Diagnostic accuracy of MALDI mass spectrometric analysis of unfractionated serum in lung cancer

Abstract

Purpose: There is a critical need for improvements in the noninvasive diagnosis of lung cancer. We hypothesized that matrix-assisted laser desorption ionization mass spectrometry (MALDI MS) analysis of the most abundant peptides in the serum may distinguish lung cancer cases from matched controls.

Patients and methods: We used MALDI MS to analyze unfractionated serum from a total of 288 cases and matched controls split into training (n = 182) and test sets (n = 106). We used a training-testing paradigm with application of the model profile defined in a training set to a blinded test cohort.

Results: Reproducibility and lack of analytical bias was confirmed in quality-control studies. A serum proteomic signature of seven features in the training set reached an overall accuracy of 78%, a sensitivity of 67.4%, and a specificity of 88.9%. In the blinded test set, this signature reached an overall accuracy of 72.6 %, a sensitivity of 58%, and a specificity of 85.7%. The serum signature was associated with the diagnosis of lung cancer independently of gender, smoking status, smoking pack-years, and C-reactive protein levels. From this signature, we identified three discriminatory features as members of a cluster of truncated forms of serum amyloid A.

Conclusions: We found a serum proteomic profile that discriminates lung cancer from matched controls. Proteomic analysis of unfractionated serum may have a role in the noninvasive diagnosis of lung cancer and will require methodological refinements and prospective validation to achieve clinical utility.

FIGURE 1

Flowchart describing the study design. A set of matched cases and controls (n = 288) was split into a training set (n = 182) and a test set (n = 106). A model signature defined in a training set was tested in a test set of matched cases and controls.

FIGURE 2

Workflow of serum profiling by MALDI MS acquisition and data processing. One-microliter samples were diluted 1:10 in water before printing to a MALDI plate. One microliter of 10 mg/ml 3,5-dimethoxy-4-hydroxy-cinnamic acid in 50:50 acetonitrile/0.1% trifluoroacetic acid was placed on the surface and allowed to dry. MALDI-MS spectra were acquired using a PerSeptive Voyage Elite time-of-flight mass spectrometer equipped with delayed extraction and a nitrogen laser (337 nm). All spectra were acquired in the linear mode with delayed extraction, using a laser intensity of 2300 (arbitrary units). Signal-to-noise (S/N) calculations included both chemical and electronic noise. Data were then processed for calibration, smoothing, baseline correction, normalization, feature selection by S/N ratio, binning, alignment, and normalization. Data were analyzed, and discriminatory signatures were obtained after statistical tests for class comparison. Discriminatory features were selected after statistical threshold and visual confirmation of the nature of the peak. Prediction of diagnostic accuracy was provided by the weighted flexible compound covariate method. Signature was validated in two independent sets of spectra. A subgroup of masses was identified by gel electrophoresis, followed by protein digestion and tandem MS–MS.

FIGURE 3

MALDI MS serum spectra from individuals with lung cancer and matched controls. Average intensity of spectrum analysis between matched cases (red plain line) and controls (blue dotted line) are presented. Arrows point to m/z values of discriminatory features.

FIGURE 4

(A) Receiver operating characteristic curves addressing diagnostic efficacy of cases and controls in the two datasets (182 training sample set, 106 matched test set). (B) Nonmonotonic (quadratic in ranks) generalization of the Spearman rank–correlation coefficient, for each of these five predictors. The generalized Spearman coefficient helps to describe the strength of marginal relationships between each of these predictor variables and the response (being with lung cancer or not). This plot shows that the serum profile has the strongest correlation with the response among these five predictors. (C) Odds ratio of being diagnosed with lung cancer according to serum profile quantile distribution.

FIGURE 5

(A) One-dimensional SDS-PAGE analysis of proteins from control and lung cancer serum samples. Serum proteins were separated using SDS-PAGE on a 10% to 20% tricine gel and stained using colloidal blue. The gel illustrates a band in the molecular weight range of 11 to 12 kDa that appears frequently in lung cancer samples but is absent in the controls. This region was excised from control and cancer samples, trypsin digested, and subjected to LC-MS-MS analysis. Lane 1, molecular weight markers; lanes 2 through 4, control serum; lanes 5 through 7, lung cancer serum. (B) Proteins identified from these bands included several variants of serum amyloid A (SAA) that were unique to the lung cancer samples. FRAGMINT generated candidate protein fragments from serum amyloid A1 isoform 2, which contained the SEQUEST-identified peptides and were consistent with the observed MALDI MS m/z values. Peptides identified are underlined. The peptides contained in the sequence from positions 43 to 105 were also identified but are not shown. The intact sequence for serum amyloid A1 isoform 2, with an intact m/z value of 13532, is presented below the possible truncated forms. (C) SAA protein expression assessed by Western blot analysis in 10 serum samples, five from patients with lung cancer (cases) and five from controls. (D) Immunodepletion of SAA removes the peaks at m/z of 11526 and 11682 from serum samples of patients with cancer. Twenty microliters of packed rec-protein G–sepharose 4B conjugate (Zymed Lab Inc., CA) were incubated for 2 hours with 40 μg of mouse monoclonal antihuman SAA antibody (Antigenix America Inc., NY) or with immunoglobulins (mouse IgG, Santa Cruz, CA). Antibody–protein G–sepharose complex was washed with phosphate-buffered saline and incubated with 20 μl of serum overnight at +40°C. The supernatant collected after brief centrifugation was considered as SAA-depleted serum. One microliter each of the serum, the SAA-depleted serum, and the serum incubated with control immunoglobulins were diluted 1:10 and subjected to MALDI MS analysis.