The following lot of commercially available primary human hepatocytes was used in this study: hu8391 (ThermoFisher Scientific). Human liver sinusoidal endothelial cells (LSECs) were purchased from LifeNet Health (NPC-AD-LEC-P1). The THP-1 human monoblast cell line (ATCC, TIB-202) was used to generate Kupffer-like cells and was pretreated 48 h before seeding with 200 ng/mL phorbol 12-myristate 13-acetate (PMA; MilliporeSIgma, 524400) to induce macrophage-like differentiation and inhibit proliferation. Human hepatic stellate cells (LX-2) were obtained from MilliporeSIgma (SCC064). Cell culture conditions were as follows: LSECs: cultured in endothelial cell basal medium-2 (EBM-2; Lonza, CC-3162). THP-1 cells: maintained in suspension in RPMI 1640 (Cytiva, SH30096.FS) with 10% fetal bovine serum (FBS; Corning, MT35010CV), 100 μg/mL penicillin-streptomycin (Cytiva, SV30010), and 2 mM L-glutamine (Cytiva, SH30034.01). LX-2 cells: cultured in DMEM (ThermoFisher Scientific, 11965118) supplemented with 2% FBS, 100 U/mL penicillin, and 100 μg/mL streptomycin [20, 36–38, 40, 46].

ht-LAMPS were assembled and maintained as previously described [20, 36–38, 40, 46] except using the Fluidic 557 Reaction Chamber Chip (Microfluidic ChipShop #557). The ht-LAMPS were constructed using four liver cell types at the following cell densities: primary cryopreserved human hepatocytes (ThermoFisher lot hu8391 2.75 × 10⁶ cells/mL), primary liver sinusoidal endothelial cells (LifeNet Health; LSECs 1.5 × 10⁶ cells/mL), and THP-1 (ATCC; 0.4 × 10⁶ cells/mL) and LX-2 (MilliporeSigma; 0.2 × 10⁶ cells/mL). The percentages of hepatocytes (56%), THP-1 (18%), LSEC (22%), and LX-2 (4%) cells are consistent with the scaling used in our previously published models. The interior of the devices were coated for 2 h at RT or overnight at 4 °C with 100 μg/mL bovine fibronectin (MilliporeSigma, F1141) and 150 μg/mL rat tail collagen (Corning, 354249) in PBS prior to cell seeding. For all steps involving injection of media and/or cell suspensions into LAMPS devices, 90–100 μL per device was used to ensure complete filling of fluidic pathways and chamber. The devices were then overlayed with 1.5 mg/mL rat tail collagen I (Corning) and maintained with the perfusion of different conditions for 8 days at a flow rate of 5 μL/h to recapitulate zone 3 oxygen tension [38, 40].

MASLD disease phenotypes in LAMPS were driven using defined media formulations to mimic disease progression from normal fasting (NF) to early metabolic syndrome (EMS) and then late metabolic syndrome (LMS) for 8 days [20, 38]. These media formulations were developed using glucose-free Williams E base medium (ThermoFisher, ME18082L1) supplemented with physiologically relevant levels of glucose (MilliporeSIgma, G8644), insulin (ThermoFisher, 12585014), and glucagon (MilliporeSIgma and molecular drivers of fibrosis including TGF-β1 (ThermoFisher, PHG9214) and lipopolysaccharide (MilliporeSIgma, L2654). The specific concentrations of all media components used for the NF, EMS, and LMS formulations are provided in Supplementary Table 1.

Devices were stained using established protocols [8, 20, 37]. Cells were fixed with 4% paraformaldehyde (ThermoFisher Scientific, AA433689M) in PBS for 30 min, followed by two washes with PBS. For lipid staining, LipidTOX Deep Red Neutral Lipid Stain (1:500; Invitrogen, H34477) was perfused into devices and incubated overnight at 4 °C. For immunofluorescence, chambers were labeled with either rabbit polyclonal anti-cytokeratin 18 (CK-18) antibody (ThermoFisher Scientific, PA5-14263; 1:200) and mouse monoclonal anti-α-smooth muscle actin (αSMA) antibody (Sigma-Aldrich, A2547; 1:100), or with rabbit polyclonal anti-collagen 1A1 (COL1A1) antibody (Proteintech, 14695; 1:100), and incubated overnight at 4 °C. The following day, devices were washed twice with PBS and then incubated for 2 h at room temperature with Alexa Fluor goat anti-rabbit 488 (1:250; Invitrogen, A-11029) and Alexa Fluor goat anti-mouse 568 (1:250; Invitrogen, A-11004) secondary antibodies, along with Hoechst (5 μg/mL; Invitrogen, H1399). Devices were washed three times with PBS prior to imaging.

Mitochondrial membrane potential was assessed in live LAMPS using tetramethylrhodamine ethyl ester (TMRE; ThermoFisher Scientific, Cat. No. T669). Devices were equilibrated in pre-warmed culture medium containing 100 nM TMRE for 15 min at 37 °C. Following incubation, channels were gently rinsed with fresh medium to remove excess dye and immediately imaged under live-cell conditions using confocal fluorescence microscopy (ex/em = 549/575 nm). For endpoint live cell imaging, cultures were stained with Calcein AM (ThermoFisher C1430) and Hoechst nuclear dye (ThermoFisher 62249). A 1 mg/mL Calcein AM stock solution was prepared in DMSO and diluted 1:1000 in normal fasting medium to yield a final concentration of approximately 1 μM, together with Hoechst (1:2000; ∼9 µM final). Following chip disconnection, 50 µL of staining solution was applied to both the inlet and outlet, and chips were incubated for 45 min at 37 °C. Imaging was performed at ×10 magnification using the FITC (Calcein AM) and DAPI (Hoechst) channels.

Imaging was conducted on the Phenix High-Content Imaging system (Revvity) using a 20×/0.4 NA air objective. Z-stacks covering 35 μm depth (5 μm steps) were acquired across 7 × 15 adjacent fields (total area: 44 mm²; 105 total fields per chamber). All images were collected using consistent exposure and laser power settings to maintain signal intensity within 50–90% of the dynamic range. Image analysis was performed using Harmony software (Revvity, v5.1).

Live cell quantification was performed on Day 8 using an automated image analysis workflow implemented in Harmony (PerkinElmer). The analysis utilized the DAPI (nuclear) (ex: 405/em: 435–480) and calcein (live cell) (ex: 488/em: 500–550) fluorescence channels to identify total and live cell populations. Nuclei were segmented using the “Find Nuclei” block set to method M in the Harmony 5.1 software, and calcein-positive regions were detected using optimized intensity thresholding within the “Find Image Region” block. Then nuclei not overlapping with calcein-positive regions were excluded using the “Select Population” block. The percentage of live cells was determined as: %  Calcein ‐ positive cells = C a l c e i n − p o s i t i v e   n u c l e i T o t a l   H o e c h s t − p o s i t i v e   n u c l e i × 100

Mitochondrial membrane potential was quantified on Day 8 from TMRE fluorescence images using automated segmentation and intensity-based analysis. Nuclei were first identified using the DAPI channel using the “Find Nuclei” block in the Harmony 5.1 software, and the corresponding perinuclear cytoplasmic regions were defined for each cell using the “Find Surrounding Region” block on method A. TMRE-positive (ex: 561/em: 570–630) signal intensity and area were measured within these regions, and integrated density was calculated as the product of TMRE-positive area and mean fluorescence intensity to represent mitochondrial activity per cell.

Steatosis was quantified on Day 8 from LipidTOX-stained images using automated segmentation and object detection workflows (Harmony). Nuclei were identified from the DAPI channel, and lipid-positive regions were segmented based on LipidTox fluorescence (ex: 640/em: 650–760). Individual lipid droplets between 5 and 30 μm in diameter were detected and counted. Lipid accumulation was report both in terms of the average number of lipid droplets per cell (total droplets divided by total nuclei) and the mean lipid-positive area per imaging field.

Albumin (ALB), blood urea nitrogen (BUN), lactate dehydrogenase (LDH), and pro-collagen Iα1 (COL 1A1) levels were assessed as previously described [20, 36, 37, 46]. Efflux from the LAMPS was collected on days 2, 4, 6, and 8 for analysis. ALB levels were quantified using an enzyme-linked immunosorbent assay (ELISA) on 1:100 efflux dilutions with commercial antibodies (Bethyl Laboratories, A80-129A and A80-129P) and an ELISA accessory kit (Bethyl Laboratories, E101), with a human albumin standard prepared in-house (MilliporeSigma, 126658). Secreted COL 1A1 was quantified using the human pro-collagen Iα1 ELISA kit (R&D Systems, cat. no. DY6220-05) in a 1:50 efflux dilution. BUN levels were determined using the Stanbio BUN Liquid Reagent for Diagnostic Set (Stanbio Laboratory, cat. no. SB-0580–250), while LDH levels were assessed using the CytoTox 96 Non-Radioactive Cytotoxicity Assay (Promega, cat. no. G1780). The BUN and LDH assays were adapted to a 384-well microplate format without efflux dilution.

The levels of IL-6, IL-8, and MCP-1 were determined in efflux collected on Day 8 using a custom version of the Human XL Cytokine Performance Panel (R&D systems). Assays were completed according to the manufacturer’s instructions at The University of Pittsburgh Cancer Proteomics Facility Luminex® Core Laboratory on the xMAP platform.

Oxygen tension was quantified in the ht-LAMPS as previously described [10, 16, 40] and validated via ratio imaging using oxygen-sensitive and oxygen-insensitive fluorescent beads. Oxygen supply in the sealed microfluidic LAMPS device was modeled to determine the flow rate required to achieve target oxygen tension. Using a 3D COMSOL Multiphysics model (v5.2a) [10, 16, 40], flow was described by Brinkman equations through the ECM and cell layer, and oxygen transport/consumption was modeled using the “transport of diluted species” module with a maximum primary hepatocyte OCR of 450 pmol s⁻¹ 10⁶ cells⁻¹. The model predicted that 5 µL h⁻¹ would maintain 6–8% oxygen tension, which was confirmed experimentally with oxygen-sensitive beads (Bangs Laboratories). PdTFPP-loaded oxygen-sensitive beads (ex/em 405/706 nm) and polystyrene oxygen-insensitive beads (Bangs Laboratories) (ex/em 405/455 nm) were obtained from Bangs Laboratories and prepared as previously described [10, 16, 40]. Validation was performed in simplified ht-LAMPS containing only hepatocytes and the bead mixture. This configuration was used to isolate the dominant oxygen-consuming hepatocyte cell population [48, 49] and enable controlled calibration of the bead-based oxygen measurements. At the seeding ratios used here, hepatocytes are the primary drivers of oxygen consumption in the system, and prior experimental and computational studies support hepatocyte-driven oxygen gradients under these conditions. Chips were pre-treated with ECM overnight, seeded with hepatocytes, and incubated for 4 h. The bead mixture were added in LECM and allowed to settle overnight. Chips were overlaid with collagen, polymerized after 4 h, and perfused at 5 μL/h in NFM (18% dissolved oxygen) for 7 days. Imaging was performed using the IN-Cell Analyzer 6000. Calibration was done using the same chips: Max O₂ (18%): Achieved by increasing flow to 50 μL/h. Min O₂ (0%): Achieved by adding 200 μg/mL glucose oxidase (MilliporeSigma) to the media. Final oxygen tension was determined as ratio between the oxygen sensitive and insensitive beads [40].

For comparisons between two groups, the student’s t-test with Welch’s Correction was used, while multigroup comparisons used a One-Way ANOVA with Tukey’s test for multiple comparisons. For all statistical analyses, a p-value of 0.05 was used as criteria for statistical significance. Reproducibility analysis was assessed using the Eve Analytics™ platform following the Pittsburgh Reproducibility Protocol (PReP) [43]. Intraclass correlation coefficient (ICC) was used to assess both intra- and inter-study reproducibility over time for the efflux metrics. Model performance reproducibility was classified as: Excellent: ICC ≥0.8 Acceptable: 0.2 ≤ ICC <0.8 Poor: ICC <0.2. The coefficient of variation (CV) was used to assess the reproducibility of the single time point measurements model performance reproducibility was classified as: Excellent: CV ≤ 5%; Acceptable: 5% ≤ CV < 20%; Poor: CV ≥ 20%.

To increase experimental throughput while preserving the biological complexity of the original LAMPS platform, the single-chamber device (Figure 1A) was adapted to a commercially available seven-chamber microfluidic format (Figure 1B). The resulting higher-throughput LAMPS (ht-LAMPS) maintains the same layered, multicellular architecture as the original LAMPS, as confirmed by 3D imaging of cell-type distribution within the device using CellTracker labeling (Supplementary Figure 1), and supports continuous perfusion under flow conditions corresponding to Zone 1 and Zone 3 oxygen tensions. Each chamber functions as an independent model, enabling parallel experimental conditions within a single device. This configuration increases experimental capacity while using the same cell composition, media formulations, and perfusion parameters established for the standard LAMPS platform.

A key functional feature of the LAMPS platform is the ability to reproduce hepatic zonation through controlled oxygen tension through defined media flow conditions [19, 31, 46]. Computer Solutions (COMSOL) modeling was used to estimate the perfusion rates required to achieve zone-specific oxygen levels in the ht-LAMPS platform (Figure 2A) [19, 31, 46]. Liver zonation was induced in ht-LAMPS by modulating oxygen tension through controlled perfusion rates. COMSOL simulations predicted that a flow rate of 5 µL/h produces zone 3–like oxygen tension (∼6% O₂), whereas 15 µL/h yields zone 1–like oxygen tension (∼15% O₂) (Figure 2A).

These predictions were validated experimentally using oxygen-sensitive beads in hepatocyte-containing chips, which serve as a simplified model for the full system given the dominant contribution of hepatocytes to oxygen consumption [16, 40, 48, 49], with oxygen levels quantified by ratiometric imaging based on the ratio of oxygen-sensitive to oxygen-insensitive fluorescence intensity (Figure 2B; p < 0.0001, n = 21chambers) [19, 31, 46]. In addition, ht-LAMPS model viability under both oxygen conditions was comparable after 8 days of culture under flow, as assessed by Calcein AM staining, with no significant difference in the percentage of Calcein-positive cells (Figures 2C,D; n.s., n = 21 chambers) nor in lactate dehydrogenase (LDH) secretion, a marker of cytotoxicity (Figure 2E; n.s., n = 21chambers). Together, these data confirm that stable, zone-specific oxygen conditions can be established in ht-LAMPS without adverse effects on model viability or cytotoxicity.

Oxygen gradients across the liver lobule drive zonation-dependent differences in hepatocyte function and protein secretion [16, 31, 40, 50]. To evaluate whether the ht-LAMPS recapitulates zone-specific hepatocyte functions, effluent media were collected and analyzed for albumin and urea nitrogen as indicators of hepatocyte secretory and metabolic activity. Consistent with clinical observations [51], albumin secretion was significantly higher in Zone 1 than in Zone 3 throughout the experimental time course (Days 2–8) (Figure 3A; p = 0.003, n = 21chambers). Similarly, urea nitrogen secretion was significantly greater in Zone 1 ht-LAMPS compared with Zone 3 over the same period (Figure 3B; p = 0.019, n = 21 chambers), consistent with zonation-dependent secretion patterns previously reported for in vivo human liver and the LAMPS platform [20, 38, 40, 52].

To further assess zonation-dependent phenotypes, mitochondrial activity and steatosis, which are known to differ by liver zone [24, 46, 53], were quantified in ht-LAMPS maintained at Zone 1 or Zone 3 flow rates for 8 days. Tetramethylrhodamine ethyl ester (TMRE) staining intensity was significantly higher in Zone 1 ht-LAMPS compared with Zone 3 on Day 8 (Figures 4A,B; p = 0.0039, n = 21 chambers), consistent with increased mitochondrial activity in Zone 1 [40]. In contrast, lipid accumulation, assessed using the neutral lipid dye LipidTOX, was significantly greater in Zone 3 ht-LAMPS than in Zone 1 on Day 8 (Figures 4C,D; p = 0.0091, n = 21 chambers), consistent with known zone-specific patterns of hepatic steatosis and with our prior LAMPS-based studies [20, 40, 54]. Together, these results demonstrate that ht-LAMPS reproduces physiologically relevant, zone-specific functional phenotypes consistent with those previously observed in the LAMPS platform [20, 40].

Reproducibility was assessed in the ht-LAMPS for the experimental metrics used to generate Figures 2–4 using the Pittsburgh Reproducibility Protocol (PReP) [43] in combination with the EveAnalytics™ platform [41, 42]. Multi–time point metrics (albumin, urea nitrogen, and lactate dehydrogenase as well as single–time point metrics collected on Day 8 (Calcein AM, TMRE, and LipidTOX) were analyzed to quantify both intra- and inter-study reproducibility (Table 1). For multi–time point metrics, intraclass correlation coefficients (ICC) indicated strong intra- and inter-study reproducibility for albumin and urea nitrogen secretion. In contrast, LDH exhibited poor inter-study reproducibility overall, although intra-study reproducibility was generally acceptable except for Study 2. For single–time point metrics (Day 8), coefficients of variation (CV) indicated high inter-study reproducibility for Calcein AM, TMRE, and LipidTOX measurements. Calcein AM and LipidTOX also showed high intra-study reproducibility, while TMRE exhibited greater variability within individual studies. Together, these results indicate that ht-LAMPS exhibits reproducible performance for several key functional and phenotypic assay metrics, while also revealing greater variability in selected readouts, including LDH and TMRE. Notably, TMRE measurements showed higher intra-study variability but consistent directional trends across studies, whereas LDH exhibited variability across both intra- and inter-study comparisons. These findings highlight both the strengths of the platform and areas where additional assay optimization may be beneficial.

To directly compare the performance of the ht-LAMPS with the previously published single-chamber LAMPS platform [20, 40, 42, 47], zonation-dependent functional metrics measured at Day 8 were compared across platforms using ht-LAMPS data generated in this study and single-chamber LAMPS datasets from prior studies (Supplementary Table 2; Supplementary Figure 2). Supplementary Figure 2 shows these comparisons, allowing for the visualization of the zonation-dependent trends wthin each platform. Ratios of Zone 3 to Zone 1 responses for albumin, urea nitrogen, mitochondrial activity (TMRE), and steatosis (LipidTOX) showed consistent directional trends between the two platforms. Across these metrics, ht-LAMPS recapitulated key zone-specific patterns previously observed in the single-chamber LAMPS, including higher metabolic activity in Zone 1 and increased steatosis in Zone 3. While these comparisons are not derived from fully matched head-to-head experiments, they indicate that scaling the LAMPS setup to a multi-chamber format recapitulates key zonation-dependent functional relationships for the metrics evaluated in this study.

We have previously developed and published a series of media formulations, normal fasting media (NF), early metabolic syndrome media (EMS), and late metabolic syndrome media (LMS), to experimentally model lifestyle-driven progression of MASLD (Supplementary Figure 1) [20, 38]. These formulations were designed based on clinical blood chemistries and have been used to characterize multiple MASLD-associated phenotypes, including steatosis, secretion of pro-inflammatory cytokines, and pro-collagen 1α1 (COL 1A1) in the LAMPS platform.

To determine whether the ht-LAMPS recapitulates these MASLD-associated phenotypes, models were cultured in NF, EMS, and LMS media (Supplementary Figure 1) and evaluated on Day 8. Because MASLD-associated injury and lipid accumulation are most pronounced in the pericentral (Zone 3) region of the liver [53, 55], ht-LAMPS were maintained under Zone 3–like flow conditions, consistent with the experimental design used in our previous MASLD studies with the LAMPS platform [20, 36, 38]. Both EMS and LMS media induced significantly greater lipid accumulation compared with NF media (Figure 5A; p < 0.0001, n = 7 chambers). In parallel, secretion of inflammatory cytokines increased under disease-inducing conditions, with IL-6 showing a significant elevation in the LMS condition relative to NF (Figure 5B; p = 0.0001, n = 7 chambers). CCL2 and IL-8 also exhibited increasing trends in the LMS condition compared with NF (Figure 5B). These cytokines (IL-6, IL-8, and CCL2) were selected as representative inflammatory mediators based on their established roles in MASLD progression and hepatic inflammatory responses [56, 57], and their use in prior LAMPS studies [20, 36, 38]. In addition, secretion of the pro-fibrotic marker COL 1A1 increased in a stepwise manner from EMS to LMS (Figure 5D; p < 0.0001, n = 7 chambers). Complementing these efflux-based measurements, immunofluorescence imaging of ht-LAMPS (Figure 5C) demonstrated increased expression of α-smooth muscle actin (αSMA) and COL1A1 under EMS and LMS conditions relative to NF. These data provide cell-associated evidence of increased stellate cell–associated and fibrogenic marker expression under disease-inducing conditions.

To assess whether MASLD-associated phenotypes observed in ht-LAMPS are consistent with those previously reported in the single-chamber LAMPS [20, 36, 38], disease-response metrics measured at Day 8 were compared between platforms under matched media conditions (Supplementary Table 3). Both platforms exhibited similar increases in steatosis and secretion of the pro-fibrotic marker COL 1A1 in response to EMS and LMS media, indicating preserved induction of key metabolic and fibrotic MASLD progression phenotypes in the ht-LAMPS configuration. Cytokine responses were also broadly comparable between platforms under LMS conditions, with both systems showing increased IL-6 secretion and increasing trends in CCL2 and IL-8. While cytokine induction under EMS conditions was attenuated in ht-LAMPS relative to the single-chamber LAMPS, the overall pattern of MASLD-associated metabolic, fibrotic, and inflammatory responses was consistent between platforms.