Gfi1 and Gfi1b act equivalently in haematopoiesis, but have distinct, non-overlapping functions in inner ear development

Generation of Gfi1:Gfi1b and Gfi1:P2A knock-in mice

To test the extent of functional overlap between Gfi1 and Gfi1b, we generated a mouse mutant, in which the Gfi1 coding region is replaced by Gfi1b. For this purpose, a gene-targeting vector was created in which exons 2–5 of Gfi1 were deleted and the murine Gfi1b cDNA was inserted directly 3′ to the Gfi1 translation initiation codon (Fig 1B). In this construct, 5′ regulatory sequences of the Gfi1 locus as well as exons 6 and 7 with the polyadenylation signal and the 3′ untranslated region were left intact. Several correctly targeted embryonic stem (ES) cell clones were identified by Southern blotting, and injection of these cells into mouse blastocysts was carried out (supplementary Fig 1A online). The functionality of the newly introduced Gfi1:Gfi1b knock-in allele was confirmed by expression analysis of Gfi1 and Gfi1b in thymocytes (supplementary Fig 1B,C online).

In addition, we generated a second knock-in mouse mutant by introducing the crippling P2A mutation into the Gfi1 locus to assess the significance of the N-terminal SNAG domain that mediates the repressor function in Gfi1 and Gfi1b. The gene-targeting construct replaced a region spanning exons 2–5 by the same sequence carrying a C to G substitution at position 4 of the Gfi1 coding sequence, with the consequence that alanine instead of proline is encoded (Fig 1C). ES cells in which a homologous recombination event occurred were identified by Southern blotting (supplementary Fig 2A online) and were injected into mouse blastocysts to generate Gfi1^P2A/P2A mutants. To exclude that remaining flox and frt sites affected the proper expression of the Gfi1:Gfi1P2A knock-in allele, we prepared thymocytes from these animals, compared the length of wild-type and knock-in transcript by reverse transcription–PCR and sequenced it (data not shown). Both transcripts had the same length and the knock-in transcript had no other mutations except the P2A exchange, indicating that the transcription of the Gfi1:Gfi1P2A allele was not disturbed (data not shown). Although it cannot be entirely ruled out that the remnant loxp and frt sites interfere with gene regulation, it is unlikely that they have an important role in our Gfi1:Gfi1P2A mutants, because we detect messenger RNA of proper length and sequence.

Expression of the Gfi1 P2A mutant protein was detectable in thymocytes of Gfi1^+/P2A mice at about the levels of Gfi1 found in wild-type (Gfi1^+/+) thymus (supplementary Fig 2B online). However, in thymocytes from homozygous Gfi1^P2A/P2A mice, the expression level of the Gfi1 mutants was significantly higher than wild-type levels (supplementary Fig 2B online). This upregulation of expression is most probably because of a loss of Gfi1 autoregulation in Gfi1^P2A/P2A mutants. It has been shown in vitro that Gfi1 is able to repress its own promoter and that this activity depends on the presence of the SNAG domain (Doan et al, 2004; Yücel et al, 2004). Hence, mutation of the SNAG domain should lead to an upregulation of the Gfi1 promoter in the absence of other intact Gfi1 proteins. Therefore, the elevated Gfi1 expression levels in thymocytes of Gfi1^P2A/P2A mutants illustrate well the loss of repression activity of the P2A knock-in allele, and allow to draw the first conclusion that, not only in transfected cells but also in vivo, an intact SNAG domain is required for the function of Gfi1 as a transcriptional repressor.

Gfi1 domain requirement in haematopoiesis

Both mutants were used to further test the significance of the SNAG and intermediary domains in Gfi1 and to determine whether Gfi1b can replace Gfi1. As the development of pre-T-cells and granulocytes is severely affected in mice lacking Gfi1, we scrutinized both cell types in thymus and bone marrow. Comparison of thymocyte numbers of Gfi1^+/+, Gfi1^−/− and Gfi1^1b/1b thymocytes indicated that Gfi1b was able to almost fully replace Gfi1, whereas the Gfi1^P2A/P2A mutant showed a significant loss of thymic cellularity, similar to the Gfi1 null mutants (Fig 2A). The reduced number of thymocytes in Gfi1-deficient mice had previously been shown to be caused by defects in early, uncommitted c-Kit-positive T-cell progenitors (Yücel et al, 2003). Whereas wild-type mice showed about 67% c-Kit⁺ cells among the CD4⁻/CD8⁻ pre-T-cell subset, Gfi1^−/− animals showed a significantly reduced frequency (25%) of this subpopulation, as expected (Fig 2B). Replacement of Gfi1 by Gfi1b almost completely restored the frequency of c-Kit⁺ double-negative cells to wild-type levels, whereas the Gfi1^P2A/P2A mutant remained at low knockout levels (Fig 2B). In contrast to previous studies (Doan et al, 2003) in which a unique function of Gfi1 and Gfi1b was proposed, the results presented here indicate that Gfi1b, which normally is absent or at least not readily detected in thymus (supplementary Fig 1C online), could replace Gfi1 almost completely in the development of the T-cell lineage. Nevertheless, it cannot be entirely ruled out that some aspects of T-cell differentiation are differentially regulated by the two proteins. However, it has to be concluded that an intact SNAG domain is absolutely necessary for the function of Gfi1 as a regulator of T-cell development, because the Gfi1^P2A/P2A mutant showed the same defect as the complete knockout. As Gfi1b was able to replace Gfi1 almost fully in pre-T-cell development, it is likely that the intermediary domain is dispensable for the function of Gfi1 in the development of pre-T-cells.

The other main defects documented in Gfi1-deficient mice were a severe neutropenia and a pronounced increase in cells with monocytoid features (Karsunky et al, 2002; Hock et al, 2003). In bone marrow, Gfi1^1b/1b knock-in mice had slightly reduced frequencies of granulocytes and showed a mild accumulation of cells with the immunophenotype of immature myeloid cells or monocytes (Mac-1^lo, Gr-1^lo) as compared with wild-type mice (Fig 2C). Although this phenotype resembled more the wild-type configuration than the drastic effects seen in the Gfi1 knockout mouse, a complete rescue in terms of cell frequencies was not achieved by Gfi1b. Nevertheless, the typical signs that accompany the loss of neutrophils, as seen in Gfi1 null mice, such as shortened lifespan, granuloma formation or an inherent predisposition to infections (Karsunky et al, 2002; Hock et al, 2003), were not observed in Gfi1^1b/1b knock-in mice. In peripheral blood, Gfi1^1b/1b knock-in and wild-type animals showed a similar distribution of granulocytes (Mac-1^hi/Gr-1^hi) or monocytes (Mac-1⁺; Fig 2C, lower row). Neutropenia or a noticeable expansion of monocytoid cells, which was clearly present in the Gfi1-deficient mice, could not be detected in Gfi1^1b/1b knock-in mice (Fig 2C), suggesting a full rescue of neutrophil development by Gfi1b. In sharp contrast, Gfi1^P2A/P2A knock-in mice showed all phenotypes of the Gfi1 null mouse, such as severe neutropenia and pronounced increase in monocytoid cells in all tissues examined (Fig 2C).

Blood smears and cytospins from bone marrow indicated that the Gfi1^P2A/P2A mutant showed the Gfi1 null phenotype, in that almost no mature neutrophilic granulocytes (cells with clearly segmented nuclei) were found and a large population of atypical monocytoid cells that is not detected in the wild-type controls (Fig 3A–C; supplementary Table 1 online). These cells resembled monocytes in that they had greyish-blue, often vacuolated cytoplasm and an indented, bean-shaped or lobed nucleus, but they were also reminiscent of metamyelocytes, frequently exhibiting a nuclear hole (Fig 3C). In both Gfi1 null and Gfi1^P2A/P2A mice, a small percentage of almost mature appearing neutrophilic granulocytes with subtle nuclear abnormalities (abortive segmentation, plump segments; Fig 3B,C) was detectable. Concurrent with this and the previously described Gfi1 null phenotype, the proportion of lymphocytes was reduced in both blood and bone marrow of Gfi1^P2A/P2A mutant to the same degree as in the full Gfi1 knockout (supplementary Table 1 online). This suggests that an intact SNAG domain is an absolutely essential domain in Gfi1 for neutrophil development.

In contrast, the Gfi1^1b/1b knock-in mutants were almost indistinguishable from the wild-type control, both in the differential cell count of the peripheral blood smear (supplementary Table 1 online) and in the appearance of the bone marrow cytospins, which showed regular maturation within the granulocytic lineage up to the stage of a mature neutrophil with a segmented nucleus (Fig 3A,B; supplementary Table 1 online). However, the proportion of mature granulocytes in the peripheral blood seemed to be slightly reduced in the Gfi1^1b/1b knock-in mice as compared with the wild-type controls, and, in accordance with the fluorescence-activated cell sorting analysis, in the bone marrow a left shift towards more immature granulocytic precursors was observed (Fig 2C; supplementary Table 1 online). In addition, a few ‘atypical monocytes', typical for Gfi1 null mice, were also present in the Gfi1^1b/1b knock-in mice (Fig 3A). Nevertheless, Gfi1^1b/1b knock-in mice produced mature granulocytes in numbers that vastly exceeded those found in Gfi1 null mice (Figs 2C, 3A–C; supplementary Table 1 online). This indicated that the intermediary domain in Gfi1 is not required for the proper formation of mature, differentiated neutrophils, but may be required for proliferative expansion of granulocytic/monocytic cell types, which assures the high frequencies seen in wild-type mice.

The ‘atypical monocytes' that emerge in Gfi1 null cells and in Gfi1^P2A/P2A mutants were clearly a heterogenous population, rarely resembling aberrant neutrophils, more frequently reminiscent of atypical metamyelocytes and sometimes evoking more immature granulocytic precursors (Fig 3C). These observations suggest that a loss of Gfi1 or expression of a Gfi1 with a mutated SNAG domain maintains granulocytic maturation up to the myelocyte stage only and then leads to the development of aberrant forms that cannot be readily classified because the cells show features of both atypical metamyelocytes and atypical monocytes. The introduction of a Gfi1b gene was sufficient to largely rescue the loss of Gfi1, supporting the view that the intermediary domain of both proteins is less relevant for neutrophil differentiation, whereas an intact SNAG domain is essential.

Gfi1 and Gfi1b functions diverge in inner ears

It has previously been described that Gfi1 null mice show deafness and vestibular dysfunction (Wallis et al, 2003). Hearing loss in these animals is caused by late maturation defects and degeneration of inner ear hair cells beginning already before birth (Wallis et al, 2003; Hertzano et al, 2004). Gfi1^P2A/P2A mice presented with the full Gfi1 knockout phenotype, including hearing loss, assessed by Preyer reflex and vestibular dysfunction (data not shown). Surprisingly, however, the inner ear defect could not be compensated for by Gfi1b, in contrast to all other Gfi1 deficiency phenotypes tested here. Gfi1^1b/1b mice were clearly deaf, because they did not show a Preyer reflex, and no reaction to sound could be recorded by auditory brainstem response measurements in mice of 3–3.5 months of age (Fig 4A). They also showed an overt vestibular dysfunction manifested by head bobbing, an abnormal reaching response and a positive swimming test (data not shown), similar to Gfi1^−/− mice. To exclude that this effect was due to a lack of expression of the Gfi1^1b/1b knock-in allele, we isolated RNA from inner ears and tested Gfi1b expression by real-time PCR. Clearly, a substantial expression of Gfi1b noticeably higher than wild-type levels was detected (supplementary Fig 3 online), demonstrating the functionality of the knock-in allele. In addition, we had previously shown that a Gfi1:GFP knock-in allele that was constructed in an analogous manner to the Gfi1:Gfi1b knock-in allele was properly expressed in inner ear hair cells (Hertzano et al, 2004; Yücel et al, 2004), which would also exclude that a lack of Gfi1b expression could account for the observed phenotype in Gfi1^1b/1b mice.

Whole-mount immunohistochemistry of inner ears from newborn mice stained with myosin VI, a hair cell marker (Fig 4B–E), or with an antibody for acetylated tubulin that labels hair cell kinocilia, pillar cells and other tubulin-based structures in the cochlear sensory epithelium (Fig 4F–I), demonstrated a typical pattern of a single row of inner hair cells (IHCs) and three rows of outer hair cells (OHCs) in wild-type cochlear epithelia (Fig 4B,F). However, only the single row of IHCs was clearly seen in the cochlear epithelia from Gfi1^−/−, Gfi1^P2A/P2A and Gfi1^1b/1b mice (Fig 4C–E,G–I), and it appeared morphologically immature when compared with the wild-type littermates. The region where OHCs develop was disorganized in both Gfi1^1b/1b and Gfi1^P2A/P2A mutants. The hair cells appeared smaller, the stereociliary bundles were hardly identifiable and the overall myosin VI staining (a hair cell marker) was diffuse, suggesting a continuing degenerative process in that region. The phenotype in the OHC region seemed, however, more pronounced in the Gfi1^−/− and Gfi1^P2A/P2A mice than in the Gfi1^1b/1b knock-in mice (Fig 4C–E,G–I), suggesting that the OHCs might degenerate at a slightly slower pace when Gfi1b is expressed instead of Gfi1. The results indicate clearly that Gfi1b cannot compensate for the loss of Gfi1 in the cochlear hair cells as it apparently can in haematopoietic cells. However, as both IHCs and OHCs of the Gfi1^1b/1b mice seemed to be slightly less severely affected compared with the Gfi1^−/− mice and as it has not been evaluated whether the Gfi1b knock-in allele is expressed with exactly the same spatiotemporal expression pattern as Gfi1 in the inner ear during fetal development, it cannot be entirely ruled out that some functions of Gfi1 that are necessary for hair cell survival can be taken over by Gfi1b. Despite such a potential partial compensation, Gfi1b was apparently not able to prevent hair cell death in newborn mice.

Speculation

Our results indicated a cell type-specific requirement of the intermediary domains in Gfi1 and Gfi1b. Until now, the only protein that interacts with Gfi1 through this intermediary domain is the STAT3 inhibitor PIAS3 (Rödel et al, 2000). We therefore assessed whether Gfi1b is unable to interact with PIAS3, as a basis for the functional divergence of both proteins in inner ear hair cells. Indeed, co-immunoprecipitation experiments with extracts from cells expressing high levels of Gfi1b demonstrated that Gfi1b does not physically interact with PIAS3 (supplementary Fig 4 online). Because STAT3 is expressed in the same hair cell subtype (OHC) as Gfi1 (Hertzano et al, 2004), it is possible that Gfi1, but not Gfi1b, could promote OHC differentiation or survival by sequestering PIAS3 and enhancing STAT3 signalling (Hertzano et al, 2004). Therefore, PIAS3 could be the factor responsible for mediating the cell type specificity of both Gfi1 and Gfi1b. However, several alternative possibilities exist, and it is also likely that the intermediary domain mediates the differential abilities of Gfi1 and Gfi1b to associate with specific cofactors through which accessibility to target genes or the ability to modify chromatin constituents is conferred in a cell type-specific manner.