|Home | About | Journals | Submit | Contact Us | Français|
Neural basic helix-loop-helix (bHLH) transcription factors are crucial in regulating the differentiation and neuronal subtype specification of neurons. Precisely how these transcription factors direct such processes is largely unknown due to the lack of bona fide targets in vivo. Genetic evidence suggests that bHLH factors have shared targets in their common differentiation role, but unique targets with respect to their distinct roles in neuronal subtype specification. However, whether neuronal subtype specific targets exist remains an unsolved question. To address this question, we focused on Atoh1 (Math1), a bHLH transcription factor that specifies distinct neuronal subtypes of the proprioceptive pathway in mammals including the dorsal interneuron 1 (dI1) population of the developing spinal cord. We identified transcripts unique to the Atoh1-derived lineage using microarray analyses of specific bHLH-sorted populations from mouse. Chromatin immunoprecipitation-sequencing (ChIP-seq) experiments followed by enhancer reporter analyses identified five direct neuronal subtype specific targets of Atoh1 in vivo along with their Atoh1-responsive enhancers. These targets, Klf7, Rab15, Rassf4, Selm, and Smad7, have diverse functions that range from transcription factors to regulators of endocytosis and signaling pathways. Only Rab15 and Selm are expressed across several different Atoh1-specified neuronal subtypes including external granule cells (EGL) in the developing cerebellum, hair cells of the inner ear, and Merkel cells. Our work establishes on a molecular level that neuronal differentiation bHLH transcription factors have distinct lineage-specific targets.
First discovered in Drosophila, neural specific basic-helix-loop-helix (bHLH) transcription factors are crucial for determining proper neural cell fates (Jan and Jan, 1994). In vertebrates, bHLH transcription factors are essential for the general neuronal differentiation as well as neuronal subtype specification of diverse cell types in the peripheral and central nervous systems (Bertrand et al., 2002). They are thought to share activity in inducing neuronal differentiation, but have distinct functions in specifying neuronal subtypes (Parras et al., 2002; Nakada et al., 2004a). While several studies have found targets of bHLH transcription factors, they have mostly focused on their common role in neurogenesis (Bertrand et al., 2002; Castro et al., 2006; Seo et al., 2007).
Elegant genetic studies in Drosophila and mouse suggest that in addition to shared downstream transcriptional targets, bHLH transcription factors have unique targets relevant for the function or development of that specific neuronal subtype. Studies misexpressing scute or ato (Chien et al., 1996; Jarman and Ahmed, 1998), or substituting Neurog2 with Ascl1 (Parras et al., 2002) respecified neurons in a context-dependent manner. Similarly, overexpression of Ascl1 and Atoh1 in the chick spinal cord induces progenitors to differentiate into specific neuronal subtypes (Gowan et al., 2001; Nakada et al., 2004a).
We focused our study on Atoh1 (mammalian atonal (ato) homolog 1), a bHLH transcription factor required for the formation of different proprioceptive neuronal subtypes (Bermingham et al., 2001). Due to its discrete expression in defining progenitors to the dorsal interneuron 1 (dI1) population of the developing spinal cord (Bermingham et al., 2001; Gowan et al., 2001), Atoh1 was an ideal bHLH to identify neuronal subtype specific targets. In addition to dI1 neurons, Atoh1 specifies progenitors to the granule layer of the cerebellum (Ben-Arie et al., 1997), several hindbrain neurons (Ben-Arie et al., 1997; Machold and Fishell, 2005; Wang et al., 2005; Maricich et al., 2009b; Rose et al., 2009a; Rose et al., 2009b), sensory hair cells of the inner ear (Bermingham et al., 1999; Zheng and Gao, 2000; Izumikawa et al., 2005; Raft et al., 2007), and Merkel cells in the skin and vibrissae (Ben-Arie et al., 2000; Maricich et al., 2009a; Morrison et al., 2009; Van Keymeulen et al., 2009). However, fundamental mechanistic understanding of how Atoh1 directs specification of these neuronal subtypes is lacking in the spinal cord since the only known direct Atoh1 targets in vivo besides Atoh1 itself (Helms et al., 2000) are transcription factor, Barhl2 in dI1 neurons (Saba et al., 2005). In contrast, in the developing cerebellum a variety of direct Atoh1 targets were recently identified (Klisch et al., 2011) adding to the previously known targets, Barhl1 and Gli2 (Kawauchi and Saito, 2008; Flora et al., 2009).
In this study, we identified unique targets of Atoh1 by comparing sorted Atoh1-lineage cells in the developing dorsal neural tube with a neighboring population defined by the expression of the bHLH factor Neurog1 (Ngn1, neurogenin1). We identified transcripts enriched in Atoh1-lineage cells and biased against identifying common bHLH targets. Using ChIP-seq data from a FLAG-tagged Atoh1 knock-in mouse, we identified five new direct lineage-specific in vivo targets of Atoh1 whose enhancers respond to Atoh1 expression: Klf7, Rassf4, Rab15, Selm, and Smad7.
Atoh1BAC-GFP (Math1GFP-BAC) (Raft et al., 2007) and dNeurog1-GFP (TgN1-13G) (Nakada et al., 2004b) transgenic mice were generated previously. Atoh1BAC-GFP transgenic mice contain the bacterial artificial chromosome RPCI-23318G16 with the Atoh1 coding sequence replaced with nuclear localized GFP. dNeurog1-GFP transgenic mice contain an enhancer from the Neurog1 gene that directs GFP reporter expression primarily to the dP2/dI2 domains in the dorsal neural tube. Transgenic embryos were identified by GFP fluorescence. Tail clips and yolks sacs of transgenic mice were PCR genotyped for GFP.
Transgenic mice of Klf7 site A and Rassf4 site A enhancers cloned into the BgnEGFP vector (Lumpkin et al., 2003) were generated using standard procedures (Brinster et al., 1985). Each transgene was isolated from recombinant plasmid on a standard agarose gel and microinjected at 1–3 ng/μL into pronuclei of fertilized eggs from B6SJLF1 (C57BL/6JxSJL) crosses by the Transgenic Core Facility of University of Texas Southwestern Medical Center at Dallas. All animal experiments were approved by the Institutional Animal Care and Use Committee at UT Southwestern.
Atoh1BAC-GFP and dNeurog1-GFP transgenic mouse E10.5 neural tubes from forelimb to hindlimb were dissected in DMEM/F12 (Gibco) and dissociated with 0.25% Trypsin-EDTA (Gibco). Fluorescence activated cell sorting (FACS) was performed by the UT Southwestern Flow Cytometry Core Facility. Cells were sorted into GFP − and + cells with a MoFlo (Dako/Beckman Coulter, 100 μm nozzle) or a FACSAria (BD Biosciences) cell sorter directly into ZR RNA buffer (Zymo Research). RNA was extracted using the Mini RNA Isolation II Kit (Zymo Research).
cDNA for RT-PCR was prepared from TURBO DNase (Ambion)-treated GFP+ and − cell sorted RNA from E10.5 neural tubes using Omniscript or Sensiscript reverse transcription kits (Qiagen) with oligo dT primer. Quality of GFP sorted cells was verified by RT-PCR (Qiagen Taq) using primers to Atoh1 (5′-GCT GGT AAG GAG AAG CGG CTG TG-3′ and 5′-TGT ACC CCA TTC ACC TGT TTG C-3′) or Neurog1 (5′-CCA CTG TGG CAT CAC CAC TC-3′ and 5′-GCG TCG TGT GGA GCA GGT CTT TG-3′), GFP (5′-CAG AAG AAC GGC ATC AAG GTG AAC-3′ and 5′-GGG TGC TCA GGT AGT GGT TG-3′), and GAPDH (5′-ACC ACA GTC CAT GCC ATC AC-3′ and 5′-CAG CTC TGG GAT GAC CTT GC-3′) mRNA and visualized on a 2% agarose gel with ethidium bromide.
cDNA for quantitative PCR (qPCR) was prepared from TURBO DNase (Ambion)-treated pooled GFP+ and GFP− cell sorted RNA from E10.5 mouse neural tubes using the Superscript III First Strand Synthesis System (Invitrogen) with random hexamers. qPCR was carried out with Fast SYBR Green Master Mix (ABI) on the 7500 Fast Real-Time PCR System (ABI) using the following primers: Atoh1 (5′-GTA AGG AGA AGC GGC TGT G-3′ and 5′-AGC CAA GCT CGT CCA CTA-3′), Rab15 (5′-GGC TTG GGC TGT GTC ATT G-3′ and 5′-GGC AGA CAG GCC AGG AAA-3′), Selm (5′-TCG TGC TGT TAA GCC GAA ATT-3′ and 5′-CCG GGT CAT TTG GCT GAG T-3′), Smad7 (5′-CGA AGA CAG GAA ACG AGA GTC A-3′ and 5′-GGT GGT GCC CAC TTT CAC A-3′), Neurog1 (5′-CTG CGC TTC GCC TAC AAC TAC-3′ and 5′-ATC TGC CAG GCG CAG TGT-3′), and Ppib (Peptidylprolyl isomerase B, formerly known as Cyclophilin B) (5′-GGA GAT GGC ACA GGA GGA A-3′ and 5′-GCC CGT AGT GCT TCA GCT T-3′). The ΔΔCt method was used to determine the mRNA fold change of transcripts in the Atoh1BAC-GFP+ to dNeurog1-GFP+ cells (Figure 1D) and the Atoh1BAC-GFP+ to GFP− cells (Figure 2B) using Ppib as the endogenous control gene. Graphs were made using Prism 5 software.
Integrity of the isolated total RNA was analyzed using a 2100 Bioanalyzer (Agilent). One microgram of RNA pooled from Atoh1BAC-GFP+ or dNeurog1-GFP+ sorted cells from multiple embryos from multiple litters was processed for Affymetrix Mouse 430 2.0 microarrays by the UT Southwestern Microarray Core Facility using standard protocols. Two biological samples per transgenic mouse type were processed for microarrays, preprocessed with MAS 5.0 to generate normalized signal intensity data, and analyzed using GeneSpring Agilent GX 7.3 Expression Analysis software. Signals <0.01 were set to 0.01 and each chip was normalized to the 50th percentile and analyzed for probes with over two-fold change in signal where the Affymetrix probes had to be present or marginal in at least one of the Atoh1BAC-GFP or dNeurog1-GFP samples. Individual arrays from Atoh1BAC-GFP+ cells versus the dNeurog1-GFP+ cells were compared. Fold differences in transcript levels are presented separately since reliable statistical analyses cannot be performed on only two arrays. The raw microarray data has been deposited in NCBI’s Gene Expression Omnibus database (Edgar et al., 2002) (GEO Series accession number GSE23089) (http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE23089). The processed data will be provided upon request or can be found at http://www8.utsouthwestern.edu/utsw/cda/dept120915/files/150735.html.
Whole mouse male and female embryos from wildtype or Atoh1 knockout mice (Ben-Arie et al., 1997) were dissected at E10.5 where E0 was timed as halfway through the dark cycle of the morning that the vaginal plug was detected. Mouse and chick embryos were fixed in 4% paraformaldehyde for 1.5–2 hours, washed, cryoprotected, embedded, and cryosectioned (transverse, 20–30 μm). Heads of E16.5 embryos were fixed overnight. The noses were cut and embedded for coronal sections while the rest of the head was cut at a 45 degree angle between coronal and transverse sections to obtain the appropriate visualization for the developing inner ear. P0 pups were perfused with 2 mL PBS, brains dissected out, fixed overnight, and embedded for sagittal sections.
Mouse or chick sections were incubated in primary antibody and the appropriate secondary fluorophore conjugated antibodies (Alexa fluorophores 488, 568, 594, 647, Molecular Probes/Invitrogen). Primary antibodies used were: 1:100 rabbit anti-Atoh1 (Helms and Johnson, 1998), 1:4000 rabbit anti-Lhx2/9 (Liem et al., 1997), 1:500 rabbit anti-Neurog1 (Gowan et al., 2001), 1:500 chick anti-GFP (Aves), 1:500 mouse anti-BOSS (Kramer et al., 1991), 1:1000 mouse anti-c-myc (Santa Cruz Biotechnology), 1:100 mouse anti-Islet1/2 (40.2D6)(DSHB and Ericson et al., 1992), 1:100 mouse-anti Lhx1/5 (4F2)(DSHB and Tsuchida et al., 1994), and 1:200 goat anti-Neurod (Santa Cruz Biotechnology). Sections were imaged using a Zeiss LSM510 confocal microscope. For Figures 5G, J, and K, the endogenous GFP fluorescence gain and offset was kept constant and 16-bit tiff images taken on the same day within one set of chick electroporation experiments of an enhancer-GFP construct plus control, BOSSAtoh1, or mycAscl1. Mean GFP pixel intensity/cell of colocalized GFP and BOSS- or myc-tagged bHLH was processed using ImageJ (Abramoff, 2004). The threshold for colocalization was optimized for each image. In the cases where the GFP fluorescence was below detectable levels (Figure 5D, D″, F), the GFP fluorescence of the entire image was calculated. Insets in Figure 5 were manipulated with Photoshop to show that GFP fluorescence was detectable in some cases and that the neural tube was adequately injected. S.E.M. are given and p-values were determined in Excel using a two-tailed two-sample unequal variance (heteroscedastic) t-test.
In situ probes were cloned from E10.5 neural tubes or P0 cerebella cDNA. Enhancers were cloned from whole genomic DNA of ICR wildtype mice into the BgnEGFP vector (Lumpkin et al., 2003). Primer sets, cloning sites, and parent plasmid vector used for each probe or enhancer construct will be provided upon request or at http://www8.utsouthwestern.edu/utsw/cda/dept120915/files/150735.html. PCR products were obtained using standard protocols from the iProof kit (Bio-Rad), KOD Xtreme HotStart DNA Polymerase (Novagen) kit, or Expand High Fidelity (Roche) kit.
ISH was performed as per standard protocols. A detailed protocol is available upon request. Digoxygenin (DIG)-labeled antisense RNA probes (1–5 mg/mL) were hybridized overnight at 65°C, incubated with anti-digoxygenin AP antibody (Roche), and then incubated with NBT/BCIP (Roche). The sections were immediately imaged or counterstained with eosin before imaging with a Zeiss Discovery Stereomicroscope V12.
Chromatin immunoprecipitation of p300, a transcriptional co-activator, followed by high throughput sequencing (ChIP-seq) of E11.5 neural tubes from the posterior hindbrain to the hindlimb was performed as previously described for E11.5 forebrain, midbrain, and limb (Visel et al., 2009). The Atoh1-FLAG ChIP-seq from P5 cerebella was performed as described (Flora et al., 2009). The recovered DNA was submitted for Solexa sequencing at the Center for Cancer Epigenetics Solexa Sequencing Core at MD Anderson (Houston, TX), where the sequencing libraries were prepared and sequenced on the Illumina Solexa GAII pipeline according their standard operating procedures (Klisch et al., 2011). ChIP regions along with 30 vertebrate species conservation are viewed in UCSC browser, NCBI37/mm9 mouse build (Kent et al., 2002).
The following enhancer sequences of Atoh1 regulated genes in the dorsal neural tube were analyzed with MEME (Bailey and Elkan, 1994). Atoh1 enhancer A and B (Chr6: 64683369–64684709), Barhl1 enhancer, Barhl2 enhancer, Klf7 subset of site A, Rassf4 subset of site A (Chr6: 116620232–116621468), Selm site B, Smad7 subset of site A (Chr18: 75594185–75594975), and Rab15 site A (see Table 2 for coordinates). MEME was asked to find ten, eight base pair motifs that occurred in every sequence (one per sequence, min and max width = 8). As a control, sequences 2000 base pairs upstream of the enhancers were subject to the same analysis.
Fertilized White Leghorn chick eggs were obtained from the Texas A&M Poultry Department. The neural tube of chick embryos staged HH14-17 (Hamburger and Hamilton, 1992) were injected with 1 μg/μL of the enhancer-GFP plasmid construct and 1 μg/μL of the appropriate bHLH transcription factor plasmid and electroporated as previously described (Nakada et al., 2004a). Embryos were harvested after 36–48 hours at 38°C. bHLH plasmids used for this study were made in the pMiWIII expression vector: myc-tagged rat Ascl1-AQ (Mash1 NR-AQ, myccontrol) inactive mutant control (Nakada et al., 2004a), BOSS-tagged mouse Atoh1 (BOSS-Math1, BOSSAtoh1) (Helms et al., 2000), and myc-tagged rat Ascl1 where five myc tags were inserted at the N-terminus of Ascl1 (Mash1, mycAscl1) (Nakada et al., 2004a).
To determine downstream targets of Atoh1 unique to the Atoh1 lineage, we identified transcripts enriched specifically in the dorsal progenitor and interneuron 1 (dP1/dI1) populations located adjacent to the roof plate in the developing neural tube. The dP1 domain begins expressing Atoh1 and differentiates into the dorsal interneuron 1 (dI1) population marked by LIM-HD transcription factors, Lhx2 and Lhx9 (Bermingham et al., 2001; Gowan et al., 2001). Similarly, the neighboring progenitor population (dP2) is marked by Neurog1 and differentiates into the dorsal interneuron 2 (dI2) population as marked by Lhx1 and Lhx5. To identify transcripts present in the Atoh1-derived (dP1/dI1) domains that are distinct from the Neurog1-derived (dP2/dI2) domains, we compared transcripts in these two related, but discrete, cell populations.
Two transgenic mouse lines, Atoh1BAC-GFP (Raft et al., 2007) and dNeurog1-GFP (Nakada et al., 2004b), drive GFP either to the dP1/dI1 domains or the dP2/dI2 domains, respectively. Immunostaining with an antibody to each neighboring population, Neurog1 antibody for Atoh1BAC-GFP and Atoh1 antibody for dNeurog1-GFP in E10.5 neural tubes, demonstrates the restriction of GFP to dP1/dI1 or dP2/dI2, respectively (Figure 1A). GFP + and GFP− cells from E10.5 neural tubes from Atoh1BAC-GFP and dNeurog1-GFP mice were separated by fluorescence activated cell sorting (FACS). RT-PCR of RNA extracted from these populations showed good separation between GFP+ and − cells (Figure 1B). In Atoh1BAC-GFP sorted cells, Atoh1 and GFP transcripts are enriched in GFP+ cells while GAPDH was present in both populations. Likewise, RT-PCR of RNA sorted from the dNeurog1-GFP population showed enrichment in transcripts of Neurog1 and GFP. Two microarrays were performed from RNA of GFP+ cells from the Atoh1BAC-GFP and dNeurog1-GFP sorts to determine transcripts enriched specifically in the Atoh1-derived population rather than general neuronal expressed genes at this stage (GEO Series accession number GSE23089).
The intersection of two independent microarray experiments comparing Atoh1BAC-GFP and dNeurog1-GFP sorted cells found 520 Affymetrix probes were over two fold enriched in the Atoh1 population, corresponding to 443 genes (Figure 1C). Genes known to be enriched in the dI1 population, Atoh1 (Helms et al., 2000), Lhx2/9 (Bermingham et al., 2001; Gowan et al., 2001), Barhl1 (Bermingham et al., 2001), and Barhl2 (Saba et al., 2005), were over four-fold enriched in the Atoh1 marked population (Table 1). This finding confirms successful isolation of dP1/dI1 cells and illustrates the quality of our microarray analyses. The microarray data were further validated by RT-qPCR of Atoh1 and Neurog1 in the Atoh1BAC-GFP+ cells relative to the dNeurog1-GFP+ cells (Figure 1D). The RT-qPCR confirms we have good enrichment of Atoh1 in the Atoh1BAC-GFP+ cells (ten-fold) and Neurog1 in the dNeurog1-GFP+ cells (thirteen-fold). In situ probes were generated for twenty-one genes that were over two-fold enriched in the Atoh1BAC-GFP+ cells, had fluorescence values of >100 in at least one of the microarray samples, and were of biological interest. Fourteen of these (67%) gave detectable in situ hybridization (ISH) signal in E10.5 neural tubes and/or P0 cerebella. Ten of these fourteen candidate genes gave clear ISH signal in the dorsal most domain at E10.5 or by RT-qPCR (Figures 1D, 2A, B) and were pursued for further analysis (Table 1). A complete list of genes enriched in the Atoh1 population will be provided upon request or at http://www8.utsouthwestern.edu/utsw/cda/dept120915/files/150735.html.
Transcripts enriched in the Atoh1-derived population (Table 1) were tested for their specificity to the Atoh1-lineage by comparing their expression in wildtype versus Atoh1 knockout mice by ISH (Ben-Arie et al., 1997). In Atoh1 mutants, dI1 interneurons are not generated, but rather transfate to cells with either roof plate or dI2 identity (Bermingham et al., 2001; Gowan et al., 2001). ISH probes to Gmpr, Grem2, Gsg1l, Klf7, Ntrk3 (TrkC), Rab15, Rassf4, and Tle4 all gave ISH signal in the dorsal-most domain of E10.5 neural tubes that disappeared in the Atoh1 knockout (Figure 2A), demonstrating that these transcripts are in the dP1/dI1 populations and require Atoh1 for expression. Note that some of these genes, Klf7, Ntrk3, Rassf4, and Tle4 have mRNA expression in other domains of the dorsal neural tube and may be activated by other bHLH factors. ISH probes to Selm and Smad7 gave robust signal in P0 cerebella (Figure 2C), but did not give detectable signal at E10.5 (data not shown) even though they had clear signals in the microarray experiments (see GEO Series accession number GSE23089). By qPCR, Selm and Smad7 transcripts are enriched in Atoh1BAC-GFP+ sorted cells compared to GFP− cells five to six-fold (Figure 2B). In comparison, Atoh1 and Rab15 transcripts (which are clearly detectable by ISH) had 172-fold and 42-fold changes, respectively (Figure 2A, B). Thus, using microarray analyses, ISH, and RT-qPCR, we identified ten genes enriched in the dP1/dI1 populations that are potential subtype specific targets of Atoh1 (Table 1, Figure 2A, B).
To assess whether these Atoh1 downstream genes are direct Atoh1 targets and to identify Atoh1-responsive dI1-specific enhancers, we used ChIP-seq data obtained from FLAG–tagged Atoh1 knockin mice (Flora et al., 2009; Klisch et al., 2011). Due to the paucity of Atoh1-expressing cells in the E10.5 neural tube, Atoh1-expressing granule precursor cells from postnatal day 5 cerebella were used to identify Atoh1-bound sites in vivo that we further test for function in the neural tube. Thus, this analysis will identify targets that are common to both tissues. Identification of Atoh1-FLAG ChIP-seq bound sites at previously characterized Atoh1 enhancers A and B (Helms et al., 2000) (Figure 4D), Barhl1 enhancer (Kawauchi and Saito, 2008), and Barhl2 enhancer (Saba et al., 2005) (Table 2), confirms these genes as direct targets of Atoh1 in vivo, and demonstrates the robustness of the ChIP-seq experiment.
To identify candidate enhancer regions in the Atoh1-specific lineage targets, we searched for Atoh1-FLAG binding sites located within 200 kb 5′ and 3′ of each gene (Table 1) (Klisch et al., 2011). Given that there are on average eleven Atoh1-FLAG binding sites per gene, we limited our analysis to genes that had an experimentally tractable number of binding sites surrounding the gene of interest (Figures 3 and and4,4, Table 2). These Atoh1 DNA binding regions were tested for their activity during neural tube development by assaying their ability to drive GFP expression in enhancer-reporter constructs introduced into the chick neural tube by electroporation (Timmer et al., 2001). Five of the eleven genomic regions tested, Klf7 site A, Rassf4 site A, Smad7 site A, Selm site B, and Rab15 site A drove appropriate GFP expression to dI1 interneurons (Figure 3B, D, G, K, Figure 4A) as determined by co-expression of GFP with the dI1 lineage markers Lhx2/9.
Atoh1 bound regions that gave enhancer activity had some shared properties. For example, four of the five active enhancers are within introns of their respective genes (Klf7 site A, Rassf4 site A, Selm site B, and Rab15 site A) (Figure 3C, F, I, Figure 4B). The exception is Smad7 site A, which is located approximately 38 kb 3′ of Smad7 in the Gm672 gene (Figure 3H). Gm672 is expressed in the dP1/dI1 Atoh1 population based on the microarray data, but it is not specific to this population. One tested intronic Atoh1-FLAG binding region, Grem2 site A, did not give enhancer activity (Figure 4E, F). Notably, three of the five enhancers, Klf7 site A, Rassf4 site A, and Smad7 site A are also identified as active enhancers since they are also bound by the histone acetyltransferase, p300, as detected in ChIP-seq from E11.5 neural tubes (A. Visel and L. Pennacchio, unpublished observations) (Goodman and Smolik, 2000; Visel et al., 2009). However, the presence of p300 on a site does not ensure efficient dI1 expression as demonstrated by Selm site A (Figure 3I, J). Furthermore, there are many genomic regions where Atoh1 is bound in cerebellar tissue that do not drive significant enhancer activity to the dI1 domain: Rassf4 site B, Selm site A, Selm site C, Selm site D, Atoh1 site C, and Grem2 site A (Figure 3E, J, L, M, Figure 4C, E). Whether these regions can drive expression in the developing cerebellum is not known. However, the inability of Atoh1 site C to direct dP1/dI1 specific expression is consistent with the inability of a 15kb sequence 5′ of the Atoh1 gene, which includes the Atoh1 site C, to direct LacZ reporter expression in transgenic mice (Helms et al., 2000). Taken together, five new dI1 enhancers were identified, four of which are located in introns: Klf7 site A, Rassf4 site A, Selm site B, Smad7 site A, and Rab15 site A.
The identified enhancers, Klf7 site A, Rassf4 site A, Selm site B, Smad7 site A, and Rab15 site A, were tested for their response to Atoh1. Co-electroporation of enhancer-GFP constructs with an epitope-tagged Atoh1 expression vector (Helms et al., 2000) into chick neural tubes gave a marked increase in GFP fluorescence intensity for each of the enhancers tested compared to an inactive bHLH mutant control (Figure 5). GFP fluorescence was quantified for cells that co-expressed both GFP and the bHLH factor (Figure 5G). To test the specificity of this response, we also tested the responsiveness of the enhancer to another neural bHLH factor, Ascl1. An epitope-tagged Ascl1 did not significantly activate any of the enhancers except for Rassf4 site A and Rab15 site A, highlighting the specificity of these enhancers for Atoh1 (Figure 5B″, C″, D″, E″, F″, G). Of the two exceptions, Rassf4 site A may be responsive to Ascl1 in this context since this regulatory element drives GFP expression to some dI3 Ascl1-lineage cells in transgenic mice (Figure 6C), Rassf4 has faint expression in the Ascl1 domain by ISH (Figure 2A), and this site is bound by Ascl1 by ChIP (M.D. Borromeo and J.E. Johnson, unpublished observation). The lack of specificity in the response of the Rab15 enhancer is not known, however, it may be due to a missing negative regulatory element from the sequence used here.
Class II (Massari and Murre, 2000) tissue-specific bHLH factors form heterodimers with E-proteins (E2-2, E2A, and HEB) to bind an E-box defined as CANNTG where N is any base. We looked to see if we could find a common E-box motif in the enhancer sequences. Using the Atoh1-FLAG binding limits of these five enhancers combined with the previously identified, Atoh1 enhancer A and B, Barhl1 enhancer, and Barhl2 enhancer for a total of eight enhancer sequences (Table 2), we searched for common eight base pair motifs using MEME (Bailey and Elkan, 1994). One of the common motifs was an extended E-box (Figure 5H), AMCAGMTG where M is A/C. This is a subset of the extended E-box identified from genome-wide analysis of Atoh1 binding sites in the cerebellum called AtEAM, RMCAKMTGKY, where R is G/A, K is G/T, and Y is C/T (Figure 5I) (Klisch et al., 2011). The same MEME analysis performed on control sequences 2000 bp upstream of each enhancer did not give any recognizable E-box motif, indicating that the motif identified is enriched in Atoh1-responsive enhancers.
The functionality of the common E-box was tested in the context of Klf7 site A and Rassf4 site A. Klf7 site A has two E-boxes meeting the general CANNTG consensus (Figure 3C) designated Emut 1 and E*mut 2 where the * indicates the extended common E-box found in Figure 5H. These sites were mutated and tested for their sensitivity to Atoh1 compared to an inactive bHLH mutant control in the chick enhancer assay. Mutation of either E-box causes a significant decrease in the ability of the enhancer to be induced by Atoh1. However, even with both E-boxes mutated the Klf7 enhancer is still responsive to Atoh1 (Figure 5J), suggesting Atoh1 may also indirectly regulate this enhancer. Also notable is the lack of distinction between Atoh1 responsiveness of the two E-boxes, even though only one of them matches the shared motif. Similarly, the Rassf4 site A enhancer does not require the E-box with the shared motif (Figure 5K), however, this enhancer has a cluster of 11 E-boxes (Figure 3F) that likely contribute to the activation of this enhancer.
Rassf4 site A and Klf7 site A were tested in transgenic mice for their ability to drive GFP expression to the Atoh1-derived dorsal neural tube (dP1/dI1). Notably, Rassf4 site A recapitulates the expression of the Atoh1 autoregulatory enhancer (Helms et al., 2000) (Figure 6A) and drives prominent GFP expression to the Atoh1-derived domain as marked by Atoh1 and Lhx2/9 (Figure 6B, C). Only faint GFP expression colocalizes with Lhx1/5 (Figure 6E) marking dI2 interneurons, and Islet1/2 (Figure 6C, inset) marking dI3 interneurons and can only be seen upon increasing the GFP gain or adding GFP antibody to increase the fluorescence signal. Furthermore, Rassf4 site A drives GFP to the external granule cell layer (EGL) of the developing cerebellum marked by Atoh1 antibody and differentiating granule cells marked by Neurod antibody (Figure 6F–H) (Helms et al., 2001) confirming this enhancer is active in other Atoh1-derived domains.
Klf7 site A drives GFP to Atoh1+ and Lhx2/9+ cells (Figure 6J, K) marking the dP1 and dI1 domains. This enhancer, however, also drives GFP reasonably well to Lhx1/5+ and somewhat to Islet1/2+ cells (Figure 6K, L). This is consistent with the ISH of Klf7 (Figure 2A) where it appears much of the Klf7 transcript is expressed laterally in the mantle zone of the E10.5 neural tube. Taken together, two Atoh1-responsive enhancer elements identified by in vivo binding of Atoh1 are sufficient to direct expression of a reporter gene in an Atoh1-like pattern in transgenic embryos.
As discussed above, Klf7, Rab15, Rassf4, Selm and Smad7 are direct transcriptional targets of Atoh1 in the developing dorsal neural tube. Analysis of mRNA expression of these genes by ISH found that all of these genes are expressed in the developing cerebellum (Figure 2C), and disappear in the Atoh1 mutant that lack a cerebellar EGL (Ben-Arie et al., 1997). Furthermore, Rab15 and Selm are also found in Atoh1-lineage cells in the inner ear and Merkel cells in the vibrissae (Figure 2D, E). Strikingly, these two genes were also found to be in common among Atoh1-lineages by intersecting genes identified in our microarrays of the dP1/dI1 lineage with microarray results of Atoh1-GFP sorted populations from the inner ear (Neil Segil and colleagues, House Ear Institute, unpublished) and Merkel cells from the skin (Haeberle et al., 2004).
bHLH transcription factors have common roles in inducing neuronal differentiation, but distinct roles in neuronal subtype specification, functions that are contingent on developmental context (Parras et al., 2002; Nakada et al., 2004a; Powell et al., 2004; Reeves and Posakony, 2005). To determine Atoh1-specific targets, we first identified transcripts specific to the Atoh1-lineage and not common to the neighboring dorsal Neurog1-lineage. Significantly, we identified five new Atoh1-specific targets and their responsive enhancers using a combination of microarray expression data, ChIP-seq experiments, and enhancer-reporter assays.
Previously, known direct targets of Atoh1 in vivo in the developing neural tube or cerebellum included the homeodomain transcription factors, Barhl1 and Barhl2 (Saba et al., 2005; Kawauchi and Saito, 2008), the Sonic hedgehog transcriptional effector, Gli2 (Flora et al., 2009), and Atoh1 itself (Helms et al., 2000). The direct Atoh1 targets identified here have diverse functions that go beyond the identification of transcription factor cascades. Klf7, Kruppel-like factor 7, a transcription factor implicated in nociceptive neuron development in the dorsal root ganglion (Lei et al., 2005), upregulates the cyclin-dependent kinase inhibitor, p21 (Laub et al., 2005). Interestingly, in Merkel cell carcinomas where Atoh1 plays a tumor suppressor role, Atoh1 upregulates Ntrk1 (TrkA) and p21 expression leading to cell cycle arrest (Bossuyt et al., 2009) which together with our evidence could be through Klf7. Notably, in dI1 neurons, Ntrk3 (TrkC), is enriched in the Atoh1-derived domain (Figure 2A) indicating that Atoh1 may activate different neurotrophic receptor tyrosine kinases under different contexts.
Two of the target genes discovered are associated with the Ras pathway. Activated Ras proteins are usually associated with growth and proliferation; however, some Ras effector proteins such as Rassf4, Ras association (RalGDS/AF-6) domain family 4, are thought to be tumor suppressor genes that bind activated K-Ras (Eckfeld et al., 2004). As a target of Atoh1, Rassf4 may reduce proliferation allowing the differentiation of dP1 cells into dI1 neurons. Rab15, a small GTPase that is a member of the RAS oncogene family, appears to inhibit early endocytosis and recycling in cultured cells (Zuk and Elferink, 2000). An attractive hypothesis is that expression of Rab15 in dP1 cells may inhibit the endocytosis of a receptor, perhaps Notch or BMP receptor, and allow for differentiation of the cell. However, it also may play a role in neuronal migration as has been implicated for Rnd2, the small GTPase found as a Neurog2 target (Heng et al., 2008), and other Rab GTPases (Kawauchi et al., 2010).
The last two genes discovered likely play a role in the proliferation versus differentiation decision during development. Selm, selenoprotein M, is enriched in the brain where it may serve a protective role in Alzheimer’s disease possibly by inhibiting β/γ-secretase activity and decreasing Tau phosphorylation (Korotkov et al., 2002; Hwang et al., 2005; Yim et al., 2009). The role of Selm in inhibiting β/γ-secretase may be reconciled with a possible developmental signaling mechanism that is activated by Atoh1 to inhibit γ-secretase, thereby preventing the cleavage of the Notch intracellular domain allowing for differentiation of progenitor cells (Louvi and Artavanis-Tsakonas, 2006). Lastly, Smad7 inhibits TGFβ signaling through interactions with the type I receptor (Hayashi et al., 1997) and can even interact with β-catenin in cancer cells to promote cell adhesion (Hoover and Kubalak, 2008). Future work will be required to address exactly what these Atoh1 targets are doing in the Atoh1-lineages and if their expression is specifically needed for Atoh1 neuronal subtypes to develop and function in proprioceptive neuronal circuitry.
Other studies have identified genes downstream of Atoh1, but they likely represent downstream effectors of the differentiation role shared by other bHLH factors (Castro et al., 2006; Krizhanovsky et al., 2006; Seo et al., 2007; Miesegaes et al., 2009). For example, Hes5, is induced by Atoh1 in E14.5 cerebellum (Krizhanovsky et al., 2006), but is also induced by Neurog2 in P19 cells suggesting it is a common target of these bHLH transcription factors (Seo et al., 2007). Furthermore, the Hes5 Drosophila homolog, E(Spl), was found to be a target of both atonal and Scute (Reeves and Posakony, 2005; Aerts et al., 2010). Although multiple genes have been identified downstream of Atoh1 (such as Nr2f6 and Cbln2) or atonal (mouse homologs Cdkn1a and Tacr3) (Powell et al., 2004; Krizhanovsky et al., 2006; Sukhanova et al., 2007; Miesegaes et al., 2009; zur Lage and Jarman, 2010) additional experiments are required to determine if these are direct and constitute neuronal subtype specific targets. Recently, genome-wide scale identification of atonal targets has suggested that atonal does not directly activate terminal differentiation genes, but instead activates molecules in major signaling pathways (Aerts et al., 2010); however, another analysis found that atonal was able to directly activate at least one differentiation gene (Cachero et al., 2011) and it is known that the related bHLH factor, chick Atoh7, can directly activate a terminal differentiation gene in the retina (Skowronska-Krawczyk et al., 2005). Together, analysis of targets for atonal and Atoh1 indicate that these transcription factors can turn on genes with wide array of functions including transcription factors, signaling pathways, and terminal differentiation genes, which has also been shown for the cerebellum (Klisch et al., 2011).
With the identification of several Atoh1-specific targets, we have laid the foundation for understanding how Atoh1 can activate specific targets relative to other bHLH transcription factors. Two models have been proposed for the activation of Atoh1-specific targets: either Atoh1 binds a unique E-box consensus that is different from other bHLH proteins as has been suggested in Drosophila (Powell et al., 2004), or co-regulatory sites are required to bring in cofactors that work with Atoh1 to drive cell-type specific expression similar to Ascl1 and POU domain transcription factors (Castro et al., 2006) or the Drosophila ETS transcription factor, Pointed, and atonal (zur Lage et al., 2004; Sukhanova et al., 2007).
A common extended E-box, AMCAGMTG, where M is A/C (Figure 5H), was identified using the sequence from eight Atoh1-responsive enhancers. This is a subset of the AtEAM motif identified in Atoh1 bound regions from cerebellum genomic analysis (Figure 5I) (Klisch et al., 2011). This common E-box was highly conserved in six of the eight enhancers (Figures 3, ,4)4) (Kent et al., 2002; Rhead et al., 2010). Species conservation often highlights regulatory areas of interest (Visel et al., 2008), but conservation is not detected in all regulatory binding sites (Jeong et al., 2008; Wilson and Odom, 2009; Schmidt et al., 2010). Furthermore, although we identified a shared E-box among these Atoh1 target enhancers, no distinct activity could be attributed to this E-box over other E-boxes present (Figure 5J, K). An alignment of the Atoh1 common E-box to consensus binding sites identified for atonal, Ascl1, Neurog/Neurod1, and MyoD (Bertrand et al., 2002; Powell et al., 2004; Castro et al., 2006; Seo et al., 2007; Cao et al., 2010) reveals only subtle differences (Figure 5I). The differences between the Atoh1 and ato E-boxes (Figure 5I, arrowhead) (Ben-Arie et al., 1996; Chien et al., 1996; Ben-Arie et al., 2000; Wang et al., 2002) may be due to the differences in function with which the targets were identified (neuronal subtype specification versus differentiation) or the few targets used to form both E-box sequences (Powell et al., 2004; zur Lage and Jarman, 2010). Slight differences in the bHLH consensus sequences may represent true binding preferences in vivo; however, it seems more likely that bHLH factors work with other factors to carry out neuronal subtype specific programs. Indeed, mutating both E-boxes in the Klf7 site A enhancer did not completely abolish enhancer activity (Figure 5J) suggesting Atoh1 may activate an intermediate cofactor that contributes to tissue specific expression.
To identify potential transcription factors that work with Atoh1, we performed a preliminary MEME analysis and uncovered several motifs enriched in the eight test enhancers over control sequences (H.C. Lai and J.E. Johnson, unpublished observation); however, these motifs were novel sequences with no known transcription factor binding sites making it difficult to identify a good candidate co-factor. The ETS binding motif, which we might expect to find due to Pointed being a cofactor of atonal (zur Lage et al., 2004; Sukhanova et al., 2007), was found in both test and control sequences. Lastly, Zic-related factors have been implicated in regulating Atoh1-related cell-type gene expression (Ebert et al., 2003; Bertrand and Hobert, 2009) and six Atoh1 responsive enhancers contain the motif, GGAGCWG where W is A/T, which is within the sequence identified as the Zic1 binding region in the Atoh1 enhancer (Ebert et al., 2003).
In summary, we identified five in vivo targets of Atoh1 in the developing spinal cord that represent genes enriched in the Atoh1-expressing cells in the dorsal neural tube, and demonstrate the proneuronal bHLH factors have unique targets. Finding neuronal subtype specific targets is essential for a basic understanding of neuronal specification processes and will allow for a better framework to understand differentiation of specific neuronal subtypes from embryonic stem cells. Furthermore, identifying the functions of these Atoh1 specific targets within the Atoh1 lineage may reveal tractable therapeutic targets for medulloblastomas (Wechsler-Reya, 2003; Zhao et al., 2008; Flora et al., 2009) or Merkel cell carcinomas (Bossuyt et al., 2009) where Atoh1 is misregulated.
We are very grateful to Drs. Axel Visel and Len Pennacchio for sharing E11.5 neural tube p300 ChIP-seq data, and Drs. Neil Segil, Angelika Doetzlhofer, Patricia White, and Andrew Groves for sharing Atoh1-GFP inner ear microarray data before publication. We thank Dr. Lei Lei for Ntrk3 (TrkC) and Klf7 ISH probes and Dr. Thomas Jessell for the rabbit anti-Lhx2/9 antibody. Thanks to M. Borromeo, P. Mayer, D. Meredith, and T. Vue for critical reading of the manuscript, Z. Barnett, L. Dickel, J. Dumas, and T. Savage for technical assistance, and E. Kim and members of the Johnson Lab for their support. This work was supported by NIH grants F32 NS059165 to H.C.L., RO1 NS048887 to J.E.J., and HHMI to H.Y.Z.
The authors declare no conflict of interest.