RESEARCH ARTICLE SUMMARY◥

NEUROGENOMICS

An environment-dependenttranscriptional network specifieshuman microglia identityDavid Gosselin,* Dylan Skola,* Nicole G. Coufal,* Inge R. Holtman,†Johannes C. M. Schlachetzki,† Eniko Sajti, Baptiste N. Jaeger, Carolyn O’Connor,Conor Fitzpatrick, Martina P. Pasillas, Monique Pena, Amy Adair, David D. Gonda,Michael L. Levy, Richard M. Ransohoff, Fred H. Gage, Christopher K. Glass‡

INTRODUCTION:Microglia play essential rolesin central nervous system homeostasis and in-fluence diverse aspects of neuronal function,including refinement of synaptic networks andelaboration of neuromodulatory factors formemory and motor learning. Many lines ofevidence indicate that dysregulation ofmicrog-lial functions contributes to the pathogenesisof neurodegenerative diseases, including Alz-heimer’sdiseaseandParkinson’sdisease.Emergingevidence frommouse and human studies also sug-gests thatmicroglia influence neurodevel-opmental and psychiatric disorders suchas schizophrenia anddepression.Most dis-ease riskallelesassociatedwithneurodegen-erativediseases reside innoncoding regionsof the genome, requiring thedelineation offunctionalgenomicelements in therelevanthuman cell types to establishmechanismsof causation. The recent observation thatmouse brain environment strongly influ-ences microglia-specific gene expressionhas implications for understanding path-ogenic responses of microglia in diseasesand disorders andmodeling their pheno-types in vitro.

RATIONALE:Although dysregulation ofmicroglial activity is genetically linked toneurodegenerativediseasesandpsychiatricdisorders, no systematic evaluations of hu-man microglia gene expression or regu-latory landscapes are currently available.In addition, the extent to whichmice pro-vide suitablemodels for humanmicrogliais unclear. The major goals of this studywere to define the transcriptomes andDNA regulatory elements of human mi-croglia ex vivo and in vitro in comparisonto the mouse and to systematically relatethese features to expression of genes asso-ciatedwith genome-wide association study(GWAS) risk alleles or exhibiting alteredexpression in neurodegenerative diseasesand psychiatric disorders.

RESULTS: We used RNA sequencing, chro-matin immunoprecipitation sequencing, andassay for transposase-accessible chromatinsequencing to characterize the transcriptomesand epigenetic landscapes of human microg-lia isolated from surgically resected brain tis-sue in excess of that needed for diagnosis.Although some effects of underlying diseasecannot be excluded, the overall pattern of geneexpression was markedly consistent. Microglia-enriched genes were found to overlap signif-

icantly with genes exhibiting altered expressionin neurodegenerative diseases and psychiatricdisorders and with genes associated with a widespectrum of disease-specific risk alleles. Humanmicroglia gene expression was well correlated

with mouse microglia geneexpression, but numerousspecies-specific differenceswere also observed thatincluded genes linked tohuman disease. More thanhalf of the genes associ-

ated with noncoding GWAS risk alleles forAlzheimer’s disease are preferentially expressedin microglia. DNA recognition motifs enrichedat active enhancers and expression of the cor-responding lineage-determining transcriptionfactors were very similar for human and mousemicroglia. Transition of human and mouse mi-croglia from the brain to tissue culture revealedremodeling of their respective enhancer land-scapes and extensive down-regulation of genesthat are induced in primitive mouse macro-phages following migration into the fetal brain.Treatment of microglia in vitro with trans-forming growth factor b1 (TGF-b1) had rel-atively modest effects in maintaining theex vivo pattern of gene expression. A signif-icant subset of the genes up- or down-regulatedin vitro exhibited altered expression in neu-

rodegenerative diseases and psychiatricdisorders.

CONCLUSION: These studies identifycore features of human microglial tran-scriptomes and epigenetic landscapes. In-tersection of the microglia-specific genesignature with GWAS and transcriptomicdata supports roles of microglia as bothresponders and contributors to diseasephenotypes. The identification of anenvironment-sensitive program of geneexpression and corresponding regulatoryelements enables inference of a conservedand dynamic transcription factor networkthat maintains microglia identity andfunction. The combinations of signalingfactors in the brain necessary to maintainmicroglia phenotypes remain largely un-known. In concert, these findings willinform efforts to generate microglia-likecells in simple and complex culture sys-tems and understand gene-environmentinteractions that influence homeostaticand pathogenic functions of microglia inthe human brain.▪

RESEARCH

Gosselin et al., Science 356, 1248 (2017) 23 June 2017 1 of 1

The list of author affiliations is available in the fullarticle online.*These authors contributed equally to this work.†These authors contributed equally to this work.‡Corresponding author. Email: ckg@ucsd.eduCite this article as D. Gosselin et al., Science 356,eaal3222 (2017). DOI: 10.1126/science.aal3222

Brain environment specifies gene expression inmicroglia. Human microglia transcriptomes andenhancer landscapes were defined ex vivo followingpurification from surgically resected brain tissue andin vitro after transfer to a tissue culture environment.Dynamic changes in these features enabled delineationof transcription factors controlling an environment-dependent program of gene expression that overlapswith genes that are dysregulated in brain pathologies.

ON OUR WEBSITE◥

Read the full articleat http://dx.doi.org/10.1126/science.aal3222..................................................

on January 6, 2021

http://science.sciencemag.org/

Dow

nloaded from

RESEARCH ARTICLE◥

NEUROGENOMICS

An environment-dependenttranscriptional network specifieshuman microglia identityDavid Gosselin,1* Dylan Skola,1* Nicole G. Coufal,2,3* Inge R. Holtman,1,4†Johannes C. M. Schlachetzki,1† Eniko Sajti,3 Baptiste N. Jaeger,2 Carolyn O’Connor,2

Conor Fitzpatrick,2 Martina P. Pasillas,1 Monique Pena,2 Amy Adair,2 David D. Gonda,5

Michael L. Levy,5 Richard M. Ransohoff,6 Fred H. Gage,2 Christopher K. Glass1,7‡

Microglia play essential roles in central nervous system (CNS) homeostasis and influencediverse aspects of neuronal function. However, the transcriptional mechanisms thatspecify human microglia phenotypes are largely unknown. We examined thetranscriptomes and epigenetic landscapes of human microglia isolated from surgicallyresected brain tissue ex vivo and after transition to an in vitro environment. Transfer to atissue culture environment resulted in rapid and extensive down-regulation of microglia-specific genes that were induced in primitive mouse macrophages after migration into thefetal brain. Substantial subsets of these genes exhibited altered expression inneurodegenerative and behavioral diseases and were associated with noncoding riskvariants. These findings reveal an environment-dependent transcriptional networkspecifying microglia-specific programs of gene expression and facilitate efforts tounderstand the roles of microglia in human brain diseases.

Microglia are the resident macrophagepopulation of the central nervous system(CNS) that perform functions requiredfor typical development, as well as ho-meostasis, plasticity, immunity, and repair

(1–5). Dysregulation of microglia functions con-tributes to the pathogenesis of neurodegenerativediseases, includingAlzheimer’s disease (AD) (6–8),Parkinson’s disease (PD) (9–12), and Huntington’sdisease (HD) (13–15). In addition, microglia mayplay roles in neurodevelopmental and psychiatricdiseases such as autism, schizophrenia (Scz), anddepression (16, 17). Collectively, these observationsprovide a compelling incentive to define themech-anisms that control microglia functions and toinvestigate the potential to alter these functionsfor therapeutic purposes.

Microglia possess distinctive ontogeny, mor-phology, gene expression patterns, and functionalcharacteristics and account for 5 to 20% of allglial cells, depending on brain region. Studiesin mice indicate that microglia are derived fromyolk sac progenitor cells that populate the brainearly in development (18, 19). Microglia sharemany traits with other subsets of tissue-residentmacrophages, including dependence on the CSF1receptor for differentiation and survival (18, 20),a requirement for PU.1 as an essential lineage-determining transcription factor (LDTF) (21, 22),highly efficient phagocytic and tissue debris clear-ance abilities (2, 5), and the ability to quickly trig-ger an inflammatory response following detectionof pathogens or tissue damage (23).Microglia also express hundreds of genes at

higher levels than other macrophage subsets(24–26) thatmay be influenced by factors presentin the CNS microenvironment (25, 27), suggest-ingmechanisms bywhich the brain environmentinfluences the activity and functions of one of itsprimary support cells. As microglia participate inthe developmental refinement of synaptic net-works (28) and elaborate neuromodulatory fac-tors for adult motor learning (29), alterations inthe brain microenvironment may lead to patho-genic programs of gene expression in microglia.Insights into mechanisms that control the

development and function of specific cell typescan be obtained by analyses of their enhancerrepertoires and corresponding epigenetic land-scapes (30). Microglial enhancers are primed byrelatively simple combinations of LDTFs (31, 32)

to become sites of activity of broadly expressedsignal-dependent transcription factors (SDTFs).These enhancers thereby integrate cellular signal-ing pathways with DNA. Within cell-specificenhancers, DNA sequences provide informationabout the identities of key LDTFs and the activitystates of signaling pathways that control thefunctions of SDTFs (31, 33–35). Furthermore,the locations and sequence content of enhancersprovide insights into how noncoding geneticvariationmay affect gene expression and cellularphenotypes (32, 36–38).

Human microglia transcriptomes

Microgliawere isolated frombrain tissue resectedfor treatment of epilepsy, brain tumors, or acuteischemic events in 19 individuals. Patient charac-teristics, medications, and assays performed foreach sample are provided in table S1. For patientswith epilepsy, the resected tissue provided did notinclude the brain area displaying the strongestepileptogenic activity. Althoughmost specimensfrom individuals with epilepsy exhibited somedegree of cortical dysplasia, three samples weredetermined tobehistologicallynormal. Forpatientswith brain tumors (8), samples were consideredto be tumor negative by pathological examination.The two brain samples resected to treat acuteischemic injuries were from individuals who hadno prior evidence of brain pathology. As judgedby neuropathological diagnostic evaluation, noneof the samples displayed evidence of pathologicalmicrogliosis (i.e., loss of tiled distribution, pro-cess retraction, and enlarged cell bodies) despitethe occasional presence of reactive cells (fig. S1).Tissues were immediately placed on ice in the

operating room and processed for microglia iso-lation within a few hours of surgical resection(25). Amicroglia-rich cell fractionwas isolated byPercoll gradient centrifugation and further puri-fied by fluorescence-activated cell sorting,with finalgating on live-CD11b+CD45LowCD64+CX3CR1High

single cells (fig. S2A). This strategy excluded cellsexpressingmoderate to high levels of CD45,whichwere present in variable amounts depending onthe specific sample (fig. S2A). Microglia were thentriaged to specific analysis workflows dependingon yields and depth of existing data sets (table S1).For five individuals, RNA sequencing (RNA-seq)was also performed from a portion of the intactcortical brain tissue used to isolatemicroglia (here-after referred to as “cortex”).Isolated microglia exhibited a high degree of

viability and RNA integrity. In one case, braintissue was simultaneously obtained from theoccipital and parietal lobes of the same individ-ual, enabling independent isolation of microgliafrom each region. RNA expression levels for thesetwo samples exhibited a Pearson correlation co-efficient of 0.929 (fig. S3A). In a second case, anindividualunderwent sequential surgical resectionsapproximately 1 year apart. The RNA expressionlevels inmicroglia obtained from this individualat these two time points exhibited a Pearson cor-relation coefficient of 0.905 (fig. S3B). On the basisof these observations, the variance in gene expres-sion noted in these samples likely overestimates

RESEARCH

Gosselin et al., Science 356, eaal3222 (2017) 23 June 2017 1 of 11

1Department of Cellular and Molecular Medicine, Universityof California, San Diego, 9500 Gilman Drive, La Jolla, CA92093-0651, USA. 2Laboratory of Genetics, The SalkInstitute for Biological Studies, 10010 North Torrey PinesRoad, La Jolla, CA 92037-1002, USA. 3Department ofPediatrics, University of California, San Diego, 9500 GilmanDrive, La Jolla, CA 92093-0651, USA. 4Department ofNeuroscience, section Medical Physiology, University ofGroningen, University Medical Center Groningen, Groningen,Netherlands. 5Department of Neurosurgery, University ofCalifornia, San Diego–Rady Children’s Hospital, San Diego,CA 92123, USA. 6Neuroimmunology, Biogen, 225 BinneyStreet, Cambridge, MA 02142, USA. 7Department ofMedicine, University of California, San Diego, 9500 GilmanDrive, La Jolla, CA 92093-0651, USA.*These authors contributed equally to this work. †These authorscontributed equally to this work. ‡Corresponding author. Email:ckg@ucsd.edu

on January 6, 2021

http://science.sciencemag.org/

Dow

nloaded from

the variance in the isolation of microglia and im-plies ahighdegree of precision in theRNA-seqdataset. Comparisons of the RNA-seq profiles of anytwo individuals for genes expressed >2 log2 tran-scripts per million (TPM) yielded Pearson corre-lation coefficients of ~0.8, as exemplified by acomparison ofmicroglia isolated froma 1-year-oldfemale patient diagnosed with atypical rhabdoidtumor to microglia isolated from a 7-year-oldfemale patient with epilepsy (fig. S3C). When alltranscripts were considered (i.e., including genesexpressed at ≤2 log2 TPM), Pearson correlation

coefficients were in the range of 0.94 to 0.98 (fig.S3D), indicating no substantial outliers amongsamples. RNA-seq data indicated low-to-absentlevels of neuronal, astrocyte, oligodendrocyte, andendothelial-specific genes, confirming insignificantlevels of contaminating cells (fig. S2B and table S2).Hierarchical clustering of expressed genes from

all microglia samples in comparison to cortex isillustrated in Fig. 1A. Although there are likelyto be alterations in gene expression related todisease, no strong patterns emerged that wererelated to diagnosis or age (Fig. 1A), perhaps due

in part to exclusion of most reactive cells at thetime of sorting (fig. S2A). A small set of genesexhibited significant sex-specific differences ingene expression, and these were mostly relatedto X- and Y-linked genes (fig. S4). Relative ex-pression values for the 30 most highly expressedgenes across each of the patients are illustratedin Fig. 1B, indicating substantial individual var-iation for some transcripts. This analysis con-firmed that humanmicroglia expressed very highlevels of numerous genes associated with regu-lation of microglia ramification and motility,

Gosselin et al., Science 356, eaal3222 (2017) 23 June 2017 2 of 11

Fig. 1. Human microglia transcrip-tomes. (A) Heat map of mRNAexpression values determined inmicroglia isolated from 19 individualsand in cortex from 5 of these individ-uals. A total of 881 genes exhibiting>10-fold-higher average expression inmicroglia compared to cortex areindicated at the top of the heat map.(B) Individual variation in gene expres-sion of the 30 most highly expressedgenes in human microglia. (C) Enrich-ment of the 881 microglia-enrichedgenes up-regulated in HIV-associatedneurocognitive disorder (HAND) incentrum semiovale (CS) (48),up-regulated in schizophrenia (Scz) inthe anterior prefrontal cortex (PFC)(49), positively correlated with Braakstage of AD in PFC (50), up-regulatedor down-regulated in PD in cortex (51),up-regulated in frontotemporal lobardegeneration (FTLD) in frontal cortex(52), or up-regulated in AD in theCA1 region of the brain (53). (D) Piechart depicting the relative expressionof genes associated with AD riskvariants in brain cortex and microglia.(E) Combined bar graphs and dot plotsof the TPM expression levels of the21 most highly expressed genesassociated with AD risk variants inmicroglia and cortex. Red stars indicatesignificant differential expression(>2-fold, FDR 0.05).

RESEARCH | RESEARCH ARTICLEon January 6, 2021

http://science.sciencemag.org/

Dow

nloaded from

including the fractalkine receptor CX3CR1,purinergic receptor P2RY12, and genes involvedin synaptic remodeling, such as complement fac-tors C3, C1QA, C1QB, and C1QC (table S2).Nearly all studies of differentially regulated

genes in human neurodegenerative or behav-ioral disorders are based on RNA profiling ofintact tissues. As gene expression profiles forindividual cell types in the human brain are poor-ly characterized, it is difficult to assign cell typesresponsible for disease-related changes. Using astringent cutoff of 10-fold–increased expression

relative to cortex tissue at a false discovery rate(FDR) of 0.05, we defined amicroglial gene signa-ture in the human brain comprising 881 tran-scripts (Fig. 1A and table S2).We then intersectedthis gene list with 46 publicly available data setsfor differentially regulated genes in neurodegen-erative and behavioral disorders derived frommicroarray or RNA-seq analysis of intact tissue(39–53). Two-sided Fisher’s exact tests revealedsignificant enrichment or depletion of microg-lia signature genes in 28 of the 46 data sets(table S3). For example, hundreds of genes in

the microglia gene signature exhibited signif-icant overlaps with sets of genes that were posi-tively correlated with Braak stage in AD, wereup-regulated in frontotemporal lobar degenera-tion (FTLD) in frontal cortex, were up-regulatedinHIV-associatedneurocognitivedisorder (HAND),or were up-regulated in Scz (Fig. 1C). Notably,different subsets of the microglia signature geneset were up-regulated or down-regulated in cor-tex in PD (Fig. 1C).Characterization of the microglia transcrip-

tome, in conjunction with parallel whole-brain

Gosselin et al., Science 356, eaal3222 (2017) 23 June 2017 3 of 11

Fig. 2. Comparison of human andmouse microglia transcriptomes.(A) Scatter plot depicting mRNAexpression levels of mouse andhuman genes with one-to-oneorthologs, highlighting significantlydifferentially expressed genes inhuman (blue) and mouse microglia(green) (>10-fold, FDR 0.05).(B) Identification of a mousemicroglia gene signature consistingof 900 mRNAs expressed >10-foldhigher in microglia (FDR 0.05)compared to the average expressionvalues of macrophages from skin,lung, liver, and kidney. (C) Func-tional annotations of the microgliagene signature. (D) Scatter plotdepicting mRNAs expression levelsof human and mouse microgliawith orthologs intersected with themicroglia gene signature expressionprofile from (B). Genes with con-served expression are depicted inred, genes more highly expressedin human are in blue, and genesmore highly expressed in mice are ingreen. (E) Pie chart showingrelative expression of genes associ-ated with AD risk alleles in humanand mouse microglia. (F) Combinedbar graphs and dot plots illustratingthe expression of the top 21 mostabundant genes associated with ADrisk variants in human and mousemicroglia (average TPM-rankordered). Red stars indicatesignificant differential expression(>2-fold, FDR 0.05).

RESEARCH | RESEARCH ARTICLEon January 6, 2021

http://science.sciencemag.org/

Dow

nloaded from

cortical gene expression profiling, also enabledevaluation of the relative expression levels of genesassociated with AD, PD, multiple sclerosis (MS),and Scz risk alleles (54). Many of these genes werepreferentially expressed inmicroglia, with the rela-tive proportion varying by disorder (Fig. 1D andfig. S5). With respect to AD, for example, expres-sion for 28 of 48 ADgeneswas higher inmicroglia,including TREM2, SORL1, INPP5D, MEF2C, andCD33 (Fig. 1E). For MS, 48 of 81 genes associatedwith risk alleles were preferentially expressed inmicroglia (fig. S5). By contrast, approximately one-

third of the genes associated with PD and Scz riskalleles exhibited preferential expression in microg-lia (fig. S5). As the risk alleles for these disordersare largely nonoverlapping, these findings sug-gest different roles for microglia in each disordercontext.The extent to which findings in mice and hu-

mans are similar can vary dependingon the systemevaluated (55). We compared average transcriptlevels of humanmicrogliamRNAswithorthologousmRNAs frommicroglia isolated from 7- to 10-week-old male C57BL/6 mice using an identical protocol

(25) (Fig. 2A, fig. S6, and tables S4 and S5). Themicroglia transcriptomecomparisonbetweenthe twospecies revealedoverall similarities inpatterns of geneexpression (Pearson’s r = 0.806), with themajorityof orthologous genes pairs (13,253 of 15,768)expressed within a fourfold range. Nonetheless,this analysis also revealed genes with species-specific bias in expression magnitude. At a cut-off of 10-fold difference and an FDR of 0.05, 400orthologousmRNAswere preferentially expressedin human microglia and 293 mRNAs were morehighly expressed in mouse microglia (Fig. 2A).

Gosselin et al., Science 356, eaal3222 (2017) 23 June 2017 4 of 11

Fig. 3. Active genomic regula-tory regions of human microglia.(A) ATAC-seq and ChIP-seq datasets for PU.1, H3K27ac, andH3K4me2 in the vicinity of theBIN1 gene locus. Each colorrepresents a different individual.rs6733839 is a significant riskallele associated with AD and islocated within a region highlybound by PU.1. (B) Motifs enrichedat distal ATAC-seq peaks associ-ated with H3K4me2 in human andmouse microglia. (C) Heat mapof mRNA expression values ofmost highly expressed tran-scription factors recognizing motifsshown in (B). (D) Heat map ofmRNA expression values oftranscription factors associatedwith SEs in human and mousemicroglia. (E) Scatter plot depict-ing mRNA expression values oforthologous genes encodingtranscription factors in human andmouse microglia relative tooverall transcriptome profiles(gray dots). Orthologoustranscription factors similarlyexpressed are denoted in black;transcription factors more highlyexpressed (>10-fold change, FDR0.05) in human are colored inblue, and those more elevated inmouse are colored in green.(F) Scatter plot of H3K27ac tagcounts at human and mouse pro-moters. Promoters associated withRNA transcripts expressed 10-foldhigher in human microglia arecolored in blue, and promotersassociated with RNA transcriptsexpressed 10-fold higher in mousemicroglia are colored in green.

RESEARCH | RESEARCH ARTICLEon January 6, 2021

http://science.sciencemag.org/

Dow

nloaded from

Notable examples included higher expressionin human microglia of regulators of the comple-ment system (e.g., C2, C3, VSIG4, SERPING1) andgenes involved in brain structure development,includingSYNDIG1,GLDN,CTTNBP2, andROBO3.By contrast, genesmorehighly expressed inmousemicroglia included Hexb, Sparc, and Sall3. Ofthe subset of these genes that also exhibitedmicroglia-specific expression, a similar species-specific pattern of expression was observed inthe intact cortical tissue used for microglia iso-lation, indicating that these differences were nota consequence of the isolation procedure (fig. S7).We compared the levels of mRNA transcripts

in C57BL/6microgliawith different tissue-residentmacrophage populations (26, 56) and observed900 mRNAs with levels of expression in microg-lia that were 10-fold greater than the integratedmean of expression levels in the other macro-phage subsets at a FDR of 0.05 (Fig. 2B). This def-inition captures amousemicroglia gene signatureset that is larger butmostly inclusive of previouslydescribedmicroglia signature sets (27, 57, 58). Geneontology analysis of this extended microglia genesignature returned significant functional enrich-ment for CNS-related categories such as regulationof neurogenesis, synaptic signaling, CNS devel-opment, response to transforming growth factorbeta (TGF-b), and regulation of neurotransmittertransport (Fig. 2C). RNA-seq data sets are notavailable for a sufficient number of human res-ident macrophage populations to generate a cor-responding human microglia signature gene list.However, the relative expression levels of the hu-man andmouse orthologous genes correspondingto the mouse microglia signature genes revealedthat ~53% (477 of 900) of the genes that werehighly specific to microglia in the mouse were ex-pressed at comparable levels in human microg-lia (Fig. 2D).Comparing the expression of genes linked to

human neurodegenerative and behavioral dis-orders, of the 39 AD-associated and expressedgenes, 11were higher in humanmicrogliawhereasthreewere increased inmousemicroglia (Fig. 2E).Notably, of the top 25 most expressed, two genes,APOC1 andMPZL1, were essentially not expressedin C57BL/6 microglia; the expression of a thirdone, SORL1, was nearly 20 times lower in micethan in humans (Fig. 2F). PD-associated genesalso displayed similar human versus mouse dis-tributions (fig. S8). PD-associated genes SNCA,LRRK2, CD46, and CXCL12 were expressed to agreater extent in humanmicroglia than inmousemicroglia. Corresponding comparisons for allexpressed genes linked to AD, PD, MS, and Sczare illustrated in fig. S8. Collectively, these resultshighlight human versus mouse similarities anddifferences in the expressionprofile of coremicrog-lia genes.

Enhancer landscape of human microglia

We performed assay for transposase-accessiblechromatin–sequencing (ATAC-seq) to define openregions of chromatin and chromatin immunopre-cipitation–sequencing (ChIP-seq) for histone H3lysine 4 dimethylation (H3K4me2) and histone

H3 lysine27acetylation (H3K27ac) todefine regionsof poised/active and active chromatin, respectively,in bothhumanandmousemicroglia.MouseATAC-seq data were aligned with previous H3K4me2and H3K27ac data (25). ChIP-seq and ATAC-seqdata were highly correlated across human indi-viduals andmouse biological replicates (fig. S9). ArepresentativeUniversity of California SantaCruz(UCSC) genomebrowser track for humanmicrogliadata in the vicinity of the BIN1 gene is illustrated inFig. 3A, and for CX3CR1 and TREM2 in fig. S10A.rs6733839 is a noncoding risk allele for AD (59) thatlieswithinanenhancer-like region~25kbupstreamof the BIN1 transcriptional start site, suggestingregulatory function. De novo motif analysis ofATAC-seq peaks associated with H3K4me2 wasperformed to identify the most highly enrichedtranscription factor recognitionmotifs associatedwith poised or active chromatin. These analysesreturned very similar motifs for both mouse andhuman microglia. A dominant signature was ob-served for PU.1 in both species, with highly signif-icant enrichment also observed formotifs assignedto CTCF, IRF, RUNX,MEF2, C/EBP, AP-1, SMAD,and MAF factors (Fig. 3B and fig. S10B).Expression levels of transcription factors re-

cognizing the motifs identified in Fig. 3B areillustrated in Fig. 3C. SPI1 (encoding PU.1) wasthe most highly expressed ETS domain tran-scription factor, and the top motifs were nearlyperfect matches for its consensus recognitionsequence. The assignment of the second-mostenriched motif to CTCF was unexpected becauseCTCF is ubiquitously expressed and is thoughtto primarily play roles in the organization ofbroad domains of chromatin architecture. How-ever, the identified motif was a near-perfect matchfor the CTCF consensus recognition site, andCTCF was the only transcription factor expressedin microglia known to recognize this motif. Bycontrast, the remaining motifs were clearly recog-nized by multiple related members of transcrip-tion factor families that were reliably coexpressedin microglia.We used H3K27ac as a means of identifying

super-enhancers (SEs), which are regions of thegenome exhibiting a disproportionately high den-sity of active regulatory marks and transcriptionfactor binding (60–62). Common SEs were asso-ciated with genes that are known to play essentialroles in the development and function of virtual-ly all macrophages, whereas mouse microglia-specific SEs were associated with genes knownto play restricted or selected roles in this cell type(25). Corresponding analysis of the H3K27acHigh

regions in human microglia identified ~700 SEsthat were also associated with active regions ofthe genome overlapping or in close proximity togenes highly expressed in or specific to microglia,including CX3CR1, P2RY12, and TMEM119 (tableS6). The intersection of transcription factor genesassociated with SEs in human andmouse microg-lia captured many of the transcription factors illus-trated in Fig. 3C. An additional 35 factors wereassociated with SEs, some of which have beenlinked to functional roles in microglia, includ-ing SALL1, STAT3, and RELA (Fig. 3D). These

factors also exhibited similar expression levelsin mouse and human microglia.Evaluation of mRNAs annotated as encod-

ing orthologous transcription factors indicatedsimilar expression levels in human and mousefor the great majority. However, using a 10-foldcutoff at an FDR of 0.05, 20 factors were pref-erentially expressed in human microglia, includ-ing CIITA, RUNX2, EGR3, andNR4A2. By contrast,16 factors were preferentially expressed in mousemicroglia, including Sall3 and Smad1 (Fig. 3E).To establish the relationship between enriched

motifs and actual transcription factor binding,we performed ChIP-seq for PU.1 in microgliaisolated from six individuals. These experimentsyielded a highly consistent pattern of genomicbinding (Fig. 3A and fig. S10A), with PU.1 beinglocalized at 51.2% of ATAC-seq–defined regionsof open chromatin associated with H3K4me2and at 60.9% of regions exhibiting these featuresplus H3K27ac. De novo motif enrichment anal-ysis identified the PU.1 motif as the top hit, fol-lowed bymotifs for IRF, AP-1, MEF2, C/EBP, andRUNX (Fig. 3B and fig. S10B). Similar resultswere observed for Pu.1 binding sites in mousemicroglia. The frequency distribution of many ofthesemotifs is consistent with the correspondingtranscription factors forming ternary complexeswith PU.1 (e.g., the IRF motif; fig. S10C) or func-tioning as collaborative binding partners (e.g.,CTCF, AP-1, and RUNX motifs; fig. S10C).We also investigated the relationships between

PU.1 binding, open chromatin, and H3K27ac forgenes that are highly differentially expressedbetween human and mouse microglia. H3K27acat promoters was strongly correlated with differ-ential gene expression (Fig. 3F). Differential geneexpression andH3K27acwere also associatedwithspecies-specific PU.1 binding and ATAC-seq signalexemplified by the SALL1, SPARC, VSIG4, andAPOC1 loci (fig. S11). Collectively, these findingsprovide evidence that the transcriptional mech-anisms that control gene expression are largelyconserved between human and mouse microgliaand that differences in gene expression are pri-marily driven by species-specific organization ofcis-regulatory elements at target genes. However,as a small number of transcription factors thoughtto play important roles in microglia are diver-gently expressed, such as CIITA and RUNX2 pro-teins, there is also the potential for trans effectsat the subsets of genes regulated by these factors.

Alterations of the human microgliatranscriptome in vitro

Tissue environment has emerged as an impor-tant determinant of microglia identity, but manystudies ofmicroglia are performed in vitro. Trans-criptomic analysis of mouse microglia transferredto a tissue culture environment for 7 days docu-mented significant changes in expression of hun-dreds of genes, with a preferential reduction inexpression of microglia-specific genes (25). Sim-ilar to what was observed with mouse microglia,transfer of human microglia to the in vitro envi-ronment resulted in significant changes in theexpression of thousands of genes. Responses of

Gosselin et al., Science 356, eaal3222 (2017) 23 June 2017 5 of 11

RESEARCH | RESEARCH ARTICLEon January 6, 2021

http://science.sciencemag.org/

Dow

nloaded from

human microglia isolated from different indi-viduals to the in vitro environment were concor-dant (Fig. 4A, iv). For the individuals with bothex vivo and in vitro samples available, gene ex-pression within culture conditions (mean Pear-son correlation = 0.95) was more correlated (P =0.008 by permutation test) than between the exvivo and in vitro samples of the same individual(mean Pearson correlation = 0.89). A paired anal-ysis of samples for which both ex vivo and 7-dayin vitro microglia populations were obtained re-vealed more than 300 genes that were inducedgreater than 10-fold and more than 700 genesthat were repressed more than 10-fold in vitro(FDR 0.05). At a twofold cutoff, 2263 genes wereup-regulated in vitro and 3702 genes were down-regulated at a FDR of 0.05. Thirty-three percentof the conserved signature gene set and 31% ofthe genes associated with AD risk variants exhib-ited >2-fold reduction in expression in vitro (Figs.4, B to D). A similar proportion (30%) of genesassociated with PD also displayed such reduc-tion in vitro, whereas more limited changes forgenes associated with MS and Scz were observed(fig. S12).We performed RNA-seq in human microglia

at 6, 24, and 168 hours following transfer totissue culture. Highly dynamic responses wereobserved beginning at 6 hours and lasting through-out the time course (Fig. 4D). Cultured microgliaremained positive for Iba1 staining, whereas noexpression of markers for astrocytes or neuronswas observed, indicating that changes in RNAexpression were not due to contaminating cells(fig. S13A). A total of 1957 mRNAs were inducedmore than fourfold by 6 hours after plating. Geneontology analysis of highly induced genes indi-cated strong enrichment for terms related to in-flammation and stress responses (fig. S13B, C).Notably, more than 2000 genes were down-regulated more than fourfold at 6 hours afterplating, with some genes exhibiting partial re-covery at the 7-day time point (Fig. 4E and fig.S13D). Down-regulated genes were strongly en-riched for terms related to immune cell functionand signaling, as well as immune, blood vessel,and brain development (fig. S13E). Similar ob-servations of gene expression over time weremade in mouse microglia cultured for 6, 24, and168 hours (i.e., 7 days; fig. S14 and table S4).

An environment-dependenttranscriptional network

The perturbation in gene expression resultingfrom the transition to an in vitro environmentprovided an opportunity to gain insight intothe transcriptional network controlling humanmicroglia identity. Although SPI1 increased slightly,most of the mRNAs encoding transcription factorsthat recognize enriched motifs in microglia en-hancers exhibited significant reductions in vitro.In particular, AP-1 and MEF2 family members,as well asMAF, RUNX, and SMAD3, were highlysensitive to the environmental perturbation (Fig.5A). Similarly, the majority of transcription factorsassociated with SEs were also down-regulated(Fig. 5B).

Gosselin et al., Science 356, eaal3222 (2017) 23 June 2017 6 of 11

Fig. 4. Influence of tissue culture environment on microglia gene expression. (A) Comparison ofgene expression of ex vivo microglia from two individuals (panel i), effect of transition to an invitro environment in the presence of IL-34 for 7 days on microglia from each individual (panels ii andiii), and comparison of in vitro gene expression profiles (panel iv). (B) Scatter plot illustratingeffects of culture environment on the conserved microglial gene signature (red) in human microglia.(C) Pie chart of the genes associated with AD risk variants that are expressed in ex vivo or invitro microglia above a log2 (TPM +1) value of 2. Genes more highly expressed ex vivo are in blue, andgene more highly expressed in vitro are in purple. (D) Combined bar graphs and dot plots of the21 most expressed genes, indicating effects of in vitro culture on expression of genes associatedwith AD risk alleles. Red stars indicate significant differential expression (>2-fold, FDR 0.05).(E) Heat map illustrating changes in microglia gene expression 6 hours, 24 hours, and 7 daysafter transfer to culture environment.

RESEARCH | RESEARCH ARTICLEon January 6, 2021

http://science.sciencemag.org/

Dow

nloaded from

We performed ATAC-seq and ChIP-seq forH3K27ac in microglia placed into culture for7 days (Fig. 5C). Overall, of the 31,164 ATAC-seqpeaks associated with at least 16 normalizedH3K27ac tags (i.e., high confidence sites), ~14%exhibited a loss of H3K27ac and ~20% a gain ofH3K27ac (Fig. 5D), indicating remodeling of theregulatory landscape in vitro. Extending thisanalysis, 424 of the 740 SEs defined ex vivo werelost and 541 new SEs were observed (Fig. 5E).De novo motif analysis of the ATAC-seq peaksassociated with a >2-fold decrease in H3K27acreturned recognition elements for PU.1, IRF,CTCF, AP-1, SMAD, andMEF2motifs as themostsignificantly enriched (Fig. 5F), consistent withPU.1 establishing microglia enhancers and re-duced expression and/or activity of many of thecollaborative and signal-dependent factors thatrecognize these motifs (Fig. 5A). Notably, loss of

the SMADmotif is consistent with reduced TGF-bsignaling in the in vitro environment.We compared the expression of the mouse

orthologs of transcription factors with those pre-sent in primitive yolk sac macrophages and theirprogeny in developing and adult mouse brain(63). A subset of mRNAs encoding these factorswas expressed in primitive macrophages andmaintained high levels of expression in brainmicroglia (Fig. 6A). By contrast, an alternativesubset of factors that recognize DNA motifs en-riched in microglia enhancers and/or were pre-ferentially expressed in adult human and mousemicroglia exhibited increases in expression fromthe embryonic to the adult stage of brain devel-opment (Fig. 6A). Notably, a majority of the adultmicroglia transcription factors that exhibitedreduced expression after transfer to the in vitroenvironment were induced following migration

of primitive macrophages into the developingbrain, consistent with their induction by localenvironmental factors.We compared the set of genes robustly up-

regulated during brain development in mousemicroglia to those consistently down-regulatedupon transition to the in vitro environment. Ahighly significant overlap was observed: Of 3517genes that increased during development, 1459intersected with the 2398 genes “down” in cul-ture (P < 1E-308; Fig. 6B). This finding illustratesthat a substantial component of the microgliaunique signature that emerges during develop-ment is down-regulated in vitro.The correlation between genes up-regulated

in fetal microglia compared to yolk sac macro-phages and genes down-regulated upon trans-fer of adult microglia to an in vitro environmentprovided evidence that the changes observed in

Gosselin et al., Science 356, eaal3222 (2017) 23 June 2017 7 of 11

Fig. 5. Environment-dependentenhancer landscapes. (A) Heatmap illustrating the effect ofculture environment on humanand mouse microglia mRNAexpression of transcription factorsrecognizing motifs highly enrichedex vivo. (B) Heat map illustratingthe effect of culture environmenton human and mouse microgliamRNA expression of transcriptionfactors associated with SEsex vivo. (C) UCSC browser visual-ization of regulatory elementsnear genes SPI1 and P2RY12, andthe SALL1 SE in human microglia.Top panels display regions ofaccessible chromatin in microgliaex vivo, as defined by ATAC-seq.Bottom panels display H3K27acabundance ex vivo (blue) and invitro (yellow). (D) Fractions ofATAC-seq + H3K27ac regions inhuman microglia exhibiting similar,gained, or reduced H3K27ac signalafter transfer to a cultureenvironment for 7 days. (E) Venndiagram of SEs identified in ex vivoand in vitro human microgliabased on H3K27ac signal.(F) Motifs enrichment at distalaccessible chromatin regions inex vivo human microglia defined byATAC-seq that are associated withtwofold loss of H3K27ac signalsafter maintenance in cultureenvironment for 7 days.

RESEARCH | RESEARCH ARTICLEon January 6, 2021

http://science.sciencemag.org/

Dow

nloaded from

vitro result from loss of signals that are part ofthe brain environment. We cultured mouse mi-croglia in vitro for 7 days and treated them withTgf-b1 cytokine (Fig. 6C and fig. S15), which isnecessary for maintenance of the in vivo mousemicroglia phenotype (27, 64). Of the 2398 genesdown-regulated >2-fold in vitro, Tgf-b1 inducedexpression >2-fold of 429, including Sall1, Tmem119,and P2ry12. Corresponding studies of TGF-b1 treat-

ment of human microglia in vitro yielded similarresults, although variation between individualslimited the number of genes achieving statisticalsignificance after correction for multiple testing.Nonetheless, SALL1was among the genes exhibit-ing significant up-regulation (fig. S15B). Becauseof the relatively modest effects of TGF-b1, weevaluated several other tissue culture conditionsto investigate whether any might be more effec-

tive in maintaining an ex vivo gene expressionsignature. Although most treatment conditionsproduced some effect on gene expression, noneresulted in preservation of the ex vivo molecularphenotype (fig. S16). It has not yet been possibleto maintain primary humanmicroglia in a serum-free environment. Recent studies have derivedmicroglia-like cells from human induced pluri-potent stem cells in a defined media that more

Gosselin et al., Science 356, eaal3222 (2017) 23 June 2017 8 of 11

Fig. 6. A network of develop-mentally programmed andenvironment-dependent tran-scription factors establishesmicroglia identity and functions.(A) Heat map displaying changes inmRNA expression of key microgliatranscription factors as a functionof development of the mouse brain.(B) Venn diagram illustrating sig-nificant overlap of mouse microgliagenes repressed (>2-fold,FDR 0.05) in culture and genesthat increased (>2-fold, FDR 0.05)in expression during brain develop-ment. P-value of the overlap isprovided (Fisher’s exact test).(C) Venn diagram illustrating sig-nificant overlap of mouse microgliagenes repressed (>2-fold,FDR 0.05) in culture after 7 daysand genes maintained or induced(>2-fold, FDR 0.05) by chronicstimulation with Tgf-b1. P-value ofthe overlap is provided (Fisher’sexact test). (D) Pie chart ofmicroglia-enriched genes com-pared to brain cortex overlappingwith differentially expressedgenes in neurodegenerative orbehavioral diseases from Fig. 1Ccolor-coded according to whetherthey are increased, decreased,or unchanged after transfer to atissue culture environment.(E) Diagram illustrating mainfeatures of transcription factornetwork regulating microglia cellidentity and functions.

RESEARCH | RESEARCH ARTICLEon January 6, 2021

http://science.sciencemag.org/

Dow

nloaded from

closely resembles cerebrospinal fluid (65). How-ever, these cells exhibited a transcriptomic phe-notype that was much more similar to humanmicroglia cultured in vitro than to ex vivo microg-lia (fig. S17).Lastly,we investigatedwhether genes expressed

at least 10-fold higher in ex vivo microglia com-pared to brain cortex tissue and that are as-sociated with differentially regulated genes inneurodegenerative or behavioral disorders (Fig.1C) were sensitive to the environment. Of thenonredundant set of 291 microglia transcriptsassociated with differentially regulated genesin AD, PD, HAND, Scz, or FTD, 117 were down-regulated in vitro whereas 29 were up-regulatedin vitro (Fig. 6D). These observations provide evi-dence that differential expression of microglia-enrichedgenes in thesedisorders is at leastpartiallydue to changes in brain environment.

Discussion

The transcriptomes and epigenomic features ofhuman microglia show that disease status, age,sex, and treatment history had a relatively minorinfluence on the overall pattern of gene expres-sion. It is thus likely that the composite transcript-omic and epigenomic profiles identified largelycapture a representative human microglial genesignature and its underlying network of genomicregulatory elements. We find that mouse and hu-man microglia transcriptomes are largely con-served, but they also exhibit large differencesin the expression of several hundred genes thatappear to be primarily driven by cis-regulatorydifferences.Genetic evidence suggests thatmicroglia dysreg-

ulation is a core feature of numerous chronicneurodegenerative and psychiatric diseases. Tothis end, our profiling of the human microgliatranscriptome enabled us to estimate a weightedinfluence of microglia on the development of dif-ferent brain disorders. In addition, by defining amicroglial gene signature in the human brain, itis possible to link changes in gene expression inneurodegenerative and behavioral disorders toalterations in the number and/or function ofmicroglia. The observation that the microglialgene signature overlaps with genes that are bothup- and down-regulated in PD provides strongevidence that differential gene expression at thewhole-tissue level reflects changes in microgliagene expression and not simply changes in themicroglia population number. This interpreta-tion is further reinforced by the observation thatmany of themicroglia-enriched genes that changein the context of neurodegenerative and be-havioral disorders also show significant changesin expression upon transfer to an in vitro en-vironment. Although we also observed substan-tial individual variation in genes linked todisease risk, more individuals will be required topower analyses linking genetic variation to geneexpression.Examining differences in ex vivo and in vitro

microglia, we observe rapid and profound alter-ations in gene expression by 6 hours, charac-terized by activation of genes associated with

acute inflammatory responses and decreasedexpression of genes associated with diverse im-mune functions. The incomplete recovery ofdown-regulated genes at 7 days was at least inpart due to the absence of brain-specific signals.At present, the spectrum of brain-specific signalsthat are necessary to maintain microglia pheno-types remains largely unknown, but this findinghighlights the limitations of investigatingmicrog-lia biology in a tissue culture setting. Notably,current efforts to reprogram various starting cellpopulations to microglia-like cells largely achievean in vitro phenotype. It is likely that more com-plex culture conditions, such as co-culture withneurons and astrocytes in three-dimensional ororganoid contexts, will be required to more pre-cisely model the in vivo phenotype. Integrationof these findings with recent studies of yolk sacand embryonicmicroglia (56, 63) suggests amodelby which intersecting developmentally deter-mined and environment-dependent transcrip-tional networks establish microglia identity andfunction (Fig. 6E). We speculate that regulato-ry elements are targets of environmental per-turbations associated with disease and are alsoimportant sites of action of natural geneticvariation.In sum, ourworkdemonstrates that humanand

mousemicroglia display relativelywell-conservedtranscriptomic and epigenomic phenotypes, andit provides a critical resource that will allow for abetter understanding ofmousemodels of humanbrain disorders, guide reprogramming efforts forgeneration of human microglia in vitro, and en-able a better interpretation of the functions ofDNA variants associated with disorders of thecentral nervous system.

MethodsHuman tissue

Microglia were isolated from brain tissue (inexcess of that needed for pathological diagno-sis) resected for treatment of epilepsy, braintumors, or acute ischemia in 19 individuals. Forpatients with epilepsy (9), the resected tissue dis-played epileptogenic activity. However, the epi-leptic focus exhibiting the strongest epileptogenicactivity was not provided and was separated atthe time of resection for pathological analysis.For patients with brain tumors (8), samples con-sisted of incidentally resected brain tissue, i.e.,brain tissue resected to gain access to the tumorand considered to be tumor negative by the neu-rosurgeon intraoperatively. Two brain sampleswere resected to treat acute ischemic injuriesfrom individuals who had no prior evidence ofbrain pathology. Brain tissue was obtained withinformed consent under a protocol approvedby the UC San Diego and Rady Children’s HospitalInstitutional Review Board. Resected brain tissuewas immediately placed on ice and transferredto the laboratory for microglia isolation within3 hours after resection.

Human and mouse microglia isolation

Resected human brain tissues and mouse brainswere homogenized by gentle mechanical disso-

ciation, as performed in Gosselin et al. (25). Mi-croglia were then enriched by Percoll gradient,washed, and stained with cell surface markersfor final purification by flow cytometry, using aBDInflux cell sorter.Humanmicrogliaweredefinedas live/DAPI– CD11b+CD45LowCD64+CX3CR1High

single cells. C57BL/6 mouse microglia were de-fined as live/DAPI– CD11b+CD45Low single cells.

Primary culture of isolated microglia

Isolated microglia were maintained in culturefor 6 hours, 24 hours, or 7 days. Culture mediaconsisted of Dulbecco’s modified Eagle medium(DMEM)/F12 (Life Technologies, 113300-032)supplemented with 5% fetal bovine serum (heatinactivated, Omega Scientific, FB-12) and 1× Anti-Anti (Life Technologies, 152240-062). No mediachanges were performed for the duration of theculture experiments. CSF1R ligand Interleukin-34 (R&D, human 5265-IL-010; mouse 5195-ML-010) was supplemented daily at a concentrationof 20 ng/ml. For experiments investigating theeffects of TGF-b1 on microglia gene expressionin culture, recombinant human or mouse TGF-b1 cytokine (R&D, human 240-B-002; mouse 7666-MB-005) was added daily at a concentration of10 ng/ml.

RNA-seq library preparation: Isolationand fragmentation of poly(A) RNA andsynthesis of cDNA

Isolated microglia were pelleted and put into150 ml of lysis/Oligo d(T) Magnetic Beads bindingbuffer and stored at –80°C until processing. Forhuman brain cortical RNA, 500 ng of RNA wasdiluted with proper volume of 2× lysis/Oligo d(T)Magnetic Beads binding buffer to a final concen-tration of 1× lysis/Oligo d(T) Magnetic Beadsbinding buffer. mRNAs were enriched by incu-bation with Oligo d(T) Magnetic Beads (NEB,S1419S) and then fragmented and eluted by in-cubation at 94°C for 9 min. cDNA was then syn-thesized with Superscript III (first-strand synthesis;Life Technologies kit 18080-044) and then DNAPolymerase I (second-strand synthesis; Enzymat-ics P7050L)

Chromatin immunoprecipitation

For ChIP-seq experiments, a cross-linking stepwith paraformaldehyde was performed imme-diately after staining but before microglia sort-ing. Chromatin immunoprecipitation for histonemodification H3K4me2, H3K27ac, and transcrip-tion factor PU.1 was performed as described inGosselin et al. (25).

RNA-seq and ChIP-seq final librarypreparation and sequencing

Sequencing libraries were prepared from recov-ered DNA (ChIP) or generated cDNA (RNA) byblunting, A-tailing, and adapter ligation as pre-viously described in Heinz et al. (31). Prior tofinal polymerase chain reaction amplification,RNA-seq libraries were digested by 30 min ofincubation at 37°C with Uracil DNA Glycosylase.RNA-seq and ChIP-seq libraries were single-endsequenced for 51 cycles on an Illumina HiSeq

Gosselin et al., Science 356, eaal3222 (2017) 23 June 2017 9 of 11

RESEARCH | RESEARCH ARTICLEon January 6, 2021

http://science.sciencemag.org/

Dow

nloaded from

4000 or NextSeq 500 (Illumina, San Diego, CA)according to the manufacturer’s instruction.

Data analysisPreprocessing

FASTQ files from sequencing experiments weremapped to either the UCSC genome build hg38(for human) or mm10 (for mouse). STARwith de-fault parameters was used to map RNA-seq ex-periments. Bowtie2 with default parameters wasused tomapATAC-seq andChIP-seq experiments.HOMER was used to convert aligned reads into“tag directories” for further analysis.

RNA-seq

The “analyzeRepeats” function of HOMER wasused to generate a table of read counts for each ex-periment using theparameters “-raw -count exons-strand both -condenseGenes” and a table of tran-scripts per million (TPM) values using “-tpm-normMatrix 1e7 -count exons -strand both-condenseGenes.” The base-2 logarithm of theTPM values was taken after adding a pseudo-count of 1 TPM to each gene. Within-species dif-ferential gene expression analysis was performedusing the R package limma with the voom nor-malization.Mappingof orthologous genes betweenspecies was done using the one-to-one orthologsfrom theEnsembl (version 84) Compara database.

ChIP-seq

Tag directories for the ChIP-seq experimentswerecombined into ex vivo and in vitro pools for eachtarget in both mouse and human. The tag direc-tories for the input DNA were likewise com-bined. Peaks were then called on the pooled tagswith the pooled input DNA as background usingHOMER’s “findPeaks” command with the fol-lowing parameters for the PU.1 ChIP-seq: “-stylefactor -size 200 -minDist 200”; and the follow-ing parameters for the H3K27ac and H3K4me2ChIP-seq: “-style histone -size 500 -minDist 1000-region.” The “-tbp” parameter was set to the num-ber of replicates in the pool.

Motif enrichment

Peaks were considered overlapping if theircoordinates overlapped by at least one base pair.De novomotif enrichment analysiswas performedon ATAC-seq and PU.1 ChIP-seq peaks that over-lapped either H3K4me2 (ATAC-seq) or H3K27ac(PU.1), usingHOMER’s “findMotifsGenome” com-mandwith the following parameters: “-size given-len 14,12,10,8 -mask.”

REFERENCES AND NOTES

1. S. Rivest, Regulation of innate immune responses in the brain.Nat. Rev. Immunol. 9, 429–439 (2009). doi: 10.1038/nri2565;pmid: 19461673

2. A. Katsumoto, H. Lu, A. S. Miranda, R. M. Ransohoff, Ontogenyand functions of central nervous system macrophages.J. Immunol. 193, 2615–2621 (2014). doi: 10.4049/jimmunol.1400716; pmid: 25193935

3. A. Shemer, D. Erny, S. Jung, M. Prinz, Microglia Plasticity DuringHealth and Disease: An Immunological Perspective. TrendsImmunol. 36, 614–624 (2015). doi: 10.1016/j.it.2015.08.003;pmid: 26431939

4. D. P. Schafer, B. Stevens, Microglia Function in CentralNervous System Development and Plasticity. Cold Spring Harb.

Perspect. Biol. 7, a020545 (2015). doi: 10.1101/cshperspect.a020545; pmid: 26187728

5. A. M. Casano, F. Peri, Microglia: Multitasking specialists of the brain.Dev. Cell 32, 469–477 (2015). doi: 10.1016/j.devcel.2015.01.018;pmid: 25710533

6. T. M. Malm, T. R. Jay, G. E. Landreth, The evolving biology ofmicroglia in Alzheimer’s disease. Neurotherapeutics 12, 81–93(2015). doi: 10.1007/s13311-014-0316-8; pmid: 25404051

7. N. C. Derecki, N. Katzmarski, J. Kipnis, M. Meyer-Luehmann,Microglia as a critical player in both developmental and late-lifeCNS pathologies. Acta Neuropathol. 128, 333–345 (2014).doi: 10.1007/s00401-014-1321-z; pmid: 25056803

8. K. I. Mosher, T. Wyss-Coray, Microglial dysfunction in brainaging and Alzheimer’s disease. Biochem. Pharmacol. 88,594–604 (2014). doi: 10.1016/j.bcp.2014.01.008;pmid: 24445162

9. E. C. Hirsch, S. Vyas, S. Hunot, Neuroinflammation inParkinson’s disease. Parkinsonism Relat. Disord. 18 (suppl. 1),S210–S212 (2012). doi: 10.1016/S1353-8020(11)70065-7;pmid: 22166438

10. E. Croisier, L. B. Moran, D. T. Dexter, R. K. Pearce,M. B. Graeber, Microglial inflammation in the parkinsoniansubstantia nigra: Relationship to alpha-synuclein deposition.J. Neuroinflammation 2, 14 (2005). doi: 10.1186/1742-2094-2-14;pmid: 15935098

11. L. M. Collins, A. Toulouse, T. J. Connor, Y. M. Nolan,Contributions of central and systemic inflammation to thepathophysiology of Parkinson’s disease. Neuropharmacology62, 2154–2168 (2012). doi: 10.1016/j.neuropharm.2012.01.028;pmid: 22361232

12. K. Saijo et al., A Nurr1/CoREST pathway in microglia and astrocytesprotects dopaminergic neurons from inflammation-induced death.Cell 137, 47–59 (2009). doi: 10.1016/j.cell.2009.01.038;pmid: 19345186

13. A. Crotti et al., Mutant Huntingtin promotes autonomousmicroglia activation via myeloid lineage-determining factors.Nat. Neurosci. 17, 513–521 (2014). doi: 10.1038/nn.3668;pmid: 24584051

14. M. Politis et al., Microglial activation in regions related tocognitive function predicts disease onset in Huntington’sdisease: A multimodal imaging study. Hum. Brain Mapp. 32,258–270 (2011). doi: 10.1002/hbm.21008; pmid: 21229614

15. Y. F. Tai et al., Microglial activation in presymptomaticHuntington’s disease gene carriers. Brain 130, 1759–1766(2007). doi: 10.1093/brain/awm044; pmid: 17400599

16. P. M. Bilimoria, B. Stevens, Microglia function during braindevelopment: New insights from animal models. Brain Res.1617, 7–17 (2015). pmid: 25463024

17. L. R. Frick, K. Williams, C. Pittenger, Microglial dysregulation inpsychiatric disease. Clin. Dev. Immunol. 2013, 608654 (2013).doi: 10.1155/2013/608654; pmid: 23690824

18. F. Ginhoux et al., Fate mapping analysis reveals that adultmicroglia derive from primitive macrophages. Science 330,841–845 (2010). doi: 10.1126/science.1194637;pmid: 20966214

19. C. Schulz et al., A lineage of myeloid cells independent of Myband hematopoietic stem cells. Science 336, 86–90 (2012).doi: 10.1126/science.1219179; pmid: 22442384

20. M. R. Elmore et al., Colony-stimulating factor 1 receptorsignaling is necessary for microglia viability, unmasking amicroglia progenitor cell in the adult brain. Neuron 82,380–397 (2014). doi: 10.1016/j.neuron.2014.02.040;pmid: 24742461

21. A. M. Smith et al., The transcription factor PU.1 is critical forviability and function of human brain microglia. Glia 61,929–942 (2013). doi: 10.1002/glia.22486; pmid: 23483680

22. K. Kierdorf et al., Microglia emerge from erythromyeloidprecursors via Pu.1- and Irf8-dependent pathways.Nat. Neurosci. 16, 273–280 (2013). doi: 10.1038/nn.3318;pmid: 23334579

23. C. K. Glass, K. Saijo, B. Winner, M. C. Marchetto, F. H. Gage,Mechanisms underlying inflammation in neurodegeneration.Cell 140, 918–934 (2010). doi: 10.1016/j.cell.2010.02.016;pmid: 20303880

24. E. L. Gautier et al., Gene-expression profiles and transcriptionalregulatory pathways that underlie the identity and diversityof mouse tissue macrophages. Nat. Immunol. 13, 1118–1128(2012). doi: 10.1038/ni.2419; pmid: 23023392

25. D. Gosselin et al., Environment drives selection and function ofenhancers controlling tissue-specific macrophage identities.Cell 159, 1327–1340 (2014). doi: 10.1016/j.cell.2014.11.023;pmid: 25480297

26. Y. Lavin et al., Tissue-resident macrophage enhancerlandscapes are shaped by the local microenvironment. Cell

159, 1312–1326 (2014). doi: 10.1016/j.cell.2014.11.018;pmid: 25480296

27. O. Butovsky et al., Identification of a unique TGF-b-dependentmolecular and functional signature in microglia. Nat. Neurosci.17, 131–143 (2014). doi: 10.1038/nn.3599; pmid: 24316888

28. D. P. Schafer et al., Microglia sculpt postnatal neural circuits inan activity and complement-dependent manner. Neuron 74,691–705 (2012). doi: 10.1016/j.neuron.2012.03.026;pmid: 22632727

29. C. N. Parkhurst et al., Microglia promote learning-dependentsynapse formation through brain-derived neurotrophic factor.Cell 155, 1596–1609 (2013). doi: 10.1016/j.cell.2013.11.030;pmid: 24360280

30. S. Heinz, C. E. Romanoski, C. Benner, C. K. Glass, The selection andfunction of cell type-specific enhancers. Nat. Rev. Mol. Cell Biol.16, 144–154 (2015). doi: 10.1038/nrm3949; pmid: 25650801

31. S. Heinz et al., Simple combinations of lineage-determiningtranscription factors prime cis-regulatory elements required formacrophage and B cell identities. Mol. Cell 38, 576–589(2010). doi: 10.1016/j.molcel.2010.05.004; pmid: 20513432

32. S. Heinz et al., Effect of natural genetic variation on enhancerselection and function. Nature 503, 487–492 (2013).doi: 10.1038/nature12615; pmid: 24121437

33. M. U. Kaikkonen et al., Remodeling of the enhancer landscapeduring macrophage activation is coupled to enhancertranscription. Mol. Cell 51, 310–325 (2013).doi: 10.1016/j.molcel.2013.07.010; pmid: 23932714

34. A. Kundaje et al., Integrative analysis of 111 reference humanepigenomes. Nature 518, 317–330 (2015).doi: 10.1038/nature14248; pmid: 25693563

35. C. E. Romanoski, C. K. Glass, H. G. Stunnenberg, L. Wilson,G. Almouzni, Epigenomics: Roadmap for regulation. Nature518, 314–316 (2015). doi: 10.1038/518314a; pmid: 25693562

36. M. Kasowski et al., Extensive variation in chromatin statesacross humans. Science 342, 750–752 (2013). doi: 10.1126/science.1242510; pmid: 24136358

37. G. McVicker et al., Identification of genetic variants that affecthistone modifications in human cells. Science 342, 747–749(2013). doi: 10.1126/science.1242429; pmid: 24136359

38. H. Kilpinen et al., Coordinated effects of sequence variation onDNA binding, chromatin structure, and transcription. Science342, 744–747 (2013). doi: 10.1126/science.1242463;pmid: 24136355

39. D. Avramopoulos, M. Szymanski, R. Wang, S. Bassett, Geneexpression reveals overlap between normal aging and Alzheimer’sdisease genes. Neurobiol. Aging 32, 2319.e27–2319.e34 (2011).doi: 10.1016/j.neurobiolaging.2010.04.019; pmid: 20570407

40. E. M. Blalock et al., Incipient Alzheimer’s disease: Microarraycorrelation analyses reveal major transcriptional and tumorsuppressor responses. Proc. Natl. Acad. Sci. U.S.A. 101, 2173–2178(2004). doi: 10.1073/pnas.0308512100; pmid: 14769913

41. A. Capurro, L. G. Bodea, P. Schaefer, R. Luthi-Carter,V. M. Perreau, Computational deconvolution of genome wideexpression data from Parkinson’s and Huntington’s diseasebrain tissues using population-specific expression analysis.Front. Neurosci. 8, 441 (2015). pmid: 25620908

42. V. Colangelo et al., Gene expression profiling of 12633 genes inAlzheimer hippocampal CA1: Transcription and neurotrophicfactor down-regulation and up-regulation of apoptotic and pro-inflammatory signaling. J. Neurosci. Res. 70, 462–473(2002). doi: 10.1002/jnr.10351; pmid: 12391607

43. W. S. Liang et al., Altered neuronal gene expression in brainregions differentially affected by Alzheimer’s disease: Areference data set. Physiol. Genomics 33, 240–256 (2008).doi: 10.1152/physiolgenomics.00242.2007; pmid: 18270320

44. J. N. McClintick et al., Ethanol treatment of lymphoblastoid celllines from alcoholics and non-alcoholics causes many subtlechanges in gene expression. Alcohol 48, 603–610 (2014).doi: 10.1016/j.alcohol.2014.07.004; pmid: 25129674

45. M. Mishra et al., Gene expression analysis of frontotemporallobar degeneration of the motor neuron disease type withubiquitinated inclusions. Acta Neuropathol. 114, 81–94 (2007).doi: 10.1007/s00401-007-0240-7; pmid: 17569064

46. J. Satoh, Y. Yamamoto, N. Asahina, S. Kitano, Y. Kino, RNA-Seqdata mining: Downregulation of NeuroD6 serves as a possiblebiomarker for alzheimer’s disease brains. Dis. Markers 2014,123165 (2014). doi: 10.1155/2014/123165; pmid: 25548427

47. J. A. Webster et al., Genetic control of human brain transcriptexpression in Alzheimer disease. Am. J. Hum. Genet. 84,445–458 (2009). doi: 10.1016/j.ajhg.2009.03.011;pmid: 19361613

48. A. Borjabad et al., Significant effects of antiretroviral therapyon global gene expression in brain tissues of patients with

Gosselin et al., Science 356, eaal3222 (2017) 23 June 2017 10 of 11

RESEARCH | RESEARCH ARTICLEon January 6, 2021

http://science.sciencemag.org/

Dow

nloaded from

HIV-1-associated neurocognitive disorders. PLOS Pathog. 7,e1002213 (2011). doi: 10.1371/journal.ppat.1002213;pmid: 21909266

49. P. R. Maycox et al., Analysis of gene expression in two largeschizophrenia cohorts identifies multiple changes associated withnerve terminal function. Mol. Psychiatry 14, 1083–1094 (2009).doi: 10.1038/mp.2009.18; pmid: 19255580

50. B. Zhang et al., Integrated systems approach identifies geneticnodes and networks in late-onset Alzheimer’s disease. Cell153, 707–720 (2013). doi: 10.1016/j.cell.2013.03.030;pmid: 23622250

51. B. E. Riley et al., Systems-based analyses of brain regionsfunctionally impacted in Parkinson’s disease reveals underlyingcausal mechanisms. PLOS ONE 9, e102909 (2014).doi: 10.1371/journal.pone.0102909; pmid: 25170892

52. A. S. Chen-Plotkin et al., Variations in the progranulin geneaffect global gene expression in frontotemporal lobardegeneration. Hum. Mol. Genet. 17, 1349–1362 (2008).doi: 10.1093/hmg/ddn023; pmid: 18223198

53. J. A. Miller, R. L. Woltjer, J. M. Goodenbour, S. Horvath,D. H. Geschwind, Genes and pathways underlying regional andcell type changes in Alzheimer’s disease. Genome Med. 5,48 (2013). doi: 10.1186/gm452; pmid: 23705665

54. D. Welter et al., The NHGRI GWAS Catalog, a curated resourceof SNP-trait associations. Nucleic Acids Res. 42, D1001–D1006(2014). doi: 10.1093/nar/gkt1229; pmid: 24316577

55. T. C. Burns, M. D. Li, S. Mehta, A. J. Awad, A. A. Morgan, Mousemodels rarely mimic the transcriptome of humanneurodegenerative diseases: A systematic bioinformatics-basedcritique of preclinical models. Eur. J. Pharmacol. 759, 101–117(2015). doi: 10.1016/j.ejphar.2015.03.021;pmid: 25814260

56. E. Mass et al., Specification of tissue-resident macrophagesduring organogenesis. Science 353, aaf4238 (2016).doi: 10.1126/science.aaf4238; pmid: 27492475

57. I. M. Chiu et al., A neurodegeneration-specific gene-expressionsignature of acutely isolated microglia from an amyotrophiclateral sclerosis mouse model. Cell Reports 4, 385–401 (2013).doi: 10.1016/j.celrep.2013.06.018; pmid: 23850290

58. S. E. Hickman et al., The microglial sensome revealed by directRNA sequencing. Nat. Neurosci. 16, 1896–1905 (2013).doi: 10.1038/nn.3554; pmid: 24162652

59. J. C. Lambert et al., Meta-analysis of 74,046 individualsidentifies 11 new susceptibility loci for Alzheimer’s disease.Nat. Genet. 45, 1452–1458 (2013). doi: 10.1038/ng.2802;pmid: 24162737

60. W. A. Whyte et al., Master transcription factors and mediatorestablish super-enhancers at key cell identity genes. Cell 153,307–319 (2013). doi: 10.1016/j.cell.2013.03.035;pmid: 23582322

61. J. Lovén et al., Selective inhibition of tumor oncogenes bydisruption of super-enhancers. Cell 153, 320–334 (2013).doi: 10.1016/j.cell.2013.03.036; pmid: 23582323

62. D. Hnisz et al., Super-enhancers in the control of cell identity anddisease. Cell 155, 934–947 (2013). doi: 10.1016/j.cell.2013.09.053;pmid: 24119843

63. O. Matcovitch-Natan et al., Microglia development follows astepwise program to regulate brain homeostasis. Science 353,aad8670 (2016). doi: 10.1126/science.aad8670;pmid: 27338705

64. A. Buttgereit et al., Sall1 is a transcriptional regulator definingmicroglia identity and function. Nat. Immunol. 17, 1397–1406(2016). doi: 10.1038/ni.3585; pmid: 27776109

65. J. Muffat et al., Efficient derivation of microglia-like cells fromhuman pluripotent stem cells. Nat. Med. 22, 1358–1367(2016). doi: 10.1038/nm.4189; pmid: 27668937

ACKNOWLEDGMENTS

We thank J. Collier for technical assistance, L. Van Ael forassistance with manuscript preparation, and M. L. Gage forassistance with manuscript editing. We also thank D. Malicki forassistance in obtaining patient samples for Iba-1 immunohistologicalanalysis. We thank J. A. Miller (Allen Institute for Brain Science) forassistance with the analysis of brain disease-associated gene sets.These studies were supported by the Larry L. Hillbloom Foundation;NIH grants NS096170, DK063491, and GM085764; and by the FlowCytometry Core Facility of the Salk Institutes with funding from NIH–

National Cancer Institute CCSG: P30 014195. D.G. was supportedby a Canadian Institutes of Health Research fellowship and a MultipleSclerosis Society of Canada fellowship. D.S. was supported byUniversity of California, San Diego, institutional predoctoral traininggrant (T32DK007541). I.R.H. was supported by University MedicalCenter Groningen Institutional postdoctoral traveling grant, the DutchMS Research Foundation, and the Gemmy and Mibeth TichelaarFoundation. J.C.M.S. was supported by the DeutscheForschungsgemeinschaft (SCHL2102/1-1.) B.N.J. was supportedby a European Molecular Biology Organization Long-term Fellowship,the Bettencourt-Schueller Foundation, and the Philippe Foundation.F.H.G. is supported by the JPB Foundation, Dolby FamilyVentures, The Paul G. Allen Family Foundation, and the EngmanFoundation. C.K.G. is supported by the Ben and Wanda Hildyard Chairin Hereditary Diseases. D.G., N.G.C., F.H.G., and C.K.G. conceivedthe study. N.G.C. coordinated tissue acquisition, clinical chart review,and institutional review board approval. M.L.L. and D.D.G. obtainedpatients’ informed consent, and resected and evaluated brain tissue.D.G. and J.C.M.S. isolated microglia, with assistance from B.N.J.,M.P., and A.A. D.G., J.C.M.S., C.O., C.F., and M.P.P. performedmicroglia sorting. J.C.M.S. isolated monocytes. N.G.C. and M.P.performed histological analyses. D.G. performed culture experiments.D.G., J.C.M.S., and E.S. prepared, respectively, microglia, monocyte,and intact brain sequencing libraries. D.G. and C.O. analyzed flowcytometry data. D.G., D.S., I.R.H., and C.K.G. analyzed sequencingdata. D.G., I.H.R., D.S., and C.K.G. wrote the manuscript, withcontributions from N.G.G., R.M.R., and F.H.G.

SUPPLEMENTARY MATERIALS

www.sciencemag.org/content/356/6344/eaal3222/suppl/DC1Materials and MethodsFigs. S1 to S17Tables S1 to 6References (66–76)

6 November 2016; accepted 12 May 2017Published online 25 May 201710.1126/science.aal3222

Gosselin et al., Science 356, eaal3222 (2017) 23 June 2017 11 of 11

RESEARCH | RESEARCH ARTICLEon January 6, 2021

http://science.sciencemag.org/

Dow

nloaded from

An environment-dependent transcriptional network specifies human microglia identity

Richard M. Ransohoff, Fred H. Gage and Christopher K. GlassCarolyn O'Connor, Conor Fitzpatrick, Martina P. Pasillas, Monique Pena, Amy Adair, David D. Gonda, Michael L. Levy, David Gosselin, Dylan Skola, Nicole G. Coufal, Inge R. Holtman, Johannes C. M. Schlachetzki, Eniko Sajti, Baptiste N. Jaeger,

originally published online May 25, 2017DOI: 10.1126/science.aal3222 (6344), eaal3222.356Science 

, this issue p. eaal3222Sciencehumans and mice or after culturing were often implicated in neurodegenerative diseases.surgery (ex vivo) or after culturing (in vitro). Interestingly, those genes exhibiting differences in expression between expression profiles, as well as a shared environmental response among microglia collected either immediately afterdifferences in age, sex, and diagnosis. Mouse and human microglia demonstrated similar microglia-specific gene

despitethe epigenetics and RNA transcripts from single microglial cells and observed consistent profiles among samples examinedet al.Microglia are immune system cells that function in protecting and maintaining the brain. Gosselin

Of mice and men's microglia

ARTICLE TOOLS http://science.sciencemag.org/content/356/6344/eaal3222

MATERIALSSUPPLEMENTARY http://science.sciencemag.org/content/suppl/2017/05/24/science.aal3222.DC1

CONTENTRELATED

http://stke.sciencemag.org/content/sigtrans/12/569/eaar2124.fullhttp://stm.sciencemag.org/content/scitransmed/8/369/369ra178.full

REFERENCES

http://science.sciencemag.org/content/356/6344/eaal3222#BIBLThis article cites 74 articles, 11 of which you can access for free

PERMISSIONS http://www.sciencemag.org/help/reprints-and-permissions

Terms of ServiceUse of this article is subject to the

is a registered trademark of AAAS.ScienceScience, 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 U.S. Government WorksCopyright © 2017 The Authors, some rights reserved; exclusive licensee American Association for the Advancement of

on January 6, 2021

http://science.sciencemag.org/

Dow

nloaded from