In vivo detection of dysregulated choline metabolism in paclitaxel-resistant ovarian cancers with proton magnetic resonance spectroscopy

All cells were cultured in RPMI 1640 medium (Sigma Aldrich, St Louis, MI, USA) with 10% fetal bovine serum (FBS, Gibco, Thermo Fisher Scientific, Waltham, MA, USA) and maintained at 37 °C under 5% CO₂. PTX-resistant OVCAR-3 (OV_PTX) cells were generated in the laboratory and kindly provided by Dr. GX Xu (Fudan University, China) [23]. OV_PTX cells were derived from parental OVCAR-3 (OV) cells by treating cells with the PTX (Sichuan Taiji Pharm, China) regimen through a gradually increasing PTX dose in RPMI 1640 medium with 10% FBS.

Cell proliferation was measured by Cell Counting Kit-8 (CCK-8, Dojindo Molecular Technologies Inc., Shanghai, China). OV and OV_PTX cells were plated in 96-well plates with a density of 5 × 10³ cells per well and incubated for 24 h. PTX was added with increasing concentrations from 0.001–10 μmol/mL to cells, which were then incubated for 48 h. The cells were then incubated with 10 μL of CCK-8 per well for 1 h at 37 °C. Absorbance was measured at a wavelength of 450 nm using a Bio-Tek ELX808IU absorbance microplate reader (Bio-Tek Instruments Inc., USA).

Female BALB/c nude mice (Jiesijie Laboratory Animal Company, Shanghai, China; age, 4–5 weeks; weight, 12–15 g) were used under approved animal care. In vivo experiments were performed in accordance with the guidelines formulated by the Ethics Review Committee of China Animal Experimental System and were approved by the ethics committee of Shanghai Municipal Public Health Clinical Center (No. 2020-A019-01). Twenty nude mice were randomly divided into the OV group and OV_PTX group (n = 10 in each group). Briefly, 5 × 10⁶ OV or OV_PTX cells/mouse were suspended in serum-free medium, and 0.1 mL of the cell suspension was injected subcutaneously in the right anterior limb. Mice were monitored daily and weighted every two days. Tumor size was measured with a caliper every two days for the greatest longitudinal diameter (length) and greatest transverse diameter (width). Volume was calculated using the modified elliptical formula (length × width²)/2 [24]. Tumors were allowed to grow for 19 days after injection until a diameter of approximately 1.0 cm was measured, which were suitable for in vivo ¹H-MRS. One mouse in the OV_PTX group died before imaging.

Mice with OV or OV_PTX tumors were anesthetized with isoflurane (1.5–2%) in oxygen (1 L/min) and imaged in the prone position in a 7.0 T Biospec small-animal MRI scanner (Bruker Corporation, Billerica, MA, USA). The imaging protocol included the following sequences:

The tumors were scanned on the transverse, coronal and sagittal planes using the T2 rapid acquisition relaxation enhance sequence for the three-dimensional positioning of ¹H-MRS. Metabolite spectral fitting was performed using LCModel Version 6.3-0I, with a basis set provided by the LCModel software for a 7.0 T Bruker MRI scanner with time of echo = 16 ms [25]. Relative metabolite concentrations and their uncertainties were estimated by fitting the spectrum to a linear combination of basis spectra of each individual metabolite. The unsuppressed water spectrum was used to normalize the initial fit to generate a first estimate of metabolite concentration by scaling the relative areas and chemical shifts across the two sets of spectra. The spectral range for the analysis was set to 0.2–4.0 ppm to contain most peaks of interest: alanine (Ala), aspartate (Asp), creatine (Cr), phosphocreatine (PCr), glycerophosphocholine (GPC), phosphocholine (PC), inositol (Ins), lactate (Lac), taurine (Tau), N-acetylaspartate (NAA), N-acetylaspartylglutamate (NAAG), macromolecules (MM) 09, lipid (Lip) 09, Lip 13a, Lip 13b, MM12, MM14, MM20, Lip 20. The numbers after MM and Lip indicated the approximate chemical shift in ppm of the peaks; e.g., MM 14 for the macromolecule peak near 1.4 ppm. Only metabolite concentrations quantified with Cramèr-Rao lower bounds below 20% on average were included in further analysis [26]. With software (Fire Voxel, CAI2R, New York University, NY, USA), the region of interest was manually delineated slice-by-slice along the contour of the tumor on the transverse T2-weighted images (T2WIs) (M.X.L. and L.J., with 5 and 7 years of experience in gynecological imaging, respectively). Then, the volume of interest was postprocessed automatically for tumor anatomic measurement.

In order to acquire a true map of in vivo metabolism, tumors were collected as soon as possible after MRI scanning with mice euthanized by excess CO₂ exposure. Tumor samples were divided into multiple parts. One part was fixed with 4% paraformaldehyde and then embedded in paraffin and stained using hematoxylin and eosin for histological feature analysis. The other parts of tumor samples were flash-frozen in liquid nitrogen and were used for metabolomics analysis, proteomics analysis and quantitative RT-PCR, respectively.

Tumor samples (100 mg) were extracted for analysis using by liquid chromatography-mass spectrometry (LC–MS). Pooled quality control samples were also prepared by combining the same volume of each sample and repeatedly injected during the assay to monitor instrumental stability and avoid systematic bias. The acquired LC–MS data was pretreated using XCMS software. Features with < 50% of quality control samples or 80% of test samples were removed, and values for missing peaks were extrapolated with the k‐nearest neighbor algorithm to further improve the data quality. The group datasets were normalized before analysis. The detailed parameters were described in Additional file 1: Supplementary of metabolomics analysis method. Statistical analysis included principal component analysis (PCA), orthogonal partial least squares discriminant analysis (OPLS-DA), Student’s t-test and fold change analysis. The P value obtained by Student’s t‐test was then adjusted for multiple tests using a false discovery rate-reducing process (Benjamini-Hochberg) and was used to determine differential metabolites. The following criteria were used to screen the differential metabolites: variable importance in projection (VIP) sores ≥ 1, P value < 0.05, and fold change ≥ 2 or ≤ 0.5. Pathway enrichment analysis of differential metabolites was performed using the Kyoto Encyclopedia of Genes and Genomes and MetaboAnalyst 3.0 (Montre al, QC, Canada) databases. Bioinformatic analysis was performed using the OmicStudio tools at https://​www.​omicstudio.​cn/​tool.

Samples were lysed in sodium dodecyl sulfate buffer and homogenized. Proteins were digested overnight by trypsin (Promega, Madison, WI, USA), and the resulted peptides were collected as a filtrate. Pooled peptides from all samples were fractionated by reversed-phase chromatography using an Agilent 1260 infinity II HPLC (SCIEX, Framingham, MA, USA). The detailed parameters were described in Additional file 2: Supplementary of proteomics analysis method. Raw data of data-independent acquisition were processed and analyzed by Spectronaut 14.6 (Biognosys AG, Switzerland) with default settings. Data extraction was determined by Spectronaut X based on extensive mass calibration. Precursors which passed the filters were used for quantification.The average top 3 filtered peptides that passed the 1% Q- value cutoff (false discovery rate) were used to calculate the major group quantities. Significantly enriched proteins were selected using Student’s t-test.

Total RNA was extracted from tumors using Trizol reagent and reverse-transcribed into cDNA with 500 ng of total RNA using the PrimeScript RT reagent kit (Thermo Fisher Scientific, Waltham, MA, USA) according to the manufacturer’s instructions. Gene expression assays were performed on an ABI StepOnePlus Real-Time PCR instrument (Thermo Fisher Scientific, Waltham, MA, USA). The following custom designed primers were purchased from GENEWIZ Inc. (Suzhou, China): GPCPD1, GTTTTTGCGATATGTGGAAGCTG (forward) and AGCGATACTGAACTGATACTCCT (reverse); GDE1, GACTGGGCGATTGTGTGATTT (forward) and AGGGTAGGGATCTTTTCATCAGG (reverse); and β-Actin, GCCGTGGTGGTGAAGCTGT (forward) and ACCCACACTGTGCCCATCTA (reverse).

Continuous variables with normal distribution were presented as the mean ± standard deviation; nonnormally distributed variables were presented as the median (interquartile range). All statistical analyses were performed with SPSS (version 23.0, SPSS, Inc., Chicago, IL, USA). The data were analyzed using Student’s t-test or Mann–Whitney U test. Receiver operating characteristic curve analysis (MedCalc Software, Mariakerke, Belgium) was used to assess the diagnostic performance and determine a cutoff value for the significant metabolites from ¹H-MRS to differentiate the PTX-sensitive and PTX-resistant tumors. The correlation of metabolites between in vivo ¹H-MRS and ex vivo metabolomics analysis and the correlations between protein expressions and metabolite levels were analyzed using Spearman correlation tests. Differences with a P < 0.05 were considered statistically significant.

We performed the cell proliferation assay to confirm the PTX resistance, with a resistance index of 484.3 ± 166.8 (Fig. 2A). Cell atypia was more obvious in the OV_PTX tumor than that in OV tumor (Fig. 2B). Xenograft tumors were significantly larger in the OV group (1028.5 ± 370.5 mm³) than in the OV_PTX group (337.2 ± 224.0 mm³) (P < 0.001).

Eight mice from each group were successfully imaged. A total of 13 metabolites (Cramèr-Rao lower bounds < 20%) were analysed, but no significant differences in frequencies of metabolite peaks were observed between the OV and OV_PTX tumors (Table 1; Fig. 2C). Thereinto, PC was very difficult to resolve from GPC, as was PCr from Cr. The sum of GPC + PC was much more accurate than the individual concentrations. Usually, the units of the absolute concentrations are unknown, and only concentration ratios are meaningful. LCModel metabolite ratios relative to Cr + PCr were used for our analysis (as outlined in detail in the LCModel manual). The metabolite ratios and their differential levels in the OV and OV_PTX tumors are shown in Table 1 and Fig. 2D. The ratio of (GPC + PC) to (Cr + PCr) (i.e. Cho/Cr) in the OV_PTX group (1.64 [0.69, 4.18]) was significantly higher than that in the OV group (0.33 [0.10, 1.13]) (P = 0.04). The metabolites of interest are displayed in Fig. 3A and B. For differentiating the two groups of tumors, the optimal cutoff value of the (GPC + PC)/(Cr + PCr) ratio was 1.216, which yielded the sensitivity, specificity, and area under the curve (AUC) values of 100%, 80%, and 0.85 (95% confidence interval, 0.550–0.982), respectively (Fig. 3C). No significant differences in other 4 metabolite ratios were found between the OV and OV_PTX tumors, although higher ratios were detected in the OV_PTX tumors.

To decipher the mechanisms of in vivo metabolic imaging alterations associated with treatment resistance, we characterized the metabolome of the OV and OV_PTX tumors from xenograft models. As shown in the PCA plot, the metabolites of the OV group and the OV_PTX group were well separated (Fig. 4A; Additional file 3: Fig. S1A). Next, the OPLS-DA model was used to further analyze the compounds responsible for the differences between the two groups. Goodness of fit values and predictive ability values indicated that the model possessed a satisfactory fit with a good predictive power (Fig. 4B, C; Additional file 3: Fig. S1B, C). Differential metabolites were screened using VIP and fold change as thresholds. Metabolites that met VIP ≥ 1, fold change ≥ 2 or ≤ 0.5, and P < 0.05 were considered to be differential metabolites (Fig. 4D, E). Forty-five ex vivo metabolites were identified to be significantly different between the OV tumors and the OV_PTX tumors, with the majority involving of lipids and lipid-like molecules (66.67%, Fig. 5A, Additional file 4: Table S1). Compared to those in the OV tumors, the levels of 14 metabolites increased while 31 ones decreased in the OV_PTX tumors. Metabolomics analysis showed the top ten differential metabolites according to VIP values: spermine, taurocholate, spermidine, vitamin C, 1-methyl-6-phenyl-1H-imidazo[4,5-b]pyridin-2-amine, acylcarnitine 19:0, glycerophosphocholine, acylcarnitine 14:0, acylcarnitine 17:0, and palmitoylcarnitine (Fig. 5B). Those may be potential biomarkers to identify the PTX-resistant EOC.

Pathway enrichment analysis was performed to identify dysregulated pathways in PTX-resistant tumors. It showed that the commonly dysregulated pathways included lipid metabolism, glycometabolism, amino acid metabolism, nucleotide metabolism, and energy metabolism (Fig. 5C). The top 3 related metabolism pathways were taurine and hypotaurine metabolism, glutathione metabolism and glycerophospholipid metabolism.

The correlation between in vivo differential metabolite (GPC + PC) and every ex vivo differential metabolite was investigated by calculating the Spearman’s correlation coefficient. There were 30 ex vivo differential metabolites that correlated significantly with the differential metabolite GPC + PC (|r|> 0.5, P < 0.05) observed on in vivo ¹H-MRS, exhibiting the highest positive correlation with GPC (r = 0.885, P < 0.001) (Fig. 5D). Overall, these data indicated that abnormal Cho concentrations were detected in the PTX-resistant EOCs because of the dysregulated glycerophospholipid metabolism, especially Cho metabolism.

According to the major metabolic reprogramming observed in the PTX-resistant EOCs, some essential enzymes might have modulated. Thus we analyzed the proteomics data to identify potential target proteins. A total of 233 differentially expressed proteins, which met fold change ≥ 2 or ≤ 0.5, and P < 0.05, were detected between the OV and OV_PTX tumors, without any enzyme of the Cho metabolism (Fig. 6A). GPCPD1 and GDE1 which are responsible for the cleavage of GPC [27] (Fig. 6B) were detectable in the OV tumors, but not in the OV_PTX tumors. Undetectable data in the OV_PTX tumors represented the proteins undetected in over half of the tumors under this experimental methods and conditions. In view of this, we tried to identify GPCPD1and GDE1 since little is known about them in the PTX-resistant EOCs. The expressions of GPCPD1 and GDE1 mRNA were significantly lower in the PTX_OV tumors compared to the OV tumors (both P < 0.01) (Fig. 6C).

Next, to address the question that whether gene expression levels of GPCPD1 and GDE1 were related to metabolic outcome, Spearman correlation coefficients were calculated between protein expressions and differential metabolite levels. GPC + PC, the differential metabolite observed on in vivo ¹H-MRS, correlated negatively with the mRNA levels of GPCPD1 and GDE1, with r = -0.698 (P = 0.01) and -0.775 (P = 0.003). There were 18 and 26 ex vivo differential metabolites correlated significantly with mRNA levels of GPCPD1 and GDE1 (|r|> 0.8, P < 0.05), with a closely negative correlation with GPC (r = -0.841 and -0.912, both P < 0.001, respectively) (Fig. 6D). Expressions of enzymes responsible for GPC decomposition reduced in the OV_PTX tumors, leading to the accumulation of GPC.