Obtaining public datasets

Expression array datasets GSE21257 (including 53 osteosarcoma biopsy samples) [7], GSE42352 (including 84 pretreatment biopsy samples) [11], and GSE39055 (including 37 osteosarcoma biopsy samples) [8] with accompanying clinical information were obtained from the Gene Expression Omnibus [12] (GEO, http://www.ncbi.nlm.nih.gov/geo/).

Cell lines and cell culture

All cell lines were confirmed to be mycoplasma-free based on the VenorGeM OneStep detection kit (Minerva Biolabs). The cell lines U2OS and SaOS2 were provided by Dr. K. Matsuda, Department of Computational Biology and Medical Sciences, Graduate School of Frontier Sciences, The University of Tokyo, while the HuO3N1 line was provided by the Department of Cell Biology, Okayama University Graduate School of Medicine, Dentistry, and Pharmaceutical Sciences. G292 was provided by Dr. J. Toguchida, Institute for Frontier Medical Sciences, Field of Clinical Application Department of Tissue Regeneration, Kyoto University. IMR-32 was provided by RIKEN Cell Bank. All cell lines were grown in RPMI 1640 medium (Thermo Fisher Scientific) supplemented with heat-inactivated 10% fetal bovine serum (Gibco) and 100 units/mL penicillin‒streptomycin (Gibco) at 37 °C under a humidified 5% CO2 atmosphere.

Small interfering (si)RNA transfection and cell viability assays

All in vitro experiments were conducted in triplicate and repeated at least twice to validate the results to ensure the reproducibility and validity of the results. For siRNA knockdown assays, osteosarcoma cell lines (U2OS, HuO3N1, SaOS2, or G292) were seeded in 96-well plates (Corning) at 3000–4000 cells per well and transfected 24 hours later with two Silencer Select siRNAs for each target gene and Silencer Select Negative control siRNA No. 1 using Lipofectamine RNAiMAX (all Thermo Fisher Scientific) [13]. In the rescue experiment, PDGF-BB (MBL QK044-0050) or EGF (Abcam ab259398) was added to the medium the day after siRNA knockdown. The number of viable cells was estimated at selected times posttransfection using the Cell Counting Kit 8 (CCK-8) according to the manufacturer’s protocol (Dojinbo). In other experiments, osteosarcoma cells were seeded as described, cultured for 24 h, and treated with itraconazole (Cayman I0732), MMP3 inhibitor VIII (Cayman 17246 [14]), and/or furin inhibitor I (Cayman 14965 [15, 16]) as indicated. Cell viability was evaluated using CCK-8 (Dojinbo). Cell cycle analysis was performed using the Cell-Clock Assay Kit (Biocolor).

Preparation of shRNA interference vectors

A specific shRNA targeting human C1GALT1 (shC1GALT1) was designed and subcloned, and inserted into pENTR4-H1tetOx1, CS-RfA-ETV, and CS-RfA-ETBsd vectors (RIKEN BRC) as previously reported [17], while a nontargeting control shRNA was designed against luciferase (shLuc). The target sequences are provided in Supplemental Table S1. Cell lines were transfected as described in the previous sections.

Production and transduction of lentivirus

For the production of lentiviral vectors, HEK293T cells were transiently transfected with the packaging construct (pCAG-HIVgp), the VSG-G and Rev-expressing construct (pCMV-VSV-G-RSV-Rev), and the self-inactivating lentiviral vector construct as previously reported [18]. The viral supernatant was then collected and concentrated using Amicon Ultra15 Centrifugal Filter Units (100 K) (Millipore, C7715). Following measurements of viral titer using the Lenti-X p24 Rapid Titer Kit (Takara Bio, 632200) and Lenti-X GoStix Plus (Takara Bio, 631280), G292 cells were infected and selected by continuous culture in blasticidin S (Cayman, 14499). For the in vitro study of shRNA effects, the cells were incubated with doxycycline for 2 days in the wells of 6-well plates (Corning) or 4 days in the wells of 96-well plates. Knockdown of C1GALT1 was confirmed by Western blotting. The number of viable cells was estimated using CCK-8 (Dojinbo) according to the manufacturer’s instructions.

Xenograft mouse model

To establish a xenograft mouse model of osteosarcoma, male NOG mice aged 6 to 8 weeks (CLEA Japan, Inc.) were injected in the flanks with 1 × 106 G292 cells infected with either shC1GALT1 lentiviral vector (experimental group) or shLuc lentiviral vector (control group). In all experiments, experimental and control groups were formed by selectively injecting individual littermates with shC1GALT1 or shLuc. Tumors were measured with a caliper, and volume was calculated according to the formula (length × width2)/2. Mice were given oral doxycycline through drinking water starting on the day the tumor size reached 100 mm3. Ten mice were assigned to each group. Randomization and blinding were not used. Mice were treated for up to 34 days after tumor cell inoculation and euthanized if the tumor size reached 2000 mm3. Tumors were immediately resected for immunohistochemistry.

Immunohistochemistry

Resected tumors were formalin-fixed and paraffin-embedded using standard techniques and then cut into 5-µm-thick sections. Sections were deparaffinized, rehydrated in gradient ethanol, heated in citrate buffer (pH 6, Genostaff #ARSC6-01) for antigen retrieval, incubated in 0.3% hydrogen peroxide in methanol for 30 min to quench endogenous peroxidase activity, and incubated with G-Block (Genostaff #GB-01) and avidin/biotin blocking kit reagent (Vector #SP-2001). Blocked sections were incubated with primary mouse anti-human C1GALT1 monoclonal antibody (Santa Cruz, sc-100745, dilution 1:100) at 4 °C overnight, washed, incubated with biotin-conjugated goat anti-mouse IgG (Vector #BA9200) for 30 min at RT, and then treated with peroxidase-conjugated streptavidin (Nichirei #426062) for 5 min. Peroxidase activity was visualized by diaminobenzidine. The sections were counterstained with Mayer’s hematoxylin, dehydrated, and then mounted under a cover glass with malinol for examination under light microscopy.

Western blotting

Cellular proteins were extracted from cell cultures and freshly excised tumor tissue using RIPA lysis buffer, separated on 4–12% Mini-PROTEAN TGX Precast Gels (Bio-Rad), and transferred onto nitrocellulose membranes (Millipore). The membranes were then incubated with antibodies against α-tubulin (Abcam, ab7291), C1GALT1 (Santa Cruz, sc-100745), ERK (CST, 4695), phospho-Erk1/2 (CST, 4370), AKT (CST, 9272), phospho-AKT (CST, 9271), PDGFR (CST, 3169), phospho-PDGFR Tyr740 (CST, 3168), EGFR (CST, 4267), and phospho-EGFR Thr992 (CST, 2235). Membranes were washed, incubated with secondary antibodies (Cytiva, NA931 and NA934) for 1 hour, and treated with chemiluminescence reagent to visualize target protein bands.

Comprehensive protein quantification analysis using protein arrays

The osteosarcoma cell line U2OS was seeded on 6-well plates (Corning) at 1.8–2.25 × 105 per well, cultured for 24, and then transfected with siRNA as described above. Three wells were used for each condition. After 48 hours, cellular proteins were extracted using RIPA lysis buffer and used as input (200–500 μg) for the Proteome Profiler Human Phospho-Kinase Array or Human RTK phosphorylation array (RayBiotech). The chemiluminescent signal was detected using the ImageQuant LAS 4000 mini-imager (GE Healthcare). Signal blots were quantified and standardized with ImageQuant TL version 8.1 (GE Healthcare).

Glycosylation profiling

Human glycosylation antibody arrays 493 and 507 (GAH-GCM-493 and GAH-GCM-507, RayBiotech) were used for glycosylation profiling of U2OS cells with or without prior C1GALT1 knockdown. Briefly, U2OS cells were collected 48 hours after siRNA transfection (as detailed in the previous sections) and lysed for protein extraction. Proteins were immunolabeled on glass array slides according to the manufacturer’s protocol, and slides were scanned using GenePix4100A (Molecular Devices). After subtracting background signals and normalization to positive controls, signal intensities were compared between control and C1GALT1 knockdown conditions. A ≥ 1.5-fold increase or ≤0.65-fold decrease in signal intensity was considered a significant difference in glycosylated protein expression provided that both signals were well above the background (mean background +2 standard deviations).

Gene expression analyses of osteosarcoma cell lines

To examine the effects of C1GALT1 knockdown on the gene expression profile, U2OS cells were transfected with siRNAs as described and collected 48 hours later for RNA extraction. To evaluate the effect of itraconazole treatment on gene expression, U2OS cells were treated with 2.5 μM itraconazole or vehicle (DMSO) for 48 h starting 24 h after seeding. After the indicated treatment, RNA was extracted using NucleoSpin RNA (MACHEREY-NAGEL), and libraries for RNA sequencing were prepared using the NEBNext Ultra RNA Library Prep kit from Illumina (New England Biolabs). Next-generation sequencing was performed using the Illumina HiSeq 2000 or 2500 platform with a standard 100-bp paired-end read protocol according to the manufacturer’s instructions. Reads were aligned, quality checked, and counted using our Genomon pipeline (http://genomon.hgc.jp/exome/en/index.html). Read counts were normalized by variance-stabilizing transformation using the R package DESeq2 application [19]. Differential expression was analyzed by the Wald test using negative binomial generalized linear model fitting. Gene set enrichment analysis was conducted using GSEA software version 4 [20].

Statistical analyses

All statistical analyses were performed using R v3.5.3 software [21]. Survival times were estimated using the Kaplan–Meier method, and group values were compared using the log-rank test. Differentially expressed genes (DEGs) identified using the Wald test were used to form expression matrices. Matrices were evaluated by unsupervised consensus clustering using the ConsensusClusterPlus package (RRID:SCR_016954) to identify stable clusters [10]. Independent samples Student’s t tests were used to compare functional assay results for proliferation and tumor size. Beta regression was used for comparisons of percentages that sum to 100%. A P < 0.05 (two-tailed) was considered statistically significant for all tests. The center values of continuous variables are expressed as the median. Error bars in graphical representations are calculated and displayed as the standard error of the mean.