Modelling human liver fibrosis in the context of non-alcoholic steatohepatitis using a microphysiological system

Measurement of fibrosis in liver microtissues loaded with fat

MPS models involving primary human hepatocytes exposed to FFA in culture media, containing physiological levels of glucose and insulin, have been shown to accumulate three times more fat compared to hepatocytes not exposed to FFA whilst maintaining their viability and functionality over a 2-week period¹⁸. In this way, features of NAFL are recapitulated and herein we refer to these microtissues as steatotic. But in vivo the liver is a multicellular organ and crucially the presence of Kupffer cells and hepatic stellate cells (HSC) have been shown to play a pivotal role in the progression of NAFL into NASH^2,21,28,29. Indeed, one of the distinguishing histological features of NASH is the presence of fibrosis owing primarily to the activation of HSC, which causes increased collagen-1 deposition. However, in published MPS studies collagen-1 has not been directly measured, instead expression of surrogate markers such as secreted pro-collagen 1, TGFβ and osteopontin are reported^19,20, in addition to gene expression changes^19,20. Here, we have developed a high content, automated imaging application that enables quantification of collagen-1 deposition and spatial localisation of α-SMA expression within HSC, an indicator that quiescent HSC has dedifferentiated into motile myofibroblasts (Supp. Fig. 1). Compared to control and steatotic microtissues (PHH with/without FFA exposure) there was a sixfold increase in α-SMA expression and an eightfold increase in collagen-1 deposition in NASH microtissues (Fig. 1a, b). These changes were corroborated by measurement of secreted clinical fibrotic markers TIMP-1, pro-collagen 1, YKL-40 and fibronectin, which were all significantly increased in NASH compared to steatotic and control microtissues (Fig. 1c) and the increased expression of genes associated with fibrosis in the NASH microtissues (Fig. 1d and Supp. Table 1). During the progression of NAFL to NASH, there is an increase in production of pro-inflammatory cytokines, most likely initiated by activated KC. As previously reported our NASH microtissues produced significantly higher levels of a wide variety of inflammatory cytokines associated with NASH including IL-1β, IL-6, TNFα, MCP-1 and MIP-1β (Supp. Fig. 2a), without affecting the cell health assessed by tissue density of the microtissues and albumin production (Supp. Fig. 2b, c). Fat loading in the NASH microtissues was confirmed using Oil Red O staining (Supp. Fig. 2d, e). We additionally established that the phenotypic changes observed in the NASH microtissues were not affected by the biological donors used, particularly as non-donor matched material was always used to generate NASH microtissues (Supp. Fig 3). The same pattern of expression was observed, with some variation in absolute values, for a wide variety of cytokines and biomarkers when comparing NASH microtissues produced from three separate donors of PHH, KC and HSC (Supp. Fig 3).

PHH alone (Control and Steatosis) or PHH, KC and HSC co-cultures (NASH) were cultured in the MPS platform for 14 days under standard media (control) or high fat conditions (Steatosis and NASH). a Liver microtissues were stained for cytoskeleton (phalloidin), collagen-type I and α-SMA and imaged by confocal microscopy. Representative images shown and scale bars 200 µm. b Staining of microtissue was quantified by measuring total fluorescence intensity throughout individual microtissues, each data point represents an average of all microtissues within a FOV (min 8, max 12) and two FOV per scaffold. c Secreted fibrosis biomarkers TIMP-1, Pro-collagen 1, YKL-40 and Fibronectin were all measured in cell culture medium by ELISA, at the end of the culture. d The expression of fibrosis genes was analysed in total RNA from steatosis and NASH models using Human Fibrosis RT2 Profiler PCR Arrays. Gene expression levels are expressed as Log10 relative expression compared to housekeeping genes (GAPDH/B2M/HPRT1) and compared between the NASH and steatosis models (red = upregulation, green = downregulation, black = no change, upper line—1.5-fold increase, lower line 1.5-fold decrease). All data points shown with means ± SD highlighted, data generated from a minimum of nine independent cultures (three donors per condition and n = 3 per donor).

Transcriptomic signature of NASH microtissues aligns with human data and is significantly different to that from commonly used rodent models of the disease

The molecular events that drive NASH are poorly understood, but the transcriptomic signatures derived from NASH livers offer an opportunity to identify molecular pathways and their contribution to NASH pathogenesis. Using RNAseq, we conducted transcriptomic profiling of our MPS model and transcriptomic profiles of control lean microtissues mapped to non-diseased human liver profile (Supp. Fig. 4a). The control MPS retained expression of PHH, HSC and KC markers (Supp. Fig. 4b–d) providing confidence that the cells in the microtissues together with the microenvironment created by the MPS device combine to give a relevant model system to study disease. Comparing the profile of NASH microtissues (PHH, KC, HSC with FFA) with the Human Gene Atlas database (www.ebi.ac.uk/gxa/home) identified a human liver profile (Fig. 2a) and comparison to the DISEASES database (https://diseases.jensenlab.org/Search) produced a close correlation with human fatty liver disease (Fig. 2b). These results provide confidence that the microtissues are exhibiting good similarity to the disease.

PHH, KC and HSC co-cultures were cultured in the MPS platform under high fat conditions for 14 days and total RNA extracted and analysed by RNAseq to determine overall transcriptomic profile. Transcriptional profile of liver NASH MPS compared to the a Human gene atlas and b the DISEASES database using the Enricher gene enrichment analysis tool (maayanlab.cloud/Enrichr/)⁶⁸. 300 highest expressed DEGs in the NASH model were compared to control PHH samples, ranking by p value (probability of genes being associated with the gene set. c Breakdown of Top 100 genes expressed in liver MPS NASH model. d, e Published studies identify key differentially expressed genes specifically in NAFLD/NASH patient samples. d Advanced vs mild fibrosis in NAFLD³⁰, e NASH vs NAFLD vs obese control¹⁷. Heatmaps demonstrate Log2 fold change in transcripts for these genes in transcriptome from liver MPS NASH model compared to control PHH samples. f Comparison of correctly overlapping DEGs in NASH pre-clinical models, murine models from literature¹⁷ and NASH MPS model (data from e). Transcriptomic data were generated from nine independent replicate samples and averaged.

The breakdown of the top 100 genes expressed in the NASH microtissues (Supp. Table 2) indicated that the vast majority (77%) of genes were associated with hepatic function, metabolism, inflammation, and fibrosis (Fig. 2c). To further demonstrate the translational relevance of the MPS NASH model we compared the transcriptional profile to gene lists from literature^17,30 that are differentially expressed in human NASH patients (Fig. 2d, e). For the dataset from Moylan et al.³⁰, of all the differentially expressed genes (DEGS) from NASH patients, 75% were also differentially expressed in the NASH MPS microtissues (Fig. 2d). In the study by Teufel et al.¹⁷, 193 genes were identified as representing key changes in human NAFLD/NASH and these genes were then compared to nine standard mouse pre-clinical NASH models. The study demonstrated between 0.01% and 10% of the DEGs in NASH patients were also differentially expressed in the mouse models indicating their low human relevance. Comparison to the transcriptomic signature of our NASH microtissues indicated a far greater relevance to the patient samples with up to 45% of the DEGs being correctly differentially expressed (Fig. 2e, f).

Reversibility of disease phenotype and pharmacological modulation in NASH liver microtissues loaded with fat

Despite many on-going clinical trials, there is still no approved treatment for NASH, but emerging data provide confidence that NASH is a pharmacologically responsive disease. As such, the drug discovery process would significantly benefit from easy-to-use, human-relevant models. Having demonstrated that our NASH MPS microtissues are representative of the human disease, we next assessed its pharmacological responsiveness to two agents in late-stage clinical development, OCA and ELF. First, we set out to determine an optimal dosing strategy for each of the treatments based on an assessment of their physicochemical properties and pharmacokinetic behaviour within the MPS device (Table 1). Non-specific binding to the MPS device was determined after a 72 h incubation with each candidate drug at 1 μM (Table 1). Both candidate drugs were calculated to have less than 0.1% loss as a consequence of non-specific binding to the MPS device (Table 1). Protein binding of each candidate drug within the fat culture medium was also measured and we found that OCA and ELF were highly protein-bound, highlighting that higher concentrations of the drugs would need to be added to the microtissues to ensure sufficient free drug availability for pharmacological effect (Table 1). The protein binding calculation for OCA is comparable to reported plasma protein binding³¹, but an equivalent figure for ELF was not available. Because the hepatocytes within the MPS device are metabolically competent, we evaluated the half-life for each candidate drug in NASH microtissues. OCA was found to have a half-life of 12 h, whereas ELF had a half-life of 16 h (Supp. Fig. 5). Using these data, dosing strategies for both compounds were identified which maintain drug concentrations close to clinically relevant levels throughout the treatment period (Table 1).

Following 10 days of compound treatment (Supp. Fig. 6a), NASH microtissues were assessed for the production of inflammatory and fibrotic biomarkers (Fig. 3a), the expression of pro-fibrotic genes (Fig. 3b) and the expression of collagen-1 and alpha SMA (Fig. 3c, d). Both OCA and ELF caused a concentration-dependent decrease in inflammatory cytokines, including: CXCL1, IL1-Rα, IL-6 and MCP-1 (Fig. 3a). Both compounds also reduced expression of genes associated with ECM remodelling (e.g. MMP1, MMP9 and SERPINA1), TGFβ signalling (CEBPB, DCN) and inflammation (CCL2, CCL3, CXCR4) (Fig. 3b). Neither compound significantly affected cell health or microtissue formation, but higher doses of ELF did slightly reduce albumin production (Supp. Fig. 6b, c). Crucially using the developed quantitative imaging approach, both OCA and ELF abrogated α-SMA expression in a concentration-dependent manner by up to 50% and collagen-1 deposition was reduced by up to 35% (Fig. 3d).

PHH, KC and HSC co-cultures were cultured in the MPS platform under high fat conditions and dosed with varying concentrations of Obeticholic acid (OCA) and Elafibranor (ELF) QD or vehicle control (control) for 10 days, following an initial 4-day pre-culture phase. a The secreted cytokine profile from liver microtissues was compared by Luminex to determine the effects of each compound on the inflammatory profile of the liver with samples analysed at day 8 and 14 of the culture. Data were normalised by Z-transformation to allow comparison of all analytes and presented as a heatmap. b The expression of fibrosis-associated genes was analysed in total RNA from control, OCA and ELF treated liver microtissues (highest concentration for each compound) using Human Fibrosis RT2 Profiler PCR Arrays. Gene expression changes are expressed as a fold change over control. c Liver microtissues were stained for collagen-type I, α-SMA, nuclei (DAPI – blue), phalloidin (green) and imaged by confocal microscopy. Representative images are shown and scale bars 200 µm. d Staining of microtissue was quantified by measuring total fluorescence intensity throughout individual microtissues, each data point represents an average of all the microtissues imaged within an MPS scaffold (min 8, max 20 microtissues). e PHH, KC and HSC co-cultures were cultured in the MPS platform under high fat or lean conditions for 30 days. Cultures either remained in the same type of media throughout or were switched at day 15 from fat media to lean media (Fat -> Lean). Cell culture medium samples were analysed for the presence of IL-6, f MCP-1, g TIMP-1, which were measured by ELISA. h Microtissues were stained with Oil-Red O to determine fat loading during culture period, total stain was quantified by absorbance at 510 nm. All datapoints shown, either as box-whisker plots highlighting mean and min-max or with error bars highlighting means ± SD. All data from a minimum of three independent cultures; statistical comparisons made to control samples unless other comparison shown.

Bariatric surgery can significantly reduce NASH and liver fibrosis in patients with obesity, as described in recent meta-analyses^32,33, so in addition to pharmacological intervention we also explored whether the disease phenotype in the NASH microtissues could be reversed by reducing the FFA in the culture medium, as a mimic of dietary changes in humans. NASH microtissues were cultured in lean or fat conditions for a total of 30 days, with some microtissues cultured in fat medium for the first 15 days and then switched to lean medium for the remaining time. At day 15 and at day 30 the phenotype of the microtissues was assessed. By switching the microtissues from fat to lean media inflammation (IL-6 and MCP-1 production) was reduced (Fig. 3e, f), markers of fibrosis (TIMP1) were reduced (Fig. 3g) and fat loading was halted (Fig. 3h). Together these data demonstrate that the NASH microtissues are well placed to assess the efficacy of a range of therapeutic interventions and assess the full spectrum of disease endpoints from fat loading/steatosis through to liver fibrosis.

Systems biology interrogation of NASH microtissues demonstrates a spectrum of disease phenotypes can be generated

There is only limited understanding of the pathogenic drivers of NASH and their interaction, principally due to a lack of translational models that are amenable to detailed molecular analysis^2,34. Here, a systems biology ‘cue-signal-analysis’ paradigm^35,36 was utilised to produce large factorial datasets to enable the identification of molecular pathways of importance. We first compared monocultures of PHH with co-cultures with and without both KC and HSC to determine which combination gave the phenotype most akin to the human disease and observed that all NASH biomarkers were only produced with all three cell types present in the microtissues (Supp. Fig 7). To create varying set ups of the NASH microtissues we created hepatocyte microtissues in the MPS with high numbers of NPC (60,000 KC and HSC—equivalent to 10% of microtissue population) and low numbers of NPCs (6000 KC and HSC—equivalent to 1% of microtissue), representing physiological and non-physiological cell ratios respectively. Next, we took these different microtissues and cultured them in media with or without FFA. Finally, we supplemented the media with lipopolysaccharide (LPS), TGFβ, fructose or cholesterol either as single ‘cues’, in pairs or all together (Fig. 4a). ‘Cues’ were selected for clinical association with NASH. LPS, is a pleiotropic inflammatory molecule, it is proposed to reach the liver via the hepatic portal vein after release from the gastrointestinal tract, contributing to disease progression^37,38. TGFβ has been extensively studied in relation to fibrosis due to its regulation of genes controlling collagen and matrix-modifying enzymes³⁹. Fructose is a key substrate for de-novo lipogenesis, a process that has been identified as a key event in lipid accumulation within NAFLD patients⁴⁰. Finally, cholesterol accumulation has also been linked to NASH disease aetiology⁴¹.

PHH, KC and HSC co-cultures were cultured in the MPS platform under a variety of conditions for 14 days and transcriptomic profile of microtissues were compared. a Study was designed to test number of NPCs, presence of fat, fructose, cholesterol, LPS and TGFβ for effects on the transcriptional profile of the liver MPS NASH model. b Differentially expressed genes (defined by comparison to low NPC lean) from selected conditions were compared to identify enriched pathways and biological processes, using PANTHER database and mapped using CompareCluster, with size of cluster representing number of genes involved (GeneRatio) and colour identifying confidence interval (p.adjust). Only high NPC conditions shown (low NPC data in Supp. Fig. 8) and some conditions did not produce any gene clusters with significant p value < 0.05 and are not shown on plot. c Expression of markers of NAFLD activity score progression⁴⁵. d Expression of markers of fibrosis stage progression in NAFLD⁴⁵. e Expression of liver-specific genes⁴⁶. Heat maps used to compare different culture conditions with data normalised to z-scores to show variance from mean of all observations for each gene. Transcriptomic data were generated from a minimum of three independent replicate samples per condition.

We compared the transcriptional profile of the NASH MPS model with and without each cue to determine how they affected the disease state. Using GO term enrichment and cluster analysis indicated that the cues affected processes implicated in the multi-hit hypothesis of NASH pathogenesis⁴²: extracellular matrix organisation, mitochondrial function, inflammation and responses to stress (Fig. 4b). These effects and clusters were substantially much more prominent in the high NPC condition than the low NPC condition (Supp. Fig. 8). LPS only induced pathways associated with inflammatory signalling, not all of which are relevant to NASH, whilst TGFβ and combinations of cues including TGFβ induced a wide range of clusters associated with collagen formation and matrix reorganisation (Fig. 4b).

We also assessed translational relevance of the microtissue model by comparison of differential expression changes to marker gene sets derived in NAFLD patient cohorts. We firstly determined if the microtissues recapitulated the molecular signatures of processes that occur early in the development of NAFLD in humans, namely lipid accumulation, insulin resistance, mitochondrial dysfunction and changes to the poly (ADP-ribose) polymerases (PARP) pathway. FFA and the combination of FFA plus cholesterol, but not the other single cues, affected lipid droplet-associated genes⁴³ in a manner resembling simple steatosis in NAFLD (Supp. Fig. 9). Fructose and fat (FFA) challenge combined with fructose-induced differential expression changes consistent with insulin resistance⁴⁴ (Supp. Fig. 10). Most cues also reduced the expression of mitochondrial associated genes (Supp. Fig. 11) and challenge with FFA was sufficient to modulate PARP pathway, but changes were moderate and additional cues did not cause further changes (Supp. Fig. 12). Interestingly, there were differential gene expression changes with the expected directionality to NAFLD in the microtissues with the lower number of HSCs and KCs (6 K) as well as the microtissues with a more physiological number (60k) of non-parenchymal cells (Supp. Figs. 9, 10) and the changes in mitochondrial genes were most pronounced in the low NPC group (Supp. Fig. 11).

Next, we assessed if the NASH microtissues re-capitulated molecular features of disease progression. TGFβ and combination of cues with TGFβ mirrored a gene signature for NAFLD progression⁴⁵ (Fig. 4c) and fibrosis stage progression⁴⁵ (Fig. 4d) and a reduced signature for liver-specific genes⁴⁶ (Fig. 4E), reflecting decline in normal liver function in advanced disease. Confirmatory analyses indicated that TGFβ and, to a lesser extent, fructose triggered gene expression changes resembling fibrosis⁴⁷ (Supp. Fig. 13), overexpression of stellate cell signature⁴⁸ (Supp. Fig. 14) as well as differential expression changes of markers associated with HCC risk^49,50 (Supp. Fig 15). In particular, genes associated with collagen (e.g. COL3A1) whose deposition is associated NASH progression⁴⁷ were elevated (Supp. Fig. 13). TGFβ and cues combined with TGFβ reduced expression of core metabolism and transporter genes⁵¹ (Supp. Fig. 16), mimicking reduced detoxication ability of NASH livers⁵². Treatment with LPS did not mimic transcriptomics profiles of human NASH or NAFLD.

Assessment of gene expression changes was supplemented by assessment of the secretome produced by NASH microtissues in the MPS after exposure to the various cues. Increased pro-inflammatory cytokine secretion is a defining characteristic in NASH patients, so we conducted a Luminex cytokine array covering 42 different biomarkers, cytokines and chemokines secreted into the culture media of the different microtissues at multiple time points. We consistently detected 27 out of the 42 biomarkers and observed that there was differential expression of cytokines and chemokines driven by the different cues applied to the micro-tissue (Fig. 5a, b). The LPS cue and the presence of higher NPC numbers were responsible for driving the largest increase in cytokine and chemokine release, with the largest increases in cytokines measured on day 14 (Fig. 5b). Specifically, for LPS dosed samples IL-4, GM-CSF and MIP1α showed the biggest increase compared to control. IL-6 and IL-8 also have recognised chemoattractant properties and are up-regulated in NAFL/NASH patients² but interestingly, although LPS caused an increase in their secretion, it was not as high as IL-4, GM-CSF and MIP1α (Fig. 5b). Fructose caused the biggest increase in IP-10 production, occurring on day 15 (Fig. 5b). The greatest increase in TNFα also occurred on day 14 in response to a cholesterol cue (Fig. 5b). However, the data also indicated that there was inter-cue crosstalk with respect to cytokine production, with most cues in combination resulting in a suppression of inflammation markers (Supp. Fig. 17). For the pro-fibrotic biomarkers, again the increased number of NPCs, TGFβ and to a lesser extent fructose caused the most significant increases in pro-collagen 1, fibronectin and TIMP-1 expression (Fig. 5c). Similarly, to the cytokine production, these changes were not further enhanced by combining cues, as all the cue combinations containing TGFβ had the highest expression (Supp. Fig. 17c). Albumin production was similar between lean and fat microtissues irrespective of what cue was studied, but the presence of TGFβ and to a lesser extent cholesterol reduced albumin production (Supp. Fig. 18), an observation which mirrors the transcriptomic data suggesting TGFβ in particular causes a decline in liver function as expected in more advanced NASH⁵².

PHH, KC and HSC co-cultures were cultured in the MPS platform under a variety of conditions for 14 days to test number of NPCs, presence of fat, fructose, cholesterol, LPS and TGFβ for effects on the soluble biomarker profile of the liver MPS NASH model. Expression of biomarkers, cytokines and chemokines were compared, all concentrations were normalised by Z-transformation to allow comparison and shown relative to mean for each marker (white = not detected). a Cytokine expression compared with varying levels of NPCs; b for the high NPC group in the presence of additional cues; c Expression of pro-fibrotic biomarkers with varying NPC content and with additional varying cues in high NPC condition (additional data in Supp. Fig. 17). d Graphical summary of the changes that occur in the NASH model in the presence of Fat and Fat + TGFβ. All data were generated from a minimum of three independent replicate samples per condition.

By using this systems biology ‘cue-signal-analysis’ paradigm to interrogate the MPS NASH model we have identified that the MPS NASH model can be created with varying levels of NPCs but with higher and more physiologically relevant NPC numbers are all soluble biomarkers detectable. FFA and the combination of FFA with cholesterol can be used to model steatosis. Treatment of the MPS NASH model with TGFβ and to a lesser extent fructose, could be used to model inflammation and fibrosis in NASH and potentially study the development of NASH-induced hepatocellular carcinoma, because TGFβ may be pro-carcinogenic⁵⁰.

Pharmacological modulation of NASH microtissues with enhanced fibrosis

The systems biology approach identified TGFβ to be a key driver in creating a more advanced NASH phenotype in the MPS model, with enhanced fibrotic markers and reduced liver function. Transcriptomic effects were observed with both high and low NPC groups, so we compared the fibrosis by microscopy of NASH microtissues with varying numbers of NPCs in the presence of fat and TGFβ (Supp. Fig. 19). Microtissues with enhanced NPC numbers had significantly higher levels of collagen-I and alpha SMA expression and were more responsive to TGFβ (Supp. Fig. 19). Therefore, we explored how effective pharmacological intervention would be for NASH microtissues with high NPC numbers supplemented with TGFβ and exposed them to varying concentrations of OCA and ELF QD as previously described (Supp. Fig. 6a). Similarly, to previous observations, both candidate drugs were shown to have profound inhibitory effects on inflammatory cytokine production (Fig. 6a). OCA had the greatest effect on IL-1Rα whereas ELF had the greatest effect on MCP-1 reflecting the fact that while both candidate drugs have anti-inflammatory properties, but their mechanisms of action are different. The differences in the compounds were further illustrated when exploring effects on the soluble biomarkers TIMP-1 and pro-collagen 1, both of which were reduced by ELF treatment but not OCA treatment (Fig. 6b). Finally, we used our quantitative fibrosis imaging assay to assess the effect of each candidate drug on collagen-1 deposition and α-SMA expression (Fig. 6c). We observed a significant increase in collagen-1 and α-SMA signal in NASH microtissues treated with fat and TGFβ compared to just fat alone (Fig. 6c). Both candidate drugs caused a significant reduction in collagen-1 deposition and α-SMA expression, with the highest concentrations of OCA and ELF causing an 80% reduction in α-SMA expression and a 75% reduction in collagen-1 deposition (Fig. 6c). These reductions are more significant than those previously observed in the fat only NASH model (Fig. 3) and demonstrates the value of the MPS NASH model to be able to investigate the effects of therapeutic intervention on varying stages of disease from simple steatosis, to NAFLD and through to advanced NASH with significant fibrosis.

PHH, KC and HSC co-cultures were cultured in the MPS platform under high-fat conditions and dosed with varying concentrations of Obeticholic acid (OCA) and Elafibranor (ELF) QD or vehicle control for 10 days, following an initial 4-day pre-culture phase. Throughout the compound dosing period TGFβ was also dosed onto all microtissues (NASH only vehicle control samples were also included in the study for comparison, which were not treated with TGFβ or dosed with any compound). a The secreted cytokine profile from treated liver microtissues was compared to determine the effects of each compound on the inflammatory profile of the liver with samples analysed at day 8 and 14 of the culture. Data were normalised by Z-transformation to allow comparison of all analytes (white = undetected). b The expression of fibrosis-associated biomarkers was analysed on day 14 media samples and compared to NASH + TGFβ control samples. c Liver microtissues were stained for collagen-type I, α-SMA, nuclei (DAPI), phalloidin (green) and imaged by confocal microscopy. Representative images are shown and scale bars 200 µm. Staining of microtissues was quantified by measuring total fluorescence intensity throughout individual microtissues, each data point represents an average of all microtissues per MPS culture (min 8, max 20). All datapoints shown with error bars highlighting means ± SD, all data from a minimum of three independent cultures: all comparisons were made to NASH + TGFβ control samples.