Deep brain stimulation induces rapidly reversible transcript changes in Parkinson's leucocytes

Gene expression patterns distinguish Parkinson's status

To detect gene expression patterns associated with Parkinson's disease, leucocyte mRNA of seven male PD patients and of six age-matching male healthy controls (HC) was examined using exon arrays [16]. Patients’ blood samples were taken in three states: prior to DBS neurosurgery (‘pre-DBS’), following DBS ON electrical stimulation (‘post-DBS’, 2.2 ± 0.9 months after DBS neurosurgery, upon disease motor symptoms stabilization) and following 1 hr of OFF stimulation (see Fig. 1 for experimental design and study workflow). RNA integrity numbers (RIN) ranged from 6.2 to 9.3 (with average of 8.5 and standard deviation of 0.85; Table ST1B). Exon arrays contain three main annotation levels for each probe set: core, extended and full. The core probe sets correspond to well-annotated exons and only those were analysed; in this study, we referred to the specific genes represented by these core exons as unities.

Before the statistical analysis, two filtering steps were conducted on all samples: first, removal of all un-annotated transcripts, and then variance-based filtration (var) to remove those transcripts with variance less than the 15th percentile that were of no interest. Calculating the median expression signal from all the core probe sets that interrogate each gene yielded gene level signals, and differentially expressed genes were identified by t-tests. To select those changes, which would not be found under random shuffling of the treatment and the control samples, the t-test was repeated 1000 times by permuting the columns of the gene expression data matrix [17] (Fig. 2). Of the resultant gene lists, we selected the shortest one; this identified 173 transcripts as non-randomly differentially expressed between PD patients to controls (Table ST4). The differentially expressed transcripts included the PD-associated genes SNCA (also designated PARK1) [18, 19] and PARK7 [13, 20] in which mutations are linked to early appearance of PD. Also, we compared the current results with those of a previously published independent cohort of 98 early PD patients, neurological and healthy controls [7, 8]. The full lists of differentially expressed transcripts were then subjected to post-hoc and ad hoc gene ontology classifications and to biological validation by quantitative real-time PCR (qRT-PCR; Fig. 2 and Table ST3).

Transcript profiles of PD patients differ from controls

To test if the identified transcript modifications would distinguish PD patients from controls, we applied two clustering approaches on the expression signals of the detected genes in a manner blind to clinical diagnosis. Unsupervised classification by hierarchical classifier (HCL) segregated correctly all of the samples by the modified expression signals (Fig. 3A) using the expression signals of the 173 detected transcripts. Principal component analysis (PCA) [21], classified as well all the samples correctly by type (patients/controls; Fig. SF1). In addition, the HCL clustering segregated all the detected transcripts by their expression pattern (Fig. 3A, right-side dendrogram). To examine functional relevance of the gene-level classification, we applied post-hoc functional analysis [22] on the six top-level gene clusters created by the classifier. This identified enriched neuron development function [22] (Fig. 3A, ST4), suggesting detection of functional changes in the parkinsonian brain as reflected in patient leucocytes. Of note, the modified PD leucocyte transcripts (as compared with matched controls) did not predict the clinical outcome of the DBS surgery or the subsequent DBS-induced expression changes (Fig. SF2–4).

STN-DBS affects a wide range of transcripts including disease-modified ones

Next, we identified PD patient genes that are differentially expressed between post- to pre-DBS states. Following exhaustive permutation tests, we detected 465 genes as differentially expressed after DBS surgery (Table ST2). Thus, the DBS stimulus induced almost three times more leucocyte transcriptional changes than the disease itself. Hierarchical classifier classification analysis distinguished all of the pre-DBS samples correctly from the post-DBS ones based on the expression signals of these 465 detected genes (Fig. 3C, right-side dendrogram). Principal component analysis classification as well correctly segregated all of the samples by state (Fig. SF1). We then composed the transcripts differentially expressed in PD compared to control to those modified in the post-DBS compared to pre-DBS samples. Twenty-two (13%) of the 173 PD modified transcripts, including SNCA were among the 465 DBS-modified transcripts (Fig. 3B). The probability that 22 of the 173 transcripts that were detected as changed in PD compared to controls will also change post-DBS was calculated using the binomial coefficient with the equation , where q= (1 −p) and the probability for a gene to change post-DBS is . The outcome of the test is p= 6.661 χ 10⁻⁹ demonstrating that it is highly unlikely to observe a random overlap between 13% of the transcripts modified in PD and those changed following DBS.

In all patients, the Unified PD motor Rating Scale (UPDRS-III) improved (i.e. decreased) following STN-DBS (t-test P= 0.0008; Fig. 3C). To better characterize the change we conducted post-hoc GO functional analysis [23] on the detected genes. Genes that exhibit similar patterns of expression were clustered together by the classifier. Analysing each of the 10 top-level clusters revealed enriched decreases in DNA replication transcripts, T-helper immune response, response to metal ions, mitochondrial transport, interferon-gamma biosynthesis, protein kinase cascade and cation homeostasis (Fig. 3C and Table ST5). There was no correlation between pre-operative levodopa responsiveness and STN-DBS efficacy (Fig. SF4), similar to findings in a larger group of DBS-treated PD patients [24].

The post-DBS stimulation state differs from HC

Given the improved motor symptoms following DBS, we proceeded to test if the leucocyte post-DBS transcript profiles regained similarity to those of HC. Surprisingly, PD patients’ post-DBS on stimulation exhibited distinct expression as compared with HC. Permutation t-tests identified 321 transcripts as changed between PD patients’ post-DBS to controls, including PARK7 and SFRS7 which maintained their PD-characteristic changes. Nevertheless, all post-DBS samples were correctly classified from controls by both HCL (SF1) and PCA (Fig. SF1) classifiers. Post-hoc functional analysis revealed enrichment of dopaminergic synaptic transmission in the list of detected transcripts (Table ST5).

Two central PD genes and one splicing factor, all detected as changed in patients and following DBS were selected for qRT-PCR validation: SNCA (PARK1), PARK7 and SFRS1. The SNCA gene consists of six exons creating six different splice variants [25], three of which encode protein. Quantitative RT-PCR validated Twofold disease-associated decreases in both exons 2–3 and 4–5, which are included in different SNCA variants. These decreases were fully correctible by STN-DBS, with similar trends detected by qRT-PCR.

The PARK7 gene up-regulates human tyrosine hydroxylase by inhibiting the splicing factor SFPQ [26]. PARK7 covers four splice variants [27] and an additional 5’ promoter variant [28]. PARK7 exhibited disease-induced increases and qRT-PCR validated those in both the junctions linking exons 4–5 and 6–7. Among several splice factors which changed, SFRS1 (ASF/SF2) has two ultra-conserved splice variants differing in their 3’-UTR. Only SFRS1 transcripts including full-length 3’-UTR encode the intact ASF protein and are rescued from mRNA degradation [29]. The arrays detected treatment-associated increases in the SFRS1 3’-UTR as compared to controls, which were validated by qRT-PCR (Fig. SF3).

Rapid, immediate, reversal of transcript modifications upon stimulation cessation

We hypothesized that the reaction to DBS neurosurgery is subjective to a dynamic, short-term change that is dependent on the existence of the electrical stimulus. Furthermore, the changes observed post-DBS could tentatively be due to the electrical stimulation, to the operation itself or to both. To discriminate between these possibilities, we tested RNA samples extracted 1 hr after the electrical stimulation was turned off (OFF state). ON- and OFF-sample sets were both derived from patients on the same dose of DRT, which is considerably lower than that administered to pre-surgery patients. Therefore, the OFF state also served to assess the contribution of medication dose to the observed changes. Under OFF state, the major disease symptoms rapidly re-occur. For example, a PD patient pre-operation and in OFF-state (both lacking the DBS stimulation, but with different medication doses) will have limited mobility, as reflected by a worsened UPDRS-III (Fig. 3A–D and Table ST2). The OFF-state was accompanied by differential expression of 351 transcripts (Table S4). Unsupervised HCL fully distinguished PD patients in stim-ON state from stim-OFF state. The classified clusters were enriched in DNA binding and nucleotide binding (Fig. 3D and Table ST4). PCA further segregated all samples correctly (Fig. SF1).

Globally, the post-DBS state exhibited more transcript decreases than increases compared to pre-DBS patients, inverse to the trend for increases between PD patients and HC (Figs 4A and SF4). Furthermore, post- to pre-surgery comparison presented more divergent functional categories than those enriched in PD compared to controls or to OFF-state (Fig. SF4). Of the stim-OFF changed transcripts, 105 (30%) (Fig. 4B) also changed following STN-DBS compared to the pre-surgery state, mostly in an inverse direction (Fig. SF4A), reflecting a molecular manifestation of the reversibility of motor function observed with DBS and verifying that the effect of the stimulus was larger than this of the medication dosage. The likelihood for an overlap of 105 common transcripts to occur in random is

nearly zero. Considering that the patients received the same reduced medication dose under OFF- and ON-state further attributes the observed change to the cessation of the stimulus alone. Also, comparing the pre-DBS to post-DBS OFF stimulation profiles (once again, using permutation t-tests) identified 114 transcripts as differentially expressed, much less than those 465 transcripts modified following DBS stimulation; demonstrating that the pre-DBS and post-DBS-OFF stimulation states were more similar to one another than any other two states among those tested. Of the above 114 differential transcripts, 27 (24%) were also modified under DBS stimulation and reduced medication.

Next, we performed multi-comparisons. Of the post-DBS transcripts different from controls 114 (35.5%), also changed in at least one of the three major comparisons (PD compared to controls, post-DBS compared to pre-DBS or post-DBS stim-OFF compared to stim-ON). Of those, 46 (23%) changed both from PD to post-DBS and to HC, although pre- and post-DBS patients are on different medication dosages, again highlighting the reversible mode of the observed changes and their dependence on the stimulus itself.

Rapid functional reversibility of the DBS stimulus

Post-hoc functional analyses detect robust changes in differentially expressed genes, but disregard subtle changes in functionally related transcript groups, which are not robustly modified [30]. To detect subtle increases and decreases in all transcript classes, the expression signal of each gene was calculated based on all of its core-annotated probe sets for both biological process (BP) and molecular function (MF) terms. Then functional ad hoc GO analysis was applied by both Kolmogorov–Smirnov (KS) statistics and threshold-dependent Fisher's exact test [8]. The KS analysis is not based on pre-defined cut-off by fold-change threshold. Therefore, it is completely gene list independent and is able to detect subtle expression changes in groups of genes that function together [31]. Each GO term was separately tested for increase and decrease by the above tests. Nine of the reversed terms increased stim-ON and decreased stim-OFF, and 16 terms decreased in stim-ON and increased upon stim-OFF, including the Wnt receptor signalling pathway. Of note, five disease-increased BP terms resisted stim-OFF.

Next, we searched for GO terms that changed between all of the tested stages (controls to PD, pre-DBS to ON- and ON- to OFF-stimulation). These included 16 MF and 4 BP common terms (of those, 12 MF and 3 BP terms changed in the same direction between controls to PD, as between the DBS-ON to the OFF-stimulation state (Fig. 4B and Table ST5). Most of the disease-increased BP terms reversed under stim-ON and reversed again under stim-OFF, reinforcing the observation of a reversible clinical effect of stimulus cessation toward a pre-treatment disease state. Comparing each GO term to its direct GO tree ‘parents’ enabled a zoom-in (magnified) focus on specific changes (Fig. SF4).

Transcripts modified in all comparisons reflect the neurological efficacy of DBS

To gain translational utility, we searched for a selected group of transcripts, which could constitute a ‘molecular signature’, and would react both to the disease and to the DBS procedure. Twenty-nine transcripts changed (P < 0.01) both in PD patients compared to controls and between ON and OFF stimulation. Clustering the expression values of those 29 genes classified, DBS-treated patients together with controls, and OFF-stimulus ones with pre-operated patients (Fig. 5A). Principal component analysis classified these two groups similarly, except for two apparent control samples which were classified with the patients group and one stim-OFF patient maintaining a stim-ON pattern (Fig. 5B). Future follow up will be required to determine if the few false positive control subjects detected in this analysis reflected early development of disease biomarkers. Intriguingly, the 29 classifier sequences included six ‘signature’ transcripts which we also identified as changed in a web-deposited 3’ data set of a larger cohort of early PD patients compared to controls [7, 8] (Fig. 5A, double stars) or to patients with other neurological diseases (neurological controls, NC; ANOVA FDR P < 0.05) [8] (Fig. 5A, triple stars). Clustering analysis by HCL revealed that these six transcripts classify in the early PD cohort data set PD patients both from healthy controls and from neurological controls (SF5).

The six ‘signature’ transcripts were all regulatory in nature: the transcription factor NR2F1 [32]; the PCBP2 poly (rC) binding protein that affects RNA turnover [33]; the PD-associated PTPN1 protein tyrosine phosphatase [34]; the HNRPDL member of the heteronuclear proteins which control exon inclusions [35], the protein ubiquitination controller PJA1 [36]; and the endoplasmic reticulum membrane glycoprotein TRAM1 involved in protein translocation [37], which induces interferon-β[38] through coupling with Toll-like receptor 4 (TLR4) that mediates bacterial infection responses [39].

Post-DBS patients, either ON- or OFF-stimulus, all showed improved motor function as independently evaluated by a neurologist (based on the UPDRS-III scale), albeit to different extents (25–88.9% improvement; Table ST2). In each patient, the magnitude of change of all of the DBS-modified transcripts was correlated with the individual neurological improvement (Fig. SF4) but not with the extent of reduction in DRT or with the extent of pre-DBS change in gene expression compared to controls (Fig. SF4). Linear regression test validated this correlation (R²= 0.679). Furthermore, the expression changes observed in the 29 signature transcripts alone showed tight correlation with the motor clinical improvement in each patient as reflected by their UPDRS-III (R²= 0.902, ANOVA P= 0.005; Fig. 5C), identifying these 29 transcripts as an independent ‘molecular signature’ reflecting treatment success.

Inverse cholinergic and inflammatory changes

In PD, the dopaminergic–cholinergic balance is impaired by dopamine depletion and acetylcholine increase. Correspondingly, anti-cholinergics were the first drugs available for the symptomatic treatment of PD [40]. In this context, we were intrigued to detect disease-associated reductions in the KS analysis of transcript categories of cholinergic signalling. The category of muscarinic acetylcholine signalling and activity, for example, showed DBS-inducible enhancement and OFF-stimulus reduction. Compatible with the reported cholinergic blockade of inflammation responses [39, 41], we further observed an inverse trend of PD-associated increases, DBS-repressible inflammation and immune-related categories and re-induction of those by OFF state, suggesting cholinergic-mediated changes in peripheral inflammation as a mechanism underlying at least part of the observed differences in leucocyte transcripts (Fig. 6). Changed categories (having P < 0.05 in at least one of the four ad hoc tests) and direction of change are given under Figure SF6. The DBS-induced decreases in the T-helper-1 immune response, which was not detectable in pre-surgery PD patients may balance part of this effect.