Rachel Ostroff

Unlocking Biomarker Discovery: Large Scale Application of Aptamer Proteomic Technology for Early Detection of Lung Cancer

Background Lung cancer is the leading cause of cancer deaths worldwide. New diagnostics are needed to detect early stage lung cancer because it may be cured with surgery. However, most cases are diagnosed too late for curative surgery. Here we present a comprehensive clinical biomarker study of lung cancer and the first large-scale clinical application of a new aptamer-based proteomic technology to discover blood protein biomarkers in disease. Methodology/Principal Findings We conducted a multi-center case-control study in archived serum samples from 1,326 subjects from four independent studies of non-small cell lung cancer (NSCLC) in long-term tobacco-exposed populations. Sera were collected and processed under uniform protocols. Case sera were collected from 291 patients within 8 weeks of the first biopsy-proven lung cancer and prior to tumor removal by surgery. Control sera were collected from 1,035 asymptomatic study participants with ≥10 pack-years of cigarette smoking. We measured 813 proteins in each sample with a new aptamer-based proteomic technology, identified 44 candidate biomarkers, and developed a 12-protein panel (cadherin-1, CD30 ligand, endostatin, HSP90α, LRIG3, MIP-4, pleiotrophin, PRKCI, RGM-C, SCF-sR, sL-selectin, and YES) that discriminates NSCLC from controls with 91% sensitivity and 84% specificity in cross-validated training and 89% sensitivity and 83% specificity in a separate verification set, with similar performance for early and late stage NSCLC. Conclusions/Significance This study is a significant advance in clinical proteomics in an area of high unmet clinical need. Our analysis exceeds the breadth and dynamic range of proteome interrogated of previously published clinical studies of broad serum proteome profiling platforms including mass spectrometry, antibody arrays, and autoantibody arrays. The sensitivity and specificity of our 12-biomarker panel improves upon published protein and gene expression panels. Separate verification of classifier performance provides evidence against over-fitting and is encouraging for the next development phase, independent validation. This careful study provides a solid foundation to develop tests sorely needed to identify early stage lung cancer.

Early Detection of Malignant Pleural Mesothelioma in Asbestos-Exposed Individuals with a Noninvasive Proteomics-Based Surveillance Tool

Background Malignant pleural mesothelioma (MM) is an aggressive, asbestos-related pulmonary cancer that is increasing in incidence. Because diagnosis is difficult and the disease is relatively rare, most patients present at a clinically advanced stage where possibility of cure is minimal. To improve surveillance and detection of MM in the high-risk population, we completed a series of clinical studies to develop a noninvasive test for early detection. Methodology/Principal Findings We conducted multi-center case-control studies in serum from 117 MM cases and 142 asbestos-exposed control individuals. Biomarker discovery, verification, and validation were performed using SOMAmer proteomic technology, which simultaneously measures over 1000 proteins in unfractionated biologic samples. Using univariate and multivariate approaches we discovered 64 candidate protein biomarkers and derived a 13-marker random forest classifier with an AUC of 0.99±0.01 in training, 0.98±0.04 in independent blinded verification and 0.95±0.04 in blinded validation studies. Sensitivity and specificity at our pre-specified decision threshold were 97%/92% in training and 90%/95% in blinded verification. This classifier accuracy was maintained in a second blinded validation set with a sensitivity/specificity of 90%/89% and combined accuracy of 92%. Sensitivity correlated with pathologic stage; 77% of Stage I, 93% of Stage II, 96% of Stage III and 96% of Stage IV cases were detected. An alternative decision threshold in the validation study yielding 98% specificity would still detect 60% of MM cases. In a paired sample set the classifier AUC of 0.99 and 91%/94% sensitivity/specificity was superior to that of mesothelin with an AUC of 0.82 and 66%/88% sensitivity/specificity. The candidate biomarker panel consists of both inflammatory and proliferative proteins, processes strongly associated with asbestos-induced malignancy. Significance The SOMAmer biomarker panel discovered and validated in these studies provides a solid foundation for surveillance and diagnosis of MM in those at highest risk for this disease.

Protein signature of lung cancer tissues.

Lung cancer remains the most common cause of cancer-related mortality. We applied a highly multiplexed proteomic technology (SOMAscan) to compare protein expression signatures of non small-cell lung cancer (NSCLC) tissues with healthy adjacent and distant tissues from surgical resections. In this first report of SOMAscan applied to tissues, we highlight 36 proteins that exhibit the largest expression differences between matched tumor and non-tumor tissues. The concentrations of twenty proteins increased and sixteen decreased in tumor tissue, thirteen of which are novel for NSCLC. NSCLC tissue biomarkers identified here overlap with a core set identified in a large serum-based NSCLC study with SOMAscan. We show that large-scale comparative analysis of protein expression can be used to develop novel histochemical probes. As expected, relative differences in protein expression are greater in tissues than in serum. The combined results from tissue and serum present the most extensive view to date of the complex changes in NSCLC protein expression and provide important implications for diagnosis and treatment.

Validation of a blood protein signature for non-small cell lung cancer

BackgroundCT screening for lung cancer is effective in reducing mortality, but there are areas of concern, including a positive predictive value of 4% and development of interval cancers. A blood test that could manage these limitations would be useful, but development of such tests has been impaired by variations in blood collection that may lead to poor reproducibility across populations.ResultsBlood-based proteomic profiles were generated with SOMAscan technology, which measured 1033 proteins. First, preanalytic variability was evaluated with Sample Mapping Vectors (SMV), which are panels of proteins that detect confounders in protein levels related to sample collection. A subset of well collected serum samples not influenced by preanalytic variability was selected for discovery of lung cancer biomarkers. The impact of sample collection variation on these candidate markers was tested in the subset of samples with higher SMV scores so that the most robust markers could be used to create disease classifiers. The discovery sample set (n = 363) was from a multi-center study of 94 non-small cell lung cancer (NSCLC) cases and 269 long-term smokers and benign pulmonary nodule controls. The analysis resulted in a 7-marker panel with an AUC of 0.85 for all cases (68% adenocarcinoma, 32% squamous) and an AUC of 0.93 for squamous cell carcinoma in particular. This panel was validated by making blinded predictions in two independent cohorts (n = 138 in the first validation and n = 135 in the second). The model was recalibrated for a panel format prior to unblinding the second cohort. The AUCs overall were 0.81 and 0.77, and for squamous cell tumors alone were 0.89 and 0.87. The estimated negative predictive value for a 15% disease prevalence was 93% overall and 99% for squamous lung tumors. The proteins in the classifier function in destruction of the extracellular matrix, metabolic homeostasis and inflammation.ConclusionsSelecting biomarkers resistant to sample processing variation led to robust lung cancer biomarkers that performed consistently in independent validations. They form a sensitive signature for detection of lung cancer, especially squamous cell histology. This non-invasive test could be used to improve the positive predictive value of CT screening, with the potential to avoid invasive evaluation of nonmalignant pulmonary nodules.

Life's simple measures: unlocking the proteome.

Using modified nucleotides and selecting for slow off-rates in the SELEX procedure, we have evolved a special class of aptamers, called SOMAmers (slow off-rate modified aptamers), which bind tightly and specifically to proteins in body fluids. We use these in a novel assay that yields 1:1 complexes of the SOMAmers with their cognate proteins in body fluids. Measuring the SOMAmer concentrations of the resultant complexes reflects the concentration of the proteins in the fluids. This is simply done by hybridization to complementary sequences on solid supports, but it can also be done by any other DNA quantification technology (including NexGen sequencing). We use measurements of over 1000 proteins in under 100 μL of serum or plasma to answer important medical questions, two of which are reviewed here. A number of bioinformatics methods have guided our discoveries, including principal component analysis. We use various methods to evaluate sample handling procedures in our clinical samples and can identify many parameters that corrupt proteomics analysis.

Discovery and Validation of a Six-Marker Serum Protein Signature for the Diagnosis of Active Pulmonary Tuberculosis

ABSTRACT New non-sputum biomarker tests for active tuberculosis (TB) diagnostics are of the highest priority for global TB control. We performed in-depth proteomic analysis using the 4,000-plex SOMAscan assay on 1,470 serum samples from seven countries where TB is endemic. All samples were from patients with symptoms and signs suggestive of active pulmonary TB that were systematically confirmed or ruled out for TB by culture and clinical follow-up. HIV coinfection was present in 34% of samples, and 25% were sputum smear negative. Serum protein biomarkers were identified by stability selection using L1-regularized logistic regression and by Kolmogorov-Smirnov (KS) statistics. A naive Bayes classifier using six host response markers (HR6 model), including SYWC, kallistatin, complement C9, gelsolin, testican-2, and aldolase C, performed well in a training set (area under the sensitivity-specificity curve [AUC] of 0.94) and in a blinded verification set (AUC of 0.92) to distinguish TB and non-TB samples. Differential expression was also highly significant (P < 10−20) for previously described TB markers, such as IP-10, LBP, FCG3B, and TSP4, and for many novel proteins not previously associated with TB. Proteins with the largest median fold changes were SAA (serum amyloid protein A), NPS-PLA2 (secreted phospholipase A2), and CA6 (carbonic anhydrase 6). Target product profiles (TPPs) for a non-sputum biomarker test to diagnose active TB for treatment initiation (TPP#1) and for a community-based triage or referral test (TPP#2) have been published by the WHO. With 90% sensitivity and 80% specificity, the HR6 model fell short of TPP#1 but reached TPP#2 performance criteria. In conclusion, we identified and validated a six-marker signature for active TB that warrants diagnostic development on a patient-near platform.

Development of a Protein Biomarker Panel to Detect Non–Small-Cell Lung Cancer in Korea

Background Lung cancer screening using low‐dose computed tomography reduces lung cancer mortality. However, the high false‐positive rate, cost, and potential harms highlight the need for complementary biomarkers. We compared the diagnostic performance of modified aptamer‐based protein biomarkers with Cyfra 21‐1. Patients and Methods Participants included 100 patients diagnosed with lung cancer, and 100 control subjects from Asan Medical Center (Seoul, Korea). We investigated candidate biomarkers with new modified aptamer‐based proteomic technology and developed a 7‐protein panel that discriminates lung cancer from controls. A naive Bayesian classifier was trained using sera from 75 lung cancers and 75 controls. An independent set of 25 cases and 25 controls was used to verify performance of this classifier. The panel results were compared with Cyfra 21‐1 to evaluate the diagnostic accuracy for lung nodules detected by computed tomography. Results We derived a 7‐protein biomarker classifier from the initial train set comprising: EGFR1, MMP7, CA6, KIT, CRP, C9, and SERPINA3. This classifier distinguished lung cancer cases from controls with an area under the curve (AUC) of 0.82 in the train set and an AUC of 0.77 in the verification set. The 7‐marker naive Bayesian classifier resulted in 91.7% specificity with 75.0% sensitivity for the subset of individuals with lung nodules. The AUC of the classifier for lung nodules was 0.88, whereas Cyfra 21‐1 had an AUC of 0.72. Conclusion We have developed a protein biomarker panel to identify lung cancers from controls with a high accuracy. This integrated noninvasive approach to the evaluation of lung nodules deserves further prospective validation among larger cohorts of patients with lung nodules in screening strategy. Micro‐Abstract We investigated candidate biomarkers for the Korean population using a new aptamer‐based proteomic technology and developed a 7‐protein panel that discriminates lung cancer from controls. This 7‐marker panel was able to discriminate lung cancer from controls with an area under the curve of 0.82/0.77 in the train/verification set respectively. This panel may have clinical utility in risk‐stratifying screen‐detected lung nodules.

An integrated approach to blood-based cancer diagnosis and biomarker discovery.

Disrupted or abnormal biological processes responsible for cancers often quantitatively manifest as disrupted additive and multiplicative interactions of gene/protein expressions correlating with cancer progression. However, the examination of all possible combinatorial interactions between gene features in most case-control studies with limited training data is computationally infeasible. In this paper, we propose a practically feasible data integration approach, QUIRE (QUadratic Interactions among infoRmative fEatures), to identify discriminative complex interactions among informative gene features for cancer diagnosis and biomarker discovery directly based on patient blood samples. QUIRE works in two stages, where it first identifies functionally relevant gene groups for the disease with the help of gene functional annotations and available physical protein interactions, then it explores the combinatorial relationships among the genes from the selected informative groups. Based on our private experimentally generated data from patient blood samples using a novel SOMAmer (Slow Off-rate Modified Aptamer) technology, we apply QUIRE to cancer diagnosis and biomarker discovery for Renal Cell Carcinoma (RCC) and Ovarian Cancer (OVC). To further demonstrate the general applicability of our approach, we also apply QUIRE to a publicly available Colorectal Cancer (CRC) dataset that can be used to prioritize our SOMAmer design. Our experimental results show that QUIRE identifies gene-gene interactions that can better identify the different cancer stages of samples, as compared to other state-of-the-art feature selection methods. A literature survey shows that many of the interactions identified by QUIRE play important roles in the development of cancer.

Improving Assessment of Drug Safety Through Proteomics: Early Detection and Mechanistic Characterization of the Unforeseen Harmful Effects of Torcetrapib

Background: Early detection of adverse effects of novel therapies and understanding of their mechanisms could improve the safety and efficiency of drug development. We have retrospectively applied large-scale proteomics to blood samples from ILLUMINATE (Investigation of Lipid Level Management to Understand its Impact in Atherosclerotic Events), a trial of torcetrapib (a cholesterol ester transfer protein inhibitor), that involved 15 067 participants at high cardiovascular risk. ILLUMINATE was terminated at a median of 550 days because of significant absolute increases of 1.2% in cardiovascular events and 0.4% in mortality with torcetrapib. The aims of our analysis were to determine whether a proteomic analysis might reveal biological mechanisms responsible for these harmful effects and whether harmful effects of torcetrapib could have been detected early in the ILLUMINATE trial with proteomics. Methods: A nested case-control analysis of paired plasma samples at baseline and at 3 months was performed in 249 participants assigned to torcetrapib plus atorvastatin and 223 participants assigned to atorvastatin only. Within each treatment arm, cases with events were matched to controls 1:1. Main outcomes were a survey of 1129 proteins for discovery of biological pathways altered by torcetrapib and a 9-protein risk score validated to predict myocardial infarction, stroke, heart failure, or death. Results: Plasma concentrations of 200 proteins changed significantly with torcetrapib. Their pathway analysis revealed unexpected and widespread changes in immune and inflammatory functions, as well as changes in endocrine systems, including in aldosterone function and glycemic control. At baseline, 9-protein risk scores were similar in the 2 treatment arms and higher in participants with subsequent events. At 3 months, the absolute 9-protein derived risk increased in the torcetrapib plus atorvastatin arm compared with the atorvastatin-only arm by 1.08% (P=0.0004). Thirty-seven proteins changed in the direction of increased risk of 49 proteins previously associated with cardiovascular and mortality risk. Conclusions: Heretofore unknown effects of torcetrapib were revealed in immune and inflammatory functions. A protein-based risk score predicted harm from torcetrapib within just 3 months. A protein-based risk assessment embedded within a large proteomic survey may prove to be useful in the evaluation of therapies to prevent harm to patients. Clinical Trial Registration: URL: https://www.clinicaltrials.gov. Unique identifier: NCT00134264.

Factorized sparse learning models with interpretable high order feature interactions

Identifying interpretable discriminative high-order feature interactions given limited training data in high dimensions is challenging in both machine learning and data mining. In this paper, we propose a factorization based sparse learning framework termed FHIM for identifying high-order feature interactions in linear and logistic regression models, and study several optimization methods for solving them. Unlike previous sparse learning methods, our model FHIM recovers both the main effects and the interaction terms accurately without imposing tree-structured hierarchical constraints. Furthermore, we show that FHIM has oracle properties when extended to generalized linear regression models with pairwise interactions. Experiments on simulated data show that FHIM outperforms the state-of-the-art sparse lear-ning techniques. Further experiments on our experimentally generated data from patient blood samples using a novel SOMAmer (Slow Off-rate Modified Aptamer) technology show that, FHIM performs blood-based cancer diagnosis and bio-marker discovery for Renal Cell Carcinoma much better than other competing methods, and it identifies interpretable block-wise high-order gene interactions predictive of cancer stages of samples. A literature survey shows that the interactions identified by FHIM play important roles in cancer development.