Expression Profiling of Immature Thymocytes Revealed a Novel Homeobox Gene That Regulates Double-Negative Thymocyte Development¹

Intrathymic development of thymocytes from their bone marrow progenitors is a critical process for the generation of mature T cells, through which the immature thymocyte progenitors, as identified by the absence of mature T cell markers (CD4/CD8 double-negative (DN)³ T cells), differentiate into the CD4/CD8 double-positive (DP) cells, and finally produce CD4 or CD8 single-positive T cells. Although accounting for <5% of total thymocytes in mice, the DN thymocytes undergo a dynamic developmental process that is essential for the subsequent expansion into the DP population (1). Among these DN thymocytes, the earliest chronological subset (DN1) is recognized as a CD44^highC-Kit⁺CD25⁻ population (2, 3). In the first wave of cytokine-dependent pre-T cell expansion, the DN1 cells begin to proliferate with concomitant up-regulation of CD25, giving rise to the DN2 population showing the CD44^highC-Kit⁺CD25⁺ phenotype (4). The DN2 thymocytes then start to rearrange their TCR genes and down-regulate the CD44 expression to generate the CD44^lowCD25⁺ DN3 subset (5). In the DN3 stage, thymocytes are subjected to a process called β selection and only those DN3 cells that have productively rearranged their TCRβ gene can survive and transit into CD44^lowCD25⁻ DN4 thymocytes, followed by rapid expansion into the DP population (6, 7, 8).

During these early developmental processes, immature thymocyte progenitors lose their multilineage plasticity and exclusively commit to T cell differentiation. Under appropriate conditions, the DN1 population can generate multilineage hemopoietic components in vitro (9, 10), although there still remains some controversy with regard to their potential to B lineage differentiation (11). The potential of the multilineage commitment is more restricted in the DN2 stage, where cells can still give rise to NK cells and thymic dendritic cells, but no more B cells (10, 12, 13, 14), and after the DN2/DN3 transition and the succeeding β selection, thymocytes mostly lose their potential to multilineage plasticity and totally commit to T cell lineage (10).

Because these processes are thought to take place under tightly controlled gene expression, it is of particular importance to clarify the nature of this gene regulation and the key regulators involved in that regulation. To date, a number of molecules have been identified that regulate these developmental processes (15). Notch1-deficient thymocytes, for example, are not able to produce T cells but differentiate into B cells (16), while pTα, TCRβ, Lck, SLP76, and Lat are shown to be indispensable for β selection and their knockout mice show a severe maturational block at the DN3/DN4 transition (17). Similarly, the DN2/DN3 transition is completely blocked in Runx1-deficinet mice (18, 19) and also in double knockout mice of pTα and common cytokine receptor γ-chain genes (20). In contrast, it is well anticipated that these developmental processes in DN thymocytes should involve regulation of much larger numbers of genes than those related to these known molecules. To understand the molecular mechanisms of DN thymocyte development, it may be also of use to clarify how these developmental processes are regulated in terms of their entire gene expression, to which cell differentiation is ultimately ascribed.

In the current study, we approached this issue by investigating gene expression profiles in discrete subsets of DN thymocytes under development in which DN2, DN3, and DN4 thymocytes were sorted and subjected to expression profiling analysis with high-density oligonucleotide microarrays. Clustering of differentially expressed genes among these DN thymocyte subsets demonstrated that during DN development, regulation of gene expression is predominantly repressive rather than inductive, in which multiple lineage-affiliated genes expressed in immature thymocytes are down-regulated during the course of DN thymocyte development. Functional mapping of clustered genes also revealed a possible involvement of previously uncharacterized functional gene regulations in thymocyte development. Finally, we identified a novel homeobox gene (Duxl), transiently expressed in DN3 thymocytes. We showed that Duxl is induced by Runx1 and regulates DN thymocyte development by promoting the DN2/DN3 transition, while deregulated expression of Duxl resulted in impaired β selection and severely compromised production of DP thymocytes, indicating its critical roles in DN thymocyte production.

All Abs used for cell sorting were purchased from BD Pharmingen. Thymocytes were harvested from 5- to 6-wk-old female C57BL/6 mice. Four independent cell sortings were performed and four mice were sacrificed for each experiment. Before cell sorting, CD4⁺ cells and CD8⁺ cells were depleted using the MACS LD system (Miltenyi Biotec). The remaining fraction was stained with anti-CD44 and anti-CD25 Abs conjugated to FITC or PerCP-Cy5.5, respectively, and also with PE-conjugated Abs to CD4, CD8, CD3, NK1.1, and TCRγδ and sorted using a FACSAria cell sorter (BD Biosciences). DN2, DN3, and DN4 subsets were identified as FITC⁺PE⁻PerCP⁻Cy5.5⁺, FITC⁻PE⁻PerCP⁻Cy5.5⁺, and FITC⁻PE⁻PerCP⁻Cy5.5⁻ populations, respectively. For expression analysis of various hemopoietic lineages, mononuclear cells were separated from a single-cell suspension of bone marrow of 5- to 6-wk-old female C57BL/6 mice by centrifugation on a Histopaque-1083 (Sigma-Aldrich). c-kit⁺ cells were obtained by positive selection for the c-kit Ag with MACS magnetic beads. The remaining fraction was stained with FITC-conjugated Ab to Ter119, PerCP-conjugated Ab to B220, and PE-conjugated Abs to Mac1, and Ter119⁺ fraction, B220⁺ fraction, and Mac1⁺ fraction were sorted. B220+ splenocytes and CD3+ splenocytes were collected by positive selection of splenocytes for the B220 Ag or CD3 Ag with MACS magnetic beads. RNA was extracted from sorted cells using an RNeasy Mini kit (Qiagen) according to the manufacturer’s instruction.

Biotin-labeled cRNA probes were prepared using a Two-Cycle cDNA Synthesis Kit (Affymetrix). Following fragmentation, biotin-labeled cRNA was hybridized to the Mouse Genome 430 2.0 Array (Affymetrix) for 16 h at 45°C as recommended by the manufacturer. Washing was performed using an automated fluidics workstation, and the array was immediately scanned with GeneChip Scanner 3000 7G. Expression data were extracted from image files produced on Affymetrix GeneChip Operating software 1.0 (GCOS). The absolute detection call (present, absent, or marginal) for each probe set was determined on GCOS. Normalization and expression value calculation were performed using a DNA-Chip Analyzer (www.dchip.org) (21). The invariant set normalization method (22) was used to normalize arrays at probe cell level to make them comparable and the model-based method (22) was used for computing expression values. These expression levels were attached with SEs as measurement accuracy, which were subsequently used to compute 90% confidence intervals of fold changes in two group comparisons (22). The lower confidence bounds of fold changes were conservative estimate of the real fold changes. Differentially expressed probe sets were identified as those sets whose mean signals showed >1.5-fold difference between DN2 and DN3 and DN3 and DN4. The probability of false discoveries with this threshold was calculated by random permutations as described previously (23). Raw microarray data can be found at http://www.ncbi.nlm.nih.gov/geo/, GEO accession no. GSE7784.

One thousand five hundred differentially expressed genes that were identified as described above were clustered using a K-means (24) server on Gene Expression Pattern Analysis Suite (http://gepas.bioinfo.cnio.es/) (25, 26). Molecular network of each cluster was analyzed by KeyMolnet software, which was developed by the Institute of Medicinal Molecular Design, Inc. (IMMD) (27). Known molecular data were curated by IMMD and the obtained gene list of each cluster was combined with this software and shown as molecular networks. Significance of a determined pathway to obtained networks was determined. To ascertain whether any molecular pathways determined by IMMD annotate a relation between molecules in the networks at a frequency greater than that would be expected by chance, this software calculates a p value using the hypergeometric distribution as described previously (28).

cDNAs of Duxl and FLAG-tagged Duxl having NotI and XhoI sites on their 5′ and 3′ terminus, respectively, were PCR amplified from template cDNA prepared from total thymic RNA using TaKaRa LA Taq (Takara Bio) with the following sets of primers: 5′-AAAAGCGGCCGCATCGATACCATGGAGCTGAGCTGCAGTACT-3′ (for Duxl sense), 5′-AAAAGCGGCCGCATCGATACCATGGACTACAAGGACGACGATGACAAGATGGAGCTGAGCTGCAGTACT-3′ (for FLAG-tagged Duxl sense), and 5′-AAAACTCGAGCTACGGAGTTTGGTGTGCTT-3′ (for the common antisense for both). Each PCR product was digested with NotI and XhoI and cloned into the NotI-XhoI site located at the 5′ upstream of IRES-GFP or IRES-NGFRt of the pGCDNsam (a gift from Dr. H. Nakauchi, University of Tokyo, Tokyo, Japan) retrovirus vector. Nucleotide sequences of these plasmids were confirmed by resequencing. Retrovirus was prepared by transfecting the PlatE (a gift from Dr. T. Kitamura, University of Tokyo) cell line with each construct.

Total cellular RNA was converted into cDNAs by reverse transcriptase (Superscript III; Invitrogen Life Technologies) with random primers. cDNAs were amplified in triplicate for 40 cycles at 95°C for 15 s and 60°C for 60 s using an Applied Biosystems PRISM 7000 Sequence Detection System according to the manufacturer’s instructions. Predesigned TaqMan primer and probe sets for 1110051B16Rik (Mm00841823_m1; Applied Biosystems) (see Fig. 4), for PU.1 (Mm00488140_m1), and for 18S rRNA (no. 4308329; Applied Biosystems) were used for the assay. PCR amplification of GAPDH, 1110051B16Rik (see Fig. 5 B) and Rag1 was performed using Platinum SYBR Green qPCR SuperMix-UDG with ROX (Invitrogen Life Technologies) and the following primer sequences: GAPDH (forward, 5′-GAATCTACTGGAGTCTTCACC-3′; reverse, 5′-GTCATGAGCCCTTCCACGATGC-3′), 1110051B16Rik (forward, 5′-GGGAAAACTGGCTCAACAA-3′; reverse, 5′-GTGTTCTGTCCTGGGTCTGG-3′), Rag1 (forward, 5′-CTGAAGCTCAGGGTAGACGG-3′; reverse, 5′-CAACCAAGCTGCAGACATTC-3′). Significant PCR fluorescent signals were normalized for each sample to a PCR fluorescent signal obtained using GAPDH or 18S rRNA as control.

OP9-DL1 is a bone marrow stromal cell line that expresses a Notch ligand, Delta-like 1, and supports development of DP thymocytes from FL-derived hemopoietic progenitors (29) and was provided by Dr. J. C. Zúñiga-Pflücker (University of Toronto, Toronto, Canada). In our OP9-DL1 assay, FL cells were harvested from E14.5 embryos and cultured on OP9-DL1 cells in combination with retroviral gene transfer as previously described (19). Briefly, mononuclear cells were separated from FL cells of C57BL/6 mice. In brief, 5 × 10⁴ mononuclear cells were cultured on confluent OP9-DL1 cells in flat-bottom 24-well culture plates with 500 ml of MEM (Invitrogen Life Technologies) supplemented with 20% FCS, penicillin/streptomycin, and 5 ng/ml recombinant human (rh) IL-7 (Techne Laboratories). After 5 or 6 days of culture, 5 × 10⁴ cells were passed onto newly prepared OP9-DL1 cells in the presence of 5 ng/ml rhIL-7, and retrovirus infection was performed using polybrene (final concentration, 8 mg/ml), followed by another 5 or 6 days of culture. In brief, 5 × 10⁴ cells were again passed onto newly prepared OP9-DL1 cells and cultured for another 5 or 6 days, but in rhIL-7-free culture medium.

PCR for TCRβ rearrangements was performed as described elsewhere (30) on DNA isolated from FL cells cultured on OP9-DL1 using the following primers: Dβ2, GTAGGCACCTGTGGGGAAGAAACT; Jβ2, TGAGAGCTGTCTCCTACTATCGATT; and Vβ5.1, GTCCAACAGTTTGATGACTATCAC. After 40 cycles of amplification (10 s at 98°C, 2 min at 68°C), PCR products were separated on a 4% agarose gel.

The vector backbone was RNAi-Ready pSIREN-RetroQ-ZsGreen (BD Clontech). The RNA interference target sequences were GGAGCAGGATAAACCTAGA (sequence 1), GACTGATATTCTAATTGAA (sequence 2), and GTTCCAGACTGATATTCTA (sequence 3). The small hairpin RNA (shRNA) were designed by a shRNA design algorithm, which was developed by Dr. M. Miyagishi (31). Oligonucleotides used for construction were GATCCGGGGTAGGATAAACTTAGAACGTGTGCTGTCCGTTCTAGGTTTATCCTGCTCCTTTTTACGCGTG (oligonucleotide 1, sense), AATTCACGCGTAAAAAGGAGCAGGATAAACCTAGAACGGACAGCACACGTTCTAAGTTTATCCTACCCCG (oligonucleotide 1, antisense), GATCCGATTGATGTTCTAGTTGAAACGTGTGCTGTCCGTTTCAATTAGAATATCAGTCTTTTTACGCGTG (oligonucleotide 2, sense), AATTCACGCGTAAAAAGACTGATATTCTAATTGAAACGGACAGCACACGTTTCAACTAGAACATCAATCG (oligonucleotide 2, antisense), GATCCGTTTCAGATTGATGTTCTAACGTGTGCTGTCCGTTAGAATATCAGTCTGGAACTTTTTACGCGTG (oligonucleotide 3, sense), and AATTCACGCGTAAAAAGTTCCAGACTGATATTCTAACGGACAGCACACGTTAGAACATCAATCTGAAACG (oligonucleotide 3, antisense). Retrovirus was prepared by transfecting the PlatE cell line with the knockdown vectors.

The DN2, DN3, and DN4 populations were FACS sorted from DN thymocytes harvested from four C57BL/6 mice and analyzed by an Affymetrix Mouse Genome 430 2.0 Array for gene expression (Fig. 1^,A). Four independent experiments were performed using 16 mice. After normalizing the array signals using an invariant gene set (22), we extracted differentially expressed probe sets whose signals showed ≥1.5-fold difference between DN2 and DN3 or between DN3 and DN4. With this threshold, 1,901 probe sets (or 1,500 nonredundant genes) were extracted as “differentially expressed” from a total of 27,330 probes (16,131 genes) expressed in either of the three DN subsets, with a false discovery rate of 0.007 as determined by random permutation tests (23). In hierarchical clustering, the four independent array data sets for each subset were correctly clustered into the same clusters, validating the reproducibility across the experiments (Fig. 1^,B). In contrast, the K-means clustering of the 1,500 differentially expressed genes identified six clusters, CL1–CL6, showing discrete temporal expression profiles: genes expressed higher in DN2 and down-regulated in DN3 (CL1), those expressed higher in DN2 and gradually down-regulated in DN3 and DN4 (CL2), those expressed in DN2 and DN3 but down-regulated in DN4 (CL3), those only transiently expressed in DN3 (CL4), those expressed both in DN3 and DN4 (CL5), and those showing low expression in DN2 and DN3 and up-regulated in DN4 (CL6) (Fig. 1 C). The lists of genes in these clusters were presented in supplementary Table S1, a–f.⁴

With regard to the temporal profiles of gene expression in DN thymocytes, our first note is that 1123 (74.9%) of the 1500 differentially expressed genes are initially expressed in DN2 but eventually down-regulated during the course of DN thymocyte development (CL1–3), while only 79 (5.27%) and 89 (5.93%) genes are up-regulated in DN3 and DN4 (CL5 and CL6), respectively (Fig. 1^,C). The other set of genes (CL4) are transiently expressed in DN3 but have low expression in DN2 and DN4. Thus, gene regulation during DN thymocyte development is largely repressive rather than inductive. Our next note was that these down-regulated genes contain a variety of genes whose expression is relatively characteristic to one or more hemopoietic lineages (lineage affiliated), including NF-E2 and thrombin receptor (megakaryocytes), PU.1, lysozyme, and myeloperoxidase (myeloid lineages), Lyn, Syk, Btk, Bcl3, BCAP, and Bach2 (B cells), and 2B4, Fas ligand, and granzyme B (mature T cells) (Fig. 2^,A). This suggested that “promiscuous” expression of multiple lineage-affiliated genes in early thymocyte progenitors and their repression during the course of their commitment to mature T cells could be one of the characteristics of immature thymocytes. To confirm this in more detail, we identified those genes whose expression was thought to characterize a variety of mature hemopoietic lineages and tracked their expression levels during the course of DN thymocyte development. Because no comprehensive gene expression database was available for mice, we did this using a human tissue expression database at the Genomics Institute of the Novartis Research Foundation web site (http://symatlas.gnf.org/SymAtlas/) (32) by translating mouse genes into their human counterparts. A gene is considered to be affiliated with a lineage if its human counterpart shows >10 times higher expression in that lineage than their median expression among 79 different tissues. Among the 1500 differentially expressed genes, the human counterparts were uniquely identified for 744 (49.6%) genes (Fig. 1^,C), of which 133 (17.9%) satisfied the above criteria for being affiliated with one or more hemopoietic cell lineages (Fig. 3^,A), and lists of genes affiliated with respective lineages are presented in supplementary Table S2, a–g.⁴ Among the 133 lineage-affiliated genes, 110 (83%) are expressed in DN2 and down-regulated during the course of the DN thymocyte development (CL1–CL3), accounting for 20% of 548 down-regulated genes assigned to CL1–3 (Fig. 3,B). More genes are rapidly down-regulated during the DN2/DN3 transition (CL1), whereas as indicated from the relatively high CL3 component, down-regulation of genes affiliated with NK cells seems to occur more slowly (Fig. 3,C). Although most of these down-regulated genes are affiliated with lineages other than T cells, many T cell-affiliated genes are also prematurely expressed in immature thymocytes and undergo down-regulation before they are definitely expressed later in thymocyte maturation (Figs. 2,A and 3,C). Only 12 (9%) lineage-affiliated genes were newly up-regulated until the DN4 stages and mostly related to T cells (Fig. 3, B and C). Note that our criteria for lineage relatedness may be too conservative, since many of genes presumed to be specific to mature T cells, such as 2B4, Fas ligand, and granzyme B, were not extracted as T cell-related genes.

The functional features of genes showing discrete expressional time courses are of interest to understand the regulation of DN development. To address this, we statistically explored possible functional links among genes within each cluster by evaluating a probability that a given set of genes on the known functional pathway was grouped together by chance within that cluster, showing a similar temporal expression profile using KeyMolnet software (IMMD) (27). Because of the small numbers of genes within the clusters CL5 and CL6, both clusters were analyzed after being combined together. In this pathway analysis, several sets of functionally related genes or pathways were extracted from each cluster. Table I shows a list of pathways extracted in high significance values (p < 0.01). For examples, genes involved in c-kit signaling (p = 0.00147) and CD44 signaling (p = 0.0086) were extracted from the CL1 cluster, reflecting the fact that c-kit and CD44 are down-regulated in DN3 and also indicating that the pathways extracted from CL1 are expected to be inactivated in DN3. Other pathways that are known to undergo dynamic regulations during DN thymocyte development were also extracted from different clusters, including the Runx1 pathway from CL2 (gradually down-regulated), NFAT and AP-1 pathways from CL3 (down-regulated in DN4), TCRαβ, NF-κB, and Notch pathways from CL4 (transiently activated in DN3), and Ets and TCRαβ pathways from CL5/CL6 (up-regulated in DN3 or DN4). The other pathways that were extracted with high significance values but have not been previously implicated in T cell development include those related to platelet-derived growth factor, γ-aminobutyric acid, and peroxisome proliferator-activated receptor α.

In view of clarifying gene regulations that operate during thymocyte development, of particular interest are genes included in CL4, because they show a unique temporal expression profile of transient induction or up-regulation in DN3 and contain key genes for the thymocyte development or β selection, including pre-TCRα, SpiB, Egr3, and Notch3 (33, 34, 35, 36, 37) (Fig. 2,B). Especially, we were interested in genes that were thought to be involved in transcriptional regulations (Table II). Among these genes, we focused on a gene, 1110051B16Rik, which encodes a putative transcription factor with unknown functions (Table II and Fig. 4,A). It has an open reading frame of 1050 nt and the predicted protein shares structural features and sequence similarities with the families of double homeobox proteins that are thought to have rapidly diverged during evolution (Fig. 4,B) (38). Although currently no definite human ortholog is uniquely identified due to incomplete sequence homology to human sequences, it shows the highest similarity with human DUXA or DUX4 in their amino acids sequence and gene structure with DUXA (Fig. 4, B and C), and thus, was named as Duxl (Dux in lymphoid lineage). In hemopoietic compartments, expression of Duxl is largely restricted to DN3 thymocytes and B220-positive B cells in bone marrow and spleen, implicating their functional roles in both subsets of lymphocytes, although it is also expressed in embryos from mid to late gestation, as well as in other non-hemopoietic adult organs, including brain, prostate, and uterus (Fig. 4 D).

To get an insight into the role of Duxl in thymocyte development, we first examined its expression among the subpopulations of DN3 thymocytes by FACS sorting DN3 cells into two populations according to their expression of intracellular TCRβ (iTCRβ) chains (Fig. 5,A, upper panel) (8, 39). In quantitative PCR (qPCR) analysis, Duxl transcripts were detected in the iTCRβ ⁻ fraction (DN3a) but not in the iTCRβ⁺ fraction (DN3b), indicating that the Duxl expression is largely restricted to the DN3a fraction (Fig. 5 A, lower panel).

To explore the effect of constitutively expressed Duxl on DN thymocyte differentiation, we transduced the Duxl cDNA into mouse FL cells using a bicistronic retrovirus vector with a marker GFP cDNA in the second cistron, and the Duxl-transduced FL cells were then assayed on the OP9-Delta-1 (OP9-DL1) culture system that can mimic intrathymic development of DP thymocytes from the FL hemopoietic progenitors (29). As previously described, the mock-infected FL cells generated substantial numbers of GFP⁺ DP thymocytes after 15 days of culture on OP9-DL1 (Fig. 5,B, upper central panel) (19). In contrast, the number of GFP⁺ DP cells was dramatically reduced in the culture from the Duxl-transduced FL cells (Fig. 5,B, upper right panel). Although the total cell numbers were not significantly different between both cultures, the GFP⁺ cells from the Duxl-transduced culture generated an increased proportion of the CD25⁺CD44^low cells, compared with mock-infected cells (Fig. 5,B, middle panels) that produced a similar proportion of CD25⁺CD44^low cells as GFP ⁻ cells (data not shown). The increased proportion of CD25⁺CD44^low in Duxl-transduced culture was accompanied with a reduced DN2 population. The GFP⁺CD25⁺CD44^low cells in the Duxl-transduced FL cells had a Thy1⁺ CD11c ⁻ Mac1⁻NK1.1⁻B220 ⁻ phenotype. We extracted total RNAs from mock-infected DN2 and DN3 and Duxl-introduced DN2 and DN3 cells cultured on OP9-DL1 and analyzed their expression of Rag1, PU.1, and c-kit using real-time PCR and flow cytometry, respectively. In Duxl-introduced DN2 and DN3 cells, expression of Rag1 was enhanced and expression of PU.1 was reduced, whereas no difference was observed in c-kit expression between mock-infected and Duxl-introduced cells (Fig. 5 C). Therefore, Duxl is supposed to promote the DN2/DN3 transition in some respects, while not in others.

Mock-infected and Duxl-transduced CD25⁺CD44^low cells were sorted and their genomic DNA was analyzed by PCR for TCRβ rearrangement. Interestingly, although the fraction of iTCRβ-positive cells in Duxl-transduced DN cells was substantially reduced compared with that in control DN cells (Fig. 5,B, lower panels), the extent of DJβ and V-DJβ rearrangement in Duxl-transduced CD25⁺CD44^low cells was comparable to those of control cells or rather slightly accelerated compared with control cells (Fig. 5,D), indicating that Duxl actively represses expression of rearranged TCRβ genes. Indeed, expression of surface TCRβ in the mouse thymoma cell line AKR1, which exhibits the DP phenotype (CD4⁺CD8⁺TCR^int), was reduced after Duxl was introduced using retroviral vector (Fig. 5 E). Taken together, these data suggest that the increased CD25⁺CD44^low population in the Duxl-transduced cells show a DN3a-like phenotype and that constitutive expression of Duxl promotes the DN2/DN3 transition but compromises the process of β selection.

To further investigate the role of Duxl in DN2/DN3 transition, we knocked down the expression of Duxl in FL cells using RNA interference, in which FL cells were transfected with a retrovirus that produces shRNA directed against Duxl and examined for their development in the OP9-DL1 culture. Among three different shRNAs (sequences 1–3), only sequence 1 shRNA showed significant RNA interference as confirmed by the reduced cell surface expression of a truncated form of nerve growth factor receptor (NGFRt) in NIH3T3, where the NGFRt was translated from a bicistronic message of the Duxl-IRES-NGFRt (Fig. 6,A). When transduced with sequence 1 shRNA, FL cells showed a reduced DN2/DN3 transition in OP9-DL1 culture as determined by the higher levels of CD44 expression in GFP⁺CD25⁺ cells compared with GFP⁻CD25⁺ cells within the same culture or with GFP⁺CD25⁺ cells in the αLuc (mock)-infected culture (Fig. 6 B).

It was of particular interest to explore a possible functional link between Duxl and Runx1, because we showed that Runx1 is essential for the DN2/DN3 transition (18). Thus, we tested whether Duxl can rescue the phenotype of Runx1 deficiency in developing thymocytes by transducing Duxl into FL cells from Runx1-deficient mice in OP9-DL1 culture. Although mock-transduced Runx1-deficinet FL cells showed the defective DN2/DN3 transition with impaired down-regulation of CD44 (19), Duxl transduction clearly increased the CD44^lowCD25⁺ population (Fig. 7, A and B), although the latter population still did not express intracellular TCRβ chains (Fig. 7,A, lower panels). Thus, Duxl can induce down-regulation of CD44 in a Runx1-independent manner. We further examined whether Duxl is induced by Runx1 or not, where Runx1 was transduced into normal FL cells in OP9-DL1 culture or an AKR1 cell line, and expression of Duxl was measured by quantitative PCR 72 h after the transduction. In both experiments, Runx1-transduced GFP⁺ cells showed a marked increase in Duxl expression (Fig. 7 C). These results indicated that Runx1 promotes the DN2/DN3 transition by, at least in part, regulating expression of Duxl, although it is not clear whether this regulation is direct or indirect.

Several groups have investigated the gene expression profiles of developing thymocytes (40, 41, 42, 43). Despite the differences in analytical methods, panels of genes to be examined, microarray used, thymocyte subpopulation analyzed, or species, the reported expression pattern of some representative genes such as pTα, Nothc3, Egr3, or SpiB were reproduced in our study (15, 40, 42). Furthermore, our study revealed a new aspect that was not described so far and adds to knowledge of the development of thymocytes. Intrathymic development of thymocytes is a dynamic process, during which immature thymocytes progressively commit to mature T cell differentiation accompanied by explosive expansions of their population. As such, we initially expected that this process be driven by induction of a large number of regulatory genes. Unexpectedly, however, our expression profiling clearly showed that gene regulation during early thymocyte development is characterized by mostly repressive activities of transcription, where 90% of the differentially expressed genes were eventually down-regulated. This means that as thymocytes differentiate from their progenitors, they become to use more limited sets of genes at least before differentiating into DP cells.

Among these down-regulated genes, a notable subset is a group of genes that show lineage-promiscuous expression. According to our somewhat arbitrary criteria, these genes account for a substantial proportion (20%) of down-regulated genes in DN2. In other words, immature thymocytes in the DN2 subset simultaneously express genes from different lineage-affiliated programs. Such lineage-promiscuous states were previously implicated for multipotent hemopoietic progenitors based on the observation for limited numbers of lineage-affiliated genes (44, 45, 46). We confirmed this for DN thymocytes by analyzing expression of a large number of genes and also evaluated their temporal changes during the course of their development. Of note is that mature T cell-related genes expressed in DN2 are also transiently repressed during DN2 to DN4 progression. Because our criteria of >10 times higher expression than the median could be too strict, more DN2 genes undergoing down-regulation thereafter could be explained under this framework.

The interpretation of this lineage-promiscuous gene expression in very immature thymocytes is elusive, but it may be speculated that these immature cells are conditioned or primed before their differentiation into particular lineages and that these “primed” states are represented by expression of multiple lineage-affiliated genes. In this scenario, the lineage-promiscuous expression of DN thymocytes could be related to plasticity of these cells. DN2 thymocytes are committed to T lineage to a certain extent but still can give rise to other lineages such as NK cells (10, 12, 13, 14), which might reflect the observation that NK cell-affiliated genes were more slowly down-regulated during DN thymocyte development (Fig. 3 C). Alternatively, lineage-promiscuous gene expression could be explained by the heterogeneous lineage potential at the level of cellular complexity. Because CD44⁺LCD25^low DN1 cells are composed of multiple subsets with distinct differentiation capacity that could be identified with additional c-kit and CD24 staining (11), DN1 cells express a large number of lineage-promiscuous genes down-regulated at the DN1/DN2 transition (40). This may be also the case with DN2 cells we have sorted without c-kit and CD24 staining in this study. Although, the observation that differentiation capacity of DN1 cells with lysozyme expression was similar to that of DN1 cells without lysozyme expression (47) strongly supports the former interpretation, it is necessary to examine the gene expression profiles of each single DN cell, as well as more precise sorting with c-kit and CD24 staining, to know which interpretation is more accurate.

Computational mapping of differentially expressed genes showing similar temporal profiles (CL1–CL6) to known functional pathways seems to be effective to extract the relevant pathways in DN thymocyte development, because it successfully extracted well-characterized pathways in developing thymocytes, such as those related to Runx1, TCRβ, NF-κB, NFAT, and Notch genes. In addition, it also implicated the presence of several previously unknown pathways that might be regulated during thymocyte development. Although further evaluations are required, finding of these potentially relevant pathways will provide valuable clues to the exploring molecular mechanisms of regulation of DN thymocyte development.

During DN thymocyte development, only a minority of genes was newly induced, among which those included in the CL4 cluster were of our particular interest, because they were transiently expressed within a window in DN3 and contained several well-known pathways and genes that are critical for thymocyte development. From a few candidate genes within this cluster that are potentially involved in transcriptional regulations, we identified a novel double homeobox gene, named Duxl, which is transiently expressed in DN3a and could have critical roles in regulation of DN thymocyte development. When constitutively expressed in FL cells in OP9-DL1 culture, Duxl enhances the proportion of CD44^lowCD25⁺ thymocytes that mimic the phenotype of DN3a cells with severely reduced DP cells, while the knockdown of Duxl in FL cells impaired the proper DN2/DN3 transition. Thus, Duxl is thought to play an important role in promoting the DN2/DN3 transition in the development of DN thymocytes.

In early thymocyte development, Duxl has several functional similarities to SpiB, in that it is highly expressed in DN3a as well as B cells, and that normal thymocyte development is impaired when constitutively expressed or knocked down in early thymocytes (33). These observations may implicate a possible functional link between SpiB and Duxl, which should be addressed in further analysis. For example, SpiB also regulates B cell differentiation, implicating that Duxl is also involved in development of B cells.

Of particular interest is the finding that Duxl is induced by Runx1 expression and can partially rescue the Runx1-deficient phenotype with regard to the down-regulation of cell surface CD44, indicating that Duxl is one of the effecter molecules involved in the DN2/DN3 transition that is regulated by Runx1. The detailed molecular mechanism of the DN2/DN3 transition is largely unknown because only limited numbers of mice strains, namely, Runx1-deficient mice (18) and pTα/common cytokine receptor γ-chain double knock-out mice (20), exhibit the maturational block between the DN2 and the DN3 stage. However, considering the fact that during the transition from the DN2 to the DN3 stage, the TCRβ gene is rearranged and αβT cell, γδT cell, and NK cell lineages begin to diverge (10, 48, 49), the DN2/DN3 transition, on which our findings could shed light, is supposed to be a crucial developmental step.

On the other hand, the interpretation of the severely reduced production of DP cells associated with constitutive expression of Duxl in FL cells is complicated in a context of its physiological roles. Since the Duxl-transduced FL cells seem to show a maturational block at the transition between DN3a and DN3b, during which Duxl undergoes down-regulation in vivo, it may be postulated that Duxl being normally down-regulated in this step is important for the β selection and subsequent DP thymocyte production. We may safely conclude that the precise regulation of Duxl expression is essential for DP thymocytes to be normally generated in OP9 culture, but its physiological functions in β selection and DP thymocyte generation are still elusive. To address this, more sophisticated experimental approaches with precisely targeted expression of Duxl in vivo should be required.

In conclusion, through the gene expression profiling of chronologically discrete subsets of DN thymocytes, we demonstrated the predominantly repressive gene regulation during DN thymocyte development along with its implication in lineage-promiscuous gene expression in immature thymocytes. Among the gene cluster showing transient expression in DN3, we identified Duxl, a novel double homeobox gene, that is induced by Runx1 and involved in regulation of DN thymocyte development.

We thank J. C. Zúñiga-Pflücker for OP9-DL1 stromal cells, H. Nakauchi for pGCDNsam retroviral vector, M. Miyagishi for designing shRNA, and T. Kitamura for PlatE packaging cells.

The authors have no financial conflict of interest.