Multi-modular structure of the gene regulatory network for specification and commitment of murine T cells

Multi-modular structure of the

gene regulatory network for

speci

cation and commitment

of murine T cells

Boyoung

Shin

and Ellen V.

Rothenberg

Division of Biology and Biological Engineering, California Institute of Technology, Pasadena,

CA, United States

T cells develop from multipotent progenitors by a gradual process dependent on

intrathymic Notch signaling and coupled with extensive proliferation. The stages

leading them to T-cell lineage commitment are well characterized by single-cell

and bulk RNA analyses of sorted populations and by direct measurements of

precursor-product relationships. This process depends not only on Notch

signaling but also on multiple transcription factors, some associated with

stemness and multipotency, some with alternative lineages, and others

associated with T-cell fate. These fa

ctors interact in opposing or semi-

independent T cell gene regulatory network (GRN) subcircuits that are

increasingly well de

fi

ned. A newly comprehensive picture of this network has

emerged. Importantly, because key factors in the GRN can bind to markedly

different genomic sites at one stage than they do at other stages, the genes they

signi

fi

cantly regulate are also stage-speci

fi

c. Global transcriptome analyses of

perturbations have revealed an under

lying modular structure to the T-cell

commitment GRN, separating decisions to lose

“

stem-ness

”

from decisions to

block alternative fates. Finally, the updated network sheds light on the intimate

relationship between the T-cell program, which depends on the thymus, and the

innate lymphoid cell (ILC) program, which does not.

KEYWORDS

gene regulatory network, early T cell development, transcription factors, gene expression

program, cell fate decision, thymus, epigenetic control, multipotency

1 Introduction

Conventional T cells provide lifelong pro

tection against infection and cancer by

recognizing their cognate antigens and mediating effector functions. To ensure that the

host can exert various immune responses in a context-speci

fi

c manner, T cells have extensive

diversity in their sub-lineages. The unique properties of T cells stem from the intersections of

functional similarities with different types of immune cells (

). Both T and B cells utilize

antigen receptors whose diversities are achieved by DNA recombination and selection

mechanisms. Distinct from B cells though, T cells can differentiate to various subsets

Frontiers in

Immunology

frontiersin.org

OPEN ACCESS

EDITED BY

Mihalis Verykokakis,

Alexander Fleming Biomedical Sciences

Research Center, Greece

REVIEWED BY

Barbara L. Kee,

The University of Chicago, United States

Avinash Bhandoola,

National Institutes of Health (NIH),

United States

*CORRESPONDENCE

Boyoung Shin

boyoung@caltech.edu

Ellen V. Rothenberg

evroth@its.caltech.edu

SPECIALTY SECTION

This article was submitted to

T Cell Biology,

a section of the journal

Frontiers in Immunology

RECEIVED

26 November 2022

ACCEPTED

11 January 2023

PUBLISHED

31 January 2023

CITATION

Shin B and

Rothenberg EV (2023)

Multi-modular structure of the gene

regulatory network for speci

fi

cation

and commitment of murine T cells.

Front. Immunol.

14:1108368.

doi: 10.3389/fimmu.2023.1108368

open-access article distributed under the

terms of the

Creative Commons Attribution

License (CC BY).

The use, distribution or

reproduction in other forums is permitted,

provided the original author(s) and the

the original publication in this journal is

cited, in accordance with accepted

academic practice. No use, distribution or

reproduction is permitted which does not

comply with these terms.

TYPE

Review

PUBLISHED

31 January 2023

DOI

10.3389/fimmu.2023.1108368

showing functional parallelism with distinct types of helper-like

innate lymphoid cells (ILCs) and natural killer (NK) cells, which

lack recombined antigen receptors but produce effector cytokines

and/or perform cytolytic functions in response to environmental

signals. In addition, T cells poss

ess proliferative potential and

generate self-renewing long-lived populations, which are features

shared with the multipotent hematopoietic progenitor cells and

stem cells, respectively.

A recent global comparison of a wide range of hematopoietic cell

types found that lineage-speci

fi

c transcription factor motif

“

signatures

”

distinguished the active chromatin patterns for nearly

every major analyzed hematopoietic lineage, suggesting the impacts of

distinct lineage-determining transcription factors, but that T-lineage

cells notably failed to show any cell type-speci

fi

“

signature

”

(

). How,

then, is the T-cell identity established and robustly maintained,

despite functions broadly overlapping with those of other immune

cells? We propose that this distinctive positioning as T cells can be

supported by combinatorial actions of transcription factors, instead of

relying on a lineage-speci

fi

“

master regulator

”

. T cells utilize many

transcription factors that are commonly employed by other types of

hematopoietic cells for their own respective lineage speci

fi

cation and

function. However, by precisely regulating combinatorial actions of

these transcription factors over time under the in

fl

uence of Notch

signaling, and possibly also using epigenetic chromatin changes to

inhibit reversibility, intrathymic hematopoietic precursors launch and

lock down the speci

fi

c T-lineage program. In this analytical review, we

bring together recent evidence that assembles a newly clear picture of

how this comes about.

2 Gene regulatory network

models as explanations of

developmental pathways

2.1 Gene regulatory networks

in development

Developmental progression to generate a specialized cell type

requires continuous, ordered changes in gene expression (

). As a

regulatory program, development must thus be distinct both from the

stable epigenetic mechanisms that maintain a mature cell type

’

identity (e.g., superenhancers) a

nd from random transcriptional

“

noise

”

, both of which have been much studied in cell lines (

While environmental signals are often essential to trigger and support

development, the unidirectional regulatory cascade that emerges

depends on the transcription factors that are present in the cell

receivingthesignalsandtheirimpactsonothertranscription

factors and cellular chromatin

states. The genes encoding

transcription factors and signa

ling components collectively

determine the regulatory state of the cell, and then the genes they

control (downstream genes) encoding effector molecules determine

the functional identity of a cell.

All these genes, whether they encode signaling receptors,

transcription factors, or effector proteins, are controlled by

cis

regulatory modules encoded in the genomic DNA, such as

enhancers, silencers, and insulators, largely

via

interactions with

trans

-acting sequence-speci

fi

c transcription factors (

Speci

fi

cally activated long noncoding RNAs (lncRNAs) can

contribute to chromatin states as well, although their actions are

still only characterized in a few cases (

–

). Therefore, the

important components driving development, the genes and their

regulatory modules, transfer information in a directional manner.

For example, transcription factor

“

”

binds to the regulatory elements

of target genes

and

, which in turn induces expression of other

regulatory factors,

“

”

and

“

”

. Importantly, the activities of these

cis

regulatory elements are never driven by single transcription factors

alone, but rather by combinations of factors, even though one or

another factor can be rate-limiting in a particular experimental

situation (

–

). As a result, gene regulation cascades are not

explained by a linear pathway but by a hierarchical network (

). These features of developmental regulatory networks can be

effectively captured by using topological network models, in which

the functional interactions are represented as inputs and outputs.

While there are some caveats about the interpretation of these

networks for hematopoietic development (

), topological

models are indispensable for compiling evidence to explain cell

state differences in terms of gene regulation mechanisms.

In this review, we will focus on the gene regulatory programs

utilized in the early stages of thymic T-cell development, in which

multipotent progenitor cells undergo de

fi

nitive T-lineage

commitment and establish T-cell identity. Some gene regulatory

network (GRN) circuits have been shown to promote cell type

stability, as in the pluripotency state of ES cells, while others have

been shown to drive ultra-rapid, deterministic cascades of change, as

in the early Drosophila embryo (

). As shown below, early T cell

development falls between these models. It includes both regulatory

subcircuits that resist change and regulatory connections that enforce

change; the stochastic balance between these network subcircuits is

likely to underlie the distinctively asynchronous kinetics of T cell

program entry (

). We review the major regulatory genes that are

involved in different sub-programs to establish T-cell fate, resolve

coherent but distinct program modules that need to be deployed, and

propose an updated gene regulatory network (GRN) model.

2.2 Technical requirements and challenges

for accurate GRN construction

There are signi

fi

cant challenges and caveats of experimental

strategies that are utilized to understand developmental GRNs. As

transcription factors need to bind to speci

fi

c sequences in the DNA in

order to work, genome-wide transcription factor occupancy data

should in principle help to map where the

“

direct

”

interactions of

the putative regulatory factor occur. In the past

fi

fteen years it has

become relatively easy to map transcription factor binding across the

genomebyChIP-seq(orCUT&RUN,orCUT&Tag)(

–

Potential transcription factor inputs to active regulatory elements

can also be mapped in the DNA even without direct evidence of

transcription factor binding, based on the enrichment of their motifs

in accessible chromatin in a given cell type (

–

). Mapping open

chromatin by DNase-seq or ATAC-seq and using the cell type-

speci

fi

c enrichment of motifs predicted to be bound by given

transcription factors (

–

) in the open chromatin has become a

Shin and Rothenberg

10.3389/

mmu.2023.1108368

Frontiers in

Immunology

frontiersin.org

powerful way to predict which transcription factors may be important

in that cell type (

–

). Even without any evidence that a particular

target gene actually responds to the presence of the transcription

factor itself, such predictive analysis can be a valuable step toward

perceiving network relationships at a global level (

). Adding

measured evidence for actual transcription factor occupancy at sites

around key genes, when possible, enables speci

fi

c network predictions

to be made (

However, actually confronting binding-based or motif-based

predictions with empirical tests of individual target gene responses

to transcription factor activit

y perturbations has given a more

complex picture. Binding shou

ld not be overinterpreted.

Transcription factor-DNA interaction sites in a cell type are often

much more numerous (order of 10

sites) than the number of genes

that change expression in response to the loss or gain of a given

transcription factor in that cell type (order of 10

genes) (

–

). This

indicates that a given transcription factor

’

soccupancyisoften

dispensable for regulation of most of the genes that it binds. Even

when the transcription factor may be required for the existence of the

cell type, its binding to the promoter of a particular gene in that cell

type can be functionally irrelevant for that gene. It is thus challenging

to predict which binding sites are actually functional, or what mode of

actions they mediate (activation vs. inhibition), solely based on the

transcription factor occupancy pattern. In addition, it is not simple to

assign a binding site to potential target genes, especially if the binding

site is surrounded with multiple genes or when the binding site is

distant from a promoter. Developmentally important enhancers of

key genes in the T-cell gene regulatory network can be hundreds of

kilobases (kb) away from the promoters (e.g.

Bcl11b

Ets1, Gata3, Id2;

also

Myc

)(

–

). Disrupting individual regulatory elements by

deletion or a mutation of a speci

fi

c motif sequences can reveal the

enhancer-target gene link, but functional redundancy of regulatory

elements can lead to underestimation (

–

For identi

fi

cation of the gene network linkages described here,

therefore, we have required evidence from functional tests that acutely

perturbed the levels of transcription factor proteins in a speci

fi

developmental context. This has been achieved by germline/

conditional deletion using a Cre-

loxP

or CRISPR-Cas9 system,

knocking down using RNA interference (RNAi), repression using

CRISPR interference (CRISPRi),

or acute overexpression/ectopic

activation of the target gene, each of which was then followed by

vivo

and/or

in vitro

phenotype scoring at the cellular, transcriptomic,

and epigenomic levels. Furthermore, we have given greater weight to

results from experiments where the role of a transcription factor was

tested in a precise, relevant T-cell developmental context and time

window. This was important because recent results of stage-speci

fi

perturbation tests have shown that the same transcription factor

’

function may change or disappear entirely in a different context, or

even at a different stage within the same lineage (

–

). We

have sought to apply consistent statistical criteria to these expression

differences and to emphasize relationships that have been

independently con

fi

rmed.

Precise controlling of perturbation timing can be experimentally

challenging, especially in the context of a developmental process, as

the cells are constantly progres

sing forward to the next stage.

Inadequately de

fi

ned perturbation time-windows could lead cells to

developmental deviation toward irrelevant fates prior to reaching the

stages of interest, or allow cells to activate compensatory mechanisms.

Wide perturbation time windows could also span states in which the

same transcription factor plays different roles, which could

complicate data interpretation. Also, although stage-speci

fi

perturbation approaches can capture the functional consequences of

the perturbation effectively, it is dif

fi

cult to dissect whether the

phenotype is resulting from direct effect, indirect effect

via

gene

network, or indirect effect

via

differential population survival. As

noted above, simply detecting transcription factor binding in the

vicinity of a possible target gene is not enough to prove a direct

functional relationship.

These limitations and challenges are not completely avoided

within the data sources that we have analyzed to construct the

update of the early T-development GRN. However, support for

speci

fi

c sites through which a transcription factor could exert direct

control of a target gene can be gleaned from the recent increase in

available genome-wide transcription factor binding data together with

measurements of local chromatin states such as chromatin

accessibility, 3D chromatin structure, and changes in histone

marks, when these data are coupled with analyses of transcription

factor perturbation effects.

2.3 Construction of an updated T-cell

speci

cation GRN model

The current gene network model we present differs from previous

versions (

–

) in several ways. In particular, initial models for

early T-cell developmental GRNs were based primarily on candidate

gene measurements, due to technical considerations. Targeted assay

systems such as qPCR were used to examine perturbation effects on

sets of only 100-150 genes out of 10,000 expressed transcripts, focused

mainly on high sensitivity monitoring of transcription factor coding

genes. It is now routine to use RNA-seq to measure the entire

transcriptome quantitativel

y in an unbiased manner with low

numbers of input cells, both at the bulk population level and at the

single-cell level. This reveals whole batteries of genes coregulated by a

given transcription factor perturbation in the speci

fi

c context, which

help to identify the changes in developmental status that have been

induced. Genome-wide transcriptome data processing pipelines also

standardize accepted statistical criteria. Where available, single-cell

transcriptome analyses are also useful to separate perturbation effects

on cells within a lineage from perturbation effects on population

balances between the lineage of interest and contaminants.

Another change has been the advent of better technology for acute

loss of function as well as acute gain of function of transcription factor

genes within a well-de

fi

ned developmental time window. Whereas

Cre excision required a separate mouse strain to be developed for each

gene to be targeted, using Cas9-transgenic mice (

) as cell sources for

in vitro

T-cell differentiation has made it possible to use guide RNAs

(gRNAs) targeting one or several genes anywhere in the genome to be

introduced, to disrupt genes ef

fi

ciently at any stage desired in the

same genetic background. Previous analyses of transcription factor-

target linkages in early T cell development have often depended on

gain of function or ectopic expression experiments because these

techniques change transcription factor levels acutely in a speci

fi

c cell

type with even faster kinetics. However, multiple recent examples

Shin and Rothenberg

10.3389/

mmu.2023.1108368

Frontiers in

Immunology

frontiersin.org

have shown that transcription factor impacts on a network can differ

markedly depending on the level of expression of the transcription

factor protein (GATA3, PU.1, Runx), and levels may be hard to

control in gain of function. While loss-of-function experiments can

pose other problems (asynchronous, sometimes slow loss of targeted

protein; potential viability losses in the affected cells; potential

masking by redundancy), they measure effects of factors at their

normal levels of expression. Thus, gain of function data have needed

to be re-evaluated in light of corresponding loss of function data, and

the manipulated levels of factor protein have needed to be compared

to physiological levels.

To construct the network models shown below, we have compiled

data from several sources where developmentally well-de

fi

ned gain or

loss of function perturbations were carried out, with all signi

fi

cantly

responding genes from each perturbation study tabulated in

Table S1

The studies used are described in

Box 1

Table S1

also compiles

evidence of local binding by the transcription factor of interest around

each functional target gene, wherever this evidence was available.

Note that different perturbation experiments used as sources focused

on different developmental time windows and were more or less

sensitive to loss or gain of a given factor

’

s activity, depending on the

normal expression baseline at that timepoint. Where there was

variation between different controls or inadequate expression of a

gene within the time window tested, even repeatedly observed effects

may have missed statistical signi

fi

cance cutoffs. Therefore, while we

have generally depended on more than one corroborating piece of

evidence for each connection shown, we have not required that all

RNA-seq sources should score the same genes as

“

signi

fi

cantly

”

affected. Taken together, however, these results now provide a

clearer view of the architecture of the T-cell speci

fi

cation gene

regulatory network, showing its modular construction, and the

coordination of changes in activities of its component subcircuits

from stage to stage.

3 Overview of early thymic

T-cell development

Thymic T-development begins as hematopoietic progenitor cells

possessing lymphoid potentials migrate into the thymus and interact

with the cortical epithelial cells providing Notch ligands, growth

factors, and cytokine signaling (

106

–

109

). This drives the

developmental program shown in

Figure 1A

. At early stages,

intrathymic precursor cells undergo extensive cell proliferation and

upregulate some of the T-lineage associated genes. However, they still

preserve multipotentiality and can differentiate into non-T lineage

cells, especially if Notch signaling is withdrawn. This uncommitted

stage is referred to as

“

Phase 1

”

, which includes double-negative 1

(DN1, or Early T-cell precursor, ETP) and DN2a stages. In Phase 1,

both ETP and DN2a cells express high levels of cKit and CD44, but

CD25 surface expression distinguishes ETP (CD25

) from DN2a

(CD25

)(

Figure 1A

). Recent single-cell transcriptomics studies

reveal heterogeneity in Phase 1 population both in human and

mouse. The pro-T cells comprising Phase 1 actually include

multiple subsets of ETPs and populations transitioning to DN2a

cells that display distinct gene regulatory programs, with patterns

mostly conserved between mouse and human based on single-cell and

bulk analyses (

–

110

). Of interest, the intermediate-stage ETP

populations, more than the most primitive ETPs, transiently express a

set of non-T lineage-associated genes (e.g.

Mpo

,encoding

myeloperoxidase) even though these cells are on the T-cell pathway

(

). This suggests that these genes are induced as a part of a normal

developmental progression program in ETPs, and multilineage

priming occurs before T-lineage commitment.

Upon sustained exposure to Notch-ligand and other thymic

microenvironmental signals, progenitor cells intrinsically commit to

a T cell fate, and the developmental plasticity to non-T-lineages is

terminally blocked. After T-lineage commitment, pro-T cells establish

a T-cell identity gene expression program and start to rearrange some

forms of T cell receptor (TCR) genes. For conventional T cells,

successful gene rearrangement for expression of TCR

chain is

assured by quality control at the

-selection checkpoint during

DN3 stage. Other T cell precursors rearrange and express genes

encoding TCR

and TCR

instead, to become

T cells. These stages

from commitment to

-selection are collectively referred to as

“

Phase

”

, which is comprised of DN2b (cKit

int

CD25

cells) and DN3a

(cKit

low

CD25

CD28

CD27

low

)stages(

Figure 1A

)(

111

–

113

Further development depends on the cells

’

TCR interactions.

Following

-selection, most developing T cells accumulate as

CD4

CD8

(

“

”

) cells, while they complete their TCR

gene

Box 1: Sources of data compiled in network models

Table S1

presents the main data and sources used to establish speci

fi

c connections in the GRN models that follow. These sources constitute RNA-seq and microarray results

from studies of germline deletion of genes encoding the high mobility group factor TCF1 (encoded by

Tcf7

)(

), the basic helix-loop-helix E proteins E2A (

Tcf3

) and

HEB (

Tcf12

)(

), and the later-activated zinc

fi

nger factor associated with T-cell lineage commitment, Bcl11b (

). In addition, we used studies of acute, stage-speci

fi

deletions of the ETS family subgroup member PU.1 (encoded by

Spi1

, previously called

Sfpi1

in mice) (

); of Lmo2 (

); of GATA3 (

); of Bcl11a (

); of Erg (

);

of Notch1 and Notch2 together (

); and of Runx1 and Runx3 together (

). Although data for cells in the same developmental stages with complete disruption of Ikaros

(

Ikzf1

) were not available, differentially expressed genes that responded to Ikaros (

Ikzf1

) zinc

fi

nger 4 deletion were also added (

). Finally, we included data from studies of

acute gain of function of factors at stages after they would normally have been shut down, including PU.1 (

Spi1

)(

), and the transcription factor adaptor Lmo2 (

–

). In

addition, supporting results came from studies introducing into pro-T cells acute antagonists of key transcription factors, including the natural

E protein antagonist ID2 (

)or

an arti

fi

cially constructed dominant repressor form of PU.1 (

). Additional supporting results came from earlier perturbation studies knocking out E protein genes

Tcf3

(E2A)

and

Tcf12

(HEB) or

Bcl11b

(

) and studies utilizing progenitor or pro-T cell lines and acute T-cell malignancies to interrogate roles of the early-acting transcription factors

Lmo2 and Hhex (

). For data on normal developmental expression dynamics of these genes in pro-T cells, RNA-seq and single-cell RNA-seq datasets were used (

–

), corroborated by highly curated microarray data (

). Data used were all from experiments in the mouse system, but the underlying gene expression patterns involved are

largely conserved in human data (

–

103

)(rev. in (

103

)). In addition to the data incorporated into

Table S1

, we consulted data from other studies for TCF1, GATA3, and

Bcl11b target gene regulation as well (

104

105

). Finally, note that positive and negative regulatory connections shown in the models indicate a measurable effect in

the indicated developmental window, but usually not a pure Boolean function.

Shin and Rothenberg

10.3389/

mmu.2023.1108368

Frontiers in

Immunology

frontiersin.org

rearrangements and express for the

fi

rst time the TCR

that they

will use forever, if they are allowed to live. However, they undergo an

ultra-stringent selection process to reject all cells with inadequate or

highly autoreactive TCR speci

fi

city. The rare surviving cells can

fi

nally mature, undergoing divergent programs of positive selection

into CD4 or CD8 single-positive cells before they emerge from the

thymus, associated with

“

helper

”

“

cytotoxic

”

function respectively.

Notably however, most core T-identity program genes that pro-T

cells activate in Phase 2 are irreversibly maintained throughout all

later stages of T cell development and immunological responses.

4 Gene regulatory network models

for early T cell development

through commitment

As the thymus-seeding precursor cells migrate from the bone

marrow and

fi

rst enter the thymus, they express transcription factors

inherited from their progenitor cells in the bone marrow. Expression

of many such legacy transcriptio

n factors is maintained across

multiple cell divisions. While many progenitor-associated factors

FIGURE 1

T cell development stages and transcription factor expression kinetics.

(A)

Diagram depicts different stages of early thymic T cell development that T-

progenitor cells sequentially go through. Informative surface proteins that are utilized to de

fi

ne each stage are indicated (cKit, CD44, CD25 for ETP to

DN3; CD4 and CD8 for DP). Developmental plasticity to generate alternative, non-T-lineage cells is shown with dotted arrows. Note that these

alternative lineage potentials are silenced after T-lineage commitment. Lineage commitment to a T cell fate distinguishes Phase 1 (before T-lineag

commitment) and Phase 2 (after T-lineage commitment). CLP, Common lymphoid progenitor, LMPP, lympho-myeloid primed progenitor.

(B)

Graphs

show mRNA expression kinetics of important transcription factors involved in early T cell gene regulation programs. Left: Transcription factors in

herited

from the bone-marrow progenitor cells whose expressions gradually decline during T-development (

Spi1, Bcl11a, Erg, Lyl1

, and

Lmo2

). Middle:

transcription factors upregulated in pro-T cells by thymic microenvironment (

Tcf7, Gata3

, and

Bcl11b

). Right: transcription factors expressed from the

bone-marrow progenitor cells and stably and maintained during T cell development (

Ikzf1, Tcf3, Tcf12, Runx1

, and

Runx3

). Gene expression data was

plotted by utilizing publicly available mRNA expression datasets for immune cells with curve smoothing (

https://www.immgen.org

)(

(C)

Diagram

illustrates transcription factors (arrows) providing distinct forces to different gene expression program modules in individual cells.

Shin and Rothenberg

10.3389/

mmu.2023.1108368

Frontiers in

Immunology

frontiersin.org

are turned off eventually, some legacy transcription factors maintain

their expression throughout thymic T cell development, but these

often adopt new roles during stage transitions by occupying different

sets of genomic regions from those in hematopoietic progenitor cells.

The co-existence of

“

inherited transcription factors

”

together with

“

newly induced regulators

”

in the thymic precursor cells generates

numerous possible combinatorial inputs to different target genes

which dynamically change their expression during developmental

progression in Phase 1 and Phase 2 (

Figure 1B

). This review focuses

on key driving regulatory factors and responders composing the

Phase 1- and Phase 2-GRNs, in which many components show

dynamic changes throughout developmental progression.

4.1 Gene regulatory network modules in

Phase 1 and Phase 2

The newly proposed GRN that we present here highlights the

observation that in each stage of early T-lineage development, subsets

of target genes for a given transcription factor are often regulated in

parallel ways, collectively composing a speci

fi

c module that often has

coherent biological function (

114

). While the de

fi

nition of a

module is not precise, we use this framework to describe fairly

discrete components of the dev

elopmental process which are

regulated distinctly even if the cell is expressing other sets of genes

at the same time. One module (de

fi

ned by expression of a group of

genes) may remain active consistently throughout a series of stages

while other modules are sharply changing activity, or a module can be

affected coherently by a perturbation that does not affect genes in

other modules within the same cells. As shown below, this gives the

overall process of early T cell development an

“

assembled

”

quality.

Key transcription factors often appear to play roles in activity of an

entire module, even while the individual target genes they act upon

also have other regulatory inputs. However, it is noteworthy that a

transcription factor does not de

fi

ne the module on its own. The same

transcription factor may even work in more than one module, as

shown below. Instead, different transcription factor ensembles can be

seen to de

fi

ne distinct modules, working together to drive expression

of shared target genes. They establish positive and negative feedback

loops within a module, for stabilization, or between different modules,

which results in dynamic gene network behavior. This modular

structure is signi

fi

cant because the timing of transitions that

individual cells make along the pathway is very asynchronous, with

cells showing an ability to linger in any of several states for several cell

cycles before progressing (

).Theslowpaceofdifferentiation

implies that regulation of the modules that predominate in any

particular stage is fairly stable; however, developmental progression

depends on eventually breaking this stability. In developmental gene

networks, a regulatory state is often stabilized when different

transcription factors with concordant effects on the same module

also support each other

’

s expression (

115

), and some examples of this

are shown below. In contrast, inter-module inhibition or repression

between regulators can make expression of different modules unstable

and eventually incompatible.

Mutually exclusive expression of different regulators generates

sharp cell-fate boundaries in bin

ary cell fate decisions and in

embryos, for example (

115

116

). However, one notable feature

of the T cell developmental networ

k linkages is that the repression

which is detected is often incomplete. Many repressive interactions

in this system cause a dampenin

g of expression levels but not a

silencing of the target genes. This

“

soft

”

repression function

enables factors with mutually anta

gonistic activities to coexist in

the cells for days and multiple cell divisions, and it enables

activities of opposing modules to overlap within an individual

cell at the same time. Therefore, distinct sets of transcription

factors can pull and push activities of their target modules in

different directions in the same cell (

Figure 1C

), and the resultant

sum of the vectorized

“

forces

”

instructs cell fate (

). In the

following, we review both the regulatory circuits that maintain

coherent module activity, and also

the sources of antagonists that

fi

nally keep them from persisting.

The architecture of the modular subprograms within the T-cell

GRN has become much clearer through recent work. The distinct

subprograms that comprise the early T-cell GRNs can be categorized

as 1) the T-identity module 2), the stem or progenitor module, 3)

conditional access to alternative fates as a side effect of the stem/

progenitor module, and 4) a cell survival and proliferation module.

While the T-identity module needs to be successfully installed to

provide constitutive expression of the core T-cell genes, the second

module must be silenced in order to convert multipotent precursors

to T-lineage committed cells. Finally, the cell survival and

proliferation program in Phase 1 and Phase 2 ensures the

production of suf

fi

cient number of immature T cells to

accommodate later positive and negative selection. Because the T-

cell identity module introduces the initiators and overall structure of

the entire process, it is discussed

fi

rst.

4.2 Installation and speci

cation of the T-

identity module in Phase 1 and Phase 2

Commitment to the T-cell fate involves two major events: 1)

terminal blockade of alternative lineages, concomitant with 2)

acquisition of T-identity gene expression. The gene network

instructing the constitutive expression of the T-lineage de

fi

ning

geneswillbereferredtoastheT-identitymodule.Successful

establishment of the T-identity module is pivotal because the core

T-identity established in early

thymic developmental phases is

robustly maintained even after these progenitor cells become

mature T cells, long after they leave the Notch ligand-rich thymic

environment. The

“

T-cell markers

”

include CD3 clusters and TCR

signaling mediators, as well as transcription factors necessary for the

induction and maintenance of these marker genes. The T-identity

module requires activities of Notch signaling, TCF1, GATA3, Runx

transcription factors, and Ikaros starting from Phase 1, then receives

critical positive inputs from the E protein complex and Bcl11b as pro-

T cells progress through Phase 2. Expression patterns of the key

factors in different immune cell populations are shown in

Figure 2A

(data from

www.immgen.org

(

)).

4.2.1 Notch signaling as an indispensable driver

Notch signaling is an absolute requirement for initiation and

establishment of the T-identity module. In the absence of Notch

signaling, B cells develop in the thymus instead of T cells, whereas

Shin and Rothenberg

10.3389/

mmu.2023.1108368

Frontiers in

Immunology

frontiersin.org

constitutive expression of intracellular domain of the Notch (ICN) in

the bone-marrow progenitor ce

lls induce extrathymic T cell

development (

118

–

120

). In addition, many if not all thymic seeding

progenitors need to be primed by some level of Notch signaling in the

bone marrow. Whereas Jagged-class Notch ligands expressed in the

bone marrow do not signal as strongly as the Delta-class Notch

ligands in the thymic microenvironment (

121

–

123

), this prior

experience seems to be important for the cells to acquire

competence to initiate the T cell program (

124

) A thymic seeding

progenitor population that had received Notch signaling before

thymic entry also exists in human

s, supporting a physiological

contribution of Notch signaling to T-lineage speci

fi

cation from the

prethymic stages in both humans and mice (

). The stromal cells in

the thymic cortex express the strongest ligands for signaling through

Notch1, mainly Delta-like 4 (Dll4), as well as Dll1 and to a lesser

extent Jagged 2 (Jag2). These engage with Notch receptors (Notch 1-

Notch3) on the thymic precursor cells with varied af

fi

nities (

122

125

This ligand-receptor interaction induces proteolytic cleavages in the

Notch receptor, releasing the ICN to the cytoplasm. Then, the Notch

ICN interacts with DNA-binding transcription factor RBPJ

and

FIGURE 2

Gene regulatory networks driving the T-identity module.

(A)

Heatmap shows expression patterns of key transcription factors supporting the T-identity

module across different immune cell types, using the ImmGen MyGeneSet tool (

https://www.immgen.org

)(

). Bone marrow progenitor (BM progen), B

cells precursors (B progen), B cells, T cells,

T cells ("gd T cells"), natural killer cells (NK), Innate lymphoid cells (ILC)s, Dendritic cells (DC), Macrophage

(MF), Monocyte (Mo), Granulocyte (GN), Mast cell (MC). See

Table S2

for the number keys.

(B, C)

BioTapestry models (

117

) of gene regulatory network

relationships described in the text. Evidence for all connections shown is in

Table S1

. Structure of BioTapestry models: Genes are shown with regulatory

regions (horizontal lines) distinct from their encoded outputs (bent arrow at

“

promoter

”

). Connections with arrows represent positive regulation.

Connections ending in blocking lines represent negative regulation. When gene products combine to produce a functional unit, or when activity is not

simple function of transcriptional output, a

“

bubble

”

symbol is placed to represent the emergent function from their collective activities. Both negative

and positive regulation can impinge on such activity bubbles, e.g. ID factors inhibiting E protein activity without necessarily inhibiting the expr

ession of

the E protein coding genes themselves. Double chevron symbol indicates ligand-receptor interactions, here used to represent Notch ligand

—

Notch

interaction as a source of regulatory input to other genes. Arrows consisting of dotted lines show decreasing or weak activities. These conventions a

used also in

Figures 4B

(B)

Current model for the T-identity module regulator transcription factors in Phase 1 (top) and Phase 2 (bottom).

(C)

Current model for the T-identity module downstream molecules in Phase 1 (top) and Phase 2 (bottom). Expression of

Il2ra

(CD25) indicates

developmental progression from ETP to DN2 and DN3 stages in mice. Surface markers:

Thy1

Cd3

clusters genes (

Cd3d, Cd3e, Cd3g

), Cd247; TCR

signaling molecules:

Lck, Itk, Fyn, Fyn, Trat1, Lat, Zap70, Ptcra

; TCR rearrangement molecules:

Rag1, Rag2, Dntt

Shin and Rothenberg

10.3389/

mmu.2023.1108368

Frontiers in

Immunology

frontiersin.org

functions as co-activator, recruiting other transcriptional coactivators

and chromatin modifying enzymes (

122

126

Although Notch1 is most strongly expressed throughout, acute

deletion of different Notch family genes in pro-T cells shows that

Notch1 and Notch2 cooperate to induce the T-identity module in

Phase 1 cells by activating genes encoding essential transcription

factors (e.g.,

Tcf7

Myb

Gata3

Hes1

), the core-T cell marker genes

(

Cd3g

Cd3e

Thy1

) and the useful DN2-DN3 stage marker

Il2ra

(in

mouse; not in human at this stage). Later, in Phase 2, Notch signaling

induces genes necessary for TCR rearrangement and signaling (

Rag1

Rag2

Lck

Ptcra

)

(

)(

Figures 2B

). Notch actions in the T-cell

identity module are not hit-and-run; the signals must be sustained

through lineage commitment and then to sustain most cells

’

viability

into the beginning of

-selection [rev. by (

127

)]. Notably, however,

the target genes regulated by Notch signals either positively or

negatively change markedly between Phase 1, early Phase 2, and the

end of Phase 2 (

). The shifting but essential roles of Notch signaling

in pro-T cells provide an example of context-dependent shifts in

regulator deployment which are seen for other factors as well (

4.2.2 Notch-induced effectors TCF1 and GATA3,

cooperative but nonredundant

Notch-activated targets in Phase 1 cells include genes encoding

TCF1 (

Tcf7

) and GATA3, which are pivotal for instituting the T-

identity module (

Figures 2A

top). The activation of

Tcf7

by Notch

signaling appears to be direct, although its maintenance becomes

Notch-independent in Phase 2 (

). The functional importance of

TCF1 and GATA3 in T-lineage speci

fi

cation is demonstrated by

transgenic animal models that lack

Tcf7

Gata3

expression.

Tcf7-

Gata3

-de

fi

ciency in pro-T cells abrogates T-cell development from

the earliest Phase 1 stage (

104

128

–

130

Tcf7

deletion in

bone-marrow derived progenitors using

Vav1-

Cre caused

developmental arrest at ETP stage and allowed abnormal

transcriptome clusters to accumulate among thymocytes in steady

state, based on single-cell RNA-seq (scRNA-seq) (

). In accord with

these results, another recent single-cell transcriptome study using

dual guide-RNAs (gRNAs) to disrupt

Tcf7

Gata3

speci

fi

cally in

Phase 1 cells showed that the precursor cells lacking TCF1 completely

failed to enter the normal T-cell developmental trajectory, while those

lacking GATA3 failed to progress properly (

). While

Tcf7

and

Gata3

both depend on inputs from Notch signaling and Runx family

transcription factors to be turned on, they also create a possible

stabilization circuit f

or early T-cell speci

fi

cation by positively

regulating each other as well (

)(

Figure 2B

Despite these positive feedbacks, acute

Tcf7

deletion results in

different impacts on the developing pro-T cell population than acute

Gata3

deletion in the same Phase 1 developmental time window,

based on scRNA-seq data. TCF1 and GATA3 regulate distinct target

genes, indicating that these factors do not perform redundant

functions. Also, studies using gain-of-function approaches suggest

that TCF1 overexpression is completely different from the effect of

high dosage GATA3 in pro-T cells. A high level of TCF1 in the mouse

bone-marrow progenitor cells can upregulate essential genes in the T-

identity module, such as

Gata3

and

Bcl11b

, even without Notch

signaling, causing cells apparently to bypass the ETP stage (

). In

contrast, elevated GATA3 levels block T cell development, promoting

deviation to an alternative, mast-cell lineage in the absence of Notch

signaling, and killing pro-T cells if Notch signaling is sustained (

131

). This difference in part involves the different impacts

these factors have on genes within proliferation and survival modules

used in Phase 1 cells, including effects on

Kit

Il7r

and

Spi1

(see below).

In pro-T cells, TCF1 and GATA3 instruct T-cell development by

upregulating many T-program genes (

Notch1, Notch3, Hes1

Gata3

Bcl11b

Lef1, Ets2

Il2ra

, all

Cd3

genes,

Cd247

Tcrb

Lat, Fyn, Rag1,

Rag2

, and

Trat1

by TCF1;

Myb, Ets1, Tcf7

, and

Bcl11b

by GATA3)

(

Figure 2C

). Importantly, both TCF1 and GATA3 are required

together for the initial induction of Bcl11b, which will be important

in Phase 2, suggesting that multiple transcription factor inputs are

non-redundantly required (

As TCF1 and GATA3 are necessary to initiate the T-lineage

speci

fi

cation program, the positive regulatory factors inducing these

transcription factors are also critical. TCF1 and GATA3 positively

regulate each other and Runx transcription factors also provide

supportive inputs, as described b

elow. The critical regulatory

elements of the

Tcf7

gene in T-lineage cells, 30-40kb upstream of

its promoter regions (

132

), are also occupied by RBPJ

, Runx factors,

GATA3, and TCF1, consistent with direct positive regulation by these

factors (

Figure 3

left, in "DN2b/DN3" samples; note that both

Tcf7

and

Gata3

in this

fi

gure are transcribed from right to left). Similarly,

an enhancer region 280 kb downstream of the

Gata3

gene, which is

known to be necessary for

Gata3

expression in the T-lineage (

), is

occupied by RBPJ

, Runx factors, TCF1, Bcl11b, and GATA3 by

Phase 2 (

Figure 3

right,

“

[i.e., "DN2b/DN3"] samples).

The strong force that TCF1 exerts to drive the T-cell program

involves reprogramming of chromatin accessibility and long-range

looping. TCF1 overexpression in

fi

broblasts opens the chromatin

regions near the T lineage-associated genes, which are naturally

demarcated by repressive histone marks in

fi

broblasts (

135

). Recent

studies demonstrate a potent role of TCF1 in chromatin architecture

remodeling in pro-T cells, later DP thymocytes, and mature T cells

(

136

137

). In Phase 2, TCF1 occupies key sites in evolutionary

conserved topologically associating domains (TAD), i.e. regions of

chromatin containing clusters of regulatory elements that interact

within the TAD but are usually insulated from other TADs. When

TCF1 binds to the inter-TAD sites along with CTCF (a transcription

factor that can anchor chromatin architecture), this co-occupancy

weakens insulation of the TAD boundaries and enables intermingling

of TADs, potentially allowing new enhancer-promoter interactions.

In addition, TCF1 establishes long-range looping around the T cell

genes and marks the surrounding regions with H3K27ac (

136

4.2.3 Multitasking positive roles of Runx

family factors

Runx transcription factors are broadly expressed in all

hematopoietic lineage cells, but they exert context-speci

fi

c functions

by switching their interaction sites across the genome. The expression

of Runx1 and Runx3 is established prior to the thymic entry, and

these two paralogs are co-expressed within individual Phase 1 and

Phase 2 pro-T cells alike (

). The sum of Runx1 and Runx3

activities, measured by total binding by ChIP-seq, is maintained

stably throughout the early T-li

neage developmental process,

suggesting that overall Runx availability between Phase 1 and Phase

Shin and Rothenberg

10.3389/

mmu.2023.1108368

Frontiers in

Immunology

frontiersin.org