A wireless, battery-free device enables oxygen generation and immune protection of therapeutic xenotransplants in vivo

The O₂-Macrodevice is a stacked, multilayer assembly (Fig. 1A) that combines device elements for i) housing therapeutic cells and isolating them from the host immune system; ii) electrolytic oxygen generation; iii) oxygen storage and delivery to transplanted cells; iv) wireless power harvesting; and v) mechanical support. Oxygen generation relies on diffusive water vapor transport to the site of a proton exchange membrane (PEM)–based electrolyzer assembly [MEA (membrane electrolyzer assembly)] (Fig. 1B) that splits water into oxygen and hydrogen on the application of a voltage above the water-splitting voltage [1.23V (38)]. Oxygen resulting from water splitting is stored in a silicone (polydimethylsiloxane, PDMS) gas diffusion chamber (GDC) bonded directly to the anode side of the MEA and separated from a cell housing chamber (CHC) (SI Appendix, Fig. S1) by a thin (100 µm), oxygen-permeable PDMS diffusion membrane, while hydrogen produced at the cathode is allowed to dissipate away. Here, the high O₂ permeability of PDMS (SI Appendix, Fig. S2) represents a key enabling feature, allowing for continuous oxygenation of cells and preventing direct contact between cells and electronic components. The CHC comprises a lithographically defined, rectangular PDMS structure with a circular loading port for cell loading and bounded on the tissue-facing surface by a porous immune-isolating, track-etched polycarbonate (PCTE)–based nanoporous membrane. Fin-shaped protrusions from the side prevent channel collapse. The choice of porous PCTE (pore size ~ 400 nm) is informed by prior work (3) that demonstrates successful immune isolation and operation in IP sites. Cells transplanted in the CHC are therefore bounded by the immune-isolating membrane on the tissue-facing surface and by a thin, PDMS gas diffusion membrane on the MEA-facing surface, directly above the anode where the oxygen evolution reaction takes place. This multilayer design therefore allows for direct oxygenation of transplanted cells (Fig. 1C). A 2-part polyamide frame and PDMS spacers provide compression-based electrical connections between the PEM and metallic electrodes (SI Appendix, Fig. S3). Vapor-based PEM electrolysis (39, 40) represents an important enabling aspect of the technology by obviating the need for liquid electrolytes that require refilling, liquid handling, and pumping. The MEA comprises a multilayer stack of a chemically (41) and thermally (42) stable sulfonated polytetrafluorotheylene PEM located between gold-plated perforated stainless steel mesh electrodes and PDMS spacers. Iridium-ruthenium oxide (IrRuOx) forms the anodic catalyst while platinum black (Pt-B) forms the cathodic catalyst. PEM materials exhibit highly selective transport of protons (H⁺) relative to other species and, accordingly, supports the following electrolytic reaction on application of a voltage (V_Applied > 1.23 V)

The water-vapor reactant mass diffuses directly to the PEM surface through the PDMS (water vapor transmission rate ~ 10 mg/cm²/day) based on a pervaporation process (43). Diffusive water vapor transport to the PEM also ensures adequate hydration, which is essential to its proton-transport properties (44). The oxygen production rate is directly related to electrolytic current and given by: M_O2= I_MEA/n_O2F (37) where M_O2 is the molar oxygen generation rate, I_MEA is the electrolytic current through the MEA, n_O2 is the number of electrons required to make a molecule of oxygen (n_O2= 4), and F is Faraday’s constant (96,500 A/mole-s) (37) (SI Appendix, Fig. S4A). The entire assembly is encapsulated in a thin (<30 µm) PDMS layer (SI Appendix, Fig. S4B) that prevents direct contact exposure of the PEM to biofluids, containing ionic species such as Na⁺ and Cl⁻ suppressing the potential for parasitic chlorine evolution reactions, a well-known challenge in salt-water electrolysis (38).

Power management is an important design consideration. Batteries have several disadvantages in this context: They require recharging, significantly increase the overall weight and size of the device, and require hermetic encapsulation to prevent water ingress and the leakage of potentially toxic battery materials. In contrast, wireless power harvesting via resonant inductive coupling offers options in fully wireless, reliable, safe, lightweight battery-free operation for implantable devices (45, 46). Our system comprises an externally located primary transmitter and an implanted secondary receiver, designed to resonate at 13.56 MHz, chosen due to its low absorption in biological tissue and compatibility with near-field communication (NFC) protocols (47). The circuit is constructed from a multilayer flexible printed circuit board (fPCB) (total thickness ~ 150 µm), with surface-mount components for power harvesting, tuning, and regulation for stable 2 V DC output (Fig. 1D and SI Appendix, Fig. S5). The thin, flexible construction of the device affords advantages in size, weight, and inductive coupling efficiency and provides the basis for potential future embodiments that are entirely flexible. The calculated required current (I_MEA) for curative doses of rat islets chosen as a test system due to their relatively high oxygen demand, in small animal models is expected to be ~ 3 mA (M_O2~ 0.25 nmol/s) (25). At 2 V, this corresponds to a power requirement (P_O2 = V_AppliedI_MEA) of 6 mW: the system described here can support up to 20 mW (SI Appendix, Fig. S6), suggesting the feasibility of the concept. Here, the MEA load (~1 k Ω) allows for peak power transfer efficiency, with a sharp drop-off in power harvesting above this load observed due to the Q-factor of the antenna. In practice, the antenna area and number of inductive coils can be scaled up to increase power transfer. Wireless, battery-free oxygen generation results from placing a fully assembled O₂-Macrodevice (SI Appendix, Fig. S7A) in the rf field of a primary transmitter coil, creating a rapid rise in oxygen partial pressure in the GDC (pO_2,GDC) from ambient air values (~20%) to 45% in 80 min (SI Appendix, Fig. S7B and Movie S1), with wireless oxygen generation performing equivalently to wired oxygen generation from a 2-V voltage source (SI Appendix, Fig. S7C). The system also offers capabilities in pulsed operation that can maintain pO_{2, GDC} without the need for active control electronics on-board the implantable device (SI Appendix, Fig. S7D and Movie S2). Pulsed mode oxygenation and control of primary transmitter coil power each offer opportunities to mitigate the potential for hyperoxia. Operation of the device results in a minimal temperature rise, <1.5 °C during DC oxygen generation (SI Appendix, Fig. S8).

A human embryonic kidney (HEK-293T) cell line engineered to secrete erythropoietin (EPO) as a surrogate model therapy cell allows for testing the effects of wireless oxygen generation in vitro and in vivo. Extensive experimental and modeling efforts have characterized the effects of cell cluster size (48), loading density, (49, 50) and fibrotic tissue formation (51) on encapsulated transplanted cells in SC sites, suggesting local oxygen concentrations of 0.5% < pO₂ < 5%. Accordingly, in vitro studies involved testing within this range. Culturing ~2 × 10⁶ HEK-EPO cells in hypoxic conditions (pO_2,culture = 0.5%) over a 12-h period in O₂-Macrodevices operated at 2V (wireless power transfer system in SI Appendix, Fig. S9) revealed two important findings: First, O₂-Macrodevices can maintain pO_2,GDC ~ 45% and a stable current I_MEA ~ 4 mA (M_O2 ~ 0.33 nmol/s) over the entire measurement period (Fig. 2A). Second, O₂-Macrodevices can prevent hypoxia in these conditions, as measured by fluorescence from a hypoxic indicator dye with an undetectable fluorescence signal that compares favorably with a normoxic (pO_2,culture = 21%) control (Fig. 2B). An expanded set of studies involving culturing at pO_2,culture = 1% over 72 h and quantitative fluorescent imaging via fluorescence-activated cell sorting (FACS) (extended data in SI Appendix, Fig. S10) following staining with the same hypoxic indicator dye establish i) equivalence between wired and wireless O₂-Macrodevices, and significantly reduced hypoxia in O₂-Macrodevices relative to naked cells (NC) and EC in nonoxygenated control devices cultured in the same hypoxic conditions (pO_2,culture = 1%) and ii) equivalent levels of hypoxia as in EC and NC cultured in normoxic conditions (pO_2,culture = 21%) (Fig. 2C). The hypoxic conditions studied here correspond to expected oxygen concentrations in densely packed cell configurations in vivo (SI Appendix, Table S1).

Experiments in awake, freely moving animals with battery-free O₂-Macrodevices implanted in SC sites validate device performance in vivo. Here, mice are housed in standard enclosures with dual-coil wire loops around the outer perimeter to create a uniform (SI Appendix, Fig. S11) region of rf-power within the cage. O₂-Macrodevices and nonoxygenated control devices each housing ~5 × 10⁶ HEK-EPO cells and sham devices containing no cells or electronics are transplanted into dorsal SC sites in healthy, C57BL/6J mice (detailed surgical procedures can be found in the Methods section), and direct oxygen measurements at these sites 4 d after implantation reveal elevated values of pO₂ in animals with O₂-Macrodevices (pO_{2,SC Space} = 13.7 ±1.7%) relative to animals with sham devices (pO_{2,SC Space} = 9.1 ± 1.0%) (Fig. 2D). Weekly measurements of serum EPO concentrations following transplantation over 6 wk reveal two important trends (Fig. 2E): Firstly, the O₂-Macrodevice group exhibits a threefold to fourfold increase in serum EPO immediately following transplantation over the nonoxygenated group; second, the O₂-Macrodevice group exhibits steady serum EPO levels of ~160 pg mL⁻¹ while the nonoxygenated group exhibits a steady, monotonic decline over time. We note that the EPO gene in our engineered cells is expressed under a cytomegalovirus (CMV) promoter (3), leading to constitutive expression irrespective of local or systemic oxygenation level. Accordingly, changes in protein production are expected to be largely driven by cell viability. As expected, the sham group does not exhibit significantly elevated serum-EPO levels throughout the posttransplantation experimental period, suggesting that surgeries and blood collections do not influence EPO measurements. Area under the curve (AUC) calculations (extended data in SI Appendix, Fig. S12) are an indicator of total EPO secretion over the entire 4-wk period (Fig. 2F). AUC measurements reveal significantly higher total EPO output in the O₂-Macrodevice group relative to the nonoxygenated and sham controls. These results suggest the importance of supplemental oxygen provided by O₂-Macrodevices in boosting protein production and maintaining cell viability in minimally invasive implants containing therapeutic cells.

Pancreatic islets are an important target for the oxygenation strategy developed in this work: They are highly metabolically active and account for 10 to 15% of arterial blood supply to the pancreas in vivo despite making up 1% of pancreatic mass (52). Poor oxygen tension in transplanted islets has been implicated in graft failure across a range of transplant sites and strategies (25) by significantly reducing islet viability. Culturing in hypoxic conditions (pO_2,culture = 1%) results in significant cell death (viable cells/islet ~55% for NC and 65% for EC) (Fig. 3A). In contrast, islets in O₂-Macrodevices cultured in pO_2,culture = 1% exhibit significantly higher viability (~90%), similar to islets cultured in normoxic conditions (~100%) (Fig. 3B). Islets subjected to hypoxic conditions also exhibit reduced functionality, as measured by glucose responsiveness (30). Here, glucose-stimulated insulin secretion (GSIS) assays quantify functionality beyond simple viability stains. Successive static incubation in conditions with low (2 mM) and high (20 mM) glucose, respectively, produce differential rates of insulin production in healthy islets cultured in pO_2,culture = 21% conditions and islets in O₂-Macrodevices cultured in pO_2,culture =1% conditions over 72 h. Islets subjected to hypoxia without supplemental oxygen generation exhibit no detectable glucose-responsive insulin secretion (Fig. 3C). Taken together, these results suggest that despite hypoxic ambient conditions, islets cultured in operational O₂-Macrodevices exhibit both viability and functionality that are similar to islets cultured in normoxic conditions and significantly elevated relative to those of islets subjected to hypoxic conditions without oxygenation.

Curative islet transplantation for humans has required islet masses of >350,000 IEQ (25). Encapsulating this volume of cells within translationally relevant device sizes requires islet packing densities that cannot be achieved without addressing oxygen limitations. This remains a challenge in the encapsulation field, with loading density scaling inversely with device size (SI Appendix, Fig. S13). O₂-Macrodevices offer opportunities to support high islet loading densities (~5,000 IEQ/cm²) by mitigating the potential for local hypoxia. In vitro culturing at low (600 IEQ/cm²), medium (2,500 IEQ/cm²), and high (5,000 IEQ/cm²) densities in moderately hypoxic conditions (pO_2,culture = 5%) in custom-designed devices (SI Appendix, Fig. S14) over a 7-d period mimic conditions in the SC space (30). Interestingly, these studies yield no detectable differences in the viability of islets cultured in O₂-Macrodevices over different densities compared to islets cultured in control devices from live/dead staining and scoring (Fig. 3D). However, islets cultured in O₂-Macrodevices secrete significantly more insulin in a glucose-responsive manner compared to control islets without supplemental oxygen, as revealed by GSIS assays performed at days 4 and 7 (Fig. 3E). These trends are captured in a combined stimulation index (SI) metric (53) across both time points (Fig. 3F), defined as the ratio between insulin secretion at high glucose and low glucose conditions. O₂-Macrodevices demonstrate significantly higher stimulation indices across loading densities (SI = 8.6 ± 3.0) relative to controls (SI = 1.27 ± 0.1).

Functional testing in SC sites in immune-competent diabetic mice (Fig. 4A) via a retrieval study with our best-performing device design reveals the effect of in vivo oxygen generation on encapsulated rat islets (~850 IEQ/device, ~1,000 IEQ/cm²). STZ-induced diabetic mice exhibiting hyperglycemia (nonfasting blood glucose, BG > 400 mg/dL) implanted with O₂-Macrodevices exhibit normoglycemia beginning the day after transplantation, as shown in Fig. 4B. A control group with an identical number of islets but without oxygen generation exhibits a decline in BG levels until day 5, followed by a return to pretransplantation hyperglycemic levels by day 12. Notably, in the control group, average BG values never go below 200 mg/dL, the threshold for normoglycemia, throughout the experimental period. Overall, the O₂-Macrodevice group exhibited significantly lower BG levels and tighter BG control across all animals over the transplant period (173 ± 12 mg/dL) relative to nonoxygenated controls (444 ± 23 mg/dL). Importantly, explanting O₂-Macrodevices results in a return to pretransplant hyperglycemic BG values 2 d after devices are retrieved. AUC analysis (Fig. 4C) also reveals significant differences between O₂-Macrodevices and controls in maintaining normoglycemia, with healthy naive C57BL/6J mice and STZ-diabetic mice serving as additional positive and negative controls, respectively (SI Appendix, Fig. S15). A larger experimental cohort (SI Appendix, Fig. S16) revealed that ~ 70% (6/9 animals) of mice in the O₂ group were normoglycemic at the end of the study while all non-O₂ controls had failed after 15 d (0/5 animals normoglycemic). One of the failures in the O₂ group was due to device migration following surgery to configuration that was parallel to the direction of magnetic field lines, resulting in no net power transfer. Postexplantation analysis of the remaining two failures revealed electronics failure, resulting in nonfunctional devices. Implants were well tolerated with surgical sites in the O₂-Macrodevice group exhibiting good healing (optical images in Fig. 4D) and stable weights (Fig. 4E). However, mice in the non-O₂ group exhibited a steady reduction in weight that correlated to their overall declining health, resulting from prolonged hyperglycemia (extended data in SI Appendix, Fig. S17). While BG measurements are a point-indicator of device performance, glycated serum protein (GSP) concentrations are an indicator of glycemic control over several weeks (54): As shown in Fig. 4F, O₂-Macrodevices (25 ± 14 µmol l⁻¹) exhibit GSP levels similar to those of naive C57BL/6J mice (46 ± 18 µmol l⁻¹) and significantly lower than those of the nonoxygenated control group (378 ± 100 µmol l⁻¹) and a group of STZ-Diabetic mice (322 ± 13 µmol l⁻¹).

In vivo glucose responsiveness is an important measure of islet functionality, and testing involves GTT wherein glucose boluses are injected intraperitoneally [IPGTT (IP glucose tolerance tests)], followed by BG monitoring over a 2-h period. The O₂-Macrodevice group again closely follows trends observed in healthy, naive animals, wherein BG values decline back to their baseline values after an initial spike that peaks t = 30 min after glucose injections. Nonoxygenated controls and STZ-diabetic mice experience a spike, followed by persistent hyperglycemia with BG values that do not decline over the 2-h experiment (Fig. 4G and extended data in SI Appendix, Fig. S18). AUC analysis serves as an additional indicator of total time spent in hyperglycemic conditions and reveals significant differences between O_2-Macrodevices and both nonoxygenated controls and STZ-diabetic controls, respectively (Fig. 4H). Significant differences in serum insulin concentrations before and after the IPGTT study (Fig. 4I) verify that islets in O₂-Macrodevices can dynamically regulate insulin secretion in response to changes in BG.

A final set of experiments on explanted devices confirm significant differences in both viability and functionality of islets encapsulated in O₂-Macrodevices relative to nonoxygenated controls. First, GSIS studies on explanted devices demonstrate islet responsiveness (Fig. 4J), with an average SI (2.3 ± 0.1) that is significantly higher than that of nonoxygenated controls (0.97), as shown in Fig. 4K. Live stains on explanted devices (Fig. 4L) confirm clear differences in islet outcomes in O₂-Macrodevices relative to nonoxygenated controls, with O₂-macrodevices supporting healthy, living islet masses, despite the formation of significant fibrotic tissue around the device (SI Appendix, Fig. S19).

Placing assembled boards with MEA assemblies between the two components of mechanical frames and applying firm pressure snaps the frame closed and applied compressive pressure to the MEA via the PDMS spacers. Separately, a PDMS prepolymer (Sylgard 184, 10:1) is drop-cast to a thickness of 0.5 mm and allowed to partially cure, creating adhesive PDMS surfaces. Assembled devices were placed anode-side down onto these adhesive surfaces, followed by drop-casting a further 0.5 mm of PDMS prepolymer (Sylgard 184, 10:1) and completely curing at 60 °C for 1 h, leaving behind entirely silicone-encapsulated devices. Devices are then cut out with scalpels and demolded from their hydrophobic surfaces to leave smooth, PDMS-encapsulated anodes. Separately, GDC layers are covalently bonded to CHC (on the PDMS surface) by successively treating the bonding surfaces with O₂-Plasma (Harris, Inc.,) and curing at 60 °C for 3 h. The GDC-CHC assembly is similarly bonded to the PDMS-coated anode via successively treating the bonding surfaces with O₂-Plasma (Harris, Inc.,) and curing at 60 °C for 3 h, completing O₂-Macrodevice assembly.

The O₂-Macrodevice was powered by resonant inductive coupling with an external primary antenna. The system was designed to resonate at 13.56 MHz (46, 67). Resonance was achieved via an L-C circuit, where the resonant frequency is given by

The inductance L of the secondary receiver coil was first computed to ±20%. Tuning capacitor values were then varied until optimal power transfer was achieved via resonance at 27 pF. The total harvested power is proportional to the magnetic flux passing through the coil loops, Φ_B and given by $\iint B.dA$ , where A is the total area enclosed by the 12-loop, dual-side inductive coil. We used two power transfer systems for our studies here, both designed for resonant power transfer at 13.56 MHz. For benchtop and in vitro studies, we used an NFC-card reader expansion board (X-NUCLEO-NFC06A1, STMicroelectronics). For in vivo studies, we used a power transfer system compatible with standard mouse enclosures (45) (Neurolux Inc., Skokie, IL), comprising an AC-voltage source, a power amplifier, and a set of wire loops (2/cage) wrapped around the outside of the enclosure (SI Appendix, Fig. S11).

We used oxygen probes with two form factors. For invasive measurements, we used a retractable optical sensor with a needle form factor based on fluorescence quenching (NTH-PSt7, PreSens, Regensburg Germany) inserted directly into the GDC. Noninvasive measurements relied on attaching a fluorescent sticker to the inner surface of the transparent PDMS GDC via a transparent adhesive (Dowsil, Dow Corning) and measuring via an integrated fiber system (Oxygen Sensor Spot, SP-PSt3-NAU, PreSens, Regensburg, Germany). For in vivo measurements, sterile optical oxygen sensors housed in retractable, 23G needles (NTH-PSt7, PreSens, Regensburg, Germany) were used. Wired voltage sources supported early characterization studies. MEA assemblies were wired in series with digital ammeters (NI-USB 4065, National Instruments) with data recording on a custom program on LabView (National Instruments). A programmable voltage source (Rigol, DP800) provided a stable, DC voltage.

Normoxic conditions involved culturing in 21% O₂ conditions, consistent with oxygen concentrations in ambient air. Hypoxic conditions involved culturing in 0.5% and 1% O₂ to replicate conditions within large islets in densely packed configurations. Each of the groups listed involved three separate devices for statistical computations. Cells were cultured for 72 h, followed by staining with a 5 µM hypoxic indicator dye (BioTracker, Green Hypoxia Dye, Sigma Aldrich, St. Louis, MO) in DMEM for 1 h. Cells were then washed with PBS, replaced with DMEM, and observed under a fluorescence microscope (Fig. 2B) after 3 h of incubation time.

For quantitative studies, devices were removed from media and washed with PBS. Devices were submerged in TrypLE and tweezers were used to manually separate the membrane from each device. After opening the membranes, devices were treated with TrypleE and incubated for 10 min to dissociate cells.

Cells were resuspended in a solution of zombie NIR (biolegend) to identify viable cells. After 15 min, cells were centrifuged and resuspended in cell staining buffer prior to analysis on a flow cytometer (BD LSR II, BD Biosciences). Cells were first gated to exclude those that were stained with the zombie NIR dye which represented dead cells. Within the viable cells, the level of hypoxic dye was measured in the FITC channel. FlowJo software was used for analysis.

First, the electronics portion of devices was carefully removed using a scalpel prior to staining in order to visualize islets within the cell chamber. Devices containing islets were rinsed twice membrane down with 20 mL Hank’s Balanced Salt Solution (HBSS) (Mediatech). Devices were then flipped membrane up and incubated with 200 µL viability dyes calcein AM (2 µM working concentration) and Ethidium homodimer-1 (4 µM working concentration) in HBSS (Invitrogen, cat# L3224). An epi-fluorescent microscope (Nikon TS2) with filters for calcein AM (ex/em ~495 nm/~515 nm) and EthD-1 (ex/em ~495 nm/~635 nm) was used to acquire images of devices at 4× magnification, which were then stitched together. Following imaging, percentages of total viable cells within 50 whole islets per device (n = 3 devices and ~150 total islets scored per condition) were estimated by a single-blinded operator trained in human islet isolation protocols utilized in the NIH-funded clinical trial by the Clinical Islet Transplantation consortium and University of Illinois at Chicago clinical trials (68).

GSIS was used to quantify islet functionality for in vitro experiments in Fig. 3 and in vivo experiments in Fig. 4. In vivo devices were explanted and placed in 21% O₂ culture overnight with continuous power supply to the oxygen-generating devices. This recovery period from retrieval procedures was previously found to stabilize insulin responses during the GSIS assay. For in vitro GSIS, hypoxia incubators were found to equilibrate to 5% O₂ after ~5 min and 1% after ~10 min. Islets in devices were washed three times with 25 mL saline, followed by 3 successive 10-min 25-mL washes in a 2 mM glucose Krebs-Ringer solution (KR2) to remove residual insulin produced during culture. Devices were then placed in 5 mL KR2 solution for two 30-min prewash steps. O₂-Macrodevices were placed on primary power transfer coils within the incubators during these prewash incubations onward. Finally, devices were incubated in 5 mL KR2 for 1 h (low 2 mM glucose) and 5 mL KR20 for 1 h (High 20 mM glucose). Supernatants were collected from the last wash KR2, to ensure that sufficient washing removed the residual insulin from media, and Low KR2 and High KR20 for analysis. The amount of insulin secreted into the supernatants was then quantified using a high-range rat insulin enzyme-linked immunosorbent assay (ELISA) (Mercodia, cat#10-1145-01) in duplicates according to manufacturer’s recommendations. SI was then computed as insulin measured at in high-glucose conditions divided by insulin measured in low-glucose conditions.

Rat islets were loaded into devices with specialized chambers (3 chambers/device, SI Appendix Fig. S14A) based on the procedures listed above at 3 densities: low (1,000 Islets/cm²), medium (2,500 islets/cm²), and high (5,000 islets/cm²) and cultured in RPMI 1640 and 10% FBS in a specialized 5% O₂ incubator for 7 d. At days 4 and 7, we performed GSIS assays on all devices. On day 7, devices were stained with viability dyes calcein AM ethidium homodimer-1 to reveal live islets and imaged on the epi-fluorescent microscope (SI Appendix, Fig. S14B). Finally, total insulin content from islets within the devices at day 7 was determined using acid/ethanol extraction. The contents of each device chamber were macerated and sheared through an insulin syringe 10 times in acid/ethanol in order to fully break up islets. Samples were then incubated overnight at 4 °C under continuous mixing. Samples were spun the next day 4,000 rcf for 5 min to remove cell debris and the supernatant containing insulin was collected. The amount of insulin per device was quantified via ELISA (Mercodia, cat#10-1145-01) (SI Appendix, Fig. S14C).

All surgical procedures were approved by the Massachusetts Institute of Technology’s (MIT) committee on animal care, as detailed in approved protocol number 0620-045-023. Animal studies were supervised by veterinary staff at MIT’s Division of Comparative Medicine. C57BL/6J mice and STZ-treated C57BL/6J mice were the only animals used in this work. All mice were purchased from Jackson Labs (Bar Harbor, ME).

For surgical implantation, mice were anesthetized under continuous flow of isoflurane and oxygen. Following hair removal, the entire shaved area was aseptically prepared povidone-followed by rinsing and 70% alcohol. A sharp surgical scissor was used to cut a 0.5 to 0.75 cm incision through the skin. A SC pocket of ~1 cm × 1 cm was created via blunt dissection and devices were carefully placed inside the pocket via blunt tweezers. The membrane side was inserted facing down onto the muscle bed, while the cathode side of the device faced the skin. Sutures then closed the skin following implantation. After the surgery, animals were placed back in cages on a heat pad or under a heat lamp and monitored until they came out of anesthesia. Device retrieval proceeded similar to implantation, with care taken not to disrupt fibrotic tissue.

Cell-loaded devices were incubated overnight in cell media and washed with HBSS (Gibco by Life Technologies) before transplantation surgery in healthy C57BL/6J mice. On the day of surgery and every week thereafter, ~100 μL blood was collected from the medial saphenous vein in blood collection tubes (catalogue number 365967; BD Microtainer). After serum separation, the EPO concentration was measured using a Mouse EPO ELISA kit (catalogue number 442708; BioLegend).

Mice were rendered diabetic using a single IP injection of STZ (150 mg/kg). Mice were first weighed and the dosage of STZ calculated. STZ aliquots were wrapped in foil and kept on ice. Immediately before injection, ice-cold STZ buffer (0.57% wt/vol Citric Acid, 0.67% wt/vol sodium citrate, ph 4.5, in DI H20 and sterile filtered) was added to STZ aliquots. Once STZ dissolved ~3 min, animals were injected within 7 min due to the short half-life of STZ. Diabetes occurs approximately 48 h after STZ injection. Blood glucose and weights were evaluated daily, and mice were considered diabetic after two consequent blood glucose of >350 mg/dL. A subset of STZ-diabetic mice (STZ-control groups shown in Fig. 4) were purchased directly from a vendor (Jackson Labs, Bar Harbor, ME) with induction of diabetes validated via successive blood glucose measurements that confirmed fasting BG values >350 mg/dL.

First, 0.5 µL volumes of blood were collected via tail-pricks with 25-27G and measured with a commercial handheld blood glucose meter (Clarity, BG1000). The upper limit of measurements is 600 mg/dL on these systems, and values above this were recorded as 600 mg/dL for analysis. Nonfasting values were recorded for 7 to 10 d following implantation, after which BG values were recorded following a 2.5-h fast. Mice were weighed before and after fasting periods.

For the IPGTT assays shown in Fig. 4, mice were fasted for 8 h and injected with a glucose bolus (2 g/kg in sterile saline) intraperitoneally via 25G needle. Blood samples (~100 µL) were collected into blood collection tubes (BD Microtainer; cat# 365967) via saphenous vein bleeds before and 2 h after glucose injection, followed by serum extraction and insulin ELISA to quantify insulin concentration. BG measurements were performed 15, 30, 45, 60, 90, and 120 min after glucose injection on all groups.

GSP levels were measured from each of the mice at the end of the 28-d study. Two blood samples (each ~100 µL) were collected at day 28 (Fig. 4) in blood collection tubes (BD Microtainer, cat# 365967) via saphenous vein bleeds. Serum was then assayed for mouse-specific GSP using a commercially available ELISA kit (Crystal Chem, cat# 80420) in singlet but using 2 serum collections per mouse according to the manufacturer’s recommendations.

Experimental numbers for all studies are indicated in figure legends. Statistical analysis was performed on OriginPro 2019 (OriginLabs, Northampton, MA). When comparing two independent experimental groups, statistical significance was computed using unpaired, 2-tailed t tests. When comparing independent experimental groups, statistical significance was computed via ANOVA with Bonferroni corrections. When comparing longitudinal data from the same experimental group, statistical significance was computed via paired, 2-tailed t tests. Statistical significance was based on measured P-values of <0.05.

S.R.K., M.A.B., S.B., R.L., and D.G.A. designed research; S.R.K., C.L., M.A.B., N.K., B.W., L.O., A.F., R.L., and D.G.A. performed research; S.R.K., C.L., M.A.B., N.K., R.L., and D.G.A. analyzed data; and S.R.K., R.L., and D.G.A. wrote the paper.

S.R.K, M.A.B, S.B, N.K, R.L., and D.G.A are inventors on a patent application relevant to the technology described in the above work. D.G.A is on the Scientific advisory board of Sigilon Therapeutics, a biotechnology company based in Cambridge, MA, that develops antifibrotic materials for microencapsulated cell-based therapies. R.L. receives licensing fees (to patents in which he was an inventor on) from, invested in, consults (or was on Scientific Advisory Boards or Boards of Directors) for, lectured (and received a fee), or conducts sponsored research at MIT for which he was not paid for a large number of entities. A full list can be found in SI Appendix.