Runx factors launch T cell and innate lymphoid programs via direct and gene network-based mechanisms

Runx factors launch T cell and innate lymphoid programs via

direct and gene network-based mechanisms

Boyoung Shin

Wen Zhou

1,2,3

Jue Wang

1,2

Fan Gao

1,4,5

Ellen V. Rothenberg

1,*

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

91125, USA

Program in Biochemistry and Molecular Biophysics, California Institute of Technology

Current address: BillionToOne, Menlo Park, CA 94025, USA

Bioinformatics Resource Center, Beckman Institute of California Institute of Technology

Current address: Lyterian Therapeutics, South San Francisco, CA 94080, USA

Abstract

Runx factors are essential for lineage specification of various hematopoietic cells, including T

lymphocytes. However, they regulate context-specific genes and occupy distinct genomic regions

in different cell types. Here, we show that dynamic Runx binding shifts in murine early T cell

development are mostly not restricted by local chromatin state but regulated by Runx dosage

and functional partners. Runx co-factors compete to recruit a limited pool of Runx factors

in early T-progenitors, and a modest increase in Runx protein availability at pre-commitment

stages causes premature Runx occupancy at post-commitment binding sites. This increased Runx

factor availability results in striking T-lineage developmental acceleration by selectively activating

T-identity and innate lymphoid cell programs. These programs are collectively regulated by Runx

together with other, Runx-induced transcription factors that co-occupy Runx target genes and

propagate gene network changes.

Introduction

Runx family transcription factors (Runx1, Runx2, Runx3, and their cofactor CBF

) are

important for T cell development from the earliest steps in the lineage, playing partially

redundant roles

–

. However, they are also vital for the establishment of hematopoietic

Corresponding author. evroth@its.caltech.edu.

Author contributions Statement

B.S. and E.V.R. conceptualized the project, wrote the paper, and edited the paper. B.S. performed the experiments, and analyzed

data with W.Z. and J.W. F.G. wrote the in-house bioinformatic pipeline for hashtag alignment and provided further analysis. E.V.R.

supervised research, acquired funding, and provided additional data analysis. All authors edited the paper and provided helpful

comments.

Competing Interests Statement

WZ is employed by BillionToOne, Inc. and has been employed by 10X Genomics (CA 94588). FG is employed by Lyterian

Therapeutics. EVR was a member of the Scientific Advisory Board for Century Therapeutics and has advised Kite Pharma and A2

Biotherapeutics. The other authors declare no competing interests.

CODE AVAILABILITY

All code used for data analysis in this work is publicly available and listed in the Methods and Reporting Summary.

HHS Public Access

Author manuscript

Nat Immunol

. Author manuscript; available in PMC 2024 March 01.

Published in final edited form as:

Nat Immunol

. 2023 September ; 24(9): 1458–1472. doi:10.1038/s41590-023-01585-z.

Author Manuscript

stem cells in early embryos

and for generation of B cells and megakaryocytes throughout

life

–

, quite different programs. Runx target motifs are consistently highly enriched

around open chromatin sites and lineage-specific transcription factor (TF) binding sites

in multiple hematopoietic lineages

–

, suggesting a common contribution to active

enhancers generally. However, Runx factors regulate different target genes to mediate

distinct stage-specific functions, by switching the genomic sites they occupy at different

stages of early T cell development

. Thus, key questions are how Runx factors guide

their contributions to distinct cell programs, whether by intrinsic DNA-binding sequence

specificity, epigenetic constraints, or interactions with other partner factors.

Notch signaling and other microenvironmental cues within the thymus convert multipotent

progenitor cells to T cells. They traverse a series of CD4

−

CD8

−

double negative (DN) pro-T

stages (DN1-4), then CD4

CD8

double-positive (DP) stage, before becoming mature CD4

or CD8 single-positive (SP) T cells. These stages are distinguished by changes in chromatin

states and changes in expression of a discrete set of regulatory factors

–

. Pro-T cells

in the c-Kit

DN1 (or “ETP”) and DN2a stages (“Phase 1”) still resemble hematopoietic

stem and progenitor cells (HSPC) in regulatory gene expression and chromatin state and

can still produce non-T lineage cells. T-lineage commitment normally occurs in transition

from DN2a to DN2b stages with up-regulation of T cell identity genes (“Phase 2”), and this

post-commitment stage extends till successful assembly of T cell receptor

(TCR

) in DN3

stage. Runx factors are crucial for progression through both Phases

. Although Runx1 and

Runx3 act differently in other contexts

, they appear functionally redundant in pro-T cells

However, they both bind to different genomic sites and regulate different target genes from

Phase 1 to Phase 2.

Here, we evaluated the chromatin constraints on Runx action across the Phase 1 and

Phase 2 stages of T cell development and tested the hypothesis that Runx binding site

shifts depend on competition between Phase 1 and Phase 2 partners for a limiting amount

of Runx protein. We found that at modest excess, when no longer titrated by Phase 1

partners, Runx factors directed a distinctive accelerated form of early T and innate-like

cell development, relieving the need to repress most Phase 1 regulators before T-lineage

regulatory genes could be upregulated. Both direct (binding site-mediated) and indirect

mechanisms propagated through a Runx-dependent gene regulatory network drove this

acceleration. Thus, availability of Runx factors for their target sites at each stage is a major

timing controller of the deployment of the T cell specification gene regulatory network.

Results

Runx TFs repeatedly shift binding sites during T development

Stages in T cell development can be distinguished by changes in cell-surface markers

plus

Bcl11b

expression, which marks lineage commitment

(Fig. 1a). Runx1 and Runx3

change expression only slightly from DN1 stage to T-lineage commitment, while Runx2

levels are substantially lower, and the total numbers of Runx1 and/or Runx3 genomic

occupancy sites remain constant

. However, Runx factors interact with different genomic

regions in pre-commitment stages than in post-commitment stages

. Cell type-specific

binding is seen especially at non-promoter sites, which are much better correlated with

Shin et al.

Page 2

Nat Immunol

. Author manuscript; available in PMC 2024 March 01.

Author Manuscript

target gene regulation by Runx factors than the promoter sites

. Fig. 1b shows that

in different hematopoietic lineage contexts from HSPCs to mature T cells, B lineage

and megakaryocyte-precursor cells

–

, Runx factors preferentially occupied distinct

genomic regions in each (Fig. 1b, A–F), with only ~10% of sites shared in all (Fig. 1b,

G). This site infidelity of Runx factors contrasted with binding patterns in pro-T cells of

PU.1, a critical Phase 1-specific TF, which showed more similar binding site choices from

HSPCs to DN2b pro-T cells (Extended Data Fig. 1a). Importantly, each cluster of Runx

binding regions from HSPC to mature T cells harbored a distinct set of motifs in addition

to the Runx motif, in which motifs for EBF, PU.1, E2A, TCF1, GATA, and ETS factors

were differentially enriched in each cluster (Extended Data Fig. 1b). The shifts were not an

artifact of indirect binding via protein-protein crosslinking, which could have contributed

to previous data using ChIP-seq for Runx

. Here, all Runx DNA binding profiles in

pro-T cells were newly defined using cross-linking-independent CUT&RUN

(C&R)(see

Supplementary Note 1, Extended Data Figs. 1c–e and 2a,b for detailed comparisons). These

results suggested that Runx factors changed binding sites during T cell development to

interact with distinct regions potentially occupied by different TF partners.

Runx factors do not follow local chromatin accessibility

Runx binding shifts could be caused by chromatin state changes at large or local scales,

or by individual-site interaction with partner factors. To test whether Runx TFs were

constrained or redirected by large-scale chromatin remodeling during commitment, the

non-promoter Runx binding sites were categorized into three groups: Phase 1-preferential

binding sites (Group 1), Phase 2-preferential binding sites (Group 2), and stably occupied

sites (Group 3) (Fig. 1c). We analyzed how Runx binding was correlated with “active”

(A) or “inactive” (B) large-scale nucleome compartments

by comparing the principal

component 1 (PC1) values of the previously reported Hi-C correlation matrices from ETP

(DN1), DN2, and DN3 cells

. Runx binding sites were preferentially enriched within

the A compartment (84-92%) and were depleted from the B compartment (3-7%) in all

site Groups, regardless of whether C&R or ChIP-seq was used to detect Runx binding

(Extended Data Fig. 2b–e; note

Ets1

flanking regions). The great majority were also scored

as locally accessible as defined by assay of transposase accessible chromatin (ATAC), even

the Runx binding sites in the “inactive” B compartment (Extended Data Fig. 2c). As pro-T

cells developed from ETP to DN3 stages, Runx occupancy tended to follow the “active”

compartments (Extended Data Fig. 2d). Among the minority of genomic regions changing

compartment, those switching from active to inactive (A to B trend) included more Group 1

(4.75%) than Group 2 Runx sites (1.32%), whereas compartments becoming active (B to A

trend) included more Group 2 sites (4.56%) than Group 1 (2.35%)(Extended Data Fig. 2d).

Multiple Group 2 sites appeared with a B to A compartment flip in the extended flanking

region of

Bcl11b

(Extended Data Fig. 2e). However, most Runx site shifts occurred within A

compartments.

To examine more local changes around Runx sites, we coded individual chromatin states

across the genome from pre-commitment to post-commitment stages using ChromHMM

with published datasets for chromatin state marks in pro-T cells

(see Methods).

Although Runx factors can work both as repressors and as activators

–

, Runx binding

Shin et al.

Page 3

Nat Immunol

. Author manuscript; available in PMC 2024 March 01.

Author Manuscript

sites overall were preferentially enriched in open/active chromatin regions (states 1-3,

5-10), or in weakly accessible regions harboring H3K4me2 marks representing likely poised

regions (state 13)( Extended Data Fig. 2f).

However, the developmental changes of Runx binding site choices did not strictly depend on

local DNA accessibility. At genomic sites where Runx factors changed occupancy between

Phase 1 and Phase 2 (Fig. 1d, Extended Data Fig. 2e, g), Runx binding often changed

even when ATAC accessibility of those sites did not change. Of Group 1 (Phase 1-specific)

sites, only 43.8% were open in a Phase 1-restricted way, and only 21.4% of the Group 2

(Phase 2-specific) sites were open selectively in Phase 2: many Group 1 and a majority of

Group 2 sites did not change accessibility. Where Group 1 Runx occupancies disappeared

near progenitor regulatory genes

Hhex

and

Meis1

from DN1 to DN3, accessibility was

reduced (Fig. 1e, Extended Data Fig. 2g). However, at

Ets1

multiple sites only gained Runx

occupancies from DN1 to DN3 although these sites were already accessible in DN1 (ETP)

stage (Fig. 1e; zoom-out in Extended Data Fig. 2e). Furthermore, over 1/3 of Group 2

sites remained closed in both Phases (Fig. 1d). Thus, local chromatin accessibility failed to

explain why Runx occupancy was delayed at Group 2 binding sites.

Co-factor motifs distinguish sites with dynamic Runx binding

Site choice shifts could alternatively be due to differences in Runx binding avidity (site

affinity times site density) which could make Group 2 sites highly sensitive to changes in

Runx concentration, or collaborations with different stage-specific partners

. We evaluated

these options by quantitative motif analysis, focusing exclusively on Runx sites that were

consistently “open” to minimize chromatin effects (Fig. 1f, g). Runx binding sites mapping

to open promoter regions had negligible Runx motif frequencies and poorer Runx motif

quality scores (Fig. 1f). Stably open non-promoter Runx binding regions in Groups 1, 2,

and 3 had much higher Runx motif frequencies and motif quality than promoter sites, but

to different degrees. Consistently occupied Group 3 sites had the highest scores. Group 1

and Group 2 sites were similar in Runx motif frequencies and motif scores, although lower

than the Group 3 sites. Thus, at non-promoter sites without chromatin barriers, stage-specific

redistribution of Runx factors occurred most readily between “modest” Runx motif sites

without strong advantages for recruiting Runx factors via DNA recognition

per se

(Fig. 1f).

Consistent with previous results

, we found distinct partner motifs enriched at Runx sites in

different stages (Fig. 1g, Extended Data Fig. 1b).

De novo

motif enrichment analysis of the

open sites confirmed that the Group 1 sites were highly enriched for PU.1 (ETS subfamily)

motifs whereas the Group 2 sites were highly enriched for E2A (basic helix-loop-helix,

bHLH) motifs. Although C&R Runx peaks did not show the extreme enrichment of ETS

motifs seen with ChIP-seq (Extended Data Fig. 1e), canonical (non-PU.1) ETS factor motifs

were still enriched, and were found at similar frequencies in all classes of non-promoter

sites (Fig. 1g). Thus, at sites that were stably accessible throughout Phases 1 and 2, different

ensembles of TFs might recruit Runx factors stage-specifically.

Shin et al.

Page 4

Nat Immunol

. Author manuscript; available in PMC 2024 March 01.

Author Manuscript

Partner proteins shift Runx binding but are titratable

A central question is whether a single mechanism could cause Runx factors to shift

in concert from Group 1 vs. Group 2 sites during commitment. Nearly half the pre-

commitment-specific binding sites for Runx factors (Group 1 sites) are actual co-binding

sites with PU.1

. In later pro-T cells other TFs such as Bcl11b collaborate with Runx

factors at different sites

. Notably, the presence of PU.1 can redirect Runx1 occupancy to

PU.1 sites, while depleting Runx1 binding (“theft”) from alternative Runx sites

. If Runx

factor levels are truly limited such that partners have to compete to recruit Runx to different

sites, then the tipping of a balance between partners might cause concerted occupancy

switches from Group 1 to Group 2 sites.

We hypothesized that if such competition occurs, it could be overridden if Runx availability

were increased. We first tested this hypothesis in a PU.1 “theft” model

(Fig. 2a–b,

Extended Data Fig. 3a–c). The DN3-like Scid.adh.2C2 pro-T cell line, representing a

PU.1-negative Phase 2 state, was retrovirally transduced with exogenous PU.1, with or

without additional exogenous Runx1 (Fig. 2a, Extended Data Fig. 3a–b). As expected

PU.1 activated myeloid markers in the cells with or without exogenous Runx1 (Extended

Data Fig. 3b) and recruited endogenous Runx1 to a set of new co-occupancy sites with PU.1,

most of which had been closed before (Fig. 2b, PU.1-induced). As previously reported

PU.1 also caused a loss of Runx1 occupancy from ~55% of the normal endogenous

Runx binding sites (Fig. 2b, PU.1-depleted). However, when extra Runx1 was added

(OE), although PU.1 still recruited Runx1 to the PU.1-induced sites, occupancy of the

PU.1-depleted sites was fully rescued (Fig. 2b). The extra Runx1 also occupied a set of

novel sites (OE new). These had high quality Runx motifs at high frequency (Extended Data

Fig. 3c), but were mostly sequestered in closed chromatin in the normal Scid.adh.2C2 (Fig.

2b). These results suggest that either Runx1-PU.1 complexes or high-level Runx1 alone

could gain access to normally closed chromatin, and that the ability of PU.1 to remove

Runx1 from its default binding sites was based on competitive titration only when Runx1

was limiting.

Increased Runx1 prematurely induced key T-developmental TFs

If titration of potentially competing partner factors affects Runx site choice, Runx

concentration might affect the T-lineage specification program in early progenitor cells.

To test this, we exploited the OP9-Delta-like ligand 1 (Dll1)

in vitro

differentiation system

as in our previous studies

. Briefly, bone-marrow derived progenitor cells expressing a

Bcl11b

-mCitrine reporter and

Bcl2

transgene were co-cultured with OP9-Dll1 cells, and

exogenous Runx1 was retrovirally delivered to pro-T cells when the progenitor cells were

still at DN1 stage (Fig. 2c). The

Bcl2

-transgene does not have effects on normal T cell

development

in vivo

in vitro

(Supplementary Note 2). Then, we measured T-development

markers (cKit, CD44, and CD25) and

Bcl11b

-mCitrine expression, normally a marker for

T-lineage commitment (Cmt)(see Fig. 1a, 2c)

. At day 2 and day 4 after exogenous Runx1

introduction (overexpression, OE), total Runx1 protein levels were 2-3× increased relative

to controls; however, Runx1 OE cells had a ~1.5× decrease in levels of Runx3, the other

major Runx factor in pro-T cells, as compared to control cells (Fig. 2d, Extended Data

Fig. 4a). Downregulation of Runx3 was most pronounced in the cells expressing more

Shin et al.

Page 5

Nat Immunol

. Author manuscript; available in PMC 2024 March 01.

Author Manuscript

Runx1 (Fig. 2d, left panel). Extra Runx1 dampened mRNA expression of all Runx paralogs:

endogenous

Runx1

transcript (not measuring exogenous

Runx1

Runx2

, and

Runx3

were

all significantly decreased (Extended Data Fig. 4b). Hence, total Runx availability was likely

≤2× increased, and this modest degree of increase was important for the health of the cells

Runx1 OE strikingly accelerated

Bcl11b

induction 2-3 days after transduction: ~20% of

control cells but ~50% of Runx1 OE DN2 cells expressed

Bcl11b

-mCitrine by day 3

(Fig. 2e). Furthermore, increased Runx1 levels caused premature appearance of DN3-like

cells (CD44

low

CD25

)(Fig. 2e, middle). Among other developmentally important TFs,

Runx1 OE upregulated TCF1 and GATA3 protein expression within the cKit

CD25

−

DN1

population at both days 2 and 4 post-infection (Fig. 2f), anticipating increases that occurred

in controls only at day 4 when the cells became CD25

(DN2a cells)(Fig. 2f). Runx1

normally represses PU.1 (encoded by

Spi1

) only in Phase 2

(DN2b, DN3 cells). PU.1

expression was not affected by Runx1 OE at day 2 post-infection, but became significantly

downregulated even in the DN1 population by day 4 (Fig. 2f). All these responses to Runx1

OE were more pronounced in the cells expressing higher Runx1 protein, suggesting Runx

dosage effects (Extended Data Fig. 4c–d). Hence, a mild increase in Runx factor availability

in Phase 1 pro-T cells could accelerate aspects of early T cell development.

Runx perturbations change single-cell transcriptomes

Runx level effects on initiation of the T cell specification program as a whole were analyzed

by single-cell RNA-seq (scRNA-seq). To identify targets affected by Runx1 OE that were

also dependent on normal Runx levels in controls, we compared

Runx1/Runx3

double

knockout (dKO) cells

with Runx1 OE and empty vector control cells. We delivered Runx1

OE vector or empty vector control into

Bcl2

-transgenic progenitor cells to test OE, or

guide-RNAs (gRNAs) against

Runx1

and

Runx3

or control irrelevant gRNAs into

Cas9;Bcl2

transgenic prethymic progenitor cells to test dKO. These samples were each co-cultured

with OP9-Dll1 cells to two different timepoints (day 2 and day 4 post-infection for OE; day

3 and day 6 post-infection for dKO), then marked by unique hashtag oligos, and then all

three OE, dKO, and control conditions were pooled and subjected to scRNA-seq together in

the same reaction (Fig. 3a, see Methods).

In a low dimensional transcriptome representation from two independent 10X runs, the first

dimensions in t-distributed stochastic neighbor embedding (tSNE1) and Uniform Manifold

Approximation and Projection (UMAP1) reflected cell-cycle phases (Extended Data Fig.

5a, b, top panels) and poorly separated developmental states; thus we utilized UMAP2

and UMAP3 for data exploration. After cell-cycle regression, UMAP2 (x axes, Fig. 3b–d)

approximately represented real time developmental progression for normal, control pro-T

cells, while UMAP3 (y axes) reflected perturbation; note that within each population the

cells progressed asynchronously. In controls, cells with low-UMAP2 values expressed DN1

signature genes (

Lmo2, Spi1, Bcl11a, Cd34, Mef2c, Hhex

). Genes transiently expressed

during DN1 to DN2a transition (

Mycn, Fgf3

) were maximally expressed at UMAP2-

intermediate values (Fig. 3b); while DN2a/b marker genes (

Il2ra, Tcrg-C1, Gata3, Tcf7,

Thy1, Cd3g

) were expressed in UMAP2-intermediate and high control cells. At later

timepoints, genes associated with T-lineage commitment and the DN2b stage (

Bcl11b, Ly6d,

Shin et al.

Page 6

Nat Immunol

. Author manuscript; available in PMC 2024 March 01.

Author Manuscript