Substituting mouse transcription factor Pou4f2 with a sea urchin orthologue restores retinal ganglion cell development

Pou domain transcription factor Pou4f2 is essential for the development of retinal ganglion cells (RGCs) in the vertebrate retina. A distant orthologue of Pou4f2 exists in the genome of the sea urchin (class Echinoidea) Strongylocentrotus purpuratus (SpPou4f1/2), yet the photosensory structure of sea urchins is strikingly different from that of the mammalian retina. Sea urchins have no obvious eyes, but have photoreceptors clustered around their tube feet disc. The mechanisms that are associated with the development and function of photoreception in sea urchins are largely unexplored. As an initial approach to better understand the sea urchin photosensory structure and relate it to the mammalian retina, we asked whether SpPou4f1/2 could support RGC development in the absence of Pou4f2. To answer this question, we replaced genomic Pou4f2 with an SpPou4f1/2 cDNA. In Pou4f2-null mice, retinas expressing SpPou4f1/2 were outwardly identical to those of wild-type mice. SpPou4f1/2 retinas exhibited dark-adapted electroretinogram scotopic threshold responses, indicating functionally active RGCs. During retinal development, SpPou4f1/2 activated RGC-specific genes and in S. purpuratus, SpPou4f2 was expressed in photoreceptor cells of tube feet in a pattern distinct from Opsin4 and Pax6. Our results suggest that SpPou4f1/2 and Pou4f2 share conserved components of a gene network for photosensory development and they maintain their conserved intrinsic functions despite vast morphological differences in mouse and sea urchin photosensory structures.


Introduction
The deep conservation of regulatory genes for eye development has amply demonstrated an underlying framework for eye diversification [1,2]. However, the developmental and evolutionary mechanisms that led to this remarkable diversity remain vague. Although changes in gene regulatory networks are probably to be the drivers of eye diversification, very little is known about the level at which conserved gene networks might have diverged to produce different structures and functions [2]. A potentially informative but largely unexplored phylum for investigating eye development and evolution is echinoderms. Echinoderms are basal deuterostomes that develop in ways similar to chordates but are distinctly different in their adult body plan. Unlike all other deuterostomes, echinoderms lack an obvious anterior-posterior axis. Instead, they exhibit highly derived body plans that are organized along a radial axis [3]. Notably, echinoderms lack any structures that even remotely resemble a vertebrate eye. Nevertheless, many behavioural patterns in adult sea urchins are attributable to highly developed photoreception [4].
We recently addressed the question of whether adult echinoderms have distinct photosensitive neurons that are separate from the diffuse surface-wide neural network already known to exist [5][6][7]. Using probes that were Strongylocentrotus purpuratus orthologues of mouse genes expressed in the retina [8], Agca et al. [9] and Ullrich-Lü ter et al. [10] showed that small groups of photoreceptor neurons were clustered around the periphery of the tube feet disc. This conclusion was based on the fact that many genes expressed in the mouse retina had homologues that were expressed in the tube feet neurons. The genes included retinal-expressed transcription factors [8][9][10][11][12] and an orthologue of opsin, Opsin4, which had signature features of a light-transducing opsin [8][9][10][11]13]. Many of the transcription factors have important roles in the developing mouse retina. Of particular note, the S. purpuratus orthologue of Pou4f2, SpPou4f1/2, has a 91% and 96.6% sequence match with mouse Pou4f2 in its Pou-specific and Pou-homeodomains, respectively. However, it is highly divergent outside of these domains (electronic supplementary material, figure S1a). Phylogenetic analysis of Pou-class genes showed that SpPou4f1/2 clustered with the Pou4f2 family over these domains [8]. We have shown that SpPou4f1/2 is expressed in the tube feet [9].
Although the results of Agca et al. [9] and Ullrich-Lü ter et al. [10] are correlative, they suggest the presence of functional photoreceptor neurons in the tube feet. However, the function of SpPou4f1/2 in the tube feet is unknown. Given the technical difficulties in directly determining the function of SpPou4f1/2 in the adult sea urchin tube feet, we chose a more feasible, albeit less direct, approach using a mouse knock-in (KI) strategy. We asked whether SpPou4f1/2 could function in the context of the developing mouse retina. Pou4f2 is essential for retinal ganglion cell (RGC) differentiation and survival [14][15][16]. In the developing retina, Pou4f2 expression is restricted to newly differentiated RGCs, and its expression is maintained throughout adult life. If SpPou4f1/2 functionally replaced Pou4f2, it would suggest that SpPou4f1/2 binds to and activates a similar set of genes in tube feet neurons, despite more than 540 Myr of divergence from the common ancestor of sea urchins and mice.
As demonstrated further in the article, our experiments in both mice and sea urchins support this hypothesis and indicate that conservation of Pou-specific and Pouhomeodomains in SpPou4f1/2 is, to high degree, sufficient to support RGC development.

Material and methods (a) Pou4 class protein sequence analysis
We performed protein sequence analysis with Megalign software (DNASTAR, Madison, WI, USA) using the ClustalW multiple sequence alignment method using four protein sequences: mouse Pou4f1, Pou4f2, Pou4f3 and S. purpuratus SpPou4f1/2. Phylogeny tree was constructed with MEGA5 software [17] using neighbour-joining method with 1000 bootstrap repeats. Amino acid sequences from multiple POU families were used to construct a phylogeny tree [18]. Sequences from Ciona intestinalis, Danio rerio, Mus Musculus, Strogylocentrotus purpuratus and Xenopus tropicalis were used for comparison.
(d) Isolation and processing of embryos, embryonic retinas, and adult eyes, optic nerves and retinas Embryos or adult eyes with optic nerves were harvested from different stages and processed in different ways. For histological analysis, tissues were fixed overnight in 4% paraformaldehyde (PFA) and 3% glutaradehyde in phosphate-buffered saline (PBS), subjected to PBS washing and methanol dehydration, and finally embedded in paraffin for sectioning. For immunolabelling of sections, tissues were fixed in 4% PFA for 30 min, washed three times with PBS, and embedded in optimal cutting temperature compound (Fisher Scientific).

(f ) Electroretinogram recordings (i) Subjects
Subjects were 2.5-to 3.5-month-old mice with the following genotypes:

(ii) Electroretinogram recordings
Mice were dark-adapted overnight and preparations for recording were all performed under dim red illumination (l . 650 nm) as previously described in [21] and the electronic supplementary material. Stimuli were provided from light-emitting diodes (l max ¼ 462 nm) over a range of time-integrated flash illuminances (stimulus strengths) from 26.7 to 2.3 log scotopic (sc) cd-s m 22 . The inter-flash interval was adjusted to allow the electroretinogram (ERG) response to return to baseline between flashes.

(iii) Data analysis
Amplitudes (microvolts) of a-waves were measured on the leading edge of the wave, at a fixed time (7 ms) after the brief flash, which was close to the peak amplitude for the strongest stimuli. Amplitudes of b-waves were measured between the a-wave trough and the b-wave peak after applying a low-pass 60 Hz filter to remove oscillatory potentials (the electronic supplementary material).
(g) Expression of SpPou4f1/2, Pax6 and Opsin4 in Strongylocentrotus purpuratus tube foot (i) Tube feet preparations Tube feet were dissected from live S. purpuratus and immediately fixed in 4% PFA in PBS at room temperature for 2 -4 h. After washing the collected tube feet several times in PBS, we transferred them to 100% methanol and stored them at 2208C until experimental processing.
(ii) In situ hybridization and immunohistochemistry SpPou4f1/2 probes were generated by cloning full-length SpPou4f1/2 cDNA into pIRES-hrGPF-2a vector (Agilent). The Pax6 probe has been described in Ullrich-Lü ter et al. [10]. Both antisense-and sense-digoxigenin-labelled SpPou4f1/2 probes were obtained using a digoxigenin-RNA labelling kit (Roche), following the manufacturer's instructions by using 1 mg of linearized plasmids. The Pax6 RNA probe was similarly prepared using unlabelled ribonucleotides and was subsequently labelled with 2, 4-dinitrophenyl using a Label-it kit (Mirus) following the manufacturer instructions. Whole-mount in situ hybridization and immunostaining against Sp-Opsin4 followed the protocol of Ullrich-Lü ter et al. [10]. Two-colour in situ staining was performed as described in Cole et al. [22]. After staining, samples were mounted in glycerol and analysed on a Leica TCS SP2 confocal laser-scanning microscope.
We first determined whether SpPou4f1/2 could restore RGCs in mature adult retinas in the absence of Pou4f2. Retinas of Pou4f2 Z/Z mice at 60 days of age (P60) lacked approximately 70% of their RGCs compared with the retinas of wild-type mice (Pou4f2 þ/Z ) (figure 2a,c) [15,16]. This depletion of RGCs resulted in a thinner retina (figure 2a,c). The retinas of Pou4f2 SpPou4f1/2/þ mice were normal, indicating that the SpPou4f2-KI allele had no dominant phenotype (data not shown). Most importantly, Pou4f2 SpPou4f1/2/Z retinas were not notably different from Pou4f2 SpPou4f1/2/þ or Pou4f2 þ/Z retinas (data not shown). SpPou4f1/2-expressing retinas appeared to have a normal ganglion cell layer with a full complement of RGCs (figure 2b). This result suggests that the Pou4f2 SpPou4f1/2 allele was able to replace Pou4f2's function in forming RGCs during retinal development and in maintaining the survival of RGCs in adult retinas.
Immunostaining of flat-mounted P60 Pou4f2 SpPou4f1/2/Z retinas with either SMI32 or NF-L antibodies showed the presence of many well-bundled axons emanating from SpPou4f1/2-expressing RGCs, whereas in Pou4f2 Z/Z there was little detectable staining (figure 2d -i). The number of axons emitted from Pou4f2 SpPou4f1/2/Z RGCs was substantially greater than the number observed in Pou4f2 Z/Z mice (figure 2e,f and h,i), and their appearance was not qualitatively different than those of Pou4f2 þ/Z axons (figure 2d,e and g,h). These results suggested that SpPou4f1/2 expression rescued the phenotype generated by the absence of Pou4f2. Moreover, optic nerves of Pou4f2 SpPou4f1/2/SpPou4f1/2 mice were indistinguishable in thickness and appearance from those of wild-type mice whereas Pou4f2 Z/Z mice had only a thin sheath largely devoid of axon fibres (cf. arrowheads in the electronic supplementary material, figure S2a -c).
To further pursue the properties of SpPou4f1/2-expressing axons, we traced their path into the brain with alkaline phosphatase (AP) using a Pou4f2-AP allele. Pou4f2-AP mice were bred to Pou4f2 þ/SpPou4f1/2 mice and retinorecepient regions in the brain were stained for AP activity. Pou4f2 þ/AP and  figure S3), indicating that axons emanating from RGCs expressing SpPou4f1/2 were capable of extending into and connecting with the primary visual centres in the brain. Although axonal connections were intact in AOS, we observed lower levels of AP staining. This suggested a reduction of RGC axons at the AOS.
To further inspect the functionality of these SpPou4f1/2expressing RGC axons, we injected CTB-488 into the left eyes and traced their propagation into the major brain targets. We compared coronal brain sections from wild-type and Pou4f2 SpPou4f1/2/SpPou4f1/2 mice, and detected comparable CTB-488 labelling in the LGN (the electronic supplementary material, figure S2p,q), SCN (electronic supplementary material, figure S2r,s) and OPN (electronic supplementary material, figure S2t,u). The CTB-488 signal in the SC, however, was relatively weaker in Pou4f2 SpPu4f1/2/SpPou4f1/2 mice than that in control mice (cf. arrowheads in electronic supplementary material, figure S2f,g), yet more CTB-488 signal was retained in the retinas of Pou4f2 SpPu4f1/2/SpPou4f1/2 mice than in control retinas (electronic supplementary material, figure S2l,m). This difference was even more evident in the SC regions of Pou4f2 SpPu4f1/2/Z and Pou4f2 þ/Z control mice (data not shown). These results indicated that axons emanating from RGCs expressing SpPou4f1/2 were capable of extending into and connecting with the primary visual centres in the brain; however, they are less effective in transporting macromolecules along the path.

(b) SpPou4f1/2] in the developing retina
Without Pou4f2, RGC precursors cannot perform axon related tasks including axon outgrowth, pathfinding and final targeting. Most undergo apoptosis [16]. This is probably due to a failure to activate critical genes required for RGC differentiation and survival. We compared lacZ expression in Pou4f2 þ/Z , Pou4f2 SpPou4f1/2/Z and Pou4f2 Z/Z E15 retinas to determine whether SpPou4f1/2 was able to generate a full complement of RGCs. lacZ expression in SpPou4f1/2 retinas was indistinguishable from that in Pou4f2 retinas; both were heavily labelled while slightly less labelling was found in Pou4f2 Z/Z retinas (electronic supplementary material, figure S4a-c). TUNEL analysis showed that while many RGCs were apoptotic in the ganglion cell layer of Pou4f2 Z/Z retinas (figure 3c), only an occasional apoptotic RGC was observed in Pou4f2 SpPou4f1/2/Z retinas, which looked identical to control Pou4f2 þ/Z retinas ( figure 3a,b). These results indicated that SpPou4f1/2 could replace Pou4f2 in sustaining RGC survival.
Several RGC-expressing genes directly activated by Pou4f2 have been identified, including another member of the Pou4f family, Pou4f1, which is expressed in many RGCs [23], and the T-box gene Tbr2/Eomes, whose expression is restricted to a small number of RGC subtypes [20]. Both of these genes were activated in RGCs of E15 Pou4f2 SpPou4f1/2/Z retinas with expression levels equal to those of Pou4f2 þ/Z retinas (figure 3d,e and g,h). As expected only background expression was observed in Pou4f2 Z/Z retinas (figure 3f,i). Thus, at least for Pou4f1 and Tbr2, SpPou4f1/2 can perform the necessary functions for binding to Pou4f consensus sites in the cis-regulatory regions of these genes and activating gene expression.
To quantify the extent to which SpPou4f1/2 could rescue the RGC differentiation in mature retinas, Pou4f1 and Tbr2 were chosen for the analysis because their expression is mutually exclusive [24]. Pou4f1 is expressed in approximately    [24,25]. By contrast, Pou4f1 expression marks distinct subtypes of RGCs [26]. Another essential RGC gene, Isl1, which is widely expressed in RGCs [27], has an expression pattern partially overlapping with that of Pou4f2. The precise timing and spatial expression patterns of Pou4f1 and Isl1 are required for the normal development of RGCs [26,27], and the expression of Tbr2 and Opn4 are indicators of a conserved Pou4f2-Tbr2-Opn4 genetic regulatory cascade as well as regulatory cascades for other Tbr2-expressing RGC subtypes [24]. We asked whether SpPou4f1/2 was able to reproduce this spatio-temporal programme in mature retinas in the absence of Pou4f2. This turned out to be the case. Double immunostaining P30 SpPou4f1/2-expressing retinas for Pou4f1 and Tbr2 showed the expected pattern of staining (electronic supplementary material, figure S5b). This pattern was very similar to that of control P30 retinas (electronic supplementary material, figure S5a).
SpPou4f1/2-expressing retinas co-immunostained with Opn4 and Isl1 also showed spatial expression patterns that were identical to those of wild-type mice (electronic supplementary material, figure S5c,d). As expected only a minor fraction of Isl1-expressing RGCs were co-expressed with Opn4. Notably, Opn4 staining reveals Opn4-expressing RGCs extending an extensive meshwork of neurite processes. These were readily observed in both Pou4f2 þ/Z and Pou4f2 SpPou4f1/2/Z retinas (electronic supplementary material, figure S5c,d).

(c) Electroretinogram responses for
Pou4f2 SpPou4f1/2/SpPou4f1/2 mice In Pou4f2 þ/þ mice, the dark-adapted ERG in response to weak stimuli, e.g. was dominated by two signals from the inner retina, positive ( p) and negative (n) scotopic threshold responses (STR) [21] (figure 4a). The very sensitive positive STR, present at stimulus strengths lower than those for which the b-wave contributes to the positive peak, relies essentially entirely on the integrity of RGCs for its generation  rspb.royalsocietypublishing.org Proc. R. Soc. B 283: 20152978 [21,28]. The negative STR amplitude is also impacted by the loss of RGCs, but the reported extent of loss has varied [28]. Pou4f2 2/2 mice lacked both waves originating from the inner retina that were present in the other genotypes ( figure 4a). The ERG responses to the weakest stimuli in Pou4f2 SpPou4f1/2/SpPou4f1/2 mice (figure 4a) were much more similar to those in Pou4f2 þ/þ mice than to those in Pou4f2 2/2 mice, indicating the presence of RGC function. In all three groups, the ERG responses to stronger stimuli, e.g. 21.4 log sc cd-s m 22 , were similar: the b-wave, thought to be generated by rod-driven bipolar cells increased in amplitude with stimulus strength, and for the strongest stimuli, e.g. 1.4 log sc cd-s m 22 , a negative-going a-wave, which reflects photoreceptor currents, was present at the beginning of the response [29].
To quantify the ERG responses, stimulus versus ERG amplitude plots were constructed based on measurements of the ERG waves, pSTR and b-wave, measured at the peak (figure 4b) and trough of the response (figure 4c), respectively. Figure 4d shows the amplitude measured at 7 ms after the flash on the leading edge of the a-wave. As predicted by the ERG traces (figure 4a; electronic supplementary material, table S1), the pSTR amplitudes were significantly lower in the Pou4f2 2/2 mice than in the other groups. For pSTR, amplitudes for the different groups were compared only in the boxed region in the pSTR/b-wave plot. The nSTR amplitudes of the Pou4f2 2/2 mice were also significantly lower than in Pou4f2 SpPou4f1/2/SpPou4f1/2 mice, and just missed being significantly lower the Pou4f2 þ/þ mice, whereas p-and nSTRs in Pou4f2 þ/þ and Pou4f2 SpPou4f1/2/SpPou4f1/ mice were not significantly different from each another. No other ERG amplitude measures were significantly different across the groups, although the b-wave amplitudes for the Pou4f2 2/2 mice tended to be lower than those for the other groups. Implicit times for measured responses also did not differ significantly across groups.

(d) SpPou4f1/2 expression in Strongylocentrotus purpuratus tube feet
Our results suggest that SpPou4f1/2 performs similar regulatory functions to those of Pou4f2 in photoreceptor neurons of S. purpuratus tube feet. It is also likely that SpPou4f1/2 is regulated directly or indirectly by the S. purpuratus orthologue of Pax6, as Pax6 is expressed in neuronal cells of the tube feet disc [9,10]. Our expectation was that SpPou4f1/2 would be expressed in tube feet photoreceptor cells and its expression would overlap with that of both Pax6 and the sea urchin photopigment Opsin4 [10]. Accordingly, we performed in situ hybridization with adult S. purpuratus tube feet using SpPou4f1/2 and Pax6 RNA probes and immunolabelling using an anti-Opsin4 antibody [10]. SpPou4f1/2 transcripts were detected in neuronal tissue associated with the tube feet disc skeletal rosettes (figure 5b,c,e-h). The SpPou4f1/2 staining pattern viewed from the top of the tube feet resembled that of an antibody against S. purpuratus synaptotagmin, which labels neuronal processes [8] (figure 5b,c). When viewed from the stalk side of the disc (figure 5b), SpPou4f1/2 expression was far more intense, as the majority of the nerve fibres lies on this side and enclose the skeleton from underneath, a finding confirmed by previous electron microscopic studies [10]. SpPou4f1/2 sense control showed no detectable signal over background (electronic supplementary material, figure S6). From SpPou4f1/2 expression, a more precise location of the previously reported expression of Pax6 could also be obtained [10]. Comparison of the two expression patterns showed that Pax6 was expressed in proximity to, but not overlapping with, SpPou4f1/2 expression (figure 5b,c). Probably due to reported differences in expression quantities between the tube feet stalk and the disc [9,12] (figure 5b,c), single-cell resolution could not be obtained for Pax6 expression in the tube feet disc.
A TEM three-dimensional reconstruction of serial sections of a tube feet disc illustrates cell position and identity in the area of SpPou4f1/2, Pax6 and Opsin4 localization (figure 5d). The distal portion of a disc rosette is accompanied by a massive nerve tissue. Two photoreceptor cells are adjacent to and above the large nerve tissue, which send cell extensions into it [10]. SpPou4f1/2 expression can be clearly correlated to the massive nerve tissue, while Pax6 expression correlates with the region where the cell bodies of photoreceptors are localized. Higher magnification reveals that SpPou4f1/2 expression colocalizes with Opsin4 protein in the same cells (figure 5f-h). While Opsin4 protein was localized in a portion of the photoreceptor cell bodies and in their dendrites, which contact the epidermal surface (figure 5e,g and h), SpPou4f1/2 expression is restricted to a smaller area of the photoreceptor cell body, close to the nucleus.
These results showed that SpPou4f1/2 mRNA and Opsin4 protein colocalized within the same photoreceptor cells. However, our results suggested that SpPou4f1/2 and Pax6 transcripts were not necessarily expressed within the same photoreceptor cell or alternatively that they were segregated into distinct domains within the same photoreceptor cell. Pax6 and SpPou4f1/2 are transcription factors that are largely localized to the cell nucleus. The non-overlapping or partially overlapping expression of Pax6 and SpPou4f1/2 transcripts might suggest that the two proteins are being translated in distinct regions of tube feet photoreceptor cells, or that Pax6 expression is not persistent in fully differentiated S. purpuratus Opsin4 positive photoreceptor cells.

Discussion
Our results indicate that SpPou4f1/2 can substitute for Pou4f2 and that there is a high degree of functional conservation between the two genes. In contrast with the much later role that Pou4f2 plays in RGC differentiation, Pax6 sits at the top of the hierarchical tier for eye development, setting up the eye field and activating a highly conserved gene regulatory network [30,31]. The subsequent processes that lead to the formation of the lens, retina and other tissues of the eye are regulated by transcription factor networks downstream of the Pax6 network. Our results support the notion that many genes downstream of Pou4f2 and SpPou4f1/2 are held in common.
Although Pou4f2 and SpPou4f1/2 share high sequence similarity in their Pou-specific and Pou-homeodomains, there is only weak similarity outside of these domains. The bipartite DNA-binding domains of POU domain factors confer versatility by flexible interactions with their DNA target sites [32]. Our finding that SpPou4f1/2 binds to DNA at Pou4f2 binding sites and activates many or all the genes that are activated by Pou4f2 may not be surprising, given the high degree of sequence rspb.royalsocietypublishing.org Proc. R. Soc. B 283: 20152978 similarity in the DNA-binding domains of the two proteins. However, POU domain transcription factors must interact with complex transcriptional machineries in order to function [33]. Several proteins are known to interact with POU domain factors at sequences mapping outside of the bipartite DNA-binding domains [33,34]. These interactions provide further functional specificity to target gene selection. SpPou4f1/2 is probably to interact with a multitude of proteins in tube feet photosensory neurons. The observed functional equivalence of Pou4f2 and SpPou4f1/2 implies that their interactions with co-activators, co-repressors and other components of the transcriptional machinery are also functionally conserved. Given the high degree of sequence divergence of Pou4f2 and SpPou4f1 outside of their Pou-specific and Pou-homeodomain, interactions with other proteins are likely to be confined to these conserved domains. However, the lack of sequence similarity outside the Pou-specific and Pou-homeodomains does not preclude a conserved role for at least some amino acids in these regions. We think it is unlikely that a Pou4f class protein with sequences chosen at random outside the conserved domains would properly fold into a functional protein.
Our experiments support the view of functional equivalence of the Pou4f factors. While many genes expressed in photosensitive neurons in the tube feet and RGCs in the retina are likely to be held in common, there are likely to be genes specifically required for the specialized neurons of each species. Tube feet-specific genes might have acquired consensus Pou4f DNA-binding sites at some point during rspb.royalsocietypublishing.org Proc. R. Soc. B 283: 20152978 echinoderm evolution. The corresponding orthologous genes in the mouse genome would not have Pou4f DNA-binding sites and would not be expressed in RGCs. Conversely, for genes specifically expressed in RGC differentiation, there would be a set of corresponding orthologous genes in the S. purpuratus genome that would not have SpPou4f1/2 DNA-binding sites. Low levels of SpPou4f1/2 expression in the developing retina were not surprising as its expression is under the control of the relatively weak Pou4f2 promoter and intron 1 was deleted in the KI allele. Nevertheless, the low expression of SpPou4f1/2 was sufficient to form fully functional RGCs, even in the complete absence of Pou4f2. Pou4f2 Z/Z mice are genetically null and Pou4f2 protein cannot be detected in mutant retinas [15,16]. However, Pou4f2 þ/Z mice are phenotypically wild-type. As Pou4f2 SpPou4f1/2/Z expression levels are significantly lower than Pou4f2 þ/Z expression in the developing retina, the threshold level for Pou4f2 necessary to function in RGC development is probably to be substantially lower. Data accessibility. The Pou4f2 SpPou4f1/2 mouse line generated in this study is freely available to all investigators upon request as stipulated in MD Anderson's Memorandum Transfer Agreement policies.