Functional, patient-derived 3D tri-culture models of the uterine wall in a microfluidic array

The human uterus is a central organ in a woman’s reproductive system, responsible for fundamental physiological processes including menstruation, pregnancy, and childbirth. The uterus comprises an inner mucosal lining (endometrium), a smooth muscle layer (myometrium), and outer serosa. The endometrium is the site of embryo implantation and pregnancy establishment, whilst the myometrium functions both as a protective shell for the foetus before labour and provides the muscular contractions to expel the pregnancy during labour ( Hertelendy and Zakar, 2004 ; Tempest et al. , 2022 ). The endometrium is arranged in two sub-luminal layers composed primarily of epithelial glands and stromal cells; the superficial functionalis and underlying basalis. Both the lumen and functionalis are shed during menses and are regenerated during the proliferative (follicular) and secretory (luteal) phases in response to ovarian sex hormones ( Tempest et al. , 2022 ; Maclean et al. , 2020a ). The endometrial–myometrial junction (EMJ) is the junctional zone between the mucosal and smooth muscle zones of the uterus. The smooth muscle layer directly adjacent to the endometrium, the inner myometrium, is physiologically distinct from the outer myometrium; contractions originating from the inner myometrium regulate uterine peristalsis in a cyclical rhythm orchestrated by oestrogen and progesterone ( Naftalin and Jurkovic, 2009 ). The EMJ is implicit in physiological preservation of basalis endometrium and in several pathologies, including adenomyosis and placenta accreta disorders ( Tempest et al. , 2022 ). The endometrial basalis harbours stem/progenitor cells responsible for iterative endometrial regeneration, whilst the inner myometrium/EMJ is proposed to maintain basalis cells quiescent by being less responsive to ovarian hormonal signals ( Kamal et al., 2016 ; Maclean et al. , 2020b ). Adenomyosis occurs when endometrial tissue is present deep within the myometrium, likely via invagination of the endometrial basalis through the EMJ. This process may be initiated by both physical and physiological trauma to the EMJ ( Kunz et al. , 2005 ; Leyendecker and Wildt, 2011 ; Hao et al. , 2020 ; Yamaguchi et al. , 2021 ). The placenta accreta spectrum describes a range of pathologies characterized by abnormal invasion of the placenta into the myometrium. This invasion is patently linked to processes that disrupt the EMJ, such as Caesarean section (CS) ( Jansen et al. , 2020 ). Three-dimensional (3D) in vitro models better recapitulate structural organization and functional cellular interactions than traditional two-dimensional (2D) cultures, and a variety of 3D uterine models exist. Most, however, incorporate endometrial cells alone, in many cases using cell lines rather than patient-derived cells ( Gołąbek-Grenda and Olejnik, 2022 ) and were often developed as models of implantation ( Rawlings et al. , 2021 ; Li et al. , 2022 ) or for studies into gynaecological cancers ( Park et al. , 2021 ; Yang et al. , 2021 ). Far fewer studies have reported on the development of 3D myometrial models ( Vidimar et al , 2018 ; Xie et al , 2018 ). Lü et al (2009) previously reported on a three-layered rabbit uterine tissue model and Kuperman et al (2020) have recently described a tri-culture uterine wall macro-model based on cell lines. However, despite the significance of the EMJ in uterine physiology and prevalent pathologies, no patient-derived in vitro models containing all key EMJ cellular participants (i.e. epithelia, stroma and myocytes) currently exist. The development of myometrial-endometrial tri-cultures from small quantities of uterine tissue are urgently required to mimic the uterine wall with a patient-specific focus, facilitating the study of physiological and disease mechanisms. Miniaturization of in vitro models using microfluidic, organ-on-a-chip (OOAC) technologies offers significant opportunities ( Campo et al. , 2020 ) to better mimic the in vivo microenvironment and to scale-up the screening of limited but highly relevant patient-derived models. Such OOAC systems have recently been utilized to generate endometrium models ( Gnecco et al. , 2019 ; Campo et al. , 2020 ), including a 3D model of vascularized endometrium that incorporated human epithelial, stromal and endothelial cell lines within a fibrin hydrogel ( Ahn et al. , 2021 ). More complex systems have also been established, including a physio-mimetic model of the menstrual cycle comprising a multi-chamber platform containing cells of the ovaries, oviducts, cervix and endometrium ( Xiao et al. , 2017 ). However, none of these systems were used to model a 3D uterine wall by incorporating myometrial cells and, importantly, did not take advantage of the OOAC miniaturization and multiplexing capabilities, which are key to maximizing the utility of patient tissue when developing screening and functional assays using patient-derived cells. In this study, we have validated for the first time a 3D, premenopausal, patient-derived tri-culture model of the uterine wall, sequentially incorporating endometrial epithelial and stromal cells, and myometrial cells (the three principal cell types of the EMJ). The 3D models, which were cultured for up to 13 days, were created within arrays of microwells embedded within a multi-channel microfluidic device, allowing multiple miniaturized assays to be performed in a single experiment for investigating structural and functional phenotype. Here, we present this new methodology and describe its optimization, the characterization of cellular organization, and subsequent functional assays, including hormonal stimulation to elicit a decidualization response. The 3D multicellular cultures derived from patient biopsies offer an effective route to better mimic the biological complexity of uterine tissue, and our model provides a new tool for understanding cellular behaviours in uterine pathologies. Furthermore, with substantial donor–donor heterogeneity in disease presentation and in response to treatment, our patient-derived models ultimately have potential for application in personalized screening of therapies.

Three cellular fractions ( Fig. 1a ) (endometrial epithelial cells, endometrial stromal cells and myometrial SMCs) were isolated from patient tissue and used in a range of different seeding scenarios to create multicellular 3D cultures. The creation and multiplexing of these cultures were facilitated by use of a microfluidic microwell array device, which enables the simultaneous formation of multiple individual 3D cultures. The device architecture ( Fig. 1b ) consists of 5 × 5 microwell arrays at the base of a microfluidic channel, accessed via open loading wells for straightforward cell injection. When a cell suspension is pipetted into one open well, the resulting capillary pressure difference creates a flow through the microfluidic channel, with cells being allowed to sediment into the microwells ( Mulholland et al , 2018 ; Paterson et al. , 2022 ). The ultra-low adhesion surface of the device promoted the formation of 3D cellular aggregates within 24–48 h inside the microwells ( Fig. 1c , showing the formation of myometrial spheroids), with no cells able to attach to the inner device surface. Sequential cell loading, washing and incubation stages could then be used to create multi-layered 3D cell aggregates. On-chip approach to the creation of patient-derived 3D uterine models. ( a ) Three separate cell fractions (epithelia and stroma from dissected endometrial tissue and smooth muscle cells from myometrium) are produced from biopsy tissue by a combination of enzymatic and mechanical digestion. ( b ) These fractions are sequentially seeded (by injection) into the microfluidic device, at the base of which is a microwell array, into which cells sediment and subsequently aggregate. Scalebar is 1 cm. ( c ) An example of a microwell array seeded with SMCs that aggregate to form 3D cell cultures in the non-adherent microwells. Scale bar is 250 µm. We first characterized the 3D culture of individual cell types ( Fig. 2a–c ). The SMC and endometrial stroma single cell suspensions aggregated within 24 h to form compact spheroids, whilst the endometrial epithelial cells formed a looser structure. To prevent the epithelium from becoming diffuse, supporting cells were needed; when in co-culture with stromal cells, the epithelial cells either wrapped around the stromal cell spheroids or formed a distinct compact cluster attached to the aggregated stromal cells ( Supplementary Fig. S1 , with the markers used in all experiments validated in 2D culture as shown in Supplementary Fig. S2 ). To explore the formation of a uterine wall-like tri-culture model, a range of cell seeding strategies were investigated, sequentially or simultaneously seeding the different cell fractions as described in Fig. 2d , with an example time-course of cell aggregation shown in Fig. 2e . In addition, Supplementary Movie S1 (a time-lapse recording of fluorescently labelled cells) illustrates the aggregation of SMCs around a pre-formed epithelial-stromal cluster. Sequential seeding strategies for the creation of 3D uterine wall models. Whilst smooth muscle ( a ) and stromal ( b ) cells readily aggregated to form compact spheroids, cells from the epithelial fraction alone ( c ) aggregated to form a looser, sheet-like structure. ( d ) The series of sequential seeding strategies for the formation of uterine tri-cultures investigated (Scenarios A, B, and D), alongside the simultaneous seeding of all three cell types to investigate their self-organization (Scenario C). ( e ) An example timeline from a sequential seeding scenario (Scenario B) illustrating cellular aggregation and organization at different time-points. ( f ) The cellular organization of the tri-cultures, at Day 10, resulting from the different seeding scenarios (A–D as marked) as assessed by staining for epithelial (pan-cytokeratin (pCK), magenta) and smooth muscle (smooth muscle actin (SMA), green) markers, alongside nuclear staining (Hoechst, blue). Scale bar is 100 µm. The different seeding order did not affect the overall pattern of cellular organization and broadly similar organization was observed for all sequences investigated. Although there was noticeable variation between individual cultures for any given protocol, in most cases the epithelia either formed the outer layer of the 3D cultures ( Fig. 2f ) or segregated into a separate cluster within a 3D aggregate ( Supplementary Fig. S3a ). Surprisingly, the most robust uterine mimics, that is a single compact 3D aggregate characterized by high cell viability ( Supplementary Fig. S4 ) with a clear outer epithelial layer and SMC core, were produced when SMCs were seeded last (i.e. scenarios B and D). The largest number of these models was found with scenario D (where approximately 50% of 3D cultures were either wholly or partially surrounded by an epithelial layer). The use of the microwell array format facilitated the assessment of the phenotype of individual cultures, allowing the observation of rarer alternative 3D architectures ( Supplementary Fig. S3 ). For example, with seeding scenario B (initially seeding with endometrial stroma/epithelia followed by SMCs), there were infrequent (2 out of 97 cultures, fixed across Day 6–10) examples of 3D cultures with an epithelial core surrounded by myometrial cells ( Supplementary Fig. S3c ). No examples of this architecture were seen in any other seeding scenario. To ascertain whether the provision of extracellular matrix (ECM) proteins promotes the growth of robust 3D cultures with uterine-like organization and enhanced reproducibility, the incorporation of a solid gel support and of a low-viscosity matrix (both Matrigel ® , 100% and 5% gels, respectively) were investigated. The same order of cell addition as in scenario D was used and live-cell fluorescence labelling was incorporated to better monitor cellular organization ( Fig. 3a and b ). The incorporation of 100% Matrigel ® did not result in the formation of uterine-like 3D models ( Fig. 3c and d ), rather it promoted separation of cell types and pronounced outgrowth of stromal cells, with clear signs of stromal cell migration into the Matrigel ® in nearly all cases (90% of cultures from day 9 immunocytochemistry; see also Supplementary Fig. S5c ). However, the 5% Matrigel ® condition ( Fig. 3c and d ) was highly effective in promoting the formation of 3D uterine-like models, resulting in a clearly polarized cell distribution with distinct outer epithelial layers. These improved 3D cultures are particularly apparent when comparing an array incorporating the 5% gel to an array without the gel (comparing Supplementary Fig. S5a and b , both have the same cell seeding and imaging timeline). Incorporation of low-viscosity ECM facilities the formation of 3D tri-cultures with uterine-like organization. ( a ) The experimental timeline used for forming gel-supported 3D cultures. ( b ) Example time course, showing cells labelled with CellTrace™ reagents (stromal fraction, blue; epithelial fraction, magenta; smooth muscle cells (SMCs), green), illustrating cellular organization prior to gel addition (scale bar corresponds to 100 µm). ( c ) The different cellular architectures obtained on day 7 following the addition of either 100% Matrigel ® , 5% Matrigel ® or no gel (cells labelled with CellTrace™ reagents, colours as noted in b). ( d ) Staining of 3D cultures on day 10 to determine the organization of the epithelial cells (pan-cytokeratin (pCK), magenta), stromal cells (vimentin, blue) and SMCs (calponin, green) in response to 5%, 100%, or no Matrigel ® conditions (DAPI staining shown in yellow). Note the stromal cell migration (yellow arrow) in the 100% gel example (observed in 90% of cultures). All microwells are 250 µm wide. Having identified the key parameters required for obtaining 3D cultures with a uterine-like organization, i.e. the use of a low viscosity ECM support and the seeding of SMCs last, we further quantified important features of the resulting cultures: epithelial encapsulation, culture size, and expression of functional epithelial and stromal markers. Epithelial encapsulation was confirmed by optical sectioning (Apotome imaging in Fig. 4a and b and Supplementary Fig. S6 ). When comparing wide-field and Apotome imaging, epithelial encapsulation can be clearly identified by wide-field imaging, which allows for faster assessment of entire arrays containing cultures of different dimensions. Validation and quantification of epithelial encapsulation and expression of functional markers. ( a ) Confirmation of epithelial encapsulation by optical sectioning (Apotome images shown in lower panels; upper panels showing standard wide-field imaging). ( b ) 3D reconstruction of the lower half of a 3D culture, showing x – z (green box) and y – z (yellow box) planes. ( c ) Quantification of the extent of encapsulation resulting from three different patient biopsies and two different seeding scenarios (n refers to the number of individual 3D cultures assessed). Day 15 samples were subjected to 6 days of hormonal stimulation, Day 9 samples were not stimulated. ( d ) Expression of the junctional marker E-cadherin (E-Cad) and oestrogen receptor beta (ER-beta (Day 15) cultures). Images obtained by optical sectioning, with the upper panels showing an optical section through the base of the 3D culture and lower panels a section through the centre. All scale bars correspond to 100 µm. The distribution of individual culture dimensions reflects a range of parameters, including the precise cell seeding distribution of each cell type across an array and the cell growth rates. Individual aggregate dimensions were quantified ( Supplementary Fig. S7 ) and compared across three different patient donors (three different donors for endometrial tissue, with a common myometrial donor) and for the two seeding scenarios, where SMCs were added last (B and D). This was assessed for cultures grown with (fixed on Day 15) and without (fixed on Day 9) hormonal stimulation that mimicked the secretory phase of the menstrual cycle. Differences in the mean values of the estimated diameter of the aggregates were observed between the two scenarios. With hormone stimulation and a 15-day growth period, the diameter was larger for Scenario D than for Scenario B for all donors, though this was not significant in the case of patient 1 ( P  = 0.953/0.013/<0.001 for patient 1/2/3; two-sample t -test). The opposite trend was observed for the non-stimulated cultures assessed at Day 9, where the Scenario B cultures had larger mean equivalent diameters (significantly different for patient 1 and 2; P  = 0.021/0.005/0.051 for patient 1/2/3). The percentage of individual cultures encapsulated with epithelium was then assessed for the same conditions. The smallest cultures, those with effective diameters ≤70 µm (6–8% of total cultures for Day 9/15), were excluded from the analysis (as their central focal planes were much lower than the majority of cultures being imaged and, therefore, difficult to accurately classify). Because some microwells contained more than one 3D cell aggregate, two levels of encapsulation were considered: ‘Complete’, where any 3D aggregate within the well was encapsulated; and ‘Partial’, for wells where two 3D aggregates were present and the larger of the two was encapsulated. In Fig. 4c , ‘All’ refers to the sum of Complete (%) and Partial (%) cases and shows that high levels of encapsulation were obtained, particularly following hormone stimulation (Day 15, >80%), where the highest percentages were obtained with Scenario D. There was not, however, a significant difference in encapsulation between the two seeding scenarios ( P  = 0.058/0.661, two-sample t -test for ‘All’ Day 15/Day 9, respectively). A visual comparison of cultures obtained from these three patients can be seen in Supplementary Fig. S8 . When assessing a metric for the fraction of epithelial cells in a culture ( Supplementary Fig. S9 ), with respect to the total cell population, no clear trend was identified when comparing the two different seeding scenarios across the three patients. However, the variation in the data for scenario D was significantly lower than for scenario B ( P  = 0.018, two-sample t -test comparing the variances obtained for each scenario, n = 3 patients). With hormonal stimulation, which is required to demonstrate model functionality, Scenario D appeared to be most consistent. Therefore, we investigated the expression of both the epithelial junctional marker E-Cadherin, which plays a key role in epithelial barrier function, and oestrogen receptor (ER-beta), which is expressed by both stromal and epithelial cells in the uterus ( Hapangama et al. , 2015 ). Hormone-stimulated Day 15 cultures exhibited E-cadherin expression at cell-cell junctions (visible in bottom plane) and ER-beta staining in the outer (endometrial) layers of the 3D culture ( Fig. 4d ). To assess the functionality of our microfluidic tri-culture models, secretion of the decidualization markers insulin-like growth factor binding protein-1 (IGFBP-1) and osteopontin were measured in response to hormonal stimulation. The 3D cultures with 5% Matrigel ® support were produced using cells isolated from secretory phase patient tissue and stimulated as noted above (seeding and stimulation timeline used here shown in Fig. 5a ). In response to hormonal stimuli, IGFBP-1 secretion was observed to increase over time for two of the three donors (relative to untreated controls, Fig. 5b ), with the third showing a response up to day 4. Osteopontin secretion was consistently higher in hormone treated tricultures compared to controls ( Fig. 5c ). The decidualization response of the 3D uterine wall models. ( a ) Timeline of the hormone stimulation workflow. ( b ) IGFBP-1 secretion in response to hormone stimulation as measured by ELISA. Data are expressed as fold change relative to untreated controls. ( c ) Osteopontin secretion by hormone treated and control tricultures derived from three donors. Further functional analysis was performed by assessing the calcium response of SMCs within the 3D cultures to stimulation with the agonists ET-1 and OT, measuring fold changes in cytoplasmic calcium concentration ([Ca 2+ ] c ) by fluorescence microscopy. When 3D cultures were formed using SMC fractions only, all cultures recorded (n = 15) showed a clear response to ET-1 stimulation ( Fig. 6a and b ), with a rise in [Ca 2+ ] c for most individual cells. However, when the optimized uterine tri-culture models were formed (using 5% Matrigel ® ), a different picture emerged ( Fig. 6c and d ). Clear responses to ET-1 were observed for 3D cultures that lacked an encircling outer epithelial layer but, in contrast, no responses were observed from larger cultures with an apparently complete outer epithelial layer. When compared to ET-1, the responses obtained to OT stimulation were substantially lower, in terms of the number of individual cells responding within a 3D culture (4 out of 12 SMC-only cultures showing >3 responding cells; 11 showing ≥1 cells), with a comparison between the two shown in Supplementary Fig. S10a . Interestingly, monolayer SMC cultures showed a stronger response to OT (with the majority of cells responding) than to ET-1 (with the minority of cells responding), as demonstrated in Supplementary Fig. S10b . Calcium responses of 3D uterine models to stimulation with endothelin-1 (ET-1). ( a ) Smooth muscle cell (SMC)-only cultures responded strongly to stimulation with ET-1, with the majority of cells exhibiting a rise in intracellular calcium. The colour coding in the map of responding cells (right, with left showing bright-field image) indicates temporal variations in the response time of individual cells, grouping cells according to whether they responded in the first (magenta), second (green) or third (blue) 20 s period after the onset of the response. ( b ) Example traces illustrating the temporal response of individual cells within the 3D culture indicated by the yellow arrow in (a), with the position of the specific cells tracked indicated by the regions of interest (ROIs) highlighted in the right-hand image. ( c ) Map of responders for two 3D tri-cultures supplemented with 5% Matrigel ® , with the left-hand culture responding strongly to ET-1 and the right-hand culture showing no response, aside from a small cluster of cells that were loosely attached to the main 3D culture (the position of latter is highlighted by the yellow circle). ( d ) Bright-field and immunofluorescence images for cultures in (c), with the responding culture at the top. The latter shows segregation of the epithelial cells (pCK) within the 3D structure, whilst the non-responding culture was encircled by an epithelial layer.

The collection and use of human tissue was approved by the Liverpool Adult Research Ethics Committee (LREC: 19/SC/0449) and all women recruited gave informed written consent. Endometrial pipelle biopsies were collected from premenopausal women undergoing gynaecological surgery for benign conditions who were not taking hormonal medications. Myometrial biopsies were collected during pre-labour elective CS deliveries of women at full-term pregnancies (>37 weeks). An overview of participant demographics and the experiments for which their samples were used is given in Supplementary Table S1 . Tissue was minced into small pieces (<1 mm) using a scalpel blade and digested with 1 mg/ml Dispase II (Gibco™, Thermo Fisher Scientific, UK), 2 mg/ml collagenase type I (Gibco™, Thermo Fisher Scientific) and 80 µg/ml deoxyribonuclease (DNase) I (Merck, UK) for ∼1 h at 37°C in a shaking water bath as previously described ( Valentijn et al. , 2013 ). Digests were periodically triturated to enhance tissue breakdown and observed under a microscope to check for the presence of whole tissue fragments and free epithelial glands. The endometrial digests were passed through a 40 µm cell sieve (Falcon™) to separate glandular (retentate) and stromal (flow-through) elements. Erythrocytes were removed from the myometrial digests and endometrial stromal fractions using Ficoll-Paque (Merck) density gradient centrifugation. Intact endometrial glands, endometrial stroma and myometrial cells were frozen and stored at −80°C prior to use. Primary endometrial tissue samples and myometrial smooth muscle cells (SMCs) were thawed rapidly at 37°C and immediately washed in 5 ml cell culture medium (complete for SMCs and stroma, basal for epithelia). To further digest large epithelial glands, pellets were resuspended in 1 ml serum-free media (SFM; phenol red-free DMEM/F12, Merck, UK), 1 ml 0.25% trypsin (Thermo Fisher Scientific) and 100 µl DNase I (4 mg/ml, Merck) and incubated at 37°C for 20 min. Glands were gently triturated using a narrow bore pipette tip to yield single cells. Following trituration, the digestion reaction was quenched with 1 ml foetal bovine serum (FBS, Thermo Fisher Scientific). The digested epithelial fraction was washed with complete cell culture medium (phenol red-free DMEM/F12, 10% [v/v] FBS (Merck), 2 mM L-glutamine and 100 µg/ml Primocin (InvivoGen, France)) supplemented with 20 ng/ml recombinant human epidermal growth factor (EGF) (Merck). Myometrial SMCs and endometrial stromal cells were maintained in complete culture media in monolayers to up to 70% confluency prior to seeding into microfluidic devices. For all co-/tri-cultures containing epithelial cells, media was always supplemented with 20 ng/ml EGF. Multi-layered microfluidic devices were fabricated using standard soft lithography techniques, following established protocols ( Mulholland et al. , 2018 ; Paterson et al., 2022 ). A device consisted of up to 24 microfluidic channels, each of which was connected by two open wells, with channels hosting one or three arrays of 25 square microwells (250 × 250 × 200 µm). Briefly, a 10:1 ratio of poly(dimethylsiloxane) (PDMS) prepolymer (Sylgard 184, Dow Corning) to curing agent was mixed and dispensed onto patterned silicon wafers. The wafers were degassed and subsequently incubated at 85°C for a minimum of 3 h to allow curing of the PDMS solution. PDMS layers were then cut from the wafers and open wells were formed using a surgical biopsy punch (Miltex). PDMS layers were cleaned and treated with oxygen plasma (Pico plasma cleaner, Diener electronic, Germany) to permanently bond the layers together, after which a 1% solution of Synperonic ® F108 (Sigma Aldrich, UK) was injected to create ultra-low adhesion conditions. Subsequently, devices were washed using phosphate-buffered saline (PBS) and used for cell culture. For seeding of single-cell suspension (SMCs and stroma) or small epithelial cell clusters into microfluidic channels, cells were resuspended to a concentration of 2–4 × 10 6 cells/ml. Cell solution volumes (5 µl) were seeded in the microfluidic device and incubated for 5 min to ensure settling of the cells to the bottom of the microwells. Excess cells were removed from the inlet and outlet open wells, which were then filled with 40 µl culture medium. For co-/tri-cultures of epithelia, stroma and SMCs, cells were seeded sequentially as described below. For injection of 100% Matrigel ® Basement Membrane Matrix (# 356237 Corning, UK), cell culture media was removed from the inlet and outlet and the device placed between two cool packs for 10 min. 20 µl of Matrigel ® was then injected into the inlet, avoiding any air bubble formation. The device was further incubated for 15 min at 4°C before injecting 30 µl of ice-cold media to remove Matrigel ® from the channel above the microwells. After the media flow had stopped, the inlet and outlet were completely emptied again and 40 µl of cell culture media simultaneously injected into both. The device was incubated for 30 min at 37°C to allow complete gelation of Matrigel ® . For the injection of 5% Matrigel ® , the appropriate volume of Matrigel ® was mixed into ice-cold medium. Cell culture media from inlet and outlet was fully removed and 5 µl of the 5% Matrigel ® solution injected into the inlet, followed by a 5 min incubation. The same process was repeated with the outlet, then both the inlet and outlet were topped up to 40 µl with the 5% Matrigel ® solution. Stromal cells and SMCs were thawed 72 h prior to seeding into microfluidic devices. Epithelial glands were thawed and digested on the day of seeding. Stromal and epithelial cells were seeded together in a 1:1 ratio. After 24 h, SMCs were added to the existing co-culture and this tri-culture was cultured for 24 h, followed by exchange of the culture medium for medium supplemented with 5% [v/v] Matrigel ® . All cell culture medium used subsequently was supplemented with 5% Matrigel ® . 24 h before hormone treatment, the culture media was replaced with media containing 2% [v/v] charcoal-stripped (CS) FBS (Thermo Fisher Scientific). For hormone treatment, during the first 24 h of hormone exposure, cells were cultured in 2% CS-FBS DMEM/F12 containing 10 nM β-estradiol (E2, Merck) to mimic the end of the proliferative phase. The initial 24 h hormone treatment was followed by treatment with 10 nM E2, 100 nM medroxyprogesterone acetate (MPA, Sigma Aldrich) and 500 µM cyclic adenosine monophosphate (cAMP, Merck). Conditioned medium was collected and replaced every two days over the course of six days by performing a 50% volume exchange (20 µl) with new hormone-containing media. All hormone treatments were conducted in three technical replicates. IGFBP-1 and osteopontin concentrations were measured from conditioned medium using a human IGFBP-1 DuoSet sandwich ELISA kit (R&D Systems, UK) and human osteopontin enzyme-linked immunosorbent assay (ELISA) kit (ThermoFisher Scientific), respectively, according to the manufacturer’s protocols. Standard curves were generated using seven point, 2-fold serial dilutions and samples were assayed in duplicate. Absorbance was measured at 450 and 570 nm using a FLUOstar Omega microplate reader (BMG Labtech, UK). Optical imperfections were corrected by deducting the 570 nm value. Standard curves were generated using a four-parameter logistic regression model in GraphPad Prism Version 9.0. To distinguish between different cell types by live-cell imaging, stroma, epithelia and SMCs were labelled with CellTrace dyes (InvitroGen, Thermo Fisher Scientific). Stromal cells were labelled as described by the manufacturer with CellTrace Far Red (CTFR), 72 h after thawing. SMCs were labelled with carboxyfluorescein succinimidyl ester (CFSE) (green) using the same dye concentrations. The epithelial cells were labelled in suspension with CellTrace Violet (CTV) immediately after the digestion of the glandular fraction. Monolayer cell cultures were fixed with 4% [w/v] paraformaldehyde (PFA, Thermo Fisher Scientific) for 25 min, quenched with 100 mM glycine, and permeabilized with 0.2% [v/v] Triton X-100 in PBS for 10 min before blocking with 2% [w/v] bovine serum albumin (BSA, Merck) in PBS for 30 min. Afterwards, cells were incubated with primary or conjugated-primary antibodies for 24 h at 4°C, with secondary antibody incubations being 120 min long. Staining of 3D cell cultures in microfluidic devices was performed as described by Pettee et al. (2019) . In brief, cells were washed with PBS, fixed with 4% PFA, permeabilized with 0.5% Triton X-100/PBS and blocked with 3% BSA. Primary antibodies were incubated for up to 48 h and secondary for another 24 h at 4°C in the dark before imaging. The antibodies used in the work were: anti-pan-cytokeratin (p-CK)-Alexa647 (SC8018AF647) 1:100 dilution (Santa Cruz, USA); anti-calponin-1 (CAL)-Alexa 488 (SC-58707AF488) 1:50 dilution (Santa Cruz); anti-smooth muscle actin (SMA)-Cy3 (C6198) 1:100 dilution (Merck); anti-vimentin (SC-6260) 1:50 dilution (Santa Cruz); anti-vimentin (ab45939); 1:200 dilution (Abcam); anti-CD324 (E-Cadherin) 1:200 dilution (eBioscience, Invitrogen); anti-estrogen receptor beta (sc-390243) 1:200 dilution (Santa Cruz Technologies); donkey anti-rabbit Alexa555 (Invitrogen); goat anti-rat Alexa488 (Invitrogen); goat anti-mouse Alexa647 (Invitrogen); and donkey anti-mouse-Alexa555 1:200 (Thermo Fisher Scientific). Cultures were loaded with a fluorescent calcium indicator using a 5 µM Cal-520™ AM (Stratech, UK) solution containing 0.02% [v/v] Pluronic F-127 and incubated for 60 min at 37°C followed by 15 min at room temperature in the dark. Prior to imaging, cells were washed with SFM and incubated for 20 min. Immediately before imaging, the inlet and outlet of the device were emptied. Changes in measured fluorescence levels in response to the addition of either endothelin-1 (ET-1, 100 nM) or oxytocin (OT, 100 nM) (both Merck) were then recorded (2.6 fps). The response to OT was measured first, followed by two washes with SFM and a 5 min rest period, before measuring the response to ET-1. For Ca 2+ imaging, an inverted Observer A1 microscope (Zeiss) with an Andor LucaR EMCCD camera and a ×10 0.25 NA objective was used. All other imaging was performed using an Observer Z1 (Zeiss) microscope, with an Apotome 2 structured illumination module used for optical sectioning (sections acquired every 7 µm in the vertical axis, moving up from the base of the 3D aggregate), connected to an Orca Flash 4.0 sCMOS camera (Hamamatsu) with an x20 0.8 NA, a ×10 0.3 NA or a ×5 0.16 NA objective. Data were analysed using either ZEN Blue version 3.4 (Zeiss) or ImageJ version 1.53f (NIH, USA). Data stacks obtained from Ca 2+ imaging experiments were processed by first performing an eight-frame sequential subtraction (e.g. the pixel intensity values for each frame were subtracted from the values of the image eight frames ahead to produce an image where only changes in intracellular fluorescence levels are shown) and then applying a 3 × 3 median filter. Beginning at the onset of the response, the resulting images were averaged across three 20 s periods and the three images used to produce a colour map of responding cells that indicates variation in the response time of individual cells. Regions of interest (ROI) defining individual responding cells were manually drawn and maps of all viable cells were created from a 20-frame average of the baseline fluorescence prior to agonist application. To quantify overall dimensions and cell composition of individual 3D aggregates, fluorescent images were first thresholded in Zen Microscopy software (Zeiss) and then converted to binary images. These were analysed using a Matlab (Matlab R2023a) routine to extract area values. This procedure was performed to estimate both the overall aggregate dimensions (merging all fluorescent channels into the same binary image), and for each individual cell type (using individual fluorescent channel separately). Pixel area values were then converted to µm 2 according to the objective used during image acquisition.

Model systems are important for understanding human physiology and pathologies, while being essential for mechanistic studies and drug development. There are inherent limitations to existing ex vivo and in vitro experimental approaches of studying human uterine physiology and prevailing pathologies. These include animal models not accurately recapitulating human reproductive physiology, simplified systems (2D models) being unable to resemble the complexity of in vivo cell–cell interactions, cell line-based models that are less relevant to non-transformed cells present in benign pathologies, and 3D models disregarding the significance of the EMJ and containing only endometrial components. To overcome these limitations, we have established a novel 3D uterine microfluidic tri-culture model using biopsy-derived cells and comprising both the myometrial and endometrial cellular participants of the uterine wall; thus, our model shows improved physiological relevance when compared with many existing models. Recent studies in oncology have demonstrated the superior predictive power of biopsy-derived 3D models in assessing treatment response prior to clinical use ( Snijder et al. , 2017 ; Vlachogiannis et al. , 2018 ). Our study fills the gap in the current literature, by exploiting a similar approach in devising a multi-cell type, patient-derived 3D model of the uterine wall that will be suitable and relevant for benign, chronic gynaecological conditions, where heterogeneity in disease presentation is well established ( Tempest et al. , 2021 ; Maclean et al. , 2023 ; Powell et al. , 2023 ). In addition to improving physiological mimicry, for which the importance of using primary human cells has previously been highlighted in an endometrial model ( Cook et al , 2017 ), our 3D model created from biopsy-derived cell populations would also enable patient-specific models of disease to be created, ultimately leading to personalized tailoring of therapies. Microfluidic technology offers several advantages in miniaturization and multiplexing experimental capabilities, compared with standard culture techniques. The specific microfluidic methodology employed in this article enables control of sequential cell and reagent loading, maximizing the use of small patient tissue samples (which is particularly important when expanding cell numbers in vitro is not possible due to the potential loss of phenotype) and offering a medium throughput screening tool to test a range of therapeutic approaches using small quantities of patient-derived cells. This approach promotes observation of patient-to-patient differences with statistical relevance, enables the robust implementation of advanced culture protocols, and could ultimately facilitate disease modelling of individual patients and thus, personalized drug screening. To create a robust 3D uterine wall model, with an outer epithelial layer enclosing an inner myocyte core, optimum results were obtained when SMCs were seeded last and by the inclusion of a low-viscosity ECM support (5% Matrigel ® ). Our observation agrees with Hirokawa et al. (2021) , who previously reported that low viscosity ECM support facilitated the efficient establishment of large intestine organoids, removing the requirement for a solid matrix. Our use of a low viscosity ECM support to establish robust individual multicellular 3D structures with distinct uterine-like cellular organization is favourable to the study of endometrial physiology and contrasts to the use of a solid matrix (100% Matrigel ® ), where segregation of individual cell types and outward migration of stromal cells was pronounced. Recent single cell studies have amply demonstrated human endometrial transcriptomic heterogeneity at the cellular level, in health and in pathologies, such as adenomyosis and endometriosis ( Ma et al. , 2021 ; Goad et al. , 2022 ; Yildiz et al. , 2023 ). An additional benefit of a microwell array format, therefore, is the possibility to study distinct cell subtype-specific behaviours. Such functional differences can be readily identified when screening large numbers of cultures across multiple arrays. As an example, we observed rare instances of an epithelial core surrounded by a layer of myometrial cells, which is potentially useful in modelling the pathogenesis of adenomyosis ( Gnecco et al , 2020 ). Identification of behaviours of less abundant cellular subtypes would improve the understanding of their specific contribution to disease pathogenesis and differential responses to medications. To validate the physiological relevance of our model, two functional assays were performed. The first of these was a decidualization assay. Ovarian hormone-driven decidualization facilitates human pregnancy establishment by transforming the endometrium into a receptive and permissive environment for embryo implantation ( Gellersen and Brosens, 2003 ), and cAMP-mediated protein kinase A (PKA) is also crucial for the decidualization process ( Kusama et al , 2014 ). Triculture models derived from endometrial biopsies from three donors in the secretory phase of the cycle, demonstrated a decidual response in the presence of E2, MPA, and cAMP. Relative levels of secreted IGFBP-1 varied between donors, however, all three donor-derived tricultures demonstrated some recapitulation of response. The best decidualization response according to IGFBP-1 secretions was seen in the late secretory sample, whereas the weakest response was produced by the early secretory sample. This may demonstrate either the expected patient-specific heterogeneity in cellular response ( Abbas et al. , 2020 ), or reflect the exact menstrual cycle phase-specific differences in response. Future work including multiple samples at different phases of the cycle from the same donor and scrutiny of a larger number of samples from different donors will be required to further confirm these possibilities. Osteopontin is expressed by the endometrial glandular epithelium during the secretory phase of the menstrual cycle and is regulated by progesterone ( Johnson et al. , 2000 ; Apparao et al. , 2001 ; von Wolff et al. , 2004 ; Hapangama et al. , 2012 ). Tricultures consistently exhibited increased osteopontin secretion in the presence of hormones, thus demonstrating the expected epithelial differentiation response in uterine tricultures. In 3D culture, the patient-derived myocytes showed strong responses to ET-1, compared with OT, a contrasting observation to the response observed in 2D culture of the same myocytes. Differences in gene and protein expression levels between 2D and 3D cultures of human myometrial cells have previously been reported, with Malik et al. (2014) demonstrating that the culture architecture alters cellular behaviour. The higher sensitivity of pregnant myometrial cells to OT compared with non-pregnant myometrium is also well-established ( Arrowsmith et al. , 2012 ; Fuchs and Fuchs, 1963 ). Although we obtained myocytes from pregnant women, their response to OT was reduced with the addition of endometrial cellular components in 3D culture, indicative of adapting a non-pregnant phenotype. Furthermore, tri-cultures also demonstrated an architecture-dependent response. The lack of Ca 2+ response to ET-1 in larger, uterine-like models, where myometrium is encapsulated with endometrial epithelium, may be a consequence of barrier function due to a complete epithelial layer with junction formation. Although this model contains the main primary cell types constituting the uterine wall, endometrial cells demonstrate region-specific cellular differences. Therefore, future work is required to ascertain whether specific cells obtained from the endometrial sub-regions (e.g. basalis) are needed or whether endometrial tissue from any region will acquire properties of the basalis, which is adjacent to the myometrium in vivo , when co-cultured with myometrium. High-resolution imaging techniques, such as electron microscopy, as well as spatial transcriptomics or single cell sequencing, could be employed in future studies to further ascertain the functional relevance, organization and cellular phenotype of uterine tricultures. Together, the results of morphological characterizations and functional assays presented validate this microfluidic triculture approach (using endometrial epithelial and stromal cells, and myocytes) for the creation of a physiologically relevant human model of uterine function, providing the first myocyte-containing, patient-derived organ-on-a-chip uterine model. Our human in vitro model provides a new tool that could be further developed and employed to improve our understanding of the cellular behaviours underlying chronic gynaecological disorders, such as endometriosis, adenomyosis and leiomyoma, with applications in personalized drug screening and the assessment of patient-to-patient heterogeneity in responses.