Search

Base editors are capable of installing precise genomic alterations without creating double-strand DNA breaks. In this study, we targeted critical motifs regulating γ-globin reactivation with base editors delivered via HDAd5/35++ vectors. Through optimized design, we successfully produced a panel of cytidine and adenine base editor (ABE) vectors targeting the erythroid BCL11A enhancer or recreating naturally occurring hereditary persistence of fetal hemoglobin (HPFH) mutations in the HBG1/2 promoter. All 5 tested vectors efficiently installed target base conversion and led to γ-globin reactivation in human erythroid progenitor cells. We observed ~23% γ-globin protein production over β-globin, when using an ABE vector (HDAd-ABE-sgHBG-2) specific to the –113A>G HPFH mutation. In a β-YAC mouse model, in vivo hematopoietic progenitor/stem cell (HSPC) transduction with HDAd-ABE-sgHBG-2 followed by in vivo selection resulted in >40% γ-globin+ erythrocytes in the peripheral blood. This result corresponded to 21% γ-globin production over human β-globin. The average –113A>G conversion in total bone marrow cells was 20%. No alterations in hematological parameters, erythropoiesis, and bone marrow cellular composition were observed after treatment. No detectable editing was found at top-scoring, off-target genomic sites. Bone marrow lineage–negative cells from primary mice were capable of reconstituting secondary transplant-recipient mice with stable γ-globin expression. Importantly, the advantage of base editing over CRISPR/Cas9 was reflected by the markedly lower rates of intergenic HBG1/2 deletion and the absence of detectable toxicity in human CD34+ cells. Our observations suggest that HDAd-vectorized base editors represent a promising strategy for precise in vivo genome engineering for the treatment of β-hemoglobinopathies.

Our goal is the development of in vivo hematopoietic stem cell (HSC) transduction technology with targeted integration. To achieve this, we modified helper-dependent HDAd5/35++ vectors to express a CRISPR/Cas9 specific to the “safe harbor” adeno-associated virus integration site 1 (AAVS1) locus and to provide a donor template for targeted integration through homology-dependent repair. We tested the HDAd-CRISPR + HDAd-donor vector system in AAVS1 transgenic mice using a standard ex vivo HSC gene therapy approach as well as a new in vivo HSC transduction approach that involves HSC mobilization and intravenous HDAd5/35++ injections. In both settings, the majority of treated mice had transgenes (GFP or human γ-globin) integrated into the AAVS1 locus. On average, >60% of peripheral blood cells expressed the transgene after in vivo selection with low-dose O(6)BG/bis-chloroethylnitrosourea (BCNU). Ex vivo and in vivo HSC transduction and selection studies with HDAd-CRISPR + HDAd-globin-donor resulted in stable γ-globin expression at levels that were significantly higher (>20% γ-globin of adult mouse globin) than those achieved in previous studies with a SB100x-transposase-based HDAd5/35++ system that mediates random integration. The ability to achieve therapeutically relevant transgene expression levels after in vivo HSC transduction and selection and targeted integration make our HDAd5/35++-based vector system a new tool in HSC gene therapy.

Summary: During mammalian embryogenesis, changes in morphology and gene expression are concurrent with epigenomic reprogramming. Using human embryonic stem cells representing the preimplantation blastocyst (naive) and postimplantation epiblast (primed), our data in 2iL/I/F naive cells demonstrate that a substantial portion of known human enhancers are premarked by H3K4me1, providing an enhanced open chromatin state in naive pluripotency. The 2iL/I/F enhancer repertoire occupies 9% of the genome, three times that of primed cells, and can exist in broad chromatin domains over 50 kb. Enhancer chromatin states are largely poised. Seventy-seven percent of 2iL/I/F enhancers are decommissioned in a stepwise manner as cells become primed. While primed topologically associating domains are largely unaltered upon differentiation, naive 2iL/I/F domains expand across primed boundaries, affecting three-dimensional genome architecture. Differential topologically associating domain edges coincide with 2iL/I/F H3K4me1 enrichment. Our results suggest that naive-derived 2iL/I/F cells have a unique chromatin landscape, which may reflect early embryogenesis. : Human ESCs grown in 2iL/I/F are useful for studying the transition from naive to primed pluripotency. Battle et al. show that the naive epigenome is globally enhanced for H3K4me1. Many of the H3K4me1 regions are putative enhancers and are lost, or decommissioned, during the transition to primed. Differential topologically associating domains (TADs) exist between states and contain H3K4me1 enrichment at TAD boundaries. Keywords: enhancers, 3D genome architecture, naive hESCs, embryogenesis

While the chicken (Gallus gallus) is the most consumed agricultural animal worldwide, the chicken transcriptome remains understudied. We have characterized the transcriptome of 10 cell and tissue types from the chicken using RNA-seq, spanning intestinal tissues (ileum, jejunum, proximal cecum), immune cells (B cells, bursa, macrophages, monocytes, spleen T cells, thymus), and reproductive tissue (ovary). We detected 17,872 genes and 24,812 transcripts across all cell and tissue types, representing 73% and 63% of the current gene annotation, respectively. Further quantification of RNA transcript biotypes revealed protein-coding and lncRNAs specific to an individual cell/tissue type. Each cell/tissue type also has an average of around 1.2 isoforms per gene, however, they all have at least one gene with at least 11 isoforms. Differential expression analysis revealed a large number of differentially expressed genes between tissues of the same category (immune and intestinal). Many of these differentially expressed genes in immune cells were involved in cellular processes relating to differentiation and cell metabolism as well as basic functions of immune cells such as cell adhesion and signal transduction. The differential expressed genes of the different segments of the chicken intestine (jejunum, ileum, proximal cecum) correlated to the metabolic processes in nutrient digestion and absorption. These data should provide a valuable resource in understanding the chicken genome.

The post-translational modification of histones by the small ubiquitin-like modifier (SUMO) protein has been associated with gene regulation, centromeric localization, and double-strand break repair in eukaryotes. Although sumoylation of histone H4 was specifically associated with gene repression, this could not be proven due to the challenge of site-specifically sumoylating H4 in cells. Biochemical crosstalk between SUMO and other histone modifications, such as H4 acetylation and H3 methylation, that are associated with active genes also remains unclear. We addressed these challenges in mechanistic studies using an H4 chemically modified at Lys12 by SUMO-3 (H4K12su) and incorporated into mononucleosomes and chromatinized plasmids for functional studies. Mononucleosome-based assays revealed that H4K12su inhibits transcription-activating H4 tail acetylation by the histone acetyltransferase p300, as well as transcription-associated H3K4 methylation by the extended catalytic module of the Set1/COMPASS (complex of proteins associated with Set1) histone methyltransferase complex. Activator- and p300-dependent in vitro transcription assays with chromatinized plasmids revealed that H4K12su inhibits both H4 tail acetylation and RNA polymerase II-mediated transcription. Finally, cell-based assays with a SUMO-H4 fusion that mimics H4 tail sumoylation confirmed the negative crosstalk between histone sumoylation and acetylation/methylation. Thus, our studies establish the key role for histone sumoylation in gene silencing and its negative biochemical crosstalk with active transcription-associated marks in human cells.

The inner ear is a complex structure responsible for hearing and balance, and organ pathology is associated with deafness and balance disorders. To evaluate the role of epigenomic dynamics, we performed whole genome bisulfite sequencing at key time points during the development and maturation of the mouse inner ear sensory epithelium (SE). Our single-nucleotide resolution maps revealed variations in both general characteristics and dynamics of DNA methylation over time. This allowed us to predict the location of non-coding regulatory regions and to identify several novel candidate regulatory factors, such as Bach2, that connect stage-specific regulatory elements to molecular features that drive the development and maturation of the SE. Constructing in silico regulatory networks around sites of differential methylation enabled us to link key inner ear regulators, such as Atoh1 and Stat3, to pathways responsible for cell lineage determination and maturation, such as the Notch pathway. We also discovered that a putative enhancer, defined as a low methylated region (LMR), can upregulate the GJB6 gene and a neighboring non-coding RNA. The study of inner ear SE methylomes revealed novel regulatory regions in the hearing organ, which may improve diagnostic capabilities, and has the potential to guide the development of therapeutics for hearing loss by providing multiple intervention points for manipulation of the auditory system.

Disorders involving β-globin gene mutations, primarily β-thalassemia and sickle cell disease, represent a major target for hematopoietic stem/progenitor cell (HSPC) gene therapy. This includes CRISPR/Cas9-mediated genome editing approaches in adult CD34+ cells aimed toward the reactivation of fetal γ-globin expression in red blood cells. Because models involving erythroid differentiation of CD34+ cells have limitations in assessing γ-globin reactivation, we focused on human β-globin locus-transgenic (β-YAC) mice. We used a helper-dependent human CD46-targeting adenovirus vector expressing CRISPR/Cas9 (HDAd-HBG-CRISPR) to disrupt a repressor binding region within the γ-globin promoter. We transduced HSPCs from β-YAC/human CD46-transgenic mice ex vivo and subsequently transplanted them into irradiated recipients. Furthermore, we used an in vivo HSPC transduction approach that involves HSPC mobilization and the intravenous injection of HDAd-HBG-CRISPR into β-YAC/CD46-transgenic mice. In both models, we demonstrated efficient target site disruption, resulting in a pronounced switch from human β- to γ-globin expression in red blood cells of adult mice that was maintained after secondary transplantation of HSPCs. In long-term follow-up studies, we did not detect hematological abnormalities, indicating that HBG promoter editing does not negatively affect hematopoiesis. This is the first study that shows successful in vivo HSPC genome editing by CRISPR/Cas9.

Summary The establishment of DNA methylation patterns in oocytes is a highly dynamic process marking gene-regulatory events during fertilization, embryonic development, and adulthood. However, after epigenetic reprogramming in primordial germ cells, how and when DNA methylation is re-established in developing human oocytes remains to be characterized. Here, using single-cell whole-genome bisulfite sequencing, we describe DNA methylation patterns in three different maturation stages of human oocytes. We found that while broad-scale patterns of CpG methylation have been largely established by the immature germinal vesicle stage, localized changes continue into later development. Non-CpG methylation, on the other hand, undergoes a large-scale, generalized remodeling through the final stage of maturation, with the net overall result being the accumulation of methylation as oocytes mature. The role of the genome-wide, non-CpG methylation remodeling in the final stage of oocyte maturation deserves further investigation., Highlights • Broad-scale CpG methylation is largely stable during oocyte maturation • Non-CpG methylation accumulates as oocytes mature • Non-CpG methylation changes are genome wide, CpG methylation is established globally in adult immature oocytes and remains stable during maturation. Non-CpG methylation undergoes a genome-wide, generalized remodeling through the final stage of maturation, with the net overall result being accumulation of methylation as oocytes mature.

Summary: Activating germline mutations in STAT3 were recently identified as a cause of neonatal diabetes mellitus associated with beta-cell autoimmunity. We have investigated the effect of an activating mutation, STAT3K392R, on pancreatic development using induced pluripotent stem cells (iPSCs) derived from a patient with neonatal diabetes and pancreatic hypoplasia. Early pancreatic endoderm differentiated similarly from STAT3K392R and healthy-control cells, but in later stages, NEUROG3 expression was upregulated prematurely in STAT3K392R cells together with insulin (INS) and glucagon (GCG). RNA sequencing (RNA-seq) showed robust NEUROG3 downstream targets upregulation. STAT3 mutation correction with CRISPR/Cas9 reversed completely the disease phenotype. STAT3K392R-activating properties were not explained fully by altered DNA-binding affinity or increased phosphorylation. Instead, reporter assays demonstrated NEUROG3 promoter activation by STAT3 in pancreatic cells. Furthermore, proteomic and immunocytochemical analyses revealed increased nuclear translocation of STAT3K392R. Collectively, our results demonstrate that the STAT3K392R mutation causes premature endocrine differentiation through direct induction of NEUROG3 expression. : Saarimäki-Vire et al. use iPSCs derived from a patient with permanent neonatal diabetes to demonstrate that an activating STAT3 mutation leads to premature NEUROG3 expression and concomitant differentiation of pancreatic progenitors through increased nuclear shuttling of the mutant protein. Keywords: monogenic diabetes, STAT3, NEUROG3, iPSC, stem cells, CRISPR, genome editing, pancreatic differentiation, beta cell, endocrine cells

Early embryonic development is characterized by a dynamic epigenome. Some of the most drastic changes occur as the mammalian embryo transitions from the free-floating, preimplantation blastocyst to the early embryo embedded in the uterus. Efforts to understand early embryonic molecular phenotypes have driven the derivation of mammalian embryonic stem cells that represent the pre- and postimplantation embryo, termed naive and primed embryonic stem cells. Here, we describe in detail mouse and human embryonic stem cell derivation as it pertains to pre- and postimplantation development. We look at the epigenetic research conducted in these systems and what we have learned to date. We highlight whenever possible in vivo embryo studies as a comparison with the in vitro embryonic stem cell studies. Where data are lacking, we describe what is currently known. Altogether, we hope to focus attention on the research that has led to major discoveries in naive and primed pluripotent stem cells.

Oocyte maturation is a coordinated process that is tightly linked to reproductive potential. A better understanding of gene regulation during human oocyte maturation will not only answer an important question in biology, but also facilitate the development of in vitro maturation technology as a fertility treatment. We generated single-cell transcriptome and used our previously published single-cell methylome data from human oocytes at different maturation stages to investigate how genes are regulated during oocyte maturation, focusing on the potential regulatory role of non-CpG methylation. DNMT3B, a gene encoding a key non-CpG methylation enzyme, is one of the 1,077 genes upregulated in mature oocytes, which may be at least partially responsible for the increased non-CpG methylation as oocytes mature. Non-CpG differentially methylated regions (DMRs) between mature and immature oocytes have multiple binding motifs for transcription factors, some of which bind with DNMT3B and may be important regulators of oocyte maturation through non-CpG methylation. Over 98% of non-CpG DMRs locate in transposable elements, and these DMRs are correlated with expression changes of the nearby genes. Taken together, this data indicates that global non-CpG hypermethylation during oocyte maturation may play an active role in gene expression regulation, potentially through the interaction with transcription factors.

Chromosomal deletions associated with human diseases, such as cancer, are common, but synteny issues complicate modeling of these deletions in mice. We use cellular reprogramming and genome engineering to functionally dissect the loss of chromosome 7q (del(7q)), a somatic cytogenetic abnormality present in myelodysplastic syndromes (MDS). We derive del(7q)- and isogenic karyotypically normal induced pluripotent stem cells (iPSCs) from hematopoietic cells of MDS patients and show that the del(7q) iPSCs recapitulate disease-associated phenotypes, including impaired hematopoietic differentiation. These disease phenotypes are rescued by spontaneous dosage correction and can be reproduced in karyotypically normal cells by engineering hemizygosity of defined chr7q segments in a 20-Mb region. We use a phenotype-rescue screen to identify candidate haploinsufficient genes that might mediate the del(7q)- hematopoietic defect. Our approach highlights the utility of human iPSCs both for functional mapping of disease-associated large-scale chromosomal deletions and for discovery of haploinsufficient genes.

BackgroundThe mammalian inner ear is a complex morphological structure responsible for hearing and balance, and its pathology is associated with deafness and balance disorders. To evaluate the role of epigenomic dynamics in the development and maturation of mouse inner ear sensory epithelium, we performed whole-genome bisulfite sequencing on inner ear tissue, yielding temporal base-pair resolution methylomes at key developmental time points.ResultsWe found a late accumulation of non-CpG methylation, indicating a similarity between the inner ear sensory epithelium and neuronal tissue. Moreover, annotation of both unmethylated and low methylated regions pointed to regulatory elements active in the inner ear in proximity of and distal from transcriptional units. Finally, we identified differentially methylated regions across the transition periods. An analysis of these regions led us to identify several novel candidate regulatory factors, connecting regulatory elements from specific time points in development to molecular features that drive the development and maturation of the inner ear sensory epithelium. The GJB6 locus putative regulatory region was shown to upregulate distal GJB6 gene expression and a non-coding RNA.ConclusionsOur analysis of inner ear sensory epithelium DNA methylation sheds light on novel regulatory regions in the hearing organ, and may help boost diagnostic capabilities and guide the development of therapeutics for hearing loss, by providing multiple intervention points for manipulation of the auditory system.

Gene replacement therapies utilizing adeno-associated viral (AAV) vectors hold great promise for treating Duchenne muscular dystrophy (DMD). A related approach uses AAV vectors to edit specific regions of the DMD gene using CRISPR/Cas9. Here we develop multiple approaches for editing the mutation in dystrophic mdx4cv mice using single and dual AAV vector delivery of a muscle-specific Cas9 cassette together with single-guide RNA cassettes and, in one approach, a dystrophin homology region to fully correct the mutation. Muscle-restricted Cas9 expression enables direct editing of the mutation, multi-exon deletion or complete gene correction via homologous recombination in myogenic cells. Treated muscles express dystrophin in up to 70% of the myogenic area and increased force generation following intramuscular delivery. Furthermore, systemic administration of the vectors results in widespread expression of dystrophin in both skeletal and cardiac muscles. Our results demonstrate that AAV-mediated muscle-specific gene editing has significant potential for therapy of neuromuscular disorders., CRISPR/Cas9-mediated gene editing is an emerging strategy to treat Duchenne muscular dystrophy. Here the authors develop multiple CRISPR/Cas9-based approaches to correct different dystrophin gene mutations, and show significant restoration of dystrophin expression in skeletal and cardiac muscle in mice.

We currently lack a comprehensive understanding of the mechanisms underlying neural tube formation and their contributions to neural tube defects (NTDs). Developing a model to study such a complex morphogenetic process, especially one that models human-specific aspects, is critical. Three-dimensional, human embryonic stem cell (hESC)-derived neural rosettes (NRs) provide a powerful resource for in vitro modeling of human neural tube formation. Epigenomic maps reveal enhancer elements unique to NRs relative to 2D systems. A master regulatory network illustrates that key NR properties are related to their epigenomic landscapes. We found that folate-associated DNA methylation changes were enriched within NR regulatory elements near genes involved in neural tube formation and metabolism. Our comprehensive regulatory maps offer insights into the mechanisms by which folate may prevent NTDs. Lastly, our distal regulatory maps provide a better understanding of the potential role of neurological-disorder-associated SNPs.

Autoimmune diseases are complex disorders of largely unknown etiology. Genetic studies have identified a limited number of causal genes from a marginal number of individuals, and demonstrated a high degree of discordance in monozygotic twins. Studies have begun to reveal epigenetic contributions to these diseases, primarily through the study of DNA methylation, but chromatin and non-coding RNA changes are also emerging. Moving forward an integrative analysis of genomic, transcriptomic and epigenomic data, with the latter two coming from specific cell types, will provide an understanding that has been missed from genetics alone. We provide an overview of the current state of the field and vision for deriving the epigenomics of autoimmunity.

Loss of auditory hair cells (AHCs) is a major cause of human deafness. Considerable effort has been devoted to unraveling how these mechanotransducers of sound are specified, with a view to correcting hearing loss by gene or stem cell therapies. Recent work on signaling cascades, particularly lateral inhibition and planar cell polarity, has begun to tie together some of the known pathways. Mutations in mice and humans that cause hearing and/or balance disorders are also shedding light on how AHCs are specified and, maintained and handle ion flux. Studies on some of these genes are beginning to provide insights into the more complex genetics of later onset forms of hearing loss. Progress has also been made in solving some long-term goals of auditory biology. Cadherin23 has been identified as a component of AHC stereocilia tip links, and progress has been made towards identifying the elusive AHC mechanoreceptor channel. Preliminary steps have also been taken towards inner-ear gene therapy, and in the engineering of embryonic stem cells for eventual cell therapies. Mammals cannot regenerate AHCs, but birds and other lower vertebrates can. Genomic tools have now been brought to bear on this problem with the aim of deciphering the molecular basis of this regenerative capability. The combination of new genomic tools and the many mouse and chicken embryological and genetic resources should increasingly provide new insights into how AHCs are programed and maintained.

Nature Communications 8: Article number: 14454 (2017); Published: 14 February 2017; Updated: 23 June 2017 This Article contains an error in Fig. 4, for which we apologize. In panel a, the image reporting dystrophin labelling following SaCas9Δ5253 treatment was inadvertently duplicated from the corresponding image following SpCas9/Δ5253 treatment.

We have tested the specificity and utility of more than 200 antibodies raised against 57 different histone modifications in Drosophila melanogaster, Caenorhabditis elegans and human cells. Although most antibodies performed well, more than 25% failed specificity tests by dot blot or western blot. Among specific antibodies, more than 20% failed in chromatin immunoprecipitation experiments. We advise rigorous testing of histone-modification antibodies before use, and we provide a website for posting new test results (http://compbio.med.harvard.edu/antibodies/).

Loss of inner ear sensory hair cells (HC) is a leading cause of human hearing loss and balance disorders. Unlike mammals, many lower vertebrates can regenerate these cells. We used cross-species microarrays to examine this process in the avian inner ear. Specifically, changes in expression of over 1700 transcription factor (TF) genes were investigated in hair cells of auditory and vestibular organs following treatment with two different damaging agents and regeneration in vitro. Multiple components of seven distinct known signaling pathways were clearly identifiable: TGFbeta, PAX, NOTCH, WNT, NFKappaB, INSULIN/IGF1 and AP1. Numerous components of apoptotic and cell cycle control pathways were differentially expressed, including p27(KIP) and TFs that regulate its expression. A comparison of expression trends across tissues and treatments revealed identical patterns of expression that occurred at identical times during regenerative proliferation. Network analysis of the patterns of gene expression in this large dataset also revealed the additional presence of many components (and possible network interactions) of estrogen receptor signaling, circadian rhythm genes and parts of the polycomb complex (among others). Equal numbers of differentially expressed genes were identified that have not yet been placed into any known pathway. Specific time points and tissues also exhibited interesting differences: For example, 45 zinc finger genes were specifically up-regulated at later stages of cochlear regeneration. These results are the first of their kind and should provide the starting point for more detailed investigations of the role of these many pathways in HC recovery, and for a description of their possible interactions.

We applied a micro-cDNA-based subtraction method to identify genes expressed in the regenerating sensory epithelia (SE) of the chicken inner ear. Sensory hair cells in the avian utricle SE are in a constant state of turnover, where dying hair cells are replaced by new ones derived from supporting cells. In contrast, hair cells in the cochlea remain quiescent unless damaged. We used this difference to enrich for utricle-specific genes, using reiterative cDNA subtraction and demonstrate enrichment for utricle-specific sequences. A total of 1710 cDNA sequence reads revealed the presence of many cDNAs encoding known structural components of the SE (for example, Harmonin and beta-tectorin), proteins involved in cellular proliferation, such as P311, HIPK2, and SPALT1, among many others of unknown function. These libraries are the first of their kind and should prove useful for the discovery of candidate genes for hearing disorders, regenerative and apoptotic pathways, and novel chicken ESTs.

The sensory receptors for hearing and balance are the hair cells of the cochlea and vestibular organs of the inner ear. Permanent hearing and balance deficits can be triggered by genetic susceptibilities or environmental factors such as infection. Unlike mammalian hair cells that have a limited capacity for regeneration, the vestibular organ of the avian ear is constantly undergoing hair cell regeneration, whereas the avian cochlea undergoes regeneration only when hair cells are damaged. In order to gain insights into the genetic programs that govern the regenerative capacity of hair cells, we interrogated custom human cDNA microarrays with sensory epithelial cell targets from avian inner ears. The arrays contained probes from conserved regions of approximately 400 genes expressed primarily in the inner ear and approximately 1500 transcription factors (TF). Highly significant differences were observed for 20 inner-ear genes and more than 80 TFs. Genes up-regulated in the cochlea included BMP4, GATA3, GSN, FOXF1 and PRDM7. Genes up-regulated in the utricle included SMAD2, KIT, beta-AMYLOID, LOC51637, HMG20B and CRIP2. Many of the highly significant changes were validated by Q-PCR and in situ methods. Some of the observed changes implicated a number of known biochemical pathways including the c-kit pathway previously observed in melanogenesis. Twenty differentially expressed TFs map to chromosomal regions harboring uncloned human deafness loci, and represent novel candidates for hearing loss. The approach described here also illustrates the power of utilizing conserved human cDNA probes for cross-species comparisons.

Background:Discovering patient subtypes and molecular drivers of a subtype are difficult and driving problems underlying most modern disease expression studies collected across patient populations. Expression patterns conserved across multiple expression datasets from independent disease studies are likely to represent important molecular events underlying the disease.Methods:We present the INSPIRE (INferring Shared modules from multiPle gene expREssion datasets) method to infer highly coherent and robust modules of co-expressed genes and the dependencies among the modules from multiple expression datasets. Focusing on inferring modules and their dependencies conserved across multiple expression datasets is important for several reasons. First, using multiple datasets will increase the power to detect robust and relevant patterns (modules and dependencies among modules). Second, INSPIRE enables the use of multiple datasets that contain different sets of genes due to, e.g., the difference in microarray platforms. Many methods designed for expression data analysis cannot integrate multiple datasets with variable discrepancy to infer a single combined model, whereas INSPIRE can naturally model the dependencies among the modules even when a large proportion of genes are not observed on a certain platform.Results:We evaluated INSPIRE on synthetically generated datasets with known underlying network structure among modules, and gene expression datasets from multiple ovarian cancer studies. We show that the model learned by INSPIRE can explain unseen data better and can reveal prior knowledge on gene functions more accurately than alternative methods. We demonstrate that applying INSPIRE to nine ovarian cancer datasets leads to the identification of a new marker and potential molecular driver of tumor-associated stroma - HOPX. We also demonstrate that the HOPXmodule strongly overlaps with the genes defining the mesenchymal patient subtype identified in The Cancer Genome Atlas (TCGA) ovarian cancer data. We provide evidence for a previously unknown molecular basis of tumor resectability efficacy involving tumor-associated mesenchymal stem cells represented by HOPX.Conclusions:INSPIRE extracts a low-dimensional description from multiple gene expression data, which consists of modules and their dependencies. The discovery of a new tumor-associated stroma marker, HOPX, and its module suggests a previously unknown mechanism underlying tumor-associated stroma.