Ptf1a is essential for the differentiation of GABAergic and glycinergic amacrine cells and horizontal cells in the mouse retina

The mammalian retina consists of six major neuronal cell types (cone photoreceptors, rod photoreceptors, horizontal cells, bipolar cells, amacrine cells and ganglion cells) and one type of glia (Mueller cells) that originate during retinogenesis from a common population of undifferentiated multipotent retinal progenitor cells (RPCs)(Marquardt, 2003). Both the chronological sequence of cell generation and the spatial expansion of these diverse cell types are strictly regulated. In mice, retinal cell differentiation starts with the generation of ganglion cells at embryonic day 11.5 (E11.5), followed in a temporally overlapping sequence by horizontal cells, cone photoreceptors, amacrine cells, rod photoreceptors, bipolar cells,and finally Mueller glia cells (Marquardt,2003). Retinogenesis in mice is completed at approximately postnatal day 10 (P10). These functionally different cells form three cellular(nuclear) layers and two synaptic (plexiform) layers in the mature retina. The outer nuclear layer (ONL) contains rod and cone photoreceptor cell bodies. The inner nuclear layer (INL) is comprised of horizontal, bipolar, amacrine and Mueller cell bodies. The ganglion cell layer (GCL) includes both nuclei of displaced amacrine cells and ganglion cells. The synaptic connections of these neuronal cells are localized in the outer plexiform layer (OPL) and the inner plexiform layer (IPL).

The molecular basis of pattern formation and cell-type specification in the vertebrate retina has been intensively investigated. Accumulating evidence indicates that cell-intrinsic regulators, such as transcription factors, and cell-extrinsic signals, such as neurotrophic factors, play important roles in progenitor cell fate determination and their subsequent differentiation(Malicki, 2004; Yang et al., 2003). Recent studies have demonstrated that basic helix-loop-helix (bHLH) transcription factors regulate the determination and differentiation of multiple neuronal cell types during retinogenesis (Akagi et al., 2004; Hatakeyama and Kageyama, 2004; Vetter and Brown, 2001). The gene Ptf1a (also known as PTF1-p48) encodes a bHLH protein of 48 kD. Ptf1a, together with two proteins [RBP-L and a common (type I) bHLH protein] comprises the three subunits of the pancreas transcription factor (PTF1)(Roux et al., 1989; Beres et al., 2006). Ptf1a plays a fundamental role in exocrine and endocrine pancreas development in mice (Kawaguchi et al.,2002). In a recent study the expression of Ptf1a has also been detected in the retina of developing zebrafish embryos by in situ hybridization (Zecchin et al.,2004). Using a positional candidate gene approach, Sellick et al.(Sellick et al., 2004) have identified mutations in the Ptf1a gene of patients with pancreatic and cerebellar agenesis, and Hoshino and colleagues(Hoshino et al., 2005) could demonstrate that Ptf1a is involved in GABAergic neuronal cell specification in the cerebellum. Very recently, it has been shown that Ascl1 (previously Mash1) controls the expression of Ptf1a. Furthermore, Mash1 is expressed in sensory interneuron progenitors and is involved in the switch between excitatory and inhibitory cell fates in the developing mouse spinal cord(Mizuguchi et al., 2006).

In the present study, we show that the bHLH transcription factor Ptf1a is expressed in the neuroretina of developing mice. Furthermore, inactivation of the Ptf1a gene produces severe cellular defects of the inner retina as a result of inhibition of differentiation of GABAergic and glycinergic amacrine precursor cells and horizontal precursor cells.

The Ptf1a locus was targeted with a vector replacing part of exon 1 (ex1) with the Cre recombinase and neo-resistance genes. Following removal of the neo-resistance gene by transient transfection of a plasmid encoding the FLP recombinase, embryonic stem (ES) cells were injected into C57BL/6 E3.5 blastocysts. The integration of Cre at the ATG-start codon of the Ptf1a gene was verified directly by DNA sequencing. All procedures were approved by the local animal care committee.

The Gt(ROSA)26Sortm1Sor (R26R) reporter mouse line was purchased from The Jackson Laboratory (Bar Harbor, Maine, USA)and the Tg(ACTB-Bgeo/GFP)21Lbe (Z/EG)reporter mouse line was kindly provided to us by Dr Mori Tetsuji, Institute of Stem Cell Research, GSF-National Research Center for Environment and Health(Soriano, 1999; Novak et al., 2000). Heterozygous R26R and Z/EG mice were crossed with Ptf1a-Cre(ex1)(Ptf1a^+/Cre(ex1))to generate Ptf1a^+/Cre(ex1)R26R and Ptf1a^+/Cre(ex1)Z/EG mice.

All offspring were genotyped by PCR of genomic DNA from mouse tail clips with primers specific for the Ptf1a, Cre, R26R and Z/EGtransgenes as described by Krapp et al.(Krapp et al., 1996), Gu et al. (Gu et al., 1993), Soriano(Soriano, 1999) and Novak et al. (Novak et al., 2000)respectively.

Genomic DNA from ES clones were digested with SacI, separated by agarose gel electrophoresis, blotted to nylon membranes, and detected with the radioactive labeled 5′ flanking external probe.

Real-time PCR was performed in a Perkin-Elmer 7700 Sequence Detection System. Total RNA was isolated by RNeasy Mini Kit (Qiagen) according to the manufacturer's manual. Two micrograms of total RNA was first reverse transcribed with SSIII (Invitrogen) and random primer in a total volume of 20μl for 2 hours at 50°C. SSIII was inactivated by heating at 75°C for 15 minutes. The cDNA was diluted fivefold, and 5 μl was used for each 30 μl PCR using the SYBR GREEN PCR Master Mix (Applied Biosystems). The Taqman primers were designed using Primer Express Software (Applied Biosystems). The primer sequences are listed in Table 1. The PCR conditions for all genes were as follows: 50°C for 2 minutes hold, 95°C for 2 minutes hold and 40 cycles of 95°C, for 15 seconds and 60°C for 30 seconds. For each gene, the real-time PCR assay was performed four times with four different batches of total RNA. The cyclophilin gene served as an RNA input control. Relative gene expression ratio (ER) was calculated on the basis of differences between expression level of cyclophilin and genes analyzed using the following formula: ER=2^-ΔCt, whereΔCt is the difference of threshold cycles between the gene of interest and the control gene (cyclophilin).

Eyes were isolated from E18.5 embryos. Dissected explants of neural retina were placed into a Millicell CM chamber (Millipore), with the ganglion cell layer upward. These retinal explants were cultured in 50% DMEM (Gibco)supplemented with HEPES, 25% Hank's solution, 25% heat-inactivated horse serum, 200 μmol/l L-glutamine and 6.75 mg/ml glucose at 34°C in a 5%CO₂ incubator (Hatakeyama and Kageyama, 2002).

Sections were air-dried for 20 minutes, fixed in paraformaldehyde, washed in PBS and stained with X-Gal reaction buffer (2 mmol/l MgCl₂, 5 mmol/l potassium ferrocyanide, 5 mmol/l potassium ferricyanide, 0.5 mg/ml X-Gal in PBS) at 37°C overnight.

Pregnant mice were injected with 0.14 mg/g body weight BrdU 2 hours before they were sacrificed. BrdU incorporation was detected on 10 μm cryostat sections by antibody directed against BrdU (Sevotech), and at least four different eyes from wild-type embryos were analyzed.

After enucleating the mouse eyes, we quickly removed the anterior segment(cornea and lens) leaving half eyecups. We analyzed sections, which were through the central retina. Histological analysis of retinal explant cultures was as previously described (Hatakeyama and Kageyama, 2002). Tissues and whole mounts were fixed in 4%paraformaldehyde for 30 minutes on ice, equilibrated overnight in 30% sucrose at 4°C, frozen in OCT compound (Leica) and stored at -80°C. Tissues were sectioned (8-10 μm) with a cryostat (MICROM, Laborgeraete GmbH,Walldorf, Germany).

Immunohistochemical and immunofluorescence analyses were carried out as previously described (Hsu,1990). Briefly, sections were incubated with cell-type-specific antibodies in blocking serum overnight at 4°C. We used the following primary antibodies: anti-Ptf1a (a kind gift from R. J. MacDonald), anti-Thy1.2(CD90; Promega), anti-syntaxin HPC 1 (Sigma), anti-γ-amino butyric acid(GABA; Sigma), anti-glycine transporter 1 (GlyT1; Chemicon), anti-Lhx1(anti-Lim1; Chemicon), anti-Hu/D (Elavl4; Molecular Probes), anti-Brn3 (clone C-13; Santa Cruz) anti-Pax6 (Developmental Studies Hybridoma Bank),anti-N-cadherin (Transduction Laboratories), anti-recoverin (Chemicon),anti-β-galactosidase (from rabbit; ICN), anti-β-galactosidase (from mouse; Developmental Studies Hybridoma Bank), anti-GFP (Molecular Probes),anti-calbindin-D28K (Sigma), anti-PKCα (Zymed) and biotin-conjugated secondary antibodies (Dianova), as well as the avidin-biotin-peroxidase and avidin-biotin-alkaline phosphatase complex (Vectorstain ABC and ABC-AP),according to the manufacturer's instructions (Vector Labs). Antibodies were detected by BCIP (5-bromo-4-chloro-3-indolyl phosphate) and NBT (nitro blue tetrazolium), TSA (Tyramide Signal Amplification) (Molecular Probes). The cell nuclei were stained with DAPI(4′,6-diamidine-2-phenylindol-dihydrochloride; Boehringer, Mannheim). Tunnel assay was performed using the In Situ Cell Death Detection Kit TMR red(Roche Diagnostics) following the manufacturer's protocol, except that all sections were counterstained with DAPI. At least three to five independent retinae from knockout and wild-type mice were collected and analyzed. For each retina, positive cells were counted under the microscope from four different sections in an area extending from the outer segments of the photoreceptor cells (osp) to the inner limiting membrane (ilm). The length of osp and ilm was 200 μm. Images were acquired using a Zeiss ApoTome microscope (Axiovert 200M) and captured by a CCD cool digital camera (Zeiss). Software module AnxioVision (Zeiss) was used for subsequent 3D reconstruction of the images.

Data from quantitative PCR and immunohistochemistry were analyzed by Student's t-test or Mann-Whitney U-test, as appropriate. A P-value of <0.05 was considered statistically significant. All analyses were performed using SPSS 12.0 software (SPSS, Chicago, IL).

To study the function of the Ptf1a gene in the developing mouse,we generated a Ptf1a-Cre knock-in mouse line by homologous recombination in ES cells. To prevent a loss of potential regulatory elements in the first intron of Ptf1a, we replaced precisely only the coding sequence in exon1 (ex1) with the coding region of a Cre recombinase targeted to the nucleus (Gannon et al.,2000) and obtained a Ptf1a null allele[Ptf1a^tm1(Cre)Nakor Ptf1a-Cre(ex1)] (see Fig. S1 in the supplementary material). After verifying germline transmission, we crossed Ptf1a^+/Cre(ex1)heterozygous mice with Gt(ROSA)26Sortm1Sor(R26R) or Tg(ACTB-Bgeo/GFP)21Lbe(Z/EG) reporter mouse lines(Soriano, 1999; Novak et al., 2000) to determine Cre activity. The R26R mice carry a modified lacZgene driven by the cell-type-independent ROSA26 promoter and the Z/EG reporter mouse line contains a CMV enhancer/chicken β-actin promoter. In Ptf1a^+/Cre(ex1)R26R animals, Ptf1a-driven expression of Cre excises a stop cassette upstream from the lacZ gene and thereby activates the expression of β-galactosidase. Expression of β-galactosidase therefore marks all the cells in which Ptf1a is activated and the active β-galactosidase expressing locus is inherited by their descendant cells. These cells can be efficiently detected by staining with 5-bromo-4-chloro-3-indolyl-β-D-galactoside (X-Gal). In offspring from the cross between Ptf1a^+/Cre(ex1)and Z/EG mice line, lacZ expression is replaced with the enhanced green fluorescent protein (GFP) expression in cells expressing Cre.

Examination of Ptf1a^+/Cre(ex1)R26R mice, in which all cells carry a Ptf1a heterozygous null allele, revealed that the Ptf1a-Cre-mediated R26Rrecombination and the subsequent activation of β-galactosidase expression was not restricted to the developing pancreas(Kawaguchi et al., 2002),neural tube (Obata et al.,2001) and cerebellum (Hoshino et al., 2005; Sellick et al.,2004) (see Fig. S2A,B in the supplementary material). Moreover, we detected β-galactosidase-expressing cells in the neuroretina of developing mice (see Fig. S2C-F in the supplementary material). These X-Gal positive cells were first detectable at E14.5 (see Fig. S2C,E in the supplementary material). At this early stage of embryonic development, the mouse retina consists of an inner and an outer neuroblastic layer (NBL), which contains RPCs (Dyer et al.,2003). X-Gal labeled cells of the Ptf1a lineage were located within the inner NBL. At E18.5, β-galactosidase expressing cells were primarily found in the inner NBL of the developing retina. Small numbers of X-Gal positive cells were also detected in the GCL and the outer region of the neuroretina (data not shown). In adult mice, most of the cells expressingβ-galactosidase were localized in the innermost zone of the INL. Someβ-galactosidase-expressing cells were found in the GCL and in the outermost zone of the INL (see Fig. S2D,F in the supplementary material). These data led us to a detailed analysis of the expression and function of the Ptf1a gene during retinal development in mice.

To identify the cells expressing Ptf1a in the developing retina,we immunolabeled retinal cryosections of wild-type mice at embryonic and postnatal stages of development. Ptf1a-expressing cells were first detected at E12.5 (Fig. 1A,D). From E12.5 to P1, Ptf1a was expressed in the nuclei of retinal precursor cells scattered in the NBL (Fig. 1A-F). The number of Ptf1a-expressing cells increased from E12.5 to P1. After P1, the number of Ptf1a-expressing cells decreased gradually, and from P3 onward,Ptf1a expression was not discernible in the retina.

To identify the specific neuroretinal cells that are derived from Ptf1a-expressing cells in the developing mouse, we analyzed the adult retina of Ptf1a^+/Cre(ex1)R26R mice by double-labeling experiments using antibodies againstβ-galactosidase and cell-type-specific markers: Thy1.2 (ganglion cell marker), syntaxin (amacrine cell marker), GABA (GABAergic amacrine and displaced amacrine cell marker), glycine transporter 1 (GlyT1, glycinergic amacrine cell marker), calbindin (horizontal cell marker) and PKC-α(bipolar cell marker). These results showed that Ptf1a expression, as assayed by Cre-mediated β-galactosidase expression, was observed in the amacrine, displaced amacrine and horizontal cells(Fig. 2A-R).β-Galactosidase was not coexpressed with vimentin, a marker for Mueller glia (data not show). These lineage-tracing data, taken together with the immunolocalization results of Ptf1a, indicate that Ptf1a is expressed in precursors of amacrine, displaced amacrine and horizontal cells.

To determine whether cells expressing Ptf1a were actively dividing, we pulse-labeled retinae of wild-type mice with the thymidine analog BrdU for 2 hours and carried out double-immunofluorescence staining using anti-Ptf1a and anti-BrdU antibodies at E13.5 and P1. We found that neither E13.5 nor P1 S-phase cells were Ptf1a immunopositive (see Fig. S3 in the supplementary material). Thus, in summary, Ptf1a expression is initiated in the postmitotic amacrine, displaced amacrine and horizontal cell precursors.

To study the biological role of the Ptf1a gene in retinal development, we then investigated the retina of Ptf1a^{Cre(ex1)/Cre(ex1)}R26R mice. In these mice, all cells are homozygous for the Ptf1a null mutation, but the retinal cells in which Ptf1a is transcriptionally activated can be identified by their expression of lacZ derived from Ptf1a-Cre-mediated R26Ractivation. As the Ptf1a^{Cre(ex1)/Cre(ex1)}R26R pups die approximately three hours after birth, we examined the morphology of the retina in Ptf1a^{Cre(ex1)/Cre(ex1)}R26R embryos at E18.5. After verifying the absence of Ptf1a protein in Ptf1a^{Cre(ex1)/Cre(ex1)}R26R retinae (Fig. 3A,E) by immunohistochemistry, we compared cryosections of Ptf1a^+/Cre(ex1)R26R with Ptf1a^{Cre(ex1)/Cre(ex1)}R26R retinae. In retinae of heterozygous Ptf1a^+/Cre(ex1)R26R mice, the IPL formed a continuous layer between the GCL and NBL(Fig. 3B,C,D). X-Gal staining of Ptf1a^+/Cre(ex1)R26R retina showed that the majority ofβ-galactosidase-expressing cells were localized in the innermost zone of the NBL (Fig. 3B,C). By contrast, in Ptf1a^{Cre(ex1)/Cre(ex1)}R26R retina, the GCL and NBL were fused, resulting in loss of the IPL(Fig. 3F-H). We confirmed these results using immunostaining with anti-N-cadherin antibody and 3D reconstruction by ApoTome microscopy (Fig. 3D,H). In addition, the β-galactosidase-expressing cells were scattered in the inner retina of Ptf1a^{Cre(ex1)/Cre(ex1)}R26R mice (Fig. 3G). These data suggested a misplacement of β-galactosidase-expressing cells in Ptf1a^{Cre(ex1)/Cre(ex1)}R26R toward the inner retina.

To identify the cells expressing β-galactosidase in Ptf1a-deficient retinae and to further elucidate the defect in formation of the IPL in the Ptf1a-deficient retinae, we labeled retinal cryosections with the cell-specific antibodies Pax6 and Hu/D.

At E18.5, Pax6 is mainly expressed in differentiated amacrine cells and to a lower level in differentiated ganglion cells, as well as in RPCs(Marquardt et al., 2001). We found that Pax6 and β-galactosidase were coexpressed in differentiated amacrine cells of Ptf1a^+/Cre(ex1)R26R mice (Fig. 4A-C). By contrast, in Ptf1a-deficient retinae, the expression of Pax6 was restricted to cells in the NBL and showed no coexpression withβ-galactosidase (Fig. 4D-F). These data indicate that retinae of Ptf1a-deficient mice fail to generate differentiated amacrine cells.

During retinal development, the RNA-binding protein Hu/D is expressed in differentiating amacrine and ganglion cells, but not in differentiated amacrine cells (Link et al.,2000). At E18.5, immunostaining with anti-Hu/D antibody of Ptf1a^+/Cre(ex1)R26R retinae showed Hu/D expression mainly in the GCL, which was not co-localized with β-galactosidase-expressing cells(Fig. 4G-I). However, in Ptf1a-deficient retinae we found that the large majority of the cells expressing β-galactosidase also were positive for Hu/D(Fig. 4J-L). These results suggest that β-galactosidase-expressing cells are amacrine cell precursors and/or ganglion cells in Ptf1a knockout retina.

Based on these data, we hypothesized that in Ptf1a knockout retina, the β-galactosidase-expressing cells, which would differentiate to amacrine or horizontal cells in the course of normal retinal development,were either inhibited in their differentiation or transdifferentiated to ganglion cells.

To test the hypothesis that Ptf1a-deficient precursor cells transdifferentiated to ganglion cells, we analyzed the cell fate of these cells by using the Z/EG reporter mouse line, which expressesβ-galactosidase before Cre excision and enhanced GFP after Cre excision. The double immunostaining of the Ptf1a^+/Cre(ex1)Z/EG retinae with anti-GFP and anti-Brn3 showed no co-labeling(Fig. 5A-C). However, in Ptf1a^{Cre(ex1)/Cre(ex1)}Z/EG retinae, GFP and Brn3 were coexpressed in approximately 30% of the cells (Fig. 5D-F,H). Taken together, these data provide evidence that a proportion of Ptf1a-deficient precursor cells transdifferentiate to ganglion cells. In line with these results, we observed a significant increase of Brn3-positive cells in Ptf1a^{Cre(ex1)/Cre(ex1)}(Cre/Cre) compared with wild-type retinae(Fig. 5G).

To further determine the cell fate of Ptf1a-deficient precursor cells that did not transdifferentiate to ganglion cells, we examined Ptf1a knockout cells at the postnatal stage of development. As homozygous null mice for the Ptf1a gene die about three hours after birth with pancreatic and cerebellar agenesis(Hoshino et al., 2005),retinal explants were prepared from E18.5 embryos and cultured for 12 days[E18.5+12 days in culture (DIC)]. We analyzed cryosections of retinal explants from Ptf1a^+/Cre(ex1)R26R and Ptf1a^{Cre(ex1)/Cre(ex1)}R26R mice using specific antibodies against each retinal cell type(Fig. 6, Table 2).

In both heterozygous and homozygous Ptf1a null mutant retinal explants, Thy1.2 (Fig. 6A,B), a marker expressed in ganglion cells, and Hu/D(Fig. 6C,D), a marker for ganglion and amacrine cells, were present. However, the number of these cells in Ptf1a^{Cre(ex1)/Cre(ex1)}R26R retinal explants was significantly decreased.

In retinal explants of Ptf1a-deficient mice, we found almost no GABAergic (Fig. 6E,F) and glycinergic-positive cells (Fig. 6G,H). In addition, in Ptf1a^{Cre(ex1)/Cre(ex1)}R26R retinal explants, the number of X-Gal-positive cells was significantly reduced (P=0.001, Fig. 6Q,R), whereas that of apoptotic cells was significantly increased (P=0.001, Fig. 6S,T) compared with Ptf1a^+/Cre(ex1)R26R retinal explants. Therefore, we suggest that in Ptf1a-deficient retinae, the precursors of GABAergic and glycinergic amacrine cells either transdifferentiate to ganglion cells or undergo apoptotic cell death during postnatal development. If transdifferentiation occurs, it should be accompanied by an increase of X-Gal-positive cells. We detected only a few X-Gal positive cells in Ptf1a-deficient retinal explants, and we conclude that most of the precursors of GABAergic and glycinergic amacrine cells undergo apoptotic cell death rather than transdifferentiate to ganglion cells.

To further characterize the defects of Ptf1a-deficient retinae, we investigated the genesis and differentiation of the remaining retinal cells. The presence of syntaxin- and calbindin-positive cells in Ptf1a-deficient retinal explants(Fig. 6J,L) indicates that not all of the amacrine precursor cells undergo apoptotic cell death or transdifferentiate to ganglion cells. Thus, we propose that a subtype of amacrine precursor cells do not need Ptf1a for differentiation.

Furthermore, we observed a disorganization of syntaxin-positive cells in Ptf1a^{Cre(ex1)/Cre(ex1)}R26R compared with Ptf1a^+/Cre(ex1)R26R retinal explants (Fig. 6I,J). The same was true for calbindin (amacrine and horizontal cells) and PKCα (bipolar cells) (Fig. 6K-N) immunoreactive cells. Recoverin staining showed less evident changes in Ptf1a-deficient retinal explants(Fig. 6O,P).

In summary, retinal explant results show: (1) a small population of differentiated amacrine cells are present in the Ptf1a null mutant;(2) GABAergic and glycinergic amacrine cells are absent in the Ptf1a-deficient mice; (3) a disorganization of the inner neuroretina;and (4) an increase of apoptotic cells in Ptf1a knockout mice.

To consolidate our findings, we performed real-time PCR analysis on total RNA extracted from wild-type and Ptf1a-deficient retinae at E18.5 using a set of marker genes that regulate amacrine [Foxn4, Neurod1(formerly NeuroD), Neurod4 (formerly Math3), Barhl2] and horizontal [Foxn4, Lhx1 (also known as Lim1)] cell development plus glutamic acid decarboxylase 1(Gad1), an enzyme that catalyzes the synthesis of the inhibitory neurotransmitter GABA (Fig. 7I). Foxn4 controls the genesis of amacrine and horizontal cells by activating the expression of Neurod1 and Neurod4 (Li et al.,2004). The expression level of Foxn4 remained unchanged in wild-type and Ptf1a-deficient retinae, indicating that Ptf1a acts chronologically later on amacrine cell genesis than Foxn4. Both bHLH transcription factors, Neurod1 and Neurod4, are expressed in differentiating amacrine cells and regulate amacrine cell fate specification (Inoue et al., 2002). The absence of Ptf1a does not alter Neurod1 and Neurod4 expression, suggesting that the involvement of Ptf1a to amacrine cell fate specification takes place at later stages of amacrine cell development, as is the case for Neurod1 and Neurod4. By contrast, in Ptf1a-deficient retinae the expression of Barhl2 and Gad1 transcripts were downregulated, indicating that Ptf1ais involved in the molecular mechanism governing the specification of subtype identity of amacrine cells. The result that the expression level of Barhl2 in Ptf1a-deficient retinae is only slightly decreased is consistent with the fact that Barhl2 is also expressed by ganglion cells. Also in line with our immunohistochemical data was the finding that the transcription factor Lim1 (horizontal cell marker) is not detectable in Ptf1a-deficient retinae (Fig. 7II).

Using a novel Ptf1a-Cre knock-in mutation in combination with a ROSA26 reporter transgene, we could analyze the expression of Ptf1a in the mouse retina. The experiments reported here show that Ptf1a expression is first detectable at around E12.5 in the developing retina. As development proceeds, Ptf1a-positive cells spread from the centre to the inner retina and become primarily localized to the outer NBL. However, Ptf1a expression is absent in adult retina. Our cell-lineage-tracing experiments revealed that transcriptional activation of Ptf1a-Cre occurs in amacrine and horizontal precursor cells. Analysis of the cell cycle state of Ptf1a-expressing cells showed that these cells were postmitotic. Our data suggest that, in neuroretinal cells, Ptf1a expression commences at a developmental state in which these cells are specialized but not terminally differentiated. This is in line with the finding that Ptf1a expression is absent in terminally differentiated amacrine and horizontal cells. The early and cell-specific expression of Ptf1a suggests an important role for Ptf1a in inducing the differentiation of amacrine and horizontal cells.

Amacrine cells can be divided into two major nonoverlapping subpopulations classified by neurotransmitter production. In the mouse retina, GABAergic amacrine cells comprise ∼35% and glycinergic amacrine cells ∼40% of all amacrine cells (Mo et al.,2004). We presented several lines of evidence that Ptf1ais involved in the terminal differentiation of GABAergic and glycinergic amacrine cells. First, birth-dating studies of retinal cells have revealed that during mouse retinogenesis amacrine cells are generated over a time period from E11 to P4 (Young,1985). We demonstrated that Ptf1a expression starts at around E12.5 and ceases gradually at P3 in the developing retina of wild-type mice, which is concomitant to amacrine cell genesis. Second, we could show that inactivation of the Ptf1a gene leads to downregulation of genes(Barhl2, Gad1) contributing to mature amacrine cell phenotype and to complete absence of GABAergic and glycinergic amacrine cells in retinal explants.

Recent studies have shown that the bHLH transcription factor Ptf1ais involved in driving neural precursors to differentiate into GABAergic neurons in the cerebellum, and that Ptf1a is required for the formation of GABAergic neurons in the dorsal horn of the spinal cord. The absence of Ptf1a expression in this tissue causes a transdifferentiation of GABAergic neurons to glutamatergic neural cells(Glasgow et al., 2005; Hoshino et al., 2005). By contrast to spinal cord, lack of Ptf1a activity lead to differentiation arrest of amacrine precursor cells and partial transdifferentiation of amacrine precursor cells to ganglion cells, which results in loss of GABAergic and glycinergic amacrine cells in mature retina. The fact that in the Ptf1a-deficient retinae the number of differentiated amacrine cells is significantly decreased and the number of ganglion cells is increased could lead to loss of IPL. These data are in line with the results that in Math3/NeuroD double-mutant retina the IPL is not formed (Inoue et al.,2002).

The amacrine and horizontal cells are important interneurons that process and transmit visual input in the retinal circuitry. It has been demonstrated that amacrine and horizontal cells begin to exit the cell cycle at E11(Young, 1985). In our analysis, Ptf1a expression first appears at E12.5 in postmitotic cells, indicating that Ptf1a is unlikely to be involved in initial generation of amacrine and horizontal cells. We have also followed the cell fate of horizontal cells in Ptf1a-deficient retina. Ptf1a is important for differentiation of horizontal precursor cells. First, our lineage-tracing analysis revealed that amacrine and horizontal cells express Ptf1a. Second, in Ptf1a-deficient retina the expression of Lim1 is totally absent, which indicates that horizontal cells are not generated. Our observation of calbindin-positive cells in Ptf1amutant retinal explants, although the horizontal cells are deficient, may be attributed to the presence of a subtype of amacrine cells as previously reported (Uesugi et al.,1992).

Foxn4 is involved in the genesis of horizontal cells(Li et al., 2004). As the absence of Ptf1a does not affect the expression level of Foxn4 (Fig. 7I), we conclude that Ptf1a acts downstream in the development of horizontal cells, then Foxn4.

In Caenorhabditis elegans, genetic analyses of Lim and homeodomain genes have demonstrated a prominent role for them in terminal differentiation of specific neurons. For instance, mec-3 is required for differentiation of mechanosensory neurons, and lim-6 regulates neurite outgrowth and function of GABAergic motoneurons(Hobert et al., 1999; Way and Chalfie, 1988).

In Ptf1a knockout retinae, we observed a co-localization of GFP and Brn3 expression in approximately 30% of GFP-positive cells at E18.5(Fig. 5). This strongly indicates that a proportion of Ptf1a- deficient precursor cells transdifferentiate to ganglion cells. As the number of apoptotic cells in Ptf1a knockout retina is increased, we assume that Ptf1a-deficient cells that do not transdifferentiate undergo apoptotic cell death. Regarding the transdifferentiation of retinal cells, Ptf1a knockout resembles Math3/NeuroD double-mutant retina(Inoue et al., 2002). However,there are several significant differences concerning amacrine cell differentiation and the number of retinal cells. First, in INL of Ptf1a-deficient retinal explants, some amacrine cells(syntaxin⁺, calbindin⁺) were generated. By contrast, in Math3/NeuroD null mutant retinal explants, amacrine cells are completely missing (Inoue et al.,2002). Second, whereas the number of ganglion cells was significantly increased in Math3/NeuroD double-mutant retinal explants, that of Ptf1a-deficient retinal explants was strongly reduced. Third, in Ptf1a null mutant retinal explants, the number of apoptotic cells was increased, bipolar cells were significantly reduced and the architecture of the INL was disrupted. In Math3/NeuroDdouble-mutant retinal explants, bipolar cells were normally generated and positioned and the number of apoptotic cells was not increased. Therefore, we conclude that there are more defects in Ptf1a-deficient retina compared with Math3/NeuroD double-mutant retina.

To summarize previous results and our present study, we propose a hypothetical model for cell differentiation in the retinae of embryonic mice(Fig. 8). The transcription factors Neurod1 and Neurod4 specify RPCs to amacrine precursor cells. It has been demonstrated that the Barhl2 homeobox gene is involved in the specification of glycinergic amacrine cells(Mo et al., 2004). As the inactivation of Ptf1a leads to loss of GABAergic and glycinergic amacrine cells, as well as to the downregulation of Barhl2 and Gad1 transcripts, Ptf1a promotes the differentiation of amacrine cell precursors to GABAergic and glycinergic amacrine cells. Furthermore, we found that in the Ptf1a null retina a small number of amacrine precursor cells differentiated to amacrine cells. Thus, we conclude that Ptf1a contributes to the specification of amacrine cell subtypes rather than to the generation of amacrine cells. Ptf1a is expressed in postmitotic horizontal precursor cells. As the expression level of Foxn4 in Ptf1a-deficient retinae did not change(Fig. 7I), we propose that Ptf1a acts downstream to Foxn4 in the cell signaling cascade regarding the generation of horizontal cells. Even though there are gaps in our knowledge about retinogenesis, we provide evidence that Ptf1aplays an important role in retinal cell differentiation.

We are grateful to Raymond J. MacDonald for providing the Ptf1a antibody(Department of Molecular Biology, the University of Texas Southwestern Medical Center, Dallas). We thank Peter K. Wellauer (Swiss Institute for Experimental Cancer Research, Epalinges, Switzerland) for the genomic Ptf1asequence, Klaus Rajewsky (CBR Institute for Biomedical Research, Harvard Medical School, Boston) for the Cre construct and Bernhard Luecher(Institute of Pharmacology, University of Zurich, Switzerland) for the neomycin-resistance cassette flanked by FRT sites. We are grateful to T. Marquardt for helpful discussions and A. Dietrich as well as N. Buentig for comments on the manuscript. This work was supported in part by the Deutsche Forschungsgemeinschaft (SFB 518) and the Wilhelm-Roux program (FKZ 10/41) from Martin Luther University Halle-Wittenberg.