Reprogramming of T Cells to Natural Killer–Like Cells upon Bcl11b Deletion

See allHide authors and affiliations

Science  02 Jul 2010:
Vol. 329, Issue 5987, pp. 85-89
DOI: 10.1126/science.1188063


T cells develop in the thymus and are critical for adaptive immunity. Natural killer (NK) lymphocytes constitute an essential component of the innate immune system in tumor surveillance, reproduction, and defense against microbes and viruses. Here, we show that the transcription factor Bcl11b was expressed in all T cell compartments and was indispensable for T lineage development. When Bcl11b was deleted, T cells from all developmental stages acquired NK cell properties and concomitantly lost or decreased T cell–associated gene expression. These induced T-to–natural killer (ITNK) cells, which were morphologically and genetically similar to conventional NK cells, killed tumor cells in vitro, and effectively prevented tumor metastasis in vivo. Therefore, ITNKs may represent a new cell source for cell-based therapies.

T cell development involves progenitor homing, lineage specification, and commitment and requires a complex interplay among key transcription factors (1, 2). In the periphery, the cytokine interleukin 7 (IL-7) and the constant interaction of T cells with self peptide–major histocompatibility complex (MHC) play a critical role in T cell maintenance (3). Reverse transcription–polymerase chain reaction (RT-PCR) analysis indicates that many genes important for T cell commitment start to increase their expression in the transition through the double-negative (DN) stages from DN1 to DN2, and Bcl11b is the most up-regulated transcription factor (4). In bony fish, Bcl11b is shown to be required for T cell–precursor homing to the thymus (5). In the mouse, Bcl11b has critical roles in fetal thymocyte development and survival, for positive selection, and in survival of double-positive (DP) thymocytes (6, 7).

To determine Bcl11b expression in T cells at the single-cell level, we produced and analyzed a Bcl11b-tdTomato knock-in mouse (fig. S1, A and B) (8). In hematopoietic lineages, Bc11b was not expressed in B or myeloid cells, whereas almost all DN2 to DN4 and DP thymocytes, CD4+ and CD8+ T cells, γδ T cells, and natural killer T cells (NKTs) expressed Bcl11b (figs. S2, A to C, and S3, A to C). In DN1 thymocytes, very little to no expression of Bcl11b was detected in CD117++ cells [known as early T cell lineage progenitors (2)] (figs. S2A and S3A). During NK development, transient low Bcl11b expression was observed in immature NK cells but not in NK precursors or mature NK cells (figs. S2D and S3D). In contrast, the majority of thymic NK cells, identified by CD127 (9), expressed Bcl11b (figs. S2D and S3E). Moreover, using quantitative real-time PCR (QRT-PCR) analysis, we showed that, in both CD4+ and CD8+ splenic T cells, Bcl11b transcription in naïve (CD44CD62L+) T cells was roughly two times that in activated (CD44+CD62L) T cells (figs. S2E and S3F), and activated T cells exhibited a bimodal pattern of expression (fig. S2F).

To further determine Bcl11b functions in T cells, we generated the conditional knockout mice (Bcl11bflox/flox) (fig. S4A), which were crossed to the Rosa26Cre-ERT2 mice (10). Consequently, in CreERT2; Bcl11bflox/flox mice (the PLBD line, referred to hereafter as flox/flox), Bcl11b could be deleted by treating cultured cells or mice with 4-hydroxytamoxifen (OHT). Using OHT-treated whole thymocytes from these and the control (CreERT2; Bcl11bflox/+, referred to as flox/+) mice, we sorted and subsequently cultured DN1 and DN2 cells in T cell media on OP9-DL1 stromal cells (fig. S4B) (11), which support T cell development but suppress NK cell development from the progenitors (12). All stromal cells were killed in the OHT-treated flox/flox DN1 thymocyte culture.

Flow cytometry showed that 24% of cells in this culture expressed NKp46, which is primarily expressed on NK cells (Fig. 1A) (13). These NKp46+ cells did not express T cell genes for CD3 or the β T cell receptor (TCRβ) (fig. S4C) and had lost both alleles of the Bcl11b exon 4 (fig. S4D), which indicated that they did not acquire or had lost T cell features, even though they were cocultured with OP9-DL1 stromal cells for 14 days. However, the control OHT-treated flox/+ and untreated flox/flox DN1 cells proliferated rapidly, and many acquired CD3 expression but not NKp46+ (Fig. 1A and fig. S4E). These data demonstrated that Bcl11b deficiency caused production of the NKp46+ cells from DN1 thymocytes and that Bcl11b was required early in T cell development. Similar to cultured DN1 thymocytes, OHT-treated flox/flox DN2 thymocytes also produced NKp46+CD3 cells, which killed the stromal cells, whereas control DN2 thymocytes did not (Fig. 1A and fig. S4E). Growth of NK-like cells from Bcl11b-deficient DN1 or DN2 thymocytes appeared to be independent of Notch signaling, because NKp46+ cells were readily produced from DN1 or DN2 thymocytes cultured on OP9 stromal cells without IL-2 or IL-15 (fig. S4F).

Fig. 1

Bcl11b is essential for T cell development and for maintaining T cell identity. Thymocytes from flox/flox or flox/+ control mice were treated, or not, with OHT, then sorted into DN1 or DN2 subsets and cultured on OP9-DL1 stromal cells. (A) Flow cytometry profiles of cultured DN1 and DN2 thymocytes (+OHT) in the absence of IL-2. Numbers refer to percentage of cells in the gate. Data are representative of three experiments. (B) Flow cytometry profiles of cultured flox/flox DN3 thymocytes (±OHT) supplemented with IL-2. Data are representative of three experiments. (C) Killing of OP9-DLI stromal cells by OHT-treated flox/flox DN3 thymocytes. Scale bar, 40 μm. (D) DNA from purified NKp46+ cells was prepared and subjected to PCR to detect DJ (top) and V(D)J (bottom) recombination at the TCRβ locus. T, T cells growing from untreated DN3 thymocytes; N1 and N2, sorted NKp46+ cells growing from OHT-treated flox/flox DN3 thymocytes; Thy, wild-type whole thymocytes; B, B cells; and GL, germline band. H2O: no DNA template in PCR. Numbers indicate DJ recombination products. (E and F) Microarray analysis of gene expression in NKp46+CD3± ITNK cells from DN3 thymocytes (columns I1 to I4), IL-2–expanded NK cells (LAK; L1 to L4) and sorted DN3 flox/flox thymocytes (DN3; D1 to D4) were subjected to expression. (E) Two-way hierarchical cluster map of the array data. Column numbers (I1 to I4 for instance) refer to four independent RNA samples for each cell type, and rows represent individual transcripts. Scale indicates the log2 value of normalized signal level. (F) QRT-PCR validation of gene expression of selected genes among ITNKs, LAKs, and DN3 cells. Bars are means ± SD of three samples.

We subsequently deleted Bcl11b in DN3 thymocytes. Again, stromal cell–killing NKp46+CD3 cells appeared (Fig. 1, B and C, and fig. S4G). The reprogramming also worked in myeloid or B cell culture media (fig. S4, H and I), which demonstrated that reprogramming to NKp46+ cells was intrinsic to the Bcl11b-deficient thymocytes. To confirm the T cell origin of NKp46+CD3 cells, we examined their TCRβ locus. These cells exhibited TCRβ V(D)J recombination [recombination of the variable (V), diversity (D), and joining (J) gene segments] even though TCRβ was not expressed (Fig. 1D). We therefore named these killer cells that were reprogrammed from T cells induced T-to–natural killer or ITNK cells.

We next compared, using microarray analysis, the expression profiles of DN3 thymocytes; normal splenic NK cells that were expanded in vitro after enrichment (lymphokine-activated killer, or LAK cells, composed of >90% NK cells); and ITNKs reprogrammed from DN3 cells (Fig. 1E). Consistent with the killing ability of ITNK cells, their expression profile was more similar to that of LAK cells than to their parental DN3 thymocytes. QRT-PCR validation showed that expression of many T-lineage genes—such as Notch1, Est1, Hes1, Gata3, Dtx1, and Tcf1—was decreased, whereas expression of genes usually associated with NK cells—such as Id2 (14), IL2rβ (CD122), Zfp105 (15), and E4bp4 (16)—was up-regulated (Fig. 1F and table S1). Zbtb32 (Rog, Repressor of GATA), which is not normally expressed in DN3 cells but plays important roles in regulating T cell activation and suppresses Gata3 activity (17), was highly expressed in ITNKs. Expression of Cdkn1c (p57KIP2), a putative direct downstream target gene of Bcl11b (18), was also drastically increased in ITNKs (Fig. 1F). These results collectively demonstrated that Bcl11b was essential for maintaining the T cell expression profile and for suppressing NK cell gene expression.

ITNKs could also be produced from mature T cells. We OHT-treated sorted DP thymocytes, CD4+ and CD8+ T cells, and γδ T cells from flox/flox mice. Many ITNKs (NKp46+) were found growing in DP thymocytes and CD8+ T cell cultures (fig. S4, K and L), which effectively killed stromal cells. These ITNKs, in contrast to those reprogrammed from DN1 to DN3 thymocytes, retained TCRβ on the cell surface. We were unable to obtain consistent production of NKp46+cells from splenic or thymic CD4+ T cells or from γδ T cells, because these cells appeared prone to cell death in vitro once Bcl11b was deleted.

To estimate the reprogramming efficiency, we sorted single DN3 thymocytes from OHT-treated flox/flox thymocytes into individual wells of 96-well plates preseeded with OP9-DL1 stromal cells in T cell medium (fig. S5A). Out of the 79 wells that had cells growing, 36 wells had many fast-proliferating T cells (Fig. 2A), which deleted only one flox Bcl11b allele (Fig. 2B, lanes T1 and T2). These cells (flox/–) nevertheless served as excellent controls for Cre toxicity because they had activated Cre recombinase. In the other 43 wells, thymocytes were reprogrammed to NKp46+ stromal cell–killing ITNKs (Fig. 2A). IL-2 was clearly able to greatly promote proliferation of ITNKs, because, from one DN3 thymocyte, up to 0.5 million ITNKs were obtained with IL-2, but only ~50,000 cells without IL-2. All ITNK cells had lost both Bcl11b alleles (Fig. 2B, lanes I1 and I2), and ITNKs of individual wells had unique rearranged TCRβ loci, which confirmed their independent origins (Fig. 2C). Therefore, once Bcl11b was deleted, the reprogramming efficiency of DN3 thymocytes to ITNKs could reach 100%. ITNKs from DN3 thymocytes not only expressed NK cell surface receptors and had similar cytotoxic functions, but were morphologically similar to LAK cells. ITNKs were larger than thymocytes and had granules and showed evidence of high protein-synthesis activity with abundant endoplasmic reticulum (Fig. 2, D and E). Besides NKp46, ITNKs expressed NKG2A/C/E, TRAIL, perforin, and interferon-γ, but not some other key NK cell function genes, such as members of the Ly49 family (fig. S5, B and C). Similar observations were made with in vitro reprogrammed ITNK cells from DP thymocytes (table S2 and fig. S5D). ITNKs were unlikely to be related to thymic NK cells because they did not express CD127 (fig. S5E). Moreover, unlike conventional mature NK cells, most ITNKs did not express CD11b, rather, they expressed CD27 and retained their killing ability even after being cultured in vitro for 1 month (fig. S5F).

Fig. 2

Efficient reprogramming of T cells to ITNKs. (A) Representative flow cytometry profiles of ITNKs reprogrammed from single flox/flox DN3 cells. Numbers refer to percentage in total cells. T: T cells that did not have complete Bcl11b deletion. Data are representative of three experiments. (B) PCR genotyping of Bcl11b deletion in two representative T cell (T1 and T2) and ITNK (I1 and I2) wells. flox, floxed allele; del, deletion allele. –OHT: no OHT treatment; H2O: no template control. (C) DJ recombination at the TCRβ locus of five ITNK wells (I1 to I5) showing unique DJ recombination. L, DNA ladder; Thy, wild-type thymocytes. (D) Giemsa stain of parental DN3 thymocytes (T) and ITNK cells. Scale bar, 20 μm. (E) Transmission electron micrograph of an ITNK cell. 1, Nucleus; 2, Golgi body; 3, granule; 4, endoplasmic reticulum. Scale bar, 2 μm. (F) Cytotoxicity of ITNKs (labeled as “+OHT”) and LAKs measured in standard 51Cr-release assays with B16F10, RMA, and RMA-S tumor cell targets at the indicated effector-to-target (E:T) ratios. –OHT: flox/flox T cells. Data are means of triplicate wells.

We next measured the killing ability of the DN3-reprogrammed ITNKs by performing standard 51Cr-release assays with three cell lines: B16F10 melanoma, which are MHC class I (MHC I) low or negative (19); RMA lymphoma, which express MHC I; and RMA-S lymphoma (TAP-1–deficient variant), which have reduced MHC I presentation (20, 21). LAK cells generally only killed MHC I–negative cells (Fig. 2F). Similar to LAKs, ITNKs also selectively killed MHC I–negative B16F10 and RMA-S cells, although they had slightly lower killing potency (Fig. 2F).

To exclude the possibility that ITNKs were in vitro artifacts, we deleted Bcl11b in vivo (fig. S6A). Two to 3 weeks after OHT treatment, ITNKs were detected in both the spleen (NKp46+CD3+) and the thymus (NKp46+) from flox/flox mice but not from the flox/+ controls (Fig. 3A). Bcl11b was found deleted in these in vivo reprogrammed ITNKs (fig. S6B). Both CD4+ and CD8+ ITNKs (NKp46+) were found (fig. S6C). Some wild-type γδ T cells expressed NKp46; however, Bcl11b deletion caused a threefold increase in the NKp46+ γδ T cells (Fig. 3B), which suggested that all T cell populations have reprogramming potential. The in vivo reprogrammed ITNKs could readily be expanded in NK culture conditions (fig. S6D), but they were not NKT cells (fig. S6, E and F). Besides expressing NK cell–associated genes, the in vivo reprogrammed ITNKs also lost or decreased some key T cell genes such as Il7r, Tbx21 (T-bet), and Cd8a (fig. S6G). Consequently, TCR signaling in ITNKs appeared to be compromised (fig. S6H).

Fig. 3

ITNKs reprogrammed in vivo are potent tumor cell killers. (A) Flow cytometric analysis of thymocytes and splenocytes from OHT-treated flox/flox and flox/+ mice. Numbers refer to the percentage in the lymphocyte gate. Data are representative of four mice. (B) Analysis of ITNKs from thymic γδ T cells in OHT-treated flox/flox mice. Data are representative of two mice. (C) ITNKs production in Rag2−/−Il2rg−/− recipients injected with flox/flox DP thymocytes. Two weeks after injection, donor (CD45.2+) and host (CD45.1+) splenocytes were analyzed. Numbers refer to the percentage of lymphocyte gate. Plots are representative of 15 mice from three independent experiments. (D) Ex vivo expansion of ITNKs in IL-2 from splenocytes of the recipient mice. Viable cells were counted (top) at the indicated time points and analyzed (bottom). Numbers refer to percentages. Most cells in the culture were ITNKs because they expressed NKp46, TCRβ, NK1.1, and NKG2D. Bars are means ± SD of four samples. Data are representative of three experiments. (E) The ex vivo expanded ITNKs (labeled as “+OHT”) were used in 51Cr-release killing assays with B16F10, RMA, and RMA-S tumor cell targets at the indicated effector-to-target (E:T) ratios. –OHT: flox/flox T cells. Data are means of triplicate wells. Results are representative of three experiments. (F) ITNKs prevented tumor metastasis. Rag2−/−Il2rg−/− recipients were first transplanted with treated (+OHT) or untreated (-OHT) flox/flox DP thymocytes or phosphate-buffered saline. Recipients were subsequently injected intravenously with 50,000 B16F10 melanoma cells. Lung tumor colonies were enumerated 2 weeks after tumor challenge. Data are from individual mice, and bars represent the means.

The in vivo analysis of ITNKs in flox/flox mice was complicated by the presence of many host T cells and NK cells (Fig. 3A). To address this problem and also to investigate whether in vivo reprogramming upon Bcl11b loss is cell autonomous, we transplanted OHT-treated DP thymocytes from flox/flox mice (CD45.2+) into Rag2−/−Il2rg−/− mice (CD45.1+) that lack B, T, and NK cells (fig. S7A) (22). We chose DP thymocytes because they usually account for more than 75% of total thymocytes and could be efficiently reprogrammed in vitro to ITNKs (fig. S4K). Two weeks after transplantation, ~5% of splenocytes were found to be from the donor cells (CD45.2+) (Fig. 3C), and ~47% of them expressed NKp46 and thus were ITNKs. ITNKs lost both copies of Bcl11b, and the majority of them expressed CD8 (fig. S7, B and C). The other 53% of cells (NKp46) were T cells and still retained the Bcl11b floxed allele (fig. S7C). ITNKs were also found in the bone marrow and peripheral blood (fig. S7D). We estimated that there were ~200,000 ITNK cells in the spleen. No NKp46+ cells were found in control mice transplanted with untreated DP thymocytes (Fig. 3C). ITNK cells were maintained in the recipients for at least 3 months without change in cell number, perhaps representing a dynamic balance in their numbers. The recipient mice did not show any noticeable abnormality, which indicated that ITNK cells did not indiscriminately kill normal cells, nor were they malignantly transformed.

Compared with those reprogrammed in vitro, the in vivo reprogrammed ITNKs expressed NK surface receptors of the Ly49 family, including Ly49C/I and Ly49G2 (fig. S7E) (table S2), and could be extensively expanded ex vivo with IL-2 or IL-15 for at least 3 weeks while still retaining their killing ability (Fig. 3D).

The ex vivo expanded ITNKs were then assessed for their tumor cell–killing capability. These cells exhibited elevated cytotoxic potential and were also generally more potent than both in vitro ITNKs and LAKs against each of the target cells (Fig. 3E and Fig. 2F). Moreover, they killed RMA cells with almost the same efficiency as they killed RMA-S cells (Fig. 3E), despite expression of some inhibitory Ly49 receptors, which recognize MHC I.

Transplantable murine melanoma B16 cell lines are well-established models for studying experimental cancer therapies and NK cell tumor surveillance function (23). Injection of B16 cells into Rag2−/−Il2rg−/− mice leads to rapid formation of metastatic foci in the lungs (24). To investigate the tumor-killing ability of the ITNK cells in vivo, we injected 2 million OHT-treated or untreated DP thymocytes from flox/flox mice into Rag2−/−Il2rg−/− recipients to allow reprogramming of thymocytes to ITNKs in vivo (fig. S7F). Two weeks later, each recipient was injected with B16F10 melanoma cells. Four weeks after the initial thymocyte transplantation, recipients were killed and analyzed. Mice injected with phosphate-buffered saline or with untreated DP cells had ~200 metastatic foci in the lungs. In contrast, mice injected with OHT-treated DP thymocytes had ~20 tumor colonies in the lung (Fig. 3F and fig. S7G). Therefore, ITNKs were potent killers of tumor cells in vivo and prevented cancer progression.

Deletion of Bcl11b led to the rapid disappearance of Bcl11b protein (Fig. 4A). Microarray analysis showed that, in OHT-treated flox/flox thymocytes, expression of T cell genes such as TCRβ and CD3 was already down-regulated within 24 hours (table S3). In another 24 hours, many genes associated with NK cells were expressed (table S3). Bcl11b is proposed to be regulated by Notch signaling in T cell development (25). The Drosophila ortholog of Bcl11 genes, CG6530, is shown to be a direct downstream target gene of Notch signaling (26). To confirm this, we identified putative CSL-binding sites (CGTGGGAA) (27) at the Bcl11b locus, which were conserved between mouse and human Bcl11b genes (Fig. 4B) (table S4). A chromatin immunoprecipitation (ChIP) assay performed with a CSL polyclonal antibody pulled down genomic DNA fragments from T cells. Three genomic regions were greatly enriched in the T cell samples when we used the CSL antibody compared with the control (Fig. 4C). The ChIP result thus confirmed that the canonical Notch signaling directly regulated Bcl11b in T cells (fig. S8).

Fig. 4

Bcl11b is a direct downstream target gene of Notch signaling. (A) Bcl11b protein in T cells after OHT treatment detected by Western blot. (B) Schematic of the Bcl11b locus showing putative CSL binding sites (BS) and that of an irrelevant control binding site (CTL). (C) Genomic DNA was prepared from immunoprecipitation of thymocytes, by using CSL or control immunoglobulin G (IgG) antibodies, and was amplified by using primers flanking the putative CSL or the control binding sites at the Bcl11b locus. Three Bcl11b-binding regions: Region 1, about 1.8 kb from the start of the transcription; region 2, 5.4 kb downstream of exon 1; region 3, about 600 base pairs downstream of exon 2. CSL, CSL antibody; IgG, control IgG. Fold-enrichment was calculated relative to the IgG control (set to 1). Bars are means ± SD of triplicate samples.

NK cell–based therapies hold promise in cancer treatment. We are now able to reprogram T cells to ITNKs, which can be extensively expanded but are not malignantly transformed. Rather, they effectively killed tumor cells in vitro and eliminated metastatic cells in mice but did not appear to attack normal cells. Therefore, ITNK cells may serve as a new cell source for cancer immunotherapy and other cell-based therapies.

Supporting Online Material

Materials and Methods

SOM Text

Figs. S1 to S8

Tables S1 to S4


References and Notes

  1. Materials and methods are available as supporting material on Science Online.
  2. This work is supported by the Wellcome Trust to P. Liu, from MRC and BBSRC to F. Colucci, and by MRC to B. Huntly. We thank core facilities at Sanger Institute (RSF and microarray), Babraham Institute and CIMR for technical support. Microarray data have been submitted to GEO (Gene Expression Ominibus). Accession number: GSE21016. A patent application has been filed by the Wellcome Trust for production and application of ITNKs (U.K. patent application No. 0912287. U.S. Provisional Application No. 61/225779), and P.L. and P.L. are listed as authors on this application. A Mouse/Tissue Transfer Agreement is required for the use of PLBD mice.
View Abstract

Navigate This Article