Germ‐free and microbiota‐associated mice yield small intestinal epithelial organoids with equivalent and robust transcriptome/proteome expression phenotypes

Abstract Intestinal epithelial organoids established from gut tissue have become a widely used research tool. However, it remains unclear how environmental cues, divergent microbiota composition and other sources of variation before, during and after establishment confound organoid properties, and how these properties relate to the original tissue. While environmental influences cannot be easily addressed in human organoids, mice offer a controlled assay‐system. Here, we probed the effect of donor microbiota differences, previously identified as a confounding factor in murine in vivo studies, on organoids. We analysed the proteomes and transcriptomes of primary organoid cultures established from two colonised and one germ‐free mouse colony of C57BL/6J genetic background, and compared them to their tissue of origin and commonly used cell lines. While an imprint of microbiota‐exposure was observed on the proteome of epithelial samples, the long‐term global impact of donor microbiota on organoid expression patterns was negligible. Instead, stochastic culture‐to‐culture differences accounted for a moderate variability between independently established organoids. Integration of transcriptome and proteome datasets revealed an organoid‐typic expression signature comprising 14 transcripts and 10 proteins that distinguished organoids across all donors from murine epithelial cell lines and fibroblasts and closely mimicked expression patterns in the gut epithelium. This included the inflammasome components ASC, Naip1‐6, Nlrc4 and Caspase‐1, which were highly expressed in all organoids compared to the reference cell line m‐ICc12 or mouse embryonic fibroblasts. Taken together, these results reveal that the donor microbiota has little effect on the organoid phenotype and suggest that organoids represent a more suitable culture model than immortalised cell lines, in particular for studies of intestinal epithelial inflammasomes.


| INTRODUCTION
Epithelia constitute essential barriers that protect the inner organs of the body, facilitate uptake and secretion and coordinate immune responses (Allaire et al., 2018). Consequently, their biology has received significant attention. Due to the difficulty of keeping primary epithelial cells in culture, mechanistic studies of epithelial cell biology and physiology have traditionally relied on epithelial cell linestransformed cultures established from carcinomas or produced by introducing oncogenes (Bens et al., 1996;Fogh & Trempe, 1975;Fogh, Wright, & Loveless, 1977;Scherer, Syverton, & Gey, 1953). Cell lines are easy to grow, can be maintained in culture indefinitely and allow flexible genetic and pharmacological manipulation. However, the transferability of results to the in vivo scenario is often limited (Antoni, Burckel, Josset, & Noel, 2015;Ben-David et al., 2018;Niepel et al., 2019). This can be explained by poor mimicking of the complexity and interconnectedness inherent to epithelia in vivo (Antoni et al., 2015), the disruptive effects of cellular transformation and the gradual accumulation of genetic anomalies during prolonged culture (Ben-David et al., 2018;Foulke-Abel et al., 2014;Liu et al., 2019). Therefore, new, more stable and possibly more representative experimental models are needed.
Gut epithelial organoids offer an attractive alternative. Protocols for culturing and differentiation of primary blood-derived cell types have existed for decades (Sallusto & Lanzavecchia, 1994;Stone & Takemoto, 1970). More recently, the cumulative knowledge of the signals that maintain stem cells, drive epithelial cell growth and promote differentiation has allowed analogous protocols to be developed for epithelia from humans and mice. This progress has been driven by studies of the gut stem cell niche (Sato et al., 2009;Sato & Clevers, 2013;Stappenbeck & Virgin, 2016). Embedding of extracted intestinal epithelial stem cells in a matrix overlaid with a growth factor-enriched culture medium (containing, e.g., Wnt, Noggin, EGF, R-spondin; Sato et al., 2011;Sato & Clevers, 2013) results in the outgrowth of threedimensional primary epithelial structures-so called intestinal epithelial organoids (hereafter simply referred to as "organoids"). These organoids comprise a single layer of epithelial cells, with their apical side oriented towards a central lumen, while the basal side faces the extracellular matrix. In further similarity to the intact gut, organoids feature crypt invaginations harbouring the stem cell compartment (Sato & Clevers, 2013). These stem cells divide and give rise to epithelial cell precursors, which differentiate into paneth cells, enteroendocrine cells, goblet cells and enterocytes, hence recapitulating much of the complexity of co-existing cell types in the gut mucosa (Foulke-Abel et al., 2014;Sato et al., 2011). For these reasons, organoids have since their conception become a widely used and realistic model to study the role of intestinal epithelial cells in, for example, gut physiology (Almeqdadi, Mana, Roper, & Yilmaz, 2019;Gunasekara et al., 2018;Williamson et al., 2018), cancer biology (Drost et al., 2015;Tuveson & Clevers, 2019), pharmacology (Takahashi, 2019;Walsh, Cook, Sanders, Arteaga, & Skala, 2016) and infectious disease (Co et al., 2019;Foulke-Abel et al., 2014; Hausmann & Hardt, 2019; Sun, 2017; Zhang, Wu, Xia, & Sun, 2014).
It appears conceivable that organoids over time will replace traditional cell lines as the main tissue culture model of choice for mechanistic studies. However, it has remained unclear if the donor gut environment, in particular microbiota exposure, affects the organoid phenotype. These factors are difficult to control in organoids derived from human donors. To address the influence of the donor microbiota on organoid cultures, we have here compared organoids from wellcontrolled colonies of genetically identical mice, either germ-free or colonised with two different microbiotas.
Especially in the fields of gut inflammation and infection biology, the necessity for littermate-controlled in vivo experiments to normalise for such microbiota effects has become pressingly evident (Mamantopoulos et al., 2018).
The stem cell-containing crypts that make up the starting-material for intestinal epithelial organoids derive directly from this complex in vivo niche (Sato & Clevers, 2013). This raises the question whether environmental/microbial stimuli within the donor animal impact the long-term phenotype of established organoid cultures, for example, by epigenetic mechanisms (Foster & Medzhitov, 2009) and whether experiments in genetically modified murine organoids require wildtype littermate-derived control organoid cultures. Moreover, the organoid establishment procedure itself might impose bottlenecks and promote drifts between independently generated cultures that could affect the long-term phenotype. Hence, the impact of in vivo environmental factors, the amplitude and causes of organoid culture variability and the possible implications for experimental reproducibility remain poorly understood. This complicates the interpretation and comparability of results obtained in this emerging tissue culture model.
To assess reproducibility, faithful recapitulation of responses to relevant biological stimuli and stability towards confounding factors, we generated multiple independent organoid cultures from intestinal epithelial crypts of genetically identical mice housed in two distinct specific pathogen-free (SPF) facilities and one germ-free (GF) facility.
By combining proteomics and transcriptomics, we compared the global expression profiles of the organoid cultures among each other, to their tissue of origin, and to widely used epithelial cell line and fibroblast models. Strikingly, organoids established from germ-free or colonised mice exhibited basal expression profiles that co-cluster together, rather than forming separate subgroups. Instead, the modest variability in expression between organoid cultures could be traced to stochastic sources during establishment and in-culture maintenance.
Also, the specific expression program induced by a defined stimuluslow-dose TNF-appeared similar between organoid cultures from germ-free and colonised animals, but differed markedly from TNFinduced changes in a transformed intestinal epithelial cell line. Finally, our work uncovered an organoid expression signature that highlights significant expression of inflammasome signalling components in the primary intestinal epithelium, which is not detectable in commonly used cell lines.
2 | RESULTS 2.1 | Proteome profiles of independently established organoid cultures reveal a limited impact of the donor's microbiota A tissue culture model should ideally exhibit limited variability and recapitulate the properties of the corresponding in vivo tissue. We have focused on murine intestinal epithelial organoids, as these provide an easily accessible system which allows precise control for impacts of the microbiota and the genetic background of the host, in contrast to human material. Using this system, we assessed reproducibility from genetically identical animals with the same life history, reared in the presence or the absence of a microbiota. Proteins carry most cellular functions and are tightly associated to specific phenotypes (Aebersold & Mann, 2016). Thus, as a start, we used proteome profiling to systematically probe the main sources of variability among intestinal epithelial organoid cultures and to address the relatedness of organoids to the gut epithelium.
To assess the effects of different microbiota exposures, we chose C57BL/6J wild-type mice which were bred in parallel for >2 years (>10 generations) in two separate SPF facilities featuring two different microbiotas (SR and SE), and one germ-free facility (GF). Organoid cultures were established from the jejunum of three 8-12 weeks old cohoused male littermate mice from each facility. During organoid establishment, samples corresponding to whole intestinal tissue (distal jejunum; contains epithelium, lamina propria and submucosa) (Tissue) and the isolated epithelial fraction (Epithelium) were also collected ( Figure S1, Supporting Information). To avoid batch-to-batch medium variation, all organoid cultures were maintained using commercially available reagents (see section 4) purchased in bulk. Organoid cultures were grown to purity, cryopreserved in liquid nitrogen, revived and grown to passage 5-8 before sample collection (Organoid, see section 4). This sample set allowed us to probe the relatedness between primary intestinal epithelial cells and the corresponding organoids, and to assess the source(s) of inter-sample variability in the absence of genetic diversity.
As reference samples, we employed an immortalised murine small intestinal epithelial cell line (m-IC c12 ; Bens et al., 1996). Mouse embryonic fibroblasts (MEF; C57BL/6 mesodermal origin) were chosen as an outgroup representing primary cells from a different mouse organ. Using SWATH mass spectrometry (SWATH-MS; Aebersold & Mann, 2016;Gillet et al., 2012;Liu et al., 2015Liu et al., , 2019Williams et al., 2016), a proteomic data acquisition method that generates highly reproducible datasets between multiple samples, randomised sample processing and downstream analysis in OpenSWATH (Röst et al., 2014(Röst et al., , 2016, we were able to reproducibly quantify 3,653 Swissprot murine proteins (i.e., 3,331 unique proteins matching to the transcriptomics data below) across the entire sample set. Analysis of technical SWATH-MS replicates confirmed a minimal variability stemming from the proteomics procedure itself (average Pearson correlation between technical replicates: 0.999).
Input from luminal microbiota, ingested chemicals and food particles may have profound effects on epithelial cell physiology and may imprint long-lasting characteristics, for example, by epigenetic processes (Allaire et al., 2018;Lotz et al., 2006;Pan et al., 2018). We To gauge the level of experimental noise, we next assessed the variability in protein expression between replicates within each sample group (measured as dispersion coefficient; i.e., standard deviation divided by the mean, in percent). As expected, the two reference cell lines (m-IC c12 and MEF) displayed a low variability between replicate samples (disp. coeff. of 13.27% and 11.77%, respectively;  F I G U R E 1 Donor microbiota minimally impacts the global protein expression pattern of small intestinal epithelial organoid cultures. (a) Unsupervised hierarchical clustering analysis of the proteome data set including tissue (Tissue_I-III, star symbol), epithelial cellenriched fraction (Epithelium_I-III, circle symbol) and organoid (Organoid_I-III, hexagon symbol) samples from mice raised in SPF facility 1 (_SR, dark green), SPF facility 2 (_SE, light green) or the germ-free facility (_GF, yellow), as well as MEF (MEF_I-III, blue triangle symbol) and m-IC c12 cell (m-IC c12 _I-III, red square symbol) samples. Correlation matrix depicts Pearson correlation values between indicated samples. (b) Principal component analysis of the proteome data set as described in (a) 19.70%) sample groups. By comparison, the variability within the organoid sample group was lower than within the epithelial and tissue sample groups (disp. coeff. 16.12%; Table 1). Two of the organoid subgroups even displayed a variability close to the one of the m-IC c12 sample group (disp. coeff. Organoid_SR 13.58%; Organoid_SE 17.08%; Organoid_GF 14.93%; Tables 1 and 2). Considering that each organoid sample stems from a unique establishment, cryopreservation, revival and~3-4 additional weeks in separate culture, this degree of variability can be considered modest, and close to the variability noted for homogenous cell line cultures (m-IC c12 and MEF).
Moreover, the variability within the tissue and epithelium samples is higher than within the organoid sample group, indicating that environmental cues influencing mouse-to-mouse variations may be partially eliminated in culture.
Taken together, we conclude (a) that murine small intestinal epithelial organoids exhibit a distinct proteome profile; (b) which resembles that of the in vivo epithelium more closely than an immortalised epithelial cell line; (c) that in vivo environmental factors including previous exposure to microbiota in the murine gut have a negligible impact on the global proteome of organoids; and (d) that the intersample variability between independent organoid cultures is only modestly higher than for commonly used cell lines.

| Contrasting stochastic organoid culture variation to the impact of a physiological stimulus
In a next step, we sought to contrast the stochastic variation between independently established organoid cultures to the impact of a subtle physiological stimulus. For this purpose, we stimulated organoids with the cytokine tumour necrosis factor (TNF) (5 ng/mL, 8 hr), known to induce a defined gene expression program in epithelial cells. Notably, we chose a low TNF concentration which would induce a distinct proinflammatory response rather than cell death (Janes et al., 2006;Vlantis et al., 2016). This treatment led to the significant up-or downregulation of 15 proteins in the organoid sample group, including upregulation of typical marker proteins such as Nfkb2 (Mukherjee et al., 2017; Figure S3A and Table S1).
Again, the global unsupervised clustering was used to assess the relative impact of stochastic culture-to-culture variability and TNF-  Table S2). Among these are previously described TNF-target genes, including Nfkb2, Tnfaip3, C3 and Relb (Mukherjee et al., 2017;Sheerin, Zhou, Adler, & Sacks, 1997;Vlantis et al., 2016;Zhao et al., 2015). Organoid_SE_I_T; Figure 2c). This is well in line with the subtle, physiological nature of the TNF stimulus employed in our experiment, in analogy to a typical specific biochemical perturbation, which is expected to affect only a very small set of selected genes in epithelial cells (Janes et al., 2006;Vlantis et al., 2016). Similar conclusions could be drawn both at the proteome and transcriptome level when the analysis was redone for the 100 proteins/transcripts contributing most to variation ( Figure S4a-

| Robust induction of a TNF-induced gene expression program in organoids from differentially colonised mice
The data above reveal a modestly elevated variability in baseline organoid gene expression, as compared to cultured cell lines (Figures 1 and 2). For an experimental model system to be useful, another key aspect is the ability to respond reproducibly to a given stimulus. To assess this, we estimated the similarity in specific gene (d) different origin, that is, from the germ-free facility (mean of Organ-oid_GF) and the two SPF facilities (means of Organoid_SR and Orga-noid_SE; "Organoid_SPF"). This analysis revealed that the organoids derived from germ-free mice responded to TNF with a robust degree of similarity to those derived from SPF mice (R 2 = .563; Figure 3a).
This implies that neither prior in vivo stem cell exposure/nonexposure to gut microbes, nor variability in the organoid production process, imprint differences that may preclude interpretation of the small intestinal organoid responses to the prototypical stimulus TNF.
Traditionally To test if the identified organoid signature agreed with expression levels in the gut epithelium, we reassessed expression of the 10 protein hits (Figure 4) in the entire proteome data set, that is, including also the tissue and epithelium sample groups ( Figure S6). Strikingly, these identifier proteins showed highly similar expression levels in the epithelium samples and in organoids ( Figure S6)  passages. It is likely that microbiota imprints are detectable in earlier passages (Janeckova et al., 2019).
The observed culture-to-culture variation in expression profiles may stem from bottleneck effects during early organoid establishment and/or adaptation to the culture conditions. In addition, the differentiation state and cell type composition of organoids is highly sensitive to the concentration of growth factors provided in the culture medium (e.g., Noggin, R-spondin and EGF), or produced by the organoids themselves (e.g., Wnt3a; Farin et al., 2016;Kim et al., 2014;Lehmann et al., 2019;Lindeboom et al., 2018;Sato et al., 2009;van der Flier & Clevers, 2009). Fluctuation of these, often unstable, proteinaceous factors provides another plausible source of culture-toculture variability. While the exact impact of these and potentially other causes remain to be examined, the net effect is a moderately this study. We cannot exclude that persisting microbiota effects would be more pronounced in colon organoids, due to higher microbial exposure within this gut region in vivo. Nor do our data refute that some specific organoid signalling pathways can be affected by the tissue donor's microbial status, especially during early culture passages (Janeckova et al., 2019) or in cases of pathobiont exposure.
Human inter-individual variation also exceeds that of genetically inbred animals, which has repercussions for experimental design in patient-derived organoids (Cristobal et al., 2017).
By integration of the transcriptome and the proteome datasets, we were able to identify a physiologically relevant organoid-typic expression signature, distinguishing the full set of organoid cultures across all three SPF/GF mouse colonies from the reference cell lines. Interestingly, this signature highlighted high expression of ASC, a central scaffolding protein for inflammasome signalling pathways (Richards et al., 2001). Our follow-up analysis extended this finding to also encompass the transcripts for a range of other inflammasome receptors (e.g., Naip1, 2, 5, 6, Nlrc4), inflammatory caspases (e.g., Caspase-1) and downstream executors (e.g., GsdmD), which all exhibited high expression in organoids and low to undetectable expression in epithelial cell line m-IC c12 and fibroblast reference cells. Importantly, high expression of these inflammasome components in epithelial cells were reported previously (Hausmann, Sellin, & Hardt, 2020;Winsor, Krustev, Bruce, Philpott, & Girardin, 2019), further indicating that organoids more realistically represent the in vivo situation. The differential regulation of inflammasome components upon exposure to the proinflammatory cytokine TNF likely represents a preparation of epithelial cells to microbial exposure. Upon sensing of PAMPs or DAMPs, inflammasomes drive acute pro-inflammatory and anti-microbial responses (Broz & Dixit, 2016). However, earlier studies have also implicated, for example, ASC, NAIPs, NLRP3 and NLRP6 as tumour suppressors (Allam et al., 2015;Allen et al., 2010;Das et al., 2006;Normand et al., 2011). A hallmark feature of inflammasome activation is the prompt induction of cell death machinery in the activated cell (Aglietti et al., 2016;Kayagaki et al., 2011;Knodler et al., 2010Knodler et al., , 2014Miao et al., 2010;Rauch et al., 2017;Richards et al., 2001;Sellin et al., 2014;Shi et al., 2015). It therefore seems conceivable that upon transformation/immortalization of epithelial cell lines, there would be a strong selective pressure to lose or downregulate inflammasome pathway components and thereby dampen cell death effects. By contrast, organoids grown under optimal conditions retain expression also of such potential tumour suppressor genes. Notably, with regard to the widely discussed reciprocal interactions between microbiota and inflammasomes in the gut (Mamantopoulos et al., 2017;Robertson et al., 2019;Seo et al., 2015;Winsor et al., 2019), the expression of inflammasome components appears unaffected by the donor microbiota in small intestinal epithelial organoids.
Thus, compared to classical tissue culture cell lines, organoids should be more realistic models to study the function of epithelial inflammasomes.
While the impact and mechanisms of inflammasome signalling in typical immune cells (e.g., macrophages, dendritic cells) have been thoroughly documented (Boyden & Dietrich, 2006;Franchi et al., 2006;Mariathasan et al., 2004;Martinon, Pétrilli, Mayor, Tardivel, & Tschopp, 2006;Miao et al., 2006), the importance of intestinal epithelial inflammasomes in tissue homeostasis and defence has become evident only recently (Allam et al., 2015;Harrison et al., 2015;Knodler et al., 2010Knodler et al., , 2014Nowarski et al., 2015;Rauch et al., 2017;Sellin et al., 2014;Winsor et al., 2019). Tumour-derived or immortalised cell lines have traditionally been used as proxies for molecular studies in intestinal epithelia, which may in part explain why intestinal epithelial inflammasomes have for long been overlooked. We anticipate that the transition into primary organoids as the tissue culture models of choice will reshape our understanding of these and other physiological signalling circuits in the gut mucosa and beyond.

| RNA sequencing
Reads were quality-checked with FastQC. Sequencing adapters were removed with Trimmomatic (Bolger, Lohse, & Usadel, 2014) and reads were hard-trimmed by 5 bases at the 3 0 end. Successively, reads at least 20 bases long, and with an overall average phred quality score greater F I G U R E 5 Inflammasome components are highly expressed in organoids compared to m-IC c12 cells and fibroblasts. Heat map depicting expression levels of several inflammasome components in untreated or TNF-treated (_T) organoids (Organoid_I-III, hexagon symbol) samples from mice raised in SPF facility 1 (_SR, dark green), SPF facility 2 (_SE, light green) or the germ-free facility (_GF, yellow), as well as untreated or TNF-treated (_T) m-IC c12 cell (m-IC c12 _I-III, red square symbol) and MEF (MEF_I-III, blue triangle symbol) samples than 10 were aligned to the reference genome and transcriptome of Mus musculus (FASTA and GTF files, respectively, downloaded from GRCm38, Release 91) with STAR v2.5.1 (Dobin et al., 2013) with default settings for single end reads. Distribution of the reads across genomic isoform expression was quantified using the R package GenomicRanges (Lawrence et al., 2013) from Bioconductor Version 3.0.

| Differential expression
Differentially expressed genes and proteins were identified using the R package edgeR (Robinson, McCarthy, & Smyth, 2010)

| Transcriptomics-proteomics integration
The integration of the transcriptomics and the proteomics data was performed using the DIABLO framework (Singh et al., 2019) from the CRAN package mixOmics (Rohart et al., 2017).

| Proteome analysis
All the biological samples were suspended in 10 M Urea with complete protease inhibitor cocktail (Roche) and ultrasonically lysed in a VialTweeter device (Hielscher-Ultrasound Technology), as previously described (Collins et al., 2017;Liu et al., 2019). The mixtures were centrifuged at 21,000 g for 1 hr and the supernatant protein amount was quantified by Bio-Rad protein assay. Protein samples were reduced by 10 mM Tris-(2-carboxyethyl)-phosphine (TCEP) for 1 hr at 37 C and 20 mM iodoacetamide (IAA) in the dark for 45 min at room temperature. All the samples were further diluted by 1:6 (v/v) with 100 mM NH4HCO3 and were digested with sequencing-grade porcine trypsin (Promega) at a protease/protein ratio of 1:25 overnight at 37 C. The amount of the purified peptides was determined using Nanodrop ND-1000 (Thermo Scientific) and 1.5 μg peptides were injected per a LC-MS run. The peptide samples were stored in −80 C before measurement.
An SCIEX 5600 TripleTOF mass spectrometer was interfaced with an Eksigent NanoLC. Peptides were directly injected onto a 20 cm PicoFrit emitter (New Objective, self-packed to 20 cm), and then separated using a 90 min gradient at a flow rate of 300 nL/min (Gillet et al., 2012;Liu et al., 2019). For shotgun sequencing mode, MS1 spectra were collected in the range 360-1,460 m/z with 250 ms per scan. The 20 most intense precursors triggered MS2 spectra were collected (50-2,000 m/z for 100 ms). For SWATH mode, 64-variable window schema was used (Collins et al., 2017;Liu et al., 2019;Ludwig et al., 2018). A dwell time of 50 ms was used for all MS2 scans after a survey MS1 scan of 250 ms, resulting in a duty cycle of~3.45 s.
OpenSWATH (Röst et al., 2014) was used to identify peptides from all SWATH maps with statistical control at 1% FDR and then to align between SWATH maps using a novel TRIC with requantification option enabled (TRansfer of Identification Confidence; Röst et al., 2016). Because we had nine samples for tissue and epithelium type of samples analysed in this dataset (18 samples for organoid type), to further increase the protein confidence, only those peptide signals identified in more than eight samples were accepted for protein identification and quantification. To quantify the protein abundance levels across samples, we used the Top3 method (Grossmann et al., 2010;Liu et al., 2015;Ludwig, Claassen, Schmidt, & Aebersold, 2012;Williams et al., 2016). The quantitative protein matrix was rounded to the full integer value for further analysis. The input for the downstream analysis of the protein expression data was the matrix obtained by applying TMM-normalisation to the raw count matrix.
The proteome data set was uploaded to the PRIDE database (project ID PXD016339).

ACKNOWLEDGMENTS
We thank the Hardt, Sellin and Aebersold labs for discussions and input, and Miguel Cuenca, Shinichi Sunagawa, the Functional Geno-  (Edgar, Domrachev, & Lash, 2002) with the dataset identifier GSE140703.

CONFLICT OF INTEREST
The authors declare no potential conflict of interest.