Editor’s Highlight: Modeling Compound-Induced Fibrogenesis In Vitro Using Three-Dimensional Bioprinted Human Liver Tissues

Abstract

Compound-induced liver injury leading to fibrosis remains a challenge for the development of an Adverse Outcome Pathway useful for human risk assessment. Latency to detection and lack of early, systematically detectable biomarkers make it difficult to characterize the dynamic and complex intercellular interactions that occur during progressive liver injury. Here, we demonstrate the utility of bioprinted tissue constructs comprising primary hepatocytes, hepatic stellate cells, and endothelial cells to model methotrexate- and thioacetamide-induced liver injury leading to fibrosis. Repeated, low-concentration exposure to these compounds enabled the detection and differentiation of multiple modes of liver injury, including hepatocellular damage, and progressive fibrogenesis characterized by the deposition and accumulation of fibrillar collagens in patterns analogous to those described in clinical samples obtained from patients with fibrotic liver injury. Transient cytokine production and upregulation of fibrosis-associated genes ACTA2 and COL1A1 mimics hallmark features of a classic wound-healing response. A surge in proinflammatory cytokines (eg, IL-8, IL-1β) during the early culture time period is followed by concentration- and treatment-dependent alterations in immunomodulatory and chemotactic cytokines such as IL-13, IL-6, and MCP-1. These combined data provide strong proof-of-concept that 3D bioprinted liver tissues can recapitulate drug-, chemical-, and TGF-β1-induced fibrogenesis at the cellular, molecular, and histological levels and underscore the value of the model for further exploration of compound-specific fibrogenic responses. This novel system will enable a more comprehensive characterization of key attributes unique to fibrogenic agents during the onset and progression of liver injury as well as mechanistic insights, thus improving compound risk assessment.

Chronic liver injury progressing to fibrosis and liver failure can result from a wide range of insults including drug or chemical exposure, metabolic disease, alcoholism, or viral infection, and is a major health burden worldwide with 2% of all deaths attributable to liver cirrhosis (Lozano et al., 2012; Murray et al., 2012). Whereas the major precipitating factors underlying drug- and chemical-induced fibrosis have been gleaned from animal models, the key initiating and series of adaptive events that perpetuate this response, especially in humans, are still not well understood. Regardless of etiology, progressive fibrotic liver injury is orchestrated by complex intercellular interactions among hepatocytes (HCs), endothelial cells (ECs), hepatic stellate cells (HSCs), Kupffer cells (KCs), and recruited inflammatory cells (Bataller and Brenner, 2005).

Animal models of chronic liver injury with fibrosis partially recapitulate the human condition, but may fail to provide robust human translation due to species differences in metabolism, injury response, and capacity/mechanisms of repair and regeneration (Liu et al., 2013). Cell culture has been used as a surrogate to dissect the mechanistic details underlying HC dysfunction and fibrogenic outcome; however, conventional two-dimensional (2D), cell-based hepatic model systems do not reliably recapitulate liver structure, function, and its inherent multicellular architecture (LeCluyse et al., 2012). This is largely due to the absence of non-parenchymal cells (NPCs) relevant to liver injury and fibrogenic response. HSCs are recognized key effectors in the development and progression of hepatic fibrosis (Puche et al., 2013) however, they also help to define the molecular and structural microenvironment of the parenchymal compartment and space of Disse via the production of soluble and insoluble cues, including growth factors, inflammatory cytokines, and deposition of extracellular matrix (ECM) (Friedman, 2008). These microenvironments mediate requisite gene expression patterns for metabolic homeostasis, cellular differentiation, and maturation (Guillouzo et al., 1993; Rogiers and Vercruysse, 1993) and thus modulate the liver’s response to both acute and chronic injury. These observations suggest that current in vitro models used to evaluate potential fibrogenic agents lack fundamental cellular components that may moderate or exacerbate hepatocellular injury, an event strongly associated with the initiation of fibrogenesis (Canbay et al., 2004). Furthermore, the appearance and progression of basic fibrogenic features such as inflammation, tissue remodeling, collagen accumulation, and compensatory hepatocellular regeneration are best detected and interpreted in the context of a three-dimensional (3D) tissue environment. As such, these components are required in order to fully understand quantitative and temporal relationships underlying complex processes, such as fibrosis arising from chronic liver injury.

The recent availability of bioprinted human liver tissue models that incorporate both parenchymal (ie, HCs) and NPCs (ie, HSCs and ECs) in a 3D context has created the opportunity to examine progressive liver injury in response to known pro-fibrotic modulators and compounds (Nguyen et al., 2016). In this study, we utilized this novel model system to establish conditions for monitoring tissue responses after treatment with fibrogenic agents, including transforming growth factor-β1 (TGF-β1), and prototype fibrogenic compounds methotrexate (MTX), and thioacetamide (TAA). Significant concentration- and time-dependent elevations of lactate dehydrogenase (LDH) were observed for both MTX and TAA and were accompanied by the acquisition of a fibrogenic phenotype as supported by tissue remodeling, NPC migration/activation, histologic evidence of collagen deposition, transient elevations in proinflammatory, immunomodulatory, and chemotactic cytokines, and the expression of ACTA2 and COL1A1. In comparison, treatment with TGF-β1, a known pro-fibrogenic cytokine, yielded moderate fibrotic change in the tissue with little evidence of hepatocellular damage. Taken together, these data demonstrate the utility of novel 3D bioprinted tissues to further evaluate compound-induced liver fibrosis in a more defined and systematic fashion.

MATERIALS AND METHODS

Tissue production

Three-dimensional bioprinted liver tissues were manufactured by Organovo (San Diego, California) using primary cryopreserved human HCs (Life Technologies, Carlsbad, California), HSCs (ScienCell, Carlsbad, California), and human umbilical vein ECs (Becton Dickinson, Tewksbury, Massachusetts), using patented protocols (U.S. Patents 8,241,905 B2; 8,852,932; 9,222,932 B2; 9,222,932 B2) as described previously (Forgacs et al., 2012, 2014; Murphy et al., 2015; Nguyen et al., 2016; Shepherd et al., 2015). Each commercial cell supplier provides assurances that the cells come from tissues collected in compliance with applicable laws and provided based on informed consent by the donors. Briefly, HSCs and ECs were cultured prior to tissue fabrication and cryopreserved HCs were thawed and prepared for use according to the manufacturer’s instructions. Separate bio-inks comprising parenchymal cells (100% cellular paste, generated via compaction) or NPCs (150e⁶ cells/mL formulated in NovoGel 2.0 Hydrogel) were prepared and loaded into separate heads of the NovoGen Bioprinter platform (Organovo) housed within a standard biosafety cabinet (Forgacs et al., 2012, 2014; Jakab et al., 2008; Murphy, et al., 2015; Nguyen, et al., 2016; Shepherd, et al., 2015). An automated computer script was then executed to precisely deposit the bio-inks in a two-compartment planar geometry onto the membranes of standard 24-well 0.4 µm transwell membrane inserts (Corning, Tewksbury, Massachusetts) via continuous deposition, with NPCs comprising the border regions of each compartment and HCs filling each compartment such that the cell ratios roughly approximated physiologic ratios and the final tissue thickness was approximately 500 µm (Murphy and Atala, 2014). Following fabrication, the tissues were cultured in DMEM supplemented with Primary Hepatocyte Maintenance Supplements (Life Technologies, Carlsbad, California) and EGM-2 (Lonza, Basel, Switzerland) and maintained in a 37 °C incubator under humidified atmospheric conditions with 5% CO₂. Liver tissues were allowed to coalesce into tissue-like structures for a minimum of three days with the daily replacement of medium prior to treatment with compounds.

Compound exposure

Concentration ranges of MTX (Sigma-Aldrich, St. Louis, Missouri) and TAA (Sigma-Aldrich) were selected based on plasma C_max values of effective doses reported in clinical studies and animal models of fibrotic injury (Chilakapati et al., 2005; Shiozawa et al., 2005) and at an estimated sinusoidal concentration (Ferslew and Brouwer, 2014). For TAA, the plasma C_max values were further benchmarked against toxicity studies performed in vitro where TAA did not elicit LDH release or evidence of cellular necrosis at concentrations up to 50 mM in cultured primary rat HCs (Hajovsky et al., 2012) to select the final concentration range. TGF-β1 (Miltenyi Biotec Inc., San Diego, California), a well-recognized potent, pro-fibrogenic cytokine that directly stimulates collagen synthesis in HSCs (Leask and Abraham, 2004), was evaluated at an estimated physiologically relevant concentration (0.1 ng/mL) and at a concentration traditionally employed in in vitro model systems (10 ng/mL) as a positive control (Fogel-Petrovic et al., 2007). All dosing solutions were prepared immediately prior to addition to liver tissue constructs. Stock concentrations of MTX prepared in 100% dimethyl sulfoxide (DMSO; Sigma-Aldrich) were diluted in 3D Liver Tissue Medium (Organovo, San Diego, California) to final concentrations of 0.1 and 1.0 μM (final DMSO concentration, 0.1%). A 25 mM dosing solution of TAA was prepared directly in the culture medium and further diluted to prepare a 5.0 mM dosing solution. Lyophilized TGF-β1 was reconstituted in Corning USP/EP Certified Sterile WFI-Quality Water (Fisher Scientific, Pittsburgh, Pennsylvania) according to the product data sheet recommendations and added to the medium to prepare the 0.1 ng/mL and 10 ng/mL dosing solutions. To ensure vehicle consistency across treatment groups, 100% DMSO was spiked into the TAA and TGF-β1 dosing solutions and standard culture medium (vehicle control) such that the final DMSO concentration was 0.1%. Liver tissues were treated daily for either 7 or 14 days starting on the third day post-printing.

Lactate dehydrogenase assay

Spent medium samples collected on alternate treatment days were analyzed fresh for LDH activity using a commercially available colorimetric assay (Abcam, Cambridge, Massachusetts). The assay was performed per the manufacturer’s instructions using a CLARIOstar microplate reader (BMG Labtech, Germany) with minor modifications. Briefly, a half area polystyrene high-bind 96-well plate (Sigma-Aldrich) was employed, allowing the volumes of the kit reagents to be reduced by half. Samples were diluted to obtain readings within the linear range of the NADH standard curve corresponding to LDH activity between 1.0 and 100 mU/mL.

Albumin immunoassay

Spent medium samples from treatment days 1, 7, and 14 (ie, Tx1, Tx7, and Tx14) were analyzed for albumin content by a plate reader-based sandwich ELISA (Bethyl Laboratories, Montgomery, Texas) per the manufacturer’s instructions with minor modifications as described above. Samples were tested at different dilutions to obtain readings within the range of the standard curve generated from Human Reference Serum (1.6–1200 ng/mL; Bethyl Laboratories).

Cytokine measurements

Spent medium samples from select time points throughout the exposure period were aliquoted and stored at −80 °C until further analysis. The levels of cytokines released into the medium were assayed on the MESO QuickPlex SQ 120 Instrument using the Meso Scale Discovery (MSD) V-PLEX Human Biomarker kit (MSD, Rockville, Maryland), per the manufacturer’s instructions. MSD Discovery Workbench software (version 4.0) was used to generate a standard curve with a 4 parameter logistic fit and 1/y² weighing (R²>0.998). Cytokine concentrations in unknown medium samples were then interpolated from the standard curve. Cytokine heat maps exhibiting cytokine concentrations for each treatment relative to time-matched, vehicle-treated control were constructed using the JMP statistical software package Graph Builder (SAS Institute, Inc., Cary, North Carolina).

Histology

At the conclusion of the study, a subset of bioprinted liver tissues from each treatment group were formalin-fixed in a 2% paraformaldehyde solution (ie, 2% paraformaldehyde, 10 mM calcium chloride, 50 mM sucrose in phosphate buffered saline) for 24 h at 4 °C and transferred to 70% ethanol for 24 h. After processing and embedding tissues, blocks were sectioned at a 5.0 μm thickness using a rotary microtome (Jung Biocut 2035; Leica Microsystems, Buffalo Grove, Illinois). Sections were stained with Gill 3 Formulation Hematoxylin (Ricca Chemical Company, Arlington, Texas) and Eosin Y Solution, 1% Aqueous (Electron Microscopy Sciences, Hatfield, Pennsylvania). Additional slides were developed using Gomori’s One-Step trichrome (American MasterTech, Lodi, California) to evaluate collagen content. Slides were imaged using a Zeiss Axioskop microscope (Zeiss, Jena, Germany) and acquired with a Zeiss ICM-1 camera using Zen Pro software (blue edition).

RNA isolation and measurement of fibrosis-associated genes using quantitative real-time PCR

At the conclusion of the study, tissue lysates were prepared for each treatment group by homogenization in TRIzol Reagent (ThermoFisher Scientific, Waltham, Massachusetts). Total RNA was isolated by performing a phenol chloroform extraction/phase separation facilitated by Phase Lock Gel Heavy (5 PRIME, Inc., Gaithersburg, Maryland) and column purified using the RNeasy Plus Mini Kit (Qiagen, Valencia, California) according to the manufacturer’s protocol. RNA purity and yield was assessed using the NanoDrop 1000 version 3.5.2 (ThermoFisher Scientific, Wilmington, Delaware). RNA samples were reverse transcribed with the High Capacity RNA-to-cDNA Kit following the manufacturer’s instructions (ThermoFisher Scientific). Real-Time qRT-PCR was performed using TaqMan Universal PCR Master Mix (ThermoFisher Scientific) and manufacturer recommended “Best Coverage” TaqMan Gene Expression Assays (ThermoFisher Scientific) for hypoxanthine guanine phosphoribosyltransferase (HPRT; housekeeping), α-smooth muscle actin (ACTA2) and collagen, type 1, α1 (COL1A1), two genes known to be up-regulated during fibrogenesis. Triplicate reactions were carried out using a 7900HT Fast Real-Time PCR System with sample analysis performed using ABI PRISM Sequence Detection System software version 2.4 (ThermoFisher Scientific). Relative quantities (RQ) were calculated for each gene of interest by normalizing to HPRT and are represented as fold change relative to vehicle-treated control for each set of treatments (n = 2).

Immunostaining

Deparaffinized, formalin-fixed normal native and untreated bioprinted liver tissue sections harvested 60 h post-printing were subject to heat-mediated antigen retrieval in 1X citrate buffer solution, pH 6.0 (Diagnostic BioSystems, Pleasanton, California) and immunolabeled using primary antibodies against E-cadherin (ab1416 [1:100], Abcam) and vimentin (ab8978 [1:100], Abcam) to highlight the formation of cellular junctions and distribution of NPCs within the tissues. Additional tissue sections were incubated with primary antibodies against albumin (A6684 [1:500], Sigma), CD31 (ab76533 [1:250], Abcam), desmin (ab15200 [1:200], Abcam), and α-smooth muscle actin (α-SMA; ab7817 [1:200], Abcam) to demarcate the compartmentalized architecture of bioprinted liver and activation status of HSCs in a 3D context. A subset of bioprinted liver sections obtained from tissues exhibiting prominent features of fibrotic change with trichrome staining compared to vehicle-treated control tissues (ie, 1.0 µM MTX and 25 mM TAA) were immunolabeled for collagen I (ab34710 [1:500], Abcam), collagen IV (ab6586 [1:200], Abcam), vimentin, and α-SMA to examine the prevalence and distribution of collagen subtypes in the tissue constructs and migration/activation of NPCs throughout the tissue constructs. Vehicle-treated control tissue was used to assess tissue non-specific antibody staining within the tissue constructs. A secondary antibody control was also performed on successive bioprinted liver tissue sections as a procedural control (Supplementary Fig. 1). ImageJ software (Schneider et al., 2012) was used to quantitatively measure the percent area covered by immunoreactive collagens I and IV in three fields of view from representative tissue sections.

Statistical analysis

Unless otherwise noted, results are expressed as the mean of 4–5 replicates ± standard deviation (SD). Replicates refer to the same lot of bioprinted livers from a particular batch or print. The bioprinting process affords high reproducibility from each batch of bioprinted tissues. Statistical significance of treatment-induced differences relative to vehicle-treated control was determined using a two-way or one-way analysis of variance (ANOVA) where appropriate, with post hoc Dunnett’s multiple comparisons test using GraphPad Prism version 6.0 (Graph Pad Software, Inc., La Jolla, California). A P-value < .05 was considered statistically significant. Outliers were identified using Grubbs’ test to identify samples that fell one SD outside of the mean of the data (α = .05) using GraphPad Prism version 6.0 (Graph Pad Software).

RESULTS

Bioprinted Constructs Exhibit Key Features of Native Liver

Bioprinting is the automated fabrication of multicellular tissue that mimics the three-dimensional (3D) architecture and complexity of native tissue via the spatially defined deposition of cells in a proprietary bio-ink (Fig. 1A and B). Culturing cells in this 3D context facilitates the formation of parenchymal tissue architecture and polarization of epithelial cell membranes as evidenced by E-cadherin staining between parenchymal cells that resembles in vivo tissue density and localization (Fig. 1C and D). The punctate patterning and distribution of mesenchymal marker vimentin is preserved in 3D bioprinted liver and facilitates key heterotypic cell–cell interactions critical for supporting phenotypic features of uninjured liver (Fig. 1E and F), and sustained viability for at least 4 weeks post-printing (Nguyen, et al., 2016). Tissues comprising cryopreserved primary HCs, HSCs, and ECs were fabricated reproducibly on the membranes of standard 24-well culture inserts, thus enabling the use of this system to conduct routine in vitro toxicity testing (Fig. 2A). During the initial 3-day culture period, bioprinted cells coalesce and remodel to form a tissue-like construct with the retention of parenchymal (HC) and non-parenchymal (NPC) compartments. These compartments are illustrated with the hepatocellular marker albumin and the NPC markers CD31 and vimentin which stain ECs and HSCs, respectively (Fig. 2B). A proportion of vimentin positive HSCs were observed both within the NPC compartment and scattered throughout the parenchymal compartment making key heterotypic contacts with HCs.

FIG. 1.

3D bioprinted tissue recapitulates the tissue-like density and architecture of normal liver. (A) Transverse cross-sections of native human liver and (B) bioprinted human liver tissue stained with H&E. (C and D) Formation of hepatocellular junctions is shown with E-cadherin and the mesenchymal marker (E and F) vimentin is used to highlight distribution patterns within the parenchyma analogous to native liver. Scale bar = 50 µm.

FIG. 2.

3D bioprinted tissue exhibits a compartmentalized architecture and maintains hepatic stellate cells in a quiescent-like phenotype. (A) Illustration of a transverse cross-section of bioprinted tissue on a transwell insert comprising hepatocytes (HCs) and compartmentalized endothelial cells (ECs) and hepatic stellate cells (HSCs). (B) The organization of non-parenchymal cells (NPCs) is depicted with CD31 and vimentin staining to mark ECs and HSCs, respectively. Albumin is used to denote the hepatocellular compartment (HC). Scale bar = 100 µm, inset scale bar = 25 µm. (C) HSC activation status was examined using desmin (generic marker) and α-SMA (activation marker). Quiescent HSCs are denoted with white arrows. Scale bar = 50 µm.

Hepatic Stellate Cells Exhibit a Quiescent-like Phenotype in a Three-Dimensional Context

Hepatic stellate cells (HSCs) and their activation state within the center of the tissue construct were tracked in 3D culture by staining for desmin, an intermediate filament present in HSCs (Puche et al., 2013; Schmitt-Graff et al., 1991) and α-SMA, a marker of activated HSCs (Friedman, 2008). Prior to incorporation in 3D bioprinted tissues, HSCs were propagated through multiple population doublings and serial passages in 2D culture. HSCs typically reside in a quiescent state in uninjured liver but undergo activation in response to injury or 2D culture (ie, culture activation on collagen or plastic surfaces) as demonstrated by increased expression levels of activation markers, such as α-SMA (Carpino et al., 2005; Friedman, 2008). When cultured in a 3D context, HSCs embedded within the tissue architecture exhibited a more quiescent-like phenotype as illustrated by the retention of desmin and lack of α-SMA positivity (Fig. 2C, merge; white arrows). Whereas a majority of the HSCs embodied in the tissue are desmin(+), activated desmin(+)/α-SMA(+) HSCs were noted mainly at the periphery of the tissue (ie, the apical capsular region and basolateral edge of the tissue in contact with the culture medium and transwell membrane), consistent with a typical culture-activated phenotype (data not shown) and previously reported observations (Nguyen et al., 2016).

Effects of Fibrogenic Agents on Markers of Hepatocellular Injury and Function

Drug- or chemical-induced fibrosis is a complex and progressive process that usually occurs as a result of chronic exposure to low levels of compounds. To evaluate the culture model as a platform for studying drug- and chemical-induced fibrosis, bioprinted tissues were exposed to known fibrogenic agents, MTX and TAA, for up to 14 days. MTX is a folate antagonist effectively used to manage inflammatory disorders (ie, rheumatoid arthritis, psoriasis) at low doses for extended periods of time. However, it is known to cause elevations in alanine aminotransferase (ALT) and fibrosis in a subset of patients over prolonged treatment periods (Lindsay et al., 2009; Maybury et al., 2014). TAA is a prototypical fibrogenic agent extensively used in rodent models to study the development of liver injury and fibrosis (Starkel and Leclercq, 2011).

During the course of the study, the tissues remained intact macroscopically, with a marked dose- and treatment-dependent reduction in tissue size noted for 1.0 µM MTX and TAA-treated groups by treatment day 14 (Tx14; Fig. 3A). LDH release was measured in the culture medium to assess the impact of repeated exposure on tissue viability. During the initial 7-day exposure period, LDH release for MTX- and TGF-β1-treated groups remained consistent with vehicle control levels (Fig. 3B). Within the same time period, elevations in LDH were observed for 25 mM TAA beginning at Tx3 (Fig. 3C). At later treatment time points (>Tx9), both 0.1 µM and 1.0 µM MTX-treated groups exhibited a time-dependent 2- to 3-fold increase in LDH release relative to vehicle control (****P < .0001). TAA exhibited similar trends at the lower concentration beginning on Tx5; however, LDH release measured from the 25 mM TAA-treated group exhibited a monophasic increase that peaked around Tx5 (****P < .0001) and then declined again by Tx11. These results were further supported by parallel trends in ALT release at similar time points during the exposure period (Supplementary Figure 2). By Tx4, elevations in ALT were noted for the 1.0 µM MTX- and 5.0 mM TAA-treated groups and sustained for the remainder of the treatment time course. A similar monophasic increase in ALT release was noted for 25 mM TAA, peaking around Tx4.

FIG. 3.

Impact of fibrotic agents on biochemical markers of liver tissue viability and functionality. (A) Gross images of tissues following 14 days of treatment. Scale bar = 2.5 mm. (B and C) LDH release during an extended 14-day treatment with MTX, TAA, and TGF-β1 (n = 9 for Tx1–Tx7, n = 5 for Tx9–Tx14). (D) Albumin production as a measure of hepatocellular function is depicted at key time points during the treatment period (n = 5). Significance was determined using a one-way ANOVA with post hoc Dunnett’s multiple comparisons test (*P < .05, **P < .01, ***P < .001, ****P < .0001).

Albumin output was measured at time points defined by the LDH results (ie, prior to elevations in LDH, mid-way through the treatment period, and time points at which statistically significant elevations in LDH were observed) during the exposure period as a measure of hepatocellular function (Fig. 3D). Albumin output (ng/mL/million cells) for most of the treatment groups remained within vehicle-treated control levels during the treatment time course with the exception of the 0.1 ng/mL and 10 ng/mL TGF-β1 (Tx14 increased; *P < 0.05 and *****P < 0.0001, respectively) and the 25 mM TAA (Tx7 and Tx14 decreased; ****P < .0001) treatment groups. The measured depreciation in albumin output for 25 mM TAA at Tx7 and Tx14 complements the LDH and ALT results (Fig. 3C and Supplementary Fig. 2B, respectively) further suggesting a perturbation in tissue function as a result of hepatocellular injury. Because TAA requires CYP2E1-mediated bioactivation to elicit hepatotoxicity, the expression of CYP2E1 was verified in untreated bioprinted tissues spanning the timeframe of exposure used in the current studies (Supplementary Fig. 3).

Evidence of Collagen Deposition in Tissues Treated With Fibrogenic Agents

For complex disease processes such as fibrosis that lack early and informative biomarkers, histological assessment remains the gold standard for detecting and evaluating the progression of fibrotic injury (Barker et al., 2011; Sebastiani and Alberti, 2006). In order to assess the effects of compound treatment, a subset of tissues from each treatment group were examined histologically. Transverse cross-sections of formalin-fixed, paraffin embedded tissues were stained with hematoxylin and eosin (H&E) to evaluate overall cell and tissue morphology (Fig. 4). Distinct compartments were evident within the tissue constructs with delineation of HC and NPC compartments (Fig. 4B). Treatment with 1.0 µM MTX (Fig. 4G) and TAA at both concentrations (Figs. 4D and E) resulted in a compact and rounded mass of cells compared to the vehicle-treated group (ie, approximate 50% reduction in tissue size) suggesting possible cell death and tissue degeneration (which was also supported by the biochemical data), ECM deposition, and enhanced contraction of tissue architecture. Invasion of NPCs into the HC compartment was observed for all tissues exposed to fibrogenic agents. For both TGF-β1-treated groups (Fig. 4B and C), the basal surface of the tissue in contact with the transwell membrane exhibited concentration-dependent differences in the thickness of scar-like tissue (white arrows). This phenomenon was also observed for the MTX- (Fig. 4F and G) and TAA-treated (Fig. 4D and E) groups with a more extensive replacement of the tissue with fibrous-scarring.

FIG. 4.

H&E and trichrome staining reveals key features consistent with clinical fibrosis in bioprinted tissues following 14 days of treatment with select fibrogenic agents. Representative sections of bioprinted liver treated with (A) 0.1% DMSO vehicle, (B) 0.1 and (C) 10 ng/mL TGF-β1, (F) 0.1 and (G) 1.0 µM MTX, and (D) 5.0 and (E) 25 mM TAA. (B and C) A circle is used to delineate the non-parenchymal (NPC) from the parenchymal (HC) compartments and white arrows denote the basolateral edge of the tissue in contact with the transwell membrane. Collagen deposition was visualized (blue) in successive sections of bioprinted tissue stained with Gomori’s trichrome. Entrapped hepatocytes (EH), nodular areas of collagen deposition (NF), pericellular fibrosis (PF; F and G inset, 150% enlarged), (G) yellow arrows denote bridging fibrosis. Scale bar = 100 µm. (F) Expression of fibrosis-associated genes at Tx7 and Tx14 in MTX-treated tissue.

Gomori’s trichrome stain (Fig. 4) revealed stark differences between vehicle-, TGF-β1-, MTX-, and TAA-treated tissues. In the vehicle-treated control, there was faint evidence of collagen deposition (blue) throughout the construct (Fig. 4A). TGF-β1 treatment caused diffuse areas of collagen deposition localized to the NPC compartment (Fig. 4B; outlined), a dose-dependent thickening of the basolateral edge of the tissue (Fig. 4B and C; indicated with white arrows), and a generally preserved HC mass, a histological outcome mirrored in the LDH data (Fig. 3). MTX (0.1 µM) caused mild hepatocellular damage and evidence of pericellular fibrosis (Fig. 4F; PF; inset 150% enlarged) that progressed with 1.0 µM MTX treatment to include areas of nodular fibrosis (NF) and a compacted mass of cells separated by fibrotic septae bisecting the hepatocellular compartment (Fig. 4G; yellow arrows). Treatment with TAA diminished the parenchymal compartment in the tissues in a concentration-dependent manner by 14 days, with entrapment of remaining HCs (Fig. 4D and E; EH) and replacement of the parenchymal compartment with extensive whorls of scar-like tissue. Preliminary analysis of two fibrosis-associated genes, ACTA2 and COL1A1 in MTX-treated tissues exhibited a 2- to 3-fold increase in expression relative to vehicle-treated control by Tx14 concordant with histological findings (Fig. 4H).

Cross-sections of tissues from treatment groups that exhibited pronounced evidence of collagen deposition as evidenced by Gomori’s trichrome staining were further examined immunohistochemically for the prevalence of collagens I and IV and the distribution and activation of HSCs within the tissue using vimentin and α-SMA (Fig. 5). Relative to vehicle-treated control, collagen production (Fig. 5A and B) was visually upregulated in tissues treated with 1.0 µM MTX and 25 mM TAA with both collagen subtypes most prevalent in the 25 mM TAA-treated tissue. The average percent area of collagen in the 25 mM TAA-treated group compared to vehicle-treated control was 25.784% versus 7.529% for collagen I and 29.832% versus 2.121% for collagen IV (Fig. 5A vs B), respectively.

FIG. 5.

Increased deposition of collagens I and IV and expression of vimentin and α-SMA in tissues exhibiting pronounced fibrogenic change. Tissues treated with 0.1% DMSO vehicle, 1.0 µM MTX, and 25 mM TAA were further assessed immunohistochemically for (A) collagen I, (B) collagen IV, (C) vimentin, and (D) α-SMA. White and yellow arrows denote the apical and basolateral edges of the tissue, respectively. Areas of fibrillar ECM deposition are outlined in successive collagen I and collagen IV stained sections. Nodular areas of collagen deposition (NF). The percent area covered by collagens I and IV is depicted in the bottom left-hand corner of the photomicrographs. (C) Punctate areas of vimentin positivity in control tissue (white arrows) and diffuse patterning in treated tissue (yellow arrows). The black and white inset accentuates the shift in vimentin patterning observed with treatment. (D) α-SMA(+) HSCs were mainly noted at the periphery of the tissue in the vehicle-treated control (white arrows). Increased SMA(+) HSCs in the center of treated tissues and altered distribution of the cells corresponding to areas of collagen deposition (white arrows). Scale bar = 25 µm.

Whereas collagen I and collagen IV exhibit similar patterns at first glance, closer examination shows distinguishing patterns in the localization of greatest staining intensity. Collagen I, positive areas were localized to the apical face of the tissue (ie, capsular region) of the tissue (Fig. 5A; white arrow) and at the basolateral edge of the tissue in contact with the transwell membrane (Fig. 5A; yellow arrow). In addition, collagen I staining was prominent in the septae traversing the parenchymal compartment, thicker ECM fibers present within the tissue constructs (outlined), and in areas of nodular fibrosis (NF) concordant with collagen positive areas in Gomori’s trichrome-stained sections (blue staining; Fig. 4). In comparison, collagen IV positive areas were mainly localized to the immediate periphery of cells in the HC compartment with little to no staining of fibrillar collagen (Fig. 5B; corresponding areas of collagen I- versus collagen IV-stained sections are outlined). Similar to collagen I, collagen IV was also prominent in nodular areas of collagen deposition, particularly in the NPC compartment.

The distribution and patterning of vimentin in the tissue constructs mirrors the results obtained from collagen immunohistochemistry (IHC) and varies with treatment and the degree of fibrotic injury (Fig. 5C). In the vehicle-treated group, vimentin positivity appears as small punctate spindle shaped areas with the greatest prevalence in the non-parenchymal compartment and even distribution throughout the parenchymal compartment (white arrows). Following 14 days of treatment, vimentin transitions to a more extensive and diffuse patterning throughout the tissue constructs (series of yellow arrows) compared to vehicle-treated control, particularly in areas corresponding to nodular areas of collagen deposition (NF). In addition to measuring the distribution of NPCs within the tissue constructs, the activation of HSCs in bioprinted liver was also assessed by staining for α-SMA (Fig. 5D). Activated α-SMA(+) HSCs were mainly noted at the periphery of the tissue in the vehicle-treated control (white arrow) with the minimal activation of HSCs within the center of the tissue construct. Treatment with fibrogenic agents resulted in an increase in SMA(+) HSCs in the center of the tissue (white arrows) and altered distribution of the cells corresponding to areas of collagen deposition. Overall, these histological features were consistent with the upregulation of ACTA2 and COL1A1 relative to vehicle-treated controls at Tx7 and Tx14 (Fig. 4H and Supplementary Table 1).

Cytokine Profiles Are Indicative of a Fibrogenic State

Because inflammation is closely tied to fibrogenesis (Pellicoro et al., 2014), the abundance of pro- and anti-inflammatory cytokines (pg/mL) released into the culture medium from treated tissues throughout the exposure period (ie, alternate treatment days) was measured. A general decrease in cytokines starting at Tx1 is apparent for all treatment groups including vehicle-treated control (Supplementary Fig. 4). By Tx7, the initial spike in cytokine production subsides (Supplementary Fig. 4) and treatment-dependent differences become perceptible.

In order to assess treatment-dependent effects over time, the fold change in cytokine levels relative to time-matched vehicle-treated control was determined (Fig. 6). A subset of cytokines depicted in Figure 6A changed consistently across replicates and illustrate treatment- and concentration-dependent effects over the course of the 14-day exposure period. Other cytokines detected from the cytokine panel exhibited similar trends although they were admittedly variable. During the initial treatment period (Tx1 and Tx3), deviations in the prevalence of specific cytokines from vehicle-treated control are not readily apparent with the exception of IL-13, an important inflammatory mediator, in treated tissues (Fig. 6A; decreased). Starting at Tx7, elevations in proinflammatory IL-6 were evident for 1.0 µM MTX and both TAA-treated groups with slight elevations observed for 0.1 µM MTX starting at Tx9. IL-6 regulates acute phase response proteins in response to injury (Choi et al., 1994) and in part, coincides with the biochemical data for MTX and TAA (Fig. 3 and Supplementary Figure 2). Furthermore, a more general decrease in cytokines observed starting at Tx7 with 25 mM TAA treatment, was concordant with the timeframe of LDH and ALT release (Fig. 3C and Supplementary Fig. 2B), and decreased albumin production at Tx7 and Tx14 (Fig. 3D) suggesting tissue damage and perturbation of hepatocellular function. In comparison, the trends in cytokines observed with MTX and TAA treatment were not evident for TGF-β1-treated tissues. This difference in profiles is not entirely unexpected, considering the apparent absence of hepatocellular damage seen with TGF-β1 treatment (Fig. 3) and the different mechanisms of action of cytokine- versus xenobiotic-induced liver fibrosis.

FIG. 6.

Subset of cytokines exhibiting treatment-dependent differences over time and at select treatment time points. (A) The upregulation (red) and downregulation (green) of proinflammatory, immunoregulatory, and chemotactic cytokines relative to vehicle-treated control was represented in a heat map. (B) Samples collected at mid (Tx7) and late (Tx14) treatment time points were profiled for additional cytokines and chemokines. Values outside the range of the standard curve or excluded via Grubb’s outlier analysis are shaded grey.

After the initial assessment of selected cytokines, additional cytokine profiling was performed at Tx7 and Tx14 (Fig. 6B; cytokines exhibiting consistent changes across replicates are depicted). Measurement of cytokine levels at these time points showed treatment- and time-dependent differences in acute phase response, immunomodulatory, angiogenic, and chemotactic cytokines. A Log2(Fold Change) of 2 or −2 was considered statistically significant. IL-6 was significantly increased at Tx7 and Tx14 for 1.0 µM MTX and both TAA treatment groups consistent with initial temporal observations from Figure 6A. Fms-related tyrosine kinase-1 (Flt-1), involved in cell proliferation, differentiation, and monocyte activation/recruitment (Motomura et al., 2005), is statistically increased at Tx7 for 25 mM TAA-treated tissues and then returns to vehicle-treated levels by Tx14. Monocyte chemotactic protein-1 (MCP-1), involved in facilitating macrophage/monocyte infiltration to perpetuate an adaptive response to continued insult (Baeck et al., 2012), increases at Tx7 for 1.0 µM MTX and 5.0 mM TAA treatment and continues to increase by Tx14 with the exception of 25 mM TAA. Interestingly, the abundance of eotaxin, a mediator of inflammatory cell infiltration and recruitment, was significantly increased in the culture medium of tissues treated with 10 ng/mL TGF-β1 suggesting a possible direct-acting stimulation of eotaxin expression in the absence of overt hepatocellular injury (Matsukura et al., 2010).

DISCUSSION

Recently, liver fibrosis secondary to compound-induced liver injury has become an interest to the Joint Research Centre (JRC) and other regulatory organizations focused on adverse outcomes (Ankley et al., 2010). Because fibrosis develops over time from a sequence of complex and cumulative interactions between HCs and NPCs, it has proven challenging to model using standard in vitro and preclinical in vivo models. The development of an effective Adverse Outcome Pathway framework depends on the employment of models that overcome these translational challenges and provide a test bed that is multicellular, compatible with chronic exposure testing regimens, and able to reveal a full spectrum of relevant outcomes from initiation through progression, including biochemical, genomic, and histologic endpoints (Van de Bovenkamp et al., 2007). Here we evaluated the potential of a novel bioprinted in vitro tissue model of human liver to model compound-induced fibrosis. This approach represents a significant innovation in the study of progressive liver injury and in vitro toxicity testing, as it addresses many of the shortcomings associated with traditional models.

The constitutive activation of HSC monocultures has been a significant barrier in the in vitro assessment of potential fibrogenic agents, as compound-related effects are confounded by the culture-activated HSC phenotype. Whereas there have been some recent advances in the study of hepatic fibrosis in vitro using precision cut liver slices (PCLS) or spheroids (Leite et al., 2016; Thiele et al., 2015; Van de Bovenkamp et al., 2007; Westra et al., 2016), there still exist limitations in the application of these platforms to understand the progression of events underlying fibrogenesis. PCLS have a short life span ex vivo (generally <1 week) and develop early onset fibrogenic changes, irrespective of treatment, which may confound the interpretation of a causal relationship after compound exposure (Westra et al., 2016). As such, chronic exposure studies aimed at modeling progressive fibrogenic features over weeks to months have not been feasible with PCLS. Finally, whereas the fixed configuration of the cells comprising each slice preserves the normal tissue architecture initially, the ability to tease apart the roles of the different cell types is severely limited. Bioprinting is an efficient and reproducible means of establishing key architectural relationships between cells and preserving tissue-level functions over prolonged periods of time (Nguyen et al., 2016). The 3D nature and substantial biomass of the model enable histological assessment of treated liver tissues, which remains the diagnostic gold standard for the accurate detection and staging of fibrosis (Sebastiani and Alberti, 2006). The unique compartmentalized architecture of bioprinted tissues compared to other 3D models, such as spheroids (Leite et al., 2016), facilitates the temporal assessment of progression by revealing specific patterns of collagen deposition that are analogous to patterns described in human biopsy samples.

Importantly, the incorporation of HSCs into 3D bioprinted tissue re-establishes a quiescent-like phenotype and thus uniquely enables the model to be used in the assessment of compound effects on early fibrogenic processes—something that has not been feasible to date using conventional approaches. This phenomenon is consistent with the outcome of elegant fate-mapping studies in mice that demonstrated a subset of HSCs are able to revert to a quiescent-like phenotype (Kisseleva et al., 2012) during the resolution of fibrotic injury. HSCs that have been previously activated exhibit a primed phenotype with rapid and robust patterns of reactivation in response to subsequent injury (Kisseleva et al., 2012; Taghdouini et al., 2015). This observation could explain the accelerated fibrogenic features observed in the current study, compared to the clinical setting in which fibrosis can take months or even years to develop. Regardless of whether the desmin(+)/α-SMA(−) HSCs represent quiescent or inactivated HSCs, the results presented herein demonstrate their clear capacity to mount a measurable and progressive response to fibrogenic insults.

Following 14 days of exposure, the extent of LDH release and time to peak release was compound-, concentration-, and time-dependent. The monophasic increase in LDH release and subsequent return to vehicle-treated levels for 25 mM TAA most likely reflects the outcome of prior hepatocellular damage, as supported by the corresponding loss of albumin production and decrease in ALT release. In contrast, TGF-β1 treatment did not elicit elevations in LDH release during the entire exposure period. TGF-β1 is a well-established pro-fibrogenic mediator, directly triggering the activation of HSCs and synthesis of ECM (Leask and Abraham, 2004). Albumin production was not significantly perturbed with the exception of the TGF-β1 (increased production) and the 25 mM TAA (decreased production) treatment groups. Whereas previous studies have reported TGF-β1 inhibits albumin RNA and protein synthesis in primary HCs, it should be noted that these studies were conducted in HC monocultures (Busso et al., 1990). We hypothesize the lack of concordance may reflect TGF-β1-induced secretion of ECM proteins and other factors by the NPC compartment that further support HC function in the absence of overt injury. Taken together, these data demonstrate that the exposure conditions described herein are able to produce the mild/moderate HC injury associated with low-concentration, chronic exposure, which sets the stage for the development and detection of more complex adverse outcomes such as fibrosis.

Histological assessment of treated tissues revealed the initiation and progression of fibrogenic processes in response to insult. The degree of collagen deposition in treated tissues correlated with biochemical evidence of hepatocellular damage, with 1.0 µM MTX and 5.0 and 25 mM TAA-treated tissues exhibiting a disrupted architecture with prevalent collagen deposits that progressed to a scar-like matrix and displaced the HC compartment over time. Interestingly, patterns of collagen deposition and tissue injury observed in the MTX- and TAA-treated groups were analogous to those reported in clinical biopsy samples of MTX-induced fibrosis and preclinical animal models of TAA exposure (Müller et al., 1988; Osuga et al., 2015), which suggests the model may serve as a translational tool for mechanistic and interventional studies involving fibrogenic agents and modulators. Changes in ACTA2 and COL1A1 expression, further confirmed the progression- and concentration-dependent nature of the fibrogenic response, particularly in MTX-treated tissues. The treatment-induced mobilization of NPCs and activation of HSCs within the tissue were consistent with findings from published clinical studies of progressive fibrosis (Attallah et al., 2007; Veidal et al., 2011). Whereas future studies will expand the genomic, proteomic, and histologic characteristics of the model during progressive fibrotic injury, these initial observations provide encouraging evidence that the model has translational utility.

Interestingly, compound-dependent fibrogenic responses were elicited in the absence of liver resident macrophages. While inflammation typically precedes or accompanies liver fibrosis and is recognized as a driver of fibrogenesis (Czaja, 2014; Pellicoro et al., 2014), the precise role of KCs in mediating this process remains elusive. This is largely due to the inability to target specific macrophage subpopulations (Tacke and Zimmermann, 2014), a question which can be addressed using this model. We anticipate that KCs will play an important tolerogenic role in attenuating the tissue response during early exposure to pro-fibrogenic agents (Ju and Pohl, 2005). We sought to validate the established 3D liver model, which consists of two key NPC constituents, namely, ECs and HSCs, but also is amenable to modifications in cellular composition. These studies lay the foundation for the future assessment of the role of KCs and other cell types relevant to the response (ie, sinusoidal endothelial cells) in exacerbating or remediating tissue injury and their impact on fibrogenic outcome.

Cytokine profiles differed between early and late time points of exposure to fibrogenic agents, which likely reflects the modulatory role(s) of specific cytokines that mark liver injury and influence fibrogenic outcomes in response to insult. The induction of proinflammatory cytokines during early exposure (Tx1–Tx7) mimics some features of the classic wound-healing response (Pellicoro et al., 2014), with a surge in the production of proinflammatory cytokines that drive tissue remodeling and regeneration and may aid in the formation of a cohesive tissue-like mass after bioprinting. It is likely that both the HSC and EC components of the bioprinted tissues contributed to the elevations of IL-6 and IL-8, which enhance EC survival, proliferation, angiogenesis, and recruitment of inflammatory cells during wound healing (Qazi et al., 2011). IL-1 is also rapidly released in response to tissue damage (Gieling et al., 2009) and could explain the transient increase in IL-1β early in the exposure period.

When cytokines approached steady-state levels in vehicle-treated tissues (10 days post-printing) treatment-dependent spikes in proinflammatory cytokines were detected. Increased chemotactic cytokines, such as MCP-1, at later time points for some treatment groups suggest an adaptive shift in response to persistent tissue stress/injury that results in recruitment of inflammatory cells to the site of damage (Baeck et al., 2012). Observed differences in the cytokine profiles in TGF-β1- versus compound-treated tissues is likely due to mechanistic differences between the TGF-β1 response (little to no hepatocellular injury) and compound-induced responses to hepatocellular injury and/or the transient nature of cytokine profiles. The global decline in cytokine production observed at 25 mM TAA by Tx14 was consistent with the decline in viability and functionality for that time point. Significant differences in cytokine levels among treatment groups were not noted during early exposure, likely due to their masking by the observed wound-healing response post-fabrication. Furthermore, cellular interactions and cross-talk that occur during tissue formation could influence susceptibility or magnitude of response to particular insults. Future studies will examine alterations in response profiles in maturing (3–5 days old) versus matured (7–10 days old) tissues. Nonetheless, the cytokine data in conjunction with the gene expression and histological data supports the hypothesis that these tissues are actively engaging in fibrogenic processes in response to compound-induced injury.

In summary, the outcomes from these studies support continued development of 3D bioprinted human tissues as in vitro surrogates for studying compound-induced liver fibrosis. Future studies will provide new insights into early initiating and adaptive events underlying fibrogenic responses, help identify both common and distinct pathways of compound-induced effects, and improve compound risk assessment. While there exist a number of challenges towards developing effective treatment strategies (ie, causation, stage of fibrotic injury, co-morbidities), these studies bridge a critical gap that could inform effective therapeutic approaches (ie, novel biomarkers and interventional strategies) for treatment at early and late stages of fibrogenesis during which different hepatic cell types may be involved and targeted to prevent or reverse liver fibrosis.

SUPPLEMENTARY DATA

Supplementary data are available online at http://toxsci.oxfordjournals.org/.

FUNDING

This work was supported by the National Institutes of Health (NIH), Initiative for Maximizing Student Diversity (IMSD) at University of North Carolina at Chapel Hill [R25 GM055336 to L.M.N.], and National Institute of Environmental Health Sciences (NIEHS) Toxicology Training Grant [T32 ES007126 to L.M.N.], and Organovo, Inc. The content is the authors’ own and does not reflect the views of the NIH. The research described in this article has been reviewed by Organovo, Inc., San Diego, and approved for publication. Approval does not signify that the contents necessarily reflect the views and policies of the Company, nor does the mention of trade names of commercial products constitute endorsement or recommendation for use.