Selective HIF stabilization alleviates hepatocellular steatosis and ballooning in a rodent model of 70% liver resection

Background: Small-for-size syndrome looms over patients needing liver resection or living-donor transplantation. Hypoxia has been shown to be crucial for the successful outcome of liver resection in the very early postoperative phase. While poorly acceptable as such in real-world clinical practice, hypoxia responses can still be simulated by pharmacologically raising levels of its transducers, the hypoxia-inducible factors (HIF). We aimed to assess the potential role of a selective inhibitor of HIF degradation in 70% hepatectomy (70%Hx). Methods: In a pilot study, we tested the required dose of roxadustat to stabilize liver HIF1α. We then performed 70%Hx in 8-week-old male Lewis rats and administered 25 mg/kg of roxadustat (RXD25) at the end of the procedure. Regeneration was assessed: ki67 and EdU immunofluorescent labeling, and histological parameters. We also assessed liver function via a blood panel and functional gadoxetate-enhanced magnetic resonance imaging, up to 47 hours after the procedure. Metabolic results were analyzed by means of RNA sequencing. Results: Roxadustat effectively increased early HIF1α transactivity. Liver function did not appear to be improved nor liver regeneration to be accelerated by the experimental compound. However, treated livers showed a mitigation in hepatocellular steatosis and ballooning, known markers of cellular stress after liver resection. RNA sequencing confirmed that roxadustat unexpectedly increases lipid breakdown and cellular respiration. Conclusions: Selective HIF stabilization did not result in an enhanced liver function after standard liver resection, but it induced interesting metabolic changes that are worth studying for their possible role in extended liver resections and fatty liver diseases. panel and death rate. Effectiveness was assessed through HIF1α measurement in cytosolic and nuclear liver extracts, and the evaluation of expression of HIF-regulated genes. RNAseq approach. The analysis is reported in Supplemental Information 3. We detected 2,521 genes whose expression was significantly different in 70%Hx-RXD25 compared to 70%Hx-placebo. The principal component analysis separated the samples according to the treatment. We employed ORA to determine if the significantly modulated genes likely originated from any known gene pathways, while GSEA was performed to assess the enrichment of up- or down-regulated genes in these pathways. We identified nine clusters of cellular functions that were recurrently enriched in 70%Hx-RXD25 compared to 70%Hx-placebo. Three of those pathway clusters appeared generally upregulated: HIF signaling, energy production via oxidative phosphorylation, and lipid metabolism, while six pathway clusters were downregulated: cell growth and death, oxidative stress, xenobiotic metabolism, transmembrane transporters, inflammation, endocrine signaling (Supplemental Information 4). The comparison between the 70%Hx-RXD25 group and the sham-placebo group showed that the former exhibits an upregulation of cell cycle genes and the HIF1α pathway. The comparison between the 70%Hx-placebo group and the sham-placebo group showed that the former exhibits an upregulation of cell cycle genes. Both resected groups showed a down-regulation in lipid metabolism genes as opposed to the sham group (Supplemental Information 5).


Introduction
In liver surgery, the management of liver resection in cirrhotic patients or small-size hepatic remnants even in healthy livers still represents a challenge. The postoperative conundrum alternatively called posthepatectomy liver failure (PHLF) or SFSS is the major cause of death after liver resection [1].
In recent years, selective liver hypoxia has been shown to be beneficial after extended hepatectomy. In rat models of human associating liver partition and portal vein ligation technique for two-stage hepatectomy (ALPPS), parenchymal hypoxia, resulting from a reduced arterial flow, has been associated with accelerated liver regeneration after portal vein ligation and parenchymal transection [2], and evidence showed that Downloaded from http://portlandpress.com/clinsci/article-pdf/doi/10.1042/CS20210183/921101/cs-2021-0183.pdf by guest on 23 September 2021 5 ALPPS curbs mortality after extended liver resection [3,4]. Indeed, post-hepatectomy hypoxia-driven response induces new vessels formation and reduces the shift between hepatocellular and vessel proliferation by slowing down the former and accelerating the latter [4,5]. Tissue hypoxia activates the α subunits of HIF transcription factors, HIF1α and HIF2α, which mediate a wide array of responses to parenchymal low oxygen pressure. Accordingly, increased HIF-α transactivity leads to new blood vessel formation in response to hypoxic injury [6]. In view of the translational relevance of these findings, introducing hypoxia in hepatobiliary surgery is an appealing perspective, though apparently unrealistic because it is known that systemic hypoxia impairs hepatic regeneration [7], and that the human biliary tract is exquisitely sensitive to hypoxia [8]. The inhibition of prolyl hydroxylase domain (PHD) proteins effectively increases HIF-α activity. PHDs require oxygen, iron, and α-ketoglutarate, to hydroxylate HIFs and steer them towards degradation. In this regard, PHDs are considered intracellular oxygen sensors. Several molecules inhibit PHD function, such as deferoxamine, dimethyloxalylglycine (DMOG), ethyl-3,4dihydroxybenzoate (EDHB), and selective PHD inhibitors [6,[9][10][11]. Indeed, deferoxamine protects from oxidative stress and IRI in animal models of liver resection [12][13][14][15][16][17][18][19]. DMOG increases sinusoidal endothelium density [11], preserves sinusoidal diameter, and significantly improves survival after 87% hepatectomy in rats [4]. EDHB acts as a preconditioning agent and improves liver regeneration after ALPPS-like procedure, by inducing cell cycle-promoting cyclins [9]. Because of their mechanism of action, the cited PHD inhibitors lack selectivity [20]. DMOG and EDHB compete with ketoglutaric acid and inhibit other α-ketoglutaratedependent dioxygenases [21,22]. Deferoxamine sequesters iron, which is required for PHD activity, but it affects any iron-requiring enzymes. Recently, a new class of selective PHD inhibitors has been released: the isoquinolines, collectively known as HIF stabilizers because they selectively compete for the active site of PHDs. Among them, roxadustat (FibroGen, San Francisco, US-CA), a small (352.34 Da) orally active molecule, appears particularly promising and has been already licensed in the European Union and China to treat anemia in end-stage renal disease as erythropoietin release is controlled by HIF-α [23,24] Besides, the effects of this entire class of molecules have not yet been reported in the framework of liver resection.
Roxadustat reported half-life in Sprague Dawley rats is 3.4-5.6 hours and is not dose-dependent. To obtain a steady hematocrit increase, it is chronically administered to rats three times a week [25]. Chronic Downloaded from http://portlandpress.com/clinsci/article-pdf/doi/10.1042/CS20210183/921101/cs-2021-0183.pdf by guest on 23 September 2021 6 administration of roxadustat >15 mg/kg yields polycythemia and associated adverse events in rats, while doses up to 10 mg/kg are active and do not induce significant mortality [25][26][27]. The intraperitoneal injection of roxadustat is known to effectively target the liver and stabilize HIF-α with a peak at six hours after administration in mice [24], while residual HIF1α stabilization is detectable after 30 hours in an ex-situ liver graft perfusion model [26].
Thus, we hypothesized that the beneficial effects elicited by hypoxia after liver resection are linked to HIFassociated response pathways. We chose a standard rat model of 70% hepatectomy (70%Hx), known to induce intense liver regeneration without mortality [28]. We aimed to assess whether a selective pharmacological HIF stabilization, obtained by the administration of roxadustat, affects liver function and regeneration postoperatively.

Animals
Male Lewis rats came from Janvier Labs (Le Genest-Saint-Isle, France). At the time of the procedure, they were eight weeks old and weighed 244 g (interquartile range, IQR=234-247). The animals had been housed for at least six days in the local animal facility (Laboratory of Experimental Surgery and Transplantation, Harvey Tower, 4 th floor, 55 Avenue Hippocrate, 1200 Brussels, Belgium) before the experimental procedures, which took place in the same laboratory. The animals were kept in standard cages, in twelvehour light-dark cycle, and received standard chow and drinking water at libitum. Our institutional review board approved this protocol number 2019/UCL/MD/043 on December 17, 2019.

Pharmacological pilot study
Dosage, timing of injection, and effectiveness of intraperitoneally administered roxadustat in inducing liver HIF stabilization was assessed through a pharmacological study that entailed three groups of three rats each: the first receiving roxadustat 10 mg/kg (the RXD10 group), the second receiving RXD 25 mg/kg (the RXD25 group), and the third undergoing hepatic artery ligature at the hilum (the HAL group). Animals were sacrificed at three time points: baseline, 60 minutes and 6 hours. Toxicity was assessed via a biochemical Downloaded from http://portlandpress.com/clinsci/article-pdf/doi/10.1042/CS20210183/921101/cs-2021-0183.pdf by guest on 23 September 2021 7 panel and death rate. Effectiveness was assessed through HIF1α measurement in cytosolic and nuclear liver extracts, and the evaluation of expression of HIF-regulated genes.
Seventy percent hepatectomy was performed by removing the median and the left lateral lobes after median laparotomy, according to the principles described by Higgins and Anderson, [29]. Since the ligation at the basis of the median lobe might constrict the lumen of the inferior vena cava, we ligated the vascular pedicles individually, we divided the median lobe into its right and left segments, finally the lobes were ligated and severed one by one, according to Kubota et al [30].
In the sham procedure, rats underwent median laparotomy and dissection of liver ligaments. The liver was wrapped in humid gauzes for the duration of the procedure.
All the procedures were carried out during the afternoon (between 3:30 p.m. and 10:00 p.m.) to ensure that animals had a spontaneously empty foregut.

Study groups and postoperative follow-up
At the end of surgery, a single intraperitoneal dose of roxadustat 25 mg/kg was injected. The drug was dissolved in dimethyl sulfoxide (DMSO, Sigma-Aldrich, St. Louis, US-MO) and then diluted in pH 7.4 phosphate-buffered saline (PBS) to a 2% concentration of DMSO. The placebo consisted of 2%DMSO-PBS, in the same volume as the experimental drug. Postoperatively, all rats received a single subcutaneous shot of 4 ml/kg 2.1% NaHCO3 and had free access to food and 20% glucose as drinking water to avoid postresection hypoglycemia [28].

Tissue assays
Liver nuclear and cytosolic protein extracts were obtained with the Nuclear Extract Kit (#40010, Active Motif, Carlsbad, US-CA) and concentrations were measured via a bicinchoninic acid assay. HIF-1α and -2α were quantified in subcellular extracts using an enzyme-linked immunosorbent assay (#LS-F11633 and #LS-F9117, LSBio, Seattle, US-WA), as per manufacturer's instructions. Total liver lipids were extracted with water, methanol and chloroform and quantified by the vanillin-phosphoric acid reaction.

Assessment of the hepatic function with magnetic resonance imaging (MRI)
MRI experiments were performed on a Bruker Biospec 11.7 Tesla (Bruker Biospin GmbH, Ettlingen, Germany) equipped with a volume coil (74 mm of inner diameter). Animals were anesthetized during imaging session with isoflurane mixed with air at 3.5% for induction and 2% for maintenance (duration about 1 hour). Estimation of global hepatocellular function was obtained through gadolinium ethoxybenzyl diethylenetriamine pentaacetic acid (Gd-EOB-DTPA, Primovist, Bayer, Leverkusen, Germany)-enhanced MRI using a dynamic contrast-enhanced sequence [31,32]. The contrast agent was injected intravenously via a catheter in the tail vein, at a dose of 0.025 mmol/kg 30 sec after the beginning of the acquisition [33][34][35].
Acquisition parameters were as follows: TE: 1.2 msec, TR: 9.62 msec, NA: 3, number of repetitions: 500, TA: 33 min, FA: 15°, FOV: 60x60 mm², Mat: 128x128, slice thickness: 1 mm. The resulting temporal resolution was 4 sec. In post-processing, we selected the regions of interest (ROIs) in the right lobe from which we averaged the signal. The hepatic function was estimated with a homemade algorithm on Matlab (MathWorks, Natick, US-MA) by fitting the signal from the ROI [36]. We obtained the following parameters from the curve: time to peak, i.e. the delay between the arrival time (t 0 ) of the contrast agent and the maximum signal in the ROI, time to 30% signal decay, i.e. the delay between the maximum signal and its reduction of 30%, and the areas under the curve between t 0 and t 0 +60 sec (AUC60), and between t 0 and t 0 +90 sec (AUC90) [37]. We calculated the AUC between t 0 and the time to peak and the whole AUC.

Histology
An experienced liver pathologist evaluated hematoxylin-eosin (H&E)-stained whole-liver sections following these parameters: presence of mitoses, inflammation, endothelial denudation, microvascular thrombosis, parenchymal hemorrhage, hepatocellular ballooning, steatosis, and necrosis. Endothelial denudation was defined as enlargement of endothelial nuclei or endothelial dislocation or extravasation of red blood cells in the Disse space. Hepatocellular ballooning was defined as clear-cell change with doubling in size of hepatocytes. Steatosis was further classified based on its location [38], its proportion (<5%, 5-33%, 33-66%, >66%), and the type of lipid droplets (micro-or macrovesicular). Steatosis involving ≥5% of hepatocytes was considered relevant.

Fluorescence immunostaining, whole-slide imaging, and quantitative evaluation of immunostaining
Paraffin-embedded sections were deparaffinized. Endogenous peroxidases were inhibited with 3% hydrogen peroxide in methanol for 20 minutes. Antigens were retrieved in pH 5.7 10 mM citrate buffer.
Aspecific antigen binding sites were blocked in tris-buffered saline (TBS) + 5% bovine serum albumine + 0.1% Tween20. Slides were then submitted to Ki67 staining or EdU revelation. Anti-Ki67 primary antibody (rabbit, clone SP6, #ab16667, 1/100 dilution, Abcam, Cambridge, UK) was incubated for 90 min at room temperature (RT). Slides were then incubated with anti-rabbit secondary antibodies (#K4003, Dako, Glostrup, Denmark) for 60 min at RT. This reaction was visualized using Alexa555-conjugated tyramide Fluorescence immunostained liver sections were digitalized using a Pannoramic 250 FlashIII scanner (3DHistech, Budapest, Hungary) at X20 magnification. Scanned slides were then analysed using the image analysis tool Author version 2017.2 (Visiopharm, Hørsholm, Denmark). On each slide, tissue sections were automatically surrounded at low magnification. Delineations were visually checked and manually corrected if required. Cells were then detected at high resolution (X20) with a nuclear-based cell classification relying on Hoechst labeling. Following segmentation, post-processing steps were applied to separate Ki67+/-and EdU+/-cells. The parameters were constant for all slides. Results were expressed as permillage of stained cells. We assumed that hepatocellular nuclei appear as vague spheres and, on histological sections, as circles. Thus, V= 4 3 πr 3 , where V is the volume of the sphere, π is the Archimedes' constant, r is the radius, and A=πr 2 , where A is the area of the circle. Therefore, A=π( 3 4 V/π)^⅔, where V is the mean volume of non-neoplastic hepatocytes (287 µm 3 ) reported by Jack et al [39]. The average nuclear area of rat hepatocytes is consequently 50-53 µm 2 [39,40]. Nuclei larger than 40 µm 2 were considered to pertain to hepatocytes. This threshold was chosen to accommodate for the different levels at which nuclei might have been cut.

RNA extraction, cDNA synthesis and real-time quantitative polymerase chain reaction (RT-qPCR)
Total RNA was extracted from homogenized rat liver with the RNeasy Mini kit (Qiagen, Hilden, Germany)

RNA sequencing (RNAseq)
Sequence libraries were prepared from the liver RNA extracts with the Lexogen QuantSeq 3' mRNA-Seq library prep kit according to the manufacturer protocol. Samples were indexed to allow for multiplexing.
Library quality and size range was assessed using a Bioanalyzer (Agilent Technologies, Santa Clara, US-CA) with the DNA 1000 kit (Agilent Technologies). Libraries were subsequently sequenced on an Illumina HiSeq4000 instrument. Single-end reads of 50 bp length were produced with a minimum of 1M reads per sample. Quality control of raw reads was performed with FastQC 0.11.7 [41]. Adapters were filtered with ea-utils fastq-mcf v1.05 [42,43]. Splice-aware alignment was performed with HiSAT2 against the Rattus norvegicus reference genome (genome build=Rnor6.0). Reads mapping to multiple loci in the reference genome were discarded. Resulting BAM files were handled with Samtools v1.5 [44]. Quantification of reads per gene was performed with HT-seq Count v2.7.14. Count-based differential expression analysis was done with R-based Bioconductor package DESeq2 [45]. Reported P values were adjusted for multiple testing with the Benjamini-Hochberg procedure, which controls false discovery rate (FDR). The raw RNAseq data are available from http://www.ncbi.nlm.nih.gov/bioproject/705692.
Results from the DESeq2 package were used to perform Over-Representation Analysis (ORA) and Gene Set Enrichment Analysis (GSEA) with the WebGestaltR package [46]. These analyses were made on six from Gene Ontology (GO) [47,48], the Kyoto Encyclopedia of Genes and Genomes (KEGG) [49], the Reactome [50], and from the Panther databases [51].

Statistical analysis
Continuous variables were reported as medians and interquartile ranges (IQR). Differences in medians were tested with the Mann-Whitney U test. Overall differences in distributions between groups were tested with the Kruskal-Wallis omnibus test. Post-hoc tests were run to identify single differences by adjusting for multiple testing (Dunn's test). Dichotomous variables were explored using the maximum likelihood χ2 test or the Fisher's exact test as appropriate. Significance was retained at P<0.05. Analyses were run with SPSS 25.0 (IBM Corp., Armonk, US-NY) and GraphPad Prism 8.0 (GraphPad Software Inc., San Diego, US-CA).

Pharmacological pilot study
Roxadustat was well tolerated. HIF1α nucleus-cytoplasm ratio was higher, one hour after injection, only in RXD25 compared to baseline. VEGF gene expression was higher in RXD25 than in RXD10 or in HAL, one hour after injection. Likewise it was higher than in HAL six hours after injection. At 6h, PDGFβ gene expression was higher in RXD25 than in HAL, the expression of AGPT2 was higher in RXD25 than in RXD10 ( Figure 2). As roxadustat stabilized liver HIF1α and drove an HIF-related gene signature only at 25 mg/kg, this dosage was chosen to proceed further.

Surgery
Operations took 40 (IQR=38-45) min to be completed, with no differences between groups (Supplemental Information 1). The hepatectomy entailed a similar removal of 6.2 g of mass (IQR=6.0-6.8, P=0.229). The remnant liver, estimated from a pool of rats of comparable age and body weight, amounted to 2.7 g (IQR=2.6-2.9). The absolute and relative growth of the liver remnant was comparable between the two operated groups at all time points except for d1, when 70%Hx-RXD25 rats showed a reduced liver weight gain compared to 70%Hx-placebo (Figure 3). The HIF1α nucleus-cytoplasm ratio was greater in 70%Hx-RXD25 livers compared to 70%Hx-placebo at 1h (P=0.022). After this first time point, no differences between groups were detected. At no time point, the HIF2α nucleus-cytoplasm ratio was significantly different between the three groups ( Figure 4).

Complete blood count
The red blood cell count showed that, only in 70%Hx-RXD25, there was a surge in red blood cell count on d2 compared to d1 (P<0.001), and to 6h (P=0.003). Concordantly, hemoglobin concentration was greater on d2 compared to d1 (P<0.001), and almost significantly greater if compared to 6h (P=0.050), hematocrit was greater on d2 compared to d1 (P<0.001), and to 6h (P=0.003) Platelet count was greater in 70%Hx-RXD25 compared to 70%Hx-placebo and sham-placebo on d1. Peripheral white blood cells count was higher in 70%Hx-RXD25 compared to 70%Hx-placebo on d2. This divergence is accounted for by differences in lymphocytes and monocytes ( Figure 5).

Plasma analytes
Resected rats exhibited a similar rise in transaminases compared to sham-placebo, with a peak on d1 and a partial decline on d2 ( Figure 6). While albumin levels are overall stable after liver resection, resected rats showed lower levels on d1 compared to sham-placebo. We observed an increase in factor V concentration in resected rats on d1, with no difference between 70%Hx-RXD25 and 70%Hx-placebo. Ammonia levels were higher in 70%Hx-RXD25 than in sham-placebo at one, 22, and 47 hours after hepatectomy. On d1 and d2, lactate levels were higher in 70%Hx-RXD25 than in sham-placebo. Similarly, total bilirubin in 70%Hx-RXD25 showed a modest but significant increase compared to sham-placebo on d1 and d2, and to 70%Hxplacebo only on d1. On d1, pH was higher in 70%Hx-RXD25 than in sham-placebo. Creatinine was within normal range at all time points.

Gd-EOB-DTPA MRI
On d1, the time to peak was similar among groups. AUC60 was significantly reduced in 70%Hx-RXD25 compared to 70%Hx-placebo and to sham-placebo. AUC90 data followed a similar trend (not shown).
Accordingly, the AUC from t0 to peak, which accounts for contrast uptake, was reduced in 70%Hx-RXD25 compared to sham-placebo. The total AUC, which accounts for total contrast exposure, was one third in 70%Hx-RXD25 compared to 70%Hx-placebo and sham-placebo, though the time to 30% decrease was Downloaded from http://portlandpress.com/clinsci/article-pdf/doi/10.1042/CS20210183/921101/cs-2021-0183.pdf by guest on 23 September 2021 similar among groups. On d2, except for the time to peak that was longer in 70%Hx-RXD25 than in 70%Hxplacebo, no parameter showed significant differences (Figure 7).

Survival
We recorded comparable survival rates, with only 4 events out of 34 animals in the 70%Hx-RXD25. All the events happened during or soon after d1 MRI (Supplemental Information 2).
Immunofluorescence (Figure 8) While unaffected on d1, the proportion of Ki67-stained cells, increased significantly on d2 in regenerating lobes, with no differences between 70%Hx-RXD25 and 70%Hx-placebo. Hepatocytes, i.e., liver cells with large round nucleus, were the main contributor to this increase. A similar behavior was observed at EdU immunofluorescence, labeling cells in S-phase of cell cycle (Figure 9).

Histology
Mitoses were not detected on d1 and were extensively displayed on d2 in resected rats without differences between 70%Hx-RXD25 and 70%Hx-placebo. On d1, inflammation, endothelial denudation, microvascular thrombosis, parenchymal hemorrhage, and necrosis were virtually absent from all sections. Hepatocellular ballooning was detected in all 70%Hx-placebo samples but only in 2/13 70%Hx-RXD25 and in 1/8 shamplacebo. Steatosis occurred in all but one 70%Hx-placebo samples, mainly as microvesicular and midzonal.
In contrast, fatty infiltration was detected in only 5/13 of liver remnants in the 70%Hx-RXD25 group. When steatosis was present, its degree was less severe in 70%Hx-RXD25 than in 70%Hx-placebo ( Figure 10).
On d2, inflammation, endothelial denudation, parenchymal hemorrhage, and necrosis were virtually absent from all sections. Only one instance of microvascular thrombosis was detected in the 70%Hx-placebo group. Hepatocellular ballooning was milder and clearly detectable only in one case of 70%Hx-RXD25.
Steatosis lost zonation and was generally diffuse, appearing in all cases of resected rats without differences in proportion or in degree (Table 1).

RNAseq (Figures 11-12)
The relief from hepatocellular ballooning and steatosis, the only significant effect of roxadustat as opposed to placebo in the context of 70%Hx, was identified after 22h from roxadustat administration. Liver samples coming from the three experimental groups underwent analysis of global transcriptome changes with an RNAseq approach. The analysis is reported in Supplemental Information 3. We detected 2,521 genes whose expression was significantly different in 70%Hx-RXD25 compared to 70%Hx-placebo. The principal component analysis separated the samples according to the treatment. We employed ORA to determine if the significantly modulated genes likely originated from any known gene pathways, while GSEA was performed to assess the enrichment of up-or down-regulated genes in these pathways. We identified nine clusters of cellular functions that were recurrently enriched in 70%Hx-RXD25 compared to 70%Hx-placebo.
Three of those pathway clusters appeared generally upregulated: HIF signaling, energy production via oxidative phosphorylation, and lipid metabolism, while six pathway clusters were downregulated: cell growth and death, oxidative stress, xenobiotic metabolism, transmembrane transporters, inflammation, endocrine signaling (Supplemental Information 4). The comparison between the 70%Hx-RXD25 group and the sham-placebo group showed that the former exhibits an upregulation of cell cycle genes and the HIF1α pathway. The comparison between the 70%Hx-placebo group and the sham-placebo group showed that the former exhibits an upregulation of cell cycle genes. Both resected groups showed a down-regulation in lipid metabolism genes as opposed to the sham group (Supplemental Information 5).

Discussion
Liver resection causes a rapid induction of HIF1α-associated main intracellular signaling pathways in a murine model of both standard (68%) and extended liver resection (86%) [52]. Activation of HIF-driven pathways could be a consequence of the immediate post-resection hypoxic conditions, caused by the overflow of oxygen-poor portal blood in a partially resected liver and the consequent buffering reduction in oxygen-rich arterial supply. This ischemic phase is probably required to trigger the early phase of liver regeneration [5]. Interestingly, no later than 32 hours after standard two-third hepatectomy, hypoxiainduced pathways are rapidly downregulated. Hypoxia may thus be an important fine-tuning mechanism to induce liver regeneration [52], which acts in the very first days after liver resection. This might explain why even extreme resections performed in the setting of a reinforced hypoxia evolve favorably [4]. This is the first reported experience with roxadustat, a selective HIF stabilizer, in partial liver resection. In terms of safety, roxadustat tolerance was unexplored in this kind of hepatobiliary surgery. Data in the literature report that healthy adult rats should well tolerate a single dose of Roxadustat 25 mg/kg [25]. A few roxadustat-treated rats, but none of the placebo-treated, died on the first day after liver resection.
Remarkably, death occurred during of immediately after MRI imaging, suggesting that roxadustat might result from a reduced tolerance to prolonged anesthesia after liver resection. Besides, we cannot fully exclude an unexpected toxicity from the association between roxadustat and Gd-EOB-DTPA.
While liver HIF1α stabilization vanishes by the sixth hour from roxadustat administration, RNAseq confirmed that the roxadustat-induced upregulation of HIF1α-associated pathways is detectable at least until 22 hours later. Similarly, the rise in the erythroid lineage on d2 along with the increased platelet release on d1 [53] support the evidence for an efficient HIF1α stabilization. Roxadustat has been indeed developed to raise the levels of erythropoietin and, upon roxadustat administration, we registered a blood cell response. Moreover, erythropoietin per se exerts an influence on liver regeneration and metabolism, by inducing hepatic proliferation and decreasing lipid accumulation in the liver [54,55]. Then we additionally measured the plasma levels of erythropoietin to investigate whether our findings could be partly explained by a raise in erythropoietin release but the results did not confirm this hypothesis (Supplemental Information 6).
At variance with data concerning genetic PHD1 silencing in a murine model of liver ischaemia [56], we could not detect a roxadustat-induced differential HIF2α stabilization. Other less selective PHD inhibitors prompted a response similar to ours: nucleus pulposus cells, once exposed to DMOG, showed an accumulation of HIF1α and an upregulation of its target genes, but not of HIF2α [57]. These data, the evidence that both HIF1α and HIF2α are stabilized after hepatectomy [4], along with our experience suggest that proteasomal degradation of HIF2α is only partly or not primarily mediated by classical oxygendependent PHD pathway after liver resection.
Overall, hypoxia is supposed to coordinate murine liver regeneration by coupling parenchymal growth to vascular expansion [4,5]. However, because fatalities occur during the first 48 hours after liver resection in murine models and hypoxia-driven response rapidly fades out after 32 hours, the putative protective effect of HIF stabilization should entail very early functional adjustments consequent to changes in hepatocellular energetic status [58]. Therefore, beside morphological regeneration, we evaluated function, essential to claim that a treatment has an effect whatever on postoperative liver failure [59]. More than crude mortality rate, liver function is a reproducible and meaningful endpoint because it reflects the pathophysiology of human hepatocellular dysfunction in terms of synthesis, detoxification, inflammation, necrosis, and collateral damage [60].
Gd-EOB-DTPA uptake appeared generally hindered in roxadustat-treated resected rats as assessed by the reduction in AUC60 and AUC90 on d1, and the delay of the time to peak on d2 compared to 70%Hx-placebo and sham-placebo. Gd-EOB-DTPA employs organic anion-transporting polypeptides (OATPs) to enter hepatocytes following the electrochemical gradient across plasma membrane [61][62][63]. Thus, decreased AUC could correspond to a decreased uptake. Correspondingly, RNAseq confirmed that roxadustat specifically inhibited the expression of Slco1a1, the gene encoding OATP1, in the context of liver resection (Supplemental Information 3). Inversely, we did not detect a significant effect on Gd-EOB-DTPA biliary excretion, which relies on the ATP-dependent canalicular membrane multidrug resistance proteins [61][62][63].
We did not gather evidence that roxadustat improved hepatocellular function postoperatively. Yet, to our knowledge, this is the first established rat model of standard hepatectomy, in which hepatocellular function was detailed through contrast-enhanced MRI and which showed a substantial invariability in MRI parameters between sham-operated and 70%-liver resected rats. Previously, in a rat model of 70% and 90% liver resection, the vascular clearance of Gd-EOB-DTPA was utilized to identify liver function differences in a plasma pharmacokinetic study [64]. In our experience, other measures of liver function did not appear to be influenced by roxadustat compared to 70%Hx-placebo. In particular, synthesis of albumin or factor V, clearance of lactate, ammonia or bilirubin, and transaminases showed no differences.
The current literature concerning the models of liver regeneration after hepatectomy almost invariably emphasizes results in terms of death rate, immunohistological measures of cell proliferation, and increase in remnant size [56], as proxies for improved liver regeneration or reduced SFSS [4,56]. However, while death is an inhumane endpoint [65], evidence coming from SFSS-associated rodent mortality can hardly be translated to humans [66]. Moreover, graft size is influenced by inflammation and edema, and it does not necessarily correlate with function. Finally, the fine-tuned study of liver function, much advocated in order to produce meaningful insights into the physiology of liver regeneration, should be routinely reported [67]. After extreme resections, liver regains mass for up to seven days and the difference between livers exposed to hypoxia or not peaks on postoperative day three [4]. In the standard model of liver resection we used, there is little room for improvement because the hepatostat is reached within three days after 70%Hx [68].
Liver growth, as we assessed by size measurement, proved actually similar between the treatment groups, in keeping with the evidence coming from ki67 staining, EdU labeling and mitoses count. Our results are not in line with previous research with PHD inhibitors. Harnoss et al. observed that EDHB improved hepatocellular proliferation, measured on d2 after 75% hepatectomy [9]. Correspondingly, EDHB induced cyclin D1 and cyclin D2 upregulation after liver resection. These effects are HIF2α-dependent [69], and, while the authors did not show the differential profile of HIF1α-HIF2α stabilization in their experience, it is safe to assume that the growth-promoting effects of EDHB are mediated by an increase transactivity of HIF2α via a non-selective inhibition of its degradation. Whether this HIF2α stabilization is obtained via the inhibition of PHDs or other dioxygenases or a mix of the two remains unclear, because in our experience roxadustat, which effectively and selectively inhibits PHDs, stabilized only HIF1α and did not promote liver regeneration after hepatectomy. Conversely, the data coming from RNAseq unanimously underline that the HIF1α-stabilizing treatment dampens cell cycle progression.
The most relevant effect observed upon the administration of roxadustat in 70%Hx was a marked improvement in hepatocellular ballooning and steatosis, known hallmarks of hepatocellular stress, on d1.
Fatty infiltration of the liver after partial resection has been known for 70 years [70]. After hepatectomy, circulating fatty acids are inefficiently metabolized by the remnant liver, and infiltrate parenchyma, resulting in diffuse hepatocellular steatosis and, in some instances, to cell death [70][71][72]. This peculiar postresection injury has been shown to peak between day one and day two after 70%Hx, while it resolves and disppears thereafter in surviving rats. The RNAseq performed on day-one liver samples showed that roxadustat tilts the balance of cellular functions, favoring lipid metabolism rather than cell replication, exactly by the moment of the worst lipid overload. Whether such a shift streamlines the postoperative recovery has to be ascertained. Still, Vic et al. previously proved that relieving steatosis via testosterone administration raised rat survival rate after 90% hepatectomy [71]. In set of additional experiences, we evaluated whether the level of intrahepatic or circulating portal and peripheral lipids might contribute to Downloaded from http://portlandpress.com/clinsci/article-pdf/doi/10.1042/CS20210183/921101/cs-2021-0183.pdf by guest on 23 September 2021 the explanation of our findings but the results of these extra measurements did not show differences between 70%Hx-RXD25 and 70%-placebo groups (Supplemental Information 6).
The improvement in cellular stress obtained after roxadustat treatment was an unforeseen finding and it may pave the way for the exploration of the role of selective HIF1α stabilization after liver resection. In other experimental settings, Roxadustat has shown to alleviate liver fatty infiltration in two zebrafish atp7b deficiency models of human Wilson's disease [73]. Since, in our surgical model, transient steatosis tends to physiologically vanish in surviving rats after d2, we could not expect to find a statistically significant difference after this time point, and we decided to discontinue the experiment on the grounds of ethical considerations.
It remains to be established whether isoquinolines are a treatment for fatty liver infiltration at large thanks to their action on steatosis and ballooning. Nonetheless the effect of roxadustat on steatosis and ballooning looks promising for the constellation of non-alcoholic fatty liver disease and steatohepatitis, whose incidence is alarmingly soaring in the general population. These conditions are known to be associated with defective liver regeneration after resection and with increased rate of primary non-function after transplantation if steatotic livers are used as grafts. The mechanism by which roxadustat might prompt the liver to process lipids is elucidated by the RNAseq analysis: a selective HIF1α stabilization largely elicited the transcription of genes coding for lipid metabolism and oxidative phosphorylation. What is more elusive is the possible significance of this effect in the context of hypoxia. Anaerobic metabolism is usually preferred in conditions of hypoxia [26,74], and, in fact, PHDs deletion [75][76][77] and vHL loss [78][79][80] originate the classical cellular response to hypoxia: decreased mitochondrial activity, increased glycolysis, and glycogen and lipid accumulation. However, increased cellular respiration results in a positive energy balance, a prerequisite for an improved liver function during regeneration. While not manifested by indicator other than decreased cell stress and ballooning in a paraphysiological standard 70%Hx, the exploration of roxadustat in hepatic surgery on steatotic livers is an intriguing perspective.
In conclusion, it remains unproven whether a roxadustat-induced selective HIF stabilization enhances function or proliferation during the early postoperative period after standard hepatectomy, when hypoxia responses are at their acme. However, roxadustat improved hepatocellular ballooning and fatty infiltration,

Clinical Perspectives
1. Hypoxia triggers liver regeneration after resection but it is controversial to actively induce hypoxia in patients undergoing hepatectomy. However, a hypoxic response can be simulated by means of roxadustat, a selective prolyl hydroxylase inhibitor that artificially stabilizes hypoxia inducible factor-1α (HIF1α) levels.
2. This study showed the administration of roxadustat concomitant with liver resection alleviated postoperative steatosis and ballooning, two hallmarks of liver suffering after resection, by raising HIF1α and increasing oxidative phosphorylation.
3. Selective HIF stabilization does not accelerate liver regeneration but promotes cellular respiration and mitigate steatosis, suggesting a possible role for roxadustat in the surgery of steatotic livers and in the treatment of fatty liver disease at large.

Data availability statement
The data that support the findings of this study are available from the corresponding author, upon reasonable request. The raw RNAseq data are available from http://www.ncbi.nlm.nih.gov/bioproject/705692.