Optimization of an O2-balanced bioartificial pancreas for type 1 diabetes using statistical design of experiment

A bioartificial pancreas (BAP) encapsulating high pancreatic islets concentration is a promising alternative for type 1 diabetes therapy. However, the main limitation of this approach is O2 supply, especially until graft neovascularization. Here, we described a methodology to design an optimal O2-balanced BAP using statistical design of experiment (DoE). A full factorial DoE was first performed to screen two O2-technologies on their ability to preserve pseudo-islet viability and function under hypoxia and normoxia. Then, response surface methodology was used to define the optimal O2-carrier and islet seeding concentrations to maximize the number of viable pseudo-islets in the BAP containing an O2-generator under hypoxia. Monitoring of viability, function and maturation of neonatal pig islets for 15 days in vitro demonstrated the efficiency of the optimal O2-balanced BAP. The findings should allow the design of a more realistic BAP for humans with high islets concentration by maintaining the O2 balance in the device.


Results
Screening of O 2 supply strategies for BAP containing MIN6 pseudo-islets (MPIs). The first objective of this study was to screen the selected O 2 strategies (HEMOXCell and silicone-CaO 2 ) in two different O 2 tension environments (20% O 2 and 1% O 2 ) for their impact on the viability and function of encapsulated MPIs using a 2 3 screening experimental design after 6 days of culture (Fig. 1A). The DoE was analyzed using variance analysis (Supplementary Tables 1-3). The models seemed to adjust well to the experimental data with determination coefficients (percentage of total variations explained by the model, R 2 ) higher than 0.80 (Supplementary  Tables 1-3). The experimental data obtained ( www.nature.com/scientificreports/ pseudo-islet viability, with a significant positive effect of this factor on the ATP content (p < 0.05, Table 2). The positive effects of the silicone-CaO 2 disk on the ATP content and the ATP/LDH viability ratio were evident (p < 0.001, Table 2), while no significant improvement of the insulin stimulation index (p = 0.0633, Table 2). As expected, the hypoxic environment had a strong negative impact on pseudo-islet viability and function, with significant effects of O 2 tension on ATP content, ATP/LDH ratio, and insulin index (p < 0.01, Table 2). A significant  www.nature.com/scientificreports/ interaction was evident between silicone-CaO 2 and the O 2 tension on the ATP content (p < 0.001, Table 2). A strong positive effect of silicone-CaO 2 on the MPI ATP content was observed in the hypoxic environment, while a negative effect of this factor was observed under normoxic conditions (Fig. 1B, silicone-CaO 2 and O 2 tension interaction). Interestingly, a significant positive interaction was evident between HEMOXCell and silicone-CaO 2 on the ATP content (p < 0.05, Table 2), as was a benefit on the ATP/LDH viability ratio (p = 0.0890, Table 2). The effect of HEMOXCell was higher in combination with silicone-CaO 2 than alone and vice-versa for these two responses (Fig. 1B). Even if not significant, we also observed a positive interaction between HEMOXCell and the O 2 tension on the insulin stimulation index, with a higher effect of HEMOXCell in the normoxic environment. A multi-response optimization was performed to find the best condition to maximize pseudo-islet viability and function in the BAP in the hypoxic environment ( Table 3). The best conditions under hypoxia were found in the presence of HEMOXCell and silicone-CaO 2 . These optimal conditions resulted in an increase of ATP content and ATP/LDH viability ratio by 14% and 48% respectively, while a decrease of insulin stimulation index by 19% was observed compared to the normoxic control without the ISO ( Table 1).

Optimization of the BAP design regarding O 2 balance.
In the second part of the study, the objective was to optimize the configuration of the BAP carrying the previously defined O 2 strategy in a hypoxic environment to increase the density of viable islets in the device. To incorporate the O 2 strategy, the BAP was made of two sheets of alginate encapsulating islets and HEMOXCell. The sheets were placed on either side of the silicone-CaO 2 disk. A central composite experiment was designed to maximize the density of viable islets in the BAP (ATP content and ATP/LDH ratio) by tuning the HEMOXCell concentration and the islet seeding density (Fig. 2C). As primary islets used in the BAP in clinical settings would not proliferate, we optimized the device configuration on MPIs over a short (24 h) period, where differences in MIN6 cell proliferation were negligible ( Supplementary Fig. 1). A HEMOXCell concentration ranging from 50 to 500 µg/mL was defined according to the literature 24,30,31 . The islet seeding density range was chosen on the basis of the O 2 balance in the BAP. One silicone-CaO 2 was able to produce a mean of 11.9 ± 0.3 nmol/min of O 2 over 12 days ( Fig. 2A). MPI OCR was estimated to be 1.04 ± 0.48 pmol/min.IEQ (Fig. 2B), giving a maximal islet density of 11,500 IEQ that could be supplied with O 2 per silicone-CaO 2 disk. In the literature, OCR was typically found to be approximately 1.64 ± 0.36 pmol/min.IEQ for human pancreatic islets 28,32-35 and 2.2 ± 0.42 pmol/min.IEQ for neonate pig islets 33,36 , suggesting a maximal islet density of 7300 and 5400 per silicone-CaO 2 disk, respectively. Based on these estimations, we decided to evaluate the islet seeding density range of 300 to 7000 IEQ in the BAP device ( Table 4). The optimization DoE was analyzed using variance analysis (Supplementary Tables 4 and 5). The resulting models adjusted well to the experimental data regarding ATP content according to the high R 2 value obtained (0.76, n = 6). In contrast, a low R 2 value was observed for the ATP/LDH ratio (0.39, n = 5), suggesting that factors other than those tested are involved in the response variations observed. Nevertheless, the lack-of-fit tests remained non-significant for both ATP content and ATP/LDH ratio (Supplementary Tables 4 and 5). The analysis of diagnostic plots confirmed the aforementioned results concerning model validation with overall good prediction for the ATP content, while poorer results were obtained for the ATP/LDH ratio ( Supplementary Fig. 2). This was consistent with the higher experimental variability for the ATP/LDH ratio of the raw data obtained ( Supplementary Fig. 2). where Y is the response; X A and X B are the linear variables associated with factors A (HEMOXCell) and B (islet seeding density), X A 2 and X B 2 the quadratic variables associated with factors A and B, β the coefficient associated with these variables, and ε is the residual variation. Estimates of regression coefficients and response surfaces are presented for each biological parameter tested in Supplementary Table 6 and Fig. 2D. We observed a significant quadratic effect of HEMOXCell concentration on ATP content (p < 0.05, Table 5). As expected, a strong positive effect of the islet seeding density on ATP content was observed (p < 0.0001, Table 5). Moreover, a negative quadratic effect of the islet seeding density was also observed for this parameter (p < 0.01, Table 5). No significant interaction was observed between the HEMOXCell concentration and the islet seeding density in the device.
According to these data, increasing the islet seeding density increased the viable cell content in the BAP but reached a plateau for the higher densities tested (Fig. 2D). Moreover, high islet density also resulted in a decreased islet viability ratio in the device (Fig. 2D). Multi-response optimization was used to determine the optimal HEMOXCell concentration and islet seeding density to maximize the ATP content and the ATP/LDH ratio in the BAP (Table 6). To this end, a desirability function was defined to find the best compromise between   Supplementary Fig. 3). An estimated cell number equivalent to 3572 IEQ was viable after 24 h in the hypoxic environment with the O 2 strategy. As a comparison, the multi-response optimization was also performed ( Table 6) with data obtained from the same DoE performed in hypoxia without the silicone-CaO 2 disk (Supplementary Tables 6, 7, 8 and 9). The analysis of the central composite experimental design without silicone-CaO 2 was carried out in the same way as with the O 2 generator (data not shown). In the absence of the O 2 generator, the optimal BAP configuration was defined with the 500 µg/mL HEMOXCell and 375 IEQ/device ( Table 6).

Validation of O 2 balanced BAP design on MPIs.
The optimal BAP configuration was first validated in vitro on MPIs in the hypoxic environment (1% O 2 ) for 3 days. The seeding density of MPIs embarked in the O 2 -BAP was set to rounded value of the maximal density defined in the RSM optimization in presence of silicone-CaO 2 (3000 IEQ) and the HEMOXCell concentration of 500 µg/mL was chosen. Indeed, as the oxygen released by the silicone-CaO 2 will progressively decrease until exhaustion, the concentration of HEMOXCell    Fig. 3A) and by 65% (p < 0.05, Fig. 3B) compared to the high O 2 tension condition. In the case of the O 2 balanced BAP (1% O 2 + ISO), the ISO significantly increased the ATP content (p < 0.05) and the ATP/ LDH ratio (p < 0.05) of MPIs compared to those cultured without. This allowed the ATP content and ATP/LDH ratio to reach the level observed in the positive control. Insulin secretion by the alginate encapsulated MPIs was assessed following glucose plus theophylline (G + T) stimulation after 3 days of culture (Fig. 3C,D). In high O 2 tension, the production of insulin by MPIs after G + T stimulation was significantly increased (p < 0.01, Fig. 3C) and reached a stimulation index of 12.65 ± 3.52 (Fig. 3D). In hypoxia, the MPIs totally lost their ability to secrete insulin in response to G + T (1% O 2 ). Interestingly, in the O 2 balanced BAP, the adverse effect of hypoxia on the insulin-secretory function of MPIs was significantly prevented (p < 0.05, Fig. 3C,D). In fine, the O 2 -balanced BAP allowed reaching a stimulation index of 10.80 ± 2.78 close to the index observed in the positive control.

Validation of O 2 balanced BAP design on NPIs.
Finally, the optimal BAP configuration was validated in vitro on primary NPIs in a hypoxic environment mimicking the acute hypoxic period (1% O 2 ) before graft neovascularization for 15 days. As the maximal seeding density obtained using RSM methodology with MPIs was lower than the maximal theorical seeding density of NPIs that was estimated to be 5400 IEQ, we set as well the density of NPIs in the BAP to 3000 IEQ and the HEMOXCell concentration to 500 µg/mL.  www.nature.com/scientificreports/ As expected, adverse effects of hypoxia without the ISO on the NPI ATP content were observed compared to the high O 2 tension condition on the different days of analysis (Fig. 4A). Indeed, while the total metabolic activity of NPIs under high O 2 tension was almost completely maintained during the 15 days of culture, it decreased under hypoxia by 15% (p < 0.05), 37% (p < 0.05), and 63% (p = 0.0625) after 3, 8, and 15 days, respectively. Interestingly, in the case of the O 2 balanced BAP, the ISO significantly increased the ATP content of NPIs on day 3 (p < 0.05) and 8 (p < 0.05) compared to NPIs cultured without the O 2 strategy. This allowed the ATP content to reach that observed in the positive control. After 15 days, the O 2 strategy still seemed to improve the NPI ATP content as a 25% increase was observed compared to NPI cultures in hypoxia without the O 2 strategy (p = 0.1167). However, at this time, a drop of up to 40% ATP was observed compared to the NPIs cultured under high O 2 tension. The viability (ATP/LDH ratio) of NPIs followed the same trends as observed for the total metabolic activity of NPIs under different conditions (Fig. 4B). Hypoxia decreased the viability of NPIs and the O 2 -strategy seemed to improve the viability. However, due to the high variability of LDH levels encountered in cultures, differences between groups were not found significant. The effect of O 2 tension on NPI morphology was characterized using hematoxylin and eosin immunohistochemical staining after 8 days of culture (Fig. 4C). NPIs under hypoxia showed altered nuclei, while islets grown under hypoxia with the ISO had a similar morphology to those cultured under 20% O 2 .
Insulin secretion by the alginate encapsulated NPIs was assessed following glucose plus theophylline (G + T) stimulation after 3, 8, and 15 days of culture (Fig. 5). From day 3 of culture in hypoxia, the NPIs lost their ability to secrete insulin in response to G + T. In high O 2 tension, NPIs remained functional until day 15. In the O 2 balanced BAP, the adverse effect of hypoxia on the insulin-secretory function of NPIs seemed to be highly mitigated, although it failed in some experiments for which lower insulin responses were observed, even in NPIs cultured under high O 2 tension (Fig. 5A). At any day of culture, a significant adverse effect was evident for NPI insulin stimulation indexes triggered by hypoxia from 13.5 ± 5.0 to 1.6 ± 0.4 (p < 0.005). This was significantly reversed by the addition of the ISO to 7.2 ± 2.5 (p < 0.05) (Fig. 5B). The optimized O 2 strategy attained a NPI stimulation index that was not significantly different (p = 0.262) from the NPIs cultured under high O 2 tension.
The effect of O 2 tension on NPI maturation was assessed by the quantification of the expression of insulin, glucagon, and PDX-1 after 8 or 15 days of culture compared to day 1 ( Fig. 6 and Supplementary Fig. 4). The percentage of insulin-stained area in NPIs significantly increased from day 1 to day 8 regardless of O 2 tension (p < 0.0001), although the observed increase was lower in 1% O 2 compared to 20% O 2 (p < 0.05, Fig. 6A,B). In addition, under 20% O 2 conditions, we observed a maturation of alpha cells from day 1 to day 8 of culture, as shown by the enlarged glucagon-stained areas (p < 0.0001) and the low O 2 pressure still seemed to hinder this  Fig. 4). We also performed quantitative real-time PCR for insulin, glucagon, PDX1 and NKX6.1 on NPI cultured in BAP for 15 days (Fig. 6D). All transcripts showed enhanced expression between day 1 and 15 days of culture in 20% O 2 condition, confirming maturation and functionality gain over culture (p < 0.05). Except for glucagon, there was a drop in all studied transcripts expression when NPI were cultured in 1% O 2 (p < 0.05 for INS and NKX6.1 and p = 0.057 for PDX1). Interestingly, the ISO significantly mitigated the hypoxic effect observed on PDX1 and NKX6.1 expression at the transcriptional level in NPIs (p < 0.05, Fig. 6D) with a trend to improve insulin transcript expression (p = 0.068, Fig. 6D) and protein (Fig. 6B,C). Surprisingly, ISO also caused glucagon expression in 1% O 2 -cultured NPI to rise beyond the level observed in 20% O 2 condition after 15 days of culture (p < 0.05, Fig. 6D), while a decrease of the protein expression was observed after 8 days in culture (Fig. 6B). Altogether, these results underlined the influence of O 2 tension on NPIs maturation and suggested the benefit of the oxygenation strategy to increase NPIs maturation in BAP after transplantation. As part of the hypoxia response, we evaluated VEGF release by pancreatic islets in the O 2 -balanced BAP or negative and positive controls (Fig. 7A). As expected, hypoxia enhanced the VEGF/ATP ratio by 132% (p < 0.01) and 192% (p = 0.0625) after 3 and 8 days, respectively, compared to the positive control (NPIs cultured under high O 2 tension) (Fig. 7A). At day 3, the hypoxia-driven VEGF production seemed to be prevented by ISO (p = 0.0625). With more time, however, no significant differences were observed between both conditions despite the oxygen supply. In addition, the relative expression of HO-1 in NPIs cultured for 15 days under the different conditions compared to day 1 was quantified to assess islet oxidative stress in response to hypoxia (Fig. 7B). The level of HO-1 mRNA expression in NPIs cultured under high O 2 tension was stable from day 1 to day 15, while hypoxia increased HO-1 expression by 4 (p < 0.01). Despite the benefits previously observed on NPI viability, function, and maturation, the O 2 strategy did not mitigate the effect of hypoxia on HO-1 expression, as similar mRNA levels were observed as in hypoxia.

Discussion
Encapsulation in a biomaterial is essential to isolate transplanted cells from the immune or autoimmune response of the host and allows low or no immunosuppression regimens. However, encapsulation aggravates the O 2 diffusional limitations. Low O 2 tensions of approximately 10 mmHg are encountered in the graft after transplantation during the critical 7 to 14 day period preceding the BAP surface neovascularization 9,13 . This acute hypoxic period is responsible for massive pancreatic islet dysfunction and cell death 8,14 . A diffusion-based device design usually results in an inadequate transplant size with low encapsulated islet density, making it difficult to scaleup BAPs designed for small animals to large animal models or for human recipients 4 . Using the factorial design and RSM methodologies, we designed an O 2 -balanced BAP carrying islets at high density under low O 2 tension    17 and rat 15 islets. By increasing the O 2 supply to the encapsulated islets with the combination of an O 2 carrier and generator, we increased the pseudo-islet seeding density to 3284 IEQ/cm 2 in the BAP composed of two 680 µm-thick alginate sheets (21,893 IEQ/mL) cultured in a hypoxic environment (1% O 2 ). Maintenance of the O 2 balance in the device allowed almost all the seeded MPIs to remain viable over the 24 h culture period. In contrast, the optimal seeding density without silicone-CaO 2 was only 375 IEQ/cm 2 . Although the presence of HEMOXCell increased the viability of MPIs in the device, the tested concentrations of hemoglobin did not significantly affect the percentage of viable islets in the BAP. Thus, we selected the highest concentration of hemoglobin assessed to improve O 2 diffusion throughout the BAP once the O 2 generator becomes exhausted.

in vitro decreases cell viability and function, and induces pro-inflammatory responses in human
After the validation with MPI model, the ISO BAP configuration was extrapolated to BAPs containing primary pancreatic islets isolated from neonate pigs. NPIs remain the most promising alternative to human islets, as pig insulin and metabolic characteristics are very similar to human characteristics 39 . Moreover, naked NPIs exhibit natural resistance to hypoxia in terms of survival and function 20 . The endocrine portion of NPIs is immature and requires a long period before reaching functional maturity 22,[40][41][42][43] . This maturation process may further be impacted by low O 2 tensions in the graft 44,45 . Placing the presently designed BAP in an environment of low O 2 tension to mimic the O 2 tension in the graft before its vascularization resulted in a significant impairment of the NPIs viability and of their ability to secrete insulin in response to glucose plus theophylline. The discrepancy with previously published results might be explained by the negative effect of the macrocapsule and the high islet density on the O 2 availability in the device versus the use of low-density naked NPIs. Hypoxia also prevented the NPI maturation process observed in the BAP cultured under high O 2 tension, suggesting that the hypoxic environment before graft neovascularization should delay the time needed for NPIs to become functional in vivo. Interestingly, our BAP design including the oxygenation system mitigated the adverse effects triggered by hypoxia by improving NPI viability, function, and maturation. Nevertheless, the O 2 strategy failed to prevent NPI upregulation of mRNA expression of the oxidative stress marker HO-1. This could be linked to the production of ROS by silicone-CaO 2 disk 24,46 . Finally, VEGF was increased in response to hypoxia by NPIs. In the O 2 balanced BAP, this pro-angiogenic response was maintained from days 3 to 8. We 24 and other 30 have already shown that hemoglobin may have a specific proangiogenic effect on pancreatic islets. The Hyperoxic-Hypoxic paradox 47 , suggesting that fluctuation in the free O 2 concentration rather than the absolute level of O 2 can be interpreted at the cellular level as a lack of O 2, could also explain the proangiogenic effect triggered by the oxygenation strategy. VEGF secretion could improve BAP engraftment by maintaining a beneficial proangiogenic signal in the absence of O 2 limitation.
The DoE methodology was used to design an O 2 balanced BAP allowing a high islet density by preventing the adverse effect of low O 2 tension encountered in the graft before its vascularization. Use of such an O 2 balanced BAP would lead to an acceptable device size of approximately 180 cm 2 that would carry the 600,000 IEQ needed to achieve normoglycemia in adult human patient 5,19 . Moreover, the total number of islets required to reach therapeutic efficiency in humans might be decreased in this O 2 -balanced BAP 4,28,48,49 . In the past, the amount of islets necessary has been defined based on naked pancreatic human islets transplanted into the liver 19 . In this www.nature.com/scientificreports/ environment, islets are exposed to a hostile microenvironment leading to the rapid death of a large portion of the grafted cells 9,50-52 . By preventing immune reactions and hypoxia-induced damage by encapsulation and adequate O 2 supply, the O 2 balanced BAP could reduce post-transplant cell death and thus the number of pancreatic islets required per human patient. Finally, increasing the O 2 tension in the BAP after transplantation promoted the differentiation and functional maturation of neonate pig islet beta cells in our settings. These results suggest that our oxygenation strategy may improve maturation defects of other promising alternative beta cell sources, such as insulin-producing cells derived from human stem cells 44,53 . Our study has several limitations. Future work must focus on the in vivo evaluation of the efficacy of the O 2 balanced BAP in diabetic mice in allo and xenotransplantation model. The use of genetically modified pig knockout for main xenoantigens, such as Neu5Gc and alpha1-3 GAL, could be useful to prevent the specific humoral response 54 and to improve long term engraftment. Another important consideration is the capacity of the O 2 regulated BAP to adapt to the varying O 2 tensions encountered in the transplant site during the graft surface revascularization phase. Indeed, the O 2 tension in the extravascular BAP is expected to rise from 10 mmHg (1% O 2 ) 10,52 after transplantation to 30 to 40 mmHg (5% O 2 ) after revascularization of the device 10,11,55,56 . Presently, the O 2 production rate from the silicone-CaO 2 disk was almost constant over 12 days in vitro, suggesting that the O 2 balance in the BAP should be maintained during the acute hypoxic period. Thereafter, the O 2 supply from the vasculature may not be sufficient to maintain fully viable and functional cells in the BAP embarking islets at high density. The presence of HEMOXCell could improve the O 2 supply from the graft vascularization during silicone-CaO 2 depletion due to its capacity to potentiate the islet VEGF secretion, to increase O 2 diffusivity through the alginate hydrogel 24 and to better balance the O 2 concentration in the device. The precise control and measurement of cell exposure to O 2 are limited by O 2 diffusion through the BAP and the high O 2 uptake by islets. Further work is necessary to confirm the stability of HEMOXCell and its capacity to provide sufficient O 2 from the vasculature to maintain the O 2 balance in our high islet density BAP.

Methods
The study was carried out in compliance with the ARRIVE guidelines.
Animals and ethical concerns. The data were the collective results gathered from eight neonate pigs.
Mini pigs aged 2 to 12 days were obtained from the INRAE PEGASE unit (Rennes, France). Experiments using pigs were approved by the Pays de la Loire Ethic Committee (Approval 01074.01/02) and were carried out in accordance with the relevant French (2013-118) and European regulations (2010/63 EU Directive). All efforts were made to minimize animal suffering and to restrict the number of experimental animals. Analgesia and anesthesia were provided by IM injection of methadone, midazolam and ketamine, and maintained with 2% isofluran. Piglets were subjected to laparotomy, and the pancreas was removed after exsanguination via the aorta causing euthanasia.
Pseudo-islet formation. The mouse MIN6 beta cell line was kindly provided by Pr. Jun-ichi Miyazaki (Osaka University Medical School, Japan) 58 . MIN6 cells were expanded in DMEM medium (Dutscher, Brumath, France) supplemented with 10% heat-inactivated calf serum (Invitrogen, Villebon-sur-Yvette, France), 1% penicillin/streptomycin/neomycin, and 50 µM 2-mercaptoethanol. To generate MIN6 pseudo-islets (MPIs), 10 6 MIN6 cells/mL were cultured in non-treated culture petri dishes for 3 days at 37 °C in normoxic conditions. Alginate encapsulation. Clinical grade low viscosity and high guluronate sodium alginate 2.2% (w/v) (PRONOVA UP LVG, Novamatrix, Sandvika, UK), later called "hydrogel" was used for islet (NPIs and MPIs) encapsulation. Alginate was solubilized in 0.9% NaCl (w/v) by gentle stirring overnight at 4 °C and sterilized using 0.2 µm filtration. NPIs and MPIs were quantified using canonical standardized Islets Equivalent Quantities (IEQ) 59 . For encapsulation in macrobeads, islets were gently mixed in the hydrogel at 2500 IEQ/mL alginate. Macrobeads 3 mm in diameter were obtained by alginate extrusion through a 23 G needle using a syringe driver into a 100 mM CaCl 2 gelation bath for 5 min. Alginate sheets (680 µm thickness, 1.  . Insulin secretion stimulation indexes were calculated as the ratio of the glucose + theophylline-stimulated insulin secretion level over the basal level of the encapsulated islets. Theophylline was used to potentiate insulin secretion as NPIs are immature islets containing insulin precursor cells, whose spontaneous secretion is notoriously low 57 .

Intracellular insulin content.
NPIs were recovered from the alginate sheet by incubation for 20 min at 37 °C in a decapsulating solution by calcium chelation in 5 mM citrate and 1 mM EDTA in PBS followed by mechanical dissociation and centrifugation. Proteins from naked or decapsulated NPIs were extracted by repeated pipetting and incubation steps at − 20 °C in 50 µL of an ethanol-HCl solution. Protein extracts were neutralized by adding 25 µL of Tris-HCl 1 M (pH = 7.5). Pig intracellular insulin was assayed in protein extracts by ELISA Absorbance was evaluated using a FLUOstar OPTIMA luminometer.
Transcriptomic analysis. Islets were recovered from alginate sheets as previously described and frozen at − 80 °C in NucleoZOL reagent (Macherey-Nagel, Düren, Germany). Total RNA was isolated according to the manufacturer's instructions and reverse transcribed using MLV reverse transcriptase (Invitrogen, Carlsbad, CA, USA). Pig primer sequences as described earlier 24,60 were purchased from Eurogentec (Angers, France). Validated TaqMan Gene Expression Assays (Thermofisher) were used for targets listed in Table 7. Real-time quantitative polymerase chain reaction (RT-qPCR) was performed on a CFX 96 Touch instrument (Bio-Rad, Hercules, CA, USA) using Hot FirePol qPCR reagents (Solis BioDyne, Tartu, Estonia). No template or samples processed without reverse transcriptase were included as negative controls. For each sample, the relative quantity was inferred from a standard curve created through the amplification of serial dilutions of a pool of representative samples. Whenever necessary, the target gene expression in the standard pool was artificially increased by spiking 5 µL of amplification product sequences diluted 1:1250. The expression of porcine genes encoding RPL19 (ribosomal protein L19) and PPIA (peptidylprolyl isomerase A) were used to normalize the expression of porcine PDX1 (pancreatic and duodenal homeobox 1), HO1 (heme oxygenase 1), NKX6-1 (NK6 homeobox I), INS (Insulin) and GCG (glucagon).
Insulin and glucagon stained area quantification. Images were acquired using an AxioVert microscope and Zen lite software (Carl Zeiss, Jena, Germany). Photos of representative fields of the slices were taken under both white light and fluorescence using the same exposure time for all images taken with the same staining. The percentage of insulin and glucagon staining per NPI area was quantified using ImageJ software (NIH, Bethesda, MD, USA). with the pseudo-islets in the alginate before crosslinking into macrobeads or sheets. The O 2 -generating biomaterial was prepared by mixing calcium peroxide (Sigma-Aldrich) in polydimethylsiloxane (silicone, Sylgard ® 184, Sigma-Aldrich) in a 1:3 ratio (weight/weight) as previously described by Pedraza 27 . A volume of 100 µL per well of silicone-CaO 2 was degassed using vacuum bell and cross-linked in a P48 plate for 24 h at 60 °C.
Silicone-CaO 2 O 2 production rate. The O 2 production rate (OTR) of the silicone-CaO 2 disks was followed during culture for 12 days by placing four silicone-CaO 2 disks in 200 mL of PBS (Eurobio, Courtaboeuf, France) at 37 °C. The PBS was first deoxygenated by stirring at 200 rpm in a hypoxic atmosphere (N 2 ). After reaching 0% O 2 , the container was sealed and the dissolved O 2 concentration was measured in the PBS using a Clark electrode and Rhapsody software. OTR (nmol/min/disk) was calculated as the initial slope of the dissolved O 2 concentration curve in the PBS. A negative control without a silicone-CaO 2 disk was performed in parallel.

Design of Experiment (DoE). Screening. Screening of the oxygenation strategies was performed on
MPIs encapsulated in alginate macrobeads using a full factorial design 2 3 (Fig. 1). The influence of the main factors and their first-order interactions was analyzed. The three factors studied and their levels were: without/ with HEMOXCell, without/with silicone-CaO 2 and normoxic/hypoxic O 2 tension. The response variables were intracellular ATP content per well (RLU), ATP/LDH viability ratio per well (RLU/AU), and insulin stimulation index. These response variables were assessed after 6 days of culture under the different conditions defined by the experimental design (Table 2).

Response surface method (RSM).
Based on the screening of the O 2 strategies, RSM was used to optimize the BAP configuration concerning the O 2 balance in the alginate sheet device. The objective was to maximize the density of viable MPIs in the BAP by tuning the HEMOXCell concentration and the islet seeding density in the hypoxic environment in the presence of the silicone-CaO 2 disk. A central composite design (2 2 factorial design with 4-star points and four replicates of the central point) was used to fit a second-order polynomial model (Fig. 2C). The model validation was performed by analyzing the lack-of-fit test results, the determination coefficient R 2 value, and the diagnostic plots. The two factors studied were HEMOXCell concentration and islet seeding density, and their ranges were determined according to the literature and the determination of silicone-CaO 2 OTR and pseudo-islet OCR. The response variables were intracellular ATP content (RLU) and ATP/LDH viability ratio (RLU/AU) per device. These response variables were assessed on pseudo-islets encapsulated in alginate sheets cultured for 24 h under the different conditions defined by the experimental plan ( Table 5). The optimum values were determined by solving the regression equations and analyzing the response surface plots. A multi-response optimization was performed to achieve the best compromise to maximize viability and function of the encapsulated pseudo-islets.
Statistical analyses. The experimental design for the screening and the optimization steps were repeated independently at least three times. Analysis of variance (ANOVA), regression analysis, and graphical display of DoE results were performed using the Statgraphics Centurion software 18. www.nature.com/scientificreports/ BAP design on NPIs was performed on a minimum of four independent experiments. The significance of differences between groups was evaluated using a non-parametric test (Mann-Whitney or paired Wilcoxon tests) or a parametric unpaired t-test. A p-value < 0.05 was considered statistically significant. Graphs' formatting was performed using Graphpad Prism software 8.0.2.