Temporal requirement of the alternative-splicing factor Sfrs1for the survival of retinal neurons

As the genomes of different organisms are sequenced and annotated it is becoming apparent that the complexity of an organism does not depend on the total number of protein-coding genes. The current estimate for the number of protein-coding genes in the mouse (∼23,049)(http://www.ensembl.org/Mus_musculus/index.html)is similar to that in Arabidopsis thaliana(Seki et al., 2002). Thus, it has been proposed that the complexity of higher metazoans must arise via the regulation of these genes at the transcriptional and post-transcriptional level. Alternative splicing (AS) is the mechanism by which exons of a single gene can be spliced in various combinations to encode a diverse set of proteins. Indeed, 74% of all human genes are known to be alternatively spliced and different tissues exhibit varying degrees of AS(Cheng et al., 2005; Johnson et al., 2003; Kampa et al., 2004). Moreover,when one considers that a single gene can produce multiple isoforms, the number of proteins encoded by the genome will most likely be much higher than the current estimates. In humans, the mRNAs that are expressed in the central nervous system (CNS) are subjected to the highest degree of AS when compared with other mature tissues (McCullough et al., 2005). However, the role of AS in CNS development is relatively unexplored. Given the complexity of AS, combined with that of CNS development, it is important that a system employed to investigate this question should be accessible, well characterized during development and relatively amenable to functional perturbation.

The vertebrate retina is part of the CNS, yet is a relatively simple tissue with six neuronal cell classes (rod photoreceptors, cone photoreceptors,horizontal cells, bipolar cells, amacrine cells and ganglion cells) and one glial type (Müller glia) organized in a stereotypic manner. The birth order of each cell type is conserved, such that ganglion cells, cone photoreceptors and horizontal cells are among the first-born cell types,followed by amacrine cells, rod photoreceptors, bipolar cells and Müller glia (Rapaport et al., 2004; Sidman, 1961; Young, 1985a). The production of each postmitotic cell type begins in the central retina and expands from the center to the periphery (Rapaport et al., 2004; Young,1985a; Young,1985b). Moreover, retinal development has been the focus of many studies, leading to a better understanding of mechanisms that govern cell fate determination and differentiation (Cepko,1996; Livesey and Cepko,2001). Thus, the current investigation focuses on understanding the role of an alternative-splicing factor (ASF) called splicing factor arginine/serine-rich 1 (Sfrs1) in retinal development.

Sfrs1 belongs to a highly conserved arginine/serine-rich (SR)protein family of RNA processing factors found throughout metazoans and in plants (Zahler, 1999; Zahler et al., 1992). The role of Sfrs1 in splicing has been well documented, but only recently has its role in any developmental context been investigated(Xu et al., 2005). In C. elegans, ablation of the Sfrs1 homolog resulted in late embryonic lethality, suggesting that its function is non-redundant in at least one critical stage of development (Kawano et al., 2000; Longman et al.,2001). In mice, the loss of Sfrs1 also resulted in embryonic lethality (Xu et al.,2005). A conditional knockout (cKO) mouse in which the Sfrs1 gene was flanked with loxP sites(Sfrs1^fl/fl) has enabled functional studies of Sfrs1. Xu et al. have shown that loss of Sfrs1 does not cause aberrant proliferation and/or cell death of cardiac progenitor cells and that embryonic heart development proceeds normally(Xu et al., 2005). However,during postnatal remodeling of the heart, aberrant splicing of specific target genes such as cardiac troponin T (cTnT; Tnnt2 - Mouse Genome Informatics), the Z-line protein cypher (Ldb3) and Ca²⁺/calmodulin-dependent kinase II δ(CaMKIIδ; Camk2d), resulted in physiological defects in the heart (Xu et al.,2005).

In this report, we show that Sfrs1 is expressed in the developing mouse retina and is itself regulated by AS. We have identified a new isoform that is expressed during late embryonic development and continues to be expressed during postnatal retinal development. This novel isoform lacks the SR domain that is crucial for the nuclear localization of the Sfrs1 protein(Kataoka et al., 1999; Lai et al., 2000; Lai et al., 2001). Furthermore, the present investigation employs the Sfrs1^fl/fl mice created by Xu et al. to determine the role of Sfrs1 in the developing mouse retina. To specifically ablate Sfrs1 function during retinal development, the Sfrs1^fl/fl mice were crossed to mice that express Cre recombinase under the regulation of the ceh-10 homeodomain-containing homolog(Chx10; Vsx2), a gene that is expressed in retinal progenitor cells (Rowan and Cepko,2004). The loss of Sfrs1 function resulted in a small eye at birth. We found that loss of Sfrs1 function did not have a significant effect on proliferation. However, neurons generated during early embryonic development underwent apoptosis, whereas those generated after birth did not. Consequently, neurons generated in the embryo, such as ganglion cells, cone photoreceptors, horizontal cells and amacrine cells, were significantly reduced in the Sfrs1-cKO retina. By contrast, rod photoreceptors, bipolar cells, late-born amacrine cells and Müller glia survived in the Sfrs1-cKO retina. The subset of susceptible neurons was defined primarily by the time of their birth, a possible reflection of the temporal heterogeneity in gene expression that has been defined for retinal progenitor cells (Trimarchi et al.,2008).

To generate Sfrs1-cKO retinae, Sfrs1^fl/fl mice were crossed to Chx10::Cre mice. Mice were genotyped for Sfrs1 by PCR (Xu et al.,2005). The presence of the Cre allele was also confirmed by PCR (Rivera-Feliciano and Tabin,2006). When mice were subjected to surgical procedures, all IACUC protocols to minimize pain during and after surgery were observed.

Retinae from different developmental time points were harvested and total RNA prepared in Trizol following the manufacturer's protocol (Invitrogen). For cDNA synthesis, 5 μg of total RNA from retinae harvested at various time points was used (Kanadia et al.,2006). PCR to detect Sfrs1 isoforms was performed with a forward (5′-ATGTCGGGAGGTGGTGTGATCC-3′) and a reverse(5′-CCAATCATCTTATGTACGAGAGCGAGATC-3′) primer for 30 cycles(95°C for 35 seconds; 58°C for 25 seconds; 68°C for 2 minutes). PCR to detect Tnnt2 isoforms was performed as described previously(Kanadia et al., 2003).

In situ hybridization (ISH) on 16 μm cryosections and dissociated cells was performed as described previously(Trimarchi et al., 2007). All the probes used in this report are as published previously(Trimarchi et al., 2008; Trimarchi et al.,2007). For Sfrs1 we employed a 3′UTR probe corresponding to bp 1620-2639 in the clone NM_173374 in the NCBI database.

For immunofluorescence (IF), 16 μm cryosections were first hydrated in phosphate-buffered saline (PBS, pH 7.4), followed by the immunohistochemistry protocol of Kim et al. (Kim et al.,2008). The dilutions of the primary antibodies were: chicken anti-GFP (1:2000) (Abcam); mouse anti-Pax6 (1:300) (Covance); rabbit anti-Chx10 (1:300) (Cepko laboratory); mouse anti-rhodopsin (4D2; 1:300)(Molday and MacKenzie, 1983);mouse anti-glutamine synthetase (1:300) (Chemicon); rabbit anti-red/green opsin (1:300) (Chemicon); and mouse anti-Ki67 (1:250) (BD Pharmingen).

The nuclear/cytoplasmic extraction protocol from Pierce was employed on retinae (n=10) from different stages. Upon fractionation, 30 μg of protein was resolved on a 4-20% Tris-glycine gradient gel (Invitrogen),followed by transfer of the proteins to a positively charged nylon membrane(Invitrogen), which was then subjected to immunoblot analysis as described previously (Kanadia et al.,2006). The primary antibodies used were mouse anti-Sfrs1 (1:1000)(Lifespan) and mouse anti-Cugbp1 (1:500) (Abcam).

P0 pups were harvested in 0.1 M sodium cacodylate buffer (pH 7.4), followed by fixation in 2% paraformaldehyde (PFA) and 2.5% glutaraldehyde in 0.1 M sodium cacodylate buffer. Retinae were then processed by the Harvard Medical School Electron Microscopy Core Facility.

P0 pups were electroporated with pCAG-Cre and pCAG-LoxP-Stop-LoxP-GFP plasmids or pCAG-GFP plasmid alone(Matsuda and Cepko, 2004).

Retroviral vectors were generated by cloning Cre after the IRES in the pQC-H2B-GFP-IRES-MCS vector (Punzo and Cepko, 2008). All viruses were prepared as described previously(Cepko, 1989). E10 viral infections by ultrasound-assisted delivery in timed-pregnant Sfrs1^fl/fl and/or Sfrs1^wt/wt females were performed as described previously(Punzo and Cepko, 2008). P0 viral infections were performed as described previously(Matsuda and Cepko, 2004).

A pregnant female at E16 was weighed and then injected with 3 μg of BrdU/g body weight. At P7, retinae were harvested and processed for BrdU staining by antigen retrieval as described by the manufacturer (Vector Labs). Next, the slides were fixed in 4% PFA for 20 minutes followed by two 5-minute washes with PBS. Slides were treated with 2M HCl for 30 minutes at room temperature, followed by a brief wash with 0.1 M boric acid (pH 8.5) and IF was performed as described previously (Kim et al., 2008).

To understand the role of Sfrs1 during mouse retinal development,we first characterized its expression pattern. RNA ISH was employed on sections from embryonic day (E) 10.5 to postnatal day (P) 30(Fig. 1A). At E10.5 and E11.5,expression of Sfrs1 was detected in progenitor cells. However, in the peripheral retina, Sfrs1 expression did not extend to the edge, a portion of the retina that is known to give rise to the ciliary body(Davis-Silberman and Ashery-Padan,2008). In addition, the ISH signal for Sfrs1 was enriched in cells at the apical end of the retina(Fig. 1A, E10.5 arrowhead),where the retinal progenitor cells undergo the M phase of their cell cycle(Sidman, 1961; Young, 1985b). Similar asymmetric enrichment of the ISH signal for Sfrs1 was observed in the olfactory epithelium (see Fig. S1 in the supplementary material). ISH for cell division cycle 20 (Cdc20), a marker of the G2-M phase of the cell cycle, and IF for phosphorylated histone H3 (PH3), a marker of M phase, were employed (see Fig. S1 in the supplementary material) to determine whether the region of Sfrs1 expression in the olfactory epithelium was also the region of M phase (Kawauchi et al.,2005; Trimarchi et al.,2008; Weinstein,1997). As predicted, both the ISH for Cdc20 and IF for PH3 confirmed that the mRNA for Sfrs1 was enriched in areas expressing these genes, thereby suggesting the involvement of Sfrs1in cells in the G2-M phase of the cell cycle. At E12.5, Sfrs1expression was expanded to the entire neuroblastic layer (NBL)(Fig. 1A) and was also observed in the lens epithelium. In addition, at E13.5, expression of Sfrs1was transiently enriched in cells lining the vitreal side of the peripheral retina (Fig. 1A, E13.5,arrowhead), but this expression pattern was no longer observed by E14.5, at which time the expression of Sfrs1 in the ONBL persisted. At P0, the expression of Sfrs1 was again in the ONBL(Fig. 1A). During postnatal development, the expression of Sfrs1 was restricted to the newly forming inner nuclear layer (INL) in the central retina, while in the periphery the expression of Sfrs1 was observed in progenitor cells. At P6, Sfrs1 expression became restricted to the lower half of the INL, where mostly amacrine and displaced ganglion cells are found, and to the ganglion cell layer (GCL), where ganglion and displaced amacrine cells are found (Fig. 1A). This pattern was also observed at P10 and P30 (Fig. 1A).

The expression of Sfrs1 was also investigated by RT-PCR analysis. Recent reports have indicated that the protein level of most of the SR-protein family members is regulated by AS (Lareau et al., 2007; Ni et al.,2007). Thus, we investigated whether AS might modulate the protein levels of Sfrs1 during retinal development. To determine the AS status of Sfrs1, which has 5 exons (Fig. 1B), a forward primer in exon 1 (start codon) and reverse primer in exon 4 (stop codon) (Fig. 1B, red arrows) were utilized. Indeed, Sfrs1 was found to be regulated via AS in a temporal manner, such that only one isoform (Sfrs1b)was observed in the embryo, whereas an additional isoform at a higher molecular weight (Sfrs1a) was also expressed postnatally(Fig. 1C). The temporal regulation of Sfrs1 via AS led us to investigate whether a similar form of regulation was also employed during the development of other tissues. Because the role of Sfrs1 in heart development has been investigated(Xu et al., 2005), this tissue was examined. RT-PCR analysis with the same primers showed that during embryonic heart development, both isoforms of Sfrs1 were present. By contrast, the predominant isoform at P2 was the Sfrs1b isoform, which was also the only one expressed in the adult (Fig. 1D).

Sequence analysis revealed that Sfrs1b was the canonical isoform, whereas Sfrs1a retained intron 3. Consequently, in the Sfrs1a isoform there was a frame shift in the coding sequence resulting in the truncation of the RS domain. This change is of functional significance because Sfrs1 shuttles between the cytoplasm and the nucleus and the phosphorylation of the RS domain is required for its nuclear localization(Caceres et al., 1998; Ma et al., 2008). Thus, during retinal development, the AS of Sfrs1 produces an isoform that would most likely fail to translocate into the nucleus. In agreement with the RT-PCR analysis, immunoblot analysis showed two immunoreactive bands in the cytoplasmic fraction from the postnatal, but not the embryonic, retina(Fig. 1E). In addition, the lower molecular weight isoform (Sfrs1a) of Sfrs1 was significantly more abundant in the cytoplasmic than in the nuclear fraction(Fig. 1E). Since the Sfrs1a isoform was predicted to be cytoplasmic, the low-level detection of Sfrs1a in the nuclear fraction might be due to cross-contamination between the two fractions. The inability of the Sfrs1a isoform to shuttle into the nucleus was further investigated by fusing the coding sequence of Sfrs1a and Sfrs1b to GFP, followed by transfection of each plasmid into NIH3T3 cells. As predicted,the canonical isoform, Sfrs1b, was predominantly located in the nucleus,whereas Sfrs1a was not observed in the nucleus(Fig. 1F). In summary, Sfrs1 is expressed throughout retinal development and is regulated by AS in a temporal manner.

The role of Sfrs1 during retinal development was investigated by genetic ablation of Sfrs1 function in retinal progenitor cells by crossing the Sfrs1^fl/fl mouse to a Chx10::Cre transgenic mouse(Rowan and Cepko, 2004). Consequently, the majority of the developing retina lacked Sfrs1function. The Sfrs1-cKO retinae underwent aberrant embryonic development, which was inferred from the observation that Sfrs1-cKO mice were born with small eyes (Fig. 2A). Sfrs1-cKO eyes also had a thin optic nerve and the total eye weight was half that of the wild type(Fig. 2B,C). Hematoxylin and Eosin (H&E) staining of the retinal sections revealed a significant decrease in the size of the retina, but not in the retinal pigmented epithelium (RPE) or the lens (Fig. 2D,E). The retina was detached from the RPE(Fig. 2E, arrowhead) while abutting the lens and exhibited a marked decrease in the cellularity of the GCL, along with a perturbed inner limiting membrane (ILM)(Fig. 2G, arrowhead). Quantification of the number of nuclei in the GCL(Fig. 2F,G, arrowhead)confirmed the decrease in GCL cellularity(Fig. 2H), which in turn was reflected in the thin optic nerve (Fig. 2B, insets). Ultrastructural analysis showed that cells in the mutant retina were highly disorganized, with vacuolar inclusions, unlike the wild-type retina, which had elongated nuclei with a defined polarity(Fig. 2I-L). The absence of the ILM in the H&E-stained sections was confirmed by ultrastructural analysis(Fig. 2L, arrowhead). The mutant retina exhibited a dysmorphic fibrous layer instead of the well-organized ILM seen in the wild-type control retina(Fig. 2K). In summary, loss of Sfrs1 function had a profound effect on the embryonic development of the retina and, possibly, also on that of the ciliary body, which together might have contributed to the microphthalmia observed at birth.

To determine the cause of the small retina in the Sfrs1-cKO mice at P0, we investigated whether the mutant retina had a proliferation defect during embryonic development. Ki67 antibody, a well-characterized proliferation marker (Gerdes et al.,1991), was employed on retinal sections from E12.5, E13.5, E18.0 and P0. No proliferation defect was observed in the Sfrs1-cKO retina(Fig. 3A-H). This was confirmed by quantification of the PH3 staining in the Sfrs1-cKO retina at E12.5 (see Fig. S2 in the supplementary material). Therefore, cell death was investigated as a cause of the small eye phenotype in Sfrs1-cKO mice. Terminal dUTP nick end labeling (TUNEL) assay revealed a significant increase in the number of TUNEL⁺ cells at E12.5, E13.5 and E18.0 in the mutant retina (Fig. 3I-N). TUNEL⁺ cells observed in Sfrs1-cKO retina at E12.5 and E13.5 were predominantly in the central retina(Fig. 3J,L), whereas the majority of the TUNEL⁺ cells at E18.0 were in the periphery(Fig. 3N). At P0, few TUNEL⁺ cells were observed in the periphery(Fig. 3P). Furthermore,quantification of the TUNEL⁺ cells in the mutant retina at each stage revealed that the cell death peaked at ∼E13.5, followed by a decrease in the number of TUNEL⁺ cells at E18, with few TUNEL⁺ cells at P0 (Fig. 3Q-T). In summary, loss of Sfrs1 function during embryonic development caused cell death and did not cause an observable reduction in proliferation; however, a subtle reduction in proliferation could not be ruled out.

There were three different scenarios that could explain the small eye phenotype. The dying cells might be progenitor cells, they might be postmitotic cells, or they might comprise both types of cells. Since proliferation was not reduced, it was unlikely that progenitor cells died. Nonetheless, this was investigated further because Sfrs1 is expressed in progenitor cells during embryonic development. For this, Sfrs1function was ablated in retinal progenitor cells utilizing a retrovirus expressing Cre along with nuclear GFP (histone H2B-GFP fusion). Because retroviruses can only infect mitotic cells, the retinal progenitor cells would lose Sfrs1. Moreover, the viral genome inserts into the cell genome and thereby marks all its progeny. This retrovirus was delivered to E10.5 Sfrs1^fl/fl embryos with the aid of an ultrasound-guided injection device, followed by detection of GFP immunofluorescence at P14(Fig. 4). A virus encoding membrane-GFP (mGFP), without Cre, was also injected as a control for the infection. An additional control was to repeat the same experiment in Sfrs1^wt/wt mice. Analysis at P14 showed several clones with nuclear GFP in the retinae of infected Sfrs1^fl/flmice (Fig. 4C), indicating the survival of the initially infected progenitor cells. However, there was a difference in the cell-type composition of the clones in the Sfrs1^fl/fl and Sfrs1^wt/wt mice. Based on their position, the majority of GFP⁺ cells in the Sfrs1^fl/fl mice were either rod photoreceptors, bipolar cells or Müller glia (Fig. 4C). By contrast, the cell types observed in the Sfrs1^wt/wt mice, as determined by the position of the nuclei, included rod photoreceptors, horizontal cells, bipolar cells,Müller glia, amacrine cells and ganglion cells(Fig. 4B). The Cre-negative mGFP⁺ clones were similar in composition to those of the wild type(data not shown). In conclusion, the loss of Sfrs1 did not cause retinal progenitor cell death, at least not in all infected progenitor cells.

This conclusion was further bolstered by the genetic ablation strategy that employed the Chx10::Cre line along with a reporter line, RC::PFWE, in the Sfrs1^fl/fl background. The RC::PFWE strain reports Cre activity by expressing nuclear β-galactosidase (nlacZ)(Farago et al., 2006). Thus,every progenitor cell that expresses Cre should be positive for nlacZand negative for Sfrs1 function. The results showed that at least some retinal progenitor cells did not die, as there were nlacZ⁺ neurons in the P7 mutant retina(Fig. 4E). These neurons, as judged from their position, were mostly photoreceptors, bipolars and Müller glia (Fig. 4E). By contrast, the wild-type littermate retina showed many cells expressing nlacZ in the ONL (outer nuclear layer) (rod photoreceptors, cone receptors), INL (bipolar cells, horizontal cells, amacrine cells and Müller Glia) and in the GCL (ganglion and amacrine cells)(Fig. 4D). In both the wild-type and the mutant retinae there were cells that were negative for nlacZ, which was likely to result from the mosaic expression of Chx10::Cre (Rowan and Cepko, 2004). In summary, loss of Sfrs1 function did not lead to the death of all progenitor cells, although some progenitor cell death could not be ruled out.

To examine whether the postmitotic cells were dying in the Sfrs1-cKO retinae, we first determined whether postmitotic cells were produced. For this, RNA ISH with probes that label differentiated neurons was employed on E13.5 retinal sections. First, loss of the Sfrs1transcript in the Sfrs1-cKO retina was confirmed(Fig. 5B). Although the majority of the retina lacked Sfrs1, a few cells(Fig. 5B, arrowhead) were positive for Sfrs1, most likely reflecting the Chx10::Cre mosaicism(Rowan and Cepko, 2004). ISH analysis was performed on serial sections with probes that mark postmitotic cells, including neurofilament-like light chain 68 (Nf68; Nefl - Mouse Genome Informatics), brain 3b (Brn3b; Pou4f2) and thyroid hormone receptor beta 2(Trβ2; Thrb). Ganglion cells were produced in the mutant retina, as shown by the strong ISH signal for Nf68 and Brn3b (Fig. 5D,F). Similarly, cones were produced, as shown by the ISH signal for Trβ2 in the Sfrs1-cKO retina(Fig. 5H). However, Trβ2⁺ cone photoreceptors in the mutant retina were located throughout the ONBL(Fig. 5J, arrowhead), rather than abutting the RPE as was seen in the wild type(Fig. 5I). In summary, these data suggest that at least some neurons can be produced and initiate differentiation in the Sfrs1-cKO retina, with cones showing aberrant localization.

To investigate the persistence of the neurons generated in the embryo, the above-mentioned strategy was employed on P0 retinal sections. This analysis showed that at P0, the ISH signal for Brn3b⁺ and Nf68⁺ ganglion cells was reduced in the Sfrs1-cKO retina (Fig. 5B′,D′). At P0, Nf68 also labels horizontal cells (Fig. 5C′,arrowhead), which were significantly reduced in the Sfrs1-cKO retina(Fig. 5D′, arrowhead). Similarly, the ISH signal for rabphilin 3a (Rph3a) and Ap2b,which mark amacrine cells, was also reduced(Fig. 5F′,H′). When we employed ISH with these and other probes at P7 to detect neurons produced in the embryo, very few of these neurons were detected. Specifically, there were few ganglion cells in the Sfrs1-cKO retina, as detected by ISH for Brn3b (Fig. 5B″, arrowhead) or Nf68(Fig. 5D″), the latter showing a few more ganglion cells than the former. In addition, cone cells [as determined by the S opsin (opsin 1) signal] were significantly reduced in number in the mutant retina (Fig. 5F″). Similarly, amacrine cells and horizontal cells were significantly reduced in the Sfrs1-cKO retina as determined by Ndrg4 ISH (Fig. 5H″). Immunofluorescence with Pax6 revealed more amacrine cells than did the Ndrg4 ISH, although numbers were still reduced compared with the wild type (Fig. 5M,N, open arrowhead). At P7, Pax6 antibody also marks horizontal cells, which were significantly decreased in the mutant retina(Fig. 5N, white arrowhead). By contrast, a similar reduction was not observed in cells produced postnatally. As shown by Nr2e3 and Otx2 ISH signal, the mutant retina did not exhibit a significant decrease in the thickness of the ONL, where the Nr2e3⁺ rod photoreceptors primarily reside, or in the scleral half of the INL, where the postnatally generated Otx2⁺ bipolar cells reside(Fig. 5J′,L). The significant decrease in the embryonically generated neurons was further quantified by dissociated cell in situ hybridizations (DISH)(Trimarchi et al., 2007). As shown in Fig. 6, by P7 there was a significant reduction in the number of embryonically generated neurons in the Sfrs1-cKO retina, which suggests that the cell death observed during embryonic development was most likely that of the postmitotic cells.

The conclusion that the loss of Sfrs1 function led to the death of postmitotic cells was further investigated in a birthdating experiment. A pregnant female at E16 was injected with BrdU, followed by analysis at P7 of the Sfrs1-cKO retinae stained for BrdU and Pax6. The results showed a significant decrease in the number of heavily labeled BrdU⁺ cells in the ONL, INL and GCL (Fig. 7). Here, the Pax6 antibody was used to mark amacrine cells, which are one of the major cell types produced at the time of the BrdU pulse. Quantification revealed that in the Sfrs1-cKO retina, there was a twofold decrease in the number of heavily labeled BrdU⁺ cells as compared with its wild-type littermate(Fig. 7). This is in keeping with the death of postmitotic cells, although a failure to produce postmitotic cells in the normal numbers by Sfr1-cKO progenitor cells might also contribute to this deficit.

The neuronal composition of wild-type and Sfrs1-cKO retinae was compared at P14, a time when the production of retinal cell types is complete. Analysis at this stage revealed that the Sfrs1-cKO retina had undergone further degeneration and exhibited rosettes(Fig. 8D, arrowhead). Furthermore, in the mutant, the ciliary body was fused to the retina, such that it could not be separated from the neural retina(Fig. 8J, arrowhead), in contrast to the situation in the wild-type littermate where it was easily removed (Fig. 8I, arrowhead). Despite the severe morphological deterioration of the mutant retina at P14, it still appeared to have a relatively unaltered percentage of Müller glia as shown by IF with glutamine synthetase antibody(Fig. 8A,B). Similarly, rod photoreceptors also appeared relatively unaltered in the mutant retina(Fig. 8F, upper arrowhead; see Fig. S3 in the supplementary material). In contrast to the IF signals for Müller glia and rods, IF for Pax6(Fig. 8F, lower arrowhead),Chx10 and red/green opsin showed a significant reduction in the Sfrs1-cKO retina (Fig. 8C,H). The reduction in the bipolar cells could be secondary to the loss of cone photoreceptors. This dysmorphic retina ultimately underwent complete degeneration by P30, as seen by the absence of the entire eye in the Sfrs1-cKO mouse (Fig. 8L). In summary, the data gathered at P14 indicate that despite the severe morphological defects observed in the mutant retina, cell types including rod photoreceptors, Müller glia and bipolar cells were relatively unaffected, whereas others, such as ganglion cells, amacrine cells,cone photoreceptors and horizontal cells, seemed to have undergone further loss.

P0 in vivo electroporation was used to investigate whether rod photoreceptors, Müller glia and bipolar cells were resistant to the loss of Sfrs1 function. This strategy used two plasmids, one expressing Cre and the other expressing GFP in a conditional manner, such that it reported Cre activity. As shown in Fig. 8, production of rod photoreceptors was normal, and the rod photoreceptors had proper outer segments(Fig. 8M). In addition,amacrine cells appeared to be produced normally. Although bipolar cells and Müller glia were also produced during this time, the promoter driving GFP failed to express reproducibly in these two cell types, making it difficult to assess their relative levels. To investigate the production and maintenance of bipolar cells and Müller glia, the same virus used for embryonic ablation of Sfrs1 was used to infect P0 retinal progenitor cells, followed by analysis at P55. Several nuclear GFP⁺ clones were observed in the Sfrs1-cKO retina (Fig. 8O-Q). Based on the position and shape of the nuclei, the GFP⁺ cells were judged to be rod photoreceptors, bipolar cells and Müller glia (Fig. 8O-Q;see Fig. S4 in the supplementary material). In summary, Sfrs1function is required in a temporal manner, such that neurons born postnatally are resistant to the loss of its function.

During early embryonic development, Sfrs1 RNA was enriched in the apical side of the retina (Fig. 1A), which suggests that it is associated with the M phase of the cell cycle. This was consistent with the overlapping expression patterns of Cdc20 and PH3, both of which mark the M phase (see Fig. S1 in the supplementary material). Interestingly, a similar apical expression pattern for Sfrs1 was observed in the olfactory epithelium (see Fig. S1 in the supplementary material). These data suggest that Sfrs1 might regulate the G2-M phase transition in early progenitor cells. This is indeed the case in chicken DT-40 cells, in which loss of Sfrs1 function results in G2 cell cycle block (Li et al.,2005).

Sfrs1, an ASF, is not only regulated at the transcriptional level during retinal development, but is also alternatively spliced in a temporal manner (Fig. 1). The newly identified isoform of Sfrs1 includes intron 3, which produces a truncated isoform of Sfrs1 that would fail to translocate into the nucleus. This truncated isoform is observed by immunoblot analysis and the GFP fusion version of this isoform is seen only in the cytoplasm. However, we cannot rule out the possibility that the new isoform is subject to nonsense-mediated mRNA decay and so the true identity of the lower molecular weight isoform warrants further investigation (see Fig. S5 in the supplementary material).

The importance of Sfrs1 in retinal development is evidenced by the small and dysmorphic retinae in the Sfrs1-cKO mice(Fig. 2A). This phenotype is primarily a result of cell death during embryonic development(Fig. 3). Interestingly, this is in contrast to the role of Sfrs1 in cardiac development, which proceeds normally during the embryonic period, whereas the postnatal remodeling of the heart is abnormal in Sfrs1-cKO mice, resulting in cardiomyopathy and death (Xu et al.,2005). It is interesting to note that the presence of the phenotype in the two tissues is correlated with the absence of the Sfrs1a isoform. In other words, the times at which the Sfrs1a isoform is expressed normally are the times when loss of Sfrs1 function does not result in a phenotype. One possibility is that another ASF that regulates the AS of Sfrs1 normally compensates for the loss of Sfrs1 function. Thus, the presence of Sfrs1a serves as a proxy for the presence of the compensatory ASF, which in the heart is present in the embryo, whereas in the retina it is present postnatally.

Progenitor cells in the Sfrs1-cKO retina did not die in significant numbers, as proliferation remained unaltered in the Sfrs1-cKO retina (Fig. 3A-H). In addition, infection of Sfrs1^fl/flprogenitor cells with retrovirus expressing Cre and nuclear-GFP produced clones consisting mainly of later-born neurons(Fig. 4C). Similarly, several nlacZ⁺ neurons were detected in the P7 mutant retina(Fig. 4E). The presence of nlacZ⁺ cells reflects a history of Cre activity in the progenitor cells. These data indicate that the loss of Sfrs1 function does not cause the death of the majority of retinal progenitor cells.

If the loss of Sfrs1 function does not lead to the death of progenitor cells, then it must result in the death of postmitotic cells. This possibility is consistent with the pattern of TUNEL⁺ cells in the Sfrs1-cKO retinae. Specifically, the pattern of cell death overlaps with the production of postmitotic cells during development(Fig. 3)(Farah and Easter, 2005; Young, 1985a). Furthermore,ISH and IF analysis to detect differentiated neurons at P0 and P7 revealed that ganglion cells, horizontal cells, cone photoreceptors and amacrine cells were significantly reduced (Fig. 5). Interestingly, these neurons are produced in the embryo,coincident with the period of greatest cell death in the Sfrs1-cKO retina. By contrast, the later-born rod photoreceptors, bipolar cells and Müller glia were not as severely reduced (Figs 5 and 6). Taken together, these data indicate that in the Sfrs1-cKO retina, the postmitotic cells die in the embryo, which in turn could affect proliferation, as previous reports indicate that ganglion cell death can indirectly affect proliferation(Mu et al., 2005). However, in the case of the Sfrs1-cKO retina, which had reduced numbers of ganglion cells, there was little effect on proliferation(Fig. 5D). It is possible that the reduced numbers of ganglion cells were sufficient to produce the signals that might be crucial for proliferation.

The peak of amacrine cell production is ∼E16.5-17.5, and production ends at ∼P4 (Farah and Easter,2005; Young,1985a; Young,1985b). Given that postnatally generated neurons are not affected in the Sfrs1-cKO retina, postnatally generated amacrine cells were not expected to die, which was reflected in the modest reduction in the Pax6 IF signal. By contrast, another marker of amacrine cells, Ndrg4, was significantly reduced in the mutant retina, which suggests that it marks the types of amacrine cells produced in the embryo. Based on this interpretation,we posit that there are at least two classes of amacrine cells as defined by their birthdate and subsequent dependence on Sfrs1, which warrants further investigation.

Unlike the aforementioned cell types, rod photoreceptors are born in significant numbers throughout retinal development. As discussed above, many of the early-born cells died prior to P0, which was also true for rod photoreceptors. As shown in Fig. 7, the number of BrdU⁺ cells in the ONL of the P7 Sfrs1-cKO retina was significantly reduced, suggesting that rod photoreceptors must have died. The remaining rod photoreceptors could be a consequence of Chx10::Cre mosaicism. It is unlikely that this reduction was due to the death of cone photoreceptors because very few(<5%) are produced at E16 (Carter-Dawson and LaVail, 1979). By contrast, rod photoreceptor production is initiated at ∼E13 and by E16 more than 5% of the rod photoreceptors have been produced. Thus, like amacrine cells, rod photoreceptors produced in the embryo are most likely to die in the Sfrs1-cKO retina.

In summary, the current investigation suggests that the key determinant of whether cells survive in the Sfrs1-cKO retina is when they are born,as opposed to the subtype of neuron. Several lines of evidence point to this conclusion. First, the number of TUNEL⁺ cells decreased late in embryonic development. Second, the genetic ablation of Sfrs1 early in development led to the loss of many early-born neurons. This includes the early-born, but not the late-born, amacrine cells and rods. Third, survival of postnatal neurons was demonstrated by P0 electroporation of Cre and by P0 viral infection. Survival was coincident with the presence of the Sfrs1b isoform. Finally, we propose a model in which AS mediated by Sfrs1 is required for the terminal differentiation and/or maintenance of neurons produced in the embryo, but not for neurons produced postnatally(Fig. 9). The data presented here underscore the dynamic role of ASFs during vertebrate neuronal development.

We thank Dr Xiang-Dong Fu for generously sharing the Sfrs1conditional knockout mouse. This project was funded by the Howard Hughes Medical Institute.