Using the PC to benchmark cell type representation of conventional organoids against their in vivo counterparts
Conventional intestinal organoids produced from the spontaneous differentiation of ISCs have been used to study PCs in vitro in multiple contexts [23, 24]. These in vitro PCs exist as part of a heterogeneous system, yet to be rigorously benchmarked against their in vivo counterparts. To better understand the composition of PCs within conventional organoids and how well those PCs approximate their in vivo counterparts, we sought to globally compare the conventional organoid-derived PCs and their in vivo counterparts through a single-cell transcriptomic approach (Fig. 1a).
To relate the organoid-derived PC state to in vivo PCs, we first generated an unbiased reference in vivo scRNA-seq data set. We performed massively parallel scRNA-seq using the recently developed Seq-Well platform [28] on epithelial cells from the ileal region of the small intestine acquired as two biological replicates (see Methods). We assessed quality metrics for the number of genes, unique molecular identifiers (UMIs), mitochondrial genes, and ribosomal genes, all of which fell within expectations (all cells average: 1043 genes, 2168 UMIs, 5.4% ribosomal genes, 10.4% mitochondrial genes). UMI-collapsed cell-by-gene (7667 cells × 17,505 genes) expression matrices were analyzed using Seurat (see Methods), performing dimensionality reduction, graph-based clustering, and deriving lists of cluster-specific genes in order to identify PCs. Within the spectrum of cell types, we identified two clusters (2 and 11) enriched for Lyz1 expression (Fig. 1b, c), of which we determined cluster 11 to be fully mature PCs (n = 189 cells) based on uniform expression of a set of associated antimicrobial peptide marker genes such as Defa22, Defa21, and Ang4 (receiver operating characteristic (ROC) test, area under the curve (AUC) > 0.99 for markers listed; cluster 11 average: 866 genes, 3357 UMI, 3.5% ribosomal genes, 4.8% mitochondrial genes) (Additional file 1: Table S1). We further utilized these genes (genes with AUC > 0.65 for in vivo PC) throughout our study to relate organoid-derived cell states to in vivo PCs. They are fully inclusive of the 14 high confidence markers described for Paneth cells from the terminal ileum in the recently published mouse small intestinal atlas [3]. Of note, we extended our gene list beyond truly specific marker genes that are not expressed in other cell types as we were interested in a more comprehensive set of PC-enriched genes for further comparison.
We next performed scRNA-seq using Seq-Well on conventional organoids derived from a single donor ISC-enriched state (Fig. 1a). Beginning with murine small intestinal crypts, we directly enriched for LGR5+ ISCs over 6 days following isolation within a Matrigel scaffold and medium containing recombinant growth factors EGF (E), Noggin (N), and R-spondin 1 (R), small molecules CHIR99021 (C), and valproic acid (V), as well as Y-27632 for the first 2 days to inhibit rho kinase and mitigate anoikis, as previously described (ENR+CV) [29]. To ensure reproducibility within our system and limit the risk of interference in our chemical induction approach, we conducted our study exclusively with recombinant growth factors and not cell line-derived conditioned media. Cells were passaged into conventional ENR culture for an additional 6 days to allow multi-lineage differentiation and produce stem cell-derived in vitro PCs. Following scRNA-seq, we computationally identified six clusters (amongst 2513 cells × 16,198 genes meeting quality standards, see Methods) in ENR organoids, which we label as ENR1-4, and EEC-1 and -2 for two EEC types (Fig. 1d). We identified ENR-4 as the cluster most enriched for Lyz1 and our PC reference gene set (effect size 0.721, ENR-4 vs. all ENR, *t test p < 2.2 × 10−16; for effect size details see Methods) (Fig. 1e, f). Having identified ENR-4 as the cell state of interest in organoids, we directly compared the top 200 most PC-like cells in ENR-4 to in vivo PCs by performing differential expression analysis (Fig. 1g). In comparing the two cell types, it became evident that the majority of genes enriched by in vivo PCs were defensins and antimicrobials, including Defa22, Defa21, Zg16, Ang4, Defa3, and Lyz1 (all p < 2.92 × 10−37, bimodal test, Bonferroni corrected for multiple comparisons) (Fig. 1g, h). ENR-4 cells were enriched for Chgb, an enteroendocrine marker, and translational biosynthetic genes likely indicative of the high rates of proliferation present in ENR organoids (Fig. 1g). We further note the difference in genes arising from non-sex matched comparison, like Xist, as a limitation of our comparison between a single donor for organoid derivation. Beyond these selected genes, we note a global reduction in the fraction of the transcriptome of ENR-4 cells producing the total cadre of in vivo PC marker genes (effect size 1.25, InVivo vs. ENR, *t test p < 2.2 × 10−16), suggesting that the current in vitro organoid-derived PCs are suboptimal for physiological studies (Fig. 1i).
Modulating key developmental pathways of stem cell-derived systems has emerged as a paradigm in bioengineering to rationally generate cell types for basic research and therapeutic aims [30, 31]. Specifically, modulating Wnt and Notch signaling has been suggested in the literature to increase the frequency and magnitude of Lyz1 expression and protein in ISC-derived cells [29, 32,33,34]. Leveraging the single-cell transcriptomes of our in vitro- and in vivo-derived PCs, we confirmed that Wnt target genes are enriched in vivo relative to in vitro PCs (effect size 0.559, InVivo vs. ENR, *t test p < 2.035 × 10−8) and Notch target genes were decreased (effect size −0.500, InVivo vs. ENR, *t test p < 5.25 × 10−7) (Fig. 1i, Additional file 2: Table S2). As a result, we sought to comprehensively test if driving Wnt and inhibiting Notch truly results in a more physiologically representative PC versus the organoid-derived PC, beyond bulk measures of increased Lyz1 expression.
Rationally guided chemical induction of Wnt and inhibition of Notch drives PC marker enrichment
Beginning with an LGR5+ ISC-enriched population (ENR+CV), we sought to profile how the modulation of Wnt and Notch signaling through small molecule inhibitors would alter the in vitro PC state, as suggested by our transcriptional profiling. We performed chemical induction (CI) using the previously identified compounds C to drive Wnt signaling and DAPT (D), a gamma-secretase inhibitor, to inhibit Notch (ENR+CD) (Fig. 2a) and measured bulk gene expression of PC (Lyz1, Defa1, Mmp7) and ISC (Lgr5) markers every 2 days for a total of 6 days (Fig. 2b). ENR+CD-treated cells had statistically significant increases in Lyz1 (adj. p = 0.005, see Methods) and Mmp7 (adj. p = 0.005) within 2 days compared to ENR, with differences plateauing around 4 days. Defa1 (adj. p = 0.004) expression was significantly increased by day 4 and plateaued by day 6 in ENR+CD versus ENR populations. Lgr5 expression in ENR+CD at 2 days versus ENR showed an insignificant plateau of expression, which trended down by 6 days. This may be indicative of an expansion in ‘label-retaining’ secretory precursors [35]. The precursor population ENR + CV had no significant difference in PC or ISC markers relative to ENR. The significant increase in PC gene expression in ENR + CD relative to ENR and ENR+CV over the 6-day treatment suggests rapid enrichment following CI, supporting our hypothesis that alterations in Wnt and Notch result in superior PC enrichment in vitro.
To phenotypically describe PC enrichment following CI, we performed imaging and immunocytochemistry for PC-associated features. After 6 days of ENR + CD, cell populations exhibited darkened annular morphology consistent with increased numbers of granule-rich cells (Additional file 3: Figure S1A). Confocal microscopy of whole cell clusters stained for anti-DEFA and anti-LYZ showed an increase in LYZ+ and DEFA+ cells in ENR + CD compared to both ENR and ENR + CV (Fig. 2c). Single-cell counting of confocal imaging showed a significant increase of DEFA and LYZ co-staining cells in ENR + CD (20–30% of cells) versus either ENR or ENR + CV (both < 5%; adj. p = 0.0001) (Additional file 3: Figure S1B). Additionally, normalized z-axis profiles of individual co-staining cells within cell clusters revealed a consistent distribution of DEFA (luminally polarized) and LYZ (diffuse) (Additional file 3: Figure S1C 1–3). High-resolution fluorescent imaging of individual co-staining cells from freshly isolated small intestinal crypts (in vivo equivalent) and 6-day ENR + CD-treated cells showed a similar polarized distribution of LYZ- and DEFA-stained granules, although freshly isolated cells appeared to be more granular than CI-PCs (Fig. 2d).
To confirm the extent of enrichment seen in whole population imaging, the prevalence of PCs in ENR + CD relative to ENR was assessed by flow cytometry over the course of 12 days, a longer term culture than typical for conventional organoids. We identified an in vivo PC phenotype as CD24 and LYZ co-positive cells, as per previous reports [36], and noted the presence of single-positive LYZ+ or single-positive CD24+ populations, indicative of alternative cell differentiation, immature, or non-physiological PCs (representative populations Additional file 3: Figure S1D, representative gating Additional file 3: Figure S1E). ENR + CD had substantial enrichment at all time points for double-positive, and single-positive LYZ+ or CD24+ populations relative to ENR, as well as a consistent decrease in the double negative population in agreement with the PC phenotype (Fig. 2e). Notably, both ENR and ENR + CD experience declines in total cell viability, with ENR + CD having greater survival at longer times, suggesting both a reduction in anoikis, a potentially physiological ‘long-lived’ PC phenotype in ENR + CD versus ENR, or an enhancement in niche-supporting functionality (Additional file 3: Figure S1F). Overall, imaging and flow cytometry demonstrate a significant increase in cells morphologically resembling in vivo PCs with respect to granularity, polarity, and antimicrobial co-expression in ENR + CD compared to conventional ENR organoids (Fig. 2c–e and Additional file 3: Figure S1A–F).
Chemically induced PC proteome is enriched for components of secretory lineages
With ENR + CD apparently providing a more prevalent and physiological PC population, we sought to more globally characterize the differences between in vitro PCs (ENR vs. ENR + CD) at 6 days. Because our single cell transcriptomic comparison revealed that many of the differential genes between PCs in conventional organoids and in vivo were lineage-defining protein products, we sought to assess the total intracellular proteome between the conventional organoid and our chemically induced model through liquid chromatography mass spectrometry (LC-MS/MS)-based proteomics. We quantified relative protein abundance using isobaric mass tag labeling from four ENR and four ENR + CD samples and analyzed them in a single 10-plex by LC-MS/MS (Additional file 4: Figure S2A). We identified 8015 unique proteins within all samples; each replicate pair (ENR + CD/ENR) was normally distributed (not shown) and correlated with all others, indicating consistent proteome enrichment (Additional file 4: Figure S2B). We looked at the sample pairs in aggregate and classified proteins significantly enriched in ENR + CD and ENR by a false discovery rate (FDR) < 0.05 and log fold change (± 2σ) (Fig. 2f and Additional file 5: Table S3). There were 249 ENR + CD-enriched proteins, 212 ENR-enriched proteins, and 7553 shared proteins. Known PC markers, including LYZ, DEFAs, and other secretory pathway components, were identified as significantly enriched in ENR + CD versus ENR alone. Of known antimicrobial proteins produced by PCs, we detected 10 DEFAs, 5 CRS peptides, 6 ribonucleases, 12 lectins, LYZ1, and PLA2G1B with differential abundance between ENR + CD and ENR (Fig. 2g). Each class of antimicrobials had at least one ENR + CD enriched protein (+ 2σ), with the ribonucleases significantly enriched and a majority of the lectins and DEFAs unregulated between the two conditions. Proteins associated with the EEC lineage (secretogranins, chromogranins, and neuropeptides) were also enriched in ENR + CD, in addition to multiple other secreted components, including Wnt ligands, and the complement pathway components C3 and CFI (Fig. 2h). In sum, we see a broad diversity of PC-associated antimicrobials with some enrichment of EEC-associated proteins in ENR + CD relative to ENR.
Additionally, we characterized enriched biological functions, cellular compartments, and molecular functions using DAVID v6.8 and the gene ontology database. All sets had high database coverage (greater than 85%) of queried proteins. The ENR + CD proteome is significantly enriched for extracellular and protein processing compartments and secretory-associated functions (Additional file 4: Figure S2C), while the ENR proteome favors translation, intracellular compartments, and translational activities (Additional file 4: Figure S2D). Of note, there are the extracellular exosome and calcium ion-binding associated proteins in the ENR + CD proteome that are indicative of the intestinal epithelial secretory phenotype (for a complete list of DAVID enrichments, refer to Additional file 6: Table S4). These functional enrichments further support the notion that the ENR + CD-cultured organoids are enriched in secretory cells, including PCs, although it does not rule out potential co-enrichment for the EEC lineage. Finally, we sought to identify transcription factors (TFs) that may mediate PC-specific differentiation using GSEA [37, 38] with the MSigDB transcription factor target (v5.2) gene set database [39] with a moderately conservative cutoff (see Methods). We generated an enrichment map [40, 41] of several TF targets significantly enriched in both the ENR + CD and ENR proteomes. In ENR + CD, the nuclear receptors for progesterone, aldosterone, and glucocorticoid, as well as the cellular differentiation-implicated TAL1, RP58, and NRSF, were significantly enriched. In ENR, the primary known enrichment was for the cell cycle and proliferation-related E2F TF family (Additional file 4: Figure S2E). These potential TFs are consistent with CI-PC treatment driving expected terminal differentiation of specialized cells, as opposed to conventional organoid culture, which supports a broad mix of intestinal epithelial cells, including proliferating populations. Furthermore, this analysis suggests potential targets, such as progesterone, aldosterone, and glucocorticoid, to modulate the differentiation programs of this secretory cell population in future studies.
Single-cell RNA sequencing profiles heterogeneity of chemically induced PCs, revealing subsets with improved transcriptional similarity to in vivo
With the apparent co-enrichment of canonical PC and EEC proteins in the ENR + CD proteome, we sought to identify whether we produce a homogenous population of mixed-lineage secretory cells or a spectrum of unique cell states between EEC and PC. We performed scRNA-seq using the Seq-Well platform on cells from ENR + CD and the precursor ENR + CV conditions to analyze alongside conventional ENR organoids. To ensure experimental robustness, we assessed quality metrics for the number of genes, UMIs, mitochondrial genes, and ribosomal genes by cluster, all of which fell within expectation (Additional file 7: Figure S3). UMI-collapsed digital gene expression matrices were analyzed using Seurat (see Methods), and displaying all three treatments (ENR + CV, ENR, ENR + CD) in tSNE space demonstrated clear separation between each condition (Fig. 3a), illustrating that the unique transcriptional differences induced by each treatment were conserved across all cells. Plotting key genes demonstrated that, as expected, all cells expressed high levels of Epcam, that ENR + CV cells had enhanced Mki67, a marker of proliferation, that the ENR + CD condition led to enrichment of cells expressing antimicrobial Lyz1, Defa24, Defa3, Mmp7, and EEC marker Chga, and that ENR enriched for absorptive marker Fabp2-expressing cells (Fig. 3b).
To assess subpopulation structure and provide a more robust measure of composition beyond canonical marker genes, we performed unsupervised KNN graph-based clustering on the captured cells (Fig. 3c, d and Additional file 1: Table S1 for full gene lists), distinguishing four clusters in each treatment condition. We then scored individual clusters according to the amount of the transcriptome within each cell dedicated to synthesizing the respective enriched proteins from the bulk proteome data. We observed that ENR + CD clusters yield a significant enrichment for those proteins detected in the up-regulated proteome (effect size 1.38 ENR + CD vs. ENR clusters, p < 2.2 × 10−16) and that the down-regulated proteins were enriched in the ENR and ENR + CV conditions (Fig. 3d, e and data not shown). Intriguingly, at the level of clusters, the upregulated proteome was not evenly distributed across all cells in ENR + CD, but was most enriched in cluster ENR + CD-4, which represented approximately 10% of ENR + CD cells (effect size 2.40 ENR + CD-4 vs. all cells, p < 2.2 × 10−16) (Fig. 3d, e).
To address ENR + CD composition and how it relates to conventional organoids, we interrogated the expression of Lyz1, Chga, and other selected genes across each cluster (Fig. 4a). We noted that clusters ENR-4 and ENR + CD-4 shared expression of Lyz1, Defa24, Defa3, and Mmp7, yet ENR + CD-4 cells produced significantly more of each canonical PC gene (bimodal test, p < 6.80 × 10−74 for genes listed, Bonferroni corrected for multiple comparisons). Furthermore, both ENR-4 and ENR + CD-4 cells lacked expression of EEC genes like Chga, which was observed in the EEC-1 and EEC-2 clusters arising from mixed-grouping of the sample, as well as in ENR + CD-2 and ENR + CD-3 (Fig. 4a). Altogether, this suggests that ENR + CD drives PC differentiation while also inducing a secretory transition state (ENR + CD-2 and 3) expressing a mix of PC and EEC marker genes (Additional file 1: Table S1 for full gene lists).
We next sought to compare the states generated in vitro to those observed in vivo with our refined system. Using the gene list of in vivo PC markers and further defining a list for in vivo EECs (see Methods) captured on the Seq-Well platform (Additional file 1: Table S1), we observed that the percentage of a cell’s transcriptome dedicated to synthesizing defining Paneth genes was significantly enriched relative to ENR-4 in clusters ENR + CD-2, -3, and -4 (effect size 0.15, p < 3.43 × 10−5; effect size 0.829, p < 2.2 × 10−16; effect size 2.52, p < 2.2 × 10−16, respectively) with an increase in expression of EEC genes across ENR + CD-1, -2, and -3 but not ENR + CD-4 (effect size 1.30, p < 2.2 × 10–16; effect size 1.82, p < 2.2 × 10–16; effect size 1.118, p < 2.2 × 10−16; effect size 0.0465, p = 0.2339, respectively) (Fig. 4b). Notably, ENR + CD-4 cells (~10%) had a three-fold increase in the transcriptional resemblance to in vivo PCs relative to ENR-4 (53.4% of transcriptome ENR + CD-4 vs. 16.5% of transcriptome ENR-4) (quantification of Fig. 4b). Furthermore, 45% of ENR + CD cells express a secretory PC-like transcriptional phenotype that is at least two-fold enhanced relative to conventional organoids (33.9% of transcriptome ENR + CD-3 and -4 vs. 16.5% ENR-4). Comparing the ENR + CD-4 cells relative to in vivo PCs demonstrated a striking similarity relative to the difference observed between in vivo and ENR-4 cells (PC fraction of in vivo transcriptome: effect size 0.237 InVivo vs. ENR + CD-4, p < 0.0055; effect size 1.25 InVivo vs. ENR-4, p < 2.2 × 10−16 (Additional file 1: Table S1).
In Fig. 4c, we present a heatmap of scaled expression values for the top genes (AUC > 0.65) used for the in vivo Paneth score across ENR-4, ENR + CD-4, and the in vivo cluster used to define PCs. We observed that the enhanced PC phenotype in ENR + CD-4 (effect size 1.144 ENR + CD-4 vs. ENR-4, p < 2.2 × 10−16) correlated with a greater expression of signature genes, such as Lyz1, Lyz2, and Defa5, and a greater diversity of antimicrobial peptide genes, such as Ang4, Defa3, and the metalloprotease Mmp7.
To confirm and extend our findings of pathway-based modulation, we scored clusters for enrichment or depletion of canonical growth factor-induced pathways. CHIR99021 activates the Wnt pathway, and we observed a significant enrichment for Wnt target genes in all CI-PC clusters (effect size > 0.999, p < 2.2 × 10−16 for all ENR + CD clusters vs. ENR-4) (Additional file 8: Figure S4A). While DAPT is a Notch pathway inhibitor, levels of Notch target genes were largely greater than or equivalent to ENR-4 cells across CI-PC clusters, except for significant depletion in ENR + CD-4 (effect size −0.658, p < 2.2 × 10−16 ENR + CD-4 vs. ENR-4) (Additional file 8: Figure S4B). This suggests that complete Notch suppression is key for PC differentiation distinct from an EEC fate. Additionally, given the recognized role for distinct respiratory potential in enterocytes, ISCs, and PCs, we scored cells across respiratory electron transport genes [42, 43]. ENR + CD-4 had the lowest cluster score relative to all cell subsets (effect size −1.4649, p < 2.2 × 10–16) (Additional file 8: Figure S4C). Together, this suggests that Wnt signaling is necessary but not sufficient to specify the mature PC phenotype and that Notch and metabolic conditions play a larger role in the decision between PC and EEC fates.
Chemically induced PCs mimic in vivo stimulant-induced secretion and demonstrate selective modulation of bacteria in co-culture
In addition to our morphological, proteomic, and transcriptional characterization of PC phenotype in ENR + CD and ENR, we sought to measure physiological function by assessing stimulant-induced secretion of antimicrobials. We assessed the dynamics of LYZ accumulation in the supernatant media of cultures following media wash, basally and after stimulation with carbachol (CCh), a cholinergic agonist known to induce PC secretion [44]. CCh (10 μM) induced a rapid accumulation of LYZ within 2 hours that plateaued around 6 hours post-wash (two-way ANOVA, stimulant p < 0.0001, time-point p < 0.0001) (Fig. 5b). The observed PC secretion in response to CCh is consistent with observations made in ex vivo crypts, though over appreciably longer time scales, likely due to the added diffusion barrier of the organoid structure and matrigel [44]. We next identified how LYZ secretion changes over the course of differentiation. Beginning with an ISC-enriched population, we assayed for secreted LYZ in cell culture supernatants every 2 days for 6 days of ENR + CD culture, following a 24-h stimulation with CCh or without (basal collection/non-stimulated). Notable increases in functional secretion (stimulated relative to basal) occurred at days 4 and 6 (two-way ANOVA, stimulant p < 0.0001, time-point p < 0.0001) (Fig. 5a). Compared to conventional organoids and ISC-enriched precursors, ENR + CD secreted significantly more basal LYZ (p < 0.0001) and was the only population that showed grossly measurable CCh-induced secretion (adj. p = 0.03) (Fig. 5c). This result is consistent with the observed enrichment between chemically induced populations relative to conventional.
Based on the broad spectrum of antimicrobials detected proteomically, transcriptionally, and functionally, we hypothesized that ENR + CD possess greater bactericidal effects than conventional organoids. We assayed for bacterial growth modulation by suspending cell clusters with common laboratory strains of gram-negative and gram-positive bacteria in exponential growth. CI-PCs significantly suppressed growth of gram-positive L. lactis MG1363 (adj. p = 0.0001), which did not occur with conventional organoids, indicative of increased PC-associated antimicrobial activity (Fig. 5d). Both ENR (adj. p = 0.0005) and ENR + CD (adj. p = 0.01) co-culture showed significant increase in gram-negative E. coli MG1655 growth but no appreciable effect was observed on the growth of gram-positive E. faecalis V583 versus bacteria alone (Fig. 5d). While this assay simplifies the PCs’ physiological environment and may not be a direct proxy for strain-specific growth modulation, it does demonstrate that the PC-enrichment of ENR + CD versus conventional organoids enables detectable in vitro bacteria species-specific PC antimicrobial response, opening avenues for future experimentation.
Chemically induced PCs provide niche support and enhance conventional organoid survival
Beyond the generation of antimicrobial peptides, PCs provide niche support for ISCs. We sought to test if CI-PCs provided niche factors known to drive epithelial regenerative turnover. We performed co-culture experiments, mixing and re-plating cell populations derived from 6 days of ENR or ENR + CD culture and assayed co-culture viability, caspase activity, and cytotoxicity 24 and 48 h following re-plating in ENR media. If there was no appreciable interaction, positive or negative, between the two populations we would expect to see a linear trend of every measured variable throughout mixing ratios. However, we observe a significant positive interaction where the presence of both populations drives an overall increase in cellular viability, beginning at 24 h (one-sample t test 1:1 p = 0.037) and increasing at 48 h (one-sample t test 1:1 p = 0.001 and 1:3 p < 0.001) (Fig. 5e). This is likely due to a significant decrease in overall apoptosis relative to the total cell population (one-sample t test 24 h 1:1 p = 0.004 and 1:3 p = 0.032, 48 h 1:3 p = 0.003), and unrelated to changes in cellular cytotoxicity. We believe that the presence of a PC-enriched population (from ENR + CD) is driving this effect by providing increased soluble regenerative factors to the ISC population in ENR organoids, increasing the generation of new cells, and resulting in a lower overall rate of apoptosis.
Mapping of in vivo PC-associated transcription factors to in vitro proteome and transcriptome reveals Nupr1 as important in epithelial survival
Finally, we sought to use this physiologically improved in vitro PC system (ENR + CD) to identify novel factors potentially supportive of PC survival or differentiation. Using our in vivo PC and EEC gene lists, and filtering for only TFs (using TFdb, downloaded September 2017) [45], we identified a set of PC- or EEC-specific TFs. We mapped these TFs to our in vitro proteome (Fig. 6a and Additional file 5: Table S3), which revealed the previously unreported NUPR1 as the most enriched PC-specific TF in ENR + CD. This finding was supported by differential expression between ENR + CD-2 (most EEC-like cells) and ENR + CD-4 (p < 3.12 × 10−37, bimodal test, Bonferroni corrected for multiple comparisons) (Fig. 6b). We further identified Nupr1 in our in vivo PC populations, which showed specific and enriched expression of Nupr1 by in vivo PCs (ROC test, AUC = 0.833) (Fig. 6b). Intriguingly, Nupr1 is a stress-response gene, known to promote cellular survival and senescence through mediation of autophagy, and has primarily been studied in the context of cancer [46,47,48]. Autophagy and stress response have repeatedly been implicated through GWAS study in PCs in IBD; however, Nupr1 has only ever been reported in a single IBD GWAS study, and its role in PC biology has not been formally investigated [49]. With our model, we sought to test the role of NUPR1 on in vitro PC survival through the small molecule inhibition of NUPR1 with trifluoperazine (TFP) [50, 51]. While genetic perturbation may provide for more specific effect measurement, we chose to use TFP as a simple, albeit less specific, modulator, as the complexity involved in selecting for a genetically perturbed population of PCs in organoids, if Nupr1 is a survival gene, is beyond the scope of the present study. We first tested how different dosages impact PC differentiation in combination with ENR + CD for 6 days, where doses above 1 μM led to near total cell death, and where the few surviving cells were primarily non-PC (Fig. 6c). This suggests that Nupr1 is likely critical to cellular survival during the CI differentiation process. We also tested the addition of TFP for 2 days following a 6-day course of ENR + CD, where again a profound, but not total, decline in cellular viability was observed. Further, it appears that TFP treatment is selectively more toxic to PC and PC-progenitor populations relative to non-PC populations (Fig. 6d). In total, this initial investigation suggests that NUPR1 may be a critical TF in PC development and survival, which carries therapeutic implications which we will seek to validate with expanded gene-perturbation studies in vitro and in vivo in future work.