Virginia Savova

Single-cell barcoding and sequencing using droplet microfluidics

Single-cell RNA sequencing has recently emerged as a powerful tool for mapping cellular heterogeneity in diseased and healthy tissues, yet high-throughput methods are needed for capturing the unbiased diversity of cells. Droplet microfluidics is among the most promising candidates for capturing and processing thousands of individual cells for whole-transcriptome or genomic analysis in a massively parallel manner with minimal reagent use. We recently established a method called inDrops, which has the capability to index >15,000 cells in an hour. A suspension of cells is first encapsulated into nanoliter droplets with hydrogel beads (HBs) bearing barcoding DNA primers. Cells are then lysed and mRNA is barcoded (indexed) by a reverse transcription (RT) reaction. Here we provide details for (i) establishing an inDrops platform (1 d); (ii) performing hydrogel bead synthesis (4 d); (iii) encapsulating and barcoding cells (1 d); and (iv) RNA-seq library preparation (2 d). inDrops is a robust and scalable platform, and it is unique in its ability to capture and profile >75% of cells in even very small samples, on a scale of thousands or tens of thousands of cells.

Chromatin signature of widespread monoallelic expression

In mammals, numerous autosomal genes are subject to mitotically stable monoallelic expression (MAE), including genes that play critical roles in a variety of human diseases. Due to challenges posed by the clonal nature of MAE, very little is known about its regulation; in particular, no molecular features have been specifically linked to MAE. In this study, we report an approach that distinguishes MAE genes in human cells with great accuracy: a chromatin signature consisting of chromatin marks associated with active transcription (H3K36me3) and silencing (H3K27me3) simultaneously occurring in the gene body. The MAE signature is present in ∼20% of ubiquitously expressed genes and over 30% of tissue-specific genes across cell types. Notably, it is enriched among key developmental genes that have bivalent chromatin structure in pluripotent cells. Our results open a new approach to the study of MAE that is independent of polymorphisms, and suggest that MAE is linked to cell differentiation. DOI: http://dx.doi.org/10.7554/eLife.01256.001

Genes with monoallelic expression contribute disproportionately to genetic diversity in humans

An unexpectedly large number of human autosomal genes are subject to monoallelic expression (MAE). Our analysis of 4,227 such genes uncovers surprisingly high genetic variation across human populations. This increased diversity is unlikely to reflect relaxed purifying selection. Remarkably, MAE genes exhibit an elevated recombination rate and an increased density of hypermutable sequence contexts. However, these factors do not fully account for the increased diversity. We find that the elevated nucleotide diversity of MAE genes is also associated with greater allelic age: variants in these genes tend to be older and are enriched in polymorphisms shared by Neanderthals and chimpanzees. Both synonymous and nonsynonymous alleles of MAE genes have elevated average population frequencies. We also observed strong enrichment of the MAE signature among genes reported to evolve under balancing selection. We propose that an important biological function of widespread MAE might be the generation of cell-to-cell heterogeneity; the increased genetic variation contributes to this heterogeneity.

Somatic Cell Fusions Reveal Extensive Heterogeneity in Basal-like Breast Cancer

Basal-like and luminal breast tumors have distinct clinical behavior and molecular profiles, yet the underlying mechanisms are poorly defined. To interrogate processes that determine these distinct phenotypes and their inheritance pattern, we generated somatic cell fusions and performed integrated genetic and epigenetic (DNA methylation and chromatin) profiling. We found that the basal-like trait is generally dominant and is largely defined by epigenetic repression of luminal transcription factors. Definition of super-enhancers highlighted a core program common in luminal cells but a high degree of heterogeneity in basal-like breast cancers that correlates with clinical outcome. We also found that protein extracts of basal-like cells are sufficient to induce a luminal-to-basal phenotypic switch, implying a trigger of basal-like autoregulatory circuits. We determined that KDM6A might be required for luminal-basal fusions, and we identified EN1, TBX18, and TCF4 as candidate transcriptional regulators of the luminal-to-basal switch. Our findings highlight the remarkable epigenetic plasticity of breast cancer cells.

Chromatin Signature Identifies Monoallelic Gene Expression Across Mammalian Cell Types.

Monoallelic expression of autosomal genes (MAE) is a widespread epigenetic phenomenon which is poorly understood, due in part to current limitations of genome-wide approaches for assessing it. Recently, we reported that a specific histone modification signature is strongly associated with MAE and demonstrated that it can serve as a proxy of MAE in human lymphoblastoid cells. Here, we use murine cells to establish that this chromatin signature is conserved between mouse and human and is associated with MAE in multiple cell types. Our analyses reveal extensive conservation in the identity of MAE genes between the two species. By analyzing MAE chromatin signature in a large number of cell and tissue types, we show that it remains consistent during terminal cell differentiation and is predominant among cell-type specific genes, suggesting a link between MAE and specification of cell identity.

Osteoblasts remotely supply lung tumors with cancer-promoting SiglecFhigh neutrophils

A bona fide portrayal of tumor growth Bone has a well-established role in advanced cancer. It provides a supportive microenvironment for the growth of metastatic cells that escape the primary tumor, which ultimately leads to loss of bone mass. Engblom et al. show that bone may also contribute to early-stage tumorigenesis through a mechanism that leads to an increase in bone mass (see the Perspective by Zhang and Lyden). In mouse models of lung adenocarcinoma, primary tumor cells remotely activated bone-resident cells called osteoblasts, which have a bone-building function. The activated osteoblasts in turn triggered production of a certain type of neutrophil that infiltrates the primary tumor and promotes its growth. Patients with early-stage lung cancer were also found to have an increase in bone density, consistent with the findings in mice. Science, this issue p. eaal5081; see also p. 1127 Systemic cross-talk between tumor and bone can boost the growth of early-stage lung cancer in mice. INTRODUCTION Myeloid cells have emerged as key regulators of cancer growth because of their abundance in the tumor stroma in a broad range of cancers, their association with clinical outcome, and their ability to modulate tumor progression. Most tumor-infiltrating myeloid cells derive from circulating precursors, which are produced in distant tissues, and some tumors amplify myeloid cell activity by skewing hematopoiesis toward the myeloid lineage or increasing myeloid cell populations in the periphery. For example, patients across diverse cancer types can present with elevated levels of myeloid progenitor cells in peripheral blood. Additionally, increased numbers of circulating myeloid cells, such as neutrophils, often correlate with poorer clinical outcome. It is therefore important to consider host changes that occur away from the tumor stroma to more fully understand the biological processes underlying tumor growth. RATIONALE The bone marrow is a tissue of particular interest as it is the main production site for hematopoietic cells corresponding to all circulating blood lineages in the adult. The marrow contains resident cell components, such as osteoblasts, which not only participate in bone maintenance but also regulate hematopoiesis and immune cell fate. However, our understanding of bone dynamics in the context of cancer (growing at sites distant from the local bone microenvironment) and related immune responses remains limited. To address this knowledge gap, we explored whether a common solid cancer—lung adenocarcinoma—remotely affects bone tissue and how this might shape tumor-associated hematopoietic responses and tumor growth. RESULTS We show in different mouse models and in cancer patients (n = 70) that lung adenocarcinomas increase bone stromal activity even in the absence of local metastasis. Animal studies further reveal that the cancer-induced bone phenotype involves bone-resident osteocalcin-expressing (Ocn+) osteoblastic cells. Ocn+ cells affect distant tumor progression because experimentally reducing the number of these cells limits lung tumor growth. Also, Ocn+ cells are required for full-fledged tumor infiltration by a distinct subset of neutrophils that are defined by their high expression of the lectin SiglecF (sialic acid–binding immunoglobulin-like lectin F). Compared to other neutrophils, SiglecFhigh cells express genes associated with cancer-promoting processes, including angiogenesis, myeloid cell differentiation and recruitment, extracellular matrix remodeling, suppression of T cell responses, and tumor cell proliferation and growth. Additionally, SiglecFhigh neutrophils have increased reactive oxygen species production, promote macrophage differentiation, and boost tumor progression in vivo. We further report that the soluble receptor for advanced glycation end products (sRAGE) is up-regulated in the circulation of tumor-bearing mice and fosters osteoblastic activity and osteoblast-dependent neutrophil maturation in vitro. CONCLUSION This study identifies systemic cross-talk between lung tumors and bones: Lung tumors can remotely activate Ocn+ osteoblastic cells in bones even in the absence of local metastasis. In turn, these Ocn+ cells supply tumors with SiglecFhigh neutrophils, which foster cancer progression. The findings bear scientific and therapeutic importance because they reveal contributions of the host systemic environment to tumor growth and they position Ocn+ cells, SiglecFhigh neutrophils, and sRAGE as candidate clinical biomarkers and possible intervention points for anticancer therapy. Systemic cross-talk between lung tumors and bones. Lung adenocarcinomas can remotely activate Ocn+ osteoblastic cells in bones even in the absence of local metastasis. In turn, these osteoblasts supply tumors with SiglecFhigh neutrophils, which exhibit cancer-promoting functions (left). By contrast, the bone marrow in steady state only produces SiglecFlow neutrophils (right). Bone marrow–derived myeloid cells can accumulate within tumors and foster cancer outgrowth. Local immune-neoplastic interactions have been intensively investigated, but the contribution of the systemic host environment to tumor growth remains poorly understood. Here, we show in mice and cancer patients (n = 70) that lung adenocarcinomas increase bone stromal activity in the absence of bone metastasis. Animal studies reveal that the cancer-induced bone phenotype involves bone-resident osteocalcin-expressing (Ocn+) osteoblastic cells. These cells promote cancer by remotely supplying a distinct subset of tumor-infiltrating SiglecFhigh neutrophils, which exhibit cancer-promoting properties. Experimentally reducing Ocn+ cell numbers suppresses the neutrophil response and lung tumor outgrowth. These observations posit osteoblasts as remote regulators of lung cancer and identify SiglecFhigh neutrophils as myeloid cell effectors of the osteoblast-driven protumoral response.

Autosomal monoallelic expression: genetics of epigenetic diversity?

In mammals, relative expression of the two parental alleles of many genes is controlled by one of three major epigenetic phenomena: X chromosome inactivation, imprinting, and mitotically stable autosomal monoallelic expression (MAE). MAE affects a large fraction of human autosomal genes and introduces enormous epigenetic heterogeneity in otherwise similar cell populations. Despite its prevalence, many functional and mechanistic aspects of MAE biology remain unknown. Several lines of evidence imply that MAE establishment and maintenance are controlled by a variety of genetic elements. Based on known genomic features regulating X-inactivation and imprinting, we outline likely features of MAE-controlling elements. We also assess implications of MAE for genotype-phenotype relationship, with a focus on haploinsufficiency.

dbMAE: the database of autosomal monoallelic expression

Recently, data on ‘random’ autosomal monoallelic expression has become available for the entire genome in multiple human and mouse tissues and cell types, creating a need for better access and dissemination. The database of autosomal monoallelic expression (dbMAE; https://mae.hms.harvard.edu) incorporates data from multiple recent reports of genome-wide analyses. These include transcriptome-wide analyses of allelic imbalance in clonal cell populations based on sequence polymorphisms, as well as indirect identification, based on a specific chromatin signature present in MAE gene bodies. Currently, dbMAE contains transcriptome-wide chromatin identification calls for 8 human and 21 mouse tissues, and describes over 16 000 murine and ∼700 human cases of directly measured biased expression, compiled from allele-specific RNA-seq and genotyping array data. All data are manually curated. To ensure cross-publication uniformity, we performed re-analysis of transcriptome-wide RNA-seq data using the same pipeline. Data are accessed through an interface that allows for basic and advanced searches; all source references, including raw data, are clearly described and hyperlinked. This ensures the utility of the resource as an initial screening tool for those interested in investigating the role of monoallelic expression in their specific genes and tissues of interest.

Risk alleles of genes with monoallelic expression are enriched in gain-of-function variants and depleted in loss-of-function variants for neurodevelopmental disorders

Over 3000 human genes can be expressed from a single allele in one cell, and from the other allele—or both—in neighboring cells. Little is known about the consequences of this epigenetic phenomenon, monoallelic expression (MAE). We hypothesized that MAE increases expression variability, with a potential impact on human disease. Here, we use a chromatin signature to infer MAE for genes in lymphoblastoid cell lines and human fetal brain tissue. We confirm that across clones MAE status correlates with expression level, and that in human tissue data sets, MAE genes show increased expression variability. We then compare mono- and biallelic genes at three distinct scales. In the human population, we observe that genes with polymorphisms influencing expression variance are more likely to be MAE (P<1.1 × 10−6). At the trans-species level, we find gene expression differences and directional selection between humans and chimpanzees more common among MAE genes (P<0.05). Extending to human disease, we show that MAE genes are under-represented in neurodevelopmental copy number variants (CNVs) (P<2.2 × 10−10), suggesting that pathogenic variants acting via expression level are less likely to involve MAE genes. Using neuropsychiatric single-nucleotide polymorphism (SNP) and single-nucleotide variant (SNV) data, we see that genes with pathogenic expression-altering or loss-of-function variants are less likely MAE (P<7.5 × 10−11) and genes with only missense or gain-of-function variants are more likely MAE (P<1.4 × 10−6). Together, our results suggest that MAE genes tolerate a greater range of expression level than biallelic expression (BAE) genes, and this information may be useful in prediction of pathogenicity.

High prevalence of clonal monoallelic expression

To the Editor — In recent years, there has been substantial interest in clonally stable monoallelic expression of autosomal genes in mammalian cells. This ‘autosomal analog of X-chromosome inactivation’ has been observed to affect hundreds of genes in a variety of cell types (overview in ref. 1). These genes tend to encode cell-surface proteins2,3 and to show high heterozygosity in human populations4. This observation has intriguing implications regarding the role of monoallelic expression in biological variation, especially if such genes are abundant. In a recent paper5, Sandberg and colleagues have proposed that transient monoallelic expression due to transcriptional bursts is abundant, whereas clonally stable monoallelic expression (random monoclonal expression of autosomal genes (aRME), a term also used herein) is “surprisingly scarce (< 1% of genes).” They go on to state that their observations “[call] into question the notion of widespread clonal aRME affecting thousands of genes”1–3,6, suggesting a sharp contrast to previous work from several groups, including ours. Upon careful analysis, we argue that that the findings that they report are consistent with the literature on aRME, and apparent discrepancies are due to issues with either semantics or simple methodological choices. It is outside the scope of our comments to discuss complex technical issues involved in allele-specific analysis of single-cell RNAsequencing data. Thus, we will take the factual findings by Sandberg and colleagues as reported and will focus on how these findings relate to previous claims regarding clonal aRME (recent overview in ref. 7). First, we examine the meaning of ‘prevalence’ in the context of aRME. On the one hand, prevalence could refer to the number of aRME genes per clone, which can be relatively low. On the other hand, it could apply to the number of genes in the genome that are subject to this mode of regulation, which is much greater. In each individual clone, relatively few genes are classified as monoallelically expressed, and the same genes can be stably biallelically expressed in other clones. When multiple clones are assessed, however, the cumulative number of genes exhibiting aRME reaches into the hundreds. For example, in the first genomewide analysis of aRME in human cells, we used SNP-array analysis in bulk clonal cell populations to identify 30–50 genes with monoallelic expression per lymphoblast clone, with a total of approximately 400 observed across 12 clones2. We and other groups have also used RNA-seq analysis in bulk clonal populations of different mouse cell types (refs. 8–10 among others). These studies were designed to identify clonally stable aRME, which is consistent across most cells of a given clone, but would not detect transient monoallelic expression, which varies. By applying to these studies the same 98:2 allelic bias cutoff used by Sandberg and colleagues5, we found 362 monoallelic genes per clone in mouse B cells (701 over two clones)9, 178 per fibroblast clone (330 over two clones)9, and 301 per neuronal progenitor clone (1,079 genes over eight clones)10. These numbers are in line with the observations of clonal aRME by Sandberg and colleagues, especially considering the challenges in detecting the allelic bias of all but the highest expressing genes in single cells11. In their study5, Sandberg and colleagues reported 41 and 47 clonally stable aRME genes in two fibroblast clones, and five of these genes showed aRME in both clones. A straightforward extrapolation from these numbers (with a 10–15% probability that an aRME gene that is monoallelic in one clone would also show monoallelic expression in the second clone) would bring the total estimated number of aRME genes in fibroblasts to 300–400; these genes should be detected if a sufficiently large number of clones were analyzed. In addition, the reported overlap of aRME genes between the two clones suggests (on the basis of a simple binomial model) that as few as 10–20 independent clones may be sufficient to identify most informative aRME genes. The estimate of the potential genomewide prevalence of clonal aRME may be even higher, given that such genes show highly cell-type-specific expression8,9. Thus, a union over multiple cell types would cumulatively reach ~30% of all protein-coding genes in humans and mice2,4,9, as we have reported before by using a different approach. The second issue leading to apparent discrepancies in the numbers of aRME genes arises from a straightforward question of methodology: the number of genes classified as aRME will obviously strongly depend on the allelic bias threshold used in analysis. The choice of the 98% threshold by Sandberg and colleagues appears appropriate given the challenges of single-cell RNA-seq, which make confident detection of less extreme bias difficult11. However, given that this issue is one of measurement, the threshold imposed is by necessity arbitrary. Sandberg and colleagues advocate for a more stringent threshold; whether that threshold is functionally relevant is unclear. For instance, there is no biological reason to expect a dramatic functional difference between ‘monoallelic’ expression with a 98:2 allelic bias compared to a 97:3 bias. Thus, in settings that allow for more precise measurement of bias, the rationale for choosing any particular threshold depends on the biological question asked. We and other groups have often used more permissive thresholds, which are robust in bulk RNA-seq analysis. For instance, an RNA-seq study of neuronal lineage cells10 has applied an 85% allelic bias threshold and reported up to 2,444 genes with clonal monoallelic expression across eight clones. Defining larger, more inclusive sets of aRME genes allows the sets to be interrogated in genome-wide analyses, thus yielding new biological insights. For example, we have recently found that in neurodevelopmental disease, point mutations, but not copy number variants, are linked to pathology12. We have also reported an unexpected observation that this group of genes has been subject to large-scale balancing selection4. After carefully considering the definitions of aRME, we believe that the findings on clonally stable aRME reported by Sandberg and colleagues using single-cell analysis confirm, rather than call into question, previous analyses performed in bulk clonal cell populations. These findings all suggest that aRME is a phenomenon that affects a large fraction of genes in the genome. ❐