Background
Lung cancer is the leading cause of cancer death in men and the second leading cause of cancer death in women worldwide [
1]. It was estimated that 1.8 million new lung cancer cases and 1.6 million lung cancer death occurred in 2012 worldwide, accounting for almost 19% of all cancer deaths [
2].
Most patients with lung cancer are diagnosed with advanced disease, resulting in a very low global 5-year survival of only 18% [
3]. Screening aims to detect lung cancer in an early stage, before patients experience clinical symptoms, and when treatment is the most effective. The principal aim of screening for lung cancer by low-dose computed tomography (CT) is to reduce lung cancer-specific death [
4,
5]. CT-imaging often identifies suspicious pulmonary nodules or focal lung lesions, but cannot verify whether these are the results of benign disease or a truly aggressive malignancy, leading to supplementary imaging techniques or additional CT scans with cumulative radiation levels or invasive procedures, such as tissue biopsies [
4,
6].
Due to limitations of radiological imaging techniques in the differentiation between benign and malignant tissue, positron emission tomography (PET) has become an additional option for the evaluation of suspicious pulmonary nodules and other focal lung lesions [
7].
Unlike normal tissue, malignant tumors are characterized by an increased glycolysis, which leads to an elevated glucose uptake.
18F-fluorodeoxyglucose (
18F-FDG) PET-CT makes use of this characteristic in order to diagnose and stage various human malignancies [
8‐
10]. The standardized uptake value (SUV) is a semi-quantitative measurement of the tissue
18F-FDG accumulation rate [
10]. The maximal standardized uptake value (SUV
max) is the voxel with the highest
18F-FDG uptake value in the region of interest.
However, regardless of its high accuracy and sensitivity, high
18F-FDG uptake is not cancer-specific. High levels of
18F-FDG uptake can also be detected in benign lesions such as inflammation, causing false-positive results and misinterpretation for diagnosis [
11]. Tremendous efforts have been reported in the literature to deal with this false-positive issue using different tracers e.g. labeled amino acids [
12]. However, these tracers have predominantly been used in the research environment with limited clinical usage thus far [
13]. In parallel with the introduction of new tracers, researchers also proposed different measuring protocols such a as dual time point imaging procedure and dynamic PET with tracer kinetic modeling [
14,
15]. Usually, such modeling procedures are complex, requiring longer scanning sessions, invasive arterial blood sampling, tracer analysis and complex data processing, making the technique less appropriate in daily clinical practice.
Taking the above into account, there is an urgent need to find complementary non-invasive, clinical biomarkers that are able to better discriminate between false positive and true positive results.
In recent years, metabolomics or metabolite profiling/phenotyping, has been used to investigate metabolic changes in plasma associated with lung cancer [
16‐
19]. Metabolomics is the study of substrates and products of metabolism, which are influenced by both genetic and environmental factors. Metabolites and their concentrations directly reflect the underlying biochemical activity of cells and represent the phenotype. Currently, mass spectrometry coupled to different chromatographic separation methods and
1H-NMR spectroscopy are the major tools to analyze a large number of metabolites simultaneously. Several research groups have developed a
1H-NMR derived metabolic signature of lung cancer in tissue or plasma [
16,
17,
19‐
21]. However, the patient populations in these studies were rather limited.
Recently, our research group was able to detect lung cancer in a population of 269 patients and 347 controls with a sensitivity of 78% and a specificity of 92% by means of the metabolic phenotype of blood plasma [
16]. In general, the principal metabolic alterations reported for lung cancer include changes in amino acid metabolism, choline phospholipid metabolism, glycolysis, one-carbon metabolism and lipid metabolism.
Metabolic phenotyping by 1H-NMR spectroscopy of patients with benign PET-positive lesions and of patients with lung cancer might result in the discovery of new selective biomarkers with diagnostic potential that can influence the decision-making in case of positive screening results.
The present study is the first in the field of metabolomics that aims to investigate whether the 1H-NMR-derived metabolic phenotype of blood plasma allows to discriminate between patients with pulmonary inflammatory disease and lung cancer, as well as between patients with lung inflammation and controls.
Discussion
In the United States, regular low-dose CT screening has been recommended for smokers and ex-smokers at high risk of developing lung cancer [
5]. However, the main challenge for lung cancer screening by CT remains the high prevalence of pulmonary nodules and/or lymph nodes, and a relatively low incidence of lung cancer in the screened population [
4,
33,
34]. This results in a low PPV after exclusion of lung cancer by additional imaging and potential harmful procedures, such as tissue biopsies. The aim of this study is to search for metabolites that discriminate between lung cancer patients and patients with lung inflammation by means of the plasma metabolic fingerprint. The metabolic phenotype or fingerprint consists of a large number of variables, each of them representing a single or several metabolite concentrations. To the best of our knowledge, this study is the first in the field of metabolomics that investigates the metabolic differences in blood plasma of patients with lung inflammation and lung cancer.
This study indicates that the metabolic phenotype of blood plasma, and particularly the region representing glutamate, allows to discriminate between patients with lung inflammation and with lung cancer, as well as between patients with lung inflammation and controls. These results strongly suggest the role of glutamate as a selective inflammatory marker in lung diseases. Ideally, after detection of a suspicious lesion on chest-CT, differences in the plasma metabolic profile in combination with PET findings may add valuable information about the underlying disease, i.e. cancer versus inflammation. This approach may reduce the need of invasive diagnostic procedures when the lesion has inflammatory characteristics.
Analytical approaches, such as
1H-NMR spectroscopy, generate a large number of variables per sample, resulting in models with a risk of overfitting. A careful selection of the appropriate statistical method is necessary as each of the techniques has advantages and disadvantages. The choice of method is dependent on the type of data: missing values, influence of outliers, predictive power, etc. [
30]. In the field of metabolomics, there is an increasing interest in PLS-LDA since it reduces the dimensionality of the spectroscopic data and can handle the noisy and collinear data from the experiment. Moreover, it is available in most of the statistical software packages.
Glutamate may have a key role in the differentiation between lung inflammation and lung cancer. Univariate t-test analysis with correction for multiple testing, shows that the glutamate concentration, represented by IR89, is the most significant variable with the smallest
p-value and a signal intensity which is significantly higher for lung inflammation as compared to cancer (Fig.
2 and Additional file
3: Figure S2). The differentiating power of this variable is stressed by multivariate PLS-LDA statistics showing an increase of the MCE with 26% (from 12% to 38%) after removing it from the dataset.
Addition of the SUV
max parameter, obtained by PET-CT, to the dataset has only a modest influence on the classification (e.g. a decrease of the MCE from 12% to 10%), indicating that the SUV
max has no significant power to differentiate between lung inflammation and lung cancer. This is supported by the limited specificity of PET-CT in excluding malignancy on the basis of the SUV
max value and the consensus that a metabolically active lesion requires histological assessment [
7].
MCE, sensitivities and specificities have the tendency to stabilize when the metabolic signature contains 16 variables. This means that despite the importance of glutamate, other IRs may have additional value in the classification process. Glutamate, however, was selected in all the LASSO models and was the most significant variable in the univariate analyses. Therefore, the diagnostic potential of glutamate as a single marker was further evaluated by a ROC curve.
To diagnose lung cancer, and in comparison with PET-CT itself, a relative glutamate level ≤ 0.31 has a lower sensitivity (85% versus 96%), a significant higher specificity (81% versus 23%), a higher PPV (92% versus 76%) and a comparable NPV (69% versus 71%). Due to this lower sensitivity (i.e. more false negative results) and the resulting NPV, glutamate as a single marker is insufficient to exclude lung cancer. To overcome these limitations, we propose to measure plasma glutamate in complement to PET-CT. In patients with both PET-positive lesions and low relative glutamate levels (suggestive for cancer), this procedure leads to a sensitivity and PPV to diagnose lung cancer of 100% (no false negatives) and 96% (higher true positive results than for PET/CT alone), respectively. In this patient group, a tissue biopsy or resection is indispensable to obtain the histology and to guide further therapy. A negative PET-CT and a high relative glutamate concentration (suggestive for inflammation) excludes lung cancer with a NPV of 100%. Here, further follow-up with CT but without invasive procedures seems to be justified. Caution is needed in patients with conflictive results, i.e. PET-positive patients with a high glutamate concentration or PET-negative patients with a low relative glutamate concentration. In these patients a tissue biopsy or more intensive follow-up is needed to exclude or confirm the presence of lung cancer since 19% of lung cancers remain undetected in this group.
As undetermined imaging results are less frequent in more advanced disease stages than in early stages, we compared the mean relative glutamate concentration in different stages by the leave-one-out-cross-validation (LOOCV) method. No significant differences were found between the glutamate levels of early (I and II), locally advanced (III) and advanced stages (IV), as demonstrated in Fig.
5 and Additional file
2.
To confirm the potential value of glutamate as a marker for lung inflammation, a PLS-LDA analysis was performed to discriminate between patients with lung inflammation and controls. The resulting model has a very small MCE of 7% and a high sensitivity and specificity. Relative glutamate concentrations were significantly higher in patients with lung inflammation compared to controls, supporting the importance of glutamate as an inflammatory biomarker. Building a ROC curve to determine an optimal cut-off in a diagnostic test for lung inflammation seems less relevant as common markers as C-reactive protein, sedimentation rate and leukocytosis are robust biomarkers.
Unfortunately, due to the retrospective nature of this study, these parameters were not available at the moment of the 1H-NMR analysis, preventing to look for possible correlations between the glutamate concentration and these markers.
Glutamate is a non-essential amino acid that accounts for 15% of the total amino acids in dietary proteins. Since the blood samples in this study were taken after an overnight fast and glutamate concentrations are normalized within 105 min after ingestion, the influence of glutamate intake should be negligible [
35]. Dysregulation of the glutamine-glutamate metabolism is reported for cancer cells [
36]. Cancer cells use glutamine as a source of carbon for further anabolic pathways (oxidation) and glutamine is hereto transported into the cells by the alanine-serine-cysteine-transporter-2. As a nitrogen donor for the synthesis of DNA and RNA building blocks, glutamine is converted into glutamate [
37,
38]. However, glutamine can also be exported out of the cell by antiporters in exchange for other non-essential amino acids through the L-type amino-acid transporter [
39]. Glutamine-derived glutamate also fulfills the role of a primary nitrogen donor for the synthesis of non-essential amino acids and is a precursor of the major cellular antioxidant glutathione (GSH) [
40,
41]. Increased GSH synthesis has been demonstrated in lung cancer tissue by Blair et al. [
42]. Higher levels of GSH have been related to apoptosis resistance [
43]. Glutamate that is not incorporated into GSH or involved in the synthesis of amino acids is converted to α-ketoglutarate (α-KG) through oxidative deamination. By this reaction, the glutamine-derived α-KG is utilized to replenish synthetic intermediates of the Krebs cycle, a phenomenon known as anaplerosis. Instead of the complete oxidation of glutamine to ATP, the mitochondria of cancer cells shunt glutamine into citrate for the production of NADPH and lipid synthesis, and into malate which can be converted into pyruvate and NADPH [
36]. The need of glutamate in the synthesis of GSH and macromolecules such as lipids and polynucleotides, may explain the lower levels of glutamate in the plasma of cancer patients compared to patients with lung inflammation. During inflammation the increase of vascular permeability facilitates the uptake of glutamate in the inflamed tissues. As part of the immune response generated by inflammation, cytotoxic T-cells are able to induce apoptosis in the inflamed tissue, thereby releasing intracellular glutamate. This process may explain the higher glutamate plasma concentration in patients with lung inflammation.
Regarding the role of glutamate in discriminating lung cancer patients from controls, the relative glutamate concentrations are not significantly different. As a marker of lung inflammation, glutamate is not able to distinguish between cancer patients and controls. Recently, our research group has demonstrated that the metabolic phenotype of blood plasma enables to distinguish lung cancer patients from controls [
16]. The fact that glutamate did not appear in the list of discriminating variables confirms our results and interpretation.
The generalizability of the results is subject to certain limitations. First, due to the retrospective nature of the study, other markers for inflammation such as C-reactive protein, sedimentation rate and leukocytosis were not available at the time of inclusion. Additionally, uncontrolled factors such as co-morbidities and their treatments might be possible confounders. It goes without saying that the role of glutamate as a potential marker of lung inflammation needs further evaluation in a prospective study with external validation and attention for possible confounders. Also the potential role of glutamate as a single biomarker for lung inflammation in a targeted approach needs to be further explored by another analytical technique such as HPLC-MS. And finally, the correlation with other markers for inflammation needs further investigation.