Background: There is an unmet need for a minimally invasive clinical test using samples collected by medical staff or by self-collection to screen persons for Alzheimer's disease (AD), Parkinson's Disease (PD) and other dementias and diseases. AD is underdiagnosed today, particularly in underserved populations where it is high incidence (non-Hispanic black and Hispanic) and in rural populations. Even in populations with access to health care, diagnosis is often delayed by years, even though the most recently approved therapies are most efficacious when given early in the progression of disease. While dementias are diseases of the brain, many publications link the underpinning inflammatory aspects to circulating neutrophils, macrophages, and lymphocytes. Thus, there is an opportunity to exploit the hematologic aspects of AD and PD in its diagnosis and potentially, therapeutically. AD gene expression signatures using RNA extracted from venipuncture whole blood have been reported. FDA-cleared tests based on circulating biomarkers are in use today but are not definitive tests for AD. We developed a direct, extraction-free TempO-Seq® assay of gene expression using blood from a finger prick spotted on filter paper, providing a simple and minimally invasive method of obtaining a blood sample and performing a test that can potentially be used to identify a host of diseases and disease risk. We then implemented a test that classified patients as having AD or PD. Method: The commercial targeted sequencing whole transcriptome TempO-Seq® gene expression assay (Yeakley, PLosOne doi.org/10.1371/journal.pone.0178302) was adapted to utilize blood from a simple fingerstick spotted on filter paper and dried (not a biohazard, simplifying transport and testing). Three independent patient cohorts of samples (~50:50 male:female) were obtained from normal volunteers and patients clinically diagnosed with AD or PD based on cognitive testing. AD, PD, and used for training/retraining and testing/retesting AD and PD classification signatures. To account for the presence of subtypes in our dataset, we contrasted AD and control samples by testing hundreds of random subsets and performing differential gene expression analyses (DESeq2). We used machine learning approaches (k-nearest neighbors, random forest, support vector machine with either linear, polynomial or rgb kernel and extreme gradient boosting) to build algorithms differentiating control, AD and PD. Randomly sampled subsets accounting for 80% of each cohort were used to train these classifiers. The accuracy of calling AD and PD samples was determined using the remaining 20% of samples as a test set. For each classifier an area under ROC curve (AUC-ROC) was calculated. Gene set enrichment was performed using the Gene Ontology database for biological processes (p value threshold 0.05) to identify expressed pathways. Result: A protocol that eliminated interference by components of whole blood with enzymes used in molecular biology assays (including PCR) was successfully implemented. This permitted a TempO-Seq assay to be carried out without extraction of RNA from the filter paper blood spots. Highly expressed genes were attenuated to increase the sensitivity with which AD biomarker genes could be measured. Within-spot reproducibility (r>0.96) and day-to-day reproducibility of blood collected from the same donor (r>0.94) was excellent. The AUC-ROC for the AD classifier was 0.86, and for the PD classifier was 0.87. Predominate pathways were immune response pathways, with the three most significantly differentially expressed pathways related to the involvement of neutrophils in the immune and inflammatory response. Conclusion: Patients with AD and PD, and potentially other dementias and diseases, can be classifiedusing a TempO-Seq gene expression signature obtained from fingerstick blood spotted on filter paper. Such samples are easily collected by medical staff or by self-collection, potentially transforming the diagnosis of AD and PD, reducing health disparities, and by speeding up diagnosis, enabling therapy for many to begin earlier. The data support that neutrophils and immune cells either impact disease and/or are impacted by disease, consequently resulting in expression of disease signatures capable of classifying patients.
Read full abstract