Transcription factor expression dynamics of early T-lymphocyte specification and commitment

Summary

Mammalian T lymphocytes are a prototype for development from adult pluripotent stem cells. While T-cell specification is driven by Notch signaling, T-lineage commitment is only finalized after prolonged Notch activation. However, no T-lineage specific regulatory factor has been reported that mediates commitment. We used a gene-discovery approach to identify additional candidate T-lineage transcription factors and characterized expression of >100 regulatory genes in early T-cell precursors using realtime RT-PCR. These regulatory genes were also monitored in multilineage precursors as they entered T-cell or non-T-cell pathways in vitro; in non-T cells ex vivo; and in later T-cell developmental stages after lineage commitment. At least three major expression patterns were observed. Transcription factors in the largest group are expressed at relatively stable levels throughout T-lineage specification as a legacy from prethymic precursors, with some continuing while others are downregulated after commitment. Another group is highly expressed in the earliest stages only, and is downregulated before or during commitment. Genes in a third group undergo upregulation at one of three distinct transitions, suggesting a positive regulatory cascade. However, the transcription factors induced during commitment are not T-lineage specific. Different members of the same transcription factor family can follow opposite trajectories during specification and commitment, while factors co-expressed early can be expressed in divergent patterns in later T-cell development. Some factors reveal new regulatory distinctions between αβ and γδ T-lineage differentiation. These results show that T-cell identity has an essentially complex regulatory basis and provide a detailed framework for regulatory network modeling of T-cell specification.

INTRODUCTION

T-cell development is an exceptionally favorable system in which to examine the mechanism of differentiation from stem cells in general. It offers the most completely identified sequence of developmental intermediates known for any post-embryonic stem-cell derivatives. Cells with T-cell precursor activity become physically segregated from other blood-cell precursors by migrating to the thymus at a very early stage of their development (Shortman and Wu, 1996; Ceredig and Rolink, 2002; Petrie and Zuniga-Pflucker, 2007; Rothenberg et al., 2008). This allows these intermediates to be purified for analysis with minimal interference from irrelevant cell populations. Also, because individual T-cell precursors can be isolated relatively easily, it has been possible to define the series of distinct steps at which particular developmental alternatives are lost during the stages leading up to lineage commitment (Rothenberg, 2007). T-lineage specification is both triggered and sustained by Notch signaling. Yet the way lineage-specific gene expression actually starts and how T-lineage identity is cemented at commitment are both questions still to be resolved.

T lymphocyte development from hematopoietic stem cells is not driven by a single “master gene”, but rather depends upon a large number of distinct regulatory factors, as shown by genetic evidence in targeted mutant mice (Anderson, 2006; Rothenberg and Taghon, 2005). Notch pathway signaling, triggered by ligands in the thymic microenvironment, is a central element, needed both to begin T-cell specification and to complete lineage commitment. However, to promote T-lineage development this signal must act on cells that also have an appropriate constellation of intrinsic regulatory activities. Among the transcription factors already known to have critical roles are GATA-3, Myb, Runx1+CBFβ, Ikaros, TCF-1 activated by Wnt, RBP-J activated by Notch, basic helix-loop-helix “E proteins”, Gfi1, and in an early hit-and-run role, hematopoietic transcription factor PU.1. Most of these regulatory inputs are needed to allow recognizable T-lineage precursors to emerge in the thymus from the earliest stages, long before the cells express T-cell receptors for antigen and undergo selection events that depend on those receptors. Most of the same factors are then used repeatedly over many cell cycles for days or weeks to define T-cell subtypes (Rothenberg et al., 2008; Ho and Pai, 2007; Tanigaki and Honjo, 2007; Anderson, 2006; Rothenberg and Taghon, 2005; Staal et al., 2001).

Although identities of these factors are clear, the transcriptional mechanisms that result in T-cell specification have remained more obscure than those involved in specification of many other hematopoietic cell types, especially as contrasted with another lymphoid cell type, namely B lymphocytes (Rothenberg, 2007; Rothenberg and Taghon, 2005). B cell identity is now known to be created through the action of a cascade of transcription factors in which key roles are played by EBF and Pax5, two dedicated factors that are B-lineage specific in expression as well as function within the hematopoietic system (Cobaleda et al., 2007; Singh et al., 2007; Nutt and Kee, 2007; Hagman and Lukin, 2006). Because of their specific expression patterns, mutations in these factors also yield selective defects in B-cell development that can be staged and characterized in molecular terms. These factors also act effectively as positive B-lineage regulators in both gain of function and loss of function experiments. The result has been a definitive elucidation of the B-lineage specification process (Cobaleda et al., 2007; Singh et al., 2007; Nutt and Kee, 2007; Hagman and Lukin, 2006). There are indeed two well-studied factors that are T-lineage specific in expression, namely GATA-3 and TCF-1 (product of the Tcf7 gene), and it has been known for years that these are both required specifically for T-cell development (Verbeek et al., 1995; Ting et al., 1996; Hattori et al., 1996). However, in marked contrast to the B-cell factors, these do not appear to upregulate T-lineage target genes specifically and do not enhance T-cell development in gain of function experiments (Taghon et al., 2007; Kirstetter et al., 2006; Scheller et al., 2006; Baba et al., 2005; Staal et al., 2004; Anderson et al., 2002a). One possible interpretation of existing data is that T-lineage choice is fundamentally different from B lineage choice; but another possibility has been that the EBF and Pax5 equivalents for T cells simply have yet to be recognized. Therefore, a crucial question is whether other transcription factors may exist in early T-lineage cells that could play more cell-type specific roles.

This study and its predecessors (Tydell et al., 2007; Anderson et al., 1999) were undertaken to identify any “missing” regulators of T-cell development that may be T-cell lineage specific, to track the major changes in transcription factors generally as cells progress toward T-lineage commitment, and to clarify how their roles and expression are coordinated with the observed landmarks in T-cell development (Rothenberg et al., 2008). In this study, we have relied on cDNA macroarray screening for gene discovery and on sensitive, quantitative realtime RT-PCR (qPCR) measurements to identify a large group of regulatory gene candidates and track the way that they and their family members are expressed in early T-lineage precursors. The results show that the T-cell specification process entails discontinuous upregulation of one group of transcription factor genes, surprisingly stable expression of a larger group of stem cell-inherited factors until after commitment, and downregulation of another substantial group of previously expressed transcription factor genes. However, few regulatory genes overall have highly specific T-cell expression, and while T-lineage commitment also causes upregulation of certain transcription factor genes, most of these are not T-lineage specific. These results add substantial detail to the transcription factor dynamics that underlie T-cell specification and strengthen the probability that there is a fundamental divergence between the overall regulatory strategies involved in T-cell and B-cell lineage choice.

MATERIALS AND METHODS

Animals

Cells were obtained from C57BL/6 mice and from mice of TCRβ^−/− and Rag2^−/− genotypes on a C57BL/6 background, all maintained in our breeding colony at the California Institute of Technology (Caltech). “Adult” samples were from 4–6 week old mice for thymocyte isolation and from 8–12 week old mice for bone marrow isolation. Fetal liver was obtained from timed pregnancies of C57BL/6 or (C57BL/6 × DBA/2) F₁ mice at E14.5 (day of plug = E0.5). Animals were kept in microisolator cages in an AAALAC-accredited facility with the immune-deficient mutants maintained with sterilized cages, water, food, and bedding. Euthanasia and animal care followed NIH guidelines, under protocols approved by the Institute Animal Care and Use Committee.

Gene discovery by macroarray screening

New cDNA clones of interest were isolated from a C.B17-Prkdc^scid/scid (SCID) mouse thymocyte cDNA library that was constructed and arrayed onto four large nylon filters as previously described (Anderson et al., 1999; Tydell et al., 2007). SCID thymocytes were used as starting material because their mutant phenotype prevents T-cell development from continuing through β-selection. Thus, >90% of these populations naturally consist of T-cell precursors in the DN1-DN3 stages. To select for cDNAs that might encode gene regulatory factors, four “macroarray” filters containing approx 73,000 clones were screened as previously described (Tydell et al., 2007) with probes designed to identify conserved domains associated with transcription factors.

To make these initial probes, gene sequences of representatives from various transcription factor families were aligned using ClustalX and primers were designed to their DNA binding domains or protein interaction domains. Primers used are tabulated in Table 1. The putative functional domain coding regions were then amplified from bulk SCID thymus cDNA and gel-purified. Domain-specific DNA probes were generated by random priming in the presence of α-³²P dCTP, using the gel-purified PCR amplification products as templates. Probe was hybridized to the macro-array filters overnight and washed with 3–4 low stringency washes to enable fully and partially matched hybrids to be detected. Macroarray filters were exposed to film and clones identified by positively hybridizing spot pairs that were picked from the archival library for further analysis as previously described (Tydell et al., 2007). A total of 1490 clones were selected into 96-well plates for sequencing and re-arraying onto smaller nylon filters.

TABLE 1.

Conserved Domain Primers for Macroarray Screen

A. AT-Hook		L. Homoeodomain
AT-Hook.For	5′-ctccgaaagagcccagtgaa-3′	HOXB4.For	5′-cggagctcggccaagtctatga-3′
AT-Hook.Rev	5′-ggtttcctccctggagctgt-3′	HOXB4.Rev	5′-ccggttatggaacgagatcttga-3′
B. ARID		M. Leucine Zipper
ARID.For	5′-ggacctttgaggagcagttcaa-3′	TSC.For	5′-gtgggaccgagccgctgcagat-3′
ARID.Rev	5′-ctcataggggtaaagatacttcat-3′	TSC.Rev	5′-gtcttcagcagattgttctcct-3′
C. bHLH		N. LIM Domain
MyoD.For	5′-ccaccaacgctgatcgccgcaa-3′	Lim-1.For	5′-cggagattaccagagtgagta-3′
MyoD.Rev	5′-cgcagcagagcctgcagacctt-3′	Lim-1.Rev	5′-gcaagctgggctcaggacta-3′
D. Bromodomain		4.5Lim.For	5′-gtcaggaatgccacaagcccat-3′
Bromo.For	5′-catcattaaacaccccatggacctcagta-3′	4.5Lim.Rev	5′-gcacaggatattcttgtccca-3′
Bromo.Rev	5′-gctcacctggcatcttggcaaatctcat-3′	O. NFAT
E. Chromodomain		NFAT.p-For	5′-ctcctctgccagcttcatt-3′
Chromo.Chd1-For	5′-caaccatttatgctgtcgaa-3′	NFAT.p-Rev	5′-ggcagaggagcaaccaaa-3′
Chromo.Chd1-Rev	5′-ctgctgcttcagggtctctt-3′	NFAT.x.For	5′-cctttgagtgcccaagtatccaa-3′
Chromo.Cbx/Mpc2-For	5′-cttcgcggtggagagcat-3′	NFAT.x.Rev	5′-gagatgctgctggcaggacta-3′
Chromo.Cbx/Mpc2-Rev	5′-cctttccctgttctggaa-3′	P. PAS
F. Cut-Homoeodomain		AHR.PAS-A.For	5′-caggcgctgaatggctttgt-3′
Cux.For	5′-gtcaaggaggtgctcaccgacaa-3′	AHR.PAS-A.Rev	5′-cttctgtatggatgagctcat-3′
Cux.Rev	5′-cagccacagctgcatacgcacaa-3′	AHR.PAS-B.For	5′-cgaaccaaaaacttcatcttca-3′
G. Ezrin-Radixin-Moesin (Protein Interaction)		AHR.PAS-B.Rev	5′-gtctgcagcatggatgaact-3′
ERM.For	5′-gacctgaagaccacatacgatgaa-3′	PER.PAS-A.For	5′-ggaacccggataccttcgctgt-3′
ERM.Rev	5′-ctgcagggtctgcctggccagtt-3′	PER.PAS-A.Rev	5′-acgtcatgaggagccaggagct-3′
H. Forkhead Box		PER.PAS-B.For	5′-gatccctcctgagaagaggat-3′
Fox.For	5′-tcacctatgccacccttat-3′	PER.PAS-B.Rev	5′-cagatcctgaggtaggtaa-3′
Fox.Rev	5′ctcctcttcttgcgaaactca-3′	Q. PDZ
I. GCM Box		PDZ.For	5′-cctctgacaatctctagtctgaa-3′
GCM.For	5′-gcgcaacaccaacaaccacaa-3′	PDZ.Rev	5′-ctttgtagagtcatattcaa-3′
GCM.Rev	5′-cttctgccttctgtctctgactt-3′	PDZk1.For	5′-gcagaggcagctggcttgaa-3′
J. HLH		PDZk1.Rev	5′-ccagcaccaacagagtagt-3′
Myf5.For	5′-caggctggccactgcctcat-3′	LAP.For	5′-gttgcgtttgctcagacgaga-3′
Myf5.Rev	5′-gttggtggtggtgcacctctt-3′	LAP.Rev	5′-caagcttgatgcaccgcgacgct-3′
K. HMG Box		WEGbox.For	5′acagataatgggggttgggct-3′
SOX.For	5′-gcaagccccatgctaagct-3′	WEG/PHD.Rev	5′-gtattgaacattgtctgcgccatt-3′
SOX.Rev	5′-ggtgattggcatggcggct-3	W. ZINC FINGERS
R. PHD		Btb/Poz
PHD.For	5′-gtgcttgcatgacatgtaataaa-3′	BTB/POZ.For	5′-catttgctgcagtgcctgaacgagca-3′
WEG/PHD.Rev	5′-gtattgaacattgtctgcgccatt-3′	BTB/POZ.Rev	5′-catctggagatagttggcagctgcaa-3′
ZIP/PHDf.For	5′-cagcagcattgaacagaagga-3′	ROG.For	5′-ggcaggagcaactaggatgat-3′
ZIP/PHDf.Rev	5′-ggtagtcccctttcatgttat-3′	ROG.Rev	5′-ggtcagtgggatcctctgat-3′
S. REL		C2H2
NFkB.For	5′-gctattcggtgagtaaagaa-3′	Multizf.For	5′-gcctcacagatcaccgaga-3′
NFkB.Rev	5′-gtttttgtagccctattttcat-3′	Multizf.Rev	5′-ggaagaggcttaaattgtt-3′
T. RNA Binding Domains		Fog
hnRNP.A2/B1-For	5′-ggctgcaaggcctcattccatt-3′	FOG.For	5′-gccccaggatgaagagaaa-3′
hnRNP.A2/B1-Rev	5′-gcctatcggtaattatttcaatagt-3′	FOG.Rev	5′-gcagtctttgcaggggaa-3′
hnRNP.U-For	5′-caggggcgaggcagggccaa-3′	Gli
hnRNP.U-Rev	5′-gacacaccgtaggaagctctt-3′	GLI.spfc-For	5′-gagacaaactgccactgggat-3′
hnRNP.x-For	5′-ggagagtggtgcacgtatcaa-3′	GLI.spfc-Rev	5′-gctcacagatgtaaggcttct-3′
hnRNP.x-Rev	5′-gcaaccaccttttccaataaga-3′	Groucho
Rbm.For	5′-gaaccagaggtgcttcaagaa-3′	Grg.For	5′-gaaatgcacaaacagactgaaat-3′
Rbm.Rev	5′-cccttggtaatggatactcat-3′	Grg.Rev	5′-ggtgcttcttgtcatcttttatt-3′
U. STAT		Ring Finger
STAT.For	5′-gaccctgtccctccctgtggt-3′	Ring/B-box.For	5′-gatcgaggatcttctgta-3′
STAT.Rev	5′-cccatgatagccccatcattcca-3′	Ring/B-box.Rev	5′-ggcctcatgattgctgta-3′
V. T-BOX		ZIC
T-box.For	5′-gagatgatcatcactaagcaa-3′	ZICspfc.GLI-For	5′-ctgttccgcaaccgtggctt-3′
T-box.Rev	5′-ggcagcctctggctctccat-3′	ZICspfc.GLI-Rev	5′-gggcagcatagtgctcggat-3′

Sequencing was performed by the modified Sanger’s method at the Institute for Systems Biology, Seattle, WA. The sequences were identified by BLASTn, BLASTx (http://blast.ncbi.nlm.nih.gov/Blast.cgi), and BLAT alignments against the Genbank “nr” databases and the mouse genome (up to builds 35 & 36; http://genome.ucsc.edu). New sequences that appeared to encode transcription factors or potential transcription-modifying factors, as well as sequences that defined new splice isoforms of known transcription factor genes, were submitted to Genbank as Expressed Sequence Tags (accession numbers listed in Table 2).

TABLE 2.

EST Sequences from SCID Thymocyte Library

Gene	Alt Name	EST ID#	GenBank Accession #	Date Submitted
ARHGEF2		39175456	EB739685	5/10/2006
		39175457	EB739686
ARID1A		39175458	EB739687	5/10/2006
		39175459	EB739688
BAHD1		39175460	EB739689	5/10/2006
		39175461	EB739690
BRD3		38305921	EB359574	4/6/2006
		38305922	EB359575
		38305923	EB359576
BTF3		39175462	EB739691	5/10/2006
		39175463	EB739692
Cbx4		39175464	EB739693	5/10/2006
CEBPd		39175465	EB739694	5/10/2006
		39175466	EB739695
CENTb1		39175467	EB739696	5/10/2006
CIZ	Zfp384	39175468	EB739697	5/10/2006
		39175469	EB739698
		39175470	EB739699
		39175471	EB739700
		39175472	EB739701
		39175473	EB739702
CRABP1		39781094	EC277794	6/8/2006
		39781095	EC277795
CSF2Ra		39781134	EC277834	6/8/2006
		39781096	EC277796
CTBP2		39781097	EC277797	6/8/2006
		39781098	EC277798
CUTL1		39781099	EC277799	6/8/2006
		39781100	EC277800
DAPLE	CCdc88c	39781101	EC277801	6/8/2006
		39781102	EC277802
		39781103	EC277803
		39781104	EC277804
DBP		39781105	EC277805	6/8/2006
		39781106	EC277806
DIMP	Dmtf1	39781107	EC277807	6/8/2006
		39781108	EC277808
DSC43		39781109	EC277809	6/8/2006
		39781110	EC277810
		39781111	EC277811
		39781112	EC277812
		39781113	EC277813
		39781114	EC277814
		39781115	EC277815
		39781116	EC277816
		39781117	EC277817
ELK3		39781118	EC277818	6/8/2006
		39781119	EC277819
		39781120	EC277820
		39781121	EC277821
		39781122	EC277822
ELK4		39781123	EC277823	6/8/2006
		39781124	EC277824
		39781125	EC277825
ESR1		39781126	EC277826	6/8/2006
		39781127	EC277827
		39781128	EC277828
		39781129	EC277829
		39781130	EC277830
		39781131	EC277831
		39781132	EC277832
		39781133	EC277833
FLI1		39877808	EC365110	6/13/2006
		39877809	EC365111
		39877810	EC365112
		39877811	EC365113
		39877812	EC365114
GATA3		39877813	EC365115	6/13/2006
		39877814	EC365116
GIMAP6		39877815	EC365117	6/13/2006
GNAI2		39877816	EC365118	6/13/2006
		39877817	EC365119
GSE1		39877818	EC365120	6/13/2006
		39877819	EC365121
HKR2	Zscan22	39877820	EC365122	6/13/2006
HKR3		39877821	EC365123	6/13/2006
HMGA1		38305924	EB359577	4/6/2006
		38305925	EB359578
		38305926	EB359579
		38305927	EB359580
		38305928	EB359581
		38305929	EB359582
		38305930	EB359583
		38305931	EB359584
		38305932	EB359585
		38305933	EB359586
IARS2		39877822	EC365124	6/13/2006
IKAROS		39877823	EC365125	6/13/2006
		39877824	EC365126
		39877825	EC365127
		39877826	EC365128
		39877827	EC365129
		39877828	EC365130
		39877829	EC365131
		39877830	EC365132
JMJD3		39877864	EC365166	6/13/2006
		39877831	EC365133
		39877832	EC365134
		39877833	EC365135
KLF2		39877834	EC365136	6/13/2006
		39877835	EC365137
		39877836	EC365138
		39877837	EC365139
		39877838	EC365140
		39877839	EC365141
		39877840	EC365142
		39877841	EC365143
		39877842	EC365144
		39877843	EC365145
KLF13		39877844	EC365146	6/13/2006
		39877845	EC365147
		39877846	EC365148
		39877847	EC365149
		39877848	EC365150
		39877849	EC365151
		39877850	EC365152
		39877851	EC365153
		39877852	EC365154
		39877853	EC365155
KLF15		39877854	EC365156	6/13/2006
		39877855	EC365157
		39877856	EC365158
		39877857	EC365159
		39877858	EC365160
		39877859	EC365161
		39877860	EC365162
		39877861	EC365163
		39877862	EC365164
		39877863	EC365155
MTA2		40881949	EE265166	8/3/2006
MTA3		40881950	EE265167	8/3/2006
		40881951	EE265168
NCOR2		40881952	EE265169	8/3/2006
PAX1		40881953	EE265170	8/3/2006
		40881954	EE265171
PCBP2		40881955	EE265172	8/3/2006
		40881956	EE265173
Per1		40881976	EE265193	8/3/2006
		40881957	EE265174
POU2F1		40881958	EE265175	8/3/2006
		40881959	EE265176
		40881960	EE265177
		40881961	EE265178
POU6F1		40881962	EE265179	8/3/2006
		40881963	EE265180
		40881964	EE265181
PPARd		40881965	EE265182	8/3/2006
		40881966	EE265183
PPRC1		40881967	EE265184	8/3/2006
PurB		40881968	EE265185	8/3/2006
RARa		40881969	EE265186	8/3/2006
		40881970	EE265187
		40881971	EE265188
RBAK		40881972	EE265189	8/3/2006
		40881973	EE265190
		40881974	EE265191
		40881975	EE265192
SBF1		41528217	EE663535	8/24/2006
SFPI1		41528238	EE663556	8/24/2006
		41528249	EE663567
SOX4		41528260	EE663578	8/24/2006
		41528271	EE663589
		41528176	EE663494
		41528187	EE663505
		41528189	EE663507
		41528190	EE663508
		41528191	EE663509
		41528192	EE663510
SPI-B		41528193	EE663511	8/24/2006
SUPT5h		41528194	EE663512	8/24/2006
		41528195	EE663513
TBX21		41528196	EE663514	8/24/2006
		41528197	EE663515
		41528198	EE663516
TCF3	E2A	41528199	EE663517	8/24/2006
		41528200	EE663518
		41528201	EE663519
		41528202	EE663520
		41528203	EE663521
		41528204	EE663522
TCF7		38305934	EB359587	4/6/2006
		38305935	EB359588
		38305936	EB359589
		38305937	EB359590
		38305938	EB359591
		38305939	EB359592
		38305940	EB359593
		38305941	EB359594
		38305942	EB359595
		38305943	EB359596
TCF12	HEB	41528205	EE663523	8/24/2006
		41528206	EE663524
		41528207	EE663525
		41528208	EE663526
		41528209	EE663527
THRAP3		41528210	EE663528	8/24/2006
		41528211	EE663529
TLE3		41528212	EE663530	8/24/2006
		41528213	EE663531
		41528214	EE663532
Tmsb4x		41528215	EE663533	8/24/2006
		41528216	EE663534
TRIM24		41528218	EE663536	8/24/2006
TRIM25	EFP	41528219	EE663537	8/24/2006
		41528220	EE663538
TRIM28		41528221	EE663539	8/24/2006
TSC-22d1		41528222	EE663540	8/24/2006
		41528223	EE663541
		41528224	EE663542
YY1		38305944	EB359597	4/6/2006
		38305945	EB359598
		38305946	EB359599
ZBTB20		39175474	EB739703	5/10/2006
		39175475	EB739704
ZBTB7a		41528225	EE663543	8/25/2006
		41528226	EE663544
		41528227	EE663545
		41528228	EE663546
		41528229	EE663547
		41528230	EE663548
		41528231	EE663549
		41528232	EE663550
		41528233	EE663551
		41528234	EE663552
		41528235	EE663553
ZC3H5		41528236	EE663554	8/25/2006
		41528237	EE663555
ZFP3		41528239	EE663557	8/25/2006
		41528240	EE663558
ZFP28		41528241	EE663559	8/25/2006
ZFP32	ZFP637	41528242	EE663560	8/25/2006
		41528243	EE663561
ZFP36 l2		41528244	EE663562	8/25/2006
		41528245	EE663563
		41528246	EE663564
		41528247	EE663565
		41528248	EE663566
ZFP41		41528250	EE663568	8/25/2006
		41528251	EE663569
ZFP90		41528252	EE663570	8/25/2006
		41528253	EE663571
ZFP95	Zkscan5	41528254	EE663572	8/25/2006
		41528255	EE663573
ZFP110		41528256	EE663574	8/25/2006
		41528257	EE663575
ZFP238		41528258	EE663576	8/25/2006
		41528259	EE663577
ZFP286		41528261	EE663579	8/25/2006
		41528262	EE663580
ZFP287		41528263	EE663581	8/25/2006
		41528264	EE663582
ZFP316		41528265	EE663583	8/25/2006
		41528266	EE663584
ZFP358		41528267	EE663585	8/25/2006
		41528268	EE663586
ZFP422		41528269	EE663587	8/25/2006
		41528270	EE663588
ZFP598		41528272	EE663590	8/25/2006
		41528273	EE663591
ZFP672		41528274	EE663592	8/25/2006
		41528275	EE663593
ZFP691		41528276	EE663594	8/25/2006
		41528277	EE663595
ZFPN1A4	Eos	41528278	EE663596	8/25/2006
		41528279	EE663597
		41528280	EE663598
		41528281	EE663599
Zip67		41528177	EE663495	8/25/2006
		41528178	EE663496
		41528179	EE663497
		41528180	EE663498
		41528181	EE663499
		41528182	EE663500
		41528183	EE663501
Zipro1		41528184	EE663502	8/25/2006
		41528185	EE663503
ZKSCAN1		41528186	EE663504	8/25/2006
		41528188	EE663505

Primary T-lineage cell samples

Cell populations of interest were collected by fluorescence-activated cell sorting (FACS), and cells were processed in Qiazol (Qiagen, USA) for RNA. RNA was purified using Qiagen RNAeasy kits (Qiagen), and treated with RNase-free DNaseI (Ambion) to remove genomic DNA contamination before conversion into first strand cDNA using Superscript Reverse Transcriptase (Invitrogen).

To isolate intermediates in T-cell development, antibody-mediated depletion with magnetic beads and multiparameter fluorescence-activated cell sorting (FACS) were used as previously described (Taghon et al., 2006; Yui and Rothenberg, 2004; Anderson et al., 1999). DN subsets were sorted from wild type C57BL/6 mouse thymi that were first depleted for CD3, CD8, TCRβ, TCRγδ, NK1.1, Gr-1, F4/80, B220, and Ter119. For “Series A”, cells were sorted into DN1 (or ETP, early T-cell precursor)(CD117+ CD44+ CD25−), DN2 (CD117+ CD44+ CD25+), DN3 (CD117− CD44− CD25+), and DN4 (CD117− CD44− CD25−). For “Series B”, cells were sorted into DN1, DN2, DN3a, and DN3b subsets using CD27 to split the DN3 subsets as described previously (Taghon et al., 2006), and isolating DN4 cells as CD44− CD25− CD24+. To examine DN subsets from Rag2^−/− mutant thymus, first 10% of dissociated thymocytes were reserved to sort for DN2 = Thy1.2+ CD44+ CD25+ and DN3 = Thy1.2+ CD44− CD25+ stages. The remaining 90% cells were depleted of CD24, Gr-1, F4/80, and Ter119 and then sorted to separate Sca-1+Thy1− (precursor-enriched) and “pre-NK” = Sca1− Thy1+/− subsets (David-Fung et al., 2006). These “pre-NK” cells are strongly enriched for expression of NK cell genes such as perforin and Id2 (Wang et al., 2006; David-Fung et al., 2006; Wang et al., 1998).

To analyze the development of fetal hematopoietic precursors developing in OP9-control or OP9-DL1 culture, Lin- c-Kit⁺ CD27⁺ multilineage precursors (Lin = Ter-119, CD19, F4/80, Gr-1) were isolated by FACS and cultures were established exactly as described previously (Taghon et al., 2005). In these experiments, OP9-control cultures generated B cells predominantly while OP9-DL1 cultures generated T cells. Samples were generated from two independent large-scale time course experiments, one of which was a 10-day time-course also used for previous analyses (Taghon et al., 2005; Tydell et al., 2007), and another set in which cells retaining precursor phenotypes were re-sorted after the first two days, returned to OP9 coculture, and harvested after just 4 days overall. Some samples from these time-courses were also sorted again at day 4 to re-isolate cells retaining the Lin⁻ c-Kit⁺ CD27⁺ precursor phenotype and to separate these from CD27⁻ Mac-1⁺/Gr-1⁺ cells that had also been generated in the cultures.

To assess gene expression in later T-cell development, thymic cells from wild type C57BL/6 mice were sorted by FACS directly to isolate double positive (DP) and single positive CD4+ cells, while another aliquot of the thymocytes was first depleted of CD4, B220 and NK1.1 expressing cells and then sorted to isolate double negative (DN) and single positive CD8+ cells. The markers used were: DN = CD24+ CD4− CD8−, DP = CD4+ CD8+, CD4 SP4 = CD4+ CD8−, CD8 SP = TCRβ + CD24− CD4− CD8+. TCRγδ cells were obtained from TCRβ^−/− thymocyte populations that were depleted for CD4 and CD8 prior to sorting for expression of CD3ε and TCRγδ. To assess non-T lineage expression in adult tissues, pro-B and pre-myeloid cells were prepared from Rag2^−/− mutant bone marrow as described previously (Tydell et al., 2007).

Quantitative real-time reverse transcriptase-dependent PCR

Quantitative reverse transcriptase-dependent PCR (qPCR) was performed to measure gene expression in different samples of cDNA from the cells described above. Primers were designed to genes of interest using the PrimerExpress^™1.5 (ABI) or Primer3 (http://fokker.wi.mit.edu/primer3/input.htm) programs. They were designed as far as possible to span introns of > 200 bp length, to avoid amplification of genomic DNA. Intron-exon information was verified from Ensembl (http://www.ensembl.org) and UCSC mouse genome assemblies. Primers utilized are shown in Table 3. Many primer sets used in this study have also been reported previously (Anderson et al., 2002b; Anderson et al., 2002a; Hernández-Hoyos et al., 2003; Yun and Bevan, 2003; Yui and Rothenberg, 2004; Anderson et al., 2004; Dionne et al., 2005; Taghon et al., 2005; David-Fung et al., 2006; Taghon et al., 2006; Franco et al., 2006; Taghon et al., 2007; Tydell et al., 2007). Standard curves were run for new primers to select sets with dilution curves giving close to the theoretical value of ΔC_T 3.3 per tenfold dilution. Note the anomalous titration behavior of two primer sets, Stat5b and Thrap3 (Table 3), which should be treated with caution, but the great majority showed amplification consistent with 1.8–2.0 fold amplification per cycle. Samples were run in triplicate on an ABI Prism 7700 or 7900HT realtime PCR machine, and thresholds were set uniformly across all genes analyzed with a particular reference standard to calculate ΔC_T = C_T(gene) − C_T(standard). Levels of expression were converted to units of β-actin or GAPDH expression using the formula: expression = (1.9)^−ΔCt.

TABLE 3.

Primer sequences used for Quantitative PCR

New1 primers	Gene tested	Alternative name	Primer Sequences (Forward; Reverse)	Comments Titration quality2
+	4632433K11Rik	AK014606	5′-ACAAGACTGACAGAGAAGACTTACAGAAA-3′ 5′-GGGCTAAAAGGACTTGGGAGTT-3′
	Actb	β-actin	5′-ACACCCGCCACCAGTTC-3′ 5′-TACAGCCCGGGGAGCAT-3′
	Aff3	LAF4	5′-TGACACCTCCCACCATGGA-3′ 5′-TTTCCTCCGTAACGCATTCC-3′	*(−)
+	ARID1a	Smarcf1	5′-ACGTGGGCGTTAGACACCAT-3′ 5′-AATATTCCACAAGGAGCTCTAGCAA-3′	*(−)
+	Atxn2l		5′-CTCTCCCTGACCCCTACAGA-3′ 5′-CAGGGACTTTCCCAGCTAT-3′
+	B2m	β2-microglobulin	5′-TCACTGACCGGCTTGTATGCT-3′ 5′-TGAGGCGGGTGGAACTGT-3′	Used in preliminary analysis only
+	Bahd1	mKIAA0945	5′-TCCCGCTTGGAGGGATTC-3′ 5′-AGTCCCCTGCTCCATATTGCT-3′
	Bcl11a	CTIP1	5′-GTCTGCACACGGAGCTCTAA-3′ 5′-CACTGGTGAATGGCTGTTTG-3′	Tydell et al., 2007
	Bcl11b	CTIP2	5′–GGGCGATGCCAGAATAGAT–3′ 5′–GGTAGCCTCCACATGGTCAG–3′	Primers from Tydell et al., 2007
+	Bcl6		5′-GTGGAGATGAAGCCTGTAGCA-3′ 5′-CACTTGAATCGTGCAGTGGTA-3′	**(−); Not expressed in thymocytes
+	Brd3		5′-AAATGCAGGTTCCCAACAAGTG-3′ 5′-CGGCAATGACGGGTGTCT-3′	*(−)
+	Brd4		5′-CCCTGAAGCCATCTACACTACGA-3′ 5′-ACCAGCAATCACGTCAACTTT-3′
+	Cbx4		5′-AGCTGATGGGATATCGCAAGAG-3′ 5′-GTCTTGAAGCCCAGTCAGAACAT-3′
+	Ccdc88c	Daple	5′-CCTGGGTGAAAACTTTTGGA-3′ 5′-CTGGGGTCTATTTGCAGCAT-3′
	Cd3e		5′-CGTCCGCCATCTTGGTAGAG-3′ 5′-ATTCAATGTTCTCGGCATCGT-3′
	Cd3g		5′-TGGAGAAGCAAAGAGACTGACA-3′ 5′-GCCATCCACTTGTACCAAATTC-3′.
	Cebpa	C/EBPα	5′–CGGTCATTGTCACTGGTCAACT–3′ 5′-GGACAAGAACAGCAACGAGTACC-3′	*(++) Only one intron
	Cebpb	C/EBPb	5′-GTTTCGGGACTTGATGCAATC-3′ 5′-CGCAGGAACATCTTTAAGGTGAT-3′
	Cebpd	C/EBPd	5′-TCCACGACTCCTGCCATGTA-3′ 5′-TGAAGAGGTCGGCGAAGAGTT-3′
	Cebpg	C/EBPg	5′-GCGCAGGTACATGTGAAGATT-3′ 5′-CTGCGACAGCTTGCTCATT-3′
+	Cep170	KAB1	5′-GCCGAGCATCCTGATCACTT-3′ 5′-AGAGAACTGAAAGACTGATGACAATAGC-3′	First thought to interact with KRAB domains
+	Csf2ra1	GM-CSFRα	5′-CACCGCGTCCTGTAACTCTT-3′ 5′-GCGCTCTGCCACTGGACCTCAAACT-3′	Data not shown; exp. declines DN1-4
+	Ctbp1		5′-TGAGTCAGAGCCCTTCAGCTTT-3′ 5′-CAATGGACGCCTGCTCACT-3′	*(−)
+	Cutl1		5′-CCATGGAAACTGTCTCTGTCTTCA- 5′-TGTCAGACTAGCTGGGTTGTTTG-3′	*(+)
+	Dmtf1	Dimp	5′-ACACGGACGGGAATCTCATTC-3′ 5′-AATGCTTTGATCATCCTCAGAGG-3′	*(+)
	Dtx1	Deltex1	5′-GAGGATGTGGTTCGGAGGTA-3′ 5′-CCCTCATAGCCAGATGCTGT-3′	*(−)
	Elf4	MEF (myeloid Ets)	5′-CAGGCTCACCAAAACTGTGA-3′ 5′-TTGGTCAGCACCGTAGTCAG-3′
+	Elk3	ERP	5′-CTGAGCAGAGCGCTGAGATAC-3 5′-TTCAGGATATCCGGGAAAGAG-3′	**(+)
+	Elk4		5′-GCTTTTCCAGTTTCCCTCTGT-3′ 5′-TCTGCAGGTCTGGAGAAAATG-3′
+	Erg		5′-GCCTCCCAATATGACCACAA-3′ 5′-TATTCTTTCACCGCCCACTC-3′
+	ESR1	Estrogen receptor 1	5′-GCTCCTGGACAGGAATCAAG-3′ 5′-CTTCTCCCTGCAGGTTCATC-3′
	Ets1		5′-AAAAGTGGATCTCGAGCTTTTCC-3′ 5′-CTTTCAAGGCTTGGGACATCA-3′
	Ets2		5′-GCAAGGCAAACCAGTTATTCCT-3′ 5′-ACTTGTCAGAGAGTAGCTCCAGAAGAA-3′
+	Etv6	TEL	5′-TCCTGCTGCTGACCAAAGAG-3′ 5′-CTGGCTTGGTGTGGATAGAG-3′
+	Fli-1		5′-ACAGACCAGTCCTCACGACTGA-3′ 5′-CTTTTGTTGAGGCCAGAGTTCAT-3′
+	FoxJ3		5′-TGATCCCGGAAAGGGTTCTT-3′ 5′-AGGCCCGTTCTACGGATCTT-3′	*(−)
+	Foxp1		5′–CTGGAAAACAGCCGAAAGAG–3′ 5′–GGTTGGAGGGGAAGGGCAGG–3′
+	Foxp4		5′-CCCACAACCCTACCTTCTTCATC-3′ 5′-GGGACCGGAGAACTAATTTTCA-3′	*(−)
	Fus		5′-CAGCAACGAGCTGGAGACTG-3′ 5′-TCTGGCTTAGGTGCCTTACACTG-3′
+	Gabpa		5′-TGCACTGGAAGGCTACAGAAAA-3′ 5′-TTACCCAAACCACCCAATGC-3′	*(−)
	Gapdh		5′-ACTCCACTCACGGCAAATTCA-3′ 5′-GCCTCACCCCATTTGATGTT-3′
	Gata2		5′-ACCACAAGATGAATGGACAGAA-3′ 5′-GTCGTCTGACAATTTGCACAAC-3′	*(−)
	Gata3		5′-GAGGTGGTGTCTGCATTCCAA-3′ 5′-TTTCACAGCACTAGAGACCCTGTTA-3′	Called “Gata3-endog” in Taghon et al., 2007
	Gata3 (alternate)		5′-CCTGCGGACTCTACCATAAAA-3′ 5′-GTGGTGGTGGTCTGACAGTTC-3′	In coding region; Taghon et al 2007
	Gfi1		5′-AGCGTCGGAGAAGTCACTGT-3′ 5′-CAGGTCAGACCCAGCAAGAC-3′
	Gfi1b		5′-CCTTTGCCTGTGATGTCTGT-3′ 5′-ATGAACGCTTGAAGGCTTTG-3′
+	Gse1		5′-GGGATTGAGGCAATCTTTGA-3′ 5′-GGCTGAGGCTGTAGTTCTGC-3′
	HEBAlt		5′-GTGCTTATCCTGTCCCTGGAATG-3′ 5′-TGGCTTGGGAGATGGGTAAC-3′
	Hes1	(Yun & Bevan)	5′-TACCCCAGCCAGTGTCAACA-3′ 5′-TTCTTGCCCTTCGCCTCTT-3′
	Hes1 (alternate)		5′-CCAAGCTAGAGAAGGCAGACA-3′ 5′-CGGTATTTCCCCAACACG-3′	Alternate primers, Yui & E.V.R. 2004; data not shown but similar
+	Hmga1	HMG-I(Y)	5′-AAAGTATCACCGGGTGGGTTTC-3′ 5′-CATTAAAGTGGGTGGAGCCAAC-3′
+	Hprt		5′-GAGCTACTGTAATGATCAGTCA-3′ 5′-ACCAGCAAGCTTGCAACCTT-3′
	Id1		5′-GGCGAGATCAGTGCCTTG-3′ 5′-AAGGGCTGGAGTCCATCTG-3′	**(+)
	Id2		5′-GTCCTTGCAGGCATCTGAAT-3′ 5′-CTCCTGGTGAAATGGCTGAT-3′	*(+) Possible detection of genomic DNA?
	Id3		5′-AGAGGAGCTTTTGCCACTGA-3′ 5′-TGGAGAGAGGGTCCCAGAGT-3′	**(−)
	Ikzf1	Ikaros	5′-TCCCAAGTTTCAGGAAAGGA-3′ 5′-TCTGCTGTGCTCCAGAGGTA-3′
	Ikzf2	Helios	5′-CACCTCAGGACCCATTCTGT-3′ 5′-TGACAGCGTTCCTTGTGTTC-3′
	Ikzf3	Aiolos	5′-CCGAGATGGGAAGTGAGAGA-3′ 5′-CGCTTCTCACCGATGAATTT-3′	*(−)
+	Ikzf4	Eos; Zfpn1a4	5′-CCCAAACAGCCAACACTCTT-3′ 5′-TTATCCAGGAGCCGTTCATC-3′
	Il2rb	IL-2/15Rβ	5′-ACGTCCATGCCAAGTCGAA-3′ 5′-GGAACGACCCGAGGATCAG-3′
	IL2-UP	Il2 flanking seq.	5′-ACCTTGGGAGCTGAAATCCT-3′ 5′-TTTTGAGGGATCGCTAATGG-3′	Detects genomic DNA; Adachi & E.V.R., 2005
	Il7ra	IL-7Rα	5′-AGTCCGATCCATTCCCCATAA-3′ 5′-ATTCTTGGGTTCTGGAGTTTCG-3′	*(−)
+	Jmjd3		5′-CCCTACCCCCAGCATCTATT-3′ 5′-GCCTAAGTTGAGCCGAAGTG-3′
+	Klf13		5′-CGGGCTGCGAGAAAGTTTAC-3′ 5′-GAGCGTGCGAACTTCTTGTT-3′
+	Klf15		5′-TTTTCCCGCTCAGATGAGTT-3′ 5′-GCGATGCACTTTGATGTGTT-3′	*(+) No expression in thymocytes
+	Klf2		5′-ATGCACCTGAGCCTGCTAGT-3′ 5′-ATAGCACTGCCCTCTCCCTCT-3′	*(+) crosses tiny intron
	Lat		5′-CTGTTGTCTCCTCTGCTCCTGT-3′ 5′-CTCACTCTCAGGAACATTCACG-3′
	Lck		5′-CTAGTCCGGCTTTATGCAGTG-3′ 5′-CCGAGGGAGTCTTGAGAAAAT-3′
	Lef1		5′-ACCTACAGCGACGAGCACTT-3′ 5′-GGGTAGAAGGTGGGGATTTC-3′
+ +	Lima1	Eplin	a: 5′-AGGAGAGCTGCGTGGAGTGT -3′ 5′-GCAGTAGGAGCATCGGAAACAG-3′ b: 5′-AGGAGAGCTGCGTGGAGTGT-3′ 5′-TGCAGTAGGAGCATCGGAAAC-3′	**(−) *(−)
	Mitf		5′-CCCCAAGTCAAATGATCCAG-3′ 5′-CCTTAGCTCGTTGCTGTTCC-3′	Taghon et al., 2007
+	Mlf2		5′-ACCAGGAGACGTCGGAGATG-3′ 5′-GCCCTATCCCGAATGTGATG-3′
	MLL1	ALL-1	5′-TGCCCATAGCCCATCATCA-3′ 5′-TCTGTGAATGAGGCCGATCTG-3′
	MLL2		5′-CCCGGTCCCGAATCAAAC-3′ 5′-GAAGAGAACCACGGTATTTTGCAT-3′
+	Mta2	Mta1l1	5′-CTGAAGACCCCTACCCAACTTG-3′ 5′-CCCGGTTAGCACTGGTTGTATAG-3′
+	Mta3		5′-CGTGCCGTGGGAACATTT-3′ 5′-GTGTGTCCATGGCATGAAACA-3′	*(+)
	Myb	c-Myb	5′-GAGCAGAAGAAGTTTCCCGATTT-3′ 5′-AGCGGGAATCGGATGAATCT-3′
	Mybl2	B-myb	5′-AAGGAGGTGCTCCGTTCTGA-3′ 5′-CCAGAGACTTGCGGACCTTCT-3′
+	Myst3	MOZ	5′-CCAGTGCTGAAGGGAAAGAG-3′ 5′-ACAGATGGGGATTGGTTCAG-3′	*(+)
+	Nasp		5′-TGTAGCCGAACTGGCACTGA-3′ 5′-TTCTACTGGCAATCATGGATACTGA-3′
+	Ncor1		5′-AATCTCATCCCCTATTAAACCAAACC-3′ 5′-GTATATTACATGGAGTGCAAGAAACCA-3′
+	Ncor2		5′-CGCACACCGGATCCTAGAAG-3′ 5′-TCCGCTTAAAGTACAAGATCAGCTT-3′
	Nfe2		5′-TATGCAGCTTTTGGCAGAGA-3′ 5′-GAGGGGCAGTGAAGACTGAA-3′
+	Pax1		5′-GGCAATGACCTTCAAACACC-3′ 5′-CTGTGAGAGGACAGCCCCTA-3′	No expression in thymocytes
	Pax5		5′-ACAGGACATGGAGGAGTGAATCA- 3′ 5′-CCTTGATGGGCAAGTTCCACTA-3′	Anderson et al., 2004; no expression in T
+	Per1	Period	5′-CCGAATACACACTTCGAAACCA-3′ 5′-CGAAACACATCCCGTTTGC-3′
+	Pou2f1	Oct1	5′-AACCACCAACCTGCAACCA-3′ 5′-TGCTGAGGTAGTTGCGTTAAAAGAT-3′	**(−)
+	Pou5f1	Oct4	5′-GGAAAGGTGTTCAGCCAGAC-3′ 5′-CTCATTGTTGTCGGCTTCCT-3′	No expression in thymocytes
	Pou6f1	Brn-5	5′-CTGCAACTCCCATCCCAATC-3′ 5′-CGCAAACTCCCGGATCTCTTCT-3′	*(−)
	Ptcra	Pre-Tα, pTα	5′-CTGGCTCCACCCATCACACT-3′ 5′-TGCCATTGCCAGCTGAGA-3′
+	Ptma	Prothymosin α	5′-GGCGTGCCCCACCAT-3′ 5′-CATTCTCTGCCTCCTCCACAAC-3′
+	Rara		5′-CTCCCAGGACCGCTTAACC-3′ 5′-GGAGCTGCAAGTCCCAAGT-3′
+ +	Rbak		5′-TAAGTTGGAGCAGGGAGAAGAG-3′ 5′-CAATTCCTTCTCTGGGGTCATA-3′ 5′-TCGTCATGCTACACTGTGCACTAC-3′ 5′-CCGGTGTGGACCCTTTGA-3′	Library-cloned new isoform Conventional isoform
+	Rnf149		5′-GGAATTGATGTCGATGCTGA-3′ 5′-GCCATGGGTCAATGCATATT-3′
+	Rreb1		5′-TCACCACCAATGGGAACATG-3′ 5′-AACCGCCTGCGCTTCAG-3′	*(−)
	Runx1		5′-CTCGGCAGAACTGAGAAATG-3′ 5′-GGTGATGGTCAGAGTGA-3′
+	Runx2		5′-GCCTCCGCTGTTATGAAAAA-3′ 5′-TGGGGAGGATTTGTGAAGAC-3′	*(+)
	Runx3		5′-GGTTCAACGACCTTCGATTC-3′ 5′-GGTCCATCCACAGTGACCTT-3′
+	Saps1	mKIAA1115	5′-TTCATTGGCTCAACGAAGAGA-3′ 5′-GGCTTAGGCGGATGATGTCA-3′	*(+)
	Satb1		5′-CCACAAACACGGAGGTCTCT-3′ 5′-GCAATCCCTGAGTTCGGTTA-3′
+	Sbf1		5′-AGGAGCTGCTGGATGTGATT-3′ 5′-CCATGCTCAGAACATTGTGG-3′	*(+)
	Sfpi1	PU.1	5′-CCCGGATGTGCTTCCCTTAT-3′ 5′-TCCAAGCCATCAGCTTCTCC-3′
	Sox4		5′-TCAAGGACAGCGACAAGATTC-3′ 5′-GCCGGTACTTGTAGTCAGGGTA-3′
	SpiB-1	SpiB, ex3-4	5′-CTGCAAGCCCTTCAGTTACC-3′ 5′-AAAGGCAGCAGTAGCAGGAT-3′	See Dionne et al. 2005; David-Fung et al 2006
	SpiB-2	SpiB, ex1-2	5′-CTTGCTCTGGAGGCTGCAC-3′ 5′-CCCCCATCTGAATCTGGGTA-3′	*(−); used in Yui & Rothenberg, 2004
+	Stat5b		5′-GCTGTATCCGGCACATTCTGT-3′ 5′-GTTTGGTTGATCTGAAGGTGCTT-3′	**(+) Serious titration anomaly
	Tal1	SCL	5′-CTCACTAGGCAGTGGGTTCTTT-3′ 5′-GGACCATCAGAAATCTCCATCT-3′	*(−)
+	Tbx21	T-bet	5′-AGGTGTCTGGGAAGCTGAGA-3′ 5′-ATTCGCCGTCCTTGCTTAGT-3′	**(+)
	Tcf12	HEB (canonical)	5′-GAGAAGAAGACCGCTCCATGAT-3′ 5′-TGGCTTGGGAGATGGGTAAC-3′
	Tcf7	TCF-1	5′-CAAGGCAGAGAAGGAGGCTAAG-3′ 5′-GGCAGCGCTCTCCTTGAG-3′	Intra-exon: detects N-term half of the HMG domain: “Tcf7.8”
	Tcf7 (alt primers)	TCF-1	5′-TGCTGTCTATATCCGCAGGAAG-3′ 5′-CGATCTCTCTGGATTTTATTCTCT-3	*(+) Crosses from exon III to IV: “Tcf7.3”; Taghon et al., 2005
	Tcfe2a	E2A, Tcf3	5′-CAGATGGTGGCCTGGATACT-3′ 5′-CATCCCTGCTGTAGCTGTCA-3′
+	Tfdp1		5′-GATAGGTGAATGGATCAGAGGTTAAGA-3′ 5′-CCTTGGAACTGGAGTCACAGACTT-3′	**(−)
+	Thrap3		5′-GTTGTGCCGTTGCGAGAT-3′ 5′-CTTGGGCGTGTACTTTGGAT-3′	**(+) Serious titration anomaly
+	Tle3		5′-TTGCGAAGAGACTGAACACAAT-3′ 5′-GCGTTCAACTCCGTCATGGT-3′
	Tox		5′-CCATGGACCTGCCAGAGATC-3′ 5′-TTCTGCGTTCCCAATCTCTTG-3′
+	Trim25	Efp	5′-CGGAAAATTCGACACCATCT-3′ 5′-CCTTGCAATTTTGCAGCTTT-3′
	Trim44		5′-TCTGTGTCCTGTGTCCAGTCATT-3′ 5′-CAGTCCACCGGAATCTTTGC-3′
+	Tsc22d1		5′-AAACGCTTCCGTGAGACTTG-3′ 5′-TCACCGCATACATCAAATGG-3′	*(+)
+	Unk		5′-GTTGGGGCAGAGTACCTGAA-3′ 5′-ACGTCCTGTTTGGGTTGAAG-3′
+	Yy1		5′-GCCAGAATGAAGCCAAGAAA-3′ 5′-GGTGTGCAGATGCTTTCTCA-3′
+	Zbtb20		5′-CAGCCAAACAGAACTACGTCAA-3′ 5′-TGCTACACTGGTACGCTCTCA-3′	**(−)
+	Zbtb48		5′-TGTCCCACATGTCACAAAAAGTT-3′ 5′-TCTCCTTGCGGAAGTAACACTTC-3′	**(+)
+	Zbtb7a	LRF, Pokemon	5′-CCCTACGAGTGTAACATCTGTAAAGTTC-3′ 5′-GTGGTTCTTCAGGTCGTAGTTGTG-3′	*(+) More sensitive due to shorter amplicon than for primers in Maeda et al. 2007
+	Zfp1	mKr1	5′-TGGAAGTGTGGAAGGCTGATG-3′ 5′-TTTCTTCAGTTGGTGGTTGGTTC-3′	*(−)
	Zfp109		5′-GCTGCTCAGAGGAAGCTGTA-3′ 5′-CCCCAGTGAAAGGCATCTTA-3′
+	Zfp110		5′-ATGATGGAGAACTACAGCAACATGA-3′ 5′-TGCATTGGCCAGCAGTCTT-3′	*(−)
+	Zfp238		5′-CGTCCTGTGATGAGAGTGATGT-3′ 5′-TGGCTCATACTGTACCCTGTTG-3′	*(+)
+	Zfp28	mKr5	5′-TGTACTCAACATCGGAGACTTCACA-3′ 5′-CTCCGTGACATCTACGATGACAG-3′
+	Zfp287		5′-CCCAAGAAGACTGGGAACTG-3′ 5′-ACCATCCACGGCTCTTCTAA-3′
+	Zfp316		5′-CCCGGGTATCTCCCACTACT-3′ 5′-TACACGGCCACATCTTCAAA-3′
+	Zfp358		5′-CATCGGGTCCTGTTCCAG-3′ 5′-GGTCTCCATAAAGGGGCCTAA-3′	*(+)
+	Zfp382	KS1	5′-AAACCTGATATGATCCGCAAGT-3′ 5′-TCACCAAAACTTCCTCCTCTTC-3′
+	Zfp384	CIZ	5′-GATTCCGCATGGTGAAGAGT-3′ 5′-GTTACCGCTACCTGCTTCCA-3′
+	Zfp407		5′-TTCAGGGAGTCCTGCAGTTT-3′ 5′-TGATGAACTGGGAGCCTTCT-3′
+	Zfp422		5′-CGAAGGGAATTCCGAGTTG-3′ 5′-AAATACCCCCTTTGGGCTTT-3′	*(+)
+ +	Zfp426		5′-AAGCCAGACCTGAAATAAACCAGT-3′ 5′ ′-AGAGGCAATCTCCATTCCTCTTAA-3′ 5′-TCCTAGTCTGACCCCCTGGTT-3′ 5′-GCTTAGAGTCACATAGGTCCTCTTCA-3′	**(−); Known mRNA *(−); Novel mRNA
+	Zfp598		5′-CAGGTGGTCTTTGGGAAGAA-3 5′-TGCAAACACCTTTCCATCTG-3′
+	Zfp748	2610014M12Rik	5′-ACACCCGCCACCAGTTC-3′ 5′-TACAGCCCGGGGAGCAT-3′
+	Zfp771	DSC43	5′-GGGCAGCGAGGGGCACTT-3′ 5′-CACTTGGGACTGGCACCTGAATTGTTCC-3′
	Zfpm1	FOG-1	5′-GCCTCACTGAGGGCCTATCC-3′ 5′-TCCTTAGCCAGCAGCTCTCATC-3′
+	Zkscan1		5′-CAGGGCTCTGAGAACAGGAATGG-3′ 5′-CTGTTCCCGTTTCTCTCCAA-3′
+	Zkscan5	Zfp95	5′-CGTGGCTGTGATAGAGAGTATCCA-3′ 5′-GTGGCTCCGGGTGGTACC-3′	**(−)
+	Zkscan21	Zipro	5′-CCGTCACTCTCCTGGAAGAC-3′ 5′-GCCGTTCCTGAAGTTGACAT-3′
+	Zscan22		5′-CCGTTCTGGTGGAGGATATG-3′ 5′-GGTGGCTCCAATTCATCTGT-3′

¹For original references to primer pairs that are not marked by “+”, see Anderson et al., 2002a, 2002b; Hernandez-Hoyos et al., 2003; Yui and Rothenberg, 2004; Anderson et al., 2004; Dionne et al., 2005; Taghon et al., 2005; Taghon et al., 2006; Franco et al., 2006; David-Fung et al., 2006; Tydell et al., 2007; and Taghon et al., 2007.

²Notations about titration curves: all slopes fall within the range of -3.0 to -3.7 ΔCT per tenfold dilution except as indicated:

^*(−) slopes −2.6 to −2.99;

^*(+) slopes −3.71 to −4.1;

^**(−) slopes shallower than −2.6;

^**(+) slopes steeper than 4.1.

Each gene was interrogated in at least two independently sorted sets of samples (two sets = 1 “series”) for verification of gene expression patterns, and key genes were analyzed in four or more independently sorted series of samples. Sorted cell cDNA inputs were routinely normalized by expression of GAPDH and β-actin; in early analyses, expression of Hprt was also assessed (not shown). To test for genomic DNA contamination in cDNA samples, primers IL2-UP, from the upstream non-transcribed region flanking the mouse IL2 gene were used (Adachi and Rothenberg, 2005). In general, patterns of expression defined by changes in level >3 fold were highly reproducible among independent experiments [also see (David-Fung et al., 2006; Rothenberg et al., 2008)]. However, “absolute” levels of expression measured relative to standards could vary over time according to detailed conditions of measurement, with the worst differences (5–10 fold) associated with certain primer sets. Thus all analyses shown are based on simultaneous measurements across a full series of samples, and patterns with little variance cannot be accurately distinguished.

Clustering Algorithms

Correlation values between the expression profiles of all genes in a data set were calculated for each Series, and they were post-processed by sorting on Hierarchical (tree)(Baldi and Brunak, 2001) or Minimum Spanning Tree (MST) algorithms as described in Supplementary Methods. The values from qPCR analyses used were log transformations of averages of the two independent sets of data. Data from Series A and from Series B were analyzed separately.

A method of visual representation of the correlation data was developed as a matrix of the Pearson correlation values between expression patterns of each gene with those of each other gene. In this matrix, each value of entry ‘ij’ (row i, column j) reports the correlation between expression of gene ‘i’ and gene ‘j’, while ‘ji’ would show the correlation between the same two genes across a 45 angle of reflection. For 60 genes processed, e.g., there are 60 entries per gene with a value of 1 for each self-correlation, with these self-correlations marking the diagonal of the matrix. A grid of values relating to colors from blue to red were assigned: blue representing a maximum correlation of +1, yellow a lack of correlation, and red a complete anticorrelation at −1. This representation was applied to the genes in lists ordered either by hierarchically clustered sorting (shown) or by minimum spanning tree sorting (not shown). In the hierarchically sorted coldmap clusters, the red numbers above the dendrogram denoted the chronological order of the links drawn, with low numbers for links between genes or groups of genes with very similar expression patterns and high numbers between groups of genes with more distinct expression patterns. Cluster or branch-cutting was based on these link-orders.

The minimum spanning tree (MST) algorithm comes from the cluster analyses of Xu and Olman (Olman et al., 2003; Xu et al., 2002); where points are ordered so that the sum of their linkages is minimized. This provides a way to view the data as a fully-connected graph with the data points as the vertexes and the distances as the edges. A tree can be obtained from this graph, such that each vertex is visited once and the sum of the edges is minimal. Details of the algorithms used for MST and hierarchical clustering are presented in the Supplementary Material.

To visualize the patterns represented by each cluster as well as their relationships, Self-Organizing Matrix (SOM) analysis (Tamayo et al., 1999) was also applied to the data for each Series separately. Log-transformed expression measurements for each gene across the DN subsets were normalized to a row mean of 0 and then organized into 10–12 clusters by GeneCluster2 software (http://www.broad.mit.edu/cancer/software/genecluster2/), using two-row and three-row organizations. The configurations presented explain >84% of the gene expression variance in each Series.

RESULTS

A global search for transcription factor genes expressed in early T-lineage cells

Most factors currently known to play a role in early T-cell development have been discovered indirectly, either by serendipity or by cis-regulatory analysis of genes expressed by mature T cells. To search more broadly for transcription factors or transcription factor variants that act during early T-cell development, we took a gene discovery approach based on de novo cloning.

As a source enriched for cDNAs expressed during the choice of a T-lineage fate, we used a macroarray of two cDNA libraries prepared from immature thymic cells whose development beyond the point of commitment was blocked by a “severe combined immune deficiency” (scid) mutation in the Prkdc gene (Anderson, 1999). These populations are ~10² fold enriched, relative to normal thymus, for immature T-lineage cells in stages between entry into the thymus and commitment (Fig. 1a). This master arrayed library has been a valuable source of stage-specific cDNA clones for previous studies (Anderson et al., 1999; Chen et al., 2001; Tydell et al., 2007). To recover candidate transcription factor coding sequences, we screened filters spotted with the arrayed library iteratively by hybridization with specific probes for diverse transcription factor domains. Probes were made to detect canonical DNA binding domain types or protein-protein interaction domain types (Table 1), as described in Materials and Methods. Clones significantly reactive with these probes were recovered from the arrayed archive and sequenced. Sequences predicted to encode regulatory factors, chromatin modifying proteins, and potentially relevant signaling molecules were submitted to GenBank as summarized in Table 2. Diverse splice isoforms of known genes (e.g. Tfdp1, Rbak, and Zfp426) and many genes predicted to encode transcriptional repressors (KRAB and SCAN domain Krüppel-type zinc finger proteins) were identified from the SCID thymus cDNA library.

Figure 1.

Landmarks for early T-cell lineage development. (A) Diagram of stages of early T-cell development. (B) Expression of T-cell differentiation genes at successive stages of T-cell development. Samples are from “Series B”, except that Il2rb and Il7ra are shown for samples from Series A. For primer sequences, see Table 3 and (Taghon et al., 2007; Franco et al., 2006). Gene expression levels are expressed in units relative to β-actin measured in the same samples.

Expression of the newly identified transcripts was monitored in purified immature thymocytes from wildtype mice by qPCR, together with expression of known T-lineage differentiation genes and previously identified T-lineage transcription factors, as described in the following sections.

Stages of developmental activation of T-cell genes

Cells in the first stages of T-cell development that enter the thymus are diagrammed in Fig. 1A. These stages are defined as “double negative” (DN), as they lack mature T-cell markers CD4 and CD8. In order of maturity, these stages are termed DN1 (or early T-cell precursor, ETP), DN2, DN3a, DN3b, and DN4. Progression through these stages is profoundly dependent on Notch/Delta signaling. Yet the maturation process is independent of the T-cell receptor (TCR) until after the DN3a stage. At that point the cells arrest at a developmental checkpoint and cannot proceed further until successful rearrangement of TCR gene segments occurs. This conditional arrest point is known as the “β-selection checkpoint”. During the initial stages of intrathymic development, T-cell precursors retain certain alternative developmental potentials. Precursors may develop along natural-killer and/or dendritic cell lineage pathways. However, by the DN3a stage, the cells are committed to some form of T-cell fate.

Because T-lineage differentiation occurs continuously in mammals throughout life beginning at mid-gestation, the normal thymus is in a dynamic steady state in vivo. To obtain a virtual timecourse of early T-cell development for gene expression analysis, we sorted different subsets of immature mouse thymocytes from postnatal, weanling mice, defining developmental stages based on expression of the growth factor receptor c-Kit and markers CD44, CD25 (Ceredig and Rolink, 2002), and where appropriate, heat-stable antigen (CD24) and CD27 (Taghon et al., 2006). Four independent series of samples were prepared from normal, young adult mice. DN1, DN2, total DN3, and DN4 cells were collected in each of the first two sets (“sets 1 & 2”, Series A), whereas in the second two sets (“sets 4 & 5”, Series B) the DN3 cells were further subdivided into DN3a (pre-selection) and DN3b (the first stage past β-selection).

The timing of major positive regulatory changes that confer T-lineage identity was revealed by expression of known T-cell differentiation genes, such as those encoding TCR/pre-TCR complex components CD3ε and CD3γ, kinases Lck and ZAP70, the specific linker LAT, and the surrogate TCR chain Ptcra, as shown in Fig. 1B. All these genes undergo substantial increases in expression from DN1 to DN2 to DN3 or DN3a stage, with only ZAP70 expression continuing to increase further beyond the DN3a stage. The most strictly T-lineage-specific of these genes, CD3ε, CD3γ, and Ptcra, show a particularly steep upregulation between the DN1 and DN2 stages. In corresponding subsets of fetal thymocytes, the major increase in expression of each T-lineage gene occurred similarly from the DN1 to the DN3 stage. One minor difference was that the DN1 populations in those rapidly-developing cohorts appeared to show slightly higher T-lineage gene expression than adult DN1 stage cells (David-Fung et al., 2006)(M.A. Yui, data not shown). Because Notch signaling is critical for T-cell specification, the samples were also assayed for expression of the Notch target gene Deltex1 (Dtx1). Deltex1 is clearly upregulated in DN1 stage cells as compared to prethymic precursors (Tydell et al., 2007) and is further upregulated during the DN2 to DN3 transition. Its pattern of upregulation, however, fails to parallel the T-lineage-specific genes in their initial increase from DN1 to DN2 (Fig. 1B). Thus additional regulatory inputs besides Notch signaling are likely to be needed to set the timing for upregulation of T-lineage genes.

Definition of the regulatory gene set

The regulatory factor set that we assembled for expression analysis included the new candidate genes and about 20 additional transcription factor genes already implicated in T-cell specification by direct genetic or biochemical evidence (Rothenberg et al., 2008; Anderson, 2006; Murre, 2005; Rothenberg and Taghon, 2005; Staal et al., 2001). The total set of genes assayed is listed in Table 3. Genes encoding factors with a close structural relationship to known T-cell transcription factors were included in the set as well, because they might contribute partially redundant or competitive activities.

Many of the newly identified candidates encoded zinc finger factors, often for which little annotation is available. These included both Krüppel-type C2H2 zinc finger factors, likely to bind DNA directly, and RING finger factors, likely to mediate ubiquitin E3 ligase functions and/or other protein-protein interaction functions. Several Krüppel-type zinc finger factors were predicted to be obligate repressors, based on the presence of KRAB and/or SCAN domains in their sequences (Collins et al., 2001), e.g. Rbak, Zfp1, Zfp28, Zfp109, Zfp110, Zfp287, Zfp316, Zfp382, Zfp426, Zkscan1, Zkscan5, Zscan21, and Zscan22. Four transcripts encoded Zn finger factors with BTB domains, i.e. Lrf (Zbtb7a), Zbtb20, Zbtb48, and Zfp238.

Other genes were expected to encode transcriptional modifiers that work without direct DNA binding (prothymosin α = Ptma), or contribute to chromatin remodeling (Jmjd3, Myst3, Mta2, Mta3, NCoR1, NCoR2, SBF1) either to activate or to repress gene expression. Some genes encoded members of known transcription factor families that had not individually been implicated in T-cell development before, such as the POU-homeodomain factor Pou6f1 (Brn-5) and the Ikaros relative Eos (Ikzf4). Transcripts encoding novel isoforms of Rbak, Zfp426, and the winged-helix transcription factor TfDp1, were also identified and later specifically assayed, although these did not reveal a different pattern from the expression of the canonical forms.

These novel regulatory candidates were monitored in parallel with the known essential factors (Rothenberg & Taghon 2005) E2A (Tcfe2a) and its partially redundant relative HEB (Tcf12), GATA-3 (Gata3), Gfi-1, TCF-1 (Tcf7), Ikaros (Ikzf1), Myb, Runx1, and PU.1 (Sfpi1), and their relatives GATA-2, Gfi1b, LEF-1, Mybl2, Runx2, Runx3, and Spi-B. We monitored Ets1 and Ets2, which may be implicated in TCR gene expression, and other Ets family members associated with stem-cell, T-cell, or NK cell function. Also analyzed were other members of the Ikaros family and members of the helix-loop-helix inhibitor (Id) family. As landmarks for differentiation in the same samples, we assayed genes encoding the T-cell receptor signaling component CD3ε; the Notch target gene Deltex1; and the growth factor receptors IL-7Rα (CD127), IL-2Rβ (CD122), and GM-CSF-Rα (Csf2ra) to track changes in the responsiveness of the cells to different environmental signals (Fig. 1B and data not shown).. In all, 80 different candidate regulatory genes were assayed in sets 1 & 2 (series A), and 56 candidate regulatory genes were assayed in sets 4 & 5 (series B). In a review (David-Fung et al., 2006) we previously described a subset of the data from series B, for 19 genes, which is also included in this analysis for context.

The great majority of the regulatory genes were detectably expressed in at least one developmental stage at levels that were calculated to be above 10⁻² times β-actin copy number. For cells with the overall RNA content of immature thymocytes (~1 pg/cell), this is consistent with a lower boundary of about 3 × 10⁻³ times β-actin level as the threshold for expression of at least one copy per cell in the population. The comparability of series A and series B was checked by assaying 13 transcription factor genes with important functions or well-marked, distinctive expression patterns in both series: these genes encoded Ets1, Ets2, FOG-1 (Zfpm1), GATA-3 (Gata3), HEB (Tcf12), Id2, LEF-1, Pou6f1, PU.1 (Sfpi1), Runx1, Runx3, SpiB, and TCF-1 (Tcf7). In general, there was excellent agreement between the overall patterns of expression measured across the developmental stages. Also, for more than half of these genes there was also excellent agreement with the absolute measured expression levels (within a factor of 2 or less).

Expression of landmark transcription factor genes and new candidates in normal early T-cell development

Figures 2–4 present analyses of the expression patterns of the regulatory genes we monitored in the immature DN T-lineage cells of series A and series B. In Fig. 2, actual qPCR results for the genes are shown with genes in alphabetical order. Data from Series A samples are joined by solid curves, data from Series B samples by broken-line curves. All the expression levels measured in Fig. 2 are presented in units relative to β-actin in the same samples. Several methods were used to group these regulatory genes based on their patterns of expression during T-cell specification. Fig 3 shows the correlations between expression patterns of different genes within series A and series B. Here, genes are listed in an order determined by hierarchical clustering (shown left of y axis), and the plot uses a color scale (“cold map”) to show pairwise correlations between expression patterns of each gene (y axis) with those of every other gene (x axis). Gene expression patterns were also organized by the Minimum Spanning Tree method of Xu et al. (Xu et al., 2002), which discerned similar groups (G. Buzi, data not shown). Finally, Figs. 4A, B show representative self-organizing matrix (SOM) clusterings of the normalized expression patterns for genes analyzed in each series (membership of clusters listed in Supplementary Table 1). The boundaries between clusters are somewhat flexible, but Fig. 4C–F show actual expression levels of genes that typify common patterns or features of regulatory factor expression: “rising”, “declining”, “DN3 peak”, and “legacy and loss”.

Figure 2.

Regulatory gene expression patterns through early T cell development. Gene expression levels are shown for the putative regulatory genes in this study, arranged alphabetically in order of gene names. Gene expression levels are measured by qPCR using primers shown in Table 3, and the geometric means of values determined from the two independent sets within each series are plotted on a log₁₀ scale relative to the expression of β-actin in the same samples (0 = log₁₀1 = level of β-actin). Data from both “Series A” and “Series B” are combined in these graphs, with Series A measurements joined by continuous line spline curves and Series B measurements joined by broken-line spline curves. One group of Series B measurements that were determined separately from all the others is indicated by dotted curves (Mitf, Sox4, Bcl11a, Cebpa, and one set of Zfpm1). Where the same genes were analyzed in both series, both sets of values are shown. In some cases (SpiB-1, SpiB-2; LRF, LRF*), the same genes are assayed with different primer sets. LRF* = primers from Maeda et al. (Maeda et al., 2007). GATA-3-r = GATA-3 reassayed on the same Series B samples with the same primers, but by a different investigator > 1 yr after initial Series B measurement (David-Fung et al., 2006); included to show reproducibility. To align the Series A samples (DN1, 2, 3, and 4) with the Series B samples (DN1, 2, 3a, 3b, and 4), values were plotted on an x/y plot in which DN1 was considered “1”; DN2 was considered “2”; DN3a was considered “3.0”; DN3b was considered “3.7”; DN3 (unseparated), which is mostly DN3a, was considered “3.2”; and DN4 was considered to be “4.5”. Thus, the β-selection checkpoint is represented by “3.5”. The same convention is followed in Figs. 4C–F and 6 as well.

Figure 4.

Clusters of genes defined by expression patterns and pattern elements. (A, B) Self-organizing matrices of expression patterns of genes assayed in Series A (A) and Series B (B). Numbers of clusters and 2-row orientations were chosen to account for at least 84% of the overall variance of gene expression in each Series. Cluster number (c) and number of genes in the cluster are indicated at the top of each panel. Names of members of each of the clusters are listed in Supplementary Table 1; examples are shown in (C–F). (C) Genes in this study showing an increase from DN1 to DN3 of >3 fold. Data extracted from Fig. 2. (D) Genes in this study showing a decrease from DN1 to DN3 of >3 fold. (E) Genes with patterns featuring a major peak in expression at DN3 (DN3a) followed by a substantial decline. (F) Examples of genes showing steady “legacy” expression from DN1 to DN3, including all those with substantial decreases at β-selection. Note that distinctions between the groups in (C)–(F) are not clearcut. Certain genes, e.g. SpiB and Id2, have a “legacy” element of expression but may appear more appropriate for different categories when different primers are used (SpiB-1 vs. SpiB-2, DN3 peak) or in different sample series (Id2, declining). In some of these ambiguous cases the same gene data are shown in two panels to illustrate different pattern features, e.g. Id2, Zfp110 and Zfp287). Additional evidence for these pattern groupings comes from other independent measurements (Rothenberg et al., 2008; Taghon et al., 2006; Yui and Rothenberg, 2004).

Figure 3.

Hierarchical clustering of putative regulatory genes based on pairwise correlations. (A) Genes tracked in Series A. (B) Genes tracked in Series B. “Cold maps” are shown to depict the correlation among gene expression patterns within each series, with blue representing full correlation and red representing full anticorrelation as described in Materials and Methods. All genes analyzed in each series are listed in the same order along both the y axis and the x axis of each plot, with the origin at the upper left, so that the diagonal represents the perfect correlations of each gene’s expression pattern with itself. The order of the genes on each axis is set by hierarchical clustering of expression patterns as shown at the left of each matrix, with red numbers indicating the closeness of relationship (low numbers: close; high numbers: remote; see Supplementary Methods). Note that in this hierarchical clustering order, the clusters drawn should be considered to have rotational symmetry. Blocks of color help to visualize genes that are regulated in parallel (blue blocks) or in significantly opposing ways (orange/red blocks) through early T-cell development. Three major groups of genes with decreasing, sustained (“legacy”), and increasing expression are identified with lines over particular blocks in Panel A. Series B was used in a targeted way to assay many genes known from sources other than this gene discovery study, and so panel B is more biased toward genes with diverse, highly inflected expression patterns.

Genes with minimally changing RNA expression: legacy genes

Each clustering method showed that the majority of genes we monitored were expressed in a sustained, minimally changing (<3 fold) pattern from DN1 to DN3 stage. This inter-correlated group is the most dominant feature shown in Fig. 3 (especially 3A), and encompasses not only the genes with “legacy” patterns shown in Fig. 4F but also many genes not shown in Fig. 4 and genes with very low amplitude DN3 peaks in Fig. 4E. Although many of these stably expressed genes are poorly characterized, it must be emphasized that this group also includes many of the factors known to be essential for T-lineage specification from genetic evidence. It includes Myb, MLL1, Ikaros (Ikzf1), Gfi-1, GABPα, Stat5b, Oct1 (Pou2f1), and TOX, all of which have critical roles in T-cell development and/or in T-cell differentiation gene expression [rev. in (Rothenberg et al., 2008; Rothenberg and Taghon, 2005)(Aliahmad and Kaye, 2008; Yao et al., 2006; Ernst et al., 2004; Xue et al., 2004; Wilkinson et al., 2002; Chen and Kuo, 2001; Ullman et al., 1990)]. The stably expressed group contains at least as many known contributors to T-cell development as the group of genes that are sharply upregulated (next section). Furthermore, in terms of the provision of regulatory inputs for other genes in the T-lineage program, genes such as FOG-1 (Zfpm1), Gse1, Yy1, and Zfp598 contribute stably as well, which show slightly different expression trends from most “legacy genes” (Supplementary Table 1) but with very small changes in amplitude. The prevalence of regulatory or candidate regulatory genes showing stable expression in early T-cell development contrasted remarkably with the massive changes in T-lineage gene expression and developmental potential that occur across this interval (cf. Fig. 1B).

Upregulated genes

A surprisingly small group of genes showed definite upregulation (Fig. 4C) during DN1 to DN3 progression. Most of the regulatory genes in this group were previously known to be involved in T-cell specification, i.e., Gata3, Ets1 and Ets2, Tcf12, Tcf7, Lef1, Satb1, Ikzf3 (Aiolos), and Id3. Of the less-studied factors added to this group, Cutl1, Hmga1, Jmjd3, Mlf2, Mybl2, Pou6f1, Ptma, and Zfp382 also appeared to increase in expression with the onset of lineage-specific gene expression. However, most of these upregulated factors increased relatively little across the DN1 to DN3 progression, with gross increases of <10-fold. Some of them peaked at DN3 and decreased again after β-selection (DN3b and DN4), a pattern feature discussed further below, for an even weaker net increase.

Within the “rising” group, the timing of upregulation of different sets of regulatory genes defined at least three different positive regulatory events. First, HEBalt (a promoter-use variant of Tcf12 (Wang et al., 2006)) and Bcl11b underwent the most dramatic portion of their upregulation at the DN1 to DN2 transition, as described elsewhere (Tydell et al., 2007). At the next transition, from DN2 to DN3, Ets2, Lef1, and Pou6f1 were sharply upregulated, apparently in parallel. Then, from DN3 to DN4 stage, Ikzf3 and Tbx21 were upregulated. Id3 appeared unique in its induction in early DN3 stage (DN3a) to a peak in newly-selected DN3 cells (DN3b), followed by a striking decline in the DN4 stage. The distinctive pattern seen for Id3 could reflect its preferential use in the TCRγδ lineage (Taghon et al., 2006). These results, in accord with some previously reported data (David-Fung et al., 2006; Rothenberg et al., 2008; Taghon et al., 2006; Tabrizifard et al., 2004; Morgan et al., 1997), emphasize that T-lineage specification occurs in an ordered cascade.

Genes downregulated during specification

A prominent feature of the DN1 to DN3 progression was the downregulation of hematopoietic transcription factor genes Gfi1b, Gata2, Bcl11a, and of genes encoding several members of the Ets, Helix-loop-helix, C/EBP, and Runx transcription factor families. Many of these downregulated genes (Sfpi1 (PU.1), Tal1 (SCL), Id2, Cebpa, Bcl11a) are functionally implicated in non-T cell hematopoietic differentiation programs (Herblot et al., 2000; Ikawa et al., 2001; Anderson et al., 2002b; Liu et al., 2003; Lefebvre et al., 2005; Laiosa et al., 2006). Another “declining” gene, Erg, was previously identified (Anderson, 2006; Anderson et al., 1999) and has recently been determined to be crucial for stem-cell maintenance (Loughran et al., 2008). Genes encoding many factors not previously associated with early T-cell precursors also fell into this group, e.g. Aff3, Elk3, Etv6, Klf2, Lima1, Rnf149, Tsc22d1, and Zfp316 (possibly also Thrap3). Downregulation of these factors again occurred in distinct stages. Some transcripts diminished mostly between the DN1 and DN2 stage, including Tsc22d1, Lima1, Gata2, Cebpa and Cebpb, while others fell most abruptly only after the DN2 stage (e.g., Tal1, Sfpi1, Gfi1b, Runx2 and Runx3, and Elk3). The most dramatic shutoffs of gene expression were seen for PU.1 (Sfpi1) and SCL (Tal1), which fell at least two orders of magnitude by the DN3 stage.

Multigene downregulation after the DN3 stage

The DN3 stage is when T-lineage commitment first appears to be complete, when proliferation stops, and when survival first becomes dependent on TCR signaling (β-selection). We expected most discontinuities to occur between DN2 and DN3, with the T-cell identity stabilizing thereafter. However, a more commonly observed pattern element was a rapid decline in expression after the DN3 stage. This was seen both for genes that had been stably expressed from the DN1 stage (Fig. 4F) and for some genes that were newly upregulated at the DN2 to DN3 transition (Fig. 4E). Downregulation ranged from a slight decrease for some genes to virtual extinction of others. The Notch-induced gene Deltex1 and the PU.1 relative SpiB showed abrupt, transient peaks of expression at the DN3/3a stage followed by a sharp decline (Fig. 4E). Other genes were downregulated sharply after more gradually reaching the high point in DN3/DN3a cells. These included Runx1, Zfp287, and the Ikaros family member Ikzf4 (Eos), as well as Hes1 and HEBalt, which were also reported previously (Wang et al., 2006; Tan et al., 2005; Yui and Rothenberg, 2004). Even among genes that were also downregulated between the DN1 and DN3 stages, Zbtb20, Klf2, Runx2, and even Id2 all showed their steepest downregulation after β-selection. The shutoff of expression after the DN3a stage was the most conspicuous feature of the expression pattern for the Ets-family gene Erg. In the cases of SpiB, HEBalt, and Erg, at least, this downregulation appears to be permanent (see below) (Rothenberg et al., 2008; Anderson et al., 1999). These results suggest that although developmental commitment appears to occur at the DN3 stage before selection, many “legacy” regulatory genes are shut off only afterwards, apparently triggered by β- or γδ-selection.

Genes specifically expressed in thymic natural killer cells

One regulatory gene, Tbx21 (T-bet), showed an apparently anomalous pattern of expression in these analyses. Although it was recovered from our macroarrayed SCID thymus library in multiple clones, qPCR showed this gene to be expressed at very low levels in the DN1-DN3 T-cell precursor stages, with most of its expression beginning only as the cells progressed to DN4. Since there is no β-selection in SCID thymocytes, this appears inconsistent with the apparent high prevalence of Tbx21 transcripts in the SCID thymus cDNA library. Fig. 5 shows that this is because Tbx21 is actually expressed selectively in thymic natural killer (NK) cells, a population that is highly enriched in SCID thymocytes (Telfer and Rothenberg, 2001; Anderson et al., 1999; Wang et al., 1998). As in normal DN thymocytes, the immunodeficient cells showed increasing Hes1 expression in DN2 and DN3 cells, a sharp rise in Ets2 expression in the DN3 stage. As expected (Anderson et al., 1999), PU.1 (Sfpi1) was expressed highly in a Rag-deficient population that contains the earliest T-cell precursors (Sca-1⁺ Thy-1^low), decreasing in DN2 and DN3 cells and in differentiating NK cells. By contrast, Tbx21 expression was highly enriched in the pre-NK population, a subset which is also marked by its very high expression of IL2rb (Fig. 5). High expression of Tbx21 was also confirmed in mature TCRγδ cells (see below), in agreement with others (Yin et al., 2002).

Figure 5.

Gene expression in thymocyte subpopulations with a genetic block to β-selection: early T vs. NK lineage gene expression. Realtime qPCR measurements are shown for subpopulations of thymocytes from B6. Rag2^−/− mice, sorted as described previously (David-Fung et al., 2006; Anderson et al., 2002b; Wang et al., 1998). True DN1 cells are very rare in these thymus populations, but the Sca-1⁺ Thy-1^low subset contains T-lineage precursor activity while pre-NK cells (Sca-1⁻ Thy-1^+/− CD24⁻) are marked by strong perforin expression (David-Fung et al., 2006). Note distinctively high Tbx21 and Il2rb expression in the pre-NK cells.

Discordant or compensatory expression within transcription factor families

Ets, Runx, Helix-loop-helix, and Ikaros family transcription factors are all implicated in stage-specific T-cell development and gene regulation. Our results, however, showed that these families are continuously well-represented at all stages in early T-cell development. It is the balance among family members that undergoes stage-specific developmental change. Fig. 6 (extracted from the data in Fig. 2) shows detailed patterns of expression for transcription factors of some of these families. Some of the strongest changes in expression occur, in opposite directions, among different members of the same structural families. Most dramatic are the expression patterns in the Ets factor family. While PU.1 (Sfpi1) plunges, SpiB spikes and then falls; and while Elk3, Etv6 (TEL), and Erg decrease, Ets1 and Ets2 increase (Fig. 6A). Concomitantly, as Runx2 and Runx3 decrease, Runx1 increases (Fig. 6B). Ikaros family members Ikaros (Ikzf1) and Helios (Ikzf2) are expressed fairly stably, but Aiolos (Ikzf3) dips and then rises while Eos (Ikzf4) peaks and falls (Fig. 6C). Note also that even though Lef1 and Pou6f1 are among the genes most sharply induced at the DN2 to DN3 transition, their respective relatives Tcf7 and Pou2f1 (Oct1) are also being expressed in a more sustained pattern, at apparently higher levels (cf. Fig. 2). The bHLH transcription factor families including SCL and the E proteins are also highly dynamic, as are their inhibitors (Wang et al., 2006)(cf. Fig. 2).

Figure 6.

Differential regulation of different members of the same transcription factor families. Gene expression data, extracted from Fig. 2, are shown for: (A) Ets, (B) Runx, and (C) Ikaros family members, and (D) for genes predicted to encode KRAB and SCAN domain-containing zinc finger factors.

As many “new” KRAB and SCAN domain zinc finger coding genes were identified in our screen, and are possible candidates for negative regulatory components of the commitment mechanism, it was also of interest to determine how these predicted repressors might be regulated. As shown in Fig. 6D), these genes also showed a mixture of expression patterns: many were expressed stably and at relatively high levels, while others, including the SCAN domain factor Gfi1b, were downregulated at the DN2-DN3 or later transitions. None of those identified were substantially upregulated during specification, however.

Similar or identical target sequences are thought to be bound by multiple members of the Ets, bHLH, Ikaros, Tcf/Lef, and Runx families. If nuclear concentrations of these transcription factor proteins follow the patterns of their RNAs, then these data imply that Ets, Runx, bHLH, and Ikaros binding sites in target genes may be continuously occupied throughout early T-cell development, but by shifting combinations of factors within these families.

Lineage specificity and kinetics of induction in vitro

The evidence presented so far suggests that stage-specific positive and negative regulatory changes during the DN1 to DN3 progression are superimposed on a rich background of gene expression continuing from the earliest stage through commitment. In this context, we revisited the question of which regulatory genes in our study might be T lineage-specific. Not only genes upregulated during commitment but also some of the genes with seemingly stable expression could be candidates. Most cells in the earliest stage of adult mouse T-cell development, the DN1 stage, have already been acted upon by the thymic microenvironment and are already distinct from multilineage prethymic precursors, e.g. with respect to expression of Dtx1, Gata3, Hes1, and Tcf7 (Heinzel et al., 2007; Sambandam et al., 2005; Tan et al., 2005; Tydell et al., 2007). Thus we investigated whether the stable expression of regulatory genes from DN1 to DN3 stage was actually continuing from prethymic precursors or was an immediate response to thymic signals. For this, we made use of an in vitro differentiation system in which a cohort of prethymic precursors can be traced over time as they enter the T-cell pathway (Schmitt et al., 2004b; Schmitt and Zuniga-Pflucker, 2006; Taghon et al., 2005). The T-cell program is triggered in hematopoietic progenitor cells by co-cultivation with stromal cells that have been forced to express a Delta-type Notch ligand. In the absence of the Notch ligand, or at reduced Notch signaling intensity, these same stromal cells support differentiation into B (and some NK) cells but not T cells (Schmitt et al., 2004a). Thus a comparison of the differentiation of the same precursor cells on stromal cells with and without Delta reveals not only discontinuities from prethymic input cells but also differences between the T and B-cell programs.

In two independent series of time course samples, we tracked genes representative of each expression pattern. The gene expression measurements from one well-characterized OP9 culture time course (Taghon et al., 2005; Tydell et al., 2007) are shown in Fig. 7. Similar results were obtained in a second timecourse of four days overall, and in samples re-sorted from the two timecourses at two or four days of culture (data not shown). B-cell specification was tracked by upregulation of the Pax5 gene. In these samples, the Notch target gene Dtx1 and previously characterized T-cell specific regulatory genes Gata3, Tcf7, and Bcl11b as well as the T lineage-specific differentiation gene Cd3e were upregulated only under T-cell conditions as previously reported (Taghon et al., 2005; Tydell et al., 2007). Those data are included here again as a standard for T-lineage specificity. None of the other genes in this study showed comparably strong T-lineage specificity. A group of other genes did show a weaker but detectable preference for T-cell conditions, with either accelerated or more extensive upregulation. These genes included Ets2, Foxp4, Myb, Rbak, Pou6f1, Satb1, Tcf12, and Tle3 (Fig. 7). In addition, “legacy pattern” representatives Mta3 and Stat5b appeared to be sustained better from stem-cell levels in the T-lineage conditions than in the B-lineage conditions.

Figure 7.

Cell type specificity of gene expression induced in hematopoietic progenitors on OP9 stroma. Gene expression levels are shown in units relative to β-actin at successive two-day time points, starting from multipotent c-Kit⁺ CD27⁺ Lin⁻ hematopoietic precursors from fetal liver. Cells were cultured for up to ten days on OP9-control stroma, which does not permit T-cell development (blue shaded bars), and on OP9-DL1 stroma, which promotes T-cell development (brown to yellow shaded bars). Black bars show gene expression in input cells. Cultures were harvested for RNA preparation every two days. At day 4, cells from the cultures were collected and those retaining a precursor phenotype (c-Kit⁺ CD27⁺ Lin⁻) were repurified before being placed in fresh cultures for the remaining 2–6 days (Taghon et al., 2005). Similar results were obtained in experiments where the cells were not repurified at an intermediate timepoint or were repurified at day 2 instead (data not shown). By 6 d, ~50% of cells in OP9-DL1 culture are at least DN2 stage, and ~30% of cells in OP9-control culture are CD19⁺ (Taghon et al., 2005). Genes transiently induced on either stroma (e.g. Il2rb, Cebpa) reflect abortive emergence of NK and myeloid cells (data not shown), which are then suppressed during sustained culture under OP9-control or OP9-DL1 conditions.

Even so, the great majority of the genes analyzed were not induced preferentially during T-lineage differentiation under T-cell conditions. Although genes such as HEBalt and Pou6f1 are markedly upregulated in vivo at particular DN stages, they also were substantially induced in the B-cell conditions. Of particular interest were the four regulatory genes, Ets1, Ets2, Lef1, and Pou6f1, which are specifically induced during commitment to the T-cell lineage at the DN3 (DN3a) stage and maintained thereafter. If this event represents the definitive activation of T-lineage specific regulators, then these genes should show clear T-lineage specificity. Indeed, Ets2 showed an increasing expression specifically in T-lineage promoting conditions. But as already noted, Pou6f1 was clearly upregulated in B-cell precursors as well, and Ets1 and Lef1 showed no sustained T-lineage preference at all. SpiB, also specifically upregulated during the DN3 stage, was expressed considerably better in the B lineage cells.

These results are consistent with a separate assessment of the T-lineage specificity of >70 of these genes, in which we compared expression in adult DN pro-T cells with expression in freshly isolated bone marrow myeloid cells and pro-B cells (Table 4B, columns 1–3). This survey also included Cd3e, as a marker for T lineage specificity, the B-lineage specification gene Pax5, and Bcl6, which encodes a factor crucial for mature B-cells and is reported to provide physiological restraint on the expression of GATA-3 in peripheral T cells (Kusam et al., 2003). The results confirmed that the overwhelming majority of the transcription factor genes assessed were not T lineage-specific in expression. Of the genes analyzed in this series, only Gata3, Tcf7, Pou6f1, Ets2, and TOX (not monitored in the OP9 culture samples) showed an early T-lineage preference, i.e. with T-cell levels >3x greater than their levels in myeloid and pro-B cells. SATB1 also nearly met this criterion (Bcl11b was not assayed in these samples). Tbx21 was expressed more strongly in thymocytes than in myeloid or pro-B cells, but on closer examination, it was strongly expressed only in TCRγδ cells and in unsorted mutant thymocyte populations with a strong NK component (sorted wildtype DN cell value <0.15 of pooled “pan-DN” reference value). In this comparison as well as in Fig. 7, Table 4 confirms that the levels of Ets1 and Lef1 were no higher in pro-T cells than in pro-B cells, in agreement with their known roles in B cell development (Wang et al., 2005; Reya et al., 2000).

TABLE 4.

Expression patterns beyond the pro-T cell stages. (A) Definition of pattern types for genes analyzed in the study. Genes were classified according to their patterns of expression in the DN1-DN4 stages, based on their expression in cluster analyses such as those in Fig. 4, with adjustment “by hand” to take into account whether the dynamic ranges of their profiles exceeded the range of inter-experiment error, or whether they were essentially flat in expression across these stages. Patterns are described at the foot of Table 4A. Dynamic range indications are as follows: 10 = greater than tenfold range from peak to trough in DN stages; 3 = between threefold and tenfold range; 1 = less than threefold range. (B) Expression of genes in non-T developmental contexts and in stages after β- or γδ-selection. Gene expression levels were assayed in 2 samples each of the following: bone marrow derived pre-myeloid cells; pro-B cells, as CD19⁺ cells from Rag2^−/− mouse bone marrow; Rag2^−/− total thymocytes; TCRβ^−/− total thymocytes; TCRγδ⁺ cells sorted from TCRβ^−/− thymus; and sorted subsets from wildtype thymocytes as follows: DN (mostly DN3 & DN4); DP (CD4⁺ CD8⁺); CD4 SP (TCR⁺ CD4⁺ CD8⁻); and CD8 SP (TCR⁺ CD4⁻CD8⁺). Data from individual samples were first normalized according to β-actin levels, then the geometric mean of independent samples was calculated. To compare patterns of use for genes that are expressed at very different levels, all values were then normalized to a reference “pan-DN” level for each gene, that is the geometric mean of the values in Rag2^−/−, TCRβ^−/−, and sorted DN wildtype thymocytes. The Table shows the resulting relative expression values. For genes that increase in expression after β-selection, the sorted DN sample value will be greater than 1; for genes that decrease in expression from the DN1 stage through β-selection, the sorted DN sample value will be less than 1. Cells are shaded to help discern different patterns of expression: orange shaded cells show expression that is more than 3x higher than the reference pan-DN value; green shaded cells show expression that is less than 30% of the level of the reference pan-DN value; purple shaded cells show expression that is less than 1/10 of the level of the reference pan-DN value. Gray shaded cells have no data.

Taken together, these results show that the regulatory genes with strongly T-lineage specific expression are a very select subset of pro-T cell transcriptional regulators, and that within our sample they are represented only by Gata3, Tcf7, TOX, and Bcl11b. These are all factors that undergo most of their upregulation before the DN3 stage, i.e. before T lineage commitment. But the notable result is that most of the regulatory factors in our set that are specifically induced during the transition to T lineage commitment appear to be factors with substantial expression in the B-cell or myeloid lineages as well.

Roles in later T-cell development for factors used in pro-T cell specification

The downregulation during β-selection that was observed for many of the new transcription factor genes raised the question of whether factors sharing particular expression profiles during the pro-T cell stages might also be used in parallel patterns in later stages of T-cell development. Therefore, we also surveyed gene expression in a collection of samples representing later, TCR⁺ stages of T-cell development, from CD4⁺ CD8⁺ (double positive, DP) to mature CD4⁺ or CD8⁺ TCRαβ⁺ cells, and TCRγδ⁺ cells (Table 4B). In Table 4A, each qualitative expression pattern as defined in the DN cells (from Figs. 2–4) is coded; then within each pattern, the genes are further classified according to the overall magnitudes of their expression changes from lowest to highest level, to distinguish weak and questionable expression changes from strong ones. As for the non-T cell samples in Table 4B, expression of each gene in TCR⁺ subsets is presented in units relative to its expression in a pool of DN cell samples, and color coded to help discern similar and different overall patterns of cell type specificity. A number of the gene expression patterns shown here have been corroborated in a separate analysis with independent sets of samples (Rothenberg et al., 2008).

While this survey was too small to establish definitive expression levels, Table 4B shows that genes representative of each DN pattern could go on after β-selection to be expressed in diverse combinations in each of these cell types. Not only genes with sharp ascending or descending patterns, but also genes with “legacy” or “flat” expression patterns (Table 4, patterns 5 & 8) showed notably divergent patterns of expression in TCR⁺ stages represented by DP, CD4 SP, CD8 SP, or TCRγδ cells. As expected, progenitor-cell genes such as Sfpi1 that were repressed during commitment were never re-expressed, but also an additional group of genes (e.g. Erg, SpiB, Hes1, FUS, TfDp1, and Hmga1) remained profoundly downregulated in all TCR⁺ cells, even though they were shut off only at β-selection, considerably later than Sfpi1 (PU.1), Lima1, Tal1, and Gata2. Thus, a limited hit-and-run role during the pro-T cell stages is suggested for at least two distinct sets of regulatory genes. Other genes were downregulated selectively in CD4⁺ CD8⁺ DP cells, in all TCRαβ⁺ mature cells, or in CD4⁺ or CD8⁺ cells specifically (Table 4B). Another, unanticipated finding was a group of transcription factors that appear to distinguish sharply between TCRγδ cells and any of the TCRαβ⁺ subsets. Period1, Stat5b, and Zbtb48 appeared to be downregulated in TCRγδ cells as compared to TCRαβ⁺ cells; Dmtf1 and NCoR2 were specifically elevated in TCRγδ thymocytes, along with known TCRγδ-enriched factors Runx3 and Tbx21 (Taghon et al., 2006; Yin et al., 2002). Thus, transcription factors inherited from prethymic precursors may still help to generate T-cell functional diversity after the pro-T cell stages.

DISCUSSION

T cell lineage choice appears to encompass two major regulatory events: an asynchronous upregulation of increasing numbers of T-cell genes from the DN1 to the DN3 stage, and a sharp discontinuity in developmental potential when the cells undergo lineage commitment during transition to the DN3 stage. These two processes are followed by a TCR signal-dependent cascade of selective and developmental changes through the β-selection transition. If T-cell development followed the model of B-cell development, then the onset of T-lineage gene expression should reflect the activity of T-lineage specific transcription factors, and commitment to the T-cell lineage would reflect the advent of the definitive T-cell specific regulatory state. In this study, however, we see that the T-lineage commitment process departs from expectation in several respects. First, the essential regulatory genes that are turned on in a truly T lineage-specific way remain rare and do not seem to be upregulated during commitment. Second, many of the regulatory changes that do occur in early T-cells are to repress a subset of prethymically inherited genes rather than to upregulate new T-cell specific factors, as noted by others (Kawazu et al., 2007). Third, at the DN3 stage when commitment is thought to occur, the regulatory genes turned on are mostly not T-lineage specific. This suggests that the detectable evidence of commitment at this stage may be a readout of a more general cell biological function rather than a cell type identity function as such. Finally, we show that the loss of factors inherited from the prethymic, multipotent progenitor state may only be completed during β-selection. The β-selection (or γδ-selection) process may therefore contribute a larger aspect of the commitment mechanism than has previously been recognized.

A few caveats need to be stated. First, our analysis is not a complete accounting for all transcription factor genes. Although the gene discovery methods we have used here and in related studies (Anderson et al., 1999; Tydell et al., 2007) have expanded the pool of known candidate regulatory molecules (and their splice isoforms) in early pro-T cells, our gene lists should be considered as a complement to those in microarray studies as in work by the groups of Petrie, Staal, Ogawa, and Melchers (Tabrizifard et al., 2004; Dik et al., 2005; Kawazu et al., 2007; Hoffmann et al., 2003). In contrast to this study, most of the recent microarray studies cited have not addressed how different patterns of expression might be related to lineage specificity. Where our data overlap with these published reports, they are in general agreement, but each kind of study includes genes that are not highlighted in the other. A second caveat is that the RNA-centered analyses used here almost certainly miss some levels of regulation that alter protein levels between stages at which RNA levels are constant. Known examples relevant to T-cell development are the cases of E2A (Engel et al., 2001) and GATA-3 (Yamashita et al., 2005; Kusam et al., 2003; Hernández-Hoyos et al., 2003), both of which could be more sharply regulated between certain cellular states than their transcript levels reveal. Nevertheless, even if we miss some changes in active factor levels when RNA expression appears “flat”, when we do see major changes in RNA level, it is very likely that protein activities are also changing.

This regulatory gene expression analysis sheds light both on the cellular states that make up the stages of T-cell development and on the status of the regulatory molecules that may promote transitions between these stages. Considering these regulatory genes simply as indicators, the gene expression patterns shown here increase greatly the number of “dimensions” of data we have to distinguish DN1 from DN2 from DN3 pro-T cells. Instead of relying only on c-Kit, CD44, and CD25 to track normal developmental progress, the distinctively timed downregulation patterns of PU.1 (Sfpi1), SCL (Tal1), Gfi1b, Erg, Zfp316, Runx2 and Runx3, and Elk3 can also be used together with the distinctively timed upregulation patterns of Bcl11b, Ets2, Lef1, Id3, and Pou6f1 to provide a much finer-grained picture of normal or abnormal developmental progression. The downregulation at β-selection that normally affects Hes1, SpiB, Erg, HEBalt, and genes like Zbtb20 and Zfp287 can also be used to diagnose abnormalities of passage through this checkpoint (Yui and Rothenberg, 2004). Consideration of this expanded set of stage markers reveals reproducible variations of normal T cell development, as between the normal adult pathway and the accelerated development in the fetal thymus (David-Fung et al., 2006). In this case, because the “stage markers” actually encode regulatory factors, the differences may eventually shed light on the mechanisms controlling the differing rates of fetal and adult development.

The products of these regulatory genes also need to be considered as “drivers”. Changes in gene expression from stage to stage ultimately arise from changes in transcription factor activity. However, the often-subtle changes in expression of known essential regulators that we report raise important questions about the amount and type of change that is needed to be significant. Transcription factors work combinatorially in order to regulate target genes, and even factors showing minimal change can be functionally important participants in the T-cell developmental process through interaction with different partners. Thus, not everything that is functionally important for activation of a particular target gene needs to change in level. Among the factors known to be critical for T-cell specification (Rothenberg and Taghon, 2005; Anderson, 2006), some like Myb and Gfi1 may increase less than three-to five-fold between the stem cell state and the committed DN3 stage (Tydell et al., 2007; Rothenberg et al., 2008). These precedents suggest that “legacy” factors generally could also be potential participants in T-lineage specification, changing their functions through interaction with different partners (Sieweke and Graf, 1998).

Our results also suggest that what turns on different genes at different stages of the T-cell development process is sometimes not so much the appearance of a new transcription factor DNA-binding activity, but rather, the change in the identity of the dominant transcription factor that can bind to a given DNA regulatory site. As we show here, the overall level of DNA binding activity by Ets, Runx, and Ikaros family factors may be virtually constant from DN1 to DN3 stage. It is mainly the balance among different members of the family that undergoes dramatic developmental change. Clearly some of these families can participate in causing these changes through their own cross-regulatory relationships: Gfi1b repressing Gfi1, Runx3 repressing Runx1, and PU.1 repressing SpiB, Ets1, and Ets2 (Doan et al., 2004; Vassen et al., 2005; Spender et al., 2005; Franco et al., 2006). But for the change among family members to make a difference, it is likely that protein-protein interactions with other transcription factors are affected by the swap of one family member for another, and not just simple affinity for a DNA target site: there are clear precedents for this among Ikaros family factors (Thompson et al., 2007). By extension, each of the members of these families may compete with the others functionally. For example, as the cells progress from DN1 to DN3, downregulation of Runx2, Runx3 and Gfi1b could greatly enhance the impacts of the slight upregulation of Runx1 and Gfi1, on target genes for which these family members are not functionally equivalent.

There are two crucial watersheds in the T-cell specification process that need to be explained in regulatory terms, and our results shed light on what they actually entail. These are, first, the onset of T-lineage gene expression from DN1 to DN2, while the cells are apparently still multipotent, and second, the loss of multipotency as the cells reach the DN3 stage. This work and previous data (Tabrizifard et al., 2004; Dik et al., 2005; Tydell et al., 2007) show that the DN1 to DN2 transition involves only modest increases in expression of two T-lineage specific regulators (GATA-3 and TCF-1) and the dramatic upregulation of a third, Bcl11b. But although our gene discovery strategy has yielded many “new” potential transcriptional regulators in early T-lineage cells, no new candidates are both T-cell specific and significantly upregulated at this time. This conclusion is also supported by published microarray data (Dik et al., 2005; Tabrizifard et al., 2004; Hoffmann et al., 2003). Could GATA-3 or TCF-1 be the main driver of T-cell gene activation at the DN2 stage? This seems questionable because the major upregulation of these factors occurs much earlier, in entry to the DN1 stage (Tydell et al., 2007), which can be multiple rounds of proliferation before the DN2 stage (Petrie and Zuniga-Pflucker, 2007). Notch inputs seem to be needed throughout, and there is little indication from Hes1 or Dtx1 expression that there is any special intensification of Notch signaling at this point [rev. in (Rothenberg et al., 2008)]. Bcl11b is of great interest, but seemingly is not required for development even to the DN3 stage (Inoue et al., 2006). While Gfi1 and E protein genes are somewhat upregulated, they are not T-lineage specific. Therefore, a major timing element in this initial phase of specification may be an increased permissiveness for the activity of T-lineage specific genes through the downregulation of multiple DN1 and prethymic regulatory genes. It may indeed be the downregulation of antagonists that enables stably transcribed factors like E2A to undergo sudden gains in potency, e.g. by post-transcriptional mechanisms, during the DN1 to DN2 transition (Engel et al., 2001).

The results also imply that the commitment process itself has a compound nature that is masked in conventional comparisons between DN2 and DN3 cell potential. Negative regulation of Gfi1b, SCL, PU.1, and other factors should play a role. But the surprise is that the factors specifically upregulated during commitment are almost all shared between the T lineage and other cell types, with little if any preferential expression in T cells as opposed to B cells. Ets2 upregulation and the transient hyperactivation of Notch pathway target genes are the two main positive events that are T lineage-biased; but the Notch activation is not a stable change and the Ets2 activity appears to remain low compared to its pan-lymphoid relative Ets1. Ets2 may have a later specific role in T cell development (Zaldumbide et al., 2002), as our OP9 time course captures only the beginning of its upregulation (M.A. Y., data not shown), but its expression is not actually T-lineage specific, as it is expressed in myeloid cells as well (Ross et al., 1998). Thus the transition from multipotent to committed may not use an exclusively T-lineage specific mechanism. Note that the DN3 stage in adult mouse thymocytes is associated with a proliferative halt (Petrie and Zuniga-Pflucker, 2007), with growth resuming only if β- or γδ-selection is successful. Conceivably the change assayed as “commitment” is primarily a loss of proliferation, before new functions are conferred during the β-selection process. A permanent lockdown of commitment might only be required to preserve lineage fidelity if and when proliferation might resume. Thus it is also notable that our results show another major phase of repression during β-selection, when a large subset of the “legacy” genes, which had been stably expressed through the DN3 stage are finally turned off. In this altered regulatory environment, the proliferation associated with β- and γδ-selection would not only dilute out pre-existing transcription factor proteins but also provide opportunities to remodel chromatin, which would not be available in G₁ arrest in the DN3 stage. Thus it may be only through β- or γδ-selection that commitment can actually be converted into a stable and heritable condition.

In perspective, the induction of Tcf7 and Gata3 and then Bcl11b, all in the earliest steps of T-cell development, remain the most T-lineage specific of all the regulatory features studied here. Their initial upregulation occurs in the stages when cells are still multipotent, and our results show that few if any other T-lineage specific regulators are likely to be added as the cells undergo lineage commitment. What occurs instead appears to be the stripping away of other alternative pathway regulators and neutralization of inhibitory influences, so that a latent identity already instilled in the earliest intrathymic precursors can be stabilized and revealed.