Serum Metabolomics Reveals Dysregulation and Diagnostic Potential of Oxylipins in Tumor-induced Osteomalacia

Abstract

Context

Excessive production of fibroblast growth factor 23 (FGF23) by a tumor is considered the main pathogenesis in tumor-induced osteomalacia (TIO). Despite its importance to comprehensive understanding of pathogenesis and diagnosis, the regulation of systemic metabolism in TIO remains unclear.

Objective

We aimed to systematically characterize the metabolome alteration associated with TIO.

Methods

By means of liquid chromatography–tandem mass spectrometry–based metabolomics, we analyzed the metabolic profile from 96 serum samples (32 from TIO patients at initial diagnosis, pairwise samples after tumor resection, and 32 matched healthy control (HC) subjects). In order to screen and evaluate potential biomarkers, statistical analyses, pathway enrichment and receiver operating characteristic (ROC) were performed.

Results

Metabolomic profiling revealed distinct alterations between TIO and HC cohorts. Differential metabolites were screened and conducted to functional clustering and annotation. A significantly enriched pathway was found involving arachidonic acid metabolism. A combination of 5 oxylipins, 4-HDoHE, leukotriene B4, 5-HETE, 17-HETE, and 9,10,13-TriHOME, demonstrated a high sensitivity and specificity panel for TIO prediction screened by random forest algorithm (AUC = 0.951; 95% CI, 0.827-1). Supported vector machine modeling and partial least squares modeling were conducted to validate the predictive capabilities of the diagnostic panel.

Conclusion

Metabolite profiling of TIO showed significant alterations compared with HC. A high-sensitivity and high-specificity panel with 5 oxylipins was tested as diagnostic predictor. For the first time, we provide the global profile of metabolomes and identify potential diagnostic biomarkers of TIO. The present work may offer novel insights into the pathogenesis of TIO.

Tumor-induced osteomalacia (TIO), is a rare paraneoplastic syndrome caused by excessive fibroblast growth factor 23 (FGF23) produced by a tumor (1). Although TIO is the most common cause of acquired hypophosphatemic rickets/osteomalacia, it is a rare disease with fewer than 1000 reported cases (2). Typical symptoms presented by TIO patients include widespread and progressive bone pain, muscle weakness, gait disturbance, and shortened height (3). Biochemical features, such as hypophosphatemia due to renal phosphate wasting, low or inappropriately normal 1,25-dihydroxyvitamin D (1,25[OH]₂D) levels, and elevated alkaline phosphatase (ALP) and FGF23 are mainly involved. After successful localization and resection of tumor, most of TIO patients could recover from hypophosphatemia and its related symptoms (4).

Although it has been more than 60 years since the first case of TIO was described by McCance in 1947 (5), except for the findings of FN1-FGFR1 or FN1-FGF1 fusions which might partially explain the excessive production of FGF23 in nearly half of studied tumors (6, 7), the mechanism behind tumorigenesis and overproduction of FGF23 remains to be explored. In addition, despite the recovery of low serum phosphate after surgery, it is currently unknown whether metabolic alteration or recovery happens after resection in TIO. More importantly, the rates of missed and delayed diagnosis have been extremely high due to the poor recognition of the disease and difficult location of tumor using conventional imaging modalities. Thus, exploring effective early diagnostic method is urgent and necessary.

Metabolomics is an emerging “omics” method, which provides in-depth investigation of the global small molecules in various biological specimen (8). It is widely used in clarification of pathogenesis, diagnosis of disease, and prediction of prognosis (9, 10). However, it is still a developing field requiring further investigation in TIO. In the present study, the serum metabolome of newly diagnosed TIO patients and age-matched healthy controls were analyzed by ultra-performance liquid chromatography–tandem mass spectrometry (UPLC-MS/MS). Through the global profiling, metabolic alteration in disease state could be illustrated, and a reliable basis for disease detection, diagnosis even treatment strategy could be provided.

Methods

Patients and Controls

Patients who had been diagnosed as having TIO by histopathological confirmation of phosphaturic mesenchymal tumor after surgery at Peking Union Medical College Hospital (PUMCH) between January 1, 2018, and January 31, 2019, and who met the following criteria were enrolled in the study: (1) participants were not suffering from other metabolic, liver, or kidney diseases or any other tumors; (2) preoperative blood samples after overnight fasting were collected after withdrawal of any medication; and (3) postoperative blood samples after overnight fasting were collected within 1 to 3 days after successful surgery with the recovery of serum phosphate. Age- and sex-matched healthy controls (HC) without any metabolic disease and clinical conditions were recruited from PUMCH Outpatient Clinic Physical Examination Center between January 1, 2018, and January 31, 2019.

This study was approved by the Local Ethics Committee of the Department of Scientific Research at PUMCH. Informed consent was obtained from all of the participants before entering the study.

Serum Collection and Storage

Fasting whole blood samples of patients and controls were collected and placed at room temperature for 30 minutes and then centrifuged at 3000 revolutions/min for 10 minutes to separate the serum. Serum was transferred to polypropylene tubes and stored at −80 °C until assayed.

Biochemical Analyses

Serum calcium, phosphate, and alkaline phosphatase (ALP) were measured by an autoanalyzer (Beckman Coulter AU5800, America). Serum intact parathyroid hormone (iPTH) was measured by an autoanalyzer (Beckman Coulter DXI800, America). Measurement of 25-hydroxy vitamin D (25OHD) was performed using an automated Roche electrochemiluminescence system (E601, Roche Diagnostics, Switzerland). Serum 1,25-dihydroxy vitamin D (1,25[OH]₂D) was measured by radioimmunoassay/enzyme-linked immunoassay (DiaSorin, USA). Both fasting spot urine and full-day urine were collected and then analyzed by a urinary chemical analyzer (Clinitek 500, SIEMENS, USA) and urine flow cytometer (Sysmex UF-1000i, Sysmex, Japan). Reference ranges were obtained from the central laboratory of PUMCH. The renal tubular reabsorption of phosphate (TRP) was calculated using the following formula: %TRP = 100 × [1 − (urine phosphate × serum creatinine)/(serum phosphate × urine creatinine)]. The tubular maximum reabsorption threshold of phosphate per glomerular filtration rate (TmP/GFR) was estimated using the Walton and Bijvoet nomogram (11). Serum intact FGF23 (iFGF23) was measured by an iFGF23 ELISA Kit (Kainos Laboratories, Tokyo, Japan). The reference range for serum iFGF23 is 10 to 50 pg/mL (12).

Bone Mineral Density

Areal bone mineral density (aBMD) of lumbar spine and proximal femur were measured with a GE Lunar Prodigy Advance scanner (Prodigy Advance, GE Lunar Corporation, USA). The aBMD values were standardized as Z-scores according to the operating procedures of the manufacturer for patient samples.

Sample Preparation for UPLC-MS/MS

Serum (100 μL) was added into 400 μL methanol (prechilled to −20 °C) to make a final 80% (v/v) methanol solution. After incubation at −20 °C for 2 to 4 hours, all sample solutions were centrifuged at 12 000 × g for 20 minutes. All supernatants were dried and redissolved in 100 μL of 50% (v/v) methanol solution for UPLC-MS/MS analysis. Equal volumes of all the samples were pooled and used as the quality control (QC) for UPLC-MS/MS optimizing and normalizing. 6 QC runs were performed prior to the analysis of all samples until system equilibration was achieved. QC samples were also analyzed after every 10 randomized serum samples.

UPLC-MS/MS Analytical Procedure

The sample analysis was carried out on ExionLC (SCIEX, USA) coupled with a TripleTOF 5600⁺(SCIEX, USA) in both positive and negative ionization modes. The column was a ACQUITY UPLC HSS T3 column (1.8 μm, 2.1 × 100 mm, Waters Corp., USA). The mobile phase A was water with 0.1% formic acid and phase B was acetonitrile with 0.1% formic acid. The total analysis time lasted 20 minutes. Chromatography separation was conducted as following gradient conditions: from 0 to 1.5 minutes, 5%-20% mobile phase B; from 1.5 to 15 minutes, 20%-90% B; 15 to 18 minutes, 90% -100% B; 18 to 20 minutes, held at 5% B for re-equilibration.

For metabolomic profiling, a time-of-flight (Q-TOF) mass spectrometry was connected to the UPLC system via electrospray ionization (ESI) interface by using information dependent acquisition (IDA) high-sensitivity scanning mode with the following parameters: temperature was 600 °C for the positive mode and 550 °C for the negative mode; and the ion spray voltage floating was 5.5 kV for the positive mode and the negative was 4.5 kv. MS data were acquired from m/z 50-1200 Da for MS/MS scan. The MS/MS accumulation time was 0.05 seconds. The collision energy was 40, and the collision energy range was the theoretical frequency ± 20.

UPLC-MS Data Processing and Analysis

The raw data acquired from UPLC-MS/MS were processed by Progenesis QI software (Waters Corp., USA) to perform peak detection, alignment, and normalization. EZinfo Ver. 3.0 software (Waters Corp., USA) was used for pattern recognition and multivariate statistical analysis. The matrix was further reduced by removing peaks with missing values in more than 80% of samples and those with isotope ions from each group to obtain consistent variables. The coefficient of variation of metabolites was set at a threshold of 30%, as a standard in the assessment of repeatability in metabolomics data sets. The results were visualized in the form of score plot to show the group clusters, and the S-plots and variable importance in the projection (VIP) plots based on orthogonal partial least squares discriminant analysis (OPLS-DA) analysis were to identify and reveal differential metabolites accountable for the separation between identified groups. In addition, heatmap and metabolic pathway enrichment was achieved through MetaboAnalyst 5.0 software (http://www.metaboanalyst.ca/) and IMPaLA (http://impala.molgen.mpg.de/impala/). Metabolites for further statistical analysis were identified on the basis of VIP threshold of 1 from the OPLS-DA model, which was validated at a univariate level with adjusted P < 0.05. The molecules of interests were compared and identified using the HMDB database (http://hmdb.ca). In addition, heatmap and metabolic pathway enrichment was achieved through MetaboAnalyst 5.0 software (http://www.metaboanalyst.ca/) and IMPaLA (http://impala.molgen.mpg.de/impala/).

Statistical Analysis

For clinical parameters, statistical analyses were using IBM SPSS Statistics version 17.0 software (IBM, Armonk, NY, USA). Continuous variables were presented as mean ± SD or median (interquartile range) as appropriate, whereas categorical variables were presented as frequency (percentage). Differences between patients and controls were analyzed using independent samples T test for normally distributed continuous data, Mann-Whitney U test for nonparametric continuous data, and χ2 test for categorical variables. Paired T test or Wilcoxon signed-rank tests were used to compare variables before and after surgery.

For metabolic profiling, orthogonal partial least squares discriminant analysis (OPLS-DA) was performed on the EZinfo Ver. 3.0 software. Benjamini-Hochberg–based false discovery rate (FDR) was performed for multiple testing. For pathway enrichment, Q value was calculated by IMPaLA that adjusted P value using FDR. Classic univariate receiver operating characteristic (ROC) and multivariate exploratory ROC analysis were performed to identify and validate candidate biomarkers, achieved through MetaboAnalyst 5.0. The ROC curves are generated by Monte-Carlo cross validation using balanced sub-sampling. The data matrix was autoscaled and random forest (RF) was used for the classification method and univariate area under the ROC was used for feature ranking. Validations of selected potential biomarkers were performed by using partial least squares (PLS) model, supported vector machine (SVM) model. The area under the ROC curve (AUC) value and 95% CI and average accuracy based on 100 cross validations were calculated.

Results

Clinical and Biochemical Characteristics of Enrolled Subjects

Baseline characteristics of enrolled TIO patients and healthy controls are described in Table 1. This study included 24 male (75%) and 8 female patients (25%), with a mean age of 43.6 years and a median disease duration of 42 months. Most patients underwent height loss for a median of 4.0 cm, resulting in slightly short stature with a median Z score of height of −0.9. Laboratory investigation revealed hypophosphatemia with low TmP/GFR and elevated iFGF23, elevated serum ALP, and low 1,25 (OH)₂D and 25OHD levels. The aBMD was extensively decreased as Z-scores of femoral neck, total hip, and lumbar spine were −2.5 ± 1.2, −2.8 ± 1.6, and −2.0 ± 1.6, respectively. Healthy controls enrolled in this study had normal serum phosphate, calcium, and ALP. Age, sex ratio, and body mass index were not different between patients and controls, while patients were shorter than controls. After surgery, all patients recovered from phosphatemia with significant reduction of iFGF23. Both 1,25(OH)₂D and 25OHD levels were also elevated after surgery. In addition, there was a significant decrease in the elevated ALP.

Global Metabolomic Profiling of TIO

Enrolled participants included 32 diagnosed TIO patients and 32 age- and sex-matched healthy controls who underwent metabolomic profiling. Serum of TIO patients was collected at 2 time points: once before and once after tumor resection. Using the LC-MS/MS approach, we analyzed the global metabolomic profiles of 3 groups, including newly diagnosed TIO patients, postoperative TIO, and healthy controls (corresponding to TIO, TIO post-OP, and HC). The MS raw data of 3 groups were imported into Progenesis QI software for data processing and database searching. Statistical analyses were further performed with EZinfo and MetaboAnalyst 5.0.

OPLS-DA and sPLS-DA (sparse partial least squares discriminant analysis) were utilized to characterize the differential profiling between the HC, TIO, and TIO post-OP groups. As shown in Supplementary Fig. S1A and S1B (13), the OPLS-DA in ESI+ mode between HC and TIO showed cumulative values of R2 (Y) = 94% and Q2 = 86%. ESI− data showed R2 (Y) = 99%, Q2 = 83%. These results indicated that the OPLS-DA models were not overfitting and had high separating capacity. A similar plot pattern was observed between the HC and TIO post-OP groups (Supplementary Fig. S1C and S1D) (13). However, the TIO and TIO post-OP groups could not be discriminated (Supplementary Fig. S1E and S1F) (13). The sPLS-DA analysis was conducted to validate and provide complementary evaluation of the discriminating model, as shown in Fig. 1. In both ion modes, the HC group distributed separately from the TIO and TIO post-OP groups, but the TIO post-OP cluster (blue circle) was overlapped with TIO (green circle). The multivariate analyses suggested that distinct differences were appeared between the disease group and healthy controls, but tumor resection had little impact on global metabolism compared with disease state.

The metabolites that made significant contributions to the difference between HC and TIO were identified by S-plot of the OPLS-DA analysis. The features, with variable importance in projection score (VIP) > 1, were screened to conduct hierarchical clustering. The top 50 features were plotted (positive mode shown in Supplementary Fig. S2A, and negative mode shown in Supplementary Fig. S2B) (13). Compared with the initial diagnosis TIO group, the features in HC were significantly different, while changes in TIO post-OP were inconspicuous. This analysis revealed similar results to the multivariate analyses.

By using a mass-based search procedure, the significant features were identified. A total of 180 molecules in ESI+ mode and 314 in ESI− mode were tentatively identified from the metabolomics data as differentially expressed metabolites (DEMs). The identified molecules were subjected to IMPaLA for pathway enrichment. As shown in Fig. 2 and Table 2, there were multiple metabolic pathways perturbed in the disease state (only the top 10 most significantly enriched are listed), mainly including arachidonic acid metabolism, fatty acid and lipid metabolism, as well as sphingosine and sphingosine 1 phosphate metabolism.

Candidate Biomarkers Identification for TIO

After systematically defining the global metabolomic profiles and pathway associated with TIO, we aimed to identify candidate biomarkers capable of classifying TIO vs HC. Herein, a stringent threshold was applied to identify the most discriminatory DEMs. A set of 23 metabolites were screened and tentatively identified as potential biomarkers (FDR < 0.05, fold change > 1.5, and score > 40), as listed in Supplementary Table S1 (13). The individual discriminating performance of these potential biomarkers was further supported by classical univariate receiver operating characteristic (ROC) analysis. The top 15 most relevant prognostic markers ranked based on area under ROC curve were listed in Supplementary Table S2 (13), and all features had AUC > 0.80. Five oxylipins, 4-hydroxydocosahexaenoic acid (4-HDoHE), leukotriene B4 (LTB4), 5-hydroxyeicosatetraenoic acid (5-HETE), 17-hydroxyeicosatetraenoic acid (17-HETE) and 9,10,13-trihydroxy-octadecenoic acid (9,10,13-TriHOME), were the top ranked metabolites to discriminate patients and healthy controls, indicating high diagnostic capability as TIO biomarkers (Table 3).

To optimize the diagnostic panel model, multivariate exploratory ROC analysis was further implemented based on a random forest (RF) algorithm. The ROC curves established by using a variable number of features showed excellent performance, as displayed in Fig. 3A. The AUC score ranged from 0.943 to 0.996, and the AUC gradually increased along with the number of variables. Considering the predicted accuracy, it achieved the excellent classification accuracy of 94% using the top 5 features, and accuracy slightly declined upon adding more metabolites (Fig. 3B). While the highest accuracy was achieved when using all 23 potential biomarkers, the 5-features model performed on par with the 23-features predictive model. Therefore, the combination of the top 5 metabolites constructed a proper model with an AUC value of 0.951 (95% CI, 0.827-1) (Fig. 3C). The probabilities of prediction were displayed in Fig. 3D. As shown in Fig. 3E and 3F, based on the average importance and selected frequency, ROC analysis sorted the most relevant biomarkers, and the result was similar with that of classic univariate ROC analysis. 4-HDoHE, LTB4, 5-HETE, 17-HETE, and 9,10,13-TriHOME were ranked top in average importance of the RF model. To further evaluate the predictive performance of the diagnostic panel, 2 other algorithms were applied: a supported vector machine (SVM) model and partial least squares (PLS) model, as displayed in Supplementary Fig. S3 (13). The AUC value and 95% CI and average accuracy based on 100 cross validations were calculated (Supplementary Table S3) (13). The AUC values were 0.948 (95% CI, 0.835-1) and 0.937 (95% CI, 0.785-1), for SVM and PLS, respectively. Additionally, we examine the precision of these oxylipins, which were all below 20% and considered acceptable (Supplementary Table S4). These results demonstrated that the diagnostic panel combining with the 5 oxylipins exhibited promising discriminative capabilities for TIO and healthy controls. We also tried to explore the correlation between the candidate biomarkers and biochemical parameters in TIO. However, no statistically significant correlation was found between them (Supplementary Table S5).

Taken together, 4-HDoHE, LTB4, 5-HETE, 17-HETE, and 9,10,13-TriHOME might serve as potential biomarkers for TIO diagnosis. The expression level of these oxylipins is summarized in Fig. 4. Compared with the HC groups, all these 5 metabolites were highly expressed in TIO. After tumor resection, the expression of these biomarkers tended to decrease back toward the HC levels. The differences of 5-HETE and 17-HETE were statistically significant (P < 0.05). Expression heatmap of other biomarkers was shown in Supplementary Fig. S4 (13), and the significance was illustrated.

Discussion

Although TIO is the most common cause of acquired hypophosphatemic rickets/osteomalacia, there is a lack of knowledge about TIO. In light of the high rates of misdiagnosis and delayed diagnosis in TIO, establishment of serum metabolic biomarkers measurable by high-throughput methods for diagnosis and investigation of the pathogenesis mechanism are of great significance. Thus, for the first time, we provide the global profile of metabolomics of TIO.

Our results showed that the metabolome of TIO was unique compared with that observed in HC. We identified several metabolites significantly related to the disease. Pathway enrichment analysis of these DEMs illustrated that arachidonic acid metabolism was most relevant with TIO. By random forest model, 5 oxylipins can be considered as potential biomarkers, namely 4-HDoHE, LTB4, 5-HETE, 17-HETE, and 9,10,13-TriHOME. Although evident elevation of FGF23 could highly suggest the diagnosis of TIO (14), we also provided a potential diagnostic panel according to ROC analysis using a combination of the 5 oxylipins, as a complementation to FGF23.

Oxylipins are oxidation products derived from polyunsaturated fatty acids (PUFAs) precursors. The n-3 PUFAs include α-linolenic acid (ALA), eicosapentaenoic acid (EPA), and docosahexaenoic acid (DHA), and the n-6 PUFAs include arachidonic acid (AA), linoleic acid (LA) and γ-linolenic acid (GLA) (15). They are efficient mediators in diverse biological processes, including inflammation (16, 17), immune response (18, 19), vasoconstriction (20), pain (21), and tumorigenesis (22, 23). Although a matter of some controversy, it has been suggested that oxylipins from n-6 PUFAs generally have pro-inflammatory and proliferative effects compared with those derived from n-3 PUFAs (15). In this study, we found 5 upregulated oxylipins that could serve as potential biomarkers. Four of them were n-6 PUFAs derivatives, including metabolites of AA (LTB4, 5-HETE, and 17-HETE) and LA (9,10,13-TriHOME). Simultaneously, the enrichment of DEMs suggests the dysregulation of AA metabolism pathway in TIO.

AA, as a major component of the cell membrane phospholipid content, is abundant in various tissues. It could be converted into various effective autocrine and paracrine bioactive mediators through 3 enzymatic pathways, including cyclooxygenase (COX), lipoxygenase (LOX), and cytochrome P450 (CYP450). The liberation and stepwise metabolization of AA could simulate numerous signaling responses, such as inflammation through Ca²⁺-activated phospholipase A₂ (24, 25) and pain sensory signals through TRPV1 activation (21, 26). This renders the AA metabolism pathway a valid target for drug development and therapeutic approaches. Three out of 5 potential biomarkers in our predictive panel are oxylipins derived from AA, including LTB4, 5-HETE, and 17-HETE. 5-HETE is one of the lipoxygenase products generated by AA and can be further converted to LTB4. Both 5-HETE and LTB4 has been reported to play essential roles in chronic inflammation (15, 27), which is a significant risk factor for tumor development (28). In addition, LTB4 and 5-HETE can stimulate proliferation and suppress apoptosis in several types of cells (29-32). These conclusions suggest that inflammation and cell proliferation modulated by LTB4 and 5-HETE may be similarly involved in TIO tumorigenesis. Furthermore, high vascularization is one of the histological features found in TIO. Previous studies confirmed that LTB4 induced VEGF-mediated angiogenesis via its receptor (BLT2) in vivo (33, 34). Upregulation of LTB4 might partially explain the angiogenesis that exists in tumors of TIO.

LA, another member of the n-6 PUFAs, is a precursor of AA. Despite the lower concern and attention paid to LA compared with AA, LA and LA-derived oxylipins have important biological effects as well. LA has been considered as a “cardiovascular-friendly” essential dietary fatty acid to decrease the risk of hypertension (35). However, LA-derived oxylipins has shown positive correlation with cancers (22, 36). The experiments indicated that dietary LA could exacerbate colorectal cancer in the animal model (37). The accumulation of LA-derived oxylipin in TIO might be attributed to the development of TIO.

Surprisingly, the DHA derivative 4-HDoHE was accumulated in TIO patients. Metabolites of DHA were demonstrated to inhibit angiogenesis, tumor growth, and metastasis (38). Numerous case-control studies revealed a positive inverse association between DHA consumption and reduced cancer risk (39-41). As a result, supplementation of DHA has been advocated to be an adjuvant to immunotherapy or chemotherapy (42). In contrast to the expected, 4-HDoHE levels were found to be increased compared with controls, which might be possibly result from feedback regulation.

Although these results are suggestive of a link between the dysregulation of the AA pathway and TIO pathogenesis, the possible influence of inflammation cannot be ruled out. On the one hand, common fracture in TIO due to osteomalacia could initiate an acute inflammatory response (43). In this case, whether AA metabolism is dysregulated in patients with X-linked hypophosphatemia and Fanconi syndrome, who also share the characteristics of osteomalacia, remains to be explored. On the other hand, chronic inflammation is one of the common features of tumor (14), which might contribute to the disorder of oxylipins. Unobserved correlation between oxylipins and biochemical parameters of TIO suggested that inflammation might be a potential confounding factor.

The generalizability of these results is subject to certain limitations. Firstly, due to the rarity of other FGF23-related hypophosphatemia (e.g., X-linked hypophosphatemia) and FGF23-unrelated hypophosphatemia (e.g., Fanconi syndrome), we failed to compare the oxylipins levels in those conditions with those in TIO, despite its importance to elucidate the role of oxylipins. Secondly, only short-term postoperative (1-3 days) serum were collected for comparison of metabolomics with that before surgery. Considering the returning trend of 4-HDoHE, LTB4, and 9,10,13-TriHOME, long-term follow-up is needed to observe the changes of metabolomics, even until bone mineral density returns to a normal level after surgery. Thirdly, sample size is insufficient since TIO is a rare disease. Thus, external replication and larger sample-sized studies are required to validate the capability of biomarkers. In addition, future work is needed to focus on targeting the oxylipins to increase the confidence in compound identification.

In conclusion, our study presented in this work provides the first global metabolomics analysis in TIO. UPLC-MS/MS based metabolomics of serum obtained from TIO patients reveals distinct compositional differences compared with control subjects. Our study suggests that the arachidonic acid metabolism pathway was significantly disturbed in TIO. The identification of a discriminatory panel consisting of 5 oxylipins holds promise for early diagnosis as a complement to the measurement of serum FGF23 and phosphate. Despite its exploratory nature and limitations, this study provides novel insights into TIO diagnosis and pathogenesis.

Abbreviations

Abbreviations
 
• 1,25(OH)2D

  1,25-dihydroxyvitamin D (active vitamin D)

 
• 17-HETE

  17-hydroxyeicosatetraenoic acid

 
• 25OHD

  25-hydroxy vitamin D

 
• 4-HDoHE

  4-hydroxydocosahexaenoic acid

 
• 5-HETE

  5-hydroxyeicosatetraenoic acid

 
• 9,10,13-TriHOME

  9,10,13-trihydroxy-octadecenoic acid

 
• AA

  arachidonic acid

 
• aBMD

  areal bone mineral density

 
• ALP

  alkaline phosphatase

 
• AUC

  area under the ROC curve

 
• DEM

  differentially expressed molecules

 
• DHA

  docosahexaenoic acid

 
• ESI

  electrospray ionization

 
• FDR

  false discovery rate

 
• FGF23

  fibroblast growth factor 23

 
• GFR

  glomerular filtration rate

 
• HC

  healthy controls

 
• iFGF23

  intact fibroblast growth factor 23

 
• LA

  linoleic acid

 
• LTB4

  leukotriene B4

 
• OPLS-DA

  orthogonal partial least-squares discriminant analysis

 
• PLS

  partial least squares

 
• PUFA

  polyunsaturated fatty acid

 
• PUMCH

  Peking Union Medical College Hospital

 
• QC

  quality control

 
• ROC

  receiver operating characteristic

 
• sPLS-DA

  sparse partial least squares discriminant analysis

 
• SVM

  supported vector machine

 
• TIO

  tumor-induced osteomalacia

 
• TmP

  tubular maximum reabsorption threshold of phosphate

 
• TRP

  tubular reabsorption of phosphate

 
• UPLC-MS/MS

  ultra-performance liquid chromatography-tandem mass spectrometry

 
• 9,10,13-TriHOME

  9,10,13-trihydroxy-octadecenoic acid

Acknowledgments

We thank the patients and healthy volunteers for their participation in this study and their blood donation.

Financial Support

This work was supported by grant from Beijing Municipal Natural Science Foundation (7214246), the National Key R&D Program of China (2018YFA0800801), the National Natural Science Foundation of China (81970757).

Disclosures

All authors state that they have no conflicts of interest.

Data Availability

Some or all data generated or analyzed during this study are included in this published article or in the data repositories listed in References.

References

Author notes