Chronophin regulates active vitamin B6 levels and transcriptomic features of glioblastoma cell lines cultured under non-adherent, serum-free conditions

We used five adherent glioblastoma cell lines cultured in serum: T98G, TP365, U87, U251, A172. T98G (#CRL-1690), U87 (#HTB-14) and A172 (#CRL-1620) were from ATCC (Manassas, VA, USA). TP365 and U251 were kindly provided by Prof. V. P. Collins, Cambridge, UK. Identity of all cell lines was confirmed by short tandem repeat analysis. The cells were cultured in complete medium consisting of high glucose (4.5 g/L) Dulbecco’s modified Eagle’s medium with stable glutamine (PAN Biotech, Aidenbach, Germany) supplemented with 100 U/ml penicillin, 100 μg/ml streptomycin (GIBCO) and 10% fetal bovine serum (PAN Biotech). Five previously characterized [12] human primary stem-like glioblastoma cell lines (NCH1425, NCH421k, NCH465, NCH601, NCH644, all supplied by Prof. Christel Herold-Mende, Heidelberg, Germany) were cultured in DMEM/Ham’s F-12 medium with stable glutamine supplemented with 100 U/ml penicillin, 100 μg/ml streptomycin, 20% BIT admixture supplement (Pelo Biotech, Planegg, Germany), 20 ng/ml epidermal growth factor (EGF) (ReliaTech, Wolfenbuettel, Germany) and 20 ng/ml basic fibroblast growth factor 2 (bFGF2) (ReliaTech). Cells were handled under aseptic conditions and grown at 37 °C and 5% CO₂. All cells were tested for mycoplasma contamination.

Plasmids coding for different validated CIN targeting shRNAs in a pLKO.1-puro backbone (TRCN0000050044, TRCN0000050046) and the SHC002 plasmid, a control shRNA (Sigma-Aldrich, St. Louis, USA), were transfected in HEK293T cells with lipofectamine 2000 (Life Technologies) together with a third generation lentiviral packaging system (Addgene, Cambridge, MA, USA) and the p-advantage vector (Promega, Fitchburg, WI, USA). 72 h and 96 h after transfection, supernatants were harvested and remaining cells were removed from the solution by centrifugation for 5 min at 100 x g. To remove any serum remnant from the HEK293T cell medium, which would induce differentiation of the stem-like cells, we purified the lentiviral particles twice with PEGit (System Biosciences, Palo Alto, CA, USA) according to the manufacturer’s instructions. Briefly, for one round of purification, 4 parts supernatant were mixed with 1 part PEGit, and the solution was incubated O/N at 4 °C. Viral particles were collected by centrifugation at 1500 x g, for 30 min at 4 °C. The supernatant was discarded and the remaining pellet and liquid recentrifuged for 5 min and 1500 x g at 4 °C. After complete removal of the supernatant, the remaining pellet was dissolved in 4.5 ml DMEM/F12 for further purification or transduction of stem-like cells. We determined the concentration of the lentiviral particles with a p24 ELISA kit (Cellbiolabs, San Diego, CA, USA) and infected the cells at an MOI of 5. Two days after transduction 1 μg/ml puromycin (Sigma-Aldrich, St. Louis, USA) was added to the culture medium to select for shRNA expressing cells. After two weeks of selection we obtained robustly proliferating cell cultures. For simplicity, TRCN0000050044 is hereafter referred to as CIN shRNA #1, TRCN0000050046 as CIN shRNA #2 and SHC002 as CTRL.

For proliferation assays, 2000 stably transduced NCH421k and NCH644 cells were seeded in five separate 96-well plates in a final volume of 100 μl. Every day, 10 μl resazurin (R&D Systems, Minneapolis, MN, USA) were added to one plate, incubation was performed for 3 h at 37 °C and 5% CO2 and fluorescence intensity was measured at in a FLUOstar Omega microplate reader at Ex544nm/Em590nm (BMG Labtech, Ortenberg, Germany). After background (medium w/o cells plus resazurin) substraction the values were expressed as fold of the intensity at day 1. The chemotherapeutic agent temozolomide (Sigma-Aldrich) was dissolved in DMSO at concentrations of 200 mM. The ROCK-inhibitors Y-27632 (Sigma-Aldrich) and fasudil (Tocris Bioscience, Bristol, UK) were dissolved in sterile ultrapure water (Carl-Roth, Karlsruhe, Germany) at a concentration of 10 mM. The LIMK-inhibitor LIMKi3 (Tocris Bioscience) was dissolved in DMSO at a concentration of 10 mM. All reagents were thawed three times at maximum. For chemosensitivity assays, 1000 NCH644 or NCH421k cells were seeded per well on a 96-well plate in stem cell medium. The cells were treated with 10 serial dilutions of temozolomide ranging from final concentrations of 1000 to 0.01 μM. Then, Y-27632, fasudil or LIMKi3 were added in a final concentration of 10 μM (in a final volume of 200 μl), a concentration chosen based on literature reports [22‐24]. DMSO and water served as a control. The plates were incubated for 96 h, 20 μl of resazurin were added and measurement of resazurin fluorescence intensity was performed as has been described above. For the chemosensitivity assays of shRNA cells the protocol was performed without the inhibitor treatement.

For western blot cells were washed in DPBS supplemented with 1% BSA and lysed in 150 μl of RIPA lysis buffer with added phosphatase and protease inhibitor cocktail (Roche, Basel, Switzerland) and kept on ice. The lysates were mixed with Laemmli buffer, denatured at 90 °C for 5 min. DNA was sheared with a 20G × 1.5″ needle and the samples were run on 8–15% SDS-PAGE gels depending on the size of the analyzed protein. MagicMark™ Western Protein Standard (Life Technologies) or Color Prestained Protein Standard, Broad Range (NEB, Ipswich, MA, USA) were used as a molecular weight marker. Gels were run at a constant voltage of 80 V for 30 min (stacking gel) followed by 140 V for 60–70 min (separating gel), dependent on the polyacrylamide concentration of the gels. For separation of cofilin and phosphocofilin PhosTag was added to the gels as has been described previously [8]. Protein was blotted from the SDS-PAGE gels on 0.45 μm nitrocellulose membranes (Bio-Rad, Munich, Germany) with a semi-dry Fastblot B44 (Biometra, Goettingen, Germany). Afterwards, the membrane was blocked using 5% non-fat dry milk for 1 h followed by incubation in primary antibody over night at 4 °C. The primary antibodies were diluted 1:10,000 (Tubulin, mouse antibody [Clone DM1A], Sigma-Aldrich) or 1:1000 for CIN (rabbit antibody [clone C85E3], Cell Signaling Technologies, Danvers, CO, USA), p-Ser3-cofilin (rabbit antibody [clone 77G2], Cell Signaling Technologies) and cofilin (rabbit antibody [clone D3F9], Cell Signaling Technologies). The next day, the membrane was washed three times in TBS-T for 2 min and then the primary antibody was detected by anti-rabbit or anti-mouse IgG linked to horseradish peroxidase (Santa Cruz, Dallas, Texas, USA) diluted 1:10,000 in a solution containing 5% non-fat dry milk for 1.5 h at RT. Picoluminescence substrate (Thermo Fisher Scientific, Waltham, Massachusetts, USA) was used for development on a LAS4000 imaging system (GE healthcare, Munich, Germany). Quantification was performed with ImageQuant TL ver. 7.0 (GE healthcare).

RNA and DNA were isolated with the RNA/DNA Allprep kit (Qiagen, Hilden, Germany) according to the manufacturer’s instructions. cDNA synthesis was performed from 1 μg total RNA using random hexamer primers (Gene Link, Hawthorne, NY, USA) and the SuperScript™ II Reverse Transcriptase (Life Technologies). RNA from normal human astrocytes was commercially available (ScienCell Research laboratories, Corte Del CedroCarlsbad, CA).

Real-time RT-PCR was performed with the SensiFAST™ SYBR Hi-Rox Kit (Bioline, London, UK) on the StepOnePlus™ cycler (Life Technologies). Relative expression values were calculated with the ΔΔC_T (analysis of relative gene expression) method [25] using ARF1 as the reference transcript. Primers used for CIN were 5′-CTGGAGACCGACATCCTCTTT (forward) and 5′-TTCTAGGCGGGAGACTCCTG (reverse), for ARF1 5′-GACCACGATCCTCTACAAGC (forward) and 5′-TCCCACACAGTGAAGCTGATG (reverse), for c-Myc 5′-TCGGATTCTCTGCTCTCCTC (forward) and 5′- TCATCTTCTTGTTCCTCCTCAGA (reverse) for NES 5′-ATCGCTCAGGTCCTGGAA (forward) and 5′-AAGCTGAGGGAAGTCTTGGA (reverse) and for GFAP 5′-TGAAGCCGAAGAGTGGTACC (forward) and 5′-GGTAGTCGTTGGCTTCGTG (reverse). TUBB3 primers were described by others [26]. Analysis of the CDKN2A deletion status was performed with four different primer sets targeting the CDKN2A locus as has been described by others [27, 28] with the real-time PCR conditions as described above and primers for human B2M as reference [29] as well as a commercially available human genomic DNA (Roche) as control.

Metabolites were isolated as has been described by others [30] with the exception that we used DPBS supplemented with 0.9% NaCl and 1% fraction V BSA for quenching, which was performed at 4 °C. These changes were necessary because extensive cell rupture occurred without BSA and by incubation on ice. It should be noted that vitamin B6 binds to BSA, albeit with lower affinity than to human serum albumin [31]. However, all BSA containing media were carefully removed before metabolite extraction and the amount of BSA used for quenching and washing was identical for every cell line. Therefore, the interference of BSA with the measured vitamin B6 level is identical across samples. The extracts were dried in a Savant SpeedVac-concentrator and finally resuspended in water to equal 20,000 cells/μl H₂O. Active vitamin B6 levels were measured with an enzymatic kit (Buehlmann Laboratories, Schoenenbuch, Switzerland) according to the manufacturer’s instructions. The kit was validated to yield comparable results to HPLC by the manufacturer and enzymatic methods have been shown to yield comparable results to HPLC in general [32]. Briefly, substrate was added to each well of a 96-well plate and an equal volume of 1:40 diluted sample was added. Afterwards, apoenzyme was added, and the plate was shaken for 15 s at 400 rpm. The mixture was incubated for 30 min at 37 °C, enzyme was added and the plate was shaken for 15 s. Then, another incubation was performed for 15 min. at 37 °C and the OD546 was measured.

Bisulfite conversion of 1 μg genomic DNA was performed with the EpiTect bisulfite kit (Qiagen, Hilden, Germany) according to the manufacturer’s instructions. Global DNA-methylation levels were estimated via LINE-1 bisulfite PCR [33]. Amplification of LINE-1 elements was performed with HotStarTaq DNA polymerase (initial denaturation at 95 °C for 15 min, 35 cycles of 94 °C for 90s, 50 °C for 60s, 72 °C for 60s and a final elongation step at 72 °C for 10 min). Sequencing was performed on a Pyromark Q24 instrument (Qiagen) following standard protocols.

Cells were counted and ~ 1.5 × 10⁵ cells were washed and resuspended in DPBS containing 1% BSA, fixed with 4% PFA for 15 min and collected by centrifugation at 500 x g for 5 min. Afterwards, the cells were blocked and permeabilized in DPBS supplemented with 0.5% TritonX-100 and 5% normal goat serum (PAN Biotech). After two washing steps with DPBS supplemented with 0.5% TritonX-100, the cells were stained with 1 μg Histone-K27me3 antibody (rabbit polyclonal, Millipore, Billerica, MA, USA) and a rabbit isotype control (rabbit clone [DA1E], Cell Signalling Technologies) for 90 min. in 100 μl DPBS supplemented with 0.5% TritonX-100 and 1% normal goat serum. Afterwards, cells were washed twice, stained with a 1:200 diluted Alexa-Fluor 488 conjugated anti-rabbit secondary antibody (Thermo Fisher Scientific) for 90 min, counterstained with 1 μg/ml DAPI for 10 min and washed again twice. Finally, the cells were resuspended in DPBS and analyzed on a BD FACS CantoII flow cytometer. Here, DAPI positive single cells were gated and analyzed for their fluorescence signal.

Colony formation was assessed with the reagents and procedures as described previously [12] with the following modifications. We increased the cell number to 3000 cells per well, but reduced the incubation time to two weeks. For inhibitor treatment assays, Y-27632 was added to the collagen solution as well as to the feeding medium at the concentrations indicated. Water was used as a control.

Libraries for next-generation sequencing were prepared from 600 ng total RNA with the TrueSeq RNA library preparation kit v2 as has been described previously [34]. Illumina deep sequencing was performed at a genomics core facility: Center of Excellence for Fluorescent Bioanalytics (KFB, University of Regensburg, Germany) on a HiSeq1000 instrument.

Analysis of NGS data was performed using the Genomatix software (Genomatix, Munich, Germany). First, the .fastq files were mapped to the human genome assembly GRCh38 (annotation based on ElDorado 6–2015) using the Genomatix Mining Station Mapper v3.7.6.3 allowing one mismatch. All unique hits were further processed using the Genomatix Genome Analyzer v3.51106 which was used to create count tables for all samples. Reads were counted locus-based, i.e. for unions of exons of genes. All further analyzes were performed with the free software R v3.1.1, Bioconductor v3.0 [35] and the package DESeq2 v1.6.3 [36]. Gene set enrichment analysis [37] was performed with the ssGSEA module v7 [38] with RPKM (reads per kilobase of exon model per million mapped reads) values [39]. Subtype prediction was performed with the ssGSEA module and the gene sets proposed by ref. [40].

All analyzes were performed with GraphPad Prism 5.0 and R ver. 3.1.1. If not otherwise indicated, two-sided t-tests were used for statistical analysis of two groups. For three or more groups, a one-way ANOVA followed by Dunnett’s multiple comparison test was applied. A result was accepted as significant if p was < 0.05 and significant differences were indicated where present (* p < 0.05, ** p < 0.01 and *** p < 0.001). For the determination of IC₅₀ values, the concentration of temozolomide used was log-transformed, the fluorescence values were normalized and a fit with variable slope was performed.

As a starting point, we performed an extensive molecular characterization of five non-adherent cell lines cultured in serum-free medium and five adherent cell lines cultured in serum-containing medium. For this purpose, we determined their expression subtype by next generation sequencing followed by a single sample gene set enrichment analysis. In addition, we determined CDKN2A deletion, IDH1/2 mutation and the TP53 mutational status of our cell lines (Fig. 1a and Additional File 1: Table S1). We found that TP53 mutations were very common alterations in both serum-cultured and cell lines cultured under serum-free conditions and that all lines tested carried at least a hemizygous CDKN2A deletion (Fig. 1a). Interestingly, the three best characterized glioblastoma expression subtypes (proneural, classical, mesenchymal) were all present in our cell line cohort (Fig. 1a). We also confirmed that the bona fide stem cell markers PROM1 (CD133), NES (nestin), SOX2 [41] as well as MYC (c-Myc) were overexpressed in the cells cultured under serum-free conditions (Fig. 1b), although the difference was only significant for PROM1 as determined by DESeq2 (adj. p < 0.001). When we examined differential expression patterns between serum-cultured cells and cells cultured in serum-free medium, we found that proteins regulating cofilin phosphorylation were indeed deregulated in stem-like cells, in a way favoring lower cofilin phosphorylation (Fig. 1c). This is in accordance with studies examining cofilin phosphorylation in pluripotent cells and colon cancer [9, 10, 42]. To corroborate these findings and excluding media artifacts, we reexamined the NGS-data from another study [43] that established cultures from tumor propagating glioma cells and differentiated glioma cells. Indeed, we found similar effects (Fig. 1d). Especially CIN was highly and significantly upregulated in stem-like cells (p < 0.001, two-sided t-test, Bonferroni-corrected). We then verified our sequencing data with real-time PCR and western blotting. There was a significant reduction in CIN mRNA as judged by real-time PCR in serum-cultured lines (two-sided Mann-Whitney test, p < 0.05) although levels were variable (Fig. 1e). Normal human astrocytes were used as a reference for this analysis. CIN showed a higher abundance on protein level in every cell line cultured under serum-free conditions and the difference in protein levels between the culture conditions was highly significant (Fig. 1f, p < 0.0005).

We analyzed the levels of active vitamin B6 and expression changes in the pathway regulating vitamin B6 metabolism. Active vitamin B6 levels were significantly higher in serum-cultured cell lines and the expression of AOX1 that metabolizes the precursor of pyridoxal 5′-phosphate, the active form of vitamin B6, was strongly downregulated (Fig. 2a and b). We then examined cofilin phosphorylation taking advantage of the PhosTag compound [44], which separates specifically p-Ser3-cofilin from unphosphorylated cofilin in glioma cells as established previously [8]. We found that p-cofilin levels are indeed strongly reduced in glioblastoma cell lines cultured under non-adherent, serum-free conditions (Fig. 2c and d; p < 0.01). Importantly, we also verified this result with a p-Ser3-cofilin specific antibody and found that results from PhosTag blotting and blotting with a p-Ser3-cofilin specific antibody were highly correlated (Additional file 2: Figure S1). We therefore hypothesized that CIN might regulate glioblastoma stem cell identity via its function in cofilin phosphorylation and/or vitamin B6 metabolism.

We established CIN-knockdown cell lines from two stem-like cell lines representing the proneural expression subtype, NCH644 and NCH421k. CIN levels are higher in this subtype as compared to the classical and mesenchymal subtype in brain tumor samples [8] and this model system should be closest to the tumor situation.

There was a highly significant reduction in CIN mRNA levels, but not in the mRNA levels of the stem cell markers c-Myc or NES (two-way ANOVA, Bonferroni-corrected post-test, p < 0.01 in NCH644 and NCH421k for both CIN shRNA #1 and CIN shRNA #2; p > 0.05 for c-Myc and NES, Fig. 3a and b). However, the differentiation markers GFAP and TUBB3 were overexpressed in every knockdown cell line (Additional file 3: Figure S3). The CIN knockdown was also confirmed on protein level in both NCH644 and NCH421k (Fig. 3c and d). Proliferation was neither affected in NCH421k nor NCH644 and knockdown cell lines showed robust in vitro growth with no differences to controls (repeated measures, two-way ANOVA, Bonferroni-corrected post-hoc tests, p > 0.05; Fig. 3e and f).

We then analyzed if one of the two substrates of CIN, active vitamin B6 or phosphocofilin showed altered abundance after knockdown of CIN. Indeed, active vitamin B6 levels increased significantly in NCH644 and NCH421k cells (Fig. 4a and b: one-way ANOVA followed by Dunnett’s multiple comparison’s test, p < 0.05 and p < 0.001 for both shRNAs in NCH644 and NCH421k, respectively), whereas phosphocofilin/total cofilin ratios remained unaltered (Additional file 4: Figure S2, one-way ANOVA followed by Dunnett’s multiple comparison’s test, p > 0.05). It should be noted that alterations in phosphocofilin/total cofilin ratios are often absent in serum-cultured cell lines, too [8].

We then examined if the highly clonogenic cell lines NCH644 and NCH421k showed an alteration in colony formation ability on CIN-knockdown. Indeed, the colony numbers where significantly reduced by both CIN targeting shRNAs in NCH644 (Fig. 4c, generalized linear model, family = poisson, link-function = log, p < 0.01 for both shRNAs) but not in NCH421k (Fig. 4d, generalized linear model, family = poisson, link-function = log, p > 0.05 for both shRNAs). Importantly, treatment with the ROCK-inhibitor Y-27632 led to a significant increase in colony numbers (Fig. 4e, generalized linear model, family = poisson, link-function = log, p < 0.01 for Y-27632 at 10 μM) supporting our hypothesis that phosphoregulation of cofilin is responsible for this phenotype. In addition, in NCH421k cells, no significant phenotype was found in analogy to the CIN knockdown results.

As cell proliferation itself was not perturbed by CIN knockdown, we performed poly-A RNA-Seq in NCH421k and NCH644 cells to determine if CIN might influence stem cell-related transcriptional activity. We deposited this dataset under GEO accession GSE98797. On the single gene level, quality control plots showed a high enrichment of small p-values after standard filtering [36], indicative of a significant effect of CIN on the cellular transcriptome and many deregulated genes (Additional files 5, 6, 7: Figure S4, Table S2 and Table S3).

In NCH644, there were 409 upregulated and 258 downregulated genes and in NCH421k there were 172 upregulated and 66 downregulated genes (Fig. 5a and b: logFC ≥0.6, adjusted p < 0.1). Hierarchical clustering based on rlog normalized expression values of the Top100 significantly deregulated genes (sorted by adjusted p-value) separated CIN shRNA and control cells efficiently in both NCH644 (Fig. 5c) as well as NCH421k (Fig. 5d). There was a limited overlap of 28 upregulated and 8 downregulated genes, when NCH644 and NCH421k were compared (Fig. 6a and b). Among the common upregulated genes was the inhibitor of hedgehog signaling TULP3 [45], which was significantly overexpressed in all CIN knockdown cell lines (Fig. 6c, adjusted p-values calculated by DESeq2, · = p < 0.1, * = p < 0.05, ** = p < 0.01, *** = p < 0.0001). The expression patterns of other glioma stem cell promoting proteins that were found in the list of deregulated genes, ITGA6 [46] and BMI1 [47], where variable. While ITGA6 was significantly downregulated in NCH421k there was only a trend in NCH644 cells. BMI1, as well as other tumor promoting genes (like EGFR and LEF1) were significantly upregulated in NCH644 in contrast to our expectations (Fig. 6d, adjusted p-values calculated by DESeq2, · = p < 0.1, * = p < 0.05, ** = p < 0.01, *** = p < 0.0001). CIN knockdown was highly efficient for both shRNAs in both cell lines (Fig. 6e and f, adjusted p-values calculated by DESeq2).

Global histone and DNA methylation, two possible modifications linking the transcriptomic changes and the cellular function of CIN, remained unaltered as determined by flow cytometry and pyrosequencing of LINE-1 elements (Additional file 8: Figure S5).

Finally, as turnover of p-cofilin can be more important than absolute levels of the phosphorylated protein and altered absolute p-cofilin levels are not present in all cell lines cultured in serum after CIN deregulation [8], we tested if inhibitors of the upstream kinases ROCK1/2 or LIMK1/2 are able to sensitize the cell lines NCH421k and NCH644 to chemotherapeutic agents. Treatment with these inhibitors alone had an effect -if any at all- at excessively high concentrations (Fig. 7a and b, Additional file 9: Figure S6, one-way ANOVA followed by Dunnett’s multiple comparison test, significant changes indicated). However, there were opposite effects in NCH644 and NCH421k: While after inhibitor treatment of NCH644 cells there was a significant chemosensitization towards temozolomide, NCH421k cells instead showed an increase in resistance to temozolomide (Fig. 7c-e, n = 3, one-way ANOVA followed by Dunnett’s multiple comparisons test, significant changes are indicated). Cofilin phosphorylation in our hands was very efficiently abolished by 10 μM Y-27632 and LIMKi3 and less efficiently by fasudil in stem-like cells. U87, a classical serum cultured glioblastoma cell line with a very high phosphocofilin level was used as a positive control (Fig. 7f, Additional file 9: Figure S6). In contrast, there was no change in chemosensitivity after CIN knockdown (Additional file 10: Figure S7, one-way ANOVA followed by Dunnett’s multiple comparison test, significant changes indicated).