Remodeling the chromatin landscape in T lymphocytes by a division of labor among transcription factors

Summary

An extraordinary degree of condensation is required to fit the eukaryotic genome inside the nucleus. This compaction is attained by first coiling the DNA around structures called nucleosomes. Mammalian genomes are further folded into sophisticated three-dimensional (3D) configurations, enabling the genetic code to dictate a diverse range of cell fates. Recent advances in molecular and computational technologies have enabled the query of higher-order chromatin architecture at an unprecedented resolution and scale. In T lymphocytes, similar to other developmental programs, the hierarchical genome organization is shaped by a highly coordinated division of labor among different classes of sequence-specific transcription factors. In this review, we will summarize the general principles of 1D and 3D genome organization, introduce the common experimental and computational techniques to measure the multilayer chromatin organization, and discuss the pervasive role of transcription factors on chromatin organization in T lymphocytes.

Introduction

The adaptive immune response defined by the presence of T or B lymphocytes is initiated when a pathogen overwhelms innate defense mechanisms. The development of T cells from blood progenitors exclusively occurs in the thymus in a highly controlled manner. To confer adaptive immunity, a naïve T cell leaves the thymus and interacts with its specific antigen presented to it as a peptide:MHC complex on the surface of an antigen-presenting cell. This interaction will induce the proliferation and differentiation of the naive T cell into progeny with new properties that contribute to removal of antigen. Both T cell development in the thymus and T cell differentiation in the periphery occur in response to environmental signals including Notch and the combinatorial effect of cytokines. Changes in the cellular environment command the expression of a combination of sequence-specific transcription factors, which have the intrinsic ability to “transcribe” the instructions encoded in our genomes.

Our molecular understanding of how transcription factors control T cell fate has drastically improved over the last few decades. The availability of DNA microarrays and technologies for engineering the mouse genome enabled the immunology community to unmask the functional roles of a cadre of transcription factors in T cells. The rush to discover novel T cell subsets and their “master regulators” revealed context-specific effects of transcription factors such as T-bet, Gata3, Rorc, Foxp3, and many more. Early reductionist strategies proposed that these proteins control cell fate by binding to promoters of a small number of target genes. For example, T-bet is the master regulator of T helper 1 cells and the ‘mechanism’ through which it controls Th1 cells was linked to its binding to the promoter of Ifng and a few of its enhancers. In a dramatic contrast to these early perspectives, the latest technological advances of the past decade unraveled a much more complex picture. The glimpse on how transcription factors are involved in chromatin organization, both on linear packaging of DNA around nucleosomes and on the 3D genome architecture inside the nucleus, indicate a more widespread way these proteins supervise T cell fate and function. In this review, we will summarize the general principles of genome organization in 1D and 3D, introduce major techniques to measure multiple layers of chromatin organization, and end by describing the latest discoveries about sequence-specific proteins controlling T cell fate and function.

1. Principles of the mammalian genome organization (BWF Figure)

1.1. Nucleosome positioning and chromatin accessibility

Eukaryotic genomes must be condensed by extraordinary orders of magnitude to fit into the nucleus of a cell¹. This compaction is attained by first coiling the DNA around structures called nucleosomes. Nucleosomes are comprised of 147 base-pair of DNA, an octamer of the four core histones, and in some contexts a linker histone². Wrapping of DNA around nucleosomes enables the extreme compression of eukaryotic genomes into condensed chromatin fibers. However, nucleosomal DNA can be inherently restrictive to access of transcriptional machinery³ since within a nucleosome, part of the DNA faces the globular domains of the histones, causing the DNA sequence on that side to be hidden sterically⁴. Supplementing nucleosomal restriction, posttranslational modification of histones can further facilitate chromatin compaction. For example, the histone H3’s tail can be covalently methylated at lysine 9 (H3K9me2 or H3K9me3)⁵ or lysine 27 (H3K27me3)^6,7, and these modifications are linked to additional chromatin compaction. The lamin-enriched nuclear periphery⁸ associates with H3K9me2 heterochromatin to silence genes⁹. On the other hand, H3K9me3-associated heterochromatin occurs in apparent phase-separated globules throughout the nucleus¹⁰. In contrast to repressive histone modifications, chromatin decompaction and nucleosome-free regions can be flanked by unique combinations of histone modifications. A different lysine residue, lysine 4 on histone H3’s tail can be covalently methylated (H3K4m1, 2, or 3) or lysine 27 can get acetylated, and nucleosomes carrying these modifications flank accessible regulatory elements acting as enhancers and promoters^11,12.

1.2. Transcription factors and chromatin remodelers

In every developmental program, the chromatin landscape is shaped by a highly coordinated division of labor among different classes of transcription factors. The first task is carried out by a small number of lineage-determining transcription factors often referred to as “pioneer transcription factors” (a comprehensive review can be found here⁴). These transcription factors are unique because of their intrinsic ability to bind their cognate DNA sites directly on the nucleosome. Lineage-determining transcription factors, such as EBF1^13-15, TCF-1¹⁶, and FoxA1¹⁷, initiate the cooperative interactions with chromatin remodelers to evict nucleosomes and create accessible chromatin regions. The second task force is coordinated by transcription factors such as JUNB, BATF, and IRF4^18,19, who add permissive modifications to flanking histones of accessible chromatin regions. Although such poised regulatory elements are not immediately involved in gene regulation, they are primed for future transcription, enabling a rapid response to changes in the cellular environment. The final task of firing transcription, particularly essential for both CD4⁺ and CD8⁺ T cells, is executed by transcription factors downstream of signal transduction pathways and is often carried out through acetylating histones. These signal-dependent transcription factors are in place to respond quickly to cytokines due to changes in the cellular environment. For instance, the JAK/STAT pathway required for CD4⁺ T helper differentiation directly alters the histone acetylation landscape, while the NF-κB factors pervasively change histone modifications in response to inflammation^18,20–22.

Since most transcription factors have limited enzymatic activities, they are required to cooperate with chromatin remodeling proteins to alter nucleosome positions. Moreover, transcription factors can direct enzymes to write, read, or erase a unique combination of histone modifications in a cell-type specific manner. The orchestrated division of labor by transcription factors instructs chromatin remodelers to design the chromatin landscape for a specific T cell program in various time scales. While chromatin remodelers, such as those in the SWI/SNF family²³, have ATPase activities and can displace nucleosomes, other proteins, such as histone deacetylases (HDACs), have enzymatic activities to write or erase covalent histone modifications. The conditional deletion of few chromatin remodelers in T cells has been investigated. It has been shown that histone deacetylase (HDAC) 3 is required for T cell development²⁴. More specifically, HDAC3’s function in restraining CD8-lineage genes in DP thymocytes for the generation of CD4 T cells has been recently reported²⁵. Conditional deletion of Jmjd3 and Utx, the H3K27me3 demethylases, in CD4⁺ T-cell precursors demonstrated that these enzymes are redundantly required for H3K27me3 deposition and gene expression²⁶. Despite these reports (reviewed here²⁷), there is a paucity of information on transcription factors and chromatin remodelers working together to endow competence for inducing T cell fate. Although it has been shown that changes in metabolism play a role in regulating T helper cell differentiation programs by changing the CTCF-binding events and genome interactions^28,29, the extent by which changes in metabolism instruct transcription factors to recruit chromatin remodelers and alter genome topology for a specific T cell response remains to be understood. Whether dedicated histone variants³⁰ or T-cell specific chromatin remodelers are involved at various stages of T cell responses remains to be explored. The new windfall of unbiased CRISPR screening efforts in T cell subsets^31–33 may further unravel the underappreciated roles of chromatin remodelers in T cells.

1.3. Three-dimensional chromatin architecture

Nucleosome wrapping is not adequate to fit the 2-meters of DNA into the micrometer space of the nucleus. Mammalian genomes are further folded into sophisticated three-dimensional (3D) configurations, dictating a diverse range of cell fates³⁴ (Figure 1). The continuous advances of genomic and imaging technologies have enabled the query of higher-order chromatin architecture at an unprecedented resolution and scale^35–37. The emerging theme indicates that the genome is hierarchically packaged into territories³⁸, compartments³⁹, topologically associating domains^40,41, in addition to cell-type specific features including DNA loops⁴², 3D cliques^43,44, and architectural stripes⁴⁵. Evident from early imaging studies, every interphase chromosome occupies separated chromosome territories in the nucleus⁴⁶. For each chromosome territory, the positioning of DNA is dictated by transcriptional activity^34,47, partitioning the entire genome into active (A) and inactive (B) compartments. Genomic regions predicted to be in the A compartment are generally in transcriptionally active euchromatin state, and additionally contain transcribed genes, accessible chromatin, and histone modification signatures of active enhancers and super-enhancers⁴⁸. In contrast, genomic regions labeled as the B compartment contain inactive genes with histone modifications associated with heterochromatin and repressed chromatin states. The heterochromatin often identified by H3K9me2 and H3K9me3 occurs at the nuclear lamina⁴⁹ and is typically localized in the B compartment.

Figure 1. Hierarchical genome folding.

10 nm chromatin fiber wrapped around nucleosomes, loops, TADs, A/B compartments, territories, and nucleus.

At the sub-megabase scale, the genome is organized into segments variously called topologically associating domains (TADs)^40,41, contact domains⁵⁰, or insulated neighborhoods⁵¹. These regions are defined such that there is a higher probability of interactions within them than with their neighboring regions. Most recent high-resolution imaging experiments revealed that TAD boundaries are stochastically present in single cells⁵². Despite the conservation of TADs across cell types and species, the detailed molecular processes through which TADs are formed and their functional relevance have been the topic of intense debates^53,54. The enrichment of CTCF and the cohesin complex at TAD boundaries implies the roles of these structural proteins in TAD formation^41,42. The loop extrusion model is among few plausible models explaining how TADs are formed^55,56. In this model, the ring-shaped cohesin complex is involved in extruding a DNA loop, similar to threading yarn through the eye of a needle⁵³, until it reaches an extrusion barrier associated with the DNA which is likely to be CTCF binding sites with convergent motifs⁴². The field started questioning if TADs are functionally important for cells when cohesion degradation wiped out TADs but led to an underwhelming effect on the global transcriptional outputs⁵⁷.

Despite the long-standing debates on whether form dictates function, there is considerable evidence that TAD formation restrains enhancer-promoter loops⁵⁸. The cell-type specific enhancer-promoter interactions also engage transcriptional cofactors such as Mediator, cohesion complex, and CTCF⁵⁹. The latest reports also confirm the formation of communities of enhancers and promoters variously called cis-regulatory domains (CRDs)⁶⁰, activation hubs⁶¹, or 3D cliques⁴⁴, all contained within TADs. In addition to the idea of communities of enhancers, architectural stripes⁴⁵ (also referred to as ‘flames’⁵⁹ or ‘lines’⁵⁶) were recently characterized. Stripes refer to the observation that one region can act as an anchor, interacting with the entire domain at high frequency, a process which can be explained by a one-sided loop extrusion model. The extent by which such hierarchical organization is linked to function is an active area of investigation.

2. Experimental and analytical techniques to map genome organization

2.1. Genomic technologies to map transcription factor binding and nucleosome positioning

The arsenal of molecular biologists started to transform with the advent of next-generation sequencing more than a decade ago⁶². Hundreds of protocols have been proposed to measure various aspects of genome organization and gene regulation including chromatin accessibility, nucleosome positioning, transcription factor binding, histone modification, and DNA methylation. Although summarizing these techniques is beyond the scope of this review, here we will discuss the two most widely used assays by immunologists: ChIP-seq and ATAC-seq. Chromatin immunoprecipitation followed by sequencing (ChIP-seq) is undoubtedly the most popular sequencing-based technique. ChIP-seq is used for mapping DNA-binding proteins or modified histones in a genome-wide manner⁶³. In this protocol or its numerous variations, a critical step is to crosslink cells with formaldehyde. The next step involves solubilizing the entire cellular content to fragment the chromatin. Finally, the chromatin fragments are pulled down with an antibody for the histone modification or transcription factor of interest⁶⁴. These major steps make ChIP-seq an inefficient protocol since it requires a significant number of cells, e.g. 10⁷ cells, as the starting material. The recent competitor of ChIP-seq is the Cleavage Under Targets and Release Using Nuclease (CUT&RUN) proposed by the Henikoff lab^65,66. In this protocol, intact permeabilized cells are incubated with an antibody against a transcription factor or histone modification, after which, a protein A-MNase fusion directs the cleavage of chromatin footprint. In contrast to ChIP, CUT&RUN is performed in situ in the absence of formaldehyde crosslinking without total genome fragmentation and solubilization. The in situ nature of CUT&RUN allows high-resolution footprinting of protein-DNA interactions with minimal background, such that as few as 100 cells or even single cells⁶⁷ can be used for mapping protein-DNA interactions. The low cell number requirement of CUT&RUN makes this protocol appealing for studying direct transcription factor targets in T cell subsets.

The assay for transposase-accessible chromatin using sequencing (ATAC-seq)⁶⁸ was described as an alternative advanced method for DNase-seq also called DNase I hypersensitivity assay^69,70. In DNase-seq, the nuclease DNase I leads to double-strand breaks since it is able to nick the complementary strands of DNA one strand at a time⁷¹. In ATAC-seq⁶⁸, the transposase Tn5 cleaves DNA with some degree of sequence specificity. Although DNase-seq protocols were popular during the first phase of ENCODE project⁷⁰, the input material requirement of conventional DNase-seq experiments hindered their utilization in biological contexts such as studying primary cells of the immune system where millions of cells are not available. The ATAC-seq protocol exploited the idea of ‘tagmentation’, which is to simultaneously cut and ligate adapters for high-throughput sequencing at open chromatin regions. The paired-end sequencing of insertion ends charts chromatin accessibility of the entire genome. The primary reasons for the explosive utilization of this technique (~12,000 entries in NCBI GEO and ~4000 pubmed publications since 2013) relates to (a) a relatively simple and short protocol and (b) the low input requirement of ~ 50,000 cells and even fewer cells in more recent variations of this protocol. It has also been suggested that the position of nucleosomes can be inferred from ATAC-seq⁷². A comprehensive overview of ChIP-seq, ATAC-seq and their variations can be found here⁷³. Adapting these assays at the single-cell level and combining molecular features such as gene expression and chromatin accessibility are the intense focus of technology developments of the last couple of years^74–76.

2.2. Genomic and imaging Technologies to map 3D genome folding

Two complementary techniques each with different strengths and weaknesses are used to study the 3D genome organization: microscopy-based imaging assays and chromosome conformation capture (3C)-based sequencing methods. The imaging tools directly assess spatial distances between genomic loci at the single-cell level but are often limited in genome coverage. The complementary approaches are the numerous genome-wide variations of 3C-technologies, which measure contact frequencies of two genomic loci as a proxy for spatial proximity across a population of cells.

2.2a. Imaging-based tools

Prior to the invention of genome-wide techniques, the major assay for measuring nuclear organization was the revolutionary fluorescence in situ hybridization of DNA (DNA-FISH)⁷⁷. DNA-FISH fluorescently labels select genomic loci using probes that hybridize to DNA sequences in the nucleus⁷⁸. In the hybridization step, a single-stranded probe enters the nucleus facilitated by permeabilizing cells with a detergent or organic solvent, such as methanol. The DNA is typically denatured enabling the probes to bind to their target sequences. Microscopy is then used to visualize the genomic loci labeled by the hybridized fluorescent probes. The key limitations of DNA-FISH are its low resolution and restricted genome coverage. Improvement in the FISH resolution has been obtained by implementing super-resolution microscopy (reviewed in⁷⁹) and short oligonucleotide-based probes also called Oligopaints^80–82. A multiplexing FISH strategy aims to tackle the limited genome coverage by incorporating multiple rounds of hybridizations and sequential imaging, tracking the spatial positions of an increasing number of genomic loci simultaneously^52,80. Although super-resolution imaging and major technical advances in DNA-FISH allowed researchers to investigate chromatin organization at unprecedented resolutions, a major inherent feature of DNA-FISH experiments relates to the high levels of cell-to-cell variations. Hence, a large number of cells is required to rigorously study various features of the nuclear organization across individual cells.

2.2b. Chromosome Conformation Capture (3C) technologies

Complementary to visualizing the 3D genome topology using image-based tools, the 3C technology and its numerous variations⁸³ offered a way to measure pairwise contact frequencies between two genomic loci as a proxy for their spatial proximity. In this genomic strategy relying on a population of cells, the nucleus is chemically fixed to maintain the 3D genome conformation, followed by DNA fragmentation and religation. When two DNA fragments are spatially proximal during the cross-linking step, through a process known as proximity ligation, DNA fragments are ligated⁸⁴. The frequency of the ligation products also referred to as the “contact frequency” can be measured by PCR, DNA microarrays, or sequencing techniques^39,84,85. In recent years, numerous optimizations of the 3C approach resulted in an explosion of techniques that map genomic organization with various scales of coverage and resolution.

While reviewing the evolution of these techniques is beyond the scope of this review, we will discuss two widely used assays, Hi-C³⁹ and HiChIP/PLAC-seq^86,87. The genome-scale version of the 3C technique, Hi-C, offers an unbiased measurement of any potential interaction across the genome³⁹. A major advancement in the original dilution Hi-C technique³⁹ was the in situ Hi-C⁴². The critical modification was to perform the ligation step in situ inside the nucleus, a constrained space, instead of in solution, where DNA fragments from different cells may ligate incorrectly. This modification led to a major reduction in the frequency of random ligation. This in situ ligation step also made the identification of long-range interactions associated with specific proteins of interest feasible. PLAC-seq and HiChIP were introduced as the optimization of the relatively inefficient ChIA-PET^86,87. The trick to improve efficiency of ChIA-PET in PLAC-seq and HiChIP protocols was following the in situ Hi-C approach. First DNA fragmentation and in situ proximity ligation are performed in the cross-linked cells. To enrich for interactions associated with a specific protein or histone modification, chromatin immunoprecipitation is performed for antibodies detecting protein or histone modifications of interest and is followed by the enrichment of biotinylated ligation junctions⁸³. Although the ideal strategy is to study the 3D genome interactions in an unbiased manner, Hi-C’s primary limitation is its requirement for an unreasonably high sequencing coverage. Strategies such as PLAC-seq and HiChIP are fitting when high-resolution dissection of enhancer-promoter interactions is required but sequencing coverage is limited.

2.3. Computational approaches to analyze genomic measurements

2.3a. ChIP-seq, CUT&RUN and ATAC-seq.

The analysis of 1D genomic data generated by ChIP-seq, CUT&RUN, and ATAC-seq involves the chained execution of many computational tasks, often using command line applications. A critical step for the analysis of these assays is to perform peak calling. The peak-calling step identifies significantly enriched loci for the protein or histone modification of interest (peaks) in the genome. While numerous peak-calling tools have been developed (reviewed here⁸⁸), macs2⁸⁹ is the most commonly used peak-calling tool for ChIP-seq and ATAC-seq. A selective peak calling technique called SEACR, which uses the global distribution of background signal to calibrate a simple threshold for peak calling, has been recently developed for CUT&RUN⁶⁵.

In analyzing large numbers of ChIP-seq or ATAC-seq data sets across different genotypes or experimental conditions, the main source of computational irreproducibility arises from a lack of good practice pertaining to software and database usage⁹⁰. In particular, different parameters or even software versions may alter key biological findings. For example, an alternative processing strategy for RNA-seq and H3K27ac datasets using stringent alignment parameters for different strains of mice revealed the importance of controlling for sequence variation in different mouse models in particular the congenic and backcross mice⁹¹. Such concerns about the reproducibility of data analysis has led to the development of in silico workflow management systems such as Nextflow⁹² and Snakemake⁹³. These strategies empower rapid pipeline development through the adaptation of existing pipelines written in most scripting languages.

2.3b. 3D genome organization data analysis.

Thanks to international consortiums like ENOCDE⁹⁴, various standards and guidelines⁹⁵ have been proposed for ChIP-seq and ATAC-seq data analysis. However, computational strategies for the 3D genome topology measurements are at its early stages mostly due to the complexity of the genome architecture and sparsity of measurements. The goal of the 4D Nucleome Network⁹⁶, now entering its second phase, is to develop computational and experimental standards to map the structure and dynamics of the human and mouse genomes in space (three dimensions) and time (the 4^th dimension). Although there is some consensus on the analysis of A/B compartments or TADs, detection of higher resolution (Kb) interactions are much more challenging across data sets. From the original Hi-C publication in 2009, the plaid pattern of Hi-C data after normalization and conversion of a raw contact matrix into an observed over expected matrix revealed the segregation of mammalian genomes into two compartments (A/B)³⁹. Genomic regions with a positive value for the first principal component of this normalized matrix are assigned to the A compartment while a negative value is assigned to the B compartment. A relatively common approach to quantify TADs is to calculate an ‘insulation score’ to genomic intervals along the chromosome. The score reflects the aggregate of interactions occurring across each interval where minima of the insulation profile denote areas of high insulation and can be classified as TAD boundaries⁹⁷. Analytical strategies for detecting other architectural features such as 3D cliques^43,44 or stripes⁴⁵ have also been presented. Long-range interactions within TADs or chromatin loops are found in most Hi-C contact maps. These features are characterized by the presence of ‘foci’ or ‘corner dots’⁵⁴—a group of adjacent pixels with significantly higher interaction frequency compared with that of the neighboring genomic regions. Despite multiple methods^98–101, loop calling is the most challenging part of Hi-C or HiChIP/PLAC-seq data analysis since it represents the major resolution-sensitive step where high sequencing coverage is required.

3. Transcription factors as genome organizers in T lymphocytes

Although the mammalian genomes potentially contain millions of regulatory elements mostly in the form of enhancers¹⁰², only a small subset of them is engaged in gene regulation in any particular cell type such as T lymphocytes. It is much appreciated that a network of transcription factors expressed at various stages work together to set up the 1D chromatin landscape of T cells. The evidence that enhancer-promoter contacts are only occasionally linked to CTCF-CTCF interactions suggests the possibility that cell-type specific transcription factors may participate in enhancer-promoter interactions⁵⁹. Despite these recent findings, transcription factors with structural roles in T cells are yet to be identified. There are some indications that transcription factors such as Bcl11b, which is the critical regulator of T cell commitment¹⁰³, are able to change the 3D genome organization¹⁰⁴. Yet, the detailed molecular mechanism through which this occurs remains to be understood.

3a. Hyperconnectivity of 3D genome harboring T-lineage transcription factors

To globally model the 3D genome organization of regulatory elements in thymocytes, our own group recently developed an analytical approach and algorithmically searched for communities of densely connected enhancers and promoters measured by HiChIP^43,44. Although HiChIP for the H3K27ac modification or cohesin subunit Smc1 can generate a biased view of enhancer-promoter interactions, they enabled us to dissect higher resolution short-range interactions with a reasonable sequencing coverage (~500 million reads). Referring to these interconnected elements as “3D cliques”, we observed asymmetry in the connectivity distributions with distinct 3D community topology. The majority of 3D cliques contained very few interactions. However, more than 100 cliques, including thousands of regulatory elements, were categorized as “hyperconnected” in DP T cells. Here, we would like to propose that this unbiased strategy to prioritize the genome of T cells based on 3D connectivity may reveal salient yet uncharted regulatory nodes in T cells. Supporting this notion, the megabase pair genomic region containing Bcl11b was the most hyperconnected 3D clique in T cells. Strikingly, the non-coding RNA called ThymoD at the Bcl11b locus has been shown to be essential in changing the 3D genome organization of this locus during T cell development¹⁰⁵ (Figure 2). Moreover, Runx1, Ets1, and Bcl6 loci scored as hyperconnected with intriguing 3D genome architectures in thymocytes (Figure 2). The functional relevance of Bcl6 as a boundary gene, Ets1 and Fli1 forming a hyperconnected 3D clique, and Runx1’s end forming long range interactions with its megabases away promoter are yet to be determined. We put forward an idea that such complex architectural patterns at genomic loci harboring transcription factors with salient roles in T cells may endow T cells to precisely control the levels of these transcription factors at the right time in the right place during antigen-specific responses. Further genetic perturbation studies are required for fully evaluate this idea and the functional relevance of such hyperconnected regions.

Figure 2. Hyperconnected 3D cliques in thymocytes.

Contact frequencies from Smc1 HiChIP were normalized by juicebox at key T cell transcription factors: Bcl11b, Ets1 and Bcl6 loci.

3b. Shaping genome organization in T lymphocyte development

The role of lineage-determining transcription factors on establishing the chromatin accessibility landscape of B cells and macrophages has been the focus of numerous studies^{13,14,21,106–110}. A pioneering study from the Natoli lab demonstrated that PU.1 binding events directly map to open chromatin regions in macrophages¹⁰⁷. Remarkably, PU.1, which is also expressed in early stages of T cell development, participates in gene regulation of T cells by recruiting transcription factor collaborators to its recognition sites and by depleting the collaborators from their preferred sites when PU.1 is not expressed¹¹¹. In B cells, EBF1 has the ability to alter the chromatin state at least in part through its C-terminal domain recruiting the SWI/SNF family proteins^13–15. The lineage-determining transcription factors with key roles in normal development also enable changes in cell fate during cellular reprogramming. An example includes the transcription factor C/EBPα, which can initiate the transdifferentiation of B cells towards macrophages by activating regulatory elements of macrophages¹⁰⁹.

Although epigenetic landscapes of CD4⁺ T helper cell differentiation^{20,48,112–114} and CD8⁺ T responses^19,115–119 have been extensively studied, only recently transcription factors with parallel roles on the chromatin accessibility during T cell development have been reported. Two studies including one from our group revealed the role of Tcf-1 and its collaborator in setting up the chromatin accessibility landscape of T cells^16,120. A network among transcription factors, including Tcf-1, Gata3, and Bcl11b, regulates the distinct phases of T cell development in the thymus. In addition to the expression of these essential transcription factors, restriction of alternative-lineage factors, for example PU.1 and Bcl11a, is also necessary for proper T lineage determination¹²¹. Notch1 signaling induces Tcf-1, encoded by Tcf7, in early T cell progenitors and the Tcf-1 expression level remains high until T cell maturation. The expression of Bcl11b, which is necessary for T lineage commitment, in addition to Gata3 are dependent on Tcf-1 expression^103,122. The constitutively expressed transcription factors in the hematopoietic system such as Ets1, E2A and Runx1 are not only enriched at T cell specific regulatory elements but are also required for proper T cell development^123,124.

Our group mapped open chromatin regions at eight stages of thymic T cell development in mice using ATAC-seq. Our integrative analysis revealed a significant enrichment of the HMG motif, which is Tcf-1’s recognition site, at genomic regions that become accessible at the earliest stage of T cell development and persist until maturation¹⁶. Germline deletion of Tcf-1 leads to a major reduction in thymocyte numbers¹²⁵ while overexpression of Tcf-1 in bone marrow progenitors can upregulate Bcl11b, Gata3, and other T cell specific genes¹²⁶. Despite these functional reports dating back to twenty years ago, detailed molecular mechanisms through which Tcf-1 controls T cell identity remained elusive. A small number of T cells are able to develop in Tcf-1 deficient mice although they have major functional defects¹²⁷. Supporting the functional reports, the chromatin accessibility and transcriptional profiles of Tcf-1-deficient mice were very different from those of normal T cells. Moreover, Tcf-1 binding events, in a dramatic contrast to those of Runx1 or Gata3, showed an interesting pattern across single cells when chromatin accessibility was mapped by scATAC-seq¹⁶. Tcf-1 bound regions were revealed to be more homogenously accessible across individual cells, suggesting a deterministic effect of this lineage-determining transcription factors across individual cells. Unexpectedly, the ectopic expression of Tcf-1 in fibroblasts unmasked the ability of Tcf-1 to bind to nucleosome-occupied regions, generating de novo chromatin accessibility. In contrast to other well-established pioneer factors such as FoxA1, Tcf-1 is able to erase the pre-existing repressive marks in fibroblasts, implying that such major chromatin remodeling events may also be Tcf-1 dependent in T cell development. Moreover, Gounari and colleagues discovered the cooperation between Tcf-1 and the E protein, HEB, encoded by Tcf12, in shaping the chromatin accessibility landscape during T cell development¹²⁰. It was demonstrated that the binding of both Tcf-1 and HEB is required at shared binding sites for epigenetic and transcriptional gene regulation. Cooperative binding of Tcf-1 and HEB to their conserved recognition sites with T cell-specific enhancers promoted their expression. A large portion of these binding events occured at promoter regions and lacked the consensus recognition site of Tcf-1 or HEB. The analysis of gene expression after conditional deletion of Tcf-1 suggested that those binding events occurring at promoters of cell-cycle-associated genes can limit cell proliferation. Together, these two complementary studies corroborate that Tcf-1 and its close collaborators establish the epigenetic and transcription profiles of DP thymocytes. Despite these reports, what specific chromatin remodelers collaborate with Tcf-1 and how the genome organization is remodeled by this protein remain unknown. It is evident from this study and other reports that a combination of transcription factors collaborate on the chromatin to establish the permissive chromatin state. Relying on natural genetic variation in inbred strains of mice is a great unbiased approach which can define transcription factors and their collaborators with significant effects on chromatin accessibility^43,60,91,128.

Elegant strategies from the Rothenberg laboratory has revealed an asynchronous combinatorial relationship among Notch signaling, Tcf-1, and Bcl11b for T cell commitment¹²². Three distinct, asynchronous mechanisms are proposed between Notch signaling and Notch-activated transcription factors, e.g. Tcf-1. Tcf-1 and GATA-3 can work through an early locus priming function, corroborating the reports on Tcf-1 and chromatin accessibility alteration. Notch signaling is in charge of a stochastic-permissivity function while Runx1, which is not a T-cell specific transcription factor, controls the level of gene expression. This gene regulatory network is stage specific and can provide a hierarchical mechanism for developmental gene regulation¹²².

An integrative analysis of 1D and 3D genome organization data i.e. chromatin accessibility, topologically associating domains, AB compartments, and gene expression from HSPCs to DP T cells revealed that unexpected genome-wide changes at all three levels of chromatin organization can occur during the transition from double-negative stage 2 (DN2) to DN3¹⁰⁴. The transcription factor Bcl11b, was shown to be associated with increased chromatin interaction, and Bcl11b deletion limited 3D genome interaction at Bcl11b dependent genes. Together, a combination of T-cell specific and hemetopoesis-related transcription factors set up the epigenetic landscape of T cells during development. Whether and how poising of the chromatin from 1D to 3D perspectives prepares T cells for their specialization in periphery remains to be investigated.

3c. Shaping genome organization in CD4⁺ T helper cell differentiation

Understanding the principles of genome organization is a relevant topic for host defense in particular CD4⁺ T helper cell differentiation. For balanced immunoregulation, naive CD4⁺ T cells should make numerous developmental decisions. Such calculated decisions aim to eliminate microbial pathogens but not cause autoimmunity. Indeed, the variety of T helper cell programs from T helper type 1 (Th1), Th2, and Th17 cells to regulatory T (Treg) cells, and many more new subsets, is overwhelming. Remarkably, the Th1/Th2 paradigm has been extensively used as early as 1990s as a model to study gene regulation and chromatin organization even in the absence of today’s modern techniques¹²⁹.

Plasticity of CD4⁺ T cells has been the focus of many debates. Early in vitro reports suggested that CD4⁺ T cell fates are fixed, even in the presence of cytokines that can drive differentiation to the opposing populations^130–132. This notion has been changed due to numerous reports demonstrating that polarized T cells can change cell fate towards mixed or alternative lineages¹³³. An example includes lineage-tracing experiments in mice, which revealed that distinct CD4⁺ T helper subsets can alter their lineage specification during their lifespan^134–137. Other studies showing the phenotypic plasticity of regulatory T cells in parallel with their cognate T helper cell subsets posits the idea that T cells are inherently flexible, being both inflammatory and regulatory, to adapt to the ever-changing environments¹³³. Considering the lessons learned from cellular reprogramming, we would like to propose that CD4⁺ T cells can be reprogramed to any subset as long as they express the cytokine receptors responding to changes in the cytokine environment. The most extreme case of cell fate change, which is the ability to induce pluripotency in fibroblasts by only four transcription factors, advocate for transcription factors with the ability to alter the chromatin state as the molecular mechanism of CD4⁺ T cell plasticity.

T cell receptor (TCR) can change the combinatorial landscape of histones in T cells¹³⁸. The lineage-specific cytokines with their downstream transcription factors can lead to genome-scale chromatin reorganization^20,112. For any modification of the extracellular environment, cytokine signaling is propagated through available cytokine receptors and cytoplasmic signaling events, all of which are required to interpret the instructions embedded in genomic DNA sequences. Downstream of these signal transduction pathways, there are transcription factors with diverse abilities to remodel the chromatin²⁰. Branding transcription factors to ‘master’ regulators was popular among T cell biologists¹³⁹. Nonetheless, we believe that semantics such as ‘pioneer’ or ‘master’ regulator in the absence of quantitative definitions can be misleading. Moreover, the coexpression of such transcription factors now corroborated by the avalanche of single-cell RNA-seq reports in mice and humans questioned the concept that the expression of a single master regulator can be used to define a CD4⁺ T cell phenotype¹⁴⁰. We suggest that high-resolution molecular measurements such as chromatin accessibility or genome folding across CD4⁺ T cell subsets can be used to quantitatively define a lineage.

One might expect the defined master regulators cause widespread effects on active enhancer landscapes, but early genome-wide maps in T helper cells showed that this is not the case. Loss of T-bet did not change the Th1-specific enhancers²⁰. In contrast, T-bet had a generally negative impact on p300 binding suggesting its role as a repressor rather than an activator based on p300 binding data²⁰. Similarly, Ciofani et al demonstrated that the absence of Rorγt had a minor effect on p300 occupancy in Th17 cells¹¹². Around the same time, Rudensky and colleagues reported that Foxp3 is dispensable for chromatin accessibility of Treg cells¹¹³. Samstein et al revealed that Foxp3 defines Treg cell functionality by exploiting the already accessible enhancer landscape instead of establishing a new one¹¹³. Recently, the same group added more mechanistic insights into this observation¹²⁸. The Rudensky lab leveraged known differences between inbred strains of mice and examined disruption in which transcription factor binding sites due to sequence variation between alleles can be linked to changes in chromatin accessibility. Strikingly, sequence variation within binding sites of HMG-proteins largely affected Treg cell-specific chromatin accessibility. Moreover, expression levels of the HMG transcription factor, Tcf-1, and Treg cell-specific chromatin accessibility were highly correlated. These data suggest that a relatively small reduction in Tcf-1 expression can cause large-scale loss of chromatin accessibility in Treg cells. Hence, in order to create a distinct cell fate, not all lineage-determining transcription factors seem to have the biochemical capability to remodel the chromatin in a direct manner. This study further supports the notion that there is a division of labor among transcription factors when it comes to altering the chromatin landscape. Although open chromatin regions define the regulatory potential of a cell and hence they are often used as a proxy to evaluate cell fate, whether Foxp3 can control other epigenomic features such as unknown histone modifications or unique histone variants necessary for Treg cell fate remains to be studied.

Signal transduction pathways through downstream transcription factors such as STAT3, STAT4, and STAT6 can pervasively change histone acetylation deposition in T helper cells¹¹². Importantly, ectopic expression of T-bet or GATA3 in STAT4- or STAT6-deficient cells failed to recover STAT-dependent enhancer landscapes. These large-scale histone acetylation changes were mirrored in changes of gene expression. The roles of STATs in both limiting and promoting gene expression became clear by the demonstration that STATs can also lead to unloading of p300 at enhancers of opposite lineages. Other reports suggest that IL-2 signaling is essential for maintaining accessible chromatin regions at hundreds of gene regulatory elements of T cell differentiation pathways¹⁴¹. More specifically, IL-2 signaling can activate AP-1 and STAT5 proteins that can subsequently bind lineage-determining transcription factors, depending on other proteins downstream of extracellular environments of T cells. Although earlier work was only possible to study the role of STATs in vitro, more recent studies also revealed the in vivo effect of STAT proteins in particular STAT5. Persistent STAT5 activation is able to reprogram the epigenetic landscape in CD4⁺ T cells to drive polyfunctionality and antitumor immunity¹⁴². Together, these studies suggest that signaling-dependent transcription factors such as STATs directly convey changes of the cellular environment to the chromatin environment.

3d. Shaping genome organization in CD8⁺ T cell responses

Naive CD8⁺ T cells are able to clonally expand and differentiate into effector CD8 (Teff) cells during acute infection or after vaccination. The major outcome of this expansion is to directly kill target cells and control infections. Teff differentiation is mediated by transcriptional, epigenetic, and metabolic reprogramming as well as the acquisition of effector signatures such as the ability to produce cytokines, chemokines, and cytotoxic molecules¹⁴³. Most of activated T cells die after antigen clearance and resolution of inflammation. Despite a dramatic degree of cell death, a small subset of cells contract to a long-lived pool of memory T (Tmem) cells. Tmem cells downregulate their effector program and acquire a stem cell–like ability to survive¹⁴⁴. During chronic infections and in cases where antigen stimulation persists such as cancer, T cell memory formation is unsuccessful and T cells become exhausted. A comprehensive review on T cell exhaustion can be found here¹⁴³.

The feasibility of collecting millions of cells from the in vitro CD4⁺ T cell differentiation made T helper cells an early sought-after model for applying modern epigenetic techniques to dissect the role of transcription factors in gene regulation. However, the low cell requirement of ATAC-seq (~ 50,000) revolutionized our mechanistic understanding of CD8⁺ T cells in response to infection. Efforts from multiple groups mapping chromatin accessibility of CD8⁺ T cells after acute or chronic infection revealed that the acquisition of effector or memory phenotypes was associated with stable changes in chromatin accessibility away from the naive T cell state^{115,117,118,145}. Unexpectedly, the chromatin landscape of exhausted T cells was distinct from functional memory CD8⁺ T cells, suggesting that exhausted T cells are a distinct developmental lineage. Reproducibly described in these early studies differentially accessible chromatin regions among distinct in vivo CD8⁺ T cell subsets were enriched for recognition sites of transcription factors known to regulate each subset. The exhaustion-specific accessible regions were enriched for consensus binding sites of NFAT and Nr4a family members, implying that these transcription factors, whose expression levels are not exhaustion specific, may play a role in shaping the chromatin accessibly landscape¹¹⁷.

To study the role of NFAT in exhausted T cells, it was shown that the ectopic expression of a modified form of NFAT1, which is not able to interact with AP-1 transcription factors, can limit T cell receptor (TCR) signaling, but enhance the expression of inhibitory cell surface receptors, and interfere with the ability of CD8⁺ T cells to protect against infection and weakened tumor growth in vivo¹⁴⁶. Comparing the genomic regions occupied by endogenous and the modified form of NFAT1 in primary CD8⁺ T cells revealed that genes directly induced by the modified form of NFAT1 had a significant overlap with the signature genes of exhausted CD8⁺ T cells in vivo. It was suggested that NFAT promotes T cell exhaustion in addition to T cell anergy by binding at genomic loci that do not require the cooperation with AP-1 family transcription factors¹⁴⁶.

The enrichment of the Nur77 motif within exhausted-specific open chromatin regions in multiple independent studies suggested the role of Nr4a family members in T cell exhaustion through remodeling the chromatin landscape^{115,117,118,145}. This hypothesis, generated by the comparison of ATAC-seq datasets across cell types, was tested by two groups and corroborated the role of Nr4a on defining the chromatin signature of exhausted T cells^147,148. Phenotypes and gene expression profiles of Nr4a triple knockout chimeric antigen receptors (CAR) tumor-infiltrating lymphocytes were similar to those of CD8⁺ effector T cells. Moreover, the uniquely accessible chromatin regions in Nr4a triple knockout CAR tumor-infiltrating lymphocytes were enriched for recognition sites of NF-κB and AP-1, which are transcription factors involved in activation of T cells¹⁴⁷. NR4a1 binding also promoted H3K27ac, leading to the activation of tolerance-related genes¹⁴⁸.

The exhaustion-specific expression of the HMG transcription factor, TOX, was reported in a 2012 study from the Wherry lab¹⁴⁹. This work aimed to dissect pathways centrally involved in exhaustion versus memory by defining the transcriptional coexpression networks of CD8 T cells using microarray analysis. It also revealed differences between exhausted and memory CD8⁺ T cells and reported TOX, among many other genes, as a centrally connected hub gene in exhaustion versus memory¹⁴⁹. Seven years later, an unprecedented number of publications^150–154 from independent groups reported that TOX is functionally required for CD8⁺ T cell responses during chronic infection and in cancer. Using the chronic infection model, it was shown that TOX was largely dispensable for the formation of Teff and Tmem cells, but it was critical for exhaustion: in the absence of TOX, Tex cells do not form¹⁵¹. Although TOX can be induced by calcineurin and NFAT2 signaling, it can become calcineurin-independent and sustain in Tex cells. Overexpression of TOX alone can create the chromatin accessibility landscape of exhausted T cells. Robust expression of TOX led to commitment to Tex cells by creating the transcriptional and epigenetic developmental landscapes of Tex cells. The importance of TOX on exhausted T cells was also established using a tumor specific T cell response¹⁵². It was shown that TOX is highly expressed in dysfunctional tumor-specific T cells from tumors. Deletion of Tox in tumor-specific T cells abrogated the exhaustion program, such as Cd244, Pdcd1, and Tigit, the chromatin of which remained largely closed, and retained high expression of Tcf-1. Furthermore, positive regulation of NR4A by TOX and of TOX by NR4A was reported¹⁵³.

Transcription factors with key roles on the chromatin of memory T cells have also been identified. A comprehensive examination of the chromatin accessibility landscape of naive T cells during TCR stimulation revealed that Runx3 governs chromatin accessibility during TCR stimulation and enforces the memory cytotoxic developmental program¹⁵⁵. Runx3, essential for memory T cell differentiation, can promote accessibility to memory cytotoxic-specific cis-regulatory regions before the first cell division. Runx3 was specifically required for accessibility to regions highly enriched with IRF, bZIP and PRDM1-like recognition binding sites, upregulation of IRF4 and Blimp1, and activation of cytotoxicity attributes in early effector and memory precursor cells. Hence, it was suggested that Runx3 governs chromatin accessibility during TCR stimulation and enforces the memory T cell developmental program.

Genome-wide profiling of active (H3K27ac) and repressed (H3K27me3) histone modification in naive, Tmem, and Teff CD8⁺ T cells during viral infection also unmasked the large H3K27me3 deposition at multiple pro-memory and pro-survival genes in Teff cells. These histone modification data suggest fate restriction at both pro-memory and pro-effector genes in Tmem cells¹⁵⁶. Loss of polycomb repressive complex 2 led to limited clonal expansion and Teff cell differentiation, but did not affect CD8⁺ memory T cell maturation. Remarkably, abundant H3K27me3 deposition at pro-memory genes occurred late during Teff cell development, and it was suggested that diminished transcription factor FOXO1 expression can be responsible for this gain in repressive histone modification. These findings proposed a dynamic model for loss of memory cell potential through selective epigenetic silencing of pro-memory genes in effector T cells. Together, recent studies unmasked prominent insights describing the mechanisms through wich multiple transcription factors establish the 1D chromatin organization of CD8 T cells in viral infection and antitumor immunity.

4. Concluding remarks

The last decade has witnessed tremendous improvements in genomics, imaging, genome editing, and phenotyping techniques, which can aid researchers to better study T cell fate determination in humans and mice. We may define ourselves as a molecular immunologist, focusing only on what happens inside the nucleus, or as a cellular immunologist, focusing only on the cell surface and the presence or absence of proteins to define T cell identity. However, our T cells are oblivious about these academic definitions. They utilize proteins expressed on their surface for communications and all other cellular compartments are essential to effectively clear pathogens. Their genomes are tightly folded in the 3D space of the nucleus, yet they are able to effectively express the right amount of proteins at the right time. We would like to argue that it is time to have wholistic views for T cells. The explosion of multimodal single-cell assays such as CITE-seq, where epigenetic or transcriptomic features can be combined with cell surface markers across individual cells, have made the divide between molecular and cellular immunology blurrier. It is now easier more than ever to measure the 3D genome organization of T cell subsets. Remarkable discoveries of factors defining T cell fates relied on first generating the “descriptive” reports of epigenomics or transcriptomics alteration across T cell subsets. Many years of investigations, sometimes by many groups using different genetic models, are required to confirm the functional relevance of major findings in descriptive studies. The integration of screening techniques in mice and human cells, quantitative strategies, mouse genetics, and genomics and imaging techniques available in single-cell or bulk levels can help us unmask major regulatory circuitry involved in establishing T cell fate.