RASSF1A independence and early galectin‐1 upregulation in PIK3CA‐induced hepatocarcinogenesis: new therapeutic venues

Aberrant activation of the phosphoinositide 3‐kinase (PI3K)/AKT/mTOR and Ras/mitogen‐activated protein kinase (MAPK) pathways is a hallmark of hepatocarcinogenesis. In a subset of hepatocellular carcinomas (HCCs), PI3K/AKT/mTOR signaling dysregulation depends on phosphatidylinositol‐4,5‐bisphosphate 3‐kinase, catalytic subunit alpha (PIK3CA) mutations, while RAS/MAPK activation is partly attributed to promoter methylation of the tumor suppressor Ras association domain‐containing protein 1 (RASSF1A). To evaluate a possible cocarcinogenic effect of PIK3CA activation and RASSF1A knockout, plasmids expressing oncogenic forms of PIK3CA (E545K or H1047R mutants) were delivered to the liver of RASSF1A knockout and wild‐type mice by hydrodynamic tail vein injection combined with sleeping beauty‐mediated somatic integration. Transfection of either PIK3CA E545K or H1047R mutants sufficed to induce HCCs in mice irrespective of RASSF1A mutational background. The related tumors displayed a lipogenic phenotype with upregulation of fatty acid synthase and stearoyl‐CoA desaturase‐1 (SCD1). Galectin‐1, which was commonly upregulated in preneoplastic lesions and tumors, emerged as a regulator of SCD1. Co‐inhibitory treatment with PIK3CA inhibitors and the galectin‐1 inhibitor OTX008 resulted in synergistic cytotoxicity in human HCC cell lines, suggesting novel therapeutic venues.


Introduction
Hepatocellular carcinoma (HCC) is the most frequent primary malignant tumor of the liver and a leading cause of cancer mortality worldwide [1]. North America and Europe experienced a constant rising incidence of HCC over the last decades [2]. Systemic therapeutic options for advanced disease stages are scarce, albeit the combination treatment of atezolizumab and bevacizumab is one recent glimmer of hope [3]. The pathogenesis of HCC is a multistep process involving the progression from cirrhosis to low-grade dysplastic nodule, high-grade dysplastic nodule, and, ultimately, HCC [4,5]. The precise molecular mechanisms and their specific contributions to hepatocarcinogenesis remain poorly understood. Ras/mitogen-activated protein kinase (MAPK) signaling is among the heavily implicated pathways in HCC, despite the absence of activating RAS mutations [6,7]. Instead, diminished activity of Ras antagonists represents a plausible explanation for Ras/MAPK pathway unconstrained activity [8][9][10]. A critical physiological counterpart of Ras is Ras association domain-containing protein 1 (RASSF1a), which is an established tumor suppressor embedded into an intricate regulatory network [11]. RASSF1A gained importance as it is frequently epigenetically silenced by aberrant hypermethylation in diverse types of cancer [12], including HCC [6,13]. In addition to being a crucial negative-feedback downstream effector of Ras Signaling [14], RASSF1A is a regulator of microtubule stability, Rho GTPases, and the Hippo pathway [15][16][17]. Thereby, multiple functions are exerted, such as cell cycle arrest, inhibition of migration, and metastasis through microtubular stabilization, reduction of epithelial-mesenchymal transition (EMT), and induction of apoptosis [18][19][20]. Interestingly, RASSF1A hypermethylation can be therapeutically reversed with panobinostat, a pan-deacetylase inhibitor, which affects DNA methyltransferases' activity and expression, as demonstrated in a HepG2 xenograft model [21]. In HCC, transfection of RASSF1A yielded growth retardation in cell lines and the respective murine xenografts [22]. Furthermore, homozygous deletion of RASSF1A elicited late liver tumor susceptibility [23] and accelerated diethylnitrosamine-induced HCC formation via autophagy promotion [24]. This body of evidence leaves little doubt on the oncogenicity of RASSF1A knockout (KO). However, the magnitude of the observed effects in the latter experimental models was somewhat limited, so that concurrent oncogenic stimuli are necessary together with RASSF1A loss to induce hepatocarcinogenesis.
Activation of the phosphoinositide 3-kinase (PI3K)/ RAC-alpha serine/threonine-protein kinase (AKT)/ mTOR pathway could well qualify as a promising therapeutic candidate. Indeed, the PI3K cascade is one of the most deregulated pathways along tumorigenesis [14]. PI3K is a heterodimeric lipid kinase composed of a catalytic and a regulatory subunit (p110a or b or d and p85a or b, respectively, for class Ia PI3K), which generates phosphatidylinositol (3,4,5)-trisphosphate (PIP 3 ) [25]. A pivotal downstream effector is AKT, which binds to PIP3 and is subsequently activated by phosphorylation. Once induced, AKT mediates various targets and regulates numerous cellular functions, such as cell cycle progression, growth, survival, and metabolism [26,27]. Several genetic events triggering PI3K/ AKT/mTOR activation have been reported [28]. Among them, activating mutations of the PI3K 110a subunit (PIK3CA) occupy a prominent position [29]. Important hot spot mutations include PIK3CA E542K, E545K, and H1047R, where the former affect the protein's helical domain and the latter its kinase domain [30][31][32]. The reported PIK3CA mutational frequency in human HCCs ranges from 4% to 6% [7,33]. Murine models have only demonstrated a hepatic cocarcinogenic effect for activated PIK3CA in conjunction with either activated yes-associated protein 1 (YAP), RasV12, or c-Met to date [34,35]. Indeed, injection of PIK3CA mutant forms alone promoted hepatic steatosis but did not induce hepatocarcinogenesis [35].
Here, we investigated a possible interplay between RASSF1A loss and the PI3K/AKT/mTOR pathway using an animal model of hydrodynamic tail vein injection with sleeping beauty-mediated somatic integration. E545K and H1047R PIK3CA mutant forms were injected into homozygous RASSF1A KO mice. The resulting neoplastic lesions were subjected to a gene expression microarray analysis to shed light on PIK3CA-driven hepatocarcinogenesis and identify putative therapeutic targets.

Mouse breeding and genotyping
A RASSF1A KO founder breeding pair was kindly provided by L. van der Weyden (Wellcome Trust Sanger Institute, Research Support Facility, Hinxton, Cambridge, CB10 1SA, UK). The genetic background of RASSF1A wild-type (WT) and KO mice was C57BL/6J 9 129Sv. RASSF1A WT mice were obtained by crossing RASSF1A KO mice with C57BL/ 6J mice purchased from Charles River Laboratories (Sulzfeld, Germany). Genotyping was performed on tail biopsies after DNA extraction. Polymerase chain reactions to detect the RASSF1A gene were carried out according to a previously established protocol [36]. The forward primer RSF-5 (5 0 -CTC GCC CCT GTC AGA CCT CAA TTT CCC-3 0 ) was applied together with the reverse primer RSF-3 (5 0 -CCA GGC TTC CTT CTC ACT CCT CTG CCG C3 0 ), which yields a 400 base pair product in RASSF1A KO mice, where Exon 1 a has been deleted. The product length was evaluated by gel electrophoresis.
All experimental mice were kept and bred under standard conditions and stored in type III Makrolon cages with 12-h light/dark cycles. The maximum number of mice was limited to five per cage. The animals were fed autoclaved food and water ad libitum.

Hydrodynamic injections, mouse monitoring, and tissue sampling
Male mice of 6-8 weeks were subjected to hydrodynamic tail vein injections in any of the following groups: untreated, 19 PBS, EV; pT3-EF1a-PIK3CA E545K, or pT3-EF1a-PIK3CA H1047R. Hydrodynamic tail vein injections were performed as described elsewhere [37]. Briefly, 10 µg of pT3-EF1a-PIK3CA, pT3-EF1a-PIK3CA H1047R, or pT3-EF1a was combined with pCMV/sleeping beauty transposase in a ratio of 25 to 1 in a total volume of 2 mL 0.9% sodium chloride. The solution was filtered with 0.22µm mesh size. The total volume was injected into the lateral tail vein within 5-7 s. To attain at least 6-8 evaluable mice per time point and group, 10 animals per time point were injected in the experimental groups. Since control groups could be subjected to a combined analysis, only five animals were injected per condition and timepoint (Table 1).
Animals were excluded from the evaluation if the injected total volume was below 2 mL. Another criterion to assess the quality of injection was the mouse behavior following the injection. Successfully injected mice displayed a decreased activity for at least 60 min due to the systemic volume challenge. Mice that did not show a similar behavior were excluded from the experiment. Mice were monitored daily and kept until the prespecified experimental time point. Respiratory distress, lethargy, and palpable liver masses equivalent to a size of~3.5-4 cm were defined as termination criteria. The cervical dislocation was applied. A photodocumentation of the livers was performed, after  (N)  1 week  10  10  10  10  5  5  5  5  5  5  1 month  10  10  10  10  5  5  5  5  5  5  3 months  10  10  10  10  5  5  5  5  5  5  6 months  10  10  10  10  5  5  5  5  5  5  9 months  10  10  10  10  5  5  5  5  5  5  12 months  10  10  10  10  5  5  5  5  5  5  Sum of animals per experimental group  60  60  60  60  30  30  30  30  30

Histology, immunohistochemistry, image acquisition, and proliferation index
At least two board-certified pathologists and liver experts conducted the histopathological assessment of the liver lesions (KU, KE). The liver tissue was processed following the standard diagnostic institutional guidelines. Two-micrometer-thin histological sections were cut from formalin-fixed, paraffin-embedded tissue samples. The sections were deparaffinized through a series of xylene and gradient alcohols to water. For antigen retrieval, the slides were heated in a microwave oven for 10 min, while 10 mM sodium citrate buffer with a pH of 6.0 was applied. Afterward, the samples were cooled down to room temperature. The slides were treated with 19 Dako Peroxidase-Blocking Solu-tionÒ (cat S2023; Agilent Technologies Inc., Santa Clara, CA, USA) for 10 min. The primary antibody was diluted and applied in Dako Antibody DiluentÒ (cat. S2022, Agilent Technologies Inc.); the slides were placed in a humidity chamber and incubated at room temperature overnight (see Table 2 for a list of all primary antibodies Images were acquired with the slide scanner Pannoramic 250 Flash IIIÒ (Sysmex, Kobe, Japan) with a 209 objective. Subsequently, images were displayed after stitching using the software CaseViewerÒ (Sysmex). With 300 pixels per inch, resolution screenshots were taken.
The stitched images of Ki67 immunohistochemistry were imported to the DeePathology TM STUDIO software (DeePathology.ai; Raanana, Israel). A boardcertified pathologist (KU) trained the algorithm to discern hepatocyte nuclei from nuclei of other cell types. These were determined as background cells (such as lymphocytes and Kupffer cells) and excluded from the analysis. Regions of interest were manually selected. A graphical representation of the calculation results was visually assessed for further validation.

Western blot analysis
For protein extraction, lysis of cells and mouse liver tissues was achieved using the Mammalian Protein Extraction Reagent (cat 78501; Thermo Fisher Scientific, Waltham, MA, USA) associated with the Halt Protease Inhibitor Cocktail (cat 78429; Thermo Fisher Scientific). An additional mechanical force was applied for tissue homogenization with the Next Advance Bullet BlenderÒ Storm 24 (Next Advance, Inc.; Troy, NY, USA). The lysates were incubated for 30 min at 4°C while vortexing every 5-10 min. A centrifugation step for 30 min (> 20 000 g, 4°C) ensued. Concentrations of total protein were assessed using the Bradford Protein Assay [38]. To attain a standard curve for linear regression, serial dilutions of BSA were employed.
For western blot analysis, protein lysates were separated by sodium dodecyl sulfate/polyacrylamide gel electrophoresis. 2.5 lg of total proteins was loaded onto BoltTM 4-12% Bis-Tris Plus gels (Thermo Fisher Inc.) at 150 V for 30-60 min. Next, gels were incubated in 10% ethanol for dehydration before blotting the proteins to a membrane using the BlotTM 2 Gel Transfer Device (Thermo Fisher Inc.). After staining the membranes in Ponceau solution, they were placed in EveryBlot Blocking Buffer (Bio-Rad Laboratories, Inc.; Hercules, CA, USA) for 15-30 min at room temperature. Primary antibodies ( Table 2) were diluted in the blocking buffer and incubated overnight at 4°C. The next day, membranes were washed in Tris-buffered saline with TweenÒ (Cell Signaling Technology, Inc., Cambridge, UK) for 5 min at room temperature. Then, the secondary antibody was applied at room temperature for 1 h. The membranes were washed with Tris-buffered saline with TweenÒ. The chemifluorescent signal was visualized by Clarity Max TM Western ECL Substrate (Bio-Rad Laboratories) on a ChemiDoc TM MP Imaging System (Bio-Rad Laboratories). For quantitative analysis of band intensities, version 6.1 of the software IMAGELAB (Bio-Rad Laboratories) was employed. Values were normalized to the corresponding b-actin bands.

Transmission electron microscopy
Liver tissue samples were fixed in 0.1 M cacodylatebuffered Karnovsky fixative containing 2.5% glutaraldehyde and 2% paraformaldehyde (Electron Microscopy Sciences, Hatfield, PA, USA) overnight at room temperature with a subsequent postfixation in 1% osmium tetroxide (Electron Microscopy Sciences), which was applied for 2 h. Next, the samples were dehydrated in graded ethanol (Sigma-Aldrich). Afterward, they were embedded in an EMbed-812 epoxy resin (Electron Microscopy Sciences). Following 2 days of heat polymerization at a temperature of 60°C, 0.8 µm thin sections were prepared. These were stained with toluidine blue (Agar Scientific; Essex, UK) and basic fuchsine solution (Polysciences Inc.; Warrington, PA, USA). Subsequently, the epon block was adjusted to allow ultrathin sectioning. Eighty-nm sections were cut with a diamond knife on a Reichert Ultracut-S ultramicrotome (Leica, Wetzlar, Germany). These were double contrasted using aqueous 2% uranyl acetate (Honeywell International Inc., Morristown, NJ, USA) and lead Germany) operated at 100 kV was used for imaging the ultrathin sections.

Gene expression microarray hybridization and analysis
For comparative microarray of mouse livers, the RNA was isolated from liver tissue using the NucleoSpinÒ RNA Plus Kit (Macherey-Nagel GmbH & Co. KG, D€ uren, Germany), following the manufacturer's instructions. Gene expression microarray analysis was performed with mouse SentrixÒ BeadChips (Illumina, San Diego, CA, USA). We followed the workflow established by the manufacturer. Briefly, cDNA synthesis, in vitro transcription, and cleanup were conducted using the AmbionÒ RNA Amplification Kit (Illumina). The samples were quantified after adding Molecular Probes Ribo GreenÒ (Thermo Fisher Scientific) by fluorometric measurement. Afterward, the samples were hybridized, and an image was extracted with the BeadArray Reader (Illumina). The data were analyzed with the BeadStudioÒ application (Illumina). Raw data were background corrected, quantilenormalized, and log2-transformed using the limma package [39] provided by R/Bioconductor (https:// www.bioconductor.org). Subsequently, differentially expressed genes were calculated, using the limma package. Concerning multiple testing, Benjamini and Hochberg's method was applied to adjust the P-values. Heatmaps were generated with the pheatmap package provided by R/Bioconductor (https://www.bioconductor. org). Furthermore, preranked Gene Set Enrichment Analysis (GSEA; Hallmark gene sets) was performed using the official software tools from the Broad Institute (Boston, MA, USA, https://www.gsea-msigdb.org) [40]. For this purpose, all probes representing the same gene symbol were averaged.
For comparative microarray of stably transfected HCC cell lines, the sample preparation for microarray hybridization was conducted as described in the Applied Biosystems TM GeneChip TM Whole Transcript (WT) PLUS Reagent Kit User Guide (Thermo Fisher Scientific). In brief, 200 ng of total RNA was used to generate double-stranded cDNA. Twelve microgram of subsequently synthesized cRNA was purified and reverse transcribed into single-stranded (ss) cDNA, with unnatural dUTP residues incorporated. Purified ss cDNA was fragmented using uracil DNA glycosylase (UDG) and apurinic/apyrimidinic endonuclease 1 (APE 1) followed by terminal labeling with biotin. Subsequently, 3.8 µg of fragmented and labeled ss cDNA was hybridized to Applied Biosystems TM Gene-Chip TM Clariom S human arrays for 16 h at 45°C and 60 r.p.m. in an Applied Biosystems TM GeneChip TM hybridization oven 640. Hybridized arrays were washed and stained in an Applied Biosystems TM Gene-Chip TM Fluidics Station FS450, and the fluorescent signals were measured with an Applied Biosystems TM GeneChip TM GeneChip Scanner 3000 7G System. Fluidics and scan functions were controlled by the Applied Biosystems TM GeneChip TM Command CONSOLE v5.0 software (Thermo Fisher Scientific Inc.).
Sample processing was performed at a Genomics Core Facility, 'KFB-Center of Excellence for Fluorescent Bioanalytics' (Regensburg, Germany; www. kfb-regensburg.de). For the data analysis, summarized probe set signals in log 2 scale were calculated by using the GCCN-SST-RMA algorithm with the Applied Biosystems TM GeneChip TM Expression CONSOLE v1.4 Software. After exporting into Microsoft Excel comparison, fold changes were calculated. Probe sets with a fold change above 2.0 fold were considered significantly regulated.
Heatmaps were generated with the Transcriptome Analysis Console (Applied Biosystems, Waltham, MA, USA). GSEA analysis was performed using the WEBbased GEne SeT AnaLysis Toolkit [41] by selecting the Panther pathway functional database.
We performed cell viability assays using the xCEL-LigenceÒ real-time cell analysis dual plate (RTCA DP) device (OLS OMNI Life Science GmbH & Co KG; Bremen, Germany). For impedance-based realtime cell index measurement, cells were grown on E-Plate 16 PET (Agilent Technologies, Inc.). Measurement sweeps were acquired every 15 min. Six thousand two hundred fifty cells suspended in a total volume of 150 µL of growth medium were seeded in each well. After 24 h, varying concentrations of dimethyl sulfoxide (Sigma-Aldrich), the inhibitory compounds alpelisib (MedChemExpress; LLC., Monmouth, NJ, USA), and/or OTX008 (MedChemExpress) were added. Afterward, the measurement was acquired for a total of 72 h. Raw data were analyzed with the RTCA software (OLS OMNI Life Science GmbH & Co KG). The data were normalized to the timepoint of inhibitor addition. Synergism was evaluated 24 h after adding the inhibitory compounds using the software COMPUSYN (ComboSyn, Inc., Paramus, NJ, USA), which generates dose-effect curves and isobolograms, and determines the combination index.
For alpelisib single treatment, cells were seeded into 6-well plates at a density of 3-5 9 10 5 cells in 2 mL medium per well. The next day, either alpelisib in varying concentrations or matched DMSO was added. After an incubation of 48 h, cells were harvested, centrifuged, and the cell pellet was used for further processing.
For galectin-1 (Gal-1) silencing, cells were seeded at a density of 3 9 10 5 cells in 2 mL of medium per well in six-well plates. Cells were transfected with SilencerÒ Select Negative Control #1 siRNA (Thermo Fisher Inc) or LGALS1 siRNA (Eurofins Genomics, Ebersberg, Germany) with the sense sequence 5 0 -[UUGCUGUUGCACACGAUGGUGUUGG]-3 0 the following day using LipofectamineÒ RNAiMAX (Thermo Fisher Inc) according to the manufacturer's instructions. Lipofectamine and siRNA were diluted in OptiMEMÒ Reduced Serum Medium (Thermo Fisher Inc.) and combined. Medium in the wells was discarded, and cells were washed with 19 PBS before adding the transfection solution. After an incubation period of 48 h, the transfection was repeated once more. Cells were harvested after an additional 24 h of incubation using cell scrapers. Harvested cell suspensions were centrifuged (300g, 5 min). The pelleted cells were used for further analyses.
For OTX008 treatments, cells were seeded at a density of 3 9 10 5 cells in 2 mL of medium per well in six-well plates. The next day, OTX008 (or an adjusted amount of DMSO) was added to a total concentration of 20 µM. After an incubation period of 48 h, cells were harvested using cell scrapers. Harvested cell suspensions were centrifuged (300 g, 5 min). The pelleted cells were used for subsequent western blot analysis.
Subsequently, SNU387 and SNU449 cells were infected with the virus and fresh Roswell Park Memorial Institute 1640 (RPMI 1640) medium at the volume ratio of 1 : 1. After 48 h of transfection, cells were treated with puromycin-containing media (1.5 lgÁ mL À1 ) to select cells with stable expression of the target gene with the puromycin resistance.
To determine the half-maximal inhibitory concentration (IC50) values for OTX008 and alpelisib and perform combination treatments in the stably transfected cell lines, these were seeded into 24-well plates at 2.5 9 10 5 and treated with escalating concentrations of alpelisib/OTX008 or both compounds for 48 h. Subsequently, the cells were washed in PBS 3 times, followed by crystal violet staining for 10 min. After washing, the stained cells were incubated in lysis solution for 20 min. Next, the diluted solution was added to 96-well plates to measure OD values at 590 nm by the BioTek ELX808 Absorbance Microplate Reader (Thermo Fisher Scientific). The IC50 values were calculated based on the OD values. These experiments were repeated in triplicates.
Expected AB = E A + E B À EA EB E A and E B are the effects of the two compounds in monotherapy, measured as the reduction in viability. According to the Bliss model, the observed effect of the combination therapy is directly compared with the expected effect and calculated as the excess over the Bliss score (EOB) [43]. If EOB is > 0, the combined compounds have an effect more potent than if they acted independently.
A compound library including 315 approved anticancer drugs was purchased from TargetMol (L2110; Target Molecule Corp., Boston, MA, USA), diluted to a concentration of 1 mM, and 40 nL was dispensed in 384-well plates (Greiner µClear #781091, Greiner Bio-One GmbH) using an acoustic liquid handler (EchoÒ 550, Beckman Coulter Life Sciences, Brea, CA, USA). PLC/PRF/5 cells were detached and separated in two tubes containing either OTX008 at a final concentration of 20 µM or DMSO vehicle. The cells were then seeded using a liquid dispenser (Multidrop TM Combi, Thermo Fisher Scientific) at 2500 cells per well in prespotted assay plates at a final volume of 40 µL per well to reach a final concentration of 1 µM of the compounds. Cells were incubated at 37°C, 5% CO 2 for 72 h. At the end of the incubation time, 10 µL of a solution of PBS and Hoechst 33342 (2 dropsÁmL À1 , NucBlue TM Live ReadyProbes TM Reagent; Thermo Fisher Scientific) was added to each well using a liquid dispenser (Multidrop TM Combi; Thermo Fisher Scientific), incubated for 1 hour and imaged for brightfield and Hoechst staining using the Operetta CLS TM High-Content Analysis System (PerkinElmer, Inc.). Stained nuclei were counted using the HarmonyÒ software (PerkinElmer, Inc.), and the number of nuclei measured cell viability and proliferation. For each plate, the proteasome inhibitor Carfilzomib was used as a positive control and DMSO as a negative control to calculate the Z 0 value as a plate QC criteria: All plates had a Z 0 > 0.7, above the widely accepted threshold of 0.5. Data from two independent experiments were averaged (correlation coefficient between replicates 0.96). The number of nuclei in each well was normalized to DMSO-treated controls, and normalized viability was used to calculate the EOB as described above. Compounds were defined as a hit if the observed normalized viability upon treatment in combination with OTX008 was below 0.6 and the Z score of EOB was > 1.5. Information regarding the mode of action and the targeted genes for each compound was obtained from the drug repurposing hub database [44].

Statistical analyses
Descriptive statistics compared tumor occurrence frequency. Comparisons between two groups were conducted with nonparametric Mann-Whitney U-tests due to small sample sizes (quantification of lipogenic enzymes in Gal-1 silencing experiments). For multiple comparisons, nonparametric data were compared using the Kruskal-Wallis test without adjustment (liver/ weight over bodyweight comparison; quantification of western blots for Gal-1 in PIK3CA E545K liver lysates; quantification of western blots in alpelisib treatment experiments).
Kaplan-Meier curves and log-rank tests were used to compare survival between RASSF1A WT and KO mice with PIK3CA E545K injection. A mixed linear model (maximum-likelihood estimation, unstructured repeated covariance type) was applied to evaluate repeated proliferation measures between RASSF1A KO and WT mice and assess the proliferation within normal tissue, preneoplastic lesions (PREs), and tumors. Moreover, corresponding injection groups were compared against each other separately in RASSF1A WT versus KO mice. GRAPHPAD PRISM version 9, GRAPHPAD PRISM R version 4.0.3 (GraphPad Software; San Diego, CA, USA), and SPSS version 26 (IBM; Armonk, NY, USA) were employed. All P-values were obtained in two-tailed tests, and P ≤ 0.05 was considered statistically significant.
Microarray data were analyzed using the R package Limma. Moderated contrast t-tests were computed, and Benjamini and Hochberg's method to adjust for multiple testing was applied to microarray expression data [45,46].
Contrary to our expectations from previous injections of mutant PIK3CA containing plasmids [34,35], mice in all PIK3CA injection groups developed liver tumors within the prespecified observation period of 12 months. This unexpected finding presumably depended on the mixed background used for the experiments (C57BL/6J 9 129Sv), which differs from the previously employed FVB/N inbred mouse strain [35].
Based on combined gross and histological examination, the observed lesions were stratified into the following categories: preneoplastic lipid-rich clusters, preneoplasias (criteria of expansive growth with initial compression of surrounding tissue (SR) and estimated cell content > 100), and tumors (irregular borders, presence of necrosis, expansive growth with evident compression or diffuse infiltration of SR, macroscopic correlation, cytologic signs of malignancy). Tumors in PIK3CA E545K-injected mice were already detectable after 3 months, instead of a 12-month latency of tumorigenesis in the PIK3CA H1047R injection groups. Surprisingly, H1047R injections yielded discernible tumors in only three of 112 mice within the defined observation time of 12 months. In contrast, E545K injections resulted in numerous tumors (tumors in 56 of 113 mice), which frequently necessitated a premature termination. Serving as solid evidence against cooperativity, neither clusters, PREs, nor tumors displayed a marked difference in the occurrence frequency when comparing PIK3CA mutant forms injected in RASSF1A WT and RASSF1A KO mice. A combined control group (~5 mice each without transfection, injection of 19 PBS, and transfection of the EV) did not demonstrate spontaneous tumorigenesis in RASSF1A WT mice. However, the RASSF1A KO mouse control group developed a total of two tumors at the experimental time points of 9 and 18 months (extended period of observation; Fig. 1B-F).
To assess differences in tumor burden between experimental groups, the liver weight/body weight ratio for the combined experimental time points ranging from 3 to 12 months was compared using the Kruskal-Wallis test (Fig. 1G). The post hoc test showed that PIK3CA E545K had increased oncogenic potency compared to H1047R and control in RASSF1A WT and KO mice (P-values < 0.0001). PIK3CA H1047R-injected mice were neither significantly different from control in RASSF1A WT nor KO mice (P-values > 0.05). Moreover, the lack of significant differences between RASSF1A KO and WT mice substantiated that RASSF1A loss does not increase tumor burden (P-values > 0.05). Next, we conducted a survival analysis in mice with only tumor-related deaths in PIK3CA E545K subgroups. Animals were deliberately euthanized based on the assessment of the following termination criteria: respiratory distress, lethargy, and palpable liver masses equivalent to a size of~3.5-4 cm. Premature termination occurred at the formerly denoted time points 3-6, 6-9, and 9-12 months. A minor, more prolonged survival of PIK3CA E545K-injected mice was detected in RASSF1A KO mice (median survival = 6.9 months) compared with RASSF1A WT mice (median survival = 6.5 months, P = 0.0330; Fig. 1H).
Altogether, these data indicate that RASSF1A loss does not increase the susceptibility to PIK3CA mutant form-mediated hepatocarcinogenesis and even shows a tendency to improve survival in the PIK3CA E545K group. Of note, the only two tumors observed in the control group emerged in RASSF1A KO mice. Even if these tumors were indeed RASSF1A related, a very long latency was required for tumorigenicity.
Next, we searched for potential histological differences between RASSF1A WT and KO mice. Histological analysis revealed that proliferative clusters, pericentrally located PREs, and tumors were characterized by a lipid-rich phenotype and possibly shared the common ancestor of disseminated lipid-rich cells, which can be observed in the liver acinus zone 3, corresponding to singular transfected cells. PIK3CA H1047R-injected mice preferentially developed discontinuous but expansive PREs in the sense of confluent clusters, while PIK3CA E545K-injected mice developed focal, coherent, and more rounded PREs. Tumors were predominantly well-differentiated and displayed a relatively low nuclear-cytoplasmic ratio and a low degree of nuclear pleomorphism while maintaining their lipid-rich phenotype. Visual examination of Ki67 revealed an increase in proliferation in tumors of E545K-injected mice, which was not readily apparent for the observed lipid-rich clusters in the H1047R injection group. There was no histologically evident difference in the morphology of PREs and tumors between RASSF1A WT and KO mice ( Fig. 2A,B).
To ascertain the preneoplastic nature of the described lesions, we conducted a deep learning-based analysis using the software DeePathology TM STUDIO after training the recognition of Ki67-positive and Ki67-negative hepatocyte nuclei and the omission of background cells (such as lymphocytes and Kupffer cells) on manually selected regions of interest (Fig. S1). A mean amount of 5543 cells for PIK3CA H1047Rinjected mice and 1650 cells for PIK3CA E545Kinjected mice were analyzed per animal.
Mixed linear models (Fig. 2C,D) demonstrated that PREs and tumors have a significant increase in proliferation compared with normal-appearing tissue for RASSF1A WT and KO in PIK3CA E545K-injected mice. At the same time, there was no significant difference in proliferation between PREs and tumors for RASSF1A WT and KO in PIK3CA E545K-injected mice. Also, the proliferation was not significantly different between RASSF1A WT and KO in PIK3CA E545K-injected mice in normal-appearing liver tissue and preneoplastic tissue (P-values > 0.05). However, when comparing the proliferation in tumors between RASSF1A WT and KO in PIK3CA E545K-injected mice, a statistically significant lower proliferation rate could be detected in RASSF1A KO mice (P = 0.0280).
For PIK3CA H1047R-injected mice, a significant increase in proliferation of PREs compared to normalappearing liver tissue could only be found for RASSF1A KO. In PIK3CA H1047R-injected mice, tumors had a significantly higher proliferation than PREs. Moreover, for these mice, a significant difference could be found by comparing RASSF1A WT and KO in PREs, with the restriction that the numbers of animals analyzed for these subgroups are small.
Finally, the two tumors spontaneously arising in RASSF1A KO mice in the combined control group were histologically examined. These tumors were extremely well-differentiated, pure HCC with a low proliferation rate demonstrated by Ki67 immunohistochemistry. Interestingly, one of these tumors featured a lipid-rich component with focally increased proliferation. This component displayed a hinted upregulation of stearoyl-CoA desaturase-1 (SCD1). In contrast, phosphorylated/activated phosphorylated extracellular signal-regulated kinase (pERK) and phosphorylated RAC-alpha serine/threonine-protein kinase (pAKT) did not show an increased immunoreactivity (Fig. 2E).
Altogether, the low-grade histology following a long latency confirms the limited oncogenic potential of RASSF1A inactivation alone. Proliferation showed a statistically significant tendency to be lower in PIK3CA E545K mutant tumors in RASSF1A KO mice than RASSF1A WT mice (which was not detectable in PIK3CA H1047R-injected mice). Thus, irrespective of RASSF1A, we could define the multistep nature of PIK3CA mutant forms induced hepatocarcinogenesis.

PIK3CA-induced tumors display a strong upregulation of canonical effectors
Due to the absence of overt differences in tumorigenesis between RASSF1A WT and KO mice, we focused on RASSF1A WT mice. To confirm the hepatocellular nature and the activity of PIK3CA effectors in the PREs and tumors developed in PIK3CA mutant mice, we performed immunohistochemical analyses. Immunohistochemistry demonstrated that pAKT, the primary downstream target of PIK3CA, was strongly upregulated in preneoplastic clusters and tumors. Similarly, pERK was markedly induced, implying the activation of the ERK-MAPK signaling pathway. Moreover, the master regulators of lipogenesis, SCD1 and fatty acid synthase (FASN), were strongly positive in E545K-and H1047R-injected cohorts. Intense positivity for carbamoyl phosphate synthetase I (CPS1) protein, a highly specific hepatocyte marker, confirmed the hepatocellular nature of the neoplastic lesions. Compellingly, foci with cholangiocellular differentiation developed in PIK3CA E545K and H1047R mouse livers, as indicated by cytokeratin 7 (CK7) immunoreactivity. An overlap of CK7 with CPS1 was detected, which agrees with the hypothesis that the cholangiocellular components originate from transdifferentiation rather than from separate cholangiocarcinogenesis (Fig. 3A,B) [52]. At the protein level, we verified the gradual upregulation of PIK3CA, pAKT, p-serum/ glucocorticoid-regulated kinase family member 3 (SGK3), pERK1/2, p-signal transducer and activator of transcription 3 (STAT3), FASN, and SCD1 in PREs and tumors in PIK3CA E545K-injected mice. The activation of the aforementioned phosphorylated proteins was also evident in terms of an increased ratio over the total amount of the respective proteins.
Notably, prostaglandin-endoperoxide synthase 2 (COX2), known to induce tumor-promoting inflammation through increased prostaglandin E2 synthesis [53], was also upregulated. Recently, eicosanoid metabolism has been increasingly recognized as an essential mechanism of PIK3CA-mediated oncogenicity (Fig. 3C-E) [54]. When looking into COX2 expression on an immunohistochemical level, enhanced immunoreactivity could be detected preferentially in tumor vasculature, while an upregulation in neoplastic cells was only hinted (Fig. S2). Interestingly, overexpression of COX2, which produces prostaglandin E2, an important angiogenic factor, in tumor-associated blood vessels, has been previously described in human HCC [55].
Next, we carried out transmission electron microscopy to evaluate the tumors on an ultrastructural level. In concordance with the histologically visible empty vacuoles, tumor cells showed abundant intracytoplasmic microvesicular lipid droplets. These differed markedly from an example of spontaneous liver steatosis in a control mouse, where only a few scattered, larger intracytoplasmic lipid droplets could be found. The extensive cytoplasmic accumulation of lipid vesicles underlines the profound reliance of these tumors on lipogenesis (Fig. S3A). Subsequent Sudan IV histochemical staining on fresh frozen tissue confirmed the extensive presence of intracytoplasmic triglycerides in the tumors (Fig. S3B).
To assess the potential involvement of fibrosis and stellate cells on hepatocarcinogenesis in our model, we performed Picro Sirius Red histochemical stainings on untreated mice and mice harboring PREs or tumors. These stainings did not show an increase in fibrosis along the process of hepatocarcinogenesis. Moreover, to test for a potential activation of stellate cells, desmin and alpha-smooth actin immunohistochemistry was carried out, which did not demonstrate an increase in the number or activation status of stellate cells in the neoplastic lesions (Fig. S4A). In concordance with these results, electron microscopy did not show increased or activated stellate cells when comparing tumors and normal liver tissue (Fig. S4B).   Overall, we observed a marked upregulation of PIK3CA canonical downstream effectors. Massive lipogenesis, paralleled by robust upregulation of FASN and SCD1 lipogenic enzymes, was one of the earliest events in PIK3CA-mediated hepatocarcinogenesis.

Identification of additional putative targets in PIK3CA-driven hepatocarcinogenesis
Although lipogenesis is an established mediator of hepatocarcinogenesis, it remains challenging to target it selectively without significant systemic adverse effects due to its ubiquitous involvement in metabolism [56]. Therefore, the search needs to be widened to identify novel targets that can be pharmacologically manipulated. Thus, we performed gene expression microarray analyses of PIK3CA H1047R-and E545Kinduced PREs and tumors (Fig. 4A). Lgals1, which encodes Gal-1, was among the few genes concomitantly upregulated in PIK3CA-induced PREs and tumors compared with age-matched livers from EVinjected mice. We reached this conclusion based on the depicted Venn diagram [57] (Fig. 4B). This finding was intriguing for several reasons. Gal-1, being a 14 kDa beta-galactose-specific binding protein [58], is known to be increased in numerous neoplasms, including primary hepatic tumors [59]. Moreover, Gal-1 is an adverse prognostic marker in HCC [60,61]. And finally, Gal-1 has been linked to the PI3K-AKT-mTOR signaling, thereby enhancing in HCC cell lines [62]. Given that Gal-1 expression was already elevated in preneoplastic tissue and retained during tumor progression, it was selected for further investigation.
In accordance with immunohistochemical results, induction of the PI3K-AKT-mTOR pathway was detected by the Hallmark Gene Set Enrichment Analyses (GSEA) [40]. The latter revealed an upregulation in PREs and/or tumors of pathways such as inflammatory response, interferon-gamma response, angiogenesis, cholesterol homeostasis, E2F targets, EMT, KRAS signaling, and Myc targets. Surprisingly, fatty acid metabolism was exclusively upregulated in PREs, suggesting a transition in the direction of increased independence from lipid metabolism in the later stages of carcinogenesis. In addition, several genes in the Hallmark PI3K-AKT-mTOR gene set (upregulation of Tnfrsf1a, Trib3, and Rps6ka1; downregulation of Gna14, and Ecsit) and the Hallmark Fatty Acid Metabolism gene set (upregulation of Cbr3, Cbr1, Ube2l6; downregulation of Rdh16, Retsat, Glul, Aadat, Idi1, Ehhadh, Decr1, Hmgcs2, Cpt2, and Mgll) met the following criteria |log 2 FC| > 1, adj. P ≤ 0.05 and counts > 3 as determined in the E545K 5 months of comparison group (Fig. 4C,D).
Further analysis of the microarray data for targetable proteins with a cutoff of Log2 fold change and adj P < 0.05 in the PIK3CA E545K-injected group revealed Lgals-3 and ZIP4 as additional compelling targets (Fig. 4E). Lgals-3 encodes galectin-3 (Gal-3), another member of the lectin family, implicated in the development of liver cirrhosis [63], immunosuppression in cancer [64], and with a negative prognostic value in HCC [65]. Zrt-Irt-like protein 4 (ZIP4) is a cellular zinc transporter [66,67], upregulated in HCC, where it represses apoptosis and increases migration [68]. Also, ZIP4 is correlated with poor survival in HCC patients [69].

Galectin-1, ZIP4, and galectin-3 are highly expressed in PIK3CA-induced lesions
To validate the gene expression microarray analysis results, the expression of Gal-1, ZIP4, and Gal-3 was Overview of samples and comparison groups for cDNA microarray analysis. Preneoplastic lesions were compared to age-matched controls injected with the EV. EV-injected 3-month time points were contrasted to EV-injected 6 months to rule out age-dependent effects. Only one tumor at 12 months was available from the H1047R injection arm. The asterisk indicates a PIK3CA H1047R-induced confluent preneoplastic lesion analyzed at 6 months. (B) Venn diagram of cDNA microarrays exhibiting overlap of regulated genes in the prespecified comparison groups. Selection criteria were Log 1.5 change and P < 0.01 for H1047R 3 months, E545K 3 months and control 6 versus 3 months, Log 2 fold change and adj P < 0.05 for E545K 5 months. The numbers of overlapping and uniquely altered genes are given. The three genes conjointly called in the PIK3CA injection groups are mentioned, while Lgals1 encoding for Gal-1 is written in bold. (C) Heatmap of differentially expressed genes with selection criteria |log 2 FC| > 1, adj. P < 0.05 and counts > 3 as determined in the E545K 5 months of comparison group. Specified genes are comprised in the GSEA hallmark gene sets PI3K-AKT-MTOR signaling (left panel) and fatty acid metabolism (right panel). Gradient scale color codes for Log 2 fold change. A cluster dendrogram is shown on the left and above. (D) GSEA was performed to identify significantly upregulated (red) or downregulated (green) biological processes in PIK3CA E545K 3 months, H1047R 3 months, and E545K 5 months of injection groups. Particularly relevant processes implicated with hepatocarcinogenesis are reported in bold. (E) Target genes glypican-3 (Gpc3), Lgals-3 (Gal-3), and ZIP4 fulfilled the selection criteria prementioned above in (B) and (C).
LGALS1 (Gal-1) met the selection criteria described in (B). Gpc3 serves as an internal positive control as upregulation is expected in HCC. Target genes were manually chosen based on novelty, potential pharmacological targetability (Gal-1 and Gal-3), and early upregulation in preneoplastic lesions (Gal-1). Presentation as a heatmap with identical color-coding as in Panel (C). investigated by western blot analysis. All proteins were strongly upregulated in PIK3CA E545K-induced tumors, and it was statistically significant for Gal-1. Moreover, a tendency of upregulation occurred in PREs (Fig. 5A). To corroborate these findings, we performed immunohistochemistry. Gal-1 appeared robustly increased in tumors and PREs from PIK3CA mutant-injected mice (Fig. 5B,C).
To investigate a potential link with the PIK3CA-AKT-mTOR pathway, we additionally examined different mouse models that were induced by the injection of myristoylated AKT either alone [70] or in combination with other genes, such as b-catenin [71], F-box and WD repeat domain-containing 7 (FBXW7) [72], and YAP [73]. Notably, the latter two combinations induce cholangiocarcinogenesis [73]. In all mouse models, tumors exhibited strong immunoreactivity for Gal-1 compared to the surrounding liver tissue.
Remarkably, Gal-1 positivity was also evident in isolated lipid-rich preneoplastic hepatocytes (Fig. 5D). Furthermore, we conducted immunohistochemical analyses of ZIP4 protein expression. Pronounced membranous immunolabeling for ZIP4 was detected in tumors and PREs of PIK3CA E545K-injected mice. Interestingly, there was also a corresponding upregulation of the Hippo pathway effector YAP and active YAP (Fig. 5E).

The galectin inhibitor OTX008 synergizes with PI3K inhibitors in vitro
In light of Gal-1 overexpression in PIK3CA-AKTdependent tumors, we assessed the importance of Gal-1 on HCC cell growth in vitro. Thus, we treated the PLC/PRF/5 and HLE HCC cell lines with the PI3K inhibitor alpelisib, either alone or associated with the Gal-1 inhibitor OTX008. Alpelisib has proven its clinical actionability in the large phase 3 trial SOLAR-1, which resulted in the drug's approval by the Food and Drug Administration for hormone receptor-positive, human epidermal growth factor receptor 2-negative breast cancer [74]. In contrast, OTX008, a calixarene derivative designed to bind the Gal-1 b-sheet conformation [75], has only undergone early clinical testing in the form of a dose-finding phase 1 study in patients with advanced solid tumors [76].
Combining the two drugs led to significantly (by evaluation of 95% confidence intervals) [77] higher antigrowth properties than either of the two drugs alone. This effect was evident for combined concentrations of 20 and 30 µM in PLC/PRF/5 and 20 and 50 µM in HLE cell lines. Notably, OTX008 only displayed relevant antigrowth effects as a single drug in PLC/PRF/5 at the examined concentrations up to 50 µM (Fig. 6A). To confirm the synergistic effect, the combination index was calculated based on the median-effect principle of the mass-action law [78]. A combination index < 1 signifying synergism was attained for most combined concentrations in both PLC/PRF/5 and HLE cells (Fig. 6B).
To determine whether alpelisib treatment influences Gal-1 expression, we treated the two cell lines with 5 and 20 µM, respectively. Western blot analyses indicated an upregulation of Gal-1 in response to alpelisib treatment after 48 h. However, this trend was only statistically significant (P = 0.0141) in the quantification at an alpelisib concentration of 5 µM in the HLE cell line (Fig. 6C). Although this finding contrasted our expectations, it might still explain the synergistic effects observed for combination treatments with OTX008, given that previous studies inferred a correlation between the antiproliferative effect of OTX008 and Gal-1 expression [79].
To study the observed synergy in a mutation-specific background, the SNU387 and SNU449 HCC cell lines were lentivirally transduced with PIK3CA E545K, PIK3CA H1047R, and EGFP plasmids. The yielded stably transfected cell lines were functionally characterized using western blots. Increased levels of PIK3CA, phosphorylated AKT at Ser473 and Thr308 residues, and phosphorylated ERK1/2 (Thr202/Tyr204) confirmed the activation of the PI3K pathway (Fig. S5). The IC50 values of the stably transfected HCC cell lines were calculated using the crystal violet cytotoxicity assay after applying incremental concentrations of either alpelisib or OTX008. Transfection with the two PIK3CA mutant   forms markedly decreased alpelisib IC50 values compared to EGFP-transfected cell lines, while a reduction of OTX008 IC50 values based on the transfected constructs was not apparent (Fig. S6). Finally, to evaluate the combined cytotoxicity of alpelisib and OTX008 based on the PIK3CA transfection status, incremental concentrations of the two drugs were applied either alone or in combination, and combination indices were calculated based on the Chou-Talalay method. A trend toward a higher synergy (i.e., CI < 1) could be observed in all of the PIK3CA stably transfected cell lines, with the most pronounced effects for SNU387. Notably, EGFP-transfected cells even displayed an antagonism when combining the two drugs, while in PIK3CA E545K transfected cells, a high synergy could be observed for all tested concentrations (Fig. S7).
To evaluate the generalizability of the observed synergy with OTX008 to other PI3K inhibitors, treatments of PLC/PRF/5 cells with 1 µM taselisib, 1 µM buparlisib, or 1 µM alpelisib with or without the addition of 20 µM OTX008 were conducted for 72 h. The mean excess over Bliss score was 0.1685 for alpelisib, 0.1164 for buparlisib, and 0.1548 for taselisib in three internal replicates, indicating a similarly strong synergistic effect for all three of these compounds (Fig. S8).
Moreover, as a discovery approach for more potential interaction partners of OTX008 (at a concentration of 20 µM), we conducted a compound library screening including 315 approved drugs (at a concentration of 1 µM). All four PI3K inhibitors included in the screening demonstrated a synergistic effect in combination with OTX008. The excess over Bliss score was 0.0931 for Duvelisib, 0.0924 for Copanlisib, 0.0723 for Idelalisib, and 0.0189 for buparlisib (Table S2). Using the threshold of viability from combination treatment < 0.6 and Z score of EOB > 1.5, several more putative synergistic interaction partners of OTX008 could be identified ( Fig. S9 and Table S1). Pacritinib, a Janus-Kinase 2inhibitor, emerged as the compound with the highest synergistic effect. On top of that, several other tyrosine kinase inhibitors, such as Axitinib, Apatinib, Crizotinib, Sorafenib, Crenolanib, and Nintedanib, and the two EGFR inhibitors Neratinib and Afatinib also reached the defined threshold. Thus, in advanced clinical trials, several approved drugs could be considered worthwhile candidates for combination treatments with OTX008 and warrant further investigations.
3.6. Differential effectors of PIK3CA E545K and PIK3CA H1047R In an attempt to shed light on the observed difference in oncogenicity and tumor latency between PIK3CA E545K and H1047R, we performed comparative gene expression microarray analyses (using the Applied Biosystems TM GeneChip TM Clariom S human array) of the stably transfected SNU387 and SNU449 HCC cell lines. Among the significantly upregulated genes in comparing the two mutant forms against EGFP and WT, PIK3CA could be detected in both cases, serving as an internal positive control. Moreover, vascular cell adhesion molecule 1, encoding for a protein connected to angiogenesis and metastasis [80], was strongly downregulated on a single gene level in H1047R compared with WT/EGFP and E545K. An overview of all significantly regulated genes is provided in Figs S10 and S11. To search for potentially different effector pathways, we performed GSEA analyses based on the Panther pathway gene set [81] (Fig. S12). Commonly upregulated pathways for the two PIK3CA mutant forms included the following: apoptosis signaling pathway, hypoxia response via HIF activation, and insulin/ IGF pathway-protein kinase B signaling cascade. Differentially activated pathways comprised the PI3 kinase pathway (which by this method was significantly upregulated only in H1047R cells), the interleukin signaling pathway, chemokine and cytokine signaling, and the plasminogen activating cascade, which were upregulated in H1047R cells. In contrast, in E545K cells, the Toll receptor signaling pathway, CCKR signaling, angiogenesis, and VEGF cascade were induced. The only significant hit (false discovery rate > 0.05) for the direct comparison of the two PIK3CA mutant forms was the plasminogen activating cascade, with increased activation in H1047R cells.
To summarize, PIK3CA H1047R preferentially induced inflammatory pathways (interleukin and chemokine signaling), while PIK3CA E545K instead promoted angiogenic pathways (angiogenesis and VEGF signaling). Albeit no distant metastases were detected in our mouse model, microarray analyses in human HCC cell lines hint at a 'neovascular/metastatic phenotype' for E545K and an inflammatory phenotype for H1047R.

An interdependence with SCD1 links galectin-1 to lipogenesis
In the search for a potential mechanism of cytotoxicity, we assessed whether Gal-1 is associated with lipogenesis in HCC. A functional connection between Gal-1 and the expression of lipogenic factors has recently been elucidated in livers of Gal-1 KO mice, where FASN, acetyl-CoA carboxylase 1 (ACAC), and SCD1 were downregulated in animals fed a high-fat diet [82]. For this purpose, we performed western blot analysis of SCD1 expression following Gal-1 siRNA-mediated silencing (Fig. 7A). As expected, Gal-1 levels were downregulated in all tested HCC cell lines (Pvalues < 0.05). Furthermore, decreased SCD1 levels in PLC/PRF/5, HLF, and Snu182 HCC cell lines subjected to Gal-1 knockdown were observed. Quantitative analysis of band intensities supported this observation in PLC/PRF/5 and HLF and reached statistical significance (P-values < 0.05), which was, however, not met in Snu182 (P = 0.34). In addition, we tested the effect of Gal-1 downregulation on other lipogenic enzymes, including peroxisome proliferatoractivated receptor gamma (PPARc), ATP citrate synthase (ACLY), FASN, and ACAC. Among these, only PPARc showed a statistically significant downregulation (P = 0.0173) upon Gal-1 silencing in the PLC cell line (Fig. 7B).
As a further validation for the observed effect on SCD1 protein levels, OTX008 treatments were performed to inhibit Gal-1 functionally. Similar to the Gal-1 silencing data, a downregulation of SCD1 protein levels was evident and statistically significant for PLC/PRF/5 (P < 0.0001) and HLF (P = 0.0315) cell lines. In contrast, no significant effect could be observed for the Snu182 cell line (Fig. 7C).
Taken together, we could deduce a dependence of SCD1, one of the master regulators of lipogenesis, on Gal-1 expression in HCC.

Discussion
Both the PI3K-AKT-mTOR and the Ras-MAPK pathways have been commonly implicated in hepatocarcinogenesis [83,84]. This finding raised the question of possible interdependence and cooperativity between these signaling networks. RASSF1A as an inhibitor of Ras activity with a high methylation frequency in HCC [13] represented a promising candidate as an oncogenic signaling node, mediating the two pathways' effects. We addressed the question of RASSF1A oncogenicity using a murine RASSF1A KO model with hydrodynamic tail vein injection of PIK3CA H1047R and E545K mutant forms. Surprisingly, we found that PIK3CA mutant forms elicited hepatocarcinogenesis irrespective of RASSF1A status, precluding hypothesized cooperativity.
Another unanticipated conclusion from these experiments was that PIK3CA mutant forms H1047R and E545K alone could induce hepatocarcinogenesis. Previous experiments suggested that combination partners such as activated NRAS or c-Met are necessary for PIK3CA to drive hepatocarcinogenesis [35]. Notably, PIK3CA E545K exhibited a much stronger oncogenic potential, giving rise to HCCs in most mice as early as 3 months after injection. PIK3CA H1047R, in contrast, only produced sporadic tumors after 12 months, which is opposite to findings in breast cancer [85]. The generation of this novel purely PIK3CA-induced mouse model opens the door to further investigations into the PI3K-AKT-mTOR pathway in hepatocarcinogenesis. The availability of such a model is of high interest. First, because 4-6% of human HCCs harbor a PIK3CA mutation [7,33], which could be amenable to targeted therapies. Second, the PI3K-AKT-mTOR pathway activation often occurs in HCCs regardless of the PIK3CA mutational status [32]. Importantly, we could establish PIK3CA-mediated hepatocarcinogenesis as a multistep process including PREs, enabling the study of early predominant mechanisms. This finding paralleled observations that have been made in mice hydrodynamically injected with activated AKT [70]. When we examined the tumors induced in this model, we found an increase in the canonical effectors of PIK3CA, such as pAKT, pERK, p-SGK3, and p-STAT3.
In addition to these canonical targets, we revealed a pronounced increase in lipogenesis in tumors and even PREs, which was evident on several experimental layers: an increase in the level of the master regulators of lipogenesis SCD1 and FASN, an upregulation of lipid metabolism by gene expression analysis, and a lipidrich phenotype to an extent, where tumor cells were packed with lipid droplets as seen by electron microscopy. De novo lipogenesis is a cancer hallmark. The synthesis of fatty acids provides tumor cells with metabolic intermediates essential for synthesizing the membrane lipid bilayer, energy storage in the form of triglycerides, and signaling molecules [56]. In a related murine HCC model, it has antecedently been demonstrated that the deletion of FASN abolishes AKTinduced carcinogenesis [86]. Therefore, inhibitors of FASN and other lipogenic enzymes can represent enticing therapeutic options, was it not for doselimiting toxicity, which can arise in adipose and liver tissue [56]. Consequently, it is an important endeavor to widen the search for potential therapeutic targets that can influence lipogenesis.
Using gene expression analyses, we detected several candidates, which could serve as potential targets. Among them, we identified Gal-1 as a protein already strongly upregulated in PREs, underlining its putative importance in hepatocarcinogenesis. A captivating finding was the interrelation of Gal-1 expression and lipogenesis by regulating SCD1 protein levels. The incentive to analyze a potential link between Gal-1 and lipogenesis came from a recent report showing the downregulation of FASN, ACC1, and SCD1 expression in adipose and hepatic tissues from Gal-1 KO mice [82]. This regulation could well play a critical role in hepatocarcinogenesis and warrants further experimental elucidation.
Intriguingly, Gal-1 is accessible to pharmacological targeting, such as the calixarene derivative OTX008, already tested in an early clinical trial [76]. In this study, the drug was administered to HCC cell lines. While there were limited effects when OTX008 was applied alone, a strong synergism could be detected in combination with the PIK3CA inhibitor alpelisib. Indeed, previous reports directly put Gal-1 in the context of PI3K-signaling [87]. In our hands, alpelisib treatment increased Gal-1 expression, which might sensitize cells to OTX008 treatment. In the phase 1 trial mentioned before, plasma concentrations '> 1 µM over 12 h after administration' [76] were achieved. Given that the concentrations of OTX008 were comparatively high in our experiments (~20 µM), we must exert at least some caution in interpreting these findings. Of note, synergistic effects of OTX008 with either Sorafenib or Everolimus have been reported in HCC cell lines before [79], which further substantiates the interaction of Gal-1 with the PI3K-AKT-mTOR pathway. In addition, multidrug pharmacologic screening revealed several more candidate drugs acting synergistically with OTX008, including JAK, EGFR and other tyrosine kinase inhibitors, highlighting its potential in combination therapies.
Finally, two additional upregulated targets identified in this study with potential therapeutic implications were Gal-3 and ZIP4. Besides being repeatedly implicated in inflammation [88], Gal-3 is also linked to lipid metabolism since FASN, proliferator-activator receptor gamma, and fatty acid-binding protein 4 were significantly reduced in Gal-3 KO mice [89]. ZIP4 emerged as a highly interesting target protein since we found a correlating upregulation of YAP and active YAP. ZIP4 and YAP have recently been demonstrated to be commonly responsible for EMT control in pancreatic cancer [90]. Further research is required to verify the context of these proteins in PIK3CA-mediated carcinogenesis and determine the effect of ZIP4 on EMT in HCC.
Overall, the present study has determined the independence of RASSF1A KO-mediated and PIK3CAdriven hepatocarcinogenesis. We have established both PIK3CA stably transfected cell lines that reveal the targetability of oncogenic PIK3CA mutations and an HCC mouse model of high translational value concerning PI3K-targeted therapies. Furthermore, this model allowed us to confirm the paradigm of PIK3CA-driven carcinogenesis (in terms of canonical effectors of the pathway) and enhance our understanding of PIK3CA oncogenic properties. We ascertained Fig. 8. Graphical depiction of proposed PIK3CA-induced hepatocarcinogenesis. H1047R and E545K mutations result in a receptorindependent constitutive activation of PIK3CA. In complex with the regulatory subunit p85, the conversion of phosphatidylinositol (4,5)bisphosphate (PIP 2 ) into PIP 3 is catalyzed. PIP 3 activates the downstream effectors AKT, ERK1/2, phosphorylated SGK3, and STAT3, either directly or via the mediation of phosphoinositide-dependent protein kinase 1. AKT has a pivotal role in the pathway by inducing antiapoptosis and cell proliferation. Cellular metabolism is veered toward lipid biosynthesis by increased expression of the master lipid regulators SCD1 and FASN. Gal-1 upregulation acts in unison by eliciting the upregulation of SCD1 and could account for an accessible therapeutic target. Furthermore, the elevation of ZIP4 encompasses YAP activation, thereby potentially providing a bridge to the Hippo signaling pathway. Finally, in correlation with the macroscopic steatohepatitic phenotype, pro-inflammatory proteins such as COX2 and Gal-3 are stimulated. Cooperatively, these alterations induce the formation of HCC. the pivotal role of lipogenesis and discovered novel putative effectors, including Gal-1, with therapeutic actionability (Fig. 8). These findings pave the way for ensuing investigations, such as searching for further combination partners of Gal-1 and the study of PIK3CA-dependent oncogenicity in Gal-1 KO mice.

Conclusion
PIK3CA E545K and H1047R mutant forms confer oncogenic potential irrespective of RASSF1A in vivo. Induced HCCs showed a concomitant upregulation of lipogenesis and Gal-1 expression. The regulation of SCD1 by Gal-1 and the synergism of alpelisib and OTX008 suggest Gal-1 as a potential therapeutic target in HCC.

Supporting information
Additional supporting information may be found online in the Supporting Information section at the end of the article.  Fig. S2. Upregulation of COX2 protein levels in PIK3CA-induced neoplastic lesions shows a vascular distribution. Fig. S3. Morphological and histochemical evidence of increased lipids in PIK3CA-induced neoplastic lesions. Fig. S4. Stellate cells in PIK3CA mutant form injection-induced lesions. Fig. S5. Upregulation of PIK3CA canonical effectors in PIK3CA mutant form stably transfected HCC cell lines. Fig. S6. Enhanced sensitivity of PIK3CA mutant form stably transfected HCC cell lines to Alpelisib. Fig. S7. Combination indices of Alpelisib and OTX008 treatments in PIK3CA mutant form stably transfected HCC cell lines. Fig. S8. PLC/PRF/5 cells treated with different PI3K inhibitors and OTX008 in mono-or combination therapy for 72 h. Fig. S9. Drug screening of 315 approved anti-cancer compounds with and without OTX008. Fig. S10. Microarray analysis of PIK3CA E545K and H1047R effectors in stably transfected HCC cell lines SNU387 and SNU449. Fig. S11. Differential effectors of PIK3CA E545K and H1047R in stably transfected HCC cell lines SNU387 and SNU449. Fig. S12. Gene Set Enrichment Analysis of PIK3CA E545K and H1047R in stably transfected HCC cell lines SNU387 and SNU449. Table S1. Overview of drug targets with ≥ 1 hit. Table S2. Combinatorial drug screening results for PLC/PRF/5 cells.