Amplifying the Impact of Kidney Microphysiological Systems: Predicting Renal Clearance Using Mechanistic Modelling Based on Reconstructed Drug Secretion *

Accurate prediction of pharmacokinetic parameters such as renal clearance is fundamental to the development of effective and safe new treatments for patients. However, conventional renal models have a limited ability to predict renal drug secretion, a process that is dependent on transporters in the proximal tubule. Improvements in microphysiological systems (MPS) have extended our in vitro capabilities to predict pharmacokinetic parameters. In this study a kidney-MPS model was developed that successfully recreated renal drug secretion. Human proximal tubule cells grown in the kidney-MPS, resembling an in vivo phenotype, actively secreted organic cation drug metformin and organic anion drug cidofovir, in contrast to cells cultured in conventional culture formats. Metformin and cidofovir renal secretory clearance were predicted from kidney-MPS data within 3.3-and 1.3-fold, respectively, of clinically reported values employing a semi-mechanistic drug distribution model, using kidney-MPS drug transport parameters together with in vitro to in vivo extrapolation. This approach introduces an effective application of a kidney-MPS model coupled with pharmacokinetic modelling tools to evaluate and predict renal drug clearance in humans. Kidney-MPS renal clearance predictions can potentially complement pharmacokinetic animal studies and contribute to the reduction of pre-clinical species use during pre-clinical drug development.


Introduction
In the kidneys, secretion plays a major role in eliminating xenobiotics and metabolic by-products from the systemic circulation into the urine (Feng et al., 2010). The pharmacokinetic (PK) profile of renally cleared small molecule therapeutics is impacted by the activity of an array of both basolateral and apical transporters expressed in the proximal tubules (Figure 1-A). By actively removing a broad variety of substrates, these transmembrane proteins are determinants of drug distribution and efficacy (Yin and Wang, 2016;Moss et al., 2014). Drug-drug interactions leading to altered PK or nephrotoxicity (Sjögren et al., 2018) illustrate the significance of renal drug transporters and the importance of their functional evaluation during drug development (Zamek-Gliszczynski et al., 2018).
During pre-clinical drug development, the prediction of renal clearance has long-relied on established in vitro and in vivo models. Cell lines are easily accessible and popular tools for characterizing drug uptake and PK parameters. However, a key challenge with the use of established cell lines is their limited ability to model and predict normal physiological conditions where transepithelial flux of molecules is mediated by a complex and coordinated system of transporters from multiple families. Immortalized proximal tubule cell lines (iRPTEC) that exhibit a more native phenotype are a valuable tool for nephrotoxicity assays. Nonetheless, epithelial membrane polarity is limited as well as the expression of key uptake drug transporters (e.g. organic anion transporters, OATs) (Caetano-Pinto et al., 2016;Jenkinson et al., 2012). Membrane polarity can be improved with the use of conventional transwell systems or capillary hollow fibres (Jansen et al., 2015). Growing cells on permeable porous membranes enables their basolateral and apical sides to face different compartments. However, in the hollow fibres characterized by Jansen et al. (2016), the basolateral membrane of the iRPTEC faces the lumen of the fibres, in

2.3.
Fluorescent functional and permeability assays Fluorescent probes were used to evaluate the activity of RPTEC cultured in both microplates and MPS, as well as the permeability of the kidney-MPS renal tubule. P-gp and MRP activity was determined using calcein-AM and carboxyfluorescein diacetate succinimidyl ester (CFDA-SE), respectively. Both molecules diffuse through the cell membrane and are subsequently metabolized to their fluorescent forms (calcein and CFDA). Efflux activity was determined by blocking P-gp and MPRs with selective inhibitors valspodar and MK571, respectively, resulting in increased intracellular fluorescence. Assays were performed by incubating cells with 1 μM calcein-AM ± 5 μM valspodar or 2 μM CFDA-SE ± 50 μM MK571. In 2D assays, cells were incubated for 30 min in HBSS at 37°C and cellular fluorescence intensities were subsequently quantified using a Polar-star Optima microplate reader (BMG Labtech). In the kidney-MPS, the substrates (± inhibitors) were perfused via the cell-free channel for 2 h and fluorescence accumulation in the RPTECs or matrix was determined using a CV7000 (Yokogawa) automated high-content confocal microscope. To determine monolayer integrity in the kidney-MPS, 25 mg/mL inulin-FITC in RPTEC medium was perfused through the cell-free channel. Renal tubule perfusate was collected daily, for 7 days, and fluorescence quantified using a Polar-star Optima microplate reader). Furthermore, inulin-FITC (25 mg/mL in HBSS) was perfused either through an empty kidney-MPS-chip or the lumen of the renal tubule for 3 h. Fluorescence in renal tubule or extracellular matrix was evaluated using an Evos TM FL microscope (Invitrogen).

2.4.
Immunofluorescence RPTEC cultured in microplates were washed with HBSS, fixed with 4% formaldehyde for 10 min at room temperature and permeabilized using 0.1% Triton X (v/v)/2% BSA (w/v) in HBSS (permeabilization-buffer) for 1 h. Kidney-MPS chips were washed with HBSS, under continuous perfusion (flow rate: 1 µL/min) for a minimum of 2 h and subsequently fixed with 4% formaldehyde for 30 min at room temperature (flow rate: 2.5 µL/min). Following an additional HBSS wash, chips were permeabilized with permeabilization-buffer for 1 h (flow rate: 2.5 µL/min). Incubation with the primary antibodies in both microplates and MPS chips was performed static (no perfusion) overnight at 4°C in permeabilization-buffer. Subsequently, fluorescently conjugated secondary antibodies (Invitrogen) were incubated for 1 h at room temperature, together with Hoechst 33342 (1:1000, Invitrogen) and Thermofisher). In Figure S1 1 shows the specificity of the OAT1, OCT2 and P-gp antibodies used and Table S2 1 lists antibodies and dilutions used. Image acquisition was carried out on a CV7000 automated confocal microscope (Yokogawa) or an LSM 700 confocal microscope (Zeiss).

2.5.
Immunofluorescent image analysis Acquired images were processed and analysed using the open-source imaging software Fiji (Schindelin et al., 2012). The number of cells in the kidney-MPS renal tubule was estimated by counting Hoechst-stained nuclei in three selected tubule sections, reconstructed from z-stacks comprising 25 images. Cilium length was estimated from z-stack images by counting positive images for TBN. Cell height and tubule diameters were estimated from the pixel/µm ratio using images representative of the renal tubule mid-section. Fluorescence intensity was measured in sets of 3 individual images per replicate to estimate calcein and CFDA retention levels.

2.6.
Gene expression analysis Expression of pivotal proximal tubule markers was analysed using custom-designed TaqMan™ gene expression array cards (Thermo Fisher, Waltham, USA; Tab. S3 1 ). RPTEC from a single donor were cultured in kidney-MPS or in plastic transwell ALTEX, accepted manuscript published November 3, 2022 doi:10.14573/altex.2204011 5 microplates (pore size 0.4 µm; Millipore); three biological replicates, corresponding to different RPTEC aliquots were used. 2D and trans-wells-cultured cells were harvested after DPBS washing (37°C), followed by directly adding RLT lysis buffer (RNeasy® kit, Qiagen, Venlo, The Netherlands). Cells in MPS chips were perfused with HBSS (37°C) at 1 µL/min for 2 h, followed by perfusion with RLT lysis buffer at 5 µL/min for 20 min collecting lysate at the renal channel. After incubating at room temperature for 30 min, kidney-MPS chips were perfused with additional RLT lysis buffer at 5 µL/min for 30 min to collect additional cell lysates. Two human kidney cortex samples were obtained frozen (BioIVT, Burgess Hill, UK). One fresh tissue kidney sample was obtained from a nephrectomy (Sahlgrenska University Hospital, Gothenburg, Sweden) for which collection complied with the Swedish Biobanks in Medical Care Act. Tissue samples (3x3x3 mm) were transferred to microcentrifuge tubes containing RLT lysis buffer (600 µL) and lysed for 2 min at 25 Hz using a TissueLyser (Qiagen). Total RNA of all samples was isolated using the RNeasy® kit (Qiagen) and cDNA was generated using the High Capacity Reverse Transcriptase kit (Applied Biosystems) in a Eppendorf Master-cycler. Taqman Array Cards were loaded with 300 ng of cDNA using TaqMan Fast Advanced Master Mix (Qiagen) according to the manufacturer's instructions. The quantitative polymerase chain reaction was performed on a Quantstudio 7 Flex thermocycler (Applied Biosystems). Data was analysed by the 2 -ΔCt and 2 -ΔΔCt methods (Livak and Schmittgen, 2001) to compare absolute and relative gene expression, respectively, using HPRT1 as the housekeeping reference gene.

2.7.
Statistical analysis Gene expression data was analysed using GraphPad Prism 8.00 (GraphPad Software, La Jolla California USA). An unpaired t-test corrected for multiple comparisons using the Holm-Sidak method with statistical significance of 0.05 was used to compare the different experimental groups. The 2D-Plastic samples were used as the reference group and analysis was performed based on the ΔCt values.

2.8.
Perfusion and drug transport assays For each drug transport assay, a maximum of four chips were coupled to 1 mL syringes (BD) mounted on a multi-lane syringe pump (Perkin Elmer). Temperature was maintained at 37°C using a heating-mat. Both microfluidic circuits were perfused at a flow rate of 1 µL/min, with flow initiated simultaneously upstream of the collagen chamber ( Figure 4, A2-A3). The shear stress experienced by cells lining the tubule under assay conditions is estimated at ~0.7 dyne/cm 2 (Tab. S7 1 ). Perfusate samples from both channels were collected from the chip outlets in 30 min intervals over a period of 360 min. Perfusate volume was permanently monitored and chips excluded from the analysis if the perfusion became inconsistent. Drug transport assays were performed in HBSS/1% FCS/10 mM HEPES with different pH (basolateral: 7.4, apical: 6.5). [ 14 C]-metformin and [ 3 H]-cidofovir (Perkin Elmer) were prepared in solution with a ratio of 1:100 for radioactive:non-radioactive, each with a total concentration of 10 µM. This concentration was chosen to be below the Km values reported for both drugs for their primary uptake transporters (Morrissey et al., 2012) Drug exposure was achieved by perfusing the radiolabelled solutions via the cellfree channel (loading channel -basolateral compartment) and buffer via the renal channel (apical compartment). Assays were performed in the presence or absence of selective drug transport inhibitors imipramine (500 µM) or probenecid (1 mM), for metformin and cidofovir, respectively. These concentrations were selected to be substantially higher than the reported IC50 values for OCT1 and OCT2 (Morrissey et al., 2012) to maximize the inhibitory potential. Inhibitors were used both apically and basolaterally to achieve maximal inhibition. 5 µL of each perfusate sample was added to 5 mL of Ultima Gold scintillator (Perkin Elmer) and radioactivity quantified in a Tri-Carb 2100TR (Perkin Elmer). To determine transepithelial transport in static conditions, transwell microplates (Millipore) were used to perform assays under the same experimental conditions aforementioned, but with an incubation period of 60 min instead. Determination of drug flux across the tubule and in transwell cultures is detailed in the supplementary information: Section 4 (Tab. S4, Fig. S3 1 ).

2.9.
Mathematical modelling and in vitro to in vivo extrapolation A semi-mechanistic mathematical model of drug disposition in the kidney-MPS platform was developed and implemented in MATLAB (Release 2020b, The MathWorks, Inc., Natick, Massachusetts). Briefly, the model considers the net dispersion of drug across the physical extracellular matrix separating the loading and renal microfluidic channels, and, in addition, the basolateral-to-apical transport into the renal channel ( Fig. 6A; Tab. S8 1 ). Fitting the model to the outlet concentration-time data allows for estimation of the key in vitro kinetic parameters Qtr, PSp and PSa, representing the first-order rate constant for flux through the ECM, passive and active permeability-surface area product, respectively. Human renal clearance was predicted from PSp and PSa by accounting for derived surface area in vivo (Tab. S9 1 )and expressing kidney organ clearance as the net result of glomerular filtration, tubular secretion, and fractional tubular reabsorption following the method established by Kunze et al. (2014). Predictability was assessed by comparing estimates to reported clinical values before and after correction for filtered fraction. Since secretory clearance is not directly given from observations, calculation of this component was based on assuming a fraction reabsorbed estimated from MPS (Eq. S9 1 ) and a filtration clearance equal to fu x GFR (Eq. S10 1 ). A comprehensive description of the modelling and in vitro to in vivo extrapolation strategy, including parameters, assumptions and equations is provided in the supplementary information 1 .

3.1.
Kidney-MPS recreates renal proximal tubule morphology A comparative immunofluorescence analysis of RPTEC grown in conventional two-dimensional culture (2D-Plastic) and the kidney-MPS revealed that culture under microfluidic conditions enhanced crucial renal proximal tubule cell features, as summarized in Figure 1. In 2D-Plastic, RPTEC maintained dense tight junctions (Zonula Occludens 1: ZO1) present between cells, clearly highlighting cellular boundaries. Cells presented a single primary cilium (acetylated tubulin: TBN) of ~10 μm in length. Cytoskeleton filaments were concentrated in the cell periphery (f-actin) and co-localized with ZO1. The two major renal uptake transporters Organic Anion Transporter 1 (OAT1) and Organic Cation Transporter 2 (OCT2) were identified intracellularly in the perinuclear space and were seemingly not present in the cell membrane ( Figure S2-A 1 ). The efflux transporter P-glycoprotein (P-gp) was detected both in the cell membrane and intracellularly ( Figure S2-B 1 ). These observations demonstrate that, despite retaining some key features, RPTEC polarity is reduced in conventional 2D culture. In the kidney-MPS, RPTEC populate a channel with ~125 μm in diameter and form a highly polarized, monolayered tubule. This 3D renal RPTEC tubule comprises approximately 10,000 cells ( Fig. S4 1 ) with a height of about 20 µm and with well-defined basolateral and apical membranes. The renal tubule experiences shear stress between 0.3 and 0.7 dyne/cm 2 (Tab. S7 1 ), which is within the previously reported physiological range (Vedula et al., 2017).
Automated confocal microscopy image acquisition (Peel et al., 2019) determined that the tight junction marker ZO1 was homogenously expressed in all cells along the tubule and predominantly localized to the apical membrane. F-actin was mainly expressed at the cell periphery and cells displayed a single primary cilium extending into the lumen, with a length of ~20 μm. The drug uptake transporters OAT1 and OCT2 were consistently observed in the basolateral membrane whilst the efflux pump P-gp was localized in the apical membrane. Furthermore, the ion channel Sodium-Potassium ATPase (Na + K + -ATPase) was also determined to be localized in the basolateral membrane. From these findings it is evident that RPTEC exhibited strong epithelial polarization when arranged in a tubular conformation and grown under continuous flow. Moreover, the evaluated drug transporters were correctly localized in the membrane, a characteristic which was absent in conventional 2D culture.

3.2.
Kidney-MPS drug efflux transport activity and permeability The functional activity of RPTEC efflux transporters was demonstrated using fluorescent substrates in combination with prototypical inhibitors, in both 2D-Plastic and the kidney-MPS (Figure 2-A1-B1).
Activity of Multidrug-Resistance-associated Proteins (MRPs) and P-gp was evaluated using the substrates carboxyfluorescein diacetate succinimidyl ester (CFDA-SE) combined with the inhibitor MK571 or calcein-acetoxymethyl (calcein-AM) combined with valspodar, respectively. In 2D-Plastic, the selective inhibition of MRPs and P-gp led to a 5.7 ± 0.4 and 3.5 ± 0.3-fold increase in fluorescent substrate retention in the cells, respectively (Figure 2-C). In the kidney-MPS, CFDA efflux under control conditions resulted in minimal intracellular retention which could not be quantified. Therefore, MRP efflux activity was defined as CFDA extracellular fluorescence accumulation, for which an 8.1 ± 0.1-fold increase was observed when un-inhibited. P-gp inhibition led to a 5.4 ± 0.1-fold increase in fluorescence retention in the renal tubule ( Figure  2-D).
Inulin-FITC diffusion takes place solely paracellular and is independent of active transport. Inulin transfer from the basolateral to the apical side of the renal tubule decreased substantially over the initial culture period. The trace levels of inulin-FITC fluorescence detected in the luminal effluent after day 3 reflect the presence of dense tight-junctions in the kidney-MPS and a highly impermeable epithelial barrier. Conversely, when perfused apically inulin-FITC is contained inside the renal tubule, while in a cell-free chip (therefore without barrier function) extensive inulin-FITC fluorescence was observed throughout the extracellular matrix. In the kidney-MPS, RPTECs grown into a self-assembled renal tubule showed substantial apical drug transporter activity while forming an impervious monolayer. These results show that the key traits crucial for the study of drug transport across renal proximal tubule epithelium are present in this kidney-MPS.

MPS platform enhances and potentiates the expression of renal drug transporters
To further compare the kidney-MPS to 2D culture conditions, the gene expression profile of key proximal tubule genes, including those encoding drug transporters, membrane channels and RPTEC markers, was assessed across the culture formats. Hypoxanthin-Guanin-Phosphoribosyl-transferase (HPRT1) was used as the housekeeping reference gene for data normalization (Tab. S5 1 ). The analysis was performed using RPTEC from a single donor, grown in 2D-Plastic, on transwell filters (2D-Transwell) or in the kidney-MPS for 2 or 7 days (MPS Day 2 and MPS Day 7). Human renal kidney cortex samples, obtained from 3 donors, were used as an in vivo reference.
This analysis revealed significant differences in mRNA expression levels between 2D-Plastic, 2D-Transwell and kidney-MPS for the analysed genes (Tab. S4 1 ), which is summarized in Figure 3. For drug uptake transporters, gene expression of OCT2 was similar in MPS and 2D-Transwell but slightly reduced compared to 2D-Plastic. OAT3 expression was undetectable in the 2D culture formats as well as the kidney-MPS. OAT1 showed the starkest difference observed between culture conditions. Completely absent in 2D, OAT1 gene expression was substantially recovered in the kidney-MPS. For drug efflux transporters, MRP4 and P-gp showed the highest absolute expression levels of all drug transporters tested. MRP4 expression was maintained in 2D-Plastic, 2D-Transwell and kidney-MPS, while P-gp expression was decreased in the kidney-MPS compared to 2D-Plastic and 2D-Transwell. Relative expression of MATE1 and MATE2-K was strongly increased from 2D-Plastic to kidney-MPS Day 7 (86.8 ± 48.4-fold and 13.7 ± 4.6-fold, respectively).
In addition, the expression of the key renal proximal tubule markers Aquaporin 1 (AQP1), Sodium-Glucose Transport Protein 2 (SGLT2), Sodium-dependent Phosphate Co-transporter (NaPi), Megalin (LRP2) and Cubilin (CUBN) was confirmed in the MPS-cultured cells. Interestingly, NaPi was expressed in the kidney-MPS and 2D-Transwell but not in 2D-Plastic. AQP1 showed a high expression in 2D and was further upregulated in the kidney-MPS. The Hepatocyte Nuclear Factor 4-Alpha (HNF4α), that regulates the expression of renal drug transporters, was also substantially increased in kidney-MPS Day 7 (47.9 ± 24.0-fold relative to 2D-Plastic). These results show that the expression of specific proximal tubule drug transporters and membrane proteins is maintained, enhanced and even recovered in the kidney-MPS, with the exception of Pgp which appears downregulated. Gene expression levels of all genes investigated were evidently lower in the kidney-MPS compared to human renal cortex (Tab. S5 1 ). This observation supports previous reports describing a reduction in mRNA transcripts when RPTECs are isolated and cultured in vitro (Verhulst et al., 2008). Nevertheless, the kidney-MPS expressed all drug transporter genes analyzed except OAT3. MRP4, P-gp and AQP1 maintained high expression levels, while the most pronounced reduction was observed for OAT1. Overall, these gene expression results demonstrate that, despite losses in expression relative to renal tissue, kidney-MPS cultured RPTECs demonstrate improved gene expression of drug transporters compared to 2D cultures.

3.4.
Kidney-MPS enables drug secretion through active transepithelial transport Activity of drug transporters is fundamental for renal secretion. To determine transepithelial transportmovement of compounds across the cellular barrierthe secretion of prototypical renally cleared drugs, was assessed in both 2D-Transwell ( Fig. 5A; Tab. S6 1 ) and kidney-MPS cultures. All assays were performed using RPTEC from a single donor. For every experiment, a different batch of cells (aliquots) was expanded once after thawing. Each experimental condition was performed independently using a total of 3 to 4 kidney-MPS chip replicates.
In conventional 2D culture, RPTECs did not demonstrate any drug secretion capability ( Figure 5-C1, C2). Transepithelial transport was assessed by exposing RPTECs, grown on semi-permeable transwell inserts, to metformin or cidofovir. Limited drug flux was observed from the basolateral to the apical compartment (B2A) which would model the movement of drug from blood to urine. Further, no differences were observed in the apparent permeability coefficients (Papp; Tab. S6 1 ) for both drugs from B2A and vice-versa (A2B), either in the presence or absence of inhibitors. Active drug secretion would result in an increased (and inhibitable) permeability from B2A compared to A2B.
In contrast to 2D setups, the compartments adjacent to the basolateral and apical cell surfaces in the MPS are not directly accessible for drug exposure or sampling. Instead, drugs are dispensed via inlet ports into the chips and perfusate is collected from the outlet ports downstream of the extracellular matrix chamber, where exchange of constituents between the drug donating (loading) and the receiving (renal) channels occurs (  (1) were filled with the radio-labelled drugs or buffer and coupled to a 10lane syringe pump (2). Flex-tubing (3) was used to connect two syringes per kidney-MPS chip (4). Each chip comprises two independent microfluidic circuits, corresponding to a loading channel (used to perfuse drug, basolateral compartment) and a renal channel (apical compartment), populated or not with RPTEC. Buffer was perfused via the renal channel inlet (5) and radio-labelled drugs via the loading channel inlet (6). Samples of perfusate were collected from both renal and loading channel outlets (7, 8) every 30 min for 6 h. Temperature was maintained during the assays at 37°C using a heating mat (9). Chips were perfused at a flow rate of 1 μL/min which enables the collection of 30 μL of perfusate every 30 min. The chip perfusion channels converge into a collagen I filled chamber (A3) where drug exposure takes place. The compounds perfused via the loading channel diffuse through the collagen matrix and reach the renal channel, perfused with buffer. To determine the perfusion profiles at the loading (B) and renal (C) channel outlets, corresponding to the drug concentrations available in the collagen matrix, 10 μM of metformin (red circles) or cidofovir (blue diamonds) were perfused via the loading channel of cell-free microfluidic chips over 6 h. Cell-free chips do not incorporate RPTEC (absent barrier function) and allow the evaluation of perfusion driven drug distribution in the chip. Experimental values are representative of the mean and associated standard error of the mean (n=3). It is evident that the perfusion in both the loading and renal channels follows a similar profile. The drug concentration at the outlet of the loading channel approaches levels similar to the input from approximately 200 min, while at the renal channel outlet the concentration is substantially lower.
with no barrier function, metformin and cidofovir perfusion displayed overlapping concentration-time profiles at the renal outlet, reaching detectable levels from approx. 1.5 h and an apparent steady-state plateau within approx. 4 h of perfusion ( Figure 4B-C), indicating that in absence of an RPTEC barrier drug dispersion in the MPS is governed primarily by the MPSchip flow rate. RPTECs lining the renal channel form a selective barrier to enable the translocation of drug into the renal lumen. Co-perfusion of selective inhibitors of OCT2 (imipramine) or OATs (probenecid) were used to evaluate the contribution of active drug secretion to the overall transepithelial transport of metformin and cidofovir (Wang et al., 2014;Takeda et al., 2001). The levels of both substrates in the renal perfusate were reduced in the presence of the renal tubule in relation to the cell-free set-up ( Figure 5-D1, D2), reiterating the barrier function in the presence of RPTEC and, further diminished in the presence of the inhibitor, corresponding to approx. 13% and 21% of the steady-state concentration reached after perfusion of metformin and cidofovir alone (Figure 6-B1, C1).

3.5.
Semi-mechanistic modelling of Kidney-MPS drug-transport combined with in vitro to in vivo extrapolation can correctly predict renal drug clearance While the empirical data clearly displayed the activity of key renal transporters in the kidney-MPS, inferring a quantitative meaning to observations requires deconvolution of outlet drug concentration-time profiles. To this end, a bespoke mathematical modelling framework encompassing the temporal aspects of drug distribution along with the fundamental pathways governing transepithelial renal transport was developed (supplementary Information, Sections 8,9 and 10 1 ). Sequential fit of the semi-mechanistic model to kidney-MPS data from the cell-free channel and renal tubule in absence and presence of specific inhibitors (Figure 6-B2, C2) allowed for estimation of the active and passive renal permeability-surface area products, PSa and PSp, respectively. In turn, the human renal clearance of each drug was predicted by scaling the MPS permeability parameters on basis of the tubular surface area in vitro relative to that of the in vivo physiology (Kunze et al., 2014). Table 1 summarizes the MPS parameter estimates along with the predicted and clinically observed total renal clearance (CLr). Remarkably, mean CLr predicted from the kidney-MPS matched clinical observations for cidofovir (1.3-fold), while ALTEX, accepted manuscript published November 3, 2022 doi:10.14573/altex.2204011

Fig. 5: Transepithelial flux in 2D-Transwell and Kidney-MPS. (A)
Schematic representation of the cross-section a 2D conventional transwell system. RPTEC grow on a semi-permeable membrane which delineates the static basolateral and apical compartments. Drugs are exposed directly to the cells in the basolateral compartment. In the continuously perfused kidney-MPS (B) RPTEC form a renal tubule embedded in an extracellular matrix (ECM), with cells acting as a barrier thereby defining the basolateral (ECM) and apical (renal tubule lumen) sides. Drug exposure is continuous into the ECM via a channel parallel to the renal tubule. Metformin flux over a period of 60 min, from the basolateral to the apical side, in 2D-Transwell was 8.2 ± 5.0 µmol cm -2 min -1 and 7.9 ± 5.9 µmol cm -2 min -1 when inhibited by imipramine (n=3) (C1). In the kidney-MPS, metformin flux was 73.8 ± 52.6 µmol cm -2 min -1 (n=4) and 4.9 ± 1.5 µmol cm -2 min -1 when inhibited (n=3) (D1). Cidofovir flux in 2D-Transwells, over 60 min, was 5.0 ± 2.9 µmol cm -2 min -1 and 4.2 ± 0.3 µmol cm -2 min -1 when inhibited (n=3) (C2). In the kidney-MPS, cidofovir flux was 21.8 ± 19.7 µmol cm -2 min -1 (n=4) and 6.6 ± 2.5 µmol cm -2 min -1 when inhibited (n=3) (D2). Flux in the kidney-MPS reflects the activity observed over the course of 180 min after continuous drug flow was started. Flux is calculated as the quantity of a drug crossing the area of the barrier (RPTEC monolayer) from the donor compartment (loading channel) to the recipient compartment (renal channel effluent) over time, and it is expressed as mean ± standard error of the mean. Transwell assays were performed with cells from a single donor and represent three individual experiments comprising three technical replicates each.
metformin CLr was within 2.2-fold of reported values. Predicted secretion clearance CLr,sec was similarly in agreement with estimated observed values for cidofovir (1.3-fold), while metformin predictions were 3.3-fold off the observations. This suggest that the enhanced representation of the renal physiology in the MPS setting provides more physiologically relevant in vitro parameters, which can be used in combination with quantitative modelling approaches to improve the prediction of renal transporter-mediated drug clearance.

Discussion
Accurate prediction of clearance by the kidneys is essential for drugs in development. Although the renal clearance of compounds that are mainly eliminated by glomerular filtration is reasonably estimated by established approaches, clearance of low permeability candidate drugs that undergo active tubular secretion is poorly predicted (Mathialagan et al., 2017). The main reason for this is the limited active drug secretion capacity of current in vitro systems, mostly 2D-cultured cells (Scotcher et al., 2016). Due to their ability to partially mimic in vivo physiology, MPS models have received interest from the biotech and pharmaceutical sectors, with the goal to enhance in vitro cultures and ultimately improve in vitro to in vivo translation of transporter-mediated drug clearance (Cirit and Stokes, 2018;Ewart et al., 2017;Neuhoff et al., 2013;Huang and Isoherranen, 2018). In this study, a kidney-MPS model that demonstrates a tight and polarized human renal proximal tubule epithelium which recapitulates the expression of cationic and anionic drug transport pathways, enables robust evaluation of transepithelial flux. When transepithelial flux measurements were combined with a bespoke mathematical modelling strategy, transportermediated renal clearance of metformin and cidofovir was accurately predicted.

ALTEX, accepted manuscript published November 3, 2022 doi:10.14573/altex.2204011
Tab. 1: Estimated model parameters and scaled renal clearance Calculated assuming physiological parameters as specified in supplementary information 1 . Clinical PK as reported for metformin and cidofovir by Graham et al.(2011) andCundy et al.(1995), respectively. Secretion clearance calculated from total renal clearance assuming the fraction reabsorbed estimated from MPS and a filtration clearance = fu x GFR.  (Cundy et al., 1995) ml/min 481 (180 -782) 69 (0 -189) Obs. / pred. Conventional renal proximal tubule cultures demonstrate poor epithelial polarization (Figure 1-B) and limited secretion capacity of cationic drugs, such as metformin and cisplatin (Motohashi and Inui, 2013;Elsby et al., 2017). By incorporating continuous luminal shear stress through perfusion and a tubular architecture, the kidney-MPS enables cells to polarize with spatially correct localization of drug transporters in the basolateral and apical cell membranes (Figure 1-C). These features likely enable the substantial improvement in renal drug transporter activity (Caetano-Pinto and Stahl, 2018), responsible for the transepithelial transport of clinically relevant drugs in our kidney-MPS model. Shear stress was implicated in facilitating the transport of organic cations, demonstrated by increased OCT2 and MATE1 transport activity in transfected MDCK cell lines in dynamic culture (Jayagopal et al., 2019), as well as enhanced MATE2-K mediated transport in primary RPTEC under flow (Fukuda et al., 2017). Recently, cisplatin nephrotoxicity was only observed after basolateral but not apical exposure in a similar kidney-on-a-chip, an indirect indication of cationic uptake activity that validates the physiological polarisation of the kidney-MPS model (Nieskens et al., 2020). The enhanced metformin secretion observed in our kidney-MPS model is postulated to be the result of OCT2 localization to the basolateral membrane and the increased activity of the OCT2 -MATE1 -MATE2-K cation transport pathway. The upregulation of both MATE1 and MATE2-K at the gene level observed in the kidney-MPS further validates the impact of fluidic conditions on the expression of renal drug transporters but does not represent a functional gain in itself. Previously, transport of the anionic substrate para-amino hippuric acid (PAH) was demonstrated in a similar kidney-MPS model using freshly isolated renal proximal tubule cells (Weber et al., 2016). Moreover, the excretion of morphine and its active metabolite morphine-6-glucuronide was also recently reported (Imaoka et al., 2021). With the determination of transcellular metformin and cidofovir flux, we reiterate the presence of both anion and cation secretion activity in this model. This shows that endogenous function of this key renal uptake transporter can be reinstated in vitro using environmental conditions, without depending on overexpression systems. Freshly isolated renal tubule cells can retain OAT1 and OAT3 expression and activity (Brown et al., 2008;Van der Hauwaert et al., 2014;Lash et al., 2006). However, cells dedifferentiate when expanded and cryopreserved, losing or substantially reducing the expression of key tubular features (Sakolish et al., 2018). This transformation is evident in the cryopreserved RPTEC used (Figure 3-A). HNF4A activity is involved in the expression of Solute Carrier (SLC) transporters during renal development (Martovetsky et al., 2013(Martovetsky et al., , 2016. Its upregulation in the kidney-MPS (Figure 3-H) may suggest that this nuclear transcription factor plays a role in the recovery of OAT1 expression indicating that cryopreserved RPTEC, under dynamic 3D culture, can re-differentiate, following similar pathways to those involved in kidney maturation (Martovetsky et al., 2013).

System
Applications of modelling-based methodologies to translate MPS in vitro data to vivo renal parameters predictive of either renal clearance or drug induced toxicity were recently demonstrated (Imaoka et al., 2021;Maass et al., 2019). These approaches maximize the impact of using MPS models in translational research and follow initial studies that explored the use of multi-organ MPS platforms encompassing liver and gut models to predict drug metabolism and absorption (Maass et al., 2017). More complex designs incorporating interconnected models were used to predict the pharmacokinetic profiles of different drugs (Maass et al., 2017;Edington et al., 2018). While these studies factor in different parameters towards a 'physiome-on-a-chip' (Edington et al., 2018) our approach is directed at drug transport and renal secretion as means to predict renal clearance CLr. The kidneys are responsible for the elimination of a considerable number of drugs and a fifth of all new compounds are estimated to fail in late development due to renal attrition (Brater, 2002;Redfern et al., 2010). Quantifying CLr in early drug development is important in such aspects as evaluating drug-drug interactions and considering dose adjustments in patient cohorts, pivotal to the success of new compounds (Feng et al., 2010;Brater, 2002). The endpoints required to model CLr include the glomerular filtration rate (GFR), active or passive permeability and fraction excreted in urine. Given the multifactorial complexity of reabsorption and secretion in the kidneys, the processes behind these endpoints may differ substantially. Preceding mechanistic models based on 2D in vitro data often only consider either active secretion or passive drug reabsorption, in combination with GFR, to predict CLr (Scotcher et al., 2016;Mathialagan et al., 2017). These static models rely on different in vitro systems and distinct modelling strategies to predict drug clearance, estimating either active transport or passive permeability separately. A limitation to static approaches is the lack of the physiological compartmentalization that would allow for drug and pH gradients across blood, tubular cells and tubular lumen to be appropriately considered. Mechanistic kidney models have the potential to incorporate human-like physiology whereby impact of e.g. water reabsorption from tubular fluid on passive drug reabsorption can be considered in the in vitro-in vivo extrapolation (Scotcher et al., 2016;Neuhoff et al., 2013). Moreover, previous approaches do not reasonably predict the clearance of anionic substrates, given the lack of OAT activity in the cell models used (Kunze et al., 2014). This kidney-MPS recapitulates drug transport at the renal proximal tubule level, with both anionic and cationic secretory routes expressed and active. The semi-mechanistic model used enables the derivation of both active and passive drug permeability parameters and captures the temporal nature of the kidney-MPS perfusion. Human CLr of cationic and anionic drugs could therefore be predicted using one strategy that incorporates both active secretion and reabsorption derived from our kidney-MPS model. In our study, transporter-mediated cidofovir and metformin flux could not be demonstrated in 2D, likely to the lack of OAT1 expression and limited OCT2 membrane expression, in combination with the low permeability of both molecules. Therefore, our CLr predictions were modelled solely on MPS data.
Despite the advantages of employing kidney-MPS in CLr predictions challenges persist. Drug concentration in the loading channel outlet takes about 4 hours to reach a similar level as the concentration in the input solution/buffer, which is the combined result of transit through the chip micro-perfusion circuits and bubble traps, perfusion rate and dispersion into the matrix. Within this time frame the concentration in the receiver outletrenal tubuleis about 10-20% of the loading concentration in a cell free-MPS (Figure 6-B1, C1). This phenomenon is the result of both drug diffusion and convection across the matrix, meaning that a substantial amount of the drug is dispersed in the collagen matrix in the MPS and that the renal tubule is exposed to a lower drug concentration, over a significant period of time. Therefore, transepithelial transport may not be a rate-limiting step for compounds actively transported with high efficiency, where drug permeability (µL/min) is higher than the drug flux (µL/min) through the matrix (PS > Qtr). In our approach, perfusion data from chips without cells were important to understand the drug distribution profile in the MPS and determine this characteristic of the system, which may be behind the apparent underprediction of metformin CLr. Drug concentrations at the loading outlets of the MPS-chip are overlapping (with and without inhibition), showing that the cell monolayer forms a rate-limiting barrier to drug translocation. With increasing efficiency of transporter-mediated transport drug supply will, eventually, become a limiting factor, and transporter kinetic parameters will not be identifiable. This is a practical limitation applicable to any dynamic system of this nature.
Kidney-MPS recreates only the proximal tubule secretion and reabsorption, the other central CLr parameter for IVVIEglomerular filtrationdepends on previously reported values. In order to maintain the simplicity of our approach, only CLr was attempted. A PBPK modelling strategy would require extensive use of non-MPS derived parameters. This study ALTEX, accepted manuscript published November 3, 2022 doi:10.14573/altex.2204011 13 focused on well described reference drugs, essentially with no protein binding (Scheen, 1996;Cundy et al., 1995) and expected to have minimal binding to the chip's PDMS body (van Meer et al., 2017), an event that would hamper precise transepithelial flux determinations and conflict with the modelling strategy employed. Further testing should be extended to include compounds secreted by different transport pathways (e.g. OATP4C1-P-gp) and with different physicochemical properties (e.g. high lipophilicity, protein binding). The addition of an adjacent endothelial tubule, can be used to study permeability across the renal vasculature and the proximal tubules, as recently demonstrated (Chapron et al., 2020). The MPS chip model used incorporates a single renal tubule, which can be a limiting factor when it comes to the number of experimental replicates and data variability. In our study, the passive permeability data retrieved showed high variability, however, this fact has a lower impact on the overall prediction accuracies since the active secretion component is the main contributor to drug clearance of the tested drugs. In future pre-clinical studies data variability could be mitigated by expanding the experimental replicates, employing cells derived from multiple donors. Prospective kidney-MPS designs could incorporate multiple renal tubules per chip, thus increasing the throughput, and minimizing these issues. Further, scaling strategies employing proteomics-based relative expression factors (REFs), could improve IVIVE predictions since they can incorporate the abundance of drug transporters at the protein level, relative to surface area or cell number. However, the use of this approach is still limited by the biological specimen sizes in this MPS. A multisegment chip recreating also glomerular and distal tubule components, could capture filtration and reabsorption and enable all renal clearance endpoints to be derived from the same model. Apart from the proximal tubules, advances have been made towards developing a glomerulus-on-a-chip (Petrosyan et al., 2019;Doi et al., 2022); nonetheless, a multi-segment kidney-MPS is still in the future. With segments recreated on different MPS platforms with no interconnectivity, modelling strategies, such as ours, offer a powerful tool to integrate PK parameters derived from different models and different sources.
A potential strategy to increase the experimental throughput of this kidney-MPS is to use a version of the semimechanistic micro-perfusion model, where the parameters are derived from the MPS data at steady-state (Tab. S10 1 ) when the drug concentration in the loading channel outlet reaches the levels of the input. This approach estimates permeability within 25% of the perfusion-derived parameters and simplifies the experimental procedure, requiring far less samples to be collected and analysed.
With this study we demonstrate that renal total and secretory clearance can be predicted within an acceptable 5-fold range, an industry standard commonly employed in pre-clinical development (Davies et al., 2020), using a model-based approach applied to a continuously perfused kidney-MPS that recapitulates endogenous drug transport activity. As MPS applications progress and a substantial body of evidence validates their usage, kidney-MPS can bridge the gaps in renal clearance predictions derived from conventional in vitro systems and play a major role in the reduction and potential future replacement of animals in PK studies.