SUPPLEMENTARY INFORMATION 1. SUPPLEMENTARY FIGURE LEGENDS Supplementary Figure 1. Long-term exposure to sorafenib increases the expression of progenitor cell-like features. A) mRNA expression levels of PROM-1 (CD133), THY-1 (CD90), EpCAM, KRT19, and VIM assessed by quantitative real-time PCR. Data represent the mean expression value for a gene in each phenotypic type of cells, displayed as fold-changes normalized to 1 (expression value of its corresponding parental non-treated cell line). Expression level is relative to the GAPDH gene. Bars indicate standard deviation. Significant statistical differences are set at p<0.05. B) Immunocitochemical staining of CD90 and vimentin in Hep3B sorafenib resistant cell line and its parental cell line. C) Western blot analysis comparing protein levels in resistant Hu6 and Hep3B cells vs their corresponding parental cells lines. Supplementary Figure 2. Efficacy of gene silencing of IGF1R and FGFR1 and evaluation of MAPK14 signaling activation. IGF1R and FGFR1 knockdown expression 48h after transient transfection with siRNAs (50 nM), in non-treated parental cells and sorafenib-acquired resistant tumor derived cells was assessed by quantitative RT-PCR (A) and western blot (B). C) Activation status of MAPK14 signaling was evaluated by western blot analysis in vivo, in tumors with acquired resistance to sorafenib in comparison to non-treated tumors (right panel), as well as in in vitro, in sorafenib resistant cell lines vs parental non-treated. Supplementary Figure 3. Gene expression levels of several pro-angiogenic factors. mRNA expression levels of FGF1, FGF2, VEGFA, IL8, ANGPT2, KDR, FGFR3, FGFR4 assessed by quantitative real-time PCR in tumors harvested from mice. Plotted data represent the mean expression value for a gene in each experimental group, displayed as fold-changes normalized to 1 (mean expression value in non- treated animals). Expression level is relative to the GAPDH. Bars indicate standard deviation. Significant statistical differences are set at p<0.05. Supplementary Figure 4. Increased expression and activation of FGFR1 and IGF1R in acquired resistance tumors. A) Immunohistochemical analysis of FGFR1 and IGF1R protein levels in sorafenib-sensitive and sorafenib-acquired resistant tumors. B) Immunostaining showing activation (phosphorylation) of FGFR1 and IGF1R in viable clusters of cells close to necrotic areas in sorafenib-resistant tumors. 2. SUPPLEMENTARY MATERIAL & METHODS Sphere formation assay To obtain single cell suspension, subcutaneous Huh7-derived tumors were excised and mechanically dissociated by gentle pipetting in Ca/Mg-free PBS on ice, and filtered through a 40 µm cell strainer (BD Falcon). Then, cells were exposed to red cell lysis buffer (Sigma-Aldrich) to remove red blood cells. No enzymatic digestion was used to isolate cells. Cell viability was quantified in a Neubauer cytometer using the trypan blue (1%; Sigma Aldrich) exclusion method. After tumor harvest and generation of single cell suspensions tumor cells where directly cultured into non-adherent conditions to allow the formation of spheroids as previously described(1). Briefly, cells were grown in serum free epithelial growth medium (MEGM, BioWhittaker) supplemented with B27 (Invitrogen), 20 ng/mL EGF and 20 ng/mL bFGF (BD Biosciences), and 4 μg/mL insulin (Sigma). Live cells were then plated in 96-well ultra-low attachment plates (Corning) at limiting dilution (10, 102, 103 and 104 cells) and sphere formation was assessed microscopically after 10 days of growth. Cell culture, reagents and drugs Parental Huh7, Hep3B and H6 cell lines were cultured in DMEN with 10% fetal bovine serum (Gibco, Grand Island, NY). The Hep3B-sorafenib resistant cell line(2) and H6- sorafenib resistant cells line were kindly provided Dr. Desbois-Mouthon (INSERM, Paris) and grown in the presence of sorafenib at a concentration of 4 µM and 3 µM, respectively. Tumor-derived spheres were grown under non-adherent culture conditions in serum free epithelial growth medium (MEGM, BioWhittaker) supplemented with B27 (Invitrogen), 20 ng/mL EGF and 20 ng/mL bFGF (BD Biosciences), and 4 μg/mL insulin (Sigma). Cells from sorafenib-acquired resistant tumors were grown in the presence of 4 µM of sorafenib. Sorafenib was purchased from LC Laboratories (Woburn, MA). Linsitinib (IGF1R inhibitor) and BGJ398 (pan inhibitor of FGFR1, -2, -3) were purchased from BioVision (Milpitas, CA). Brivanib (FGFR inhibitor) was provided by Bristol-Myers Squibb (BMS, New York, NY). In vitro and the in vivo doses of the compounds were chosen after revision of published preclinical studies, as well as, by following the manufacturer’s recommendations(3-5). For in vivo oral administration, sorafenib was dissolved in ethanol 95%/cremophor EL/sterile water (12.5:12.5:75%) at 30 mg/kg/day dosage. Brivanib was dissolved following BMS recommendations and orally dosed at 100 mg/kg/day. For MTT assay dissociated spheres were plated at a density of 5,000 cells/well in 96-well ultra-low attachment. Huh7, Hep3B, Hep3B-sora, Huh6 and Huh6-sora cells were seeded at 2,500-3,000 cells/well density under adherent conditions. Then cell viability was evaluated after 48-72 hours in the presence of sorafenib or/and other inhibitors (0.1-10 M). Each experimental condition was measured in triplicate in 3 independent experiments. siRNA transfection Cells were plated on 96-well plates or 6-well plates and transiently transfected with Silencer® Select siRNAs from Ambion (Life technologies, Inc.) (cat. #4390824 ID: #s7211 for IGF1R siRNA, ID: #s5164 for FGFR1 siRNA, cat. # 4390649 for control GAPDH siRNA, cat. 4390646 negative control #2) at 50 nM using Lipofectamine® 3000 tranfection reagent (Life technologies, Life Technologies) following manufacturer’s intructions. 4 hours post transfection sorafenib at 3 µM was added to the medium and cells were allowed to grown for additional 48 hours. Protein extraction and Western blot analysis Cells were plated into 15 cm culture dishes and grown to subconfluency. Total protein was quantified using Bradford Reagent (Sigma, Saint Louis, MO). To isolate cytoplasmic proteins cells were lysed with RIPA buffer plus protease and phosphatase inhibitors (50 mM Tris pH=7.4, 150 mM NaCl, 1% Trithon X-100, 0.1% SDS, 0.25 mM EDTA, 1% Sodium deoxycholate). A modified lysis buffer was used to better preserve IGF1R and FGFR1 in whole lysates (Hepes pH=7.5 50 mM, NaCl 150 mM, MgCl2 1.5 mM, EDTA 1 mM, 10% Glycerol, 1% Trithon X-100). Both buffers contained a mixture of protease and phosphatase inhibitors; Cocktail Tablets (Roche, Mannheim, Germany), NaF 1 mM, NaPiP 1 mM and Na3VO4 0.2 mM. For western blot analysis equal amounts of total protein were resolved in polyacrylamide gels and transferred to nitrocellulose membranes (Pierce, Rockford, IL). Membranes were BSA-blocked and hybridized overnight at 4ºC with the primary antibodies; IGF1R, phospho-IGF1R, FGFR1, phospho-FGFR1, ERK, phospho-ERK, Akt, phospho-Akt, MAPK14, phospho-MAPK14, EpCAM, vimentin, CK19 (Cell Signaling) and later at room temperature with HRP-conjugated secondary antibodies. To evaluate the activation status of FGFR1, immunoprecipitation with total anti-FGFR1 and western blot analysis using the anti-phosphotyrosine clone PY20 HRP (Santa Cruz) was performed as previously described(5). Signals were visualized with Amersham ECL Prime (GE Healthcare, Buckinghamshire, UK) using a LAS-4000 imaging system (Fujifilm, Tokyo, Japan). Immunohistochemical analysis For immunohistochemistry, 5-m paraffin-embedded sections of mice xenograft tumors were baked at 55 C for 1 hour, deparaffinized in xylene, and rehydrated in a graded series of ethanol solutions. Antigens were unmasked by microwave heating the samples in 10 mM sodium citrate buffer (pH 6.0) for 15 minutes (5 minutes, 3 times), and the reaction was quenched using hydrogen peroxide 3%. After washing with PBS, samples were incubated with the following primary antibody concentrations: p-FGFR1 (Tyr766) 1:50, FGFR1 1:100 (both from Abcam plc, Cambridge, UK), anti-p-RPS6 (phosphoSer240/244) 1:200, anti-pERK (phosphoThr202/Tyr204) 1:100, anti-ERK 1:100 (all from Cell Signaling, Danvers, MA), and anti-p-IGF-1R (phosphoTyr1316)(6) 1:100, at 4ºC overnight. DAB (3,3'-diaminobenzidine) was used as a detection system (EnVision+ System-HRP, Dako). Samples were contrasted with Gill’s hematoxilyn (Panreac, Castellar del Valles, Spain), dehydrated and mounted with DPX (Richmond, IL). Immunoreactivity was graded by a liver pathologist (MS). The variables measured were immunostaining intensity and staining pattern. For immunocytochemistry cells were seeded on poly-L lysine treated coverslips on 24- well plates for 16h, fixed, permeabilized and incubated O/N at 4ºC with primary antibodies for CK19 (Dako), EpCAM, CD133, CD90 (Abcam), and vimentin (Cell Signaling). Gene expression analysis profiling Briefly, 1 g of total RNA was hybridized to the Human Gene 1.0 ST platform. Differential expression among experimental groups was analyzed using Array File Maker 4.0 (AFM) software. Data was normalized by RMA method and analyzed with GenePattern software (www.broadinstitute.org/genepattern). The dataset is deposited at GEO Omnibus (http://www.ncbi.nlm.nih.gov/geo) under accession number GEO (GSE73571). For relative mRNA quantification TaqMan® Gene Expression Assays were used following manufacturer’s instructions (Applied Biosystems, Foster City, CA). GAPDH and ribosomal RNA (18S) were chosen as endogenous reference genes. TaqMan® probes are listed in Supplementary Table 2. For the purpose of the study, we used data from the transcriptomic analysis of 442 HCCs performed by the HCC Genomic Consortium in previous studies (7-11). Computation of sphere- and tumor-initiating cell frequencies To compute the frequency of sphere- and tumor-initiating cells, the maximum-likelihood estimation method was used(4, 12). Similar results were obtained using a second method(13), and internal consistency was validated by chi-squared analysis (data not shown). Bioinformatic and statistical analysis Gene expression profiling among experimental groups was analyzed with the module Comparative Marker Selection as implemented in Gene Pattern, and Array File Maker 4.0 (AFM) tools.