Simultaneous Quantification of Serum Nonesterified and Esterified Fatty Acids as Potential Biomarkers to Differentiate Benign Lung Diseases from Lung Cancer

In this study, we have employed graphene oxide as a matrix to simultaneously and directly quantify serum nonesterified and esterified fatty acids (FAs) using matrix-assisted laser/desorption ionization-Fourier transform ion cyclotron resonance mass spectrometry (MALDI-FTICR MS). Twelve serum nonesterified FAs combined with their individual esterified FAs (i.e., C16:0, C16:1, C18:0, C18:1, C18:2, C18:3, C20:2, C20:3, C20:4, C20:5, C22:5, and C22:6) were quantified based on their calibration curves with the correlation coefficients of >0.99, along with the analytical time of <1 min each sample. As a result, serum levels of twelve total FAs (TFAs) in 1440 serum samples from 487 healthy controls (HCs), 479 patients with benign lung diseases (BLDs) and 474 patients with lung cancer (LC) were determined. Statistical analysis indicated that significantly increased levels of C16:0, C16:1, C18:0, C18:1, C18:3, C20:3, and C22:6 and decreased levels of C20:5 were observed in LC patients compared with BLDs. Receiver operating characteristic (ROC) analysis revealed that panel a (C18:2, C20:3, C20:4, C20:5, C22:5, and C22:6), panel b (C18:0, C20:4, C20:5, and C22:6), and panel c (C16:1, C18:0, C18:1, C20:3, and C22:6) have exhibited good diagnostic ability to differentiate BLDs from LC relative to clinical uses of tumor markers (CEA and Cyfra 21-1).


Results
Generating serum TFAs using optimal MS condition. To establish an optimal MS condition for fragmenting phospholipids and generating serum TFAs using the QC sample as a model sample, we first optimized two important parameters: MALDI laser power and the skimmer 1 voltage of FTICR MS using GO as a MALDI matrix 31 . When the laser power was set to 30%, 50%, or 90% at both the skimmer 1 voltage of − 10 V and− 45 V, respectively, the intensities of different serum phospholipids were decreased gradually, while those of serum TFAs were increased (Fig. 1A-D). It should be noted that when the skimmer 1 voltage was set up to − 100 V, the intensities of serum TFAs were dramatically decreased and phospholipids almost disappeared (Fig. 1E). Finally, it is found that most of the serum phospholipids were fragmented and abundant serum TFAs were generated with laser power of 90% and the skimmer 1 voltage of − 45 V (Fig. 1D). As a result, the MS condition was optimal as laser power of 90% and the skimmer 1 voltage of − 45 V.
To assess the effect of different MALDI matrixes on the degree of the fragmentation of different phospholipids, several standard phospholipids (i.e., PA(18:1/18:1), PE(18:1/18:1), PC(18:1/18:1), PI(16:0/18:2) and PG(16:0/16:0)) were selected as model compounds. As shown in Fig. 2, a comparison of commonly used matrixes (i.e., 9-AA, DMAN, and NEDC) and GO shows that GO matrix has exhibited the best performance to fragment all of the above-mentioned standard phospholipids, along with the fragmentation degree of > 80%, while NEDC matrix shows the lowest degree of fragmentation for all above-mentioned standard phospholipids compared with other matrixes. Therefore, GO was selected as the MALDI matrix to analyze serum extracts.
Simultaneous quantification of serum TFAs. Based on the working standard solutions, the calibration curves of C 16:0 , C 16:1 , C 18:0 , C 18:1 , C 18:2 , C 18:3 , C 20:4 , and C 22:6 were constructed with their correlation coefficients of > 0.99 (Table 1), and their linearity ranges, LODs, and spike-and-recovery are also listed in Table 1. The reproducibility of the eight FAs is less than 18.0% (Table 1). Their intraday RSDs were from 6.2% to 10.7% and their interday RSDs were from 9.5% to 15.1% (Supplementary Table S1). The spike-and-recovery in triplicate at three different concentrations of all eight FAs were between 78.8% and 110.0%.
Representative mass spectra of serum TFAs from one HC, one BLD patient, and one LC patient are shown in Fig. 3. Serum TFAs quantified in this study were identified based on their observed accurate m/z values relative to the theoretical values with a mass error of < ± 0.00025 Da and reliable isotopic distributions with the RSDs of < 2.0% between the observed and theoretical intensities (Supplementary Table S2). The levels of serum TFAs were calculated based on their respective calibration curves (Table 1) and the resulting data are shown in Fig. 4.

Associations of the levels of serum TFAs with gender and age. Comparison of the levels of serum
TFAs between female and male in each physiopathological state (For HCs, females: n = 212, age: 47.3 ± 10.3 years old; males: n = 275, age: 48.5 ± 10.1 years old. For BLD patients, females: n = 212, age: 55.7 ± 9.0 years old; males: n = 234, age: 55.2 ± 9.5 years old. For LC patients, females: n = 238, age: 57.5 ± 8.4 years old; males: n = 236, age: 57.9 ± 8.3 years old) and all participants (females: n = 745, age: 53.4 ± 10.2 years old; males: n = 695, age: 53.7 ± 10.2 years old) was performed using Mann-Whitney U test. The statistical analysis indicated that there is no statistical significance between the levels of serum TFAs and gender in each physiopathological state and between three different states (p > 0.05, Supplementary Table S3).
The effect of age on the levels of serum TFAs for HCs was also analyzed based on four different age groups (i.e., group 1, 30~39 years old (n = 114); group 2, 40~49 years old (n = 157); group 3, 50~59 years old (n = 138), and group 4, 60~70 years old (n = 78)) using one-way ANOVA with LSD test after data were transformed to normal

Association of changes in the levels of serum TFAs with physiopathological states. Based on
the above-mentioned results, the 914 age-matched participants, including 304 HCs, 311 BLD patients, and 299 LC patients were selected to screen biomarkers for differentiating different physiopathological states ( Table 2). In order to obtain high accurate diagnostic biomarkers, these participants were further classified randomly into a training set and a validation set (Fig. 5). In the training set study, as shown in Fig. 4, the levels of C 18:2 , C 20:2 , C 20:3 , C 20:4 , C 20:5 , C 22:5 , and C 22:6 in BLD patients were significantly decreased relative to HCs. Significant increase in the levels of C 16:0 , C 18:0 , C 18:1 , and C 18:3 and decrease in the levels of C 20:2 , C 20:4 , C 20:5 , C 22:5 , and C 22:6 in LC patients were observed compared with HCs. However, remarkable increase in the levels of C 16:0 , C 16:1 , C 18:0 , C 18:1 , and C 18:3 and decrease in the levels of C 20:3 and C 22:6 in LC patients were detected compared with BLD patients. In addition, an independent validation set also proved the above-mentioned change trends of serum TFAs in different physiological states (Fig. 4), and all p values are listed in Supplementary Table S7.
Diagnostic ability of serum TFAs. The AUC values, sensitivities, specificities, and cut-off values of serum TFAs panels are listed in Table 3. For the training set, a combination of C 18:2 , C 20:3 , C 20:4 , C 20:5 , C 22:5 , and C 22:6 , namely panel a, has shown a powerful capability to differentiate HCs from BLD patients, with the AUC value of 0.863. A combination of C 18:0 , C 20:4 , C 20:5 , and C 22:6 , namely panel b, has a powerful ability to differentiate HCs from LC patients, with the AUC value of 0.729. A combination of C 16:1 , C 18:0 , C 18:1 , C 20:3 , and C 22:6 , namely panel c, is a good predictor for distinguishing BLD from LC patients, with the AUC value of 0.752. To validate the diagnostic ability of the above-mentioned panels, an independent validation study was performed. As shown in Table 3, the panels a, b, and c all have good capability to differentiate between HCs, BLDs, and LC based on the cut-off values obtained in the training set, with the AUC values of 0.781, 0.759, and 0.703, respectively. In addition, based on these cut-off values, each of three panels has shown a good capability to differentiate HC from BLDs plus LC, with the AUC values of > 0.74 (Table 4), and it should be noted that the AUC values of three individual panels to distinguish HC plus BLDs from LC were still more than 0.64 ( Table 4).

Levels of serum tumor markers.
In this study, serum tumor markers, CEA and Cyfra 21-1, were also measured in accordance with the manufacturer's instructions and their median and ranges are listed Table 5. It is found that both CEA and Cyfra 21-1 were significantly increased in BLD or LC patients relative to HC (p < 0.001). However, no statistical differences in the level of serum Cyfra 21-1 was observed between BLD patients and LC patients.

Comparison of diagnostic ability between serum tumor markers and serum TFA panels.
Based on the cut-off values of serum CEA of 5.0 ng/mL, Cyfra 21-1 of 3.5 ng/mL and a combination of CEA and Cyfra 21-1 of 0.5, their AUC values were calculated to differentiate HC from BLDs, HC from LC or BLDs from LC. As shown in Table 6, CEA and Cyfra 21-1 present a similar AUC values as the serum TFA panels to differentiate HCs from BLDs patients or LC patients, along with high specificities and low sensitivities for CEA and Cyfra 21-1 compared with serum TFA panels. It is worth noting that serum TFA panels with the AUC values of 0.706~0.732 have shown high diagnostic accuracy to differentiate BLDs from LC relative to serum tumor markers with the   (Table 7), while serum TFA panels have exhibited a slightly better diagnostic ability to differentiate HC plus BLDs from LC compared with serum tumor markers (Table 7).

Discussion
The detection of serum TFAs is usually performed using gas chromatography-mass spectrometry, along with a complicated and time-consuming sample preparation: lipid extraction, hydrolysis, and methylation 32 . In the present study, we developed a highly efficient method to generate esterified FAs and to detect directly and simultaneously serum nonesterified and esterified FAs through optimizing the skimmer 1 voltage of MS in combination with GO as a MALDI matrix. The calibration equations of C 16:0 , C 16:1 , C 18:0 , C 18:1 , C 18:2 , C 18:3 , C 20:4 , and C 22:6 were constructed with the correlation coefficients of > 0.99 based on their mixture standard working solutions. Their LODs were between 0.1 μ M and 2.2 μ M. 144 mass spectra of the QC sample were analyzed with a relative standard deviation of 18% for all analytes. The spike-and-recovery experiments on the basis of three different concentrations of FAs indicate that their recoveries were in the range from 78.8% to 110.0%. Our data indicate that the stability and precision of this method are acceptable for complex biological sample analysis. In term of the established method, twelve TFAs from 1440 serum samples were rapidly and simultaneously quantified and the measured concentrations of serum TFAs in the present study are similar to the previously published results obtained using gas chromatograph-mass spectrometry 13,33 . Statistic analysis indicated that gender-specific difference was observed neither in the levels of twelve serum TFAs in each physiopathological state nor between three physiopathological states (Supplementary Table S3).    It is worth noting that age-specific differences were only detected in the levels of some TFAs including C 18:2 , C 20:2 , C 20:3 , C 20:4 , C 20:5 , and C 22:6 in HCs and C 18:3 in BLD patients (Supplementary Tables S4 and S5), while for LC patients, no age-specific differences was observed, which are in line with previous reports 19,34 . Our findings indicate that different metabolic mechanisms between HC, BLDs, and LC might exist. Based on the age-matched samples (Table 2), statistic analysis indicated that the changes in the levels of serum TFAs between three different physiological states still present significantly statistical differences (Supplementary Table S7), further indicating the different metabolic mechanisms of serum TFAs between three physiopathological states.
Previous studies indicate that FA synthase has been found to be a hyperactivity in many cancers including LC 35,36 . Increase in the levels of C 16:0 , C 18:0 , and C 18:1 in LC patients may be ascribed to the overexpression of FA synthase to sustain the increasing demand of energy during cancer cell proliferation 37 . It is found that α -linolenic       These results further prove that lipid metabolism in HC and BLDs is significantly different from that in LC. The human body can produce all FAs except for C 18:2 n-6 and C 18:3 n-3 45 . C 18:2 n-6, which affects gene expression 46 , is the precursor of n-6 series of FAs. Lipid mediators (e.g., prostaglandins, thromboxanes, and leukotrienes) play proinflammatory and angiogenic functions and involve in several pathologic progresses, which are generated primarily through oxidative pathways from C 20:4 28 . In the present study, decreased C 18:2 in BLDs and decreased C 20:4 in both BLDs and LC might be associated with changes in inflammatory response. In addition, the combinations of serum TFAs have exhibited powerful diagnostic ability to differentiate BLDs from LC with the AUC values of 0.706~0.732 relative to serum tumor markers with the AUC values of 0.521~0.588 (Tables 6 and 7), indicating that lipid mechanisms are closely correlated with different physiological states.
Our study has some limitations. First, due to the lack of the detailed clinical information on LC staging, early stage diagnostic ability of serum TFAs could not be preformed. Second, the location of the double bonds of each unsaturated FAs has not been designed due to no gas chromatography separation and corresponding standard compounds. Finally, non complete fragmentation of serum phospholipids may affect the diagnostic capability of serum TFAs.

Conclusions
In the present study, based on the optimized parameters of MS in combination with the physicochemical properties of GO as a MALDI matrix, we developed a rapid and simultaneous quantification method of 12 serum FAs including nonesterified and esterified FAs via the direct fragmentation of phospholipids using MALDI-MS without hydrolysis and methylation of FAs. The correlation coefficients of the calibration curves were large than 0.99, with a linear dynamic range of 3 orders of magnitude and the LODs of 0.1~2.2 μ M. The levels of serum TFAs in 1440 serum samples from three different physiopathological states reveal that lipid metabolic mechanisms are closely correlated with the physiopathological states. ROC analysis indicated that three different TFA panels have exhibited good diagnostic capability to differentiate among three different physiopathological states with the AUC values of > 0.7 compared with serum tumor markers. Especially for differentiating BLDs from LC, the AUC values of serum TFA panels are 0.706~0.732, while those of serum tumor markers are 0.521~0.588. Taken together, our findings indicate that lipid metabolism is deeply involved in changes in physiological state, and our data may offer a stepping stone of new biomarker panels for differentiating BLDs from LC.     Participants and study design. In this study, a total of 1440 overnight (more than 10 hours) fasting serum samples were collected in Peking Union Medical College Hospital (Beijing, China). Serum samples are the remaining sera after routine physical examination or clinical examination. All samples were stored under − 80 °C until use. Healthy controls (HCs) without any aberrant clinical appearance and pathomorphology were included. BLDs were diagnosed based on the clinical diagnostic criterion and LC was further confirmed by cytological or histological examination of tumor tissue. Study design is shown in Fig. 5. Age-unmatched participants (i.e., 183 HCs, 168 BLD patients, and 175 LC patients ) were excluded, and the clinical characteristics of the age-matched participants are shown in Table 2 (1) Statistical analysis. Mass spectral data were obtained using ApexControl 3.0.0 (Bruker Daltonics). After isotopic deconvolution, the resulting data were transferred to Microsoft Excel, and the half of baseline strength in each spectrum was adopted as their intensities of missing serum TFAs. Univariate analysis was performed using non-parametric Mann-Whitney U test. One-way analysis of variance (ANOVA) with Fisher's least significant test was used to evaluate the effect of age on the levels of serum TFAs. Non-normally distributed data were transformed into normal distribution before statistical analysis. Receiver operating characteristics (ROC) curve analysis was used to calculate the area under the ROC curve (AUC), cut-off values, sensitivities, and specificities. The prediction model was further confirmed by an independent validation set based on the cut-off values obtained in the training set. Statistical analyses were performed using SPSS software (version 16.0, Chicago, IL, USA). In all cases, a p value less than 0.05 was considered to be statistically significant.

Methods
Identification of serum TFAs. Serum TFAs were identified as previously described 19 . Briefly, serum TFAs were identified based on their observed accurate m/z values relative to theoretical values of < ± 0.00025 Da and their observed distributions of isotopic abundance relative to theoretical distributions of < 2.0%.

Method validation for quantitative analysis. The reliability of MALDI-MS for quantitative analysis of
FAs was validated based on the linearity, limit of detection (LOD), stability, precision, and spike-and-recovery experiments.
To construct the calibration curves of each FA, the mixed stock standard solution was diluted to four different concentrations (i.e., 3.3, 6.6, 16.5, and 82.5-fold), respectively, and finally, five working standard solutions were obtained. The calibration curves between the intensity ratios of individual FAs to ISs (the final concentrations of 37.5 μ M C 17:0 and 7.5 μ M C 21:0 ) versus their corresponding concentration ratios were constructed based on the above-mentioned working standard solutions. C 17:0 as an IS was for quantifying the levels of C 16:0 , C 16:1 , C 18:0 , C 18:1 , C 18:2 , and C 18:3 , and C 21:0 as an IS was for quantifying the levels of C 20:2 , C 20:3 , C 20:4 , C 20:5 , C 22:5 , and C 22:6 . In addition, the calibration curve of C 20:4 was also used for quantifying C 20:2 , C 20:3 , and C 20:5 and the calibration curve of C 22:6 was for quantifying C 22:5 because their commercial standards are not available. Each of the working solution was analyzed three times and the results were shown as mean ± standard deviation. The LOD is defined as the concentration of each analytes at the signal-to-noise ratio of 3.
A pooled quality control (QC) serum sample obtained from the mixture of 5 HCs and 5 LC patients sera was analyzed once every 10 test samples. Finally, a total of 144 spectra of the QC sample were obtained in this study. The experimental stability and reproducibility were evaluated using the relative standard deviation (RSD) obtained based on the intensity ratios of the detected TFAs relative to their corresponding ISs. The experimental precision was assessed based on the intraday precision of three measured values of the QC sample on the same day and the interday precision of three measured values of the QC sample on three consecutive days.
To assess the extraction efficiency of serum phospholipids and FAs, the spike-and-recovery experiment was employed. Briefly, 10 μ L of the QC serum sample in triplicate was mixed with 10 μ L ISs, 127.5 μ L hexane/isopropanol (2:1, v:v), and 42.5 μ L water, and then three resulting solutions were spiked with 10 μ L of three different concentrations of standard FAs, respectively.