Williams Medium E, penicillin, streptomycin, L-glutamine, and HEPES were purchased from Invitrogen (Carlsbad, CA). Insulin (Humulin®) was purchased from Eli Lilly (Indianapolis, IN), and calf serum was obtained from HyClone Laboratories (Logan, UT). Tissue culture dishes were from Corning Glass Works (Corning, NY).

This study was carried out in accordance with the recommendations of the National Institutes of Health. The protocol was approved by the Institutional Animal Care and Use Committee of the University of Pittsburgh. The liver cell isolation and culture procedures were as follows:

-Primary Hepatocytes (HC) were harvested from adult C57BL/6 male mice (Charles River Laboratories, Wilmington, MA). Cells were isolated by collagenase perfusion using the method of Seglen (1976) and purified to >98% purity by repeated centrifugation at 50 g, followed by further purification over 30% Percoll. Viability at time of plating was checked by trypan blue exclusion. Highly purified HC (>98% purity and >95% viability by trypan blue exclusion) were suspended in Williams' E medium supplemented with 10% heat-inactivated calf serum, 15 mM HEPES (pH 7.4), 16 units insulin, 2 mM L-glutamine, 100 units/ml penicillin, and 100 μg/ml streptomycin. The cells were then plated on collagen-coated cell culture dishes (3 × 10⁶ cells/6-cm dish) or plates (250,000 cells/well in 6-well plates) and cultured overnight at 37°C. The old medium was then removed and cells were further incubated with fresh media containing 5% heat-inactivated calf serum under 21% O₂ (C, Control), 6 h hypoxia (H) or 6 h hypoxia followed by reoxygenation as previously described (Metukuri et al., 2009). Hypoxic conditions were obtained by placing the cells into a modular incubator chamber (Billups-Rothenburg, Del Mar, CA) flushed with a hypoxic gas mixture containing 1% O₂, 5% CO₂ and 94% N₂. Duplicate hypoxic cultures were returned to 21% O₂ for reoxygenation overnight (18 h). At the end of the experiments, both cell culture media (CCM) and whole cell lysate (WCL) were collected and stored at −80°C until analysis. Total protein isolation and determination was done using the BCA protein assay kit from Pierce (Rockford, IL) with bovine serum albumin as standard as previously described (Metukuri et al., 2009). All data were normalized and the final mediator concentrations were expressed in pg/mg total protein. For data analysis and unless otherwise indicated, the number of independent experiments (n) refers to the number of separate mice from which HC or HSC were harvested. Experimental data are presented as mean ± SEM.

-Primary Hepatic Stellate Cells (HSC) were isolated as previously described (Dangi et al., 2012). Briefly, mouse livers from C57BL/6 mice were perfused in situ through the inferior vena cava with 30–40 ml HBSS (without Ca²⁺), then digested with 30–40 ml HBSS (with Ca²⁺) containing collagenase type IV (0.25 mg/ml) (Worthington, Lakewood, NJ) and protease (0.50 mg/ml) (Sigma, St. Louis, MO). The cell suspension was filtered through 100 μm nylon mesh. Following removal of HC and cell debris by low speed centrifugation (50 g; 2 min), HSC were purified by Histodenz density gradient centrifugation, and suspended in DMEM containing 100 U/ml penicillin, 100 μg/ml streptomycin, 10% FBS and 10% horse serum. Cells were seeded in gelatin (0.1% in PBS)-coated plates at a density of 0.5 × 10⁶/cm² and 20 min later, loosely adherent HSC were harvested and re-seeded in new 6 or 96-well flat-bottom plates. Cell purity, as assessed by vitamin A autofluorescence, desmin and glial fibrillary acidic protein (GFAP) immunostaining (Sumpter et al., 2012), was >98%. HSC were cultured under 21% O₂, 6 h hypoxia or 6 h hypoxia followed by reoxygenation for 18 h, then harvested for protein isolation and analysis as described above for HC.

Mouse inflammatory mediators were measured using a Luminex® 100 IS apparatus (Luminex, Austin, TX) and the BioSource 20-plex™mouse cytokine bead kit (BioSource-Invitrogen, San Diego, CA) as per manufacturer's specifications. The antibody bead kit included: Basic fibroblast growth factor (FGF basic), Granulocyte-Macrophage Colony-Stimulating Factor (GM-CSF), Interferon-γ (IFN-γ), Interleukin (IL)-1α, IL-1β, IL-2, IL-4, IL-5, IL-6, IL-10, IL-12 (both p40 and p70 subunits), IL-13, IL-17A, Interferon-γ-inducible Protein 10 (IP-10/CXCL10), Keratinocyte-derived Cytokine (KC/CXCL1), Monocyte Chemoattractant Protein-1 (MCP-1/CCL2), Monokine induced by Interferon-γ (MIG/CXCL9), Macrophage Inflammatory Protein-1α (MIP-1α/CCL3), Tumor Necrosis Factor-α (TNF-α), and Vascular Endothelial Growth Factor (VEGF).

We applied a series of statistical and data-driven modeling techniques aimed at discovering principal drivers, interconnected networks, and potential key regulatory nodes of inflammation in HSC vs. HC following increasing levels of cellular stress. We applied these methods in a stepwise fashion based on our concept of the way liver cells respond to inflammatory stimuli (Figure 1, adapted from Namas et al., 2015). Importantly, we also sought to address intracellular information processing (“thinking”) as well as extracellular inflammatory mediator production (“talking”) carried out by each cell type individually, as well as inferring how these two processes may interact in each cell type. In brief, we hypothesize that low-level stresses associated with cell culture, as well as more stressful stimuli such as hypoxia with or without reoxygenation, will cause liver cells to produce a panoply of inflammatory mediators in order to restore homeostasis. We hypothesize that this dynamic production of inflammatory mediators takes the form of inflammatory networks, and thus can be elucidated using network inference algorithms. As these early inflammatory responses progress, and in the face of continued hypoxia or the onset of reoxygenation, these networks coalesce around an intermediate, core set of inflammation-associated cellular functions, which we hypothesize can be inferred via PCA and related methods. These intermediate processes evolve to form a differential set of inflammatory responses characterized by distinct mediators that characterize the autocrine response of each liver cell type. These differential responses can be discerned both by standard statistical approaches such as Two-way ANOVA as well as data-driven modeling tools such as hierarchical clustering (Figure 1). Accordingly, we followed the stepwise series of analyses detailed below:

The production of inflammatory mediators in both CCM and WCL under the three experimental conditions, along with the corresponding P-values for comparison of the changes in HSC vs. HC by Two-Way ANOVA, can be found in Supplementary Figure 1. Key examples of inflammatory mediators whose intracellular and extracellular levels are significantly different between HSC and HC are shown in Figure 2. In general, on a pg protein per mg total cell protein basis, HSC both produced and secreted much larger quantities of inflammatory mediators than the rather indolent response of HC. To identify those inflammatory mediators that showed similar expression or secretion behavior in each cell type, these patterns were then compared and grouped using hierarchical clustering. The resulting dendrograms showed a clear cluster separation between HC and HSC both in the CCM (Figure 3A) and WCL (Figure 3B) under the three experimental conditions studied. From this analysis, we infer that there is more heterogeneity in the intracellular (“thinking”) vs. the extracellular (“talking”) levels of inflammatory mediators in both cell types.

Next, we sought to define an intermediate, core set of inflammation-associated characteristics of HSC and HC in response to increasing levels of stress, which we hypothesize represent different inflammation-related functions that ultimately become manifest in the expression and secretion of the defined subset of inflammatory mediators studied. To do so, we utilized PCA (Figure 1), a technique (or variants thereof) used by multiple investigators to delineate the core characteristics of a multivariate biological response (Janes et al., 2005; Janes and Yaffe, 2006; Mi et al., 2011; Nieman et al., 2012; Azhar et al., 2013; Ziraldo et al., 2013). We assessed several properties of the inflammatory responses under control, hypoxia, and hypoxia/reoxygenation: (1) the total number of mediators contributing to the total variance (i.e., mediators with sums of eigenvalues > 0; (2) the number of inferred principal components (represented as stacked bars), which we hypothesize represent distinct characteristics or inflammation-related functions; and (3) the sums of the eigenvalues (amplitude) for each experimental condition, which we hypothesize represents the magnitude of the inflammatory response to the indicated level of stress. The number of mediators and number of principal components expressed by each cell type under each condition is shown in Figure 4. An overall comparison of the inflammatory mediators showed that the inflammatory response of HSC was clearly different from that in HC both in CCM (Figure 4A) and WCL (Figure 4B) under each experimental condition. But certain similarities were seen. The major similarity was that the amplitude of the response was greatest when both HSC and HC were subjected to hypoxia/reoxygenation. Moreover, in general, the number of mediators involved as well as the inferred functions (number of principal components) in both WCL (“thinking”) and CCM (“talking”) were lowest in both HSC and HC subjected to hypoxia/reoxygenation as compared to control or hypoxia. A notable exception was the “thinking” response in HC WCL (Figure 4B), in which all three experimental conditions were characterized by two principal components, and the overall number of mediators was least in the hypoxia setting. Together, these analyses suggest that above a certain threshold of cellular stress, the inflammatory response generally becomes focused on a limited number of functions and associated mediators at a high intensity of response.

The final analysis we carried out was aimed at discovering the initial networks of inflammation induced under control conditions, hypoxia, or hypoxia/reoxygenation (Figure 1). We have previously employed Dynamic Network Analysis (DyNA) to define the dynamic interconnections among inflammatory mediators in a mouse model of trauma/hemorrhagic shock (Mi et al., 2011), in isolated mouse HC subjected to hypoxic stress (Ziraldo et al., 2013), and to define the dynamics of systemic inflammation after trauma (Abboud et al., 2016; Namas et al., 2016) or during acute liver failure (Zamora et al., 2016). For the current analyses, we used a modified version of this method, in which data from a single experimental condition (rather than a time interval) were used.

This network analysis suggested the presence of more complex extracellular inflammatory networks in HSC than in HC under all experimental conditions (Figures 5A,B and Supplementary Figure 2A) but roughly similarly complex intracellular networks (Figures 5C,D and Supplementary Figure 2B). Despite this difference in network complexity, multiple, shared inflammatory mediators were involved in both “thinking” and “talking” in HSC and HC. Secreted mediators shared by networks in both HSC and HC under control conditions included VEGF, MCP-1, KC, and IL-17A. Intracellularly under control conditions, only MIG and IL-4 were shared by HSC and HC networks. Under hypoxia, intracellular HSC and HC networks shared the mediators TNF-α, KC, MIG, and MCP-1, while intracellular networks only shared the mediators MIP-1α and IP-10. Under hypoxia/reoxygenation, secreted HSC and HC networks shared the greatest number of mediators: IL-1α, IL-2, IL-4, IL-6, IL-17A, MIG, KC, IP-10, GM-CSF, VEGF, and TNF-α; in contrast, shared intracellular mediators included TNF-α, MIG, MCP-1, IL-1α, and IL-17A. Control experiments comparing single network analysis of both isolated HC and HSC cultured under 21% O₂ for 24 h confirmed the presence of more complex extracellular inflammatory networks in HSC than in HC under both control and hypoxia/reoxygenation (Supplementary Figure 3A) and similar complex intracellular networks after hypoxia/reoxygenation (Supplementary Figure 3B).

Based on PCA (Figure 4), we hypothesized that network interactions would lead to a reduced set of principal mediators and inflammatory functions in both CCM and WCL of HSC as they transition from hypoxia to hypoxia + reoxygenation. In support of this hypothesis, the networks observed under hypoxia/reoxygenation in HSC CCM as well as WCL exhibited multiple anti-correlated nodes (indicated in red; Figures 5A,C). We interpret this finding to mean that both intracellular and extracellular HSC mediators induced in the setting of hypoxia followed by reoxygenation regulate each other in a negative fashion that leads ultimately to the evolution of a reduced number of inflammatory functions relative to hypoxia alone. In the extracellular milieu of HSC, the chemokine MIG; the cytokines TNF-α, IL-6, and IL-17A; and the hypoxia-inducible growth factor VEGF appear to coordinate some of these downregulatory responses (Figure 5A). VEGF, TNF-α, MCP-1, and IL-17A also appear to play a more limited negative intracellular role (Figure 5B).

The foregoing analyses suggested that there may be an interplay between intracellular responses (“thinking”) and extracellular responses (“talking”) in each cell type. We therefore next sought to use network analysis to gain insights into the autocrine information flow within each cell population. Accordingly, we repeated the network analysis, but this time using both CCM and WCL data together for HSC vs. HC (Figure 6). Surprisingly, this analysis suggested the presence of multiple negative interactions in control HSC. This analysis also inferred the presence of a large number of negative interactions in both HSC and HC exposed to hypoxia/reoxygenation (Figure 6A,Table 1). Network analysis also demonstrated a higher complexity of autocrine networks involving both intracellular (“thinking”) and extracellular (“talking”) in HSC as compared to HC (Figure 6B), similar to the results obtained separately in CCM and WCL (Figures 5B,D).

We next assessed the flow of information from the intracellular (WCL) to extracellular milieu (CCM) in HSC vs. HC in greater detail (Figure 6 and Supplementary Table 1) Under control conditions, inflammatory networks in HC were comprised solely of extracellular mediators (IL-1β, IL-13, IL-17A, MCP-1, KC, VEGF, and TNF-α). In contrast, the combined network analysis inferred multiple positive and negative connections among a large number of intracellular and extracellular inflammatory mediators in HSC, suggesting a much more robust “thinking” and “talking” interplay in this cell type at baseline. Under hypoxia alone, both HSC and HC exhibited only positive, extracellular inflammatory connectivity, though the number of connections was higher in HSC (a large number of mediators) as compared to HC (only TNF-α, IL-17A, KC, MCP-1, and MIG). Interestingly, when subjected to hypoxia/reoxygenation, both HSC and HC exhibited a large number of positive and negative connections in the intracellular as well as extracellular compartments. This suggests a major adaptation to stress in HC, while this phenotype in HSC remained qualitatively similar to that observed under control conditions.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.