Trans‐omic profiling between clinical phenoms and lipidomes among patients with different subtypes of lung cancer

Abstract Lung cancer has high mortality, often accompanied with systemic metabolic disorders. The present study aimed at defining values of trans‐nodules cross‐clinical phenomic and lipidomic network layers in patients with adenocarcinoma (ADC), squamous cell carcinomas, or small cell lung cancer (SCLC). We measured plasma lipidomic profiles of lung cancer patients and found that altered lipid panels and concentrations varied among lung cancer subtypes, genders, ages, stages, metastatic status, nutritional status, and clinical phenome severity. It was shown that phosphatidylethanolamine elements (36:2, 18:0/18:2, and 18:1/18:1) were SCLC specific, whereas lysophosphatidylcholine (20:1 and 22:0 sn‐position‐1) and phosphatidylcholine (19:0/19:0 and 19:0/21:2) were ADC specific. There were statistically more lipids declined in male, <60 ages, late stage, metastasis, or body mass index < 22 . Clinical trans‐omics analyses demonstrated that one phenome in lung cancer subtypes might be generated from multiple metabolic pathways and metabolites, whereas a metabolic pathway and metabolite could contribute to different phenomes among subtypes, although those needed to be furthermore confirmed by bigger studies including larger population of patients in multicenters. Thus, our data suggested that trans‐omic profiles between clinical phenomes and lipidomes might have the value to uncover the heterogeneity of lipid metabolism among lung cancer subtypes and to screen out phenome‐based lipid panels as subtype‐specific biomarkers.

cancer (SCLC) cells altered after mitogen-activated protein kinase (MAPK) kinase module MEK5/ERK5 was blocked, accompanied by dysfunctions of several lipid metabolism pathways like the mevalonate pathway for cholesterol synthesis. 1 Lipids, mainly including subclasses of phosphatidic acid (PA), phosphatidylcholines (PC), phosphatidylethanolamine (PE), phosphatidylglycerol (PG), phosphatidylinositol (PI), and phosphatidylserine (PS), have multiple important biological functions, such as biomembrane composition, vesicular trafficking, adhesion, migration, apoptosis, energy storage, neurotransmission, signal transduction, and posttranslational modification. They have alterations under circumstance of lung cancer. Circulating levels of PCs and PEs in patients with nonsmall cell lung cancer (NSCLC) differed from those with noncancer lung diseases or health and were suggested as diagnostic biomarkers of early NSCLC. 2 The heterogeneity of circulating lipidomic profiles was found to exist among patients with squamous cell carcinomas (SCC), adenocarcinoma (ADC), or SCLC, and there was a clear correlation between genomic and lipidomic profiles of lipid-associated proteins and enzymes. 3 As the part of clinical trans-omics, the lung cancer-specific and subtypespecific lipidomics in the circulation were defined and evidenced by integrating lipidomic data with genomic expression of lipid proteins among lung cancer subtypes.
Clinical trans-omics is defined as a new subject to integrate clinical phenomes with molecular multi-omics for understanding molecular mechanisms of diseases in multiple dimensions. 4 Clinical trans-omics becomes more important as a new and novel approach for the discovery of disease-specific biomarkers and therapeutic targets, although there are still many obstacles to be overcome, for example, specificity and decisive role of trans-nodules among multi-omic networks for intra-and intercellular communication. 5,6 Recent studies applied the transomics among phosphor-proteomics, transcriptomics, gene sequencing, and genomics for new molecular category of liver cancer, to provide a new therapeutic strategy. 7 As the part of clinical trans-omics, clinical lipidomics was considered as one of major metabolic profiles for identification and validation of lung cancer-specific biomarkers by integrating clinical phenomes with lipidomic profiles. 8,9 Clinical lipidomics could demonstrate the complexity of the lipidome in metabolic diseases and lung cancer and presented the variation among diseases and subtypes of lung cancer. [10][11][12] Our previous study demonstrated the difference of lipidomic profiles among patients with different lung cancer subtypes and the potential association between lipidomic phenotypes and gene expression of lipid metabolism-associated proteins and enzymes as a concept evaluation. 3 The present study furthermore investigated the values of trans-nodules cross-clinical phenomic and lipidomic network layers in the recognition of lung cancer subtypes (ADC, SCC, and SCLC), in order to understand clinical phenome-associated lipid changes or lipidomic phenotype-associated clinical phenomes. We also evaluated the differences of lipidomic profiles between male and female, various ages, early and late stages, with or without metastasis, body mass index (BMI) <22 or >22, and digital evaluation scores less or more than 90.

Digital evaluation score system
The Digital Evaluation Score System (DESS) is a score index system by which clinical descriptive information of each phenome can be translated into clinical informatics. 14 When the severity of each component was scored as 0, 1, 2, or 4, of which 4 represented the most severe condition, whereas 0 indicated normal physiological range. The gross DESS scores ranged from 0 to 584 points; the higher the score, the severer the condition. A total of 194 clinical phenomes were collected and scored in each of three lung cancer group, including 30 histories, 30 symptoms, 18 signs, 27 laboratory measurements, 49 image features, and 40 pathologic indexes, as listed in Table S1.

Lipid extraction for mass spectrometry analysis
About 200 µL plasma was collected into a glass tube, into which 10 µL internal standard was added and then 5 mL of methanol:chloroform:formic acid (10:10:1), as reported previously. 3,15 This mixture was incubated at -20 • C overnight after vigorous shaking. Two milliliters of Hajra's reagent (0.2 M H 3 PO 4 , 1 M KCl) were dropped, blended, and centrifuged at 3000 rpm for 5 min. After stratification, chloroform in the lower layer was pipetted to another glass tube and concentrated to 200 µL with the nitrogen flow, where the liquid with isopropyl alcohol:hexane:100 mM ammonium acetate at the ratio of 58:40:2 was added till 1 mL. The sample was then centrifugated at 14 000 rpm at 4 • C for 20 min. The normal-phase liquid chromatography coupled Triple-Quadrupole mass spectrometer (QTRAP 6500, SCIEX, Framingham, MA, USA) was used for lipid extraction by the positive and negative electrospray ionization mode. In the multiple reaction, Q-Trap was utilized to scan the precursor/product ion and examine the mode operation. Each test was repeated three times. The peak area of each pair was quantified with multiple reaction monitoring data by the software MultiQuant (AB SCIEX).
In the meanwhile, 2.0 µL was added with the split ratio of 50:1 at the ignition chamber temperature of 220 • C and the injection port temperature of 150 • C. It was started at temperature 150 • C, which gradually increased (4 • C/min) to 250 • C, and kept for 5 min. The mass spectrometry was subjected to liquid chromatography-mass spectrometer analysis (FOCUS DSQTM II, Thermo Fisher Scientific), mainly under the following conditions: Electron Ionization (EI) as ionization source, ion source temperature at 200 • C, ion-ization voltage at 70 eV, multiplier voltage at 0.9 kV, 4 min solvent delaying, and 50-650 amu of scan range.

Identification of lipidomic profiles
Lipid extracts were loaded onto an Ultremex silica column (250 mm × 4.6 mm, 5 µm), which was fitted with a 2 mm × 4 mm silica guard cartridge (Phenomenex, Torrance, CA, USA), and then eluted. The sample was enriched at a gradient of 300 nL/min. In the 50 min's run, B phase was 50% from 0 to 5 min, then rose to 100% from 5 to 30 min, linearly ramped for 10 min, as to return 50% from 40 to 41 min until the end. The Q-Trap was conducted in the multiple-reaction monitoring mode, and the different precursor/product ion pairs were scanned in the positive/negative electrospray ionization mode. Up to 502 lipids of plasma samples were carried out to get possible lipids chemical structures.

Comprehensive analyses of lipidomic profiles
MultiQuant software (AB SCIEX) was used to process data, after lipids were identified by mass spectrometry. Further, MetaboAnalyst software 4.0 (www.metaboanalyst.ca) was utilized for conducting multivariate statistical analysis, cluster analysis, dimensionality reduction, and making heat map.

Trans-nodule analyses cross phenome and lipidome network layers
The type-specific lipids were identified as more than two times elevated or declined significantly compared with other lung cancer subtypes (fold change > 2 and Pvalue < .05), whereas the co-expression lipids were identified as those similarly changed in all lung cancer subtypes as compared with healthy controls. The expression quantitative trait locus (eQTL) model was utilized to evaluate trans-nodules between lipidomic profiles and clinical phenomes.

Statistical analysis
Data were presented as mean ± SE. The means of each group were used for calculation and comparison. Statistical significance of differences between two groups or among multiple groups was determined by Student's t-test or one-way ANOVA test, respectively. Statistical significance was affirmed when P-value < .05. We also separately calculated mean values of each phenome's DESS score in different lung cancer subtypes, which were ranked to obtain top 10 clinical phenomes of those three groups of patients. Volcano maps showed the significantly elevated or declined lipids in ADC, SCC, or SCLC patients. A VIP plot was further exploited to sort the lipids according to their importance to differentiate the four groups. To explore the correlation between lipid elements and clinical phenomes, we applied the lipid-quantitative trait loci model modified from eQTL model. Besides, MatrixlQTL R package was used to acquire the significant phenome-lipid pairs and corresponding P-values. Moreover, GraphPad Prism was utilized to make the receiver operating characteristic curve to evaluate the early-diagnostic value and accuracy of clinical phenome-specific lipid elements in ADC, SCC, or SCLC. The present study furthermore analyzed the significant differences of lipids among different ages (eg, <60, 60-70, and >70), between female and male, early and late stage, metastasis and non-metastasis, high and low DESS scores (≤90 and >90), and high and low BMI (≤22 and >22).

Clinical phenomes of patients with lung cancer subtypes
Eighteen female and 36 male lung cancer patients were enrolled in the present study, aged from 43 to 80 (63 ± 10) years old, including 28 ADC, 15 SCC, and 11 SCLC. The total scores of DESS were 72 ± 28, 79 ± 25, and 98 ± 19 in patients with ADC, SCC, and SCLC, respectively. The DESS values of SCLC group were significantly higher than those of healthy control group (P < .05). Top 10 clinical phenomes of ADC, SCC, or SCLC patients as well as patients survived or nonsurvived during study period were listed in Table 1. Stages at primary diagnosis and recruitment period for the study, lymphatic metastasis (N1-2) in ipsilateral paratracheal hilum or mediastinum, and enhanced images (eg, focus enhanced in CT or hypermetabolism in PET/CT) were shown in all three subtypes of lung cancers. In addition, thyroid transcription factor-1 (TTF-1), Napsin A, keratin 7, and location of tumor were noticed in ADC; obscure boundary, emphysema, tumor size, the cycle number of first line chemotherapy, obstructive pneumonia atelectasis, and pulmonary nodule in SCC; as well as number of metastatic lymph nodes in SCLC, separately. Top 10 clinical phenomes were similar between survived and nonsurvived patients, but the total amounts of DESS of nonsurvived patients were significantly higher than those of survived patients (Table 1). Of 194 total clinical phenomes, 46 had the statistical significance of each two groups in-between (Table 2; P < .05 or less).
As compared with the healthy control ( Figure 4A), we noticed that PI mainly declined in ADC ( Figure 4B), PA Abbreviations: N, degrees of lung cancer metastasis to lymph nodes of TNM category; N1, degree 1 that has metastatic lymph nodes near pulmonary center and side of main bronchia; N2, degree 2 that has metastatic lymph nodes in the same side of the mediastinum as lung cancer; TTF-1, thyroid transcription factor-1 as an immunohistochemical biomarker for adenocarcinoma; CK7, keratin 7 as an immunohistochemical biomarker for epithelial cells; L1 cycle, number of the first line chemotherapy cycles.
in ADC and SCC ( Figure 4C), and lysoPG in SCLC (Figure 4D), whereas PG and TAG increased in ADC and SCC, PE in SCLC, and PC in SCC. The volcanic map demonstrated the clear patterns of lipid elements significantly increased or declined between heathy controls with ADC ( Figure 4E), SCC ( Figure 4F), or SCLC ( Figure 4G) and varied among different subtypes of lung cancers.

Different lipidomics between patient genders
About 81 or 79 lipid elements significantly increased and 38 or 21 declined more than twofold in male or female lung cancer patients, as compared with male or female healthy controls, respectively (Tables S4 and S5). Of those, PC and PE mainly elevated in male and female patients, whereas PA declined in both, although the number of PA in male patients was more than in female patients. Table 4 demonstrates gender-specific lipid elements identified by those lipid elements elevated or declined exclusively in either male or female lung cancer patients, for example, some of lysoPS, PC, and PS elevated in male patients, whereas lysoPI and PE in female patients. There were about 45 or 28 increased or declined lipid elements differed between male and female lung cancer patients (Table 4).

TA B L E 2
Comparisons of clinical phenomes in increased folds and statistical significance (P-values) between each two groups of adenocarcinoma (ADC), squamous cell carcinoma (SCC), or small cell lung cancer (SCLC) patients

Different lipidomics among patient ages, stags, metastases, and survival status
About 106, 88, and 82 lipid elements significantly elevated or 47, 44, and 18 declined in lung cancer patients at <60, 60-70, >70 years old, respectively, as compared with healthy controls (P < .05). We noticed that elements of PG and PS mainly increased in lung cancer patients at all age groups: for example, 19% and 18% at <60-year group; 17% and 20% at 60-to 70-year group; and 16% and 16% at >70year group; lysoPC and PC increased at <60-year group (18% and 16%) and at >70-year group (16% and 24%); PE increased at 60-to 70-year group (17%) and at >70-year group (22%), as detailed in Table S6. Elements of PA mainly declined in lung cancer patients at all ages (Table S7). Of those significantly altered lipid elements, 13%, 20%, and 15% appeared only at <60-year, 60-to 70-year, and >70-year groups, respectively, and considered as age-specific lipid elements (Table 5). LysoPC and lysoPI mainly increased in <60-year and 60-to 70-year old patients, whereas lysoPE declined in <60-year group. We also compared lipidomic profiles between patients at early and late stages of lung cancer, and found 72 and 86 lipid elements significantly increased at early and late stages, respectively, of which PE, PG, ad PS increased in both stages, lysoPI in early stage, and PC in late stage (Table S8). About 10 and 42 elements declined at early and late stages, where the majority was PA (Table S9). Table 6 demonstrates stage-specific lipid elements identified by those lipid elements elevated or declined exclusively at early and late stages of lung cancer, for example, some of lysoPI and PE elevated at early stage, and lysoPC and PE at late stage.
We noticed about 54 or 89 lipid elements significantly increased in patients without or with metastasis, of which lysoPI mainly elevated in patients without metastasis, whereas PC and PE in patients with metastasis (Table S10). The declined number of lipid elements, especially PA in patients with metastasis, was significantly higher than in patients without metastasis (Table S11). There were about 58 or 25 elevated or declined lipid elements in patients with metastasis, of which PA was majority of declined elements in patients with metastasis (Table 7).
Lipidomic panel also differed between survived and nonsurvived patients. There were only eight lipids exclusively elevated in nonsurvived patients, that is, lysoPS14:0,  (Table 8).

Trans-omic profiles between clinical phenomes and lipidomes
We also compared the difference of lipidomic profiles between general metabolism statuses of patients indicated by BMI and between degrees of clinical phenomes measured by DESS scores. Levels of lysoPC or lysoPI mainly elevated in patients with BMI ≤ 22 or > 22, respectively (Table S12), whereas the number of declined PA in patients with BMI ≤ 22 was higher than that in patients with BMI > 22 (Table S13). About 42 BMI-associated lipid elements significantly elevated or declined exclusively in patients with BMI ≤ 22, and about 22 in patients with BMI > 22 (Table 9). Levels of lysoPC and PE or PG and PS mainly increased in patients with DESS ≤ 90 or > 90 (Table  S14). The number of declined PA (n = 19) in patients with DESS ≤ 90 was lower than that in patients with DESS > 90 (n = 44). Figure 5 demonstrates the variation of trans-omic profiles among lung cancer subtypes indicated by transomic nodules cross significant networks of clinical phenome and lipidome layers.

DISCUSSION
The present study preliminarily found the differences of lipidomic profiles among patients of different lung cancer subtypes, genders, ages, stages, metastatic status, body qualities, and clinical phenome severities. Besides, it initially demonstrated clinical phenome-associated lipid TA B L E 5 Age-associated lipid elements significantly elevated or declined alone in each age group of patients with lung cancer (more than twofold) as compared with healthy control (P-values) elements and lipid element-associated clinical phenomes using clinical trans-omics. Studies on lipidomic profiles of lung cancer patients have experienced three phases to detect the difference of lipidomic profiles between healthy and lung cancer patients, 16 the association of multi-omics among lung cancer subtypes, 3 and the molecular mechanism of clinical lipidomics-based target lipid elements. 17 Of those lipidomics-based data, limited information could be adopted to understand the disease occurrence and development, phone progression, and TA B L E 6 Stage-associated lipid elements significantly elevated or declined alone in patients with lung cancer at the early stage or late stage (more than two folds), respectively, as compared with healthy control (P-values) response to therapy, due to the lack of link between omics data and clinical phenomes. Like other omics investigations, most genomic data were not tied with clinical information, so that with little values to be understood and applied for clinical precision medicine. 18 In order to face the major challenge that most clinical information was descriptive and unmatched with the digital quantity of omics data, clinical phenomes were scored by DESS and integrated with genomic and proteomic data of patients with acute respiratory distress syndrome and chronic obstructive pulmonary disease. [19][20][21][22] Clinical phenomes were furthermore integrated with lipidomic profiles in patients with pulmonary embolism, acute pneumonia, and acute exacerbation of chronic obstructive pulmonary diseases, based on clinical trans-omics principle. 15 Lipidomic profiles difference between health and lung cancer has been defined and it depends upon methodologies of measurement and analysis, sample preparations and sources, and patient populations and status. 8 For example, serum levels of lysoPC (C26:0 and C26:1) and PC (C42:4 and C34:4) were different between stage I NSCLC and healthy patients. 22 Some elements and pathways of serum PC and PE profiles increased in patients with lung benign disease and early-stage NSCLC, as compared with healthy, whereas few (e.g., PC 15:0/18:1) significantly elevated in early-stage NSCLC patients alone. 2 It seems that patterns of lipid elements may be associated with the specificity of lung cancer and stage, rather than the intact lipid pathways. We performed a pilot prospective study to compare the variation of lipidomic profiles between health and lung cancer, and among lung cancer subtypes, as well as to correlate lipidomics with circulating leukocyte transcriptional profiles and clinical phenomes. 3 The present study further investigated the difference of lipidomics among lung cancer subtypes with a special focus on subtype-specific lipid elements. Of measured lipid elements, levels of PE elements (36:2, 18:0/18:2, and 18:1/18:1) were significantly higher in SCLC than in healthy control and other lung cancer subtypes, whereas lysoPC (20:1 and 22:0 sn-position-1) and PC (19:0/19:0 and 19:0/21:2) levels in ADC were more than 1500-fold higher than in SCLC. Similar results were also seen in other studies: PCs (34:1, 36:2, 36:3, and 32:0) were found upregulated in NSCLC, 23 whereas the latter (PC 32:0) was identified as a diagnostic factor for ADC in Hall's study. 24 Marien also found an increase in several PC species. 25 The specificity of human breast cancer cell subtypes was also noticed by measuring the quantitation of lipid C = C location and sn-position structure isomers. 26 The monounsaturated/saturated PC ratios were proposed to distinguish ADC histologic subtype-dependent variations among lepidic, acinar, papillary, micropapillary, solid, and mucinous phenomes. 27 The heterogeneity of lipidomic profiles and lipid metabolism could impact the specificity for lung cancer subtype, severity, stage, and drug efficacy, although characteristics of lipidomic profiles of lung cancer subtypes varied among studies because of different subjects, designs, methods, and analyses. 8 In addition, multiple factors of lung cancer patients, for example, age, gender, stages, metastatic status, body condition, and clinical phenotyping, should be considered to evaluate values of lipidomic profiles. Lipidomic profiles with about 300 lipids in 11 classes in homogenized human lung cancer tissue were correlated with clinical and historical phenotypes, and it turned out that they had high links with stroma and tumor contents, inflammation, aging, and emphysema. 28 Some of lung surfactant components changed in lung cancer tissues, including triacylglycerols, cholesteryl esters, and PGs. It was suggested that lipid panels might have the specificity to differentiate between ADC and SCC, and the association between lipid profiles of tumor-free alveolar tissues and phenotypes. We found that the total number of changed lipid elements was obviously higher in younger patients, late stage, metastasis, or BMI ≤ 22. Of those changes, the decline number of lipids in male was 50% higher than female, <70 ages 60% than >70 ages, late stage 75% than early stage, metastasis 37% than nonmetastasis, or DESS > 90 58% than DESS ≤ 90.
Multidimensional analyses of lipidomic profiles in the present and previous studies strongly indicate that identified target panels of lipid elements have great impacts of identifying influencing factors in early diagnosis severity, progression, and therapeutic responses for lung cancer patients, by using multivariate classification or trans-omic network models. Choline-containing phospholipids of serum lipidomic profiles were considered as the potential TA B L E 7 Metastasis-associated lipid elements significantly elevated or declined alone in lung cancer patients with metastasis (more than twofold), as compared with healthy control (P-values) biomarkers to differentiate between stage I NSCLC and noncancer subjects, especially lysoPC (C26:0 and C26:1) and PC (C42:4 and C34:4). 26 The panel of lysoPC (18:0, 18:1, and 18:2) was suggested to detect early lung cancer patients from population screen, asymptomatic imaging detection, or stage IA-IIB. 16 We also selected several target panels with specificity on the basis of the fact that some lipids only altered in certain population of patients, for example, lysoPS ( To differ from the correlation strategy between lipidomic profiles and other phenomes and reduce potential subjective biases, we furthermore analyzed the trans-omic nodules cross lipidomic and clinical phenomic layer networks of mechanistic interactions, which is consisted of measures from patients with lung cancer subtypes with sta-tistical difference from healthy patients. Yugi and Kuroda demonstrated that metabolism-centric trans-omics contained multiple nodules on each omic layer; cross multilayers presented global regulatory mechanisms of metabolic pathways, targets of direct/indirect regulations and interactions, and spatiotemporal dynamics of cell functiondependent networks. 29 Differentiated from multi-omics, trans-omics can be a potent approach to screen out and develop disease-and function-specific biomarkers and therapeutic targets and provide more precise insights and predictions for disease progression and prognosis, although clinical applications are needed to evaluate the practical values, outcomes of trans-nodules vary among methodologies, and repeatability and reliability of selected crossing nodules are validated. 4,30,31 It is highly expected to link molecular omics with biological phenotypes, clinical phenomes, and environmental factors. We introduced clinical phenomics into trans-omics to evaluate network nodules cross lipidomic and phenomic layers and define phenome-specific lipid panels and lipid-specific phenome groups in acute lung injury patients with or without infections and with or without chronic diseases, as well as in patients with lung cancer subtypes. 8,15,17 To gather our present and previous studies, we found that the variation of trans-omic profiles in patients was TA B L E 8 Lipid elements significantly elevated or declined alone in lung cancer patients survived or nonsurvived (more than twofold), respectively, as compared with healthy control (P-values) Those lipid metabolites and related limiting enzymes can be regulated by up/down-expression of the corresponding genes and mitochondrial DNA methylation and fusion/ fission. 34,35 There were also limitations in this study. First, the sample size was small, especially as for healthy control and for SCC and SCLC lung cancer groups. Second, therapeutic factors were not explored, because most lung cancer patients included underwent chemotherapy. For more and more effective molecular target medications and immunotherapies that were applied in lung cancer, corresponding groups of patients should be studied in future researches to explore medical influence on lipidomics. Last but not least, there might be latent confounding effects of complications to lipidomic metabolism, such as diabetes, hyperlipemia, and atherosclerosis, which were not strictly excluded in the study.
In conclusion, we qualitatively and quantitatively measured plasma lipidomic profiles in lung cancer patients and found that altered lipid panels and concentrations varied among lung cancer subtypes, genders, ages, stages, metastatic status, nutritional condition, and clinical phenome severity. Levels of PE elements (36:2, 18:0/18:2, and 18:1/18:1) were found to be SCLC specific and lysoPC (20:1 and 22:0 sn-position-1) and PC (19:0/19:0 and PC 19:0/21:2) ADC specific. We found that the declined number of lipids was obviously higher in male, <70 ages, late stage, metastasis, or DESS > 90. Clinical trans-omics analyses demonstrated that one phenome in lung cancer subtypes may be generated from multiple metabolic pathways and metabolites, whereas a metabolic pathway and metabolite can contribute to different phenomes among subtypes, although those need to be furthermore confirmed by studies of larger population of patients in multicenters. Thus, our data suggested that trans-omic profiles between clinical phenomes and lipidomes might have the value to uncover the heterogeneity of lipid metabolism among lung cancer subtypes and to find phenome-based, subtypespecific lipidomic biomarkers. TA B L E 9 BMI-associated lipid elements significantly elevated or declined alone in lung cancer patients with BMI < 22 or > 22 (more than twofold), respectively, as compared with healthy control (P-values)

D ATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available from the corresponding author upon reasonable request.