miR-122 Release in Exosomes Precedes Overt Tolvaptan-Induced Necrosis in a Primary Human Hepatocyte Micropatterned Coculture Model

Abstract

Idiosyncratic drug-induced liver injury (IDILI) is thought to often result from an adaptive immune attack on the liver. However, it has been proposed that the cascade of events culminating in an adaptive immune response begins with drug-induced hepatocyte stress, release of exosomal danger signals, and innate immune activation, all of which may occur in the absence of significant hepatocelluar death. A micropatterned coculture model (HepatoPac) was used to explore the possibility that changes in exosome content precede overt necrosis in response to the IDILI drug tolvaptan. Hepatocytes from 3 human donors were exposed to a range of tolvaptan concentrations bracketing plasma C_max or DMSO control continuously for 4, 24, or 72 h. Although alanine aminotransferase release was not significantly affected at any concentration, tolvaptan exposures at approximately 30-fold median plasma C_max resulted in increased release of exosomal microRNA-122 (miR-122) into the medium. Cellular imaging and microarray analysis revealed that the most significant increases in exosomal miR-122 were associated with programmed cell death and small increases in membrane permeability. However, early increases in exosome miR-122 were more associated with mitochondrial-induced apoptosis and oxidative stress. Taken together, these data suggest that tolvaptan treatment induces cellular stress and exosome release of miR-122 in primary human hepatocytes in the absence of overt necrosis, providing direct demonstration of this with a drug capable of causing IDILI. In susceptible individuals, these early events may occur at pharmacologic concentrations of tolvaptan and may promote an adaptive immune attack that ultimately results in clinically significant liver injury.

Tolvaptan is a vasopressin V₂-receptor antagonist marketed for the treatment of hyponatremia. Tolvaptan also has been approved for the treatment of autosomal dominant polycystic kidney disease (ADPKD) in some countries (Torres et al., 2012). However, a small number of idiosyncratic drug-induced liver injury (IDILI) cases have been associated with tolvaptan use in ADPKD patients (Watkins et al., 2015).

It now appears that many and perhaps most IDILI events involve an adaptive immune attack on the liver (Usui and Naisbitt, 2017). This is supported, in part, by the identification of human leukocyte antigen (HLA) alleles as significant risk factors for IDILI reactions (Urban et al., 2014). The current theory is that proteins encoded by these HLA alleles present drug-associated neoantigens that stimulate a cytotoxic T-cell attack on the liver (Mosedale and Watkins, 2017). However, presentation of neoantigen alone would not be sufficient to initiate a robust immune response (Matzinger, 1994). Therefore, it is hypothesized that IDILI reactions begin with some level of direct drug-induced stress to the hepatocyte, including activation of specific molecular response pathways such as mitochondrial dysfunction, oxidative stress, and alterations in bile acid homeostasis (Cho and Uetrecht, 2017; Mosedale and Watkins, 2017). It has been further proposed that activation of stress response pathways must result in the release of “danger signals,” which are necessary to target cytotoxic T-lymphocytes to the liver (Mosedale and Watkins, 2017).

Recent evidence suggests that the release of danger signals may occur in the absence of hepatocyte necrosis, and that even minor serum alanine aminotransferase (ALT) elevations that are often observed in a large percentage of patients treated with IDILI drugs reflect hepatocyte death due to an adaptive immune attack on the liver. For example, drug-reactive T cells are detected in patients experiencing mild and/or transient elevations in response to treatment with the well-known IDILI drug isoniazid (Warrington et al., 1978, 1982). In addition, drug-reactive T cells can be detected in the blood of patients with isoniazid-induced liver injury prior to the onset of ALT elevations (Warrington et al., 1982). Furthermore, specific HLA alleles have been associated with transient elevations in serum ALT due to several other drugs capable of causing more serious IDILI reactions (Kindmark et al., 2008; Singer et al., 2010; Spraggs et al., 2011).

Several recent studies support the idea that danger signals released from viable hepatocytes come in the form of small extracellular vesicles called exosomes. It has been shown both in vivo and in vitro that exposure to acetaminophen changes the RNA content of hepatocyte-derived exosomes in the absence of overt hepatocyte necrosis (Holman et al., 2016). It also has been demonstrated that hepatocyte-derived exosomes released in response to ethanol exposure can be internalized by monocytes and increase proinflammatory cytokine production in response to LPS challenge (Momen-Heravi et al., 2015). In this study, the effects on the monocytes could be attributed directly to the hepatocyte-specific microRNA-122 (miR-122) contained in the exosomes. miR-122 is therefore both a specific marker of hepatocyte-derived exosome release and a likely danger signal itself.

Although hepatocyte death during an IDILI event may require the presence of specific HLA risk alleles (Usui and Naisbitt, 2017), early events may not be as dependent on host susceptibility factors and therefore may be replicated in hepatocytes obtained from random donors with sufficient drug exposure. Micropatterned cocultures (MPCCs) of primary human hepatocytes and 3T3-J2 murine embryonic fibroblasts have good retention of hepatocyte functionality (Khetani and Bhatia, 2008) and have been used extensively for long-term studies of perturbed hepatocellular pathways in vitro (Davidson et al., 2015; Khetani et al., 2013; Ramsden et al., 2015; Trask et al., 2014; Ware et al., 2015, 2017). Tolvaptan is metabolized by cytochrome P450 3A (CYP3A) enzymes (Shoaf et al., 2012) with biliary excretion as the predominant route for elimination (Furukawa et al., 2011). The ability to retain CYP450 enzyme activity and transporter function (Lin et al., 2016) was another reason MPCCs were selected for this study. Furthermore, it may be important that this culture system does not incorporate an extracellular matrix protein gel overlay, which could interfere with exosome release into the culture medium.

In this study, a commercially available MPCC model called HepatoPac was used to examine early hepatocellular events in response to tolvaptan. It was hypothesized that an increase in tolvaptan-induced exosomal miR-122 release would be observed prior to overt necrosis and may provide insight into the pathogenesis of IDILI. High-content imaging (HCI) and toxicogenomic analyses also were performed to identify mechanisms contributing to the initiating cellular stress.

MATERIALS AND METHODS

Micropatterned cocultures of human hepatocytes

HepatoPac MPCCs in 96- and 24-well formats were purchased from Hepregen Corporation (now Ascendance Biotechnology, Medford, Massachusetts). Cryopreserved primary human hepatocytes were supplied by Hepregen (previously purchased from a commercial vendor) or purchased from a commercial vendor (QPS Hepatic Biosciences) and provided to Hepregen. Lot numbers and donor information are described in Table 1.

Table 1.

Donor Information for Hepatocytes Used in MPCCs

Donor	Supplier	Lot No.	Sex	Age	Race	BMI	Cause of Death
1	QPS	QHu0030	Male	30	Caucasian	20.9	Anoxia (hanging)
2	Hepregen	3121A	Female	52	Caucasian	19.4	Anoxia (2nd to cardiac arrest)
3	Hepregen	4202	Female	39	Caucasian	35.2	Anoxia (2nd to cardiac arrest)

Hepatocyte culture treatment

Micropatterned cocultures were maintained in HepatoPac Maintenance Medium (Hepregen) with medium replacement every 2–3 days. The plates were kept in a 37°C humidified incubator with 10% CO₂. After shipping, MPCCs were allowed to acclimate for at least 48 h prior to drug treatment. On the morning of dosing (day 0), MPCCs were washed and incubated for 2 h with serum-free HepatoPac Toxicity Application Medium (Hepregen). The medium then was replaced with dosing solutions (prepared in serum-free HepatoPac Toxicity Application Medium) and the cells were exposed continuously for 4, 24, or 72 h. Hepatocytes were exposed to either tolvaptan (0.01, 0.1, 1, 20, or 50 µM; Otsuka Pharmaceutical Co Ltd, Rockville, Maryland) or vehicle (0.5% vol/vol DMSO; Sigma, St Louis, Missouri). In HCI experiments, the use of 96-well plates allowed for 2 additional concentrations of tolvaptan (3 and 10 µM) to be assayed. The selection of concentrations was based on clinical and nonclinical study data provided by Otsuka Pharmaceutical Co Ltd.

Toxicogenomic and exosome experiments

Gene expression and exosome analyses were multiplexed in 24-well plates. Two rounds of dosing were performed for each donor and time point (ie, two plates treated at different times); the samples were pooled for analysis. Acetaminophen (Sigma) was used as the reference control (results shown in Supplementary Information). At the desired time point, photomicrographs of wells 2 and 6 of each treatment condition were taken at 4X, 10X, and 20X using a Nikon Eclipse TS100 microscope equipped with a DS-Fi1 digital camera (Nikon Instrument, Melville, New York). Medium from each treatment condition was collected and replicate wells were pooled in a microcentrifuge tube to create N = 1 sample per concentration, time, and donor. Medium samples were centrifuged at 3000 × g for 15 min to pellet cellular debris. The supernatant then was transferred into 2 microcentrifuge tubes and stored at  ≤ −70°C for downstream assays.

Cytotoxicity

One of the two medium aliquots was used to assay for ALT, urea, and miR-122. Alanine aminotransferase was assayed on a SpectraMax microtiter plate reader (Molecular Devices, Sunnyvale, California) using the Infinity ALT Liquid Stable Reagent (Thermo Scientific, Waltham, Massachusetts) as described previously (Harrill et al., 2012). Urea production was measured via a colorimetric endpoint analysis with diacetylmonoxime, acid, and heat (Stanbio Labs, Boerne, Texas) as described previously (Ware et al., 2015). miR-122 was assayed using qPCR as described previously (Kroh et al., 2010) with some modifications. Briefly, total RNA was isolated using the miRNeasy Serum/Plasma Kit (Qiagen, Valencia, California) and reverse transcription reactions were performed using the TaqMan miRNA Reverse Transcription Kit and miRNA-specific stem-loop primers for miR-122 (Applied Biosystems, Foster City, California). Standard curves with known concentrations of miR-122 were generated in parallel using miScript miRNA mimics (Qiagen). qPCR was performed on both samples and standards using the TaqMan Universal PCR Master Mix II with no uracil-N-glycosylase and miRNA-specific TaqMan primer/probe mixes (Applied Biosystems) on a 7900HT Fast Real-Time PCR System (Applied Biosystems). Absolute values of miR-122 copy number were determined using the synthetic miR-122 standard curve.

Exosome enrichment and qPCR

The second medium aliquot was used for evaluation of mRNA content released in exosomes. Extracellular vesicles were pelleted using Exoquick-TC (System Biosciences, Palo Alto, California) according to the manufacturer’s instructions and as previously described (Holman et al., 2016). Vesicles enriched by this protocol have been previously characterized and shown to be highly enriched for exosomes (Holman et al., 2016). Therefore, the term exosome is used henceforth to describe the extracellular vesicles analyzed in this study. Total RNA was isolated from the exosomal pellet using the miRCURY RNA Isolation Kit (Exiqon, Woburn, Massachusetts) as described previously (Eldh et al., 2012). mRNA was reverse-transcribed using the High Capacity cDNA RT kit (Applied Biosystems) according to the manufacturer’s protocol. miR-122 from exosomal RNA was quantified as described earlier.

Cellular RNA isolation and microarray processing

Total cellular RNA was isolated from MPCCs using an RNeasy total RNA isolation kit and the QIAcube automated system (Qiagen) according to the manufacturer’s instructions. RNA from replicate wells (8 total) was pooled to create N = 1 sample per concentration, time, and donor. RNA integrity was verified spectrophotometrically with an Agilent 2100 Bioanalyzer. For microarrays, double-stranded cDNA was synthesized from 150 ng of total RNA, then transcribed to biotin-labeled cRNA using the 3′ Express IVT kit (Affymetrix, Santa Clara, California). Fifteen micrograms of labeled cRNA were fragmented and prepared for hybridization. The Affymetrix HT HG-U133+ PM peg arrays were used with the Affymetrix Gene Titan system. Gene Titan robotic instrumentation was used to perform the hybridization, washing, and scanning of the peg arrays. Two microliters of total RNA also were used for analysis of cellular miR-122 as described earlier.

High-content imaging experiments

High-content imaging assays were performed following the adoption of the National Institutes of Health/National Center for Advancing Translational Sciences HCI guidelines (Buchser et al., 2004; Trask, 2004) and a recent publication describing HCI methodology to assess liver toxicity in MPCCs (Trask et al., 2014). High-content imaging assays were conducted in 96-well microplates. Valinomycin (Sigma) was used as the reference control compound. At each time point, medium was removed and the cells were stained and fixed as follows. MitoTracker Orange and TOTO-3 (Life Technologies, Carlsbad, California) dyes were introduced for approximately 30 min followed by cell fixation with 4% (vol/vol) paraformaldehyde and permeabilization with 0.1% (vol/vol) Triton X-100. Subsequently, cells were blocked with 3% (wt/vol) BSA and labeled with an intracellular human cytochrome c antibody (Enzo Life Sciences, East Farmingdale, New York, clone 2B5.F8), followed by an Alexa Fluor 488 secondary and a Hoechst 33342 nuclear dye counterstain (Life Technologies). At the completion of the staining protocol, phosphate buffered saline was added to each well, and the plates were sealed and maintained at 4°C until image acquisition.

Image acquisition and analysis

Image acquisition took place within 72 h of staining. Images were acquired using a modified Cellomics ArrayScan XTI (Thermo Scientific) reader equipped with a Zeiss 10X/0.45 NA Plan-ApoChromat objective lens. Fluorescence signals were triggered using an LED light source (Lumencor, Beaverton, Oregon) and collected through a BrightLine quad-band filter set (Semrock, Rochester, New York). The following fluorescent probes were measured: Hoechst 33342 (ex 386/em 446); Alexa Fluor 488 (ex 485/em 523); MitoTracker Orange (ex 549/em 600); and TOTO-3 (ex 650/em 677). A total of at least 12 image fields per well were captured for each fluorescent probe using a 14-bit X1 camera (Photometrics, Tucson, Arizona).

The compartmental analysis bioapplication algorithm (Thermo Scientific) was used to quantitatively measure cell size, shape, intensity, and texture as described previously (Trask et al., 2014). Responses at the individual cell level were initially calculated and then summarized at the well-level to determine the number or percentage of objects at a defined threshold of response criteria. Thresholds for well-level responses were set using the lower or upper 10th percentile of vehicle control responses. Results were recorded for the individual HCI features described in Table 2.

Table 2.

Individual High-Content Imaging Features

Name	Primary Probe(s)	Description/Biological Meaning
Valid object count	Hoechst	Number of all objects (nuclei); decrease indicates either cell loss or cell detachment
Low object size	Hoechst	Percentage of individual nuclei < pixel threshold; higher number indicates shrinkage of cell nuclei; likely programmed cell death
High object variance	Hoechst	Percentage of individual nuclei with variability in intensity > threshold; higher number is an indicator nuclear texture relating to nuclear changes; likely programmed cell death
Cell permeability	TOTO-3	Percentage of individual cells whose nuclear area intensity > threshold; increase indicates cell membrane permeability
High cytochrome c	Cytochrome c	Percentage of individual cells whose cytoplasmic area has intensity spots > threshold; increase is an indicator of mitochondrial-induced apoptosis
Low mito membrane potential	MitoTracker Orange	Percentage of individual cells whose cytoplasmic area has intensity < threshold; increase indicates decreased mitochondrial membrane potential
Subpopulation 1	Hoechst	Subpopulation analysis; percentage of individual cells that are both low object size and high object variance; increase confirms nuclear condensation; confirms programmed cell death
Subpopulation 2	Cytochrome c and MitoTracker Orange	Subpopulation analysis; percentage of individual cells that are both high cytochrome c and low mito membrane potential; increase confirms mitochondrial-induced apoptosis

Statistical analyses

Statistical analyses for cytotoxicity, exosomal, and HCI endpoints were performed using GraphPad Prism statistical software version 5.03 (GraphPad Software, La Jolla, California). Sample material from the 2 rounds of exosome and toxicogenomics dosing was pooled creating an N = 1 per donor, time point, treatment, and concentration. Copies of miR-122 in the cell and the medium samples were used to calculate total miR-122 in the system for individual treatments. Vesicular and medium miR-122 then were divided by total miR-122 in the system to calculate the relative percentages of miR-122 released into the medium in each fraction. Fold-change values for cytotoxicity and exosome endpoints were calculated for each donor, time point, treatment, and concentration by dividing the values for each concentration over the value of the corresponding vehicle control. Mean values of fold changes for each donor, time point, and treatment were compared across concentrations using a 2-way analysis of variance. Comparisons between each concentration and the vehicle control within a treatment were performed using a Dunnett’s test.

Mean values for HCI endpoints were calculated for each donor, time point, treatment, and concentration from replicate wells on each plate. Fold-change values for each donor, time point, treatment, and concentration then were calculated by dividing the mean values for each concentration over the mean value of the corresponding vehicle control. Mean values of fold changes for each donor, time point, and treatment were compared across treatment concentrations using a 2-way analysis of variance. Comparisons between each concentration and the vehicle control within a treatment were performed using a Dunnett’s test.

Gene expression analysis was performed in Partek Genomics Suite version 6.6 (Partek, St Louis, Missouri). For gene expression profiling, Affymetrix CEL files (54 total, representing N = 1 per donor, time point, treatment, and concentration) were normalized using the Robust Multiarray Average method with a log base 2 (log₂) transformation (Irizarry et al., 2003). A filtering step was performed to remove control probe sets, probe sets without mRNA annotation, and low expression probe sets. Gene expression differences were identified using analysis of variance models with linear contrasts. Both array scan date and donor were identified as significant factors contributing to the variability in probe intensity levels and were therefore used to normalize analyses where appropriate. Probability values were adjusted for multiple comparisons using a false discovery rate of 5% (Benjamini and Hochberg, 1995). An absolute value fold change cutoff of > 1.5 also was used as noted below. Pathways and toxicity lists enriched among statistically significant, differentially expressed genes in the data were identified using the Tox Analysis module in Ingenuity Pathway Analysis (Ingenuity Systems, Redwood City, California; Build version 400896M; Content version: 28820210). Gene expression data generated for this article can be downloaded in its entirety from the Gene Expression Omnibus repository under the accession number GSE99878. All data are MIAME compliant.

RESULTS

Tolvaptan-Induced Hepatocellular Stress Precedes Overt Cytotoxicity and Is Associated With an Increase in Exosomal miR-122

Hepatocytes from 3 adult donors were exposed to tolvaptan (0.01–50 µM) or DMSO vehicle control (0.5% vol/vol) continuously for 4, 24, or 72 h. Initially, cell stress and cytotoxicity were evaluated using hepatocyte-specific endpoints including hepatocyte morphology, urea, ALT, and miR-122. Some morphological changes, including increases in vesiculation and granularity within the cytoplasm as well as the loss of well-delineated plasma membrane borders and bile canaliculi, were observed with 50 µM tolvaptan after a 72-h exposure (Figure 1). Tolvaptan exposure also resulted in a concentration-dependent decrease in urea production at both 24 and 72 h (Figure 2A). However, markers of overt hepatocyte necrosis, ALT and total miR-122 release into the medium, were not affected significantly at any concentration of tolvaptan at any time point of exposure (Figs. 2B and 2C).

Figure 1.

Morphological changes only observed in hepatocytes exposed to 50 μM tolvaptan at 72 h. Representative phase contrast photomicrographs from donor 3. Images were captured using a 10X objective lens.

Figure 2.

Exosomal miR-122 release is associated with hepatocellular stress in response to tolvaptan. Average fold change (compared with control) in medium levels of urea (A), alanine aminotransferase (ALT) (B), total microRNA-122 (miR-122) (C), and exosomal miR-122 (D). Data are represented as mean + SE of N = 3 biological replicates. *p < .05, **p < .01, and ****p < .0001 indicate the difference in responses observed between individual drug concentrations and vehicle control for a given exposure time (Dunnett’s test).

Exosomes were enriched from the conditioned cell culture medium and used to assay for exosomal miR-122. Previous studies have shown that exosomal miR-122 may be released from hepatocytes prior to and in the absence of overt drug-induced cytotoxicity (Holman et al., 2016). Exosomal miR-122 was increased 7.7-fold in response to 50 µM tolvaptan at 72 h (Figure 2D). A trend toward increased exosomal miR-122 release into the medium also was observed with increasing tolvaptan concentrations at 24 h. On average, exosomal miR-122 only accounted for approximately 4% of total miR-122 released into the culture medium (data not shown).

High-Content Imaging Identifies Apoptosis and Mitochondrial Dysfunction Following Tolvaptan Exposure

High-content imaging was used to determine if exosomal release was associated with the activation of stress response pathways in MPCCs exposed to tolvaptan. The HCI panel consisted of four probes designed to detect hepatocyte number and nuclear morphology, membrane permeability, cytochrome c release, and mitochondrial membrane potential (Table 2). Detectable fluorescent signals were observed for the four probes used in the HCI experiments. A composite image of a representative control well is shown in Figure 3. The 72-h time point of donor 3 was chosen as a representative well to demonstrate that appropriate morphology was observed throughout the entire culture timeline.

Figure 3.

Composite image of a representative control well from donor 3 at 72 h. The 4 fluorescent probes are shown in different colors: Hoechst 33342 (blue), TOTO-3 (cyan), MitoTracker Orange (red), and cytochrome c 1° antibody with an AlexaFluor488 2° antibody (green).

Although no overt cytotoxicity was observed by biochemical measures (ALT and total miR-122), cellular imaging revealed a small (1.7-fold) but statistically significant increase in the number of hepatocytes with increased permeability at the highest tolvaptan concentration at 72 h (Figure 4A). Subpopulation analyses then were used to evaluate stress response pathways of programmed cell death (ie, hepatocytes with both small and textured nuclei) and mitochondrial-induced apoptosis (ie, hepatocytes with both high cytoplasmic cytochrome c spot intensity and low mitochondrial membrane potential). Tolvaptan induced time- and concentration-dependent elevations in programmed cell death, with the most significant effect occurring in response to 50 µM tolvaptan at 72 h (Figure 4B). At 4 and 24 h, programmed cell death was associated with significant increases in mitochondrial-induced apoptosis (Figure 4C).

Figure 4.

Tolvaptan-induced programmed cell death is associated with cell permeability and mitochondrial-induced apoptosis. Average fold change (compared with control) in cell permeability (A), subpopulation 1 (hepatocytes with both small and textured nuclei; ie, programmed cell death) (B), and subpopulation 2 (hepatocytes with both high cytoplasmic cytochrome c spot intensity and low mitochondrial membrane potential; ie, mitochondrial-induced apoptosis) (C). Data are represented as mean + SE of N = 3 biological replicates. **p < .01 and ****p < .0001 indicate the difference in responses observed between individual drug concentrations and vehicle control for a given exposure time (Dunnett’s test).

Microarray Profiling Identifies the Activation of Oxidative Stress, Protein Ubiquitination, and Xenobiotic Metabolizing Pathways in MPCCs Exposed to Tolvaptan

Global gene expression profiling was used to identify gene and pathway-level changes contributing to the tolvaptan response. This approach provides an unbiased (ie, not restricted to known mechanisms of toxicity) assessment of the molecular stress response that may occur prior to changes observed at the cellular level (Ware et al., 2017). The 3T3-J2 murine embryonic fibroblasts only contributed nominally to the signal detected by the human-specific Affymetrix HT HG-U133+ PM peg arrays and did not influence the overall interpretation of differential hepatocyte gene expression (complete Methods and Results are described in Supplementary Information); therefore, total RNA from the MPCCs (including the fibroblasts) was used for microarray analysis. Differential expression was determined by linear contrasts between each tolvaptan concentration and the vehicle control at each time point of exposure (false discovery rate p < .05 and |fold change| > 1.5). A total of 3271 probe sets were differentially expressed between tolvaptan and vehicle control at ≥ 1 concentration and duration of exposure. The greatest numbers of differentially expressed probe sets were observed in response to 50 µM tolvaptan at 24 h (2542) and 72 h (789), followed by 20 µM tolvaptan at 24 h (352) and 72 h (135) (Figure 5A). Hierarchical clustering of all significant probe sets demonstrates time- and tolvaptan concentration-dependent responses, both up and down for these transcripts (Figure 5B). The majority of differentially expressed probe sets at 24 h were upregulated (2089/2542 at 50 µM, 82.2%), while the majority of probe sets at 72 h were downregulated (478/789 at 50 µM, 60.6%). The majority of differentially expressed probe sets overlapped between 20 and 50 µM tolvaptan at 24 h (220/352 at 20 µM, 62.5%) and between 20 and 50 µM tolvaptan at 72 h (93/135 at 20 µM, 68.9%). However, only a minority of differentially expressed probe sets were overlapping between 24 and 72 h at 20 µM tolvaptan (52/135 at 72 h, 38.5%) and between 24 and 72 h at 50 µM tolvaptan (195/789 at 72 h, 24.7%).

Figure 5.

Gene expression changes identified by microarray profiling of MPCCs exposed to tolvaptan. A, The number of probe sets significantly differentially expressed (false discovery rate p < .05 and |fold change| > 1.5) at each concentration of tolvaptan (0.01, 0.1, 1, 20, or 50 μM) and duration of tolvaptan exposure when compared with vehicle. B, Hierarchal clustering of all significant probe sets as a function of tolvaptan concentration and exposure time. The shading reflects the directionality (red, upregulated and blue, downregulated) and magnitude (darker, larger fold change) of the log₂(ratio) for the contrast.

Pathway enrichment analysis of genes that were differentially expressed between the highest concentrations of tolvaptan (50 µM) and vehicle control at each time point of exposure demonstrated temporally regulated transcriptional responses (Figure 6). At 4 and 24 h, the strongest pathway enrichment in response to 50 µM tolvaptan exposure was observed in the “Nrf2-mediated oxidative stress response,” “protein ubiquitination,” and “superpathway of cholesterol biosynthesis” pathways. However, at 72 h, stronger enrichment was observed in pathways containing xenobiotic metabolizing enzymes such as “PXR/RXR activation,” “Bupropion degradation,” and “acetone degradation.” The most significantly affected genes in these pathways were CYP450s, such as CYP2C9 which was increased 5.3-fold at 72 h (p = 1.62E-12) versus 4.1-fold (p = 9.74E-11) at 24 h, CYP3A4 which was increased 8.0-fold at 72 h (p = 4.85E-10) versus 4.6-fold (p = 2.91E-07) at 24 h, and CYP3A5 which was increased 4.6-fold at 72 h (p = 2.69E-06) but only 3.5-fold (p = 4.44E-05) at 24 h. Enrichment of “FXR/RXR activation” also was observed in response to 50 µM tolvaptan at 72 h. This pathway contains genes coding for proteins involved in the regulation of intercellular bile acid concentration such as organic anion-transporting polypeptides 1B1 and 1B3.

Figure 6.

Pathway enrichment analysis demonstrates temporally regulated transcriptional responses in MPCCs treated with tolvaptan. Comparison analysis of the top pathways (p < 1.0E-05) enriched among genes significantly differentially expressed in response to 50 μM tolvaptan at 4, 24, and 72 h. Significant differential expression was defined as false discovery rate p < .05 and |fold change| > 1.5. The shading reflects the significance (darker, lower P value) of the enrichment for each pathway. Pathways are ordered from top to bottom by lowest combined P value at each time point of exposure.

DISCUSSION

Significant evidence suggests that many forms of IDILI begin with drug-induced hepatocyte stress, release of danger signals, and activation of the innate immune system, all steps necessary but not sufficient to promote an adaptive immune attack on the liver (Cho and Uetrecht, 2017; Mosedale and Watkins, 2017). It also has been proposed that IDILI drugs may cause the release of danger signals from hepatocytes in the absence of hepatocyte necrosis (Mosedale and Watkins, 2017). In this study, this hypothesis was explored with tolvaptan, a drug that causes IDILI in patients with ADPKD.

An almost 8-fold increase in the release of hepatocyte-derived exosomal miR-122 was observed in response to tolvaptan in the absence of overt hepatocyte necrosis as assessed by biochemical markers such as ALT. Although other miRNAs may be released in exosomes, miR-122 was the focus of this study for the following reasons: (1) miR-122 is a marker for hepatocyte-derived exosomes (Holman et al., 2016) and (2) a recently published study has shown that exosomal transfer of miR-122 from ethanol-treated hepatocytes can inhibit antioxidant signaling and sensitize cells to LPS stimulation, causing increased release of proinflammatory cytokines from nonethanol-treated monocytes (Momen-Heravi et al., 2015). The authors conclude that this may be an important mechanism for immune cell activation and liver injury in response to acute and/or chronic alcohol consumption in vivo. It is possible that a similar mechanism promotes innate immune activation in the livers of patients treated with tolvaptan. Other nonexosomal danger signals such as the well-known damage associated molecular pattern molecule high mobility group box 1 and/or other miRNAs could also be involved in innate immune activation.

The most significant increase in exosomal miR-122 release was observed at the highest dose and longest duration of tolvaptan exposure. At this point, hepatocyte stress was advanced and programmed cell death was underway. The small but significant increase in cell membrane permeability also may indicate the onset of cell death more necrotic in nature. These observations are consistent with findings from clinical trials of the IDILI drug lumiracoxib. A retrospective analysis of archived DNA revealed a strong association between the HLA allele DQA1*0102 and the risk of transient elevations in serum ALT observed in some patients during clinical trials (Singer et al., 2010). However, the lumiracoxib association was stronger in the subset of patients who experienced higher peak serum ALT values relative to those with lower peak values, suggesting that some drug-induced liver injury was independent of an adaptive immune response. These clinical observations, together with our tolvaptan observations in vitro, suggest that cell stress sufficient to initiate an adaptive immune response occurs near a tipping point toward hepatocyte necrosis.

Early changes in exosome miR-122 were most associated with mitochondrial-induced apoptosis, of those tested by HCI. Microarray profiling also demonstrated tolvaptan-induced upregulation of genes involved in NRF2 response, protein ubiquitination, and cholesterol metabolism, all of which are indicative of oxidative stress (Gu et al., 2014; Ma, 2013; Malhotra and Kaufman, 2007). These results are supported by recent findings from HepG2 studies that also demonstrated mitochondrial-induced apoptosis and oxidative stress in response to tolvaptan exposure (Wu et al., 2015).

Statistically significant effects of tolvaptan on apoptosis and oxidative stress were observed at similar concentrations (between 20 and 50 µM) in both primary human hepatocytes in this study and HepG2 cells, as previously reported (Woodhead et al., 2017; Wu et al., 2015). These concentrations are much higher than the reported 1.7 µM median plasma C_max in ADPKD patients on a 90/30 mg split-dose regimen (Boertien et al., 2013), raising the question of whether or not the in vitro findings have in vivo relevance. The concentration of tolvaptan in the human liver is currently unknown and it is often suggested that when accounting for potential liver accumulation of a drug, interindividual variability, and first-pass exposure, concentrations as high as 100-fold plasma C_max should be tested (Atienzar et al., 2016). This threshold also doesn’t compromise the specificity of DILI assessment in vitro (Khetani et al., 2013; Ware et al., 2015; Xu et al., 2008), suggesting that it is a reasonable range to account for these factors. Furthermore, hepatocellular accumulation of tolvaptan (approximately 10-fold at 1.5 µM) has been reported in sandwich-cultured hepatocytes (Lu et al., 2016). However, in a clinical trial of healthy subjects, repeated exposure of tolvaptan at 60 mg/day did not affect the pharmacokinetics of R-warfarin, a CYP3A4 substrate (Shoaf and Mallikaarjun, 2012). Given that tolvaptan is > 98% protein bound (Furukawa et al., 2011) and in vitro concentrations of 20 and 50 µM in serum-free medium resulted in increased CYP3A4 mRNA this would suggest that the in vitro concentrations used in this experiment are likely higher than the liver concentrations achieved in ADPKD patients at the 90/30 mg dose. Nevertheless, increased mRNA levels don’t always correspond to changes in protein levels and/or activity and neither CYP3A4 induction nor warfarin metabolism/clearance were evaluated here.

Regardless, the use of in vitro concentrations above plasma C_max and/or liver concentrations achieved in ADPKD patients at the 90/30 mg dose may still be relevant. First, the tolvaptan and warfarin interaction study was conducted in a small trial (N = 24) of healthy subjects which may not accurately reflect variability in tolvaptan pharmacokinetics among ADPKD patients (Woodhead et al., 2017). Second, all experiments in this study were performed in hepatocytes from random healthy donors. The observation that IDILI due to tolvaptan has not been reported outside the ADPKD patient population supports the possibility that patient-specific susceptibility factors required for the tolvaptan response may not be represented among the donors selected for this study. For example, ADPKD is associated with increased bile acid concentrations in liver cyst fluid (Munoz-Garrido et al., 2015) and impaired mitochondrial function (Hajarnis et al., 2017; Ishimoto, 2016), suggesting there may be a predisposition of ADPKD patients to tolvaptan-induced liver injury by these mechanisms. Both mechanisms have been previously implicated in tolvaptan hepatotoxicity (Mosedale et al., 2017; Woodhead et al., 2017), and the transcriptional enrichment of “FXR/RXR activation” and decreased urea production observed in response to tolvaptan treatment here support these findings. Furthermore, the risk of toxicities by these mechanisms would increase with disease progression, which has been hypothesized to explain DILI susceptibility observed > 1 year into treatment (Watkins et al., 2015; Woodhead et al., 2017). As a result, it may be necessary to increase in vitro exposures beyond what is pharmacologically relevant to achieve relevant responses (susceptibility) in healthy donors. Finally, we conducted relatively short-term treatments (< 72 h) with no medium change and no repeat dosing to accommodate the use of exosome release as an endpoint. Long-term exposures of IDILI compounds with repeat dosing result in a leftward shift of the EC₅₀ for toxicity (Bell et al., 2016, 2017). Therefore, it is possible if longer exposures and repeat dosing were used in this study, responses would be seen at lower concentrations.

In conclusion, at concentrations approximately 30-fold above median plasma C_max tolvaptan-induced exosome release of miR-122 from primary human hepatocytes in the absence of overt necrosis, providing the first direct demonstration of this event with a drug capable of causing IDILI. The early exosomal miR-122 release was associated with mitochondrial-induced apoptosis and oxidative stress. It is likely that tolvaptan-induced hepatocyte stress in combination with the release of exosomal miR-122 and probably other danger signals helps promote an innate immune response. In susceptible individuals, these early events should set up the liver for an adaptive immune attack that ultimately results in clinically significant liver injury. Because miR-122 is a marker for hepatocyte-derived exosomes (Holman et al., 2016), and because hepatocyte-derived exosomes can reach systemic circulation (Wetmore et al., 2010), serum levels of exosomal miR-122 may provide a minimally invasive, nongenetic, and mechanistic biomarker for an early liver response to tolvaptan that could be used to identify patients susceptible to IDILI in the clinic.

FUNDING

This work was supported by Otsuka Pharmaceutical Development & Commercialization, Inc.

SUPPLEMENTARY DATA

Supplementary data are available at Toxicological Sciences online.