to perform clonal marking studies on fish progenitor-like cells to test their ability to generate a mature epithelium, and to resolve the roles of Wnt-Lgr activity in zebrafish epithelial homeostasis.
Separate to cycling, Notch-positive cells, we identified transcriptional markers for secretory enteroendocrine cells. As a caveat, we failed to identify mucin-producing goblet cells (Crosnier et al., 2005;Wallace et al., 2005) in either of our data sets, indicating a gap in our resolution of secretory lineages that merits addressing in follow-up studies. Comparisons between CV and GF fish revealed pronounced effects of the microbiome on secretory activity within the endocrine lineage. Most notably, we observed a precipitous drop in expression of the insulin ortholog, ins, supporting earlier work that established essential requirements for gut microbes to drive insulin production in larvae (Hill et al., 2016).
Upon examination of absorptive lineages, we identified transcriptional signatures of regionally and spatially specialized enterocytes in CV larvae, as well as an extensive profile of gene expression in lysosome-rich enterocytes that mediate protein absorption and metabolism (Park et al., 2019). Our findings provide a molecular underpinning of the regionally-controlled nature of nutrient metabolism in the zebrafish gut. Perhaps more intriguingly, we also uncovered two absorptive lineages that were unknown in the zebrafish gut. For example, we identified Best4/Otop2-positive enterocytes, a minimallycharacterized cell type only recently discovered in humans (Parikh et al., 2019;Smillie et al., 2019). Given the utility of zebrafish for examination of gut development, particularly in the context of host-microbe interactions, we believe fish will be of considerable value for in vivo characterization of Best4/Otop2 cells.
In addition to metabolically active lineages, we identified a separate absorptive cell type that displays transcriptional and developmental similarities to mammalian M cells, a cell type dedicated to transcytosis of microbial antigens from the lumen to lymphoid tissue (Dillon and Lo, 2019). There is little data on M cells in zebrafish. However, M cells have been identified in salmonids (Fuglem et al., 2010), and our results match an earlier study that posited the existence of posterior M-like cells in the larval zebrafish gut (Løvmo et al., 2017). Thus, it will be of interest to functionally test the actions of putative zebrafish M cells in the future, a strategy that may permit development of a genetically accessible model for M cell function.
The availability of a high-resolution transcriptional atlas of the zebrafish intestine, and associated tissue, allowed us to map microbial effects on each cell type. We successfully validated many known microbe-dependent events, confirming the validity of our data set. Importantly, we also uncovered a large number of unknown microbe-driven processes in the host, and resolved each process to the level of distinct cell clusters. Our work shows that the microbiota regulates growth, developmental, metabolic and immune processes across the host, displaying remarkable cellular specificity. To provide one example, we will discuss effects of the microbiota on host immune activity, however, we note our data permit identification of microbial impacts on a large number of physiological processes.
Our work revealed a hitherto unknown complexity of germline-encoded immune gene expression patterns in CV fish. Absorptive intestinal epithelial cells expressed enriched amounts of detoxifying alkaline phosphatases (Bates et al., 2006), and myeloid-activating serum amyloid A (Kanther et al., 2011;Murdoch et al., 2019). In contrast, progenitor and endocrine cells expressed elevated levels of the bacterial peptidoglycan sensor nod2, key NF-kB pathway elements, and il22, a cytokine that activates epithelial defenses (Dudakov et al., 2015), suggesting a refined partitioning of immune functions among epithelial cell types. Myeloid cells are characterized by elevated expression of pro-inflammatory cytokines such as il1b, tnfa and tnfb, whereas mesenchymal cells are prominent sources of immune-regulatory TGFbeta class cytokines. Comparisons between CV and GF fish uncovered a remarkable input from the microbiome on all these processes, with cell-specific expression of many immune effectors and mediators declining, relocating, or disappearing almost entirely in GF fish. This was particularly evident in the EC6 population of CV fish that were characterized by high levels of interferon activity in conventional conditions. Removal of the microbiome led to complete loss of EC6 enterocytes, suggesting that microbes are either critical for cell type-restricted interferon responses in the intestine, or that microbial products dictate specification of a dedicated class of interferon-positive intestinal epithelial cells.
To test developmental consequences of microbial removal on larvae, we focused on intestinal angiogenesis. In fish, the intestinal vasculature arises from angioblasts that migrate ventrally from the posterior cardinal vein, and establish a plexus that gradually resolves into the dorsal supra-intestinal artery, the ventral sub-intestinal vein, and a series of parallel vessels that connect artery and vein (Goi and Childs, 2016;Isogai et al., 2001;Lenard et al., 2015;Nicenboim et al., 2015). We noted diminished expression of key angiogenesis regulators in GF larvae, particularly VEGF-class receptors with established roles in formation of connecting vessels (Goi and Childs, 2016). Examination of GF fish showed that the microbiota is dispensable for positioning and spacing of the artery and vein. In contrast, removal of the microbiota had deleterious effects on connecting vessels, confirming a role for the microbiome in establishing the intestinal vasculature. Our results match observations from mice, where germ-free growth also diminishes villus angiogenesis (Reinhardt et al., 2012;Stappenbeck et al., 2002), suggesting a conserved requirement for microbial cues to direct intestinal angiogenesis in vertebrates. We believe the advances made in this study will allow us to trace the molecular, and cellular networks that control intestinal vasculogenesis in a developing vertebrate.

Zebrafish strains and maintenance
Zebrafish were raised and maintained using protocols approved by the Animal Care & Use Committee: Biosciences at the University of Alberta, operating under the guidelines of the Canadian Council of Animal Care. TL strain zebrafish were used for single-cell RNA sequencing and in situ hybridization experiments, and the Tg(kdrl:mCherry) line (Wang et al., 2010a) was used for analysis of intestinal vasculogenesis. Fish were raised and maintained within the University of Alberta fish facility at 28°C under a 14 hour/ 10 hour light/ dark cycle as previously described (Westerfield, 2000).

Generating germ-free zebrafish
Fish embryos were made germ-free essentially as in Melancon et al., 2017. Embryos were collected then washed and kept in embryo media (EM) supplemented with ampicillin (100 µg/mL), kanamycin (5 µg/mL), amphotericin B (250 ng/mL), and gentamicin (50 µg/mL). Embryos were washed every 2 hours with EM plus antibiotics. Once at 50% epiboly, embryos were successively washed three times in EM, then 2 minutes in 0.1% polyvinylpyrrolidone-iodine (PVP-I) in EM, followed by three EM washes, then a 20minute incubation with 0.003% sodium hypochlorite (bleach) in EM. Embryos were washed three more times then transferred into tissue culture flasks with sterile EM. At 4 days post-fertilization, EM was collected from culture flasks then plated on TSA plates and subjected to 16S rDNA PCR to confirm absence of bacteria.

Whole mount in situ hybridization
Antisense probe generation and in situ hybridization were performed essentially as in Balay et al., 2020, where fabp2 and icn were transcribed from gene-specific PCR products with an integrated T7 RNA polymerase site (TAATACGACTCACTATAGGG) that were purified by gel extraction (Qiagen 28704). Genespecific primer sequences were as follows: fabp2 -Forward: CCTGGAAAGTCGACCGCAATG; Reverse: Colorization reactions were stopped with four washes in PBST, pH 5.5. Larvae were then cleared in 30%, 50% and 70% glycerol in PBS, then eyes were removed, and larvae were mounted in 70% glycerol for imaging on a ZEISS Axio Observer. Image brightness and contrast was adjusted in FIJI.

Imaging and quantifying intestinal vasculature
Tg(kdrl:mCherry) fish (Wang et al., 2010a) were raised under conventional or germ-free conditions for 6 dpf, then euthanized with tricaine and fixed overnight at 4°C. Larvae were washed 3X in PBS then embedded in 0.7% UltraPure low melting point agarose (Invitrogen 16520) on a glass bottom dish. Tile and Z-stack images (5 µm sections) of whole fish were captured on a Leica Falcon SP8 equipped with a 25x 0.95NA Water HC Fluotar objective lens. Images were stitched with Leica Application Suite X software (Leica) and imported to FIJI to produce maximum intensity Z-projection images that were adjusted for brightness and contrast, as well as false color manipulations. To quantify intestinal vasculature, corresponding brightfield images were used to set intestinal boundaries in FIJI. Fluorescent images were then converted to binary images and the area of kdrl:mCherry signal relative to the area of the whole intestine was measured. Box plots were generated in RStudio with ggplot2.

Generating single-cell suspension for single cell RNA-seq
Zebrafish 6 days post-fertilization were euthanized in PBS plus tricaine, then intestines were immediately dissected and placed into PBS on ice until 25 intestines were collected per condition, with dissection time kept below 2 hours. Intestines were incubated in a dissociation cocktail of 1 mg/mL fresh collagenase A, according to the manufacturer's protocol then sent to Novogene for sequencing on the HiSeq system.

Processing and analysis of single cell RNA-seq data
For single cell analysis, Cell Ranger v3.0 (10X Genomics) was used to demultiplex raw base call files from Illumina sequencing and to align reads to the Zebrafish reference genome (Ensembl GRCz11.96). These matrices were analyzed using the Seurat R package version 3.1.1 (Butler et al., 2018) in RStudio. Cells possessing fewer than 200 UMIs, greater than 2500 UMIs, or greater than 50% mitochondrial reads were removed to reduce the number of low-quality cells and doublets. Seurat was then used to normalize expression values and perform cell clustering at a resolution of 1.0 with 20 principal components. The same parameters were used for independent and integrated sample analyses. After using the "FindMarkers" function in Seurat to identify marker genes for each cluster, clusters were annotated according to known epithelial or extra-intestinal markers in zebrafish, or orthologous markers in mammals. Where cluster identities were not immediately obvious, functional analysis via gene ontology enrichment scores was used to ascribe cell identity.

Guided identification of LREs
To manually cluster LREs in the conventionally reared dataset, we subset cells with log normalized expression of dab2 > 0.1 and slc15a2 > 1.0. This method isolated five LREs. Seurat requires 10 cells to make a cluster, thus LREs were not identified during initial clustering.

Lineage trajectory construction
Developmental trajectory analysis was performed in the STREAM (Single-cell Trajectories Reconstruction, Exploration And Mapping) package for bioconda, version 0.4.0 (Chen et al., 2019). Analysis was run on single epithelial cells profiles exported from Seurat post-QC. Lowly expressed genes and genes expressed in fewer than 5 cells were filtered further in STREAM, then variable genes were identified. The dimensionality of the dataset was reduced using the top 10 principal components, and trajectory inference was run using k-means clustering with 5 initial clusters. Inferred trajectories were then visualized using stream plots, with progenitor-like cells at the root. Transition genes for each branch of the trajectory ( Figure S4) were detected with "st.detect transition markers", using default parameters.

Gene ontology (GO) enrichment analysis
Marker genes from single datasets (p-value cut-off < 0.05), as well as up-and down-regulated gene lists from the integrated dataset (p-value cut-off < 0.05) were analyzed in GOrilla (Gene Ontology enRIchment anaLysis and visuaLizAtion tool) to determine GO term enrichment (Eden et al., 2009). Genes were analyzed in a two-list unranked comparison using the whole dataset gene list as background. To remove redundant GO terms, enriched terms with associated p-values from GOrilla were run through REVIGO (REduce and VIsualize Gene Ontology) using SimRel semantic similarity metric with an allowed similarity of 0.4 (Supek et al., 2011). Only top enriched terms are shown in each plot, as well as those same GO terms when found in other clusters (for Figures 2 and 4). Bar plots and bubble plots were manually generated using ggplot2 in RStudio.        Comparison of log normalized expression for known microbe-responsive genes in all intestinal cell types under CV and GF conditions. Genes whose expression increases in GF relative to CV fish intestines are shown in (A) while genes whose expression decreases in GF relative to CV fish intestines are shown in (B). Figure S10. Microbes stimulate IFN-enriched enterocytes. UMAP plots of enterocytes are shown under CV or GF conditions, with color coded annotated clusters or with relative cell-specific expression of genes involved in IFN signaling that are markers of CV EC6. ifit15 is orthologous to human IFIT2, isg15 is orthologous to human ISG15, and trim25 is orthologous to human TRIM25.