Transcription factor autoregulation is required for acquisition and maintenance of neuronal identity

Autoregulation is a feature common to many prokaryotic and eukaryotic transcriptional regulatory systems (Bateman, 1998; Crews and Pearson, 2009; Ptashne et al., 1976). It was first discovered in phage lambda where cI protein, a DNA binding transcription factor, represses the lytic stage of a bacteriophage infection, thereby promoting the lysogenic state. To maintain this specific regulatory state, cI binds to its own promoter to ensure its own continuous expression (Ptashne et al., 1976). The maintenance of cellular states by autoregulating transcription factors has subsequently been observed in a wide variety of cellular contexts in prokaryotic, eukaryotic and multicellular organisms. For example, in multicellular organisms, cell type-specific transcription factors define the identity of specialized, differentiated cell types via the activation of cell type-specific effector gene batteries (Bateman, 1998; Davidson et al., 2003; Hobert, 2016; Levine and Tjian, 2003) and such transcription factors are often also continuously expressed throughout the postembryonic life of a differentiated cell. In many cases, such maintained expression is ensured via autoregulation, in which the respective transcription factor binds to cognate cis-regulatory sites in its own locus (Bateman, 1998; Crews and Pearson, 2009; Schier and Gehring, 1992).

Even though many studies documented the existence of autoregulation (Bateman, 1998; Crews and Pearson, 2009), there have been few genetic loss-of-function studies that formally demonstrated that continuous expression of identity-specifying transcription factors is indeed required to maintain the differentiated state (e.g. Deneris and Hobert, 2014; Etchberger et al., 2009; Kadkhodaei et al., 2009; Kratsios et al., 2011; Liu et al., 2010; Serrano-Saiz et al., 2018, 2013; Song et al., 2011). In these studies, the identity-specifying and often autoregulating transcription factors were removed post-developmentally, i.e. after the differentiated state has been reached, and an ensuing loss of cellular identity was observed. However, formal proof of the importance of autoregulation should leave the transcription factor intact and only remove its autoregulatory aspect, ideally via mutating the cis-regulatory motif that is predicted to confer autoregulation. We provide here such formal proof in the context of neuronal cell type specification in the nematode Caenorhabditis elegans.

In this simple model system, a host of transcription factors, called terminal selectors, have been identified as key regulators of neuronal cell type differentiation (Hobert, 2008, 2016). Neuron type-specific terminal selectors are activated around the time of neuronal birth and control large batteries of effectors genes that impose specific phenotypic features onto individual neuron types. All known terminal selectors are continuously expressed throughout the life of the respective neuron type (Hobert, 2016). Although many terminal selector indeed autoregulate (Baumeister et al., 1996; Hobert et al., 1997; Masoudi et al., 2018; Way and Chalfie, 1989), the functional necessity of the autoregulatory phenomenon per se has not yet been examined.

We used the C2H2 zinc-finger type terminal selector che-1, which is required for the differentiation of the gustatory ASE sensory neuron pair (Fig. 1A) (Chang et al., 2003; Etchberger et al., 2007; Uchida et al., 2003), as a paradigm for investigating the importance of autoregulation. Using a fosmid-based reporter transgene, we have previously shown that che-1 is exclusively expressed in the ASE sensory neuron pair, starting during embryogenesis when the ASE neurons are born, and being subsequently maintained through larval and adult stages (Sarin et al., 2009; Tursun et al., 2009). Initiation of che-1 expression in the embryo requires the orphan nuclear hormone receptor nhr-67/Tailless/Tlx (Sarin et al., 2009). nhr-67 is only transiently expressed in ASE until the first larval stage (Sarin et al., 2009), suggesting that nhr-67 cannot be responsible for the continuous expression of che-1 throughout larval and adult stages. We have also previously shown that a <700 bp region upstream of the che-1 locus, when fused to a fluorescent reporter, recapitulates the continuous ASE-specific expression observed with a fosmid reporter transgene (Etchberger et al., 2007). This upstream region behaves like many other cis-regulatory control regions of transcription factors: (1) it is turned off in ASE neurons in a che-1 mutant background, demonstrating that che-1 autoregulates its expression (Etchberger et al., 2007); and (2) in addition to putative initiator sites, it contains a single copy of a biochemically defined CHE-1 binding site (the ‘ASE motif’; Fig. 1A) (Etchberger et al., 2007), suggesting that autoregulation is direct.

The advent of CRISPR/Cas9-based genome engineering technologies (Arribere et al., 2014; Doudna and Charpentier, 2014; Kim et al., 2014; Ward, 2014) has offered us the opportunity to examine the functional relevance of this autoregulatory cis-acting motif in the context of the endogenous che-1 locus. We demonstrate here that this autoregulatory motif is indeed required for maintaining ASE neuronal identity and function, and, in addition, we have uncovered evidence for a role of autoregulation in the initial amplification of che-1 expression.

We used CRISPR/Cas9 to generate che-1(ot856[che-1::gfp]) animals in which we tagged the che-1 locus with gfp. We then used CRISPR/Cas9 again in these animals to generate two independent mutant alleles of the ASE motif in the 5′ promoter region of che-1, called che-1(ot908 ot856) and che-1(ot871 ot856) (Fig. 1A). The ASE motif is twelve nucleotides long, with each of the four Zn fingers of CHE-1 binding four adjacent sets of three nucleotides (Etchberger et al., 2007) (Fig. 1A). Based on previous extensive in vitro analysis of CHE-1 binding to ASE motifs (Etchberger et al., 2007), both CRISPR/Cas9-generated cis-regulatory alleles are predicted to affect CHE-1 binding. One of the two alleles still retains an invariant 3 nucleotide motif normally binding the fourth Zn finger (Etchberger et al., 2007) and could potentially display a weaker effect than the other allele, which disrupts almost all nucleotides known to be involved in CHE-1 binding (Fig. 1A).

To assess the functional consequence of mutating the che-1 autoregulatory motif, we first investigated the main animal behavioral output requiring ASE neuron function: salt chemotaxis (Fig. 1B). Sensory perception of sodium chloride induces ASE sensory neuron activity (Suzuki et al., 2008) and sodium chloride chemotaxis is abolished in animals in which either the ASE neuron is microsurgically removed (Bargmann and Horvitz, 1991) or which carry deletion or coding mutations of the che-1 locus (Uchida et al., 2003; Ward, 1973). We find that adult animals that carry either ASE cis-regulatory mutations fail to properly exhibit chemotaxis (Fig. 1C). In older stage animals, the effect of the strong cis-regulatory allele is almost as strong as a che-1 null allele, whereas the weaker cis-regulatory allele results in the retention of some chemotactic ability (Fig. 1C). Chemotaxis defects of the cis-regulatory alleles are less pronounced in earlier stage animals but are already detectable in first larval stage animals in more-challenging dilute NaCl gradients (Fig. 1C,D). Overall, the progressive worsening of the chemotactic defects during postembryonic maturation of the animal clearly demonstrate that the ASE motif and, hence, che-1 autoregulation, is required to maintain the functional properties of the ASE neurons. However, chemotaxis defects observed at earlier stages already suggest that ASE neurons may have actually never attained a fully functional state.

We also tested the effect of mutations in the autoregulatory motif of che-1 locus on the expression of four molecular markers of the ASE differentiated state: two chemoreceptor-encoding genes, the receptor-type guanylate cyclases gcy-5 and gcy-7 (both exclusively expressed in ASE; Yu et al., 1997); and two neuropeptide-encoding genes, the FMRFamides flp-4 and flp-6 (both expressed in ASE, as well as in other neurons; Kim and Li, 2004). All four genes are targets of the che-1 transcription factor, i.e. their expression is completely abrogated in che-1-null mutants, and three of them contain ASE motifs that have been shown to be required for expression in the ASE neurons (Chang et al., 2003; Etchberger et al., 2007; Uchida et al., 2003) (Fig. 1A). We found that the expression of gcy-5 in adult worms is significantly reduced in both cis-regulatory alleles, with close to undetectable expression levels in the stronger allele (Fig. 2A,B). Similar results were observed when gcy-7 and flp-6 expression was analyzed: a severe reduction in expression in the stronger allele and a milder defect in the weak cis-regulatory allele both in adult and larval stages (Fig. 2C-F). flp-4 expression is affected equally by both alleles (Fig. 2G,H).

In line with the chemotaxis defects of young larvae, we also noted in the molecular marker analysis that removal of the autoregulatory motif affects expression of differentiation marker at early larval stages. Specifically, we observed that during embryonic and larval stages, gcy-5 expression levels in mutant worms never reached those observed in wild-type animals (Fig. 2B). Likewise, defects were already detected at the time when gcy-7 or flp-6 expression is activated (Fig. 2D,F). In summary, although effector gene expression defects observed upon disruption of che-1 autoregulation are generally more pronounced in the adult stage (corroborating the importance of autoregulation in maintaining the differentiated state of the ASE neurons), defects were already observed at much earlier developmental stages.

After characterizing the downstream effects of removing the che-1 autoregulatory motif, we examined che-1 expression, using the gfp-tagged che-1 locus (ot856 allele). We observed the same expression pattern as that found for the previously published fosmid-based reporter construct (Sarin et al., 2009). Specifically, expression was exclusively observed in the two ASE neurons, starting around the time of ASE cell birth in the embryo, and continuing throughout all ensuing embryonic, larval and adult stages (Fig. 3A-C). Examination of the expression of the gfp-tagged che-1 locus in which the ASE motif was mutated, che-1(ot908 ot856) and che-1(ot871 ot856), revealed an intriguing pattern, one part of it expected, another part unexpected. In larval and adult stages, we observed an expected reduction of che-1::gfp expression (Fig. 3B,C). Adult animals carrying the allele that disrupts the binding of all four zinc fingers exhibit a significantly greater decrease of che-1::gfp expression compared with the cis-regulatory allele in which one zinc finger of CHE-1 is still predicted to bind DNA. In larvae, we also observe a significant reduction in che-1::gfp expression in both cis-regulatory alleles compared with wild-type animals at all larval stages (Fig. 3B,C). This reduction is similar in the first (L1) and second (L2) larval stages, whereas the reduction in expression is significantly higher in the stronger allele starting at the L3 larval stage. Among all post-embryonic stages, animals carrying the weaker cis-regulatory allele always showed detectable che-1::gfp expression levels, possibly owing to residual binding of CHE-1 to the mutated ASE motif. In both alleles, che-1 expression defects are less pronounced at earlier stages, perhaps due to the perdurance of the embryonic initiators of che-1 expression, such as nhr-67, the expression of which fades postembryonically (Sarin et al., 2009). The overall pattern of decreases in che-1 expression correlates with the decreases in chemotactic behavior described above (Fig. 1), such that defects are generally less pronounced at early larval stages and more pronounced in late larval or adult stages. However, as the animals progresses through larval development, the precise timing of che-1 expression decreases is not well correlated with losses of chemotactic ability. For example, although animals carrying the stronger allele show a progressive reduction on che-1 expression levels, the chemotactic response drops abruptly in L2 larvae. As che-1 controls the expression of hundreds of target genes, many of which involved in controlling ASE function, the phenotypic consequence of partial losses of subsets of targets upon partial loss of che-1 expression is hard to predict.

The unexpected part of the che-1::gfp expression pattern was revealed during embryogenesis. We found that in the two ASE motif-mutant strains, che-1::gfp expression never reaches wild-type levels even at those earlier embryonic stages at which che-1 expression first becomes detectable (Fig. 3B and 4A). Specifically, we find that, in wild-type animals, che-1::gfp expression is normally initiated at the pre-bean stage, and amplified from this stage to the bean stage. The observed onset of CHE-1 expression at the late pre-bean stage agrees with previous reports in which the onset was determined by fosmid reporter expression (Sarin et al., 2009) and by smFISH (Cochella et al., 2012), suggesting that our analysis captures the very onset of CHE-1 expression. Transiently expressed initiators of che-1 expression, such as nhr-67, trigger this initiation (Sarin et al., 2009) but, as we show here, this initial amplification is defective in both cis-regulatory alleles (Fig. 4A). These results indicate that che-1 autoregulation is not only required to maintain wild-type che-1 expression levels at larval and adult stages but that it may also be required to amplify che-1 expression during initial specification of ASE neuron identity.

One caveat of this interpretation is that the ASE motif may not operate as an autoregulatory motif at the stage of initiation of che-1 expression, but that it may rather serve as a binding site for distinct upstream factors that initiate che-1 expression. To test this possibility, we used CRISPR/Cas9 genome editing to engineer gfp into the che-1(ot63) loss-of-function allele that carries a missense mutation in the fourth zinc-finger domain of CHE-1 (Cys255Tyr) (Chang et al., 2003). This fourth zinc finger is essential for CHE-1 binding to the ASE motif in vitro (Etchberger et al., 2007) and che-1 gene function in vivo (Chang et al., 2003). We reasoned that if che-1 had no role in the initial establishment of che-1 expression, we would observe normal initiation of che-1 expression, followed by a failure to maintain its expression; if, in contrast, che-1 had a role in the initial establishment of che-1 expression, we should observe little to no che-1 expression. We indeed found that che-1(ot63 ot941) animals show a substantially reduced level of gfp expression in the embryo, failing to reach che-1 expression levels detected in wild-type animals (Fig. 4A). The effect of disabling CHE-1 protein function on the initiation of che-1 gene expression (i.e. initial amplification) is even stronger than mutation of the ASE motif (Fig. 4A). We ascribe this effect to the presence of an additional ASE motif in the second intron in the che-1 locus, which may contribute to CHE-1 autoregulation. Thus, disabling CHE-1 protein function may completely eliminate the ability to properly amplify che-1 expression through either of the two ASE motifs. We conclude that che-1 is indeed required to amplify its own expression, likely pursuant to a weak initial regulatory input.

Although transcription factor autoregulation is a pervasive gene regulatory phenomenon, the functional relevance of autoregulation in cellular differentiation has not been extensively probed through the genetic removal of autoregulatory, cis-acting transcription factor binding motifs. The only instance that we are aware of is the removal of a large cis-regulatory element with several Krox20-binding sites from the mouse and zebrafish Krox20 loci, which implicates Krox20 autoregulation in rhombomere specification (Bouchoucha et al., 2013; Torbey et al., 2018). We have undertaken a more fine-grained analysis in a distinct cellular and organismal context, addressing the importance of autoregulation during the execution of a terminal differentiation program in the nervous system. Through constitutive or post-developmental removal of che-1 gene activity, this transcription factor locus was previously known to be required to initiate and maintain the differentiated state of a single chemosensory neuron in C. elegans (Etchberger et al., 2009). If maintenance of expression were the only function of autoregulation, mutation of the autoregulatory motif would be predicted to result in effects schematized in Fig. 4B. Transiently expressed factors (such as nhr-67)(Sarin et al., 2009) initiate che-1 expression and levels of CHE-1 are sufficient to initially trigger the onset of expression of che-1 target genes and, hence, ASE differentiation. Subsequently, owing to the lack of autoregulation, che-1 expression, as well as effector gene expression, would fade out. Although we clearly do observe a progressive worsening of the loss of che-1 and effector gene expression, as well as neuron function (corroborating the previously described maintenance role of che-1; Etchberger et al., 2009), we also observed effects that we interpret as schematized in Fig. 4C. In the absence of the autoregulatory motif, che-1 expression never reaches normal wild-type levels of expression, even early in the embryo when che-1 expression is initiated. Hence, the fully functional differentiated state is never attained. We interpret this to mean that the factor(s) that initiate che-1 expression (such as nhr-67) (Sarin et al., 2009) may be weak activators that are alone not sufficient to produce the amounts of che-1 expression required to activate che-1 target genes. Autoregulation of che-1 seems to ensure that che-1 expression is amplified to reach the critical threshold for activation of downstream target genes. This mechanism contrasts with other autoregulation cases that follow the more conventional model shown in Fig. 4B, where the autoregulation phase is clearly separated from the initiation phase. For example, in the tunicate Ciona intestinalis, maintenance of Ebf expression in atrial siphon muscle founder cells relies on autoregulation, while the initial amplification depends on an initiation event that requires MAPK signaling (Razy-Krajka et al., 2018). In contrast, in mouse, Krox20 autoregulation is required for the conversion of a transient input into a stable fate commitment (Bouchoucha et al., 2013). However, Krox20 expression is transient and its function is not required for the maintenance of the differentiated state. We describe here a case in which transcription factor autoregulation is required both to maintain the differentiated neuronal state in adult animals and to amplify transcription factor expression during embryonic development to initially specify neuronal identity.

In conclusion, the function of transcriptional autoregulation goes beyond a mere maintenance function by ‘boosting’ transcription factor expression beyond a critical threshold required to activate its downstream effectors. This auto-amplifying, boosting mechanism enables the detection of weak and transient regulatory inputs to eventually lock them into a stably maintained, terminal state. Considering the previously reported case of vertebrate Krox20, for which a similar auto-amplification has been reported in the context of rhombomere development (Bouchoucha et al., 2013), we propose that the functional duality of transcriptional autoregulation, initial amplification and ensuing maintenance, constitutes a widely used gene regulatory principle during animal development.

The C. elegans strains used in this study were: OH14130, che-1(ot856[che-1::gfp]) (Leyva-Diaz et al., 2017); OH15579, che-1(ot908 ot856) (generated in this study); OH15683, che-1(ot871 ot856) (generated in this study); OH15815, che-1(ot63 ot941) (generated in this study); OH610, che-1(ot63) (Chang et al., 2003); and OH13098, che-1(ot75) (Chang et al., 2003). The reporter transgenes were: OH3192, ntIs1[gcy-5^prom::gfp] (Chang et al., 2003); OH1092, otIs131[gcy-7^prom::rfp] (Chang et al., 2003); OH12372, otIs494[flp-6^fosmid::sl2::1xNLS::mChOpti]; and NY2030 ynIs30[flp-4^prom::gfp] (Kim and Li, 2004). Animals were maintained at 20°C with abundant E. coli OP50 as food following standard conditions (Brenner, 1974).

The CRISPR/Cas9-mediated modification of the genome to mutate the ASE motif on the che-1 promoter was performed using a Co-CRISPR screening strategy (Arribere et al., 2014; Kim et al., 2014; Ward, 2014). This screening strategy uses a visible phenotype at one locus to help identify edits at a second locus of interest. Briefly, che-1(ot856[che-1::gfp]) adult hermaphrodites were used for gonad injection of a DNA injection mix containing: eft-3::Cas9 plasmid (50 ng/µl), che-1 sgRNA plasmid (70 ng/µl), che-1 homology directed repair (HDR) plasmid (40 ng/µl), Co-CRISPR sgRNA plasmid (40 ng/µl) and myo-2::mCherry plasmid (3 ng/µl). We used a guide RNA that targets a sequence 108 bp downstream the ASE motif on the che-1 promoter (target sequence: ATCACAAAAATAAAGAGGG). The che-1 HDR plasmid contains a 2.2 kb homology region centered around the ASE motif on the che-1 promoter (−1486, +758). In this plasmid, the ASE motif was substituted into an EcoRI restriction site for the che-1(ot908) mutation and into a PmeI site for the che-1(ot871) mutation, while the PAM site was mutated to prevent Cas9 from cutting the repair template (TGG to TAG). che-1(ot908) was generated using unc-22 mutations for Co-CRISPR, while che-1(ot871) was generated using a gain-of-function dpy-10(cn64) Co-CRISPR approach. When dpy-10(cn64) Co-CRISPR was used, a dpy-10 repair oligo (Arribere et al., 2014) (14 ng/µl) was included in the injection mix instead of myo-2::mCherry. For the che-1(ot908 ot856) mutation, transformed F1 progeny of the injected animals were identified based on the myo-2::mCherry co-injection marker and singled into independent plates. F1 worms that produced F2 progeny with a ‘twitching’ phenotype (unc-22 phenotype) were screened by PCR to identify animals carrying the desired modification. For the che-1(ot871 ot856) mutation, che-1(ot908 ot856) and che-1(ot871 ot856) mutations were outcrossed at least five times. CRISPR/Cas9-mediated gfp tagging of the che-1(ot63) locus was carried out with the same gfp cassette present in che-1(ot856[che-1::gfp]), using a previously described protocol (Dokshin et al., 2018). che-1(ot63) adult hermaphrodites were used for gonad injection of a mix containing S. pyogenes Cas9 protein (250 ng/µl, IDT), tracrRNA (100 ng/µl, IDT), che-1 crRNA (56 ng/µl, IDT), dsDNA donor cocktail (200 ng/µl) and PRF4::rol-6(su1006) plasmid (40 ng/µl). We used a crRNA that targets a sequence 21 bp upstream the che-1 STOP codon (target sequence: CACAGAGTGGGAACTTGCAT).

Worms were anesthetized using 100 mM sodium azide (NaN₃) and mounted on 5% agarose pads on glass slides. Images were acquired using a Zeiss confocal microscope (LSM880). Several z-stack images (each ∼0.7 µm thick) were acquired with the ZEN software. Images were reconstructed via maximum intensity z-projection of 2-10 µm z-stacks using the ImageJ software (Schneider et al., 2012). Representative images are shown following orthogonal projection of 2-10 µm z-stacks.

For fluorescence intensity quantification, reporter expression levels were analyzed using the mean gray value of ImageJ version 1.51. Fluorescence intensity was measured in the focal plane with the strongest ASE expression within the z-stack. For each fluorescent reporter, the ASE neuron area was delineated (region of interest, ROI). For each image, the same ROI was also used to measure the background next to ASE in the same focal plane, and this value was then subtracted from the reporter fluorescence intensity value in ASE. For all images used for fluorescence intensity quantification, the acquisition parameters were maintained constant among all samples (same pixel size, laser intensity, PMT voltage, pinhole, etc.) for each of the fluorescent reporters. The fluorescence intensity of the gcy-5^prom::gfp and flp-4^prom::gfp reporters is much higher than that of che-1::gfp. We tested that che-1::gfp fluorescence was not detectable with the acquisition settings used for gcy-5^prom::gfp or flp-4^prom::gfp so that che-1::gfp fluorescence did not interfere with the reporters fluorescence intensity acquisition.

The response to NaCl gradients was assayed as previously described (Bargmann and Horvitz, 1991). Briefly, 10 ml of buffered agar (20 g/l agar, 1 mM CaCl₂, 1 mM MgSO₄ and 5 mM KPO₄) was poured into 10 cm diameter petri dishes. To establish the chemical gradient, we applied 10 μl of NaCl solution to the attractant spot and 10 μl of double-distilled H₂O to the control spot. The NaCl was allowed to diffuse for 14-16 h at room temperature before the assay. To increase the steepness of the gradient, another 4 μl of NaCl solution or water was added to the same spots 4 h before the assay. We applied a 1 μl drop of 1 M sodium azide to both attractant and control spots 10 min before the assay to immobilize worms that reached these areas. Synchronized animals were washed three times with CTX solution (1 mM CaCl₂, 1 mM MgSO₄ and 5 mM KPO₄) and 100-200 animals were placed in the center of the assay plate in a minimal volume of buffer. Animals were allowed to move about the agar surface for 1 h, after which assay plates were placed at 4°C overnight. For worms assayed at all stages in Fig. 1C, a 2.5 M NaCl solution was applied to the attractant spot. Additionally, for L1 worms, different salt dilutions (1.25 M, 0.75 M and 0.25 M NaCl) were used in Fig. 1D.

The distribution of animals across the plate was then determined and a chemotaxis index was calculated as the number of animals at the NaCl spot minus the number of animals at the control spot, divided by the total number of animals. Animals that did not leave the initial inner circle were not included in the count of total number of animals, as these animals were dead or had movement defects. Chemotaxis assays were performed on at least three independent days with at least three assays per day.

For quantification data shown in graphs of all figures, dot symbols represent individual values, the black line indicates the mean value and error bars represent the s.e.m. Statistical analyses were carried out using a two-way ANOVA followed by Tukey's multiple comparisons test. Wild type versus weak, wild type versus strong, wild type versus null and weak versus strong allele comparisons are shown if significant (Figs 1–3); only comparisons with wild type are shown in Fig. 4; *P<0.05, **P<0.01, ***P<0.001. The Shapiro-Wilk normality test was performed on all data to test whether the values follow a normal distribution. All statistical analyses were performed using GraphPad Prism 7.0d for Mac OS X.

We thank Chi Chen for the C. elegans injections, Dylan Rahe for providing the GFP-tagged che-1(ot856) allele, Tulsi Patel for providing the flp-6 fosmid reporter, and Mark Ptashne, Steve Crews and members of the Hobert lab for comments on this manuscript.