Analysis of Gene Expression in Parkinson's Disease: Possible Involvement of Neurotrophic Support and Axon Guidance in Dopaminergic Cell Death

Abstract

Parkinson's disease (PD) is a neurodegenerative disorder characterized by the progressive loss of dopaminergic neurons in the substantia nigra. We have studied alterations in gene expression in the substantia nigra, the caudate nucleus and putamen of four PD patients and four matched controls using custom designed Agilent microarrays. To gain insight into changes in gene expression during early stages of dopaminergic neurodegeneration, we selectively investigated the relatively spared parts of the PD substantia nigra, and correlated gene expression changes with alterations in neuronal density. We identified changes in the expression of 287 transcripts in the substantia nigra, 16 transcripts in the caudate nucleus and four transcripts in the putamen. For selected transcripts, transcriptional alterations were confirmed with qPCR on a larger set of seven PD cases and seven matched controls. We detected concerted changes in functionally connected groups of genes. In the PD substantia nigra, we observed strong evidence for a reduction in neurotrophic support and alterations in axon guidance cues. As the changes occur in relatively spared parts of the PD substantia nigra, they suggest novel disease mechanisms involving neurotrophic support and axon guidance in early stages of cellular stress events, ultimately leading to dopaminergic cell death in PD.

INTRODUCTION

Parkinson's disease (PD) is, after Alzheimer's disease, the second most prevalent age‐related neurodegenerative disorder, affecting approximately 1% of the population above the age of 65. The main clinical features of PD are resting tremor, rigidity, slowness of voluntary movement and postural instability 19, 53). Neuropathologically, PD is characterized by the progressive loss of nigrostriatal dopaminergic (DAergic) neurons in the presence of two types of neuronal inclusion bodies: spindle or threadlike Lewy neurites in cellular processes, and globular Lewy bodies in neuronal somata 11, 27). The loss of DAergic neurons in the substantia nigra pars compacta (SN) results in reduced dopamine (DA) levels in the striatum, which in turn causes the majority of the clinical symptoms of PD. Patients with severe neuronal loss in the SN show, generally, also pathology in the neocortex (11). Interestingly, the SN is affected in PD in a heterogeneous way. For example, DAergic neurons in calbindin‐poor regions (the nigrosomes) of the SN are more prone to degenerate than neurons in the calbindin‐rich nigral matrix (18). This suggests that interactions of DAergic neurons with their surroundings might be important in neuronal vulnerability in PD.

In the last decade, mutations in the SNCA, LRRK2, PARK2, PINK1 and DJ‐1 genes have been found to cause inheritable forms of PD, with differences in age of onset and other specific clinical manifestations of the disease. Furthermore, association studies revealed several other genes that alter PD susceptibility, such as UCHL1 and NURR1 (32). On first sight, the genes that are involved in PD are unrelated in function. For example, LRRK2 functions as a cytosolic kinase, PARK2 is an E3 ubiquitin ligase that tags ubiquitin molecules to proteins destined to be degraded by the proteasome, and DJ‐1 is involved in signaling in response to oxidative stress (32). The function of SNCA, although extensively studied, remains relatively unknown. It is likely to be involved in vesicle trafficking and synaptic transmission (13). For SCNA, both a gene triplication and point mutations lead to familial PD 62, 71). These alterations make SCNA more prone to aggregate, and SNCA is indeed a major component of Lewy bodies. Interestingly, UCHL1 and PARK2 are both involved in protein (de)ubiquitination. Dysregulated protein aggregation therefore appears to be an important common factor in PD pathogenesis (33).

Apart from genetic mutations, toxins such as 1‐methyl‐4‐phenyl‐1,2,3,6‐tetrahydropyridine (MPTP), rotenone and 6OH‐DA are known to induce Parkinson‐like pathology (9). Induction of oxidative stress is a common mode of action of these neurotoxins. Mutations in DJ‐1 and PINK1 are also associated with elevated levels of oxidative stress and/or mitochondrial dysfunction. Although discoveries of gene mutations in familial PD and the existence of specific neurotoxic substances inducing parkinsonian phenotypes have provided important insights into familial and sporadic PD, there are still many open questions regarding the specific molecular mechanisms underlying the sporadic form of PD. The etiology of sporadic PD is most likely multifactorial, and the heterogeneous vulnerability of SN DAergic neurons suggests involvement of the surrounding cellular environment. Also, in PD, the loss of presynaptic terminals in the striatum precedes the loss of neuronal cell bodies in the SN (5). We therefore hypothesize that DAergic neurons are under the control of environmental cues in the SN as well as terminal areas. These environmental cues, together with factors intrinsic to DAergic neurons, make up the unique conditions in which SN DAergic neurons selectively degenerate in PD. To investigate these conditions, we have generated gene expression profiles of the caudate nucleus, putamen and SN of four clinically and neuropathologically well‐characterized PD patients and matched controls using custom‐designed high‐density microarrays, and validated genes possibly implicated in PD on a larger set of seven PD patients and seven matched controls.

EXPERIMENTAL PROCEDURES

Subjects

Post‐mortem human brain tissue was obtained from The Netherlands Brain Bank, Netherlands Institute for Neuroscience, Amsterdam (NBB). All material has been collected from donors from whom a written informed consent for a brain autopsy and the use of the material and clinical information for research purposes had been obtained by the NBB. All subjects were clinically diagnosed and treated PD patients, a diagnosis that was neuropathologically confirmed (n = 7), or were controls without neurological or psychiatric disorders and without neuropathological alterations (n = 8). All PD patients received DA replacement therapies during the course of their disease. Both PD patients and controls were neuropathologically systematically investigated as described earlier (75). The controls did not exceed a Braak pathology score for neurofibrillary tangles of 2 (10). Care was taken to match subjects as closely as possible for age, sex, post‐mortem interval and brain pH (see Table 1). As agonal state is one of the most important factors influencing RNA quality, and RNA expression profiles are correlated with brain pH (74), samples with a cerebrospinal fluid pH below 6.3 were excluded. In addition, we excluded patients who were treated with Prednisone at the time of death, unless stated otherwise. Only subjects from whom freshly frozen tissue samples of the caudate nucleus, putamen, as well as SN were available, were included. There were no significant differences between PD and control groups for age, pH, RNA integrity, post‐mortem interval and brain weight (Mann‐Whitney U test, Table 1).

Table 1.

Clinicopathological data of human postmortem samples. Abbreviations: PD/DEM = Parkinson's disease with dementia; PMI = post‐mortem interval (h); BW = brain weight (g); RIN = RNA integrity number; P‐value = Mann‐Whitney U P‐value; SD = standard deviation.

Subject	Diagnosis	Sex	Age	PMI	pH	BW	RIN	Cause of death	Remarks
00–115	PD/DEM	M	70	9:05	6.33	1258	6.2	Pneumonia, septic shock
04–045	PD/DEM	M	71	6:58	6.55	1358	8.4*	Pneumonia
00–139	PD/DEM	M	72	7:15	6.55	1546	6.7	Uremia
02–003	PD	F	75	5:00	6.52	1218	9.6*	Euthanasia
02–011	PD	F	79	5:45	6.37	1203	8.7*	Myocard infarction
00–034	PD	M	86	8:30	6.52	1178	9.2*	Unknown
02–064	PD	M	87	7:20	6.37	1166	7.4	Respiratory insufficiency
Mean ± SD		5M/2F	77.1 ± 7.1	7.1 ± 1.3	6.5 ± 0.1	1275 ± 136	8.0 ± 1.3
98–126	CTRL	M	71	6:00	6.54	1385	8.8	Respiratory insufficiency
00–049	CTRL	M	78	6:55	6.42	1332	9.2*	Cardiac failure
97–144	CTRL	M	78	4:00	6.43	1160	9.0*	Pulmonary carcinoma	Prednisone
00–142	CTRL	F	82	5:30	6.60	1280	9.2*	Myocardial infarct
00–022	CTRL	F	83	7:45	6.52	1102	9.2*	Acute myocard infarction
98–062	CTRL	M	85	4:35	6.95	1332	7.5	Respiratory insufficiency secondary to metastasized adenocarcinoma
99–046	CTRL	F	89	5:10	6.62	1168	9.5	Cardiac arrest
01–029	CTRL	F	90	5:25	6.58	1066	7.6	Myocard infarction
Mean ± SD		4M/4F	82 ± 6.3	5.4 ± 1.1	6.6 ± 0.2	1228 ± 119	8.8 ± 0.8
P‐value			0.22	0.06	0.08	0.49	0.22

^* Samples used in microarray experiment.

Tissue dissection

For the caudate nucleus and putamen tissue samples, structures were identified macroscopically in the frozen tissue by the gross anatomical features and surrounding white matter. A cylindrical punch of 50–100 mg gray matter was obtained by a hollow dry‐ice cooled drill, collected in prechilled 2 mL tubes and immediately put on dry ice. The SN tissue samples were first treated for 48 h with RNAlater^®‐ICE (Ambion, Austin, TX, USA) at −20°C to stabilize the RNA and to soften the tissue to enable dissection using scalpels. The area of interest was the least affected part of the SN of each PD patient that was macroscopically defined by the presence of black pigmentation in the unstained tissue, which represents the neuromelanin in DAergic neurons. Although it was apparent that the SNs from PD patients lost the majority of its pigmentation, for all individuals, pigmented areas could easily be identified. Small portions of the pigmented areas (∼10 mg) were rapidly dissected out by hand on a dry‐ice chilled metal plate, and transferred to prechilled 2 mL tubes and immediately put on dry ice. Dissected samples were stored at −80°C until use.

RNA isolation

Total RNA was extracted from all brain regions using a hybrid protocol of Trizol (Life Sciences, Grand Island, NY, USA) and Qiagen RNeasy Mini Kit™ (Qiagen, Valencia, CA, USA) RNA isolation methods. Briefly, samples were homogenized in Trizol (3 mL per 100 mg tissue). After phase separation by addition of chloroform, the aqueous phase was transferred to a new tube and mixed with an equal amount of 70% ethanol. This mixture was then applied to an RNeasy Mini™ column, and processed according to the RNeasy Mini™ Protocol for RNA Cleanup (version June 2001, from Step 3). RNA purity was determined using a NanoDrop^® ND‐1000 spectrophotometer (Nanodrop Technologies, Wilmington, DE, USA). RNA integrity was determined by the RNA Integrity Number (RIN) as measured on the Agilent™ 2100 bioanalyzer (Agilent Technologies, Palo Alto, CA, USA). The RIN values varied, ranging from 6.2 (moderately degraded) to 9.6 (intact RNA). Because RNA quality has strong effects on gene expression profiles, only samples with high RNA quality were selected for microarray experiments (four PD patients and four controls, RIN > 8.3, see Table 1). RNA quality was sufficient for qPCR analysis for all samples under investigation.

Sample labeling and microarray hybridization

For microarray analysis, Agilent custom‐made 22K 60‐mer oligonucleotide arrays were used. Array probes were designed to contain all homo sapiens unigene clusters consisting of at least one mRNA molecule (∼15 K probes) and ∼7.5 K probes for expressed sequence tags known to be expressed in the brain or eye. Array hybridization was performed according to the manufacturer's instructions. Briefly, equal amounts (500 ng for caudate and putamen, 1 µg for substantia nigra) of Cy3 and Cy5‐labeled RNA were hydrolyzed for 30 minutes at 60°C in 1x fragmentation buffer (Agilent Technologies). The fragmented targets were hybridized to a microarray by incubating for 17 h in 1x target solution (Agilent Technologies) at 60°C in a rotating hybridization chamber. Microarrays were washed in 6xSSPE/0.005% N‐Lauroylsarcosine (Sigma‐Aldrich, St Louis, MI, USA) for 5 minutes at room temperature (RT) and 0.06xSSPE/0.005% N‐Lauroylsarcosine for 1 minute at RT. Finally, microarray slides were washed in acetonitril (Sigma‐Aldrich) for 30 s and dried in a nitrogen flow. Microarrays were scanned using an Agilent™ DNA Microarray Scanner (Agilent Technologies) at 5 µm resolution and 100% PMT setting. Microarray scans were quantified using Agilent™ Feature Extraction Software (Agilent Technologies, version 7.5.1). The hybridization setups can be found in Supplementary Table S1.

Single gene analysis

Raw expression data generated by the Feature Extraction Software were imported into the R statistical processing environment (http://www.r‐project.org), and analyzed using the LIMMA package in Bioconductor (http://www.bioconductor.org). Intra‐array normalization was performed using the loess algorithm. The intensity distributions between arrays were normalized using the Aquantile scaling algorithm. For each brain area, significant genes were determined by fitting a linear model to the normalized ratios and determining the contrast between PD and controls. Uncorrected P‐values were corrected for multiple testing using the Bonferroni correction algorithm. Genes with a corrected P‐value < 0.05 were considered significant.

Gene group analysis

For analysis on a gene group level, probes on the microarray were annotated using the Gene Ontology database (GO) (2) and the Kyoto Encyclopedia of Genes and Genomes (KEGG) (59). Briefly, Ensembl gene identifiers for the probe sequences were retrieved using the ProbeLynx probe annotation tool (67). GO annotation for the Ensembl identifiers was retrieved from the Ensembl database using the Bioconductor biomaRt package. Out of the 21 939 probes on the array, 10 813 probes were annotated with one or more GO identifiers, for a total of 3941 unique GO identifiers. The KEGG annotation was retrieved from the Bioconductor KEGG package. 2663 probes were annotated with one ore more KEGG identifiers, for a total of 181 unique KEGG classes. A gene group was then defined as a group of probes with either the same GO or KEGG annotation. Only gene groups with 5 to 150 member genes were selected for subsequent analyses.

Significantly dysregulated gene groups were identified using the functional class scoring approach as described in Pavlidis et al (61). Briefly, for each gene group, a functional class score (FCS) was defined as the mean of the log transformed uncorrected P‐values of differential expression between PD and controls of the gene group members. In order to determine the background distribution of FCSs for each gene group, a random gene set of the same size as the gene group under investigation was chosen, and the FCS was calculated. This procedure was repeated 200.000 times to obtain the simulated background distribution. The FCS P‐value was then defined as the fraction of simulated FCS scores that were higher than the actual FCS score for the gene group under investigation. The smallest possible P‐value therefore was 5e‐06. Groups with smaller P‐values were set to 2.5e‐06. FCS P‐values were corrected for multiple testing with the Benjamini and Hochberg False Discovery Rate model. P‐values < 0.05 were considered significant.

Reverse transcription and quantitative PCR

For the caudate nucleus and putamen, each sample was reversely transcribed using 500 ng RNA, oligo dT primers and SuperScript II™ Reverse Transcriptase (Invitrogen, Carlsbad, CA, USA). 1/200th of the total cDNA yield was used for all quantification reactions. Because of lower yields, 250 ng of SN RNA was used as input for the cDNA synthesis reaction, and 1/100th of the total cDNA yield was used for SN quantification reactions. Transcript quantifications were carried out on an ABI 7300 sequence detection system (Applied Biosystems, Foster City, CA, USA). Each reaction was performed with the appropriate amount of cDNA, 3 pmol of forward and reverse primers and 10 uL 2x SYBR^® green ready reaction mix (Applied Biosystems) in a total volume of 20 µL. Forward and reverse primer sequences can be found in Supplementary Table S2. To check for primer dimers, dissociation curves were generated for all wells. For between‐sample normalization, genes were selected for each brain area that were not regulated between PD and control using expression stability measurements and geometric averaging of multiple internal reference genes (76). Briefly, an initial selection of 10 genes was made based on the gene expression data and reference gene data from literature. The most stable genes between paradigms were selected, and for each sample, a normalization factor was determined by geometric averaging of the expression values of these genes. For the SN, ACTB, MRPL24 and DHX16 were used for normalization purposes. For the caudate nucleus, ACTB, GAPDH, PRPSAP1 and UFM1 were used. For the putamen, ACTB, GAPDH, GOT2, DHX16 and FAM96B were used. Genes were selected for quantitative PCR (qPCR) validation in two ways: by random selection from the list of differentially expressed genes to technically validate the microarray data, and by selecting genes based upon their possible involvement in the dysregulated processes as determined by the gene group analysis. Genes of interest were quantified in the following manner. For each primer pair, relative amounts were calculated by means of the formula primer.eff ∧ (Ct‐minimal Ct). The resulting raw relative amounts were scaled using the normalization factor to yield corrected relative amounts. P‐values between PD and control were calculated with the Mann‐Whitney U test using the corrected relative amounts. P‐values < 0.05 were considered significant.

Cell density measurements

In order to estimate the neuron density in the SN, for each of the seven PD and eight control subjects, five 6‐µm thick sections were randomly taken from the formalin‐fixed, paraffin‐embedded mesencephalon contralateral to the frozen side that was used for the microarray and qPCR experiments. Sections were stained with hematoxylin‐eosin. Pictures were taken using a Zeiss Axioskop microscope with Neofluar objectives (Carl Zeiss MicroImaging GmbH, Oberkochen, Germany) and an EvolutionMP™ camera and Image Pro Plus™ (version 5, Media Cybernetics, Bethesda, MD, USA). Pictures were analyzed using a custom‐build macro in Image Pro Plus™.

Two templates were constructed that represented 2 × 1 mm or 1.5 × 1 mm rectangles, which corresponded to the dimensions of dissected tissue samples used in the microarray and qPCR studies. For each section, the part of the SN that contained the highest density of neuromelanin was identified by eye at 2.5× magnification. A composite picture of this area was taken by combining 6 to 12 pictures at 2.5× magnification. In this composite picture, the area with the highest neuromelanin density was outlined using the template best fitting this area. The outlined area was then divided into subsections by a macro for cell counting at 20× magnification. For each subsection, a picture at 20× magnification was made, and structures of interest were counted by visual identification. Two cell types were included in the counting experiment: large neuromelanin‐positive neurons with a nucleolus, and large neuromelanin‐negative neurons with a nucleolus. These two types of neurons were counted separately. To prevent double counting, only neurons containing a nucleolus were counted. This counting procedure, which was judged to be the best for the thin (6 µm) section used 3, 34), is based upon the principle that nucleoli can be considered as hard particles that will not be sectioned by a microtome knife, but instead are pushed either in or out of the paraffin when hit by the knife (12). No nuclei with double nucleoli were observed. The nucleoli could be easily identified at 20× magnification. When in doubt, identification was made using a 40× objective. The estimated densities of the two neuronal types per cubic millimeter were calculated by dividing the number of identified structures by the measured area, and by correcting for section thickness. The final density per subject was defined as the average density of the five sections. As the quantification of neuronal structures based on counting nucleoli might not be related to the actual neuronal number in a one‐to‐one fashion, we expressed all counts as percentages of the average count of all structures in the control subjects (63). P‐values between PD and control were calculated with the Mann‐Whitney U test. P‐values < 0.05 were considered significant.

RESULTS

Clustering of gene expression profiles

In order to identify differences in gene expression between PD and controls, we performed microarray experiments on the SN, caudate nucleus and putamen of four PD patients and four matched controls. First, a hierarchical cluster analysis was applied to the microarray dataset in order to detect possible outliers (Figure 1). Only genes with consistently high expression levels throughout all subjects and brain areas were used (intensity >256 for all measurements, 6952 genes in total) to reduce the possible confounding effects of genes with low signal‐to‐noise ratios. The largest differences in expression were found between brain areas: the SN samples cluster apart from the caudate nucleus and putamen. Within the SN cluster, there was also full separation between PD and control samples, indicating substantial differences in gene expression. Unexpectedly, one control SN sample (sample ID 00‐022, or CTRL2) clustered apart from all samples. A comparison of the expression pattern of CTRL2 with the average expression pattern of all other SN samples (both PD and control) revealed a strong decrease in neuron‐specific transcripts in CTRL2 (data not shown). For example, SYT13 levels were only at 5% of combined PD and control levels. Similar low levels were found for, among others, NEFH (5%), MAP2 (18%) and NSE (20%). In contrast, several white matter‐related genes were overexpressed in CTRL2. These transcripts included VIM (1132% of combined PD and control levels), PMP22 (1392%) and FN1 (373%). We therefore concluded that the sample analyzed from the CTRL2 SN consisted of a different tissue type, and discarded the sample from further analysis.

Figure 1.

Clustering of microarray data. Hierarchical clustering analysis on array probes with a background corrected intensity >256 for all measurements (6952 genes in total). Note the aberrant position of CTRL2 SN. CAU = caudate nucleus; PUT = putamen; SN = substantia nigra pars compacta.

Detection of transcriptional alterations in the SN, caudate nucleus and putamen in PD

For each area of interest, the microarray data sets were analyzed to detect transcriptional alterations between PD and control. In the SN, 287 genes out of the 21 939 genes on the array were significantly changed in PD. The majority of the significant genes (240) was downregulated in PD, whereas 47 genes were upregulated. Expression changes in the target areas of the nigrostriatal pathway were less pronounced. In the caudate nucleus, 16 genes were found to be differentially expressed (six genes upregulated in PD, 10 genes downregulated). In the putamen, we found two genes significantly upregulated in PD, and two downregulated genes (Table 2 and Supplementary Table S3).

Table 2.

Overview of the most significant expression changes in the PD SN, caudate nucleus and putamen, as determined by microarray analysis. For the SN, only the 50 most significant genes are given (the full list of significantly changed genes in the SN can be found in Supplementary Table S3). Abbreviations: DE = differential expression, as fraction of control levels; P‐value = Bonferroni‐adjusted P‐value; PD = Parkinson's disease; SN = substantia nigra pars compacta.

Substantia nigra
Systematic name	Gene name	Description	DE	P‐value
Downregulated genes
NM_000158	GBE1	Glucan (1,4‐alpha‐), branching enzyme 1 (glycogen branching enzyme, Andersen disease, glycogen storage disease type IV)	0.18	0.00002
NM_033495	KLHL13	Kelch‐like 13 (Drosophila)	0.24	0.00003
AL117636		MRNA; cDNA DKFZp434H205 (from clone DKFZp434H205)	0.34	0.00004
NM_004411	DYNC1I1	Dynein, cytoplasmic 1, intermediate chain 1	0.26	0.00005
NM_002863	PYGL	Phosphorylase, glycogen; liver (Hers disease, glycogen storage disease type VI)	0.26	0.00005
AB002376	CAST1	CAZ‐associated structural protein	0.28	0.00011
NM_000345	SNCA	Synuclein, alpha (non A4 component of amyloid precursor)	0.27	0.00016
NM_002928	RGS16	Regulator of G‐protein signalling 16	0.22	0.00017
AB014731	DENR	Density‐regulated protein	0.41	0.00025
NM_018092	NETO2	Neuropilin (NRP) and tolloid (TLL)‐like 2	0.46	0.00029
NM_014332	SMPX	Small muscle protein, X‐linked	0.31	0.00031
NM_006670	TPBG	Trophoblast glycoprotein	0.18	0.00035
NM_017999	RNF31	Ring finger protein 31	0.32	0.00037
NM_006350	FST	Follistatin	0.17	0.00037
BF035258	AKAP12	A kinase (PRKA) anchor protein (gravin) 12	0.39	0.00043
AW974789	TTMA	Two transmembrane domain family member A	0.47	0.00047
AF428101	DACH2	Dachshund homolog 2 (Drosophila)	0.35	0.00053
NM_018700	TRIM36	Tripartite motif‐containing 36	0.30	0.00054
AK024251	SLITRK5	SLIT and NTRK‐like family, member 5	0.46	0.00056
NM_005639	SYT1	Synaptotagmin I	0.23	0.00062
AB020710	EHBP1	EH domain binding protein 1	0.49	0.00063
H06068		yl72g10.s1 Soares infant brain 1NIB Homo sapiens cDNA clone IMAGE:43489 3‐, mRNA sequence	0.46	0.00064
AF225987	SCN3A	Sodium channel, voltage‐gated, type III, alpha	0.47	0.00067
AL390172		Homo sapiens genomic DNA; cDNA DKFZp547I204 (from clone DKFZp547I204)	0.35	0.00068
AK055427	DGKH	Diacylglycerol kinase, eta	0.31	0.00075
NM_022726	ELOVL4	Elongation of very long chain fatty acids (FEN1/Elo2, SUR4/Elo3, yeast)‐like 4	0.46	0.00086
NM_006186	NR4A2	Nuclear receptor subfamily 4, group A, member 2	0.20	0.00089
NM_031850	AGTR1	Angiotensin II receptor, type 1	0.07	0.00095
NM_014309	RPL41	Ribosomal protein L41	0.38	0.00101
R42166		yf99d08.s1 Soares infant brain 1NIB Homo sapiens cDNA clone IMAGE:30780 3‐, mRNA sequence	0.25	0.00103
NM_012189	CABYR	Calcium binding tyrosine‐(Y)‐phosphorylation regulated (fibrousheathin 2)	0.49	0.00112
NM_006113	VAV3	Vav 3 oncogene	0.34	0.00113
NM_014905	GLS	Glutaminase	0.43	0.00141
J04621	SDC2	Syndecan 2 (heparan sulfate proteoglycan 1, cell surface‐associated, fibroglycan)	0.37	0.00142
AF070623	LOC138046	Hypothetical protein LOC138046	0.38	0.00144
NM_004796	NRXN3	Neurexin 3	0.49	0.00148
AK025615	BCAT1	Branched chain aminotransferase 1, cytosolic	0.31	0.00153
NM_003979	GPRC5A	G protein‐coupled receptor, family C, group 5, member A	0.30	0.00170
AF339796	HS6ST3	Heparan sulfate 6‐O‐sulfotransferase 3	0.48	0.00172
AL137498	SCN2A2	Sodium channel, voltage‐gated, type II, alpha 2	0.45	0.00174
NM_000809	GABRA4	Gamma‐aminobutyric acid (GABA) A receptor, alpha 4	0.29	0.00182
AW294729		Transcribed locus, moderately similar to NP_509344.1 HIStone family member (his‐71) [Caenorhabditis elegans]	0.55	0.00185
NM_003138	SRPK2	Homo sapiens SFRS protein kinase 2 (SRPK2), transcript variant 2, mRNA	0.53	0.00186
AF397397	OLFM3	Olfactomedin 3	0.31	0.00197
NM_005100	AKAP12	A kinase (PRKA) anchor protein (gravin) 12	0.43	0.00220
AB046788	ROBO2	Roundabout, axon guidance receptor, homolog 2 (Drosophila)	0.32	0.00232
AV707343		V707343 ADB Homo sapiens cDNA clone ADBBSH02 5‐, mRNA sequence	0.25	0.00238
Upregulated genes
NM_018276	SSH3	Slingshot homolog 3 (Drosophila)	1.66	0.00222
BC013923	SOX2	SRY (sex determining region Y)‐box 2	1.86	0.00117
BI823317		603041226F1 NIH_MGC_115 Homo sapiens cDNA clone IMAGE:5182015 5‐, mRNA sequence	2.18	0.00136

Caudate nucleus
Downregulated genes
NM_001888	CRYM	Crystallin, mu	0.43	0.02542
NM_006211	PENK	Proenkephalin	0.45	0.03796
NM_030615	KIF25	Kinesin family member 25	0.56	0.01649
NM_016454	TMEM85	Transmembrane protein 85	0.61	0.00577
NM_006680	ME3	Malic enzyme 3, NADP(+)‐dependent, mitochondrial	0.62	0.02043
NM_000783	CYP26A1	Cytochrome P450, family 26, subfamily A, polypeptide 1	0.64	0.02402
AL538117	CXADR	Coxsackie virus and adenovirus receptor	0.65	0.02382
NM_015417	C20orf28	Chromosome 20 open reading frame 28	0.68	0.03102
NM_004087	DLG1	Discs, large homolog 1 (Drosophila)	0.70	0.04833
NM_021252	RAB18	RAB18, member RAS oncogene family	0.71	0.03775
Upregulated genes
AL136693	CYBRD1	Cytochrome b reductase 1	1.39	0.03087
NM_003243	TGFBR3	Transforming growth factor, beta receptor III (betaglycan, 300kDa)	1.47	0.01079
NM_005413	SIX3	Sine oculis homeobox homolog 3 (Drosophila)	1.52	0.01266
NM_005859	PURA	Purine‐rich element binding protein A	1.54	0.04095
AK026152	CTSH	Cathepsin H	1.64	0.00460
NM_002343	LTF	Lactotransferrin	2.57	0.00004

Putamen
Downregulated genes
BF381899	GNPDA2	Glucosamine‐6‐phosphate deaminase 2	0.52	0.00000
AC016696		Homo sapiens BAC clone RP11‐130P22 from 2, complete sequence.	0.71	0.00313
Upregulated genes
NM_015556	SIPA1L1	Signal‐induced proliferation‐associated 1 like 1	1.28	0.00944
NM_001124	ADM	Adrenomedullin	1.80	0.02671

To validate the transcriptional alterations observed on the microarray, and to extrapolate the observed expression changes to the general PD population, we performed qPCR experiments on an extended set of seven PD patients and seven matched controls (Figure 2 and Supplementary Table S4). We were able to verify the direction of regulation of all transcripts investigated. Importantly, we detected significant differences between PD and controls for 26 out of the 34 genes under investigation (P < 0.05), with a further three genes nearing significance (P < 0.1). Also, qPCR and microarray fold changes exhibited a very strong linear correlation (r ² = 0.86, P = 2.67e–15, Figure 3). This suggests that the observed alterations in the microarray experiment are indeed of relevance in a larger PD population.

Figure 2.

Transcriptional alterations in the PD SN. Selection of expression changes observed in the PD SN, as determined by qPCR on seven PD patients and seven controls. For each gene, the average of control measurements was set to 100%. All alterations are similar to those observed in the microarray experiment. (A) downregulated transcripts in PD. (B) upregulated transcripts in PD. (C) Alterations in truncated (T1) and full‐length (FL) forms of TrkB. Note the shift in the ratio TrkB.T1/TrkB.FL in PD. P‐values determined by the Mann‐Whitney U test. *P‐value < 0.05; **P‐value < 0.01. Abbreviations: PD = Parkinson's disease; SN = substantia nigra pars compacta.

Figure 3.

Comparison microarray and qPCR data. Correlation between microarray and qPCR fold changes for all 34 transcripts investigated by qPCR in the SN. The line represents the least squares fitted curve through the data points. Fold changes are given as log2 (PD/control). Abbreviations: SN = substantia nigra pars compacta; PD = Parkinson's disease.

Furthermore, several of the observed changes in SN gene expression corroborated earlier observations in PD. The most significantly downregulated gene in our SN dataset, GBE1, was also found to be significantly downregulated in the study performed by Grunblatt et al (30). Other examples of transcriptional alterations that were in agreement with earlier findings in PD include SYT1, TRIM36, MDH1, NSF, SASH1, ALDH1A1, DSCR1L1 and SNX10 30, 52, 81). Furthermore, we report transcriptional downregulation of both SNCA and NURR1, which are well known for their roles in familial PD. It is also noteworthy that in the studies of Miller et al and Moran et al, the number of downregulated genes in the PD SN by far surpasses that of upregulated genes, an observation we corroborated 52, 54).

We have also identified dysregulations in transcription levels of several genes previously not implicated in PD, which might have a role as new players in PD neurodegeneration. The most pronounced changes are described below.

Altered expression of genes involved in neurotrophin signaling

A substantial number of genes involved in neurotrophin signaling are transcriptionally altered in the PD SN. A reduction to 36% of control of RIT2 levels, a specific regulator of neurotrophin signaling (70), was detected. Transcript levels of MAGEE1, possibly involved in neurotrophin signaling via the p75 neurotrophin receptor (1), showed a similar decrease. The expression of NDN was reduced to 38% of normal levels in PD SN. NDN facilitates TrkA signaling to promote neuronal survival (44). Also, we observed a significant downregulation of DOK6, which is involved in the GDNF signaling pathway (17). DLK1, which is known to be regulated by GDNF (15), also exhibited a severe decline in expression (22% of normal levels). Interestingly, the expression of PHLPP was increased. PHLPP is known to decrease intracellular neurotrophin signaling by dephosphorylating Akt (28). We also observed elevated transcript levels of a truncated form of TrkB (TrkB‐T1), the receptor for BDNF. TrkB‐T1 lacks the intracellular kinase domain, and is therefore unable to induce the intracellular BDNF signaling cascade. Interestingly, the splice variants encoding for the full‐length versions of TrkB were significantly reduced in PD, as determined by qPCR (see Figure 2).

Changes in genes encoding axon guidance cues

The expression levels of several genes implicated in axon guidance are changed in the PD SN. The repulsive guidance protein RGMA was upregulated to 197% of control levels. Furthermore, transcript levels of ROBO2, the receptor for Slit2, were significantly reduced. The expression of NETO2 and SLITRK5 were both decreased in the PD SN. Although there is little knowledge on the cellular function of these proteins, their domain structure is suggestive for a role in axon guidance in development. SDC2, a heparan sulfate proteoglycan possibly involved in regulating the function of some axon guidance molecules, was downregulated in the PD SN. Finally, although not significantly changed, a subset of patients showed elevated levels of the axon guidance molecule Sema5A (see Supplementary Table S4).

Other gene changes in PD SN, caudate nucleus and putamen

One of the most strongly upregulated genes in the PD SN was the gene coding for the iron‐binding protein lactotransferrin (LTF), a finding that was replicated in the caudate nucleus. Other noteworthy gene changes in the PD SN include AGTR1, VMAT2, ALDH1A1, DSCR1L1 and PTS. In the PD caudate nucleus, in addition to the upregulation of LTF, CYBRD1, involved in iron homeostasis, was also modestly upregulated. Furthermore, we observed expression changes in the mitochondrial enzyme ME3 and PENK. In the PD putamen, the most significantly changed gene was GNPDA2. Furthermore, a strong increase in transcriptional levels of adrenomedullin was detected.

Identification of dysregulated gene groups in the PD caudate, putamen and SN

One of the main advantages of global gene expression profiling is the ability to detect concerted changes in functionally related groups of genes. We applied an unbiased FCS approach to identify those GO or KEGG classes that were significantly dysregulated in PD. Also, the average expression change for each gene group was calculated to assess its overall directional change in transcript levels. The FCS calculations identified 75 GO classes and 29 KEGG classes as being differentially expressed in the SN between PD and control. Eight KEGG classes and one GO class were found to be changed in the putamen, and for the caudate nucleus, no statistically significant differences were observed (Table 3).

Table 3.

Altered gene groups in the PD SN and putamen. Gene groups identified by functional class scoring as significantly different (False Discovery Rate‐corrected P‐value < 0.05) between PD and control, based on the microarray data. Class size: the number of genes with a particular GO or KEGG annotation on the array. Fold change: the mean fold change of all group members, as fraction of control levels. Abbreviations: PUT = putamen; PD = Parkinson's disease; SN = substantia nigra pars compacta; GO = Gene Ontology database; KEGG = Kyoto Encyclopedia of Genes and Genomes.

Identifier	Description	P‐value	Class size	Fold change	Brain area
Synaptic transmission
hsa04540	Gap junction	0.0002	71	0.92	SN
GO:0019717	Synaptosome	0.0002	17	0.66	SN
GO:0007264	Small GTPase mediated signal transduction	0.0004	149	0.94	SN
GO:0045211	Postsynaptic membrane	0.0007	60	0.83	SN
GO:0007268	Synaptic transmission	0.0016	146	0.92	SN
GO:00052161	Ion channel activity	0.0016	96	0.94	SN
GO:0045202	Synapse	0.0019	34	0.80	SN
hsa04530	Tight junction	0.0027	75	1.00	SN
GO:0008021	Synaptic vesicle	0.0146	33	0.79	SN
hsa04540	Gap junction	0.0179	70	0.99	PUT
GO:0006887	Exocytosis	0.0247	31	0.81	SN
GO:0001518	Voltage‐gated sodium channel complex	0.0247	9	0.85	SN
GO:0015075	Ion transporter activity	0.0293	9	1.01	SN
GO:0004890	GABA‐A receptor activity	0.0304	16	0.76	SN
GO:0009966	Regulation of signal transduction	0.0304	12	0.87	SN
GO:0005248	Voltage‐gated sodium channel activity	0.0304	12	0.95	SN
GO:0030426	Growth cone	0.0320	8	0.73	SN
GO:0006814	Sodium ion transport	0.0325	70	0.92	SN
GO:0008503	Benzodiazepine receptor activity	0.0331	6	0.59	SN
GO:0005234	Glutamate‐gated ion channel activity	0.0341	22	0.89	SN
GO:0030424	Axon	0.0349	8	0.76	SN
GO:0006904	Vesicle docking during exocytosis	0.0367	16	0.84	SN
GO:0030594	Neurotransmitter receptor activity	0.0393	19	0.77	SN
GO:0008308	Voltage‐gated ion‐selective channel activity	0.0440	7	0.73	SN
GO:0006836	Neurotransmitter transport	0.0440	28	0.80	SN
hsa04070	Phosphatidylinositol signaling system	0.0459	71	0.95	SN
GO:0017137	Rab interactor activity	0.0492	15	0.86	SN
Energy synthesis
hsa00193	ATP synthesis	0.0002	26	0.70	SN
hsa00190	Oxidative phosphorylation	0.0002	79	0.75	SN
GO:0046961	Hydrogen‐transporting atpase activity, rotational mechanism	0.0002	30	0.74	SN
GO:0046933	Hydrogen‐transporting ATP synthase activity, rotational mechanism	0.0002	29	0.75	SN
GO:0015986	ATP synthesis coupled proton transport	0.0002	31	0.75	SN
GO:0016469	Proton‐transporting two‐sector atpase complex	0.0002	29	0.75	SN
hsa00190	Oxidative phosphorylation	0.0004	79	0.91	PUT
GO:0015992	Proton transport	0.0007	37	0.79	SN
GO:0003954	NADH dehydrogenase activity	0.0037	30	0.79	SN
GO:0008137	NADH dehydrogenase (ubiquinone) activity	0.0040	30	0.79	SN
GO:0005743	Mitochondrial inner membrane	0.0043	58	0.86	SN
GO:0005753	Proton‐transporting ATP synthase complex (sensu Eukaryota)	0.0095	7	0.68	SN
GO:0019866	Inner membrane	0.0157	27	0.80	SN
GO:0042623	Atpase activity, coupled	0.0247	16	0.90	SN
GO:0004129	Cytochrome‐c oxidase activity	0.0304	17	0.79	SN
hsa00193	ATP synthesis	0.0336	26	0.90	PUT
GO:0006120	Mitochondrial electron transport, NADH to ubiquinone	0.0492	19	0.81	SN
Cytoskeleton
GO:0005874	Microtubule	0.0002	117	0.91	SN
hsa04810	Regulation of actin cytoskeleton	0.0026	144	1.02	SN
GO:0007018	Microtubule‐based movement	0.0182	52	0.86	SN
GO:0003774	Motor activity	0.0304	64	0.99	SN
GO:0000776	Kinetochore	0.0319	15	0.79	SN
GO:0008092	Cytoskeletal protein binding	0.0320	35	1.08	SN
GO:0005875	Microtubule associated complex	0.0407	46	0.93	SN
GO:0000146	Microfilament motor activity	0.0444	11	1.03	SN
GO:0030048	Actin filament‐based movement	0.0492	13	0.96	SN
Metabolism
GO:0006096	Glycolysis	0.0002	35	0.82	SN
hsa00710	Carbon fixation	0.0086	18	0.74	SN
hsa00020	Citrate cycle (TCA cycle)	0.0086	14	0.90	SN
hsa00260	Glycine, serine and threonine metabolism	0.0106	33	1.05	SN
hsa00564	Glycerophospholipid metabolism	0.0153	56	0.95	SN
hsa00740	Riboflavin metabolism	0.0176	11	0.88	SN
hsa00330	Arginine and proline metabolism	0.0226	40	1.01	PUT
hsa00220	Urea cycle and metabolism of amino groups	0.0336	15	1.01	PUT
hsa00620	Pyruvate metabolism	0.0385	35	0.86	SN
hsa00720	Reductive carboxylate cycle (CO2 fixation)	0.0385	7	0.89	SN
hsa00910	Nitrogen metabolism	0.0409	18	0.99	PUT
hsa00340	Histidine metabolism	0.0417	35	0.94	SN
hsa00480	Glutathione metabolism	0.0417	18	0.97	SN
hsa00410	Beta‐Alanine metabolism	0.0431	20	0.98	SN
hsa00650	Butanoate metabolism	0.0459	39	0.96	SN
hsa00271	Methionine metabolism	0.0459	14	1.02	SN
hsa00251	Glutamate metabolism	0.0476	24	0.91	SN
Extracellular matrix
hsa04512	ECM‐receptor interaction	0.0013	59	1.19	SN
hsa04510	Focal adhesion	0.0027	143	1.09	SN
GO:0005518	Collagen binding	0.0060	11	1.16	SN
hsa04514	Cell adhesion molecules (cams)	0.0122	84	1.10	SN
hsa04510	Focal adhesion	0.0179	137	1.01	PUT
GO:0007160	Cell‐matrix adhesion	0.0182	48	1.12	SN
GO:0005201	Extracellular matrix structural constituent	0.0222	52	1.07	PUT
Neurodegeneration
hsa05010	Alzheimer's disease	0.0086	15	0.93	SN
hsa01510	Neurodegenerative disorders	0.0102	26	0.89	SN
hsa05020	Parkinson's disease	0.0191	14	0.77	SN
hsa05040	Huntington's disease	0.0459	18	1.06	SN
Ubiquitin‐proteasome system
hsa03050	Proteasome	0.0002	20	0.73	SN
GO:0051082	Unfolded protein binding	0.0058	114	0.90	SN
GO:0005839	Proteasome core complex (sensu Eukaryota)	0.0146	13	0.84	SN
GO:0005838	Proteasome regulatory particle (sensu Eukaryota)	0.0325	7	0.78	SN
GO:0006511	Ubiquitin‐dependent protein catabolism	0.0393	98	0.96	SN
Other
GO:0005829	Cytosol	0.0002	134	0.89	SN
GO:0043234	Protein complex	0.0002	30	0.74	SN
GO:0007417	Central nervous system development	0.0002	68	0.95	SN
GO:0005516	Calmodulin binding	0.0002	110	1.00	SN
GO:0003924	Gtpase activity	0.0002	124	0.89	SN
hsa04020	Calcium signaling pathway	0.0013	129	0.97	SN
hsa05110	Cholera—infection	0.0016	35	0.85	SN
GO:0007517	Muscle development	0.0020	96	1.05	SN
GO:0004618	Phosphoglycerate kinase activity	0.0022	6	0.89	SN
GO:0051301	Cell division	0.0058	115	1.01	SN
GO:0005386	Carrier activity	0.0070	9	0.84	SN
GO:0004721	Phosphoprotein phosphatase activity	0.0093	37	0.99	SN
GO:0004298	Threonine endopeptidase activity	0.0094	14	0.82	SN
GO:0030145	Manganese ion binding	0.0094	84	0.99	SN
GO:0045786	Negative regulation of cell cycle	0.0157	65	1.07	SN
GO:0004252	Serine‐type endopeptidase activity	0.0157	12	1.00	SN
GO:0051258	Protein polymerization	0.0173	16	0.68	SN
hsa04110	Cell cycle	0.0176	82	1.01	SN
GO:0004553	Hydrolase activity, hydrolyzing O‐glycosyl compounds	0.0247	8	0.72	SN
GO:0030529	Ribonucleoprotein complex	0.0247	64	1.02	SN
GO:0007420	Brain development	0.0269	31	0.96	SN
GO:0016310	Phosphorylation	0.0304	10	0.84	SN
GO:0003705	RNA polymerase II transcription factor activity, enhancer binding	0.0333	11	1.05	PUT
hsa00280	Valine, leucine and isoleucine degradation	0.0336	36	0.99	PUT
hsa00970	Aminoacyl‐trna synthetases	0.0385	27	0.84	SN
GO:0008020	G‐protein coupled photoreceptor activity	0.0444	6	0.80	SN
GO:0006821	Chloride transport	0.0444	32	0.93	SN
GO:0004176	ATP‐dependent peptidase activity	0.0451	7	0.71	SN
hsa00770	Pantothenate and coa biosynthesis	0.0457	17	0.95	SN
hsa00290	Valine, leucine and isoleucine biosynthesis	0.0459	9	0.90	SN
hsa04910	Insulin signaling pathway	0.0459	97	0.94	SN
hsa00960	Alkaloid biosynthesis II	0.0496	14	1.06	SN

Alterations in gene groups involved in synaptic transmission in the PD SN

Many gene groups related to synaptic transmission are compromised in the PD SN. For example, the average transcriptional activity of the 17 genes annotated as “synaptosome” (GO:0019717) was reduced to 66% of control levels. Similar changes were detected for, among others, the classes “synapse” (GO:0045202), “long‐term potentiation” (hsa04720) and “long‐term depression” (hsa04730). The average reduction of gene expression in synaptic transmission‐related gene groups in PD was 15%. Although this seems a relatively mild change in expression levels, one should take into account that affected gene groups contain both changed and unchanged genes: the range of transcriptional changes runs up to a 77% reduction (SYT1). Related to lowered synaptic transmission in the SN, a dysregulation of “gap junction” (hsa04540) was detected in the putamen.

Reduction of gene groups involved in oxidative phosphorylation in PD SN and putamen

We observed a strong reduction in transcriptional levels of gene groups involved in oxidative phosphorylation and ATP synthesis in the PD SN. This was reflected, among others, by the dowregulation of the KEGG classes “oxidative phosphorylation” (hsa00190) and “ATP synthesis” (hsa00193), and the GO classes “ATP synthesis coupled proton transport” (GO:0015986) and “NADH dehydrogenase activity” (GO:0003954). Average transcript levels for these gene groups were reduced to ∼70% of control. Interestingly, we also detected a reduction of the KEGG gene groups “oxidative phosphorylation” and “ATP synthesis” in the putamen (∼90% of control levels). Additionally, a number of GO classes involved in oxidative phosphorylation showed a downward trend in the putamen (P < 0.08, data not shown).

Downregulation of ubiquitin‐proteasome gene groups in PD SN

The ubiquitin‐proteasome system (UPS) is dysregulated in PD. We observed a downregulation of the classes “proteasome” (hsa03050), “unfolded protein binding” (GO:0051082), “proteasome core complex (sensu Eukaryota)” (GO:0005839), “proteasome regulatory particle (sensu Eukaryota)” (GO:0005838) and “ubiquitin‐dependent protein catabolism” (GO:0006511). Average transcript levels for these groups were reduced by 15%.

Alterations in gene groups encoding for proteins involved in cytoskeletal structure and microtubule‐based transport in PD SN

Noteworthy is the dysregulation of gene groups involved in cytoskeletal structure and microtubule‐based transport in the PD SN. Some groups were downregulated, such as “microtubule‐based movement” (GO:0007018), whereas we observed an increased average expression of transcripts belonging to “cytoskeletal protein binding” (GO:0008092). Possibly related to this, the gene groups “axon guidance” (hsa04360) and “growth cone” (GO:0030426) were also changed in the PD SN.

Other gene group changes in PD

A transcriptional upregulation (average expression increase of ∼10% over control levels) in the PD SN was found for extracellular matrix (ECM) protein groups such as “collagen binding” (GO:0005518) and “ECM‐receptor interaction” (hsa04512). In the putamen, a similar increase was found for the group “extracellular matrix structural constituent” (GO:0005201). Finally, we report expression changes in the gene groups “Parkinson's disease” (hsa05020), “Alzheimer's disease” (hsa05010) and “neurodegenerative disorders” (hsa01510) in the SN of PD patients.

Cell density measurements

The most typical and striking neuropathological feature of PD is the loss of neuromelanin‐positive neurons from the SN. It is estimated that around 80% of the DAergic neurons in the SN are lost in the end stages of PD (73). We and others found that DAergic neurons that are still present in end‐stage PD tend to appear in clusters (18). We performed cell countings, in order to estimate neuronal densities in the relatively spared parts of the PD SN and in the control SN, on tissue sections of the contralateral SN of the seven PD subjects and eight controls used in the qPCR and microarray experiments. We distinguished large neuromelanin‐positive neuronal profiles and large neuromelanin‐negative neuronal profiles. The number of small neuromelanin‐negative structures was negligible. We observed a significant decrease of 51% in neuromelanin‐positive profiles in the PD SN (Mann‐Whitney U test, P = 0.015) relative to control. Surprisingly, we measured a relative increase of 104% in PD neuromelanin‐negative neurons (Mann‐Whitney U test, P = 0.049). Overall, the relative total neuronal density was estimated to be decreased by 29% in PD as compared with control (Mann‐Whitney U test, P = 0.037) (Figure 4).

Figure 4.

Changes in relative density of neuronal structures in the PD SN. Estimated neuronal densities relative to the average total densities of control, of large neurons in the relatively spared parts of the PD SN that were used in the present study (n = 7), and control SN (n = 8). A distinction was made between neuromelanin‐positive (pigmented) and neuromelanin‐negative (nonpigmented) neurons. Note the relative increase in neuromelanin‐negative neurons in PD, and the rather mild decrease in total neuronal density in the sampled areas of the PD SN. *Mann‐Whitney U test P‐value < 0.05. Abbreviations: PD = Parkinson's disease; SN = substantia nigra pars compacta.

DISCUSSION

To our knowledge, this is the first study to examine global changes in gene expression in relatively spared areas of the PD SN. Cell counting experiment enabled us to relate the microarray data to the degree of neuronal loss. We observed a moderate decrease in total neuronal density of 29% in the sampled part of the PD SN. Subpopulations of the SN are differentially vulnerable in PD. For example, the lateral and dorsal parts of the SN show a 21% and 57% loss of DAergic neurons respectively, whereas the pars compacta is more severely affected, with a decrease of 86% (18). Although differentially affected, all areas of the PD SN do exhibit cell loss during disease progression. Thus, gene expression measurements from less affected parts of the PD SN might provide insight in the molecular biological processes leading to neuronal degeneration in PD. Transcriptional alterations that occur before massive neuronal demise has taken place might represent early, causative molecular changes instead of events that are the consequence of the disease process. The observed 51% decrease in neuromelanin‐positive neurons in the SN in combination with a 104% increase in neuromelanin‐negative neurons shows that many neurons are still present, but are in the process of losing their neuromelanin content, which also explains the presence of extracellular neuromelanin granules in the PD neuropil. The observed decrease (29%) in total neuronal density in the PD SN is a confounding factor for the interpretation of gene expression data. The downregulation of a neuron‐specific transcript by 29% might be explained fully by the loss of neurons. It is thus noteworthy that all of the 240 genes significantly downregulated in the PD SN are reduced by more than 29%, and 149 genes by more than 50%. Moreover, the downregulation of the neuron‐specific transcript NURR1 by 82%, when correcting for the loss of neurons, still translates to a 58% reduction per neuron. Therefore, the observed changes in gene expression are possibly relevant for PD pathogenesis and cannot simply be explained by a loss of DAergic neurons. In apparent contrast, the average downregulation for the most significantly altered gene groups in our study usually does not exceed 29%. However, the mean expression of each gene group is calculated by averaging the expression values for all group members. Therefore, transcripts that are not altered in PD contribute to the gene group expression value, effectively underestimating the “true” transcriptional downregulation for the changed group members. For example, NURR1 is decreased by 82%, whereas the total decrease of the gene group “neurodegenerative disorders” (hsa01510) is 18%. We cannot exclude the possibility that changes in the expression of some genes in particular gene groups are caused by the loss of DAergic neurons, and do not represent PD‐specific transcriptional alterations in DAergic neurons. However, the observation of significant changes in multiple members of a specific gene group provides strong evidence for alterations in the processes in which these genes function.

Whole‐tissue expression studies possibly provide insights into disease mechanisms that would have been missed in a study on DAergic neurons alone. For example, in PD animal models, reactive astrocytes have an essential role in the trophic support of DAergic neurons by secreting neurotrophic factors such as GDNF and NGF 55, 56), and metabolizing potentially harmful extracellular DA by expression of MAOB and COMT. Interestingly, we observed a 50% increase in COMT expression in the PD SN. PD‐induced death of DAergic neurons should therefore be regarded as a disease of the entire SN, not only of the neurons.

It must be noted that we cannot exclude the possible confounding influence of DA replacement therapies on the observed transcriptional alterations between the PD and control, both on the level of the striatum, and the SN. This is, however, a limitation that is likely to be present in all studies comparing gene and protein expression between the PD and control brain.

Gene group changes in the PD SN

We detected significant alterations in several gene groups, notably those involved in oxidative phosphorylation, synaptic transmission, microtubular dynamics and microtubular‐based transport, and the UPS. The reduced levels of transcripts involved in synaptic transmission and ATP synthesis, combined with lowered expression levels of individual genes involved in DA transmission such as VMAT2 and TH, all point to a functionally compromised tissue. This corroborates the idea that the SN in PD is unable to provide sufficient levels of DA to the striatum. Adding to that, the reduced capacity of targeted protein degradation might underlie the formation of Lewy bodies and Lewy neurites 33, 60). Thus, the observed alterations in specific gene groups are consistent with the neuropathological manifestations of PD.

Several of the observed gene group changes are in line with current literature on PD. Mitochondrial dysfunction has long been implicated in PD neuropathology. Two other gene expression studies on the PD SN report reductions in ATP synthesis 30, 81). In the prefrontal cortex of PD patients, mitochondrial complex I subunits are oxidatively damaged and functionally impaired (38). Together with findings on mitochondrial complex I inhibition by rotenone and MPTP, and mutations in genes linked to oxidative phosphorylation, including DJ‐1 and PINK1, there is strong evidence for mitochondrial involvement in PD pathogenesis. Importantly, as we selectively investigated spared parts of the SN, mitochondrial alterations appear to play a role in the relatively early stages of DAergic neurodegeneration.

The severe reductions in synaptic transmission corroborate the microarray studies of Grunblatt et al and Miller et al (52), but now in an area of the SN where a significant number of DAergic neurons has survived. This makes it likely that synaptic transmission is decreased in SN neurons that are still present. Also, SNCA, reduced in the PD SN in our study and in a study by Kingsbury et al as determined by in situ hybridization (39), is thought to play a role in neurotransmission (45). It was recently shown that presynaptic aggregates of SNCA cause neurodegeneration in dementia with Lewy bodies (41). One explanation of the observed reduction might be that the surviving neurons in PD express lower levels of SNCA, making them less vulnerable to SNCA‐induced neurotoxicity.

Finally, the occurrence of protein aggregates suggests involvement of protein degradation pathways in PD neuropathology. We indeed observed reduced levels of transcripts of the UPS in surviving areas of the SN PD. Earlier studies also have shown alterations in proteasomal function within DAergic neurons in the SN in the end stage of PD (51). Duke et al reported on an overall loss of genes encoding for UPS proteins, which was correlated with expression of mitochondrial gene groups (25). In that and our own study, UCHL1 was found to be significantly decreased. UCHL1 is responsible for converting polyubiquitin chains into monomeric ubiquitin molecules. When UCHL1 levels are reduced, the available pool of ubiquitin might become too small to meet the demand for UPS‐mediated protein degradation. Indeed, reduced UPS activity leads to a familial form of PD, and mice expressing the mutant form of UCHL1 show DAergic neuron loss 69, 78).

In addition to UCHL1, several of the observed significant single gene changes are in agreement with previous reports. Examples of these genes include GBE1, SYT1, VMAT2, DSCR1L1 and ALDH1A1 30, 52, 81). There are striking similarities between the single gene and gene group alterations found in this study, literature on sporadic and familial end stage PD, and PD animal model systems. These studies have however not specifically targeted the spared parts of the PD SN. We therefore suggest that the observed alterations are already implicated before massive neuronal loss.

It is noteworthy to address the potential regulation of genes known to be involved in familial PD. Apart from the previously mentioned alterations in SCNA and UCHL1 levels, other genes implicated in familial PD could not be shown to be changed in expression: the expression of PARK2 was too low to be reliably measured, DJ‐1 is reduced by 30% (but this reduction is not statistically significant), PINK1 is not regulated, and a probe for LRRK2 was not present on the array (data not shown). Based on this experiment, it is difficult to draw definitive conclusions on the involvement of the aforementioned genes in sporadic PD, however, other genome‐wide expression studies do not report changes in the expression level of these genes either 30, 52, 81). This does not imply that the biological pathways in which familial PD genes play a role are not important in sporadic PD. For example, aggregation of aberrant proteins is a key feature of sporadic PD (33), and UCHL1 and PARK2 are both involved in protein (de)ubiquitination.

Reduction of neurotrophic support and microtubular alterations in PD

We identified a number of novel single gene changes that are possibly involved in PD pathogenesis. Several genes implicated in neurotrophic support display altered expression. Neurotrophic factors, including BDNF, are reduced in PD, and the role of neurotrophic factors in animal models for PD has been studied extensively [reviewed in (20), (40) and (72)]. GDNF is an important factor in the development and survival of mesencephalic DAergic neurons, and protects DAergic neurons against MPTP‐induced neurotoxic stress 14, 42). We observed reductions in the expression of RIT2, DLK1, NDN, MAGEE1 and DOK6, all involved in neurotrophic signaling 15, 17, 42, 43, 44, 70). DLK1 is strongly coupled to both TH expression and neurotrophic signaling. Silencing of DLK1 causes a reduction in TH and VMAT2 expression (4), and influences the GDNF pathway by a reduction in Ret expression (15). Furthermore, we observed an increase in transcript levels of PHLPP, an inhibitor of Trk signaling by dephosphorylating Akt (28). Interestingly, both NDN and MAGEE1 are located in the chromosomal region deleted in Prader‐Willi syndrome, and mouse models with a deleted paternal copy of the NDN gene show increased apoptosis in NGF‐dependent neurons (44). It might therefore be of interest to study the SN in the aging Prader‐Willi population.

Altered expression of TrkB isoforms is further evidence for a decrease in neurotrophic support in the PD SN. TrkB‐T1, the truncated isoform of the TrkB receptor, functions as a dominant‐negative receptor variant. Interestingly, haploinsufficiency for TrkB and TrkC results in cell loss and accumulation of SNCA in the SN of aged mice (8), and increased levels of TrkB‐T1 expression are known to be associated with neuronal atrophy and death 23, 24). We observed a shift in the ratio TrkB‐T1/TrkB toward more TrkB‐T1 (Figure 2).

Adding to alterations in neurotrophic support, we also observed transcriptional evidence for changes in the endocytosis of activated receptors. AMPH, DNM3, NECAP‐1 and AAK1 are all downregulated, and serve important roles in regulating endocytosis either by interacting with AP‐2 (AAK1 and NECAP‐1) 65, 66), by coupling the PSD to the endocytic zone (DNM3) (48), or by sensing and inducing membrane curvature (AMPH) (31). Interestingly, HIP1R is known to prolong the half‐life of growth factor receptors after endocytosis (35), and is increased in expression. STS‐1, a gene that negatively regulates the endocytosis of receptor tyrosine kinases, is downregulated in PD (64). The regulation of STS‐1 and HIP1R might thus represent a coping mechanism of neurons suffering from reduced neurotrophic support in this disorder. Another transcriptional alteration potentially affecting neurotrophic signaling is the severe decrease of DYNC1I1, a component of the dynein motor complex responsible for retrograde transport of activated Trk receptors (6). Altered expression levels of several gene groups involved in microtubule‐based transport may point to a more general impairment of axoplasmic transport. These simultaneous alterations in transcripts encoding for proteins involved in neurotrophic support and signaling all suggest decreased levels of neurotrophic signaling, possibly directly underlying DAergic neuronal degeneration and death in PD.

Alterations in the expression of genes encoding repulsive axon guidance and outgrowth cues in the PD SN

We observed transcriptional evidence for the involvement of axon guidance molecules in SN PD neuropathology. The role of axon guidance cues in the development of the nervous system has been long appreciated, but recently, they are also believed to be functionally relevant in the adult nervous system. Axon guidance cues might play a role in synaptic plasticity and neurodegeneration 22, 49). Increased expression of the chemorepulsive protein Sema3A or the neurite outgrowth inhibitor Nogo‐A promotes denervation in a model for amyotrophic lateral sclerosis 21, 36), and Sema3A is increased in the cerebellum in schizophrenia (26). In the PD SN, we observed a twofold increase in RGMA expression. RGMA is a membrane‐bound glycoprotein implicated in axon guidance and neural tube closure during development (57). After human brain injury, RGMA expression is enhanced (68). We also observed changes in the expression of SDC2, a member of the heparan sulphate proteoglycan (HSPG) family of extracellular matrix proteins. Some axon guidance molecules bind to HSPGs, and this modulates their repulsive properties. When the axon guidance protein Sema5A binds to the HSPG SDC3, it acts as an axon attractant, and when Sema5A is bound to chondroitin sulphate proteoglycans, it converts to a chemorepulsive protein (37). Thus, the decreased expression of SDC2 might suggest a shift to a more repulsive character of the extracellular matrix. This effect is possibly aggravated by the upregulation of Sema5A in a subset of patients (Supplementary Table S4). An association between Sema5A polymorphisms and PD has been reported, although studies trying to replicate this in different populations were not successful 7, 16, 50). The 68% decrease of ROBO2 expression also suggests a change in axon guidance cue signaling in PD. ROBO2 is the receptor for Slit2, which regulates the outgrowth of developing midbrain DAergic neurons via the ROBO receptors (47). Also noteworthy are the reductions in SLITRK5 and NETO2, which are thought to play a role in axon guidance and outgrowth.

Taken together, transcriptional alterations of multiple genes involved in axon guidance and neurite outgrowth suggest an altered and possibly more chemorepulsive environment around DAergic neurons. Indeed, a recent SNP analysis of axon guidance pathways also implicated these pathways in PD (46). Changes in chemorepulsive signaling might contribute to the loss of synaptic contacts between DAergic neurons, ultimately leading to DAergic cell loss.

Expression changes in the caudate nucleus and putamen

We did not observe large alterations in gene expression in the putamen and caudate nucleus, which may result from the very stringent Bonferroni multiple‐testing correction method applied (to eliminate potential false positives), and/or the DA replacement therapies that all PD patients received during the disease course. We detected several noteworthy changes in gene expression that are apparently not prevented by the DA replacement therapies. Remarkable is the increase of LTF expression in the PD caudate nucleus and SN. LTF is not only involved in iron homeostasis, but also has cell growth and differentiation properties (77). The neuromelanin in SN DAergic neurons contains high levels of iron. Together with the observation that surviving neurons might lose their neuromelanin, the upregulation of LTF is indicative for increased cytosolic and synaptic iron levels. Indeed, increased iron levels are observed in surviving neurons in the PD SN, and raise oxidative stress levels in SN DAergic neurons 58, 80). Furthermore, in our study, CYBRD1, mutated in some iron overload disorders (79), is also upregulated in the caudate nucleus. Thus, increased iron loads in the striatum and SN might contribute to PD‐associated neurodegeneration.

Complex interactions between multiple insults underlie SN DAergic neurons degeneration

Our microarray data point to the impairment in neurotrophic signaling pathways and retrograde transport in PD pathogenesis, and to the involvement of molecules regulating axon guidance and neurite outgrowth. These changes might interact with deleterious processes already recognized to be involved in the etiology of sporadic PD, including mitochondrial dysfunction, reduction of synaptic transmission and impairment of the UPS. Importantly, genes encoding for proteins involved in these processes were also found to be affected in our study. In a recent review, a multiple hit hypothesis for SN DAergic degeneration was proposed, in which neuronal death only occurs after several cellular processes have been compromised (73).

We hypothesize that complex interactions between altered processes such as loss of neurotrophic support and mitochondrial dysfunction together constitute the unique conditions in which SN DAergic neurons are prone to degenerate in PD. It is important to note that our study does not provide insight in the exact sequence of events ultimately leading to SN DA neuronal demise in PD. We propose the following model. Impairment of several neurotrophic support pathways reduces the viability of DAergic neurons. In conjuction, alterations in expression of axon guidance cues exert a negative influence on the synapses of SN DAergic neurons. As synaptic transmission is dependent on viable synaptic contacts, changes in neurotrophic support and axon guidance cue signaling negatively affect DA transmission to the striatum. The reduction of neurotrophic support is aggravated by the disturbance of microtubule‐based transport, which impairs the retrograde transport of activated Trk receptors necessary for neurotrophic signaling. In parallel, mitochondrial dysfunction leads to lower levels of available ATP. As SN DAergic neurons are metabolically more active than their VTA counterparts (29), the greater energy demands might make SN DAergic neurons more vulnerable to metabolic insults. Furthermore, as UPS‐mediated protein degradation is ATP‐dependent, a reduction in oxidative phosphorylation likely negatively affects the cell's capacity to break down proteins, with proteinacious inclusions as a result. The observation that some surviving neurons lose their neuromelanin content, possibly leads to increased iron contents in the SN, and, via the nigrostriatal pathway, in the striatum. This increased iron load raises oxidative stress levels. Finally, the severe reduction of VMAT2 levels causes insufficient packaging of DA in DA vesicles, effectively increasing cytosolic DA concentrations, which also induces oxidative stress. Taken together, this model describes a complex interaction between seemingly unrelated, but most likely intimately related intra‐ and extracellular events, which together lead to selective DAergic vulnerability in PD.

The purpose of this study was to identify genes and gene classes that may be functionally linked to PD neurodegeneration. We have not as yet demonstrated the biological significance of the presented alterations: our observations are predictive of the role of specific molecules and pathways in the process of PD neurodegeneration. Therefore, the challenge for the future lies in the translation of the current findings to functional studies in cellular and animal models of PD.

FUNDING

Royal Dutch Academy of Sciences, Innovation Fund; Solvay Pharmaceuticals.

Supporting information

Table S1. Hybridization setup microarray experiments.

Table S2. qPCR primer sequences.

Table S3. Significantly altered genes in the PD SN.

Table S4. qPCR validation of changed transcripts in the PD SN.

This material is available as part of the online article from: http://www.blackwellsynergy.com

Please note: Blackwell Publishing is not responsible for the content or functionality of any supplementary materials supplied by the authors. Any queries (other than missing material) should be directed to the corresponding author for the article.

Supporting info item

Supporting info item

Supporting info item

Supporting info item