Predicting abdominal aortic aneurysms by combining mass spectrometry-based proteomics with clinical data

Background

Abdominal aortic aneurysms (AAA) are in most cases asymptomatic and pose a life-threatening risk due to unexpected rupture of the aortic wall. AAA are most often detected accidently when patients undergo CT scans for other purposes, and in many countries AAA screening is considered too costly. Therefore, a simple blood test discriminating individuals with an AAA from non-AAA individuals is urgently needed.

Aim

The aim of the present study was to identify specific plasma proteins that discriminate individuals with an AAA from non-AAA individuals by applying mass spectrometry-based proteomics.

Methods

Our study involved 7082 participants, including 545 patients with AAA, and 6537 control subjects from the population-based Danish Cardiovascular Screening (DANCAVAS) trial. Plasma samples were prepared and labelled with 18-plex TMTpro prior to LC-MS/MS analysis using data-dependent acquisition on an Orbitrap Eclipse mass spectrometer. Data processing was performed using a combined Sequest HT, MSPepSearch, CHIMERYS®, and MS Amanda search in Proteome Discoverer 3.0. For MSPepSearch our recently developed TMTpro-specific spectral library was applied. For statistical model development, participants were randomly divided into either a training set (80%) or a test set (20%). Gene Ontology enrichment analysis of significant differentially regulated proteins was made with ShinyGo (v. 0.80). Logistic regression combining proteomics data and clinical data was used for the identification of individuals with AAA. Performance of our model was measured using the area under the curve (AUC) for the detection of AAAs on the independent test set.

Results

Mass spectrometry-based proteomics identified 46 proteins as significantly differentially regulated between AAA patients and control subjects (all FDR adj. p < 0.05). Up-regulated pathways included complement and coagulation cascades, cholesterol metabolism, and platelet activation, while vitamin digestion and absorption, cholesterol metabolism and fat digestion and absorption were among the down-regulated pathways. A standard model trained on clinical variables only achieved an AUC = 0.81. Inclusion of a panel of twenty-two proteins in our model significantly improved the ability to discriminate between control subjects and AAA patients with an AUC = 0.85 (DeLong, p = 0.0384).

Conclusion

We identified a panel of plasma proteins that – in combination with clinical variables - improves the prediction of individuals with an AAA and has the potential as blood protein markers for the detection of AAA outside a hospital setting.