Screening for modulators of the cellular composition of gut epithelia via organoid models of intestinal stem cell differentiation

Murine crypt isolation and organoid culture

Small intestinal crypts were isolated under animal protocols approved by the Massachusetts Institute of Technology (MIT) Committee on Animal Care (CAC). Proximal and/or distal small intestines were isolated from wild-type C57BL/6 mice of both sexes, aged between 1 and 6 months in all experiments. Small intestinal crypts were isolated as previously described²³. Briefly, the small intestine was collected, opened longitudinally and washed with ice-cold Dulbecco’s phosphate buffered saline without calcium chloride and magnesium chloride (PBS0) (Sigma-Aldrich) to clear the luminal contents. The tissue was cut into 2–4 mm pieces with scissors and washed repeatedly by gently pipetting the fragments using a 10 ml pipette until the supernatant was clear. Fragments were rocked on ice with crypt isolation buffer (2 mM EDTA in PBS0; Life Technologies) for 30 min. After isolation buffer was removed, fragments were washed with cold PBS0 by pipetting up and down to release the crypts. Crypt-containing fractions were combined, passed through a 70 μm cell strainer (BD Bioscience), and centrifuged at 300 r.c.f. for 5 min. The cell pellet was resuspended in basal culture medium (2 mM GlutaMAX (Thermo Fisher) and 10 mM HEPES (Life Technologies) in advanced DMEM/F12 (Invitrogen)) and centrifuged at 200 r.c.f. for 2 min to remove single cells. Crypts were then cultured in a Matrigel culture system (described below) in small intestinal crypt medium (100X N2 supplement (Life Technologies), 100X B27 supplement (Life Technologies), 1 mM N-acetyl-l-cysteine (Sigma-Aldrich) in basal culture medium) supplemented with differentiation factors at 37 °C with 5% CO₂. Penicillin/streptomycin (100X) was added for the first 4 d of culture post-isolation only.

Small intestinal crypts were cultured as previously described²³. Briefly, crypts were resuspended in basal culture medium at a 1:1 ratio with Corning Matrigel membrane matrix–GFR (Thermo Fisher) and plated at the centre of each well of 24-well plates. Following Matrigel polymerization, 500 μl of crypt culture medium (ENR + CV) containing growth factors EGF (50 ng ml⁻¹, Life Technologies), Noggin (100 ng ml⁻¹, PeproTech) and R-spondin 1 (500 ng ml⁻¹, PeproTech) and small molecules CHIR99021 (3 μM, LC Laboratories or Selleck Chem) and valproic acid (1 mM, Sigma-Aldrich) was added to each well. ROCK inhibitor Y-27632 (10 μM, R&D Systems) was added for the first 2 d of ISC culture only. The cell culture medium was changed every other day. After 4 d of culture, crypt organoids were expanded as and enriched for ISCs under the ENR + CV condition. Expanding ISCs were passaged every 4–6 d in the ENR + CV condition.

After 2–6 d of culture under ENR + CV condition, ISCs were differentiated to Paneth cells. Briefly, ISC culture gel and medium were homogenized via mechanical disruption and centrifuged at 300 r.c.f. for 3 min at 4 °C. The supernatant was removed and the pellet resuspended in basal culture medium repeatedly until the cloudy Matrigel was almost gone. On the last repeat, the pellet was resuspended in basal culture medium, the number of organoids counted, and the suspension centrifuged at 100 r.c.f. for 1 min at 4 °C. The cell pellet was resuspended in basal culture medium at a 1:1 ratio with Matrigel and plated at the centre of each well of 24-well plates (~100–250 organoids per well). Following Matrigel polymerization, 500 μl of crypt culture medium (ENR + CV) was added to each well. The cell culture medium was changed every 2–4 d depending on seeding density.

Human crypt isolation and organoid culture

Excess surgical tissue samples from adult human duodenum were collected for organoid culture in accordance with Massachusetts General Hospital Institutional Review Board (IRB) guidance under Mass General Brigham Protocol 2010P000632. De-identified human donor tissue was collected following medically indicated bulk surgical resection via MGH Pathology as excess tissue. Donors of both sexes, aged between 58–74 years, presented with pathologies unrelated to the duodenum. Crypts were isolated from bulk resections as follows. Bulk resections were cut into approximately 0.25 cm sections from the epithelial surface, and washed in PBS0 repeatedly by gently pipetting the fragments using a 10 ml pipette until the supernatant was clear. Fragments were rocked on ice with crypt isolation buffer (10 mM EDTA, 10 mM HEPES, 2% FCS in PBS0) for 30 min. After isolation buffer was removed, fragments were washed with cold PBS0 by vigorous shaking to release the crypts. This process was repeated with reserved crypt-laden supernatant fractions 4–6 times or until supernatant was free of intact crypts (visual inspection). Crypt-containing fractions were combined, passed through a 100 μm cell strainer (BD Bioscience), and centrifuged at 300 r.c.f. for 5 min. The crypt pellet was resuspended in basal culture medium (2 mM GlutaMAX (Thermo Fisher) and 10 mM HEPES (Life Technologies) in advanced DMEM/F12 (Invitrogen)) and centrifuged at 200 r.c.f. for 2 min to remove single cells. Crypts were then cultured in a Matrigel culture system (described previously). Organoids were cultured and passaged as described for murine organoids every 6–8 d in Matrigel domes with established media conditions meant to recapitulate a stem cell-enriched condition⁴⁷. Organoid culture media contained recombinant EGF (Thermo Fisher), FGF2 (Thermo Fisher), IGF1 (Peprotech), Gastrin (Sigma Millipore) and TGF-b inhibitor A83-01 (Tocris Bioscience) with 50% conditioned medium of L-cell line secreting Wnt3a, R-spondin3 and Noggin (L-WRN CM) supplemented with 10 μmol l⁻¹ Y-27632 (Tocris Bioscience). L-WRN CM was prepared from L-WRN (ATCC; CRL-3276) as described previously⁶⁵. L-WRN (50%) is a 1:1 mixture of 100% L-WRN and primary culture medium. Primary culture media consist of advanced DMEM/F12, penicillin/streptomycin, GlutaMAX (all from Thermo Fisher), and FBS (20%). Organoid samples grown in culture over varying periods were either maintained and passaged or treated with 160 nM KPT-330 for 6 d before assay, with media changes every other day.

High-throughput screening

For 384-well plate high-throughput screening, ISC-enriched organoids were passaged and split into single cells with TyrpLE (Thermo Fisher) and cultured for 2–3 d in ENR + CVY (Y: Y-27632 at 10 μM) before transfer to a ‘2.5D’ 384-well plate culture system. To prepare for ‘2.5D’ plating, cell-laden Matrigel and media were homogenized via mechanical disruption and centrifuged at 300 r.c.f. for 3 min at 4 °C. The supernatant was removed and the pellet washed and spun in basal culture medium repeatedly until the cloudy Matrigel above the cell pellet was gone. On the final wash, the pellet was resuspended in basal culture medium, the number of organoids counted, and the cell pellet resuspended in ENR + CD medium at ~7 clusters per μl. Plates (384-well) were first filled with 10 μl 70% Matrigel (30% basal media) coating in each well using a Tecan Evo 150 liquid handling deck, and allowed to gel at 37 °C for 5 min. Then 30 μl of cell-laden media was plated at the centre of each well of 384-well plates with the liquid handler, and the plates were spun down at 100 r.c.f. for 2 min to embed organoids on the Matrigel surface. Compound libraries were pinned into prepped cell plates using 50 nl pins into 30 μl media per well. Cells were cultured at 37 °C with 5% CO₂ for 6 d in ENR + CD medium supplemented with the tested compounds, with media change at 3 d. On day 6, lysozyme secretion and cell viability were assessed using lysozyme assay kit (EnzChek) and CellTiter-Glo 3D (CTG 3D) cell viability assay (Promega), respectively, according to the manufacturers’ protocols. Briefly, screen plates were washed 3× with FluoroBrite basal media (2 mM GlutaMAX and 10 mM HEPES in FluoroBrite DMEM (Thermo Fisher)) using a BioTek 406 plate washer with 10 min incubations, followed by a 1 min centrifugation at 200 r.c.f. to settle media between washes. After removal of the third wash, 30 μl of non-stimulated FluoroBrite basal media was added to each screen well using a Tecan Evo 150 liquid handling deck from a non-stimulated treatment master plate, and plates were incubated for 30 min at 37 °C. After 30 min, the top 15 μl of media from each well of the screen plate was transferred to a non-stimulated LYZ assay plate containing 15 μl of 20X DQ LYZ assay working solution using a Tecan Evo 150 liquid handling deck. The non-stimulated LYZ assay plate was covered, shaken for 10 min, incubated for 50 min at 37 °C, then fluorescence measured (shaken for 10 s; 494 mm/518 nm) using a Tecan M1000 plate reader. After media transfer to the non-stimulated LYZ assay plate, the remaining media were removed from the screen plate and 30 μl of stimulated FluoroBrite basal media (supplemented with 10 μM CCh) was added to each screen well using a Tecan Evo 150 liquid handling deck from a stimulated treatment master plate, and plates were incubated for 30 min at 37 °C. After 30 min, the top 15 μl of media from each well of the screen plate was transferred to a stimulated LYZ assay plate containing 15 μl of 20X DQ LYZ assay working solution using a Tecan Evo 150 liquid handling deck. The stimulated LYZ assay plate was covered, shaken for 10 min, incubated for 50 min at 37 °C, then fluorescence measured (shaken for 10 s; 494 mm/518 nm) using a Tecan M1000 plate reader. Finally, 8 μl of CTG 3D was added to each well of the screen plate, which was shaken for 30 min at room temperature, then luminescence read (shaken for 10 s; integration time 0.5–1 s) to measure ATP.

Primary screens were performed using the Target-Selective Inhibitor Library (Selleck Chem). Assays were performed in triplicate using 4 compound concentrations (0.08, 0.4, 2 and 10 μM).

Screen analysis

Analysis of all screen results was performed in R. Results (excel or.csv files) were converted into a data frame containing raw assay measurements corresponding to metadata for plate position, treatments, doses, cell type and stimulation. Raw values were log₁₀ transformed, then a locally estimated scatterplot smoothing (LOESS) normalization was applied to each plate and assay to remove systematic error and column/row/edge effects using the formula:⁶⁶

where (widehat {x_{ij}}) is the LOESS fit result, (x_{ij}) is the log₁₀ transformed value at row i and column j, and (mathrm{loess.fit}_{ij}) is the value from LOESS smoothed data at row i and column j calculated using R loess function with span 1.

Following LOESS normalization, a plate-wise fold change (FC) calculation was performed on each well to normalize plates across the experiment. This was calculated by subtracting the median of the plate (as control) from the LOESS normalized values:

Replicate strictly standardized mean difference (SSMD) was used to determine the statistical effect size of each treatment in each assay (treatment and dose grouped by replicate, n = 3) relative to the plate, using the formula for the robust uniformly minimal variance unbiased estimate (UMVUE):⁶⁷

where (overline {d_i}) and s_i are respectively the sample mean and standard deviation of d_ijs where d_ij is the FC for the ith treatment on the jth plate. (Gamma ( cdot )) is a gamma function. (s_0^2) is an adjustment factor equal to the median of all (s_i^2)s to provide a more stable estimate of variance. w_i and w₀ are weights equal to 0.5 with the constraint of w_i + w₀ = 1. n is the replicate number.

Mean FC (the arithmetic mean of all samples grouped by treatment and dose across replicates) was used to determine the z-score for each treatment and dose with the formula:

where SD_pop is the standard deviation of all mean FCs.

All calculated statistics were combined in one finalized data table and exported as a.csv file for hit identification. A primary screen ‘hit’ was defined as having SSMDs for both LYZ assays greater than the optimal critical value ((beta _{alpha _1}) = 0.997) and being in the top 10% of a normal distribution of FC values for both assays with a z-score cutoff >1.282. (beta _{alpha _1}) was determined by minimizing the false positive (FPL) and false negative (FNL) levels for upregulation SSMD-based decisions by solving for the intersection of the formulas:⁶⁷

and

where

and (F_{tleft( {n – 1,sqrt n beta } right)}) is the cumulative distribution function of non-central t-distribution (t(n – 1,sqrt n beta )), n is the number of replicates, (beta _2) is an SSMD bound for FPL of 0.25 (at least very weak effect) and (beta _1) is an SSMD bound for FNL of 3 (at least strong effect).

Hit treatments were thus selected to have a well-powered statistical effect size as well as a strong biological effect size. Optimal dose per hit treatment was determined by SSMD for both LYZ assays.

Secondary lysozyme secretion assay screen

Confirmatory secondary screening with primary hits was performed using the above 384-well plate method. The screen was conducted with 4-plate replicates with a base media of ENR+CD. Media was supplemented with compound at day 0 and day 3 (n = 8 well replicates per dose) at 4 different doses: twofold above, twofold below and fourfold below the optimal final dose for each respective treatment. Additionally, each plate carried a large number of ENR+DMSO or ENR+CD+DMSO (vehicle) control wells (n = 100 for ATP, and n = 25 for LYZ.NS and LYZ.S) for robust normalization. ATP, non-stimulated lysozyme activity and CCh-stimulated lysozyme activity were again measured and the collected data were again processed in a custom R-script per primary screen, with slight modification. Values were log₁₀ transformed, and a plate-wise FC was calculated for each well on the basis of the median value of ENR+CD+DMSO (vehicle) control wells to normalize plate to plate variability. The following formula was used:

where (x_{ij}) is the log₁₀ transformed value at row i and column j, and (x_{mathrm{POS}}) are the values of the positive control wells. For the ATP assay, all vehicle-only wells were used as the control. For the LYZ.NS assay, non-stimulated vehicle-only wells were used. For the LYZ.S assay, vehicle-only wells that were non-stimulated in the LYZ.NS assay then stimulated in the LYZ.S assay were used.

Once normalized, the replicate SSMD was calculated using equation (3) to quantify statistical effect size, with 8 replicate differences taken relative to the respective plate ENR+DMSO or ENR+CD+DMSO median value. A primary hit was considered validated when SSMDs for both LYZ assays were greater than the optimal critical value ((beta _{alpha _1})) of 0.889. (beta _{alpha _1}) was determined using equation (5), with an FPL error of 0.05 for a more stringent cutoff; FNL was not considered. Optimal doses were chosen for treatments with multiple validated doses by taking the most potent (highest mean fold change relative to ENR+CD control) dose in both LYZ assays.

Lysozyme secretion assay

ISC-enriched organoids in 3D Matrigel culture were passaged to a 48- or 96-well plate and cultured with ENR or ENR+CD media containing DMSO or each drug for 6 d. DMSO- or drug-containing media were changed every other day. On day 6, cells were washed with basal media twice and treated with basal media with or without 10 μM carbachol for 3 h in a CO₂ incubator at 37 °C. Conditioned media was collected and used for lysozyme assay (Thermo Fisher, E-22013) following the manufacturer’s instruction. The fluorescence was measured using excitation/emission of 485/530 nm. CTG 3D Reagent was added afterward, and the cell culture plate was incubated on an orbital shaker at RT for 30 min to induce cell lysis and to stabilize the luminescent signal. The solution was replaced to a 96-well white microplate, and luminescent signals were measured by a microplate reader (infinite M200, Tecan). The standard curve was prepared by diluting recombinant ATP (Promega, P1132). For both assays, a polynomial cubic curve was fitted to a set of standard data, and each sample value was calculated on the Microsoft Excel.

Flow cytometry

ISC-enriched organoids in 3D Matrigel culture were passaged to a 24- or 48-well plate and induced to differentiate for 6 d by ENR+CD media containing DMSO or each drug indicated in the figures. DMSO- or drug-containing media were changed every other day. On day 6, cells were washed twice with basal media, then collected from Matrigel by mechanical disruption in TrypLE Express (Thermo Fisher, 12605010) to remove Matrigel and dissociate organoids to single cells. After vigorous pipetting and incubation at 37°C for 15 min, the cell solution was diluted twice with basal media and centrifuged at 300 r.c.f. for 3 min. The cell pellet was resuspended in FACS buffer (PBS containing 2% FBS) and replaced into a 96-well clear round-bottom ultra-low attachment microplate (Corning, 7007). The cell solution was centrifuged again at 300 r.c.f. for 3 min at 4 °C to pellet the cells. Cells were stained with Zombie-violet dye (BioLegend, 423113, 1:100) at 100X for viability staining for 20 min at r.t. in the dark. After centrifugation for 3 min at 300 r.c.f., cells were fixed in fixation buffer (FACS buffer containing 1% formaldehyde (Thermo Fisher, 28906)) for 15 min on ice in the dark. Cells were centrifuged again for 3 min at 300 r.c.f. and blocked with staining buffer (FACS buffer containing 0.5% Tween 20 (Sigma, P2287)) for 15 min at r.t. in the dark. Cells pelleted by centrifugation for 3 min at 300 r.c.f. were stained with FITC-conjugated anti-lysozyme antibody (Dako, F0372, 1:100) and APC-conjugated anti-CD24 antibody (Biolegend, 138505, 1:100) at 100X for 45 min at r.t. in the dark. The cell pellet was washed once with FACS buffer, resuspended in FACS buffer and filtered through a 5 ml test tube with cell strainer snap cap (Corning, 352235). Flow cytometry was performed using an LSR Fortessa (BD; Koch Institute Flow Cytometry Core at MIT). Flow cytometry data were analysed using FlowJo X v10.6.1 software.

Western blotting

Organoid-containing gel was homogenized in basal medium and centrifuged at 300 r.c.f. for 3 min. The organoid pellet was lysed with ice-cold Pierce IP lysis buffer (Thermo Fisher, 87787) containing EDTA-free Halt protease inhibitor cocktail (Thermo Fisher, 87785) and incubated on ice for 20 min. The lysate was centrifuged at 17,000 r.c.f. for 10 min, and the supernatant was combined with NuPAGE LDS sample buffer (Thermo Fisher, NP0007). Protein concentration was determined by Pierce 660 nm protein assay (Thermo Fisher, 22660) and normalized to the lowest concentration among each sample set. Samples were incubated at 70°C for 10 min and resolved by SDS–PAGE using NuPAGE 4–12% Bis-Tris protein gels (Thermo Fisher), followed by electroblotting onto Immun-Blot PVDF Membrane (Biorad, 1620174) using Criterion blotter with plate electrodes (Biorad, 1704070). The membranes were blocked with 2% blotting-grade blocker (Biorad,1706404) in TBS-T (25 mM Tris–HCl, 140 mM NaCl, 3 mM potassium chloride and 0.1% Tween 20) and then probed with appropriate antibodies, diluted in TBS-T containing 2% BSA (Sigma, A7906) and 0.05% sodium azide (Sigma, 71289). The primary antibody against lysozyme was purchased from Abcam (ab108508 1:2000). HRP-linked anti-rabbit IgG antibodies were purchased from Cell Signalling Technology (7074, 1:2,000). Chemiluminescent signals were detected by LAS4000 (GE Healthcare) using Amersham ECL Select western blotting detection reagent (GE Healthcare, 45-000-999), and total protein signals were obtained by Odyssey imaging system (LI-COR Biosciences) using REVERT total protein stain kit (LI-COR Biosciences, 926-11010).

Immunofluorescent imaging

For immunofluorescence staining of organoids, intestinal organoids in Matrigel were fixed with 4% paraformaldehyde, then transferred to centrifuge tubes. After washing with PBS, the isolated organoids were permeabilized with 1% Triton X, followed by incubation with blocking buffer (1% BSA + 3% Donkey Serum + 0.2% Triton X in PBS) at r.t. The organoids were then stained with primary antibodies and fluorescent dye-labelled secondary antibodies, as well as with 4′,6-diamidino-2-phenylindole (DAPI). Slides were covered with VECTASHIELD mounting media (VECTOR). The following primary and secondary antibodies were used for the staining: rabbit anti-lysozyme (Thermo Fisher, RB-372-A, 1:1,000), rat anti-E-cadherin (Thermo Fisher, 13-1900, 1:1,000), and Alexa Fluor 488 and 568 secondary antibodies (Thermo Fisher A21208, A10042, 1:1,000). Images were acquired with a confocal laser scanning microscope (Nikon Eclipse 90i) with the following acquisition settings: DAPI exposure time 2 ms, contrast gain 0; FITC (for E-cadherin) exposure time 39 ms, contrast gain 0; TRITC (for lysozyme) exposure time 30 ms, contrast gain 0. For the analysis of lysozyme+ cells per organoid area, the number of counted lysozyme+ cells were normalized to the measured organoid surface area. Fiji v2.0 was used for quantification of lysozyme+ cells.

Animal study

All animal studies were performed under animal protocols approved by the MIT CAC. Wild-type C57BL/6NCrl male mice (8–10-week-old, 027) were purchased from Charles River. Mice were housed under 12 h light/dark cycle, provided food and water ad libitum, and kept in a 20–22 °C and 30–70% humidity environment. KPT-330 (0.01, 0.05, 0.2 or 10 mg kg⁻¹) were injected orally using a disposable gavage needle (Cadence Science, 9921) at 10 μl g⁻¹ weight. KPT-330 was dissolved in DMSO initially and further diluted in sterile PBS containing Pluronic F-68 non-ionic surfactant (Gibco, 24040032) and polyvinylpyrrolidone (PVP, Alfa Aesar, A14315, average M.W. 58,000); the final concentration of DMSO is 2%, Pluronic 0.5% and PVP 0.5%. KPT-330 was administered every other day for 2 weeks, for a total of 7 injections (days 0, 2, 4, 6, 8, 10, 12), and mice were killed at day 14.

Histology

The SI was collected from mice and divided into three parts. Only proximal and distal SI were kept in PBS, and medial SI was discarded. Each SI was opened longitudinally and washed in PBS. SI was rolled using the Swiss-rolling technique and incubated in 10% neutral buffered formalin (VWR, 10790-714) for 24 h at r.t. Fixed tissues were embedded in paraffin and 4 μm sections were mounted on slides. For immunohistochemistry, slides were deparaffinized, antigen retrieved using heat-induced epitope retrieval at 97 °C for 20 min with citrate buffer pH 6, and probed with appropriate antibodies, followed by 3,3′-Diaminobenzidine (DAB) staining. An antibody against lysozyme was purchased from Abcam (ab108508, 1:2,000), Ki67 from BD Biosciences (550609 1:40) and Olfm4 from Cell Signalling Technology (39141, 1:1,000). For McManus periodic acid Schiff (PAS) reaction, slides were deparaffinized, oxidized in periodic acid and stained with Schiff reagent (Poly Scientific, s272), followed by counterstaining with Harris hematoxylin. Slides were scanned using an Aperio slide scanner (Leica) and cells were counted on an Aperio eSlide Manager. Slides were blinded and randomized before counting, and all cell types were counted in all well-preserved crypts along the longitudinal crypt–villus axis (Paneth cell, ≥30 crypts; Olfm4+ cell, ≥17 crypts; goblet cell, ≥15 villi, per sample). For the goblet cell images, the samples that included <15 well-preserved crypt–villus axes were excluded, which was predetermined.

Murine and human scRNA-seq and alignment

A single-cell suspension was obtained from murine organoids cultured under either ENR+CD or ENR+CD+160 nM KPT-330 for the differentiation time course as detailed in Fig. 2a, or human organoids treated with 160 nM KPT-330 as detailed in Fig. 6a. For both, organoids at each sampling were collected from 4–6 pooled Matrigel domes, totalling >1,000 organoids per sample. Excess Matrigel was removed per previously described washing protocol, and organoids were resuspended in TrypLE at 37 ºC for 15 min, with vigorous homogenization through a p200 pipette tip every 5 min. After 15 min, the suspension was passed through a 30 uM cell strainer twice and counted under bright-field microscopy with trypan blue staining for viable single cells. For human organoid scRNA-seq, antibody-based cell hashing was performed, with all samples pooled following labelling and three washes in FACS buffer to remove excess antibody. Each sample was manually counted to equally weight in cell pools, and then the pool was split and processed as four identical samples.

We utilized Seq-Well S³ for massively parallel scRNA-seq, for which full methods are published³⁶ and made available on the Shalek Lab website (www.shaleklab.com). Briefly, ~15,000–20,000 cells were loaded onto a functionalized-polydimethylsiloxane (PDMS) array preloaded with ~80,000 uniquely barcoded mRNA capture beads (Chemgenes; MACOSKO-2011-10). After cells had settled into wells, the array was then sealed with a hydroxylated polycarbonate membrane with pore size of 10 nm, facilitating buffer exchange while confining biological molecules within each well. Following membrane-sealing, buffer exchange across the membrane permitted cell lysis, mRNA transcript hybridization to beads and bead removal before proceeding with reverse transcription. The obtained bead-bound complementary DNA (cDNA) product then underwent Exonuclease I treatment (New England Biolabs; M0293M) to remove excess primer before proceeding with second strand synthesis.

Following Exonuclease I treatment, the beads were mixed with 0.1 M NaOH for 5 min at r.t. to denature the mRNA–cDNA hybrid product on the bead. Second strand synthesis was performed with a mastermix consisting of 40 ul 5x maxima RT buffer, 80 ul 30% PEG8000 solution, 20 ul 10 mM dNTPs, 2 ul 1 mM dn-SMART oligo, 5 ul Klenow Exo- and 53 ul DI ultrapure water, with the mastermix being added to the beads and incubated for 1 h at 37 °C with end-over-end rotation. After the second strand synthesis, PCR amplification was performed using KAPA HiFi PCR Mix (Kapa Biosystems, KK2602). Specifically, a 40 ul PCR Mastermix consisting of 25 ul KAPA 5X Mastermix, 0.4 ul 100 uM ISPCR oligo and 14.6 ul nuclease-free water was combined with 2,000 beads per reaction. Following PCR amplification, whole transcriptome products were isolated through two rounds of SPRI purification using Ampure Spri beads (Beckman Coulter) at both 0.6X and 0.8X volumetric ratio and quantified using a Qubit. For the antibody hashed human organoid samples, the first SPRI supernatant was retained and subjected to an additional SPRI at 2X final volumetric ratio and quantified using a Qubit. The hashing library then went through a round of step-up PCR to append sequencing handles and indices, followed by a final 1.6X volumetric ratio SPRI before final pooling with the mRNA library (below).

Sequencing libraries were constructed from whole transcriptome product using the Nextera Tagmentation method on a total of 800 pg of pooled cDNA library per sample. Tagmented and amplified sequences were purified through two rounds of SPRI purification (0.6X and 0.8X volumetric ratios) yielding library sizes with an average distribution of 500–750 base pairs in length as determined using the Agilent hsD1000 screen tape system (Agilent Genomics). Murine organoid arrays were sequenced within multi-sample pools on an Illumina NovaSeq through the Broad Institute walk-up sequencing core. Human organoid arrays were sequenced within multi-sample pools on an Illumina NextSeq 550 with a v2.5 high output kit (75 cycle). The read structure was paired end with Read 1 starting from a custom read 1 primer containing 20 bases with a 12 bp cell barcode and 8 bp UMI, and Read 2 being 50 bases containing transcript information. Sequencing read alignment was performed using version 2.1.0 of the Dropseq pipeline previously described⁶⁸. For each sequencing run, raw sequencing reads were converted from bcl files to FASTQs using bcl2fastq based on Nextera N700 indices that corresponded to individual samples. Demultiplexed FASTQs were then aligned to the mm10 (murine) or hg19 (human) genome using STAR and the DropSeq pipeline on a cloud-computing platform maintained by the Broad Institute. Individual reads were tagged with a 12 bp barcode and 8 bp UMI contained in Read 1 of each sequencing fragment. Following alignment, reads were grouped by the 12 bp cell barcodes and subsequently collapsed by the 8 bp UMI for digital gene expression (DGE) matrix extraction and generation. Cell hashing FASTQs were processed with CITE-seq-Count (v1.4.2, https://zenodo.org/record/2590196) to obtain UMI-collapsed hashing DGE matrices corresponding to the 6 antibody tags.

Murine scRNA-seq analysis

Before analysis, DGE matrices were pre-processed to remove cellular barcodes with <500 unique genes, >35% of UMIs corresponding to mitochondrial genes, low outliers in standardized house-keeping gene expression⁶⁹, >30,000 UMIs and cellular doublets identified through manual inspection and use of the DoubletFinder algorithm⁷⁰. These pre-processed DGEs are deposited as GEO GSE148524 and are available with interactive visualization tools, metadata and digital gene expression matrices at the Broad Institute’s Single-Cell Portal (https://singlecell.broadinstitute.org) as study SCP1547.

After quality and doublet correction, we performed integrated analysis on a combined dataset of 19,877 cells, with quality metrics for gene number, captured UMIs and percent mitochondrial genes reported in Extended Data Fig. 2. To better control for potential batch effects that may arise in sample handling and library preparation, dimensional reduction and clustering were performed following normalization with regularized negative bionomical regression as implemented in Seurat V3 via SCTransform⁷¹. We performed variable gene identification and dimensionality reduction utilizing the first 9 principal components based on the elbow method to identify 8 clusters using Louvain clustering (Resolution, 0.45). Following UMAP visualization, we used log-normalized RNA expression for all differential gene expression tests, gene-set enrichment analyses and gene module scoring. Of the 8 original clusters, a single cluster had mixed marker expression corresponding to the secretory goblet and Paneth lineages. Accordingly, we subsetted this cluster and performed variable gene selection and dimensional reduction (14 principal components), and identified 2 previously unreported clusters corresponding to goblet and early secretory cells by Louvain clustering (Resolution, 0.3), which were annotated accordingly in the full dataset. We identified genes enriched across clusters using the Wilcoxon rank sum test, with genes expressed in at least 20% of cells and with a minimum log fold change of 0.5 to identify generic cell types, and corroborated these cell type identities relative to gene signatures coming from an established murine small intestinal scRNA-seq atlas³⁷. Gene modules were scored within each cell on the basis of enrichment in gene set expression relative to randomly selected genes of comparable expression levels in each cell⁶⁹, via the AddModuleScore function within Seurat v3. In addition to cell-type module scoring from ref. ³⁷, we incorporated gene sets for ISC sub-typing from ref. ⁶, in addition to gene sets representing ISC activity⁴⁶ and genes known to contain NES from the ValidNESs database³⁹.

To quantify enrichments in cell populations between treatment and control within the murine dataset, we utilized Fisher’s exact test for each cell type relative to all others at each timepoint. We only considered populations for testing when that cell type accounted for at least 0.5% of cells in both KPT-330 and control samples. We present the relative enrichment or depletion of a cell population with KPT-330 treatment over time as the odds ratio with a corresponding 95% confidence interval, and false discovery rate (FDR)-adjusted P values with significance denoted as ‘*’s in corresponding figure legends.

To interrogate differences in signalling pathway activity between cell types and treatment conditions in the organoid differentiation experiment, we employed the PROGENy package³⁸ to infer pathway activity across the package’s 14 supported pathways. Pathway activity was inferred on a single-cell basis without permutation and the top 300 genes were used to generate the model matrix, which was appended as a Seurat object assay in accordance with the PROGENy tutorial for scRNA-seq (https://saezlab.github.io/progeny/articles/ProgenySingleCell.html). Pathway activity for the untreated populations is presented as scaled means of pathway activity for each cell type, while Cohen’s d was calculated between the single-cell distributions of KPT-330-treated and untreated cells.

To interrogate potential differences in upstream TF activity between cell types and treatment conditions of the organoid differentiation experiment, we employed the DoRothEA package³⁸ to infer upstream TF activity in each single cell. Upstream TF activity was inferred on a single-cell basis with the default murine regulon and a minimum of 10 targets per regulon, which was appended as a Seurat object assay in accordance with the DoRothEA tutorial for scRNA-seq (https://saezlab.github.io/dorothea/articles/single_cell_vignette.html). We performed dimensionality reduction on the full DoRothEA assay utilizing the first 7 principal components based on the elbow method to identify 7 clusters using Louvain clustering (Resolution, 0.45). Following UMAP visualization, we used the DoRothEA assay to perform differential upstream TF expression testing, identifying maker TFs for each cluster. To quantify enrichments in upstream TF clusters by cell type between treatment and control, we utilized Fisher’s exact test for each cell type relative to all others for each DoRothEA cluster. We only considered populations for testing when that cell type had at least 10 cells originating from both KPT-330 and control samples within that DoRothEA cluster. We present the relative enrichment or depletion of a cell population with KPT-330 treatment in each DoRothEA cluster as the odds ratio with a corresponding 95% confidence interval, and FDR-adjusted P values with significance denoted as ‘*’s in corresponding figure legends.

GSEA was performed on the full rank-ordered list of differentially expressed genes (without fold change or P-value cutoffs) using the piano R package⁷² and the MsigDB hallmark v7 gene sets^44,45. Gene sets with at least 25 and no more than 500 matching genes were considered, and only gene sets with an FDR-corrected P < 0.05 were retained.

Human scRNA-seq analysis

Before analysis, DGE matrices were pre-processed to remove cellular barcodes with <500 unique genes, >35% of UMIs corresponding to mitochondrial genes, low outliers in standardized house-keeping gene expression⁶⁹ and >30,000 UMIs. Antibody hashed arrays were demultiplexed with doublets and negative-staining cells removed following default settings of the Seurat function HTODemux. These pre-processed DGEs are deposited in the Broad Institute Single Cell Portal (https://singlecell.broadinstitute.org) as study SCP1318.

We performed integrated analysis on a combined dataset of 2,484 cells, with quality metrics for gene number, captured UMIs and percent mitochondrial genes reported in Extended Data Fig. 6. Dimensional reduction and clustering were performed following normalization in Seurat V3 via SCTransform⁷¹. We performed variable gene identification and dimensionality reduction utilizing the first 18 principal components based on the elbow method to identify 7 clusters using Louvain clustering (Resolution, 0.5). Following UMAP visualization, we used log-normalized RNA expression for all differential gene expression tests, gene-set enrichment analyses and gene module scoring. Of the 7 original clusters, a single cluster had mixed marker expression corresponding to the secretory goblet and enteroendocrine lineages. Accordingly, we subsetted this cluster and performed variable gene selection and dimensional reduction (8 principal components), and identified 2 previously unreported clusters corresponding to goblet and enteroendocrine cells by Louvain clustering (Resolution, 0.3), which were annotated accordingly in the full dataset. We identified genes enriched across clusters using the Wilcoxon rank sum test, with genes expressed in at least 10% of cells and with a minimum log fold change of 0.25 to identify cell types, and corroborated these cell type identities relative to known gene markers. Gene modules were scored within each cell on the basis of enrichment in gene set expression relative to randomly selected genes of comparable expression levels in each cell⁶⁹, via the AddModuleScore function within Seurat v3 (for genes known to contain NES, from the ValidNESs database³⁹).

To quantify enrichments in cell populations between treatment and control within the human dataset, we utilized Fisher’s exact test for each cell type relative to all others by donor. We only considered populations for testing when that cell type had at least 1 cell in both KPT-330 and control samples. We present the relative enrichment or depletion of a cell population with KPT-330 treatment over time as the odds ratio with a corresponding 95% confidence interval, and FDR-adjusted P values with significance denoted as ‘*’s in corresponding figure legends.

To interrogate differences in signalling pathway activity between cell types and treatment conditions in the human organoid experiment, we employed the PROGENy package³⁸ to infer pathway activity across the package’s 14 supported pathways. Pathway activity was inferred on a single-cell basis without permutation and the top 300 genes were used to generate the model matrix, which was appended as a Seurat object assay in accordance with the PROGENy tutorial for scRNA-seq (https://saezlab.github.io/progeny/articles/ProgenySingleCell.html). Pathway activity for the untreated populations is presented as scaled means of pathway activity for each cell type.

Reporting Summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.