Role of Hepatic Stellate and Liver Sinusoidal Endothelial Cells in a Human Primary Cell Three-Dimensional Model of Nonalcoholic Steatohepatitis

Abstract

Nonalcoholic steatohepatitis (NASH) is an inflammatory and fibrotic liver disease that has reached epidemic proportions and has no approved pharmacologic therapies. Research and drug development efforts are hampered by inadequate preclinical models. This research describes a three-dimensional bioprinted liver tissue model of NASH built using primary human hepatocytes and nonparenchymal liver cells (hepatic stellate cells, liver sinusoidal endothelial cells, and Kupffer cells) from either healthy or NASH donors. Three-dimensional tissues bioprinted with cells sourced from diseased patients showed a NASH phenotype, including fibrosis. More importantly, this NASH phenotype occurred without the addition of disease-inducing agents. Bioprinted tissues composed entirely of healthy cells exhibited significantly less evidence of disease. The role of individual cell types in driving the NASH phenotype was examined by producing chimeric bioprinted tissues composed of healthy cells together with the addition of one or more diseased nonparenchymal cell types. These experiments reveal a role for both hepatic stellate and liver sinusoidal endothelial cells in the disease process. This model represents a fully human system with potential to detect clinically active targets and eventually therapies.

Graphical abstract

Nonalcoholic fatty liver disease is a spectrum of diseases characterized by an excess of fat in the liver (steatosis) found in approximately 30% of the worldwide population.¹ Approximately 20% of the patients with nonalcoholic fatty liver disease will develop nonalcoholic steatohepatitis (NASH), inflammation of the liver characterized by hepatocellular ballooning with or without fibrosis, that can progress further to a higher amount of fibrosis, cirrhosis, liver failure, and/or hepatocellular carcinoma. In the coming years, NASH is expected to surpass hepatitis C as the main cause of liver transplant in the United States.² The mortality rate of patients with NASH of 7.9% is twice as high as that of the general population. The prevalence of NASH in the United States is estimated to double every 10 years, with approximately 43 million Americans expected to be affected by the disease by 2025.³

Nonalcoholic fatty liver disease results from a complex interaction of metabolic, environmental, microbial, and genetic factors. Obesity, type 2 diabetes, and dyslipidemia (high serum triglyceride levels and low serum high-density lipoprotein levels) are features of the metabolic syndrome that are common comorbidities of nonalcoholic fatty liver disease.⁴ Fat accumulation in hepatocytes can cause lipotoxicity and inflammation in some individuals, leading to NASH.⁵ There is a complex interplay between the different liver cell types in the pathophysiology of the disease.⁶ Enhanced cytokine secretion from activated liver-resident macrophage Kupffer cells (KCs) and infiltrating cells of the immune system activate myofibroblast hepatic stellate cells (HSCs), inducing the secretion of collagens and other extracellular matrix components that lead to fibrosis. Chronic inflammation also changes the biology of specialized liver sinusoidal endothelial cells (LSECs) that line the liver microvessels, such as reducing their ability to allow nutrients to pass from the blood to the liver parenchyma.⁷ Thus, LSECs play an important role in the disease processes, and therapeutic efforts are underway for targeting this cell type in NASH.⁸

Although many clinical trials are ongoing, there are no US Food and Drug Administration–approved therapies for NASH. Progress in drug development efforts is hampered because preclinical animal and two-dimensional human culture models fail to recapitulate important biological and pathophysiological features of NASH,⁹ although newer animal and three-dimensional (3D) cell culture models more closely resemble human NASH.^10,11 3D bioprinted liver models using primary human liver cells have been developed12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22 that exhibit tissue-like density with highly organized cellular features, including intercellular tight junctions, microvascular networks, and a microenvironment that is more suited for in vivo–like cellular function than two-dimensional monocultures or monolayer co-cultures, as well as maintenance of a more defined architecture than what is observed in self-aggregated co-culture models. 3D bioprinted tissues preserve metabolic functions in culture, including the activities of key enzymes and molecular transporters, without significant decline for several weeks. The responses of 3D bioprinted tissues to acute or chronic exposure to drugs and known toxins resemble in vivo liver tissue. The automated bioprinting process results in scalable tissues, with tight control of the composition and morphology. Previous 3D bioprinted models used cells from healthy donors, human umbilical vein endothelial cell (HUVEC) umbilical cord endothelial cells rather than LSECs, and additives to induce the disease process, such as transforming growth factor-β.16, 17, 18, 19

To address the need for a tissue model that more directly resembles NASH, the current study used liver cells from either healthy or NASH-diseased donors and showed that 3D bioprinted tissues from NASH-diseased cells exhibit increased fibrosis compared with healthy donors without artificial disease-inducing agents. This model represents a fully human NASH system with the potential to detect clinically active drug targets. Furthermore, using chimeric bioprinted tissues containing one or more diseased cell types in the context of a healthy background of other cell types helped determine the contribution of each cell type to the NASH phenotype.

Materials and Methods

Cells

Hepatocytes were obtained from Corning (Corning, NY; donor 4180) or LifeNetHealth (Virginia Beach, VA; donor HL170057), nonparenchymal cells (NPCs) were obtained from LifeNetHealth, and HUVECs were obtained from ATCC (Manassas, VA). Cryopreserved hepatocytes and NPCs were procured from livers of deceased healthy and NASH-diseased donors based on informed consent and in compliance with applicable laws. Table 1 lists the donors for each cell type, who were carefully selected with no history of alcohol abuse, hepatic viral disease, or use of drugs that affect liver functions. Table 2 summarizes the healthy and diseased donors that were used for each cell type. LSECs, HSCs, and HUVECs were propagated per manufacturer's instructions before bioprinted tissue fabrication. Cryopreserved cells were thawed per manufacturer's instructions before bioprinting.

Table 1.

Bioprinted Tissues and Donor Cells Used for This Study

Bioprinted tissue name (composition)	Bioprinted tissue cell composition
	Hepatocyte donor (status)	HUVEC or LSEC endothelial cell donor (status)	Stellate cell donor (status)	KC donor (status)
H-HUV (fully healthy with HUVEC)	4180 (Healthy)	HUVEC (healthy)	HL160036 (healthy)	HL160036 (healthy)
H1 (fully healthy with LSEC)	4180 (Healthy)	HL160026, LSEC (healthy)	HL160036 (healthy)	HL160036 (healthy)
dK-HUV (diseased KC with HUVEC)	4180 (Healthy)	HUVEC (healthy)	HL160036 (healthy)	HL160041 (diseased)
dS-HUV (diseased HSC with HUVEC)	4180 (Healthy)	HUVEC (healthy)	HL160041 (diseased)	HL160036 (healthy)
dE (diseased LSEC)	4180 (Healthy)	HL160041, LSEC (diseased)	HL160036 (healthy)	HL160036 (healthy)
dESK1 (diseased LSEC, HSC, and KC)	4180 (Healthy)	HL160041, LSEC (diseased)	HL160041 (diseased)	HL160041 (diseased)
H2 (fully healthy with LSEC)	4180 (Healthy)	HL170044, LSEC (healthy)	HL160036 (healthy)	HL160036 (healthy)
dES (diseased LSEC and HSC)	4180 (Healthy)	HL160041, LSEC (diseased)	HL160041 (diseased)	HL160036 (healthy)
dESK2 (diseased LSEC, HSC, and KC)	4180 (Healthy)	HL190087, LSEC (diseased)	HL190087 (diseased)	HL190087 (diseased)
Initial mixture∗	HL170057 (healthy)	HL170057, LSEC (healthy)	HL160038 (healthy)	HL180073 (healthy)

Donor cells are listed as numbers for all liver cell types (hepatocytes, LSECs, HSCs, and KCs). For example, for tissue dESK1, the hepatocytes were from a healthy donor 4180, whereas all the nonparenchymal cell types were from a nonalcoholic steatohepatitis–diseased donor 041. For some tissues, HUVECs, which are umbilical cord endothelial cells, were used instead of LSECs, as indicated.

HSC, hepatic stellate cell; HUVEC, human umbilical vein endothelial cell; KC, Kupffer cell; LSEC, liver sinusoidal endothelial cell.

^∗Initial mixture was not a bioprinted tissue sample, but instead a mixture of the liver cell types mixed together, which were processed for single-nuclei RNA sequencing.

Table 2.

Summary of Cell Types and Donors for the Bioprinted Tissues Used in This Study

Cell type	Status	Donors (number of donors)
Hepatocytes	Healthy	4180 (1)
LSECs	Healthy	HUVEC, HL160026, HL170044 (3)
LSECs	Diseased	HL160041, HL190087 (2)
HSCs	Healthy	HL160036 (1)
HSCs	Diseased	HL160041, HL190087 (2)
KCs	Healthy	HL160036 (1)
KCs	Diseased	HL160041, HL190087 (2)

HSC, hepatic stellate cell; HUVEC, human umbilical vein endothelial cell; KC, Kupffer cell; LSEC, liver sinusoidal endothelial cell.

Bioprinted Liver Tissue Fabrication

The 3D liver tissues, composed of primary human hepatocytes, HSCs, KCs, and LSECs or HUVECs, were manufactured by Organovo (San Diego, CA) with a NovoGen (Pledran, France) Bioprinter onto 24-well transwell culture inserts using patented protocols,12, 13, 14, 15 as previously described.16, 17, 18, 19 Separate high-density bioinks, composed of parenchymal hepatocytes (100% cellular paste, generated via compaction) or NPCs (1.5 × 10⁸ cells/mL formulated in NovoGel 2.0 Hydrogel; Organovo), were prepared and loaded into separate heads of the NovoGen Bioprinter Instrument within a standard biosafety cabinet. A computer script was then executed to deposit the bioinks in a two-compartment planar geometry onto the membranes of standard 24-well 0.4-μm transwell culture inserts (Corning) via continuous microextrusion with the NPC bioink, comprising the border regions of each compartment, and the hepatocyte bioink, filling each compartment (Figure 1A). Following fabrication, the tissues were fed daily with 600 μL of 3D Liver Tissue Media, consisting of William's E with Primary Hepatocyte Maintenance Supplements (Life Technologies, Carlsbad, CA) and EGM-2 (Lonza, Basel, Switzerland), and incubated at 37°C under humidified atmospheric conditions supplemented with 5% CO₂. Tissues were allowed to mature in culture for at least 3 days following fabrication and were substantially free of preformed scaffold before initiation of experimentation, which was designated as day 0. For the time course, tissues were fed daily, as stated in the previous sentences, for a total of 14 days, with the spent media on each day collected and stored for subsequent biomarker analysis. On specified days, the tissues were harvested and processed for transcriptomics analysis and histology.

Figure 1.

Three-dimensional bioprinted tissue layout and cell compositions. A: Boxed area: General layout of bioprinted tissues, which contain a core hepatocyte cell paste surrounded by a nonparenchymal cell paste composed of endothelial cells [human umbilical vein endothelial cells (HUVECs) or liver sinusoidal endothelial cells (LSECs)], hepatic stellate cells (HSCs), and Kupffer cells (KCs), as described in Materials and Methods. B: Compositions of the bioprinted tissues used in this study. The general composition of a fully healthy bioprinted tissue is shown in the upper right. Cells from healthy donors are displayed in blue, whereas cells from nonalcoholic steatohepatitis (NASH)–diseased donors are displayed in red. Chimeric bioprinted tissues contain certain cell types from healthy donors and other cell types from NASH-diseased donors.

Composition of Bioprinted Tissues

Bioprinted human liver tissues were prepared in three bioprinting runs (Table 1). The cell type compositions of the bioprinted tissues are diagrammed in Figure 1B. The first run contained healthy (H) tissues with HUVECs (HUV), abbreviated H-HUV, and were composed of hepatocytes, HSCs, and KCs from healthy liver donors, and HUVECs. These were the same as produced in previous studies16, 17, 18, 19 and lacked LSECs. Diseased (d) tissues for this first group contained one or more diseased NPC types: dK-HUVs were composed of KCs from a NASH-diseased donor, hepatocytes and HSCs from healthy donors, and HUVECs; dS-HUVs were composed of HSCs from a NASH-diseased donor, hepatocytes and KCs from healthy donors, and HUVECs; dEs were composed of LSECs from a NASH-diseased donor, and hepatocytes, KCs, and HSCs from healthy donors; and dESK1s were composed of LSECs, HSCs, and KCs from NASH-diseased donors and hepatocytes from a healthy donor. For the second run, healthy bioprinted tissues were prepared using all cell types from liver, including LSECs from healthy donors, abbreviated H1. The successful production of the second run showed that viable bioprinted tissues could be made with all cell types derived from the liver. The third run used healthy (abbreviated H2) tissues with cells from the same healthy donors as with the first group, except for the LSECs, which were from another healthy donor; diseased tissues with LSECs and HSCs from the same NASH-diseased donor (labeled dES) and hepatocytes and KCs from the same healthy donors; and, finally, diseased tissues containing LSECs, HSCs, and KCs from another NASH-diseased donor (labeled dESK2) and hepatocytes from the same healthy donor.

Histologic Analysis Using Paraffin Sections

On days 8 and 14 after treatment, liver tissues were fixed in 5% formalin solution for 24 hours at 4°C. After 24 hours, the fixation solution was removed and replaced with 70% ethanol. Constructs with the attached transwell membrane were then processed for paraffin embedding using a tissue processor. Following infiltration with paraffin, liver constructs were embedded in paraffin molds, and cross-sections (5 μm thick) were prepared with a rotary microtome. For staining, sections were dewaxed in xylene and rehydrated in an ethanol gradient.

Hematoxylin and Eosin Staining

Slides were immersed in Gill hematoxylin (Fisher Scientific, Pittsburgh, PA) and eosin solution (American MasterTech Scientific, Lodi, CA), then dehydrated through an ethanol gradient, cleared in xylene, and mounted with resinous mounting media (CytoSeal; Fisher Scientific).

Trichrome Staining

Slides were incubated in Bouin Solution (American MasterTech Scientific) for 1 hour at 56°C and rinsed in running tap water until clear. Slides were immersed in Weigert Hematoxylin (American MasterTech Scientific) for 10 minutes, then rinsed in running tap water for 5 minutes. Slides were then immersed in Biebrich Scarlett solution (American MasterTech Scientific) for 5 minutes, then briefly rinsed with deionized water. Slides were cleared in Photsphotungstic/Phosphomolybdic solution (Sigma-Aldrich, St. Louis, MO) for 5 minutes, then immediately placed into two exchanges of Analine Blue solution (American MasterTech Scientific), briefly dipping for each exchange. Slides were immersed in 1% Acetic Acid Solution (American MasterTech Scientific) for 2 minutes, then rinsed briefly in deionized water. Slides were then dehydrated through an ethanol gradient, cleared in xylene, and mounted with resinous mounting media (CytoSeal).

Picrosirius Red

Slides were incubated overnight at room temperature in Bouin Fixative (American MasterTech Scientific). Following overnight incubation, slides were rinsed in two changes of deionized water. Slides were then immersed in Picrosirius Red Solution (American MasterTech Scientific) overnight at room temperature. After overnight incubation, slides were rinsed in two changes of Glacial Acetic Acid Solution (American MasterTech Scientific) for 5 seconds each. Slides were then dehydrated through an ethanol gradient, cleared in xylene, and mounted with resinous mounting media (CytoSeal).

PSR Quantitation

Digital Carl Zeiss Images (CZI file format) of picrosirius red (PSR)–stained bioprinted tissues were imported into the HALO software version 3.6.4134 (Indica Labs, Albuquerque, NM) for analysis. Annotations were generated automatically around the tissue area, and the area quantification module was used to determine the percentage area of collagen compared with the total area of the tissue. The software was trained to differentiate collagen fibers from the background staining based on threshold values for individual pixels. The percentage area of collagen data was then exported and graphed using GraphPad Prism 7 (GraphPad Software, San Diego, CA).

Immunofluorescence Staining

Sections were hydrated in deionized water for 5 minutes, followed by incubation in Dulbecco's phosphate-buffered saline. Heat-mediated antigen retrieval was performed with citrate buffer (Sigma-Aldrich), followed by endogenous peroxidase block, avidin/biotin block (Vector Labs, Burlingame, CA), and 10% normal goat serum (Vector Labs).

CD31

Samples were incubated with primary CD31 antibody (ab76533; 1:250; Abcam, Cambridge, MA) overnight at 4°C. The next day, samples were incubated with secondary antibody (Vector Labs), followed by streptavidin (Vector Labs), and amplified with tyramide signal amplification (Life Technologies). All slides were cover-slipped with Fluorogel II with DAPI (Electron Microscopy Sciences, Hatfield, PA).

Desmin/α-Smooth Muscle Actin

Immunofluorescence double staining took place on 3 separate days. On day 1, samples were incubated with desmin antibody (ab15200; 1:200; Abcam) overnight at 4°C. On day 2, samples were incubated with secondary antibody (Vector Labs), followed by streptavidin/horseradish peroxidase (Vector Labs), and amplified with tyramide signal amplification (Life Technologies). Slides were then blocked again for peroxidase, avidin, and biotin, as performed on day 1. Slides were then incubated with α-smooth muscle actin antibody (ab7817; 1:200; Mouse Anti-Alpha Smooth Muscle Actin antibody; 1A4; Abcam) overnight at 4°C. On day 3, samples were again incubated with secondary antibody, streptavidin/horseradish peroxidase, and tyramide signal amplification, as performed on day 2. All slides were cover-slipped with Fluorogel II with DAPI.

Albumin

Samples were incubated with mouse anti-albumin antibody (ab236492; 1:100; Abcam) overnight at 4°C. The next day, samples were incubated with goat anti-mouse fluorophore 594 secondary antibody (Life Technologies). All slides were cover-slipped with Fluorogel II with DAPI.

CD68

Samples were incubated with primary CD68 antibody (ab955; 1:100; Abcam) overnight at 4°C. The next day, samples were incubated with secondary antibody (Vector Labs), followed by streptavidin (Vector Labs), and amplified with tyramide signal amplification (Life Technologies). All slides were cover-slipped with Fluorogel II with DAPI. Slides were imaged using a Leica Biosystems (Wetzlar, Germany) Aperio Versa. Images were acquired with ImageScope software version 12.3.3 (Leica Biosystems).

Biomarker Immunoassays

Supernatants were thawed on ice and analyzed by enzyme-linked immunosorbent assay (ELISA) on a microplate reader (BMG Labtech, Ortenberg, Germany) for alanine aminotransferase (ALT; Biotang, Lexington, MA) or albumin (Bethyl Laboratories, Montgomery, TX), per the manufacturer's instructions with minor modifications. Briefly, a half-area 96-well plate was employed, and the volumes of the kit reagents were reduced by 50%. Data were plotted in GraphPad Prism and are expressed as the average of four tissue replicates ± SD. Individual time point statistics were completed using one-way analysis of variance with Dunnett multiple comparisons, and outliers were assessed using Grubbs test. Time course statistics used two-way analysis of variance with Dunnett multiple comparisons. All statistics were calculated using GraphPad Prism 7 software. Supernatants frozen at various time points were thawed on ice and analyzed by ELISA for procollagen type 1 N-terminal peptide (P1NP; Human Pro-Collagen I alpha 1 DuoSet ELISA; R&D Systems, Minneapolis, MN), procollagen type III N-terminal peptide (P3NP; Cisbio, Codolet, France), and arresten (DuoSet ELISA; R&D Systems), per the manufacturer's instructions. Pro-C3 was measured by Nordic Biosciences (Herlev, Denmark).

Lactate Dehydrogenase Assays

Supernatants for lactate dehydrogenase were collected at the indicated time points for each study and analyzed for lactate dehydrogenase activity colorimetrically with a commercially available reagent (Promega, Madison, WI), per the manufacturer's instructions with minor modifications. Briefly, a half-area 96-well plate was employed, allowing the volumes of the kit reagents to be reduced by 50%, and samples were diluted to obtain readings in the linear range of the standard curve. Data were plotted in GraphPad Prism 7 and are expressed as the average of four tissue replicates ± SD. Individual time point statistics were completed using one-way analysis of variance with Dunnett multiple comparisons, and outliers were assessed using Grubbs test. Time course statistics used two-way analysis of variance with Dunnett multiple comparisons. All statistics were calculated using GraphPad Prism 7 software.

Tissue Viability Assays (Alamar Blue Assay)

Tissues were analyzed for viability using resazurin conversion to resorufin (ThermoFisher, Waltham, MA). Tissues were incubated in the presence of resazurin reagent for 1 hour. Following the 1-hour incubation, the reagent was removed from the tissue and transferred to a black 96-well plate for fluorescence measurement on a microplate reader (BMG Labtech). Data are expressed as the average of four or more tissue replicates ± SD. Statistics (t-test) and outliers (Grubbs test) were calculated using GraphPad Prism 7 software.

Palmitic Acid Treatment and Oil Red O Staining

Bioprinted H-HUV tissues were incubated in Dulbecco’s modified Eagle’s medium containing high glucose (ThermoFisher) and either untreated or treated daily with 200 or 500 μmol/L palmitic acid (Sigma-Aldrich) for 21 days. Cryopreserved tissues were sectioned and stained using oil red O staining solution (Sigma-Aldrich) per manufacturer's instructions. Staining was quantified using ImageJ version 1.52k (NIH, Bethesda, MD; http://imagej.nih.gov/ij).

Preparation of Nuclei and Single-Nuclei RNA Sequencing

Snap-frozen bioprinted tissues were ground using a BelArt Scientific (Warminster, PA) Liquid Nitrogen Cooled Mortar and Pestel Set (catalog number H37260-0100). Ground tissues were transferred to ice and resuspended in 1 mL of ice-cold homogenization buffer [1% Triton X-100, 220 mmol/L sucrose, 2.2 mmol/L KCl, 4.4 mmol/L MgCl₂, 8.8 mmol/L Tris, pH 8, 0.35 U/μL RnaseIn Plus (Promega; catalog number N2618), and 0.175 U/μL SUPERaseln (ThermoFisher; catalog number AM2696)] by gentle pipetting. Samples were then centrifuged at 4°C for 5 minutes at 500 × g, supernatants were removed, and pellets were resuspended gently in 0.4 mL of resuspension buffer (2% bovine serum albumin with 0.35 U/μL RnaseIn Plus and 0.175 U/μL SUPERaseln). Then, 25 μL of 1 mg/mL propidium iodide solution was added, and the fluorescent nuclei were isolated using an Aria 2 flow cytometer (BD Biosciences, Franklin Lakes, NJ). A total of 1 mL of resuspension buffer was added to the isolated nuclei before single-nuclei sequencing, performed at the University of California, San Diego, Institute for Genomics Medicine core facility using Chromium Chip Single Cell and Chromium 3′ GEM, Library and Gel Bead Kit version 3 (10x Genomics, Pleasanton, CA) and a NovaSeq 6000 sequencer (Illumina, San Diego, CA), per manufacturer's instructions.

Analysis of Single-Nuclei RNA-Sequencing Data

Quality control, alignment, and quantification of reads were performed using Cell Ranger software version 50.1 (10x Genomics). Sequencing reads were mapped to the human genome (GRCh38) and annotated with Ensembl release 98. The R package Seurat version 4.0.6 (https://satijalab.org/seurat) was used for downstream dimension reduction, clustering, and differential expression analyses. Before downstream analyses, genes that were expressed in <0.1% of the cell were excluded from the analysis. Cells with high percentages of mitochondrial genes (≥5%) or a low number of unique genes per cell (<200) were also removed. After filtering out low-quality cells, gene expression levels were log normalized and scaled using Seurat functions NormalizeData and ScaleData, respectively. DoubletFinder was used to detect and remove any doublets from each sample. Samples were then integrated using Seurat's FindIntergrationAnchors and IntegrateData. Principal component analysis was used on the integrated scaled data. Differentially expressed genes were calculated using the FindMarkers function, which applies the Wilcoxon rank-sum test with Bonferroni correction.

Results

Characterization of 3D Bioprinted Liver Tissue Models

The 3D bioprinted liver tissues have increased fibrosis on treatment with fibrosis-inducing agents,16, 17, 18, 19 and increased steatosis and fibrosis with a nutrient overload treatment resembling a Western diet with inflammatory stimuli (Supplemental Figure S1), thereby recapitulating key features of NASH. In the context of this fully healthy ExVive model, named H-HUV in this study, the contribution of individual nonparenchymal cell types to fibrosis was assessed by replacing cells from healthy donors with cells from NASH-diseased donors. These bioprinted tissues were named dK-HUV when diseased KCs were used, dS-HUV when diseased HSCs were used, and dE when diseased LSECs were used instead of HUVECs (see Materials and Methods and Figure 1B). This allowed for an assessment of the effects of different diseased cell types on the final phenotype of the chimeric bioprinted tissue. Fully healthy, all liver bioprinted tissues were also made, which contained all liver cell types from healthy donors and no HUVECs (named H1 and H2). Finally, bioprinted tissues containing more than one cell type from diseased donors were made (dES with diseased LSECs and HSCs, and dESK1 and dESK2 with diseased LSECs, HSCs, and KCs) to measure additive effects and interactions between the cell types. Hepatocytes for all bioprinted tissues were from healthy donors. Figure 1A shows the general layout of the bioprinted tissues, as described in Materials and Methods, and Figure 1B shows the donor cell composition (healthy or diseased cells) for each bioprinted tissue. More importantly, all the 3D bioprinted tissues were cultured under the same conditions without addition of any fibrosis-inducing agents, and key NASH biomarker and fibrosis assays were conducted to assess phenotypic differences.

Standard tests measuring lactate dehydrogenase, ALT, albumin, and Alamar blue were performed to assess the health and quality of the bioprinted tissues (Supplemental Tables S1, S2, S3, and S4, respectively). Albumin, ALT, and Alamar blue levels differed across some of the samples and groups. For example, albumin and ALT levels were higher for H1 and H2 compared with H-HUV, whereas Alamar blue levels were higher for H-HUV and H1 compared with H2. Furthermore, over time, albumin levels were generally stable for seven of bioprinted tissues, as reported previously,¹⁶ with two tissues, dE and H1, showing statistically significant decreases from day 1 to day 14 (Supplemental Table S3). The variability of these biomarkers in the bioprinted tissues is currently under investigation and could be because of the use of LSECs instead of HUVECs and the use of NASH-diseased NPCs. However, the bioprinted tissues maintained their integrity, as shown by consistent phenotypic collagen biomarker production (see The Bioprinted Model Using Diseased NPCs Recapitulates the NASH Phenotype) and hematoxylin and eosin staining of sections (Supplemental Figure S2) that was similar to previously published findings,16, 17, 18, 19 and met Organovo quality control standards, and all cell types remained viable, as visualized by immunofluorescence staining (Supplemental Figure S3).

The Bioprinted Model Using Diseased NPCs Recapitulates the NASH Phenotype

High collagen production is a hallmark of fibrosis in patients with NASH. Clinically, serum biomarkers can assist in the analysis of patient status and fibrotic progression.23, 24, 25 Collagen production can be quantified by measuring soluble markers of proteolytic collagen processing and can be detected in media or sera using ELISA-based assays. Medium from the bioprinted tissue wells was sampled daily, allowing for the measurement of collagen production biomarkers over time. Initially, the authors measured the production of pro-C3, a marker for collagen III production and a well-characterized, clinically relevant biomarker of disease progression.^26,27 The medium from a set of bioprinted tissues made from either healthy or diseased cells was assessed on days 7 and 13 (Figure 2) using HUVECs to represent healthy endothelial cells (HUV). There were significantly higher levels of pro-C3 on day 7 for dE (diseased LSECs) and dESK1 (diseased LSECs, HSCs, and KCs) tissues, indicating a significantly increased disease phenotype with addition of diseased HSCs and KCs into the bioprints. dK-HUV and dS-HUV tissue media had levels of pro-C3 similar to healthy (H-HUV) tissues (Figure 2A), indicating that without LSECs, these cell types on their own do not trigger an increase in the disease phenotype. On day 13, however, pro-C3 levels were at the same low level for all samples. From this initial finding, the signals from the fully diseased NPC bioprints, dESK1, and the diseased LSEC chimeras, dE, were examined in a more detailed time course (Figure 2B). Pro-C3 levels in the diseased dE and dESK1 tissue media were significantly higher than the healthy H-HUV on the early days. Pro-C3 levels in the diseased tissue media were the highest on the earliest day examined, day 4, and gradually decreased to levels found in the healthy tissues by day 10, suggesting a slow reduction in the disease signal in these tissues.

Figure 2.

The collagen III biomarker pro-C3 is elevated in dE and dESK1 bioprinted tissue media. Medium for the bioprinted tissues was removed every 24 hours and replaced with fresh media. Enzyme-linked immunosorbent assay–based assays for pro-C3 were performed on the media from the indicated days in biological triplicates for each group and day. A: Pro-C3 assay on days 7 and 13 bioprinted tissue media for the indicated samples. B: Pro-C3 assay on a subset of the samples from A in a more detailed time course, days 4, 6, 7, 8, 10, and 12. Refer to Table 1 for the identity of each bioprinted tissue. Statistically significant differences (P < 0.05) between samples for each day are shown, assessed by two-way analysis of variance with the Tukey multiple-comparisons test. Data are given as means ± SEM (A and B). ∗∗P < 0.01, ∗∗∗P < 0.001, and ∗∗∗∗P < 0.0001.

Next, a fully healthy tissue using healthy LSECs instead of HUVECs (H1) was prepared, and a broader set of biomarkers for collagen (I, III, and IV) were analyzed. P3NP is another biomarker for collagen IIIα(1) processing that derives from the same amino-terminal prodomain as pro-C3. Similar to the pro-C3 results, P3NP was significantly elevated on the earlier days in the media of dE and dESK1 tissues compared with the media from healthy tissue (Figure 3A). And like pro-C3, P3NP levels in the diseased tissue media were higher earlier, and then gradually decreased over time, whereas the healthy tissues tended to increase over time. The soluble C-terminal noncollagenous domain 1 (NC1 domain) of collagen IVα(1), known as arresten, was also significantly elevated in dE and dESK1 bioprinted tissue media (Figure 3B), with a greater increase in expression between fully diseased nonparenchymal dESK1 chimeras and healthy tissues persisting over a longer period. In contrast, P1NP, the soluble N-terminal prodomain of collagen 1αI, had a mixed profile (Figure 3C); dESK1 tissues had higher levels of P1NP compared with H-HUV tissues on day 4 only, the earliest day measured, whereas both dE and dESK1 tissues had higher P1NP levels than H1 tissues on days 4 and 6. Therefore, P1NP levels were generally higher in the diseased tissues in the early days, similar to the other collagen biomarkers. In contrast, on the later day 10, H-HUV but not H1 tissues had higher levels of P1NP than dE and dESK tissues. Notably, the levels for all collagen biomarkers were generally similar for H-HUV and H1, showing that the HUVEC and LSEC endothelial cell types, respectively, have the same effect in the bioprinted tissues with respect to collagen production.

Figure 3.

Collagen I, III, and IV biomarker levels are elevated in the media from diseased bioprinted tissues. Enzyme-linked immunosorbent assay–based assays were performed as in Figure 1 for procollagen type III N-terminal peptide (P3NP), a biomarker for collagen III (A and D); arresten, a biomarker for collagen IV (B and E); and procollagen type 1 N-terminal peptide (P1NP), a biomarker for collagen I (C and F). Refer to Table 1 for the identity of each bioprinted tissue. Statistically significant differences (P < 0.05) between samples for each day are shown, assessed by two-way analysis of variance with the Tukey multiple-comparisons test. Data are given as means ± SEM (A–F). ∗P < 0.05, ∗∗P < 0.01, ∗∗∗P < 0.001, and ∗∗∗∗P < 0.0001.

The same experiments measuring P3NP, arresten, and P1NP were performed on an additional set of healthy bioprinted tissues (H2), fully diseased chimeric tissues (dESK2), and chimeric tissues containing diseased LSECs and HSCs (dES) (Figure 3, D–F). Arresten had the highest levels in diseased tissues on the earlier days, particularly with the dES tissues (Figure 3E). Meanwhile, P3NP (Figure 3D) and P1NP (Figure 3) levels were not the highest in the earliest days but were greater than healthy tissues later in the time course. There were statistically significant differences between the two diseased tissues for all the biomarkers. Consistent differences were seen across biomarkers and timing for healthy and disease-containing chimeras. Comparing diseased samples, dES tissues produced more P3NP and arresten on the earliest days, whereas dESK2 tissues produced more P3NP and P1NP on later days.

Standard histologic staining techniques of collagen deposits were used to visualize and compare hepatic fibrosis in healthy and diseased 3D bioprinted tissues (Figures 4, 5, and 6 and Supplemental Figure S4). Masson trichrome staining of the diseased tissues showed more blue stained areas (Supplemental Figure S4); similarly, there was more dark red staining with PSR (Figures 4, 5, and 6, A–C), which corresponded to visibly higher levels of PSR birefringence under polarized light (Figures 4, 5, and 6, D–F). Quantitation of PSR birefringence from the tissues showed that the diseased HSC-containing tissues (dS-HUV) had significantly higher PSR staining (Figure 4G) than control, showing that chimeric bioprints with diseased HSCs had higher collagen deposition. dS-HUV tissues also produced higher levels of soluble fibrinogen, an extracellular matrix protein, providing further evidence that diseased HSCs increase fibrosis in the bioprinted tissue model compared with healthy HSCs (Supplemental Figure S5). Chimeras containing diseased KCs trended higher for PSR birefringence compared with healthy tissues (H-HUV) but did not reach significance (Figure 4G).

Figure 4.

Collagen deposition is elevated in dS bioprinted tissues. A–F: Representative images for bright-field picrosirius red (PSR) staining (A–C), and PSR staining in polarized light (D–F) of bioprinted tissues H-HUV (A and D), dK-HUV (B and E), and dS-HUV (C and F). G: Quantitation and statistical analysis (two-way analysis of variance, Tukey multiple comparisons with P < 0.05 are shown) of PSR birefringence, showing significantly higher levels of collagen deposition in chimeric dS-HUV bioprinted tissues. H is healthy, dS contains diseased hepatic stellate cells, and dK contains diseased Kupffer cells. Refer to Table 1 for the identity of each bioprinted tissue. Data are given as means ± SEM (G). ∗∗P < 0.01. Scale bar = 500 μm (A–F).

Figure 5.

Collagen deposition is elevated in dE and dESK1 bioprinted tissues. A–F: Representative images for bright-field picrosirius red (PSR) staining (A–C), and PSR staining in polarized light (D–F) of bioprinted tissues H1 (A and D), dE (B and E), and dESK1 (C and F). G: Quantitation and statistical analysis (two-way analysis of variance, Tukey multiple comparisons with P < 0.05 are shown) of PSR birefringence, showing significantly higher levels of collagen deposition in chimeric dE and dESK1 bioprinted tissues. Refer to Table 1 for the identity of each bioprinted tissue. Data are given as means ± SEM (G). ∗P < 0.05, ∗∗P < 0.01. Scale bar = 500 μm (A–F).

Figure 6.

Collagen deposition is elevated in dES and dESK2 bioprinted tissues. A–F: Representative images for bright-field picrosirius red (PSR) staining (A–C), and PSR staining in polarized light (D–F) of bioprinted tissues H2 (A and D), dES (B and E), and dESK2 (C and F). G: Quantitation and statistical analysis (two-way analysis of variance, Tukey multiple comparisons with P < 0.05 are shown) of PSR birefringence, showing significantly higher levels of collagen deposition in dES and dESK2 bioprinted tissues. Refer to Table 1 for the identity of each bioprinted tissue. Data are given as means ± SEM (G). ∗∗P < 0.01. Scale bar = 500 μm (A–F).

To test the role of the other cell types in driving collagen deposition, dE and dESK1 tissues were compared with healthy H1 tissues by PSR staining (Figure 5). Chimeras containing diseased LSECs significantly increase fibrosis compared with healthy tissues. Chimeras containing all the diseased NPC types (dESK1) were significantly higher than healthy tissues and trended even higher than the diseased LSEC-containing chimeras (dE) (Figure 5G). The significantly higher PSR birefringence was replicated in diseased cell-containing chimeras relative to healthy bioprinted tissues in a second group of 3D bioprinted tissues (Figure 6). Chimeras containing both diseased HSCs and LSECs (dES) did not differ from chimeras with the addition of diseased KCs (dESK2) (Figure 6G).

snRNAseq of Bioprinted Tissues

The increases in fibrotic biomarkers and in collagen deposition in bioprinted tissues containing different diseased cell types likely reflect alterations in gene expression within selected cell types. Nuclei were isolated from bioprinted tissues shown in Table 3 and subjected to single-nuclei RNA sequencing (snRNAseq) to measure gene expression. A sample containing primary cells, all from healthy donors mixed together, was also examined (initial mixture) (Tables 1 and 3).

Table 3.

Bioprinted Tissues Used for snRNAseq

Tissue∗	Day	Replicates
Initial mixture	0	1
H-HUV	3	2
H1	3	2
dS-HUV	3	2
dE	3	2
dESK1	3	2
H2	3	2
dES	3	2
dESK2	3	2
H1	8	1
dESK1	8	1
dES	9	1
dESK2	9	1
H-HUV	14	1
H1	14	1
dE	14	1
dESK1	14	1

snRNAseq, single-nuclei RNA sequencing.

^∗Tissues are described in Table 1.

There were 54,321 nuclei in the data set. Fourteen clusters were identified, corresponding to five liver cell types, including hepatocytes, HSCs, LSECs, macrophages (KCs and recruited macrophages), and T cells. Cell types were identified using established liver cell marker genes²⁸ (Figure 7, A and B, and Supplemental Figure S6). Clusters 0, 2, and 4 expressed known HSC marker genes, including NGFR, TIMP1, COL1A1, and COL1A2. Cluster 6 expressed fibroblast markers, including FAP and PDGFRB. Clusters 3, 7, 8, and 10 expressed hepatocyte markers, including CRP, HNF4A, and ALB. Cluster 9 expressed LSEC markers, including PECAM1 and FLT4. Cluster 11 expressed T-cell markers, including CD3D, CD3E, TRAC, CD3G, and NKG7. Cluster 12 expressed KC marker genes, including CD163, C1QA, CCR2, and CD163. Clusters 1, 5, and 13 had an abnormally low number of detected features, indicating a possible quality issue for these cells (Figure 7D), and these clusters also did not express the expected pattern of marker genes (Figure 7E). For these reasons, clusters 1, 5, and 13 were excluded from further analysis. Most clusters contained some cells resembling those from the initial mixture population, except for clusters 4 and 5 (Figure 7E). Cell clusters with smaller or absent populations from the initial mixture condition may represent cell types that expanded or developed in the bioprinted or the diseased context. The proportions of normal and diseased cells were relatively even, although clusters 4 and 9 had the largest proportion of cells from diseased bioprinted tissues (Figure 7C). In addition, there were no significant differences by cluster in the distribution of cells grouped by condition (healthy versus diseased), tissue identifier (sample), or time point (Figure 7, F–H).

Figure 7.

Summary of the bioprinted tissue single-nuclei RNA-sequencing results. A: Heat map showing relative average expression of marker genes in organoid clusters. B: Uniform Manifold Approximation and Projection (UMAP) scatterplot showing the 12 identified bioprinted tissue clusters. Clusters 1, 5, and 13 were undefined for reasons explained in the text. C: Bar plot showing the composition of bioprinted tissue types in each cluster, excluding tissues containing human umbilical vein endothelial cells and initial mixture cells. D: Violin plot showing the distribution of the number of features identified per cluster, a proxy for quality. E: Bar plot showing the number of cells from the initial mixture condition in each cluster. F: Bar plot showing the composition of all organoid types per cluster. G: Bar plot showing the composition of time points per cluster. H: Bar plot showing the composition of tissue number per cluster. Refer to Table 1 for the identity of each bioprinted tissue. HSC, hepatic stellate cell.

HSCs from Bioprinted Tissues Resemble NASH HSCs

Using marker genes identified in an snRNAseq study of HSCs from human liver (S.B.R. and D.A.B., unpublished data), the three HSC clusters were characterized. Cluster 0 resembled a quiescent subtype, cluster 4 resembled an activated subtype, and cluster 2 resembled an intermediate activation subtype. Cluster 0 up-regulated markers of quiescence, including BMPR, LRAT, HGF, ETS1, and ADAMTS9, whereas cluster 4 up-regulated markers of activation, including TIMP1, COL1A1, and COL1A2 (Supplemental Figure S7). When a more global analysis was employed, the genes up-regulated in quiescent HSC cluster 0 were negatively correlated with the genes up-regulated by human activated HSCs (r = −0.05; P = 2.5 × 10⁻⁵) (Figure 8A). Similarly, the genes up-regulated in activated stellate cell cluster 4, and to a lesser extent in intermediate activated cluster 2, were positively correlated with the genes up-regulated by human activated HSCs (r = 0.068, P = 4 × 10⁻⁷; and r = 0.043, P = 0.001, for cluster 4 and cluster 2, respectively) (Figure 8, B and C).

Figure 8.

Characterization of hepatic stellate cells (HSCs) from bioprinted tissues. A–C: Scatterplot comparing the average log fold change (FC) between human activated and quiescent HSCs (S.B.R. and D.A.B., unpublished data) on the x axis, and the average log fold change between cluster 0 (A) and cluster 2 (B) and cluster 4 (C) on the y axis. D: Venn diagram depicting the overlap between activated HSC marker genes from a mouse nonalcoholic steatohepatitis (NASH) model (red), and the genes up-regulated in cluster 4 (top diagram) and cluster 0 (bottom diagram). P value computed using a hypergeometric test. E: Heat map showing the relative average expression for the activated and quiescent HSC marker genes identified from the mouse NASH model, which overlap with the marker genes of bioprinted tissue clusters 0, 2, and 4. F: Scatterplot comparing the average log fold change between human activated and quiescent HSCs (S.B.R. and D.A.B., unpublished data) on the x axis, and the average log fold change between the fully diseased dESK1 and dESK2 bioprinted tissues (dESK) versus the healthy H1 and H2 bioprinted tissues, on the y axis. G: Network showing the genes up-regulated in fully diseased dESK1 and dESK2 bioprinted tissues (dESK) versus healthy H1 and H2 bioprinted tissues, which are also in the extracellular matrix pathway, and their known interactions from the molecular interaction network STRING. A–C and F: The linear regression best fit line is shown with 95% CIs. Refer to Table 1 for the identity of each bioprinted tissue. FOZ, HSCs from Alms1 mutant mouse model.

Published marker genes of activated, quiescent, and inactivated HSCs from an Alms1 mutant (FOZ/FOZ) mouse model of NASH²⁹ provide another resource to evaluate how well NASH bioprinted tissues recapitulate NASH. Although none of the three HSC clusters identified here had significant overlap with the inactivated HSC cluster from the FOZ/FOZ mouse model (P > 0.05; data not shown), the marker genes for both HSC cluster 4 (P = 7 × 10⁻⁴⁵) (Figure 8D) and HSC cluster 2 (P = 1 × 10⁻³⁴; data not shown) overlapped significantly with the mouse HSC activation genes. Similarly, the marker genes for HSC cluster 0 overlapped significantly with mouse HSC quiescent genes (P = 2 × 10⁻⁴) (Figure 8D). The overlapping FOZ/FOZ activated HSC genes were strongly expressed in cluster 4 and weakly expressed in cluster 0, whereas the FOZ/FOZ quiescent HSC genes were strongly expressed in cluster 0 and weakly expressed in cluster 4 (Figure 8E), further supporting evidence that cluster 4 represented activated HSC, whereas cluster 0 represented quiescent HSC. Cluster 2 had an intermediate level of expression of both the FOZ/FOZ activated and FOZ/FOZ quiescent marker genes, relative to clusters 0 and 4, indicating that it may be an intermediate cell type that is not fully activated (Figure 8E). When all the HSCs were compared between the fully diseased dESK bioprinted tissues (dESK1 and dESK1) and healthy bioprinted tissues (H1 and H2), the HSCs in the diseased bioprinted tissues up-regulated the same genes that were up-regulated in human activated HSCs (r = 0.065; P = 1.8 × 10⁻⁶) (Figure 8F).

A total of 364 genes in HSC cells were significantly up-regulated in dESK compared with healthy tissues (average log fold change > 0; adjusted P < 0.05). Pathway and network analysis of these genes revealed that they were significantly connected in the interaction network, with more edges observed between genes than would be expected by chance (P < 1 × 10⁻¹⁶), and a significant enrichment for fibrogenic pathways, including extracellular matrix (P = 9.9 × 10⁻¹⁴) and collagen biosynthesis (P = 7 × 10⁻⁷) (Figure 8G and Supplemental Table S5). This confirmed that the genes altered in HSCs from diseased bioprinted tissues impact similar pathways as human NASH.

Diseased NPCs Interact with Other Cell Types to Promote NASH Gene Expression Profiles in the Bioprinted Tissues

Next, the effects of individual cell types, LSECs and HSCs, in the chimeric bioprinted tissues were analyzed. Strikingly, in dE bioprinted tissues, which were composed of LSECs from a NASH-diseased donor, and all other cell types from healthy donors (dE), the HSCs increased expression of certain fibrogenic markers, when compared with these same HSCs bioprinted into fully healthy (H1 and H2) tissues (Figure 9A). This result suggests that the diseased LSECs communicate with HSCs to drive a partial NASH disease-associated gene expression profile in the HSCs. In further support of this finding, comparison of the gene expression changes in HSCs in chimeric diseased dE bioprinted tissues to the full disease dESK bioprinted tissues revealed that dE had a significant positive correlation with dESK (r = 0.40; P < 1 × 10⁻²⁰⁰) (Figure 9B). In the dES chimera, which had diseased LSECs and HSCs, the HSC gene expression profile correlated more strongly to dESK (r = 0.82) (Figure 9B), indicating that the dE chimera captured some, but not all, of the changes observed in the full diseased model (dESK). The high correlation between dES and dESK indicates that the diseased KCs in dESK contribute only minimally to the gene expression profile from the fully diseased HSCs, because the bioprint composed of diseased LSECs and diseased HSCs (dES) altered most of the same genes, in the same direction, as the fully diseased dESK bioprints.

Figure 9.

Partial disease RNA expression subtypes in bioprinted tissues. A: Heat map showing the log fold change of dE, dES, and dESK (dESK1 and dESK2) relative to H (H1 and H2), for select known activated hepatic stellate cell (HSC) marker genes. B: Scatterplots and heat map comparing changes induced in partial disease bioprinted tissues (dE versus H and dES versus H) to changes induced by full disease bioprinted tissues (dESK versus H) in HSCs. C: Scatterplots and heat map comparing changes induced in partial disease bioprinted tissues (dE versus H and dES versus H) with changes induced by full disease bioprinted tissues (dESK versus H) in endothelial cells. D: Scatterplots and heat map comparing changes induced in partial disease bioprinted tissues (dE versus H and dES versus H) with changes induced by full disease bioprinted tissues (dESK versus H) in hepatocytes. E: Scatterplots and heat map comparing changes induced in partial disease bioprinted tissues (dE versus H and dES versus H) with changes induced by full disease bioprinted tissues (dESK versus H) in macrophages. B–E: Heat maps indicate the Pearson correlation in log fold changes between the indicated comparisons, and the scatterplots show the average log2 fold change values of every gene in the indicated comparison, with the top five currently up-regulated and down-regulated genes labeled. Refer to Table 1 for the identity of each bioprinted tissue.

The analysis to examine the effects of diseased HSCs and diseased LSECs on other tissues was extended. Regarding the relationship between full and partial diseased bioprinted tissues with respect to LSECs, dE and dES have similar correlation with dESK (r = 0.61 and r = 0.64, respectively) (Figure 9C), inferring that diseased HSCs from the dES bioprinted tissue do not drive LSECs into a more advanced diseased state, possibly because the LSECs for all these samples were from a diseased donor. However, the observation that the correlation is <1 means that the disease KCs may be having an impact, because disease KCs are present in dESK but not in the other models.

Significant positive correlations in the gene expression profile of hepatocytes (Figure 9D) and KCs (Figure 9E) were observed between dE (diseased LSECs only) versus dESK, and the correlations were even higher between dES (diseased LSECs and HSCs) versus dESK, showing that diseased LSECs and HSCs promoted a diseased gene-expression profile in hepatocytes. The correlations were <1, meaning that diseased KCs in dESK also contributed to a diseased gene expression profile in hepatocytes. The effect of diseased LSECs in the dE bioprinted tissues was more pronounced than in HSCs (r = 0.40) (Figure 9B) compared with hepatocytes (r = 0.25) (Figure 9D) or macrophages/KCs (r = 0.26) (Figure 9E), suggesting that diseased LSECs promoted more of a NASH-associated gene expression profile in HSCs compared with hepatocytes and macrophages.

Taken together, this analysis suggests that the bioprinted tissues with diseased NPCs reproduce a NASH phenotype suitable for use in further research and that each NPC cell type from NASH-diseased donors can promote a NASH-diseased gene expression profile in the other cell types.

Discussion

The compounds developed using the current drug discovery and animal efficacy models have not translated into any approved drugs to treat NASH.³⁰ There exists an opportunity to change this using high-fidelity in vitro human cell models. Using multiple types of human primary cells derived from diseased patient tissues and bioprinted into a 3D multicellular tissue for drug discovery and development provides a patient-centric approach to finding compounds that can more reliably alter the course of human NASH when tested in patients. This approach relies on the patient's own disease, excluding agents to artificially generate a disease phenotype, thus enhancing the relevance of the results. This study presents the initial characterization of bioprinted tissues from patient-derived liver cells and the NASH phenotypic and single-nuclei gene expression differences between healthy and diseased cell-containing bioprinted tissues, and uses chimeric diseased tissues containing one or more diseased cell types in the context of healthy cell types to unravel the role of specific nonparenchymal cells in driving the disease phenotype.

Tissues bioprinted using cells from healthy subjects exhibited less fibrosis when compared with bioprinted tissues prepared with nonparenchymal cells from patients with ongoing NASH (Figures 2, 3, 4, 5, and 6). A key aspect of NASH disease progression is fibrosis, characterized by inappropriate collagen deposition. Collagens are synthesized as procollagen precursors that subsequently undergo various post-translational modifications before formation of mature collagen filaments, including proteolysis to remove the prodomains that are solubilized and can be detected and quantified in media or sera. Differences between healthy and diseased NPC-containing bioprints were detected in several collagen production biomarkers indicative of disease progression, including collagen 1 (P1NP), collagen 3 (pro-C3, P3NP), and collagen 4 (arresten) (Figures 2 and 3).

Collagen production markers were increased in bioprinted tissues from diseased nonparenchymal cells and generally decrease over time during the incubation period, highlighting the dynamic nature of this system. Biomarker differences between healthy and diseased tissues were maximal early in the experiment and waned over time. This is not entirely unexpected as the disease signature was likely overridden by culture conditions. This is an area of active investigation given the value in understanding these signals. The healthy and diseased bioprinted tissues did not differ in matrix metalloproteinase–dependent collagen degradation markers (data not shown). Fibrosis is determined by the balance of production and degradation of collagen. Biomarker data suggested a net increase of collagen deposition in the disease tissues. Experiments demonstrated that bioprinted tissues containing solely diseased nonparenchymal cells (dESK1 and dESK2) had significantly higher levels of collagen deposition compared with tissues printed with nondiseased cells (Figures 5 and 6), consistent with the biomarker data. Thus, diseased NPCs retained the capability to produce disease phenotype after removal from the patient. Single-nuclei RNA sequencing in HSCs (Figure 9, B and C) confirmed that these bioprinted diseased tissues retain a gene expression profile that is similar to that found in human patients with NASH, further validating this model of NASH disease. Hepatic stellate cells in the bioprinted tissues exhibit the expected clusters, including quiescent and activated HSCs. HSCs in the diseased bioprinted tissues tended to up-regulate the same genes that are up-regulated in human activated HSCs (r = 0.065; P = 1.8 × 10⁻⁶) (Figure 8F). This highly significant correlation shows that the bioprinted tissue model shares many key features of NASH disease, although the effect size is relatively small, indicating, not unexpectedly, that differences remain between the bioprinted tissues and natural human livers.

The Uniform Manifold Approximation and Projection (UMAP) scatterplot of snRNAseq data from the bioprinted liver tissues revealed three clusters of HSCs (Figure 7, A and B). By comparing these data with data sets from human and mouse NASH livers, the three clusters were identified as quiescent, activated, and intermediate activated HSCs. The activated HSCs represent the classic liver myofibroblast in the fibrotic liver with high expression of fibrillar collagens, TIMPs, and ACTA2. The activated HSC is the major source of fibrosis in NASH, and by inference, the major source of collagen and fibrosis in the bioprinted liver tissues. Overall, these three HSC clusters resembled HSCs in vivo in NASH. In particular, the presence of the three clusters within the same in vivo liver or bioprinted liver was remarkably similar, demonstrating the heterogeneity of the HSC phenotype in NASH. Therefore, the myofibroblast/activated HSC represents the logical target for antifibrotic therapies.

Because the bioprinted liver tissues demonstrated a significant correlation with the NASH-diseased phenotype, they afford the unique opportunity to analyze the contribution of specific liver cell types to the diseased phenotype. Chimeric bioprinted tissues were built using diseased LSECs, HSCs, or KCs, with the rest of the cells sourced from nondiseased donors. This provides information on which cells can drive fibrotic disease. The collagen biomarker results for pro-C3, P3NP, arresten, and P1NP indicated that introduction of diseased LSECs enhances collagen biomarker production (Figures 2 and 3). This effect was present when comparing with both HUVECs and healthy LSECs, was maximal early in the experiment, and waned over time. The biomarker data were corroborated by quantitative collagen deposition data, indicating that introduction of diseased LSECs into a bioprinted tissue significantly increased the amount of fibrosis detected (Figures 4 and 5). Furthermore, the snRNAseq analysis demonstrated a significant role of diseased LSEC signaling to normal HSCs to induce additional fibrogenic pathways. snRNAseq data from normal hepatocytes and macrophages in the presence of diseased LSECs also indicated that LSECs signal to hepatocytes and macrophages as well. Although weaker, this resulted in the triggering of disease pathways in these originally normal cells. This is consistent with recent findings that LSECs interact more strongly with HSCs than hepatocytes and macrophages in NASH.³¹

The important changes in LSECs in response to liver disease have been described previously.^32,33 Uniquely positioned at the blood-liver interface, LSECs are the first cells to sense toxins or damaging signals. During disease initiation, LSECs undergo capillarization, thereby losing fenestrations and acquiring a more vascular profile. There is also evidence that capillarized LSECs can undergo partial endothelial-to-mesenchymal transformation and acquire collagen deposition capabilities.^34,35 This transformation by LSECs in chronic liver disease occurs before the development of fibrosis, but it was unclear if this is merely a response to disease and a loss of a regenerative signal, or an active driver of the disease process.³⁶ The results shown here clarify this point and indicate that LSECs can actively participate in disease progression and are not simply failing to repress fibrotic signals. Diseased LSECs signal to HSCs via mechanisms under study³⁷ and can induce a fibrotic phenotype in healthy stellate cells. This suggests that the LSECs could be an important target for drug discovery and disease treatment.

The addition of HSCs and KCs (dS and dESK) enhanced the disease biomarker effect beyond that of diseased LSECs alone. Furthermore, the collagen deposition data (Figure 3), which integrate deposition of many collagen subtypes beyond collagen 3, showed a trend for diseased Kupffer cells and a significantly higher signal for chimeras containing only diseased HSCs. The snRNAseq data also indicated significant phenotypic alterations of gene expression, consistent with the diseased phenotype when diseased HSCs are incorporated into the bioprints. Thus, these cells can contribute to a fibrotic phenotype in bioprinted tissues, consistent with literature supporting an important role for these cell types in fibrotic progression.^38,39

Limitations of this work include the use of nonliver HUVEC endothelial cells in the initial bioprinted tissues (H-HUV, dK-HUV, and dS-HUV). The bioprinted tissues in previous studies used HUVECs,16, 17, 18, 19 corresponding to H-HUV in this study. Bioprinted tissues containing cells from NASH-diseased donors had not been previously made, and dK-HUV and dS-HUV in this study established that viable bioprinted tissues could be produced from NASH-diseased donors. Later bioprinted tissues containing LSECs from either healthy or diseased donors showed that viable bioprinted tissues can be made using LSECs. Future studies will assess the role of diseased KCs and HSCs in the context of fully liver LSEC-containing bioprinted tissues to more properly assess the role of these cell types in NASH. Another limitation of this work is the low number of NPC donors tested as well as the use of a single healthy donor for hepatocytes as a baseline for these studies. Donor-to-donor variation may have contributed to slight differences in the levels of collagen and the other biomarkers (albumin, ALT, and Alamar blue) among the different bioprinted tissues. For example, collagen levels were slightly and consistently higher in H2 compared with H1 and H-HUV (Figures 3, 4, 5, and 6 and Supplemental Figure S4), and it is possible that the different LSEC donors used for each tissue contributed to these results. Donor-to-donor variation and other variables that could affect the performance of the bioprinted tissues are currently under investigation. More importantly, however, inclusion of diseased hepatic nonparenchymal cells into the bioprints was sufficient to induce significantly higher levels of fibrosis, as measured by collagen production and deposition. Using multiple donors of diseased NPCs will be important for future work in characterizing the bioprinted tissue model. We also plan on using diseased hepatocytes as procurement, stability, and culture condition challenges with steatotic hepatocytes are improved.

The increased disease phenotype in bioprinted tissue from patients with NASH compared with subjects without disease suggests that factors inherent to the cells play an important role in disease. Such inherent factors can be maintained during careful isolation and handling of the primary cells from the livers of donors, and the cells can be recombined in the 3D model to achieve the NASH phenotype. This result means that it may be possible to identify the genes and epigenetic changes that drive NASH in patients. These genes would be expected to alter the disease phenotype when perturbed in the bioprinted 3D tissues. Perturbations in this model can be achieved with RNA interference, tool compounds, or overexpression systems, where modulation of a specific target would change the phenotype for better or worse. These disease driver genes represent important drug targets that could provide medications that are more efficacious clinically because they were identified and validated in a patient-derived 3D model of disease.

Disclosure Statement

J.N.M. and K.M. are founders, D.A.B. is a scientific advisor, P.K.T. and H.A. are employees, T.O. is a former employee, and S.L. is a former consultant of Viscient Biosciences, a company that supported this work.

Supplemental Data

Supplemental Figure S1.

Three-dimensional bioprinted liver tissue treated with palmitic acid. Top panels: Palmitic acid (PA) induces steatosis in bioprinted tissues. ExVive (H-HUV) tissues were untreated (control; left panel) or incubated with a 200 μmol/L (low; middle panel) or a 500 μmol/L (high; right panel) dose of palmitic acid for 21 days, and then subjected to oil red O staining to visualize lipid droplets. Bottom panel: The quantitation of the oil red O staining shows a statistically significant (one-way analysis of variance with Tukey multiple comparisons) dose-responsive steatosis phenotype to palmitic acid treatment. ∗P < 0.05, ∗∗P < 0.01, and ∗∗∗∗P < 0.0001. Scale bars: 50 μm (top panels, main images); 25 μm (insets of selected regions at higher magnification).

Supplemental Figure S2.

Hematoxylin and eosin staining of bioprinted tissues. Sections from day 14 bioprinted tissues are displayed, showing that all the bioprinted tissues had good integrity. Scale bars: 200 μm (main images); 100 μm (insets of selected regions at higher magnification). Refer to Table 1 for the identity of each bioprinted tissue.

Supplemental Figure S3.

Visualization of liver cell types in bioprinted tissues. Marker proteins [albumin in red for hepatocytes, α-smooth muscle actin (α-SMA) in red and desmin in green for hepatic stellate cells, CD31 in green for liver sinusoidal endothelial cells, and CD68 in green for Kupffer cells) in healthy H-HUV, H1, and diseased dESK1 bioprinted tissues were visualized by immunofluorescence. Nuclei were stained blue using DAPI. Bioprinted tissues were from day 14. The data show that each cell type remained present and viable in the tissues for both healthy and diseased cell types. Similar results were obtained for the other bioprinted tissue samples. Refer to Table 1 for the identity of each bioprinted tissue. Scale bars: 200 μm (main images); 100 μm (insets of selected regions at higher magnification).

Supplemental Figure S4.

Masson trichrome staining of bioprinted tissues. Sections from day 14 bioprinted tissues are displayed. Blue staining indicates areas of collagen deposition. Refer to Table 1 for the identity of each bioprinted tissue. Scale bars = 500 μm.

Supplemental Figure S5.

dS-HUV bioprinted tissues have higher levels of soluble fibronectin. Fibronectin was measured in the indicated bioprinted tissue media from the indicated days using an enzyme-linked immunosorbent assay–based assay. Refer to Table 1 for the identity of each bioprinted tissue. Asterisks indicate statistically significant differences for each day assessed by two-way analysis of variance with the Tukey multiple-comparisons test. Data are given as means ± SEM. ∗P < 0.05, ∗∗P < 0.01.

Supplemental Figure S6.

PanglaoDB enrichment of known cell types for each cluster. Bar charts displaying the −log10 (P value) of the overlap between cell type marker genes and marker genes for cluster 0 (A), cluster 1 (B), cluster 2 (C), cluster 3 (D), cluster 4 (E), cluster 5 (F), cluster 6 (G), cluster 7 (H), cluster 8 (I), cluster 9 (J), cluster 10 (K), cluster 11 (L), cluster 12 (M), and cluster 13 (N). For each cluster, the top five most highly enriched cell types are shown, with the −log (P value) of the hypergeometric test shown on the y axis.

Supplemental Figure S7.

Relative expression levels of hepatic stellate cell (HSC) marker genes in HSC clusters 0, 2, and 4, and fibroblast cluster 6.