Verification of a Blood-Based Targeted Proteomics Signature for Malignant Pleural Mesothelioma

Currently, it is not possible to reliably detect malignant pleural mesothelioma (MPM) from the blood (1). MPM is mainly induced by asbestos and it is estimated that 1.3 million workers in the United States and 125 million people worldwide have a history of asbestos exposure (2–10). These numbers will likely increase in the coming years because of the ever growing use of asbestos in developing countries (11–13). A blood test for MPM would be helpful for monitoring people at risk and for early treatment decisions. However, even very well-investigated blood protein biomarkers, such as the soluble-mesothelin related peptides (SMRP) from the soluble form of mesothelin (14–16) or fibulin-3 (17), which use antibody-based immunoassays, are not routinely applied for the detection of MPM, mainly due to limits in sensitivity or absence of full validation (18–20). In our work, we intended to test a different approach for MPM biomarkers based on MS-targeted proteomics (21, 22). The technology allows the quantitative monitoring of dozens of proteins within complex samples such as blood (23–25). In this way, multiple different biomarkers are simultaneously accessible, which can increase the biological information about the cancer and its clinical impact (26–29). Such a strategy may support in particular modern personalized medicine, where clinical decisions are increasingly biology-driven. In addition, it offers a biologically more robust approach for cancer biomarkers. Indeed, the current general concept of a blood biomarkers assumes that a single protein can act as an accurate surrogate of the tumor. It is a priori unlikely that one protein alone may reliably reflect the complex phenotype of the cancer. Instead, the possibility to access multiple proteins at once will bring the observation closer to the biological reality of the cancer within the body. Furthermore, the human proteomics community has dedicated major and successful efforts to streamline and standardize the MS-based proteomics workflows, simplifying the access to the technology for nonexperts and advancing proteomics closer to clinical applications (30–37).

In our work, we therefore sought to test the use of a MS-based targeted proteomics approach for the detection of MPM from the blood using a high-risk cohort of controls, and its impact for assessing patient prognosis. As an anchoring reference, we related the exploratory results from our academic effort to a diagnostic assay in blood approved for clinical use, namely the FDA-approved immunoassay for the best validated MPM biomarker SMRP (38).

Biobanks were located at New York University (NYU) Langone Medical Center (New York, NY), at the National Centre for Asbestos Related Disease, University of Western Australia (Nedlands, Western Australia), and at University Hospital Zürich [Zürich, Switzerland, samples and clinical data were collected within study 17/04 of the Swiss Group for Clinical Cancer Research–SAKK (39), registered with ClinicalTrials.gov, number NCT00334594]. Serum samples were obtained by venipuncture from treatment-naïve donors and stored in accordance with standard operating protocols with written informed consent and in accordance with recognized ethical guidelines and IRB approvals of the collecting biobank. MPM diagnosis was histologically confirmed. Staging was performed according to the International Mesothelioma Interest Group staging system (40). The study was approved by The Ohio State University Medical Center Institutional Review Board (protocol number 2013C0182).

Serum samples were processed on 96-well plates using adapted protocols for the isolation of N-linked glycoproteins (N-glycoproteins), which enrich the serum proteome for the candidate biomarker peptides and reduces its analytic complexity (41–43). The PNGaseF enzyme cleaved the glycan as a glycosylamine, converting asparagine (N) to aspartic acid (D) in the process. Therefore, N was replaced by D in the amino acid sequences of the monitored N-linked glycopeptides (N-glycopeptides). Each 96-well plate contained aliquots (minimum two and maximum four aliquots per plate) from a commercially obtained serum sample from one healthy donor (Zenbio) to monitor the variability of the workflow. Two protein standards, namely nonhuman N-glycoproteins invertase 1 from Saccharomyces cerevisiae (UniProt entry identifier P10594, gene name SUC1) and recombinant viral B19R protein from Vaccinia virus (from R&D Systems, UniProt entry identifier P25213), were spiked into each aliquot of the serum samples of the cohort to account for discrepancies in the sample preparation (Supplementary Table S1). The tryptic digested serum peptides were resuspended in a standard mixture of isotopically labeled synthetic peptides (heavy peptides) including amino acid sequences matching the MPM candidate biomarker peptides (Supplementary Table S2), to monitor performance of the LC-SRM, and enable accurate detection and quantification of peptides, and between-run normalization. Samples were randomized, processed, and profiled over a period of approximately three months. Peptides were separated by reverse-phase liquid chromatography online with a triple quadrupole MS (Xevo TQ-S Waters) and quantified in targeted SRM mode. Additional details are reported in Supplementary Methods, online.

Immunodetection of SMRP in randomized serum samples was performed in duplicates using the commercially available Mesomark-Kit (Fujirebo Diagnostic) in accordance with the manufacturer's protocol. Results from one sample were removed from the analysis due to absorbance values consistently outside of the range proposed by the protocol.

A minimum of four and up to seven transitions per signature and protein standards peptides were monitored and a minimum of three and up to six transitions per peptides were selected for quantification (Supplementary Tables S1 and S2). Skyline-daily (from v2.1 and higher) was used for visualization and analysis of the SRM traces (34). Peak intensities were normalized with a three-step strategy based on heavy peptides and two N-glycopeptides from nonhuman spiked-in glycoproteins (HYNDITWYK, FATDTTLTK), as described in Supplementary Methods, online. Samples were randomly partitioned into a training set of 105 patients with MPM and 104 asbestos-exposed donors, and a testing set of 108 patients with MPM and 85 asbestos-exposed donors (Supplementary Table S3), while balancing MPM stages (early stages I and II and advanced stages III and IV). Parameters of the multivariate biomarker signature with six N-glycopeptides as predictors were estimated in the training set by fitting logistic regression model. The model was evaluated in the testing set. Detailed information is reported in Supplementary Methods, online.

The multicenter study population included 213 patients with MPM with early (stage I and II) and advanced (stage III and IV) stages and 189 cancer-free asbestos-exposed donors (Table 1). For 26 cases, nodal status was reported as N0/1 or N2 and disease stage was not further specified. Histology was mainly epithelioid, 17 cases were biphasic, seven were sarcomatoid, 29 cases were reported not to be epithelioid, and 12 cases were not otherwise specified. Extrapleural pneumonectomy (EPP) was performed in 81 cases and pleurectomy with decortication (P/D) in 42 cases (15 cases had not information about the surgery). The majority of patients receiving chemotherapy only were treated with platinum-based doublets (pemetrexed or gemcitabine).

The signature was derived from our previous targeted proteomics work for the identification of MPM biomarkers from the blood (44). There, we investigated the cell surface proteome of MPM and control cell lines with the hypothesis that N-glycosylated membrane proteins are prone to be shed or released into the blood and that they can therefore offer a promising source for blood biomarkers (45). We identified the MPM multivariate signature from cell line-derived N-glycoproteins detected by SRM-targeted proteomics in the serum of MPM and control donors. The original signature included seven N-glycopeptides from the proteins anthrax toxin receptor 1 (UniProt entry identifier Q9H6X2), basement membrane-specific heparan sulfate proteoglycan core protein (P98160), hypoxia up-regulated protein 1 (Q9Y4L1), intercellular adhesion molecule 1 (P05362), mesothelin (Q13421), serum paraoxonase/arylesterase 1 (P27169), and thrombospondin-1 (P07996). In our current study, the SRM detection of the peptide from the protein anthrax toxin receptor 1 (DFDETQLAR) was inconsistent, and we discarded it from the multivariate signature in this manuscript. Therefore, the multivariate signature in this work is composed of six proteotypic N-glycopeptides (Fig. 1A; Supplementary Table S2). A query of the neXtProt database (v2.23.2; ref. 46) revealed the involvement of the signature proteins in several processes important for the survival and the growth of the cancer like cell adhesion and migration, vascularization as well as metabolic processes and response to hypoxia. The STRING database (v11.0; ref. 47) contained no evidence of direct or functional association between the proteins. At the same time, thrombospondin-1 was reported to coexpress with the proteins basement membrane-specific heparan sulfate proteoglycan core protein (RNA coexpression score 0.108) and intercellular adhesion molecule 1 (RNA coexpression score 0.07). Further, we used the Protein Universal Reference Publication-Originated Search Engine (PURPOSE; ref. 48) to investigate the literature for reports of the signature proteins in relation to MPM (we used the following inputs: query type: custom search; input custom topic: mesothelioma; species: human). As expected, mesothelin had the highest number of literature reports (PURPOSE score 39.905) followed by thrombospondin-1 (score 10.403) and intercellular adhesion molecule 1 (score 9.478). We found no reports for the remaining proteins. This information underscored the independence of our approach to MPM biomarker verification from prior knowledge of proteins function. The multiplexing offered by MS-based targeted proteomics allowed us to combine proteins previously under-investigated in MPM within a single signature of cancer phenotype.

We applied an academically developed targeted proteomics workflow based on LC-SRM and performed in our translational laboratory. Samples were injected in duplicate, the mass spectrometry runs were manually examined for quality (as reported in Supplementary methods, online) and low-quality runs were removed. The procedure retained a total of 775 mass spectrometry runs. Aliquots from one healthy donor (HD) were processed in parallel with the serum samples from MPM and asbestos-exposed subjects, and a total of 21 HD replicates were available for the analysis. We used these HD samples to assess the variability of our workflow. The endogenous peptides of the multivariate signature presented coefficient of variations (CV) between 29.4% (peptide VIDETWAWK) and 48.8% (peptide VVDSTTGPGEHLR) among the HD samples (Supplementary Fig. S1). Examples of SRM elution traces of the signature peptides are reported in Supplementary Fig. S2. Skyline and raw MS files are accessible from the ProteomeXchange consortium (49) via Panorama Public (https://panoramaweb.org/meso20.url; ProteomeXchange ID: PXD018457; ref. 50).

Upon normalization and summarization of the quantified transitions into peptide-level abundances, we investigated the performance of the multivariate targeted proteomic signature in serum for discriminating between patients with MPM and asbestos-exposed donors. An unsupervised hierarchical clustering algorithm revealed no clustering of the samples with respect to the biobank location, making biobank associated classification biases less likely (Supplementary Fig. S3). We used the training set to estimate the parameters of a multivariate signature based on logistic regression, with the targeted proteins as predictors (Supplementary Figs. S4 and S5). In the testing set, the signature had an area under the receiver-operating characteristic curve (AUC) of 0.738 [99% confidence interval (CI), 0.640–0.830; Fig. 1B, Table 2]. The signature had an AUC of 0.765 (99% CI; 0.648–0.863) for discriminating specifically early-stage MPM (n = 45) from asbestos-exposed subjects (Fig. 1C, Table 2). We observed that the signature composed of multiple biomarkers had an improved predictive ability compared with the logistic regressions with single-peptide biomarkers quantified on the same LC-SRM platform. Indeed, the predictive ability of the single biomarker peptide KWDVTSLETLK from the cleaved part of the mesothelin protein was significantly inferior compared with that of the multiple biomarkers signature (Fig. 1B and C, Table 2; Supplementary Fig. S6), although it had the highest AUC in the training set when using only one biomarker at a time as a classifier (Supplementary Fig. S7).

Next, we used the FDA-approved SMRP ELISA to compare the performance of our academically developed targeted proteomics workflow with a reference assay already optimized for clinical applications. We performed SMRP ELISA measurements for the samples of the testing cohort (Supplementary Tables S4 and S5). SMRP discriminated patients with MPM from asbestos-exposed donors with an AUC of 0.795 (99% CI, 0.708–0.869) and for the early stages MPM, the AUC was 0.744 (99% CI, 0.612–0.869; Fig. 1B and C, Table 2). These AUCs obtained with the SMRP ELISA were not significantly different from those observed for the proteomic signature (P = 0.201 for early and late stages MPM and P = 0.741 for the early stage MPM subgroup, Fig. 1B and C). Overall, the two methods showed a significant although limited correlation for MPM or asbestos-exposed category assignment (Pearson correlation r = 0.386, P < 0.0001, Fig. 2). At a 2 nmol/L cutoff proposed from the literature (19), the SMRP ELISA had specificity above 95% for both groups of all stages (early and advanced) and early stages MPM (Table 2). The specificity of the proteomic signature was lower (71%), but it presented a higher sensitivity (Table 2) in particular for the early stages MPM (62%; cutoff of the signature was chosen to maximize the predictive accuracy in the training set). The single mesothelin peptide KWDVTSLETLK by LC-SRM, if assessed alone, reflected the overall abundance distribution observed for the SMRP ELISA between the two groups of MPM and asbestos-exposed subjects (although it did not show significant differences between early and advanced MPM stages, Supplementary Fig. S8A–S8D), but the correlation between the two methods was limited (r = 0.410, P < 0.0001, Supplementary Fig. S9). We used a logistic regression model, to evaluate whether the combination of the multivariate proteomic signature together with the SMRP ELISA, could improve the discrimination between patients with MPM and asbestos-exposed donors from the testing cohort. After cross-validation, the AUC of the integrated model did not differ significantly from that of the proteomic signature or the SMRP ELISA separately (Supplementary Fig. S10), although the cohort was small and possibly insufficient to conclusively answer the question.

To assess the diagnostic values of the tests, we calculated the posttest probabilities of the multivariate proteomic signature, the SMRP ELISA and also of the single mesothelin peptide KWDVTSLETLK by LC-SRM (Table 3; definitions of the quantities are reported in the Supplementary Methods, online; for the ELISA we also report results based on the literature proposing a 2 nmol/L cutoff). The positive likelihood ratio (LR +; at 95% specificity threshold) for the proteomic signature was 5.93 and for the SMRP ELISA was 8.6. The negative likelihood ratio (LR-; at 95% sensitivity) was 0.33 for the proteomic signature and 0.2 for the ELISA. If we consider a pretest probability of disease of 25% or 50% [chosen as previously proposed from the literature (19)], the posttest probabilities for a positive result using the proteomics signature are 66.4% and 85.6% respectively, and for a negative result are of 9.8% and 24.6%. Using the ELISA, posttest probabilities are 74.1% and 89.6% for a positive result and 6.2% and 16.6% for a negative result. Interestingly, for the subgroup of early-stage MPM the LR+ of the proteomic signature was 4.9, still lower than that of the ELISA which was 7.1, but the LR- of the proteomic signature was 0.11 whereas that of the ELISA was 0.25. Considering again a pre-test probability of disease of 25% or 50%, the post-test probabilities using the proteomics signature would be 3.6% and 10.1%, respectively, whereas those of the ELISA would be higher with 7.7% and 20% or even higher (19.2% and 41.7%) if applying the 2 nmol/L cutoff (instead of the 95% sensitivity threshold). These results confirmed the higher specificity of the SMRP ELISA assay for MPM and highlight at the same time the increased sensitivity of the proteomic signature for MPM at early stages.

Furthermore, we investigated whether the biological content of the proteomic signature may also be informative about the prognosis of MPM. To do so, we performed a survival analysis fitting Cox proportional hazard models based on MPM samples in the training set and using the six LC-SRM quantified candidate biomarker peptides as predictors (Fig. 3A; Supplementary Methods, Supplementary Fig. S11; Supplementary Table S5, online). We calculated the prognostic index (PI) for each subject in the training set and used the obtained median PI as the arbitrary cutoff (−0.03476) for assigning survival risk groups in both the training and testing set (low risk group having PI below the cutoff, and high-risk group having PI above the cutoff). Forty-seven patients with MPM of the testing set were assigned to the low-risk group and 61 to the high-risk group. Kaplan–Meier survival curves were significantly different between the two groups (P = 0.021 by log-rank test). The HR for risk group was 1.659 (95% CI, 1.075–2.562; P = 0.021 by Wald test; Fig. 3B). The median survival for patients of the low-risk group was 21 months (95% CI, 17–31) with an average survival of 31 months (SE = 3.63), whereas the median survival in the high-risk group was 19 months (95% CI, 17–24) with an average survival of 20.9 months (SE = 1.69). Patients within the low-risk group of the testing set were treated with surgery in 36 cases (in 29 cases in combination solely with chemotherapy and in three cases with the addition of chemotherapy and radiotherapy), with chemotherapy alone in nine cases and two patients did not receive any of the treatments. In the high-risk group, surgery was performed in 41 cases (32 patients receiving surgery and solely chemotherapy and 7 patients receiving surgery, chemotherapy, and radiotherapy), chemotherapy alone was given to 18 patients, one patient received chemotherapy together with radiotherapy and for one case, no information was reported about treatment. Furthermore, we developed a survival prediction model using age, stage, and risk group assignment in testing set and which is intended to be an example on how the signature could be used for prognostic stratification of patients with MPM (Fig. 3C; Supplementary Methods, online). Together, these results highlighted the possibility to apply the targeted proteomic signature for the stratification of patients with MPM based on their predicted survival.

In our work, we have investigated the application of a MS-based targeted proteomics workflow for the detection and the prognosis of MPM from the blood. To do so, we have applied an adapted version of our previously proposed targeted proteomic MPM signature (44) in one of the largest clinical blood sample collections for proteomics or MPM biomarker studies. Unsupervised analysis confirmed that each of the proteins composing the signature had independent classification ability with no apparent functional or physical relation among each other. Compared with standard biomarkers methods used in the clinic, the multiplexing possibility of targeted proteomics allowed grouping them into a single signature with better diagnostic potential than any of the proteins alone. The MS-based proteomics approach offers in this way a strategy to increase the biological information gleaned about the cancer and which could be important in particular in the current era of biology-driven personalized medicine. Still, the translation of MS-based proteomics for clinical oncology is proceeding only at a slow pace. This may also be related to the perception that proteomics relies on a sophisticated technology too complex for routine applications and accessible only within highly dedicated laboratories. It was therefore important for us to demonstrate with our study that larger clinical investigations can be approached by MS-based proteomics also within academic translational laboratories with no particular focus on the technology. This becomes relevant in particular when we observe the trend of other medical fields, which have now adopted mass spectrometers as their standard diagnostics. Newborn screening for metabolic disorders may be an example (51) or the routine use of MALDI-TOF (matrix assisted laser desorption ionization- time of flight) in clinical microbiology for microorganism identification (52).

In our work, we have compared the performance of the proteomic signature with that of the FDA-approved ELISA for the best investigated MPM blood biomarker SMRP. We did so because we wanted to anchor the current status of our academically developed proteomics workflow (with all the limits of an academic effort) to that of an assay already meeting clinical standards. We thought that was important to further guide the translation of the proteomics technology for routine clinical use. In our study, we could observe that the two methods did not differ significantly in their ability to discriminate MPM and asbestos-exposed subjects. This was a very promising result, considering the highly different level of technical optimization of the two platforms. Indeed, while the commercial ELISA has already been developed into a robust clinical assay (38), the CVs of the proteomic workflow, for example, are still relatively high. Also, interfering signals were still encountered for the peptides of the signature. For the mesothelin peptide KWDVTSLETLK, for example, we repeatedly observed a coeluting peak (Supplementary Fig. S2G and S2H), which could possibly be caused by an interfering signal or maybe by the oxidation of tryptophan (W) during sample processing and storage. Therefore, additional optimization will be needed before the proteomic signature could be applied for the clinic. We think that in particular, it will be crucial to further establish a standardized and robust protocol for the large-scale processing of the blood samples as well as an analytical LC-MS workflow tailored specifically for the peptides composing the signature (53). The technical limits at the current status of our academically developed proteomic workflow may also be at the base of the limited correlation we observed between the SMRP ELISA measurements and the quantification of mesothelin by LC-SRM. This observation seems to be in apparent contrast with the results from our previous work, where the two methods showed a better correlation of 0.83 (44), although here, we have to highlight that the cohort we used in our current work was much larger and more complex than the one used previously. Indeed, differences in mesothelin level are expected to be less pronounced between patients with MPM and asbestos-exposed (54) and the different composition of the current cohorts from multiple international sites and storage duration within the current study may have contributed to this reduced correlation too.

Beyond its technical characteristics, the proteomic signature presented an interesting property which, if confirmed, may become relevant for the early detection of MPM. Indeed, we observed that the signature had a higher sensitivity for the detection of early-stage MPM and a lower LR- of 0.11, compared with the SMRP biomarker. It may be speculated that this was possible due to the multiple biological information gathered within the proteomic signature and which have made possible to capture systemic events induced early by the cancer. But more importantly, in this way the proteomic signature could become an important complement to compensate for the limited sensitivity of the SMRP ELISA test and which currently restricts its clinical use. Indeed, the higher specificity of the SMRP ELISA was confirmed also within our study and the LR+ was actually even higher than the one reported in the meta-analysis of Hollevoet and colleagues (19). At the same time, the sensitivity of the assay remained lower also within our cohorts and the LR- for early stages MPM was limited, with a value of 0.25 (similar to what was reported in Hollevoet and colleagues). Indeed, considering that MPM is a highly aggressive disease with a poor prognosis, the better specificity and LR+ of the SMRP ELISA test are important to avoid wrong assignment for people who in reality are not affected by MPM (false positives). At the same time, the lower sensitivity and the limited LR- hamper the use of the SMRP ELISA test to rule out the disease among people with a negative test result because of the risk of false negatives. Still, this information would be relevant for example for the interpretation of radiologic exams from people at risk and also in the context of the increasing use of CT screenings for lung cancer, which will reveal findings not only limited to lung cancer. Radiologic false positives or ambiguous results not only may trigger unnecessary invasive diagnostics or have a negative economic impact, but they can also cause significant psychological harm for the person who become confronted with the diagnosis of a highly deadly disease (55–57). MPM has a long asymptomatic latency period and its radiologic presentation is complex (58–62). The proteomic signature could therefore represent an additional noninvasive tool to help rule out the disease among people with ambiguous radiologic findings. Indeed, the lower LR- observed for the proteomic signature may increase the confidence that a negative test result is also reflecting the true absence of the disease (true negative). The integration of the higher specificity of the SRMP ELISA together with the better sensitivity offered by the proteomic signature may therefore become an interesting approach for the early detection of patients with MPM in particular within populations at risk for the cancer. Unfortunately, although we tried in our study to combine the two tests using a logistic regression model, we did not observe the expected benefit. Still, we think that our observation may not be conclusive because our cohort was too small to appropriately address this question and an additional investigation on a larger number of contemporary samples would be required for a definitive answer. In addition, the proteomic signature demonstrated a prognostic potential, which is not seen with the commercial assay, and which could be important for treatment decisions. A more radical intervention like EPP together with chemotherapy could for example be an option for patients with MPM with a better survival as predicted by the signature, whereas patients with a predicted poor prognosis might be selected for more aggressive or experimental treatments.

In summary, we presented an approach based on targeted proteomics, which permitted to combine multiple independent units of information about the MPM cancer into a single informative signature from the blood. The diagnostic accuracy of our academically developed proteomic signature was comparable with standard biomarker assays already optimized for clinical use. Its increased sensitivity for the detection of early-stage MPM may complement current diagnostic approaches for monitoring people at risk for the cancer. Its prognostic ability may support patient's stratification and treatment selection. We believe that the experience reported in this exploratory study represents an additional diagnostic approach for informing clinical decisions for patients at risk for MPM using MS-based proteomics in blood.

F. Cerciello reports grants from Swiss National Science Foundation (SNSF) during the conduct of the study. S.L. Sinicropi-Yao reports a post-doc position from Sanofi outside the submitted work. B.W.S. Robinson reports grants from National Health and Medical Research Council during the conduct of the study. J. Creaney reports grants from National Health and Medical Research Council outside the submitted work. H.I. Pass reports grants from NCI/NIH Early Detection Research Network during the conduct of the study and outside the submitted work. D.P. Carbone reports personal fees from AbbVie, Adaptimmune, Agenus, Amgen, Ariad, AstraZeneca, Biocept, and Boehringer Ingelheim; grants and personal fees from Bristol-Myers Squibb (research grant); and personal fees from Celgene, Clovis, Daiichi Sankyo, EMD Serono, Foundation Medicine, GenePlus, Genentech/Roche, GlaxoSmithKline, Gloria BioScience, Gritstone, Guardant Health, Humana, Incyte, Inivata, Inovio, Janssen, G1 Therapeutics, Kyowa Kirin, Loxo Oncology, Merck, MSD, Nexus Oncology, Novartis, Oncocyte, Palobiofarma, Pfizer, prIME Oncology, Stemcentrx, Takeda Oncology, and Teva outside the submitted work. No potential conflicts of interest were disclosed by the other authors.

F. Cerciello: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. M. Choi: Conceptualization, resources, data curation, software, formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. S.L. Sinicropi-Yao: Investigation, writing–original draft. K. Lomeo: Investigation. J.M. Amann: Resources, investigation, writing–original draft, project administration. E. Felley-Bosco: Resources, data curation, writing–original draft, writing–review and editing. R.A. Stahel: Resources, data curation. B.W.S. Robinson: Resources, data curation. J. Creaney: Resources, data curation, writing–original draft. H.I. Pass: Resources, data curation. O. Vitek: Conceptualization, resources, data curation, software, formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. D.P. Carbone: Conceptualization, resources, data curation, supervision, funding acquisition, validation, investigation, writing–original draft, project administration, writing–review and editing.

This work was supported by the Kathy and Jay Worly Lung Cancer Early Detection Fund and by Don Ward, President, Special Claims Services, Inc. at The Ohio State University Comprehensive Cancer Center (OSUCCC) - The James (to D.P. Carbone). F. Cerciello was recipient of a postdoc mobility fellowship from the Swiss National Science Foundation. Specimens collected from New York University were funded under U01CA 111295-04 (H.I. Pass, principal investigator) from the Early Detection Research Network, NCI, NIH. The authors thank Dr. Nathalie Selevsek at the Swiss Integrative Center for Human Health (SICHH), Fribourg, Switzerland, for her critical inputs on the analysis and interpretation of SRM traces.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.