Characterization of human islet function in a convection‐driven intravascular bioartificial pancreas

Abstract Clinical islet transplantation for treatment of type 1 diabetes (T1D) is limited by the shortage of pancreas donors and need for lifelong immunosuppressive therapy. A convection‐driven intravascular bioartificial pancreas (iBAP) based on highly permeable, yet immunologically protective, silicon nanopore membranes (SNM) holds promise to sustain islet function without the need for immunosuppressants. Here, we investigate short‐term functionality of encapsulated human islets in an iBAP prototype. Using the finite element method (FEM), we calculated predicted oxygen profiles within islet scaffolds at normalized perifusion rates of 14–200 nl/min/IEQ. The modeling showed the need for minimum in vitro and in vivo islet perifusion rates of 28 and 100 nl/min/IEQ, respectively to support metabolic insulin production requirements in the iBAP. In vitro glucose‐stimulated insulin secretion (GSIS) profiles revealed a first‐phase response time of <15 min and comparable insulin production rates to standard perifusion systems (~10 pg/min/IEQ) for perifusion rates of 100–200 nl/min/IEQ. An intravenous glucose tolerance test (IVGTT), performed at a perifusion rate of 100–170 nl/min/IEQ in a non‐diabetic pig, demonstrated a clinically relevant C‐peptide production rate (1.0–2.8 pg/min/IEQ) with a response time of <5 min.


| INTRODUCTION
Type 1 diabetes (T1D) is an autoimmune disease that affects around 34 million people worldwide. 1 Clinical islet transplantation by infusion into the portal vein is an attractive treatment for T1D due to its minimally invasive nature. Though islet transplantation has successfully treated patients with unstable T1D, 2 its wider applicability is hindered by tissue donor shortage and the need for chronic immunosuppressive therapy, 3 which has been shown to negatively affect the islets and their recipients. 4 In most cases, achieving insulin independence requires more than one islet infusion, and less than 50% of patients are insulin independent 5 years after intraportal transplantation. 5 The shortage in donor islets is exacerbated by poor engraftment due to inadequate oxygenation during organ procurement and islet preparation for portal venous infusion. After intraportal infusion and before revascularization, oxygen delivery occurs by diffusion from the surrounding blood and liver tissue, resulting in critically low oxygen tensions, below 40-50 mmHg. 6 Moreover, a large percentage of islets are destroyed by the instant blood-mediated inflammatory reaction (IBMIR). 7 Despite the engraftment challenges and complications from immunosuppressive therapy, 8,9 islet transplantation remains a promising experimental treatment for T1D because of its ability to reproduce physiologic insulin secretion kinetics and eliminate hypoglycemic episodes.
Encapsulation is a promising approach to transplant islets without systemic immunosuppression. Our group is investigating the development of an intravascular bioartificial pancreas (iBAP) using silicon nanopore membranes (SNM) fabricated using microelectromechanical systems (MEMS) technology. [10][11][12][13] The SNM feature submicron pores and exhibit high hydraulic permeability at physiologic blood pressures to support increased rates of convective mass transport, and potentially, sustain clinically relevant islet densities by overcoming the limitations of diffusive transport characteristic of extravascular bioartificial pancreas devices. 8 Previous work has demonstrated that the molecular selectivity and hydraulic permeability of SNM are significantly greater than polymeric membranes. [14][15][16] Furthermore, studies with murine islets have shown that the SNM serve as an immune barrier under convective mass transport and support insulin production and islet viability both in vitro and in vivo. [10][11][12] While murine islets are useful for demonstrating preliminary device feasibility, adult human islets are a more clinically appropriate tissue for the iBAP. Hence, we transitioned our testing to adult human islets, and here, we report on the potential of the iBAP to support human islet function.
While our previous islet encapsulation studies utilized both silicon nanopore membranes (SNM) with <50 nm-wide pores 10 and silicon micropore membranes (SμM) with 1 μm-wide pores, 12 we focused this investigation around larger SNM with 450 nm-wide pores to investigate the effects of higher hydraulic permeability on islet function. First, we modeled in silico the oxygen consumption profiles 17,18 of islets seeded at three different densities within scaffolds at in vitro (pO2 = 160 mmHg, atmospheric) and in vivo oxygen levels (pO2 = 95 mmHg, arterial). Then, we evaluated their glucose-insulin kinetic profiles through in vitro glucose-stimulated insulin secretion (GSIS) assays at various density levels and ultrafiltration rates (normalized to islet quantity, also referred to as "perifusion" rates) of 14-200 nl/min/IEQ. The optimal perifusion rate was determined based on the outcomes of computational modeling and in vitro testing. Finally, a proofof-concept demonstration experiment to evaluate insulin production was conducted in vivo via implantation of an iBAP prototype in a non-diabetic pig, followed by intravenous glucose tolerance test (IVGTT).

| Governing equations
The glucose-dependent oxygen consumption model presented here is an adaptation of the work described by Buchwald. 18 This model couples convective flow across the microchannels and diffusive transport with consumption rates across the islet tissue. The Navier-Stokes and mass continuity equations for incompressible Newtonian fluids describe the velocity field (u) due to convection (Equations 1 and 2)   while the diffusion model is defined by the standard diffusion equation for incompressible fluids (Equation 3): where ρ denotes density (kg/m 3 ), η is viscosity (Pa s = kg/m s), p corresponds to pressure (Pa = kg/(m s 2 )), and F is the volume force (N/m 3 = kg/(m 2 s 2 )). In Equation (3), c refers to the concentration of the species of interest (mol/m 3 ), D is the diffusion coefficient (m 2 /s), the del operator r ¼ i ∂ ∂x þ j ∂ ∂y þ k ∂ ∂z and R represents the consumption term or the reaction rate (mol/m 3 s). Both the glucose (Equation 4) and oxygen (Equation 5) consumption rates are assumed to follow Michaelis-Menten-type kinetics. The metabolic demands of insulin production due to changes in glucose levels affect the local oxygen consumption, which is represented by a modulating function (φ o,g ) dependent on glucose concentration (Equation 6). This function is defined by a base-rate φ base ð Þ and a component that changes with metabolic demand along with the insulin secretion rate as a function of glucose concentration. As a first estimate, the base-rate was assumed to represent 50% of the total rate possible and the scaling factor φ sc ð Þ was also equal to 1.8. Furthermore, a step-down function (δ) was also included to account for cell necrosis and suppress the oxygen uptake when its local concentration dropped below the critical value (C cr,oxy ).  Table 1). All islets were assumed to be the size of an islet equivalent (1 IEQ = 150 μm in diameter) and a stepwise increase to 28 mM glucose concentration 19 was added to correlate the in vitro insulin production data to the spatial oxygen distribution at any given time in the GSIS assay.

| Human islet receipt and culture
Freshly isolated human islets were extracted from deceased donor pancreata by the UCSF Islet Production Core (San Francisco, CA). The islets were cultured overnight after isolation at 5% CO2 and 37 C in

| Glucose-stimulated insulin secretion
The islet scaffold presented here builds on the work reported by Song et al. 10 The chambers were fabricated from biocompatible 316L stainlesssteel grade metal that was CNC machined by Hayes Manufacturing Ser- This inter-microchannel distance, which resulted from constraints of the available fabrication methods, is lower than the tissue unit in the oxygen model, which can therefore be considered a "worst-case" scenario.  Table 1).    The IVGTT protocol employed here is an adaptation of the work by Hara et al. 22 Blood glucose was measured at every timepoint with an Accu-Chek Compact Plus glucometer. Due to the high cross reactivity between human and porcine insulin, a human C-peptide ELISA (Mercodia: 10-1141-01) was used to test the ultrafiltrate samples to determine contribution of the encapsulated islets.

| Statistical analysis
Results were expressed as the mean ± standard deviation of the mean (SD).
Multiple sample comparisons were done with two-way analysis of variance (ANOVA) followed by post hoc Tukey test while sample pairs were evaluated using Student's t-test with the Holm-Sidak correction method.
All statistical analyses were performed with GraphPad Prism 6 (San Diego, CA) and p values <0.05 were considered statistically significant.

| Microchannel oxygen modeling
The perifusion rates were modeled at incoming atmospheric (160 mmHg) and arterial (95 mmHg) oxygen tensions to recreate in vitro and in vivo conditions (Figure 5a). The resulting oxygen profiles were extracted 13 min after the introduction of high glucose to capture the maximal firstphase insulin release. The oxygen concentration steadily decreased with the radial distance of the tissue unit as oxygen diffused from the microchannel into the agarose-islet region where it was consumed by the islets.
As expected, the islets farthest away from the microchannel in the radial direction showed the lowest oxygen concentration and represented the worst-case scenario within the islet scaffold. Also, the oxygen concentration within the tissue unit (both axial and radial directions) increased as the incoming perifusion rate was increased. The post-processing cut-line feature in COMSOL was used to visualize the oxygen concentration in the worst-case scenario for each condition. The results obtained from the simulation at atmospheric oxygen tension indicate that 28 nl/min/IEQ (10.0% density, 50 μl/min) is the lowest tested rate supporting islet function; since its worst-case scenario drops slightly below 25 mmHg (0.034 mol/m 3 ), which corresponds to the threshold for uninhibited maximal insulin production 23,24 (Figure 5b). Simulations at arterial pO2 levels revealed that the maximal insulin production was supported by perifusion rates ≥100 nl/min/IEQ (2.5% density, 50 μl/min) (Figure 5c).

| Effect of perifusion rate in vitro, pO 2 = 160 mmHg
The five perifusion rates that were modeled were also evaluated for GSIS. In vitro results show that 28 nl/min/IEQ (10.0% density, F I G U R E 5 Simulated in vitro and in vivo oxygen concentrations in the tissue unit as a function of islet perifusion rate 10 min after introducing glucose (28 mM) in the bulk fluid (a). Surface plots of oxygen concentration gradient across the tissue unit with radial and longitudinal cross-sections halfway through the unit. The average oxygen concentration at the islet cores farthest from the microchannel (worst-case scenario) plotted at (b) 160 mmHg and (c) 95 mmHg inlet pO 2 . Simulations at arterial pO 2 levels suggest that at least 100 nl/min/IEQ (***) is required to be within the oxygen threshold of uninhibited maximal insulin production (0.034 mol/m 3 ). See Table 1 for correlation between perifusion rate and islet density loading levels. Statistical significance is expressed as *p < .05 and **p < .001 50 μl/min) is the lowest perifusion rate that can sustain islet function in the microchannel islet scaffold as a marked response to changes in glucose levels can still be observed (Figure 6a) (Figure 6b).  Table 1 for correlation between perifusion rate and islet density loading levels. Statistical significance is expressed as *p < .05 and **p < .001 subcutaneous tissue. 35 Another promising approach is the use of prevascularized devices to facilitate revascularization in the islets. 36,37 Intravascular BAP devices may improve GSIS compared to extravascular devices as they can deliver arterial pO2 levels (80-100 mmHg) to the encapsulated islets, compared to the low pO2 levels (10-50 mmHg) at the surface of extravascular devices. 38 Previous groups have achieved long-term xenogeneic islet function in intravascular diffusion-based devices without immunosuppression 39-41 ; however, the need for exogenous insulin was not fully eliminated and their translation to the clinic was obstructed by device patency issues and failure at the artery/device connection. 42 Furthermore, GSIS delays were observed in both the diffusion-based 43  where a slight delay in insulin production and a lower stimulation index were observed, the latter which is due to higher baseline insulin production at low glucose. Despite the slight delay in the GSIS profile, the 200 nl/min/IEQ perifusion rate still exhibits the desirable features of the biphasic response.

| In vivo intravenous glucose tolerance test
After connection to the vasculature of the non-diabetic swine, the SNM-encapsulated human islets, which are exposed to hemoglobin-free ultrafiltrate, exhibited promising C-peptide production. The C-peptide F I G U R E 7 In vivo intravenous glucose tolerance test (IVGTT) in a non-diabetic pig. The SNM-based iBAP with 500 IEQ and a perifusion rate of 100-170 nl/min/IEQ showed stable C-peptide production in the fasting period. Time 0 marks the administration of the glucose bolus, and elevated C-peptide production is observed as the glucose concentration increases. The C-peptide production approached its basal level as the blood glucose returned to fasting levels.
production curve, which peaked at $4.0 pg/min/IEQ, generated from the in vivo IVGTT with the non-diabetic pig corresponds to in vitro perifusion rates between 100 and 200 nl/min/IEQ. During the stabilization period at fasting glucose levels, the C-peptide production was stable but trending slightly downward and comparable to in vitro studies at low glucose levels. This data suggests that human islets within the iBAP can sense plasma blood glucose and secrete insulin in a physiologically normal pattern during fasting glucose levels. 48 Furthermore, after initiating the IVGTT, the adult human islets rapidly released insulin, as determined by human C-peptide measurements, with a time delay of <5 min.
This rapid release of insulin is a key requirement for a functional bioartificial pancreas to achieve normoglycemia. It has been estimated the GSIS response must be <15 min for normal physiologic BAP funciton. 51 Interestingly, the glucose-insulin kinetics exhibited an oscillating profile as opposed to the biphasic pattern observed during in vitro GSIS studies with the SNM-based iBAP and islet perfusion studies with adult human islets. 49,52,53 Instead, the islets displayed a pulsatile secretion pattern Our investigation is associated with several limitations that must be satisfactorily addressed for the successful development of a scaled-up iBAP suitable for future clinical translation. The in vitro studies were short-term and conducted with an iBAP prototype that held no more than 3600 islets. While the in vivo pig study with the implanted iBAP showed some promising preliminary data, it was performed with lowest islet density (2.5%) for just 90 min and with a single pig. Future work will need to examine effects of increased islet loading density levels and C-peptide trends during the low glucose phases of the IVGTT. Studies will need to be conducted with a statistically significant number of pigs and for longer periods (few hours to multiple days to many months). To compare our iBAP results more readily with published literature on encapsulated islets, experiments could be performed at low and high glucose levels of 2.8 and 16.7 mM, respectively, and use a physiologic salt solution as the medium for GSIS experiments to avoid the confounding influence of insulinotropic factors in culture media. For translational relevance to the clinical setting, the iBAP design will need to be scaled up to house an increased SNM area, and therefore generate higher ultrafiltrate volumes, to support a greater number of islets.

| CONCLUSIONS
The prototype SNM-based iBAP supported adult human islet function in vitro as well as in a healthy Yucatan pig. The oxygen profile models showed that a minimum perifusion rate of 28 nl/min/IEQ and 100 nl/min/IEQ is needed to sustain islets for glucose production in vitro and in vivo, respectively. The animal test demonstrated the potential feasibility of a future scaled-up device to provide clinically relevant C-peptide production with 100-200 nl/min/IEQ perifusion rates. Based on simulated oxygen profiles and insulin production in both basal and stimulatory phases, the results of this investigation will inform future islet dosing and device scalability studies required to systemically deliver insulin to treat pigs with chemically induced T1D, and ultimately, T1D patients.

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