H1 hESCs were obtained from WiCell and cultured under feeder-free conditions on 0.27 mg/mL reduced growth factor matrigel-coated (Corning) vessels with mTeSR™1 (STEMCELL Technologies). Cells were passaged using Gentle Cell Dissociation Reagent (GCDR) (STEMCELL Technologies) when confluence was ∼80%. Briefly, the spent media was aspirated, the monolayer was rinsed with PBS without Ca²⁺/Mg²⁺(PBS−/−) (Thermo Fisher Scientific) and incubated with GDRC at 37°C for up to 6 min. GDRC was diluted with fresh mTeSR™1 to stop further cell detachment. The cell suspension was collected and centrifuged for 5 min at 1,000 revolutions per minute (rpm). The supernatant was discarded, and the cells were gently resuspended in mTeSR™1 supplemented with 10 μM Y27632. Duplicate aliquots were collected for cell counts and viability using the Nucleocounter-200^® (NC-200™; Chemometec), and the cells were reseeded at the appropriate densities. For a 3-, 4- or 5-day schedule, the seeding densities used were 10,000 cells/cm², 15,000 cells/cm², or 20,000 cells/cm², respectively. Subsequent feeds were done with a complete media exchange using mTeSR™1 only. Cells were cultured under standard conditions: 21% O₂, 5% CO₂, 37°C, and fed daily. Following thaw from cryopreservation, the cell cultures did not exceed four passages before re-seeding for aggregation. All experiments with cell lines were approved by the Canadian Stem Cell Oversight Committee and/or UBC Clinical Research Ethics Board.

Once the cells reached ∼80% confluence, they were passaged as single cells with TrypLE™ Express (Invitrogen) using a similar workflow as outlined above. After determining the cell count, cells were reseeded in the respective cell culture platforms at the appropriate seeding density or cell concentration in mTeSR™1 supplemented with 10 μM Y27632 (Figure 1). Unless stated otherwise, 24 ± 2 h later, the media was changed to mTeSR™1, and all experiments were maintained for up to 5 days post-aggregation in mTeSR™1 only.

To determine the effect of the initial seeding density on aggregate formation using AggreWell400™ plates (STEMCELL Technologies), 200-, 400-, 800- or 1,000 cells/microwell were seeded and cultured overnight. Cell aggregates were dislodged by gently triturating each well with a 10 mL serological pipette, collected in 50 mL falcon tubes and spun at 800 rpm for 2 min. The supernatant was discarded or analyzed for cell viability. Recovered clusters were resuspended in fresh mTeSR™1, and duplicate cell counts were done using the NC-200™. Additional cell suspension aliquots were collected for aggregate sizing and viability staining. After aggregation, the cells were transferred to low attachment 6 well plates (Corning) on an orbital incubator shaker (Kuhner, LT-X) at 95 rpm (19 mm throw). Cells were fed with 5 mL/well mTeSR™1 daily.

To determine the impact of the initial seeding concentration on aggregate formation in 6 well plates on an orbital shaker, 0.5–2 × 10⁶ cells/mL were tested in a working volume of 5 mL/well. Complete daily media changes were performed. Wells were visually inspected, pooled into a 50 mL falcon tube, spun, resuspended in fresh media and evenly redistributed back into the wells. All aggregation experiments using 6 well plates or AggreWell400™ were done in parallel.

Aggregates were formed in roller bottles (Corning) using 100 mL working volume on a FlexiRoll (Argos Technologies) rotating at 31 rpm. The impact of initial seeding concentrations and subsequent cell expansion was tested using 0.5-, 1- or 2 × 10⁶ cells/mL. Media exchanges were done by collecting the cell suspension, spinning and resuspending in fresh media.

Aggregates were formed with paddle-based spinner flasks (Corning) using 75 mL working volume on a magnetic base (Chemglass), mixing at 80 rpm. The impact of initial seeding concentrations between 0.5–1 × 10⁶ cells/mL was tested. Daily media changes were done as in described with the roller bottle experiments. Some experiments performed in spinner flasks were performed in parallel using roller bottles.

Finally, the impact of either the initial seeding concentration (0.5–1.5 × 10⁶ cells/mL) or the agitation rate (40–110 rpm) was tested using PBS-Minis (PBS Biotech) with 100 mL working volume. Unless otherwise stated, daily media changes were performed as outlined above. Samples were collected for further downstream analysis. All conditions tested are summarized in Figure 1.

To evaluate the impact of each experimental condition on the growth kinetics of the cells, daily sampling was done after each complete media exchange. Briefly, the cell suspension was collected and spun for 2 min at 800 rpm. Cell aggregates were resuspended in up to 15 mL of mTeSR™1 and transferred to a 15 mL falcon tube. The suspension was mixed by gently inverting the tube before collecting several 100 µL aliquots for cell counts and viability using the NC-200™. Aggregates were also stained with LIVE/DEAD™ dyes (Thermo Fisher Scientific) using 2 µM calcein-AM and 4 µM ethidium bromide for 30 min in the dark at room temperature (RT). As a control, aggregates were incubated in either 70% ethanol or dimethyl sulfoxide (DMSO) for 15 min before staining with LIVE/DEAD™. Fluorescence images were acquired on an AxioZoom microscope (ZEISS). To assess cell recovery ∼24 h after seeding into a new cell culture vessel, the aggregation yield or plating efficiency was determined using the following equation: Aggregation yield or plating efficiency  % = N 1 / N 0   x   100

where N₀ is the initial total cell count at t = 0 h hours (h), and N₁ is the total cell count at t = 24 h post-aggregation unless stated otherwise. The aggregation yield quantifies cell recovery following the period of aggregation and can not distinguish contributions of cell growth or loss. A plating efficiency of 100% suggests that all seeded cell clumps reattached to the matrigel-coated surface. The fold expansion was calculated using the following equation: Fold change = N f / N 1

where N₁ and N_f are the total cell count at 1 day and the final day post-aggregation, respectively. The net growth rates and the doubling times were determined using the following equation: Net growth rate  µ net = In  N f   –  In  N 1 / Δ t Doubling time  T d = In  2 / µ net

Where N₁ and N_f are the total cell count at 1 day and the final day post-aggregation, respectively, and Δt is the time (h) elapsed between day 1 and the final day post-aggregation.

Aggregate morphology was captured daily using a Primovert microscope (ZEISS), and diameter was determined using a semi-automated pipeline in FIJI (ImageJ) after background subtraction using Adobe Photoshop (Adobe Inc.) (Supplementary Figure S1). Between 3-6 independent images were analyzed for a total of 150–1,500 aggregates. The area of each aggregate was determined, and the diameter was calculated using this equation: Diameter  d = 2 A / π where A is the area of a given aggregate.

RNA extraction, reverse transcription qPCR were done as previously described (Schulze et al., 2021). The list of primers is provided in Table 1.

Cell monolayers or aggregates were dispersed as single cells using TrypLE™ Express (Invitrogen; Cat# 12604021). Cells were stained for viability with LIVE/DEAD™ fixable aqua dead cell stain (Invitrogen) for 30 min in the dark. Cells were stained with SSEA4 (R&D systems FAF1435, 1:5000), in stain buffer (0.5% BSA, 0.05% sodium azide in PBS) for 30 min prior to fixation in BD Cytofix/Cytoperm™ (BD Biosciences) at RT for 20 min or overnight at 4°C. The cells were washed once with BD perm/wash™ (BD Biosciences) and stained for OCT4 (BD Biosciences 560329, 1:50) and SOX2 (R&D systems IC21018P, 1:1,000) at RT for 45 min. Stage 1 cells (S1D3) were stained for SOX17 (BD Biosciences 561591, 1:500) and FOXA2 (R&D systems IC2400G, 1:100). Flow cytometry was performed using a LSR II flow cytometer (BD Biosciences), and the data was analyzed using FlowJo™ v10 software (BD Life Sciences).

Aggregates were stained with 5 mg/mL Dithizone (DTZ, Sigma 194832). Aggregates were incubated in DTZ for 2 min at RT, and rinsed with PBS−/− until the solution was clear before imaging using an AxioZoom V16. To visualize hypoxia, aggregates were incubated in 5 µM Image-iT™ Green Hypoxia Reagent (Invitrogen I14834) for 3 h at 37°C, 21% O₂ and 5% CO₂, rinsed with PBS−/− then imaged.

Pancreatic progenitors and insulin-producing cells were generated from hPSC aggregates using a modified version of our 7-stage differentiation (Rezania et al., 2014). First, hPSC aggregates were formed from single cells in 0.1 or 0.5 PBS-Minis using 1 × 10⁶ cells/mL, mixing at 60 rpm in mTeSR™1 and 10 µM Y27632. One day after the initial seed, the media was changed to mTeSR™1 and the agitation was reduced to 40 rpm. On day 2, cell counts were determined and reseeded using the following parameters in order to start directed differentiation: 1) hPSC aggregates made in 0.1 Minis were reseeded into 0.1 Mini at 0.5, 0.75 and 1 × 10⁶ cells/mL, 2) hPSC aggregates made in 0.5 Minis were reseeded into 0.5 Mini at 0.5 × 10⁶ cells/mL and in 6WPs at 0.5 and 1 × 10⁶ cells/mL.

Statistical analysis was performed with GraphPad Prism. Data are shown as either an interquartile range with max and min values or mean +/− standard deviation. Pearson’s correlation was used to evaluate the relationship between the initial seeding conditions and either the aggregation yield or the aggregate diameter 1-day post aggregation (D1). A p-value < 0.05 was considered statistically significant.

To evaluate the growth kinetics of hESCs grown on a monolayer, we used the H1 hESC line, as it is a widely-used and well-characterized line (Thomson, 1998; Adewumi et al., 2007; Allegrucci and Young, 2007; Tosca et al., 2015). We identified 3 seeding densities that would be confluent and ready to passage at 3-, 4- or 5 days. The cells were maintained for 3 passages. In all cases, cells grew in compact colonies with phase bright borders and a high nucleus-to-cytoplasm ratio (Supplementary Figure S2A). All conditions had a steady increase in cell number by the end of the expansion period with a minimal lag growth phase and viability >80% on most days (Supplementary Figure S2B–D). The cell concentration reached ∼ 1-, 2- or 2.6 × 10⁶ cells/mL by the end of a 3-, 4 or 5-days of culture, respectively. We calculated the plating efficiency to estimate the percentage of cells attached to a matrigel-coated vessel. Plate efficiencies were between 50%–80% (Supplementary Figure S2E). Interestingly, we observed a slight increase in the plating efficiency as the initial seeding density increased. The doubling time was ∼16–18 h with a net growth rate of ∼0.04 h^-1 regardless of the initial seeding density (Supplementary Figure S2E), similar to previous reports (Panyutin et al., 2017). The fold change in cell expansion was ∼7.5X, ∼22X and ∼44.5X by days 3, 4 and 5, respectively. The lowest initial seeding density resulted in the highest fold change in cell expansion; however, more time (5 vs. 3 or 4 days) was required to reach the appropriate confluence prior to passaging. Although the highest seeding density resulted in the lowest fold expansion, it reached confluence sooner, which could make it a more suitable option for weekend expansion or experiments with a limited time frame. Furthermore, the cell yield using the highest seeding density can be increased by scaling out the cell culture with multiple vessels.

AggreWell™ plates can be used to generate and subsequently culture aggregates using a variety of cell types, including hPSCs and their derivatives (Ungrin et al., 2012; van Wilgenburg et al., 2013; Filice et al., 2020; Balboa et al., 2022), mesenchymal stem cells (Baraniak and McDevitt, 2012; Allen et al., 2019), immortalized cell lines (Wrzesinski et al., 2014) and isolated primary cells (Rettinger et al., 2014; Yu et al., 2018). To determine the impact of initial seeding density on undifferentiated H1 aggregate formation in AggreWell™ plates, 200-, 400-, 800- or 1,000 cells/microwell was seeded and statically cultured overnight. Aggregate morphology 1-day post-aggregation (D1) was compact with a smooth periphery in all conditions (Figure 2A, Supplementary Figure S3A). Aggregation yield estimates cell recovery and expansion ∼24 h later and was above 100% of seed quantity in most conditions (Figure 2B). There was a moderate correlation between the initial seed number and the aggregation yield (r ² = 0.68, p-value = 0.18). However, there was a strong correlation between the initial seed number and the aggregate diameter at D1 (r ² = 0.97, p-value = 0.01). One day post-aggregation, the aggregate diameter distribution was fairly homogenous between replicate runs (hereafter runs referred to as R1, R2 and R3) of a given condition. The D1 median aggregate diameter generated ranged between 89–104 μm, 106–126 μm, 136–149 μm, and 144–156 µm when 200-, 400-, 800- or 1,000 cells/microwell was seeded, respectively (Supplementary Table S1). We evaluated the D1 viability of the recovered cell clusters and any cells in the supernatant during the media change. The viability of aggregates from all conditions at D1 was high relative to the positive control (Supplementary Figure S3B,C), while the few single cells recovered from the supernatant were non-viable (data not shown).

Aggregates were harvested from the AggreWell™ plates and transferred to low attachment 6-well plates on the Kuhner LT-X orbital shaker for the rest of the experiment. In all conditions, the median aggregate diameter increased daily until day 3 (Figure 2C). By D3, the median cluster diameters had increased by 1.3–1.6X with a range between 155–164 μm, 159–184 μm, 188–203 μm, and 194–207 µm for the 200-, 400-, 800- or 1,000 cells/microwell conditions, respectively (Figure 2C; Supplementary Table S1). Most conditions had a slight increase in cell concentration through the duration of cell culture (Figure 2D); however, we observed higher growth during R3 for all conditions. After an initial drop, the viability in all conditions was maintained at >80% (Figure 2E, Supplementary Figure S3B). Interestingly, the steepest drop in D1 viability was in the 200 cells/microwell condition. There were also variable growth rates, doubling times and fold change expansion between replicate runs (Figure 2F). The fold change in cell expansion between all conditions ranged from 0.6 to 4.2X. The doubling time ranged from 1–4.5 days, with a couple of runs exhibiting population decline. The net growth rate was between 0.003–0.03 h^−1, with a few runs having higher death rates based on negative net growth rates. Overall, the growth kinetics of H1 aggregates generated with AggreWell™ plates was slower than H1 cells growing on a 2D monolayer. The data suggest that initial seed and, to an extent, aggregation yield, are process parameters that impact the size of aggregates generated using the AggreWell™ platform. However, the initial seed number had no obvious impact on the measured quality cell attributes (fold change in cell expansion, doubling time and net growth rate).

Six well plates are routinely used in labs to generate aggregates for downstream applications like differentiations. Undifferentiated H1 aggregates were generated from single cells in low attachment 6-well plates (6WPs) on the Kuhner LT-X orbital shaker at 95 rpm. The impact of initial seeding concentration (0.5-, 0.75-, 1-, and 2 × 10⁶ cells/mL) on aggregate formation and cell expansion were tested. At D1, the clusters formed from all conditions were compact throughout the experiment (Figure 3A). The aggregation yield was variable between conditions and within replicate runs ranging from ∼38–110% (Figure 3B) and generally lower than observed with AggreWell™ plates despite using the same cell inoculum (Figure 2B). D1 clusters had high viability with minimal viable cells lost to the supernatant (Supplementary Figure S4). There was a weak correlation between the initial cell concentration and either the aggregation yield or the median diameter of aggregates at D1 (vs. aggregation yield: r ² = 0.3, p-value = 0.45; vs. median diameter: r ² = 0.24, p-value = 0.5). Interestingly, the range of initial seeding concentrations tested did not have an obvious impact on the size of the aggregates generated at D1 (Figure 3C). At D1, aggregates made from seeding 0.5 × 10⁶ cells/mL had the following median diameters: R1- 184.7 ± 25.9 µm and R3- 198.2 ± 35.7 µm, despite a 2.5X difference in the aggregation yields between the two runs. Day one aggregate size distribution for 1 × 10⁶ cells/mL R1 and R2 were similar to R3 even though there was ∼3.2X difference in aggregation yields between those replicate groups. The median cluster diameter at D1 ranged from 177–199 μm, 181–189 μm, 155–164 μm, and 158–239 µm after an initial seeding concentration of 0.5-, 0.75-, 1-, and 2 × 10⁶ cells/mL, respectively (Figure 3C; Supplementary Table S1). There were variable growth patterns between conditions and replicate runs (Figure 3D). Regardless of growth kinetics, aggregates in all conditions maintained their compact morphology and had an increase in the diameter distribution by the end of the experiment. By D3, the median aggregate diameter was between 226–260 μm, 214–250 μm, 213–246 μm, and 176–290 µm with an initial seeding concentration of 0.5-, 0.75-, 1-, and 2 × 10⁶ cells/mL, respectively. The viability of all conditions was over 80% throughout the experiments (Figure 3E, Supplementary Figure S4). In all conditions, the fold change in cell expansion between D1 and D3 was between 0.5–2.4X (Figure 3F). The 2 × 10⁶ cells/mL condition had the lowest fold change (0.5–1.9X) relative to other conditions, with a decline in cell number in 2 of the 3 replicates (R1 and R2). Overall, the net growth rate, regardless of the initial condition, was between −0.001–0.02 h⁻¹, slower than when H1 cells were cultured on a monolayer.

Mammalian cells can be cultured in roller bottles as an adherent monolayer (Dhahri et al., 2022), in suspension using microcarriers (Jauregui et al., 2007), or cell aggregates (Schulz et al., 2012; Schulz, 2015). To evaluate how efficiently undifferentiated H1 aggregates could be made in roller bottles, single cells were seeded using an initial cell concentration of 0.5-, 1- or 2 × 10⁶ cells/mL. At D1, aggregates formed with 0.5- and 1 × 10⁶ cells/mL had a compact morphology with a fairly smooth periphery (Figure 4A). Some clusters were more elongated than spherical in shape. The initial seed of 2 × 10⁶ cells/mL resulted in cell sheets and clusters with both tight and loose irregular morphologies. By D3, both 0.5- and 1 × 10⁶ cells/mL had compact aggregates, while the 2 × 10⁶ cells/mL condition had loose aggregates with a jagged periphery. The aggregation yield from most runs was below 40% (Figure 4B). To determine whether the relatively low aggregation yield was due to the death of the remaining cells, we examined the cells in the supernatant at D1. While the viability of clusters from all conditions was high, we found smaller viable clusters as well as viable and non-viable single cells in the supernatant, suggesting that the aggregation yields reported were underestimated (Supplementary Figure S5). Interestingly, as the initial seeding cell concentration increased, so did the aggregation yield and the median diameter of aggregates at D1; however, there was no statistically significant correlation (vs. aggregation yield: r = −0.41, r ² = 0.17, p-value = 0.73; vs. median diameter: r = 0.98, r ² = 0.95, p-value = 0.14). The median cluster diameter at D1 had a range from 81–136 μm and 107–187 µm after an initial seeding concentration of 0.5-, and 1 × 10⁶ cells/mL, respectively, while aggregates made with 2 × 10⁶ cells/mL had the broadest range with a median D1 diameter of 181.3 ± 110.5 µm (Figure 4C; Supplementary Table S1). Due to the loss of smaller viable clusters to the supernatant, all reported D1 diameters do not fully represent the range of aggregates generated using roller bottles. Aggregate size increased in most conditions by D3 (median range: 132–180 μm, 170–206 μm, and 197 µm for 0.5-, and 1 × 10⁶ cells/mL conditions, respectively). We observed minimal cell growth in most conditions (Figure 4D). Viability initially dropped at D1 but increased ≥70%, thereafter in most conditions (Figure 4E). Overall, the fold change in most conditions was ≤1, with negative net growth rates indicative of a higher death rate relative to the growth rate (Figure 4F). The lack of cell growth and the subsequent population decline observed is unlikely due to the quality of the cell inoculum as the initial viability was >90%, and the aggregates generated in parallel using spinner flasks showed an increase in cell count (Figure 5).

To investigate the effects of initial seeding cell concentration on the formation of undifferentiated H1 cell clusters in spinner flasks, we tested 0.5-, 0.75-, 0.85-, or 1 × 10⁶ cells/mL. In all conditions, clusters formed after 1 day of aggregation were relatively even in size, and had smooth boundaries with a compact morphology (Figure 5A). Next, we sought to determine the efficiency of aggregate formation in this bioreactor geometry, as well as any impact the initial cell concentration may have. The aggregation yield was variable between conditions and within their respective replicates with a range between ∼50–130% (Figure 5B). We found the aggregation yield to be most similar between conditions when clusters were formed from the same cell inoculum. For example, the aggregation yield from all conditions during R3 was ∼60%. There was a weak correlation between the initial cell concentration and either the aggregation yield or the median aggregate diameter (vs. aggregation yield: r = 0.47, r ² = 0.22, p-value = 0.53; vs. median diameter: r = 0.37, r ² = 0.14, p-value = 0.63). The range of initial cell concentrations tested in this study did not have an obvious impact on the aggregate diameter distribution. The median diameter at D1 for all conditions was between 70–99 µm (Figure 5C; Supplementary Table S1). Clusters had high viability (Supplementary Figure S6A). Next, we tracked the changes in aggregate size over the course of the study. In all conditions, aggregate size increased throughout the experiment (Figure 5C). When extended to 4 days, aggregate size distribution was more heterogeneous. We found there was a deposit of cells on the impeller by the air-liquid interface in all bioreactors regardless of the condition (Supplementary Figure S6B). There were clusters stuck to the bottom of the spinner flasks during every media change (Supplementary Figure S6C). As a result, the diameters of these megaclusters were not measured. In most conditions, there was a steady increase in cell number with minimal time in lag growth phase, and the viability was ≥70% despite the initial drop at D1 (Figures 5D,E). The fold change in cell expansion for 0.5 × 10⁶ cells/mL was between 0.6–4.7X (Figure 5F). For the remaining conditions, the fold change was between 1.7–3. The doubling time, when positive, was between 1.3–3.4 days and the net growth rate was between ∼0.01–0.02 h⁻¹. In summary, the range of initial seeding densities had no obvious impact on aggregate made using spinner flasks and or subsequent growth kinetics of the cell clusters.

PBS-Mini bioreactors can be used to culture several cell types with or without microcarriers (Nogueira et al., 2019; Lembong et al., 2020; Silva et al., 2020; Miranda et al., 2022). We examined the impact of the initial seeding cell concentration and agitation rate on the formation of undifferentiated H1 cell clusters using PBS-Minis. First, we tested the effect of seeding 0.5–1.5 × 10⁶ cells/mL in PBS-Mini bioreactors mixing at 40 rpm. The aggregation yield in all conditions ranged between 49.5%–62.7% with non-viable single cells lost to the supernatant (data not shown). One day post-aggregation, clusters formed in all conditions were compact with a smooth periphery (Supplementary Figure S7A). Aggregates from 0.75–1.5 × 10⁶ cells/mL maintained their compact morphology. By D4, some aggregates generated with 0.5 × 10⁶ cells/mL had irregular periphery with a looser appearance, and cell sheets were present (Supplementary Figure S7A). Interestingly, the initial cell concentrations tested in this study did not modulate aggregate size distribution. The median aggregate diameter from each condition at D1 was as follows: 0.5 × 10⁶ cells/mL was 125.8 ± 25.8 µm, 0.75 × 10⁶ cells/mL was 132.6 ± 28.8 µm, 1 × 10⁶ cells/mL was 143.3 ± 29.3 µm, 1.5 × 10⁶ cells/mL was 147 µm ± 28.8 µm (Supplementary Figure S7B, Supplementary Table S1). Median aggregate size in all conditions steadily increased to between 158–187 µm by D4 (Supplementary Figure S7B). Cell concentration and viability initially dropped at D1 but subsequently increased to up to 1.25 × 10⁶ cells/mL and >80%, respectively, over time in all conditions (Supplementary Figure S7C,D). Overall, growth kinetics were slower than the monolayer control (Supplementary Figure S7E). The doubling time ranged between 2 to 7 days, with a net growth rate between 0.004–0.02 h⁻¹. Viability stains also showed some non-viable single cells in the cell suspension at D1 and D5 (Supplementary Figure S7F). All conditions maintained their pluripotency by the end of the expansion period with >96% OCT4+/SOX2+ cells and over 94% SSEA4+ cells (Supplementary Figure S7G). In summary, the initial seeding cell concentrations tested in PBS-Minis did not have an obvious impact on the aggregate size distributions or the aggregation yields. Furthermore, cell expansion and viability were similar between conditions.

To determine the impact of agitation rate on aggregate formation in PBS-Mini bioreactors, 1 × 10⁶ cells/mL were seeded at 40-, 60-, 80-, 100- and 110 rpm for the first 24 h. Aggregates generated from all conditions were tightly packed and fairly symmetric with a smooth periphery. There was a strong inverse correlation between the agitation rate and the D1 diameter (agitation rate vs. D1 diameter: r = −0.98, r ² = 0.97, p-value = 0.02). D1 aggregate diameter decreased as the agitation rate increased (Figures 6Ai,Bi,Ci,Di,Ei). Within any given run, the aggregation yield was quite similar, regardless of the agitation rate, with a broad range between ∼40–90% (Supplementary Figure S8A). There was a moderate correlation between the agitation rate and the aggregation yield (r = −0.81, r ² = 0.65, p-value = 0.19). Interestingly, there was some correlation between aggregation yield and the median aggregate diameter at D1 for some agitation rates tested (40 rpm: r = 0.99, r ² = 0.97, p-value = 0.01; 60 rpm r = 0.86, r ² = 0.74, p-value = 0.34; 80 rpm r = 0.17, r ² = 0.03, p-value = 0.9; 100 rpm r = −0.27, r ² = 0.07, p-value = 0.82). This suggests that aggregation yield may also affect the size of aggregates generated at 40- and 60 rpm.

The D1 aggregates recovered had high viability; however, more non-viable clusters were observed at 110 rpm (Supplementary Figure S8B). Although the supernatant from most conditions had mostly dead cells, viable cells (∼20% of the cell pellet volume) were lost in the supernatant when aggregates were generated at 110 rpm (Supplementary Figure S8B). This loss may be due to a suboptimal centrifugation time preventing the settling of these small-sized aggregates. Viable cell loss was reduced by extending the aggregation period at 110 rpm to 48 h without any media change during that time (referred to as 2D110 rpm). Compared to 110 rpm, aggregates made using 2D110 rpm had a slight dip in cell number by D3 before expanding (Supplementary Figure S8C).

Aggregate diameter increased over time for all conditions (Figures 6Aii, Bii, Cii, Dii, Eii; Supplementary Table S1). Notably, the median D1 diameter between runs varied when generated at 40- or 60 rpm possibly because of the aggregation yields (Figures 6Aii, Bii). The median diameters at D1 were between 109–198 µm for 40 rpm and 94–141 µm for 60 rpm and increased to between 168–236 μm and 172–213 µm respectively by D5. Aggregates generated at either 80- or 100 rpm resulted in fairly consistent diameter distribution between 50–80 μm at D1, and increased to between 133–159 μm and 84–133 µm for 80- and 100 rpm, respectively (Figures 6Cii). Finally, aggregates formed at 110 rpm for 48 h had similar diameters of ∼60 ± 9.4 µm in 2 of 3 runs that increased over the course of expansion (Figure 6Eii).

The growth kinetics between replicate runs was variable, with R1 having the longest lag phases (Figures 6Aiii,Biii,Ciii,Diii,Eiii). Aggregates formed at 40 rpm steadily increased in cell number and viability after the initial dip at D1. Growth kinetics were fairly similar between runs following aggregation at 40 rpm, with a maximum cell concentration at ∼1.5 × 10⁶ cells/mL as early as 2 days post-aggregation (Figure 6Aiii). Viability remained above 70% in both conditions (Figures 6Aiv,Biv,Civ,Div,Eiv). 2D110 rpm had lower viability at day 3 or 4 than other conditions. Based on the parameters characterized above, we further evaluated the pluripotency of the cells at the end of the expansion cycle. Pluripotency was maintained in all conditions. Flow cytometry analysis showed >85% OCT4+/SOX2+ cells and >98% SSEA4+ cells in all conditions at D5 (Supplementary Figure S8D). OCT4 and SOX2 expression at the end of the cell culture was similar to or higher than the input cells (Supplementary Figure S8E).

To demonstrate the integration of aggregation before initiating differentiation, hPSC clusters made in PBS-Minis were differentiated into pancreatic progenitors and insulin-producing cells (Figure 7A). hPSC aggregates were made using 0.1 and 0.5 PBS-Minis, both mixing at 60 rpm. The viability of aggregates was >90%, and while there was an initial cell loss, cell numbers increased by day 2 in both PBS-Minis, with a more modest increase observed in the 0.5 vessel (Figure 7B). Notably, aggregates made in 0.1 PBS-Minis were larger than those made in 0.5 PBS-Minis (Figures 7C,D). Aggregates from both scales of PBS-Minis had high pluripotency, with 99% SOX2+/OCT4+ cells (Figure 7E). Subsequent differentiations were started 48 h after aggregate formation using 0.1 and 0.5 PBS-Mini, and 6WPs on an orbital shaker.

hPSC aggregates generated in 0.5 PBS-Minis were reseeded into a new 0.5 vessel at 0.5 × 10⁶ cells/mL and differentiated to the pancreatic progenitor stage (S4D4). Initial seeding density can impact the formation of pancreatic progenitors (Gage, et al., 2013), therefore we simultaneously seeded 6WPs on an orbital shaker with aggregates at initial concentrations of either 0.5- or 1 × 10⁶ cells/mL. Since cell number can increase during the first stage of differentiation of hPSCs to definitive endoderm (DE) (Rezania et al., 2012), we hypothesized that the aggregates would be susceptible to hypoxia. As a control, aggregates differentiated in 6WPs were cultured under hypoxic conditions (2% O₂) for the first stage only (S1D1–S1D3). There was little increase in cell number between the initial seed and S1D3 in the 0.5 PBS-Mini condition, unlike both 6WP conditions (Supplementary Figure S9A). Cell numbers steadily declined between S1D3 and S4D4 in the 0.5 PBS-Mini. In contrast, both 6WP conditions increased in cell number up to S2D3 followed by a decline until S4D4. The viability was >90% in most conditions during the differentiation; however, there was a drop to 60% in the 0.5 PBS-Mini condition by S4D4 (Supplementary Figure S9A).

Aggregate morphology and cell fate were evaluated during the 4-stage differentiation. At S1D3, 0.5 PBS-Mini aggregates were less compact and less spherical compared to those in the 6WP (Supplementary Figure S9B). Despite being smaller, there was evidence of hypoxia in S1D3 aggregates from the PBS-Mini, albeit less so than was observed in the 6WP aggregates differentiated at 2% O₂ (Supplementary Figure S9C). At S1D3, pluripotency markers OCT4 and SOX2 were downregulated while DE-related markers SOX17 and FOXA2 were upregulated in all conditions, with >90% SOX17+/FOXA2+ cells (Supplementary Figure S9D,E). The data confirmed the successful exit from pluripotency into the DE fate for hPSC aggregate originally generated using 0.5 PBS-Minis.

The remaining stages of differentiation were done at 21% O₂ for both PBS-Minis and 6WPs. At S2D3, we observed the reorganization of clusters in all conditions (Supplementary Figure S9F). Aggregates differentiating in 0.5 PBS-Minis were predominantly cyst-like with loosely packaged regions compared to the 0.5 × 10⁶ cells/mL 6WP which had more blebs around the periphery, or the 1 × 10⁶ cells/mL 6WP which had a combination of blebs and cystic-like balloons regions. By S3D2, aggregate morphology reverted to being tightly organized in all conditions. By S4D4, aggregates in the 0.5 PBS-Mini condition were smaller compared to those in 6WPs. Expression of NKX6.1, and PDX1 were upregulated in both 6WP conditions confirming the induction of pancreatic progenitor fate (Supplementary Figure S9G). In contrast, 0.5 PBS-Mini aggregates had low NKX6.1, PDX1, and NEUROD1 expression and high CDX2 expression, suggesting an intestinal fate (Coskun et al., 2011).

Based on the generation of pancreatic progenitors (Supplementary Figure S9), we differentiated hPSC aggregates initially formed in 0.1 PBS-Minis to insulin-producing islet-like cells. hPSC aggregates were reseeded in 0.1 PBS-Minis at either 0.5, 0.75 or 1 × 10⁶ cells/mL, differentiated using a 7-stage protocol, and evaluated from the endocrine progenitor stage onwards (stage 5+). Cell number declined in all conditions between stages 5 (S5D3) and 6 (S6D7) with a yield of <1 S6D7 cells/input hPSC (Supplementary Figure S10A,B). Between S5D3 and S7D8+, aggregates from 0.75–1 × 10⁶ cells/mL got bigger while maintaining their spherical shape and viability at >90% (Figure 8A, Supplementary Figure S10C). Aggregates from the 0.5 × 10⁶ cells/mL condition decreased in size and viability, with more single cells observed by S6D7, followed by the formation of pearl-like sheets and megaclusters due to the agglomeration of several clusters at S7D8. The 0.5 and 0.75 × 10⁶ cells/mL seeding conditions were discontinued at S7D8 (Figure 8A, Supplementary Figure S10D).

The predominant endocrine cell types of pancreatic islets include glucagon-producing alpha cells, insulin-producing beta cells and somatostatin-producing delta cells. Expression of NGN3, a master regulation of endocrine fate, was upregulated in all conditions, with peak expression at S5D3 followed by a decline (Figure 8B). NEUROD1, a downstream target of NGN3, was upregulated, with peak expression at S6D7 irrespective of the seeding condition. PDX1 and NKX6.1 expression were similar to, if not higher than that measured in human islet controls. Expression of ARX, a transcription factor essential for alpha cells, declined in the 0.75 and 1 × 10⁶ cells/mL conditions compared to the 0.5 × 10⁶ cells/mL condition. However, the latter maintained stable expression from S5D3 onwards and was ∼3-fold higher than the other cultures by S7D8 (Figure 8C). Despite similar expression levels between the 0.5- and 1 × 10⁶ cells/mL conditions at S5D3, expression of PAX4, a transcription factor important for beta cell fate, decreased ∼4- and 1.5-fold, respectively, between S5D3 and S7D8. There was an increase in PAX4 expression for the 0.75 × 10⁶ cells/mL condition. At S7D8, PAX4 expression was similar between 0.75- and 1 × 10⁶ cells/mL condition and 2.7-fold lower in the 0.5 × 10⁶ cells/mL condition. Notably, PAX4 expression for all conditions was >100-fold higher than human islet controls. INS expression peaked by S6D7 for all conditions and declined thereafter. GCG was highest between S6D7 and S7D8 in the cultures seeded at 0.5 × 10⁶ cells/mL, while SST expression was similar for all conditions at all timepoint analyzed. Stage 7 was extended to 22 days (S7D22), for the 1 × 10⁶ cells/mL condition only. For most of the genes evaluated at S7D22, their expression levels remained similar to the levels at S7D8, with the exception of the hormone genes, which decreased (Figure 8B).

Given the difference in aggregate size between the conditions, we assessed the clusters for hypoxia from S6D7 onwards. A trend was observed between the initial seeding concentration and the number of positively stained S6D7 aggregates (Figure 8C). Little to no evidence of hypoxia was observed from the 0.5 × 10⁶ cells/mL condition. In contrast, high-intensity hypoxia-positive cells from the other conditions were localized to an arc on the edge of aggregates. By S7D8, hypoxia fluorescence increased in all conditions, with a more even distribution throughout most clusters (Figure 8D). Importantly, despite the presence of hypoxia, there was positive but varying DTZ staining between all conditions (Figure 8E).

Due to the focal pattern of hypoxia dye fluorescence observed at S6D7, we re-stained S7D22 clusters with DTZ after they had been stained and imaged for hypoxia. First, we confirmed that DTZ staining was consistent regardless of prior staining for hypoxia (Figure 8F). Both hypoxia and DTZ stain intensities at S7D22 were higher than those seen at S7D8. Hypoxia fluorescence was higher than human islets cultured for 21 days, and DTZ staining was less intense in the hESC-derivates. The viability of the S7D22 clusters remained high despite the increase in hypoxia intensity during the later stages of differentiation (Figure 8G).