Expression analysis of dopaminergic neurons in Parkinson’s disease and aging links transcriptional dysregulation of energy metabolism to cell death
Dopaminergic (DA) neuron degeneration is a feature of brain aging but is markedly increased in patients with Parkinson’s disease (PD). Recent data indicate elevated metabolic stress as a possible explanation for DA neuron vulnerability. Using laser capture microdissection, we isolated DA neurons from the substantia nigra pars compacta of PD patients, age-matched and young controls to determine transcriptional changes by expression profiling and pathway analysis. We verified our findings by comparison to a published dataset. Parallel processing of isolated neurons and bulk tissue allowed the discrimination of neuronal and glial transcription signals. Our data show that genes known to be involved in neural plasticity, axon and synaptic function, as well as cell fate are differentially regulated in aging DA neurons. The transcription patterns in aging suggest a largely maintained expression of genes in energy-related pathways in surviving neurons, possibly supported by the mediation of PPAR/RAR and CREB signaling. In contrast, a profound down-regulation of genes coding for mitochondrial and ubiquitin–proteasome system proteins was seen in PD when compared to the age-matched controls. This is in accordance with the established mitochondrial dysfunction in PD and provides evidence for mitochondrial impairment at the transcriptional level. In addition, the PD neurons had disrupted pathways that comprise a network involved in the control of energy metabolism and cell survival in response to growth factors, oxidative stress, and nutrient deprivation (PI3K/Akt, mTOR, eIF4/p70S6K and Hif-1α). PI3K/Akt and mTOR signaling are central hubs of this network which is of relevance to longevity and—together with induction of mitochondrial biogenesis—may constitute potential targets for therapeutic intervention.
KeywordsParkinson’s disease Aging Dopaminergic neuron Glia Gene expression Pathway analysis Mitochondria Energy metabolism PI3K/Akt mTOR Hif-1α
In the aging brain, dopaminergic (DA) neurons degenerate at an estimated rate of 4–5% per decade [25, 58]. The lifelong cell loss in the substantia nigra zona compacta (SNc) accumulates to 30–40% and is associated with declining motor function in the elderly [52, 53, 54]. A compelling explanation for the vulnerability of DA neurons in the SNc, is their high metabolic rate due to an autonomous pacemaking activity that is driven by voltage-dependent L-type Ca2+ channels and results in altered Ca2+ metabolism and cellular redox imbalance in aging [16, 33, 59]. It is assumed that over a lifetime, lipids, proteins, and DNA are damaged by reactive oxygen species (ROS), which are not sufficiently counteracted by anti-oxidative defense systems [18, 30]. As a possible consequence, high levels of clonally expanded mtDNA deletions are detected in aged DA neurons, which in turn are considered a cause of mitochondrial dysfunction, further compromising cellular energy status [6, 7].
Whole-genome expression studies of post-mortem samples have been proposed for the identification of key genes and regulatory pathways involved in idiopathic PD, as well as for the identification of potential therapeutic targets [3, 44, 51]. Hitherto, most studies were limited to the analysis of brain homogenates of the SNc or other brain areas [10, 32, 34, 38, 43, 45, 66, 68]. In a re-analysis of published datasets, common pathways such as IGF1-, VEGF- and axon guidance signaling were identified , highlighting differential regulation of growth and survival pathways in PD. More recently, by applying a comprehensive meta-analysis to an expanded dataset, Zheng et al.  identified defects of mitochondrial electron transport, glucose utilization and glucose sensing as early disease events.
To date, three studies on targeted whole-genome expression analysis of DA neurons in PD have been published [15, 23, 56], but due to differences in the study design and statistical methods they are not directly comparable. Our previous study focused on four hits significant after Bonferroni correction, which led to the identification of a new PD risk gene . Cantuti-Castelvetri et al.  reported only a few differentially regulated genes specific for PD and highlighted sex-specific differential regulation. Simunovic et al.  discovered 1,045 significantly regulated transcripts (up/down = 580/465) including prominent down-regulation of members of the PARK gene family.
Here, we extended our global gene expression analysis of individually isolated DA neurons of the SNc and applied a standardized pathway approach to integrate data of the Simunovic study , which is similar in design and statistical methods. For the first time, we included a younger control group to discriminate age- from disease-specific regulation. The pathway-based approach implicates disrupted energy metabolism and adaptation in PD that may explain the impact of aging and other known risk and protective factors.
Materials and methods
Ethics statement/inclusion criteria
Frozen human brain tissue was obtained from individuals who had a history of PD and from age matched and younger control individuals without any neurological disease. Written consent was obtained with verification/assent in writing from next of kin who confirmed the wishes at time of death. All procedures were in line with the UK Human Tissue Authority guidance and approved by the Local Research Ethics Committee. The cases were clinically well documented as having had PD symptoms prior to death. Neuropathological diagnosis demonstrated the presence of Lewy body pathology in the substantia nigra with typical pathological features, including moderate to severe neuronal loss and gliosis. Synuclein immunohistochemistry or ubiquitin immunohistochemistry was used to confirm findings on H&E stained sections and cases were graded according to published criteria for Lewy body disorders (LBD) [11, 42].
RNA quality control
Bulk tissue of cortical grey matter was dissected, weighed, transferred to RNAlater-ICE® (Ambion) and stored at −80°C. RNAlater-ICE® was removed and samples were rapidly homogenized (using an ultra-turrax homogenizer) for 10 s in pre-cooled 4°C TRI reagent (Applied Biosystems, Carlsbad, CA, USA). Total RNA was isolated using a spin column method according to the manufacturer’s instructions (RiboPure®, Applied Biosystems, Carlsbad, CA, USA). After extraction, RNA integrity of the samples was analyzed on an Agilent 2100 Bioanalyzer (Agilent, Santa Clara, CA, USA). Parallel cortical tissue samples were homogenized in distilled water (9× volume/weight) and the pH of the homogenate was measured at room temperature.
Initially, post-mortem brain tissue of 80 PD/LBD and control individuals was assessed for inclusion into the study. Brain samples of 11 PD cases (PD), 11 age-matched controls (old controls; OC), and eight young controls (YC) met the screening criteria of pH ≥ 6.3 and RNA integrity numbers (RIN) > 6 and these were used for laser-capture microdissection (LMD), RNA isolation, in vitro transcription (IVT) and microarray hybridization. Detailed information on patients and samples, as well as pre- and post microarray analysis is given in Online Resource 1.
Sectioning, staining, laser capture microdissection, in vitro transcription (IVT) and microarray
Unfixed midbrains stored at −80°C were used for analysis. All procedures were carried out under RNAse-free conditions. Frozen 20-μm midbrain sections were mounted on Leica 2 μm PEN- membrane slides (Leica Microsystems, Wetzlar, Germany), rapidly stained with toluidine blue solution (1%), dehydrated in an ethanol series and processed on a Leica AS LMD microscope (Leica Microsystems, Wetzlar, Germany). Approximately, 100 neurons per case/control were collected and RNA extracted with the Arcturus PicoPure® Kit (Applied Biosystems, Carlsbad, CA, USA) according to the manufacturers protocol. Four OC, four PD, and all YC were collected in biological replicate (one YC was sampled and processed four times) with a total of 48 individually processed samples. For some experiments, the SNc was delineated on membrane slides and a 20 μm cross-section of the SNc was removed entirely for RNA isolation. IVT comprised one round of linear amplification with Ambion MessageAmp™ II (Applied Biosystems, Carlsbad, CA, USA), followed by a second round of IVT with the Illumina® TotalPrep™ RNA Amplification Kit (Applied Biosystems, Carlsbad, CA, USA). Second round IVT yielded >3 μg cRNA with an average length of 800 bp and was used for hybridization on Illumina® WG6v1 expression chips (Illumina, San Diego, CA, USA). A detailed protocol is given in Online Resource 2.
Statistical analysis of microarray data
After hybridization of expression chips, raw data were exported from the Illumina Software “Beadstudio” to R (http://www.cran.r-project.org), log-scale transformed (log, basis 2) and normalized (nonlinear transformation employing the loess smoother). 8,491 transcripts were detected in all samples. For the detection of gene expression changes shared or different in aging and disease, we applied ANOVA with a cutoff at a false discovery rate (FDR) p < 0.05 (Benjamini and Hochberg). Probes were mapped on transcripts and validated using UCSC Genome browser. Mitochondrial localization was verified using the MitoP2 database . Detected transcripts were further classified as changed in aging and/or disease with a p value threshold <0.05.
Microarray cluster analysis
Including biological replicates, 48 microarray expression profiles were obtained from microdissected midbrain DA neurons. Cluster analysis revealed three main clusters (Online Resource 1). Cluster I contained 24 arrays, which were predominantly control samples (13 YC, 8 OC, 3 PD). Cluster II contained 12 arrays, including most PD samples (2 YC, 3 OC, 7 PD). Cluster III was distant from Clusters I and II and contained 12 mixed samples (3 YC, 4 OC, 5 PD). These 12 outliers exceeded the expected inter-individual variability and introduced an extreme level of heterogeneity that largely prohibited data analysis. When analyzing all 48 samples, only 3 transcripts were significant at FDR 5%. We were not able to clarify the cause (no clear relation to groups, pH, RIN, array batch effects, detected genes on array, etc.), although one possibility is IVT performance. Therefore, samples of cluster III were not included in any further analysis. Values of duplicate samples were averaged and after exclusion of cluster III, data from 8 PD cases, 9 OC and 7 YC were used for determination of differentially expressed genes. Post-mortem data for PD/OC/YC were age at death 78.6 ± 6.5/76.8 ± 9.8/52.7 ± 2.4 years, pH 6.5 ± 0.1/6.61 ± 0.21/6.81 ± 0.23 and RIN 7.7 ± 0.8/7.6 ± 1.1/8.2 ± 1.1. Only the pH of PD versus YC showed a significant difference (p = 0.005).
Correlation of gene expression profiles and enrichment of neuron/glial-specific gene expression
To determine the benefit of dissecting individual dopaminergic neurons from the SNc rather than using bulk tissue for gene expression analysis, we calculated the mean expression values from both approaches collected from adjacent microtome sections. The enrichment factor (EF) was determined by dividing the normalized expression values obtained for 100 pooled neurons (N) by the expression values of bulk isolation (B). EF = expression N/B. A factor of 1.0 indicates equal expression in neuronal and non-neuronal cells. Values <1.0 indicate a higher expression in non-neuronal cell types, values >1.0 a higher expression in DA neurons. The EF is provided for all genes in the supplemental gene lists and was used to correct for glial-specific signals.
Ingenuity pathway analysis (IPA®)
Pathways were generated with Ingenuity Pathway Analysis (Ingenuity® Systems, http://www.ingenuity.com). Canonical pathway analysis identifies pathways from the expert curated IPA library of canonical pathways that are most significant to the data set. The p value is determined by the probability that the association between the genes in the dataset and the canonical pathway is explained by chance alone. The significance of the association between the data set and the canonical pathway was calculated using Fischer’s exact test. For multiple testing correction, Benjamini and Hochberg (FDR 5%) was applied. In addition, network analysis was used to identify genes with a functional regulatory relationship that was most significant in the aging dataset. Networks are generated using the IPA Network Generation Algorithm and are scored based on the number of network eligible molecules they contain from the dataset .
Single neuron versus bulk nigral tissue
Differential DA neuron gene expression in aging and PD
Determination of quality parameters
RNA quality is crucial for successful microarray data generation, with pH and RIN of post-mortem tissue being the best surrogate markers [4, 12]. In particular, mitochondrial-related expression can be influenced by these factors . In our study, multiple mitochondrial genes were differentially regulated between the age-matched (old) controls (OC) and PD, but these groups did not differ in pH/RIN. An effect of pH could have been expected when comparing young controls (YC) to OC/PD, however, no pH-related effect was seen in the age-related genes (Online Resource 3, Fig. S2). Thus, differential expression in our dataset cannot be explained by agonal factors.
Indication of gliosis/inflammation in PD
Pathway analysis of gene expression changes in PD
Canonical pathways shared between both datasets
Canonical pathway (specific for PD)
Protein ubiquitination pathway
Regulation of eIF4 and p70S6K signaling
Regulation of stathmin
Renal cell carcinoma signaling
Reelin signaling in neurons
Clathrin-mediated endocytosis signaling
Huntington’s disease signaling
Axonal guidance signaling
Cardiac hypertrophy signaling
Germ cell-sertoli cell junction signaling
Virus entry via endocytic pathways
Pathway analysis can resolve inter-study variability
Age-related changes and ‘accelerated’ aging
Canonical pathways enriched in aging DA neurons
Specific for aging
GABA receptor signaling
Cardiac β-adrenergic signaling
Selenoamino acid metabolism
Glucocorticoid receptor signaling
G-Protein coupled receptor signaling
Estrogen receptor signaling
Aging and PD
Nicotinate and nicotinamide metabolism
Agrin interactions at neuromuscular junction
Regulation of actin-based motility by rho
Huntington’s disease signaling
Pantothenate and CoA biosynthesis
To identify the changes shared by aging and PD, i.e., the contribution of aging processes to protection/degeneration of DA neurons in PD, we analyzed genes with incremental or decremental changes. These genes were enriched in the canonical pathways for actin regulation, NAD-metabolism and other metabolic/biosynthetic pathways (Table 2b). The primary network generated from these genes was enriched for genes with a regulatory function associated with cellular assembly/organization, cell death and survival (insulin-like growth factor 1 receptor, IGF1R), synaptic transmission (glutamate receptor GRIA1) and gene expression (Online Resource 3, Fig. S5).
De novo gene expression, induction, and repression are rarely seen in the mature nervous system. Therefore, the expected magnitude of expression changes found with microarrays is only modest . In PD specifically, the signal associated with DA neurons might be hard to separate from the noise produced by non-neuronal cell populations due to significant neuronal loss and gliosis in the SNc. As shown by our comparison of expression profiles obtained from bulk tissue and isolated DA neurons, an over tenfold reduction of non-neuronal signal can be achieved using LMD and genes with the highest non-neuronal transcription signal can be identified and eliminated from analysis. Most of the glial-specific genes are immune-related and were up-regulated in PD, indicating gliosis/inflammation in PD. In relation to disease pathomechanisms, reactive gliosis is seen in many neurodegenerative disorders and recently a protective effect of non-steroidal anti-inflammatory drugs has been implicated in PD [17, 27].
In LMD, only pg/cell quantities of RNA are obtained and a subsequent second-round IVT might introduce error. Therefore, quality control, e.g. by scatter plot and cluster analysis is mandatory. We demonstrated excellent reproducibility of microarray expression patterns of DA neurons obtained in biological replicates from frozen post-mortem midbrains when stringent quality control was applied. Statistical analysis of differently expressed genes between the three groups YC, OC, and PD revealed non-random directions of gene expression changes that were attributable to meaningful biological patterns, i.e., aging- and PD-specific changes, as well as a stepwise decrease or increase of expression in aging and PD. However, the limited availability of suitable tissue limits the power of our study and comparisons to other studies are necessary. Gene-to-gene comparison with a published dataset revealed only 8% shared transcripts in PD . This might be explained by study differences, such as sampling methods (toluidine blue vs. unstained, 20 vs. 8 μm sections, 100 vs. 300 neurons), IVT (2 vs. 3 rounds) and—most importantly—different probe labeling, microarray platforms (Illumina® vs. Affimetrix®) and statistical methods. Therefore, pathway analysis was chosen to further analyze these gene sets and to provide a better understanding of transcriptional events in PD . A striking overlap was seen at the pathway level.
Disrupted energy metabolism and UPS in PD
Environmental and genetic factors are believed to predispose to the development of PD and the study of hereditary forms has greatly improved our understanding of PD pathogenesis . Localization and biological function of PD proteins have highlighted key pathogenic mechanisms, such as mitochondrial dysfunction and oxidative stress (DJ1, parkin, PINK1), vesicle and cytoskeletal dynamics (α-synuclein, LRRK2), MAPK signaling (LRRK2) as well as decreased microtubule stability (tau) . There is an intriguing overlap of these genetic pathways with the transcriptional changes found in our study. The most significantly altered pathways in our study include mitochondrial, in particular, oxidative phosphorylation (OXPHOS) genes, and the UPS. Since toxic and genetic dysfunction of mitochondria is well established in PD , this additionally supports the feasibility of the LMD and pathway-based approach. While the role of the UPS in PD remains somewhat controversial, mitochondria and the UPS are highly co-affected in experimental systems, with mitochondrial dysregulation leading to increased oxidative stress and proteasomal deregulation [8, 19, 20]. In a recent meta-analysis by Zheng et al. , the integration of all currently available PD transcriptome datasets also indicated downregulation of OXPHOS genes, as well as genes of the pyruvate metabolism and tricarboxylic acid cycle (TCA) early in the disease process. Our data support these findings and provide additional information on the impact of aging on gene expression of mitochondrial, glycolytic and TCA genes (Online Resource 3, Figs. S6 and S7). Some glycolytic and TCA genes display a decreased expression pattern in aging and PD, but the expression of most genes is stable during aging and decreases specifically in PD. This is complemented by the upregulation of pyruvate dehydrogenase kinase, indicating the use of alternative carbon sources in PD, i.e., of acetate and fatty acids (ketogenic) and amino acids (glutaminolysis) [5, 26, 55].
Dysregulation of energy-sensing/signaling pathways in PD
Genetic factors may principally determine the neuronal ability to adapt to age-related energetic and toxic stress in this regulatory network. Several other PD risk and protective factors, including environmental toxins, sex hormones, diabetes and the dietary stimulants coffee and nicotine might have an additional influence . For example, PD is more prevalent in males than in females and expression profiles are gender-specific, with an emphasis on genes related to OXPHOS, apoptosis and synaptic transmission . As outlined above, insulin has a prominent role in the adaption of energy metabolism and a disrupted metabolic control in diabetic patients might predispose DA neurons to metabolic failure. Furthermore, caffeine treatment stimulates mitogenesis in a (CAMK/p38)-dependent manner and the PI3K pathway might mediate the protective effect of caffeine [47, 67]. In conclusion, in addition to significant mitochondrial dysfunction we identified dysregulation of energy-sensing/signaling pathways in PD, which is also supported by recent functional studies.
Gene expression changes in aging
Aging is considered the most unequivocal risk factor for idiopathic PD, where loss of DA neurons in the SNc can exceed 80% at clinical onset . Nevertheless, the accentuated cell loss of ventrolateral portions of the SNc in PD as opposed to the dorsomedial SNc loss that occurs in aging indicates that PD is not simply explained by pronounced age-related processes [25, 53]. Stereological studies showed hypertrophy of the remaining DA neurons in aging, which may be considered a compensatory mechanism of surviving neurons rather than an indication of cell degeneration or necrosis, whereas atrophy of neurons is seen in PD [13, 54] (for review see Stark et al. ). In our study, analysis of genes specifically changed in aging revealed fewer significant hits compared to PD. While 1,185 transcripts were changed in PD, only 256 genes were significantly changed when comparing DA neurons of humans with an average age of 53 and 78 years. This implies that expression profiles are more dramatically and specifically changed in surviving DA neurons of individuals with PD compared to 25 years of aging. Moreover, most genes that are enriched in pathways significant for PD, show maintained expression in aging, emphasizing the specificity of PD-associated gene dysregulation. Pathway analysis in aging revealed enrichment of signaling pathways that may have an impact on healthy aging [35, 36, 39]. The cAMP/CREB signaling pathway is involved in the regulation of a wide range of biological functions such as growth factor-dependent cell proliferation and survival, glucose homeostasis and synaptic plasticity. In agreement with published data on cortical gene expression changes in brain aging, genes in inflammatory pathways (IL-22, TNFR2 signaling) are also up-regulated, whereas biosynthetic pathways and GABA receptor signaling predominately contain down-regulated genes [9, 40].
Genes that show stepwise expression changes in aging and PD may represent common pathophysiological or adaptive mechanisms. Pathway analysis yielded only a few canonical pathways with low significance and no overlap with PD-specific pathways. Nevertheless, networks generated from these genes are involved in regulation of cellular assembly/organization, cell death and survival. Expression changes of genes such as the DNA-damage-inducible transcript 4 (DDIT4, p = 1.75E-05, YC = 1.3 × OC = 1.3 × PD), Human Fission 1 (TTC11, p = 5.14E-06, YC = −1.2 × OC = −1.2 × PD) or Insulin-like growth factor 1 receptor (IGF1R) may reflect an increase of certain age-related degenerative or protective mechanisms.
Together with the aforementioned neuropathological findings, our data imply that DA neuron degeneration in PD cannot be explained by a simple acceleration of the aging processes, but that specific patterns of gene dysregulation may fundamentally contribute to the disease mechanism. As a cautionary note, the selective DA neuron vulnerability in PD and aging poses a problem in the interpretation of gene expression changes in the remaining neurons, since these changes may be pathogenic or protective. More LMD studies incorporating age-dependent and anatomical aspects would be needed to resolve this question.
We provide a full gene expression dataset that can be mined by researchers with an interest in transcriptome regulation of human DA neurons in PD and aging. This study underscores the value of human post-mortem transcriptome studies, which to date are hampered by the scarcity of suitable tissue resources . Pathway analysis provides an intriguing overlap with disease models deduced from genetic cases of PD as well as inter-study congruence. We propose a mechanistic cellular model that explains the vulnerability of DA neurons largely by their exceptional energy demand. Many genetic and environmental risk factors as well as protective factors for PD may act in part via their propensity to affect energy metabolism, stress and energy sensor pathways as well as pathways adapting to such stress. Further studies are needed to elucidate the detailed roles of the genes and pathways presented in this study.
The European Neurological Society funded M.E.; T.K., T.M., and H.P. are members of the German Network for Mitochondrial Disorders (mitoNET, 01GM0862 and 01GM0867); T.M. and H.P. were supported by the Impulse and Networking Fund of the Helmholtz Association in the framework of the Helmholtz Alliance for Mental Health in an Ageing Society (HA-215), the German Federal Ministry of Education and Research (BMBF) funded German National Research Network (NGFNplus #01GS08134) and Systems Biology of Metabotypes (SysMBo #0315494A). CMM gratefully acknowledges funding from the Health Protection Agency UK. Tissue for this study was provided by the Newcastle Brain Tissue Resource, which is funded in part by a grant from the UK Medical Research Council (G0400074), by the Newcastle NIHR Biomedical Research Centre in Ageing and Age Related Diseases awarded to the Newcastle upon Tyne Hospitals NHS Foundation Trust, and by a grant as part of the Brains for Dementia Research initiative funded by the Alzheimer’s Research Trust and the Alzheimer’s Society.
- 32.Grunblatt E, Mandel S, Jacob-Hirsch J et al (2004) Gene expression profiling of parkinsonian substantia nigra pars compacta; alterations in ubiquitin-proteasome, heat shock protein, iron and oxidative stress regulated proteins, cell adhesion/cellular matrix and vesicle trafficking genes. J Neural Transm 111:1543–1573PubMedCrossRefGoogle Scholar
- 37.Lee DW, Rajagopalan S, Siddiq A et al (2009) Inhibition of prolyl hydroxylase protects against 1-methyl-4-phenyl-1, 2, 3, 6-tetrahydropyridine-induced neurotoxicity: model for the potential involvement of the hypoxia-inducible factor pathway in Parkinson disease. J Biol Chem 284:29065–29076PubMedCrossRefGoogle Scholar
- 67.Wright DC, Geiger PC, Han DH, Jones TE, Holloszy JO (2007) Calcium induces increases in peroxisome proliferator-activated receptor gamma coactivator-1 alpha and mitochondrial biogenesis by a pathway leading to p38 mitogen-activated protein kinase activation. J Biol Chem 282:18793–18799PubMedCrossRefGoogle Scholar
- 68.Zhang Y, James M, Middleton FA, Davis RL (2005) Transcriptional analysis of multiple brain regions in Parkinson’s disease supports the involvement of specific protein processing, energy metabolism, and signaling pathways, and suggests novel disease mechanisms. Am J Med Genet B Neuropsychiatr Genet 137:5–16Google Scholar
- 69.Zheng B, Liao Z, Locascio JJ, et al (2010) PGC-1α, a potential therapeutic target for early intervention in Parkinson’s disease. Sci Transl Med 2:52ra73Google Scholar