A trans-Regulatory Code for the Forebrain Expression of Six3.2 in the Medaka Fish*

Background: The transcription factor Six3 is key element of forebrain specification, but its upstream regulators are unknown. Results: A systematic search in medaka fish identifies and functionally characterizes novel Six3 regulators. Conclusion: The spatio-temporal regulation of Six3 depends on a few trans-acting factors. Significance: This study provides new information on how forebrain neuronal diversity is originated. A well integrated and hierarchically organized gene regulatory network is responsible for the progressive specification of the forebrain. The transcription factor Six3 is one of the central components of this network. As such, Six3 regulates several components of the network, but its upstream regulators are still poorly characterized. Here we have systematically identified such regulators, taking advantage of the detailed functional characterization of the regulatory region of the medaka fish Six3.2 ortholog and of a time/cost-effective trans-regulatory screening, which complemented and overcame the limitations of in silico prediction approaches. The candidates resulting from this search were validated with dose-response luciferase assays and expression pattern criteria. Reconfirmed candidates with a matching expression pattern were also tested with chromatin immunoprecipitation and functional studies. Our results confirm the previously proposed direct regulation of Pax6 and further demonstrate that Msx2 and Pbx1 are bona fide direct regulators of early Six3.2 distribution in distinct domains of the medaka fish forebrain. They also point to other transcription factors, including Tcf3, as additional regulators of different spatial-temporal domains of Six3.2 expression. The activity of these regulators is discussed in the context of the gene regulatory network proposed for the specification of the forebrain.

A well integrated and hierarchically organized gene regulatory network is responsible for the progressive specification of the forebrain. The transcription factor Six3 is one of the central components of this network. As such, Six3 regulates several components of the network, but its upstream regulators are still poorly characterized. Here we have systematically identified such regulators, taking advantage of the detailed functional characterization of the regulatory region of the medaka fish Six3.2 ortholog and of a time/cost-effective trans-regulatory screening, which complemented and overcame the limitations of in silico prediction approaches. The candidates resulting from this search were validated with dose-response luciferase assays and expression pattern criteria. Reconfirmed candidates with a matching expression pattern were also tested with chromatin immunoprecipitation and functional studies. Our results confirm the previously proposed direct regulation of Pax6 and further demonstrate that Msx2 and Pbx1 are bona fide direct regulators of early Six3.2 distribution in distinct domains of the medaka fish forebrain. They also point to other transcription factors, including Tcf3, as additional regulators of different spatial-temporal domains of Six3.2 expression. The activity of these regulators is discussed in the context of the gene regulatory network proposed for the specification of the forebrain.
The progressive specification of the forebrain (or of any other developing structure) depends on the synchronized activity of intrinsic genetic programs and extrinsic signals, which form well integrated and hierarchically organized gene regulatory networks (GRNs). 3 Perturbation of the GRN status causes abnormal development of the involved structure, which is also considered a mechanism that enables the acquisition of morphological innovations during evolution (1,2). Components of a GRN with a high hierarchical and central position are known as hubs. In general, hubs are transcription factors (TFs), which regulate the many other components of the network and which are, in turn, regulated by a relatively low number of genes (1). Six3, a homeobox-containing TF belonging to the Six/sine oculis family, is one of the hubs of the GRN responsible for forebrain specification (2).
In contrast to the relatively well characterized downstream targets of Six3 activity and despite the detailed characterization of the cis-regulatory code that controls the spatial-temporal expression of the medaka Six3.2 and zebrafish Six3a paralogs (23)(24)(25), only a few of the TFs that govern Six3 expression have been so far identified. Six3 appears to regulate its own transcription with a context-dependent positive or negative feedback loop (25,26). There is also evidence that Six3 expression is restrained anteriorly by BMP-mediated Lmo4 activity (26) and posteriorly by Wnt signaling (13). In both cases, it is still unclear whether the mechanism is direct.
Taking advantage of the identified cis-regulatory code and of available TF binding site (TFBS) prediction tools, we have recently demonstrated that the neural inducing factor Sox2 directly controls the expression of the medaka Six3.2 (21). However, in silico prediction of putative candidates has limitations because it depends on available TFBSs and does not account for transcriptional modifications imposed by the interaction with a given cofactor. Thus, to extend our search and generate a more comprehensive scenario of the relevant Six3 regulators, we have reinforced our in silico prediction approach with a trans-regulation screening (TRS). This screening is a systematic survey of potential regulators, which has already been successfully used to identify trans-acting factors of the medaka Ath5 (27) and of the zebrafish eng2a genes (28). The latter study used a preselected version of the library enriched in characterized developmental regulators, thus improving the time/cost efficiency of the screening. By combining this midscale TRS with our in silico prediction-based approach over a well characterized cis-regulatory code (23), we have identified a number of potential upstream regulators of the medaka Six3.2 gene, further improving the time/cost efficacy of the approach over previous studies (21,27,28). These regulators were further validated with different combinations of expression pattern analysis, dose-response luciferase (Luc) assays, chromatin immunoprecipitation (ChIP), and functional studies. Our results indicate that Msx2, Pbx1, and Pax6 are bona fide direct regulators of Six3.2 expression at early stages of medaka fish forebrain development and point to Tcf3, Etv4/5, Nkx2.2, Prdm1/Blimp1, and Vsx1 as additional TFs responsible for Six3.2 expression at different developmental stages.

Experimental Procedures
Medaka Stocks-Wild type (WT) medaka fish (Oryzias latipes) of the Cab strain and the transgenic line Six3.2cI::EGFP, expressing the EGFP reporter under the control of the fulllength Six3.2 regulatory region (23), were maintained in an inhouse facility (28°C on a 14/10-h light/dark cycle). Embryos were staged as described previously (29).
In Silico Prediction of TFBS-Sequences of the medaka, zebrafish (Danio rerio), fugu (Takifugu rubripes), stickleback (Gasterosteus aculeatus), and tetraodon (Tetraodon nigroviridis) genomes were obtained from the USCS genome browser. Multiple alignments of the 4-kb regulatory region of the medaka Six3.2 gene (23) were performed using the Mulan and mVista tools (Fig. 1A). Evolutionarily conserved TFBSs were identified combining rVista (multi-TF), Jaspar, and Matinspector. In some cases, we found BSs with partially overlapping sequences. When overlap was more than 80% and the BS belonged to closely related TFs, we considered the presence of only one BS. Candidates were further screened according to gene expression data available in the literature and in the following databases: GeneCards, OMIM, Emage, and zFIN. Candidate TFs were considered "related" if their expression was overlapping with or complementary to that of Six3.2. The distribution of these regulators was plotted along the Six3.2 regulatory region using GraphPad software ( Fig. 1B and supplemental material), and these candidates were considered for subsequent experimental validation. Thus, relevant members of each one of the identified TF families were assayed in Luc assays (Table 1).
Plasmid Construction-For the TRS, the 4-kb regulatory region of the medaka Six3.2 gene, including the minimal promoter region and the 5Ј-UTR, was cloned into the pGL3b vector (Promega), carrying the firefly Luc gene to generate the pSix3.2-full reporter vector. Reporter constructs carrying different combinations of the Six3.2 cis-regulatory modules upstream of the thymidine kinase minimal promoter in the pGL3b-TK vector ( Fig. 2A) were used to validate in silico predicted candidates (21) ( Table 1). Information about the plasmids coding for the in silico-predicted candidate TFs used in the Luc assays is available upon request. The Tcf3-HA-VP16 and Tcf3-HA-En were kindly provided by Dr. E. Marti. Modified versions of the pCS2 plasmid, carrying either the HA-Eng or the HA-VP16 proteins, were generated to obtain constitutive repressive or activating forms of the TFs and then used for mRNA transcription and injection experiments. To this end, the HA-VP16 or the HA-Eng cassettes were amplified with PCR-specific primers and cloned into the pCS2ϩ vector using specific restriction enzymes. The CDS of the different TFs was subsequently amplified with specific primers and subcloned either in the pCS2-HA-En (Pbx1, Pax6) or pCS2-HA-VP16 (Msx2). Deletions of the Pax BS present in the D elements were produced by PCR amplification using the Phusion DNA polymerase (New England Biolabs) and PCR-specific primers. The obtained PCR product was subsequently purified, treated with the polynucleotide kinase, self-ligated using the T4 DNA ligase, and transformed in DH5a-competent cells.
TRS-The TRS was performed as previously described (27). Briefly, we individually tested 1064 clones (supplemental material) belonging to a Unigene library enriched in genes considered as "developmental regulators." BHK21 cells seeded in 96-well plates were co-transfected with 2 l (10 -300 ng) of each cDNA, 40 ng of the pSix3.2-full reporter vector, and 5 ng of pRL-CMV using the Fugene 6 transfection reagents (Roche Applied Science) according to the manufacturer's protocol. Cells were cultured and transfected with DMEM supplemented with 10% FCS. The quality threshold was set above 1,000 relative units for Renilla luminescence and above 10 times the background signal (10 raw units) for firefly luminescence. Clones that activate the reporters with lower values were discarded. Raw luminescence readings were stored in a File-Maker database. Experiments were performed in triplicate, and the median values were normalized and analyzed using MS Excel. To account for plate-to-plate variations, each firefly/Renilla ratio was normalized against the average of all ratios in the plate as described (27). Nevertheless, there was no significant difference in the average values of this ratio

trans-Regulatory Factors for Six3 Forebrain Expression
among plates (supplemental material), supporting the reproducibility and robustness of this approach. Selected candidate clones were further tested in a dose-response Luc assay performed in triplicate by co-transfecting 20, 40, 80, or 160 ng of pCMV-Sport6.1::cDNA with the reporter and constructs. The total amount of transfected DNA was kept constant by adding the necessary amount of pCS2ϩ vector. Dose-response experiments were performed with both CMV-and SV40-driven Renilla vectors to discard clones with a possible off-target effect on the Renilla promoter and to avoid the influence of possible difference in the expression efficiency of the different plasmids. Each clone and dose was assayed in triplicate, and the firefly/ Renilla ratio was statistically compared using an appropriate t test. Clones were considered positives when at least one of the tested doses induced a response at least 75% higher (activation) or 50% lower (repression) than that observed with the dose of 20 ng (supplemental material). Clones were also considered positive if the firefly/Renilla ratio at 20 ng was significantly different from the average value of the other clones at the same dose, independently of their dose response.
Cell Transfection and Luc Assay-Luc assays were performed using the undifferentiated and pluripotent P19 teratocarcinoma line cultured in minimum Eagle's medium ␣ with Glutamax (Gibco). Cells were transfected with 310 ng of DNA/well composed of the reporter plasmid (pSix3.2-cI-cVI, 50 ng), a CMV-driven Renilla Luc control vector (pRL-CMV; 10 ng), increasing amounts (25-250 ng) of the effector plasmid, and, when needed, variable amounts of the empty vector to kept the total amount of the transfected DNA constant. Transfections were performed with Fugene HD, following the manufacturer's instructions. To minimize possible bias imposed by differences in the expression level of the various plasmids, we normalized the reporter luciferase activity to a control Renilla reporter and then to the basal activity of the reporter construct when cotransfected with GFP. After normalization, the raw values of basal activity of the cI-cVI constructs were statistically indistinguishable and quite reproducible among experiments (data not shown). Cells were maintained in the same medium containing 5% FCS for an additional 48 h, when the Luc activity was measured using the Dual Luc reporter assay (Promega) following the manufacturer's instructions. Experiments were performed in triplicate and replicated at least three times. Data are presented as -fold induction of the Luc activity observed in the control vector normalized to 1.
Chromatin Immunoprecipitation Assays-ChIPs were performed as described (21), using a Myc-or HA-tagged version of the selected candidate genes. Briefly, P19 cells were transfected with reporter/effector plasmids (3:1 ratio, 10 g total). ChIP was performed using the anti-Myc 9E10 (Sigma) or the anti-HA 3B9 (Active Motif) monoclonal antibody. The relative enrichment of each one of the Six3.2 regulatory elements was determined by quantitative PCR (Roche Applied Science) using specific primers (information available upon request). Primers designed in the Luc coding sequence and in the genomic locus of the 18S RNA were used as negative controls. -Fold enrichment was expressed as the ratio of Myc to IgG signal and calculated with the expression, 2 Ϫ(⌬⌬Ct) , where ⌬⌬Ct ϭ Ct c-Myc Ϫ Ct IgG .
mRNA Overexpression Studies-The mRNA used in overexpression experiments were synthesized with mMESSAGE mMACHINE kits (Ambion) and co-injected with the RFP reporter (10 ng/l) into one-cell stage Six3.2-cI::EGFP transgenic embryos at concentrations ranging from 30 to 100 ng/l for each one of the tested mRNA. Within this range, we selected the best working concentration as that which gave the most evident phenotype without affecting embryo survival. Changes in Six3.2 expression were determined by in situ hybridization (see below) and EGFP reporter expression using a Leica fluorescent stereomicroscope. We injected and analyzed a minimum of 30 to a maximum of 250 embryos for each one of the tested mRNAs.
Whole-mount in Situ Hybridization-Antisense riboprobes for the identified candidates were synthesized from the pCMV-Sport6 clone containing the full-length mRNAs according to standard protocols. Probes for the medaka Tcf3, Etv4, Etv5, and Msx2a genes were amplified by PCR using specific primers (information available upon request) and cloned into the pSCA-Strataclone vector (Stratagene). The Six3.2 probe used in this study is as described (23). In situ hybridizations were performed as described (23).

Identification of Potential trans-Regulators of Six3.2 Expression-
The regulatory region sufficient to recapitulate the entire spatio-temporal expression of the medaka Six3.2 gene is composed of 10 conserved regulatory elements (CREs) distributed in a 4-kb region located upstream of the coding region ( Fig.  1, A and B) (23). Two of these CREs, known as D and IL elements, act like enhancers during early and late forebrain development, respectively (23). The A element instead silences Six3.2 expression in the hindbrain and neural tube, whereas the E, F, G, and H elements modulate D or IL function (23). In silico analysis with different prediction tools (see "Experimental Procedures") identified several TFBSs randomly distributed along the Six3.2 regulatory region ( Fig. 1B), but the addition of evolutionary conservation as a criterion reduced their number and restricted their localization to the A, C, D, E, G, and IL CRE ( Fig.  1, A and B). The identified TFBSs belonged to families of known forebrain regulators, including, among others, Pax, Sox, Gli, Tcf/Lef, and Six (Table 1 and supplemental material).
Because in silico prediction is limited by the existence of defined consensus BSs, we undertook a TRS with the aim of identifying additional trans-acting factors for the Six3.2 locus. To this end, we took advantage of a preselected medaka Unigene expression library, composed of 1064 full-length cDNAs mostly encoding bona fide "developmental regulators." These cDNAs were individually screened for their ability to modulate the expression of a Luc reporter upon co-expression in BHK21 cells. As demonstrated previously (27), with this approach, only a small number of tested cDNAs are likely to have an effect on the regulatory region of choice, and the variation of the normalized luminescence ratios around the average is expected to be random for most of the clones, thus fitting a Gaussian curve ( Fig. 1C and supplemental material). By contrast, values outside the normal range are likely to be specific, reflecting a direct or indirect regulatory relationship of the candidate gene with the

trans-Regulatory Factors for Six3 Forebrain Expression
JOURNAL OF BIOLOGICAL CHEMISTRY 26929 tested genomic region. Using this criterion, we considered that clones with a normalized ratio lower than 0.2859 or higher than 2.8732 could be considered putative repressors and activators, respectively, of the Six3.2 regulatory region ( Fig. 1C and supplemental material). Fifty-eight clones, with an apparent repressor (40) or activator (18) effect and representing 5.4% of

trans-Regulatory Factors for Six3 Forebrain Expression
the screened clones, fulfilled this criterion. The majority of the positive clones were TFs, signaling molecules, or signaling transduction components (Fig. 1D). With the exception of two clones, for which no information was available, the remaining clones were classified as chromatin modifiers, cell cycle regulators, or RNA-binding proteins (Fig. 1D). Notably, at least one putative conserved BS within the Six3.2 regulatory region could be identified for each one of the regulatory TFs selected in our screening for which a consensus BS had been described (supplemental material).
Dose-response Luc Assays Confirmed Several TRS Candidates as Six3.2 Regulators-Potential regulators of Six3.2 expression should activate/repress its regulatory region in a dose-depen-dent fashion. To select the candidates with this behavior among the 58 identified genes, we performed Luc assays co-transfecting BHK21 cells with increasing doses of the selected cDNAs, the pSix3.2-full plasmid, and either the CMV-or SV40-driven Renilla plasmids (Table 2; see supplemental material for a full version of the data). This assay had two additional purposes: to discard clones with an off-target effect over the CMV:Renilla plasmid and to confirm the repressive/activating nature of the candidates. The TFs Pbx1, Nkx2.2, Etv4, Prdm1, and Hoxb1b; the signaling protein Wnt9b; Cap1, a component of the cyclic AMP pathway; the laminin receptor integrin-␤4 subunit (Itg␤4); the histone deacetylase 1 (Hdac1); the Nedd4-binding partner-1 (N4bp1); and one of the unknown candidates acti-  (Table 2). Although the TF Vsx1 and the cell cycle regulator Cins1a (Cdk2) did not elicit a significant dose response, they both activated the Six3.2 promoter above the average basal activity of the remaining clones even when used at a low dose (Table 2 and supplemental material). The Fzd7 receptor and the medaka homolog of the Znf467 gene instead significantly reduced pSix3.2-full reporter activity in a dose-dependent fashion, whereas low doses of the GTPases Rac and RhoA, the apoptosis-related Bcl2l10 gene, and the RNA-binding protein Quaking repressed the firefly/ Renilla ratios below the average of the remaining clones (Table  2 and supplemental material).
To extend this analysis to the in silico predicted regulators, we transfected P19 cells with plasmids encoding the candidates and the reporter constructs harboring different combinations of the Six3.2 CREs (Figs. 2 (A-E) and 3 and Table 1). These constructs were used to identify the relevant bound regions because putative BSs for the selected TFs were often found in more than one CRE (Table 1 and Fig. 2F). As a positive control for the assay, we co-transfected an expression plasmid for Sox2 (Fig. 3A) (data not shown), previously described as activating the Six3.2 locus (21).
Pax6, Pbx1, and a constitutively active form of Tcf3 (Tcf3ca), generated by fusion of the VP16 activator domain (30), activated the Luc reporter of the constructs harboring distinct CRE combinations (Fig. 2, A-D). Pax6, but not Pax2 (Figs. 2B and  3C), which recognizes a similar BS (31), induced up to a 6-fold activation of the reporter constructs carrying the D module (cII and cIII; Figs. 2B and 3 (C and D)), in line with the presence of conserved Pax BS in the D box (Table 1 and Fig. 2F), but, as reported for Sox2 (21), it failed to activate the cI construct (Fig.  2B), containing the A repressor module. Notably, deletion of the two Pax6 BSs at the 5Ј-end of the D element strongly decreased reporter activation, whereas elimination of the additional Pax6 BSs had only a very modest additive effect (Fig. 3D), suggesting that this 5Ј portion of the D CRE mediates most of the Pax6 activity.
As observed for Pax6, Tcf3ca and the related Lef1ca activated in a dose-dependent manner all reporter constructs containing the H and L CREs (Figs. 2C and 3E), in which conserved BSs were found (Fig. 2F), suggesting that different members of the Tcf/Lef family may participate in the regulation of Six3.2 expression.
The strong effect of Pbx1 in the trans-regulatory screening and the presence of evolutionarily conserved Pbx BS in the A and L modules ( Fig. 2F and Table 1) made Pbx1 one of the strongest candidates for a Six3.2 trans-acting factor. Pbx TFs form complexes with members of the Meis family (32)(33)(34)(35)(36). We thus tested the effect of Pbx1 on different Six3.2 constructs transfected in P19 cells and asked if its activity was enhanced in the presence of Meis1b. Compared with the strong activation observed in the TRS, Pbx1 was less effective in activating reporter expression in P19 cells, although activation was significant (cI, cII, and cV; Fig. 2D). Meis1b, which alone had little

Summary of the TRS-selected candidates validated by dose-response Luciferase assays
For each of the factors, the data were normalized to the lowest dose employed in the study. Values above or below the reference value are color-coded: in graded greens for activation and in graded reds for inhibition. cDNAs were cotransfected with a CMV or a SV40 driven Renilla vector. The candidates were considered positive when they significantly activated or repressed the reported activity compared to their reference values at least at one of the doses in both assays. Their effects are represented with green or red arrow respectively. Gray arrows represent candidates that did not display significant changes in luciferase activity at the doses tested in this study but whose reference values were significantly higher (putative activator) or lower (putative repressor) than the average of the remaining clones. effect, enhanced Pbx1-induced reporter activation only modestly (Fig. 2D), raising the possibility that additional cofactors might be required for the stabilization of the Meis-Pbx1 complex, as already shown in other contexts (37,38,39). Alternatively, other TALE family members may be more suitable cofactors in this context. In a similar way, we observed mild or no cooperation between Sox2 and Pax6 or Otx2 on Six3.2 regulation (Fig. 3, F and G), in contrast to reports for other target genes (40,41). Furthermore, we could not find a previously proposed Six3 autoregulation (25,42), neither with Six3.2 nor with its Six3.1 paralog or their mammalian orthologue (Fig. 3, A  and B), suggesting that in medaka fish, other cofactors might be required or that the Six3 BS might be located outside of the 4-kb region that we have characterized. Similar reasons or an indirect regulation might explain why Irx3 did not repress the Six3.2-cI reporter activity (Fig. 3B), despite the described crossrepression between Six3 and Irx3 (43). In support of an indirect regulation, Irx1, which seems to act redundantly with Irx3 (44), was identified in the TRS but failed to regulate the Six3.2-full construct in the dose-response Luc assays (supplemental material). Instead, Msx2 significantly reduced reporter expression but only when the A CRE was present (Fig. 2E), in good agreement with the observation that the A element silences Six3.2 expression in the hindbrain and neural tube (23).
In Vivo Expression of Selected Candidates-To further narrow down the number of the predicted candidates, we searched different expression databases and followed only those genes with expression overlapping with or complementary to that of Six3.2. The expression pattern of these remaining genes, classified as "functionally related conserved TFs" (Fig. 1B), was further compared with that of Six3.2, using in situ hybridization at early (Fig. 4) and late (Fig. 5) developmental stages. Notably, Tcf3, a Wnt signaling effector, Etv5a, and to a lower extent Etv4, two Fgf signaling effectors, were distributed as Six3.2 in the prosencephalon with an anterior high to posterior low gradient (Figs. 4 (A-D) and 5 (A-D)). Pax6 expression overlapped with  NOVEMBER 6, 2015 • VOLUME 290 • NUMBER 45 that of Six3.2 in the retina, diencephalon, and developing lens but not in the telencephalon, as reported previously (21,45), whereas Nkx2.2 coincided with Six3.2 in the hypothalamus (Fig. 4, E and F). By contrast, the transcripts of the TFs Msx2, Tgf3b, and Pbx1 localized to the caudal neural tube or caudal optic vesicles, thus with a pattern complementary to that of Six3.2 at early stages (Fig. 4, A and G-I). Later, Six3.2 expression still overlapped with that of Pax6 and Pbx1 in the diencephalon, lens, and amacrine and ganglion cells of the retina and with that of Nkx2.2 in the hypothalamus and in the retinal ganglion cell layer (Fig. 5, E-G). The TFs Prdm1/Blimp1 and Vsx1 are expressed in retinal progenitors and then in photoreceptors and bipolar cells, respectively (Fig. 5, J and K), thus being likely candidates for Six3.2 trans-regulation during retinal neurogenesis.

trans-Regulatory Factors for Six3 Forebrain Expression
region, thus identifying the relevant BSs for the already reported Pax6-mediated regulation of Six3 expression (20,46,47). We also found Pbx1 specifically enriched on the Six3.2 regulatory region but in this case on the A and IL enhancers (Fig. 6B), as predicted by the distribution of its BSs (Fig. 2F). The binding of Pbx1 to these CREs supports its potential duality as an early repressor of Six3.2 expression mediated by the A box and as a late activator by binding to the IL module. This is well in line with the initial complementary distribution of Pbx1 and Six3.2 mRNA at early stages of development (Fig. 3I) and with their subsequent overlap in the retina and forebrain (Fig. 5, G and GЈ). Precipitation with anti-HA antibodies confirmed the binding of Msx2 to the Six3.2 A CRE, as expected by the presence of in silico predicted Msx BS within this element. In contrast, we failed to immunoprecipitate any of the Six3.2 regulatory elements using an HA-tagged version of Tcf3, Tcf3VP16, or Lef, transfected either alone or in combination with a constitutively active version of ␤-catenin (data not shown). Although different conserved clusters of Tcf/Lef BS were predicted within the H box, none of them presented the reported high affinity Tcf binding sequence, in which the G of the core T(C/A)AAG motif appears to be required for strong binding (48 -50). This suggests that Tcf3 might regulate the Six3.2 promoter either indirectly or through low affinity binding to the H CRE, making it difficult to confirm this interaction by ChIP.
We next asked whether overexpression or interference with the endogenous expression of these four genes indeed impaired Six3.2 expression, thereby interfering with forebrain development, as reported (21). In agreement with the activation seen in Luc assays, Pax6 mRNA overexpression (100 ng/l) increased the levels of Six3.2 mRNA and of the Six3.2cI::EGFP reporter in a large proportion of the injected embryos (85%; n ϭ 250; Fig.  7A). When fused to heterologous TFs, the engrailed (Eng) repression domain confers strong transcriptional repression, generating dominant negative (DN) forms of transcriptional activators (51,52). We thus fused this Eng domain to the N terminus of the Pax6 CDS to generate its DN version. As expected, overexpression of Pax6-Eng (100 ng/l) reduced the expression of Six3.2 and that of the Six3.2cI::EGFP reporter in a large proportion of embryos, as compared with control HA-Eng (100 ng/l)-injected embryos (77%; n ϭ 200; Fig. 7A), which was identical to Six3.2 expression observed in WT embryos (Fig. 4A). The decrease in Six3.2 expression was particularly evident in the telencephalon and optic vesicles, the latter being also reduced in size.
In contrast to what was observed with Pax6 mRNA, Pbx1 overexpression (50 ng/l) resulted in down-regulation of Six3.2 in the optic vesicle, thereby reducing their size, whereas Six3.2 telencephalic expression appeared expanded, a phenotype that was observed in 67% of the injected embryos (n ϭ 50; Fig. 7B). Notably, the Pbx1-Eng version (50 ng/l) did not induce an opposite phenotype but rather showed a somewhat similar phenotype in a comparable proportion of the injected embryos (72%; n ϭ 60; Fig. 7B). According to ChIP-quantitative PCR, Pbx1 was enriched on the A CRE (Fig. 6B), possibly explaining the reduction of Six3.2 expression in the optic vesicles. In contrast, the increased but spatially different transcriptional output of the Six3.2 gene/reporter after Pbx1 and Pbx1-Eng overexpression can be better explained by an indirect effect, such as Pbx1-mediated repression of other negative regulators.
Data from different species have demonstrated that Tcf3 antagonizes canonical Wnt signaling activity, thereby enabling forebrain specification in a pathway related to that of Six3 (11,13,(53)(54)(55). Indeed, the forebrain phenotypes caused by Tcf3 or Six3 disruption are similar (13,54,55), and Six3 rescues prosencephalic development in Tcf3 Ϫ/Ϫ /headless zebrafish mutants (11,13). According to these observations, overexpression of Tcf3 mRNA (30 ng/l) in Six3.2cI::EGFP transgenic medaka embryos (23) enlarged the forebrain domain and induced a 3-fold increase associated with a posterior expansion of both EGFP reporter expression and the endogenous Six3.2 mRNA distribution in the large majority of the embryos, when compared with RFP-injected controls (95%, n ϭ 30; Fig. 7C), which, in turn, displayed a pattern identical to that previously reported for WT embryos at similar stages (23). Consistent with the idea that Tcf3 is an activator of Six3.2, its DN form (30 ng/l) reduced the extension of both reporter and endogenous Six3.2 expression (85%, n ϭ 40; Fig. 7C), suggesting that at least part of Tcf3 function in the medaka forebrain is mediated by the direct activation of Six3.2. However, in agreement with previous studies (54,55), we also observed an overall increase of the forebrain size in the Tcf3-Eng-injected embryos compared with controls (Fig. 7C), probably due to Wnt pathway inhibition with a mechanism independent of Six3.2.
Overexpression of Msx2 mRNA (100 ng/l) caused a reduction in Six3.2 and Six3.2-cI::EGFP expression in a large proportion of the injected embryos, compared with control injected embryos (74%, n ϭ 30; Fig. 7D) that presented a pattern identi-  NOVEMBER 6, 2015 • VOLUME 290 • NUMBER 45 cal to those of WT embryos at the same stage (21). This finding is in consonance with the repressive effect observed in Luc assays and with the binding of Msx2 to the Six3.2 A CRE, characterized as a repressor element. Msx genes are downstream effectors of BMP signaling, which has been implicated in the repression of anterior neuroectodermal genes (26,56). In agreement with this idea, overexpression of a chimeric active version of Msx2, generated by fusing the VP16 activation domain (57) (Msx2-VP16, 100 ng/l), resulted in an increased expression of Six3.2 mRNA and of the reporter in the Six3.2cI::EGFP line, albeit only in a fraction of the injected embryos (24%; n ϭ 30; Fig. 7D). These findings provide a mechanism by which Msx TFs can limit the size of the neural plate by acting as redundant downstream effectors of BMP signaling during the establishment of the neural/non-neural ectodermal border (58,59).

trans-Regulatory Factors for Six3 Forebrain Expression
Altogether, the results of our study provide converging evidence indicating that Pax6 and Tcf3 are direct activators of Six3.2 expression, whereas Msx2 limits it at both the neural/ non neural border and the posterior neural tube.

Discussion
A hierarchical GRN organization progressively specifies the different structures of the prosencephalon. Central to GRN function are "hub" genes, which critically control the expres-

trans-Regulatory Factors for Six3 Forebrain Expression
sion of many target genes but, in contrast, are regulated by few trans-acting factors (1). Intensive research on forebrain patterning led to the outline of a partial GRN operating from neural induction to forebrain patterning (2). This fragmented network contains most of the central actors, but many of the regulatory relationships are still ill defined, hampering our understanding of the molecular mechanisms behind forebrain specification. A case in point is the Six3 gene. Functional studies had proven that Six3 is, from the very beginning, essential for forebrain patterning (4,13,15,18,20,21,42,60), bestowing on Six3 a hub position within the forebrain GRN (2). Despite this central position, only a few of its direct downstream targets have been identified, and the list of its direct regulators was even shorter. Our study starts filling the latter gap and identifies Msx2 and Pbx1 as bona fide regulators of early Six3.2 expression in distinct domains of the medaka forebrain (Fig. 8). It also confirms the previously proposed direct regulation exerted by Pax6 on Six3 and points to Etv5, Nkx2.2, Prdm1/Blimp1, Vsx1, and especially Tcf3 as additional regulators of different spatialtemporal domains of Six3.2 expression.
This information is the result of a previously used TRS (27,28) that overcame the limitations of and complemented the in silico prediction approaches based on the identification of TFBS. As described (27,28), we took advantage of a preselected subset of the library enriched in developmentally expressed genes, which improved the effort/cost as well as the efficiency of the original screening (27). In fact, there was a 5-fold increase (5.4% versus 1.1%) in the proportion of positive versus screened clones when we compared our data with those obtained in the Ath5 TRS (27). As in the latter study, an important proportion of the candidates activating the Six3.2 regulatory region (23) were classified as TFs, chromatin modifiers, or components of signaling pathways, and almost half of them showed a doseresponse behavior in Luc assays. In our study, the identification of trans-regulatory factors was further facilitated by the detailed functional characterization in enhancers, silencers, or silencer modulators of the CREs present in the Six3.2 locus (23). The knowledge that the A element acts as a silencer seems to account, at least in part, for the differences observed, for example, when Pax6 and Tcf3 (this study) or Sox2 (21) were tested on constructs containing different CREs; the presence of the A element counteracted the otherwise activator function of the Six3.2 regulatory region. Therefore, our approach based on a dual screening and the analysis of transcriptional regulators over different combinations of well characterized CREs seems to provide an efficient method to identify the trans-regulatory factors of a given gene.
In support of the efficiency of this approach, the majority of the candidates selected after this combined analysis showed expression patterns either overlapping with or complementary to that of Six3.2, which we interpreted as consistent with an activator or repressor role, respectively (Fig. 8, A and B). For Tcf3, Pax6, Pbx1, and Msx2, these predictions were validated by gain and loss of function assays, which, in the case of Pax6, are also supported by earlier studies (20,46,47,61). The direct binding of these TFs to the predicted CREs was further validated with ChIP experiments. Notwithstanding, a few candidates obtained with the TRS and confirmed as putative activators in dose-response Luc assays, as Hoxb1 and Prdm1, displayed an expression pattern not overlapping with that of Six3.2, hinting instead at a repressor function. This discrepancy may be explained by a differential expression of specific cofactors between the Six3-positive cells of the forebrain and the cell line used in the TRS.
Regulators of Early Six3.2 Expression-In a previous study, we have shown that differential binding of Sox2 to the Six3.2 regulatory region activates its graded expression along the anterior forebrain (21). There is also evidence that the BMP effector Lmo4 restrains Six3 expression at the neural/nonneural border (26) and that Msx2, Prox1, Pax6, Six3 itself,
Tcf3 inactivation disrupts prosencephalic development, in part because Tcf3 probably antagonizes the posteriorizing activity of Wnt signaling (53). In the zebrafish headless/Tcf3 Ϫ/Ϫ mutant, this phenotype is rescued by Six3 mRNA injection (13). Complementing this notion, we show that overexpression of the Tcf3ca form up-regulates endogenous Six3.2 expression, leading to an enlarged anterior neural plate, whereas a Tcf-DN has the opposite effect. Luc assays indicated that this effect could be direct and mediated by the H CRE, in which conserved Tcf BSs were identified. However, we could not confirm with ChIP assays this direct binding, probably because the clusters of BSs present in the H element do not correspond to high affinity sites (48 -50), probably generating unstable interactions. Similarly, our TRS identified the Fgf signaling components Fgf3, Etv5, and the Fgf signaling effector Etv4, as upstream regulators of Six3.2, suggesting that the well known Fgf-mediated telencephalic patterning (66 -68) may involve Six3 activation. Notably, the anterior forebrain determinant Foxg1 appears to act downstream of Fgf signaling (68), whereas high Six3.2 expres-sion levels specify the telencephalic field, in part through direct Foxg1 activation (21). Therefore, telencephalic development may require the Fgf/Six3.2/Foxg1 hierarchical organization (Fig. 8, A and C).
Reciprocal cross-regulation between Pax6 and Six3 is important for early eye specification (7,20,47,69,70). Our study shows that Pax6 directly binds the D element of the Six3.2 regulatory region, which accounts for its early forebrain expression. Notably, although Pax6 and Sox2 have been reported to synergize in the activation of other regulatory regions (i.e. during lens development (40,71)) they showed additive effects on the transactivation of the Six3.2 D CRE only at low doses. Considering that in medaka fish, Sox2 is predominantly expressed in the telencephalon and Pax6 in the eye, this effect on the D element probably reflects a spatially subdivided regulation of Six3.2 expression levels, which are critical to specify the relative size of these two domains (21). A similar mechanism could activate Six3.2 expression also in the lens ectoderm, in which a differential inter-regulation between Sox2 and Pax6 has been proposed (72)

trans-Regulatory Factors for Six3 Forebrain Expression
As mentioned above, BMP signaling limits the expression of Six3 at the neural/non-neural border of gastrulating vertebrate embryos (26). Accordingly, its downstream effectors Msx1 and Msx2 are expressed in the non-neural ectoderm of different species, including the zebrafish (58,59). Our molecular and functional studies support that the Msx2a gene directly represses Six3.2 expression in gastrulating embryos. Whether Msx2a or the related Msx2b is actually responsible for this function in vivo is unclear, because the Msx family is particularly diversified in medaka (six members versus the two reported in mammals and birds); thus, other family members may be the relevant regulators, probably acting redundantly, as reported for the regulation of other target genes. Nevertheless, Msx2a, localized to the mesencephalon, the hindbrain, and the posterior optic vesicles, is possibly repressing Six3.2 activity in these domains. The direct binding of Pbx1, which, like Msx2, is distributed in the hindbrain and the caudal neural tube, on the Six3.2cI A CRE suggests that this TF also represses Six3.2 in the caudal neural tube. This Boxes represent enhancers (green) and silencers and silencer blockers (red). Continuous thick arrowed lines represent positive (green) and negative (red) interactions among CREs. The positive (green) or negative (red) activity of Six3.2 trans-regulators is represented with thin continuous arrowed lines. Indirect or not fully characterized regulatory relationships are represented with discontinuous lines. C, schematic representation of the GRN controlling early forebrain development. Green arrows represent activation, and red lines indicate repression. A continuous line represents known direct regulations, whereas indirect or unproven direct relationships are indicated with dashed lines. The new regulatory inputs identified in this work are marked with a yellow asterisk. Dark green lines represent direct activation described in medaka, light green arrows represent evidence obtained in zebrafish, and the palest green arrows represent activating relationships proven in other vertebrate species. Similarly, dark red, orange, and pale red arrows represent inhibitory relationships proven in medaka, zebrafish, or other vertebrate species, respectively. The references reporting these regulatory relationships are indicated with blue numbers according to the References. NOVEMBER 6, 2015 • VOLUME 290 • NUMBER 45 idea is supported by the caudal extension of reporter expression observed in Six3.2cII::EGFP transgenic embryos, in which the A element is missing (23). However, the complex interactions of Pbx TF with other TALE family members and homeobox TFs (36,(73)(74)(75) hampers the identification of the precise transcriptional complex acting in Six3.2 regulation.

trans-Regulatory Factors for Six3 Forebrain Expression
Regulators of Late Six3.2 Expression-The role of Six3 in forebrain differentiation is still poorly addressed, and most studies have focused on its role during retinal neurogenesis, highlighting its requirement for progenitor proliferation and amacrine and ganglion cell specification in cooperation with Pax6, NeuroD, Math3, Tcf3, and Pbx1 (28, 51, 57, 76 -81), among others. The high transactivation activity, the overlapping expression domains, the presence of conserved putative BSs, and the direct binding of Pbx1 to the IL CRE, responsible for late Six3.2 expression, support the possibility that during neurogenesis, Pbx1 directly promotes Six3.2 retinal expression (Fig. 6B). This function is probably aided by Tcf3, which also transactivates the IL region. It is likely that the Pbx1/Tcf3/ Six3.2 network controls retinal progenitor proliferation because each one of these genes has been separately shown to influence the rate of retinal cell division (80). In this view, Pbx proteins would have an effect on retinal Six3.2 regulation opposite to that we propose at early stages. However, these two roles are not incompatible, given the aforementioned complexity of TALE protein interactions. It is less clear whether Vsx1 and Prdm1 are functionally related to Six3.2 activity in the retina, despite their significant transactivation of the Six3.2cI construct. The two genes are involved in bipolar versus photoreceptor cell specification (82,83), and their expression is complementary to that of Six3.2 during neurogenesis. It is thus possible that their identification in the TRS might reflect the presence of sequence similarities among the regulatory elements of Six3.2 and those of the Six3.1 paralog and the Six6 ortholog, which are instead expressed in the inner nuclear layer. 4 In support of this possibility, Six6 has been implicated in the development of photoreceptor precursors (84).
In conclusion, our study describes a time/cost-effective approach to identify trans-acting factors that can be applied to the study of many gene regulatory regions. More relevant, our study shows that Msx2, Pbx1, and Pax6 and probably Tcf3, Etv5, Nkx2.2, Prdm1/Blimp1, and Vsx1 are important regulators of Six3.2 expression during anterior forebrain development. Mutation of each one of the identified BSs would be a necessary step to further evaluate the relative contribution of each single BS to the spatio-temporal regulated expression of the Six3.2 gene. Nevertheless, altogether, our data add new elements to the complex GRNs in charge of early forebrain specification, providing new knowledge on how forebrain neuronal diversity is originated.