Single-cell RNA-seq reveals changes in cell cycle and differentiation programs upon aging of hematopoietic stem cells

There is no author summary for this article yet. Authors can add summaries to their articles on ScienceOpen to make them more accessible to a non-specialist audience.

Abstract

Both intrinsic cell state changes and variations in the composition of stem cell populations have been implicated as contributors to aging. We used single-cell RNA-seq to dissect variability in hematopoietic stem cell (HSC) and hematopoietic progenitor cell populations from young and old mice from two strains. We found that cell cycle dominates the variability within each population and that there is a lower frequency of cells in the G1 phase among old compared with young long-term HSCs, suggesting that they traverse through G1 faster. Moreover, transcriptional changes in HSCs during aging are inversely related to those upon HSC differentiation, such that old short-term (ST) HSCs resemble young long-term (LT-HSCs), suggesting that they exist in a less differentiated state. Our results indicate both compositional changes and intrinsic, population-wide changes with age and are consistent with a model where a relationship between cell cycle progression and self-renewal versus differentiation of HSCs is affected by aging and may contribute to the functional decline of old HSCs.

Related collections

Most cited references 77

Record: found
Abstract: found
Article: not found

Integrative Genomics Viewer

James Robinson, Helga Thorvaldsdóttir, Wendy Winckler … (2011)

To the Editor Rapid improvements in sequencing and array-based platforms are resulting in a flood of diverse genome-wide data, including data from exome and whole genome sequencing, epigenetic surveys, expression profiling of coding and non-coding RNAs, SNP and copy number profiling, and functional assays. Analysis of these large, diverse datasets holds the promise of a more comprehensive understanding of the genome and its relation to human disease. Experienced and knowledgeable human review is an essential component of this process, complementing computational approaches. This calls for efficient and intuitive visualization tools able to scale to very large datasets and to flexibly integrate multiple data types, including clinical data. However, the sheer volume and scope of data poses a significant challenge to the development of such tools. To address this challenge we developed the Integrative Genomics Viewer (IGV), a lightweight visualization tool that enables intuitive real-time exploration of diverse, large-scale genomic datasets on standard desktop computers. It supports flexible integration of a wide range of genomic data types including aligned sequence reads, mutations, copy number, RNAi screens, gene expression, methylation, and genomic annotations (Figure S1). The IGV makes use of efficient, multi-resolution file formats to enable real-time exploration of arbitrarily large datasets over all resolution scales, while consuming minimal resources on the client computer (see Supplementary Text). Navigation through a dataset is similar to Google Maps, allowing the user to zoom and pan seamlessly across the genome at any level of detail from whole-genome to base pair (Figure S2). Datasets can be loaded from local or remote sources, including cloud-based resources, enabling investigators to view their own genomic datasets alongside publicly available data from, for example, The Cancer Genome Atlas (TCGA) 1 , 1000 Genomes (www.1000genomes.org/), and ENCODE 2 (www.genome.gov/10005107) projects. In addition, IGV allows collaborators to load and share data locally or remotely over the Web. IGV supports concurrent visualization of diverse data types across hundreds, and up to thousands of samples, and correlation of these integrated datasets with clinical and phenotypic variables. A researcher can define arbitrary sample annotations and associate them with data tracks using a simple tab-delimited file format (see Supplementary Text). These might include, for example, sample identifier (used to link different types of data for the same patient or tissue sample), phenotype, outcome, cluster membership, or any other clinical or experimental label. Annotations are displayed as a heatmap but more importantly are used for grouping, sorting, filtering, and overlaying diverse data types to yield a comprehensive picture of the integrated dataset. This is illustrated in Figure 1, a view of copy number, expression, mutation, and clinical data from 202 glioblastoma samples from the TCGA project in a 3 kb region around the EGFR locus 1, 3 . The investigator first grouped samples by tumor subtype, then by data type (copy number and expression), and finally sorted them by median copy number over the EGFR locus. A shared sample identifier links the copy number and expression tracks, maintaining their relative sort order within the subtypes. Mutation data is overlaid on corresponding copy number and expression tracks, based on shared participant identifier annotations. Several trends in the data stand out, such as a strong correlation between copy number and expression and an overrepresentation of EGFR amplified samples in the Classical subtype. IGV’s scalable architecture makes it well suited for genome-wide exploration of next-generation sequencing (NGS) datasets, including both basic aligned read data as well as derived results, such as read coverage. NGS datasets can approach terabytes in size, so careful management of data is necessary to conserve compute resources and to prevent information overload. IGV varies the displayed level of detail according to resolution scale. At very wide views, such as the whole genome, IGV represents NGS data by a simple coverage plot. Coverage data is often useful for assessing overall quality and diagnosing technical issues in sequencing runs (Figure S3), as well as analysis of ChIP-Seq 4 and RNA-Seq 5 experiments (Figures S4 and S5). As the user zooms below the ~50 kb range, individual aligned reads become visible (Figure 2) and putative SNPs are highlighted as allele counts in the coverage plot. Alignment details for each read are available in popup windows (Figures S6 and S7). Zooming further, individual base mismatches become visible, highlighted by color and intensity according to base call and quality. At this level, the investigator may sort reads by base, quality, strand, sample and other attributes to assess the evidence of a variant. This type of visual inspection can be an efficient and powerful tool for variant call validation, eliminating many false positives and aiding in confirmation of true findings (Figures S6 and S7). Many sequencing protocols produce reads from both ends (“paired ends”) of genomic fragments of known size distribution. IGV uses this information to color-code paired ends if their insert sizes are larger than expected, fall on different chromosomes, or have unexpected pair orientations. Such pairs, when consistent across multiple reads, can be indicative of a genomic rearrangement. When coloring aberrant paired ends, each chromosome is assigned a unique color, so that intra- (same color) and inter- (different color) chromosomal events are readily distinguished (Figures 2 and S8). We note that misalignments, particularly in repeat regions, can also yield unexpected insert sizes, and can be diagnosed with the IGV (Figure S9). There are a number of stand-alone, desktop genome browsers available today 6 including Artemis 7 , EagleView 8 , MapView 9 , Tablet 10 , Savant 11 , Apollo 12 , and the Integrated Genome Browser 13 . Many of them have features that overlap with IGV, particularly for NGS sequence alignment and genome annotation viewing. The Integrated Genome Browser also supports viewing array-based data. See Supplementary Table 1 and Supplementary Text for more detail. IGV focuses on the emerging integrative nature of genomic studies, placing equal emphasis on array-based platforms, such as expression and copy-number arrays, next-generation sequencing, as well as clinical and other sample metadata. Indeed, an important and unique feature of IGV is the ability to view all these different data types together and to use the sample metadata to dynamically group, sort, and filter datasets (Figure 1 above). Another important characteristic of IGV is fast data loading and real-time pan and zoom – at all scales of genome resolution and all dataset sizes, including datasets comprising hundreds of samples. Finally, we have placed great emphasis on the ease of installation and use of IGV, with the goal of making both the viewing and sharing of their data accessible to non-informatics end users. IGV is open source software and freely available at http://www.broadinstitute.org/igv/, including full documentation on use of the software. Supplementary Material 1

0 comments Cited 4415 times – based on 0 reviews      Review now

Record: found
Abstract: found
Article: not found

Pseudo-temporal ordering of individual cells reveals dynamics and regulators of cell fate decisions

Cole Trapnell, Davide Cacchiarelli, Jonna L Grimsby … (2014)

Single-cell expression profiling by RNA-Seq promises to exploit cell-to-cell variation in gene expression to reveal regulatory circuitry governing cell differentiation and other biological processes. Here, we describe Monocle, a novel unsupervised algorithm for ordering cells by progress through differentiation that dramatically increases temporal resolution of expression measurements in a model of skeletal muscle differentiation. This reordering unmasks switch-like changes in expression of key regulatory factors, reveals sequentially organized waves of gene regulation, and exposes novel regulators of cell differentiation. A loss-of function screen revealed that many of these inhibitors act through regulatory elements also used by pro-myogenic factors to activate downstream genes. This study demonstrates that single-cell expression analysis by Monocle can uncover novel regulatory interactions governing differentiation. Cell differentiation is governed by a vast and complex gene regulatory program. During differentiation, each cell makes fate decisions independently by integrating a wide array of signals from other cells, executing a complex choreography of gene regulatory changes. Recently, several studies carried out at single-cell resolution have revealed high cell-to-cell variation in most genes during differentiation 1–5 , even among key developmental regulators. Although high variability complicates analysis of such experiments 6 , it might define biological progression between cellular states, revealing regulatory modules of genes that co-vary in expression across individual cells 7 . Prior studies have used approaches from computational geometry 8,9 and supervised machine learning 10 to order bulk cell populations from time-series microarray experiments by progress through a biological process. Applying this concept to order individual cells could expose fine-grained gene expression dynamics as they differentiate. We have developed Monocle, an algorithm that harnesses single cell variation to sort cells in “pseudo time” according to progress through differentiation. Applying Monocle to the classic model of myogenesis unveiled dynamics at unprecedented resolution and exposed novel regulatory factors. Skeletal myoblasts undergo a well-characterized sequence of morphological and transcriptional changes during differentiation 11 . Global expression and epigenetic profiles have reinforced the view that a small cohort of transcription factors (e.g. MYOD, MYOG, MRF4, and MYF5) orchestrates these changes 12 . However, efforts to expand this set of factors and map the broader myogenic regulatory network have been hampered by the temporal resolution of global expression measurements, with thousands of genes following a limited number of coarse kinetic trends 13 . Single-cell measurements of markers of myogenesis have made clear that cells do not progress through differentiation in synchrony. A population of cells captured at the same time may thus cover a range of distinct intermediate differentiation states. Drawing conclusions from a group of individuals based on the properties of their average is a hazardous practice because the average can mask important trends among the individuals, resulting in phenomena such as Simpson's paradox 14 . Experimental synchronization or stringent isolation of myogenic precursors is often challenging and dramatically alters differentiation kinetics. We hypothesized that capturing complete expression profiles of individual cells might avoid these problems and dramatically increase temporal resolution in global transcriptome dynamics. In essence, a single-cell RNA-Seq experiment might constitute a time-series, with each cell representing a distinct time point along a continuum. To test this hypothesis we investigated the single cell transcriptome dynamics during myogenesis. We expanded primary human myoblasts under high mitogen conditions (GM), and then induced differentiation by switching to low-mitogen media (DM). We then captured 50–100 cells at each of four time points following serum switch using the Fluidigm C1 microfluidic system. RNA from each cell was isolated and used to construct mRNA-Seq libraries, which were then sequenced to a depth of ~4 million reads per library, resulting in a complete gene expression profile for each cell (Fig 1a, S1). Averaging expression profiles of cells collected at the same time correlated well with the corresponding bulk RNA-Seq libraries, and moderately expressed genes were detectable (≥ 1 FPKM) in a majority of individual cells (Fig 1b, S2, S3). However, markers of mature myocytes were present at all time points following serum switch, and many other genes showed similar temporal heterogeneity (Fig 1c) We speculated that the high variability in cell-to-cell gene expression levels was due to unsynchronized differentiation, with myoblasts, intermediate myocytes, and mature myotubes residing in the same well concurrently. Indeed, large, multinucleated MYH2+ cells were abundant after 72 hours in DM, but these cells were present at lower frequency even at 24 hours (Fig 1c). We reasoned that informatically ordering the cells by their progress through differentiation, rather than by the time they were collected, would distinguish genes activated early in differentiation from those activated later. To this end, we developed a novel unsupervised algorithm, Monocle, which re-ordered the cells to maximize the transcriptional similarity between successive pairs (Fig 2a). The algorithm first represents the expression profile of each cell as a point in a high-dimensional Euclidean space, with one dimension for each gene. Second, it reduces the dimensionality of this space using Independent Component Analysis 15 . Third, Monocle constructs a minimum spanning tree (MST) on the cells, an approach now commonly used in other single-cell settings, such as flow or mass cytometry 16,17 . Fourth, the algorithm finds the longest path through the MST, corresponding to the longest sequence of transcriptionally similar cells. Finally, Monocle uses this sequence to produce a “trajectory” of an individual cell's progress through differentiation. Progress along a differentiation trajectory is measured in “pseudo-time”: the total transcriptional change a cell undergoes as it differentiates. This strategy is derived from a prior algorithm for temporally ordering microarray samples 8 , but extends it to allow for multiple cell fates stemming from a single progenitor cell type. As cells progress, they may diverge along two or more separate paths. After Monocle finds the longest sequence of similar cells, it examines cells not along this path to find alternative trajectories through the MST. These sub-trajectories are ordered and connected to the main trajectory, and each cell is annotated with both a trajectory and a pseudo-time value. Monocle thus orders cells by progress through differentiation and can reconstruct branched biological processes, which might arise when a precursor cell makes cell fate decisions that govern the generation of multiple subsequent lineages. Importantly, Monocle is unsupervised and needs no prior knowledge of specific genes that distinguish cell fates, and is thus suitable for studying a wide array of dynamic biological processes. Monocle decomposed myoblast differentiation into a two-phase trajectory and isolated a branch of non-differentiating cells (Fig 2b). The first phase of the trajectory was primarily composed of cells collected under high-mitogen conditions and which expressed markers of actively proliferating cells such as CDK1, while the second mainly consisted of cells collected at 24, 48, or 72 hours following serum switch. Cells in the second phase were positive for markers of muscle differentiation such as MYOG (Fig S4). A tightly grouped third population of cells branched from the trajectory near the transition between phases. These cells lacked myogenic markers but expressed PDGFRA and SPHK1, suggesting that they are contaminating interstitial mesenchymal cells and did not arise from the myoblasts. Such cells were recently shown to stimulate muscle differentiation 18 . Monocle's estimates of the frequency and proliferative status of these cells were consistent with estimates derived from immunofluorescent stains against ANPEP/CD13 and nuclear phosphorylated H3-Ser10 (Fig S4). Monocle thus enabled analysis of the myoblast differentiation trajectory without subtracting these cells by immunopurification, maintaining in vitro differentiation kinetics that resemble physiological cell crosstalk occurring in the in vivo niche. To find genes that were dynamically regulated as the cells progressed through differentiation, we modeled each gene's expression as a nonlinear function of pseudo-time. A total of 1,061 genes were dynamically regulated during differentiation (FDR = 0.8) to their relative order within the full data set. The algorithm retained the ability to detect dynamically regulated genes with high precision (>= 95%) over all designs and with increasing recall as more of the cells were included. (Fig S7) We next grouped genes with similar trends in expression, reasoning that such groups might share common biological functions and regulators. Clustering of genes according to direction and timing revealed six distinct trends (Fig 3). Genes downregulated early or upregulated late in pseudo-time were highly enriched for biological processes central to myogenesis, including cell-cycle exit and activation of muscle-specific structural proteins. However, the other clusters included many genes with broad roles in development, including mediators of cell-cell signaling, RNA export and translational control, and remodeling of cell morphology (Fig S8). A timeseries analysis of myoblast differentiation with bulk RNA-Seq identified up and down-regulated genes, but did not identify the transient clusters or distinguish the early from late regulation visible with pseudo-temporally ordered single cells (Fig S9). Furthermore, dynamic range of expression was compressed for most genes, confirming that failure to account for variability in progress through differentiation leads directly to the effects associated with Simpson's paradox. Pseudo-temporal cell ordering thus decomposes the coarse kinetic trends produced by conventional RNA-Seq into distinct, sequential waves of transcriptional reconfiguration. To identify factors driving myoblast differentiation, we performed a cis-regulatory analysis on genes in each pseudo-temporal cluster. Cis regulatory elements were first identified based on DNaseI hypersensitive sites in HSMM cells and HSMM-derived myotubes 20 , classified according to function according to histone marks 21 , and finally annotated with conserved transcription factor binding sites. While downregulated genes were enriched at near significant levels with binding sites for genes that play roles in proliferation (e.g. MAX, E2F, and NMYC), nearly all significantly enriched motifs fell near upregulated genes. These genes were highly enriched for regulatory elements containing binding motifs for 175 transcription factors, including numerous well-known regulators of myogenesis, such as MYOD, MYOG, PBX1, MEIS1, and the MEF2 family (Fig S10). Some, but not all, of these factors were revealed by a regulatory element analysis performed using bulk RNA-Seq data, underscoring the power of increased (pseudo) temporal resolution of single-cell analysis (Fig S11). A similar analysis of microRNA target sites identified miR-1, miR-206, miR-133, and numerous others as regulators of genes activated during myogenesis (Fig S12). Of these, only miR-1/206 target sites were significantly enriched among genes found to be transiently upregulated and then sharply downregulated. This may suggest that miR-1 and miR-206, which are expressed at an intermediate stage of myoblast differentiation, may act to strongly suppress a subset of genes activated earlier. Many of the transcription factors implicated by our cis regulatory analysis to govern differentiation had no previously appreciated role in muscle development. To test potential roles of these factors we performed an RNAi mediated loss of function screen for 11 candidates. Briefly, we virally expressed proliferating myoblasts with one of 44 distinct shRNAs targeting either one of these factors or a mock (non-targeting) control, followed by serum-induced differentiation for five days. We then measured the frequency and size of myosin heavy chain 2 (MYH2)-positive cells with a high-throughput immunofluorescence pipeline. Of the targets we tested, MZF1, ZIC1, XBP1, and USF1 showed significantly altered differentiation kinetics (Fig 4a,b, Fig S13) when depleted with two or more independent hairpins (FDR < 5%). Knockdown of XBP1, USF1, ZIC1, and MZF1 enhanced myotube formation, with larger myotubes containing a higher fraction of total nuclei than mock shRNA controls. Depletion of CUX1, ARID5B, POU2F1, and AHR also increased differentiation efficiency, albeit less significantly. Importantly, whole-well nuclei counts were similar between knockdowns and mock controls, indicating that enhanced differentiation was not simply a result of higher initial cell counts or increased proliferation. With the exception of ZIC1, forced overexpression did not substantially alter differentiation kinetics (data not shown). Notably, several of these factors have binding motifs that are highly enriched in promoters and enhancers that also have motifs for known muscle regulators (Fig 4c). For example, USF1 motifs are enriched in enhancers that also have MYOD motifs. Together, these results confirm that the transcription factors identified as possible regulators in fact play a role in myoblast differentiation, and demonstrate the power of Monocle for identifying key differentiation genes. Here, we report that individual myoblasts progress through differentiation in an unsynchronized manner, but that they can be reordered according to progress through differentiation. This pseudo-time ordering pinpoints key events in differentiation that are masked both by conventional bulk cell expression profiling, and by single-cell expression profiles ordered by time collected. The reordering resolves sequentially activated transcriptional sub-programs that are regulated by common factors. The temporal resolution offered by hundreds of ordered cells might enable future efforts to computationally infer novel gene-regulatory modules. For example, the enrichment of transiently upregulated genes for common microRNA target sites raises the question of whether those microRNAs are expressed later, curtailing what would have been higher levels of expression. Sequencing-based measurements of small RNAs and mRNAs from the same cell will provide answers to such systems-level questions. Moreover, single-cell analysis distinguishes cells of interest from contaminating cell types such as interstitial mesenchymal cells without experimental isolation that might disrupt cell-cell interactions important in the in vivo niche. We identified eight previously unappreciated transcription factors that dramatically influence the course of myoblast differentiation, thus proving the principle of pseudo-temporal analysis and expanding the catalog of regulators in this well-studied system. Several of the eight factors reported here may normally repress differentiation by competing with pro-myogenic factors for these regulatory elements. Alternatively, these inhibitors may co-occupy regulatory elements with pro-myogenic factors, preventing transactivation of their targets (Fig. 4d). Previous studies in other contexts provide mechanistic data supporting both of these models. USF1 inhibits MyoD autoactivation in Xenopus by competing with MyoD at its promoter through an alternative E-box 22 . Our results suggest that USF1 may repress a broad array of targets via E-box competition. CUX1 represses targets in several developmental contexts through binding site competition 23 . XBP1 was recently reported to inhibit myoblast differentiation in mice, potentially through the mechanisms proposed here 24 . Further experiments in these HSMM cells and myoblasts from other anatomic depots will be needed to confirm the mechanism of these factors. While the positive regulators of myogenesis have been well characterized, only a handful of inhibitors have been identified. The eight inhibitors reported here may shed light on how the balance of proliferation and differentiation is maintained during development and regeneration. Ordering the expression profiles of individual cells by biological progress is thus a powerful new tool for studying cell differentiation, and could in principle be used to map regulatory networks that govern a much wider array of biological processes. Supplementary Material 1

0 comments Cited 2448 times – based on 0 reviews      Review now

Record: found
Abstract: found
Article: found

Is Open Access

Integrative Genomics Viewer (IGV): high-performance genomics data visualization and exploration

Helga Thorvaldsdóttir, James Robinson, Jill Mesirov (2012)

Data visualization is an essential component of genomic data analysis. However, the size and diversity of the data sets produced by today’s sequencing and array-based profiling methods present major challenges to visualization tools. The Integrative Genomics Viewer (IGV) is a high-performance viewer that efficiently handles large heterogeneous data sets, while providing a smooth and intuitive user experience at all levels of genome resolution. A key characteristic of IGV is its focus on the integrative nature of genomic studies, with support for both array-based and next-generation sequencing data, and the integration of clinical and phenotypic data. Although IGV is often used to view genomic data from public sources, its primary emphasis is to support researchers who wish to visualize and explore their own data sets or those from colleagues. To that end, IGV supports flexible loading of local and remote data sets, and is optimized to provide high-performance data visualization and exploration on standard desktop systems. IGV is freely available for download from http://www.broadinstitute.org/igv, under a GNU LGPL open-source license.

0 comments Cited 2257 times – based on 0 reviews      Review now

All references

Author and article information

Journal

Journal ID (nlm-ta): Genome Res

Journal ID (iso-abbrev): Genome Res

Journal ID (hwp): genome

Journal ID (pmc): genome

Journal ID (publisher-id): GENOME

Title: Genome Research

Publisher: Cold Spring Harbor Laboratory Press

ISSN (Print): 1088-9051

ISSN (Electronic): 1549-5469

Publication date (Print): December 2015

Volume: 25

Issue: 12

Pages: 1860-1872

Affiliations

[1 ]Broad Institute of MIT and Harvard, Cambridge, Massachusetts 02142, USA;

[2 ]Division of Hematology, Department of Medicine, Brigham and Women's Hospital, Harvard Medical School, Boston, Massachusetts 02115, USA;

[3 ]Harvard Stem Cell Institute and Department of Stem Cell and Regenerative Biology, Harvard University, Cambridge, Massachusetts 02138, USA;

[4 ]Joslin Diabetes Center, Boston, Massachusetts 02215, USA;

[5 ]Paul F. Glenn Laboratories for the Biological Mechanisms of Aging, Harvard Medical School, Boston, Massachusetts 02115, USA;

[6 ]Howard Hughes Medical Institute, Massachusetts Institute of Technology, Cambridge, Massachusetts 02140, USA;

[7 ]Department of Biology, Massachusetts Institute of Technology, Cambridge, Massachusetts 02140, USA

Author notes

[8]

These authors contributed equally to this work.

Corresponding author: aregev@ 123456broadinstitute.org

Article

Medline ID: 9509184

DOI: 10.1101/gr.192237.115

PMC ID: 4665007

PubMed ID: 26430063

SO-VID: 2241f899-25ae-4bf5-9ccf-6f758565330a

License:

This article is distributed exclusively by Cold Spring Harbor Laboratory Press for the first six months after the full-issue publication date (see http://genome.cshlp.org/site/misc/terms.xhtml). After six months, it is available under a Creative Commons License (Attribution-NonCommercial 4.0 International), as described at http://creativecommons.org/licenses/by-nc/4.0/.

History

Date received : 19 March 2015

Date accepted : 30 September 2015

Page count

Pages: 13

Funding

Funded by: European Hematology Association http://dx.doi.org/10.13039/100008594

Funded by: EMBO Long Term Fellowship

Funded by: HFSP Long Term Fellowship

Funded by: Rothschild Fellowship

Funded by: German Cancer Foundation

Funded by: Leukemia and Lymphoma Society http://dx.doi.org/10.13039/100005189

Funded by: National Institutes of Health (NIH) http://dx.doi.org/10.13039/100000002

Award ID: P01 CA066996

Funded by: HHMI

Funded by: NIH CEGS http://dx.doi.org/10.13039/100000002

Award ID: 1P50HG006193-01

Funded by: Klarman Family Foundation http://dx.doi.org/10.13039/100005310

Funded by: Howard Hughes Medical Institute http://dx.doi.org/10.13039/100000011

Comments

Comment on this article

Cited by 359

See all cited by

Most referenced authors 1,817

See all reference authors

- Version 1
- Version 1