Functional genomic analysis of amniotic fluid cell-free mRNA suggests that oxidative stress is significant in Down syndrome fetuses

Donna K. Slonima,b,1, Keiko Koidec,1, Kirby L. Johnsonc, Umadevi Tantravahid, Janet M. Cowanc, Zina Jarrahc, and Diana W. Bianchic,2

aDepartment of Computer Science, Tufts University, Medford, MA 02155; bDepartment of Pathology, Tufts University School of Medicine, Boston, MA 02111; cDivision of Genetics, Department of Pediatrics, Floating Hospital for Children at Tufts Medical Center, Boston, MA 02111; and dDepartment of Pathology, Women and Infants’ Hospital, Providence, RI 02905

Communicated by Leonard A. Herzenberg, Stanford University School of Medicine, Stanford, CA, April 10, 2009 (received for review December 18, 2008) To characterize the differences between second trimester Down In the present study, functional analysis of the differentially- syndrome (DS) and euploid fetuses, we used Affymetrix microar- expressed highlights the importance of oxidative stress and rays to compare expression in uncultured amniotic fluid several immediate downstream or compensatory processes (e.g., supernatant samples. Functional pathway analysis highlighted the ion transport, signaling pathways) in the DS fetus. We also importance of oxidative stress, ion transport, and G sig- analyzed our results using the Connectivity Map (17, 18), a tool naling in the DS fetuses. Further evidence supporting these results designed to identify compounds whose molecular effects either was derived by correlating the observed gene expression patterns mimic or reverse an observed gene expression signature. The to those of small molecule drugs via the Connectivity Map. Our tool is based on a database that contains gene-expression profiles results suggest that there are secondary adverse consequences of from human cell lines treated with 164 different bioactive small DS evident in the second trimester, leading to testable hypotheses molecules (18). Our application of the Connectivity Map serves about possible antenatal therapy for DS. two purposes. Primarily, it provides further functional confir- mation of the pathways identified using other tools. Additionally,

antenatal therapy ͉ Connectivity Map ͉ gene expression ͉ in this perinatal/developmental context, these results suggest MEDICAL SCIENCES prenatal diagnosis ͉ trisomy 21 hypotheses regarding potential antenatal treatments for some of the changes identified in DS fetuses. Although fetal treatment would be unlikely to prevent the development of congenital own syndrome (DS) is the most frequently occurring live Ϸ anomalies (most of which occur in the first trimester) and mental Dborn human autosomal trisomy, with an incidence of 1in retardation, the data presented here suggest that there are 800 births (1). DS was initially described in the mid 19th century secondary adverse biological consequences of DS that are on the basis of distinctive clinical features, such as mental evident in the second trimester. retardation, ‘‘flat and broad face...narrow palpebral fis- sures...long, thick and roughened tongue...and (the skin) Results giving the appearance of being too large for the body’’ (2). Differential Expression in DS vs. Controls. We identified 2 sets of Almost a century later, the underlying basis of the condition was genes that differed between the trisomy 21 and euploid samples. recognized as being due to partial or complete trisomy of First, we found 414 probe sets whose individual expression levels 21 (3). Subsequent work showed that the charac- were significantly different via paired t tests (adjusted P value Ͻ teristic DS phenotype could be achieved by having 3 copies of the 0.05) in samples matched for sex and gestational age. We call this genes contained within chromosomal band 21q22, which became the ‘‘Individual’’ gene set (detailed in Table S1). Only 5 probe known as the DS critical region (4, 5). sets among the Individual gene set were located on chromosome Prenatal screening for DS has recently been recommended for 21, corresponding to the genes CLIC6, ITGB2, RUNX1, and 2 all pregnancies in the United States (6), with the understanding ORFs of unknown function (C21orf67, C21orf86). Four of these that many women who are found to be carrying a DS fetus will five were up-regulated in DS; the exception was RUNX1, which opt to terminate the pregnancy. Virtually no attention has been was down-regulated in the DS samples. In the full individual paid to research involving ongoing DS pregnancies (7); however, gene set, 224 (54%) of the genes were up-regulated and 190 functional analyses of the molecular changes in DS may allow (46%) were down-regulated. There was widespread differential development of testable hypotheses that could identify targets expression between trisomic and euploid fetuses, and clustering for novel treatments in utero. based on these genes alone, excluding genes, is To characterize more fully the differences between DS and sufficient to separate the euploid and trisomic samples (Fig. 1). euploid fetuses, we used Affymetrix microarrays to perform a Second, gene set enrichment analysis (GSEA) (19) identified functional genomic analytic comparison between 7 living sec- ond-trimester DS fetuses and 7 euploid controls, matched for gender and gestational age. Because cell culture may result in Author contributions: D.K.S. and D.W.B. designed research; D.K.S., K.K., K.L.J., and Z.J. performed research; U.T. and J.M.C. contributed new reagents/analytic tools; D.K.S., K.K., changes in gene expression (8, 9), we used cell-free fetal mRNA K.L.J., and D.W.B. analyzed data; and D.K.S., K.K., K.L.J., U.T., J.M.C., and D.W.B. wrote the from uncultured amniotic fluid supernatant samples, which we paper. have demonstrated to be a stable and reliable source of infor- The authors declare no conflict of interest. mation regarding fetal gene expression (10, 11). Prior studies of Freely available online through the PNAS open access option. fetal gene expression in DS have exclusively used cultured cells Data deposition: The data reported in this paper have been deposited in the Gene (12–16), which may not reflect the fetus in vivo. Amniotic fluid Expression Omnibus (GEO) database, www.ncbi.nlm.nih.gov/geo (accession no. GSE16176). is in direct contact with the fetal oropharynx, lungs, gastroin- 1D.K.S. and K.K. contributed equally to this work. testinal tract, skin, and urinary system. It is the only body fluid 2To whom correspondence should be addressed. E-mail: [email protected]. that derives from multiple tissues and can be safely studied in This article contains supporting information online at www.pnas.org/cgi/content/full/ living fetuses with a confirmed karyotype. 0903909106/DCSupplemental.

www.pnas.org͞cgi͞doi͞10.1073͞pnas.0903909106 PNAS Early Edition ͉ 1of5 Downloaded by guest on September 27, 2021 Color Key A Average Log Fold-Change of Chromosome 21 Genes in DS vs. Controls

-2 -1 0 1 2 Row Z-score Frequency 0 102030405060 0 102030405060

−2 −1 0 1 2 3 4 Average Log Fold-Change B Average Log Fold-Change of Euploid Genes in DS vs. Controls T4 T6 T7 T2 T1 T3 T5 C6 C4 C7 C1 C2 C3 C5

Fig. 1. Heatmap showing hierarchical clustering of control (C1-C7; solid red bar) and trisomic (T1-T7; solid blue bar) samples. Clustering is based on data from 409 probe sets: the 414 from the Individual gene set, minus the 5 from

chromosome 21. If these five are not removed, the dendrogram looks quite Frequency similar; in particular, it similarly separates the control and trisomic samples. Expression values in each row are z-score normalized.

a single chromosomal band, chromosome 21, band 22, whose genes were significantly up-regulated as a group [false discovery 0 1000 2000 3000 4000 5000 6000 ϭ −2 −1 0 1 2 3 4 rate (FDR) q value 0.006] in DS fetuses. For functional Average Log Fold-Change analysis, we selected the 82-gene ‘‘Leading Edge’’ subset (19) of the genes GSEA identified from band chr21q22 (see Materials Fig. 2. Histogram of fold-changes in gene expression between the average and Methods and Table S2). Only 3 genes (CLIC6, RUNX1, and expression in DS and the average in the controls. (A) Histogram for all 501 probe sets representing genes on chromosome 21. The fold-changes are C21orf87) are common to both the Individual and Leading Edge reported in log-scale (base 2), so that up- and down-regulation appear at gene sets. equal distances from zero. For example, a value of 1.0 represents 2-fold To quantify the extent of differential expression of the up-regulation in DS. The bold vertical line at Ϸ0.58 corresponds to the 1.5-fold known trisomic genes, we examined the changes in expression change expected from the gene-dosage hypothesis. (B) Histogram of all probe levels of all chromosome 21 probes on our microarrays. For sets representing genes on other than 21, shown on the same each probe set, we computed the fold-change between its scale as in A). average expression level in the DS samples and its average expression in the controls. A histogram of these changes is shown in Fig. 2A, in which a bold vertical line marks the functional annotations with a modified Fisher exact P value (the 1.5-fold up-regulation expected given the increased gene dos- ‘‘EASE’’ score) Ͻ0.1 (see Materials and Methods). The full age. Overexpression was seen for 65% of the chromosome 21 DAVID results for the 2 gene sets appear in Table S3 and Table probes (325 of 501). The mean and median fold-changes were S4. We observed several consistent patterns in differential expres- 1.44 and 1.25, respectively, but the range is quite large (from sion in both the Individual and the Leading Edge gene sets (Table 5-fold down-regulation to 16-fold up-regulation in DS). In S5). Because of the limited overlap between these 2 sets and the contrast, only 50.4% (27,299 of 54,174) of the probes from size of the functional groups considered, the 2 gene sets can chromosomes other than 21 are higher in DS (Fig. 2B); this is be seen as providing largely independent confirmation of the significantly different from the frequency of overexpression we importance of these functional processes in DS. Therefore, we see on chromosome 21 (␹2 test, P Ͻ 0.0006). chose to focus only on functional processes implicated by both gene sets. Using this criterion, the following functions appear to Functional Analysis. We performed pathway analysis of the Indi- be disrupted in DS (Table S5): oxidative stress, ion transport, G vidual and Leading Edge gene sets in Database for Annotation, protein signaling, immune and stress response, circulatory sys- Visualization, and Integrated Discovery (DAVID) (20). Follow- tem functions, cell structure, sensory perception, and several ing the suggestions of DAVID’s creators (21), we examined all developmental processes.

2of5 ͉ www.pnas.org͞cgi͞doi͞10.1073͞pnas.0903909106 Slonim et al. Downloaded by guest on September 27, 2021 Confirmation and Therapeutic Suggestions Using the Connectivity Map. In an attempt to further confirm the importance of these reactive functional processes, we used the Connectivity Map (17, 18) to oxygen identify compounds whose molecular signatures either mimic or species counteract that of DS. We found 4 compounds with average connectivity scores Ͼ0.7 (indicating a high correlation with the DS molecular signature), and 9 compounds with average con- nectivity scores less than Ϫ0.7 (indicating a high negative correlation). The full results of the Connectivity Map analysis oxidative appear in SI Appendix. stress The compounds that potentially would reverse our observed DS molecular phenotype (and thus might be candidates for further hypothesis testing in vitro) include NSC-5255229, celastrol, calmi- dazolium, NSC-5109870, dimethyloxalylglycine, NSC-5213008, ve- rapamil, HC toxin, and felodipine. Celastrol is an antioxidant and cell stress ion transport anti-inflammatory agent that has been suggested for use in treating Alzheimer disease, which prematurely affects many DS patients response (22). Calmidazolium is a calmodulin inhibitor, which decreases sensitivity to calcium ion signaling, and has been considered for use in treating osteoporosis (23). Verapamil and felodipine are both signal G - protein blockers, whereas dimethyloxalylglycine is a hy- transduction droxylase inhibitor thought to increase resistance to oxidative stress signaling pathways (24, 25). The 4 compounds that most mimic the DS phenotype also relate to potassium and calcium signaling or oxidation. The fact that these results also implicate oxidative stress and ion transport provides a third level of confirmation of the importance of these functional classes. cell structure

Discussion MEDICAL SCIENCES We have demonstrated that transcriptional profiling of RNA in uncultured amniotic fluid provides a unique molecular window into developmental disorders in the living human fetus. In addition to identifying genes relevant to the DS phenotype, we apoptosis failure of have used functional profiling to identify significantly disrupted cell integrity biological pathways. Among the functional pathway groups identified by both the individual and gene set analyses, several are amenable to a single explanation. Reactive oxygen species, especially hydrogen per- secondary adverse oxide, are known to disrupt ion transport mechanisms, leading consequences of DS to problems with signal transduction through cell membranes, cell dysfunction, structural failure of membrane integrity, and Fig. 3. Putative network of pathways implicated in DS. Significantly impli- ultimately to pathological symptoms, particularly in neural and cated processes, based on DAVID functional analysis, are shown in boxes. cardiac tissues (26). We observe consistent evidence of several Edges between boxes represent relationships between functional processes of these steps, including dysregulation of oxidative stress re- such as G protein signaling and disruptions in ion transport that result from sponse genes, phospholipids, ion transport molecules, heart oxidative stress, as suggested in refs. 26 (solid lines), 14 (dotted lines), 48 muscle genes, structural proteins, and DNA damage repair (dashed lines), and 49 (dash-dotted lines). Blue boxes with dashed borders are genes, in both the Individual and the Leading Edge gene sets processes over-represented in the Individual gene set only; magenta boxes with thin solid borders are processes over-represented in both the Individual (Table S5 and Fig. 3). and Leading Edge gene sets; and green boxes with thick solid borders are It has been suggested that oxidative stress plays an important implicated by both the Individual and Leading Edge gene sets and by the role in DS (27). Because individuals with DS demonstrate Connectivity Map. pathology consistent with Alzheimer disease at an early age (28), links to the role of oxidative stress in Alzheimer’s have been explored (27). Lockstone et al. (29) found that oxidative stress fluid samples from DS fetuses (31). Our study is the first response genes were over-represented in adult but not fetal DS functional analysis of the DS fetus that implicates not only tissue, and suggested that this response might reflect adult-onset oxidative stress, but potential intermediate consequences, such DS pathologies such as Alzheimer disease. More recently, a few as defects in ion transport and G protein signaling. groups (16, 30) have found oxidative stress response markers in In mouse models, at least one G protein coupled potassium fetal DS tissues, although neither study emphasized this partic- channel protein (GIRK2) has been implicated in DS pathology ular result or considered the potential relationship between (32, 33). Another study using adult mouse models has suggested oxidative stress and other functional pathways. Esposito, et al. a role for two other G protein dependent pathways in DS and (14) identified oxidative stress and apoptosis genes in neural Alzheimer disease (34). Our work, however, suggests a wider and progenitor cell lines generated from the frontal cortex of second more fundamental role for G protein signaling, involving a large trimester DS fetuses. They suggested that up-regulation of the number of proteins and appearing as early as the second chromosome 21 gene S100B causes an increase in reactive trimester. oxygen species and stress-response kinases, leading to an in- Our results contribute to an ongoing debate regarding the crease in programmed cell death. Using a biochemical approach, extent of transcriptional changes due to trisomy 21. Despite the other investigators demonstrated increased levels of isopros- many prior studies of gene expression in DS, the precise mech- tanes, a marker of oxidative stress, in second trimester amniotic anism by which the additional set of chromosome 21 genes

Slonim et al. PNAS Early Edition ͉ 3of5 Downloaded by guest on September 27, 2021 disrupts normal development and results in the phenotype of DS RNA, was amplified and purified using the WT-Ovation Pico RNA Amplifi- remains unknown. Consistent with several previous studies (30, cation System (NuGEN) and the DNA Clean & Concentrator-25 (Zymo 35, 36), we observe that the trisomic genes generally showed Research). The quality and quantity of amplified cDNA was measured on increased expression in DS, with average up-regulation of nearly the Agilent Bioanalyzer 2100 Expert software (Agilent) with the RNA 6000 1.5-fold (Fig. 2A). However, this effect is highly variable, with Nano kit (Agilent) . nearly a third of the trisomic genes actually down-regulated on average (but few significantly so). We see widespread differential Fragmentation, Labeling, and Hybridization. The FL-Ovation cDNA Biotin Mod- ule V2 (NuGEN) was used to obtain fragmented and labeled cDNA (minimum expression of genes from the other diploid chromosomes (Fig. 5 mcg) that was suitable for hybridization to Affymetrix U133 Plus 2.0 arrays 2B), consistent with the results of Gardiner (37) and Lockstone (Affymetrix). Arrays were washed, stained with streptavidin-phycoerythrin, et al. (29). Hierarchical clustering of our samples based on scanned with the GeneArray Scanner, and analyzed using the GeneChip expression levels of the 409 Individual genes not located on Microarray Suite 5.0 (Affymetrix). Array quality was assessed in R (version chromosome 21 completely separates the DS samples from the 2.7.2) using the simpleaffy package in BioConductor (version 1.7; available at controls (as seen in Fig. 1). In contrast, Mao et al. (30), who www.bioconductor.org). Three arrays with scaling factors Ͼ22 and Ͻ15% studied gene expression from frozen fetal heart and brain present calls were discarded. tissue, found that explicit classification of their samples Seven samples from DS fetuses remained: 5 males and 2 females. Five worked only when based on the expression levels of the gender-matched controls were matched within 4 days of gestational age of the corresponding DS samples; the other two were collected 10 and 12 days trisomic genes. Differences in expression patterns between earlier than the respective DS samples. A total of 7 DS and 7 matched controls their work and the present study may reflect the different were further analyzed. tissues involved. Although previous reports identified significant differential Analysis and Statistics. Normalization was performed with the threestep expression of trisomic genes, our analysis did not. We note that, command from the AffyPLM package in BioConductor, using ideal-mismatch because most of the amniotic fluid RNA is cell-free, care should background/signal adjustment, quantile normalization, and the Tukey bi- be taken when comparing our results to previously published weight summary method (40). This summary method includes a logarithmic transcriptomic profiles of material that used fetal cells or tissue transformation (40), improving the normality of the data. Identification of (12, 13, 15, 16, 30, 38, 39). This discrepancy may also be due to individual differentially-expressed genes was performed via 2-sided, paired t comparison with data derived from mouse models of DS, which tests using the multtest package in BioConductor, with the Benjamini– are more genetically homogeneous than our human population Hochberg adjustment for multiple testing (41). sample. However, most likely the difference is due to our use of Gene Set Enrichment Analysis (GSEA) (19) was performed using the GSEA software version 2.0 and MSigDB version 2.4. This analysis identifies consistent a strict statistical cutoff for differential expression, including Ͼ differential expression of sets of genes defined in the MSigDB database. We adjustment for multiple testing of 54,000 probe sets. We found examined both the functional, curated gene sets (MSigDB collection c2) and relatively few chromosome 21 genes with such consistent ex- gene sets defined by chromosomal bands (MSigDB collection c1), but only the pression in our diverse population that the evidence for their chromosomal band analysis yielded sets that were significant with a false moderate up-regulation exceeded this strict significance cutoff. discovery rate (FDR) Ͻ0.05. The full results of the chromosomal band analysis Fortunately, GSEA was developed precisely to detect such appear in Table S6. consistent but modest expression changes. With no a priori bias, To identify the most differentially expressed genes from these statistically the GSEA tool identified the DS critical region as the only significant gene sets, we chose the ‘‘leading edge subset,’’ a group of the strongly (q Ͻ 0.05) up-regulated chromosomal band in the DS most-up-regulated genes in the gene set (19). Specifically, the leading edge subset of a gene set contains the genes that contribute the most to the set’s samples. enrichment score (ES), a statistic reflecting the degree to which a gene set is The addition of the Connectivity Map analysis confirmed over-represented at the top or bottom of a list of genes ranked by their the pathways implicated by DAVID and suggested possible differential expression. testable hypotheses to develop novel treatments for DS, Functional analysis of gene lists was performed in DAVID (20), using the starting with an in vitro approach to explore the effects of Panther functional annotation classes (42) in addition to the default pathway either the Connectivity Map-implicated compounds or other selections, which include (GO) terms (43), pathways defined compounds with similar effects on oxidation or ion transport. from the KEGG (44) and BioCarta (www.biocarta.com) databases, InterPro The work described here serves as proof of concept that gene protein families (45), and Protein Information Resource keywords (46). We expression profiles from living second trimester human fetuses chose to use DAVID’s EASE score rather than the more stringent Benjamini– with developmental disorders can lead to a better understand- Hochberg FDR cutoff for the DAVID results because the FDR adjustment assumes independence of the functional pathways (which, in practice, overlap ing of the early etiology of disease and the secondary conse- heavily by design), and adjusting for multiple testing in such cases is contro- quences of congenital anomalies, and may suggest future versial (47). Therefore, to reduce the possibility of false-positive associations, innovative approaches to treatment. we focus on only those functional processes represented in the DAVID output for both the Individual and Leading Edge gene sets. Materials and Methods Hierarchical clustering was performed in R, using complete-linkage hier- Samples. The Institutional Review Boards at Tufts Medical Center and archical clustering (the hclust function in the stats package), and heatmaps ϭ Women and Infants’ Hospital approved the study. Residual second trimes- created via the heatmap.2 function in the gplots package, using the ‘‘scale ter amniotic fluid (AF) supernatant samples were obtained from women ‘row’’’ option to z-score normalize the rows. To identify compounds with carrying singleton fetuses undergoing fetal genetic testing for routine molecular signatures that might mimic or mitigate the effects of DS, we used clinical indications. All samples were anonymous, although the karyotype Connectivity Map build 1.0, which contains a database of 564 expression result and gestational age were known. AF was stored at Ϫ80 °C until RNA profiles representing the effects of 164 compounds on 4 cancer cell lines, using extraction. The initial study set consisted of AF samples with the following the Affymetrix U133A microarrays (18). Because the U133 Plus 2.0 arrays used confirmed metaphase karyotypes: 47,XX,ϩ21 (n ϭ 4); 47,XY,ϩ21 (n ϭ 5); 46, in the present study contain a superset of the probe sets on the U133A arrays, XX (n ϭ 6); 46, XY (n ϭ 6). The gestational ages of the samples ranged from we ran the Connectivity Map analysis using only those probe sets that were 16 4/7 to 21 4/7 weeks. common to both arrays.

RNA Extraction. RNA was extracted from 10 mL of AF with 30 mL of TRIzol ACKNOWLEDGMENTS. We thank the maternal-fetal medicine staffs of both Tufts Medical Center and Women and Infants’ Hospital for performing the LS Reagent (Invitrogen) and 8 mL of chloroform. After RNA extraction, RNA amniocenteses, Ms. Helene Stroh for technical support, and Dr. Jill Maron for was purified and DNA was removed using the RNeasy Maxi Kit, including her advice. This work was supported by Eunice Kennedy Shriver National the DNase step according to the manufacturer’s protocol (QIAGEN). RNA Institute of Child Health and Human Development Awards R01HD042053-06 was precipitated using 3M NaOAc and 100% ethanol, and 80% ethanol was (to D.W.B.) and R01HD058880-01 (to D.K.S.), and a Tufts Mellon Research added after 4-h incubation at Ϫ20 °C. cDNA, synthesized from extracted Fellowship (to D.K.S.).

4of5 ͉ www.pnas.org͞cgi͞doi͞10.1073͞pnas.0903909106 Slonim et al. Downloaded by guest on September 27, 2021 1. Collins VR, Muggli EE, Riley M, Palma S, Halliday JL (2008) Is Down syndrome a 26. Kourie JI (1998) Interaction of reactive oxygen species with ion transport mechanisms. disappearing birth defect? J Pediatr 152:20–4, 24 e1. Am J Physiol 275:C1–C4. 2. Down JL (1866) Observations on an ethnic classification of idiots. London Hospital 27. Zana M, Janka Z, Kalman J (2007) Oxidative stress: A bridge between Down’s syndrome Reports 3:259–262. and Alzheimer’s disease. Neurobiol Aging 28:648–676. 3. Lejeune J, Gautier M, Turpin R (1959) Study of somatic chromosomes from 9 mongoloid 28. Bush A, Beail N (2004) Risk factors for dementia in people with Down syndrome: Issues children. C R Hebd Seances Acad Sci 248:1721–1722. in assessment and diagnosis. Am J Ment Retard 109:83–97. 4. Korenberg JR, et al. (1990) Molecular definition of a region of chromosome 21 that 29. Lockstone HE, et al. (2007) Gene expression profiling in the adult Down syndrome causes features of the Down syndrome phenotype. Am J Hum Genet 47: brain. Genomics 90:647–660. 236–246. 30. Mao R, et al. (2005) Primary and secondary transcriptional effects in the developing 5. Shapiro BL (1997) Whither Down syndrome critical regions? Hum Genet 99:421–423. human Down syndrome brain and heart. Genome Biol 6:R107. 6. (2007) ACOG Practice Bulletin No. 77: Screening for fetal chromosomal abnormalities. 31. Perrone S, et al. (2007) Early oxidative stress in amniotic fluid of pregnancies with Down Obstet Gynecol 109:217–227. syndrome. Clin Biochem 40:177–180. 7. Rasmussen SA, Whitehead N, Collier SA, Frias JL (2008) Setting a public health research agenda for Down syndrome: Summary of a meeting sponsored by the Centers for 32. Best TK, Siarey RJ, Galdzicki Z (2007) Ts65Dn, a mouse model of Down syndrome, Disease Control and Prevention and the National Down Syndrome Society. AmJMed exhibits increased GABAB-induced potassium current. J Neurophysiol 97:892–900. Genet A 146A:2998–3010. 33. Harashima C, et al. (2006) Elevated expression of the G-protein-activated inwardly 8. Sambuy Y, et al. (2005) The Caco-2 cell line as a model of the intestinal barrier: Influence rectifying 2 (GIRK2) in cerebellar unipolar brush cells of a Down of cell and culture-related factors on Caco-2 cell functional characteristics. Cell Biol syndrome mouse model. Cell Mol Neurobiol 26:719–734. Toxicol 21:1–26. 34. Lumbreras M, et al. (2006) Brain G protein-dependent signaling pathways in Down 9. Li Y, et al. (2007) Transcriptional changes associated with breast cancer occur as normal syndrome and Alzheimer’s disease. Amino Acids 31:449–456. human mammary epithelial cells overcome senescence barriers and become immor- 35. Amano K, et al. (2004) Dosage-dependent over-expression of genes in the trisomic talized. Mol Cancer 6:7. region of Ts1Cje mouse model for Down syndrome. Hum Mol Genet 13:1333–1340. 10. Larrabee PB, et al. (2005) Global gene expression analysis of the living human fetus 36. Dauphinot L, et al. (2005) The cerebellar transcriptome during postnatal development using cell-free messenger RNA in amniotic fluid. J Am Med Assoc 293:836–842. of the Ts1Cje mouse, a segmental trisomy model for Down syndrome. Hum Mol Genet 11. Larrabee PB, Johnson KL, Peter I, Bianchi DW (2005) Presence of filterable and nonfil- 14:373–384. terable cell-free mRNA in amniotic fluid. Clin Chem 51:1024–1026. 37. Gardiner K (2006) Transcriptional dysregulation in Down syndrome: Predictions for 12. Altug-Teber O, et al. (2007) Specific transcriptional changes in human fetuses with altered protein complex stoichiometries and post-translational modifications, and autosomal trisomies. Cytogenet Genome Res 119:171–184. consequences for learning/behavior genes ELK, CREB, and the estrogen and glucocor- 13. Chung IH, et al. (2005) Gene expression analysis of cultured amniotic fluid cell with ticoid receptors. Behav Genet 36:439–453. Down syndrome by DNA microarray. J Korean Med Sci 20:82–87. 38. Gross SJ, et al. (2002) Gene expression profile of trisomy 21 placentas: A potential 14. Esposito G, et al. (2008) Genomic and functional profiling of human Down syndrome approach for designing noninvasive techniques of prenatal diagnosis. Am J Obstet neural progenitors implicates S100B and 4 in cell injury. Hum Mol Genet Gynecol 187:457–462. 17:440–457. 15. FitzPatrick DR, et al. (2002) Transcriptome analysis of human autosomal trisomy. Hum 39. Li CM, et al. (2006) Cell type-specific over-expression of chromosome 21 genes in Mol Genet 11:3249–3256. fibroblasts and fetal hearts with trisomy 21. BMC Med Genet 7:24. 16. Rozovski U, et al. (2007) Genome-wide expression analysis of cultured trophoblast with 40. Gentleman R, Carey V, Huber W, Irizarry RA, Duduoit S, eds (2005) Bioinformatics and MEDICAL SCIENCES trisomy 21 karyotype. Hum Reprod 22:2538–2545. Computational Biology Solutions Using R and Bioconductor (Springer, New York). 17. Lamb J (2007) The Connectivity Map: A new tool for biomedical research. Nat Rev 41. Benjamini Y, Hochberg Y (1995) Controlling the false discovery rate: A practical and Cancer 7:54–60. powerful approach to multiple testing. J Roy Statist Soc Ser B 57:289–300. 18. Lamb J, et al. (2006) The Connectivity Map: Using gene-expression signatures to 42. Thomas PD, et al. (2003) PANTHER: A library of protein families and subfamilies connect small molecules, genes, and disease. Science 313:1929–1935. indexed by function. Genome Res 13:2129–2141. 19. Subramanian A, et al. (2005) Gene set enrichment analysis: A knowledge-based ap- 43. Ashburner M, et al. (2000) Gene ontology: Tool for the unification of biology. The Gene proach for interpreting genome-wide expression profiles. Proc Natl Acad Sci USA Ontology Consortium. Nat Genet 25:25–29. 102:15545–15550. 44. Kanehisa M, Goto S (2000) KEGG: Kyoto encyclopedia of genes and genomes. Nucleic 20. Dennis G, Jr., et al. (2003) DAVID: Database for Annotation, Visualization, and Inte- Acids Res 28:27–30. grated Discovery. Genome Biol 4:P3. 45. Apweiler R, et al. (2000) InterPro—an integrated documentation resource for protein 21. Hosack DA, Dennis G, Jr, Sherman BT, Lane HC, Lempicki RA (2003) Identifying bio- families, domains and functional sites. Bioinformatics 16:1145–1150. logical themes within lists of genes with EASE. Genome Biol 4:R70. 46. Barker WC, et al. (2000) The protein information resource (PIR). Nucleic Acids Res 22. Allison AC, Cacabelos R, Lombardi VR, Alvarez XA, Vigo C (2001) Celastrol, a potent 28:41–44. antioxidant and anti-inflammatory drug, as a possible treatment for Alzheimer’s 47. Gentleman R (2004) Using GO for statistical analyses. COMPSTAT 2004–Proceedings of disease. Prog Neuropsychopharmacol Biol Psychiatry 25:1341–1357. the 2004 Symposium in Computational Statistics, ed Antoch J (Springer, New York), pp 23. Seales EC, Micoli KJ, McDonald JM (2006) Calmodulin is a critical regulator of oste- oclastic differentiation, function, and survival. J Cell Biochem 97:45–55. 171–180. 24. Cummins EP, et al. (2008) The hydroxylase inhibitor dimethyloxalylglycine is protective 48. Mates JM, Segura JA, Alonso FJ, Marquez J (2008) Intracellular redox status and in a murine model of colitis. Gastroenterology 134:156–165. oxidative stress: Implications for cell proliferation, apoptosis, and carcinogenesis. Arch 25. Zaman K, et al. (1999) Protection from oxidative stress-induced apoptosis in cortical Toxicol 82:273–299. neuronal cultures by iron chelators is associated with enhanced DNA binding of 49. Lehotsky J, et al. (1999) Membrane ion transport systems during oxidative stress in hypoxia-inducible factor-1 and ATF-1/CREB and increased expression of glycolytic rodent brain: Protective effect of stobadine and other antioxidants. Life Sci 65:1951– enzymes, p21(waf1/cip1), and erythropoietin. J Neurosci 19:9821–9830. 1958.

Slonim et al. PNAS Early Edition ͉ 5of5 Downloaded by guest on September 27, 2021