Arrested natural killer cell development associated with transgene insertion into the Atf2 locus

Natural killer (NK) cells constitute the third major population of lymphocytes and play critical roles in innate immune responses to pathogens and tumors.¹  They can be distinguished from other lymphocytes by their lack of B- or T-cell antigen receptor expression. Indeed, NK-cell development is independent of events required for B- and T-cell antigen receptor gene rearrangement, as illustrated by normal NK cells found in mice with defects in the recombination machinery, such as scid, and Rag-deficiency. However, NK-cell developmental pathways are just beginning to be elucidated (reviewed in Yokoyama et al²  and Colucci et al³ ).

NK cells appear to develop completely in the bone marrow (BM). Based on the differential expression of cell-surface markers and NK-cell functional assays, we previously proposed a developmental model of mouse NK cells.⁴  Cells at putative developmental stages can be distinguished from each other and from cells resembling mature peripheral NK cells. The earliest detectable precursor committed to the NK-cell lineage (stage 1) expresses CD122 (IL-2/IL-15Rβ)^5-7  that is then expressed throughout NK-cell maturation. Although this receptor subunit is associated with the interleukin-2 receptor (IL-2R) complex, its contribution to the IL-15R complex is probably the most relevant for NK-cell development because mice deficient in IL-15 or IL-15Rα completely lack NK cells with relative preservation of the development of other lymphocytes.^8,9  Developing NK cells express NK1.1 (NKR-P1C) (stages 2 to 5), consistent with studies of in vitro development of human NK cells, indicating that the NKR-P1A molecule is one of the earliest markers expressed.¹⁰  Among the NK1.1⁺ CD122⁺ population, there is an αv integrin⁺ c-kit^- population (stage 2) that expresses the CD94/NKG2 receptor without concomitant expression of Ly49 receptors. This is also consistent with observations that CD94/NKG2 receptors are found on fetal or neonatal NK cells without Ly49 receptor expression.^11,12  Subsequently, the NK1.1⁺ cells express Ly49 receptors, and also c-kit and αv (stage 3). When αv expression then decreases, α2 integrin (DX5) expression increases (stage 4). At this stage, developing NK cells undergo constitutive proliferation in the BM. Subsequently, as NK cells finally acquire expression of Mac-1 and CD43 at levels comparable to peripheral NK cells (stage 5), NK-cell proliferation wanes. Concomitant with high-level expression of Mac-1 and CD43, there is full functional maturation of NK cells with development of cytotoxicity and cytokine production that closely resembles mature, peripheral NK cells. Thus, NK cells, like other hematopoietic cells, undergo a series of developmental steps in order to become fully mature, though the molecular basis for these stages remains to be determined.

Analysis of targeted mutant and transgenic mice has revealed factors and precursor cells that contribute to NK-cell development, but most of these factors affect NK-cell development at stages before precursor commitment to NK-cell lineage, such that other lineages are severely affected (reviewed in Yokoyama et al² ). For example, Ikaros-deficient mice lack other lymphocytes as well as NK cells, supporting the concept that NK cells are lymphocytes sharing the common lymphocyte progenitor with B and T cells.¹³  The Tgϵ28 mouse contains a high copy number of transgenic human CD3ϵ genomic constructs and manifests deficiencies in T and NK cells,¹⁴  suggesting the existence of a bipotential cell that can give rise to both T and NK cells. This possibility is supported by studies indicating that the adoptive transfer of neonatal thymocytes results in donor-derived T and NK cells in recipient mice,¹⁵  and more recent in vitro studies.^16,17  Downstream of this putative bipotential T and NK-cell precursor are cells that give rise to both NK and the NK T-cell subset. Ets-1–deficient as well as lymphotoxin-α (LTα) mice have deficiencies affecting development of both NK cells and NK T cells with relative sparing of other lineages,^18,19  highlighting the even closer relationship of NK cells with NK T cells. However, few mice have been described with a selective defect affecting NK-cell development after NK cell-lineage commitment.^2,3  Phenotypic and molecular analyses of such mice will lend significant insight into NK-cell development.

We previously reported a mouse hemizygous for a Ly49A transgenic (Tg) construct that demonstrated selective and marked impairment of NK-cell killing associated with a reduction of CD3^- NK1.1⁺ NK cells in peripheral tissues.²⁰  Herein we demonstrate that our Ly49A Tg mice manifest a developmental block in NK-cell maturation in the BM that is unrelated to the function of Ly49A as an MHC class 1–specific receptor. The mice have an increase in NK-cell number in the BM due to the accumulation of functionally immature cells lacking expression of cell-surface markers associated with mature NK cells. These developmental abnormalities are associated with the insertion of the Tg construct into the gene for ATF-2, a basic-region leucine zipper (bZIP) transcription factor.

C57BL/6 mice were obtained from the Jackson Laboratory (Bar Harbor, ME) and the National Cancer Institute (Frederick, MD). C57BL/6-Ly5.1 mice were obtained from the Jackson Laboratory. The generation of Ly49A Tg, NK cell–deficient mice has been described,²⁰  and these mice are now available at the National Institute of Allergy and Infectious Diseases repository (http://www.mmrrc.org/dev/strains/129/0129.html). Briefly, this mouse was generated by microinjection of C57BL/6 fertilized eggs with a transgenic construct consisting of the cDNA for Ly49A inserted in place of the start codon of the granzyme A gene. Homozygous Atf2-hypomorphic mutant mice and mutant mice with a deletion in the DNA binding domain have been described.^21,22  Mice were housed in a specific pathogen-free facility supervised by the Division of Comparative Medicine at Washington University. All studies were approved by the Washington University Animal Studies Committee and carried out according to its guidelines.

To produce BM chimeric mice between WT and Tg mice, WT (Ly5.1) mice were reconstituted with Tg BM (Ly5.2) cells, or vice versa, after host received 9.5 Gy irradiation. To generate BM chimeric mice for Atf 2-null mutants, WT (Ly5.1) mice were reconstituted with fetal liver cells isolated from (day 15 to 17) embryos, which were obtained from heterozygote mating. The genotype of embryos was determined by polymerase chain reaction (PCR) as described previously.²¹  Chimeric mice were analyzed 6 to 12 weeks later.

Single-cell suspensions were prepared from BM (femur and tibia) and spleen, and depleted of red blood cells by standard methods. For flow cytometry, cells were stained with combinations of indicated fluorochrome-conjugated mAbs and analyzed with a FACScalibur (Becton Dickinson, San Jose, CA) as described previously.⁴  The following fluorochrome-conjugated mAbs were from Pharmingen (La Jolla, CA): phycoerythrin (PE)–or allophycocyanin (APC)–conjugated anti-NK1.1 (clone PK136), fluorescein isothiocyanate (FITC)–anti–Mac-1 (M1/70), FITC–anti-CD43 (S7), FITC–anti–LFA-1 (M17/4), FITC- or PE-DX5, PE–anti–-α_v integrin (H9.2B8), FITC–anti-Ly5.2 (104) FITC-, Cy-Chrome-, or peridinin chlorophyll protein-conjugated anti-CD3 (145-2C11) and FITC–anti–mouse IFNγ (XMG1.2). Dead cells were excluded by propidium iodide staining, as indicated.

Mice were intraperitoneally injected with 150 μg poly-I:C, and 24 hours later, freshly prepared BM or spleen cells were tested in standard ⁵¹Cr-release assays using 96-well U-bottom plate as previously described.²⁰ 

Cells prepared from the indicated tissues were incubated in the presence of IL-2 (1000 U/mL) and IL-12 (10 ng/mL) for 1 hour, and then further incubated in the presence of brefeldin A for 5 to 7 hours as previously described.⁴  Alternatively, cells were incubated with YAC-1 tumor targets at 10:1 splenocyte-to-target ratio. Higher splenocyte-to-target ratios did not change the results. For anti-NK1.1 stimulation, mAb PK136 (anti-NK1.1) was immobilized to plastic then washed and used for NK1.1 cross-linking as recently described.²³  For the staining of intracellular IFNγ, cells were first stained for surface markers, and then fixed, permeabilized, and incubated with anti-IFNγ mAb using the Cytofix/Cytoperm kit (Pharmingen), according to the manufacturer's instructions.

Genomic DNA was extracted from tail biopsies using the Puregene DNA Purification kit (Gentra Systems, Minneapolis, MN) for PCR analysis. The Tg mice were genotyped by standard PCR techniques using Taq polymerase (Promega, Madison, WI) and the following specific primers for the Ly49A cDNA: (CATTGTGATAGCTCTTGGCATCTT forward primer; GTTGATGTCACTTTGCATGTTGC reverse primer). This PCR reaction generates a 150-bp product only in the transgene-positive mice.

Genomic DNAs isolated from WT and Tg mice were digested separately with several restriction enzymes, ligated at a low concentration to allow circularization and intramolecular ligation, and subjected to PCR using sense and antisense primers corresponding to sequences of the 3′ end of Tg construct.²⁰  Enzymes for DNA digestion were chosen such that they cut within granzyme A-Ly49A Tg construct near 3′ end. This inverse PCR yielded a product from BclI-digested Tg DNA but not from WT DNA (data not shown). Sequence analysis revealed the junction sequence that contained sequences of exon 5 of ATF-2 gene. This insertion site (referred to as 5′ integration site) sequence was independently confirmed by generating a 507-bp product from genomic DNA using the forward ATF2 primer (a1, GCGTTTTACCAACGAGGATCA) and the forward granzyme A primer (a2, CCACCCTCTTGTTTTCCAGG). The integration of Tg construct also was confirmed with Southern blot analysis using an Atf 2 cDNA probe (data not shown).

To clone flanking sequences of the other side (3′ integration site), we attempted a similar intramolecular ligation-PCR strategy but failed. We then designed primers corresponding to sequences from either 5′ or 3′ end of Tg construct (since the Tg insertion probably consists of multiple copies inserted as a concatamer in random orientations) and primers corresponding to Atf 2 sequences downstream of exon 5. However, numerous PCR attempts with combinations of these primers failed to amplify specific bands from Tg DNA, suggesting possible deletions within the 3′ terminus of the Tg construct or Atf 2 during the integration. We designed a series of primers corresponding to exon sequences further downstream of exon 5 and performed 5′ rapid amplification of cDNA ends (RACE) PCR using these primers on RNA samples that were prepared from LAK cells derived from WT and Tg BM cells. Sequence analysis of the RACE PCR products that were specifically amplified from Tg RNA revealed a junction between granzyme A intron 3 and ATF2 intron 10. This insertion site (referred to as 3′ integration site) was independently confirmed by generating a 451-bp product from genomic Tg DNA using the reverse ATF2 primer (b1, TCCATGACCTCTGCATTAGCT) and the forward granzyme A primer (b2, ACATGTCCCCATCCCTCCC).

The Ly49A receptor inhibits NK-cell effector function when it engages its MHC class 1 ligand, such as H2D^d (Karlhofer et al²⁴ ). Although Ly49A is not known to recognize an MHC class 1 molecule in C57BL/6 mice (H2^b), we sought to further test whether the transgene-derived Ly49A receptor itself plays a functional role in the NK-cell defect. The Tg mice were bred to β2m-deficient mice that do not express any potential ligand for Ly49A²⁵  or mated with BALB/c mice that express H2D^d class 1 molecules, a strong ligand.²⁴  The NK cells remained decreased in Tg mice with either homozygous β2m-deficiency or Tg (B6xBALB)F₁ hybrid animals as compared to their respective non-Tg littermate controls (data not shown). Thus, the absence or presence of known or potential MHC class 1 ligands for Ly49A did not rescue NK-cell development in Tg mice, suggesting that the NK-deficient phenotype described previously²⁰  is not caused by the expression of transgenic Ly49A as a functional MHC class 1–specific receptor. Consistent with this, 2 other Ly49A transgenic mice did not have this defect.^26,27 

To explore the possibility that the abnormality in our Ly49A Tg mice is associated with an aberration in their development, we analyzed NK cells in the BM. In contrast to the reduction seen in Tg peripheral tissues, the Tg BM demonstrated increased number and frequency of CD3^- NK1.1⁺ NK cells (approximately 3-fold) as compared to WT BM (Figure 1). Furthermore, the increased NK-cell numbers in BM as well as their reduction in the periphery were inherited in a Mendelian dominant pattern for more than 10 generations (data not shown), indicating that a single transgenic integration site is responsible for the phenotype. Thus, the reduction of Tg NK cells in peripheral tissues and associated accumulation in BM represent a stable phenotype that is genetically linked to the transgenic construct.

Accumulation of NK cells in Tg BM could result from either incomplete maturation or abnormal trafficking of mature NK cells. To address these issues, we analyzed the functional activities of BM NK cells that were freshly isolated from Tg and control WT mice. Compared with WT BM cells, Tg BM cells poorly killed the prototypic NK target YAC-1 lymphoma (Figure 2A). A more marked defect was observed in cytotoxicity of the B16 melanoma target. Given that the Tg BM contains greater numbers of NK cells than WT BM, these data indicate that Tg BM NK cells are less cytotoxic than WT BM NK cells.

We previously found defective natural killing with Tg splenic NK cells that was more substantial than the reduction in NK-cell number,²⁰  also suggesting that there was a qualitative as well as quantitative defect in NK cells. To better analyze this issue, herein we analyzed NK-cell function at the single-cell level, for instance, ability of individual cells to produce IFNγ using flow cytometry. When BM cells were stimulated with IL-2 and IL-12, a substantial proportion of WT NK cells produced IFNγ (Figure 2B), as we previously reported.⁴  However, fewer Tg BM NK cells produced this cytokine as compared to WT BM NK cells. Moreover, the few Tg NK cells in the spleen produced less IFNγ than WT splenic NK cells (Figure 2B). Similarly, when splenic NK cells were incubated with YAC-1 tumor targets (Figure 2C), a substantial proportion of WT cells produced IFNγ, whereas Tg cells were less capable of IFNγ production. Moreover, Tg NK cells were defective in IFNγ production when stimulated by anti-NK1.1 cross-linking (Figure 2D). Taken together, although we have not addressed potential trafficking abnormalities, these data indicate that regardless of their tissue origins, Tg NK cells have an intrinsic qualitative defect that is independent of target recognition receptor repertoire or expression, suggesting that they are not functionally mature.

Developing NK cells in the BM can be distinguished by differential expression of several markers, including a number of integrins.⁴  Comparative analyses of WT and Tg NK cells prepared from the BM revealed clear differences in their phenotype (Figure 3A). Strikingly, very few Tg NK cells expressed Mac-1 (also known as αM/β2 or CD11b/CD18), a marker associated with functional maturation.⁴  By contrast, approximately half of BM NK cells had the Mac-1^hi phenotype. Thus, the proportion of WT NK cells with the mature Mac-1^hi phenotype was negligible in the Tg mice.

Differences also were observed between WT and the few Tg NK cells in the periphery (Figure 3A and data not shown). While the vast majority of WT splenic NK cells were Mac-1^hi, very few Tg splenic NK cells showed this phenotype. Importantly, other cell types in the Tg mice expressed Mac-1 normally (Figure 3A), demonstrating that the poor Mac-1 expression was restricted to the NK-cell lineage. Furthermore, Tg NK cells expressed LFA-1 (also known as αL/β2, CD11a/CD18) at levels comparable to WT NK cells (data not shown), indicating the integrity of β2 integrin expression on Tg NK cells. Moreover, 2B4, CD2, and CD122, which are expressed by all WT NK cells, were normally expressed by Tg NK cells (data not shown). By contrast, Tg NK cells poorly expressed CD43 (leukosialin) (Figure 3B), another marker that is highly expressed by WT mature NK cells and is associated with Mac-1 acquisition.⁴  Like Mac-1, the proportion of NK cells with the CD43^hi phenotype was negligible in BM as well as in the periphery of Tg mice. These data demonstrate that regardless of their tissue origin, Tg NK cells have a phenotype associated with an immature stage of NK-cell development. Together with results from functional assays, the data suggest that NK-cell maturation in Tg mice is arrested at an immature stage.

To further characterize the putative developmental block, we took advantage of intermediate stages defined in our proposed model of NK-cell differentiation in which developing NK cells differentiate into more mature cells and acquire high expression of α2β1 (DX5) and Mac-1 integrins in an ordered manner, while αvβ3 expression is down-regulated. Almost all Tg NK cells expressed DX5 at levels comparable to or even higher than those observed on WT counterparts in every tissue (Figure 3B), indicating that acquisition of the pan NK-cell marker DX5 was not interrupted in Tg mice. Also, Tg DX5^hi NK cells showed down-regulation of αvβ3 expression as seen in WT mice (Figure 3C). Taken together, these data suggest that final maturation into Mac-1^hi functional cells is blocked in Tg mice, while the events leading to this step, for instance, differentiation from NK precursor to DX5^hi Mac-1^lo cells with concomitant down-regulation of αvβ3 expression, are not interrupted.

Breeding experiments revealed that the genotype (Tg insertion) and phenotype (NK-cell deficiency) correspond to Mendelian inheritance of a single locus, consistent with transgene insertion into a single integration site (data not shown). Moreover, all of our observations on Tg NK cells were obtained from mice hemizygous for the Tg construct. Fluorescent in situ hybridization also was consistent with these findings, indicating a single integration site of multiple transgene copies was responsible for the NK-deficient phenotype (Supplemental Figure S1, available at the Blood website; click on the Supplemental Figure link at the top of the online article). However, we have not been able to generate viable mice homozygous for the Tg, suggesting that the Tg construct inserted in a site critical for mouse development.

To better understand the molecular basis of the developmental defect in Tg mice, we cloned the genomic segments flanking the transgene integration site using an inverse PCR approach on genomic DNA ligated in dilute conditions to promote intramolecular ligation (see “Materials and methods”). Sequence analysis revealed that the Tg construct inserted into exon 5 of Atf 2, the gene for ATF-2, a basic-region leucine zipper (bZIP) transcription factor family member (Figure 4A). Relative to transcription of Atf 2, the Tg construct is directed in the opposite orientation. Analysis of the 3′ integration site indicated a fusion of the Tg construct and intron 10. At this end, the Tg construct is oriented in the same direction as Atf 2. PCR of genomic DNA from Tg mice with primers derived from Atf 2 and the transgenic construct confirmed both integration sites into Atf 2 (Figure 4B). Thus, multiple copies of the Tg construct have inserted into Atf 2.

The insertion of the Tg construct into the Atf 2 gene suggested that the NK cell–deficient phenotype could be caused by (1) Atf 2 haplo-insufficiency, in which the phenotype arises due to the absence of one allele of Atf 2; (2) ATF-2 deficiency; or (3) abnormal Atf 2 transcripts. These possibilities are not mutually exclusive.

To address these issues, we studied mutant mice with targeted disruption of Atf2 genes.^21,22  We initially analyzed mutant mice heterozygous for disrupted ATF-2 allele (ATF-2^+/-) with a deletion in the DNA binding domain, since mice homozygous for this null mutation show perinatal lethality.²¹  It was possible that a similar phenotype could be observed in ATF-2^+/- mice, because our Ly49A Tg mice display the NK cell–deficient phenotype when hemizygous for the Tg construct. However, ATF-2^+/- mutants had normal NK-cell development with respect to both number and surface phenotype of NK cells (data not shown). Previous studies also have described another Atf2 mutant mouse with a hypomorphic allele. Homozygous Atf 2-hypomorphic mutant mice display markedly lower expression of ATF-2, have reduced viability, and show abnormalities in skeletal development.²²  These mice also have normal NK-cell number and mature marker phenotype (data not shown). Thus, neither Atf 2 haplo-insufficiency nor hypomorphic ATF2 account for the defect in Tg mice.

To further test the possibility that complete ATF-2 deficiency could account for the Tg NK-cell defect, we made chimeric mice by transferring fetal liver cells isolated from (day 15 to 17) embryo homozygous or heterozygous for the null allele of Atf 2²¹  or WT embryo into congenic WT (Ly5.1) mice. In every chimeric mouse, more than 95% of NK cells were donor derived at the time of analysis (data not shown). The frequencies and phenotypes of NK cells were similar regardless of donor cell genotypes (data not shown). Thus, the Tg phenotype is not explained by an associated Atf2 deficiency.

Another possible explanation is that there may be truncated forms of ATF-2 ectopically produced in Tg NK cells that could perturb transcription more broadly because ATF-2 forms heterodimers with other bZIP transcription factors in addition to forming ATF-2 homodimers.²⁸  Consistent with this possibility, our Tg construct integrated just upstream of exons 11 to 14 encoding the ATF-2 domains known to mediate dimerization and DNA binding, and the Tg construct at the 3′ end is oriented in the same transcriptional direction as Atf 2. To test this possibility we performed a series of RACE in the 3′-direction using primers corresponding to sequences of different ATF-2 exons and sequenced their products (data not shown). Although this approach was unbiased, it was compromised by the presence of the normal Atf 2 allele in the Tg mice. Nevertheless, we obtained frequent abnormal transcripts, consisting of Atf 2 intron 10 and only the dimerization and DNA binding domains (exons 11 to 14). We independently confirmed this observation using RT-PCR with primers for Atf 2 intron 10 and exon 12 (Figure 4C). While we were unable to detect a novel form of ATF-2 protein with currently available antibodies specific for the known C-terminus (data not shown), these data demonstrated that truncated versions of ATF-2 transcripts encoding only dimerization and DNA binding domains were expressed in Tg NK cells.

Regardless of the precise mechanism by which Tg insertion in Atf 2 causes the NK-cell defect, it is possible that the Tg insertion could affect either the NK-cell precursor or the BM microenvironment. To address these issues, we made BM chimeric mice by transferring WT congenic (Ly5.1) BM cells into Tg mice (Ly5.2) or vice versa. Eight to twelve weeks after transplantation, the proportions of donor-derived NK cells in spleens of both chimeric mice were more than 90% (data not shown). When transferred into Tg hosts, WT BM yielded normal numbers of NK cells (Figure 5A). In contrast, the transfer of Tg BM into WT mice yielded lower numbers of NK cells with low cytotoxicity. Furthermore, the NK cells in these chimeric mice showed a phenotype identical to that of unmanipulated Tg NK cells with low Mac-1 expression (Figure 5A). Moreover, splenic natural killing of YAC-1 and RMA-S tumor targets was restored when Tg mice were reconstituted with normal BM cells, while transfer of Tg BM cells resulted in defective natural killing (Figure 5B). These results indicate that Tg mice have an intrinsic developmental defect in the NK-cell lineage while possessing a microenvironment that is able to support normal NK-cell development.

Herein we characterize an NK-cell developmental defect in our Ly49A Tg mice that we previously described as having a selective deficiency in peripheral NK cells. The current data indicate that our hemizygous Tg mice have a developmental block in NK-cell development in the BM that is independent of the function of Ly49A as an MHC class 1–specific receptor because the phenotype was present regardless of whether host MHC class 1 ligands were absent (β2m-deficient) or a strong ligand (H2D^d) for Ly49A was expressed. Furthermore, 2 other Ly49A transgenic mice have been generated with expression constructs using a ubiquitously active (MHC class 1)²⁶  or CD2 promoter.²⁷  In these mice, NK-cell defects, other than functions attributable to Ly49A as an MHC class 1–specific inhibitory receptor, were not described. Instead, we found that our Tg mice manifested a cell-intrinsic defect in NK-cell development, as shown by the results from BM chimeric mice.

The phenotypic abnormalities in our Ly49A Tg mice include expression of markers associated with committed NK cells entering a terminal stage of NK-cell development in the BM, for instance, the Tg NK cells fail to express Mac-1 and CD43, markers associated with functional maturation. Interestingly, granzyme A is detectable early in NK-cell development, as early as stage 2 (data not shown), though the developmental block in our Tg mice appears somewhat after the stage. Consistent with the developmental block at this stage, there was an accumulation of cells characteristic of the prior stage, that is, Mac-1^lo NK1.1⁺ cells with poor killing and IFNγ production capabilities.⁴  The few Tg NK cells present in the peripheral lymphoid tissues also have this phenotype, indicating that the Tg NK cells are arrested at this stage of development. The lower number of peripheral NK cells in the Tg mice appears to be a consequence of this block in development since there is no clear evidence of cell death as a cause (data not shown). The accumulation of immature NK cells in the BM thus suggests the possibility that NK cells transit from Mac-1^lo to Mac-1^hi stage in BM, then acquire properties for preferential emigration to the periphery. There may be some ability of immature cells to emigrate since the absolute number of splenic Mac-1^lo NK cells in our Tg mice is comparable to or slightly higher than that in WT mice. Nevertheless, the finding of Tg NK cells blocked at this stage adds strong support to the normal presence of this NK-cell intermediate stage as originally proposed from in vivo analysis of normal developing NK cells in the BM.⁴ 

The defect in our Ly49A Tg mice is genetically linked to the insertion of the Tg construct within one allele of Atf 2 with concomitant deletion of 4 Atf 2 exons. This insertion should produce a null allele. Indeed, we have not been able to produce homozygous Tg mice, consistent with perinatal lethality of Atf 2-null mice. However, the Tg NK cells express transcripts that encode truncated ATF-2 molecules lacking the transactivation domain. Since ATF-2 molecules can form homodimers, it was possible that dimerization of normal ATF-2 molecules with truncated ATF-2 molecules could be repressor or “dominant-negative” forms and lead to secondary deficiency of ATF-2 even in the hemizygous Tg mice. Inasmuch as the Atf 2-null mice suffer from perinatal lethality,²¹  before most peripheral NK cells develop,²⁹  it was necessary to examine BM chimeric mice, rather than otherwise unmanipulated gene-targeted mice. We found no obvious abnormality of ATF-2-deficient NK cells compared with WT-derived NK cells in BM chimeric animals. Furthermore, there was no defect in NK cells derived from mice with a hypomorphic targeted allele of Atf2.²²  In addition, ATF-2 was reported to play a role in Ly49A expression in EL4 cells,³⁰  but we did not find any significant alteration in Ly49A expression on Atf 2-null NK cells (data not shown). Thus, the NK-cell defect in our Tg mice is not due to a deficiency of ATF-2 per se, despite the insertion of the Tg construct into Atf 2.

ATF-2 is known to dimerize with other bZIP family members, raising the possibility that truncated ATF-2 molecules may be repressor forms that dysregulate the function of its partner chains. Although the bZIP family is quite large (∼55 members in human), recent protein-protein interaction studies of human bZIP proteins indicate that ATF-2 shows selective heterodimerization with only c-Jun, JunB, JunD, c-Fos, Fos-B, Fra2, ATF-3, and C/EBPγ.²⁸  The mouse forms of these human proteins thus become prime candidates for being repressed by partnering with the truncated ATF-2 molecules, though further exploration will be required to determine if mouse ATF-2 possesses similar rules for pairing with other mouse bZip molecules. Regardless, C/EBPγ-deficient NK cells are defective in cytotoxicity and IFNγ production, reminiscent of our Tg phenotype, although C/EBPγ-deficient NK cells are normal in number, and other phenotypic analysis of marker expression has not been described.³¹  Taken together, these observations strongly suggest that specific members of the bZIP family may be important in lineage development of NK cells.

Interestingly, molecules belonging to the ATF/CREB family (ICER, pmLY2, I-CREB, CREB-2)^32-36  have been described in normal tissues as containing the DNA binding domains and functioning as repressors. These transcripts are generated by alternative splicing of related ATF/CREB family genes or by a second inducible promoter in a larger full-length gene. For example, ICER transcripts are derived from a promoter found in the intronic segment just upstream of first exon in the dimerization domain in the CREM gene. Such a promoter would be located just upstream of the 3′ integration site of our Tg construct in Atf 2, raising the possibility that our Tg phenotype may be related to dysregulation of a normally expressed repressor form of ATF-2. Indeed, a previous study demonstrated the presence of a truncated ATF-2 transcript (pmLY1) in normal mouse thymocytes and a similar transcript also was found in normal human thymus (phLY12).³⁵  These transcripts encode ATF-2 molecules that lack the N-terminal activation domain but contain the dimerization domain, indicating that such potential repressor forms may be expressed during physiologic development. Despite the close developmental relationship between NK and T cells, however, there was no apparent developmental or functional impairment in the T-cell compartment in our Tg mice. Furthermore, the ATF-2 knockout (ko) did not show an NK-cell phenotype even though the ko cassette also should interrupt the expression of this putative repressor. Nevertheless, it is intriguing to speculate that our Tg mice also may illuminate a normal developmental pathway in which ATF-2 repressor forms may be related to development of NK- and T-cell precursors.

Recent studies of mice deficient in 2 different transcription factors, GATA-3 and T-bet, have shown that these transcription factors play important roles in NK-cell maturation.^37,38  Interestingly, while GATA-3 belongs to a different transcription factor family that is not known to interact with bZIP family molecules, GATA-3–deficient NK cells show a phenotype similar to our Tg NK cells in terms of diminished Mac-1 and CD43 expression with decreased functional capacity to produce IFNγ.³⁷  The GATA-3 deficiency appears to affect a more mature stage because cytotoxicity is not impaired. Furthermore, GATA-3 deficiency leads to a selective defect in NK-cell migration to the liver, in contrast to our findings here, where NK cells are decreased in all peripheral organs and accumulate in the BM. Although T-bet belongs to yet another transcription factor family, detailed studies of T-bet deficiency also indicate a profound developmental effect on NK/T cells with a more subtle effect on NK cells.³⁸  T-bet deficiency affects a stage of NK-cell development akin to the current studies though the effect on NK-cell numbers, and cytolytic capacity is much more limited than what we describe here. Thus, GATA-3 and T-bet may regulate NK-cell differentiation in a manner that is related to that altered in our Tg mice, suggesting that a major step in NK-cell differentiation is affected by the interplay of GATA-3, T-bet, and bZIP transcription factors.

Finally, a detailed molecular dissection of the processes regulating the ordered progression of development of cells committed to the NK-cell lineage will have clinical implications because patients have been described with selective NK-cell deficiencies.^39-41  While rare, these patients suffer from recurrent viral infections, suggesting that other unrecognized patients may have succumbed to infection. Regardless, the molecular basis for these clinical disorders is currently unknown. Thus, molecular insight into NK-cell development, particularly of intermediate stages in lineage-committed cells, will be valuable in the clinic.

The authors thank Erika Holroyd for providing purified antibodies.