Multi-Omics Profiling With Untargeted Proteomics for Blood-Based Early Detection of Lung Cancer

As lung cancer stands as the leading cause of cancer-related deaths in the United States, ensuring the effective screening of high-risk, asymptomatic individuals for lung cancer cannot be understated to improve early detection, shift initial diagnoses to earlier stages, and reduce the mortality rate of the disease.

While a variety of blood-based cancer detection methods using individual ‘omics techniques have been explored, studying lung cancer presents unique challenges due to the complex etiological factors contributing to the disease. With the advent of technology that supports large-scale untargeted plasma proteomic studies, PrognomiQ, Inc. a healthcare company pioneering transformative test products that enable early disease detection and treatment, embarked on a multi-omics discovery study on a scale not previously reported or attempted.

Their main objective was to interrogate the biological phenomics landscape of blood plasma to identify, with high specificity and sensitivity, new biomarkers capable of detecting early-stage lung cancer at a point where clinical interventions can still impact overall survival.

Shedding new light into how we diagnose lung cancer and improve clinical outcomes, this scientific endeavor is one of the largest multi-omics observational studies underway with Seer’s Proteograph™ Product Suite providing extensive, untargeted plasma proteomics analysis at a scale previously not enabled by any other unbiased proteomics platform.

The Study Design

A lung cancer case-control cohort of 2513 subjects – with and without lung cancer, including those with non-malignant comorbid conditions – was enrolled across 77 clinical sites.

Three blood samples were collected from each subject and then used for proteomics, RNA-seq, metabolomics, and targeted immunoassays. Targeted immunoassay data focusing on 4 proteins (CA125a, CA15-3, CEA, CA19-9) were collected on all subjects.

From there, sample processing and liquid chromatography-mass spectrometry (LC-MS) proteomics data acquisition was conducted, with 2094 K2 EDTA plasma samples passing clinical eligibility requirements and being processed by Seer’s Proteograph Assay and 5-nanoparticle panel workflow. For the development of a machine learning-based classifier for lung cancer detection, the data was divided into training and validation sets to evaluate the sensitivity and specificity of the 682 multi-omics analyte features comprising the classifier model.

The Results

The high performance of PrognomiQ’s novel multi-omics method offered such unprecedented breadth and depth that classifiers trained only with untargeted proteomics features still achieved an AUC > 0.91 for all-stage lung cancer. The cancer classifier was further improved by combining additional ‘omics features and demonstrated 89%, 80%, and 98-100% sensitivity for all-stage, stage I, and stage III-IV lung cancer, respectively, at 89% specificity in a validation set.

What this study reveals is the true power of multi-omics. By integrating quantitative measurements of proteins, transcripts, and metabolites, we are not just looking at cancer in peripheral blood samples from one but multiple angles. And, if different aspects of the disease are captured by different ‘omics approaches, we are better positioned to capture the full complexity of the pathology, which can potentially lead to early-stage cancer detection and improve survival rates and the quality of life for those battling this devastating disease.

This landmark study highlights the potential broad clinical utility of a multi-omics approach and signals a paradigm shift to an anticipated growing number of population studies that will collect peripheral blood samples to more fully characterize additional complex diseases.