View metadata, citation and similar papers at core.ac.uk brought to you by CORE

provided by Elsevier - Publisher Connector

Available online at www.sciencedirect.com

Genomics 90 (2007) 647–660 www.elsevier.com/locate/ygeno

Gene expression profiling in the adult Down syndrome brain ⁎ H.E. Lockstone a,1, L.W. Harris a,1, J.E. Swatton b, M.T. Wayland a, A.J. Holland c, S. Bahn a,

a Institute of Biotechnology, University of Cambridge, Cambridge CB2 1QT, UK b Department of Biochemistry, University of Cambridge, Cambridge CB2 1QW, UK c Department of Psychiatry, University of Cambridge, Cambridge CB2 2QQ, UK Received 10 May 2007; accepted 16 August 2007 Available online 22 October 2007

Abstract

The mechanisms by which trisomy 21 leads to the characteristic Down syndrome (DS) phenotype are unclear. We used whole genome microarrays to characterize for the first time the transcriptome of human adult brain tissue (dorsolateral prefrontal cortex) from seven DS subjects and eight controls. These data were coanalyzed with a publicly available dataset from fetal DS tissue and functional profiling was performed to identify the biological processes central to DS and those that may be related to late onset pathologies, particularly Alzheimer disease neuropathology. A total of 685 probe sets were differentially expressed between adult DS and control brains at a stringent significance threshold (adjusted p value (q) b 0.005), 70% of these being up-regulated in DS. Over 25% of on 21 were differentially expressed in comparison to a median of 4.4% for all . The unique profile of up-regulation on , consistent with primary dosage effects, was accompanied by widespread transcriptional disruption. The critical Alzheimer disease , APP, located on chromosome 21, was not found to be up-regulated in adult brain by microarray or QPCR analysis. However, numerous other genes functionally linked to APP processing were dysregulated. Functional profiling of genes dysregulated in both fetal and adult datasets identified categories including development (notably Notch signaling and Dlx family genes), lipid transport, and cellular proliferation. In the adult brain these processes were concomitant with cytoskeletal regulation and vesicle trafficking categories, and increased immune response and oxidative stress response, which are likely linked to the development of Alzheimer pathology in individuals with DS. © 2007 Elsevier Inc. All rights reserved.

Keywords: profiling; Oligonucleotide array sequence analysis; Down syndrome; Trisomy 21; Brain; Alzheimer disease

Down syndrome (DS) is a common genetic disorder affecting hypotheses are currently proposed to explain the mechanism by around 1 in 800 live births in the human population. It is which trisomy 21 leads to the DS phenotype [5]. The develop- caused by a complete, or occasionally partial, triplication of mental instability theory suggests that dosage imbalance on chromosome 21 resulting in a complex and variable phenotype. chromosome 21 as a whole disrupts diverse developmental The disorder is primarily characterized by cognitive and pathways [6], while the gene-dosage hypothesis suggests that language dysfunction coupled with sensory and neuromotor dosage increase for specific genes on chromosome 21 con- deficits and a neuropathology primarily characterized by de- tributes more directly to different aspects of the disease pheno- creased brain size and weight plus abnormal gyrification and type [7]. neurogenesis [1,2]. In addition, DS individuals are likely to Global gene expression profiling techniques are well suited suffer from a broad range of symptoms outside of the nervous to the investigation of the effects of trisomy 21 because the system, including abnormal craniofacial development, con- expression levels of trisomic genes as well as those in the genital heart problems [3], and immune defects [4]. Two main wider genome can be assessed. Recently, gene expression profiling of brain and other tissues from human subjects with trisomy 21 and from mice trisomic for homologous regions has ⁎ Corresponding author. Fax: +44 1223 334162. been performed to characterize the DS transcriptome. The E-mail address: [email protected] (S. Bahn). majority of studies use mouse models for trisomy 21, which 1 These authors contributed equally to this work. have partial triplications and display some characteristic

0888-7543/$ - see front matter © 2007 Elsevier Inc. All rights reserved. doi:10.1016/j.ygeno.2007.08.005 648 H.E. Lockstone et al. / Genomics 90 (2007) 647–660 features of DS [8–14]. A small number of gene expression Results studies using human DS tissues have been conducted, profiling fibroblasts [15], whole blood [16], T cells [17], amniocytes Identification of differentially expressed genes [18], and placenta [19], but most have lacked statistical power. Only one group, using fetal samples, has previously investi- Analysis of gene expression in the adult DS dorsolateral gated global gene expression patterns in human DS brain prefrontal cortex was performed using seven DS samples and [20,21]. eight control samples that passed our stringent quality control These studies have established a primary gene dosage effect procedures. Using a Bayesian-moderated t test and Benjamini– for chromosome 21 but it remains unclear whether there are Hochberg correction to control the false discovery rate (see pervasive secondary transcriptional effects throughout the ge- Materials and methods), over 5000 probe sets were found sig- nome. The proportion of trisomic genes reported to show nificantly altered (adjusted p value (q) b 0.05) in the DS subjects increased expression levels and the degree of up-regulation also compared to controls. Due to this large number of altered probe vary between studies. The regulation and expression of chromo- sets, we compared the results using three different significance some 21 genes is likely to be both dynamic and complex levels (0.05, 0.01, and 0.005) to classify differentially expressed [22–24], and thus it is important that further global gene probe sets, which revealed several interesting trends. First, the expression studies with high statistical power are performed to proportion of up-regulated probe sets increased with decreasing characterize the DS transcriptome fully. q value (Fig. 1a); second, the proportion of up-regulated probe In this study, we chose to investigate the transcriptome of sets on chromosome 21 increased with decreasing q value, human adult DS brain. DS is classically characterized as a reaching 98% at q b 0.005 (Fig. 1a); third, probe sets localized to developmental disorder, and for this reason the majority of chromosome 21 were overrepresented among the most sig- molecular studies have been performed on fetal tissue. nificant genes and this trend was unique to chromosome 21 However, another notable feature of DS is precocious aging, (Fig. 1b). As overexpression of chromosome 21 genes is the and numerous brain pathologies develop specifically in later expected consequence of trisomy 21, these results demonstrate life, including calcification of the basal ganglia [25],an clear disease effects at the lower significance thresholds and increased risk of psychiatric disease [26], and the inevitable indicate sufficient statistical power with our sample size. development of Alzheimer disease (AD) pathology, character- A significance threshold of q b 0.005 was used to filter the ized by neuronal loss and the presence of amyloid plaques and list of differentially expressed probe sets for further analysis; neurofibrillary tangles. Overexpression of the amyloid pre- 685 probe sets were significantly altered, with 473 (69%) up- cursor (APP)geneonchromosome21hasbeen regulated in DS compared to controls and 212 (31%) down- proposed as the central event leading to AD pathology in DS regulated. These probe sets represented 400 up-regulated and [1,2], in keeping with the amyloid cascade hypothesis [27,28], 163 down-regulated genes. The bias toward up-regulation at but evidence suggests that triplication of this gene alone is not this significance threshold is only partially explained by up- sufficient for the development of AD pathology [29]. regulation on chromosome 21 and may therefore be a feature of Investigation of the adult DS brain is therefore of interest not DS. only in the context of Down's syndrome itself, but also to further our understanding of the mechanisms of later onset Effects of demographic variables on gene expression pathologies such as that of AD. The study of the disorders in this context presents a particular challenge, however, as inter- The effect of demographic variables on differential gene pretation of results may be hampered by the inability to dis- expression was assessed to determine whether factors other than tinguish the fundamental effects of DS from the effects of AD disease status could explain the results. This was particularly pathology per se. important in this study as the increased power gained from the We used microarrays to investigate the effect of trisomy on use of six additional control samples was offset by the groups the expression of genes on chromosome 21 and the wider being less well matched for known demographic variables transcriptome in the adult DS brain, using tissue from the (Supplementary Table 1). The mean age of subjects in the dorsolateral prefrontal cortex. In addition, we coanalyzed and disease group was 58.6 ± 9.4 years and that of the control group compared a publicly available microarray dataset from fetal was 47.8 ± 10.8 years (p = 0.06, Student's t test). For the four DS DS brain [20], to identify (1) robust transcriptional alterations, patients with data available for postmortem interval (PMI), mean which are detected in both fetal and adult brain and are thus PMI was 14 ± 8 h compared to 26 ± 11 h for the eight controls likely to be present from an early stage throughout the (p = 0.07, Student's t test). Gender was approximately evenly lifetime of DS individuals, and (2) those alterations that are split in the DS group (four females and three males) but not in the associated with the development of late onset pathologies controls (one female, seven males). For these variables to be such as AD in the adult. Functional profiling of the set of confounding factors in this study, they should not only vary genes altered in the DS brain using annotation data from the systematically between control and disease groups, but should Consortium [30] was conducted to investigate also show some correlation with the expression of one or more of the biological processes that are affected in DS and help the differentially expressed genes. For the continuous variables interpret how changes in gene expression can contribute to the age and PMI, Pearson's correlation test was used to measure phenotype. correlation with gene expression, and Benjamini–Hochberg H.E. Lockstone et al. / Genomics 90 (2007) 647–660 649

Fig. 1. Characteristics of the Down syndrome (DS) dataset at different significance thresholds. (a) The percentage of significantly altered probe sets that were up- and down-regulated at three significance thresholds for all chromosomes and for chromosome 21. An overall bias toward up-regulation is revealed at the lower thresholds when all significant probe sets are considered. For chromosome 21 probe sets, a strong bias toward up-regulation is observed at all thresholds, but is most pronounced at q b 0.005, at which 98% were up-regulated in DS. (b) The proportion of significant probe sets localized to each chromosome at three significance thresholds and in the genome as a whole. Chromosome 21 is highlighted in red and shows a unique trend of comprising a greater proportion of the significant probe sets as the significance threshold decreases. For all other chromosomes, representation at each significance threshold is relatively consistent and comparable to the percentage of all human genes localized to them. adjusted p values were computed. There were no significant key question of this study is how many trisomic genes show results for either age or PMI (smallest q values were 0.10 and increased expression at the mRNA level in adult DS subjects. 0.24, respectively). For the categorical variable of gender, the There were 46 significant probe sets targeting genes localized to LIMMA package from Bioconductor was used to fit a linear chromosome 21, and 45 of these were up-regulated in DS. There model to detect differential gene expression between male and are often multiple probe sets for a given gene on Affymetrix female samples. Again, no significant differences were found microarrays and in a few instances 2 or more significant probe following Benjamini–Hochberg correction for multiple testing sets were found for the same gene (Table 1). Redundant probe (smallest q = 0.63). These results indicate that these demo- sets can distort the calculation of the percentage of genes on a graphic factors did not have a significant influence on the chromosome that is differentially expressed and therefore only observed gene expression changes. the most significant probe set for each gene was retained. After redundant probe sets were excluded there were 39 up-regulated Gene expression on chromosome 21 genes and 1 down-regulated gene on chromosome 21 (q b 0.005, Table 1) and hence 27% (39/145) of trisomic genes investigated Chromosome 21 is currently estimated to contain 352 genes in this study were found to be overexpressed. To see how this (NCBI build 36), about 50 of which form two compared to other chromosomes, the percentage of differentially large clusters of keratin-associated that are not expressed genes was similarly calculated for all other chromo- represented on the Affymetrix HG-U133A chip due to high somes. It ranged from 3.0% (chromosomes 16 and Y) to 6.6% sequence similarity. A further 53 predicted genes with the prefix (chromosome 8) and the median was 4.3% (Fig. 2). Thus “LOC” are also not represented on the chip. After filtering for chromosome 21, with 27% of genes showing significant changes Class A probe sets (see Materials and methods) there are 238 in expression, had by far the highest proportion of significantly probe sets on the HG-U133A chip that target 145 unique and altered genes. All but 1 of the significant genes on chromosome generally well-characterized chromosome 21 genes. The first 21 were up-regulated and this pattern was not seen on other 650 H.E. Lockstone et al. / Genomics 90 (2007) 647–660

Table 1 Chromosome 21 genes significantly altered in adult DS brain and corresponding results in the fetal brain dataset Gene name Gene Adult DS brain Fetal DS brain Original fetal analysis symbol gene ID Significant Fold Adjusted Fold Adjusted Fold Adjusted probe sets/ change p value change p value change p value total probe (b0.005) (b0.3) (cerebrum) (b0.05) sets on chip Cystatin B (stefin B) CSTB 1476 1/1 2.15 0.0003 1.61 0.10 1.69 7.8×10−5 Transmembrane protein 50B TMEM50B 757 1/1 2.53 0.0003 1.50 0.28 Down syndrome critical region gene 3 DSCR3 10311 1/2 1.61 0.0003 1.14 0.29 1.82 1.4×10−5 Pituitary tumor-transforming 1 interacting protein PTTG1IP 754 1/1 2.79 0.0006 1.75 0.06 1.52 1.5×10−7 SON DNA binding protein SON 6651 4/4 1.75 0.0006 1.52 0.21 1.39 7.3×10−4 Transmembrane protein 1 TMEM1 7109 1/2 1.94 0.0011 1.36 0.20 1.22 5.3×10−4 Protein arginine methyltransferase 2 PRMT2 3275 2/4 1.69 0.0013 1.48 0.10 1.37 3.6×10−3 Junctional adhesion molecule 2 JAM2 58494 1/1 1.85 0.0016 1.39 0.28 Not significant Phosphatidylinositol glycan anchor biosynthesis, class P PIGP 51227 1/1 2.08 0.0018 1.69 0.12 Not significant BTB and CNC homology 1, basic leucine zipper BACH1 571 1/2 1.77 0.0018 1.16 0.28 Not significant transcription factor 1 ATP-binding cassette, subfamily G (WHITE), member 1 ABCG1 9619 1/2 2.67 0.0018 1.10 0.29 Not significant BTG family, member 3 BTG3 10950 2/3 2.48 0.0018 1.69 0.28 Not significant Chromosome 21 open reading frame 25 C21orf25 25966 1/1 1.78 0.0018 Not significant Not significant SH3 domain binding glutamic acid-rich protein SH3BGR 6450 1/1 1.62 0.0019 Not significant 2.17 7.1×10−7 Ubiquitin-specific peptidase 16 USP16 10600 1/1 2.12 0.0019 1.81 0.06 1.68 5.3×10−4 Integrin, β2 (antigen CD18 (p95), lymphocyte ITGB2 3689 1/1 2.93 0.0020 Not significant Not significant function-associated antigen 1; macrophage antigen 1 (mac-1) β subunit) High-mobility group nucleosome binding domain 1 HMGN1 3150 2/2 1.62 0.0021 1.40 0.28 Not significant U2 (RNU2) small nuclear RNA auxiliary factor 1 U2AF1 7307 1/1 1.26 0.0023 1.29 0.29 Not significant Regulator of calcineurin 1 RCAN1 1827 1/3 1.70 0.0023 Not significant Zinc finger protein 294 ZNF294 26046 1/1 1.80 0.0023 1.82 0.06 1.71 3.4×10−5 T cell lymphoma invasion and metastasis 1 TIAM1 7074 1/2 1.55 0.0023 1.90 0.07 Not significant Dual-specificity tyrosine-(Y)-phosphorylation DYRK1A 1859 1/3 1.62 0.0024 1.61 0.15 Not significant regulated kinase 1A V-ets erythroblastosis virus E26 -like (avian) ERG 2078 1/2 1.50 0.0024 Not significant Not significant ADAM metallopeptidase with thrombospondin ADAMTS1 9510 1/1 3.70 0.0024 Not significant Not significant type 1 motif, 1 MORC family CW-type zinc finger 3 MORC3 23515 1/1 1.61 0.0025 1.52 0.28 Not significant SAM domain, SH3 domain, and nuclear SAMSN1 64092 1/1 1.67 0.0027 Not significant Not significant localization signals, 1 Cystathionine-β-synthase CBS 875 1/1 1.78 0.0027 1.40 0.28 Not significant Chromosome 21 open reading frame 59 C21orf59 56683 1/1 1.47 0.0027 1.45 0.10 Not significant Ribosomal RNA processing 1 homolog B (S. cerevisiae) RRP1B 23076 1/2 1.67 0.0028 2.00 0.28 Not significant Chromosome 21 open reading frame 66 C21orf66 94104 1/2 1.79 0.0029 1.51 0.28 Not significant MCM3 minichromosome maintenance-deficient 3 MCM3AP 8888 1/3 1.41 0.0032 1.74 0.10 Not significant (Saccharomyces cerevisiae)-associated protein Mitochondrial ribosomal protein L39 MRPL39 54148 1/1 1.67 0.0032 1.35 0.26 1.44 7.5×10−4 Crystallin, ζ (quinone reductase)-like 1 CRYZL1 9946 1/1 1.66 0.0036 1.55 0.28 Not significant S100 calcium binding protein, β (neural) S100B 6285 1/1 2.28 0.0037 Not significant Not significant Leucine-rich repeat-containing 3 LRRC3 81543 1/1 −1.39 0.0038 −1.11 0.29 Not significant Phosphofructokinase, liver PFKL 5211 1/2 1.32 0.0038 1.25 0.28 1.16 1.9×10−4 Intersectin 1 (SH3 domain protein) ITSN1 6453 1/6 2.20 0.0038 1.46 0.28 Not significant Neural cell adhesion molecule 2 NCAM2 4685 1/2 1.47 0.0038 Not significant Not significant Collagen, type VI, α 2 COL6A2 1292 1/2 2.56 0.0040 Not significant Not significant Carbonyl reductase 1 CBR1 873 1/1 1.97 0.0047 Not significant Not significant Chromosome 21 genes significantly altered in adult DS brain are listed (qb0.005) together with the corresponding result for each gene in our reanalysis of the fetal brain dataset and in the original analysis of Mao et al. [21]. chromosomes (data not shown), strongly implying that the Distribution of differentially expressed genes on changes are a direct consequence of the triplication of these chromosome 21 genes in DS. Significantly up-regulated genes on chromosome 21 generally reflected the 1.5-fold increase in expression pre- To assess how the 40 significant genes localized to chromo- dicted by the gene-dosage hypothesis (Fig. 3). However, the some 21 were distributed, histogram plots showing the density median increase in expression was slightly higher at 1.76-fold distributions of (i) all chromosome 21 genes, (ii) those re- and 6 genes had fold changes in excess of 2.5. presented on the HG-U133A chip, and (iii) those that were H.E. Lockstone et al. / Genomics 90 (2007) 647–660 651

Fig. 2. The percentage of genes on each chromosome that were significantly altered in DS at the significance threshold q b 0.005. Over a quarter of the genes localized to chromosome 21 were significantly altered compared to a median of 4.4% for all chromosomes. These data support a primary dosage effect on chromosome 21 accompanied by widespread transcriptional disruption in adults with DS. significantly altered were generated. The genes represented on chromosome 21 can be drawn, but the QPCR result suggests that the HG-U133A chip show a density distribution very similar to it may be an unlikely phenomenon. that of all chromosome 21 genes (Supplementary Fig. 1), Three genes not located on chromosome 21 that were indicating no inherent bias in the locations of the subset of genes significantly up-regulated in the microarray study were also represented on the chip. The density distribution of significant investigated by QPCR, two of which were confirmed to be up- chromosome 21 genes was then compared to that of the genes regulated in DS (APOE and NOTCH2; Fig. 5). The third gene, represented on the chip. This revealed that the significantly APPBP2, failed to validate using QPCR, showing no difference altered genes were quite evenly distributed along the chromo- in expression between DS and control groups (Fig. 5); because some, relative to the underlying distribution (Fig. 4). These data QPCR probes were chosen to target similar transcripts to the suggest that primary gene dosage effects are found throughout microarray probe-sets, it is unlikely that the failure to confirm chromosome 21 and do not cluster at particular locations, in up-regulation of APPBP2 was due to differential expression of agreement with Mao et al. [21]. alternative transcripts of this gene and thus may indicate that the microarray finding was a false positive. Quantitative real-time PCR (QPCR) analysis Full details of the fold changes and associated p values for microarray data and QPCR data are given in Supplementary QPCR was performed to validate the altered expression levels Table 2. Fold changes were reasonably consistent between of 11 selected genes, using cDNA derived from total RNA of the microarray and QPCR data, although in the case of PTTG1IP a brain tissue samples. These included 6 chromosome 21 genes 2.8-fold increase in expression in the DS group was found by that showed significant up-regulation by microarray (q b 0.005); microarray compared to a 6.4-fold increase measured with 3 were confirmed with QPCR (ADAMTS1, BACE2, PTTG1IP; QPCR. Increases exceeding 2-fold for several chromosome 21 p b 0.05), while a trend toward up-regulation was observed for genes (ADARB1, ADAMTS1, BACE2, and PTTG1IP) were S100B (p = 0.07), RCAN1 (p = 0.13), and DYRK1A (p = 0.17) found by QPCR, confirming that some genes are up-regulated to (Fig. 5). A surprising result from the microarray study was that a greater extent than the predicted 1.5-fold. expression of APP, thought to be critical for the development of Alzheimer pathology in DS, was not significantly increased, Comparison with proteomics data although one probe set showed a trend for up-regulation (q = 0.06). APP expression was therefore investigated using As further validation and indication that at least some changes QPCR, which confirmed the negative microarray result (Fig. 5). at the transcript level affect protein expression levels, a number The final chromosome 21 gene investigated by QPCR was of genes identified as differentially expressed by microarray ADARB1, which the microarray analysis indicated as down- were also found to be dysregulated at the protein level in a regulated, albeit at the less stringent 0.05 significance level. In parallel study. A previous study from this laboratory [31] used all, 8 chromosome 21 genes were decreased at this significance two-dimensional fluorescence difference in-gel electrophoresis level in the adult brain as well as showing down-regulation in the (2D-DIGE) to investigate the protein profile of brain tissue from fetal brain dataset (see below), and we therefore wanted to the same DS individuals. 2D-DIGE is capable of resolving confirm if trisomic genes could be down-regulated in DS. In fact, around 2500 proteins, mainly abundant soluble proteins. This ADARB1 showed a significant up-regulation by QPCR (Fig. 5). study revealed 33 differentially expressed proteins between DS Further investigation of other genes is required before firm and controls, all of which showed increased expression in DS. A conclusions regarding the down-regulation of trisomic genes on large proportion of these were identified as being functionally 652 H.E. Lockstone et al. / Genomics 90 (2007) 647–660

involved with cytoskeletal organization and vesicle trafficking. Of these 33 proteins, 9 had at least one significantly up-regulated transcript at q b 0.005 and a further 6 at q b 0.05 (Supplementary Table 3). In the case of dynamin, it was not possible to distinguish the different isoforms at the protein level, but the transcript data suggest that DNM1 is decreased and DNM2 is increased. Only one protein, CLTC, was found to be down- regulated at the transcript level (q = 0.03) but was previously found up-regulated at the protein level.

Comparison with fetal DS brain

A comparison of expression data from adult and fetal brain tissue was performed to explore further the gene expression changes observed in DS. A publicly available dataset for the human fetal DS brain (cerebrum) was reanalyzed using the same statistical approach employed for the adult brain, allowing a direct comparison of the two datasets (see Materials and methods). In our reanalysis, only one probe set was significant at q b 0.05, but as the top 22 probe sets represented chromosome 21 genes and all were up-regulated, we considered it reasonable to relax the significance threshold. At q b 0.25, 41 probe sets were significant, rising dramatically to 3725 at q b 0.3. The relatively high q values may be due to the smaller sample size in the fetal dataset (four DS and seven control subjects) reducing statistical power. At q b 0.3, the reanalysis identified 25 of the 26 chromosome 21 probe sets reported by Mao et al. [21] (18 at q b 0.25), in addition to many others that did not pass the significance criteria in the original analysis. After removing redundant probe sets, we compared chromosome 21 genes that were significant in the adult DS brain at q b 0.005 to those significant in the fetal DS brain at q b 0.3. Twenty-eight genes were up-regulated in both the adult and the fetal DS brain and represent independent findings in human tissue (Table 1). Only 11 of these were reported in the original analysis by Mao et al. [21].

Functional profiling

Functional profiling of the 685 probe sets differentially expressed in DS (q b 0.005) using Onto-Express revealed 31 Gene Ontology (GO) categories significantly overrepresented in the list of up-regulated genes and 14 GO categories overrepresented among the down-regulated genes (Table 2). These individual categories could be further classified into several major groups: among up-regulated genes, enriched processes included immune system function, development, cell cycle/proliferation, cytoskeleton and vesicle trafficking, cell signaling, and regulation of gene expression as well as lipid transport and response to oxidative stress. Other cell signaling Fig. 3. Fold changes for the 46 probe sets significantly altered in DS (q b 0.005) categories, synaptic transmission, muscle-related, and develop- that map to chromosome 21. The 46 probe sets represent 40 unique genes; 2 mental processes were enriched among down-regulated genes. significant probe sets detecting BTG3, HMGN1, and PRMT2 were identified, as The presence of AD pathology in the adult DS brain is usually well as all 4 probe sets targeting SON. The theoretical 1.5-fold increase in considered a confounding effect; however, by focusing solely on expression is indicated by the dashed line and some genes show substantial deviation from a simple dosage increase. Only one gene localized to fetal tissue the pathogenesis of later-developing features such as chromosome 21, LRRC3, showed decreased expression (fold change b1) in AD pathology cannot be assessed. By comparing the adult and DS patients. fetal datasets, we aimed to distinguish those processes that are H.E. Lockstone et al. / Genomics 90 (2007) 647–660 653

Fig. 4. Distribution of significant genes across chromosome 21. Density distributions for the 145 chromosome 21 genes represented on the HG-U133A chip and for the 40 that were found significantly altered in DS are shown. The histograms are divided into 1-Mb bins. most likely to be central to the pathogenetic processes in the contribution to the phenotype than widespread transcriptional brain from those that may arise as a result of late onset pathology, disruption [7]. particularly AD. We therefore also functionally profiled the Using a variety of methods, including microarrays, QPCR, group of 179 probe sets that were altered in both adult (q b 0.005) and SAGE, previous studies have investigated the transcrip- and fetal (q b 0.3) DS brain. This approach should identify tional consequences of trisomy 21 in mouse models and human ongoing processes central to the DS phenotype and be tissues, establishing a primary gene dosage effect on chromo- particularly robust as the individual genes have been identified some 21 [8,10,18,20,21]. The issue of whether trisomy 21 has in two independent studies. Twelve significantly overrepre- secondary effects on transcription throughout the genome sented categories were identified, 11 of which were also found remains unresolved as conflicting data have been reported by a when the adult changes were profiled separately (Table 3). Some number of different studies [8–14,18,20,21]. Fundamental functional categories found dysregulated in the adult brain were differences between studies, including organism, tissue, and not found in the overlap between the fetal and the adult developmental age studied, as well as technology and statistical dysregulated genes, notably immune response, response to techniques used, have complicated the interpretation of these oxidative stress, and the majority of cytoskeletal/vesicle-related findings. Grouping studies according to whether they found categories. These may represent processes related to the late pervasive secondary transcriptional changes does not reveal any onset pathologies that occur in DS. common factor, such as developmental age, or combination of Complete lists of the genes contributing to the significant GO factors that might explain the contradictory findings. categories can be found in Supplementary Tables 4 and 5. The present study has shown up-regulation of a subset of chromosome 21 genes in human adult DS brain, accompanied Discussion by numerous transcriptional changes throughout the genome, lending support to the theory that secondary transcriptional It has been known for nearly half a century that DS is changes occur as a result of trisomy 21. In contrast, the original caused by a chromosomal abnormality, namely the trisomy of study of human fetal DS brain tissue concluded that few tran- chromosome 21. Despite intensive research, the mechanisms scriptional changes occur outside chromosome 21 [21]. by which the extra copy of chromosome 21 leads to the DS However, our reanalysis of this dataset suggests that transcrip- phenotype remain unclear. Only trisomies of human chromo- tional changes in the fetal brain may be more widespread than somes 21, 18, and 13 are compatible with postnatal survival. previously thought. Nonetheless, the wider effects observed Chromosome 21 has the fewest number of genes of all human throughout the genome in the adult brain could be the autosomes and individuals with DS typically survive into late cumulative effect of the many defects and pathologies middle age, whereas individuals with trisomy 13 or 18 rarely associated with DS. On chromosome 21, the gene expression survive more than a year. In DS the dosage imbalance for changes described here in the adult DS brain and by Mao et al. chromosome 21 genes is thought to affect normal interactions, [21] in the fetal DS brain are in broad agreement: a subset of for example, by altering the correct stoichiometry of protein trisomic genes, distributed throughout chromosome 21, shows complexes; the downstream consequences may be widespread, increased expression relative to normal controls. Both studies particularly in the case of transcription factor complexes [32]. reveal a relatively small fraction of trisomic genes up-regulated However, others postulate that the overexpression of individual in human DS subjects, which is surprising given that the genes on the trisomic chromosome makes a more important majority are known to be expressed in brain [33,34]. This is, 654 H.E. Lockstone et al. / Genomics 90 (2007) 647–660 however, consistent with other studies using human tissues the present study is likely to be underestimated. For example, of [15,16,18]. the 39 chromosome 21 genes significantly overexpressed in DS Trisomic genes that are not overexpressed may be subject to adult brain and the 23 genes in the fetal brain [21], only 11 were dosage compensation; for example, the dosage effect could be reported increased by both studies (although a greater degree of canceled if a transcriptional repressor were also at dosage overlap was seen when the adult data were compared to the fetal imbalance [35]. Such a mechanism has previously been data reanalyzed in the same manner as the adult-derived proposed in the mouse [11]. However, it seems unlikely that dataset). This inconsistency may reflect the biological conse- such a high proportion of chromosome 21 genes would show quence of temporal patterns of gene expression and/or dosage compensation and there are several reasons the number regulatory mechanisms but it is possible that genes identified of trisomic genes showing up-regulation at the mRNA level in in just one dataset represent false negatives in the other study H.E. Lockstone et al. / Genomics 90 (2007) 647–660 655 and may reflect limits in the sensitivity and/or dynamic range of samples by QPCR but is nonetheless surprising. A number of microarrays. Taken together, the two studies may suggest that a genes functionally linked to APP were found dysregulated at greater proportion of trisomic genes are overexpressed than in q b 0.005 in adult DS brain, including the chromosome 21 genes either study alone. It is also of note that a further 23 BACE2 and S100B (confirmed by QPCR) and apolipoproteins chromosome 21 genes were up-regulated in our dataset at APOE and CLU, plus PSEN1, PSEN2, and MAPT dysregulated q b 0.05, but did not meet the more stringent threshold used in at q b 0.05 by microarray. The combination of these changes this study. A more specific analysis of chromosome 21 genes in suggests disturbances in the normal processing and function of human DS brain tissue would be of interest. APP; however, the data also suggest that this may be a more In theory, the presence of an additional copy of chromosome complex process than simple up-regulation of the mRNA. A 21 would lead to a dosage-dependent 1.5-fold increase in the number of the processes revealed by functional profiling (see expression of all triplicated genes. Among some previous below) suggest further candidates for the predisposition of the studies, a consistent 1.5-fold increase in the expression of DS brain to the buildup of amyloid β and development of AD trisomic genes has been reported [9,10,21], while others, in- pathology and symptoms. cluding a comprehensive QPCR study of the Ts65Dn mouse model, have found substantial deviations from this theoretical Functional profiling in adult and fetus value [12,18]. These results may be due to differences in the sensitivity of gene expression techniques, the individual genes Among the biological process categories overrepresented in considered, and the complex nature of gene regulation. the list of genes dysregulated in both fetal and adult datasets In the present study, both the microarray and the QPCR were developmental pathways, lipid transport, cellular prolif- analyses of the adult DS brain showed that the degree of up- eration, and transcriptional regulation. These processes are regulation of trisomic genes was sometimes greater than the likely to be important to the core DS phenotype and may predicted dosage effect of 1.5-fold. The previous study represent predisposing factors to late onset pathology. investigating human fetal brain, on the other hand, found Disruption of the lipid transport process may be important in genes consistently increased in expression by 1.5-fold [21]. Our the context of brain development due to the important role lipids, reanalysis of the fetal data, using the same statistical approach and especially cholesterol, play in synapse formation and applied to the adult data, confirmed that up-regulation of maintenance. This includes the regulation of functional proper- chromosome 21 genes in the fetal brain fits the model ties of ion channels and transmitter receptors and control of (1.54 ± 0.05), suggesting that the deviation seen in the adult synaptic vesicle formation and transport [41], which could con- brain is not influenced by the statistical analysis. It is possible tribute to abnormal cognitive development in DS individuals. that expression of trisomic genes fits closely with the dosage- In the adult brain, early dysfunction in lipid transport may act dependent model at the fetal stage, but that other changes over as a contributing factor to the buildup of amyloid β and the human life span impact on the expression profile. development of AD pathology. Of particular interest is the up- regulation of APOE (confirmed by QPCR), which binds choles- The role of triplicated APP in DS and AD pathogenesis terol in astrocytes and regulates lipid transport into neurons via the LDL and LRP1 receptors [42], both of which also showed An interesting feature of DS is the inevitable development of altered expression. The APOE4 allele is a well-characterized Alzheimer pathology by around age 40. APP is encoded on risk factor for AD, which appears to influence the accumulation chromosome 21 and is an obvious candidate for the pathogenesis of amyloid β [43]. It would be of interest to investigate any effect of AD in DS individuals. This is supported by findings such as a of APOE genotype status on the expression of APP in a larger duplication of the APP locus, which causes autosomal dominant cohort. APOE is located on chromosome 19 and hence early onset AD with cerebral amyloid angiopathy [36,37], and represents a clear example of a factor that may predispose to that a 78 year old individual with a partial trisomy 21 excluding AD, but develops secondarily to the direct effects of trisomy 21. the APP gene did not develop AD pathology [38]. However, Alterations in a large number of genes falling into the there have been both positive [39] and negative reports [2,40] of category of development are unsurprising, given the develop- overexpression of APP mRNA in DS, and other studies suggest mental nature of the disorder. Of particular interest is the down- that overexpression of APP alone is insufficient for the develop- regulation of members of the distal-less family of genes, DLX2 ment of AD pathology [29]. (in our analysis of both fetus and adult) and DLX6 (in adult Intriguingly, APP was not found to be significantly over- only). Dlx family genes are important for the development of expressed in our dataset nor in the original analysis of fetal brain forebrain GABAergic neurons [44], deficits in which have been tissue by Mao et al. [21]. This result was confirmed in our demonstrated in certain cortical layers of the DS brain [1].As

Fig. 5. Selected genes were investigated by QPCR analysis of RNA from adult brain tissue, including the chromosome 21 genes ADAMTS1, ADARB1, APP, BACE2, RCAN1, DYRK1A, PTTG1IP, and S100B, plus non-chromosome 21 genes APPBP2, APOE and NOTCH2. All genes were normalized to the endogenous control gene PGK1 (located on chromosome X). Five genes (ADAMTS1, BACE2, PTTG1IP, APOE, NOTCH2) were confirmed significantly up-regulated in Down syndrome (pb0.05, indicated by a double asterisk) and a further 3 genes (RCAN1, DYRK1A, S100B) showed a trend for up-regulation (p≤0.17, indicated by a single asterisk). APP and APPBP2 failed to show significant dysregulation by QPCR; in the case of APP the lack of significant change was consistent with the microarray result. The chromosome 21 gene ADARB1 showed significant up-regulation by QPCR but significant down-regulation by microarray analysis. Fold changes and p values obtained from both microarray and QPCR data for these 11 genes can be found in Supplementary Table 2. 656 H.E. Lockstone et al. / Genomics 90 (2007) 647–660

Table 2 Biological processes altered in adult Down syndrome brain GO ID GO biological process category No. of significant No. of genes Corrected genes on chip p value Overrepresented among up-regulated genes Immune system GO:0019884 Antigen presentation, exogenous antigen 4 12 0.001 GO:0019886 Antigen processing, exogenous antigen via MHC class II 4 13 0.001 GO:0019883 Antigen presentation, endogenous antigen 4 9 0.004 GO:0007229 Integrin-mediated signaling pathway 5 45 0.012 GO:0019885 Antigen processing, endogenous antigen via MHC class I 3 12 0.020 GO:0006955 Immune response 14 274 0.037 Developmental GO:0001709 Cell fate determination 3 6 0.002 GO:0016477 Cell migration 3 10 0.008 GO:0007219 Notch signaling pathway 3 19 0.021 GO:0016337 Cell–cell adhesion 3 30 0.035 GO:0009887 Organogenesis 4 55 0.035 Cytoskeletal/vesicle related GO:0006897 Endocytosis 8 65 0.002 GO:0006886 Intracellular protein transport 13 174 0.007 GO:0030036 Actin cytoskeleton organization and biogenesis 6 54 0.011 GO:0008360 Regulation of cell shape 3 18 0.013 GO:0051016 Barbed-end actin filament capping 3 21 0.019 GO:0016192 Vesicle-mediated transport 4 49 0.028 Cell cycle/proliferation GO:0007049 Cell cycle 16 211 0.004 GO:0006916 Anti-apoptosis 7 80 0.019 GO:0008285 Negative regulation of cell proliferation 9 127 0.023 GO:0001558 Regulation of cell growth 5 67 0.038 Cell signaling GO:0007173 Epidermal growth factor receptor signaling pathway 3 13 0.005 GO:0007169 Transmembrane receptor protein tyrosine kinase signaling pathway 7 65 0.009 Regulation of gene expression GO:0006338 Chromatin remodeling 3 20 0.017 GO:0045449 Regulation of transcription 5 63 0.034 GO:0006445 Regulation of translation 3 29 0.040 Other GO:0001503 Ossification 4 24 0.006 GO:0008015 Circulation 5 45 0.008 GO:0006869 Lipid transport 5 46 0.009 GO:0006730 One-carbon compound metabolism 4 29 0.009 GO:0006979 Response to oxidative stress 3 37 0.029

Overrepresented among down-regulated genes Synaptic processes GO:0007268 Synaptic transmission 8 161 0.003 GO:0006813 Potassium ion transport 6 124 0.006 GO:0006812 Cation transport 5 121 0.021 Cell signaling GO:0006470 Protein amino acid dephosphorylation 7 107 0.004 GO:0007264 Small GTPase-mediated signal transduction 6 155 0.020 Muscle related GO:0008016 Regulation of heart contraction rate 3 21 0.002 GO:0006936 Muscle contraction 5 69 0.004 GO:0007517 Muscle development 5 111 0.018 Other GO:0007010 Cytoskeleton organization and biogenesis 6 52 0.001 GO:0008544 Epidermis development 4 60 0.006 GO:0006968 Cellular defense response 4 65 0.011 GO:0007605 Perception of sound 3 78 0.044 GO:0007275 Development 9 369 0.044 GO:0007283 Spermatogenesis 3 98 0.048 Details of the number of significant genes falling into each category, the total number of genes in that category represented on the chip, and the significance level are given. abnormal function of interneurons in the prefrontal cortex has the NOTCH2 gene in DS was confirmed by QPCR. In been implicated in schizophrenia [45] and other psychiatric addition to numerous roles in early embryogenesis, Notch disorders [46,47], dysregulation of Dlx genes may underlie not signaling is important in CNS development and is also of only cognitive deficits but also the increased incidence of interest in the context of the development of AD pathology. psychiatric disorders in DS patients. Dlx genes are also essential NOTCH1 protein is found up-regulated in AD [49], and more for craniofacial development [48] and thus are novel candidates recently, data from the Ts1Cje mouse model have implicated for further study in the genesis of the primary DS phenotype. Notch signaling in DS [9] with confirmation from human A further interesting group of developmental genes found samples [50]. Abnormal APP processing may explain the up-regulated relate to Notch signaling; increased expression of downstream effects seen in Notch signaling as Notch and APP H.E. Lockstone et al. / Genomics 90 (2007) 647–660 657

Table 3 Biological processes overrepresented in the list of genes dysregulated in both fetal and adult brain GO ID GO biological process category No. of significant No. of genes Corrected genes on chip p value GO:0006338 Chromatin remodeling 3 20 0.001 GO:0007275 Development 11 369 0.004 GO:0006869 Lipid transport 3 46 0.006 GO:0006886 Intracellular protein transport 7 174 0.006 GO:0007517 Muscle development 4 111 0.016 GO:0006357 Regulation of transcription from RNA 6 166 0.016 Pol II promoter GO:0045449 Regulation of transcription 3 63 0.016 GO:0007169 Transmembrane receptor protein tyrosine 3 65 0.018 kinase signaling pathway GO:0007283 Spermatogenesis 3 98 0.022 GO:0008285 Negative regulation of cell proliferation 4 127 0.027 GO:0005975 Carbohydrate metabolism 4 185 0.040 GO:0007049 Cell cycle 5 211 0.044 Details of the number of significant genes falling into each category, the total number of genes in that category represented on the chip, and the significance level are given. are cleaved by the same enzyme (γ-secretase) [51], and there vesicle transport [57] and amyloid deposition is connected with is cross talk between the pathways via direct interaction of the the accumulation of large synaptic vesicles in the perikaryon proteins [50]. We also found genes for the γ-secretase [58,59]. Cytoskeleton organization and synaptic transmission, components PSEN1 and PSEN2 dysregulated at q b 0.05 in however, were enriched among the down-regulated genes, the adult DS brain. Furthermore, Notch signaling controls which suggests that despite the presence of enlarged vesicles cellular differentiation, apoptosis, proliferation, and migration and/or activated vesicle recycling processes, these are unable to [52], all pathways we found dysregulated in adult DS brain. In compensate fully for deficits in synaptic function. postmitotic neurons, Notch affects neurite extension [53,54] The genes dysregulated in immune-related categories point and the actin cytoskeleton [52,55]; therefore, this pathway toward an inflammatory state in the adult brain, which is likely to may also contribute to neurodegenerative processes. If be primarily linked to the development of AD pathology as these systemic, Notch-mediated abnormalities in cell proliferation genes were not found dysregulated in the fetal brain. In the context and the actin cytoskeleton may underpin a number of the wide of AD pathology, amyloid β is an activator of inflammatory range of health problems seen in DS individuals, including processes. Additionally, the presence of inflammatory mediators leukemia and heart defects. The dysregulation of Notch and in the brain may be a primary cause of damage to neurons [60,61]. Dlx are likely to be interrelated: loss of Dlx function leads to It is of note that in the original analysis of data from fetal tissue an increase in Notch signaling [56], which mirrors the results [21], MHC II genes were found up-regulated when fetal astro- observed here in the adult brain. From these data, together cytes were examined. Thus an underlying proinflammatory CNS with a recent report of abnormal Notch signaling in DS [50], environment may exist in DS individuals [4], which is exa- Notch-related pathways emerge as a clear candidate for further cerbated over time, concomitant with the buildup of amyloid β. study in this disorder. Interestingly, a number of transcription In summary, we report a gene expression profiling study of factors listed in the dysregulated category regulation of gene Down syndrome, the first using human adult brain tissue, which expression (e.g., ID4, TCF12, TCFL5) are members of the has identified hundreds of genes that are statistically signifi- basic helix-loop-helix family, targeted by Notch signaling cantly altered in the disorder. The transcriptome of chromosome [52]. Activation of major signaling pathways such as Notch 21 was predominantly affected, displaying a unique profile of may underlie the spread of transcriptional dysregulation up-regulation, and was accompanied by transcriptional changes throughout the genome. throughout the genome. It is not clear if these secondary effects In the adult brain the processes described above are are the direct consequence of trisomy 21 or whether, being concomitant with disruption of those relating to cytoskeletal present in the adult brain, they reflect cumulative changes regulation and vesicle trafficking, an increased immune re- resulting from the many DS pathologies. sponse, and response to oxidative stress. These processes, being Despite some overlap with the published data from fetal DS specifically altered in the adult DS brain, may be more likely to brain, overexpression of many of the trisomic genes was relate to the late onset pathologies. previously unconfirmed in humans with DS. Functional pro- In particular, categories related to the cytoskeleton and filing of transcripts that are dysregulated in the DS brain has vesicle function/trafficking were overrepresented among up- provided some insights into processes that may be important in regulated genes, in good agreement with data obtained in a the syndrome; however, the role played by triplication of genes parallel proteomics study of AD and DS brain tissue [31], and on chromosome 21 in triggering the processes involved has yet may relate to synaptic degeneration. APP cleavage is linked with to be elucidated. 658 H.E. Lockstone et al. / Genomics 90 (2007) 647–660

Materials and methods used for age and PMI and a Student t test was used for gender. The p values were adjusted using the Benjamini–Hochberg FDR correction [69]. Tissue collection QPCR validation Fresh-frozen prefrontal cortex tissue from 10 Down syndrome and 3 matched control subjects was obtained from the Cambridge Brain Bank (Addenbrooke's For quantitative real-time PCR, complementary DNA was synthesized from Hospital, Cambridge, UK). To increase statistical power, an additional 6 control 500 ng DNA-free RNA with an oligonucleotide deoxythymidine primer and the samples were included, obtained from the Neuropathology Consortium of the Superscript First-Strand Synthesis System (Invitrogen). QPCR was performed Stanley brain collection (Stanley Medical Research Institute, USA). These were using the Applied Biosystems 7900HT Sequence Detection System following selected from a group of 34 on the basis that they showed profiles similar to those the manufacturer's instructions. Eleven genes were selected for validation using of the 3 matched control samples, assessed by principal components analysis. TaqMan gene expression assays (Applied Biosystems), based on their location on Demographic and clinical data for the samples used in the analysis are provided chromosome 21 (ADAMTS1, ADARB1, APP, DYRK1A, DSCR1, BACE2, in Supplementary Table 1. The tissue was collected with full informed consent PTTG1IP, S100B) and/or their relevance to AD pathology (APOE, APPBP2, obtained from a first-degree relative after death in compliance with the NOTCH2). To avoid any potential DNA amplification that may have remained Declaration of Helsinki. Ethics approval for brain collection is held by the after DNase treatment, chosen QPCR assays spanned intron–exon boundaries. Stanley Medical Research Institute and ethics approval was received from the Additionally, the assays were designed to detect the same or very similar Cambridge local research ethics committee for use of human postmortem tissue. transcript populations measured by the corresponding significant microarray probe set. PGK1 was chosen as an endogenous control for the normalization of Acquisition and preprocessing of microarray data all target genes as it was consistently expressed in microarray samples. There was no significant difference in the cycle threshold (Ct) values for PGK1 between All procedures have previously been described [62]. In brief, total RNA was disease and control groups (p = 0.40). Standard curves were constructed for each extracted from all samples using Tri-reagent (Sigma) or Trizol (Gibco BRL) and assay to ensure amplification efficiencies were close to 100% and comparable RNA quality was assessed using a high-resolution electrophoresis system across all assays. The comparative Ct method was employed for quantification of (Agilent Technologies). Isolated total RNA was then carried through the the transcript expression levels [70]. Triplicate Ct values were generated for all Affymetrix preparation protocol (www.affymetrix.com) and each sample was assays; the median value in each case was used for subsequent analysis and −ΔΔCt hybridized to one HG-U133A GeneChip (Affymetrix). The additional control 2 values were then calculated. Two control samples were removed from the samples from the Stanley brain collection were run in the same laboratory but dataset on the grounds of poor cDNA profiles. For each gene, a two-tailed separately from the initial set of 13 samples from the Cambridge Brain Bank. Student t test was used to determine whether the mean expression was Raw data were processed and analyzed using the R statistical program [63] and significantly different in the DS group compared to controls. Bioconductor [64] packages. Our rigorous quality control (QC) procedures [62] included assessment of pairwise correlations between chips, box plots of robust Comparison to DS fetal brain multichip average (RMA) normalized expression values for each chip, and comparison of all chips to a pseudo-median chip. Based on these analyses, as well A publicly available microarray dataset from human DS fetal cerebrum [21] as the assessment of RNA quality, three DS samples and one matched control was used for comparison with the results from the adult DS brain. CEL files for sample were considered outliers and excluded from further analysis. Thus, the four DS and seven control subjects were downloaded from the Pevsner final analysis was based upon seven DS and eight control samples. Normalized Laboratory Web site (http://pevsnerlab.kennedykrieger.org/ds_cel_files.htm) expression values (log base 2) for chips passing QC were computed using the and the dataset was reanalyzed using the methods described above. We RMA method [65], which has been shown to detect expression levels reliably, compared the genes localized to chromosome 21 that were up-regulated in both independent of signal intensity (out-performing MAS5.0 (Affymetrix) and the the fetal and the adult DS brain. These genes represent strong candidates for the Model Based Expression Index [66] as well as being more sensitive for the disorder as their overexpression is reported by independent studies and also detection of differentially expressed genes [67]). The gene expression matrix was implies that the dosage increase is not compensated for in development or filtered to exclude control probe sets and to include only probe sets that had a adulthood. “class A” assignment (Affymetrix annotation file, December 2005), which are most specific to the target transcript sequence. This filtering step reduced the gene Functional profiling expression matrix to 19,759 probe sets. Raw microarray data have been deposited with GEO (Accession No. Affymetrix probe set IDs that were differentially expressed in adult DS brain GSE5390). were mapped to Entrez IDs using Onto-Translate [71], and the list of unique Entrez IDs was submitted to Onto-Express [72]. This software associates each Detection of differentially expressed genes gene with its Gene Ontology (GO) terms and identifies GO categories that are statistically overrepresented in the list of differentially expressed genes relative to A Bayesian moderated t test was applied to identify differentially expressed their representation on the chip. Up- and down-regulated lists were submitted genes using the LIMMA (linear models for microarray analysis) package [68] separately, using default settings (hypergeometric distribution and FDR from Bioconductor. The Benjamini–Hochberg false discovery rate (FDR) [69] (Benjamini–Hochberg) correction) and the Affymetrix human HG-U133A was applied to correct for multiple testing. Adjusted p values are referred to as q array as reference. GO categories with a corrected p value below 0.05 and values throughout. To calculate fold changes, RMA expression values were containing at least three genes were considered in the analysis. unlogged and the ratio of mean expression in the DS group to mean expression in The list of genes identified as differentially expressed in both fetal and adult the control group was computed. Values less than 1, indicating decreased datasets was also analyzed using Onto-Express as described above, with the expression in DS, were expressed as negative fold changes by multiplying the exception that due to the smaller number of genes in this list, up- and down- control/DS ratio by −1. regulated genes were submitted together.

Demographic variables Acknowledgments

The inclusion of additional control samples had one adverse consequence in This work was supported by a grant from the Community that the disease and control samples were less well matched for demographic variables than the original sample set. The correlation of differentially expressed Fund to the Alzheimers Research Trust, United Kingdom, and genes with age, gender, and PMI was determined to assess whether these the Stanley Medical Research Institute (SMRI). Brain tissue was variables had any influence on gene expression. Pearson's correlation test was donated by the Cambridge Brain Bank and the SMRI brain H.E. Lockstone et al. / Genomics 90 (2007) 647–660 659 collection courtesy of Michael B. Knable, E. Fuller Torrey, [17] S. Giannone, et al., Gene expression profile analysis in human Maree J. Webster, Serge Weis, and Robert H. Yolken. The T lymphocytes from patients with Down syndrome, Ann. Hum. Genet. 68 (Pt 6) (2004) 546–554. funding bodies were not involved in any aspect of the study apart [18] D.R. FitzPatrick, et al., Transcriptome analysis of human autosomal from provision of samples by SMRI. We thank Simon Bevan of trisomy, Hum. Mol. Genet. 11 (2002) 3249–3256. Geneservices for the QPCR assays, Sandra Pietsch for technical [19] S.J. Gross, et al., Gene expression profile of trisomy 21 placentas: a assistance, Rachel Craddock and Martin Lan for helpful potential approach for designing noninvasive techniques of prenatal – discussions and comments on the manuscript, and all other diagnosis, Am. J. Obstet. Gynecol. 187 (2002) 457 462. [20] R. Mao, C.L. Zielke, H.R. Zielke, J. Pevsner, Global up-regulation of members of the Bahn laboratory for their support. We also thank chromosome 21 gene expression in the developing Down syndrome brain, the Pevsner laboratory for making their published microarray Genomics 81 (2003) 457–467. data available. [21] R. Mao, et al., Primary and secondary transcriptional effects in the developing human Down syndrome brain and heart, Genome Biol. 6 (2005) Appendix A. Supplementary data R107. [22] D.R. FitzPatrick, Transcriptional consequences of autosomal trisomy: primary gene dosage with complex downstream effects, Trends Genet. 21 Supplementary data associated with this article can be found, (2005) 249–253. in the online version, at doi:10.1016/j.ygeno.2007.08.005. [23] S.E. Antonarakis, R. Lyle, E.T. Dermitzakis, A. Reymond, S. Deutsch, Chromosome 21 and Down syndrome: from genomics to pathophysiology, Nat. Rev. Genet. 5 (2004) 725–738. References [24] J.A. Birchler, U. Bhadra, M.P. Bhadra, D.L. Auger, Dosage-dependent gene regulation in multicellular eukaryotes: implications for dosage [1] G.T. Capone, Down syndrome: advances in molecular biology and the compensation, aneuploid syndromes, and quantitative traits, Dev. Biol. neurosciences, J. Dev. Behav. Pediatr. 22 (2001) 40–59. 234 (2001) 275–288. [2] R.E. Mrak, W.S. Griffin, Trisomy 21 and the brain, J. Neuropathol. Exp. [25] D.M. Mann, Calcification of the basal ganglia in Down's syndrome and Neurol. 63 (2004) 679–685. Alzheimer's disease, Acta Neuropathol. (Berlin) 76 (1988) 595–598. [3] R.H. Reeves, L.L. Baxter, J.T. Richtsmeier, Too much of a good thing: [26] B.A. Myers, S.M. Pueschel, Psychiatric disorders in persons with Down mechanisms of gene action in Down syndrome, Trends Genet. 17 (2001) syndrome, J. Nerv. Ment. Dis. 179 (1991) 609–613. 83–88. [27] J. Hardy, D. Allsop, Amyloid deposition as the central event in the [4] Y.C. de Hingh, et al., Intrinsic abnormalities of lymphocyte counts in aetiology of Alzheimer's disease, Trends Pharmacol. Sci. 12 (1991) children with Down syndrome, J. Pediatr. 147 (2005) 744–747. 383–388. [5] M.A. Pritchard, I. Kola, The “gene dosage effect” hypothesis versus the [28] J. Hardy, Has the amyloid cascade hypothesis for Alzheimer's disease been “amplified developmental instability” hypothesis in Down syndrome, proved? Curr. Alzheimer Res. 3 (2006) 71–73. J. Neural Transm. Suppl. 57 (1999) 293–303. [29] A.M. Cataldo, et al., App gene dosage modulates endosomal abnormalities [6] B.L. Shapiro, Whither Down syndrome critical regions? Hum. Genet. 99 of Alzheimer's disease in a segmental trisomy 16 mouse model of Down (1997) 421–423. syndrome, J. Neurosci. 23 (2003) 6788–6792. [7] J.R. Korenberg, et al., Molecular definition of a region of chromosome 21 [30] Gene Ontology Consortium, The Gene Ontology (GO) project in 2006, that causes features of the Down syndrome phenotype, Am. J. Hum. Genet. Nucleic Acids Res. 34 (2006) D322–D326 (database issue). 47 (1990) 236–246. [31] L.J.W. Harris, J.E. Swatton, M. Wengenroth, M. Wayland, H.E. [8] M.C. Potier, et al., Transcriptional disruptions in Down syndrome: a case Lockstone, A. Holland, R.L.M. Faull, K.S. Lilley, S. Bahn, Differences study in the Ts1Cje mouse cerebellum during post-natal development, in protein profiles in schizophrenia prefrontal cortex compared to other J. Neurochem. 97 (Suppl. 1) (2006) 104–109. major brain disorders, Clin. Schizophr. Relat. Psychoses 1 (1) (2007) [9] L. Dauphinot, et al., The cerebellar transcriptome during postnatal 73–91 (Apr). development of the Ts1Cje mouse, a segmental trisomy model for Down [32] K. Gardiner, Transcriptional dysregulation in Down syndrome: predictions syndrome, Hum. Mol. Genet. 14 (2005) 373–384. for altered protein complex stoichiometries and post-translational [10] K. Amano, et al., Dosage-dependent over-expression of genes in the modifications, and consequences for learning/behavior genes ELK, trisomic region of Ts1Cje mouse model for Down syndrome, Hum. Mol. CREB, and the estrogen and glucocorticoid receptors, Behav. Genet. 36 Genet. 13 (2004) 1333–1340. (2006) 439–453. [11] P. Kahlem, et al., Transcript level alterations reflect gene dosage effects [33] E.S. Lein, et al., Genome-wide atlas of gene expression in the adult mouse across multiple tissues in a mouse model of Down syndrome, Genome Res. brain, Nature 445 (2007) 168–176. 14 (2004) 1258–1267. [34] J.U. Pontius, L. Wagner, G.D. Schuler, UniGene: a unified view of the [12] R. Lyle, C. Gehrig, C. Neergaard-Henrichsen, S. Deutsch, S.E. transcriptome, The NCBI Handbook, National Center for Biotechnology Antonarakis, Gene expression from the aneuploid chromosome in a Information, Bethesda (MD), 2003. trisomy mouse model of Down syndrome, Genome Res. 14 (2004) [35] J.A. Birchler, N.C. Riddle, D.L. Auger, R.A. Veitia, Dosage balance in gene 1268–1274. regulation: biological implications, Trends Genet. 21 (2005) 219–226. [13] N.G. Saran, M.T. Pletcher, J.E. Natale, Y. Cheng, R.H. Reeves, Global [36] A. Rovelet-Lecrux, et al., APP locus duplication causes autosomal disruption of the cerebellar transcriptome in a Down syndrome mouse dominant early-onset Alzheimer disease with cerebral amyloid angiopathy, model, Hum. Mol. Genet. 12 (2003) 2013–2019. Nat. Genet. 38 (2006) 24–26. [14] R. Chrast, et al., The mouse brain transcriptome by SAGE: differences in [37] K. Sleegers, et al., APP duplication is sufficient to cause early onset gene expression between P30 brains of the partial trisomy 16 mouse model Alzheimer's dementia with cerebral amyloid angiopathy, Brain 129 (Pt 11) of Down syndrome (Ts65Dn) and normals, Genome Res. 10 (2000) (2006) 2977–2983. 2006–2021. [38] V.P. Prasher, et al., Molecular mapping of Alzheimer-type dementia in [15] C.M. Li, et al., Cell type-specific over-expression of chromosome 21 genes Down's syndrome, Ann. Neurol. 43 (1998) 380–383. in fibroblasts and fetal hearts with trisomy 21, BMC Med. Genet. 7 (2006) [39] F. Oyama, et al., Down's syndrome: up-regulation of beta-amyloid protein 24. precursor and tau mRNAs and their defective coordination, J. Neurochem. [16] Y. Tang, et al., Blood expression profiles for tuberous sclerosis complex 2, 62 (1994) 1062–1066. neurofibromatosis type 1, and Down's syndrome, Ann. Neurol. 56 (2004) [40] F. Argellati, et al., Evidence against the overexpression of APP in Down 808–814. syndrome, IUBMB Life 58 (2006) 103–106. 660 H.E. Lockstone et al. / Genomics 90 (2007) 647–660

[41] F.W. Pfrieger, Cholesterol homeostasis and function in neurons of the [58] D. Praprotnik, M.A. Smith, P.L. Richey, H.V. Vinters, G. Perry, Filament central nervous system, Cell Mol. Life Sci. 60 (2003) 1158–1171. heterogeneity within the dystrophic neurites of senile plaques suggests [42] J. Poirier, Apolipoprotein E and cholesterol metabolism in the pathogenesis blockage of fast axonal transport in Alzheimer's disease, Acta Neuro- and treatment of Alzheimer's disease, Trends Mol. Med. 9 (2003) 94–101. pathol. (Berlin) 91 (1996) 226–235. [43] W.J. Strittmatter, A.D. Roses, Apolipoprotein E and Alzheimer disease, [59] G.B. Stokin, et al., Axonopathy and transport deficits early in the Proc. Natl. Acad. Sci. USA 92 (1995) 4725–4727. pathogenesis of Alzheimer's disease, Science 307 (2005) 1282–1288. [44] K. Letinic, R. Zoncu, P. Rakic, Origin of GABAergic neurons in the human [60] I. Blasko, et al., How chronic inflammation can affect the brain and support neocortex, Nature 417 (2002) 645–649. the development of Alzheimer's disease in old age: the role of microglia [45] D.A. Lewis, T. Hashimoto, D.W. Volk, Cortical inhibitory neurons and and astrocytes, Aging Cell 3 (2004) 169–176. schizophrenia, Nat. Rev. Neurosci. 6 (2005) 312–324. [61] H. Zanjani, et al., Complement activation in very early Alzheimer disease, [46] F.M. Benes, M.S. Todtenkopf, P. Logiotatos, M. Williams, Glutamate Alzheimer Dis. Assoc. Disord. 19 (2005) 55–66. decarboxylase(65)-immunoreactive terminals in cingulate and prefrontal [62] M.M. Ryan, et al., Application and optimization of microarray cortices of schizophrenic and bipolar brain, J. Chem. Neuroanat. 20 (2000) technologies for human postmortem brain studies, Biol. Psychiatry 55 259–269. (2004) 329–336. [47] P.V. Choudary, et al., Altered cortical glutamatergic and GABAergic signal [63] R Development Core Team, R: A language and environment for statistical transmission with glial involvement in depression, Proc. Natl. Acad. Sci. computing, R Foundation for Statistical Computing, Vienna Austria, 2007 USA 102 (2005) 15653–15658. (http://www.R-project.org). [48] P. Kraus, T. Lufkin, Dlx homeobox gene control of mammalian limb and [64] R.C. Gentleman, et al., Bioconductor: open software development for craniofacial development, Am. J. Med. Genet. A 140 (13) (2006) computational biology and bioinformatics, Genome Biol. 5 (2004) R80. 1366–1376 (July 1). [65] R.A. Irizarry, et al., Exploration, normalization, and summaries of high [49] O. Berezovska, M.Q. Xia, B.T. Hyman, Notch is expressed in adult brain, density oligonucleotide array probe level data, Biostatistics 4 (2003) is coexpressed with presenilin-1, and is altered in Alzheimer disease, 249–264. J. Neuropathol. Exp. Neurol. 57 (1998) 738–745. [66] C. Li, W.H. Wong, Model-based analysis of oligonucleotide arrays: [50] D.F. Fischer, et al., Activation of the Notch pathway in Down syndrome: expression index computation and outlier detection, Proc. Natl. Acad. Sci. cross-talk of Notch and APP, FASEB J. 19 (2005) 1451–1458. USA 98 (2001) 31–36. [51] D. Selkoe, R. Kopan, Notch and Presenilin: regulated intramembrane [67] H.C. Galfalvy, et al., Sex genes for genomic analysis in human brain: proteolysis links development and degeneration, Annu. Rev. Neurosci. 26 internal controls for comparison of probe level data extraction, BMC (2003) 565–597. Bioinformatics 4 (2003) 37. [52] A. Louvi, S. Artavanis-Tsakonas, Notch signalling in vertebrate neural [68] G.K. Smyth, Limma: linear models for microarray data, in: R. Gentleman development, Nat. Rev. Neurosci. 7 (2006) 93–102. (editors), Bioinformatics and Computational Biology Solutions using R [53] L. Redmond, S.R. Oh, C. Hicks, G. Weinmaster, A. Ghosh, Nuclear and Bioconductor, Springer-Verlag, New York, 2005, pp. 397–420. Notch1 signaling and the regulation of dendritic development, Nat. [69] Y. Benjamini, Y. Hochberg, Controlling the false discovery rate: a practical Neurosci. 3 (2000) 30–40. and powerful approach to multiple testing, J. R. Stat. Soc. B 57 (1995) [54] N. Sestan, S. Artavanis-Tsakonas, P. Rakic, Contact-dependent inhibition 289–300. of cortical neurite growth mediated by notch signaling, Science 286 (1999) [70] K.J. Livak, T.D. Schmittgen, Analysis of relative gene expression data 741–746. using real-time quantitative PCR and the 2(−Delta Delta C(T)) method, [55] E. Giniger, A role for Abl in Notch signaling, Neuron 20 (1998) 667–681. Methods 25 (2001) 402–408. [56] K. Yun, et al., Modulation of the notch signaling by Mash1 and Dlx1/2 [71] S. Draghici, P. Khatri, R.P. Martins, G.C. Ostermeier, S.A. Krawetz, regulates sequential specification and differentiation of progenitor cell types Global functional profiling of gene expression, Genomics 81 (2003) in the subcortical telencephalon, Development 129 (2002) 5029–5040. 98–104. [57] D. Cai, et al., Presenilin-1 regulates intracellular trafficking and cell [72] S. Draghici, et al., Onto-Tools, the toolkit of the modern biologist: Onto- surface delivery of beta-amyloid precursor protein, J. Biol. Chem. 278 Express, Onto-Compare, Onto-Design and Onto-Translate, Nucleic Acids (2003) 3446–3454. Res. 31 (2003) 3775–3781.