RESEARCH ARTICLE SUMMARY

PSYCHIATRIC GENOMICS

Neuron-specific signaturesin the chromosomal connectomeassociated with schizophrenia riskPrashanth Rajarajan Tyler Borrman Will Liao Nadine Schrode Erin FlahertyCharlize Casintildeo Samuel Powell Chittampalli Yashaswini Elizabeth A LaMarcaBibi Kassim Behnam Javidfar Sergio Espeso-Gil Aiqun Li Hyejung WonDaniel H Geschwind Seok-Man Ho Matthew MacDonald Gabriel E HoffmanPanos Roussos Bin Zhang Chang-Gyu Hahn Zhiping WengdaggerKristen J Brennanddagger Schahram AkbariandaggerDagger

INTRODUCTION Chromosomal conforma-tions topologically associated chromatin do-mains (TADs) assembling in nested fashionacross hundreds of kilobases and other ldquothree-dimensional genomerdquo (3DG) structures bypassthe linear genome on a kilo- ormegabase scaleand play an important role in transcriptionalregulation Most of the genetic variants asso-ciated with risk for schizophrenia (SZ) arecommon and could be located in enhancersrepressors and other regulatory elements thatinfluence gene expression however the role ofthe brainrsquos 3DG for SZ genetic risk architec-ture including developmental and cell typendashspecific regulation remains poorly understood

RATIONALE We monitored changes in 3DGafter isogenic differentiation of human inducedpluripotent stem cellndashderived neural progenitorcells (NPCs) into neurons or astrocyte-like glialcells on a genome-wide scale using Hi-C Withthis in vitro model of brain development wemapped cell typendashspecific chromosomal confor-mations associated with SZ risk loci and de-fined a risk-associated expanded genome space

RESULTS Neural differentiation was associ-ated with genome-wide 3DG remodeling in-cluding pruning and de novo formations ofchromosomal loopings The NPC-to-neurontransition was defined by the pruning of loopsinvolving regulators of cell proliferation mor-phogenesis and neurogenesis which is con-sistent with a departure from a precursor stagetoward postmitotic neuronal identity Loopslost during NPC-to-glia transition includedmanygenes associated with neuron-specific func-tions which is consistent with non-neuronallineage commitment However neurons togetherwith NPCs as compared with glia harbored amuch larger number of chromosomal interac-tions anchored in common variant sequencesassociated with SZ risk Because spatial 3DGproximity of genes is an indicator for potentialcoregulation we tested whether the neural celltypendashspecific SZ-related ldquochromosomal connec-tomerdquo showed evidence of coordinated tran-scriptional regulation and proteomic interactionof the participating genesTo this end we generated lists of genes an-

chored in cell typendashspecific SZ risk-associated

interactions Thus for the NPC-specific interac-tions we counted 386 genes including 146with-in the risk loci andanother 240 genes positionedelsewhere in the linear genome but connectedvia intrachromosomal contacts to risk locus se-quences Similarly for the neuron-specific inter-actions we identified 385 genes 158 within riskloci and 227 outside of risk loci Last for glia-specific interactions we identified 201 genes88within and 113 outside of risk lociWe labeled

the genes located outsideof schizophrenia risk locias ldquorisk locusndashconnectrdquowhich we define as a col-lection of genes identifiedonly through Hi-C inter-action data expandingmdash

depending on cell typemdashby 50 to 150 thecurrent network of known genes overlappingrisk sequences that is informed only by genome-wide association studies This disease-relatedchromosomal connectome was associated withldquoclustersrdquo of coordinated gene expression andprotein interactions with at least one clusterstrongly enriched for regulators of neuronalconnectivity and synaptic plasticity and anothercluster for chromatin-associated proteins in-cluding transcriptional regulators

CONCLUSION Our study shows that neuraldifferentiation is associatedwith highly cell typendashspecific 3DG remodeling This process is paral-leled by an expansion of 3DG space associatedwith SZ risk Specifically developmentally reg-ulated chromosomal conformation changes atSZ-relevant sequences disproportionally occurredin neurons highlighting the existence of celltypendashspecific disease risk vulnerabilities inspatial genome organization

RESEARCH | PSYCHENCODE

Rajarajan et al Science 362 1269 (2018) 14 December 2018 1 of 1

The list of author affiliations is available in the full article onlineThese authors contributed equally to this workdaggerThese authors contributed equally to this workDaggerCorresponding author Email schahramakbarianmssmeduCite this article as P Rajarajan et al Science 362 eaat4311(2018) DOI 101126scienceaat4311

Protein

RNA

CRISPRa

Risk locus Risk locus-connect

Chromosomal connectomeassociated with schizophrenia

Schizophreniarisk locus

Gene 1 Gene 2 Gene 3 Genes

224 240227

113

Large-scale 3D genome remodeling during neural differentiation

3DG remodeling across neuronal differentiation with parallel expansion of SZ risk space (Left) Chromatin conformation assays revealpruning of short-range loops in neurons along with widening of TADs upon differentiation from NPCs (Right) Cell typendashspecific chromatin interactionsfunctionally validated with CRISPR assays expand the network of known risk-associated genes (blue circle) which show evidence for coregulationat the transcriptomic and proteomic levels

ON OUR WEBSITE

Read the full articleat httpdxdoiorg101126scienceaat4311

on February 26 2020

httpsciencesciencem

agorgD

ownloaded from

RESEARCH ARTICLE

PSYCHIATRIC GENOMICS

Neuron-specific signaturesin the chromosomal connectomeassociated with schizophrenia riskPrashanth Rajarajan1234 Tyler Borrman5 Will Liao6 Nadine Schrode234Erin Flaherty1347 Charlize Casintildeo2 Samuel Powell1234 Chittampalli Yashaswini1Elizabeth A LaMarca1234 Bibi Kassim24 Behnam Javidfar24 Sergio Espeso-Gil24Aiqun Li34 Hyejung Won8daggerDagger Daniel H Geschwind8 Seok-Man Ho134Matthew MacDonald9 Gabriel E Hoffman3 Panos Roussos23 Bin Zhang3Chang-Gyu Hahn9 Zhiping Weng5sect Kristen J Brennand2347sect Schahram Akbarian24sectpara

To explore the developmental reorganization of the three-dimensional genome ofthe brain in the context of neuropsychiatric disease we monitored chromosomalconformations in differentiating neural progenitor cells Neuronal and glial differentiationwas associated with widespread developmental remodeling of the chromosomalcontact map and included interactions anchored in common variant sequences thatconfer heritable risk for schizophrenia We describe cell typendashspecific chromosomalconnectomes composed of schizophrenia risk variants and their distal targets whichaltogether show enrichment for genes that regulate neuronal connectivity and chromatinremodeling and evidence for coordinated transcriptional regulation and proteomicinteraction of the participating genes Developmentally regulated chromosomalconformation changes at schizophrenia-relevant sequences disproportionally occurred inneurons highlighting the existence of cell typendashspecific disease risk vulnerabilities inspatial genome organization

Spatial genome organization is highly regu-lated and critically important for normalbrain development and function (1) Manyof the risk variants contributing to theheritability of complex genetic psychiatric

disorders are located in noncoding sequences (2)presumably embedded in ldquothree-dimensionalgenomerdquo (3DG) structures important for tran-scriptional regulation such as chromosomal

loop formations that bypass linear genome ona kilobase (or megabase) scale and topological-ly associated domains (TADs) (3) that assemblein nested fashion across hundreds of kilobases(4ndash7) By linking noncoding schizophrenia-associated genetic variants with distal genetargets 3DGmapping withHi-C (3 8) and othergenome-scale approaches could inform howhigher-order chromatin organization affects ge-netic risk for psychiatric disease To date only avery limitednumber ofHi-C datasets exist for thehuman brain two generated from bulk tissue ofdeveloping forebrain structures (7) and adultbrain (9) and one from neural stem cells (10)Although such datasets have advanced our un-derstanding of the genetic risk architectureof psychiatric disease (7 11) 3DG mappingfrom postmortem tissue lacks cell typendashspecificresolution and may not capture higher-orderchromatin structures sensitive to the autolyticprocess (12) We monitored developmentallyregulated changes in chromosomal conforma-tions during the course of isogenic neuronaland glial differentiation describing large-scalepruning of chromosomal contacts during thetransition from neural progenitor cells (NPCs)to neurons Furthermore we uncovered an ex-panded 3DG risk space for schizophreniamdashwitha functional network of disease-relevant regulatorsof neuronal connectivity synaptic signaling andchromatin remodelingmdashand demonstrate neural

cell typendashspecific coordination at the level of thechromosomal connectome transcriptome andproteome

ResultsNeural progenitor differentiationis associated with dynamic3DG remodeling

We applied in situ Hi-C to map the 3DG of twomale human induced pluripotent stem cell(hiPSC)ndashderived neural progenitor cells (NPCs)(13) together with isogenic populations of in-duced excitatory neurons (ldquoneuronrdquo) generatedthrough viral overexpression of the transcrip-tion factor NGN2 (14) and differentiations ofastrocyte-like glial cells (ldquogliardquo) (Fig 1 A andB and table S1) (15) Transcriptome RNA se-quencing (RNA-seq) comparison with publisheddatasets (16) confirmed that the NPCs but notglia from subjects S1 and S2 clustered togetherwith NPCs from independent donors whereas S1and S2 NGN2 neurons closely aligned with di-rected differentiation forebrain neurons (17) andprenatal brain datasets (fig S1 A and B) As withour transcriptomic datasets hierarchical cluster-ing of our Hi-C datasets after initial processing(fig S2A) also showed clear separation by cell type(Fig 1A and fig S2B) Genome-scale interactionmatrices were enriched for intrachromosomalconformations (fig S2C) with the exception ofthe negative control (ldquoNoLigaserdquo) NPC library inwhich we omitted the ligase step (Materials andmethods) and observed an interaction map withno signal due to the loss of chimeric fragments (figS2D)Given the observed correlationbetween tech-nical replicates of Hi-C assays from the same do-nor and cell type and the correlation between celltypendashspecific Hi-C from the two donors (Pearsoncorrelation of PC1 Rtechnical replicates range = 0970to 0979 Rsubject1-subject 2 by cell type range = 0962 to0970) we pooled by cell type for subsequentanalyses (fig S2E)We first focused on intrachromosomal loop

formations which are conservatively defined asdistinct contacts between two loci in the absenceof similar interactions in the surrounding se-quences (3) Our comparative analyses includedpublished (3) in situ Hi-C data from the Blymphocytendashderived cell line GM12878 (tableS1) When analyzed with the HiCCUPS pipeline(5- and 10-kb loop resolutions combined sub-sampled to 372 million valid-intrachromosomalread pairs to reflect the library with the fewestreads after filtration) (3) 17767 distinct loopswere called n = 3118 (175) were shared amongall four cell types whereas n = 5068 (285) werespecific to only one of the four cell types (Fig 1C)Biologically relevant terms such as ldquocentral ner-vous system developmentrdquo ldquoforebrain develop-mentrdquo and ldquoneuron differentiationrdquowere amongthe top gene ontology (GO) enrichments fromgenes overlapping loops shared between NPCsglia and neurons (brain-specific) but not iden-tified in lymphocytes (Fig 1D and table S2) in-dicating strong tissue-specific loop signaturesthat were also confirmed in individual cell types(fig S3A and tables S3 to S6)

RESEARCH | PSYCHENCODE

Rajarajan et al Science 362 eaat4311 (2018) 14 December 2018 1 of 11

1Icahn School of Medicine MDPhD Program Icahn Schoolof Medicine at Mount Sinai New York NY 10027 USA2Department of Psychiatry Icahn School of Medicine at MountSinai New York NY 10027 USA 3Department of Geneticsand Genomics Icahn School of Medicine at Mount SinaiNew York NY 10027 USA 4Friedman Brain InstituteIcahn School of Medicine at Mount Sinai New York NY10027 USA 5Program in Bioinformatics and IntegrativeBiology University of Massachusetts Medical SchoolWorcester MA 01605 USA 6New York Genome CenterNew York NY 10013 USA 7Department of NeuroscienceIcahn School of Medicine at Mount Sinai New York NY10027 USA 8Neurogenetics Program Department ofNeurology Center for Autism Research and TreatmentSemel Institute David Geffen School of Medicine Universityof California Los Angeles Los Angeles CA 90095 USA9Neuropsychiatric Signaling Program Department ofPsychiatry Perelman School of Medicine University ofPennsylvania Philadelphia PA 19104 USAThese authors contributed equally to this work daggerPresent addressDepartment of Genetics University of North Carolina ChapelHill NC 27599 USA DaggerPresent address UNC NeuroscienceCenter University of North Carolina Chapel Hill NC 27599USA sectThese authors contributed equally to this workparaCorresponding author Email schahramakbarianmssmedu

on February 26 2020

httpsciencesciencem

agorgD

ownloaded from

Rajarajan et al Science 362 eaat4311 (2018) 14 December 2018 2 of 11

Fig 1 Neural differentiation is associated with large-scale remodelingof the 3D genome (A) (Top) Derivation scheme of isogenic cell types fromtwo male control cell lines Pink oval donor hiPSC orange NPC greenneuron purple glia (Bottom) Hierarchical clustering of intrachromosomalinteractions (Materials and methods) from six in situ Hi-C libraries a and bare technical replicates of the same library height corresponds to thedistance between libraries (Materials and methods) (fig S2B) (B) Immuno-fluorescent staining of characteristic cell markers for NPCs (Nestin andSOX2) neurons (TUJ1 and MAP2) and glia (Vimentin and S100b) (C) Venndiagram of loop calls specific to and shared by different subsets of cellsincluding previously published GM12878 lymphoblastoid Hi-C data (D) Geneontology (GO) enrichment (significant terms only) of genes overlappinganchors of loops shared by NPCs neurons and glia but absent in GM12878(E) (Left) Cell-type pooled whole-genome heatmaps at 500-kb resolution (fig

S2C) (Right) ldquoArc maprdquo showing intrachromosomal interactions at 40-kbresolution of the q-arm of chr17 for isogenic neurons NPCs and glia asindicated from subject 2 RNA-seq tracks for each cell type shown on topof arc maps Green neuron orange NPC purple glia (F) FPKM geneexpression of CUX2 across three cell types with heatmap zoomed in on CUX2loop (black arrow) (fig S3) (G) Number of loops specific to each cell type(not shared with other cell types) with one anchor in an A compartmentand another in a B compartment (pink) both in B compartments (red)or both in A compartments (blue) (H) (Left) Box-and-whisker distributionplot of TAD size across four cell types (Right) Median TAD length for each ofthe four cell types (I) Heatmaps at 40-kb resolution for a 3-Mb windowat the CDH2 locus on chr18 (Bottom) Nested TAD landscape in glia withmultiple subTADs (black arrows) called which (top) is absent from neuronalHi-C RNA-seq tracks green neuron purple glia (figs S1 to S5)

RESEARCH | RESEARCH ARTICLE | PSYCHENCODEon F

ebruary 26 2020

httpsciencesciencemagorg

Dow

nloaded from

Rajarajan et al Science 362 eaat4311 (2018) 14 December 2018 3 of 11

Fig 2 Cell typendashspecificchromosomal contactmaps at schizophreniarisk loci (A) Juiceboxobservedexpected inter-action heatmaps at 10-kbresolution for the risk-associated clustered PCDHlocus chr5140023665minus140222664 for NPC gliaand neurons as indicated(Far right) Grayscaleheatmap depicts areas ofhighly cell-specific con-tact enrichments up-stream genes includingANKHD1 (dotted rectangleldquoArdquo and arrowhead) anddownstream PCDH geneclusters (dotted rectangleldquoBrdquo and arrows) Clus-tered PCDH gene expres-sion patterns are availablein fig S6A (B) Violin plotsof observedexpectedinteraction values in theregions A and B high-lighted in (A) (C) Map ofcontacts identified bybinomial statistics Redbox with dashed blackline represents theschizophrenia risk locusdotted boxes regionsldquoArdquo and ldquoBrdquo in heatmaps(D) Cell-type resolvedcontact map of 10-kb bins(bold black vertical lines)within risk sequenceson chr12 (left) chrX(middle) and chr5 (right)NPC (orange) neuron(green) glia (purple) ndashlogq value significance ofcontact betweenschizophrenia risk locusand each 10-kb bin genemodels (ldquoGenesrdquo) belowwith SNP-loop targetgene highlighted in red(E) Epigenomic editing(CRISPRa with nuclease-deficient dCas9 in NPCs)for three risk SNP-targetgene pairs and their re-spective control sequences(top) measured withquantitative reversetranscription polymerasechain reaction (RT-PCR)(fold change from baseline) for VP64 (middle) and VPR (bottom) transcriptional activators (F) Quantitative PCR gene expression changes upondirecting catalytically active Cas9 to schizophrenia risk-associated credible SNPs (vertical red dashes with rsIDs) interacting via chromosomalcontacts with promoters of ASCL1 EFNB1 and MATR3 in NPCs Targeting strategy and contact distances depicted above P lt 005 P lt 001P lt 00001 (figs S6 and S7)

RESEARCH | RESEARCH ARTICLE | PSYCHENCODEon F

ebruary 26 2020

httpsciencesciencemagorg

Dow

nloaded from

Rajarajan et al Science 362 eaat4311 (2018) 14 December 2018 4 of 11

Fig 3 Expanded GWAS risk connectome is associated with genecoregulation (A) Counts of highly cell typendashspecific contacts asso-ciated with schizophrenia risk in each of the three hiPSC-derived celltypes (B) GO enrichment of genes located in schizophrenia riskcontacts in NPCs (left) neurons (middle) and glia (right) (C) (Left)Schematic workflow of analyses performed with cell typendashspecificcontact genes distinguished as ldquorisk locusrdquo and ldquorisk locusndashconnectrdquogenes (Middle) Venn diagram of genes located in the 145 loci and thosefound in cell typendashspecific contacts with numbers in blue indicatingldquorisk locusndashconnectrdquo genes (Right) Schematic workflow of analysesperformed with combined set of ldquorisk locusrdquo and ldquorisk locusndashconnectrdquo

genes (D) RNA Pearson transcriptomic correlation heatmaps consisting of risk locus and risk locusndashconnect genes derived from the cell typendashspecific contacts ofNPCs (left) neurons (middle) and glia (right) Organization scores (|r|avg) for each heatmap are reported with P values from sampling analysis Schematicsabove heatmaps are representations of each cell typersquos particular connectome (blue oval) and frequency distribution of organization scores from permutationanalyses of randomly generated heatmaps (red observed organization score of heatmap being tested) The gray bar corresponds to n genes that have at least1 count per million in RNA-seq dataset out of the total number of genes and are used to construct the heatmap red and blue bars indicate how many of thegenes in the heatmap are in a risk locus (red) and are risk locusndashconnect (blue) Fuchsia neuron connectivitysynaptic function genes yellow chromatin remodelinggenes as determined from gene ontology analysis in (E) Additional information on coexpression clusters is provided in tables S22 and S23 (figs S8 and S9)

RESEARCH | RESEARCH ARTICLE | PSYCHENCODEon F

ebruary 26 2020

httpsciencesciencemagorg

Dow

nloaded from

Unexpectedly there was a reduction (~40 to50 decrease) in the total number of chromo-somal loops in neurons relative to isogenic gliaand NPCs (fig S3 B and C) Reduced densities ofchromosomal conformations were also evidentin genome browser visualization of chromo-somal arms including chr17q (Fig 1E) Althoughboth glia and NPCs harbored ~13000 loop for-mations only 7206 were identified in neurons(Fig 1C fig S3 B and C and table S1) including442 neuron-specific loop formations One such

neuron-specific loop was at CUX2 a transcrip-tion factor whose expression marks a subset ofcortical projection neurons (18) and that is highlyexpressed in ourNGN2-induced neurons (Fig 1Fand fig S3 D and E) Examples of loops lost inneurons include one spanning the Ca2+ channeland dystonia-risk gene ANO3 (fig S3F) (19)Furthermore NPCs neurons and glia had similarproportions of loops anchored in solely active (A)compartments solely inactive (B) compartmentsor in both indicating no preferential loss of either

active or inactive loops in neurons (Fig 1G)However among the genes overlapping an-chors of loops that underwent pruning during thecourse of the NPC-to-neuron transition regula-tors of cell proliferation morphogenesis andneurogenesis ranked prominently in the top 25GO termswith significant enrichment (Benjamini-Hochberg corrected P lt 10minus6 ndash 10minus12) (fig S3Gand table S4B) which is consistent with a depar-ture from precursor stage toward postmitoticneuronal identity (20) Likewise loops lost during

Rajarajan et al Science 362 eaat4311 (2018) 14 December 2018 5 of 11

Fig 4 Expanded GWAS risk connectome is linked to protein-proteinassociation networks (A) Overview and representative examples (zoomedin) of protein-protein association networks in NPCs (left) neurons (middle)and glia (right) Numbers of edges connecting the proteins in each networkand STRING-computed P values are reported below Gray bar indicates thesubset of these geneswhose proteins are involved in the network out of the totalnumber of genes from cell typendashspecific interactions red and blue barsindicate how many of the genes in the network are in a risk locus (red) andare risk locusndashconnect (blue) (B) Comparison of organization scores betweenthe full RNA transcriptomic correlation heatmaps (brown) (Fig 3D) and theldquoSTRINGrdquo heatmaps (tan) (figs S13 to S15) consisting of only those genes inprotein networks foreach cell type Permutation test Plt001 (C)Representative

neuronal TAD landscape (chr1 ~2 Mb) depicting a schizophrenia riskndashassociatedlocus (red) with its risk locusndashconnect genes (blue)MED8MPL CDC20 andRNF220 which are members of the neuronal schizophrenia protein network(green circle) CDC20 and RNF220 interact at the protein level (green circlewith gray border) (D) (Left) Liquid chromotographyndashselected reactionmonitoring (LC-SRM)mass spectrometry (MS)was performed on dorsolateralprefrontal cortex (DLPFC) tissue from43adultpostmortembrains (23schizophrenia20 control) (Middle) 182 neuronal proteins were reliably quantified and four ofthemwereobserved tohaveassociations in theneuronproteinnetwork in (A) (Right)GABBR1GRM3GRIN2A and GRIA1 proteins were found to have significantlymore correlated expression than expected by random permutation analysisAdditional information on protein-protein interactions is provided in figs S9 to S15

RESEARCH | RESEARCH ARTICLE | PSYCHENCODEon F

ebruary 26 2020

httpsciencesciencemagorg

Dow

nloaded from

NPC-to-glia transition were significantly enriched(Benjamini-Hochberg corrected Plt 10minus3 ndash 10minus6) forneuron-specific functions including ldquotransmissionacross chemical synapserdquo ldquog-aminobutyric acid(GABA) receptor activationrdquo and ldquopostsynapserdquo (figS3G and table S4C) which is consistent withnon-neuronal lineage commitmentWe defined ldquoloop genesrdquo as genes that either

have gene body or transcription start site (TSS)overlap with a loop anchor (5- or 10-kb binsforming the points of contact in a chromatin loop)Genes with loop-bound gene bodies (one-tailedZ test Zrange = 421 to 592 P lt 10minus324 for all) orloop-bound TSS (one-tail Z-test Zrange = 152 to288 Prange lt 232 times 10minus52 to 440 times 10minus182) bothshowed significantly greater expression [meanlog10(FPKM + 1) FPKM fragments per kilobaseof exon permillion fragmentsmapped] than thatof background (all genes for all brain cell types)(fig S4A) suggesting that looping architec-ture was associated with increased gene expres-sion Furthermore 3 of loops shared by NPCsneurons and glia (brain-specific loops) inter-connected a brain expression quantitative traitlocus (eQTL) single-nucleotide polymorphism(SNP) with its destined target gene(s) repre-senting significant enrichment over backgroundas determined with 1000 randomdistance- andfunctional annotationndashmatched loop samplings(random sampling one-sided empirical P = 0012)(Materials and methods) (fig S4B)We aimed to confirm that the observed net

loss of loop formations during the NPC-to-neuron transition could be replicated across avariety of independent cell culture and in vivoapproaches and was not specific to our meth-odological choice of NGN2-induction We con-ducted an additional Hi-C experiment on cellsdifferentiated from hiPSC-NPCs by means of anon-NGN2 protocol that used only differentia-tionmedium and yielded a heterogeneous pop-ulation of hiPSC-forebrain-neurons in additionto a small subset of glia (17) In addition we re-analyzed Hi-C datasets generated from a mousemodel of neural differentiation consisting ofmouse embryonic stem cell (mESCs) mESC-derived NPCs (mNPC) and cortical neurons(mCN) differentiated from the mNPCs via in-hibition of the Sonic Hedgehog (SHH) pathway(21) To examine whether such genome-widechromosomal loop remodeling also occurred inthe developing brain in vivo we reanalyzed Hi-Cdata from human fetal cortical plate (CP) mostlycomposed of young neurons and forebrain ger-minal zone (GZ) primarily harboring dividingneural precursor cells in addition to a smallersubset of newly generated neurons (7) Acrossboth the hiPSC-NPC-to-forebrain neuron andmESC-mNPC-mCN differentiation in vitro neu-rons showed a 20 decrease in loops comparedwith their neural progenitors (fig S4 C and D)Consistent with this in vivo CP (neuron) com-paredwithGZ (progenitor) showeda 13decreasein loops genome-wide (fig S4E) The highly rep-licative cell types included here mouse ESCs andhuman lymphoblastoid GM12878 cells exhibitedloop numbers very similar to their neuronal

counterparts (fig S4 D and E) suggesting thatthe changes in 3DG architecture from NPC toneurons do not simply reflect a generalized effectexplained by mitotic potentialAlong with having fewer total loops neurons

exhibited a greater proportion of longer-range(gt100 kb) loops than did NPCs or glia (two-sample two-tailed Kolmogorov-Smirnov testKSrange = 01269 to 02317 P lt 22 times 10minus16 forthree comparisons Neu versus NPCGliaGM)(fig S5A) Likewise in each of the alternativein vitro and in vivo analyses considered aboveneurons exhibited a greater proportion of longer-range (gt100 kb) loops than didNPCs or glia [two-sample two-tailed Kolmogorov-Smirnov testKS = 00427 P = 15 times 10minus5 for hiPSC-NPC versusforebrain neuron KS = 00936 P = 11 times 10minus16 formESC-NPC versus mCN KS = 00663 P = 204 times10minus8 for fetal CP (neuron) compared with GZ(progenitor)] (fig S5 B C D and E) Thereforemultiple in vitro and in vivo approaches com-paring in human and mouse neural precursorsto young neurons consistently show a reducednumber of loops in neuron-enriched culturesand tissues primarily affecting shorter-rangeloopsConsistent with studies in peripheral tissues

reporting conservation of the overall loop-independent TAD landscape across developmen-tal stages tissues and species (when consideringsyntenic loci) (10 22) overall TAD landscapes (3)remained similar between neurons glia andNPCs Nonetheless TADs also showed a subtle(~10) increase in average size in neuronscompared with isogenic NPCs independent ofthe differentiation protocol applied (Wilcoxon-Mann-Whitney test P lt 53 times 10minus6) (Fig 1Hand fig S5 F and G) as highlighted here at a34-Mb TAD at the CDH2 cell adhesion genelocus (Fig 1I) TAD remodeling may thereforereflect restructuring of nested subdomains with-in larger neuronal TADs (tables S7 and S8) Toexamine whether such developmental reorgan-ization of the brainrsquos spatial genomes was as-sociated with a generalized shift in chromatinstructure we applied the assay for transposaseaccessible chromatin with high-throughput se-quencing (ATAC-seq) to map open chromatinsequences before and after NGN2-neuronalinduction (table S1) Genome-wide distributionprofiles for transposase-accessible chromatinwere only minimally different between NPCsand neurons (fig S5H) and further revealed thatboth NPCs and neurons showed low tomoderatechromatin accessibility [ndash25 lt log2(ATAC signal)lt 1] for ge89 of the anchor sequences compris-ing cell typendashspecific and shared ldquobrainrdquo loops inour cell culture system (fig S5I) These findingstaken together point towidespread 3DG changesduring the NPC-to-neuron transition and NPC-to-glia transition in human andmouse brain thatare unlikely attributable to global chromatinaccessibility differences This includes highlycell typendashspecific signatures in gene ontologiesof differentiation-induced loop prunings reflect-ing neuronal and glial (non-neuronal) lineagecommitment (fig S3 A and G and table S4 B

and C) and a subtle widening of average loopand TAD length in young neurons (Fig 1H andfig S5 A to G)

Chromosomal contacts associatedwith schizophrenia risk sequences

Because many schizophrenia risk variants lie innoncoding regions in proximity to several geneswe predicted that chromosomal contact map-ping could resolve putative regulatory elementscapable of conferring schizophrenia risk via theirphysical proximity (bypassing linear genome) tothe target gene as has been demonstrated intissue in vivo (7 11) We overlaid our cell typendashspecific interactions onto the 145 risk loci as-sociated with schizophrenia risk (2 23) Becauseonly very few loops (defined as distinct pixelswith greater contact frequency than neighboringpixels on a contact map) (3) were associated withschizophrenia risk loci (n = 212 81 and 17 loopsin NPC glia and neurons respectively) (tableS9) we applied an established alternative ap-proach to more comprehensively explore the3DG in context of disease-relevant sequences(7) This approach defines interactions as thosefiltered contacts that stand out over the globalbackground and applies binomial statisticsto identify chromosomal contacts anchored atdisease-relevant loci (7) To begin we examinedthe 40 loci with strongest statistical evidence forcolocalization of an adult postmortem braineQTL and schizophrenia genome-wide associa-tion study (GWAS) signal (24) Chromosomalcontacts were called for 29 of the 46 eQTLs pres-ent in the 40 loci with 8 of 29 (28) of the locishowing significant interactions (binomial testndashlog q value range = 133 to 110) between theeQTL-SNPs (eSNPs) in the one contact anchorand the transcription start site of the associatedgene(s) in the other anchor (table S10) We con-clude that ~30 of risk locusndashassociated eQTLswith strong evidence for colocalization withGWAS signal bypass the linear genome and arein physical proximity to the proximal promoterand transcription start site of the target generesonating with previous findings in fetal braintissue that used a similar contact mappingstrategy (7)Cell typendashspecific contact maps with 10-kb-wide

bins queried for the schizophrenia-associatedloci frequently revealed differential chromo-somal conformations in NPCs glia and neu-rons For example the risk locus upstream ofthe PROTOCADHERIN cell adhesion moleculegene clusters (chromosome 5) which is criticallyrelevant for neuronal connectivity in develop-ing and adult brain (fig S6A) (25 26) showedthrough both observedexpected interactionmatrix (27) and global background-filtered con-tact mapping (7) a bifurcated bundle of inter-actions in NPCs with one bundle emanating tosequences 5prime and the other bundle to sequences3prime from the locus In neurons the 3prime bundle wasmaintained but the 5prime bundle was ldquoprunedrdquowhereas glia showed the opposite pattern thesedifferences between the three cell types werehighly significant (observedexpected Wilcoxon

Rajarajan et al Science 362 eaat4311 (2018) 14 December 2018 6 of 11

RESEARCH | RESEARCH ARTICLE | PSYCHENCODEon F

ebruary 26 2020

httpsciencesciencemagorg

Dow

nloaded from

rank sum P lt 10minus9 to 10minus15) (Fig 2 A to C) Dos-age of the noncoding schizophrenia risk-SNP(rs111896713) at the PCDH locus significantlyincreased the expression of multiple PROTO-CADHERIN genes (PCDHA2 PCDHA4 PCDHA7PCDHA8 PCDHA9 PCDHA10 and PCDHA13) inadult frontal cortex of a large cohort of 579 in-dividuals including cases with schizophreniaand controls (fig S6B and table S11) (28) The af-fected genes were interconnected to the disease-relevant noncoding sequence in neurons andNPCs but not in glia (fig S6C) Therefore celltypendashspecific Hi-C identified chromosomal con-tacts anchored in schizophrenia-associated risksequences that affected expression of the targetgene(s) On the basis of earlier chromosome con-formation capture assays at the site of candidategenes the underlying mechanisms may includealterations in transcription factor and other nu-cleoprotein binding at loop-bound cis-regulatoryelements (5) or even local disruption of chromo-somal conformations (6)Transcriptional profiles of hiPSC-derived NPCs

and neuronsmost closely resemble those of thehuman fetus in the first trimester (29) more-over a portion of the genetic risk architectureof schizophrenia matches to regulatory elementsthat are highly active during prenatal develop-ment (30) We surveyed in our Hi-C datasetsseven loci encompassing 36 ldquocrediblerdquo (potentiallycausal) schizophrenia-risk SNPs with knownchromosomal interactions in fetal brain to genesimportant for neuron development and function(7) We found that risk-associated chromosomalcontacts were conserved between our hiPSC-NPCs and the published human fetal CP andgerminal zone Hi-C datasets (7) for five of theseven loci (71) tested (CHRNA2EFNB1MATR3PCDH and SOX2 but not ASCL1 or DRD2) (tableS12) To test the regulatory function of theseconserved risk sequence-bound conformationswe performed single-guide RNA (sgRNA)ndashbasedepigenomic editing experiments on isogenicantibiotic-selected NPCs that stably expressnuclease-deficient dCas9-VP64 (31 32) or dCas9-VPR (33 34) transactivators (table S13) Previousstudies in peripheral cell lines succeeded ininducing gene expression changes by placingdCas9-repressor fusion proteins at the site ofchromosomal contacts separated by up to 2Mbof linear genome from the promoter target (35)We tested ASCL1- EFNB1- MATR-3 and SOX2-bound chromosomal contacts separated by 200-to 700-kb interspersed sequences (Fig 2 D andE fig S7A and table S14) Pools of five individualsgRNAs directed against a risk-associated non-coding sequence bypassing 225 and 355 kb ofgenome consistently resulted in significantly de-creased expression of ASCL1 [one-way analysis ofvariance (ANOVA) FVP64(2 15) = 2220 P lt 00001Dunnettrsquos PVP64 = 0023] and EFNB1 target genes[one-way ANOVA FVP64(2 6) = 1447 P = 00051Dunnettrsquos PVP64 = 00356 FVPR(2 6) = 146 P =00111 Dunnettrsquos PVPR = 00088] in comparisonwith positive (promoter-bound) and negative(linear genome) control sgRNAs Epigenomicediting of risk sequence 500 to 600 kb distant

from the SOX2 and MATR3 loci did not altertarget gene expression (Fig 2 D and E andfig S7 A and B) which could reflect practicallimitations in nonintegrative transfection-based (as opposed to viral) methods impact ofepigenetic landscape or suboptimal guide RNApositioning (34) further limited by the 10-kbcontact map resolution Because portions of theMATR3-bound risk sequences are embedded inrepressive chromatin we directed five sgRNAsfor Cas9 nuclease mutagenesis toward a 138ndashbasepair (bp) sequence within a MATR3 long-rangecontact that was enriched with trimethyl-histoneH3K27me3 commonly associatedwithPolycombrepressive chromatin remodeling in order to dis-rupt it (fig S7 C to E) This strategy produced asignificant increase in MATR3 expression uponablation of the putative repressor sequencewhereas targeting MATR3 (linear genome) con-trol sequence remained ineffective (fig S7 D andE) We conducted additional genomic mutagen-esis assays with sgRNAs directly overlappingcredible SNPs participating in chromatin con-tacts with ASCL1 EFNB1 EP300 MATR3PCDHA7 PCDHA8 and PCDHA10 (table S10)Cas9 nuclease deletion of interacting credibleSNPs significantly increased gene expressionof ASCL1 EFNB1 and EP300 (Prange = 00053 to004 trange = 2449 to 4265) (Fig 2F and figS7F) Similar targeting of four credible SNPsupstream of the clustered PCDH locus signifi-cantly decreased levels by ~50 to 60 ofPCDHA8and PCDHA10 (Prange = 00122 to 00124 trange =4326 to 4343) two of the genes whose expressionincreasedwith dosage of the risk SNP rs111896713in adult postmortem brain (figs S6C and S7G)Taken together our (epi)genomic editing assays(fig S7H) demonstrate that chromosomal contactsanchored in schizophrenia risk loci potentiallyaffect target gene expression across hundredsof kilobases which is consistent with predic-tions from chromosomal conformation mapsfrom hiPSC-derived brain cells described hereand from developing (7 11) and adult (5) humanbrain tissue

Cell typendashspecific schizophrenia-relatedchromosomal connectomes areassociated with gene co-regulation andprotein-protein association networks

Having shown that the chromosomal contactmaps anchored in sequences associated withschizophrenia heritability undergo cell typendashspecific regulation (Fig 2 A to C) are reprodu-cible in neural cell culture and fetal brain (tableS12) frequently harbor risk-associated eQTLs(table S10) and bypass extensive stretches oflinear genome to affect target gene expressionin genomic and epigenomic editing assays (Fig 2D to F and fig S7) we investigated chromosomalcontacts for all 145 GWAS-defined schizophreniarisk loci together (23) (tables S15 to S17)We referto the resulting ldquonetworkrdquo of risk loci and their3D proximal genes as the ldquoschizophrenia-relatedchromosomal connectomerdquoEarlier studies in adult brain had shown that

open chromatin-associated histone modifica-

tion and other ldquolinear epigenomerdquo mappingsstrongly link the genetic risk architecture ofschizophrenia specifically with neuronal as op-posed to non-neuronal chromatin (36) whichwould suggest that similar cell-specific signa-tures may emerge in the risk-associated 3DGNeurons and NPCs but not the isogenic gliashowed a high preponderance of chromosomalcontacts with schizophrenia-associated risk loci(Fig 3A) There were 1203 contacts involvingschizophrenia risk sequences that were highlyspecific to neurons (median distance betweenrisk and target bins = 510 kb) 1100 highly specificfor NPCs (median distance between risk andtarget bins = 520 kb) whereas only 425 highlyspecific for glia (median distance between riskand target bins = 580 kb) (Fig 3A figs S8 A andB and tables S15 to S17) There were also un-expectedly robust cell typendash and gene ontologyndashspecific signatures including genes associatedwith neuronal connectivity and synaptic signal-ing (Fig 3B and tables S18 and S19) Separateanalysis of the Psychiatric Genomics ConsortiumldquoPGC2rdquo 108 risk loci (2) yielded similar results(fig S9 A and B)Because spatial 3DG proximity of genes is

an indicator for potential coregulation (37)we tested whether the neural cell typendashspecificschizophrenia-related chromosomal connectomeshowed evidence of coordinated transcriptionalregulation and proteomic interaction of the par-ticipating genes To this end we generated listsof genes anchored in the most highly cell typendashspecific schizophrenia riskndashassociated contacts(Materials and methods) (Fig 3C fig S8B andtable S18) Thus for the NPC-specific contactswe counted 386 genes including 146 within therisk loci and another 240 genes positioned else-where in the linear genome but connected viaan intrachromosomal contact to within-risk-locussequences Similarly for the neuron-specific con-tacts we identified 385 genes including 158within risk loci and 227 outside of risk loci (Fig3C) Last for glia-specific contacts we identified201 genes including 88 within and 113 outsideof risk loci We labeled the intrachromosomalcontact genes located outside of schizophreniarisk loci as ldquorisk locus-connectrdquo which we defineas a collection of genes identified only throughHi-C interaction data expandingmdashdepending oncell typemdashby 50 to 150 the current network ofknown genes overlapping risk sequences that isinformed only by GWAS (Fig 3C)To examine whether such types of disease-

associated cell typendashspecific chromosomal con-nectomes were linked to a coordinated programof gene expression we analyzed a merged tran-scriptome dataset (comprised of 47 hiPSC-NPCand 47 hiPSC-forebrain-neuron RNA-seq libra-ries from 22 schizophrenia and control donorsnot related to the those of our Hi-C datasets)(16) We examined pair-wise correlations of thecollective sets of the 386 NPC 385 neuron and201 glia genes representing ldquorisk locusrdquo and ldquorisklocusndashconnectrdquo genes (cell typendashspecific ldquoriskconnectomesrdquo) The risk connectome for eachcell type showed extremely strong pair-wise

Rajarajan et al Science 362 eaat4311 (2018) 14 December 2018 7 of 11

RESEARCH | RESEARCH ARTICLE | PSYCHENCODEon F

ebruary 26 2020

httpsciencesciencemagorg

Dow

nloaded from

correlations with two of the largest clustersvisualized on the neuron and NPC correlationmatrices involving an admixture of 354 ldquorisklocusrdquo and ldquorisk locusndashconnectrdquo genes each andsimilarly 181 genes from the glia matrix (Fig 3Dand table S20) The averaged gene-by-gene trans-cript correlation index for each matrix overalldefined here as ldquoorganization scorerdquo (|r|avg) forthe NPCs neurons and glia were 022 to 025Such levels of organized gene expression wererobustly significant for NPC and neurons aftercontrolling for linear genomic distance (1000random samplings |r|avg P lt 0001 for NPCand for neuron P = 0041 for glia) (Fig 3D figS9E and table S21) There were four large clus-ters in the correlation matrices of the neuronalandNPC risk connectome neuronal connectivityand synaptic signaling proteins (neuron cluster1 and NPC cluster 2) and epigenetic regulators(neuron cluster 2 and NPC cluster 1) For exam-ple within neuron cluster 1 (Fig 3D middle) 62of 125 genes encoded neural cell adhesion andsynaptic molecules voltage-gated ion channelsand other neuron-specific genes (Fig 3E andtables S22 and S23) We thus conclude that thechromosomal connectomes associated with schi-zophrenia risk are cell typendashspecific with theneuronal risk connectome particularly enrichedfor genes pertaining to neuronal connectivitysynaptic signaling and chromatin remodeling(Fig 3 D and E) Analyses of the subset of PGC2risk loci (108 and 145) provided similar results (figS9 C to F) Additionally organization scores forneuron cluster 1 and cluster 2 genes were sim-ilar between hiPSC-derived NPCs and forebrainneurons from schizophrenia cases (n = 47) andcontrol (n = 47) suggesting thatmany risk locusndashconnect and risk locus genes are coregulatedacross individuals (fig S9H)Numerous proteins encoded by risk locus and

risk locusndashconnect genes were associated withsynaptic signaling (table S24) The cell typendashspecific risk locusndashconnect and risk locus genesshow significant protein-protein interactionnetwork effects for NPCs (P = 00004) andneurons (P = 0009) but not glia (Fig 4A figsS10 to S12 and table S24) when examined byusing the STRING database v105 (38 39) Weobserved many proteomic clusters includinglarge groups of epigenomic regulators associ-ated with the SWISNF (SWItchSucrose Non-Fermentable) chromatin remodeling complexand histone lysinemethyltransferases and deme-thylases (Fig 4A and figs S10 and S11) many ofwhich were the genes identified in NPC cluster1 and neuron cluster 2 of the transcriptomeanalysis (Fig 3 D and E) The transcriptomiccorrelation heatmaps for these protein net-works (ldquoSTRINGrdquo genes) when compared withrandomly generated subset heatmaps from theoverall (ldquoFullrdquo) schizophrenia-related chromo-somal connectome (Fig 3D) had higher organi-zation scores in NPCs and neurons (NPC |r|avg =02963 P = 0007 neuron |r|avg = 02877 P =0008 glia |r|avg = 02225 P = 0595 STRINGversus full permutation test) (Materials andmethods) (Fig 4B figs S13 to S15 and table

S21) Because the transcriptomic correlationheatmap for the schizophrenia-related chromo-somal connectome was significantly decreasedby the removal specifically of the NPC STRINGprotein network genes (P lt 10minus3) (table S24) thissubset of STRING-interacting proteinsmay drivethe observed orchestrated coregulation Withinthese transcriptome- and proteome-based reg-ulatory networks were numerous occasions ofcoregulated (RNA) and interacting (protein)risk locus and risk locusndashconnect genes thatshare the same TAD including CDC20 whichregulates dendrite development (40 41) and isassociated at the protein level with RNF220 anE3 ubiquitin-ligase and b-catenin stabilizer(Fig 4C) (42)To examine whether such coregulation could

be representative of the prefrontal cortex pro-teome of the adult brain we screened a newlygenerated mass spectrometryndashbased dataset of182 neuronal proteins the majority of whichwere synaptic quantified from prefrontal cortexof n = 23 adult schizophrenia and n = 20 controlsubjects (table S25) (43) Among the 182 proteinsthere were four from the risk-associated neuro-nal protein network (Fig 4D) GABAB receptorsubunit GABBR1 and ionotropic (GRIA1 andGRIN2A) and metabotropic glutamate receptorsubunits (GRM3) Protein-protein correlationscores were significantly higher for these fourrisk-associated proteins than expected fromrandom permutation analysis from the pool of182 proteins (P lt 0002) across patients and con-trolsWe conclude that the schizophrenia-relatedchromosomal connectome tethering other por-tions of the genome to the sequences associatedwith schizophrenia heritability provides a struc-tural foundation for a functional connectomethat reflects coordinated regulation of gene ex-pression and interactions within the proteome

Discussion

Neural progenitor differentiation into neuronsand glia is associated with dynamic remodelingof chromosomal conformations including lossof many NPC-specific chromosomal contactswith differentiation-induced loop pruning pri-marily affecting a subset of genes important forneurogenesis (NPC-to-neuron loss) and neuro-nal function (NPC-to-glia loss) These findingsbroadly resonate with a recent report linkingneural differentiation to multiple scales of 3DGfolding governed by multiple mechanisms in-cluding CTCF-dependent loop alterations re-pressive chromatin remodeling and cell- andlineage-specific transcription factor networks(21) Our results suggest that developmental3DG remodeling affects a substantial portionof sequences that confer liability for schizo-phrenia furthermore these genes in 3D phys-ical proximity with schizophrenia-risk variantsshow a surprisingly strong correlation at thelevel of the transcriptome and proteome Howmight the disease-relevant reorganization ofthe spatial genome (the ldquochromosomal con-nectomerdquo) provide a structural foundation forcoordinated regulation of expression Recent

Hi-C studies in mouse brain showed that chro-mosomal contacts preferentially occurred be-tween loci targeted by the same transcriptionfactors (21) and likewise multiple schizophreniarisk loci could converge on intra- and interchro-mosomal hubs sharing a similar regulatory ar-chitecture including specific enhancers as wellas transcription and splicing factors (44ndash46)Intriguingly the three major functional catego-ries associated with the genetic risk architec-ture of schizophreniamdashneuronal connectivitysynaptic signaling and chromatin remodeling(47 48)mdashwere heavily represented within thecell typendashspecific chromosomal connectomesof neurons and NPCs described here (Fig 3 Band E) and in whole tissue in vivo (7 11) Celltypendashspecific 3DG reorganization during thecourse of neural progenitor differentiation asshown here could therefore have profoundimplications for our understanding of the ge-netic underpinnings of psychiatric disease Forexample inclusion of the cell typendashspecific risk(sequence)ndashassociated chromosomal connectomemay lead to refinements of cumulative schizo-phrenia risk allele burden estimates includingldquopolygenic risk scorerdquo (PRS) or ldquobiologically in-formedmultilocus profile scoresrdquo (BIMPS) whichcurrently only explain a small portion of dis-ease risk (49) Cell typendashspecific intersection of3DG and genetic risk maps are of clinical inter-est beyond psychiatric disorders for examplerisk variants that confer susceptibility to auto-immune disease were embedded in physicallyinteracting chromosomal loci in lymphoblastoidcells (50) Our 3DGmaps fromneural progenitorsand their isogenic neurons and glia are accessiblethrough the PsychENCODE Knowledge Portal(httpssynapseorg) and more than double thenumber of currently availableHi-C datasets fromhuman brain (7 9 10) providing investigatorswith a resource to chart the expanded genomespace associated with cognitive and neuropsychi-atric disease in context of cell typendashspecific re-modeling of chromosomal conformations duringearly development

Materials and methodsIn situ Hi-C from hiPSC-derived cells

In situHi-C librarieswere generated from2millionto 5 million cultured hiPSC-derived NPCs gliaand neurons as described in (3) without mod-ifications in the protocol Briefly in situ Hi-Cconsists of 7 steps (i) crosslinking cells withformaldehyde (ii) digesting the DNA using a4-cutter restriction enzyme (eg MboI) withinintact permeabilized nuclei (iii) filling in andbiotinylating the resulting 5prime-overhangs (iv)ligating the blunt ends (v) shearing the DNA(vi) pulling down the biotinylated ligation junc-tions with streptavidin beads and (vii) analyz-ing these fragments using paired end sequencingAs quality control (QC) steps we checked forefficient restriction with an agarose DNA geland for appropriate size selection in using theAgilent Bioanalyzer after steps (v) and (vi) Forthe final QC we performed superficial sequenc-ing on the IlluminaMiSeq (~2-3M readssample)

Rajarajan et al Science 362 eaat4311 (2018) 14 December 2018 8 of 11

RESEARCH | RESEARCH ARTICLE | PSYCHENCODEon F

ebruary 26 2020

httpsciencesciencemagorg

Dow

nloaded from

to assess quality of the libraries using metricssuch as percent of reads passing filter percent ofchimeric reads and percent of forward-reversepairs (supplementarymaterials table S1) For theforebrain directed differentiation neuronal libra-ry from subject S1 the Arima Hi-C kit (ArimaGenomics San Diego) was used according to themanufacturerrsquos instructions

Hi-C read mapping and matrix generation

TheHi-C libraries were sequenced on the IlluminaHiSeq1000 platform (125bp paired-end) (NewYorkGenome Center) Technical replicates of subjectS2 NPCs neurons and glia were also sequencedto enhance resolution Initial processing of theraw 2 times125 bp read pair FASTQ files was per-formed using the HiC-Pro analysis pipeline (51)In brief HiC-Pro performs four major tasksaligning short reads filtering for valid pairsbinning and normalizing contact matrices HiC-Pro implements the truncation-based alignmentstrategy using Bowtie v223 (52) mapping fullreads end-to-end or the 5prime portion of reads pre-ceding a GATCGATC ligation site that resultsfrom restriction enzyme digestion with MboIfollowed by end ligation Invalid interactionssuch as same-strand dangling-end self-cycleand single-end pairs are not retained Binningwas performed in 10kb 40 kb and 100 kb non-overlapping adjacent windows across the ge-nome and resulting contact matrices werenormalized using iterative correction and eigen-vector decomposition (ICE) as previously de-scribed (53) using HiC-Pros default settings of100 maximum iterations filtering of the sparsebins (lowest 2) and a relative result incrementof 01 before declaring convergence (httpnservantgithubioHiC-ProMANUALhtml) Data are re-ported in browser-extensible-data-like (BED) for-mat and visualized in theWashington UniversityEpigenome Browser (httpepigenomegatewaywustledu) Hierarchical clustering was performedon the ICE-corrected intrachromosomal contactmatrices after the bins with the 1most extremeinteraction values were excluded as largely arti-factual Clustering was performed using Wardsmethod on the 1 5 10 25 50 and 100 mostvariable remaining bins using (1-correlation) as adistance metric The results using the 10 mostvariable interaction bins shown here in a clusterdendrogram and a Pearson correlation matrixare representative of these results

Hi-C loop calls using Juicer

Loop calling was performed using the softwareHiCCUPS (3) To format data forHiCCUPS inputwe remapped reads from Hi-C libraries usingthe Juicer pipeline (54) Similar to HiC-Pro theJuicer pipeline performs read alignment filter-ing binning and matrix normalization Sampleswere pooled for each cell type (S1 and 2 technicalreplicates from S2) to generate the maximumamount of coverage required for accurate loopcalling The resulting hic matrix files (MAPQ gt 0)were then used as input to HiCCUPS The fol-lowing parameters were set for HiCCUPS fol-lowing the analysis in (3) FDR threshold ( f ) =

010 010 peak width (p) = 4 2 window width(i ) = 7 5 merge distance (d) = 20 kb 20 kbValues for parameters correspond to calls madeat 5kb and 10kb respectively Representativeneuronal and non-neuronal loops are presentedin fig S3 As the number of loops called is de-pendent upon the number of Hi-C contacts inthe matrix (55) we also generated matrices withequivalent total Hi-C contacts via subsamplinghiPSC-derived Hi-C interaction matrices wererandomly subsampled to 372787143 cis only con-tacts (the lowest number of cis contacts acrossall cell types) and HiCCUPS was rerun on thesubsampled matrices After loops were calledfor each cell type we performed a reevaluationon this union set of loop loci HiCCUPS was re-run using the union set of loop loci as input toproduce q-values for each loop in the union setfor every cell type By default HiCCUPS does notoutput a q-value for every pixel Hence this re-evaluation produced q-values for pixels in cellsthat did not pass the significance threshold Wethen defined any pixel from the union set with aq-value lt 010 with respect to the donut neigh-borhood surrounding the pixel to be a loop anddefined the loop to be shared with any cell typeshaving a q-value lt 010 for the same pixelThese loop calls were used for comparing

loop calls between cell types Loops were alsocalled and subsampled as above for the GM12878cell line using the processed data from (3) foundhere wwwncbinlmnihgovgeoqueryacccgiacc=GSE63525 Loop calls were overlapped withcompartment calls (supplementary materialsmaterials and methods) such that AA BB andAB refer to loops with both anchors in A bothanchors in B and one anchor in A and otheranchor in B respectively Loops in chromosomes4 18 19 and X were removed from this com-partment analysis since the first principle com-ponent most likely corresponded to p versus qarm distinctions and not A versus B compartments

Hi-C interactions at risk loci

To approach 3DG conformation in context ofthe disease-relevant sequences we adapted thebinomial statistics based mapping strategypreviously described by Won et al (7) The setof schizophrenia risk loci used in this study in-cluded the original (PGC2 Psychiatric Ge-nomics Consortium) (2) risk sequences or 108physically distinct association loci defined by128 index SNPs (corrected P 10minus8) and an ad-ditional 37 loci from the CLOZUK (a series ofUK cases registered for clozapine treatment witha clinical diagnosis of schizophrenia) study for atotal of 145 loci defined by 179 independentgenome-wide significant SNPs (corrected P lt 5 times10ndash8) determined by GWAS in 40675 cases and64643 controls (23) A risk locus is defined asa collection of (SNPs) existing in linkage dis-equilibrium ranging from 1bp to 89Mb (aver-age 2562 kb) in length and in total equivalentto approximately 0012 of human genomicsequenceTo identify significantly enriched interac-

tions involving a bin of interest with another bin

our principal approach was to first estimate theexpected interaction counts for each interactiondistance by calculating the mean of all intra-chromosomal bin-bin interactions of the sameseparation distance throughout the raw intra-chromosomal contact matrix We used theR package HiTC (56) to facilitate manipulationof our HiC-Pro-produced raw contact matricesand estimation of the expected counts at var-ious interaction distances The probability of ob-serving an interaction between a bin-of-interestand another bin was then defined as the ex-pected interaction between those two bins di-vided by the sum of all expected interactionsbetween the bin-of-interest and all other intra-chromosomal bins A P valuewas then calculatedas binomial probability of observing the numberof interaction counts ormore between the bin-of-interest and some other bin where the numberof successes was defined as the observed inter-action count the number of tries as the totalnumber of observed interactions between thebin-of-interest and all other intrachromosomalbins and the success probability as the prob-ability of observing the bin-bin interaction es-timated from the expected mean interactioncounts The Benjamini-Hochberg method wasused to control false discovery rate (FDR) forP values determined for all interactions with abin-of-interest (includes all bins 1Mb up anddownstream in our tests)

Generation of stable selecteddCas9-VP64VPR and Cas9 NPCs

All CRISPR-based epigenomic editing assayswere performed on antibiotic-selected dCas9-VP64 (VP64 as the tetrameric VP16 activatordomain) and dCas9-VPR (VPR as the tripartiteactivator VP64-p65-Rta) NPCs derived as de-scribed in (34) For generation of Cas9 stableselectedNPCs we used a plasmid of lentiCRISPRv2 gifted by Feng Zhang (Addgene plasmid 52961) DNA sequencing with a U6 primer con-firmed the identity Lentiviral production andtitration were performed as described previously(14) Control S1 and S2 NPCs were spinfectedwithlentiCRISPR v2 virus as described (34) 48 hourspost-transduction cells were selected by exposureto puromycin at 03 mgmLWithout transductionall control cells diedwithin around5days after theantibiotic addition The puromycin-selectedNPCs were subject to Western blot analysis ofCas9 expression 30 mg of proteins were electro-phoresed inNuPAGE4-12Bis-Tris Protein Gels(NP0323PK2 Life Technologies) in 1times MESrunning buffer 200 V constant 35 min Proteinswere transferred onto nitrocellulose membrane(IB23002 Life Technologies) on the iBlotreg 2 DryBlotting System (program P3 700 min) Themembranes were incubated with primary anti-bodies against Cas9 (1250monoclonal clone 7A9Millipore) and b-Actin (110000 mouse 1406030Ambion) overnight at 4degC Thenmembraneswereincubated with the IRDye-labeled secondary anti-bodies for 45 min at RT in the dark on the rockerFluorescencewas visualized using aLi-CorOdysseyImaging System

Rajarajan et al Science 362 eaat4311 (2018) 14 December 2018 9 of 11

RESEARCH | RESEARCH ARTICLE | PSYCHENCODEon F

ebruary 26 2020

httpsciencesciencemagorg

Dow

nloaded from

In vitro transcription and transfectionof gRNAsGuide RNAs (gRNAs) were designed on Benchling(wwwbenchlingcom) using the CRISPR toolgRNAswere generated via in vitro transcription(IVT) with the GeneArt Precision gRNA Syn-thesis Kit (Thermo Fisher Scientific A29377) asper manufacturer instructions Five gRNAs weredesigned per condition (ie ldquoloop-SNPrdquo negativecontrol and positive control) and pooled fortransfection The genomic ranges within whichloop-SNP gRNAs were designed (ie regionspanning the SNP of interest and all gRNAs inthe condition) were roughly 600 bp for ASCL1550 bp forMATR3 460 bp for EFNB1 (with 25gRNAs directly overlapping the SNP) 300 bpfor SOX2 Puromycin-selected (1mgmL in NPCmedia Sigma P7255) dCas9-VP64 and dCas9-VPR NPCs (34) were seeded at a density of~400000 per well on Matrigel-coated (BD Bio-sciences) 24-well plates Pooled IVT gRNAs(500 ng total RNAwell) and 2 mL EditPro Stemlipofectamine (MTI-GlobalStem GST-2174now ThermoFisher STEM00003) were dilutedin 50 mL Opti-MEM (Thermo Fisher Scientific31985062) and added dropwise to each wellCells were harvested with TRIzol for total RNAextraction 48 hours later All experiments wereconducted with 3 to 6 biological replicates from1 donor (subject S1) generated in parallel withthe donor contributing isogenic dCas9-VP64 anddCas9-VPR effector cells Each data point in Fig 2D to F represents one biological replicate withineach condition For each target gene promoterand candidate loop control gRNAs were strate-gically placed into themiddle third of the (linear)genome portion bypassed by the candidate loopCRISPRa results were analyzed on PRISMwitha one-way ANOVA across 3 conditions with aDunnettrsquos test for multiple comparisons Cas9mutagenesis was also performed as describedabove with the exception of the negative con-trol which in these experiments consisted of anempty transfection (ie lipofectamine +Opti-MEMwithout any gRNA) Cas9 results were analyzedwith an unpaired t test comparing the loop-SNPand negative control conditions

RNA transcriptomic correlation heatmaps

Pearson correlation coefficient matrices werecalculated for gene expression in the childhoodonset schizophrenia data set (16) using R fromlists of genes that are located in cell-type-specificloops anchored at schizophrenia risk loci and asa subset of this list sets of genes whose proteinsparticipate in an association network for eachof the three cell types (see below) Significancewas computed calculating the absolute meancorrelation coefficient of each correlationmatrix(ldquoorganization scorerdquo) as a test statistic againsta null distribution generated by random genesampling Randomized gene lists were drawnonly from the pool of genes with over 1 countper million (CPM) in at least 30 of the ex-periments described in (16) To generate a nulldistribution of organization scores for a givencell type that accounted for genomic distance

and neighborhood effects we began by ran-domly selecting a significant PGC interactionfor that cell type (eg random selection fromtable S12) Using the bp genomic distance ofthis interaction we randomly selected two 10kbbins from the genome separated by the samedistance All genes overlapping these bins werethen added to the list of genes with which tocalculate the organization score This processwas iterated until enough genes were added tothe list to match the number of genes used inthe original cell-type-specific organization scoreFinally this protocol was repeated 1000 times togenerate the null distribution of random organi-zation scores This distribution was then used tocalculate significance of co-regulation (ie P =number of times |r|avg of the null exceeded thatof the test heatmap 1000) Note that STRINGgene network transcriptomic analyses (Fig 4B)were performed with 1000 random permuta-tions of genes sampled from the full schizophreniarisk connectome (ie risk locus + risk locus-connect genes) for each cell type

REFERENCES AND NOTES

1 P Rajarajan S E Gil K J Brennand S Akbarian Spatialgenome organization and cognition Nat Rev Neurosci 17681ndash691 (2016) doi 101038nrn2016124 pmid 27708356

2 Schizophrenia Working Group of the Psychiatric GenomicsConsortium Biological insights from 108 schizophrenia-associated genetic loci Nature 511 421ndash427 (2014)doi 101038nature13595 pmid 25056061

3 S S Rao et al A 3D map of the human genome atkilobase resolution reveals principles of chromatin loopingCell 159 1665ndash1680 (2014) doi 101016jcell201411021pmid 25497547

4 Y Jiang et al The methyltransferase SETDB1 regulatesa large neuron-specific topological chromatin domainNat Genet 49 1239ndash1250 (2017) doi 101038ng3906pmid 28671686

5 R Bharadwaj et al Conserved higher-order chromatinregulates NMDA receptor gene expression and cognitionNeuron 84 997ndash1008 (2014) doi 101016jneuron201410032pmid 25467983

6 R Bharadwaj et al Conserved chromosome 2q31conformations are associated with transcriptional regulation ofGAD1 GABA synthesis enzyme and altered in prefrontal cortexof subjects with schizophrenia J Neurosci 33 11839ndash11851(2013) doi 101523JNEUROSCI1252-132013pmid 23864674

7 H Won et al Chromosome conformation elucidates regulatoryrelationships in developing human brain Nature 538 523ndash527(2016) doi 101038nature19847 pmid 27760116

8 E Lieberman-Aiden et al Comprehensive mapping of long-range interactions reveals folding principles of the humangenome Science 326 289ndash293 (2009) doi 101126science1181369 pmid 19815776

9 A D Schmitt et al A compendium of chromatin contactmaps reveals spatially active regions in the human genomeCell Reports 17 2042ndash2059 (2016) doi 101016jcelrep201610061 pmid 27851967

10 J R Dixon et al Chromatin architecture reorganizationduring stem cell differentiation Nature 518 331ndash336 (2015)doi 101038nature14222 pmid 25693564

11 L de la Torre-Ubieta et al The dynamic landscape of openchromatin during human cortical neurogenesis Cell 172289ndash304e18 (2018) doi 101016jcell201712014pmid 29307494

12 A C Mitchell et al The genome in three dimensionsA new frontier in human brain research Biol Psychiatry75 961ndash969 (2014) doi 101016jbiopsych201307015pmid 23958183

13 A Topol et al Dysregulation of miRNA-9 in a subset ofschizophrenia patient-derived neural progenitor cellsCell Reports 15 1024ndash1036 (2016) doi 101016jcelrep201603090 pmid 27117414

14 S M Ho et al Rapid Ngn2-induction of excitatory neuronsfrom hiPSC-derived neural progenitor cells Methods101 113ndash124 (2016) doi 101016jymeth201511019pmid 26626326

15 J Tcw et al An efficient platform for astrocyte differentiationfrom human induced pluripotent stem cells Stem Cell Reports9 600ndash614 (2017) doi 101016jstemcr201706018pmid 28757165

16 G E Hoffman et al Transcriptional signatures ofschizophrenia in hiPSC-derived NPCs and neurons areconcordant with post-mortem adult brains Nat Commun 82225 (2017) doi 101038s41467-017-02330-5pmid 29263384

17 K J Brennand et al Modelling schizophrenia using humaninduced pluripotent stem cells Nature 473 221ndash225 (2011)doi 101038nature09915 pmid 21490598

18 C Gil-Sanz et al Lineage tracing using Cux2-Cre and Cux2-CreERT2 mice Neuron 86 1091ndash1099 (2015) doi 101016jneuron201504019 pmid 25996136

19 G Charlesworth et al Mutations in ANO3 cause dominantcraniocervical dystonia Ion channel implicated inpathogenesis Am J Hum Genet 91 1041ndash1050 (2012)doi 101016jajhg201210024 pmid 23200863

20 J Wang et al Single-cell gene expression analysis revealsregulators of distinct cell subpopulations among developinghuman neurons Genome Res 27 1783ndash1794 (2017)doi 101101gr223313117 pmid 29030469

21 B Bonev et al Multiscale 3D genome rewiring during mouseneural development Cell 171 557ndash572e24 (2017)doi 101016jcell201709043 pmid 29053968

22 J R Dixon et al Topological domains in mammalian genomesidentified by analysis of chromatin interactions Nature 485376ndash380 (2012) doi 101038nature11082 pmid 22495300

23 A F Pardintildeas et al Common schizophrenia alleles areenriched in mutation-intolerant genes and in regions understrong background selection Nat Genet 50 381ndash389 (2018)doi 101038s41588-018-0059-2 pmid 29483656

24 A Dobbyn et al Landscape of conditional eQTL in dorsolateralprefrontal cortex and co-localization with schizophrenia GWASAm J Hum Genet 102 1169ndash1184 (2018) doi 101016jajhg201804011 pmid 29805045

25 W V Chen T Maniatis Clustered protocadherins Development140 3297ndash3302 (2013) doi 101242dev090621pmid 23900538

26 T Yagi Molecular codes for neuronal individuality and cellassembly in the brain Front Mol Neurosci 5 45 (2012)doi 103389fnmol201200045 pmid 22518100

27 N C Durand et al Juicebox provides a visualization systemfor Hi-C contact maps with unlimited zoom Cell Syst 399ndash101 (2016) doi 101016jcels201507012pmid 27467250

28 M Fromer et al Gene expression elucidates functional impactof polygenic risk for schizophrenia Nat Neurosci 191442ndash1453 (2016) doi 101038nn4399 pmid 27668389

29 K Brennand et al Phenotypic differences in hiPSC NPCsderived from patients with schizophrenia Mol Psychiatry 20361ndash368 (2015) doi 101038mp201422 pmid 24686136

30 S Gulsuner et al Spatial and temporal mapping of de novomutations in schizophrenia to a fetal prefrontal corticalnetwork Cell 154 518ndash529 (2013) doi 101016jcell201306049 pmid 23911319

31 P Perez-Pinera et al RNA-guided gene activation by CRISPR-Cas9-based transcription factors Nat Methods 10 973ndash976(2013) doi 101038nmeth2600 pmid 23892895

32 M L Maeder et al CRISPR RNA-guided activation ofendogenous human genes Nat Methods 10 977ndash979 (2013)doi 101038nmeth2598 pmid 23892898

33 A Chavez et al Comparison of Cas9 activators in multiplespecies Nat Methods 13 563ndash567 (2016) doi 101038nmeth3871 pmid 27214048

34 S M Ho et al Evaluating synthetic activation and repressionof neuropsychiatric-related genes in hiPSC-derived NPCsneurons and astrocytes Stem Cell Reports 9 615ndash628 (2017)doi 101016jstemcr201706012 pmid 28757163

35 C P Fulco et al Systematic mapping of functional enhancer-promoter connections with CRISPR interference Science354 769ndash773 (2016) doi 101126scienceaag2445pmid 27708057

36 K Girdhar et al Cell-specific histone modification maps in thehuman frontal lobe link schizophrenia risk to the neuronalepigenome Nat Neurosci 21 1126ndash1136 (2018) doi 101038s41593-018-0187-0 pmid 30038276

37 G Kustatscher P Grabowski J Rappsilber Pervasivecoexpression of spatially proximal genes is buffered at the

Rajarajan et al Science 362 eaat4311 (2018) 14 December 2018 10 of 11

RESEARCH | RESEARCH ARTICLE | PSYCHENCODEon F

ebruary 26 2020

httpsciencesciencemagorg

Dow

nloaded from

protein level Mol Syst Biol 13 937 (2017) doi 1015252msb20177548 pmid 28835372

38 D Szklarczyk et al STRING v10 Protein-protein interactionnetworks integrated over the tree of life Nucleic Acids Res43 (D1) D447ndashD452 (2015) doi 101093nargku1003pmid 25352553

39 D Szklarczyk et al The STRING database in 2017 Quality-controlled protein-protein association networks made broadlyaccessible Nucleic Acids Res 45 (D1) D362ndashD368 (2017)doi 101093nargkw937 pmid 27924014

40 Y Watanabe K Khodosevich H Monyer Dendritedevelopment regulated by the schizophrenia-associated geneFEZ1 involves the ubiquitin proteasome system Cell Reports7 552ndash564 (2014) doi 101016jcelrep201403022pmid 24726361

41 S V Puram et al A CaMKIIb signaling pathway at thecentrosome regulates dendrite patterning in the brainNat Neurosci 14 973ndash983 (2011) doi 101038nn2857pmid 21725312

42 P Ma et al The ubiquitin ligase RNF220 enhances canonicalWnt signaling through USP7-mediated deubiquitination ofb-catenin Mol Cell Biol 34 4355ndash4366 (2014) doi 101128MCB00731-14 pmid 25266658

43 C G Hahn et al The post-synaptic density of humanpostmortem brain tissues An experimental study paradigm forneuropsychiatric illnesses PLOS ONE 4 e5251 (2009)doi 101371journalpone0005251 pmid 19370153

44 S Lomvardas et al Interchromosomal interactions andolfactory receptor choice Cell 126 403ndash413 (2006)doi 101016jcell200606035 pmid 16873069

45 S A Quinodoz et al Higher-order inter-chromosomal hubsshape 3D genome organization in the nucleus Cell 174744ndash757e24 (2018) doi 101016jcell201805024pmid 29887377

46 N Khanna Y Hu A S Belmont HSP70 transgene directedmotion to nuclear speckles facilitates heat shock activationCurr Biol 24 1138ndash1144 (2014) doi 101016jcub201403053 pmid 24794297

47 Network and Pathway Analysis Subgroup of Psychiatric GenomicsConsortium Psychiatric genome-wide association study analysesimplicate neuronal immune and histone pathways Nat Neurosci18 199ndash209 (2015) doi 101038nn3922 pmid 25599223

48 S R Gilman et al Diverse types of genetic variation converge onfunctional gene networks involved in schizophrenia Nat Neurosci15 1723ndash1728 (2012) doi 101038nn3261 pmid 23143521

49 R Bogdan D A A Baranger A Agrawal Polygenic Risk Scores inClinical Psychology Bridging Genomic Risk to IndividualDifferences Annu Rev Clin Psychol 14 119ndash157 (2018)doi 101146annurev-clinpsy-050817-084847 pmid 29579395

50 F Grubert et al Genetic control of chromatin states in humansinvolves local and distal chromosomal interactions Cell 1621051ndash1065 (2015) doi 101016jcell201507048pmid 26300125

51 N Servant et al HiC-Pro An optimized and flexible pipelinefor Hi-C data processing Genome Biol 16 259 (2015)doi 101186s13059-015-0831-x pmid 26619908

52 B Langmead C Trapnell M Pop S L Salzberg Ultrafast andmemory-efficient alignment of short DNA sequences to thehuman genome Genome Biol 10 R25 (2009) doi 101186gb-2009-10-3-r25 pmid 19261174

53 M Imakaev et al Iterative correction of Hi-C data revealshallmarks of chromosome organization Nat Methods 9999ndash1003 (2012) doi 101038nmeth2148 pmid 22941365

54 N C Durand et al Juicer provides a one-click system foranalyzing loop-resolution Hi-C experiments Cell Syst 3 95ndash98(2016) doi 101016jcels201607002 pmid 27467249

55 M Forcato et al Comparison of computational methods forHi-C data analysis Nat Methods 14 679ndash685 (2017)doi 101038nmeth4325 pmid 28604721

56 N Servant et al HiTC Exploration of high-throughput lsquoCrsquoexperiments Bioinformatics 28 2843ndash2844 (2012)doi 101093bioinformaticsbts521 pmid 22923296

ACKNOWLEDGMENTS

The authors thank all members of the Brennand and Akbarianlaboratories for constructive comments and discussions

C OrsquoShea S R Shannon I Silverberg and A Austin for valuableassistance with PCR assays E Xia C Rosenbluh and A Chessfor generous access to sequencing equipment L Winterkorn andstaff at the New York Genome Center for logistical support andM Peters for help and support as it pertains to data accessibilityFunding This work was supported by National Institute ofMental Health PsychENCODE grants U01MH103392 (SA) andR01MH106056 (SA and KJB) and a predoctoral Ruth L Kirschsteinfellowship 1F30MH113330 (PR) Author contributions Cellculture work including Hi-C 3C RNA-seq ATAC-seq Cas9 anddCas9 (epi)genome editing PRa CC CY BK BJ SP EALBZ S-MH and AL Biocomputing informatics and genomicanalyses TB WL NS SE-G EF GEH and ZW Contributedmaterials S-MH EF HW and DHG Supervised researchZW PRo KJB and SA Conceived and designed experimentsPRa KJB and SA Wrote the paper PRa TB WL KJBand SA with contributions from all coauthors Competing interestsThe authors declare no conflicts of interest Data and materialsaccessibility Scripts used for bioinformatics analyses can be foundat httpsgithubcomtborrmanSchahram-project and httpsgithubcomwill-NYGCsigHicInteractions Hi-C data files andmetadata are accessible through an open source platformwwwsynapseorgSynapsesyn12979101 (registration required DataDownloadmdashStudy ldquoiPSC-HiCrdquo) The code for this paper has beensubmitted to Zenodo and the respective links are httpszenodoorgbadgelatestdoi68036139 and httpszenodoorgbadgelatestdoi151779353

SUPPLEMENTARY MATERIALS

wwwsciencemagorgcontent3626420eaat4311supplDC1Materials and MethodsFigs S1 to S15References (57ndash67)Tables S1 to S25

27 February 2018 accepted 7 November 2018101126scienceaat4311

Rajarajan et al Science 362 eaat4311 (2018) 14 December 2018 11 of 11

RESEARCH | RESEARCH ARTICLE | PSYCHENCODEon F

ebruary 26 2020

httpsciencesciencemagorg

Dow

nloaded from

riskNeuron-specific signatures in the chromosomal connectome associated with schizophrenia

Weng Kristen J Brennand and Schahram AkbarianGeschwind Seok-Man Ho Matthew MacDonald Gabriel E Hoffman Panos Roussos Bin Zhang Chang-Gyu Hahn ZhipingYashaswini Elizabeth A LaMarca Bibi Kassim Behnam Javidfar Sergio Espeso-Gil Aiqun Li Hyejung Won Daniel H Prashanth Rajarajan Tyler Borrman Will Liao Nadine Schrode Erin Flaherty Charlize Casintildeo Samuel Powell Chittampalli

DOI 101126scienceaat4311 (6420) eaat4311362Science

ARTICLE TOOLS httpsciencesciencemagorgcontent3626420eaat4311

MATERIALSSUPPLEMENTARY httpsciencesciencemagorgcontentsuppl201812123626420eaat4311DC1

CONTENTRELATED

filecontenthttpsciencesciencemagorgcontentsci3626420eaat6576fullhttpsciencesciencemagorgcontentsci3626420eaat6720fullhttpsciencesciencemagorgcontentsci3626420eaat8077fullhttpsciencesciencemagorgcontentsci3626420eaat8464fullhttpsciencesciencemagorgcontentsci3626420eaat8127fullhttpsciencesciencemagorgcontentsci3626420eaat7615fullhttpsciencesciencemagorgcontentsci36264201262full

REFERENCES

httpsciencesciencemagorgcontent3626420eaat4311BIBLThis article cites 67 articles 9 of which you can access for free

PERMISSIONS httpwwwsciencemagorghelpreprints-and-permissions

Terms of ServiceUse of this article is subject to the

is a registered trademark of AAASScienceScience 1200 New York Avenue NW Washington DC 20005 The title (print ISSN 0036-8075 online ISSN 1095-9203) is published by the American Association for the Advancement ofScience

Science No claim to original US Government WorksCopyright copy 2018 The Authors some rights reserved exclusive licensee American Association for the Advancement of

on February 26 2020

httpsciencesciencem

agorgD

ownloaded from