We first investigated if microbes affect protein absorption in LREs. To do this, we used established methods (Pham et al., 2008) to rear zebrafish to the larval stage in gnotobiotic conditions to compare LRE activity in GF and ex-GF conventionalized (CV) conditions. In these experiments, a portion of the GF cohort was conventionalized with microbes at 3 d post fertilization (dpf), while the rest remained in the GF condition until the experimental endpoint of 6 dpf. At that point, we gavaged GF and CV larvae with fluorescent soluble cargoes and imaged uptake in the LRE region using confocal microscopy (Figure 1A; Park et al., 2019; Shaner et al., 2004; Cocchiaro and Rawls, 2013). To test how the microbiome affects protein uptake in LREs, we gavaged larvae with the purified fluorescent protein mCherry (Shaner et al., 2004). Previous work showed that LREs readily take up mCherry following gavage with fluorescence peaking in the anterior LREs (Park et al., 2019). We found that GF and CV larvae rapidly took up mCherry in the LREs, absorbing detectable levels of mCherry within 5 min of gavage (Figure 1B-C). LREs progressively became more saturated with mCherry between 5–60 min post-gavage in both conditions (Figure 1B-C). However, mCherry fluorescence peaked in the anterior LREs of GF larvae by 40 min PG (Figure 1—figure supplement 1), but the peak did not emerge until 60 min PG in CV larvae (Figure 1—figure supplement 1). Between 5 and 60 min PG, the anterior LREs of GF larvae accumulated mCherry at a significantly faster rate than those in CV larvae (Figure 1—figure supplement 1). Across the entire LRE region, GF larvae took up significantly more mCherry than CV larvae by 1 hr PG (Figure 1D). However, LREs in CV larvae eventually reached a similar level of mCherry uptake by 5 hr PG (Figure 1E). Over the 1–5 hr time course, the anterior LREs in CV larvae gradually increased in mCherry saturation, while mCherry saturation remained at a stable, high level in GF larvae (Figure 1—figure supplement 1). We tested if the gut microbiome reduces mCherry uptake in LREs by lowering its concentration in the lumen. That did not appear to be the case because luminal mCherry concentrations were equivalent in GF and CV larvae at 1 and 5 hr PG (Figure 1—figure supplement 1), and the microbiome did not degrade mCherry over time (Figure 1—figure supplement 1). We observed a similar trend using Lucifer Yellow (LY) as a fluid phase endocytic tracer (Swanson et al., 1985; Park et al., 2019). GF larvae took up significantly more LY than CV larvae by 1 hr PG (Figure 1F) but accumulated similar amounts by 3 hr PG (Figure 1—figure supplement 1). These results suggest that microbial colonization reduces the rate of endocytosis in the LRE region.

We hypothesized that microbial burden may influence the rate of luminal uptake and that a threshold density may be needed for microbes to affect protein-uptake kinetics in LREs. To investigate this possibility, we compared mCherry uptake activity between CV larvae with different microbial densities. At 1 hr PG, mCherry uptake was not reduced in CV larvae when the density was 3×10⁵ CFU/mL in the gnotobiotic zebrafish media (GZM) (Figure 1—figure supplement 1). However, a higher density of 3×10⁶ CFU/mL in the GZM was sufficient to significantly reduce mCherry uptake in CV larvae compared to GF larvae (Figure 1—figure supplement 1).

LREs degrade much of the protein they take up from the intestinal lumen within their lysosomal vacuole (Park et al., 2019). To test if microbes affect the protein degradation process, we employed a pulse-chase assay (Park et al., 2019) to compare the rate of protein degradation between GF and CV larvae (Figure 2A). At 6 dpf, larvae were gavaged with mTurquoise (25 mg/mL), a pH-insensitive protein that degrades rapidly in LREs (Figure 2—figure supplement 1). One hour after gavage, the remaining luminal mTurquoise was flushed with PBS, and live larvae were imaged with confocal microscopy over the course of 1 hr. In both GF and CV larvae, mTurquoise degraded over the 1 hr imaging time course (Figure 2B-C), with the degradation process occurring at a significantly faster rate in GF compared to CV LREs (Figure 2D). Together, these uptake and degradation kinetics assays demonstrate that the microbiome reduces the rate of cargo uptake and lysosomal protein degradation in LREs.

Given the differences we observed in LRE kinetics between GF and CV larvae, we wanted to investigate how the gut microbiome impacts gene expression programs in LREs and other intestinal cells. To do so, we prepared GF or CV TgBAC(cldn15la-GFP) transgenic larvae (Alvers et al., 2014), in which all intestinal cells are GFP-positive, and then gavaged them with mCherry to label LREs and other cells at 6 dpf (Figure 3A). At 3 hr PG, we isolated mCherry-positive/GFP-positive and mCherry-negative/GFP-positive cells by fluorescent-activated cell sorting (FACS) as previously described (Park et al., 2019) and processed them for scRNA-seq.

Clustering analysis with Seurat (Satija et al., 2015) revealed seventeen cell clusters, including two LRE clusters (Figure 3B). Each of these clusters was characterized by transcriptomic signatures that distinguished them from other clusters (Figure 3C) (Supplementary file 1; Supplementary file 2; Supplementary file 3). Ileocytes are an ileal enterocyte population that specialize in bile salt recycling (Wen et al., 2021). Interestingly, fatty acid binding protein 6 (fabp6), a top ileocyte cluster marker, was also expressed in both LRE clusters (Figure 3C). Using transgenic reporters, previous studies showed that fabp6 expression is highest in the ileocytes, but expression was also detected in the anterior LRE region (Wen et al., 2021; Lickwar et al., 2017). We, therefore, termed the LRE cluster with the higher fabp6 expression in our scRNAseq dataset ‘anterior LREs’ and the LRE cluster with lower fabp6 expression ‘posterior LREs.’ In total, this dataset includes 131 GF and 199 CV anterior LREs, along with 359 GF and 381 CV posterior LREs (Figure 3—figure supplement 1).

GF and CV larvae had proportionally similar numbers of cells in each cluster, including LREs (Figure 3—figure supplement 1), with the exception of a cluster we called Cloaca 3, which was only present in the CV condition (Figure 3D). GO term analysis revealed that Cloaca 3 consisted of IECs that are enriched in pathways related to host defense, response to bacteria, and iron ion transport (Figure 3—figure supplement 1). Several of the genes involved in bacterial response were top cluster markers for Cloaca 3 (Figure 3—figure supplement 1). These include genes involved in sensing bacteria through lipopolysaccharide-binding, neutrophil recruitment, and inflammation (Kanther et al., 2011). Their expression was markedly higher in Cloaca 3 than in neighboring clusters (Figure 3—figure supplement 1). For example, one of the Cloaca 3 marker genes was serum amyloid a (saa), which is known to be induced by microbiota in the distal intestine and cloaca (Murdoch et al., 2019). These results suggest that the development of Cloaca 3 cells is stimulated by the microbiome and functions in microbial sensing and immune response.

Our scRNAseq data indicated that LRE numbers were similar in the presence or absence of the gut microbiome (Figure 3—figure supplement 1). To quantitatively determine if microbes affect the number of LREs, we gavaged DQ-red BSA to mark the LRE lysosomes (Marwaha and Sharma, 2017) in GF and CV larvae and computationally segmented LREs using ilastik (Sommer et al., 2011; Figure 3—figure supplement 1). This experiment revealed that there is not a significant difference in the number of active LREs between GF and CV larvae (Figure 3E). Together, these data show that the development of most intestinal cell clusters, including LREs, is not dependent on the gut microbiome. The notable exception is Cloaca 3, which only appeared in the presence of gut microbes.

Next, we explored the effects of the gut microbiome on mCherry uptake throughout the gut. Surprisingly, scRNA-seq analysis revealed that mCherry is internalized by several cell types in addition to LREs (Figure 4A–B). While LREs had the highest percentage of mCherry-positive cells, other mCherry-positive cell types included ileocytes, goblet, acinar, enteroendocrine, immune, and best4/otop2 cells in both GF and CV conditions (Figure 4C). Notably, anterior enterocytes, pharynx, and cloaca clusters contained extremely low levels of mCherry-positive cells (Figure 4C), showing that these cell types have very low protein uptake activity.

The gut microbiome increased the proportions of mCherry-positive cells in several clusters (Figure 4D). This effect could be observed in secretory cells such as enteroendocrine, goblet cells, and acinar cells, which increased by 25%, 11%, and 8%, respectively. The gut microbiome also increased the proportion of mCherry-positive neurons by 13%. Notably, the proportion of mCherry-positive anterior (84–91%) and posterior (98–97%) LREs was extremely high in both GF and CV conditions, respectively (Figure 4C). However, the gut microbiome had mixed effects on the number of mCherry-positive LREs. While anterior LREs had a slightly higher proportion of mCherry-positive cells in the CV condition (7% increase), microbes caused a small decrease in the proportion of mCherry-positive posterior LREs (1% decrease) (Figure 4C–D). Regardless of microbial colonization, the proportion of mCherry-positive LREs remained high, underscoring the robust protein-uptake program in these cells.

Since the microbiome increased the proportion of mCherry-positive cells in many clusters, we proceeded to investigate the transcriptional program they hold in common. There were many differentially expressed genes between aggregated mCherry-positive and mCherry-negative cells in the CV condition (Figure 4E). The top upregulated genes in aggregated mCherry-positive cells (ctsbb, lrp2b, ctsl.1, dab2, cpvl, cubn, tdo2b, lgmn, fabp6, sptbn5) were also significant LRE cluster markers (Figure 4—figure supplement 1). On the other hand, the most upregulated genes in aggregated mCherry-negative cells (fabp2, chia.2, fabp1b.1, apobb.1, apoa4b.1, apoa1a, tm4sf4, afp4.1, apoc2, apoc1) were also significant anterior enterocyte cluster markers (Figure 4—figure supplement 1). This presumably reflects the high relative abundance of LRE and anterior enterocytes in the mCherry-positive and mCherry-negative groups, respectively (Figure 4C). However, expression of the top differentially expressed genes also clearly delineated mCherry-positive and mCherry-negative cells in secretory cell clusters, including goblet, EEC, and acinar cells (Figure 4F, Figure 4—figure supplement 1). Differential expression of these genes between mCherry-positive and negative cells was significant in these clusters (Supplementary file 4). Upregulated genes in mCherry-negative cells were highly enriched in anterior enterocytes and mCherry-negative cloaca 3, acinar, pharynx, and best4/otop2 cells (Figure 4—figure supplement 1). These data suggest that intact protein uptake in non-LRE cells is associated with the expression of a subset of markers involved in protein uptake in LREs.

Despite similarities between LREs and mCherry-positive cells from other clusters, LREs maintained distinct transcriptional patterning that distinguished them from other cell types, including close clusters (Figure 5A). For example, KEGG pathway analysis confirmed that the lysosome pathway was strongly upregulated in anterior and posterior LREs in both GF and CV conditions (Figure 5—figure supplement 1), distinguishing them from their closest cluster neighbors (Figure 5B, Figure 5—figure supplement 1). In addition, both anterior and posterior LREs strongly expressed their characteristic endocytic machinery, composed of cubn, dab2, and amn (Figure 5—figure supplement 1).

Anterior and posterior LREs were distinguished by some notable differences in their transcriptional programs. Anterior LREs shared more transcriptional similarities to ileocytes than did posterior LREs. Bile salt transport genes, fatty acid binding protein (fabp6), and solute carrier family 10 member 2 (slc10a2), were highly expressed in ileocytes, but expression also occurred in the LRE clusters at a gradient from anterior to posterior LREs (Figure 5C–D). Further evidence of shared expression patterns between ileocytes and anterior LREs can be seen in their expression of tryptophan metabolic genes. Tryptophan metabolism was a significantly upregulated KEGG pathway in anterior LREs in the GF and CV conditions (Figure 5—figure supplement 1). Expression of tryptophan metabolic genes, including kynurenine 3-monoxygenase (kmo) and tryptophan 2,3-dioxygenase a (tdo2a), occurred in LREs and ileocytes (Figure 5D–E). Anterior LREs showed high expression of tryptophan metabolic genes (tdo2b, aldh9a1a.1, kmo, tdo2a, ddc) in the GF and CV conditions (Figure 5—figure supplement 1). Furthermore, anterior and posterior LREs were distinguished by differential expression of several peptidases (Figure 5—figure supplement 1).

Several genes were differentially expressed between GF and CV conditions in the anterior and posterior LREs (Figure 5—figure supplement 1) (Supplementary file 5). The differences between the GF and CV conditions were most apparent in posterior LREs where expression of several peptidase genes was higher in GF than in CV larvae (Figure 5F). Carboxypeptidase vitellogenic like (Cpvl), a serine carboxypeptidase, and Cathepsin La (Ctsla), a cysteine-type peptidase, are predicted to localized to lysosomes (Mahoney et al., 2001; Tingaud-Sequeira and Cerdà, 2007). Peptidase M20 domain containing 1, tandem duplicate 2 (Pm20d1.2) is an amino acid hydrolase (Long et al., 2016). These results raise the possibility that higher expression of protein-degradation genes cpvl, ctsla, and pm20d1.2 contributes to the faster degradation rates we observed in GF LREs.

In the CV condition, posterior LREs upregulated the expression of several genes involved in the immune response to bacteria (Figure 5G). These include genes for several proteins that mediate the immune response to toll-like receptor signaling from bacteria, including LPS-responsive beige-like anchor protein (lrba), myeloid differentiation factor 88 (myd88), and hsp90b1 (Karmarkar and Rock, 2013; Gibson et al., 2008; Franzenburg et al., 2012; van der Sar et al., 2006; Graustein et al., 2018; Wang et al., 2019). MyD88 signaling helps to prevent bacterial overgrowth and clears intestinal pathogens (van der Vaart et al., 2013; Karmarkar and Rock, 2013; Gibson et al., 2008; Franzenburg et al., 2012; van der Sar et al., 2006). MyD88 also regulates expression of JunD proto-oncogene, AP-1 transcription factor subunit (jund), a LPS-sensor that can induce inflammation in response to the intestinal bacteria, including Aeromonas spp. (Meixner et al., 2004; Li et al., 2022). EH-domain containing 1b (ehd1b) is upregulated in response to bacterial infection and is involved in vesicle-mediated transport (Dubytska et al., 2022). These patterns suggest that the microbiome stimulates posterior LREs to upregulate expression of genes involved in directing the immune response to intestinal bacteria.

The microbiome also stimulated expression of dopamine synthesis and signaling genes in posterior LREs (Figure 5H). These include dopamine decarboxylase (ddc), which decarboxylates tryptophan to synthesize dopamine and serotonin (Koyanagi et al., 2012). Expression of GNAS complex locus (gnas), which is involved in the dopamine receptor signaling pathway (Lu et al., 2006; Vortherms et al., 2006), was also elevated in CV posterior LREs. Interestingly, posterior LREs were the only cell cluster in which gnas and ddc expression was significantly higher in the CV than GF condition. These results suggest that the gut microbiome increases dopamine synthesis and signaling pathways in posterior LREs.

The microbiome also upregulated several genes related to iron ion transport and homeostasis in LREs (Figure 5I). In the CV condition, posterior LREs cells had significantly elevated levels of solute carrier family 11 member 2 (slc11a2), ferroportin (slc40a1), and melanotransferrin (meltf) (S5K). Anterior LRE cells also showed increased expression of slc40a1 and meltf (Figure 5—figure supplement 1). Slc11a2 helps regulate the influx of iron into the cell, while ferroportin regulates the export of iron out of the cell (Yilmaz and Li, 2018; Bao et al., 2024). Melanotransferrin has iron-binding properties and is predicted to localize to the plasma membrane, where it transports iron into the cell (Sekyere et al., 2006; Dunn et al., 2007). Interestingly, melanotransferrin was an important cluster marker in Cloaca 3, a cell cluster that only occurred in the CV condition and is directly adjacent to LREs in the intestine (Figure 3—figure supplement 1). These results suggest that upregulation of iron transport genes is a consistent response to microbiome in LREs and other epithelial cells in the distal intestine.

Together, these scRNA-seq data revealed that LREs upregulate genes in response to microbial colonization that are involved in innate immunity, iron homeostasis, and dopamine-synthesis and signaling pathways, and they downregulate several peptidases that may linked to reduced protein degradation upon microbial colonization.

We next investigated if specific bacteria alter LRE kinetics and gene expression patterns. To this end, we turned to monoassociation experiments (Pham et al., 2008), where larvae are colonized with a single strain of bacteria at 3 dpf rather than the unfractionated microbiome used to colonize CV fish.

We started by testing if commensal bacterial strains isolated from zebrafish gut microbiomes (Stephens et al., 2016; Roeselers et al., 2011) were sufficient to reduce LRE protein uptake kinetics. To address this question, we measured mCherry uptake in larvae reared either as GF or monoassociated by a single strain, including Acinetobacter calcoaceticus ZOR0008, Aeromonas caviae ZOR0002, Vibrio cholerae ZWU0020, or Pseudomonas mendocina ZWU0006 (Stephens et al., 2016). At 6 dpf, larvae were gavaged with mCherry (1.25 mg/mL), then mCherry uptake was measured at 1 hr PG (Figure 6A). Interestingly, we found that while mCherry uptake was only minimally reduced by A. calcoaceticus and moderately reduced by P. mendocina and A. caviae, colonization with V. cholerae reduced mCherry uptake severely (Figure 6B).

Next, we investigated the mechanisms by which V. cholerae reduced mCherry uptake in LREs. V. cholerae monoassociation did not significantly affect fish growth (Two-tailed t-test, p=0.66, n=18–22) (Figure 6—figure supplement 1), suggesting that interactions with LREs rather than systemic effects reduced uptake activity. To test if V. cholerae exposure induced acute effects on LRE uptake activity, we gavaged GF larvae with live V. cholerae or PBS. Thirty minutes after initial exposure, the larvae were gavaged again with mCherry (1.25 mg per mL). Interestingly, there was no difference in mCherry uptake between larvae that were initially gavaged with live V. cholerae or PBS (Figure 6—figure supplement 1), indicating that longer-term exposure to V. cholerae is required to reduce protein uptake activity in LREs. We next tested if V. cholerae monoassociation or CV conditions increased the rate of transcytosis in LREs, which could lead to reduced mCherry signal in LREs. To investigate this possibility, we gavaged GF, CV, and V. cholerae-monoassociated larvae with mCherry (1.25 mg/mL) and then measured the mCherry signal in the pronephros at 4 hr PG. There was no difference in pronephros signal between these conditions, indicating that intestinal microbes were not causing LREs to increase mCherry trans-epithelial transport through transcytosis (Figure 6—figure supplement 1). Finally, we tested if V. cholerae lowered mCherry uptake in LREs by reducing its concentration in the intestinal lumen, perhaps by promoting intestinal motility (Wiles et al., 2020; Logan et al., 2018). To do so, we analyzed confocal images of GF and V. cholerae-colonized larvae taken at 1 hr PG and measured luminal mCherry concentration in the intestinal lumen proximal to LREs. The concentration was the same in both conditions (Figure 6—figure supplement 1), showing that V. cholerae does not significantly reduce mCherry concentrations in the intestinal lumen within these timeframes.

Our monoassociation experiments suggested that microbes selectively affect LRE activity. To investigate if this effect is mediated by transcriptional regulation of specific protein uptake and degradation machinery genes, we employed quantitative hybridization chain reaction (HCR) RNA-FISH (Choi et al., 2018) to compare expression of target genes between GF and monoassociated larvae. We focused on dab2 and cubn, which encode endocytic components critical for the LRE’s ability to internalize proteins and soluble cargoes (Park et al., 2019).

Similar to what we found previously using conventional in situ hybridization probes (Park et al., 2019), the dab2 HCR probe was highly specific to LREs and pronephros (Figure 6C). By comparing the integrated signal obtained from max projection of confocal stacks, we found that GF larvae had significantly higher levels of dab2 expression than those colonized with V. cholerae (Figure 6D). Interestingly, the dab2 expression profiles in GF and V. cholerae-colonized larvae showed patterns reminiscent to those of mCherry uptake, with a peak in the anterior LREs that tapers off in the posterior LREs (Figure 6D). Since mCherry uptake is dependent on Dab2 (Park et al., 2019), this pattern may help explain why mCherry uptake tends to peak in the anterior LREs.

Next, we tested the effects of these strains on cubn expression. Similarly to dab2, the cubn HCR probe highlighted LREs and showed faint fluorescence in pronephros (Figure 6E). We found that while V. cholerae significantly reduced cubn expression, A. calcoaceticus did not (Figure 6F). These patterns are consistent with the degree to which V. cholerae and A. calcoaceticus affected mCherry uptake in LREs (Figure 6B). Interestingly, the cubn expression profile in LREs was different from dab2. It lacked an anterior peak, suggesting that cubn expression is perhaps more uniform throughout the LREs (Figure 6F).

In addition to the endocytic machinery, we also tested the effects of individual microbial strains on the expression of LRE-enriched proteases ctsh and ctsz. Our scRNA-seq data showed that these peptidases are highly upregulated in GF LREs. We found that V. cholerae reduced expression of ctsh and ctsz compared to A. calcoaceticus and GF larvae, which presented similar levels of ctsz expression (Figure 6G-H).

Together, our data reveal that microbial colonization reduces both protein uptake and degradation in LREs by modulating the expression of endocytic and lysosomal proteins. Our data also suggest that LREs have functional heterogeneity, which may represent some degree of specialization into two different clusters as suggested by the scRNAseq data.

We next investigated whether LRE activity has reciprocal effects on the gut microbiome. Our previous research demonstrated that cubn mutants have significantly reduced survival compared to heterozygotes when they are fed a low protein diet from 6 to 30 dpf (Park et al., 2019). We hypothesized that the combination of the cubn mutation with a low protein diet fostered a microbial community that further reduced the host’s protein uptake capabilities.

To explore this possibility, we tested the combined effects of the cubn mutation and custom-formulated, isocaloric high and low protein diets (see Materials and methods) on the larval zebrafish microbiome. We fed high-protein (HP) or low-protein (LP) diets to cubn heterozygote and homozygous mutant siblings from 6 to 30 dpf, then performed 16 S rRNA gene sequencing on whole larvae to identify microbial populations (Figure 7A). Initial analyses highlighted that zebrafish larvae developed bacterial communities that were distinctly different from the diets and tank water (Bray Curtis distance, p=0.0001) (Figure 7—figure supplement 1). Larval microbiomes from different tanks were not significantly different in either HP (Bray Curtis distance, p=0.08) or LP-fed (Bray Curtis distance, p=0.13) conditions (Figure 7—figure supplement 1). Diet strongly affected microbiome composition in heterozygotes (Bray Curtis distance, p=0.001), but not in cubn homozygous mutants (Bray Curtis distance, p=0.18) (Figure 7B–C). Mutants had significantly different microbiomes than heterozygotes when they were fed a LP diet (Bray Curtis distance, p=0.023), but not a HP diet (Bray Curtis distance, p=0.34) (Figure 7B–C). In contrast, beta dispersion was not significantly different between HP or LP-fed cubn heterozygotes and mutants (Bray Curtis dispersion, p=0.24) (Figure 7—figure supplement 1; Zaneveld et al., 2017). Some of these effects may reflect differences in taxonomic richness, which was significantly impacted by genotype and diet (1-way ANOVA, p=0.042) (Figure 7D). LP-fed cubn mutants had significantly lower microbial richness than HP-fed mutants (Two-tailed t-test, p=0.0094, n=6–9). These results suggest that the combination of the LP-diet and cubn mutation lowers taxa richness and affects the mutant microbiome.

The broad differences in microbiome diversity led us to investigate class and genus-level effects of the cubn mutation and LP diet on the microbiome. The effects of dietary protein on the microbiome were clearly seen at the class level (Figure 7E). Gammaproteobacteria were significantly more abundant in HP than LP-fed heterozygotes (DESeq2, padj = 0.02). LP-fed mutants were deficient in Alphaproteobacteria, which were more abundant in LP-fed heterozygotes (DESeq2, padj = 0.021) (Figure 7E). Indeed, one important trend was that mutants consistently had fewer differentially abundant taxa than heterozygotes at the class and genus levels (Figure 7F). The few taxa that were more abundant in mutants than heterozygotes emerged when they were fed the LP diet (Figure 7F).

The impacts of dietary protein and the cubn mutation on the microbiome were also apparent at the genus level (Figure 7F-G; Supplementary file 6). Pseudomonas spp. were significantly more abundant in HP than LP-fed heterozygotes (DESeq2, padj = 0.002) (Figure 7G). One genus, Hassalia, was significantly more abundant in LP-fed mutants than in heterozygotes (DESeq2, padj = 6.04E-21). Independently of diet, Aeromonas spp. were more highly abundant in cubn mutants than heterozygotes (DESeq2, padj = 0.01) (Figure 7H). Interestingly, our monoassociation experiments demonstrated that a member of Aeromonas genus, A. caviae, can reduce protein uptake activity in LREs (Figure 6B).

Together, these results suggest that LRE activity and dietary protein content can interactively affect the larval zebrafish microbiome. The combination of the LP diet and cubn mutation cause a less rich microbiome to develop. Furthermore, the cubn mutation may lead to the proliferation of certain microbes that can reduce protein uptake activity in LREs.

Animal experimentation: Zebrafish (Danio rerio) were used in accordance with Duke University Institutional Animal Care and Use Committee (IACUC) guidelines and approved under our animal protocol A072-20-03.

The Duke University Institutional Animal Care and Use Committee (IACUC) guidelines were followed in the care and use of all fish in this project. We maintained zebrafish (Danio rerio) stocks on a recirculating system at 28°C with a controlled, 14-hr light and 10 hr dark cycle (Westerfield, 2000). Breeding adult zebrafish were fed a 1:1 ratio of GEMMA Micro (Skretting) and artemia. To breed fish, males and females were placed in mating tanks with dividers overnight, and dividers were removed the following morning. Zebrafish from the Ekkwill (EK) background between 6–30 dpf were used in this study.

Previously described gnotobiotic husbandry methods were used to raise GF, CV, and monoassociated larvae (Pham et al., 2008). Briefly, embryos were treated with antibiotic zebrafish media, followed by iodine and bleach washes to eliminate microbes from their chorions at 0 dpf. Following microbe-removal steps, embryos were housed in sterile cell culture flasks containing autoclaved gnotobiotic zebrafish media (GZM) and incubated at 28 °C. Each flask contained 30 embryos in 30 mL media. At 3 dpf, gnotobiotic larvae were either conventionalized, monoassociated, or remained in the GF condition. In this process, 80% of the media was replaced. The media in GF and monoassociated flasks was replaced with 24 mL autoclaved GZM, while CV flasks received 12 mL autoclaved GZM and 12 mL 5 µm-filtered zebrafish system water. After this step, gnotobiotic zebrafish either (Protocol A) continued to be raised as described (Pham et al., 2008), or (Protocol B) they were raised by our modified protocol designed to boost the bacterial load. In Protocol A, each flask was given 100 µL ZM000 daily from 3 to 5 dpf following media changes, and 80% of the media was replaced with autoclaved GZM from 4 to 5 dpf. In Protocol B, flasks are given 150 µL ZM000 from 3 to 5 dpf, and 40% of the media was replaced with autoclaved GZM from 4 to 5 dpf. The scRNA-seq (Figure 3, Figure 4 and Figure 5), lucifer yellow gavage (Figure 1G), and long-term mCherry gavage (Figure 1D.E) larvae were raised according to Protocol A. The larvae used in rapid mCherry-uptake (Figure 1B,C), mTurquoise degradation (Figure 2), and monoassociation (Figure 6) experiments were raised according to Protocol B. At the 6 dpf endpoint, GF flasks were tested for sterility by spot-testing on TSA plates, as well as brain-heart, dextrose and nutrient broth. The bacterial density in CV and monoassociated media was tested by serial dilutions on TSA plates. Zebrafish larvae were monoassociated with Acinetobacter calcoaceticus ZOR0008, Aeromonas caviae ZOR0002, Vibrio cholerae ZWU0020, or Pseudomonas mendocina ZWU0006 that were previously isolated from conventionally-reared zebrafish (Stephens et al., 2016). Bacteria stocks used in monoassociation were kept in 50% glycerol at –80°C for long-term storage and on tryptic soy agar (TSA) plates at 4 °C for short-term storage. Bacteria were cultured by incubating picked colonies in LB liquid media on a shaker table at 30 °C for 24–72 hr to reach turbidity with an OD600 of 1–3. Cultures were re-suspended in 1 X PBS after spinning at 5000 RPM for 2 min. The bacterial density (CFU/mL) was determined by plating serial dilutions on TSA plates and incubating overnight at 37 °C. Re-suspended bacteria was added to GF flasks to colonize larvae at 3 dpf. For the mCherry gavage experiment (Figure 6B), GF larvae were colonized with approximately 6×10⁸ CFU of V. cholerae, A. caviae, P. mendocina, or A. calcoaceticus. The final bacterial density in the media was measured by plating serial dilutions on TSA plates.

Previously described methods were used to prepare mCherry and mTurquoise (Park et al., 2019).

We tested if the larval zebrafish microbiome can degrade mCherry. CV larvae (n = 27) were anesthetized with 0.22 µm-filtered Tricaine, then homogenized in GZM with a Tissue-Tearor (BioSpec Products, inc, model # 985370). The homogenate was spun at 5000 RPM for 2 min and re-suspended in 1X PBS. mCherry was added to the zebrafish microbiome mixture and the 1X PBS control (25 µg/mL). The microbiome and control were added to a 96-well plate. The mCherry fluorescence was measured in each well periodically over 2 hr. Average mCherry fluorescence (AU) over time was compared between treatments with a simple linear regression. This experiment was performed with 9-10 biological replicates for GF and CV conditions.

GF and CV larvae were anesthetized with 1X Tricaine at 6 dpf. Then, anesthetized larvae were gavaged with mCherry protein (1.25 mg/mL). Larvae were placed in a 28°C incubator to absorb the mCherry for 4 hr. At that point, larvae were fixed in 4% PFA and stored overnight at 4°C. Fixed samples were washed in 1X PBS three times before being mounted in 0.9% low-melt agarose in egg water. Imaging was done with a 25X objective and resonant scanner, and images were taken as z-stacks. Statistics were done by one-way ANOVA with Tukey’s multiple comparisons test. This experiment was replicated in GF and CV larvae two times with one representative experiment shown (Figure 6—figure supplement 1). Results were replicated when comparing trans-epithelial transport between GF, CV, and V. cholerae-colonized larvae (data not shown).

GF and CV larvae were anesthetized with 1X Tricaine at 6 dpf. Then, DQ-red BSA (50 µg/mL) was gavaged into larvae until the whole gut was filled with the gavage mixture (4 nL). After 0.5 hr of uptake, the larvae were flushed with a 1X PBS gavage until no visible red color was observed in the gut. Larvae were stored in a 28°C incubator for 3 hr after DQ-red BSA gavage. During this incubation period, DQ-red BSA fluorescence activated when the dye was catabolized by lysosomal proteases, thus marking the lysosomal vacuole of each LRE. Larvae were then anesthetized with 1X Tricaine, mounted in 1.3% low-melt agarose in egg water, and live imaged by confocal microscopy. Imaging was done on a Leica SP8 microscope using a 25X objective (Leica) and resonant scanner.

Images were imported into Ilastik version 1.3.3 and segmented by the Pixel Classification program. The program is trained to identify DQ-red BSA-filled LRE vacuoles by machine learning where 10–15 LRE vacuoles are manually marked per image. This process is carried out for 5–10 images before the program is allowed to train itself based on these manual feed-ins. Any errors made by the program are then manually corrected and used to improve the segmentation process. After the segmentation program is trained to have at least an estimated 85% accuracy, the results are exported as probability maps in.h5 format, which are then fed into Ilastik’s Object Classification program. The following parameters are used for the classification process: Method: Hysteresis; Smooth: σ=1.0 for x-,y- and z-axis; Threshold: core = 0.60, final=0.60; Don’t merge objects. After checking the accuracy of object classification, the results are then exported as.cvs files from which the numbers of objects (in this case LRE vacuoles) are extracted. Statistics were done by 1-way ANOVA with Tukey’s multiple comparisons test. Quantification was done with 6–8 biological replicates from the GF and CV conditions in this experiment.

TgBAC(cldn15la-GFP)^pd1034 larvae (Alvers et al., 2014) were raised in GF and CV conditions to 6 dpf as described above. These transgenic larvae express GFP in IECs. GF flasks were screened for sterility as described above. Bacterial density in CV flasks ranged from approximately 5×10⁴ – 5×10⁹ CFU/mL. In total, 276 CV and 277 GF, GFP-positive larvae were gavaged with mCherry (1.25 mg/mL) and incubated for 4 hr to mark LREs with mCherry in addition to GFP. Then, larvae were dissociated, and cell suspensions were sorted by FACS as described (Park et al., 2019). The cell death marker, 7AAD (5 µg/mL), was added to cell suspension prior to FACS to remove dead cells. Cells were sorted on a MoFlo Astrios EQ cell sorter (Beckman Coulter) by the Flow Cytometry Shared Resource Center (Duke University). In total, 61,000 double-positive and 156,000 GFP-positive cells were collected from GF larval samples. From CV larval samples, 51,000 double-positive and 140,000 GFP-positive cells were collected. Cells were collected in RLT Plus Buffer (Qiagen RNeasy Plus Micro Kit: QIAGEN Cat No. 74034) media.

To test mCherry-uptake activity in this cohort, GF and CV larvae were gavaged with mCherry (1.25 mg/mL) at the same time as larvae destined for scRNA-seq. These larvae were incubated for 5 hr, then preserved in 4% PFA. They were confocal imaged the following day as described above.

Immediately after cells were sorted, we began library preparation with the Chromium Next GEM Single Cell 3’ GEM, Library & Gel Bead Kit v3.1 following the kit protocols. GEMS were stored at –20°C for 3 d prior to post-GEM-RT cleanup and cDNA amplification following the kit protocol. Quality control was done by measuring cDNA concentration and size distribution with the ScreenTape assay using the Agilent TapeStation. Libraries were submitted to the Duke Center for Genomic and Computational Biology for sequencing. They were sequenced on the Illumina NovaSeq 6000 platform with 100 bp paired-end reads. Sequencing results are deposited at NCBI Sequence Read Archive (BioProject Accession #: PRJNA1192682).

Cell Ranger v3.0 (10X Genomics) was used to demultiplex sequencing files and align reads to the zebrafish reference genome, Danio rerio GRCz11. Analysis was performed in RStudio (version 2023.12.1) using the Seurat package (version 5.0.1) (Hao et al., 2024). Reads were filtered by the following criteria: UMIs/cell >500, genes/cell >250, Log10GenesperUMI >0.8, and mitoRatio <0.5. The SCTransform method was used for normalization. Cells were clustered with a resolution of 0.5. Cluster markers were identified using Seurat’s ‘FindMarkers’ function. Cluster identity was determined by cross-referencing marker genes to other scRNA-seq and RNA-sequencing datasets (Park et al., 2019; Wen et al., 2021; Willms et al., 2022). After the LRE cluster was identified, it was re-clustered with a resolution of 0.5 to form two sub-clusters. The GF and CV objects were merged to form one Seurat object.

HCR probes, hairpins, amplification, wash, and hybridization buffers were purchased from Molecular Instruments (Choi et al., 2018). Our methods were adapted from a previously published procedure for performing HCR on zebrafish embryos (Munjal et al., 2021). At 6dpf, larvae were anesthetized and fixed in 4% PFA, then incubated on a shaker table for 2 hr. Larvae were washed with 1X PBS twice, followed by two cold acetone washes. Larvae were incubated in cold acetone at –20°C for 8 min, followed by three 1X PBS washes. For the detection stage, larvae were first incubated in probe hybridization buffer at 37°C for 30 min on a shaker. Larvae were incubated in probe solution (4 nM) on a shaker at 37 °C for 24 – 48 hr. Excess probes were removed by washing larvae four times with preheated, 37°C probe wash buffer, incubating samples on a shaker table at room temperature for 15 min each time. This step was followed by two, 5–10 min SSCT washes on a shaker at room temperature. For the amplification stage, larvae were incubated in room temperature amplification buffer for 30 min. Hairpins (30 pmol) were prepared by heating at 95°C for 90 s, then snap-cooled in the dark at room temperature 30 min. Hairpins were added to amplification buffer. Larvae were incubated in the hairpin solution in the dark at room temperature overnight. Excess hairpins were removed with five SSCT washes on a shaker table at room temperature. The duration of the first two washes were 5 min, followed by two 30-min washes and one 1-min wash. Samples were protected from light during the washes. Larvae were imaged by confocal microscopy using a Leica SP8 microscope equipped with 10X and 25 X objectives. The difference in HCR probe fluorescence profiles between GF and monoassociated larvae was calculated with a two-way ANOVA. To do so, the average probe fluorescence in each segment (30 µm) of the LRE region (300 µm) was compared between the GF and monoassociated conditions. HCR probe sequences are listed in Supplementary file 7. HCR probes were tested and imaged in at least three replicate experiments (data not shown). HCR probe fluorescence was measured in GF and monoassociated larvae in two replicate experiments and multiple larvae as indicated in the figure legend.

Sibling larvae from a cubn heterozygous-mutant cross were conventionally raised from 0–5 dpf. At 6 dpf larvae were housed in 3 L tanks at a density of 10 larvae per tank and raised on our standard circulating aquarium. From 6 – 30 dpf, larvae were fed 10 mg/d of a custom-formulated high (HP) or low-protein (LP) diet daily between 11 am–12 pm. There were three tanks per diet. Diet formulations are described in detail below.

At 30 dpf, samples were collected for 16S rRNA gene sequencing. Larvae were anesthetized with 0.2 µm-filtered tricaine in autoclaved egg water. After the water was removed, larvae were flash frozen in liquid nitrogen. Tank water samples were collected by passing 50 mL through a 0.2 µm filter (Pall Corporation MicroFunnel Filter Funnels #4803), then flash freezing the filter paper in liquid nitrogen.

Samples were put on dry ice during transport to the Duke Microbiome Shared Resource for 16S rRNA gene sequencing. Larval DNA was extracted with the MagAttract PowerSoil DNA EP Kit (Qiagen, 27100-4-EP). Following the Earth Microbiome Project protocol (http://www.earthmicrobiome.org/), the V4 region was amplified by polymerase chain reaction with the 515F and 806R primers, which are barcoded for multiplexed sequencing. PCR product concentration was measured with the Qubit dsDNA HS assay kit (ThermoFisher, Q32854) on a Promega GloMax plate reader. Equimolar PCR products from all samples were pooled and sequenced with 250 bp PE reads on the MiSeq Illumina platform. Sequencing results are deposited at NCBI Sequence Read Archive (BioProject Accession #: PRJNA1188138).

Paired-end fastq files were demultiplexed in R (version 4.1.1). The dada2 package was used to denoise the sequences, filter, and trim reads, dereplicate reads, merge paired end reads, generate an amplicon sequence variant table, remove chimeras, and generate a Phyloseq object (Callahan et al., 2016a; Callahan et al., 2016b). The Silva database (version 138.1) was used to assign taxonomy. Mitochondria and chloroplasts were filtered out of the dataset using the Phyloseq package (version 1.46.0) (McMurdie and Holmes, 2013). Phyloseq was used for downstream analysis of relative abundance and taxa richness. The DESeq2 package (version 1.42.0) was used to measure differential abundance (Love et al., 2014). The ggplot2 package (version 3.5.0) was used to generate heatmap and relative abundance plots. The vegan package (version 2.6-4) was used to measure Bray Curtis distance.

The following primers were used to genotype the larvae: cubn F: 5’-ACTCTGTTCACCTGCAGTGC-3’, cubn R: 5’-TGACATCCGAGTGGAGTTCCTGCCAAGAC-3’.

These diets were custom-formulated at University of Alabama at Birmingham.