ISX-9 manipulates endocrine progenitor fate revealing conserved intestinal lineages in mouse and human

Enteroendocrine cells (EECs) survey the gut luminal environment and co-ordinate hormonal, immune and neuronal responses to it. They exhibit well characterised physiological roles ranging from the control of local gut function to whole body metabolism, but little is known regarding the regulatory networks controlling their differentiation, especially in human gut. The small molecule Isoxazole-9 (ISX-9) stimulates neuronal and pancreatic beta-cell differentiation, both closely related to EEC differentiation. We used ISX-9 as a tool to explore EEC specification in mouse and human intestinal organoids. ISX-9 increased the number of neurogenin3 (Ngn3) positive endocrine progenitor cells and upregulated NeuroD1 and Pax4, transcription factors which play roles in mouse EEC specification. Single cell analysis revealed induction of Pax4 expression in a developmentally late Ngn3+ population of cells. This coincided with enrichment of organoids with functional serotonin-producing enterochromaffin cells and was partly dependent on stimulation of calcium signalling in a population of cells residing outside the crypt base. Inducible Pax4 overexpression, in ileal organoids, uncovered its importance as a component of early human endocrine specification and highlighted the potential existence of two major endocrine lineages, the early appearing enterochromaffin lineage and the later developing peptidergic lineage which contains classical gut hormone cell types. Our data provide proof-of-concept for the controlled manipulation of specific endocrine lineages with small molecules, whilst also shedding new light on human EEC differentiation and its similarity to mouse. Given their diverse roles, understanding endocrine lineage plasticity and its control could have multiple therapeutic implications.


INTRODUCTION
The intestinal epithelium is a key interface with our external environment. It renews itself every 4e5 days, and is composed of five terminally differentiated cell types: the absorptive enterocytes and the secretory Paneth, goblet, tuft and enteroendocrine cells (EECs) [1]. These cells are constitutively generated by cycling Lgr5 þ crypt stem cells, and together they orchestrate the epithelium's major functions, nutrient absorption and defence. Despite representing only 1% of the epithelium, the EECs constitute the largest hormone-producing tissue and have been described as the gut's sentinels. They sample the luminal, circulating and local tissue environments and coordinate an appropriate response from the epithelium and the immune and nervous systems [2]. For example, they play a key role in controlling the response to a meal, fine tuning physiology to ensure optimal fuel absorption, use and storage [3]. Gut hormones exhibit actions ranging from the local control of gut motility, absorption and secretion, to the regulation of whole-body metabolism [4]. Whilst there is a large body of evidence describing the functional roles of gut hormones, comparatively little is known about the factors that control EEC differentiation and assign subset identity. EECs were originally classified by immunostaining according to their dominant hormone product. Glucagon-like peptide 1 (GLP-1), glucagonlike peptide 2 (GLP-2) and peptide YY (PYY) are secreted by L cells; glucose-dependent insulinotrophic polypeptide (GIP) is secreted by K cells; somatostatin (SST) is secreted by D cells; cholecystokinin (CCK) is secreted by I cells; secretin (SCT) is secreted by S cells; gastrin (GAST) is secreted by G cells; serotonin is secreted by enterochromaffin (EC) cells; and neurotensin (NTS) is secreted by N cells. The recent use of fluorescent reporter mice and transcriptomics has showed that EEC subsets may be less well defined [5e8]. The hormonal repertoire of an EEC is a function of its differentiation trajectory, its location within the gut and its height in the cryptevillus axis, which dictates differential exposure to Wnt and bone morphogenic protein (BMP) signalling gradients [9,10]. Mouse lineage tracing studies have identified a handful of transcription factors (TFs) regulating EEC differentiation. Cells exiting the stem cell compartment are fated to be secretory cells by Notch inhibition, followed by Atoh1 expression [11e13]. Atoh þ cells are then designated to the endocrine lineage by expression of the bHLH TF neurogenin3 (Ngn3) [14,15]. In mice, TFs downstream of Ngn3 known to be necessary for subset specification include Insm1 (substance P and NTS) [16], neurogenic differentiation 1 (NeuroD1) (SCT and CCK) [17e19], Nkx2.2 (CCK, GAST, GIP and SST) [20], Pax4 (5-HT, SCT, GIP, PYY and CCK) [21] Pax6, Foxa1 and Foxa2 (preproglucagon and its products GLP-1 and 2) [22], Arx (GLP-1, GIP, CCK, SCT, GAST and GHRL) [21], and Lmx1A (5-HT) [23]. Nevertheless, the regulatory networks controlling EEC identity have remained largely unknown, until a recent sophisticated study described a time-resolved transcriptional road map of mouse EEC fate trajectories [24]. It now appears that classical TFs are more promiscuous than lineage tracing implied. Furthermore, there is a paucity of knowledge regarding EEC specification in human intestinal epithelium due to lack of tractable model systems, although several of the classical TFs are upregulated in response to a NGN3 pulse in intestinal organoids derived from human pluripotent stem cells [25,26]. Understanding the factors that control gut endocrine pedigree has implications for several clinical conditions including diabetes, obesity, gut inflammatory disorders and perhaps cognitive disorders including depression and anxiety. Deciphering how to manipulate EECs may open novel treatment avenues and offer a clearer understanding of epithelial homeostasis. To identify a candidate molecule that might influence EEC fate, we drew parallels from other endocrine tissues. Gut endocrine specification is strikingly like that in the pancreas, and both bear close resemblance to neuronal differentiation. The small molecule isoxazole-9 (ISX-9) [N-cyclopropyl-5-(thiophen-2-yl)isoxazole-3-carboxamide] was uncovered in a chemical screen for drivers of neuronal differentiation [27]. It activates NeuroD1 and has also been used to investigate pancreatic beta-cell differentiation [28,29]. We explored the effects of ISX-9 on EEC identity in organoids derived from mouse and human tissue resident stem cells. Our data demonstrate proof-of-concept that specific EEC populations can be manipulated with a small molecule, highlight the similarities between mouse and human EEC differentiation and provide a tool to study human EC cells in vitro.

Crypt isolation and mouse intestinal organoid culture
Mouse small intestines were harvested and cleaned with cold phosphate-buffered saline (PBS) and separated into two parts: duodenum (proximal 5 cm), and jejunum and ileum. For our experiments, organoids were generated only from the jejenum/ileum part. This part was cut longitudinally, and villi were scraped with a glass slide. The tissue was cut with scissors into 2x2-mm pieces and repeatedly washed. Subsequently, the tissue pieces were incubated with 2 mM ethylenediamine tetraacetic acid (EDTA; Invitrogen) in PBS for 45 min in a rotator at 4 C. After removal of EDTA, vigorous shaking in cold PBS lead to the release of crypts. The crypts were further washed in PBS, passed through a 40-mm cell strainer, pelleted and resuspended in basal medium Eagle (BME; Amsbio). The crypts were plated in 48-well plates, with 200 crypts per 25 mL of BME. The BME was polymerised for 15 min at 37 C, and stem cell growth medium (WENR) supplemented with 10 mM Y-27632 (SigmaeAldrich) was overlaid. WENR medium consists of advanced Dulbecco modified Eagle medium (DMEM)/F12, 2 mM GlutaMAX, 10 mM HEPES, 100 units/mL penicillin/streptomycin, 50 ng/mL EGF, 1Â B27, 1Â N2 supplements (all from Gibco), 1.25 mM N-acetylcysteine (SigmaeAldrich), 100 ng/ mL Noggin (Peprotech), 50% Wnt3A conditioned medium and 10% Rspondin-1 conditioned medium (both in house production). Three days later, the medium was changed into differentiation media (ENR) with no Wnt3A or Y-27632. Organoids were passaged once a week by mechanical dissociation, at a 1:3 split ratio. Plated organoids were maintained in a CO 2 incubator with 5% CO 2 , and the media were changed every other day. The treatment of mouse small intestine organoids with ISX-9 (Tocris Bioscience) started 3 days after passaging. For doseeresponse experiments, organoids were treated with 2 mM, 20 mM, 40 mM and 80 mM ISX-9 for 48 h. For time-course experiments, organoids were treated with 40 mM ISX-9 for 24, 48 and 96 h, and samples were collected for RNA extraction at the end of each timepoint. Treatment with 10 mM KN93 or 10 mM KN92 (an inactive analogue of KN93) (Tocris Bioscience) in the presence or absence of ISX-9 was also performed for 48 h. For the remaining experiments, organoids were treated with 40 mM ISX-9 for 48 h, and then ISX-9 was removed for another 48 h (48-h ISX-9 pulse).

Generation and culture of human terminal ileal organoids
Human terminal ileum crypts were isolated from biopsies acquired from patients undergoing colonoscopy at Guy's and St. Thomas's NHS Foundation Trust with their informed consent. Biopsies were washed in cold PBS until the supernatant was clear. Following 10 min of incubation at room temperature with 10 mM 1,4-dithiothreitol (DTT), the biopsies were incubated with 8 mM EDTA in PBS and placed in a rotator for 1 h at 4 C. At the end of the incubation, the EDTA was removed and crypts were released with vigorous shaking in cold PBS. The crypts were further washed in PBS, pelleted and resuspended in Matrigel (Corning), in the same density as mouse crypts. Human intestinal crypts embedded in Matrigel were overlaid with stem cell growth medium (WENRAS) supplemented with 10 mM Y-27632 and 5 mM CHIR99021 (SigmaeAldrich). The human stem cell growth medium, in addition to the components of the previously described mouse medium, also contained 10 nM gastrin (SigmaeAldrich), 500 nM A83-01 (Bio-techne), 10 mM SB202190 (SigmaeAldrich) and 10 mM nicotinamide (SigmaeAldrich). Three days after isolation or splitting, Y-27632 and CHIR99221 were removed from the medium, and organoids were either maintained in WENRAS or transferred into differentiation medium for setting up experiments. For the differentiation of human ileal organoids, Wnt3A conditioned medium was reduced from 50% to 15%, and SB202190 and nicotinamide were withdrawn from the medium. Differentiation medium was used for 7 days, keeping the organoids in culture for 10 days. Five days after passaging, human terminal ileal organoids were treated with 40 mM ISX-9 for 48 h, and then ISX-9 was removed for another 48 h (48-h ISX-9 pulse), except if it is stated differently. MEK signalling was inhibited with 500 nM PD0325901 (SigmaeAldrich), and notch signalling was inhibited with 10 mM diaminophenylthiazole (DAPT; SigmaeAldrich). A combination of both inhibitors with ISX-9 was given in a 48-hour pulse. All control organoids were treated with vehicle.
2.4. RNA extraction and real-time quantitative PCR Total RNA was isolated from organoids [released from Matrigel or BME with Cell Recovery solution (Corning)] using the RNeasy Mini Kit (Qiagen) according to the manufacturer's instructions. On-column DNase digestion was performed for removing any residual genomic DNA (Qiagen). RNA extraction from sorted cells was performed using TRI Reagent LS (SigmaeAldrich) according to the manufacturer's instruction. cDNA was generated using the High-Capacity cDNA reverse Transcription Kit (Applied Biosystems). Real-time quantitative polymerase chain reaction (RT-qPCR) was performed using QuantiTect primers and QuantiFast SybrGreen PCR kit (both from Qiagen) on a LightCycler 480 or LightCycler 96 (Roche). The relative gene expression levels were calculated by averaging the Ct values of technical duplicates for each biological sample and normalising them to the expression of the housekeeping gene beta-2-microglobulin.

Evaluation of mouse small intestinal organoid growth
Growth and convolutedness of organoids were evaluated by collecting brightfield images of control and ISX-9-treated organoids with the EOS 600D Nikon camera on an TMS-inverted microscope (Nikon) on days 3, 5 and 7 in culture. The surface area, perimeter and number of buds were measured using ImageJ software.
2.6. Immunofluorescent staining of organoids Mouse or human small intestine organoids were fixed for 45 min in 4% formalin (SigmaeAldrich) and then washed with 2% bovine serum albumin (BSA) in PBS (SigmaeAldrich). Subsequently, organoids were blocked with blocking buffer consisting of 2% BSA and 5% donkey serum (SigmaeAldrich), and permeabilised with 0.5% Triton-X (SigmaeAldrich) for 1 h at room temperature. Primary antibody incubation was done in a rotor overnight at 4 C with primary antibodies diluted in blocking buffer. The primary antibodies used were rabbit polyclonal anti-Chromogranin A (1:800; Abcam) and goat polyclonal anti-serotonin (1:100; Immunostar). The next day, the organoids were washed and incubated with secondary antibodies, Alexa Fluor 488 Donkey Anti-Rabbit or Alexa Fluor 594-Donkey anti-goat (1:500; Jackson ImmunoResearch) for 1 h at room temperature. Nuclear counterstaining was performed in parallel with the secondary antibody incubation using Hoechst 33342 (1:2000; Invitrogen). After washing, organoids were mounted with Vectashield Vibrance Antifade Mounting Medium (Vector Laboratories). The fraction of proliferating cells in control and ISX-9-treated mouse intestinal organoids was determined using the Click-iT EdU Cell proliferation kit, Alexa FluorÔ 647 dye (Invitrogen). Organoids were preincubated for 1 h with 10 mM EdU, then fixed and permeabilised, and EdU-positive nuclei were labelled according to the manufacturer's instructions.

Immunofluorescent imaging
Live-cell fluorescence imaging was conducted in control and ISX-9treated Ngn3-RFP, Ngn3-Cre-RFP, and CCK-Cre-Rosa-eYFP organoids. A continuous z dimension stack of RFP or eYFP fluorescence and brightfield images was obtained, while organoids were still embedded on BME, using an A1 inverted confocal microscope (Nikon). Images of whole-mount organoids stained for chromogranin A and serotonin were also captured with an A1 inverted confocal microscope (Nikon). Image analysis was performed using either Nikon Elements or Image J software. Time-lapse fluorescent microscopy of doxycycline-induced human intestinal organoids for overexpression of Pax4 was performed with a BioStation IM-Q (Nikon).
2.8. Flow cytometry analysis and fluorescence-activated cell sorting (FACS) Control and ISX-9-treated Ngn3-RFP, Ngn3-Cre-RFP and CCK-Cre-Rosa-eYFP organoids were dissociated into single cells with mechanical disruption after a 5-minute incubation with TrypLE Express (Gibco) at 37 C. After washing with PBS, the cells were passed through a 40-mm cell strainer and resuspended in Advanced DMEM/ F12 medium with 4 mg/mL DNase (SigmaeAldrich), 10 mM Y-27632 and 2 mM EDTA. Next, 1 mg/mL 4,6-diamino-2-phenylindole (DAPI; Invitrogen) was added to the cell suspension to label dead cells. Viable cells were analysed in a BD FACS CantoÔ II (Beckton Dickinson). For RNA extraction of Ngn3 þ cells from control and ISX-9-treated Ngn3-Cre-RFP organoids, organoids were first dissociated into single cells as previously described and immediately sorted using a BD FACS AriaÔ II (Beckton Dickinson).
2.9. Sample preparation for scRNA-seq Ngn3-Cre-RFP organoids were treated with ISX-9 for 48 h, collected and processed for fluorescence-activated cell sorting as just described. DAPI was added immediately before sorting. RFP-positive, DAPI-negative cells were sorted into 384-well plates containing 384 unique molecular identifier (UMI) barcode primer-sets using a FACS AriaÔ II (Beckton Dickinson). Samples in plates were centrifuged and stored at À80 C. The samples were then processed by Single Cell Discoveries B.V. according the SORT-seq method [33]. Briefly, first and second strand synthesis (Invitrogen) was performed and all wells of a single plate were pooled. After in vitro transcription (Ambion), the amplified RNA was reverse transcribed and amplified for 10 to 12 cycles with Illumina Truseq primers. Finally, libraries were analysed on an Illumina NextSeq500 using 75-bp pair-end sequencing.
2.10. Analysis single-cell mRNA sequencing UMI count matrices were imported into R Studio and processed with the R package Seurat (version 3.1) [34]. For quality control (QC), we quantified the proportions of UMIs mapped to the mitochondrial genome. All the cells with mitochondrial reads greater than 10% were excluded. We further filtered cells with greater than 7500 unique features, and 206 control and 216 ISX cells were passed to analysis. ERCC92 spike-ins as well as genes associated with clustering artefacts (Rn45s, Malat1, Kcnq1ot1, A630089N07Rik) were also excluded from the final dataset. Control and treated datasets were merged after QC filtering and then split by treatment for normalisation using the SCTransform wrapper and percent mitochondrial variations regressed. We calculated a subset of 3000 features to integrate using the SelectIntegrationFeatures and ensured all Pearson residuals were calculated using PrepSCTIntegration. The datasets were then integrated using the FindIntegrationAnchors and the IntegrateData functions. We then ran an integrated analysis on all cells in the experiment using the standard workflow with the following default parameters: linear dimensional reduction using RunPCA, non-linear dimensional reduction using RunUMAP. A K-nearest neighbour's graph was constructed with the FindNeighbors function using the first 25 principle components before clustering using the FindClusters function with a resolution of 1.8. Differential gene analysis for cluster definition was run using the FindConservedMarkers, which reports the top markers conserved between the control and treated groups. The single-cell FASTQ and read count files are available in the GEO database (accession no. GSE143221).
2.11. Serotonin release from mouse and human small intestine organoids Serotonin secretion in control and ISX-9-treated mouse organoids was measured after 7 days in culture and in human organoids after 10 days in culture. A 48-hour ISX-9 pulse was given to both mouse and human organoids as just described. At the end of the culture period, organoids were released from BME and Matrigel with Cell Recovery solution with 0.5% BSA for 2 h. Serotonin secretion was stimulated with 10 mM IBMX (SigmaeAldrich) and 10 mM Forskolin (SigmaeAldrich) in the presence of 1 mM fluoxetine (SigmaeAldrich) for blocking potential serotonin reuptake via serotonin transporter (SERT) for 1 h. Supernatant and organoid lysates were collected for measuring secretion and content, respectively. Organoid lysates were prepared in 1Â PBS with protease inhibitors (Thermo Scientific). Lysates were frozen at À20 C, thawed and sonicated for 30 s (amplitude 10e14) on ice, and centrifuged for 5 min at 5000 g at 4 C. Serotonin concentration was measured using an enzyme-linked immunosorbent assay (ELISA) kit according to the manufacturer's instructions (Aviva System Biology).

Calcium imaging
Petri dishes (30 mm) were coated with BME diluted in Advanced DMEM/F12 (1:100) and left to set at 37 C for 1 h. Mouse small intestine organoids were collected and washed three times with cold PBS to dissolve BME and re-seeded in BME-coated Petri dishes with ENR medium. After 2 h, organoids were loaded with 7 mM Fura 2-AM (SigmaeAldrich) in HEPES saline buffer containing 0.01% pluronic F127 (Invitrogen) and 2 mM probenecid (Tocris Bioscience), and incubated for 30 min at 37 C. Organoids were washed three times with HEPES buffer, and images were recorded with an Axiovert 135 Ca 2þ imaging system (Zeiss), at 20Â magnification every 80 ms. Cells were excited at 340 nm and 380 nm, and emitted light was acquired at 510 nm. Calcium concentration was calculated by 340/380 fluorescence ratio using MetaFluor software. Adenosine triphosphate (ATP; 100 mM) was used as positive control. Imaging experiments were performed at least three times, and the representative time course is presented in the manuscript.

Genetic engineering of human terminal ileal organoids
For overexpression of Pax4 in human terminal ileal organoids, mouse Pax4 cDNA was cloned into the pPB-tetO-MCS-IRES-mCherry vector. Briefly, pPB-tetO-MCS-IRES-mCherry vector was first linearised following sequential digestion with XhoI and SpeI restriction enzymes (Promega), and the gel was purified using GenEluteÔ Gel Extraction Kit (Sigma). PCR primers for mouse Pax4 were designed with 15 bp extensions that are complementary to the ends of the pPB-tetO-MCS-IRES-mCherry linearised vector (Pax4 F: 5' -CAA AGA ATT CCT CGA ATG CAG CAG GAC GGA CT -3 0 , Pax4 R: 5' -AGG CCA TGG CAC TAG TTA TGG CCA GTT TGA GCA ATG GGT TGA -3 0 ). The 15 bp extension was required for directional cloning using the In-Fusion HD cloning kit (Clonetech). From this point on, the manufacturer's instructions were followed. Pax4 was amplified using as template cDNA generated from mouse intestinal organoids. The generated plasmid was named pPB-tetO-Pax4-IRES-mCherry. The simultaneous co-transfection of three different plasmids is required for the generation of inducible overexpressing Pax4 human terminal ileal organoids. These plasmids are pPB-tetO-Pax4-IRES-mCherry, pPB-CAG-rtTA-IRES-Hygro and pCAG-PBase plasmid. For electroporation of the three plasmids into human terminal ileal cells, we followed the protocol from Fujii et al. [35]. DNA was transfected at 7.2 mg for the two piggyBac vectors, and at 5 mg for the transposase vector. Five days post-electroporation, cells were selected with 100 mg/mL hygromycin (Cambridge Bioscience). Gene expression was induced using 1 mg/mL of Doxycycline (Cambridge Bioscience).
2.14. Statistics For cell counting experiments, "n" represents the number of individual organoids assessed. For RNA experiments, "n" represents the number of biological replicates. All data are presented as mean AE standard error of the mean (SEM), except violin plots, in which data are presented as median and quartiles. Each "n" is presented as a dot in graphs. Relevant tests are described in figure legends. All statistical analyses were performed using Graph Pad Prism Version 8.1.2 (GraphPad Software) for Windows or Mac except for RNA-seq, for which statistics were calculated using Seurat.

RESULTS
3.1. ISX-9 increases the expression of transcription factors associated with EEC differentiation ISX-9 increases the expression of NeuroD1 and induces differentiation of neuronal [27], cardiac [36] and islet endocrine progenitors [28]. We wondered if these properties could be harnessed to manipulate gut endocrine differentiation. Mouse small intestinal organoids exposed to increasing doses of ISX-9 (48-h treatment) had increased expression of transcription factors known to be important for endocrine specification ( Figure 1AeF). As expected, ISX-9 increased NeuroD1, mimicking its role in other tissues ( Figure 1A). Interestingly, Ngn3, a transcription factor usually associated with the earliest identifiable endocrine progenitor cell [37] and thought of being upstream of NeuroD1, was also increased ( Figure 1B). Other downstream TFs were differentially affected by ISX-9, with Pax4 being increased and Arx being unaffected ( Figure 1C,D). Atoh1 expression was inhibited at 40 and 80 mM, but there was little effect on Hes1 expression at any dose, an indirect measure of Notch signalling ( Figure 1E,F). Notch and Atoh1 control the gate between the secretory and absorptive lineages [37]. These data suggest that ISX-9 may act downstream of this node and could favour the differentiation of specific EEC subsets based on its differential effects on Pax4 and Arx.
We chose to use the 40 mM dose in further experiments, since at higher doses ISX-9 strongly inhibited Atoh1 and did not significantly increase NeuroD1. To explore the effect of ISX-9 on gut epithelial homeostasis and EEC specification, we designed the following protocol. After passage, intestinal organoid cultures were maintained in stem cell growth media (WENR) for 3 days to create a stem cell enriched baseline. We then switched to differentiation media (ENR) and exposed cultures to ISX-9 for up to 4 days, which is the previously reported timeframe for maturation of EECs in intestinal organoids [38]. ISX-9 increased the expression of Ngn3 and NeuroD1 at 24, 48 and 96 h but only increased chromogranin A (Chga), a gene selectively expressed in terminal-differentiated EECs [8], at 96 h ( Figure 1GeI). This finding was highly reflective of the known differentiation trajectory of the EEC lineage. We next refined our paradigm to consist of a 48-  hour pulse of ISX-9 at the beginning of the 4-day differentiation period, on day 3 of the protocol ( Figure 1J). This removed a direct effect of ISX-9 as a confounding factor and ensured measurements made at the end of the protocol could be more easily attributed to altered cell fate. In this paradigm, Ngn3 expression was unchanged, whereas NeuroD1, which is expressed in all early-to-late EEC subsets, remained elevated at 96 h, indicating that the ISX-9 pulse increased EEC specification in mouse intestinal organoids ( Figure 1K).
3.2. ISX-9 specifically enriches markers of EC cells and does not affect organoid growth Having established a paradigm in which ISX-9 appeared to increase EEC differentiation, there was a need to determine if this was authentic manipulation of cell fate or a consequence of increasing general organoid growth. We measured the morphological characteristics of organoids at different time points during our protocol: day 3 at baseline, day 5 immediately following the ISX-9 pulse, and day 7, the end of the differentiation period. There were no obvious differences between ISX-9-treated and control organoids in their general appearance or growth rates (Figure 2A). Equally, we found no difference between treatment and control in either surface area or perimeter of the organoids ( Figure 2B,C). We noted a trend for a decrease in the proportion of very budded ISX-9-treated organoids (those with >5 buds) at day 7 ( Figure 2D). This finding could suggest a decrease in stem cell proliferative capacity. However, in accordance with the surface area and perimeter observations, the number of EdU þ cells (a marker of cells in S-phase and therefore undergoing proliferation) did not differ between treatments immediately post ISX-9 treatment or at the end of the study (Figure 2E,F). These results expand our evidence that ISX-9 alters cell fate and increases the EEC lineage independently of organoid growth.
To probe the effect of ISX-9 on epithelial lineages and EEC subset differentiation, we analysed the expression of markers of individual cell types at the end of the differentiation protocol. In line with the growth data, the expression of villin (Vil1), a general epithelial marker, was unaltered between groups, as was the marker for goblet cells, mucin 2 (Muc2) ( Figure 2G). The marker for stem cells, leucine-rich repeatcontaining G-protein-coupled receptor 5 (Lgr5), was slightly increased but not significantly, whereas the enterocyte marker, alkaline phosphatase (Alpi), and the Paneth cell marker, lysozyme (Lyz1), were significantly decreased ( Figure 2G). Within the EEC lineage, ISX-9 increased markers of I cells (Cck) and EC cells [tachykinin precursor 1 (Tac1), tryptophan hydroxylase 1 (Tph1), chromogranin A (ChgA) and Lmx1a] ( Figure 2H). TPH1 is the rate-limiting enzyme for peripheral serotonin (5-HT) production, and EC cells are the most numerous gut endocrine cell type, producing 90% of the body's peripheral serotonin [23]. Whereas, Lmx1a is a transcription factor essential for the production of TPH1 [23]. Expression of L cell markers were unchanged (Pyy) or decreased (Gcg), as was the marker for K cells (Gip) ( Figure 2H). The markers for D cells (Sst), N Cells (Nts), X cells (Ghrl) and the promiscuous marker secretin (Sct) were all unchanged in response to ISX-9 ( Figure 2H). Together, these data imply that ISX-9 specifically increases the EEC lineage and drives cells to be fated toward EC cells and Cck-expressing cells rather than other mature EEC subtypes.

ISX-9 increases Ngn3 endocrine fated populations
To determine if we were observing specific alterations in EEC cell fate, we used lineage tracing, fluorescent activated cell sorting, single-cell RNA-seq and immunostaining. First, we generated small intestinal organoids from Tg(Ngn3-RFP) mice which mark the Ngn3 progenitor pool with turbo RFP [30]. We saw a five-fold increase in the percentage of RFP þ cells immediately after ISX-9 treatment, using flow cytometry, which corroborated our expression data ( Figure 3A). To uncover the effect of ISX-9 on sub-populations of Ngn3 þ cells and to attempt to identify its key regulatory mechanisms, we employed single cell RNA-seq on sorted RFP þ cells immediately after the ISX-9 treatment. The unsupervised clustering of cells, using conserved markers between control and ISX-9 treated cells, identified nine clusters split into two major branches, endocrine and non-endocrine ( Figure 3B,C). The endocrine lineage contained five clusters: early-Ngn3 (Hi Ngn3 and, Sox4), late-Ngn3 (Hi Ngn3, Cnot6l, Nkx2-2, Neurod1, Pax4 and Runxlt1), endocrine progenitor (Tac1, Chgb, Hmgn3 and Cited2), EC (Chga, Tph1, Lmx1a and Reg4) and EEC (Pyy, CCK, Gcg and Nts) ( Figure 3B, Suppl. Figure 1A). The EC and EEC clusters are consistent with recent reports of two major branches of endocrine development, one containing EC cells and the other containing peptidergic gut-hormone-producing cell types [24,39]. The non-endocrine branch consisted of four clusters. We designated one cluster as goblet progenitors because they exhibited strong goblet cell marker expression (e.g., Spink4, Muc2 and Tff3) but also contained Paneth cell markers (e.g., Lyz1 and Defa24). Two other clusters were identified as enterocyte progenitors on account of their expression of Cftr and Krt19; these clusters were distinguished from one another by one cluster expressing markers of proliferation (Top2A and Mik67) ( Figure 3B, Suppl. Figure 1A). An additional hard-to-define cluster was designated as non-endocrine progenitors (Fryl, Plac8, Stat1 and Grk4). Examination of the cluster distributions by treatment showed an increased proportion of cells in the developmentally early endocrine clusters, Early-Ngn3 and Late-Ngn3, in response to ISX-9 ( Figure 3D and Suppl. Figure 1B,E). Coincidentally, there was a decrease in the number of cells in the non-endocrine branch ( Figure 3D and Suppl. Figure 1B). Given the increased expression of EC markers and the I-cell marker Cck in our original experiments, ISX-9 plausibly biases endocrine lineage cells toward these pedigrees. However, a closer examination of Cck expression demonstrated a low level of promiscuous expression throughout the endocrine branch in contrast to other classical gut hormones (Pyy, Ghrl, Fabp5 and Gcg), which were restricted to the EEC cluster ( Figure 3B and Suppl. Figure 1C). We explored possible mechanisms of EC bias by examining TFs known to regulate endocrine differentiation. ISX-9 increased the expression of Ngn3 and NeuroD1 in the late-Ngn3 cluster, but notably increased both the expression and percentage of cells expressing Pax4 in this cluster (Figure 3EeG and Suppl. Figure 1E). Strikingly, ISX-9 increased the expression of genes associated with EC progenitor bias (Rfx6, Fev, Prdm16 and Hmgn3) whilst decreasing genes associated with EEC bias (Isl1, Arx and Cdkn1a) ( Figure 3E,F, Suppl. Figure 1D) [24]. Together these data suggest that ISX-9 increases the flux of cells through the endocrine branch in part at the expense of the nonendocrine route and increases the expression of an EC-biased genetic program in these endocrine progenitors. In addition, the powerful increase in both Pax4 expression and the percentage of Pax4 þ cells in the late Ngn3 cluster highlights a potentially novel mechanism of early EC bias in response to ISX-9 ( Figure 3E,G).
3.4. ISX-9 increases the Ngn3 lineage and enriches it with functional EC cells Next, we took advantage of a Tg(Ngn3-cre)C1Able/J::R26-loxSTOPlox-tdRFP (Ngn3-Cre-RFP) reporter transgenic line to examine the whole endocrine lineage including the Ngn3 þ progenitor pool and all daughter cell types ( Figure 4A) [31,40]. A 48-hour ISX-9 pulse increased the percentage of tdRFP þ cells analysed by flow cytometry    Figure 4B) and doubled the number of tdRFP þ cells per organoid ( Figure 4C). Expression analysis of sorted populations confirmed our earlier findings, with Ngn3 and NeuroD1 being increased immediately after 48-hour ISX-9 treatment whilst Lyz1 (Paneth) was decreased and Muc2 (goblet) and ChgA (EC cell) were unaltered (Suppl. Figure 2A). Forty-eight hours later, after removal of ISX-9, Ngn3 was downregulated, Muc2 was unchanged, Lyz1 remained reduced, NeuroD1 remained elevated and, as expected, ChgA was three-fold increased ( Figure 4D), providing evidence that ISX-9 increases the endocrine progenitor pool and biases a proportion of these cells to become specific EC cells. To further substantiate the expression data and demonstrate the programming of specific EEC subsets, we generated organoids from a CCK-iCre::R26-loxSTOPlox-eYFP reporter mouse (CCK-Cre-eYFP) [32,41]. ISX-9 increased the number of eYFP þ cells per organoid by almost three-fold (Suppl. Figure 2BeC) and produced a similar increase in the percentage of eYFP þ cells by flow cytometry analysis (Suppl. Figure 2D). In lieu of the availability of reporter mice for either chromogranin A or serotonin, we used immunocytochemistry to measure the effect of ISX-9 on the EC cell lineage. Double staining of whole organoids showed three cell populations: singly labelled 5-HT þ and CHGA þ cells, and cells that co-localised these antigens ( Figure 4E). ISX-9 increased the total number of 5-HT þ cells by twofold and CHGA þ cells by four-fold ( Figure 4F,G). Analysis of the proportion of single versus co-localised cells showed the increase to be driven mainly by cells expressing both 5-HT and CHGA, which are likely endocrine EC cells ( Figure 4H). ISX-9 increased organoid 5-HT content, and these newly generated cells were functional, releasing measurable 5-HT in response to stimulus ( Figure 4I). ISX-9 did not alter GLP-1 þ immunofluorescent cell number ( Figure 4K) despite Gcg expression being reduced, underpinning the selectivity of ISX-9's effect.
3.5. Enterochromaffin cell enrichment is partly dependent on ISX-9-induced calcium signalling Mechanistically, ISX-9 promotes neuronal differentiation by stimulating intracellular calcium signalling [27]. Thus, it seemed logical to consider if this was also true of ISX-9's actions on EEC differentiation. To investigate this, we used calcium fluorometry in Fura-2 AM-loaded mouse organoids. ISX-9 induced a long and slow (approx. 15 min) increase in mean basal-to-peak calcium response in 40% of cells examined ( Figure 5A and Suppl. Figure 3AeC). All cells including those that did not respond to ISX-9 produced the expected fast spike in calcium following administration of the positive control ATP ( Figure 5A). Positionally, responders were never found at the base of the crypt buds, where Lgr5 þ stem cells reside (Suppl. Figure 3DeF). In neuronal progenitors, ISX-9 increases calcium from both extra-and intracellular calcium sources [27]. We therefore chose to pharmacologically block ISX-9einduced calcium responses in organoids using the calcium calmodulin kinase II enzyme inhibitor KN93. CamKII is an important intracellular calcium signalling node. Interestingly, our single cell data demonstrated that the majority of Camk2b-expressing cells were restricted to the late Ngn3 and endocrine progenitor cell populations and showed a high degree of co-localisation with Pax4, particularly in the late Ngn3 progenitors (Suppl. Figure 3G). When given in conjunction with ISX-9, KN93 dose-dependently inhibited induction of Ngn3, NeuroD1 and Pax4, whereas KN93's inactive analogue, KN92, did not ( Figure 5B). Blocking CamKII signalling decreased the effect of ISX-9 on EEC differentiation. The expansion of Ngn3 þ and Cck þ cells was decreased by approximately 30%e40% when KN93 was present ( Figure 5C). KN92 did not affect the expansion of Cck þ cells driven by ISX-9 (Suppl. Figure 3H). KN93 also attenuated the increase in total CHGA þ cells by 50% and mildly but not significantly decreased the total number of 5-HT þ cells (Figure 5DeH). As expected, KN93 had little effect on the singly labelled populations of 5-HT þ and CHGA þ ( Figure 5E,F), but inhibited ISX-9-induced expansion of the co-localised population representing endocrine EC cells ( Figure 5G,H). These data show that ISX-9 manipulates EEC differentiation in part by producing a calcium signal, likely in a population of early endocrine destined progenitors.
3.6. Enterochromaffin cell enrichment is replicated in human terminal ileal organoids Historically, our understanding of mouse EEC differentiation has been based on knockout and lineage tracing in vivo studies, which were low throughput but provided key information regarding transcriptional regulation of EEC specification. The advent of organoid and single-cell technologies has rapidly expanded our knowledge in this area [24]. However, there is a deficit in our understanding of human epithelial EEC differentiation. Our mouse data identified ISX-9 as a useful tool to explore features of human EEC differentiation. We began by validating a differentiation protocol in organoids generated from terminal ileal (TI) biopsies. Organoids were stimulated to differentiate by decreasing Wnt signalling for 7 days. This protocol led to decreased LGR5 and LYZ1 but increased ALPI, VIL1 and MUC2 (Suppl. Figure 4). As expected, we observed strong increases in the expression of EEC markers, NGN3, NEUROD1, CHGA, TPH1, GCG, PYY, SST, GHRL and NTS, suggesting a broad augmentation of the lineage. Interestingly, CCK and TAC1 were not increased during the differentiation protocol (Suppl. Figure 4). Next, we designed an ISX-9 protocol for human TI that closely matched our mouse protocol. This consisted of a 3-day baseline period using stem cell media (WENRAS) followed by 7 days in differentiation media. A 48-hour ISX-9 pulse was delivered on day 6 ( Figure 6A). The human TI transcriptional response was remarkably similar to that of the mouse. NGN3, NEUROD1 and PAX4 expression were increased after 48-hour exposure to ISX-9 (day 8) (Figure 6B). At the end of the differentiation protocol (day 10), ALPI and LYZ1 expression were decreased, MUC2 was unchanged and LGR5 was increased ( Figure 6C). Examination of endocrine markers following the same protocol showed that the response to ISX-9 in human intestinal organoids mirrored that of the mouse. TAC1, TPH1, CHGA, LMX1A and CCK were increased, suggesting enrichment for EC and Cck þ cells, whereas markers of L cells (PYY, GCG), X cell (GHRL) and D cells (SST) were downregulated ( Figure 6C,D). Immunofluorescent staining confirmed EC cell enrichment, driven mainly by an expansion of cells expressing both 5-HT þ and CHGA þ (Figure 6EeG). These new cells were functional, secreting 5-HT into the media following stimulation ( Figure 6H). Overall our data point to ISX-9 programming endocrine progenitors toward an EC cell fate, prompting us to speculate whether combining ISX-9 with known stimulators of the whole EEC lineage would amplify the enrichment of EC cells. To do this, we compared the EEC transcriptional response between ISX-9; a combination of NOTCH inhibition (iNotch) and MEK inhibition (iMEK); and all three together. These inhibitors have been shown to drive EEC differentiation by inducing stem cell quiescence [9]. ISX-9 and the iNotch, iMEK combination produced differential transcriptional responses. The inhibitor combination decreased LGR5 and LYZ1 expression and, as expected, increased ALPI, MUC2, NGN3, NEUROD1, CHGA, TPH1 and CCK expression (Suppl. Figure 5). In comparison, the ISX-9 response was characterised by increased LGR5 and decreased LYZ1 and ALPI. ISX-9 was equally effective as iNotch and iMEK at increasing NEUROD1 and CCK Original Article 168 but produced stronger increases in NGN3, CHGA, TPH1 and TAC1, which was expected given its propensity for inducing EC cell enrichment. In combination, the two protocols powerfully and synergistically enriched organoids for markers of EC cells. Immunofluorescence staining confirmed a dramatic (100-fold) enrichment for endocrine EC cells (Figure 6IeK), providing further evidence that ISX-9 programmes early progenitors to become EC cells.
3.7. Overexpression of Pax4 in human ileal organoids partially mimics the effect of ISX-9 on EEC differentiation In our original experiments, Ngn3, NeuroD1 and Pax4 were identified as important transcription factors responding to ISX-9. Furthermore, at the single-cell level, our data highlighted a potential important role of Pax4 in the late-Ngn3 and endocrine progenitors that may bias cells toward an EC cell fate. To explore the role of Pax4, we generated human TI organoids with a doxycycline (Dox)-inducible Pax4 overexpressing transgene Tg(tetO-mPax4-IRES-mCherry)::Tg(CMV-rtTA), using a piggyBAC system ( Figure 7A). We overexpressed mouse Pax4, which is 85% homologous to human PAX4, so that we could distinguish between endogenous and transgenic PAX4. Dox treatment induced RFP expression within 24 h (Suppl. Figure 6A, Suppl. Movie 1), correlating with a 120-fold increase in mouse Pax4 expression, further increasing by 500-fold after 48 h (Pax4 HI ) (Suppl. Figure 6B). Dox-inducible transgenes have been documented to be mildly leaky [42]. Indeed, our transgenic untreated organoids exhibited a low but significant level of Pax4 expression (3-fold) (Pax4 Lo ) compared with genetically identical wild-type controls maintained at the same passage. Fortuitously, the level of leaky expression was similar in magnitude to the induction by ISX-9 ( Figure 5B), affording us the opportunity to examine Pax4 gene dosage on EEC differentiation. We mirrored our ISX-9 protocol by using a 3-day baseline followed by a 7-day differentiation period and gave a pulse of Dox for 48 h on day 6. Selected transcriptional markers of epithelial cell types and EEC markers were analysed on day 10 (48 h after removal of Dox) ( Figure 7B). In parallel with transgenic Pax4 expression, endogenous PAX4 expression was also increased, 3-fold in untreated and 100-fold in Dox-treated organoids, suggesting the existence of Pax4 selfregulating pathways. Supplementary video related to this article can be found at https://doi. org/10.1016/j.molmet.2020.01.012 Pax4 Lo had no effect on LGR5, LYZ1, VIL1 or ALPI expression but significantly increased MUC2 expression compared with the wildtype control, suggesting increased goblet differentiation (Suppl. Figure 6C). NGN3 and NEUROD1 expression were both significantly increased as were CHGA, TPH1 and LMX1A (EC cells), CCK (I cells), PYY (L cells), NTS (N cells), GHRL (G cells) and SST (D cells), whereas GCG (L cells), TAC1 (early EC cells), GIP (K cells) were unchanged compared with wild type (Figure 7B and Suppl. Figure 7A), suggesting that low levels of Pax4 overexpression induced selective specification of particular EEC subtypes including EC cells, partly mimicking the effect of ISX-9, but also inducing markers of N (NTS), D (SST) and X (GHRL) cells. Pax4 Hi significantly increased LGR5, LYZ1 and VIL1 whilst reducing ALPI and having no effect on MUC2 (Suppl. Figure 6C). Unexpectedly, compared with Dox-negative transgenic controls, Pax4 Hi did not enhance EEC specification, despite a 13-fold increase in NGN3 expression ( Figure 7B). In fact, the opposite was evident; all other EEC markers (NEUROD1, CHGA, TPH1, LMX1A, NTS, PYY, CCK, GHRL, GCG, SST and GIP) were either strongly decreased or undetectable ( Figure 7B and Suppl. Figure 7A), except for TAC1, which was increased two-fold ( Figure 7B). This suggested EEC differentiation was stalled by high expression of Pax4. To investigate this further, we measured transcriptional markers on day 12 of our protocol, 96 h following removal of Dox. At this time point, Pax4 Hi expression was decreased from 94-fold to a 3-fold induction ( Figure 7C). Accordingly, endogenous PAX4 induction was halved in the Pax4 Hi group, demonstrating that Pax4 expression was rapidly induced by Dox but was relatively slow to downregulate following Dox removal. Elevated NGN3 expression was decreased from 13-fold to 2-fold, and NEUROD1 was no longer suppressed ( Figure 7C), which was associated with normalisation of TAC1 expression and a disinhibition of EC cell markers (CHGA, LMX1A and TPH1) ( Figure 7C). All other endocrine markers (CCK, SST, NTS, GHRL, GCG, PYY and GIP) remained supressed ( Figure 7C and Suppl. Figure 7B). These data suggest that low levels of Pax4 expression enhance EEC specification but high levels trap EECs in an early progenitor state with an EC cell bias. As Pax4 expression normalises, endocrine differentiation proceeds with the appearance of EC cells. This finding helps explain ISX-9's effects on EC cell enrichment and suggests that the upregulation of PAX4 and TAC1 is important.

DISCUSSION
Enteroendocrine cells respond to diverse signals in the luminal environment including nutrients, microbial metabolites and pathogens. They play a central role in integrating these complex signals and altering physiology by modulating epithelial, immune, neuronal and hormonal functions. These features make the enteroendocrine system a potential target for treating multiple conditions and, notably, attention has focused on metabolic diseases [3]. An increasingly important feature of EECs is their plasticity, which hypothetically could be appropriated to alter their density and/or functional characteristics for therapeutic gain [43]. This is exemplified by two studies, which increased the secretory lineage in vivo using either a Notch or Rho-associated coiled-coil-containing protein kinase (ROCK) inhibitor [44,45]. The increased EEC density included GLP-1producing L cells, promoting glucose control and decreasing hyperglycaemia in models of diabetes. The caveat to these proofs of principle studies is their broad effect across the secretory lineage. Targeting specific EEC differentiation pathways might be a more suitable approach but requires a deeper understanding of the regulatory networks controlling EEC specification, particularly in the human epithelium, which had been difficult to study before the advent of organoid technology. We used ISX-9, a small molecule activator of NeuroD1, to further investigate EEC differentiation in mouse and human intestinal organoids. In the gut, NeuroD1 is downstream of Ngn3 in endocrine progenitors, and its deletion decreases the number of CCK and secretin cells in mice [46], potentially offering the opportunity to modulate specific EEC cell fates using ISX-9. Indeed, in our initial experiments, ISX-9 strongly increased classical TFs known to be key members of the regulator network controlling EEC differentiation (Ngn3, NeuroD1 and Pax4), but did not affect Arx, suggesting specificity. An analysis of organoids derived from transgenic reporter mice or immunostained for serotonin showed that ISX-9 increased markers of endocrine progenitor development and demonstrated an enrichment of functional terminally differentiated EC cells. At single-cell resolution, we were able to corroborate the presence of two major developmental endocrine trajectories, peptidergic (gut hormone) versus enterochromaffin, and identify developmentally earlier endocrine clusters. An analysis of ISX-9 treated cells provided evidence of increased cells in the endocrine branch at the expense of the non-endocrine goblet/enterocyte Original Article 170 lineages. A notable increase of Pax4 expression in the late-Ngn3 population coupled with an activation of a genetic program biased toward EC differentiation (Rfx6, Fev, Prdm16 and Hmgn3) could explain the enrichment in functional EC cells. These actions were upstream of factors important for late EC cell differentiation or identity, Lmx1a, Chga and Tph1, which were increased at the final time point (i.e., 48 h after ISX-9 was removed). Importantly, this finding appeared to be a genuine manipulation of lineage fate and not an alteration in organoid growth. Equally, organoid proliferation and Lgr5 expression were unaltered, in contrast to a previously published paradigm that promotes EEC lineage differentiation by inducing Lgr5 stem cell quiescence using NOTCH and MEK inhibitors [9]. The triggering of neuronal differentiation by ISX-9 is calcium dependent [27]. Similarly, its effect on the endocrine lineage was partly dependent on calcium signalling. Inhibition of the intracellular calcium signalling node, CamKII, attenuated EC and Cck þ cell enrichment and blocked induction of Ngn3, NeuroD1 and Pax4 expression. This corresponds with recent data in Drosophila, showing that activation of the mechanosensitive receptor Piezo in a population of mid-gut endocrine progenitor cells increases EEC differentiation through cytosolic Ca 2þ [47]. In mice, Piezo2 is found in a sub-population of EC cells that release serotonin in response to stretch [48]. Future work will be aimed at identifying in which cells ISX-9 stimulates calcium and whether calcium signalling is an important regulator of homeostatic endocrine differentiation in the gut. Most data relating to EEC differentiation have been gathered in mice or using mouse tissues. Comparatively little is known about the networks controlling endocrine cell fate in the human gut epithelium, but they are assumed to be conserved to some degree, stemming largely from the study of patients with mutations in NGN3 who exhibit intractable malabsorptive diarrhoea due to loss of EECs and a handful of recent studies in organoids [49e51]. In these in vitro studies, a pulse of transgenic NGN3 in iPSCs-derived human organoids mostly recapitulated the predicted endocrine repertoire [26]. The same transgenic construct has been deployed in organoids derived from tissue resident stem cells and similarly promoted EEC differentiation [52]. Our data shed further light on human EEC differentiation and its similarity to that of the mouse. Notably, ISX-9 produced identical effects on the EEC lineage between species. NGN3, NEUROD1 and PAX4 upregulation was associated with selective increases in CCK, CHGA, TPH1, LMX1A and TAC1. Whole organoid immunostaining confirmed enrichment for EC cells double positive for 5-HT and CHGA, and these cells functionally released 5-HT. Our data have several implications; they highlight how similar EEC specification is between the mouse and humans and that ISX-9 alone or in combination with iNotch and iMEK can be used to enrich human organoids with EC cells allowing functional exploration of these rare and difficult-to-study cells. EC cells modulate GI motility, bone formation, hepatic gluconeogenesis, thermogenesis, insulin resistance and regulation of fat mass [53]. Understanding how these cells function in humans has important implications. Our data highlighted the potential importance of Pax4, and its induction in the late Ngn3 population seemed to offer a developmentally early mechanistic explanation for EC cell enrichment by ISX-9. However, low level constitutive expression produced a generalised increase in EEC lineage markers, although notable exceptions were GCG and GIP. We also observed a surprising and powerful upregulation of NGN3, suggesting reciprocal regulation between PAX4 and NGN3. The upregulation of MUC2 and LYZ1 might also be explained by this NGN3 upregulation, as 15% of goblet and 40% of Paneth cells are derived from Ngn3 progenitors [40]. Strikingly, induction of high levels of Pax4 expression completely inhibited endocrine differentiation, trapping EEC differentiation at an early stage. Unexpectedly, induced transgenic Pax4 was persistent, remaining upregulated by 3fold 96 h after doxycycline removal, although this was reduced from a peak induction of 94-fold. Interestingly, at this time point, all markers of peptidergic EECs (CCK, SST, NTS, GHRL, GCG and PYY) remained supressed, but markers of mature EC cells (LMX1A, CHGA and TPH1) were now disinhibited, and TAC1, an early EC cell marker, was no longer upregulated. We drew several conclusions from these findings. First, our data are consistent with the existence of two major lineages of EECs in humans, as recently described in the mouse and corroborated by our single cell data [24,54]. One lineage gives rise to an early-appearing EC cell population and the other to peptidergicproducing cells, which generally appear later than EC cells. Second, PAX4, as in the mouse, is upstream of NEUROD1 and likely marks an early endocrine progenitor cell. Third, understanding the network of TFs controlling EEC differentiation is complicated by their promiscuity and the need to appreciate the timing and level of expression. Finally, it seems plausible that ISX-9 enriches the ECC lineage in part by its effects on PAX4 expression, which may represent an increase in the endocrine pool biased toward an EC cell fate. Finally, our data add to a handful of studies suggesting that small molecules could be found to selectively control EEC cell fate/specification with a view to treat various clinical conditions. However, our data also highlight the difficulties this approach faces. Understanding to what degree the targeted lineage creates a deficit in another and whether this induces unwanted physiological effects is of key importance. This seems a likely event when intervening downstream of NGN3, where one EEC type may be enriched at the expense of another. This situation could be mitigated by combining selective EEC targeting with a more generalised EEC lineage activator. Realising this potential will require a much deeper understanding of human EEC lineage specification aided by the organoid platform.

AUTHOR CONTRIBUTIONS
A. T. and P. F. P. helped in the design of experiments, collected data and contributed to the writing of the manuscript. P.P and B.H.H provided human biopsies for the isolation of crypts. G. A. B. wrote the manuscript and managed the project.