Molecular markers of early Parkinson's disease based on gene expression in blood

Abstract

Parkinson's disease (PD) progresses relentlessly and affects five million people worldwide. Laboratory tests for PD are critically needed for developing treatments designed to slow or prevent progression of the disease. We performed a transcriptome-wide scan in 105 individuals to interrogate the molecular processes perturbed in cellular blood of patients with early-stage PD. The molecular multigene marker here identified is associated with risk of PD in 66 samples of the training set comprising healthy and disease controls [third tertile cross-validated odds ratio of 5.7 (P for trend 0.005)]. It is further validated in 39 independent test samples [third tertile odds ratio of 5.1 (P for trend 0.04)]. Insights into disease-linked processes detectable in peripheral blood are offered by 22 unique genes differentially expressed in patients with PD versus healthy individuals. These include the co-chaperone ST13, which stabilizes heat-shock protein 70, a modifier of alpha-synuclein misfolding and toxicity. ST13 messenger RNA copies are lower in patients with PD (mean +/- SE 0.59 +/- 0.05) than in controls (0.96 +/- 0.09) (P = 0.002) in two independent populations. Thus, gene expression signals measured in blood can facilitate the development of biomarkers for PD.

Conflict of interest statement

Conflict of interest statement: C.R.S., S.R.G., and R.V.J. are listed as coinventors on a U.S. Letters Patent application for identification of dysregulated genes in patients with neurologic diseases held by Brigham and Women's Hospital. C.R.S. is a consultant to Link Medicine Corporation.

Figures

Fig. 1.

Molecular marker associated with PD risk. (a) Expression data matrix of eight marker genes of 66 blood samples from PD and control subjects. Each row represents a blood sample, and each column represents a gene. As shown in the color bar, overexpression is displayed in red, and underexpression is displayed in green. Blood samples are ordered by their risk score (shown on the left), which is defined as the correlation with the average profile of the PD group minus the correlation with the average profile of the controls. Twenty of 22 individuals with high risk scores ranked in the top third of the list have PD (third tertile). Twenty-one of 22 individuals with low risk scores ranked in the bottom third of the list are controls (first tertile). Solid lines designate tertiles of risk score values. The clinical diagnosis for each individual is shown on the right. (b) Validation of the risk marker on independent test samples confirms that high scores are significantly associated with increased PD risk (P for trend = 0.04). The expression data matrix is as in a, except that b uses 39 independent samples. (c) The ROC curve in the test set (blue curve) is highly consistent with the ROC curve for the leave-one-out cross-validated (LOOCV) marker in the training set (gray) confirming the risk prediction observed for different cutoffs. The nominal ROC curve in the training set represents an upper limit (red). (d) Dopamine replacement medication does not bias the risk score. There is no difference in risk scores of PD patients on dopamine medication versus unmedicated de novo patients (mean ± SE 0.06 ± 0.04 and 0.11 ± 0.1, respectively; P = 0.96). The average risk score for the overall PD group is 0.07 ± 0.03 (data not shown). Average risk scores are low (negative) in healthy controls (−0.24 ± 0.04) and neurodegenerative disease controls [AD, −0.25 ± 0.05; progressive supranuclear palsy, −0.19 ± 0.06; multiple system atrophy (MSA), −0.34 ± 0.17; corticobasal degeneration (CBD), −0.26; essential tremor (ET), −0.15]. Individual scores range from −0.43 to 0.6 for PD patients, −0.62 to 0.12 for healthy controls, and −0.59 to 0.37 for neurodegenerative disease controls. ND, neurodegenerative disease control; H, healthy control. Error bars indicate standard errors.

Fig. 2.

Heat-shock protein 70-interacting protein ST13 and biologic insights into PD-related changes in blood cells. Underexpression of ST13 in PD is confirmed by three distinct assays in two independent populations (a–c). (a) Twenty-four probe sets, including two probing for ST13, are significantly underexpressed in cellular blood of 31 PD patients (including five de novo PD patients) compared with healthy controls (FDR = 0.03). Untreated de novo patients are identified by a pink bar, and patients treated with dopamine replacement are identified by a black bar. Expression matrices are as in Fig. 1 except that genes here are shown on the vertical axis. The majority of PD patients are at early stages of the disease process as indicated by the Hoehn and Yahr (H&Y) scale. Dendrograms illustrate genes with similar expression patterns by cluster analysis. (b) Underexpression of ST13 in PD is first validated by real-time PCR in a large subset of 39 PD patients and 12 age-, sex-, and blood-count-matched healthy controls by using the comparative CT method. ΔCT values are displayed. In PD patients, ST13 amplifies at higher ΔCTs than in controls, indicating a lower abundance of ST13 mRNA in PD (fold change = 0.78, P = 0.025 by t test). GAPD mRNA levels are used to control for input RNA. (c) The log-transformed mean ratio of ST13 mRNA copies to 18S ribosomal RNA copies is lower in patients with PD (mean ± SE 0.59 ± 0.05) than in the control group (0.96 ± 0.09) (fold change = 0.6, P = 0.002 by t test) in a second, independent population by a standard curve method. The box plot shows the median (bold line) and the 75th and 25th percentile values (top and bottom of the box) for log-transformed ratios of ST13 mRNA copies to 18S rRNA copies in blood samples of 17 patients with PD and 17 age- and sex-matched healthy control subjects. The top and bottom of the whiskers show the maximum and minimum values. (d) Underexpression of the apoptosis-related gene BCL11B was also confirmed by the comparative CT method (P = 0.005).