PAX6 Isoforms, along with Reprogramming Factors, Differentially Regulate the Induction of Cornea-specific Genes

PAX6 is the key transcription factor involved in eye development in humans, but the differential functions of the two PAX6 isoforms, isoform-a and isoform-b, are largely unknown. To reveal their function in the corneal epithelium, PAX6 isoforms, along with reprogramming factors, were transduced into human non-ocular epithelial cells. Herein, we show that the two PAX6 isoforms differentially and cooperatively regulate the expression of genes specific to the structure and functions of the corneal epithelium, particularly keratin 3 (KRT3) and keratin 12 (KRT12). PAX6 isoform-a induced KRT3 expression by targeting its upstream region. KLF4 enhanced this induction. A combination of PAX6 isoform-b, KLF4, and OCT4 induced KRT12 expression. These new findings will contribute to furthering the understanding of the molecular basis of the corneal epithelium specific phenotype.

KLF4 enhances the expression of keratins. We assessed the expression of OCT4 and KLF4 in the human corneal epithelium, and discovered that, whereas OCT4 was only sparsely expressed throughout the epithelial multiple layers, KLF4 exhibited relatively high expression levels, especially in the central-apical corneal epithelium where the cells are highly differentiated ( Supplementary Fig. S2c,d). Single-cell gene expression analysis showed that KLF4 expression was positively correlated with KRT3 and KRT12 expression in limbal epithelial cells in vivo ( Supplementary Fig. S2e). In addition, KRT12 expression in the mouse embryonic corneal epithelium (i.e., murine Krt12) increased dramatically at E18.5, which suggests the differentiation of the corneal epithelium, under the influence of high expression levels of Pax6 and Klf4 ( Supplementary Fig. S2f). The expression levels of other non-corneal epithelium-specific keratins were also altered by different combinations of the transduced factors (Fig. 2k). In particular, KLF4 had a large impact on the up-regulation of the differentiation marker keratins, KRT3, KRT10, KRT12, KRT13, KRT14, and KRT76.
The region upstream of the KRT3 gene is a target of PAX6-isoform-a transduction. Reporter assays were conducted to examine the transcriptional activity of PAX6 isoforms, OCT4, and KLF4 on the expression of KRT12 and KRT3. The KRT12 reporters did not significantly respond to any overexpression conditions, whereas the KRT3 reporters responded to the overexpression of PAX6-a, PAX6-a-KLF4, and PAX6-a-OCT4-KLF4 (Fig. 3a,b). Truncated PAX6 mutants (PAX6Δ PAI, PAX6-a-Δ RED, and PAX6-bΔ RED), even when combined with OCT4 and KLF4, induced KRT12 and KRT3 expression with a low efficiency (Fig. 3c,d), suggesting that both the PAI and RED domains are necessary for the induction of KRT12 and KRT3 expression. Furthermore,  (c-f) qRT-PCR of KRT12 mRNA levels at day 3 (n = 6). (g,h) qRT-PCR of KRT3 mRNA levels at day 3 (n = 6). (i) Immunofluorescence staining of KRT12 and KRT3 in PAX6-a-PAX6-b-OCT4-KLF4-, PAX6-a-OCT4-KLF4-and PAX6-b-OCT4-KLF4-transduced OKF6/TERT-1 cells. (j) Low-power field of immunofluorescence staining of PAX6-a-PAX6-b-OCT4-KLF4-transduced OKF6/TERT-1 cells laid over a phase contrast image. (k) Effect of various patterns of transduction on keratin mRNA levels measured by qRT-PCR after adjusting the total PAX6 mRNA level in the transduced OKF6/TERT-1 cells to match that of the in vivo corneal epithelium (n = 4 to 8). The scale numbers are presented as the log 10 of the relative gene expression. Human corneal epithelium and conjunctival epithelium in vivo were used as controls. Pa, PAX6isoform-a; Pb, PAX6-isoform-b; O, OCT4; S, SOX2; K, KLF4; M, c-Myc. The data presented in (c,d), (e,f), and (g,h) are from the same experiments. The data are presented as the mean ± SEM (c-h). *p < 0.01 and **p < 0.05 versus control by Dunnett's test. Scale bars represent 50 μ m (b,i) and 100 μ m (j). (a,b) Luciferase reporter assay using 1 to 6 K base pairs upstream of KRT12 (a) and KRT3 (b). The reporters were co-transfected with the PAX6-a, PAX6-b, OCT4 and KLF4 vectors or their combinations (n = 6). The luminescence was normalized to that of the samples co-transfected with lacZ. **p < 0.05 versus control by paired t-test with a Bonferroni correction. (c) Schematic representation of the truncated PAX6 mutants. (d) qRT-PCR of KRT12 and KRT3 mRNA levels in OKF6/TERT-1 cells transduced with the truncated PAX6 mutants, OCT4 and KLF4 (n = 4 to 9). The mRNA levels of cells transduced with full-length PAX6 from Fig. 2k are included for reference. Pa, PAX6-isoform-a; Pb, PAX6-isoform-b; O, OCT4; K, KLF4; PAI, PAI domain; RED, RED domain; HD, homeodomain; PST, proline/ serine/threonine-rich transactivation domain. The data are presented as the mean ± SEM (a,b,d).
Scientific RepoRts | 6:20807 | DOI: 10.1038/srep20807 co-immunoprecipitation followed by mass spectrometry showed that PAX6 did not form protein-protein complexes with the co-transduced factors OCT4 and KLF4 (Supplementary Table S2). These results suggest that PAX6 isoforms bind to their targets via both the PAI and RED domains without forming a complex with OCT4 and KLF4, and that the region upstream of the KRT3 gene is a target of PAX6-a transduction.
PAX6 isoforms regulate different genes by forming a highly complex regulatory network in OKF6/TERT-1 cells. We performed RNA-seq analyses with transduced OKF6/TERT-1 cells and detected 16,725 RefSeq-coding genes that were expressed in at least one cell type. Among these, 1,570 genes were significantly up-regulated in the transduced cells as compared to control cells (> 2 fold change [FC], < 0.05 false discovery rate [FDR]), 875 genes were up-regulated by both PAX6-a-OCT4-KLF4 and PAX6-b-OCT4-KLF4 transductions, and 695 genes were uniquely up-regulated by each of the transductions (Supplementary Fig.  S3a). Remarkably, 24% of the up-regulated genes (382/1,570) were differentially regulated by each of the transductions (> 2 FC, Supplementary Table S3), which suggests a regulatory specificity of PAX6 isoforms, 162 by PAX6-a-OCT4-KLF4 transduction and 220 by PAX6-b-OCT4-KLF4 transduction (Fig. 4a). Interestingly, the 220 PAX6-b-OCT4-KLF4-dependent genes were involved in biological processes related to keratinocyte development and differentiation (Fig. 4a) and various keratins were differentially up-regulated (Fig. 4b), which is consistent with the results of the transduction experiments reported in Fig. 2k.
To infer the activity of potential key transcription factors that contribute to the regulation of differentially up-regulated genes (DUGs), we used a linear regression model with putative transcription factor-binding sites (TFBSs) of TRANSFAC 23 found from DUG promoters. By following a promoter regression modelling 24 , we exhaustively searched for the best combination of TFBSs that explains the expression levels of DUGs. In our dataset, we found 103 TFBSs in total, consisting of 45 identical and 40 opposite mean regression coefficients (RCs) in the transductions (Fig. 4c), of which nine were unique to PAX6-a-OCT4-KLF4-and nine to PAX6-b-OCT4-KLF4-dependent DUGs (Supplementary Table S4). Specifically, TFBSs for Pax-6 bound by PAX6-b presented positive RCs, whereas those bound by PAX6-a could be positive or negative, reflective of the broader transcription activities of PAX6-a. Pax-6 activities were dramatically altered to display negative RCs when keratins were removed from DUGs ( Fig. 4d), implying that PAX6 isoforms regulate keratins in the manner of activators.
To visualize the interactions between the potential regulators and DUGs, we built a network based on the results of the promoter modelling and information from a database. We first prepared TRANSFAC transcription factors that are known to bind to TFBSs listed in Fig. 4c. Of these, we further narrowed down to 22 transcription factors, whose coding genes were categorized into 1,570 up-regulated genes. We then linked the 22 transcription factors to 382 DUGs that present binding sites for these transcription factors (Supplementary Fig. S3b and Supplementary Table S5). This network, consisting of 5,170 links, revealed that, although a few transcription factors potentially could regulate either PAX6-a-OCT4-KLF4-or PAX6-b-OCT4-KLF4-dependent DUGs, the majority of these factors shared targeting genes. In particular, PAX6 was connected with all DUGs, and POU5F1 (i.e., OCT4) was linked to 356 out of 382 genes. In addition, we conducted this analysis using the OKF6/TERT-1 cells that had been only transduced with PAX6-a or PAX6-b ( Supplementary Fig. S3c Table S5).

KRT12 induction requires high expression of transgenes.
To elucidate the precise expression pattern of transduced OKF6/TERT-1 cells, we performed single-cell gene expression analysis (Fig. 5a). Individual KRT12-positive cells expressed this keratin as highly as that observed in the corneal epithelium in vitro (Fig. 5b). Our experiments show that more KRT3-positive cells than KRT12-positive cells were induced, suggesting that KRT3 induction was more readily achieved (Fig. 5c). KRT12-and KRT3-positive cells expressed PAX6-a, PAX6-b, OCT4, and KLF4 at higher levels than KRT12-and KRT3-negative cells (Fig. 5b,c), suggesting that the efficient transduction of transgenes supports the induction of KRT12 and KRT3.
The other expressed genes are shown in Supplementary Fig. S4a. KRT3-positive cells also expressed clusterin (CLU), one of the most abundant genes in the corneal epithelium 25 . CLU expression was positively correlated with PAX6-a (r = 0.568, p < 0.01), but not with PAX6-b (r = −0.101, p = 0.08) ( Supplementary Fig. S4b), consistent with the up-regulation by PAX6-a-OCT4-KLF4 shown by RNA-seq ( Fig. 4b and Supplementary Table S3). Among the other genes that were abundant in the corneal epithelium, ALDH3A1 and TKT were also up-regulated under certain conditions ( Supplementary Fig. S4a). KRT12-and/or KRT3-positive cells also tended to express higher levels of the differentiation markers KRT10, KRT13, and KRT14 ( Supplementary Fig. S4a).
Effects of the epigenetic state on KRT12 induction. We next examined the effect of the epigenetic state on KRT12 induction, for which a reprogramming factor, OCT4, is required. The expression of NANOG, a pluripotency marker regulated by OCT4 26,27 , was slightly elevated in KRT12-expressing OKF6/TERT-1 cells (Fig. 6a). Overall, however, the pluripotency markers, including NANOG, SSEA4, and TRA-1-60, were not strongly detected by immunofluorescence staining, suggesting that their expression levels are not as high as in induced pluripotent stem cells (iPSCs), which are in an undifferentiated state ( Supplementary Fig. S5a-c). KDR, a marker of early mesodermal cells, and SOX17, a marker of early endodermal cells, were both slightly up-regulated in KRT12-positive cells under certain conditions (Fig. 6a), suggesting that the epigenetic state modified by OCT4 is one of the possible factors that affect the expression of KRT12 28,29 .
We added small molecules that are known to contribute to the maintenance of the undifferentiated state and the improvement of the efficiency of reprogramming to iPSCs (Fig. 6b) [30][31][32] . The induction of KRT12 and KRT3  Table S1), the induction of KRT12 and KRT3 was enhanced by 6-bromoindirubin-3′ -oxime (BIO), a glycogen synthase kinase-3 (GSK-3) inhibitor (Fig. 6b), which supports the involvement of Wnt signalling in the corneal epithelial phenotype 33,34 . Corneal epithelium-specific keratins are preferentially induced in surface ectoderm-derived cells. Next, we transduced PAX6-a, PAX6-b, OCT4, and KLF4 into different types of human cells (Fig. 6c and Supplementary Fig. S5d). All surface ectoderm-derived cells examined (OKF6/TERT-2 cells from another oral mucosal epithelial cell line, N/TERT-1 and N/TERT-2 cells from dermal epithelial cell lines, and human oral keratinocytes [HOK] from a primary oral epithelium) expressed KRT12 at high levels after being transduced with PAX6-b-OCT4-KLF4. The same cells expressed KRT3 at high levels after being transduced with PAX6-a-OCT4-KLF4. Slightly lower expression levels of KRT3 were induced by PAX6-a in the absence of OCT4 and KLF4. The non-surface ectoderm-derived epithelial cells (ARPE-19, MKN1, and HepG2) tended to express KRT12 and KRT3 mRNA, but few KRT12-and KRT3-positive cells were detected by immunofluorescence staining. iPSCs that express OCT4 endogenously did not show any KRT12 and KRT3 expression following the transduction of PAX6 and KLF4, suggesting that KRT12 and KRT3 induction in the surface ectoderm-derived cells did not progress to a de-differentiated state.

Discussion
Limbal stem cell deficiency, which is very difficult to treat, is one of the most severe diseases of the ocular surface and causes a significant corneal epithelial defect and vision impairment 35 . Several research groups, including ours, have attempted to successfully transplant cultured autologous oral mucosal epithelial sheets to treat eyes with bilateral limbal stem cell deficiencies 12,13 . This approach has shown some promise; however, the reconstructed ocular surface does not allow a full, long-term improvement of visual acuity, and is prone to problems associated with neovascularization. This is probably because at least in part the oral mucosal epithelium lacks corneal epithelial specific genes, such as PAX6 and KRT12 9,10 . To minimize such limitations, a much greater understanding of the molecular mechanisms underlying the regulation of corneal epithelium-specific genes, especially the genes controlled by PAX6, is urgently required. Here, we report, for the first time, the effect of two isoforms of PAX6 on corneal epithelium-specific genes, particularly KRT3 and KRT12.
PAX6-a and PAX6-b display differential inductive properties on KRT3 and KRT12 and regulate different genes by forming a highly complex regulatory network. Thus, these two isoforms of PAX6 cooperatively regulate genes in the epithelium. Moreover, the upstream region of the KRT3 gene appears to be a direct target of PAX6-a transduction. Although previous work has concluded that PAX6-a binds to its target by PAI domain and PAX6-b binds to its target by RED domain 3,5 , we show that the entire PAI and RED domains are critical for KRT3 and KRT12 induction.
Following the addition of the Yamanaka factors, we discovered that KLF4 was important for the efficient induction of KRT3 and KRT12. At the outset of these experiments, we expected that KLF4 would work as a reprogramming factor 17,18,36 , but it appeared to act as an accelerator of the expression of differentiation marker keratins 37,38 . Additionally, a modified epigenetic state by OCT4 and BIX may be one of the interventions that improves KRT12 induction in oral mucosal epithelial cells 29,39 . The low expression levels of OCT4 in the corneal epithelium also point to the fact that it likely works indirectly on the regulatory region of KRT12, through the modification of its epigenetic state in oral mucosal epithelial cells. However, unlike KRT12 expression, sufficient KRT3 expression was induced without OCT4 (i.e., with PAX6-a or PAX6-a-KLF4), suggesting that the KRT3 regulatory region is readily accessible for the binding of transcription factors without epigenetic modification.
The efficiency of KRT12 induction was found to be relatively low, even though we used a uniform cell line. This can be explained by the requirement for high expression levels of the transgenes for KRT12 induction, which provides new insight into the low efficiency of iPSC generation by transduction with the four Yamanaka factors 17,18,40 . As most cells expressed KRT3 following PAX6-a-OCT4-KLF4 induction, the high expression of the transgenes does not seem to be necessary for the induction of KRT3. Our experiments revealed that the cell source was another important factor for the induction of tissue-specific genes, even when utilizing epigenetic modifications by OCT4. As summarized by Chin, most studies relating to direct reprogramming used fibroblasts in their experiments 28 . Our data suggest that the selection of appropriate cell types makes it easier to overcome the inherent epigenetic differences among tissue-specific genes and increases the chance of achieving direct reprogramming. In fact, PAX6 is able to induce KRT12 expression on its own in epithelial cells from the corneal pannus or eyelid, the origins of which are close to those of the corneal epithelium 34,41 .
In summary, we report that two PAX6 isoforms, alongside OCT4 and KLF4, differentially and cooperatively regulate corneal epithelium-specific genes, particularly KRT3 and KRT12, as well as many other genes in surface ectoderm-derived cells. The role of each transcription factor is summarized in Fig. 7. Our new findings will contribute to further our understanding of the molecular basis of the corneal epithelium specific phenotype.  Table S7) and a Taqman ® Fast Universal PCR Master Mix (Life Technologies). We also designed two types of PAX6 primers and probes, to distinguish the expression of the two variants of PAX6, as described in the Supplementary Information. For the SYBR green quantitative PCR, cDNA was applied to quantitative PCR  Table S8) and SYBR Premix Ex Taq TM GC (TaKaRa Bio, Otsu, Japan). cDNAs from the islet cells (Primary Cell Co., Sapporo, Japan) and ocular tissue (SightLife) were used as the positive controls for the qRT-PCR. The final mRNA levels were normalized to the GAPDH levels.

Methods
Single-cell gene expression analysis. The single-cell gene expression analysis was performed using a Fluidigm Single-Cell Gene Expression Workflow system (Fluidigm, San Francisco, CA, USA). The Taqman probes used are listed in Supplementary Table S7.
Preparation of RNA-seq libraries and analysis of RNA-seq data. RNA libraries from four replicates of each transduced cell type were assembled using ISOGEN (Wako Pure Chemical Industries) and the TruSeq RNA Sample Prep kit, v2 (Illumina, San Diego, CA, USA), according to the manufacturer's protocol. After purification and fragmentation, the mRNAs were sequenced by Illumina HiSeq 2000/2500, which generated 20.9-81.8 million 101-bp paired-end reads. The sequenced reads were mapped and quantified by the TopHat2 (v.2.0.7)/Cufflinks (v.2.0.5) pipeline 44 , with 84-91% of total reads uniquely mapped (Supplementary Table S9) and that normalized as the unit FPKM (fragments per kilobase of exon per million mapped reads). Differentially expressed genes were detected by Cuffdiff program in the Cufflinks package with parameters set as < 5% false discovery rate (FDR) and > 2.0 fold change (FC), as described in the Supplementary Information. Regression promoter modelling. The procedure of computational promoter modelling 24 was performed to detect the potential key regulators (Supplementary Information). This approach exhaustively searches well-fitted linear regression models that infer the importance of TFBSs to explain the gene expression levels. The TFBSs and the relevant transcription factors were prepared with the TRANSFAC professional 23 and MATCH tools 45 . The statistical significance of the inferred TFBSs was tested by a one-sample t-test after a Bonferroni correction (< 0.01 p-value). The networks ( Supplementary Fig. S3b,e and Fig. 4e) were visualized using Cytoscape software (www.cytoscape.org).

Co-immunoprecipitation (Co-IP) and mass spectrometry (MS). We used an EpiXplore TM Nuclear
Co-Immunoprecipitation Kit (Clontech Laboratories) to identify the protein-protein complexes. The nuclear extracts were incubated with a rabbit anti-PAX6 antibody (1:50, Abcam, Cambridge, UK) for 24 h. The purified proteins were subjected to sodium dodecyl sulfate poly-acrylamide gel electrophoresis (SDS-PAGE), followed by the extraction of the target lanes and an analysis using liquid chromatography-mass spectrometry (LC-MS/MS, Thermo Fisher Scientific, Waltham, MA, USA).