Regulation of cellular sphingosine-1-phosphate by sphingosine kinase 1 and sphingosine-1-phopshate lyase determines chemotherapy resistance in gastroesophageal cancer

Background Resistance to chemotherapy is common in gastroesophageal cancer. Mechanisms of resistance are incompletely characterised and there are no predictive biomarkers in clinical practice for cytotoxic drugs. We used new cell line models to characterise novel chemotherapy resistance mechanisms and validated them in tumour specimens to identify new targets and biomarkers for gastroesophageal cancer. Methods Cell lines were selected for resistance to oxaliplatin, cisplatin and docetaxel and gene expression examined using Affymetrix Exon 1.0 ST arrays. Leads were validated by qRT-PCR and HPLC of tumour metabolites. Protein expression and pharmacological inhibition of lead target SPHK1 was evaluated in independent cell lines, and by immunohistochemistry in gastroesophageal cancer patients. Results Genes with differential expression in drug resistant cell lines compared to the parental cell line they were derived from, were identified for each drug resistant cell line. Biological pathway analysis of these gene lists, identified over-represented pathways, and only 3 pathways - lysosome, sphingolipid metabolism and p53 signalling- were identified as over-represented in these lists for all three cytotoxic drugs investigated. The majority of genes differentially expressed in chemoresistant cell lines from these pathways, were involved in metabolism of glycosphingolipids and sphingolipids in lysosomal compartments suggesting that sphingolipids might be important mediators of cytotoxic drug resistance in gastroeosphageal cancers . On further investigation, we found that drug resistance (IC50) was correlated with increased sphingosine kinase 1(SPHK1) mRNA and also with decreased sphingosine-1-phosphate lysase 1(SGPL1) mRNA. SPHK1 and SGPL1 gene expression were inversely correlated. SPHK1:SGPL1 ratio correlated with increased cellular sphingosine-1-phosphate (S1P), and S1P correlated with drug resistance (IC50). High SPHK1 protein correlated with resistance to cisplatin (IC50) in an independent gastric cancer cell line panel and with survival of patients treated with chemotherapy prior to surgery but not in patients treated with surgery alone. Safingol a SPHK1 inhibitor, was cytotoxic as a single agent and acted synergistically with cisplatin in gastric cancer cell lines. Conclusion Agents that inhibit SPHK1 or S1P could overcome cytotoxic drug resistance in gastroesophageal cancer. There are several agents in early phase human trials including Safingol that could be combined with chemotherapy or used in patients progressing after chemotherapy. Electronic supplementary material The online version of this article (doi:10.1186/s12885-015-1718-7) contains supplementary material, which is available to authorized users.


Details of sample prepapration and gene expression profiling RNA Extraction
Cells were grown in T75 flasks to 80 -90 % confluency and washed twice in PBS. 1 ml of TRIzol (Invitrogen, Paisley, UK) reagent was added per flask and then lysates were transferred into 1.5 ml eppendorf tube and passed through a 20 gauge needle (0.9 mm diameter) at least 5-6 times. Samples were centrifuged at 1200g for 10 minutes at 4 °C and supernatants were collected and transferred into fresh tubes and incubated for 5 minutes at room temperature (RT). 200 µl of chloroform (Sigma Aldrich) was added and tubes were inverted several times and incubated for 3 minutes at RT. Tubes were centrifuged at 12000 g for 15 minutes at 4 ᵒC. Supernatants were collected and transferred into new eppendorf tubes. 500 µl of 100 % isopropanol was added, mixed and incubated for 10 minutes at RT. Samples were centrifuged at 12000 g for 10 minutes at 4 °C. Supernatants were removed and pellets washed with 1ml of 75 % ethanol, vortex and centrifuged at 9500 rpm for 5 minutes at 4 °C. Remaining supernatants were removed, pellets briefly air-dried and resuspended in 30 µl of RNase free water. RNA was quantified spectrophotometrically (Nanodrop 1000 Spectrophotometer, Thermo Scientific, Loughbourough, UK). Both A260/280 and A260/230 ratios were determined and all samples were in the range 1.9-2.3. Furthermore samples were purifed on the Mini columns (Qiagen). Quality was assessed by electrophoresis on Tapestation (Lab901 Limited, Peqlab, UK) and QC was determined by SDV (Screen Tape degradation value) that represents RNA integrity. Values 0-5 represent high quality RNA, 5-14 -partially degraded RNA and ≥ 15 degraded RNA. All RNA samples had SDV values between 0.3 and 2.6. All analyses used 3 independent replicates per cell line from three different passages.

Sample preparation for Gene expression profiling and hybridisation
500 ng of total RNA was reverse transcribed into cDNA and further amplified in vitro into cRNA using Ambion WT Expression kit (Austin, TX, USA). The quality of cRNA was determined by electrophoresis (Tapestation, Lab901 Limited, Peqlab, UK). The A260/280 and A260/230 ratios, concentration and yield were determined on a spectrophotometer (Nanodrop 10000 spectrophotometer, Thermo Scientific, Loughborough, UK). cRNA (10µg) was reverse transcribed into cDNA (Ambion WT Expression Kit, Ambion) and its quantity and quality was determined as above. Subsequently, cDNA samples (5.5µg) were fragmented and biotin labelled (WT terminal labelling and controls kit, Affymetrix, Santa Clara, CA), and 5.5 µg hybridised to Human Exon 1.0 ST GeneChip microarrays (Affymetrix, Santa Clara, CA) at 45 °C for 17 hours at 45 rpm in a hybridization oven.

Gene expression Data Quality assessment(QA) of Gene expression data
QA data for the profiling of drug resistant and parental cell lines is provided in table S4.1 and figure S4.1 below.
Core probe sets on the Human Exon 1.0 ST array were processed using a modified robust multiarray analysis (RMA16) algorithm (Affymetrix, Santa Clara, CA) that employs a non-linear per chip background correction with addtion of 16 to the expression values to attain variance stabilisation of low level signals, quantile normalisation and summarisation of multiple probe sets per transcript using median polishing of log2 transformed data. Data were transformed to the median of all samples. QA was performed by examining signal intensity (PM_mean), background signal detection (Bgrd_mean), detection of outliers followed by analysis of hybridization and labelling controls. Further determination of outlier samples was performed by analysis of probe set summarization metrics such as Pos_vs_neg_auc and All_Probe_Set_RLE_Mean.
Quality control of hybridization and labelling was determined by analysis of bacterial (bac_spike) and polyadenylated (polyA_spike) controls and polyadenylated RNA spikes such as Lys, Phe, Thr and Dap were analyzed independently. Table S4.1 Quality assessment measures from Affymetrix 1.0ST Exon arrays for all cell line samples Figure S4.1 Box plots represent the relative log expression for all the probe sets analyzed. The mean absolute RLE is proportional to the width of the box plots, or the inter-quartile range of RLE values and whiskers are 1.5x IQR.

Analysis of gene expression data
Gene expression data analysis was performed in GeneSpring GX v 11.5 using RMA16 normalization, log transformation and baseline to median of all samples. Core gene sets were analysed andentities with normalised expression levels between the 20th and 100 th percentiles in at least 1 sample were included with 16939 out of 17881 genes meeting this filter. Unpaired t-test, with Benjamini and Hochberg MTC corrected p ≤ 0.05 was performed to identify discriminatory gene profiles of each pair of drug resistant versus parental lines.

Pathway analysis
Gene set enrichment analysis using gene ontologies and mapping of gene sets of interest onto biological pathways was performed using DAVID v 6.7 (Function Annotation Bioinformatics Microarray Analysis), Bioinformatic Resources, NIAID, NIH Significantly enriched GO terms, functional networks or pathways were determined using gene set enrichment analysis (EASE score, DAVID v 6.7) and genes were visualised on BioCarta and KEGG pathways. Data was pre-processed in DAVID bioinformatics database to provide the link between probe IDs and Entrez Gene Symbols. Probe sets with multiple Entrez IDs and those which could not be found in Ensembl/Entrez ID were excluded from further analysis.