An Automation Workflow for High‐Throughput Manufacturing and Analysis of Scaffold‐Supported 3D Tissue Arrays

Abstract

Patient‐derived organoids have emerged as a useful tool to model tumour heterogeneity. Scaling these complex culture models while enabling stratified analysis of different cellular sub‐populations, however, remains a challenge. One strategy to enable higher throughput organoid cultures is the scaffold‐supported platform for organoid‐based tissues (SPOT). SPOT allows the generation of flat, thin, and dimensionally‐defined microtissues in both 96‐ and 384‐well plate footprints that are compatible with longitudinal image‐based readouts. SPOT is currently manufactured manually, however, limiting scalability. In this study, an automation approach to engineer tumour‐mimetic 3D microtissues in SPOT using a liquid handler is optimized and comparable within‐ and between‐sample variation to standard manual manufacturing is shown. Further, a liquid handler‐supported cell extraction protocol to support single‐cell‐based end‐point analysis using high‐throughput flow cytometry and multiplexed cytometry by time of flight is developed. As a proof‐of‐value demonstration, 3D complex tissues containing different proportions of tumour and stromal cells are generated to probe the reciprocal impact of co‐culture. It is also demonstrated that primary patient‐derived organoids can be incorporated into the pipeline to capture patient‐level tumour heterogeneity. It is envisioned that this automated 96/384‐SPOT workflow will provide opportunities for future applications in high‐throughput screening for novel personalized therapeutic targets.

An automation workflow for manufacturing and analysis on a scaffold‐supported platform (96‐ and 384‐well) that integrates tissue generation, maintenance, and single‐cell‐based analyses is established. The workflow on this platform is compatible with high‐content image analysis, high‐throughput flow cytometry, and multiplexed cytometry by time of flight analysis on complex 3D microtissues containing patient‐derived cells.

1. Introduction

Drug discovery, particularly in oncology, is challenging because conventional pre‐clinical models often do not accurately predict the effectiveness of therapies in patients.^{[ 1 ]} This is in large part because conventional 2D cell culture models and simplified spheroid 3D models constructed from immortalized cell lines typically fail to capture complex and dynamic cellular responses, as these models lack the cellular heterogeneity and microenvironmental cues necessary to replicate physiologically relevant cell behaviour.^{[ 2} , ^{3 ]} To address this challenge, 3D in vitro models that incorporate primary and/or patient‐derived cells or patient‐derived organoids (PDOs) have emerged as a powerful strategy to improve pre‐clinical assessment accuracy.^{[ 4} , ⁵ , ⁶ , ⁷ , ⁸ , ^{9 ]} These next‐generation engineered models better recapitulate the tumour heterogeneity and enable modelling patient‐specific disease, but present challenges to culture at scale for high‐throughput and automated screening applications.^{[ 10} , ¹¹ , ^{12 ]}

The growth of PDOs relies on the presence of a 3D bio‐matrix, such as Matrigel, which has poor structural integrity that can result in uncontrolled culture deformation when using high cell densities, during long‐term culture, or even during the frequent washing steps necessary for analysis.^{[ 13 ]} A variety of strategies have emerged to better support long‐term 3D cultures including models incorporating a solid substrate such as elastomer‐based microfluidic devices^{[ 14} , ¹⁵ , ^{16 ]} and scaffold‐supported platforms.^{[ 13} , ¹⁷ , ¹⁸ , ¹⁹ , ²⁰ , ²¹ , ^{22 ]} By improving the structural integrity and control over culture assembly, some of these models have also enabled increased culture complexity and relevance such as the incorporation of multiple cell types including cancer‐associated fibroblasts,^{[ 13} , ^{23 ]} mesenchymal stromal cells,^{[ 24} , ^{25 ]} and endothelial cells.^{[ 26 ]} However, most of these novel 3D models are generally low‐throughput, require manual manufacturing, and do not integrate easily with standard instrumentation in the existing drug discovery pipeline. Some examples have been reported where automation is beginning to be incorporated into workflows using these complex culture models, primarily with a focus on automation of 3D microtissue/tumour fabrication and/or dispensing drugs to perform a drug response assay.^{[ 4} , ⁶ , ⁸ , ¹⁵ , ²⁷ , ²⁸ , ²⁹ , ³⁰ , ³¹ , ³² , ^{33 ]} For example, engineered microtissues/tumours have been fabricated automatically using bioprinting techniques to control the deposition of cell‐laden bioinks, such as extrusion‐based bioprinting,^{[ 29} , ^{30 ]} stereolithography,^{[ 31 ]} acoustic bioprinting,^{[ 32 ]} magnetic force field,^{[ 8 ]} contact‐capillary wicking to infiltrate cell‐gel into existing scaffolds,^{[ 22} , ^{34 ]} and FRESH bioprinting using sacrificial support materials.^{[ 33 ]} However, the remaining manual downstream processes, such as maintenance, screening, and high‐content analysis, still significantly limit the throughput of these culture platforms and potentially introduce unwanted experimental variation. A need exists therefore for strategies to integrate automation into each step of the fabrication, maintenance, and high‐content analyses of these models.

Another challenge associated with next‐generation engineered culture models is the ability to perform single‐cell‐based analyses on a large‐scale.^{[ 4} , ⁶ , ⁸ , ¹⁴ , ³⁵ , ³⁶ , ^{37 ]} In a few cases, automation of the entire workflow has been achieved using complex customized systems unique to individual models^{[ 36 ]} or integration of multiple different instruments for a single model.^{[ 35 ]} Moss et al. were able to automate printing, cell seeding, media changes, media collection, and long‐term incubation by programming a robotic arm moving across 4 different stations or instruments. However, their approach still lacked a single‐cell resolution analysis.^{[ 35 ]} Bulk cellular read‐outs, such as cell viability,^{[ 4} , ⁶ , ^{8 ]} protein expression,^{[ 35} , ^{36 ]} permeability,^{[ 36 ]} and bulk fluorescence,^{[ 14 ]} offer an accumulated metric of the drug‐response from a heterogeneous cell population within each whole‐well. This makes it particularly challenging to decipher particular cellular responses from specific sub‐populations in a complex microenvironment such as a tumour where multiple cell types are typically present. In another example, Renner et al. reported a fully automated workflow using a standard liquid handler rather than custom equipment to screen neural 3D organoid structures.^{[ 38 ]} Automation was integrated throughout the generation, maintenance and downstream analysis pipeline and included the use of an imaging approach with single‐cell resolution.^{[ 38 ]} However, this workflow did not involve 3D biomatrix, which could limit the use of this technique for applications, such as tumour organoids, where a bio‐matrix is critical to maintaining cell growth.^{[ 39 ]} Further, it is known that the stiffness of the culture biomatrix affects tumour cell phenotypes, such as chemosensitivity, therefore incorporating a matrix is likely critical for cancer culture platforms.^{[ 2} , ⁴⁰ , ^{41 ]} The use of a 3D matrix, as is typical in many engineered culture models, however, significantly increases the complexity of image‐based single‐cell analysis as these models are typically formed into a thick gel plug geometry with a curved meniscus surface profile. Further, the thick geometry of the plug distributes cells into multiple image planes often necessitating the use of confocal microscopy to decipher different cell populations. Microtissues/tumours manufacturing approaches that involve geometric templating of the hydrogel matrix to ensure a flat and thin gel geometry can significantly improve the capacity to perform imaging and obtain single‐cell resolution data^{[ 13} , ^{22 ]} however to date, these systems have not been automated with a view toward scaling throughput.

Our lab previously developed one geometric templating platform that enables the formation of geometrically defined, flat thin, microtissue/tumour cultures is the scaffold‐supported platform for organoid‐based tissues (SPOT).^{[ 22 ]} SPOT has been developed in both 96‐ and 384‐ well formats and SPOT plates are pre‐assembled to enable off‐the‐shelf use, which makes it particularly advantageous for integration with standard automation instrumentation (Figure 1a). SPOT has also been shown to enable real‐time tracking of patient‐derived tumour organoids.^{[ 22 ]} Here, we describe the automation of the manufacturing and analysis workflow for the SPOT model that integrates generation, maintenance, high‐throughput treatment, automated microscopy, high‐content image analysis, gel‐digestion for cell extraction, high‐throughput flow cytometry, and barcoding for multiplexed cytometry by time of flight (CyTOF) of complex 3D microtissues containing primary or patient‐derived cells (Figure 1b). Our integrated automated workflow removes labour‐intensive and challenging microtissue‐handling steps using an open‐source and inexpensive Opentrons OT‐2 (OT2) liquid handler, thereby allowing for rapid platform scale‐up and implementation into existing screening facilities. Further the OT2 also enabled us to generate a barcode master plate for multiplexed CyTOF analysis and to assist in the downstream barcoding and washing process without the introduction of errors, which is common when creating such master plates manually. Our automated manufacturing and analysis workflow enables in vitro culture of highly heterogeneous PDO cell populations combined with bulk phenotypic or single‐cell analysis at scale.

Figure 1.

An integrated high‐throughput automation workflow to generate, perturb, and analyze 3D in vitro SPOT cultures. a) Schematic representation of the 96‐SPOT components. 96/384‐SPOT uses an embedded PMMA‐patterned scaffold sheet to support the growth of 3D cell‐gel microtissues. The PMMA‐patterned paper scaffold (shown in blue) is attached to a bottomless well plate (top, shown in black) using two layers of double‐sided tape (shown in yellow) and a semi‐rigid polycarbonate film (bottom). The semi‐rigid polycarbonate film is optically transparent and thin to enable microscopy. b) Schematic representation of the proposed automated workflow performed by the OT2 liquid handler. 1) The OT2 is used to automate the generation of engineered 3D microtissues. Immortalized cell lines or organoid‐based cells can be incorporated into SPOT using the optimized cell‐gel dispensing OT2 sequence. 2) The OT2 automates high‐throughput screening by assisting in drug and reagent addition and culture maintenance. The engineered tissue arrays can be analyzed by time‐course high‐content microscopy and plate‐reader‐based assays, such as AlamarBlue. 3) The OT2 is also used to automate gel digestion. The cell‐gel within the tissue is mechanically and enzymatically degraded using an optimized digestion sequence. Single cells can then be extracted for end‐point downstream analysis, such as high‐throughput flow cytometry (bottom). 4) 96‐barcode master plate can be generated using OT2 to multiplexed digested single cells for high‐content proteomics analysis using CyTOF.

2. Results and Discussion

2.1. Comparison of Scaffold Candidates to Identify Material Compatible with Automation

The overall goal of this study was to establish an automated pipeline that can dispense a cell‐gel mixture into 96/384‐SPOT to generate 3D tissue arrays, assist in medium/high‐throughput screening assays, and digest hydrogel to recover single cells from the SPOT for end‐point analysis (Figure 1b). The 96/384‐SPOT utilizes a poly(methyl methacrylate) (PMMA)‐patterned cellulose scaffold and capillary wicking within each well to form thin microgel tissue without a meniscus. Further, the cellulose scaffold provides structural support to the hydrogel to enable long‐term culture and media changes. The scaffold is attached using 2 layers of double‐sided tape to a commercial bottomless well plate and to a thin (0.127 mm) polycarbonate film, which acts as a plate bottom, to allow high‐content imaging and compatibility with plate‐reader‐based assays overtime (Figure 1a). To provide automated manufacturing and microtissue processing, we utilized the liquid handling automated pipetting system OT2. The OT2 system is open‐source, commercially available, and allows for additional attachments for workflow customization, that is, the HEPA filter module and temperature‐controlled modules allow the possibility for a sterile work surface and for dispensing of temperature‐sensitive hydrogels, respectively. Initial pilot studies identified that manual dispensing allowed flexible user control of the pipette tip contact angle with the scaffold during the contact‐wicking manufacturing process, while the OT2 automated pipetting system allowed only perpendicular dispensing of the cell‐gel mix. Perpendicular dispensing led to unsatisfactory gel distribution into the paper scaffold and uneven seeding during manufacturing. To address this design restriction, we therefore first had to select a paper scaffold with faster and more‐reproducible capillary wicking properties to ensure consistent cell‐gel mixture spreading within the well during automated manufacturing.

To this end, we characterized 3 scaffold candidates (II, III, and IV) and compared them with the original paper scaffold (I) used in 96/384‐SPOT.^{[ 22 ]} The 3 scaffold candidates were preliminarily selected from a larger pool of candidates based on thickness (similar thickness as scaffold I), material types (cellulose scaffolds are preferred), and material size compatibility with well‐plate dimensions. Scanning electron microscopy (SEM) images of each scaffold (Figure 2a) revealed a similar porous structure across all 3 scaffolds except scaffold IV, which exhibited a more compact cellulose fibre structure. To quantify the capillarity wicking ability, we measured the time it took for red ink diluted in PBS to cover one 96‐SPOT well fabricated from one of the 4 different scaffolds. As shown in Figure 2b, scaffold III offered the most superior and consistent wicking ability across multiple wells and sheets of paper. We speculate that these observations may arise from differences in the hydrophobicity of each scaffold candidate. Using widefield microscopy images, we also calculated the pore area coverage, which further confirmed that scaffold IV had a significantly smaller and lower number of pores: specifically, scaffold IV had ≈10% lower pore area coverage compared with scaffold I (Figure 2c). Scaffolds II and III had a similar physical structure, but exhibited a significantly lower pore area coverage of 4% compared with the original paper scaffold I. To assess how the pore coverage difference affected seeding and hence tissue homogeneity, we seeded green fluorescence protein (GFP)‐expressing KP4 cells suspended in 3 mg mL⁻¹ bovine collagen manually and assessed cell distribution just after seeding or 3 days later. All 4 scaffolds allowed KP4 cells growth without spatial restraints (Figure 2d). However, the spreading of cells in scaffold IV appeared not as homogenous as the other 3 scaffolds (I, II, and III) (Figure S1a, Supporting Information). To quantify the tissue homogeneity obtained with the different scaffolds, we measured the coefficient of variation, which is a surrogate for tissue homogeneity, using a previously described method.^{[ 22 ]} The coefficient of variation provides a measure of the variation within a well by quantifying the ratio of the standard deviation of 100 randomly selected squares’ mean gray value (MGV) within the well over the whole‐well MGV. Our measurements confirmed that the use of scaffold IV led to poor homogeneity, whereas better homogeneity was observed with both scaffold II and III compared to scaffold I (Figure S1b, Supporting Information). These results indicated that scaffold III was likely the most compatible candidate with OT2 automated cell‐gel deposition based on the wicking properties.

Figure 2.

Comparison of scaffold candidates to identify material compatible with automation. a) Representative SEM images of the I) original SPOT scaffold and II–IV) 3 paper scaffold candidates. The scale bar is 500 µm. b) Wicking time was measured by observing the amount of time required for dyed PBS to wick radially within one well in 96‐SPOT for each scaffold candidate. Scaffold III offered the most superior wicking ability (shown in green). Statistical significance was assessed using ANOVA. Mean ± SD of 3 independent experiments. c) Pore area coverage as a percentage for all 4 paper scaffolds was measured based on brightfield images taken under 4x magnification. Scaffold IV exhibited the most condensed cellulose fibres. Mean ± SD of 3 independent samples size of around 12 mm². d) Representative images showing GFP‐expressing KP4 cells on day 0 and day 3 seeded at 5 × 10⁶ cells mL⁻¹ hydrogel using 3 mg mL⁻¹ type I bovine collagen into each scaffold candidate. The scale bar is 500 µm.

One significant advantage of 96/384‐SPOT is the ability to easily track cell growth and assess drug treatment effects over time using a widefield microscope. We therefore also assessed whether the new scaffold was compatible with imaging and did not exhibit strong autofluorescence in conventional imaging channels (Figure S2a, Supporting Information). Scaffolds II and III offered lower autofluorescence across DAPI, FITC, Cy3, and Texas Red channels tested compared with scaffold I when they were all exposed for 300 ms (Figure S2b, Supporting Information). We note that scaffold IV could potentially impede high‐quality real‐time imaging due to its strong autofluorescence. As scaffold III offered a similar porous structure, faster capillary wicking, lower autofluorescence, and supported mammalian cell growth, we selected this scaffold for automated 96/384‐SPOT for further optimization.

2.2. Characterization and Optimization of Essential Parameters for Automated Cell‐Gel Dispensing

Having selected a scaffold with appropriate wicking properties, we next set out to identify appropriate liquid handling parameters to enable the generation of homogenously seeded scaffolds. Since type I bovine collagen gel, the bio‐matrix that has been used previously in the SPOT system, is temperature sensitive, a temperature control setup is needed to prevent the uneven gelation and clumping of cells during seeding. To achieve this, as previously described,^{[ 22 ]} cold PBS was added into the spaces between the wells of the 96‐SPOT to enable temperature and humidity stabilization across the well plate. Moreover, the cell‐gel mixture stock solution and the 96/384‐SPOT were kept on a temperature‐controlled module at 4 °C during the gel dispensing process in the OT2 system. To ensure direct contact between the surface of the temperature‐controlled module and the bottom surface of the well plate, a customized aluminum block was fitted underneath the plate, providing both improved temperature control and additional physical support for the polycarbonate film during seeding.

As the cell‐gel mixture is much more viscous than aqueous solutions (for which the OT2 system was designed for), several parameters needed to be optimized to ensure reproducible small volume deposition into the scaffold. First, we assessed different contact locations of the 8‐channel pipette tips and scaffolds along the z‐axis using a 96‐SPOT set up. The plate bottom (z = 0 mm) location was determined based on a regular well‐plate thus optimization was required for the SPOT system as it is an off‐the‐shelf platform. For different tip z‐locations we assessed cell‐gel infiltration of GFP‐expressing KP4 cells using fluorescent images. Specifically, the OT2 tips were lowered in increments of 0.5 mm from the default plate‐bottom position and dispensed 5 µL cell‐gel mixture into each well. Due to the viscosity of the cell‐gel mixture and interfacial tension on the tips, insufficient or completed failed deposition could occur which led to partially seeded or empty wells. Surprisingly, even small differences in z‐increment, introduced a significant difference in the number of empty wells (no GFP‐expressing KP4 cells observed) for each 8‐channel deposition across 3 z‐location settings tested (Figure 3a‐i). Dispensing the cell‐gel mixture at the plate bottom (z = 0 mm) offered the worst results as most wells contained no cells due to unsuccessful or incomplete contact with paper scaffold. Notably, lowering the tip in the z‐direction by 0.5 mm versus 1 mm showed no significant difference in the number of empty wells observed. However, the tips were noticeably bent with the 1 mm setting. Since 96/384‐SPOT uses a thin layer of semi‐rigid polycarbonate film to facilitate high‐quality imaging and plate‐reader‐based assays, we hypothesized it was essential to prevent denting of the film during the automated seeding process. More importantly, denting the film prohibited us from being capable of auto‐focusing which is required for automated imaging. Therefore, we selected a z‐displacement for the multichannel tips of 0.5 mm below the default plate bottom z‐location.

Figure 3.

Optimization of OT2 parameters to achieve robust seeding in 96/384‐SPOT. a) GFP‐expressing cells were seeded at 30 × 10⁶ cells mL⁻¹ hydrogel using 3 mg mL⁻¹ type I bovine collagen to evaluate the coverage of cell‐gel mixture inside the wells seeded using various parameters. i) The offset of tip z‐axis, ii) tip hold time on the paper scaffold after dispensing, iii) 96‐SPOT gel volume, and iv) 384‐SPOT gel volume were selected as −0.5 mm, 1 s, 5 μL, and 2 μL, respectively, for further OT2 sequence optimization. Statistical analysis was performed using ANOVA. b) Schematic representation of two OT2 cell‐gel deposition sequences i) sequence 1: droplet formation, in which the cell‐gel droplet (pink) is formed before contact with the scaffold, and ii) sequence 2: direct dispense, in which contact is made before the cell‐gel (pink) is dispensed. c) Assessment of i) intrawell variation using coefficient of variation and ii) intrawell variation using MGV for 96‐SPOT. d) Assessment of i) intrawell variation using coefficient of variation and ii) intrawell variation using MGV for 384‐SPOT. The #2 direct dispense sequence offers consistently better seeding results, especially for 384‐SPOT. Statistical analysis was performed using the t‐test. Mean ± SD of 3 independent experiments.

Next, as it takes more time for a viscous liquid to travel within the pipette tips due to higher interfacial tension, another parameter that needed to be optimized was the holding time of the pipette tip after the initiation of liquid dispensing. We observed that an additional holding time of 1 s after dispensing improved the cell‐gel spreading within the well significantly by decreasing the number of incompletely covered wells (Figure 3a‐ii). Since further increasing the holding time did not significantly improve dispensing results, we decided to apply a 1 s holding time for both 96‐ and 384‐SPOT to prevent clumping of gels inside the tips. We note however that this holding time was optimized for 3 mg mL⁻¹ type I bovine collagen gel, and further optimization might be needed for bio‐matrices with much higher viscosity.

Finally, since we selected a different scaffold and different dispensing equipment compared to the original SPOT device, we assessed the minimum volume of gel required to achieve complete infiltration of the well in the SPOT plate. Previously, with manual seeding, 5 and 1.5 µL cell‐gel mixtures were used for 96/384‐SPOT, respectively. Using the same cell‐gel materials, we tested a range of seeding volumes and imaged the wells. By quantifying the amount of incompletely seeded wells present using each volume, we determined that with the new scaffold material and OT2 manufacturing process, using a seeding volume of 5 and 2 µL for 96/384‐SPOT offered optimal and consistent coverage while still preserving cell materials, which is crucial for precious primary samples (Figure 3a‐iii,iv).

With the basic OT2 parameters optimized, we next set out to identify an appropriate dispensing sequence compatible with 96/384‐SPOT. Since the cell‐gel mixture is kept at 4 °C during the seeding process, we aimed to achieve consistent and reproducible seeding results while minimizing the run time to maintain high cell viability. We extensively assessed two dispensing sequences: #1 droplet formation and #2 direct dispense (Figure 3b). Note that we also tested in pilot studies other dispensing sequences, such as using the blow_out() function, which pushes an extra amount of air after dispensing liquid, however, none of them were optimal as they either produced extra bubbles or extended run time without a noticeable improvement in seeding homogeneity. The first sequence begins with droplet formation herein called “sequence #1,” which most closely mimicked our manual seeding process, where the cell‐gel is first dispensed from a single‐channel pipette above the scaffold in order to form a droplet and then the tip is lowered to contact the paper scaffold to initiate wicking. In contrast, to directly dispense, herein called “sequence # 2,” the tip fully contacts the paper scaffold prior to dispensing the cell‐gel, and then the tip is slowly lifted away from the scaffold to allow wicking. We compared the variation within a well (intrawell) associated with each sequence using the coefficient of variation as described previously while whole‐well MGV was used to assess the well‐to‐well variation (interwell). We found that sequences #1 and #2 offered a similar coefficient of variation for 96‐SPOT, indicative of a comparable uniform spreading of the cell‐gel mixture within each well (Figure 3c‐i). Sequence #1, however, resulted in slightly worse gel deposition into the scaffold as we observed a slightly lower but consistent MGV across wells (Figure 3c‐ii). Interestingly, we observed a more drastic intrawell difference between the dispense sequences in 384‐SPOT, likely due to the small volume used (i.e., 2 µL). Further, sequence #1 failed to deliver an adequate amount of gel into the scaffold, demonstrated by a low MGV consistently (Figure 3c‐iii,iv). We speculate that this was due to the use of such small volumes in which the formed droplet tended to distribute along the pipette tip, potentially due to the interfacial tension. This was not an issue with manual seeding on 384‐SPOT since the user could establish an angle between pipette tips and paper scaffold ensuring the entire gel volume wicked into the scaffold. In contrast, when the tip was perpendicular to the scaffold during automated seeding this was not possible. Based on these observations, we selected to move forward with sequence #2 as direct dispensing of the cell‐gel offered optimal seeding for both 96 and 384‐SPOT.

2.3. Automatically Generated Microtissues Offer Similar Homogeneity and Cell Survival as Microtissues Created Using Manual Seeding

We next set out to compare our optimized OT2 seeding sequence to manual seeding performed by an experienced user. To be robust and reliable, a screening platform needs to exhibit low intrawell and interwell variation. To measure the robustness of these two features, as described above, we used two image‐based metrics, coefficient of variation and whole‐well MGV, to quantify GFP‐expressing KP4 cells distributed within each well of 96/384‐SPOT on day 0. We observed that seeding of both 96‐SPOT and 384‐SPOT with OT2 resulted in a lower or at least similar intrawell variation compared to manual seeding (Figure 4a‐i,ii). Notably, the interwell variation was similar between OT2 seeding and manual seeding (Figure 4a‐iii,iv). The improvement in intrawell variation associated with OT2 seeding could be explained by more precise and consistent control of the pipette movement when using OT2 compared to manual seeding.

Figure 4.

Benchmarking optimized OT2 automated seeding to manual seeding. a) GFP‐expressing cells were seeded at 30 × 10⁶ cells mL⁻¹ hydrogel using 3 mg mL⁻¹ type I bovine collagen into 96/384‐SPOT either by the optimized OT2 sequence or manually by an experienced user. 96‐SPOT seeded by the OT2 offered a significantly lower intrawell variation. A similar interwell variation was observed in 96‐SPOT using both automation and manual seeding. 384‐SPOT seeded by the OT2 had similar intrawell and interwell variations. Statistical analysis was performed using a Student t‐test. b) Representative images of live/dead staining using calcein‐AM and propidium iodide on day 0 and day 3 for OT2 and manual seeded cells. c) The proportion of dead cells was estimated by obtaining the ratio of dead cells' MGV over live cells' MGV. The OT2 seeded wells showed more cell death only on day 0. d) When focusing on the live cell population only, cells seeded by manual and OT2 demonstrated a similar growth pattern characterized by the mean gray value of live cells. Statistical analysis was performed using a Student t‐test. Mean ± SD of 3 independent experiments.

While our analysis suggested our OT2 seeding sequence could produce even cell seeding, we also wanted to confirm that the automated dispensing sequence did not cause dramatic cell death and hinder cell growth. To do this, we assessed cell viability using calcium‐AM (staining live cells) and propidium iodide (staining dead cells) cell stains on both day 0 and day 3 for OT2 and manually seeded cells (Figure 4b). We measured the ratio between the MGV of dead cells over live cells and compared this metric (i.e., live/dead ratio) between manual and OT2‐seeded SPOT microtissues. We found that OT2 seeding did introduced a significantly higher percentage of dead cells at day 0 (Figure 4c). We speculate that this difference may result from the stress experienced by the cells during seeding using the OT2 as the tips are perpendicularly in contact with the paper scaffold which leads to a temporarily high internal pressure inside the tips. However, the difference in cell death between scaffolds seeded manually versus using the OT2 became negligible on day 3. We also measured cell growth over time to verify that OT2 seeding did not affect the growth kinetic of the seeded cells. To this end, we measured the MGV of tissues seeded with OT2 or manually after either 0 days or 3 days and observed no differences at any time point between both the OT2 and manual methods (Figure 4d).

2.4. Automated Dispensing Sequences are Reproducible in 96/384‐SPOT

Having finalized the various parameters for the OT2 we next wanted to assess the reproducibility of our automated OT2 seeding sequence for seeding entire plates. To do this, we seeded 3 entire plates of 96/384‐SPOT respectively using the OT2 and quantified heterogeneity using our image‐based metrics. Note that after the initial mixing steps, we added an additional mixing step of the cell‐gel stock containing the GFP‐expressing KP4 cells every six depositions to prevent cell pellet formation at the bottom of the tube. We acquired widefield images after 45 min of gelation and the addition of media. The representative images of an entire seeded plate of 96/384‐SPOT are shown in Figure 5a,b, respectively. By characterizing intrawell variance using the coefficient of variation across 3 independently assembled and seeded plates, we found no significant differences among rows or columns using one‐way ANOVA for both 96‐ and 384‐SPOT (Figure 5c,d). Similarly, we observed no significant difference in interwell variance based on MGV measurements of the whole well for both SPOT plates (Figure 5e,f). We were therefore confident that the optimized OT2 sequences could offer reproducible seeding results comparable to those achieved using manual seeding.^{[ 22 ]}

Figure 5.

Assessment of fabrication variation associated with 96/384‐SPOT generated using optimized OT2 sequence. a) Representative widefield image showing 96‐SPOT and b) 384‐SPOT seeded using the optimized OT2 sequence with GFP‐expressing KP4 cells (green) at 30 × 10⁶ cells mL⁻¹. The scale bar is 5 mm. c) Assessment of coefficient of variation measured from widefield images of 96‐SPOT on day 0. Mean ±SD of 3 independent experiments are organized by i) columns and ii) rows for 96‐SPOT (N = 3). ANOVA revealed no statistical significance between columns or rows across 3 independent experiments. d) Similarly consistent and reproducible results were obtained for 384‐SPOT in i) columns and ii) rows. e,f) The interwell variation for both 96/384‐SPOT was quantified based on the MGV. No statistically significant differences were observed between i) columns and ii) rows. Statistical analysis was performed using ANOVA. Mean ± SD of 3 independent experiments.

We also assessed the speed of our workflow to ensure it was within a range to feasibly generate multiple 96 or 384‐SPOT plates at once to facilitate larger‐scale high‐throughput screening. We found it took less than 4 min to seed one 96‐SPOT and no more than 12 min to seed one 384‐SPOT, including the minimum necessary gel mixing steps between 6‐repeated dispensing steps to ensure a homogenous cell‐gel stock. These timeframes are typically our targets when we use a manual seeding process however in reality this seeding pace presents a major logistical challenge for the manual process when attempting to seed a full plate and not simply a few rows within a plate. Further, the OT2 automated seeding enabled consistent run‐time from batch to batch, which we anticipate produced more consistent cell viability across plates. Also, the OT2 automated seeding allowed for more consistent well to well seeding run‐time which is impossible to achieve through manual seeding. Previously, 384‐SPOT manual dispensing was performed using an electronic 8‐channel pipette to minimize seeding time and variability. However, achieving optimal seeding consistently using this approach was very challenging for even an experienced user, in part because of the small well size and the risk of misplacing cell‐gel droplets to the side of the well. Since the gel dispensing location can be finely controlled by the OT2, the ease of use of the 384‐SPOT at these seeding rates was drastically improved. Further, the OT2 dispensing sequence provides control over the exact holding time and the pipette lifting speed away from the scaffold, which may also minimize potential variation.

2.5. Automated Workflow for Single‐Cell‐Based End‐Point Analyses in 96‐SPOT

It is useful to be able to capture single‐cell information on top of bulk measurements to decipher cellular heterogeneity in various tumour microenvironments, particularly a microenvironment containing multiple cell types. We therefore set out to show the reliability and power of single‐cell‐based analyses in our automated workflow and SPOT using 3 orthogonal approaches (Figure 6a): An image‐based readout, high‐throughput flow cytometry, and CyTOF. This experimental design validated the robust phenotype observed in SPOT and showed that each technique was compatible with the high‐throughput and integrated automation approaches we have optimized.

Figure 6.

The use of the OT2 to automate cell extraction and facilitate high‐throughput end‐point single‐cell analyses of the 96‐SPOT. a) The workflow of proof‐of‐value single‐cell‐based analyses compatible with OT2 and SPOT. A total cell density of 15 × 10⁶ cells mL⁻¹ in 6 mg mL⁻¹ bovine collagen was seeded into the 96‐SPOT with various tumour and stromal population ratios, KP4 monoculture, 30% PSCs, 50% PSCs, and PSCs monoculture, using the OT2. After 3 days of co‐culture, wells or cells were processed accordingly for downstream high‐content image analysis, high‐throughput flow cytometry or multiplexed CyTOF after barcoding. b) Using Ki67 expression as a surrogate for proliferative cells, the number of GFP⁺ Ki67⁺ (proliferative KP4 cells) and GFP⁺ cells (KP4 cells) in each stromal condition were quantified. After normalization to KP4 monoculture, an increasing trend of proliferative cells was observed, particularly in the 50% PSCs group. c) Cells could be digested out for high‐throughput flow cytometry analysis. After 3 days of co‐culture, an EdU assay was performed, and cells were digested out of the paper scaffold by the OT2. A similar increasing percentage of proliferative KP4 cells was also observed based on high‐throughput flow cytometry when co‐cultured with a higher percentage of PSCs. d) Similarly, multiplexed CyTOF analysis also captured the same effect of stimulated proliferation by the presence of PSCs. e–g) Assessment of EMT markers on KP4 cells using CyTOF revealed upregulation in mesenchymal markers (Vimentin and Laminin) was correlated with the increasing percentage of PSCs. Correspondingly, the epithelial marker (Pan‐CK) in KP4 cells was downregulated significantly when co‐cultured with PSCs. h) At the same time, PSCs were further activated when co‐cultured with more KP4 cells, as shown in the upregulation in FAP. Statistical analysis was performed using ANOVA. Mean ± SD of 3 independent experiments.

One advantage of SPOT is the ability to generate meniscus‐free thin microtissues that are compatible with high‐content image analysis. For the first single‐cell technique, we leverage this advantage of SPOT and perform high‐content image analysis of a co‐culture system. Previous work has shown that the presence of stromal content promotes the proliferation of cancer cells.^{[ 42 ]} Here, we hypothesized that we could capture this stromal‐driven increase in proliferation by co‐culturing GFP‐expressing KP4 cells with primary pancreatic stellate cells (PSCs) in the 96‐SPOT using the OT2 pipeline. The single‐cell analysis is particularly useful in this context because it is impossible to use bulk assays like AlamarBlue or Cell‐Titer Glo to capture changes in a specific population in a co‐culture system. As PSCs are more contractive than cancer cells, we used 6 mg mL⁻¹ of type I bovine collagen for tissue manufacturing to ensure a good distribution of PSC cells after 3 days of culture.^{[ 43 ]} Although 6 mg mL⁻¹ type I bovine collagen is slightly more viscous than the 3 mg mL⁻¹ type I bovine collagen that we optimized the OT2 protocols with, no adaptation of the seeding sequences was needed to achieve consistent deposition. In addition to KP4 cell monoculture and PSCs monoculture, we used two coculture ratios to create distinct microenvironments: 70% KP4 cells with 30% PSCs and 50% KP4 cells with 50% PSCs. After 72 h of coculture, the percentage of proliferative KP4 cells was inferred based on Ki67 expression, which is a nuclear protein related to cell proliferation. Using the image analysis pipeline as shown in (Figure S3a, Supporting Information), we were able to segment individual single cells based on 10x confocal images. Using GFP protein expression and Ki67 expression as masks, the total number of KP4 cells and the number of proliferative KP4 cells specifically were identified in the co‐cultures. As expected, after normalizing all the data to the KP4 cell monoculture, we observed an increasing percentage of proliferative Ki67⁺ KP4 cells with increasing proportions of PSCs (Figure 6b). We were also able to obtain the histogram of Ki67 expression level in all KP4 cells to assess cellular heterogeneity. A distribution shift to create a noticeable feature in the Ki67⁺ cell region was observed with an increasing ratio of PSCs, consistent with observed increased percentage of proliferative KP4 cells in our analysis above (Figure S3b–d, Supporting Information). This highlights the capability of our workflow to capture the within‐well heterogeneity of individual cells. Although only end‐point analysis was demonstrated here, this high‐content image analysis could be applied to real‐time acquisition images or analysis of single‐cell data containing an appropriate cellular fluorescence reporter. We note however for single‐cell‐image‐based analysis, imaging using a high‐throughput confocal microscope is likely required to obtain images with the appropriate resolution for currently available image segmentation algorithms.

We also obtained non‐image‐based single‐cell readouts using flow cytometry and CyTOF. Since the OT2 is designed to transfer liquids, we reasoned that in addition to microtissue manufacturing, the OT2 could assist in the other steps in these high‐throughput screening assays that require single‐cell extraction for downstream analyses. We therefore set out to optimize a digestion sequence to enable the recovery of single cells for downstream end‐point analysis. Previous work^{[ 22} , ²³ , ^{44 ]} using paper scaffolds infiltrated with 5 µL of cell‐gel showed successful gel digestion and cell retrieval using 1 mL of a solution containing collagenase, protease and DNAase enzymes incubated at 37 °C for 45 min in a thermomixer, and followed by vigorous pipetting. We applied 200 µL of the same digestion solution with adapted protocols (i.e., no need for a plate shaker during incubation) but focused on optimizing pipetting sequences to be compatible with 96‐SPOT digestion. Since the OT2 has an upper limit on liquid dispensing speed to prevent contamination between wells, it is not feasible to achieve the same level of vigorous pipetting as manual pipetting. First, we assessed whether pipetting with the maximum dispensing speed at variable locations (4 corners and center) within one well would improve the cell recovery rate compared with the same amount of pipetting at a fixed central location. On day 0, 96‐SPOT was seeded with a 5 µL cell‐gel mixture and subsequently digested using the OT2. As shown in Figure S4a, Supporting Information, pipetting at variable locations significantly improved the number of recovered cells. Previously, manual digestion required repeated pipetting 20 times to disrupt the remnant gel and release cells from the paper scaffold. We tested a range of pipetting times at variable locations within the well and found that repeated pipetting 50 times by the OT2 recovered the highest number of cells among all 3 pipetting times, which was also the only condition that offered similar recovery levels to the manual pipetting method (Figure S4b, Supporting Information). More than 30 000 cells were harvested from each well after the washing process, providing a practically useful cell quantity for common downstream analysis.

Interestingly, both insufficient and excess pipetting led to a significantly lower cell recovery rate than manual digestion. We speculate this was likely due to the shear stress the cells experienced during excessive pipetting, leading to a smaller number of recovered whole cells. Here, we did not explore the possibility of various digestion formulas, as our main aim was to replace repetitive manual pipetting with an automated sequence specifically for the 96‐SPOT with 3 mg mL⁻¹ type I bovine collagen. Similar digestion results and cell recovery rates were achieved as reported in literature using paper scaffold‐based models and collagen as matrix.^{[ 23 ]} Notably, our paper scaffold is patterned with PMMA instead of wax used in other paper‐based platforms.^{[ 20} , ^{21 ]} It was previously reported that wax‐patterned scaffold could introduce particulates during digestion steps^{[ 45 ]} which would impose a significant challenge for automating wax‐patterned platforms for downstream analysis. In SPOT, we rarely observed any obvious PMMA debris after digestion.

Having identified a usable automated digestion protocol, we next performed a similar proof‐of‐concept co‐culture experiment to test the entire single‐cell digestion and analysis platform workflow using GFP‐expressing KP4 and blue fluorescence protein (BFP)‐expressing PSCs at the same ratio mentioned previously. After seeding gels containing various stromal and cancer cell compositions, cells were co‐cultured for 72 h then EdU, a nucleoside analog that marks proliferative cells, was added for 3 h and then digestion of the tissues was performed using the OT2. We also used the OT2 to assist in fixation, washing and the EdU reaction steps to streamline the process and ensure the downstream results were more consistent. Since we used GFP‐expressing KP4 and BFP‐expressing PSCs in this study, we could easily separate the two populations during flow cytometry and only quantify the number of proliferative cells in the KP4 population. Similar to our high‐content image analysis, we observed an increase in GFP‐expressing KP4 cells proliferation when the ratio of PSCs in the coculture increased (Figure 6c).

While high‐throughput flow cytometry and high‐content image analysis offer single‐cell‐based readout, the number of markers that can be assessed simultaneously is highly limited due to the wide spectrum of fluorophores used for these techniques. CyTOF leverages metal isotopes to enable more than 40 markers per cell without concerns about spectral overlap. Recently, with automated workflow, it has also become feasible to multiplex samples for high‐throughput acquisition with metal isotope barcodes.^{[ 46} , ^{47 ]} Adapting a similar methodology but with nine lanthanide‐conjugated mDOTA barcode probes, we were able to create 96 distinct barcode combinations with 4 metal barcode probes per well using OT2.^{[ 46 ]} To verify the accuracy of the barcoding and de‐barcoding process, we created two patterns of wells with or without IdU, which functions similar to EdU but is compatible with CyTOF acquisition (Figure S5a,b, Supporting Information). After 1 h of IdU incubation, cells were digested, fixed, and permeabilized before being barcoded with various metal barcode probe combinations. Upon decoding the barcodes, we observed the intended pattern of IdU positive and negative wells confirming the accuracy of the barcoding method (Figure S5a,b, Supporting Information).^{[ 48} , ^{49 ]} Further, we confirmed that the number of cells acquired from individual wells after de‐barcoding was sufficient for downstream analyses (Figure S5c,d, Supporting Information). Moreover, the distribution of cells across wells was Normal, as shown in (Figure S5e,f, Supporting Information), indicating no cell recovery preferences in any specific wells after the digestion, barcoding, and de‐barcoding processes.

Using the same tissue fabrication workflow as described above for the high‐content image analysis, we also looked at the effect of co‐culture using CyTOF. IdU, which marks proliferative cells, was added for 3 h after 72 h of co‐culture. Then cells were digested using the optimized OT2 protocol, followed by barcoding before pooling all single cells from all wells together for multiplexed CyTOF acquisition. The KP4 cells and PSCs were distinguished based on GFP signature and fibronectin expression (Figure S6a, Supporting Information). Both KP4 and PSC cell populations were captured from the various co‐culture compositions (Figure S6b, Supporting Information). Further, an increasing number of proliferative KP4 cells were observed with an increasing ratio of PSCs in the coculture (Figure 6d), similar to our observations with both high‐throughput flow cytometry and high‐content image analysis. In addition to proliferation, we included several epithelial‐mesenchymal transition (EMT) markers in the CyTOF panels since PSCs promote the EMT of pancreatic cancer cells.^{[ 50 ]} The loss of key epithelial markers, like pan‐cytokeratin, and the gain of mesenchymal markers, like vimentin and laminin, are characteristics of EMT.^{[ 51} , ^{52 ]} As shown in Figure 6e–g, we observed a decrease in epithelial markers and an increase in mesenchymal markers that correlated with an increasing ratio of PSCs. Simultaneously, we measured increasing FAP expression level in the PSCs correlated with the concentrations of KP4 cells (Figure 6h, correlation coefficient of 0.980). FAP is a known activation marker for PSCs that should increase upon exposure to more tumour cells. Note that one limitation of this proof‐of‐concept study was that the digestion protocol was optimized for the extraction of KP4 cells and not the PSC population thus the seeding population ratio was not precisely preserved and the number of PSCs recovered was lower. We also only included six markers to show the value of multiplexed CyTOF which is capable of around 40 markers per cells, however, our study illustrates the potential high parameter phenotyping of co‐cultures in paper scaffolds using the OT2 for a single‐cell‐based high‐content and high‐throughput downstream analysis.

Here, we demonstrated that our workflow could capture physiologically relevant effects using 3 single‐cell‐based orthogonal methods. The compatibility of our workflow with multiplexed CyTOF offers great potential in deciphering biological effects within complex tumour environments, such as with the presence of stromal cells and immune cells. We expect our method to be compatible with other single‐cell‐based analyses, such as single‐cell RNAseq, to comprehensively assess gene expression on a cellular level. In addition to the whole cell analysis performed here, this automated protocol could also be applicable to bulk assays like qPCR to assess gene expression within individual wells. To perform qPCR, cells must be lysed for RNA extraction and therefore do not require the use of an optimized cell digestion protocol which preserves whole cells.^{[ 53 ]}

2.6. Automated Dispensing Sequences Support the Growth of Organoid‐Derived Cells

With OT2 workflows optimized for cell lines, we next aimed to demonstrate the applicability of this workflow to more precious and fragile primary cultures, such as PDOs. To do this, we used a previously tested hydrogel blend of 3 mg mL⁻¹ type I bovine collagen (75%) and Matrigel (25%) as the matrix for seeding pancreatic ductal adenocarcinoma (PDAC) organoid cells (3 × 10⁶ cells mL⁻¹ of collagen‐Matrigel blend) into SPOT.^{[ 17 ]} Despite the increased temperature sensitivity of the Matrigel gel blend, we observed consistent and comparable spreading of cells as measured by the coefficient of variation between manually seeded and OT2‐seeded microtissues (Figure S7a, Supporting Information). Next, we quantified organoid cell growth in SPOT using the AlamarBlue assay and the MGV of the whole well over 12 days and observed a similar growth trend using both assays (Figure S7b, Supporting Information). As shown by representative images in Figure 7a, GFP‐expressing organoid‐derived cells exhibited unrestrained growth over 12 days. The unrestrained growth was further confirmed based on MGV of GFP signals (Figure 7b). We confirmed organoid cell viability after seeding by performing a dead cell stain using propidium iodide (PI) while using the GFP expression as a surrogate for live cells every 4 days over the course of 12 days (Figure 7a). No statistically significant increase in cell death was observed over time while organoid‐derived cells continuously proliferated, as indicated by the increase in GFP MGV and increase of PI MGV in 70% ethanol‐treated apoptotic controls (Figure 7c). CK19 and ZO‐1 expressions were present as expected for PDAC organoids after 12 days of culture in SPOT (Figure 7d). Particularly, after long‐term culture, lumens were still present within PDAC organoids, indicating the health of the organoids and the possibility of SPOT sustaining long‐term culture (arrows appointed in Figure 7d). A similar amount of cell death was observed as with manual seeding up to 4 days, suggesting the slight decrease in viability after seeding was not caused by the OT2 manipulation (Figure S7c, Supporting Information) but rather likely due to the higher sensitivity of organoid cells compared to cell lines during standard culture manipulations. This was further confirmed by the similar growth trend of organoid cells over 6 days when seeded by the OT2 versus the manual method (Figure S7d, Supporting Information). Significantly, the OT2 seeding sequence was compatible with primary samples without requiring modifications. While we only tested one PDAC organoid line, we speculate that our workflow is likely appropriate for organoids originating from different PDAC patients, other cancer types, or from healthy tissues compatible with a similar matrix composition that facilitates wicking seeding. We also note that single‐cell analysis of organoids using the image‐based metric is a significant challenge and the approach described above for KP4 cells will require the integration of more effective image segmentation analysis approaches given the compact density of cells in organoid structures compared to the cell line used here. The quality of the confocal images achievable from SPOT cultures however will provide a useful dataset as such segmentation algorithms become available. Given previous work from our lab showing the compatibility of our digestion protocols for enabling flow‐cytometry and CyTOF,^{[ 43 ]} we anticipate these workflows will enable single‐cell organoid analysis.

Figure 7.

Translation of automated OT2 pipeline to incorporate fragile patient‐derived organoid cells. a) GFP‐expressing PPTO.46 cells were seeded at 3 × 10⁶ cells mL⁻¹ in a hydrogel blend (3 mg mL⁻¹ bovine collagen (75%) and Matrigel (25%). Representative images of GFP expressing PPTO.46 seeded using OT2 on days 0, 4, 8, and 12. Dead cells (red) were stained with propidium iodide, while GFP expression was used as the surrogate for live cells (green). Cells were treated with ice‐cold 70% ethanol for 10 min to induce complete cell death at each time point as a negative control. b) Growth curve for GFP‐expressing PPTO.46 cells after being seeded using the OT2 in SPOT for up to 12 days based on MGV. c) The dead cells were quantified based on the MGV of propidium iodide staining. There was no apparent increase in dead cells over the 12‐day culture compared to the dead control, confirming SPOT platforms coupling with OT2 seeding support patient‐derived organoid cell culture. d) The health/polarization of organoids on day 12 was assessed based on CK19 and ZO‐1 immunostaining. Confocal images were taken with a 20x objective. As expected, CK19 expression was seen at the edge of organoids. Lumens were observed (white arrows) based on ZO‐1 staining indicating a healthy status of PDAC organoids. Plots show the mean ±SD of 3 independent experiments.

In this report, we describe the automation of both 96 and 384‐SPOT. We envision that 384‐SPOT could serve as a platform for medium to high‐throughput screening relying on time‐course imaging analysis or plate‐reader‐based assays (such as AlamarBlue and Cell‐Titer Glo) as metrics to distinguish hits. On the other hand, 96‐SPOT contains sufficient cellular material to offer the possibility for single‐cell‐based end‐point analysis to decipher specific gene expression or analyze based on cell types, for example in a co‐culture model. Notably, the established seeding pipeline is optimized based on a P20 8‐channel pipette; thus, the number of distinct tissue compositions that can be generated within each plate is limited. In the future, it could be possible to incorporate a single channel pipette for cell‐gel deposition, however, the run time could be 8 times longer; thus, it might not be ideal for cell survival, particularly in fragile primary samples. Thus, the most optimal application for the current pipeline is likely using a few distinct tumour compositions per plate. The use of conditioned media from other cell types is also possible to add another layer of complexity to the microenvironment. This would allow the optimized OT2 downstream analysis protocols, rapid and consistent single‐cell‐based analysis, to be used to decipher the effect of specific tumour microenvironment effects on a subpopulation of cells. Further, given we have shown our approach is compatible with primary patient samples, we envision applications to investigate patient‐level heterogeneity in therapy response. This could potentially open a new avenue for screening patient‐specific drug targets in a high‐throughput but biologically relevant system.

We note that this report specifically describes our steps to optimize the automation of SPOT manufacturing and analysis for pancreatic tumour cells in a collagen matrix blend, providing a starting point with key parameters that impact manufacturing quality. Alterations in cell type or matrix will likely require slight adjustments to the parameters we selected here and optimization of alternative cell digestion protocols depending on the specific cell types and matrix being used. Conveniently, the OT2 workflow developed here was compatible with all 3 different viscous hydrogel compositions we used without the need for protocol adjustments. There is a considerable effort within the community to generate biomatrix alternatives to replace commonly used Matrigel since it cannot be easily standardized.^{[ 54} , ^{55 ]} Many of these alternative bio‐matrices are viscous hydrogels or hydrogel precursors therefore automated tissue manufacturing methods compatible with these more viscous materials will likely be important as the use of these alternative gels increases.^{[ 56 ]} We anticipate our workflow is compatible with many of these synthetic gels, particularly those with a thermal gelation step.

3. Conclusion

In this study, we present an automated pipeline using a commercially available liquid handler to streamline all the steps from the fabrication of 3D tissue arrays to their analyses. By combining the use of the SPOT and the OT2 liquid handler, we eliminated labour‐intensive steps while preserving the consistency and robustness of fabricated tissues across wells and plates. Beyond executing plate‐reader‐based assay to assess bulk response, this automated workflow allowed us to probe the effect of tumour microenvironment cues on specific cell types through high‐content image analysis, high‐throughput flow cytometry, and multiplexed CyTOF at a single‐cell level. Similar biological effects were captured in all 3 orthogonal methods. The compatibility of multiplexed CyTOF with the SPOT and automated workflow offers a great opportunity for high‐throughput proteomic screening on various tissue compositions/treatments. Furthermore, we demonstrated that this workflow supports the incorporation of patient‐derived primary samples with minimal adaptation. This highlights the potential application of this model for investigating patient‐level heterogeneity to assist in personalized medicine discovery and clinical management. Although no single‐cell analyses were performed with PDOs seeded in a paper scaffold in this study, it will be relatively simple to adapt and perform personalized single‐cell analysis using CyTOF, as similar studies have been performed on organoids but at a smaller scale.^{[ 43 ]} Further, the described workflow takes advantage of an open‐source and inexpensive OT2 liquid handler, to allow for easy implementation into existing drug discovery pipelines and research facilities with the possibility of combining with big‐data analysis in the future. We envision that our work will offer new avenues for discovering novel personalized medicines.

4. Experimental Section

Cell Culture

The KP4 pancreatic tumour cell line (JCRB Cellbank) was cultured in IMDM containing 10% fetal bovine serum (FBS) (Fisher Scientific, USA), and 1 µg mL⁻¹ penicillin and streptomycin (Sigma‐Aldrich, Canada). The PSCs (ref#3830, ScienceCell) were cultured in DMEM containing 10% FBS and 1 µg mL⁻¹ penicillin and streptomycin. PSCs were used for a maximum of 12 passages. Cells were passaged every 3 to 4 days. PDAC organoids established from PDAC patients were obtained from the UHN Biobank at Princess Margaret Cancer Centre (PMLB identifier PPTO.46 from the University Health Network, Canada) under a protocol in compliance with the University of Toronto Research Ethics Board guidelines (protocol #36107). Organoid cultures were maintained in Advanced DMEM/F‐12 (Gibco, ThermoFisher Scientific, USA) supplemented with 2 mm GlutaMAX, 0.5 µm A83‐01 (Tocris Biosciences, Bristol, United Kingdom), 10 nm Gastrin I (1‐14) (Sigma‐Aldrich, USA), 10 mm HEPES (Gibco), 1% penicillin/streptomycin, 1X W21 supplement (Wisent), 1.25 mm N‐acetyl‐l‐cysteine, 10 µm Y‐27632 (Selleck Chemicals, USA), 10 mm nicotinamide (Sigma‐Aldrich), 20% v/v Wnt‐3a conditioned media, 30% v/v human R‐spondin1 conditioned media (Princess Margaret Living Biobank, Canada), 50 ng mL⁻¹ recombinant human EGF, 100 ng mL⁻¹ recombinant human noggin, and 100 ng mL⁻¹ recombinant human FGF‐10 (Peprotech, USA). PPTO.46 PDAC organoid cells were cultured in 48‐well polystyrene plates in 40 µL domes of Growth Factor Reduced Phenol Red‐Free Matrigel Matrix (Corning Life Sciences, Corning, USA) with 500 µL of complete media. PPTO.46 PDAC organoid cells were passaged once per week (1:8 split ratio). PDAC organoids were used for a maximum of 40 passages. All cultures were maintained in a humidified atmosphere at 37 °C and 5% CO2.

Lentivirus Production and Cell Transduction

KP4 cells and PPTO.46 were transduced to express GFP, while PSCs were transduced to express BFP. GFP lentivirus was produced using calcium phosphate co‐transfection of HEK293T cells with the psPAx2 plasmid (packaging vector, Addgene #12 260), pMd2g plasmid (VSVG envelope, Addgene #12 259), and the pLenti‐CMV‐GFP‐Puro plasmid (Addgene #17 448) as previously described.^{[ 57 ]} BFP lentivirus was produced similarly with pBFP2‐IRES‐Neo (Addgene #108 175). Supernatants containing viral particles were harvested 48 h after transfection and concentrated using a 20 mL 100 000 MWCO spin column (Vivaspin, USA). Wild‐type KP4 cells were transduced with the concentrated virus, and GFP‐expressing cells were selected using puromycin selection (1 µg mL⁻¹, Sigma‐Aldrich, USA). Wild‐type PSC cells were transduced similarly and Neomycin (300 µg mL⁻¹, ThermalFisher, USA) was used for selection. GFP‐expressing PDOs were created using the same virus as KP4 but with a slightly different transduction protocol. PDOs were suspended in infection media and spinoculated for 1 h at 600 g at 32 °C, before being resuspended and incubated for an additional 6 h at 37 °C in 5% CO2. PDOs were then washed and seeded into Matrigel with fresh media. 48 h after infection, cells were sorted for the middle 80% of the GFP‐expressing population.

Scaffold‐Supported Platform for Organoid‐Based Tissues Component Fabrication and Assembly

The 96/384‐SPOT plates were fabricated as previously described.^{[ 22 ]} Cellulose scaffolds were partially infiltrated with a PMMA (120 KDa, Sigma‐Aldrich, USA) solution in acetone (Sigma‐Aldrich, USA) to block the inter‐well regions using the AxiDraw V3 (Evil Mad Scientists, USA). The PMMA‐acetone solution (1.75 g PMMA to 7.84 mL acetone) was mixed with blue nail polish (Essie) for visibility. For the 96‐SPOT, the solution was dispensed through a 20‐gauge standard Luer Lock tip (McMaster‐Carr, USA) in constant contact with the cellulose scaffold controlled by the Inkscape (https://inkscape.org/) through an AxiDraw extension (https://wiki.evilmadscientist.com/Axidraw_Software_Installation). As the printhead moved and acetone quickly evaporated, a sheet of PMMA that blocks the pores in the scaffold was left behind making it impermeable. For the 384‐SPOT, the solution was dispensed through a 16‐gauge standard Luer Lock tip (McMaster‐Carr, USA) controlled directly using an interactive python API (https://axidraw.com/doc/py_api/#introduction). For each plate, two sheets of double‐sided, poly‐acrylic adhesive tape (Adhesive Research, ARcare, Catalog # 90106NB) were cut with an array of 96 circular holes or an array of 384 square holes corresponding to the no‐bottom 96 well plate (Greiner, Catalog #82050‐714) or no‐bottom 384 well plate (Greiner, Catalog # 82051–262), respectively, using the CO₂ Laser Cutter (VLS3.60, Universal Laser Systems, USA) with 50% laser power and 35% speed. One side of the adhesive tape covers was engraved with two lines running across the interwell space between columns 4–5 and 6–7 for 96‐SPOT and columns 9–10 and 13–14 for 384‐SPOT with 7% laser power and 100% speed to assist in assembly alignment. All components were UV sterilized and assembled in a sterile environment. The double‐sided tape was applied to the PMMA‐patterned scaffold sheet with the support of PDMS, followed by another layer of double‐sided tape on the other side of the scaffold. Then, the sandwiched structure was aligned to the no‐bottom well plate with the aid of engraved lines as the middle tape cover was removed first to align and then the two side pieces were removed to adhere fully. Finally, a thin, transparent polycarbonate film (McMaster‐Carr, Catalog # 85585K102) was cut to 110 mm × 75 mm and attached to the bottom of the well plate. The assembled well plates were clamped prior to use to prevent delamination of the layers.

Paper Scaffolds Physical Properties Characterization

Scaffold I (Miniminit Products, R10, Scarborough, Canada), the original scaffold used in 96/384‐SPOT, scaffold II (Teeli Bag Filter Large Square, Riensch & Held GmbH & Co.KG, Germany), scaffold III (Finum Tea Filters XL, Riensch & Held GmbH & Co.KG, Germany), and scaffold IV (TR Coffee FIL WS 17.0 CR NAT, Twin Rivers Paper Company, USA) were assessed based on physical and physiological properties. SEM images were obtained using a Hitachi SEM SU3500 (Hitachi High‐Technologies Canada Inc., Canada). Samples were mounted onto carbon‐tape coated stubs and gold–palladium sputter coated for 55 s using a Bal‐Tec SCD050 Sample Sputter Coater (Leica Biosystems, USA) and then imaged at 10 kVD. The autofluorescence of each scaffold was imaged at the center of a 96‐SPOT well fabricated with corresponding scaffolds using an ImageXpress Pico (IXP) (Molecular Devices, USA), a high‐content imaging system using both laser‐based and image‐based autofocus settings (16‐bit), with 300 ms exposure acquired at 4X. To quantify the pore area coverage, brightfield images with 10 ms exposure were taken with the IXP microscope. The images were thresholded using OTSU methods on ImageJ to obtain particles and their coverage area. Edge particles and any particles smaller than 890 µm² were excluded from the analysis. The wicking ability of paper scaffolds was assessed on assembled 96‐SPOT. The amount of time required for 5 µL of water‐based red ink (Ecoline) to cover one well was obtained from observation of video recordings.

Cellulose Scaffold Seeding

Collagen hydrogel was prepared by mixing 8 parts 3 mg mL⁻¹ type I bovine collagen (PureCol, Advanced BioMatrix) or 6 mg mL⁻¹ type I bovine collagen (Nutragen, Advanced BioMatrix), with 1 part 10× minimal essential medium (MEM, Life Technologies, Grand Island, USA) by volume and neutralizing to pH 7 endpoint with 0.8 m NaHCO3 (Sigma‐Aldrich). The solution was kept on ice. Following a standard trypsinization protocol, the adherent cells, KP4 and PSCs, were pelleted by centrifugation (300 g, 5 min, 4 °C) and re‐suspended in an appropriate volume of 3 and 6 mg mL⁻¹ collagen, respectively, to achieve the desired cell concentration. In the case of organoids, the PPTO.46 cells were collected by mechanical disruption of the Matrigel domes followed by a 10 min incubation in TrypLE Express (Gibco) at 37 °C. This process was repeated until mostly single cells were observed under the microscope (a total of 3 times). Cells were pelleted by centrifugation (300 g, 5 min, 4 °C). In accordance with^{[ 58 ]} the PPTO.46 cell pellet was re‐suspended in an appropriate volume of a hydrogel blend to achieve the desired cell concentration. The hydrogel blend used for PPTO.46 cells comprised 25% Matrigel and 75% 3 mg mL⁻¹ collagen hydrogel, as reported previously.^{[ 58 ]}

As described previously,^{[ 22 ]} 70 µL of sterile PBS was added to each inter‐well space in the 96‐well plate to prevent evaporation while seeding before use, and then the plate was chilled for 30 min to ensure no clumping of cell‐gel solution during seeding. For manual seeding, a single‐channel micropipette (Gilson) was used to deposit 5 µL of cell‐gel into the center of each 96‐SPOT well by carefully dispensing the gel above the paper surface and gently touching the tip of the pipette to the paper. For the 384‐SPOT, a plate membrane (Sigma‐Aldrich, cat no. Z380059) was used to cover each column of wells as they were used to minimize evaporation. An 8‐channel electronic micropipette (Gilson, model no. P8×10M‐BC) was used during manual 384‐SPOT seeding. The electronic micropipette was used to dispense 2 µL of cell‐gel solution in a column of wells similar to the 96‐SPOT with cell‐gel solution first dispensed into an 8‐well PCR strip. Cell‐gel solution was prepared similarly for OT2 seeded plates (Opentrons, United States). The OT2 was calibrated based on manufacturer instruction prior to each run. Two temperature modules (Opentrons, United States) were pre‐set and maintained at 4 °C to keep the SPOT plate and the 8‐well PCR strip cold during the entire operation. A 110 mm × 75 mm × 4.1 mm aluminum plate was customized and used on top of the temperature module (Opentrons, United States) to support the thin polycarbonate plate during seeding along with offering better heat conduction. The protocol was designed based on Opentrons Python API V2 (https://docs.opentrons.com/v2/). The cell‐gel solution stored in 8‐well PCR strips was mixed before seeding and every six depositions to ensure homogenous cell‐gel mixture stock using an 8‐channel P300 pipette head (Opentrons, United States) with the default pipette speed of 92.86 µL s⁻¹. The 8‐channel P20 pipette head (Opentrons, United States) was used for both 96/384‐SPOT seeding. The cell‐gel solution was aspirated and dispensed at 5 µL s⁻¹, and the z‐axis speed of the pipette head was set to be 1 mm s⁻¹ during contacting and moving away from the scaffolds. For both 96/384‐SPOT, after seeding the final well, the well plate was allowed to incubate for 1–2 min on the cold ice pack or temperature module before incubation for 45 min for hydrogel polymerization at 37 °C. Then 200 or 60 µL of media was added to each 96‐well or 384‐well SPOT, respectively, after gelation.

Analysis of Cell Seeding

GFP‐expressing cell (KP4 and PPTO.46) fluorescence was used to assess the variation in seeding between and within wells. Widefield images of GFP fluorescence in each well were acquired at 4X with the IXP. 4 or 1 sites were acquired per 96‐ or 384‐SPOT well, respectively, to capture an entire well bottom, which was stitched automatically by the IXP software. A custom ImageJ script (adapted from refs. [22, 58]) was written to measure the MGV of the GFP signal of 100 randomly selected squares to obtain the standard deviation within one well. Standard deviations were used to calculate the coefficient of variation associated with each well using the following:

$$\begin{array}{ccc}& & \mathrm{Coefficient}\mathrm{of}\mathrm{variation}\left(\%\right)\hfill \\ & & =\frac{\mathrm{Standard}\mathrm{deviation}\mathrm{of}\mathrm{MGV}\mathrm{from}100\mathrm{ROIs}}{\mathrm{Mean}\mathrm{of}100\mathrm{MGV}\mathrm{measurements}\mathrm{from}\mathrm{ROIs}}\times 100\%\hfill \end{array}$$ (1)

Live Dead Assay

Wild‐type KP4 cells were seeded in 96‐SPOT at 5.0 × 10⁶ cells mL⁻¹ for live‐dead assays. 100 µL of 4 µm calcein‐AM (Biolegend, the United States) and 4 µg mL⁻¹ propidium iodide (ThermoFisher, the United States) in IMDM was added to each well and then incubated for 15 min at 37 °C. Then, the media was removed, and the well was washed with PBS before adding fresh media. Widefield images of GFP‐fluorescence and TexasRed‐fluorescence in each well were acquired at 4X with the IXP. The MGV was quantified. Similarly, GFP‐expressing PPTO.46 cells were seeded in 96‐SPOT at 3.0 × 10⁶ cells mL⁻¹ with 25% Matrigel and 75% collagen hydrogel for live‐dead assay. 100 µL of 4 µg mL⁻¹ propidium iodide (ThermoFisher, the United States) in organoid media was added to each well and then incubated for 15 min at 37 °C. The GFP expressed by the cells was used as a surrogate for live‐cell stain. The negative dead cell control was generated by incubating 70% ice‐cold ethanol for 10 min.

Quantification of PPTO.46 Cell Growth in 96‐SPOT

To assess the possibility of long‐term culture in 96‐SPOT, GFP‐expressing PPTO.46 were seeded at 3.0 × 10⁶ cells mL⁻¹ by OT2. Media was changed every 2 days from day 4 to sustain long‐term culture. Images were taken using IXP on days 0, 2, 4, 6, 8, 10, and 12. To quantify cell growth, the MGV of GFP signal was measured for each microtissue using ImageJ. The metabolic quantification AlamarBlue assay (Sigma), following the manufacturer's instructions, was used to assess PPTO.46 growth after being seeded by OT2 and compared with the image‐based growth metrics. Briefly, GFP‐expressing PPTO.46 cells were seeded in 96‐SPOT at 3.0 × 10⁶ cells mL⁻¹. On days 0, 2, 4, and 6, cells were incubated with 200 µL media with 10% AlamarBlue reagent per well for 3 h as determined previously.^{[ 22 ]} Fluorescence intensity was measured at a 3‐h timepoint in a microplate reader at 560/590 nm. Negative control wells containing no cells and positive control wells containing fully‐reduced 10% AlamarBlue reagent. Images of the different sets of cultures were taken using the IXP on days 0, 2, 4, and 6 as well to show that the cell growth was linear over 6‐days of culture and consistent with the AlamarBlue readout.

Immunostaining of PPTO.46

To assess the maturity and polarization of PPTO.46 after long‐term culture, immunostaining of CK19 and ZO‐1 was performed. As previously described, organoid‐derived cells were seeded in 96‐SPOT at 3.0 × 10⁶ cells mL⁻¹ by OT2 with the same media change frequency. After 12 days of culture, microtissues were fixed inside individual SPOT wells with 4% PFA for 10 min at room temperature (RT). Cells were permeabilized with 0.5% Triton‐X 100 (Millipore Sigma) or 0.1% Triton‐X 100 for 5 min at RT for CK19 or Z0‐1 staining, respectively. For CK19 staining, samples were blocked with 5% BSA in IF buffer (0.1% BSA/0.2% Triton‐X 100/0.05% Tween‐20) for 1 h at RT as previously described. Wells for ZO‐1 staining were blocked with 5% BSA in PBS for the same duration. CK19 (Abcam, Catalog# ab9221) and ZO‐1 (Thermal Fihser, Catalog# 339 100) primary antibodies was diluted 1:400 or 1:100 in their respective blocking buffers and incubator at 4 °C overnight. The same secondary antibody Alexa Fluor 594 (Thermal Fisher Scientific) were used for both staining with 1:400 dilution in their respective blocking buffers for 2 h at RT. Nuclei were stained with DRAQ5 (Cell Signalling Technology) 1:500 in PBS for 10 min at RT. Images were then acquired using Leica SP8 confocal at 20x objective (NA 0.75).

Gel Digestion

The digestion protocol was adapted from a previous protocol.^{[ 23} , ^{44 ]} Cell rinse buffer was used to dissolve and dilute collagenase XI (Millipore Sigma, USA, Catalog# C7657), DNase I (Sigma‐Aldrich, USA, Catalog# D4527), and protease (Millipore Sigma, USA, Catalog# P8811). 100 mL of cell rinse solution was prepared by combining 10 mL 10X modified Earle's medium, 5 mL of 50 mm MgCl₂ solution (BioShop, USA, Catalog# 7791‐18‐6), 83 mL of water, and 2 mL of 1 m HEPES (Thermo Fisher Scientific, Waltham, USA, Catalog# 15630‐080) in a sterile 500‐mL bottle. The 10X modified Earle's medium consists of 68 g of NaCl (BioShop, USA, Catalog# 7647‐14‐S), 10 g of glucose (Sigma‐Aldrich, USA, Catalog# 50‐99‐7), 4 g of KCl (BioShop, USA, Catalog# POC 308), and 1.22 g of anhydrous NaH2PO4 (Sigma‐Aldrich, USA, Catalog# RDD007) in 930 mL of water. Then the pH was adjusted to 7.3 by titrating with NaOH solution (Sigma‐Aldrich, USA). 12 mL of digestion solution was prepared by combining 9 mL of cell rinse solution with 1 mL of 2000 U mL⁻¹ collagenase XI, 3000 U mL⁻¹ DNase I and 3 mg mL⁻¹ protease. The 96‐SPOT wells were washed once with PBS and then incubated with 200 µL of digestion solution per well for 45 min at 37 °C. To stop the reaction, 20 µL FBS was added per well. Each well was pipetted vigorously using an 8‐channel pipette manually or by OT2 using an 8‐channel P300 pipette head at 275 µL s⁻¹ to dissociate cells from gel. After transferring cell suspension to a separate well plate, the wells were rewashed with 200 µL 3% BSA in PBS to harvest most of the cells. Cells were spun to remove supernatant at 50 µL s⁻¹ for subsequent fixing in 4% PFA for 15 min. After two runs of PBS washes, the cells were counted or stored at 4 °C for further analysis.

Measurement of Cell Proliferation using the EdU Assay

A total concentration of 1.5 × 10⁷ cells mL⁻¹ was seeded into the 96‐SPOT with various cell compositions: GFP‐expressing KP4 monoculture control, BFP‐expressing PSCs monoculture control, and GFP‐expressing KP4 and BFP‐expressing PSCs cocultured with ratios of 7:3 and 1:1. All the conditions were cultured in DMEM for 3 days before performing the EdU assay. 100 µL DMEM media from each well was discarded and replaced with 100 µL DMEM with 20 µm EdU (Abcam, UK, Catalog# ab219801) to reach a final concentration of 10 µm EdU. After 3 h of incubation at 37 °C, cells were digested out and fixed using the OT2 as described above. Adapted from the manufacturer's instruction (Abcam, UK), the supernatant was removed from fixed cells using the OT2 at 50 µL s⁻¹ after spinning at 800 × g at 4 °C for 5 min. 100 µL 1X permeabilization buffer (Abcam, UK, Catalog# ab219801)) was added to each well and incubated for 15 min at room temperature and then 100 µL of 3% BSA (Sigma‐Aldrich, USA) was added into the well followed by spinning and removing the supernatant. 125 µL EdU reaction solution was added per well, which consisted of 438 µL PBS, 10 µL 100 mm CuSO₄ (Abcam, UK, Catalog# ab219801)), 2.5 µL of 500 µm Texas Red azide (AAT Bioquest, USA), and 50 µL 20 mg mL⁻¹ sodium ascorbate (Abcam, UK, Catalog# ab219801)) per 500 µL solution, and incubated for 30 min at room temperature. Then the cells were washed twice with 3% BSA with the assistance from the OT2 using similar setting as before. Next, cells were re‐suspended and analyzed directly from a V‐bottom well plate (ThermoFisher, USA, Catalog# 249 570) using a LSR Fortessa flow cytometer (BD) coupled to a high‐throughput sampler. KP4 and PSC cells were gated based on GFP and BFP fluorescence, respectively, and EdU+ cells were determined as a percentage for each cell population. Cytometry data were analyzed using FlowJo (version 10.8.1). All the data was normalized to fold change based on (X‐Y)/Y with respect to individual biological replicates, where Y is the KP4 monoculture proliferative cell percentage.

Measurement of Cell Proliferation Using the High‐Content Image Analysis

As described above, GFP‐expressing KP4 cells and wild‐type PSCs were seeded to create various microenvironments. After culturing for 3 days, cells were fixed within the 96‐SPOT well plate and stained for Ki67 to interperate KP4 cells proliferation. Cells were permeablized with 2% Triton‐X 100 for 20 min at RT before being blocked with 3% BSA for another 20 min. Cells were incubated with 1:400 Ki67 primary antibody (Millipore, Catalog# AB9260) overnight at 4 °C. Alexa Fluor 555 (Thermal Fisher Scientific) was diluted 1:250 and incubated with cells for 1 h at RT before nuclei staining with DRAQ5 for 10 min at RT. Six frames of images per well were then acquired using Leica SP8 confocal at 10x objective (NA 0.4). The number of GFP‐expressing KP4 cells and GFP‐expressing Ki67‐positive KP4 cells were segmented and quantified using CellProfiler (4.2.5). The images were processes as shown in Figure S3a, Supporting Information. Data were normalized to fold change accordingly.

Cytometry by Time of Flight Barcoding

Nine lanthanide‐conjugated mDOTA barcodes were used at the following concentration in the creation of a barcode master plate: Y‐89 (6 µm), In115 (3 µm), La139 (2 µm), Pr141 (6 µm), Eu151 (2 µm), Tb159 (1 µm), Ho165 (2 µm), Tm169 (2 µm), and Lu175 (2 µm). Adapted from previously reported method,^{[ 46 ]} 4 out of 9 distinct mDOTA barcode probes were pipetted by OT2 into each well of the 96‐barcode master plate. To test the accuracy of barcoding using this master plate combination, checkerboard pattern and column pattern tests were performed. A total concentration of 3.0 × 10⁷ cells mL⁻¹ was seeded into the 96‐SPOT and cultured for 24 h before 1 h incubation with 50 µm IdU (5‐Iodo‐2’‐deoxyuridine, Fisher Scientific). IdU was added to every other well in a checkerboard pattern, or every other column to create a pattern of cells with or without IdU signals in the end. Cells were digested out as previously mentioned, and then fixed with 4% PFA for 15 min, permeabilized for 30 min with 0.2% saponin, barcoded, washed, and pooled with OT2 protocols. After pooling, samples were stained for DNA using 125 nm of Cell‐ID Intercalator Ir (Standard BioTools, San Francisco, USA) overnight at 4 °C. The next day, cells were washed twice with 1% BSA/PBS and then washed twice with Cell Acquisition Solution Plus (CAS⁺) (Standard BioTools, San Francisco, USA, Cat# 201 244). After the last wash, cells were counted using a hemocytometer. Samples were dispatched into tubes that eventually containing 8.0 × 10⁵ cells each. Samples were split in multiples tubes to prevent clogging and enhance resuspension efficiency during acquisition. Each tube was spun and then pellets were resuspended in 50 µL of CAS⁺. All tubes were then loaded on the carousel of the CyTOF XT (Standard BioTools, San Francisco, USA). The CyTOF XT was programmed to resuspend the 8.0 × 10⁵ cells per tube in 1 mL of CAS⁺ EQ6 calibration beads (Standard BioTools, San Francisco, USA) and acquired all tubes in one batch with a medium cleaning cycle run between each tube. The event rate during acquisition was maintained between 100 and 200 events s⁻¹. After acquisition, normalization, and randomization was applied using the algorithm present in the CyTOF Software (Standard BioTools, San Francisco, USA). All tubes were then concatenated in a single file for downstream analysis. First, FCS files were debarcoded using PREMESSA package in R and the debarcoding key used previously to program the OT2. The parameters used for debarcoding were: minimum separation equal to 0.2 and maximum mahalanobis distance equal to 30. After debarcoding each sample (corresponding to each well) was saved as a *.fcs file. Using CytoexploreR package (https://dillonhammill.github.io/CytoExploreR/), live single‐cell population was gated using Gaussian parameter discrimination as described previously.^{[ 43} , ^{59 ]}

Cytometry by Time of Flight Acquisition and Analysis for KP4 and Pancreatic Stellate Cell Coculture

Similar to high‐content imaging analysis, GFP‐expressing KP4 cells and wild‐type PSCs were seeded to create the 4 distinct microenvironments. After 3 days of culture, 50 µm IdU was added for 3 h before performing gel digestion. Harvested cells were fixed, permeabilized, barcoded, and pooled as previously mentioned. After pooling, samples were incubated with a cocktail of metal‐tag conjugated primary antibodies (Table S1, Supporting Information, conjugation was performed as previously described)^{[ 43 ]} in 3% BSA/PBS for 30 min at RT, sample were agitated after 15 min to prevent cell sedimentation that could lead to uneven staining. After staining, samples were washed twice with 3% BSA/PBS. Samples were then stained for 30 min with one time agitation in the middle of the incubation with an anti‐chicken secondary antibody conjugated to Sm149 to reveal GFP primary unconjugated antibody, samples were washed twice with 3% BSA/PBS, and stained with DNA Intercelator overnight similar to checkerboard tests and washed before acquisition. The acquisition and analysis were performed as mentioned above. A bivariate analysis using GFP and Fibronectin was used to identify PSC and KP4 cells. KP4 were defined as GFP positive and fibronectin negative whereas PSC were defined as GFP negative and fibronectin positive (Figure S6a, Supporting Information). Using a R custom analysis pipeline, the counts and abundance of each cell type was quantified for each sample. The relative abundance plots in Figure S6b, Supporting Information, were generated using the Scatterpie package. Finally, the median intensity of each marker in each cell type in each sample was measured. All plots related to CyTOF were generated using R.

Statistics

Statistical analysis was carried out in GraphPad Prism 9 (GraphPad Software, USA). Ordinary one‐way ANOVA was used for assessing the physical properties of paper scaffolds, basic parameters of OT2, reproducibility of microtissues seeded by the OT2 sequence, and the effect of coculture on KP4 cells and PSCs. A Student t‐test was used for assessing manual and OT2‐seeded KP4 microtissues or PPTO.46 microtissues. p < 0.05 was considered significant.

Conflict of Interest

The authors declare no conflict of interest.

Supporting information

Supporting Information

Cao R., Li N. T., Latour S., Cadavid J. L., Tan C. M., Forman A., Jackson H. W., McGuigan A. P., An Automation Workflow for High‐Throughput Manufacturing and Analysis of Scaffold‐Supported 3D Tissue Arrays. Adv. Healthcare Mater. 2023, 12, 2202422. 10.1002/adhm.202202422

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.