Insulin resistance disrupts epithelial repair and niche-progenitor Fgf signaling during chronic liver injury

Insulin provides important information to tissues about feeding behavior and energy status. Defective insulin signaling is associated with ageing, tissue dysfunction, and impaired wound healing. In the liver, insulin resistance leads to chronic damage and fibrosis, but it is unclear how tissue-repair mechanisms integrate insulin signals to coordinate an appropriate injury response or how they are affected by insulin resistance. In this study, we demonstrate that insulin resistance impairs local cellular crosstalk between the fibrotic stroma and bipotent adult liver progenitor cells (LPCs), whose paracrine interactions promote epithelial repair and tissue remodeling. Using insulin-resistant mice deficient for insulin receptor substrate 2 (Irs2), we highlight dramatic impairment of proregenerative fibroblast growth factor 7 (Fgf7) signaling between stromal niche cells and LPCs during chronic injury. We provide a detailed account of the role played by IRS2 in promoting Fgf7 ligand and receptor (Fgfr2-IIIb) expression by the two cell compartments, and we describe an insulin/IRS2-dependent feed-forward loop capable of sustaining hepatic re-epithelialization by driving FGFR2-IIIb expression. Finally, we shed light on the regulation of IRS2 and FGF7 within the fibrotic stroma and show—using a human coculture system—that IRS2 silencing shifts the equilibrium away from paracrine epithelial repair in favor of fibrogenesis. Hence, we offer a compelling insight into the contribution of insulin resistance to the pathogenesis of chronic liver disease and propose IRS2 as a positive regulator of communication between cell types and the transition between phases of stromal to epithelial repair.


Introduction
Successful wound healing requires coordination between stromal and epithelial cell compartments. Stromal activation lays the groundwork for epithelial repair, producing the appropriate microenvironment and growth factors to facilitate proliferation and remodeling of epithelia [1]. Metabolic disease is associated with a spectrum of chronic comorbidities, including cardiovascular disease and liver disease, as well as defects in wound healing [2,3]. However, it remains unclear whether insulin resistance affects injury-repair mechanisms in target organs such as the liver, which are otherwise at the forefront of insulin's metabolic actions. The liver parenchyma is highly regenerative and can undergo dramatic tissue remodeling to maintain parenchymal function in the face of chronic injury. Such remodeling is shaped in part by the activation of perisinusoidal cells, such as hepatic stellate cells (HSCs), and periportal mesenchymal cells, such as portal fibroblasts (PFs) [4], that expand to produce a fibrotic milieu capable of directing epithelial repair but that also contribute to long-term risk of scarring/fibrosis and hepatic dysfunction. To date, the impact of insulin resistance on the fibrotic stroma, including how it affects the ability of mesenchymal cells to communicate repair signals to the hepatic epithelia, remains unknown [5].
In this study, we investigate how insulin resistance affects stromal-epithelial repair mechanisms in the liver during chronic injury by knockout of insulin receptor substrate 2 (Irs2), a key adaptor protein that couples the insulin and insulin-like growth factor 1 (IGF-1) receptors to intracellular signaling pathways. Irs2 −/− mice have normal liver development but severe peripheral insulin resistance that leads to late-onset type II diabetes [6]. IRS2 is the principle regulator of insulin sensitivity in hepatocytes [7], cooperating closely with IRS1 to mediate the metabolic response to feeding [8]. Aberrant IRS2 expression has been associated with hepatic insulin resistance [9] and progression of chronic liver diseases, including nonalcoholic steatohepatitis (NASH) [10], hepatitis C [11], and hepatocellular carcinoma [12], in which increased IRS2 expression is associated with proliferation, increased cell survival, and disruption of cellfate signals controlling hepatocyte metaplasia and the expansion of bipotent liver progenitor cells (LPCs) [13,14]. Nevertheless, it remains to be established whether IRS2 plays any role in hepatic wound healing.
Fibroblast growth factor 7 (FGF7) is an important paracrine regulator of tissue morphogenesis during development [15,16] and during re-epithelialization of cutaneous lesions [17]. In the liver, FGF7 is expressed by HSCs [18] and Thy1 T-cell surface antigen (Thy1)-expressing PFs [4,19] that produce stromal niche signals to drive epithelial remodeling in response to chronic injury. Fgf7-knockout mice have reduced survival due to liver failure when fed a hepatotoxic diet containing 0.1% 3.5-diethoxycarbonyl-1.4-dihydrocollidine (DDC) because they are unable to support the expansion of adult liver stem cells/LPCs required for the adaptive response to injury [19,20]. LPCs are bipotent epithelial precursors capable of differentiating into cholangiocytes or parenchymal hepatocytes [21]. During DDC liver injury, they proliferate to form duct-like structures within periportal tracts, surrounded by fibrotic stromal cells expressing Fgf7. LPCs express Fgfr2-IIIb [19], the receptor for Fgf7, which is exclusively expressed in epithelia. Amplification of Fgfr2-IIIb occurs in LPCs within the expanding nonparenchymal cell (NPC) fraction during DDC injury [22], and potentiation of Fgfr2-IIIb signaling by overexpression of Fgf7 [19] or Fgf10 [22] drives the dramatic expansion of LPCs and immature hepatocytes within the liver. Fgf7/Fgfr2-IIIb interactions therefore communicate proregenerative signals between stroma and epithelia, with a range that is strictly limited to local paracrine effects because of the high affinity of Fgf7 for heparin-sulphate proteoglycans in the extracellular matrix and neighboring cells [23].
Usually, insulin signals are systemic, but in model organisms such as Drosophila, they also coordinate short-range communication between niche cells and stem cells [24]. This allows for tissue-specific responses to changing environmental conditions [25]. In this study, we show that integration of insulin/IRS2 signals by HSCs and LPCs promotes paracrine crosstalk between the fibrotic stroma and LPCs via Fgf7. We demonstrate that loss of Irs2 has a negative impact on hepatic wound healing, reducing the capacity of HSCs and LPCs to produce and respond to Fgf7 respectively.

Results
The response to DDC liver injury is blunted in Irs2 −/− mice We examined how Irs2 deletion affected the chronic liver-injury response in adult mice during 0.1% DDC feeding, in which paracrine Fgf7 signaling plays a central role in coordinating the LPC response and epithelial repair (Fig 1A). Transient induction of Irs2 occurred in wild-type (WT) livers on day 7 of the DDC diet (Fig 1B), correlating with a peak in liver injury as judged by serum aspartate/alanine transaminase (AST/ALT) activity ( Fig 1C). No differences in serum bilirubin or liver tissue bile acid levels were observed between the two groups ( Fig 1D), and initial induction of AST/ALT was comparable on day 7, suggesting the cholestatic effects of the DDC diet were equivalent in WT and Irs2 −/− mice. However, past day 7, serum transaminases were reduced in controls (days [14][15][16][17][18][19][20][21], whereas in the Irs2 −/− mice, they remained significantly elevated (Fig 1C), suggesting a role for Irs2 in the attenuation of liver damage.

Altered profile of stromal activation in Irs2 −/− mice highlights aberrant response by Fgf7-expressing mesenchymal cells
In rodent and human livers, Fgf7 is expressed by HSCs during injury [18]. However, in the DDC mouse model, resident Thy1+ PFs have also been identified as important Fgf7-expressing mesenchymal cells and first responders to cholestatic liver damage [4,19]. Profiling of stromal gene expression in WT mice during DDC injury revealed sustained induction of PF/ myofibroblast genes Thy1, Elastin, Vimentin, and alpha-smooth muscle actin (αSma/Acta2), together with profibrogenic gene Connective tissue growth factor (Ctgf) (Fig 3A), whereas in Irs2 −/− mice, the activation profile of these genes was either blunted (Thy1, Elastin), less sustained (Acta2), or down-regulated from day 7 (Vimentin, Ctgf, and Thy1), suggesting a negative impact on the PF/myofibroblast population within the activated stroma. Interestingly, we also observed a decline in expression of the HSC marker glial fibrillary acidic protein (Gfap) from day 7 (Fig 3A), coincident with significant reductions in mesenchymal marker Vimentin and in Ctgf, a regulator of epithelial mesenchymal transition. From day 7, a significant decrease in Gfap immunostaining in Irs2 −/− livers was also observed (Fig 3B and 3C). This reduction in Gfap+ cells was apparent throughout the parenchyma and in periportal areas where it colocalized with Fgf7-expressing stroma ( Fig 3C).
In contrast with the results of messenger RNA (mRNA) analysis (Fig 3A), immunostaining for Thy1, Elastin, and αSMA tended towards increased expression in Irs2 −/− livers on day 21, while mesenchymal intermediate filament protein vimentin was similarly induced (S4 Fig). Thy1+ stroma were observed in close contact with Spp1+/Epcam+ LPCs in both WT and Irs2 −/− mice as previously described [19] (S5A Fig), and a similar relationship between αSmaexpressing myofibroblasts and Spp1+ LPCs was also apparent in both groups (S5B Fig), indicating that failure to induce Fgf7 was not due to the lack of myofibroblast-like cells in the LPC niche of Irs2 −/− livers per se. In contrast, lack of contact between Gfap+ HSCs and LPCs was observed in Irs2 −/− livers (S5B Fig). Hence, failure to induce Fgf7 coincided with an overall loss of Gfap+ HSCs and loss of association between HSCs and LPCs in portal tracts, whereas we were unable to confirm that stromal PFs were negatively influenced by Irs2 deletion. However, increased Thy1 staining in Irs2 −/− livers was partially attributable to the increase in markers of bone-marrow-derived cells; cd45/protein tyrosine phosphatase receptor type C gene   suggesting that Thy1+ cells from the bone marrow, whose expression of Fgf7 is lower than that of resident Thy1+ fibroblast populations [4], played a more significant role in the altered stromal response to DDC injury in Irs2 −/− mice.

Survival of Fgf7-expressing HSCs is Irs2 dependent
We assessed the survival of Fgf7-expressing stroma during the early phase of DDC injury, when Irs2 was maximally expressed in WT livers (day 7) and before significant loss of Gfap + HSCs was observed. Analysis of cleaved caspase 3 (c-Casp3) staining revealed a 3.0-fold (P = 0.0395) increase in apoptosis in Fgf7-expressing cells in Irs2 −/− livers at this time, confirming accelerated loss of this subpopulation within the fibrotic stroma during injury ( Fig 4A). In order to test the hypothesis that IRS2 promoted survival of Fgf7-expressing stroma, we performed stable knockdown of IRS2 in the human HSC (hHSC) cell line LX-2 using lentiviral short hairpin RNA (shRNA) (Fig 4B and 4C). Silencing of IRS2 had no direct impact on LX-2 cell viability, Fgf7 expression, or fibrogenic gene expression under standard culture conditions (S7 Fig). However, exposure to cytotoxic alkylating agent mitomycin C (MitoC) resulted in cell-cycle arrest and apoptotic cell death, during which IRS2 promoted survival ( Fig 4B) by significantly reducing P53 expression and caspase 3 cleavage (Fig 4C). We confirmed a positive role for IRS2 in protecting HSCs from apoptosis by treating primary hHSC cultures with an allosteric inhibitor of IRS proteins (NT-157). NT-157 triggers proteasomal depletion of IRS1/ IRS2 [29], which resulted in dramatic activation of P53/caspase 3 cleavage and apoptosis in hHSCs ( Fig 4D). We therefore concluded that IRS2 plays an indirect role in driving Fgf7 expression during DDC injury by protecting HSCs from apoptosis.

Irs2 promotes epithelial sensitivity to Fgf7 and hepatic re-epithelialization during injury
In addition to the defect in stromal Fgf7 expression, Irs2 deletion also had a negative impact on sensitivity to Fgf7 in epithelial cells. Stromal Fgf7 drives re-epithelialization during skin wound healing, signaling exclusively to cells expressing the epithelial isoform of Fgfr2 (Fgfr2-IIIb), which is highly expressed by LPCs during DDC feeding [19]. In Irs2 −/− livers, we noted a sharp decline of Fgfr2-IIIb expression in the latter stages of DDC feeding (days [14][15][16][17][18][19][20][21], which contrasted with the steady increase in Fgfr2-IIIb observed in the WT group ( Fig 5A). In WT mice, we observed increased antibody staining for Fgfr2 in periportal ducts surrounded Quantification of Gfap immunostaining in the livers of WT and Irs2 −/− mice during the time course of DDC injury (n = 3-4). (C) Loss of Gfap + HSCs in Irs2 −/− mice parallels the failure to induce Fgf7 in periportal areas. Confocal immunofluorescence images showing reduced Gfap/Fgf7 double staining in Irs2 −/− mice after 14 days of DDC feeding (day 14). Data information: underlying data are available in S1 Data. Data are presented as mean + SEM. � P < 0.05, �� P < 0.01, and ��� P < 0.001. (A) Two-way ANOVA was used to compare means. Significance P values were calculated using Tukey's multiple comparison test. (B) Unpaired Student t test. Acta1, alpha-smooth muscle actin; Ctgf, Connective tissue growth factor; DDC, 3.5-diethoxycarbonyl-1.4-dihydrocollidine; Fgf7, fibroblast growth factor 7; Gfap, glial fibrillary acidic protein; HSC, hepatic stellate cell; Irs2, insulin receptor substrate 2; mRNA, messenger RNA; PF, portal fibroblast; RT-qPCR, reverse transcriptase-quantitative PCR; RU, relative unit; WT, wild type.   (Fig 5B). No such induction was observed in Irs2 −/− livers, consistent with a failure of FGFR2-IIIb expression in LPCs. We confirmed that diminished Fgfr2-IIIb expression resulted in reduced Fgf7-sensitivity in injured Irs2 −/− livers by testing the signaling response to intraperitoneal (i.p.) injection of recombinant Fgf7 (rFgf7) protein. In WT mice, rFgf7 robustly activated Fgf7/Fgfr2 signaling as assessed by western blot, demonstrating increased in vivo phosphorylation of extracellular signal-regulated kinases (ERK)1/2 in whole-liver lysates ( Fig 5C). This response to rFgf7 was abrogated in the Irs2 −/− livers, indicating that they were refractory to Fgf7. Taken together, our findings suggested that Irs2 played a dual role in (i) promoting stromal expression of Fgf7 and (ii) sustaining sensitivity of the hepatic epithelia to Fgf7 during chronic DDC liver injury by promoting Fgfr2-IIIb expression ( Fig 5D). Interestingly, the timing of Fgfr2-IIIb loss in Irs2 −/− mice (days 14-21) corresponded with a period during which the parenchymal architecture failed to re-epithelialize, as judged by quantitative analysis of β-catenin immunostaining in hepatocytes (Fig 5E), suggesting that in the later stages of DDC injury, Irs2 was required upstream of Fgf7-Fgfr2-IIIb signaling to driving adaptive epithelial remodeling.

Insulin/IRS2 signaling and FGF7 synergize to promote FGFR2-IIIb expression and the emergence of hepatocytic epithelia in bipotent human HepaRG cells
We examined the possibility that IRS2 played a direct role in regulating LPC sensitivity to FGF7 and epithelialization using a human model of bipotent adult liver progenitor cells (HepaRG) [30]. Upon reaching confluence, HepaRG cells spontaneously differentiate in media supplemented with insulin to produce phase-bright epithelial "islands" of hepatocytelike cells ( Using this model, we found that IRS2 was induced prior to FGFR2-IIIb and hepatocyte-specific genes such as APOA2 (Fig 6B). Using a human IRS2-promoter construct to visualize the pattern of IRS2 expression via green fluorescent protein (pIRS2-GFP), we also demonstrated colocalization of IRS2 with albumin and FGFR2 in the early stages of hepatocyte island formation ( Fig 6C). These data were consistent with an early role for IRS2 in the patterning of FGF7 sensitivity during LPC differentiation.
Removal of insulin from the media (S8B and S8C Fig Fig 6E). Importantly, the formation of epithelial islands and FGFR2-IIIb were restored by exogenous expression of a mouse Irs2 not targeted by the human IRS2 shRNA construct (Fig 6D and 6E); hence, we confirmed a cell-autonomous role for IRS2 in promoting FGFR2-IIIb expression and differentiation in LPC-like cells.
Culturing of HepaRG cells with recombinant human FGF7 (rhFGF7) accelerated the formation of islands and increased hepatocyte differentiation, as determined by increased albumin/HNF4α and E-cadherin (E-cad) immunostaining, human APOA2-promoter (pAPOA2)-GFP reporter activity (Fig 6F), and analysis of hepatocyte-specific gene expression ( Fig 6G). This occurred downstream of a rapid (3-6 h) and sustained induction in osteopontin/SPP1 (S9 Fig), consistent with our in vivo observation that osteopontin/Spp1 and Fgf7 were both impaired in Irs2 −/− livers during injury (Fig 2). Importantly, the ability of rhFGF7 to promote expression of osteopontin/SPP1, as well as hepatocyte-specific genes such as albumin and APOA2, was both insulin and IRS2 dependent (Fig 6G), revealing synergy between these two pathways in driving the differentiation of bipotent progenitor cells. Stable silencing of IRS2 or omission of supplemented insulin from the medium was sufficient to abrogate the positive long-term effects of rhFGF7 on the hepatogenic program, thus demonstrating a functional requirement for insulin/IRS2 upstream of paracrine FGF7 signaling. Interestingly, rhFGF7 treatment also increased FGFR2-IIIb expression, highlighting the potential for positive feedback between ligand and receptor-an effect that was also insulin/IRS2 dependent ( Fig  6G) and hence reinforced the conclusion that insulin/IRS2 signaling in LPC-like cells primed FGF7 sensitivity.
Our data supported the hypothesis that insulin/IRS2 signaling was functionally upstream of FGF7-mediated epithelialization of LPCs and highlighted a cell-intrinsic role for IRS2 in promoting sensitivity to paracrine FGF7 produced by the fibrotic stroma. However, they also revealed that loss of stromal FGF7 could by itself be sufficient to explain a decline in FGFR2-IIIb expression and epithelialization in a non-cell-autonomous manner as a consequence of reduced positive feedback between ligand and receptor.

Insulin/IRS2 resistance in fibrotic stroma limits hepatocyte differentiation and FGFR2-IIIb expression in cocultured LPCs
Our results highlighted the potential for a paracrine feed-forward loop in which FGF7 drove LPC differentiation and FGFR2-IIIb expression, potentiating both epithelialization and FGF7 sensitivity in an insulin/IRS2-dependent manner. To test this hypothesis further and to more rigorously address the role of insulin/IRS2 signaling in the fibrotic stroma, we performed coculture experiments with hHSCs and HepaRG cells, seeding the two cell types together in order to model the intimate heterotypic cell-cell interactions observed between LPCs and their stromal niche in periportal tracts during DDC injury ( Fig 7A). LX-2 control (sh-luc) and IRS2-knockdown (sh-IRS2) cells were used to simulate "normal" versus "IRS2-deficient" fibrotic stroma and gauge how stromal insulin/IGF-1 resistance influenced the LPC response in a non-cell-autonomous manner. Irs2 is required to sustain hepatic Fgfr2-IIIb expression and re-epithelialization during DDC injury. (A) RT-qPCR analysis of Fgfr2-IIIb expression in whole-liver mRNA from WT and Irs2 −/− mice during a time course of DDC feeding (n = 6-8). (B) Immunofluorescence images of WT and Irs2 −/− livers after 14 and 21 days of DDC feeding. Induction of Fgfr2 occurred within duct-like structures in periportal areas ( � ) surrounded by Fgf7-expressing stroma (s). The failure to sustain Fgf7 expression in Irs2 −/− mice was paralleled by concomitant failure to induce Fgfr2 in LPCs (arrows). (C) The livers of Irs2 −/− mice are refractory to exogenous Fgf7 stimulation following DDC injury. Western blot analysis of whole-liver extracts from DDC-treated WT and Irs2 −/− mice (day 21). 30 minutes prior to being killed, mice were administered rFgf7 or vehicle by i.p. injection. P-Erk levels relative to total Erk were quantified and are shown in the graph above (n = 4-7), and representative images of blots are shown below. (D) Schematic summarizing the observed negative impact of Irs2 deletion on Fgf7 signaling between the hepatic stroma and LPCs during DDC injury. P = portal vein, A = hepatic artery, C = central vein. (E) Immunohistochemical staining for β-catenin in WT and Irs2 −/− liver parenchyma during a time course of DDC injury. Modal pixel intensity was graphed and used to quantify the dramatic remodeling of β-catenin staining (days 0-14) and reestablishment of epithelial architecture in livers of WT but not Irs2 −/− mice in the latter period of the diet (days 14-21) (n = 6-9). Data information: underlying data are available in S1 Data. Data are presented as mean + SEM. � P < 0.05, �� P < 0.01, and ��� P < 0.001. (A and E) Twoway ANOVA was used to compare means. Significance P values were calculated using Bonferroni test. (C) One-way ANOVA was used to compare means. Significance P values were calculated using Tukey's test. DDC, 3.5-diethoxycarbonyl-1.4-dihydrocollidine; Erk, extracellular signal-regulated kinase; Fgfr2-IIIb, Fgf7 ligand and receptor; Fgf7, fibroblast growth factor 7; i.p., intraperitoneal; Irs2, insulin receptor substrate 2; LPC, liver progenitor cell; mRNA, messenger RNA; P-Erk, phosphorylated Erk (Tyrosine 204); RT-qPCR, reverse transcriptase-quantitative PCR; RU, relative unit; WT, wild type. Consistent with the hypothesis that FGF7 was indirectly regulated by stromal IRS2, we observed similar levels of FGF7 in the early stages of LX-2/HepaRG coculture (Fig 7B). However, while FGF7 expression was sustained in cocultures containing normal LX-2 stroma, a significant long-term decline was observed in those that contained IRS2-deficient LX-2 ( Fig 7B), demonstrating an indirect role for stromal IRS2 in promoting FGF7 expression in coculture. The decline FGF7 expression in the IRS2-deficient stromal cocultures also coincided with a striking failure to induce FGFR2-IIIb on day 14 ( Fig 7B). Interestingly, we also observed blunted induction of FGF10 within cocultures containing IRS2-deficient stroma (days 10-14), whereas FGF22 expression was unaffected (S10A Fig). These data reinforced the notion that FGFR2-IIIb ligands such as FGF7/FGF10 were required to drive expression of their epithelial receptor in LPCs both in vivo and in vitro.
Using the coculture method, we also found that FGFR2-IIIb and APOA2 induction by LPCs required insulin signaling via IRS2 within the fibrotic stroma. Knockdown of IRS2 in LX-2 cells completely abrogated the ability of insulin to potentiate epithelial sensitivity to FGF7 and hepatocyte differentiation within the cocultures, as judged by loss of time-dependent induction of FGFR2-IIIb and APOA2, respectively ( Fig 7C). These results demonstrated a requirement for stromal IRS2 in the non-cell-autonomous potentiation of FGF7 target genes in HepaRG cells. Moreover, we observed localized expression of FGFR2 in HepaRG at sites of contact with LX-2 cells, suggesting regulation of LPC differentiation by the stroma at close range, consistent with the short-rage paracrine actions of FGF7 ( Fig 7D).
In order to further test whether paracrine FGF7 signaling was downstream of IRS2-dependent stromal-epithelial crosstalk in hHSC/HepaRG cocultures, we used primary hHSCs, which expressed higher levels of endogenous FGF7 than LX-2 (Fig 8). Interestingly, we found that both FGF7 and IRS2 expression were significantly increased in hHSCs following MitoCinduced cell-cycle arrest (Fig 8A). This observation provided us with a useful means to modulate endogenous FGF7 expression prior to coculturing the cells with HepaRG. Consistent with the increase in FGF7 expression, we observed a dramatic enhancement of APOA2 expression in HepaRG cocultures using hHSCs pretreated with MitoC ( Fig 8B). Moreover, inclusion in the medium of a competitive FGF7 inhibitor, consisting of a chimeric rFGFR2-IIIb extracellular domain/Human IgG1 crystallizable fragment (Fc) domain protein (rhFGFR2-IIIb-Fc), significantly inhibited APOA2 expression by HepaRG, confirming APOA2 as a downstream target of paracrine FGF7 and demonstrating a role for FGF7 signaling in driving hepatocyte differentiation in coculture.
The diffusion of FGF7 is highly restricted because of its affinity for heparin-sulphate proteoglycans in extracellular matrix and cell membranes [23]. Consistent with this short-range mode of action, we found that hepatocyte differentiation within the cocultures (judged by line delimits "island." (D-E) IRS2 is necessary for FGFR2-IIIb expression during HepaRG differentiation: (D) phase-contrast images of differentiated cells after stable silencing for IRS2 using shRNA lentivirus (sh-IRS2). Arrowheads highlight "islands" observed in control cultures expressing sh-scram but not in sh-IRS2 cells. Specificity of the knockdown phenotype was confirmed by a "rescue" in which Lenti-mIrs2 was constitutively expressed in sh-IRS2 cells. (E) RT-qPCR analysis of FGFR2-IIIb expression (n = 3). (F-G) Insulin/IRS2 is required for functional sensitivity to rFGF7 in HepaRG cells and is necessary for feed-forward induction of FGFR2-IIIb by FGF7 ligand. (F) Phase-contrast and immunofluorescence images of HepaRG cells (day 15) cultured with rhFGF7 for 11 days, showing an increase in epithelial islands positive for Alb/HNF4α and E-cad immunostaining and pAPOA2-GFP reporter expression (H = Hoechst). (G) RT-qPCR analysis for indicated genes following long-term (11-day) rhFGF7 treatment in the presence (+) or absence (−) of supplemented insulin in control (sh-luc) or IRS2 knockdown cells (sh-IRS2) (n = 3). Data information: underlying data are available in S1 Data. Data are presented as mean + SEM. � P < 0.05, �� P < 0.01, and ��� P < 0.001. (E and G) One-way ANOVA was used to compare means. Significance P values were calculated using Newman-Keuls multiple comparison test. Alb, albumin; APOA2, apolipoprotein A2; Cont, control; E-cad, E-cadherin; EF1, Human elongation factor-1 alpha promoter; FGF7, fibroblast growth factor 7; FGFR2, fibroblast growth factor receptor 2; GFP, green fluorescent protein; HNF4α, hepatocyte nuclear factor 4-alpha; IRS2, insulin receptor substrate 2; Lenti-mIrs2, murine Irs2 transgene not targeted by humangene-silencing construct; mRNA, messenger RNA; pAPOA2, human APOA2 promoter; pIRS2, IRS2 promoter; rFgf7, recombinant Fgf7; rhFGF7, recombinant human FGF7; RT-qPCR, reverse transcriptase-quantitative PCR; RU, relative unit; sh-IRS2, shRNA-targeting IRS2; sh-luc, control luciferase; shRNA, short hairpin RNA; shscram, scrambled shRNA. HNF4α immunostaining) was spatially limited to sites of contact between HepaRG and vimentin+ hHSCs (Fig 8C). To explore the properties of the HSC signal driving differentiation within the cocultures further, we developed a short-term bioassay using APOA2 expression as a functional readout for FGF7 activity. Stimulation of HepaRG monocultures with a single dose of rhFGF7 rapidly induced APOA2 at 48 h ( Fig 8D). Hence, we collected conditioned medium from MitoC-treated hHSCs and added it to HepaRG cells; however, no induction of APOA2 was observed, suggesting the factor was either labile or unavailable in the cellular supernatant because of cell/matrix binding (Fig 8E). Using a similar approach, we were able to confirm that direct contact between HepaRG and hHSCs (achieved by short-term 48 h coculture) was sufficient to induce APOA2 expression in an FGF7-dependent manner, an effect that was blocked by inclusion of the competitive FGF7 inhibitor (Fig 8F). Importantly, the ability of hHSCs to induce APOA2 within 48 h in this model was conditional upon the pretreatment of hHSCs with MitoC, confirming that paracrine FGF7 signaling in hHSCs was enhanced by mitotic inactivation.
Taken together, these data demonstrated a positive role for paracrine FGF7 signaling by human HSC in supporting FGF7 sensitivity and hepatocyte differentiation in adjacent epithelial progenitors.

IRS2 deficiency promotes stromal activation at the expense of epithelial repair
The stromal injury response precedes epithelial repair. In the liver, this is characterized by HSC-derived myofibroblast proliferation and the laying down of extracellular matrix, which is followed by cell-cycle exit, fibrogenic reversion, and apoptotic cell clearance that accompanies tissue repair. The equilibrium within the fibrotic stroma between activated and reverted HSCs is therefore central to the regulation of the stromal injury response and has important consequences for liver fibrosis and scarring in the context of chronic injury.
Our data showed that both FGF7 and IRS2 expression were induced in hHSCs upon triggering of cell-cycle arrest using MitoC (Fig 8A). MitoC treatment was also accompanied by a phenotypic shift in hHSCs and LX-2s consistent with fibrogenic reversion (Fig 9A), as judged by loss of Collagen 1A1 (COL1A1) expression and activation of antiapoptotic heat shock 70 kDa protein 1B (HSPA1B), previously identified as a biomarker of HSC reversion [31]. Importantly, the down-regulation of myofibroblast marker protein αSMA upon cell-cycle exit was delayed in LX-2 cells in which IRS2 was silenced (Fig 9B), suggesting impaired transition from an activated to "reverted" state. These in vitro data, which suggested that IRS2 could participate in the transition of HSCs to a "reverted" phenotype, were consistent with the in vivo highlighting localized FGFR2 expression (arrows) at sites of contact between HepaRG and HSC stroma (marked by phalloidin). Data information: underlying data are available in S1 Data. Data are presented as mean + SEM. � P < 0.05, �� P < 0.01, and ��� P < 0.001. (B) Twoway ANOVA was used to compare means. Significance P values were calculated using Tukey's multiple comparison test. (C) Two-way ANOVA was used to compare means. Significance P values were calculated using Sidak's multiple comparison test. APOA2, apolipoprotein A2; Fgfr2-IIIb, Fgf7 receptor; Fgf7, fibroblast growth factor 7; H, Hoechst; HSC, hepatic stellate cell; Irs2, insulin receptor substrate 2; LPC, liver progenitor cell; mRNA, messenger RNA; RT-qPCR, reverse transcriptase-quantitative PCR; RU, relative unit; shIRS2, shRNA-targeting IRS2; sh-luc, control luciferase; shRNA, short hairpin RNA. Insulin resistance: Resisting repair observation that DDC-treated Irs2 −/− mice had increased periportal and perisinusoidal fibrosis, judged by Sirius Red collagen morphometry and αSMA immunostaining (Fig 9C).
We therefore returned to the LX-2/HepaRG cocultures to assess changes in fibrogenic markers. Using this method, we found that upon coculture with LPC-like cells, control (sh- Conditioned medium from MitoC-treated hHSCs was insufficient to illicit an APOA2 response at 48 h, whereas (F) direct addition of hHSCs to HepaRG cultures activated APOA2 expression in an FGF7-dependent manner, as judged by the ability of rhFGFR2-IIIb-Fc to block induction. Importantly, this effect was conditional upon prior treatment of hHSCs with MitoC, thus confirming the positive effect of mitotic inhibition on the ability of hHSCs to communicate with LPCs via FGF7. Data information: underlying data are available in S1 Data. Data are presented as mean + SEM. � P < 0.05, �� P < 0.01, and ��� P < 0.001. (A) Paired Student t test. (B, E, F) One-way ANOVA was used to compare means. Significance P values were calculated using Tukey's multiple comparison test. (D) Two-way ANOVA was used to compare means. Significance P values were calculated using Tukey's multiple comparison test. APOA2, apolipoprotein A2; cont, control; Fgfr2-IIIb, Fgf7 receptor; Fgf7, fibroblast growth factor 7; hHSC, human HSC; HNF4α, hepatocyte nuclear factor 4-alpha; HSC, hepatic stellate cell; Irs2, insulin receptor substrate 2; LPC, liver progenitor cell; MitoC, mitomycin C; mRNA, messenger RNA; rhFGFR2-IIIb-Fc or rhIIIb-Fc, recombinant protein comprising human Fgfr2-IIIb extracellular domain fused to Human IgG1 Fc domain used to inhibit FGF7 binding its receptor; rhFGF7, recombinant human Fgf7; RT-qPCR, reverse transcriptase-quantitative PCR; RU, relative unit.  (Fig 9D). In striking contrast to this, we found that silencing of IRS2 in LX-2 cells abrogated their ability to suppress COL1A1 in coculture, and we instead observed a dramatic induction of nonfibrillar type III collagen (COL3A1) indicative of an activated stromal injury response. Time-dependent switching from myofibroblast marker THY1 to LPC-associated SPP1 was also dampened by silencing of IRS2 in the stroma (S10B Fig), and the numbers of GFP-expressing LX-2 cells at the end of the cocultures was increased (S10C Fig). Interestingly, unlike LPC and epithelial repair genes (SPP1/APOA2/FGFR2-IIIb), whose expression in coculture was both insulin and IRS2 dependent (Fig 7C and S10 Fig), changes in fibrogenic/myofibroblast genes COL1A1, COL3A1, ACTA2, and THY1 (Fig 9D and S10 Fig) were largely unaffected by omission of insulin from the medium, suggesting the ability of IRS2 to restrain activation of the fibrogenic stroma was insulin independent. We concluded that silencing of IRS2 in HSCs prolonged their fibrogenic activation in coculture while slowing fibrogenic reversion by MitoC and reducing the survival of mitotically inactivated cells in which FGF7 expression was increased. Thus, IRS2 signaling in the fibrotic stroma served as a switch between states of stromal activation and fibrogenic reversion that potentiated heterotypic paracrine signaling between the two cell types and drove epithelial FGF7 sensitivity in LPCs. Together with our results demonstrating a cell-intrinsic role for IRS2 in LPCs in promoting FGF7 sensitivity, we proposed a model in which IRS2 exerted multiple proregenerative effects in both stromal and LPC cell compartments that favored epithelial repair over fibrogenesis during chronic liver injury (Fig 10).

Discussion
In this study, we provide compelling evidence that Irs2 is required for hepatic wound healing and epithelial repair during chronic DDC feeding in mice and for sustaining local Fgf7/ Fgfr2-IIIb signaling. Using human cell models, we have shown that IRS2 can affect multiple processes in both the fibrotic stroma and bipotent LPC-like cells that impinge upon their abilities to communicate with one another via FGF7 signaling. In addition, coculture studies confirm that IRS2 deficiency in HSCs disrupts paracrine crosstalk between the stromal niche and LPCs, resulting in failure to induce FGFR2-IIIb and epithelialization while simultaneously prolonging fibrogenic activation. Our data therefore support the hypothesis that insulin resistance could lead to defects in hepatic injury repair that exacerbate liver damage and progression of chronic liver disease because of reduced paracrine FGF signaling. These data also have profound implications for our understanding of how systemic insulin signals are interpreted by local injury-repair mechanisms and demonstrate that defects in insulin sensitivity can influence short-range communication between distinct cell types within a tissue, notably between the so-called "niche" and adult stem cells responsible for producing new epithelia.
Interestingly, insulin resistance in human liver disease is associated with increased fibrosis and the appearance of reactive periportal lesions called "ductular reactions" (DRs), in which intermediate "hepatobiliary" progenitors expand surrounded by fibrotic stroma [32]. These expression was analyzed at 2 days and 14 days after seeding the two cell types together (n = 3). Data information: underlying data are available in S1 Data. Data are presented as mean + SEM. � P < 0.05, �� P < 0.01, and ��� P < 0.001. (A) Paired Student t test. (B and D) Two-way ANOVA was used to compare means. Significance P values were calculated using Sidak's multiple comparison test. (C) Unpaired Student t test. Acta2, gene encoding αSMA; AU, arbitrary unit; COL1A1, Collagen 1A1; cont, control; hHSC, human HSC; HSC, hepatic stellate cell; HSPA1B, heat shock 70 kDa protein 1B; Irs2, insulin receptor substrate 2; MitoC, mitomycin C; mRNA, messenger RNA; pHSC, primary human HSC; RT-qPCR, reverse transcriptase-quantitative PCR; RU, relative unit; shIRS2, shRNA-targeting IRS2; sh-luc, control luciferase; shRNA, short hairpin RNA; WT, wild type; αSMA, alpha-smooth muscle actin. https://doi.org/10.1371/journal.pbio.2006972.g009 Insulin resistance: Resisting repair lesions exhibit striking similarities with those observed in the DDC-injury model [33], although DRs are a feature of disrepair because of their correlation with poor prognosis. The association between insulin resistance and DRs has been demonstrated in liver disorders including NASH and chronic hepatitis C [32,34], in which defective IRS2 expression has also been reported [10,11]. Our results suggest that reduced sensitivity to insulin in patients with chronic liver disease could alter the intrinsic abilities of LPCs to contribute to regenerative processes and lead to a derangement of heterotypic cell-cell interactions that mediate their expansion and differentiation, leading to DRs. Our discovery that IRS2 silencing in HSCs also promoted fibrogenesis while limiting hepatocyte differentiation and FGFR2-IIIb expression in coculture recapitulated the increase in liver fibrosis and reduced Fgf7/Fgfr2-IIIb expression described in Irs2 −/− mice during DDC treatment. Hence, our data suggest that lRS2 signaling may serve as a switch that regulates the transition between phases of stromal and epithelial expansion within DRs by promoting reversion and survival of FGF7-expressing stroma while simultaneously priming LPCs to express FGFR2-IIIb (Fig 10).
There is precedent that IRS proteins play an important role in driving epithelial growth via modification of the stromal microenvironment in the liver. NT-157, the allosteric inhibitor of IRS1/2, was recently found to limit colorectal cancer metastasis to the liver by reducing the trophic support received by stromal myofibroblasts or HSCs [35]. These findings exhibit striking parallels with our results and highlight the potential future impact of our feed-forward paracrine model in the field of cancer biology, in which subversion of injury-repair mechanisms by tumors are well documented. Consistent with this, Irs2 was also recently implicated in a positive feedback loop driving disease progression and hepatocarcinogenesis in rodent models of nonalcoholic fatty liver disease (NAFLD) [13]. Interestingly, the study identified hepatic Irs2 expression as a downstream target of Yes-associated protein (YAP) and Tafazzin (TAZ)-two oncogenic effectors of the Hippo signaling pathway that reversibly control hepatocellular fate [36], LPC activation [37], and liver regeneration [38]. Hence, it seems likely that subversion of Insulin resistance: Resisting repair Irs2-dependent aspects of the LPC response identified in our study also contribute to the pathophysiology of metabolic liver disease and cancer.
Insulin and FGF7 synergized via IRS2 to promote osteopontin/SPP1 expression and sustain hepatocyte differentiation, thus revealing a cell-intrinsic role for insulin/IRS2 in priming and potentiating LPC sensitivity to paracrine FGF7 and other IIIb ligands such as FGF22 and FGF10. Insulin/IRS2 signaling was required for the formation of "islands" in HepaRG cultures, which expanded because of a process of hepatocyte differentiation and epithelialization and conferred FGF7 sensitivity to the cultures by promoting FGFR2-IIIb expression. Conversely, in DDC-treated Irs2 −/− mice, the numbers of hepatocytes declined, and the process of re-epithelialization was impaired, as manifested by delayed restoration of the parenchymal architecture, delayed induction of Epcam, failure to sustain Fgfr2-IIIb, and failure to expand numbers of small hepatocytes. Thus, it is tempting to speculate that IRS2 promotes hepatocyte differentiation of LPCs in vivo, although this would need to be empirically tested.
We have identified positive feedback between FGF7 and its receptor in HepaRG monocultures as well as HepaRG/HSC cocultures and demonstrated a role for IRS2 in both cell compartments upstream of FGFR2-IIIb amplification (Fig 10). These data were highly consistent with the observed dual failure of Fgf7/Fgfr2-IIIb in Irs2 −/− mice during injury and provide new insight into the underlying synergies that help promote epithelial sensitivity to proregenerative stromal growth factors such as Fgf7. Positive feedback between Fgf10 and Fgfr2-IIIb has previously been reported during salivary gland development [15], and in nonmalignant tumors, Fgf7-mediated FGFR2-IIIb expression is thought to play a tumor-suppressive role by promoting epithelial differentiation [23,39]. Interestingly, synergy between Fgfr2-IIIb ligands and their receptor has previously been linked to crosstalk with phosphoinositide 3-kinase (PI3K)one of the principle effectors of insulin/IGF-1 signaling recruited by IRS2-binding [15]. Indeed, inhibition of PI3K/Protein kinase B (AKT) limits hepatocellular reprogramming by Fgf10/Fgfr2-IIIb during DDC injury in vivo [22], suggesting Irs2 could act upstream of this pathway in promoting cellular plasticity in the liver.
Further to impairment of Fgf7, we also describe reductions in Fgf10 during DDC injury and in LX-2/HepaRG cell coculture. Hence, we cannot rule out the possibility that the IRS2-dependent paracrine feed-forward loop we have identified also involves other stromal FGFR2-IIIb ligands. HSC-derived Fgf10 plays an important role in the expansion of bipotent LPCs during liver development [40], and both FGF10 and 22 are induced by the fibrotic stroma during DDC injury [22]. A deeper understanding of the indirect nature of how IRS2 regulates these genes within the stroma is therefore needed.
In recent years, FGFs have received increasing attention for their unexpected role in metabolic disease and for their potential to treat metabolic disorders by helping restore insulin sensitivity. FGF1, 19, and 21 have potent (albeit poorly understood) effects on metabolism and insulin signaling, largely through their putative actions in white adipose tissue and the liver [21,41], whereas impaired FGF-receptor signaling is associated with insulin resistance, liver disease [41,42], and impaired hepatic regeneration [43,44]. Blunted expression of FGF7 has also been previously linked to delayed skin wound healing in genetically diabetic leptin deficient (db/db) mice [45,46]. Our data therefore raise the intriguing possibility that a comparable wound-healing defect may also exist in the livers of Irs2 −/− mice following injury, resulting in a breakdown in stromal-epithelial paracrine signaling and consequent failure to coordinate the injury response. Combined future strategies to potentiate hepatic insulin sensitivity and FGF7 signaling may therefore improve patient outcomes by promoting epithelial repair in the liver.
The impact of insulin resistance on the stromal microenvironment in the liver during injury is poorly understood [5], as is the role played by IRS proteins in regulating HSC biology, which involves injury-dependent cycles of myofibroblast activation and cell clearance during so-called "fibrogenic reversion." The model we present to explain impaired stromal induction of Fgf7 in the Irs2 −/− mice (Fig 10) incorporates how IRS2 influenced fibrogenic reversion and survival of HSCs-only one of several FGF7-expressing stromal subpopulations that respond to DDC injury. However, we also provide evidence that the cellular composition of the fibrotic milieu was different in the livers of Irs2 −/− mice and leave open the possibility that Irs2-deletion affected other resident FGF7-expressing cells such as PFs. Finally, we observed a greater increase in T cells and the incorporation of bloodborne mesenchymal cells, possibly to the detriment of resident stromal cells or simply to compensate for the failure of FGF7-expressing stroma. Fibrocytes from type II diabetic patients have recently been shown to display altered migratory and inflammatory characteristics that could also have contributed to the altered pattern of stromal gene expression and impaired wound healing in our model [47].
We show that DDC injury had a negative impact on Gfap+ HSCs in Irs2 −/− mice. Gfap is associated with quiescent or inactivated HSCs; hence, our data are consistent with accelerated turnover/loss of these cells in the livers of Irs2 −/− mice due to a combination of reduced survival and/or reduced myofibroblast reversion (Fig 10). Our investigation using hHSCs revealed that IRS2 protected cells from apoptosis following mitotic inactivation using MitoC, providing a mechanism to explain the dampening of Fgf7 and increased apoptosis observed in the livers of Irs2 −/− mice during injury. Interestingly, both IRS2 and FGF7 were increased in primary hHSCs upon cell-cycle arrest and fibrogenic reversion while enhancing their ability to drive epithelial differentiation of HepaRG cells in an FGF7-dependent manner in coculture. Similarly, IRS2 was required in LX-2 cells to sustain both FGF7 expression and to allow a process of fibrogenic reversion to proceed in the presence of HepaRG cells. These data provide new evidence to suggest that epithelial repair signals produced by HSCs are increased by a program of cell-cycle arrest or cell death, two processes that are more closely aligned with stromal reversion than myofibroblast activation. Given that the regulation of FGF7 expression by the fibrotic stroma is poorly understood, we propose the hypothesis that fibrogenic reversion, rather than myofibroblast activation, drives FGF7 expression during injury and that IRS2 preferentially favors reversion and survival of HSCs tasked with stimulating epithelial repair. Future work by this laboratory will seek to further test this hypothesis.

Ethics statement
Mice were housed in the facility of the CIPF (Valencia, Spain), which is registered as an experimental animal Breeding, User, and Supplier Center (reg. no. ES 46 250 0001 002) under current applicable Spanish regulations (RD 53/2013). Animals were kept in ventilated racks with microisolators under specific-pathogen-free (SPF) conditions and handled by accredited personnel and treated in accordance with EU legislation and welfare guidelines. Procedures were approved by the CIPF "Comite Ético de Experimentación Animal" (CEEA) and the ministry for "Agricultura, pesca, alimentacion y agua" of the Generalitat Valenciana (Valencia, Spain), under the authorization code 2014/VSC/PEA/00062 tipo 2. Mice were killed by fentanil/pentobarbital overdose followed by cervical dislocation.

Liver injury and cholestasis markers
Serum transaminases were measured by ELISA according to manufacturer's instructions (MilliporeSigma, St. Louis, MO, USA), and serum total bilirubin (TBIL) was measured by colorimetric assay according to the Jendrassik-Grof method (MilliporeSigma). Bile acids were measured by metabolomic analysis. Briefly, samples were extracted from whole-liver tissue using the metanol-chloroform-H 2 O method [48] and analyzed on a 600 MHz NMR spectrometer equipped with a cryoprobe. Metabolites were identified with the aid of the spectral databases HMBD [49] and BMRB [50] and quantified using MestreNova8.

Cell culture
HepaRG cells (kind gift of Anne Corlu and Cristiane Guillouzo) were passaged and differentiated as previously described [30] using Williams E medium (10% fetal calf serum, 50 μM hydrocortisone hemisuccinate, 0.88 μM insulin, and 2 mM L-glutamine and penicillin-streptomycin). DMSO (2%) was added from day 14. Insulin supplement was omitted from the medium where indicated 24 h postplating. rhFGF7 (50 ng/ml; Cell Guidance Systems, St. Louis, MO, USA) or rhFGFR2-IIIb Fc Chimera (100 ng/ml; R&D Systems, Minneapolis, MN, USA) were added to media where indicated every 48 h. Cryopreserved primary hHSCs were obtained from a healthy 15-year-old Caucasian female donor using a previously described isolation method [51] (Innoprot, Derio, Spain). LX-2 cells (kind gift of Scott Friedman) and primary hHSCs were cultured in DMEM (2% fetal calf serum), and primary HSCs were used within 5 passages. LX-2/HSCs were inactivated for 3 h with 1mg/ml MitoC (Milli-poreSigma) before seeding onto 0.1%-gelatin-coated plates. After 24 h, activated/inactivated HSC cultures were switched to HepaRG medium until the end of the experiments. Stable cell lines were generated by lentiviral transduction (vectors summarized in S1 Table) using a multiplicity of infection of 0.5-20 TU/cell in the presence of 8 μM polybrene. Infections were performed 16 h postplating for 6 h in serum-free medium. Where applicable, cells were selected for 7 days in media containing 2.5 μg/ml puromycin.

Immunohistochemistry
Formalin-fixed paraffin-embedded tissue sections (4 μm) were processed automatically using the Autostainer Link 48 (Agilent, Santa Clara, CA, USA) with EnVision FLEX reagents according to manufacturers' instructions (Agilent). Antigen retrieval was performed using high-pH target retrieval solution (Agilent). Antibodies are provided in S2 Table.

Collagen morphometry
Collagen was visualized in PFE liver sections using the Picro Sirius Red technique (Abcam, Cambridge, UK) according to manufacturer's instructions. Following imaging, threshold analysis was performed using FIJI software to quantify staining in a total area comprising >10 portal fields per animal.

Image analysis
Fluorescence immunostained cells/cryosections were analyzed using the "INcell 1000" imaging platform and software (GE Healthcare) where indicated. Individual cells were identified by Hoechst staining. Threshold analysis was used to determine the expression of the following markers: pAPOA2-GFP, albumin, CYP3A4, HNF4α, Ki67, Spp1, and Vimentin. For replicate well/tissue sections, at least 20 fields were randomly analyzed for each replicate. PFE tissue sections were imaged using Pannoramic Viewer (3DHISTEC) and analyzed with FIJI software; βcatenin staining was quantified using FIJI (histogram function) to calculate modal pixel intensity.

Hepatocyte ploidy
Interpolation of DNA content in HNF4α/Hoechst-stained liver sections was performed using a methodology developed by our laboratory using INcell 1000 (GF Healthcare) that enabled in situ approximation of hepatocyte ploidy (S3 Fig). DNA content was calculated for all circular hepatocyte nuclei (nuclear elongation value > 0.8) as a combined function of Hoechst nuclear fluorescence intensity (DNA density) and nuclear volume (calculated as a spherical function of nuclear radius). Hepatocyte ploidy was then calibrated using circular HNF4α− NPC cell population within the tissue as a 2n control. The 2c hepatocyte population was defined as the population of HNF4α+ nuclei within the 19.99 μm 2 -34.99 μm 2 nuclear size range with a nuclear circularity index >0.8.

RT-qPCR
Total RNA was extracted using RNeasy Mini Kit (Qiagen, Venlo, The Netherlands) with DNa-seI Digestion. Liver tissue was preprocessed using Trizol (MilliporeSigma). For experiments using cells, RNA from duplicate wells were analyzed. First-strand synthesis was performed using EcoDry Premix (Takara, Kusatsu, Japan), and real time-PCR was carried out in LightCycler 480 II (Roche) using SYBR PreMix ExTaq (Mi RNaseH Plus, Takara). Reverse transcriptase-quantitative PCR (RT-qPCR) was performed in triplicates for all RNAs analyzed. Primers are listed in S3 Table. Relative gene expression was calculated by normalization to Glyceraldehyde-3-Phosphate Dehydrogenase (GAPDH) in mouse liver and to Ribosomal Protein L19 (RPL19) in human cell lines.