Design of Experiment Based Optimization of an in Vitro Direct Contact Triculture Blood Brain Barrier Model for Permeability Screening

Background: The in vivo restrictive properties of the blood brain barrier (BBB) largely arise 9 from astrocyte and pericyte synergistic cell signaling interactions that underlie the brain 10 microvessel endothelial cells (BMEC). In vivo relevant direct contact between astrocytes, 11 pericytes, and BMECS, to our knowledge, has not been established in conventional Transwell ® 12 based in vitro screening models of the BBB. We hypothesize that a design of experiments (DOE) 13 optimized direct contact layered triculture model will offer more in vivo relevance for screening 14 in comparison to indirect models. 15 Methods: Plating conditions including the seeding density of all three cell types, matrix protein, 16 and culture time were assessed in DOE P . DOE P was followed by DOE M1 and DOE M2 to assess 17 the influence of medium additives on barrier properties. The permeability of 4 kD dextran, a 18 paracellular marker, was the measured response to arrive at the optimal plating conditions. The 19 optimized model was further assessed for p-glycoprotein function using a substrate and inhibitor 20 along with a set of BBB paracellular and transcellular markers at varying permeation rates. 21 Results: DOE P revealed that length of culture post endothelial cell plating correlated highest 22 with paracellular tightness. In addition, seeding density of the endothelial cell layer influenced 23 paracellular tightness at earlier times of culture, and its impact decreased as culture is extended. 24 Medium additives had varying effects on barrier properties as seen from DOE M1 and DOE M2 . At 25 optimal conditions, the model revealed P-gp function along with the ability to differentiate 26 between BBB positive and negative permeants. 27 Conclusions: We have demonstrated that the implementation of DOE based optimization for 28 biologically based systems is an expedited method to establish multi-component in vitro cell models. The direct contact BBB triculture model reveals that the physiologically relevant 30 layering of the three cell types is a practical method of culture to establish a screening model 31 compared to indirect plating methods that incorporate physical barriers between cell types. 32 Additionally, the ability of the model to differentiate between BBB positive and negative 33 permeants suggests that this model may be an enhanced screening tool for potential neuroactive 34 compounds. 35

sources. (39,40) 116 Given the interaction of multiple cell types that maintain the BBB phenotype in the NVU, 117 many in vitro models include astrocytes and pericytes in conjunction with BMECs.(30,41-45) 118 Typically, these models involve seeding the endothelium on the apical surface of the filter and    Medium supplemented with 5% FBS, astrocyte growth supplement, and penicillin/streptomycin. Similarly, medium optimization was performed in two analyses using a custom design 182 DOE to determine medium conditions that resulted in the tightest barrier properties. The first 183 analysis (DOEM1) was performed using HEPES, hydrocortisone, dexamethasone, LiCl, calcium, 184 and retinoic acid, using the selected date for permeability analysis at 9 days post endothelial cell 185 plating (Table 2). A second analysis (DOEM2) was performed, based on the results of the first, 186 using hydrocortisone, dexamethasone, LiCl, and retinoic acid at both 5 and 7 days post 187 endothelial cell plating (Table 3).   For the DOEP studies, filters were pre-coated with poly-L-lysine (PLL) by pre-coating 12 200 mm, 0.4 μm pore Transwell ® inserts with 5 µg/cm 2 PLL. Astrocytes were plated at seeding 201 densities of 20,000, 40,000, or 60,000 cells/cm 2 and allowed to grow for 48 hours. After 48 hours 202 of astrocyte growth, astrocyte medium was removed and pericytes were seeded atop the astrocyte 203 lawn at seeding densities of 20,000, 40,000, or 60,000 cells/cm 2 and allowed to grow for 48 204 hours. After 48 hours of pericyte growth, apical medium was replaced with the specified ECM 205 protein solution. Astrocyte-pericyte lawn filters were coated with one of the following ECM 206 proteins at the respective concentrations: Matrigel ® 25 μL/cm 2 (2.5 μg/cm 2 ), Laminin 5 μg/cm 2 , 207 11 or Type I Rat Tail Collagen 5 μg/cm 2 . To coat inserts, Matrigel ® , Laminin, or collagen I aliquots 208 were diluted in HBSS with Ca 2+ and Mg 2+ and 0.5 mL dispensed onto to each respective 12 mm 209 insert and left to incubate with the respective ECM protein for 45 min at 37 °C. After incubation, 210 the ECM solution was removed and HBEC-5i cells were plated at seeding densities of 50,000, 211 80,000, or 110,000 cells/cm 2 and allowed to grow for 5, 7, or 9 days prior to permeability  In DOEM1/2 studies, the culturing methodology described above was used with the 218 modification that the complete HBEC-5i culture medium was supplemented with the DOE   Table 4.

296
Plating Optimization (DOEP) 297 Traditionally, a One-Factor-at-a-Time (OFAT) approach is used to assess the impact of 298 variable changes in cell-based models and processes, where one variable (e.g., cell density) is

312
Based on the data trends, the culture length between the assay day was determined to 313 have the largest impact on paracellular permeability resulting in significantly lower 4 kD dextran 314 permeability at day 9 compared to days 5 and 7. When separating the data by study day and 315 factor there are observable trends in permeability coefficients among the factors including the 316 effects of astrocyte and pericyte cell density. With extended culturing, higher seeding densities 317 of astrocytes and pericytes result in higher permeability of the dextran (Fig. 3). HBEC-5i seeding 318 16 density also shows trends towards lower permeability at higher seeding densities; however, this 319 trend is not as strong at day 9 when the cells have had sufficient time to reach confluence and 320 have a longer time to differentiate.

321
Using JMP 13.2 software, a prediction profiler was generated based on the obtained Peff 322 values for the given conditions. By maximizing Desirability to achieve the lowest possible 323 permeability, the optimal conditions were determined to be 20,000 cells/cm 2 for both astrocytes The optimal medium condition was determined to be 15 mM HEPES, 1 mM calcium, and 10 μM 344 retinoic acid, but the influence of these factors on barrier tightness was not significant (Fig. 5).

345
The full data set of DOEM1, including medium conditions and Peff, is tabled in Additional File 2.

499
Hydrocortisone has been shown to increase barrier tightness through the upregulation of tight      Cross sectional depiction of the Blood-Brain Barrier within the neurovascular unit (NVU) with the endothelium (BMECs) lining the capillary, pericytes embedded within the basal lamina, astrocytes having nearly full coverage of the BMECs and surrounding pericytes, and neurons in close contact with the astrocytes (left). The direct contact triculture model on the apical surface of a Transwell® lter support mimicking the in vivo NVU. Astrocytes are seeded rst on the lter, followed by pericytes, then BMECs to generate a fully apical, direct contact triculture model (right).

Figure 2
Papp and Peff of 4 kD FITC-Dextran across different direct contact triculture conditions of DOEP. All DOEP selected conditions were performed as n=1 for a rapid evaluation of the different parameter combinations. Condition 13 was compromised and permeability was not performed, data point was excluded from statistical analysis.

Figure 3
Peff of 4 kD FITC-Dextran for DOEP separated by factor and further by day of study showing relative trends of factor levels at increasing length of culture. All conditions are represented by single data points across the graph, n=1.    and [14C]-PEG-4000 across the optimized direct contact triculture. Error bars represent one standard deviation (n=3). (C) Apparent permeability of P-gp substrate rhodamine 123 (R123) in the presence and absence of P-gp inhibitor elacridar across the optimized direct contact triculture. Assays were run in triplicate and subjected to Student's t-test. Signi cant difference is indicated by *, p < 0.05 and **, p < 0.01. Error bars represent one standard deviation (n=3).

Figure 8
Apparent permeability of BBB positive (L-histidine, carbamazepine, and rhodamine 123 in the presence of P-gp inhibitor elacridar) and negative (colchicine, rhodamine 123, digoxin, clozapine, and prazosin) permeants across the optimized direct contact triculture. Assays were performed in triplicate. Error bars represent one standard deviation (n=3).

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.