Human UCB samples were obtained from consenting mothers after delivery at local hospitals. Mononuclear cells (MNCs) were isolated by Ficoll-Paque density centrifugation, washed, and frozen in 10% dimethyl sulfoxide and 90% fetal bovine serum (FBS) for later use. Thymi were obtained as surgical tissue discards from infants 2–9 months of age undergoing cardiac surgery at Erasmus MC Rotterdam, after informed consent from the parents or legal guardians. The children did not have immunological abnormalities. Thymocytes were isolated by cutting the thymic lobes into small pieces and squeezing them through a metal mesh and stored at −80°C until further analyses. Informed consents were in accordance with the Institutional Review Board of the Erasmus MC Rotterdam and in accordance with the Declaration of Helsinki.

Frozen MNCs were thawed, washed, and labeled with MicroBeads conjugated to the monoclonal mouse anti-human CD34 antibody according to the manufacturer’s procedure (Miltenyi Biotec). The magnetic separation was performed twice using two MACS Columns (Miltenyi Biotec), consistently reaching a CD34⁺ purity of approximately 95%.

Antibodies (with clone identification) used for flow cytometry: CD1a (HI149), CD3 (UCHT1), CD7 (M-T701), CD10 (HI10a), CD11b (ICRF44), CD13 (WM15), CD14 (MΦP9), CD20 (L27), CD22 (HIB22), CD33 (WM53), CD34 (8G12) (BD Biosciences), CD3 (BW264/56), CD4 (VIT4), CD5 (UCHT2), CD8 (BW135/80), CD19 (LT19), CD33 (AC104.3E3), CD34 (AC136), CD44 (DB105), CD45 (5B1), CD56 (REA196), CD94 (REA113), CD123 (AC145) (Miltenyi Biotec).

OP9-DL1 cocultures were performed according to the original protocol (18). Briefly, CD34⁺ human HSC isolated from UCB was plated into 100 or 145 mm culture dishes (Greiner Bio-One B.V.), which were seeded with a monolayer of OP9-DL1 cells beforehand. The α-minimal essential medium (Life Technologies) used for culture was supplemented with 20% FBS (Integro B.V.) plus 10 U/mL penicillin, 10 µg/mL of streptomycin, and 0.025 µg/mL fungizone (PSF) (Life Technologies). Recombinant human stem cell factor (SCF) (10 ng/mL) (R&D systems), FLT3L (5 ng/mL) (Miltenyi Biotec), and interleukin-7 (IL7) (2 ng/mL) (Miltenyi Biotec) were added at the initiation of coculture and every 2–3 days during transfer of HSC-derived cells onto new OP9-DL1 plated cells. Cocultures were separated by vigorous pipetting and put through a 40-µm filter to reduce stromal cell line aggregates and eliminate contaminating OP9-DL1 cells.

Multi-lineage differentiation experiments were performed with the following modifications: recombinant human SCF (10 ng/mL), FLT3L (5 ng/mL), IL7 (5 ng/mL), and IL15 (10 ng/mL) were added every 3–4 days to half of the medium that was refreshed without disturbing the cells. A continuous 10-day OP9-coculture was initiated with 1,000 OP9 (-DL1, -GFP, or 1:1 mix) cells/well plated in 96-well plates 1 day before adding 4,000/well of the sorted thymocytes.

OP9-GFP and OP9-DL1 cell lines were kindly given to us by Dr. Zúñiga-Pflücker; the identity of the cell lines was confirmed by DNA fingerprinting, and cells were regularly tested for mycoplasma contamination.

In-frame Dβ–Jβ and Vβ–Jβ gene rearrangements were determined using the BIOMED-2 multiplex PCR and visualized using GeneScan (19). Primers for the Jβ1 cluster were hexachloro-6-carboxy-fluorescein (HEX) labeled (blue traces), and primers for the Jβ2 cluster were 6-carboxy-fluorescein (FAM) labeled (green traces).

Throughout the differentiation of HSC toward the T-cell lineage, we sorted T-cell fractions at various time points (days 0, 7, 12, 18, and 27) based on the expression of cell surface markers including CD45, CD34, CD7, CD5, CD1a, CD4, and CD8. Cell sorting purity was >98% and performed using the FACSAria II (BD Biosciences), flow cytometry was performed using the MACS Quant (Miltenyi Biotec), and data analysis was performed using the FlowJo software (Tree Star Inc.). Sorted populations (range: 6.8 × 10⁴–2.68 × 10⁶ cells/population, median: 2.62 × 10⁵ cells) were collected, washed twice in PBS, lysed in buffer RLT (Qiagen) plus 1/100 β-mercaptoethanol, and stored at −80°C. RNA was isolated using RNeasy Micro Kit according to the manufacturer’s protocol (Qiagen).

The quality control, RNA labeling, hybridization, and data extraction were performed at ServiceXS B.V. (Leiden, the Netherlands). RNA concentration was measured using the Nanodrop ND-1000 spectrophotometer (Nanodrop Technologies). The RNA quality and integrity was determined using Lab-on-Chip analysis on the Agilent 2100 Bioanalyzer (Agilent Technologies, Inc.) and/or on the Shimadzu MultiNA RNA analysis chips (Shimadzu Corporation). Biotinylated cRNA was prepared using the Affymetrix 3′ IVT Express Kit (Affymetrix) according to the manufacturer’s specifications with an input of 100 ng total RNA. The quality of the cRNA was assessed using the Shimadzu MultiNA in order to confirm if the average fragment size was according to Affymetrix’ specifications. Per sample, 7.5 µg cRNA of the obtained biotinylated cRNA samples was fragmented and hybridized in a final concentration of 0.0375 µg/µL on the Affymetrix HT HG U133+ PM (Affymetrix). After an automated process of washing and staining by the GeneTitan machine (Affymetrix) using the Affymetrix HWS Kit for GeneTitan (part nr. 901530), absolute values of expression were calculated from the scanned array using the Affymetrix Command Console v3.2 software.

The data discussed in this publication have been deposited in NCBI’s Gene Expression Omnibus (20) and are accessible through GEO Series accession number GSE79379.¹

Reverse-transcriptase (RT) reactions (Promega) were performed to convert mRNA into cDNA. The real-time PCR (qPCR) reactions were performed by Taqman Custom arrays that were pre-designed according to the manufacturer instructions (Applied Biosystems). TaqMan gene expression assay targets were based on selected primers that were pre-loaded into each of the wells of a 384-well Taqman array card. TaqMan Gene Expression Assays consist of a pair of unlabeled PCR primers and a TaqMan probe with a FAM or VIC dye label on the 5′ end and minor groove binder and non-fluorescent quencher on the 3′ end. For the reaction, cDNA samples were diluted and mixed in 1:1 ratio with Taqman Fast Universal PCR Master Mix (Applied Biosystems). Relative levels of gene expression are determined from the fluorescence data generated during PCR using the 7900HT Fast Real-Time PCR System Relative Quantitation software (Applied Biosystems).

Affymetrix gene expression CEL files were processed using Partek^® Genomics Suite^® 6.6. Robust Multi-array Average (21) was applied prior to further analysis. In brief, the intensity levels were quantile normalized after background correction. Probeset-level expression values were summarized by median polish approach and eventually log₂ transformed. Probesets were filtered for high variance (top 5%) and log₂ expression values (>8), which resulted in 2,179 probesets. Using the annotation file of the array, these probesets were then summarized into 1,387 genes by taking the median of all probesets across a gene. Principal component analysis (PCA) and hierarchical clustering were done using Partek^® Genomics Suite^® 6.6. Log₂ expression values were standardized into z-scores prior to the clustering analysis. Pearson dissimilarity and Euclidean distance were used as distance metric. Gene set enrichment analysis (GSEA) (22) was performed using GSEA software downloaded from Broad Institute.² GSEA is broadly used to analyze genome-wide expression profiles from samples belonging to two classes, in order to determine whether an a priori defined set of genes is correlated with the phenotypic class distinction. The method derives its power by focusing on gene sets instead of single genes. In this study, we took an unbiased approach at defining the gene sets using an unsupervised clustering method. For the OP9-DL1 in vitro dataset, the GSEA input gene list consisted of 1,387 genes that were grouped into the 16 gene signatures identified by the clustering analysis. The T-cell populations were denoted as early or late T-cell program according to their gene expression profile similarities during T-cell differentiation. Gene sets with false discovery rate (FDR) q ≤ 0.25 were considered significantly correlated to the class distinction (22). The murine in vivo microarray data was obtained through access to the Immunological Genome Project data, GEO accession code: GSE15907 (23). For analysis of this in vivo dataset, the GSEA input gene list consisted of 547 genes enclosed in the 5 signatures that form the T-cell development gene expression signature. The human in vivo microarray data were obtained from Dik et al. (16), MIAME accession no. E-MEXP-337. Hierarchical clustering on the human thymocyte populations was performed using 399 out of 547 genes from the T-cell development gene signature. The remaining 148 genes could not profiled in the human dataset because the U133A array used in Ref. (16) contains fewer probesets than the U133 plus 2.0 array used by us.

Relative levels of gene expression data generated by the Taqman array microfluidic cards were processed using Partek^® Genomics Suite^® 6.6. The sorted in vitro populations (A–F) were subject to ANOVA analysis to remove day and pool effects prior to the hierarchical clustering. Standardized z-scores of the expression values in 47 selected T-cell development genes were clustered using Pearson dissimilarity measure and average linkage. These 47 genes are a subset of the 1,387 genes with high variation and expression levels in our in vitro T-cell differentiation dataset. The same genes and clustering method were used for the human thymocyte populations (I–III) dataset. For this dataset, the expression values obtained by the microfluidic cards were subject to quantile normalization before further analysis.

We isolated CD34⁺ HSCs from human UCB and cultured these cells on OP9-DL1 stromal cells to induce T-cell differentiation. At various time points during 27-day cocultures (performed in biological triplicate), seven distinct T-cell populations were sorted as progeny from pooled CD34⁺ HSCs from multiple cord blood donors (Figure 1A). Microarray gene expression analysis was performed on 26 OP9-DL1-generated T-cell populations collected from five independent cocultures and three CD34⁺ HSC starting pools. PCA revealed three distinct clusters (Figure S1A in Supplementary Material). After filtering for probes with high variance (top 5%) and high log₂ expression (>8), the expression profile was trimmed to 1,387 genes (2,179 probesets) that robustly separate the three PCA clusters (Figure 1B; Table S1 in Supplementary Material). One cluster (upper left corner) represents the gene expression profile of CD34⁺ cells. The other two developmental clusters characterize “early” T-cell progenitor (ETP) populations (including the CD7⁺ and part of the CD7⁺CD5⁺ sorted populations) and “late” T-cell differentiation populations (part of the CD7⁺CD5⁺ population and CD7⁺CD5⁺CD1a⁺, CD4/8 DP, and SP sorted populations). The “early” and “late” clusters represent a major change in the transcriptional program (Figures 1B,C). Of note, the CD7⁺CD5⁺ population (in orange) is divided over both clusters (Figure 1B); those fractions that were sorted on days 7 and 12 of the coculture cluster with early CD7⁺ populations, whereas those sorted on day 18 fall in the “late” cluster. Therefore, this CD7⁺CD5⁺CD1a⁻ population is not one homogeneous population but reflect two distinct differentiation stages that dramatically differ in their transcriptional programs.

The 1,387 genes of our expression profile can be further differentiated into 16 different gene signatures of co-expressed genes (Figures 1C,D; Table S1 in Supplementary Material). Based on GSEA, we identified five signatures (#2, #7, #8, #15, and #16) that are most differentially expressed between the early and late T-cell differentiation populations (Figure 1E). Signatures #7 and #8 are enriched for genes that are highly expressed in the early T-cell progenitor stages but downregulated at later T-cell differentiation stages. Signatures #2, #15, and #16 are enriched for genes that become expressed at later T-cell differentiation stages. These five gene signatures—comprising 547 genes in total—faithfully distinguish the early and late T-cell differentiation populations (Figures S1B,C and Table S1 in Supplementary Material). Early T-cell differentiation signatures #7 and #8 include transcription (co)factors that are important for HSC/ETP maintenance and/or lineage determination, such as MEIS1, LMO2, MEF2C, LYL1, and HHEX (24–26). Genes encoding transcription factors that are responsible for T-cell identity (including GATA3, TCF7, BCL11B, Figure S2 in Supplementary Material) and TCR rearrangements (RAG1/2) are present in late signatures #2, #15, or #16. Therefore, these five signatures contain essential genes that are associated with transcriptional programs required for stem/progenitor identity, T-cell specification, and lineage commitment.

We then compared our gene expression signatures to those of murine and human ex vivo sorted thymocyte subsets (16, 23, 25). We performed GSEA to determine the enrichment of our five gene signatures in the dataset of consecutive T-cell development populations from murine thymi, generated by the Immunological Genome Consortium (23). As a result, we demonstrate that our “early” signatures #7 and #8 are significantly enriched for genes that are typically expressed in uncommitted murine subsets ranging from long-term HSCs to DN1–DN2a stages (Figure 2A; FDR-corrected p-value is q < 0.001). Known pre-commitment genes such as HHEX, KIT, MEF2C, LYL1 are among the top 10 genes in our early signatures (Figure 2A: left panel; Table S2 in Supplementary Material). Our “late” T-cell development gene signatures #2, #15, and #16 are significantly enriched for genes expressed in post-committed murine thymocyte subsets (DN2b and later subsets) and include BCL11B, TCF7, LEF1, GATA3 and various molecules associated with TCR signaling (FDR q = 0.028, Figure 2A: right panel; Table S2 in Supplementary Material). We then compared our gene signatures to gene expression data of human ex vivo sorted thymocyte populations (16). Interestingly, our gene signatures that discriminate pre- and post-commitment also divide the ex vivo sorted human thymocytes into uncommitted (the CD1a⁻ populations) and T-cell committed populations (CD1a⁺ and beyond; Figure 2B; Table S3 in Supplementary Material). These results show that the pre- and post-commitment programs in murine and human in vivo T-cell development remain highly conserved in T-cells that are generated from HSCs cultured on OP9-DL1 stromal support.

Further examination of the genes that are expressed in the ex vivo CD34⁺CD1a⁻ human thymocyte populations (16) revealed expression of certain post-commitment genes [including PTCRA, RAG1, RAG2, and DNTT (known as TDT)]. This suggests that thymocytes prior to their acquisition of CD1a surface expression have initiated TCRB rearrangements and are already committed to the αβ T-lineage (Figure 2B; Table S3 in Supplementary Material). From our gene signatures, we searched for surface antigens that are differentially expressed between pre- and post-commitment stages. We observed that expression of CD44, CD33, CD63, CD69, and CD300LF were markedly reduced before the appearance of CD1a on the cell surface (Figure 1D; Figure S2 in Supplementary Material). As the DN2a–DN2b T-cell commitment point in murine T-cell development is associated with an initial reduction of CD44 expression (2, 23), we investigated whether loss of CD44^dim surface expression marks human T-cell commitment. CD44 surface expression on differentiating T-cells in the OP9-DL1 system is lost from CD5⁺ T-cells before these acquire CD1a expression (Figure 3A). We sorted subsets at days 12 and 18 from two independent OP9-DL1 cocultures with the inclusion of CD44 as an additional sorting marker and were able to purify CD5⁺CD44^dimCD1a⁻ (E1), CD5⁺CD44⁻CD1a⁻ (E2), and CD5⁺CD44⁻CD1a⁺ (F) populations (Figure 3A). Microfluidic cards-based RT-qPCR analysis of 47 pre- or post-commitment genes strongly supported a pre-commitment profile in sorted populations A through E1, and a T-lineage commitment profile in populations E2 and F (Figure 3B). Therefore, T-cell commitment of differentiating T-cells cultured on OP9-DL1 is associated with loss of CD44^dim expression before the acquisition of CD1a surface expression.

Next, we determined whether loss of CD44 during in vitro T-cell differentiation from UCB-derived CD34⁺ stem cells parallels T-cell commitment during in vivo human T-cell development. For this, we first assessed CD44 expression in human thymocytes in relation to surface expression levels of CD45, CD3, and other T-cell markers (Figure 4). CD44 expression on early, uncommitted thymocytes is dim in comparison to the bright CD44 expression on mature T-cells (27), and roughly correlates with CD45 expression levels (Figure 4). CD45^dim thymocytes are CD44^dim or CD44⁻ and predominantly consist of developmentally early thymocyte fractions that include CD4/8 DN, CD4 immature single positive (ISP), and some CD3⁻ DP cells. The DN thymocytes contain the CD44^dim, CD44⁻CD1a⁻, and CD44⁻CD1a⁺ populations, whereas the later CD4ISP and immature CD3⁻ DP cells are all CD44⁻CD1a⁺ (Figure 4). In contrast CD45^bright thymocytes are CD44^bright and mostly consist of CD3⁺ DP, CD4 SP, and CD8 SP thymocyte populations. The human thymus also contain very low numbers of mature B-, myeloid, and NK-cells (17), that are CD3/4/8 triple-negative but CD45^brightCD44^bright, and CD1a⁻ (Figure 4).

We then sorted immature (CD34⁺)CD7⁺CD5⁺ thymocytes from CD45^dim cells (lacking CD45^brightCD44^bright cells) from thymi of four independent and immunologically healthy donors. Following depletion of CD3, CD4, CD8, and CD19 cells, CD7⁺CD5⁺CD45^dim immature thymocytes were further sorted into three populations: CD44^dimCD1a⁻ cells (population I), CD44⁻CD1a⁻ cells (population II), and CD44⁻CD1a⁺ cells (population III; Figure 5A). These ex vivo-sorted populations I–III are equivalent to the in vitro populations E1, E2, and F (Figure 3A). Hierarchical cluster analysis of the gene expression levels of the panel of 47 pre- and post-commitment genes identifies population I (CD44^dim) as uncommitted thymocytes, whereas populations II and III CD44⁻ thymocytes have committed to the T-cell lineage (Figure 5B). Therefore, loss of CD44^dim expression marks the transition to human T-cell commitment in vivo, consistent with our in vitro data. Of note, expression of the stem-cell marker CD34 gradually decreased in populations I–III but was still present on a fraction of CD1a⁺ thymocytes (population III, Figure 5A), in line with previous observations (13, 16). Therefore, loss of CD34 surface expression does not mark T-cell commitment.

αβ T-cell committed thymocytes have initiated or completed TCRB gene rearrangements. To further validate T-cell commitment in sorted CD44⁻CD1a⁻ (population II) and CD44⁻CD1a⁺ (population III) human thymocytes, we analyzed Dβ–Jβ and Vβ–Jβ recombination events using GeneScan analysis (19). Analysis of sorted CD44^dim uncommitted thymocytes (population I) did not reveal detectable rearrangements (Figure 6A). In contrast, incomplete Dβ–Jβ and Vβ–Jβ recombination events were identified in the earliest population of post-committed CD44⁻CD1a⁻ thymocytes (II) with a more complete spectrum of full recombinations present in CD44⁻CD1a⁺ thymocytes (III) (Figure 6A).

To provide functional evidence that loss of CD44^dim marks T-cell commitment, we investigated the intrinsic potential of alternative cell fate decisions in the three sorted human thymocyte populations I–III. For this, thymocytes from populations I–III were cultured on OP9-DL1 stromal cells, on OP9-GFP control cells that do not support T-cell differentiation or on a 1:1 mixture of OP9-DL1/OP9-GFP stromal cells. These cultures were performed in the presence of cytokines that support differentiation into various hematopoietic lineages. After seven days of coculture, microscopic inspection already revealed the inability of committed thymocyte populations II and III to proliferate in cocultures with OP9-GFP stromal cells (Figure S3 in Supplementary Material). After 10 days of coculture under various stromal support conditions, we determined the total numbers of T-, NK-, B-, and myeloid cells based on lineage markers CD5 and CD1a (T), CD56 and CD94 (NK), CD19 (B), and CD33 (myeloid) (Figures 6B,C). The uncommitted thymocytes (population I) proliferated well and differentiated into cells of all four lineages on OP9-DL1 support. This uncommitted population efficiently gave rise to B-, myeloid, and NK-cells—but not T-cells—when cultured on OP9 control cells. In contrast, committed thymocyte populations II and III gave rise to CD5⁺CD1a⁺ T-cells only, indicating that these populations have lost the potential to differentiate into alternative cell fates (Figure 6C). The fact that both populations II (CD44⁻CD1a⁺) and III (CD44⁻CD1a⁻) have fully adopted the T-cell fate illustrates that loss of CD44—rather than the acquisition of CD1a—is the first surface marker that identifies post-committed human thymocytes.