A vascularized liver microphysiological system captures key features of hepatic insulin resistance and monocyte infiltration

Abstract

In vitro models can recapitulate aspects of human liver diseases, thereby aiding therapeutic development. Dynamic interactions with vascular and immune cells contribute to disease progression in ways that are challenging to capture in the hepatic spheroid models commonly used for assessing facets of metabolism and disease. To address this, we developed a microphysiological system (MPS) featuring multicellular human hepatic spheroids physically integrated with self-organized microvascular networks. We demonstrate this MPS’s utility by modeling an insulin resistance state, where chronic exposure to disease-mimetic conditions yields altered hepatocyte metabolism, dysregulated vascular features, and increased inflammation state. We extend this system to capture disease-relevant changes in immune cell recruitment, showing that monocytes perfused through the vasculature will extravasate toward hepatic spheroids, with insulin-resistant samples exhibiting greater infiltration. Altogether, this vascularized liver MPS captures local hepatocyte-immune-microvascular interactions in an accessible microfluidic platform, enabling the study of clinically relevant immune-tissue interactions in complex metabolic disease.

Introduction

The liver is responsible for a diverse set of vital functions, spanning plasma protein production, nutrient metabolism, immune regulation, and drug and toxin clearance^1,2. This versatility is enabled by both the presence of tissue-specific cell types as well as their complex architectural organization, with hepatocytes acting as the primary arbiter of many of the organ’s key activities. In carbohydrate metabolism, the liver helps to preserve blood sugar homeostasis by maintaining a balance between the storage and secretion of glucose. Additionally, the liver is a central regulator of lipid metabolism, with hepatocytes responsible for breaking down free fatty acids into triglycerides for storage^1,3. Given this organ’s essential involvement in mediating metabolic regulatory functions, dysregulation induces systemic changes that culminate in metabolic disorders such as type 2 diabetes (T2D) and metabolic-associated steatotic liver disease (MASLD)⁴.

A fundamental element of these diseases is insulin resistance, defined as the body’s inability to respond effectively to the peptide hormone insulin, which plays a key role in regulating energy storage⁵. In the healthy liver, insulin directly induces glucose uptake and glycogen synthesis, upregulation of lipogenic gene expression, and downregulation of gluconeogenic gene expression. In contrast, insulin resistance results in impaired insulin clearance, aberrant hepatic glucose uptake and production, and elevated lipid synthesis^6,7. Importantly, insulin resistance is associated with activation of liver-resident macrophages (Kupffer cells) and elevated pro-inflammatory cytokines such as IL-6 and TNF-α; this immune system response is considered part of a cycle via which inflammation can further drive insulin resistance and recruitment of circulating immune cells, which in turn causes chronic and systemic immune dysregulation^8,9.

Much remains to be understood about the mechanisms by which insulin signaling in the liver is altered, as well as the complex interactions between hepatic and systemic insulin resistance leading to disease progression. These knowledge gaps are in part due to a historical reliance on animal models, which often exhibit significant disparities from human phenotypes and establish pathophysiology through dissimilar mechanisms^10,11. As a result, translation of most findings to the human context has seen limited clinical success. To date, resmetirom is the only U.S. FDA-approved therapeutic for adults with metabolic dysfunction-associated steatohepatitis (MASH), an advanced stage of MASLD, exemplifying the ongoing gap in our understanding of metabolic liver disease¹².

In recent years, microphysiological systems (MPS) have emerged as a promising alternative to animal models, presenting new opportunities for advancing our understanding of human biology and disease^13,14. By enabling 3D co-culture of multiple cell types within meso- or microfluidic devices, MPS can maintain primary human liver phenotypes longer and more faithfully than culture in 2D^14,15,16,17. Further, efforts to incorporate tissue-specific non-parenchymal cells (NPCs)—including Kupffer cells, stellate cells, and liver sinusoidal endothelial cells (LSECs)—capture key functional contributions and cellular interactions that are otherwise lost in simpler culture contexts^8,18,19. Recent studies demonstrating improved primary human hepatocyte longevity and function with either synthetic vascular-like channel structures or co-culture with endothelial cells underscore the value of adding vascular elements^20,21,22. In addition, accessible microfluidic models with 3D perfusable vasculature have been employed to study immune cell trafficking for some applications^23,24,25,26, but they have not yet been used to enable direct transport of immune populations to hepatocytes themselves in situ, in part due to challenges with integrating perfused vasculature within hepatic tissue structures^27,28,29,30.

In parallel, liver MPS have been applied to a variety of disease states, including MASLD / MASH. These models employ tailored culture media compositions (e.g., high nutrient concentrations) designed to induce specific phenotypes (e.g., features of hepatocyte steatosis)^{31,32,33,34,35}. While inclusion of NPCs within these disease models has improved their fidelity to disease phenotypes, the known roles of vasculature and immune cell trafficking in disease pathogenesis motivate our building more accessible models that incorporate these elements^36,37,38,39.

Here, we establish a primary human cell-based, vascularized disease model of hepatic insulin resistance. By leveraging and improving upon self-organized in vitro vascular models, we developed a microfluidic platform where vasculature perfuses directly into multicellular hepatic spheroids comprising hepatocytes, Kupffer cells, endothelial cells, and fibroblasts, exhibiting high degrees of interaction and enabling transport directly into spheroid structures. Beyond supporting hepatocyte and vascular function, this MPS also captures features of disease phenotypes. A pathological disease media formulation induces an insulin resistant state in this vascularized liver platform, with altered insulin clearance and inflammation profiles. Finally, we demonstrate that these altered inflammation states functionally impact the relationship between the local environment and systemically circulating innate immune cells. With the addition of peripheral monocytes, we show enhanced monocyte recruitment under insulin resistance conditions, recapitulating immune-tissue interactions in metabolic disease.

Results

Multicellular hepatic spheroids integrate with self-organized vascular networks

We first generated liver spheroids using an alginate microwell system⁴⁰. These spheroids comprised donor-matched primary human hepatocytes and Kupffer cells; human umbilical vein endothelial cells (HUVECs); and normal human lung fibroblasts (NHLFs). The formation of perfusable microvascular networks in fibrin hydrogels via morphogenesis of HUVECs and NHLFs is a well-established protocol in the MPS community. We speculated that also including these two vascular elements in our hepatic spheroids would assist in establishing vascular junctions between the spheroids and their surrounding microvasculature^41,42. To create a suitable microfluidic platform, we adapted a polydimethylsiloxane (PDMS) device design previously used for vascularization studies^43,44,45,46 by bonding devices to 6-well glass bottom plates, facilitating higher throughput culture and improved long-term time-lapse imaging (Supplementary Fig. 1). After these spheroids compacted for 2 days, we harvested them, mixed them with fibrin gel precursor containing additional HUVECs and NHLFs, and introduced this mixture into a microfluidic device. After fibrin polymerization, we maintained the cultures with daily media changes for 7-14 days, using a rocker platform to promote vasculogenesis and longer-term maintenance of perfusable microvessels (Fig. 1A).

Fig. 1: Generation of a vascularized, primary human liver spheroid-based culture platform.

A Schematic of culture platform and seeding strategy, beginning with multicellular hepatic spheroid generation in alginate microwells for 2 days, followed by harvesting and co-encapsulation of spheroids, endothelial cells, and supporting fibroblasts inside of a fibrin hydrogel within polydimethylsiloxane (PDMS) microfluidic devices. Created in BioRender. Tevonian, E. (https://BioRender.com/o38c467). B Self-organized vascular network formation after 6 days in devices. Maximum intensity projection of live image taken at 10x magnification shows GFP-HUVEC networks (green) interacting with hepatocytes labeled with CellTracker Deep Red (magenta). Inset (yellow box) shows a magnified view, with arrows indicating regions where networks are in close proximity with spheroids. Scale bars = 200 µm. This experiment with CellTracker-labeled spheroids was repeated independently seven times with similar results. C Live image taken on Day 4 after device seeding shows RFP-HUVECs (false-colored cyan) migrating out of spheroids to connect with GFP-HUVECs (green), with hepatocytes labeled with CellTracker Deep Red (magenta). Scale bar = 200 µm. This experiment with RFP-HUVECs was repeated independently nine times with similar results. (D) Representative maximum intensity projection image of an immunostained device after 14 days of culture. Arginase-1 (magenta) indicates hepatocytes, CD31 (green) indicates HUVECs, and CD163 (white) indicates Kupffer cells. Scale bars = 200 µm. This staining was repeated for devices from three unique primary liver cell donors with similar results.

To enable co-culture of these primary human cell types, we first defined a media supportive of liver and vascular function, inspired by combining commercial formulations for hepatocyte and endothelial cell media, while maintaining physiological ranges of glucose, insulin, and free fatty acid levels⁴⁷ (Supplementary Table 1). Hepatic spheroid structural maintenance and microvascular network formation and perfusability were both independently assessed in this new media formulation over 1-2 weeks in culture (Supplementary Fig. 2).

Live imaging demonstrated that HUVECs self-organize into microvascular networks spanning the device tissue compartment over the first 3-5 days in culture (Supplementary Fig. 3, and Supplementary Movie 1). The microvascular networks surround the hepatic spheroids, establishing the spheroids as both interacting with and integrated within the microvasculature (Fig. 1B). To test whether these interactions were in part due to the endothelial cells within the hepatic spheroids forming links with those external to the spheroid, we generated hepatic spheroids containing RFP-HUVECs and monitored their interactions with surrounding GFP-HUVEC vascular networks. We observed that endothelial cells that originated within the spheroids not only connected the spheroids to the immediately adjacent GFP-HUVEC microvessels, but also migrated to distant locations throughout the networks after several days in culture (Fig. 1C, and Supplementary Fig. 3). Additionally, we confirmed that after 14 days of device culture, hepatocytes and Kupffer cells remain integrated with the networks (Fig. 1D).

Beyond visualizing these physical interactions in co-culture, we also evaluated critical structural and metabolic functions of the vasculature and hepatocytes. We established that the microvascular networks had open lumens by perfusing fluorescent dextran through each device and observing the dye localization within vessel structures (Fig. 2A). Notably, perfusable segments form through the hepatic spheroids (Fig. 2B, and Supplementary Movie 2) and span the entirety of the central hydrogel compartment (Supplementary Fig. 4). By having vasculature that perfuses directly through hepatic spheroids and thus enables transport into the tissue structures, this system improves upon previous vascularized liver models, which often show vasculature adjacent to liver spheroids rather than in an integrated culture format^27,29.

Fig. 2: Liver MPS supports the formation of perfusable vascular networks and functional primary human hepatocytes.

A Maximum intensity projection images of 10 kDa dextran (blue) perfusion on Day 10 of device culture demonstrates GFP-HUVEC vascular networks (green) have open lumens. Hepatocytes are labeled with CellTracker Deep Red (magenta). Scale bars = 250 µm. B Representative maximum intensity projection image shows that perfusable vessels penetrate through hepatic spheroids on Day 6. Scale bar = 100 µm. For 2A-B, this experiment involving dextran perfusion was repeated independently 12 times with similar results. C Hepatocyte secretion of albumin, normalized per million hepatocytes, shows increased albumin levels over time. Data are presented as the average of n = 11 devices per media condition across N = 3 independent experiments, with error bars representing standard deviation. D, E Hepatocytes maintain urea secretion and Cytochrome P450 3A4 (CYP3A4) metabolism after 13 days in device culture (15 days after initial spheroid formation). Each point represents one device. For urea assay, n = 6 devices per media condition from N = 2 independent experiments; for CYP3A4 assay, n = 9 devices and 12 devices for physiological and IR conditions, respectively, from N = 3 independent experiments. Error bars represent standard deviation from average values.

Hepatocellular function was assessed in both physiological baseline medium and a metabolic overload “insulin resistance”-inducing (IR) media formulation featuring high insulin, glucose, and free fatty acid levels (Supplementary Table 2). Hepatocytes in both media formulations sustained albumin production throughout the course of the experiment, with a drift upward in magnitude over 2 weeks in culture, an improvement over hepatic spheroids alone (Fig. 2C, and Supplementary Fig. 5). Further, hepatocytes exhibit robust CYP3A4 activity and urea production after 2 weeks of culture in MPS devices, with comparable values detected in both media formulations (Fig. 2D, E). We confirmed that this MPS model is generalizable to additional primary liver cell donors (Supplementary Fig. 6, and Supplementary Table 3). With these benchmarks of hepatic function meeting ranges in accordance with in vivo function⁴⁸, we then investigated additional behaviors of the vascularized liver MPS in simulated disease conditions.

Vascularized liver platform recapitulates known features of insulin resistance disease state

We specifically sought to understand whether this vascularized liver MPS can be applied to model features of hepatic insulin resistance that contribute to liver and systemic metabolic disorders. We hypothesized that culturing our devices in IR media would induce phenotypes associated with an insulin resistance disease state. We began by investigating the clearance of insulin by the cells in our system, a process known to be dysregulated in hepatic insulin resistance⁷. While both the physiological and IR media conditions initially showed equal insulin clearance rates, sustained exposure to the IR media resulted in a decline in clearance that is consistent with insulin resistance pathogenesis (Fig. 3A). This trend was recapitulated in additional primary liver cell donors as well (Supplementary Fig. 7).

Fig. 3: Establishment of an insulin resistance state in vascularized liver MPS.

A Insulin clearance diminishes over time in devices cultured in the insulin resistance (“IR”) medium formulation, compared to physiological media. Data are presented as the average of n = 8 devices per media condition across N = 3 experiments, with error bars representing standard deviation and statistical comparison via multiple Mann-Whitney tests with FDR correction. B Gluconeogenesis gene expression (PCK1, G6PC) is significantly increased in IR devices relative to physiological control devices, measured via RT-qPCR on Day 12 in culture from n = 4 devices across N = 2 experiments. Changes in genes associated with fatty acid transport and synthesis (FASN and FABP1) are upregulated to a lesser degree. Error bars represent standard deviation from average values. For each gene, a Mann-Whitney test was used to compare between the two media conditions. C Glucose production is increased in IR devices, measured over 24 h from Day 13 to Day 14 of device culture. Data are presented as the average of n = 8 devices per media condition, with error bars representing standard deviation. A Mann-Whitney test was used to compare between the two media conditions. D Row-normalized heatmap of cytokines, chemokines, and growth factors assayed via Luminex from cell culture supernatant collected on Day 10 of device culture across N = 3 experiments. Each row represents one device; n = 25 and 27 total devices for the physiological and IR conditions, respectively. Analyte labels in black indicate significant differences between physiological and IR conditions (P-adj <0.05). E Partial least squares discriminant analysis (PLS-DA) model built from Luminex dataset separates physiological and IR samples (n = 25 and 27 devices in total from N = 3 independent experiments, with each point representing the scores on latent variable 1 (LV1) and LV2 from one device. Ellipses show the 95% confidence interval for each condition, with the box plot quantifying the significance of separation between conditions on LV1 by Wilcoxon rank-sum test. The first and third quartiles correspond to the bounds of the box and the median is represented by a bolded line. Box plot whiskers extend 1.5 times the interquartile range. F Loadings on LV1 indicate features contributing to the model, with Gas6 associating with physiological conditions, and several chemokines and cytokines associated with the IR condition. For all plots, *p < 0.05, **p < 0.01, ***p < 0.001, and ****p < 0.0001.

Additionally, insulin signaling partially regulates glucose production by hepatocytes. When comparing the IR condition to the physiological condition, gene expression analysis by qPCR revealed significantly elevated levels of genes involved in gluconeogenesis (PCK1 and G6PC) and smaller changes detected in genes involved in fatty acid transport and synthesis (FASN and FABP1) (Fig. 3B). To establish whether this altered gene expression was associated with observable functional changes, we adapted a canonical assay to measure hepatocellular glucose output in a vascularized system, in which media is changed to a reduced glucose production media containing pyruvate and lactate (see Methods) for 24 h^49,50,51. The IR condition resulted in higher net glucose production compared to the physiological condition, indicating impaired metabolic regulation (Fig. 3C). We note that with this 24 h time scale, the absolute magnitude of glucose output observed in device culture is lower than in microwell culture of free hepatic spheroids (Supplementary Fig. 5E)—highlighting our MPS model’s ability to capture the physiologically relevant balance of glucose production by hepatocytes and consumption by vascular cells. To more specifically measure differences in glucose production, we also performed a shorter version of the assay and confirmed that net glucose levels in our vascularized system align closely with those in our spheroid-only benchmark condition after 2 h of culture in glucose production media (Supplementary Fig. 8). Notably, the greater net glucose levels in the IR condition—even after insulin stimulation—are consistent with the observed metabolic gene expression changes (Supplementary Fig. 8).

Due to the association between chronic inflammation and insulin resistance, we also analyzed a panel of chemokines, cytokines, and growth factors associated with inflammation state or vascular dysregulation. This panel was assayed via Luminex on supernatant collected on Day 10 of device culture (Fig. 3D, and Supplementary Fig. 9). To better identify key features associated with the IR state, we performed supervised multivariate analysis to identify features associated with each media condition using partial least squares discriminant analysis (PLS-DA). This model was able to significantly separate samples by physiological and IR conditions (Fig. 3E). Examining the PLS-DA loadings driving this separation, we see chemokines such as CXCL1, CXCL2, and CCL4 correlating with IR conditions (Fig. 3F). This aligns with clinical observations of increased hepatic infiltration by immune cells in IR and MASLD, with this acute inflammation further driving chronic insulin resistance^52,53.

Interestingly, from our Luminex panel we also saw increased secretion of the glycoprotein ICAM-1 associated with the IR devices, suggesting a pro-inflammatory endothelial cell disease state. Some patient cohort studies have suggested that elevated plasma levels of ICAM-1 and other cell adhesion molecules could be a predictor for people at increased risk of developing T2D and associated cardiovascular complications^54,55. To parse the extent to which the vascular networks themselves contribute to the disease phenotypes observed, we also performed experiments separating the two main components of our model, spheroids and microvasculature. Hepatic spheroids alone in physiological and IR conditions confirmed vasculature was not required for disease state development (Supplementary Fig. 5). Meanwhile, microfluidic culture of vascular networks alone demonstrated vascular cells contribute to some metrics, such as insulin clearance (Supplementary Fig. 10A) and inflammation profiles (Supplementary Fig. 10C-E), but features such as gluconeogenic gene expression (Supplementary Fig. 10B) did not change in the absence of liver cells, as expected.

While we initially focused on hepatocyte-specific metrics in our comparisons of the physiological vs. IR media conditions, we also noticed that IR media led to the development of consistently narrower vascular networks in our liver MPS during initial dextran perfusion tests (Fig. 4A). This observation, in combination with our PLS-DA findings of a potentially inflamed endothelial state, motivated us to further quantify macro-scale differences in vascular morphology and permeability (Fig. 4B). Imaging our vascularized liver MPS, followed by quantification of vessel features⁵⁶, showed that by Day 6, there was significantly reduced vessel coverage under IR conditions compared to the physiological baseline (Fig. 4C), as well as narrower vessels (Fig. 4D). In parallel, our permeability analysis showed that the narrowed, IR-associated microvascular networks also display increased permeability, resulting in “leakier” vessel structures (Fig. 4E). Interestingly, these results not only mimic reported effects of high glucose on 2D endothelial cultures, but also reflect microvascular complications often seen in diabetic patients, with increased capillary permeability and narrower vessels being hallmarks of diabetic microangiopathy and peripheral vascular disease^57,58,59,60.

Fig. 4: Vascular networks in liver MPS display hallmarks of peripheral disease.

A Representative images of vascular network morphology after 6 days in culture, with maximum intensity projections of the fluorescent dextran channel shown for visualizing perfusable vessels. Scale bars = 200 µm. This experiment was repeated independently at least seven times with similar results. B Schematic illustrating protocol for imaging vascular networks and performing downstream analyses. Created in BioRender. Tevonian, E. (2025) https://BioRender.com/d75u708. Vascular permeability coefficient is calculated using z-stack images taken of the same region of interest at two defined time points, imported into an ImageJ macro. Vessel morphological parameters are quantified using maximum intensity projection images of the same z-stacks, loaded into the REAVER MATLAB tool. C Insulin resistance (“IR”) culture condition results in vascular networks that cover a significantly smaller area, compared to the physiological baseline. D IR culture condition results in vascular networks that have a significantly smaller mean 2D projected diameter (“mean vessel diameter” in REAVER), compared to the physiological baseline. E Vascular networks in IR devices exhibit increased permeability to 10 kDa dextran, with z-stack images (5 µm step size, 10 slices) taken 9 minutes apart. For 4C-E, 3 regions of interest were quantified per device (see Methods) for n = 9 devices from one experiment, with each dot representing the average measurement for one device and error bars indicating standard deviation. For statistical analyses, Mann-Whitney tests were performed to compare between two groups; *p < 0.05 and **p < 0.01.

Establishing peripheral immune - liver interactions

In addition to peripheral vascular dysregulation, insulin resistance and MASLD are also clinically associated with chronic inflammation and increased immune cell recruitment to the liver^52,53. These features, in combination with our earlier results demonstrating elevated chemokine and cytokine levels in the IR liver MPS, motivated us to investigate immune cell recruitment to hepatic spheroids as a function of induced disease state.

For these experiments, we generated the vascularized liver MPS as previously described in either physiological or IR conditions. After one week, we added CD14+ monocytes freshly isolated from healthy human donors to the microvasculature and maintained this immune co-culture under static conditions for an additional 4 days (Fig. 5A). Time-lapse imaging of CellTracker-labeled monocytes showed these cells immediately travel through the vascular networks (Fig. 5B). While some quickly flowed to the other side of the devices, others attached to the microvascular networks, extravasated through vessel walls, or localized to hepatic spheroids within the first 12 h (Fig. 5C, and Supplementary Movie 3). After 4 days, we saw these monocytes persist within the vascularized liver MPS, with many remaining closely associated with the hepatic spheroids. Furthermore, a subset of these monocytes appeared to begin differentiating based on their reduced circularity and cell spreading, indicating longer term colonization of the hepatic niche (Fig. 5D).

Fig. 5: Vasculature supports immune cell trafficking and reveals disease-associated recruitment.

A Schematic illustrating addition of human PBMC-derived monocytes on Day 8 of device culture, after perfusable vasculature has been established, followed by 4 days of co-culture before endpoint fixation or tissue harvesting. Created in BioRender. Tevonian, E. (https://BioRender.com/e15y242). B Maximum intensity projection of devices 12 h after monocyte addition shows that monocytes (CellTracker Deep Red, false-colored yellow) localize within GFP-HUVEC vascular networks (green) and begin interacting with spheroids (CellTracker Red, magenta). Scale bars = 200 µm. C Representative maximum intensity projection images (insets of 5B) showing monocyte interactions with spheroids. Arrows indicate an example of a monocyte that has extravasated from the vascular networks near a hepatic spheroid (left) and a cluster of monocytes pausing in the networks at a spheroid (right). Scale bars = 200 µm. This experiment was repeated independently three times with similar results, where each experiment used a unique primary monocyte donor. D Imaging after 4 days of monocyte co-culture shows monocytes persist in the devices. Insets (right) show that most monocytes appear to localize with hepatic spheroids (visible in brightfield channel), and some monocytes appear to be differentiating based on their changed morphology. Scale bars = 200 µm (left), 50 µm (insets). This monocyte imaging experiment was repeated independently four times with similar results, where each experiment used a unique primary monocyte donor. E Flow cytometry gating strategy to quantify the frequency of monocytes (CellTracker Deep Red) in each device. F Monocytes were significantly enriched within insulin resistance (IR) devices compared to the physiological condition. Comparison to devices containing vascular networks alone (no hepatic spheroids) also demonstrates the presence of spheroids significantly increases monocyte frequency. n = 6 devices per condition across N = 3 experiments. Statistical analysis performed with Kruskal-Wallis test and post-hoc Dunn’s multiple comparison test. *p < 0.05 and **p < 0.01. G Representative histogram shows elevated CD163 on CellTracker+ monocytes cultured in MPS devices compared to naive monocytes (gray).

To discern if there were significant differences under IR conditions, we performed flow cytometry on these samples to quantify monocyte frequency. To do so, we manually cut out the central compartment from each device, digested the hydrogel to isolate single cells, and ran flow cytometry to measure the fraction of CD14+ cells residing within the tissue compartment of each device (Fig. 5A). Gating for the live, CellTracker-positive population in each device allowed for identifying changes in monocyte frequency (Supplementary Fig. 11, and Fig. 5E). We saw significantly increased monocyte frequency in the IR liver MPS samples. To further understand the role of hepatic spheroids in driving this immune cell recruitment, we also cultured monocytes in devices that contained only vascular networks. Comparisons between our full liver MPS model and this vascular networks-only control revealed the hepatic spheroids were responsible for significantly elevated monocyte recruitment (Fig. 5F). Cell surface marker characterization revealed monocytes cultured in these devices display elevated CD163 compared to the initial monocytes, indicating differentiation towards CD163+ tissue macrophages (Fig. 5G). Previous studies have examined soluble CD163 as a marker of insulin resistance^61,62. Future work characterizing the effects of the recruited CD163+ macrophages in our system, including activation and shedding of CD163, could be an interesting avenue for further investigation. Additionally, comparisons between liver MPS samples with vs. without monocytes show that the addition of monocytes leads to elevated CCL2 and vasculature-associated soluble cues (Supplementary Fig. 12). Ultimately, these altered frequencies and inflammation state between conditions and monocyte differentiation highlight the ability to examine altered recruitment and immune-tissue interactions in this MPS.

Discussion

We have established here an MPS model featuring functional primary human liver cell aggregates co-cultured with a perfusable microvasculature surrounding and permeating the hepatic spheroids. We have further demonstrated that this model can be driven by culture media modifications to recapitulate certain metabolic and immune behaviors observed in the insulin-resistant state: metabolic dysregulation, inflammation, and macroscale vascular dysfunction. This modeling was enabled by employing a physiologically relevant media formulation with insulin, glucose, and free fatty acid concentrations in line with patient values. Notably, by starting with this clinically informed baseline we enhance our ability to interrogate subsequent disease state perturbations that more closely mimic human biology—especially in comparison to traditional culture models, which often contain supraphysiological concentrations of nutrients and insulin that exceed even pathological levels.

We note that this model of insulin resistance most closely represents an earlier phase of disease, without signs of cell death and lipotoxicity for example, which may be observed as insulin resistance progresses to MASLD / MASH. With its modular media components and cell ratios, however, this liver MPS could also be adapted to model more advanced disease, where different parameters could be tuned to mimic and investigate a wider spectrum of MASLD. For example, using primary cells from diseased donors or further altering nutrient, hormone, and/or cytokine levels could recapitulate later metabolic disease stages featuring fibrosis and severe steatosis, though such alterations may also require maintaining cultures for much longer than the two weeks described here. Additionally, such longer-term studies would require continuous, unidirectional flow and replacement of fibrin with a synthetic extracellular matrix that more robustly withstands cell-mediated remodeling and flow conditions—modifications we are currently implementing.

In addition, this MPS could be applied to study liver-specific vascular features that accompany and contribute to metabolic dysregulation⁶³, using LSECs and stellate cells in the spheroids instead of HUVECs and NHLFs, and liver-derived large vessel endothelial cells for the external vasculature. Presently, incorporating these liver-specific cell types into complex in vitro models remains a major challenge across the field; primary human LSECs are notoriously difficult to properly isolate and validate⁶⁴, while standard methods for in vitro culture of stellate cells are known to result in their activation⁶⁵. While a number of groups have reported liver models that include donor LSECs or LSEC-like cells derived from stem cell populations^{64,65,66,67,68,69,70,71,72}, all of these in vitro systems lack an integrated, perfusable vasculature, ultimately limiting their translational potential. Our study seeks to narrow this gap by introducing bioengineering methods to achieve physical integration of hepatic tissue aggregates with perfusable, self-assembled microvasculature capable of exhibiting barrier integrity and signaling function.

Importantly, the highly perfusable vasculature in our system plays a key role in facilitating our studies of monocyte recruitment. While it has been clinically established that there is increased monocyte recruitment in metabolic liver diseases such as MASLD and MASH, the timing and relationship between the recruitment of this immune population and a loss of tissue-resident Kupffer cells remains incompletely understood^39,73,74. Given that our model improves upon other vascularized liver MPS by featuring donor-matched Kupffer cells^27,28,30,75, this system is well positioned for deeper investigation of immune cell dynamics following recruitment. Having addressed the technical challenge of enabling hepatocyte-immune-microvascular interactions, our group is now able to transition toward using non-parenchymal cells (NPCs) that are either liver-specific or derived from human induced pluripotent stem cells (iPSCs)⁷⁶. In particular, incorporating rigorously validated, functional iPSC-derived NPCs into these complex in vitro systems will open up new avenues for patient-specific disease modeling, allowing us to gain more physiologically relevant insights into the immune and vascular features that accompany and contribute to metabolic dysregulation in the liver.

It is also important to highlight that, when studying multicellular models, bulk measurements represent a net response of the full system that is influenced by each cell type. For example, in our liver MPS we saw that the vascular networks themselves not only exhibit disease states, but also contribute to broader phenotypes. While incorporating a greater number of cell types may ultimately improve physiological relevance of the system, it also complicates interpretations of experimental results, requiring additional specialized assays to parse cell type-specific contributions, especially given the inherent variability across devices (Supplementary Fig. 13). One approach to further examine this interplay between tissue and vasculature in vitro is through altering the ratio of hepatic spheroids to vascular network cells in our model. Initial experiments varying this ratio indicated it was an accessible approach to better infer spheroid contributions, though more extensive optimization will be required to define an optimal spheroid ratio without introducing undue vascular heterogeneity (Supplementary Fig. 14).

As we move towards deeper characterization of complex disease models—and ultimately, their adoption for new mechanistic studies—it will also be essential to carefully consider which assays and experimental design schemes allow us to accurately assess the contribution of specific cell populations. For instance, gene expression profiles are influenced not just by the presence of constituent cell types, but also cell-cell interactions and environmental cues such as flow and ECM-mediated signaling. Additionally, when these complex systems are maintained over multiple weeks, changes in the proportions of each cell type may also affect metabolic metrics such as the balance of glucose production versus consumption, underscoring the value of future studies that can examine these dynamics with higher spatiotemporal resolution. Finally, exploring how primary cell donor characteristics such as sex, age, and metabolic status influence disease phenotypes will be informative for patient-specific modeling that can guide therapeutic applications.

In conclusion, we present a liver MPS model of perfused, vascularized primary human hepatic spheroids. This platform recapitulates key metabolic and immune features of hepatic insulin resistance and establishes new methods for investigating disease-relevant monocyte infiltration. With further advances underway to incorporate liver-specific or iPSC-derived NPCs, this vascularized liver MPS holds promise for improving our ability to model complex liver disease biology and inform future therapeutic development.

Methods

Ethical statement

This study was conducted following all relevant ethical regulations and guidelines. The primary human liver cells used in this study were sourced from LifeNet Health (Virginia Beach, VA USA). Informed donor consent and research permissions were obtained by LifeNet Health for all donor cells used in this study. Primary human monocytes used in this study were sourced from buffy coats from deidentified healthy donors via the Massachusetts General Hospital Blood Transfusion Service with approval from the Massachusetts Institute of Technology Committee on the Use of Humans as Experimental Subjects (E-4269).

Cell culture

Human umbilical vein endothelial cells (HUVECs; Angioproteomie, cAP-0001) and normal human lung fibroblasts (NHLFs; Lonza, CC-2512) were cultured in flasks using VascuLife VEGF Endothelial Medium Complete Kit (VascuLife; LifeLine, LL-0003) and FibroLife S2 Fibroblast Medium Complete Kit (FibroLife; Lifeline, LL-0011), respectively. GFP- and RFP-expressing HUVECs were used in studies that involved real-time visualization of microvascular network formation (Angioproteomie, cAP-0001GFP and cAP-0001RFP). Cells were cultured at 37 °C and 5% CO2 in a humidified incubator, with media changed every other day. HUVECs and NHLFs were detached using TrypLE (ThermoFisher) and used between passages 5-9 for all experiments.

Primary human hepatocytes and Kupffer cells in this study were donor-matched from three individuals (Supplementary Table 3) (LifeNet Health LifeSciences). Experiments presented in the main text were all performed with hepatocytes and Kupffer cells from Donor #1 (53-year-old Hispanic male). Hepatocytes were thawed in Cryopreserved Hepatocyte Recovery Media (CHRM; ThermoFisher), spun down at 100 g for 8 minutes, and resuspended in our custom “physiological” William’s E / Vasculife media formulation (Supplementary Table 1). Kupffer cells were thawed in ice cold William’s E media and spun at 450 g for 5 minutes before being resuspended in physiological medium. Both hepatocytes and Kupffer cells were thawed directly before spheroid formation.

Multicellular hepatic spheroid formation

Spheroids were formed using alginate microwells fabricated using a PDMS reverse-mold as previously published⁴⁰. Alginate microwells were primed with 5% BSA for at least 2 h at 37 °C before seeding. 120,000 primary human hepatocytes, 60,000 HUVECs, 12,000 Kupffer cells, and 12,000 NHLFs were pre-mixed to a volume of 75 µL per well. After removing the BSA solution, alginate microwells were primed with 75 µL of culture media, then the spheroid cell mixture was added dropwise directly onto the alginate microwells. In total, each alginate well contains 600 spheroids, where each spheroid is composed of ~200 hepatocytes, 100 HUVECs, 20 Kupffer cells, and 20 NHLFs (with some inherent level of variability).

Microfluidic device fabrication

The microfluidic devices used in this study consist of three parallel microchannels and were fabricated using soft lithography as previously described⁴⁴. Briefly, polydimethylsiloxane (PDMS; Sylgard 184, Ellsworth Adhesives) elastomer and curing agent were mixed in a 10:1 w/w ratio, degassed and poured onto a silicon master mold, and cured at 60 °C overnight. The PDMS was removed from the mold and cut into individual devices, and biopsy punches were used to create inlet/outlet ports in the central gel channel (2 mm) and media reservoirs (5 mm). Dust was removed from the PDMS surfaces using Scotch tape before the devices were dry sterilized in the autoclave. The PDMS devices were then treated with plasma (Harrick Plasma Cleaner) for 2 minutes and bonded to 6-well glass-bottom plates (Cellvis, P06-1.5H-N) to facilitate efficient live imaging during experiments (Supplementary Fig. 1). After plasma bonding, devices were left overnight in a 60 °C oven to recover hydrophobicity before use. For this study, a device geometry was selected in which the central gel channel measures 3 mm wide and 500 µm tall, in order to provide sufficient space for microvascular networks to form around the ~150 µm spheroids⁴³. For experiments aimed at testing a higher ratio of spheroids to vascular network cells, a smaller microfluidic device geometry was used, with a central gel channel that measures 1.8 mm wide and 400 µm tall and accommodates a total of 10 µL of gel. For initial media screens to test perfusable network formation (without spheroids), AIM Biotech idenTx chips were used with the same cell concentrations.

Device seeding

After 48 hours of aggregation in alginate wells, spheroids were harvested from alginate wells by pipetting with a wide bore pipette tip and incubated with AlgiMatrix Dissolving Buffer (Gibco, A11340) for 5 minutes before spinning down at 50 g for 2 minutes. Supernatant was aspirated and cells were resuspended in 1 mL of media. After spheroids naturally settled by gravity, media was aspirated to remove loose single cells and spheroids were resuspended again before being combined with HUVECs and NHLFs.

Fibrinogen from bovine plasma (Sigma, F8630) was dissolved at 37 °C in Dulbecco’s. Phosphate-Buffered Saline (DPBS) to a concentration of 6 mg/mL. Thrombin from bovine plasma (Sigma, T4648) was reconstituted to 100 U/mL in 0.1% w/v bovine serum albumin in water, and then diluted in physiological media to a concentration of 3-4 U/mL. HUVECs, NHLFs, and spheroids were combined in a master mix and centrifuged at 200 g for 5 minutes, with the resulting cell pellet resuspended in the diluted thrombin solution. For each device, 15 µL of the thrombin cell suspension was mixed thoroughly with 15 µL fibrinogen (final concentration: 3 mg/mL) and pipetted into the central gel channel using a wide bore pipette tip. The final concentrations of HUVECs, NHLFs, and spheroids were 7 M/mL, 1.5 M/mL, and ~200 spheroids/device, respectively. Some devices included only HUVECs and NHLFs as a “vascular networks alone” control. After seeding, devices were moved to a humidified incubator for 15 minutes to allow the fibrin gel to polymerize, after which cell culture medium (either the “physiological” or “insulin resistance” (IR) formulation, as described in Supplementary Tables 1 and 2) was added to the side channels and reservoirs. To provide interstitial flow to aid with vasculogenesis, devices were placed on an OrganoFlow rocker platform (Mimetas) set to tilt 8° every 8 minutes along the axis perpendicular to the gel channel during the experiment.

Device culture

Media changes (150 µL removed from and added to the reservoirs) were performed every 24 h. Collected media was cleared of cell debris by spinning at 1000 g for 5 minutes immediately after collection, then the supernatant was frozen and stored at −80 °C until further use. Four days after initial device seeding, a monolayer of HUVECs was added (1 million/mL, 30 µL per side channel) as described previously, in order to facilitate the generation of perfusable microvascular networks and to prevent non-luminal transport of dye or immune cells between the central gel compartment and media channels in downstream assays⁴⁴.

Multicellular hepatic spheroid culture

For some experiments, multicellular hepatic spheroids were maintained in alginate wells as a spheroid-only control for our vascularized MPS model. After 48 h of aggregation, spheroids were switched into either the physiological or IR medium formulation and maintained for 12 additional days. Media changes (300 µL per well) were performed every 48 h according to previously published protocols⁴⁰.

Albumin quantification

To assess hepatocyte spheroid function, stored cell supernatant was thawed at 4 °C and albumin secretion was assayed via ELISA (Bethyl Laboratories, E80-129) according to manufacturer instructions.

CYP3A4 assay

Devices were rinsed with PBS before adding media containing Luciferin-IPA reagent at a 1:1000 dilution from CYP3A4 P450-Glo™ kit (Promega). Devices were incubated for 1 h before media collection and stored at −80 °C until downstream analysis. Media was thawed at room temperature and the assay was performed according to manufacturer protocol alongside a luciferin standard curve to calculate CYP3A4 levels. Plates were read with a Spectramax i3x plate reader luminescence cartridge.

Insulin clearance quantification

Insulin clearance was evaluated from cell supernatant via ELISA (R&D, DY8056). Physiological media samples were diluted 1:2, while IR media samples were diluted between 1:3-1:5 for each insulin ELISA. Clearance fraction was calculated by dividing the measured insulin concentration in cultured media by the measured insulin concentration in cell-free control wells.

Glucose output assay

To assess glucose production, devices were rinsed 3 times for 10 minutes in William’s E media containing no serum, glucose, or insulin. Next, devices were transitioned to a “Glucose production media” containing 50 µM glucose, 20 mM lactate, and 1 mM pyruvate for 24 h (William’s E basal medium, 3.5% dialyzed Heat Inactivated FBS, 2.5 ng/mL VEGF, 2.5 ng/mL EGF, 2.5 ng/mL FGF-b, 7.5 ng/mL IGF-1, 2 mM Glutamax, 1% Pen/Strep, 100 nM hydrocortisone, 7.5 mM HEPES, 50 µm glucose, 20 mM lactate, and 1 mM pyruvate). Collected media was spun down at 1000 g for 5 minutes. Glucose concentration was then measured with the Amplex™ Red Glucose Assay kit (A22189, Invitrogen).

Luminex assay

Supernatant was analyzed according to manufacturer instructions using R&D Systems Custom 28-plex or 12-plex analyte panels. Assay kits were adapted to fit a 384 well format, using 12.5uL of sample and antibody beads per well. Samples were run at both low (2X) and high (25-30X) dilutions in duplicate on a Bio-Plex 3D suspension array system (Bio-Rad). Downstream analysis was performed in MATLAB to fit a 5-parameter logistic standard curve. Any samples with bead count less than 30 or analytes outside the limits of detection were excluded from downstream analysis. Heatmaps were generated on row z-scores using gplots in R (4.0.3). Significance was assessed via Wilcoxon rank-sum test with Benjamini-Hochberg p-value adjustment for multiple hypothesis correction. Partial least squares discriminant analysis (PLS-DA) was performed using ropls (1.20.0) and an adaptation of the systemsseRology package with LASSO feature selection on the dataset z-scores. PLS-DA scores and loadings were plotted with ggplot2 (3.3.3) and marginal distribution was plotted with ggExtra (0.10.1).

Gel harvesting

The devices were rinsed 1X with PBS, the PDMS was cut along the media channels with a sterile scalpel, the PDMS above the central gel channel was lifted, and then the gel was scooped into a tube containing 50 µL of Liberase^TM TM (Roche, LIBTM-RO) at 0.5 mg/mL in William’s E medium. Gels were then incubated for 20 minutes at 37 °C with intermittent mixing by gently pipetting up and down until the gel flowed easily and visibly broke apart. The samples were then spun down at 400 g for 5 minutes. Supernatant was removed and cells were transferred to a 96-well plate for staining or frozen in appropriate reagent for downstream RNA or protein extraction.

RNA analysis

Harvested cells were frozen in Trizol at −80 °C until ready for RNA isolation. Samples were thawed and incubated at room temperature for 5 minutes, then spun down at 1000 g for 5 minutes to remove any residual debris. RNA was isolated with Direct-zol RNA MiniPrep kit (Zymo Research) according to manufacturer instructions, including the in-column DNase I treatment step. RNA was quantified using Nanodrop One and converted to cDNA using the High-Capacity RNA-to-cDNA Kit (ThermoFisher Scientific, 4387406). Next, qPCR was performed using TaqMan Fast Advanced Master Mix (Applied Biosystems, 4444557) with probes for PCK1 (Hs00159918_m1), G6PC (Hs00609178_m1), FASN (Hs01005622_m1), FABP1 (Hs00155026_m1), and endogenous control 18S (Hs03928990_g1). Samples were measured on a StepOnePlus Real-Time PCR system (Applied Biosystems). The delta-delta Ct method was used to calculate the relative fold change in gene expression using 18S as the internal reference.

Immune cell perfusion

Buffy coats were acquired from Massachusetts General Hospital. The blood was diluted 1X in PBS and layered over Lymphoprep (StemCell Technologies), then spun for 30 minutes with no brake. The PBMC layer was carefully removed, spun for 10 minutes, and resuspended in FACS buffer (PBS, 2% Heat Inactivated FBS, 2 mM EDTA) before proceeding directly to CD14+ cell isolation using positive selection beads (StemCell, 17858). CD14+ cells were labeled with CellTracker Deep Red (ThermoFisher, C34565) or CellTracker Red CMTPX (ThermoFisher, C34552) for 20 minutes. After resuspending monocytes to a concentration of 1 million/mL, cells were perfused through microfluidic devices by removing all media from the device, followed by addition of 40,000 monocytes to one media channel. Media was allowed to equilibrate, enabling flow of monocytes across the gel channel to the opposing media channel for 5 minutes, before adding an additional 100 µL of media to each media channel in static culture, with daily media changes for 2-4 more days.

Flow cytometry

Immediately after being harvested from devices, cells were rinsed 1X with PBS and spun at 200 g for 5 minutes. Cells were labeled with a fixable viability dye (BioLegend, 423107; BioLegend 423105) for 15 minutes at room temperature. After washing in FACS buffer, cells were incubated with an FC block for 15 minutes before being stained for 40 minutes on ice. Cells were fixed with 4% paraformaldehyde for 15 minutes and stored in FACS buffer at 4 °C until measurement on a BD FACSymphony A3 Cell Analyzer (BD Biosciences).

Vascular permeability analysis

After the microvascular networks became perfusable, vascular permeability was measured using time-lapse confocal imaging (Zeiss LSM 880) as previously published⁴⁴. Briefly, all media was removed from the devices and replaced with a solution of 0.1 mg/mL 10 kDa Cascade Blue dextran (ThermoFisher, D1976) in media. Three random, non-overlapping regions of interest (ROIs) were chosen for each device; for each ROI, two z-stacks (step size = 5 µm, 10 steps, resolution = 640 ×640 pixels, ROI dimensions >= 600 µm x 600 µm, 10x objective) were captured 9 minutes apart. Vascular permeability values were computed after a custom FIJI macro was used to calculate morphological parameters and changes in the average fluorescence intensity in the vasculature and matrix over time, as previously described. For each device, the average measurement of the 3 ROIs was reported.

Vessel morphology analysis

Maximum intensity projection images were generated from the same ROIs of dextran-perfused devices used in the vascular permeability analysis. Quantitative analyses of microvascular network morphology were performed for these ROIs using the open-source tool REAVER to obtain metrics such as “mean segment diameter” (2D projected diameter) and “vessel area fraction”⁵⁶. For each device, the average measurement of the 3 ROIs was reported. To control for the inherent inter-experiment variability that occurs during device seeding, measurements were compared between media conditions within the same experiment rather than pooled across experiments.

Immunostaining

Devices were washed 1X with DPBS and fixed in 4% paraformaldehyde for 30 minutes at room temperature on a rocker platform. After fixation, devices were washed 3X with DPBS and permeabilized with 0.5% Triton-X-100 in PBS for 2 h at room temperature. Devices were then washed and incubated with blocking buffer overnight (5% donkey serum in PBS) at 4 °C. Samples were then incubated with some combination of primary antibodies for CD31 at 1:100 (AF806, Bio-Techne, lot # DGX0523031), CD68 at 1:200 (14-0688-82, Invitrogen, lot # TE2570551), CD163 at 1:100 (AF1607, Bio-Techne, lot #JPZ0224011), and Arginase-1 at 1:200 (14-9779-82, Invitrogen, lot # 2265377), suspended in blocking buffer and incubated for 48 h on a rocking platform at 4 °C, before being washed 3 times and incubated with AlexaFluor secondary antibodies at a 1:200 (A-21447, Invitrogen, lot # 1977345) or 1:500 dilution (A11015, Invitrogen, lot # 2785566; A10037, Invitrogen, lot # 2300930) in blocking buffer for 48 h at 4 °C. All fluorescent images were taken on a Zeiss LSM 880 confocal microscope (10x, 20x objectives) or a Keyence BZ-X800 fluorescence microscope (4x, 10x objectives).

Statistical analysis

Statistical analysis was performed using GraphPad Prism 10. Bar plots show averages with error bars representing standard deviation (SD). Mann-Whitney tests were performed in comparisons of two groups. Benjamini-Hochberg (BH) multiple hypothesis correction was applied for analysis of multiple comparisons in Fig. 3. For multiple groups, two-way ANOVAs with appropriate post-hoc tests were performed. Statistical significance was defined as *p < 0.05, **p < 0.01, ***p < 0.001, and ****p < 0.0001.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.