Summary
Long term synaptic plasticity requires transcription in response to changes in neuronal activity. While genes induced by neuronal activity have been extensively studied, genes induced by hyperpolarization are largely unknown. We focused on Pura, a Rho1 GEF whose rhythmic expression drives the daily retraction of the projections of Drosophila LNv circadian pacemaker neurons. We found that Pura transcription is repressed by activity and induced by hyperpolarization in LNvs – the opposite of typical activity-regulated genes. Pura is repressed by activity-regulated transcriptional factors including Mef2 and Sr (fly Egr-1) and activated by Toy, a Pax6 transcription factor. toy transcription is also induced by inactivity. Thus toy and Pura represent a class of genes induced by hyperpolarization.
Highlights
Hyperpolarization activates transcription of Pura, a plasticity gene
This phenomenon occurs in circadian pacemaker neurons and mushroom body neurons
The Pura enhancer integrates recent neuronal activity to regulate transcription
Hyperpolarization activates transcription of toy (Pax-6), which then activates Pura
Introduction
Neuronal plasticity is essential for the nervous system to wire correctly and underlies processes including learning and memory and recovery from strokes (reviewed by (Kandel et al., 2014; Murphy and Corbett, 2009). Disruptions to plasticity are increasingly being associated with neurological disorders such as autism spectrum disorder and schizophrenia (reviewed by (Ebert and Greenberg, 2013; Glausier and Lewis, 2013).
Plasticity encompasses changes in both the excitability and structure of neurons. While changes in neuronal excitability alter the strength of the signals transmitted, structural plasticity can make new connections between neurons. New gene expression plays a major role in plasticity via a set of activity-dependent genes whose transcription is rapidly induced by neuronal firing. The products of activity-dependent genes such as Arc and Homer1a directly regulate synaptic function (reviewed by (Leslie and Nedivi, 2011), while others such as c-fos and Nur77 / Nr4a1 encode transcription factors whose targets likely include genes that alter neuronal connectivity (Yang et al., 2016).
However, neuronal plasticity also includes reducing neuronal excitability and weakening or losing connections. These aspects of neuronal plasticity are probably important in forgetting memories and also in sleep, in which widespread synaptic downscaling has been proposed to reduce overall connectivity in the brain (de Vivo et al., 2017; Diering et al., 2017). However, relatively few molecular mechanisms have so far been described to explain how neurons decrease their connectivity and the role of gene expression in this phenomenon has been largely ignored.
One example of how gene expression can weaken neuronal connections comes from Drosophila circadian pacemaker neurons, which drive ∼24hr rhythms in behavior. A subset of pacemaker neurons – the 4 small ventral lateral neurons (s-LNvs) in each brain hemisphere – show predictable and intrinsic daily rhythms in their excitability (Cao and Nitabach, 2008) and in the structure of their projections (Cao and Nitabach, 2008; Fernandez et al., 2008). We recently showed that Rho1 activity is rhythmic in s-LNv projections and this is regulated by rhythmic expression of Puratrophin-like (Pura), a Rho1 Guanine nucleotide exchange factor (GEF) that increases Rho1 activity (Petsakou et al., 2015). s-LNv plasticity is also regulated by neuronal activity, which normally peaks around dawn (Cao and Nitabach, 2008) when s-LNv projections are maximally expanded. (Sivachenko et al., 2013) found that artificially inducing s-LNvs to fire at dusk rapidly expanded their projections to the dawn-like state. Part of the mechanism that allows s-LNv projections to expand at dusk in response to firing could be that increasing neuronal activity reduces Pura RNA levels in LNvs (Chen et al., 2016; Mizrak et al., 2012). Indeed, it seems important for neuronal activity to repress genes such as Pura that retract neuronal projections since activity-induced genes tend to have the opposite effect to Pura on neuronal structure (Yang et al., 2016).
We investigated how Pura transcription responds to neuronal activity. We constructed transcriptional reporter genes for Pura and Hr38, which is a typical activity-regulated gene: Hr38 expression is induced by neuronal activity in invertebrates just like its mammalian orthologue Nr4a1 (Fujita et al., 2013; Milbrandt, 1988). Neuronal activity induced Hr38 but repressed Pura transcription in LNv clock neurons. More interestingly, hyperpolarizing LNvs activated Pura transcription and repressed Hr38 transcription. Thus transcription of Hr38 and Pura responds to both neuronal activity and hyperpolarization, but in opposite directions to each other.
Dissecting the Pura regulatory region, we identified an 116bp enhancer that gives the correct spatial and temporal expression of Pura in LNvs as well as responding to both neuronal activity and hyperpolarization. This enhancer contains binding sites for Mef2, Stripe (Sr, the Drosophila Egr-1 orthologue) and Atf-3, which are either post-translationally or transcriptionally activated by neuronal activity in mammals (Mao et al., 1999; Milbrandt, 1987; Zhang et al., 2011). All three factors repressed Pura transcription, suggesting a mechanism for how neuronal activity represses Pura. We also identified that Toy, one of the two Drosophila Pax6 transcription factors (Czerny et al., 1999), activates Pura transcription likely through a Toy binding site in the minimal Pura enhancer. Since the Toy binding site in the minimal Pura enhancer overlaps the Mef2 and Sr binding sites, the Pura enhancer seems to integrate recent neuronal activity to determine levels of Pura transcription. Strikingly, toy transcription itself is activated when LNvs are hyperpolarized. We propose that inactivity-driven gene expression is a general property of plastic neurons.
Results
Transcription of Hr38 and Pura responds in opposite directions to neuronal activity and hyperpolarization
We developed reporter genes to test if transcription of Hr38 and Pura responds to neuronal activity. To identify the regulatory regions that direct expression to LNvs, we crossed the available Hr38- and Pura-Gal4 lines (Jenett et al., 2012; Kvon et al., 2014) to UAS-GFP and identified one Gal4 line for each gene that expressed in larval LNvs (Fig. 1A and Fig. S1A). The Hr38 enhancer from this Gal4 line was fused upstream of the Hsp70 basal promoter, whereas the Pura enhancer already contained its endogenous basal promoter and transcriptional start site. Downstream we fused a codon-optimized and destabilized tdTomato fluorescent protein, which has three copies of a nuclear localization signal (NLS) to focus signals for quantification (Mezan et al., 2016). We also added an intervening sequence (IVS) to promote nuclear export of spliced mRNA (Pfeiffer et al., 2010) and a synthetic 5’ UTR (Syn21) to increase translation (Pfeiffer et al., 2012). Hr38- and Pura-Tomato constructs were inserted to the same location as each other on chromosome 2.
To test if the reporter genes follow the temporal patterns of Hr38 and Pura RNA, we measured Tomato levels in larval LNvs across a 12hr:12hr light:dark (LD) cycle. We found that Hr38-Tomato levels in LNvs peak at ZT2 (Zeitgeber Time, ZT0 = lights on, ZT12 = lights off) (Fig. 1B). This is consistent with previous reports of higher Hr38 mRNA levels in LNvs at CT3 than CT15 (Circadian Time, time in constant darkness after entrainment in LD cycles) (Mizrak et al., 2012) and with LNvs being more excitable at dawn than dusk (Cao and Nitabach, 2008). Pura-Tomato levels were maximal at ZT18 and low by ZT24 (Fig. 1B), which is very similar to Pura RNA (Ruben et al., 2012). The data shown here are for Pura-Tomato B, which has ∼1.5kb of regulatory region directly upstream of the Pura basal promoter. Pura-Tomato B showed very similar temporal dynamics as Pura-Tomato A, which has an additional ∼0.5kb downstream of the transcriptional start site (Fig. S1B). Thus the Hr38 and Pura transcriptional reporters have similar temporal dynamics to their RNAs (Mizrak et al., 2012; Ruben et al., 2012).
To test how neuronal activity regulates Hr38 and Pura transcription, we first used a UAS-NaChBac transgene (UAS-NCB) to hyperexcite LNvs. NCB is a low threshold voltage-gated bacterial Na+ channel that increases LNv excitability (Nitabach et al., 2006; Yuan et al., 2011). UAS-NCB was expressed specifically in larval LNvs using a Pigment dispersing factor-Gal4 driver (Pdf-Gal4) (Park et al., 2000). [In the figures, we use the nomenclature Pdf > X to indicate larvae with Pdf-Gal4 and UAS-X transgenes.] Larvae were entrained in LD and Hr38- and Pura-Tomato levels were measured at ZT24, just before the normal increase in Hr38 transcription after dawn. Expressing NCB in LNvs increased Hr38-Tomato levels 2.2-fold and reduced Pura-Tomato levels 1.4-fold compared to control LNvs expressing a UAS-GFP transgene (Fig. 1C).
To test how hyperpolarization affects Hr38 and Pura transcription, we used Pdf-Gal4 to express UAS-Kir2.1, a mammalian inward rectifier K+ channel (Baines et al., 2001). We again measured Hr38- and Pura-Tomato levels at ZT24 and found that Hr38-Tomato levels were 2-fold lower while Pura-Tomato levels were 3.3-fold higher in Kir2.1-expressing LNvs than in control LNvs (Fig. 1C). Thus transcription of Hr38 and Pura responds in opposite directions to long-term changes in neuronal activity: Hr38 transcription is activated when LNvs are hyperexcited and repressed when LNvs are hyperpolarized, whereas Pura is activated by hyperpolarization and repressed by increased activity.
Hr38 and Pura transcription responds in opposite directions to transient changes in neuronal activity
To test if Hr38 and Pura transcription respond to more transient changes in neuronal activity, we used Pdf-Gal4 to ectopically express UAS-dTrpA1, a heat-activated Drosophila cation channel (Hamada et al., 2008). Since TrpA1 is inactive below 23°C, larvae were entrained in LD cycles at 22°C and then shifted to 30°C for 4hr from ZT12 to ZT16. We selected lights-off to induce firing since LNvs are normally relatively inactive at this time (Cao and Nitabach, 2008). The data in Fig. 2A show that Hr38-Tomato levels at ZT16 were 3.8-fold higher in dTrpA1-expressing LNvs than in control LNvs without a dTrpA1 transgene that followed the same temperature shift. In contrast, Pura-Tomato levels were 4.6-fold lower in dTrpA1-expressing LNvs than in control LNvs.
We also tested the responses of Hr38 and Pura to transient hyperpolarization. For this, we used a temperature-sensitive Gal80 transgene (tubulin-Gal80ts, (McGuire et al., 2003)) to repress Gal4 activity and regulate expression of UAS-Kir2.1. Larvae were entrained in LD cycles at 18°C when Gal80ts is active and shifted to 30°C at ZT22 to inactivate Gal80 and induce Kir2.1 expression. Larvae were maintained at 30°C for 4hr in darkness to prevent light-induced neuronal activity (Yuan et al., 2011). The data in Fig. 2B show that Hr38-Tomato levels were 1.4-fold lower while Pura-Tomato levels were 2-fold higher in LNvs expressing Kir2.1 than in control LNvs. Thus Hr38 and Pura transcription respond in the same way to more transient changes in LNv electrical activity (Fig. 2A, B) as they do to long-term changes (Fig. 1C).
Hr38 and Pura transcription respond in opposite directions to hyperpolarization via an intact neural circuit
The DN1 dorsal larval clock neurons release glutamate and inhibit LNvs via GluCl, a glutamate-gated chloride channel (Collins et al., 2012). We therefore decided to manipulate DN1 activity and ask if changes in Hr38 and Pura transcription in LNvs could be detected when LNv activity was altered via an inhibitory neural circuit.
cry39-Gal4 was used with Pdf-Gal80 to restrict UAS-dTrpA1 and UAS-GFP expression to larval DN1s. cry39-Gal4 is expressed in larval DN1s and LNvs (Picot et al., 2007), but the temperature-independent Pdf-Gal80 transgene (Stoleru et al., 2004) constitutively blocks Gal4 activity in LNvs. We confirmed this by assaying GFP localization (Fig. S2A). Larvae with cry39-Gal4, Pdf-Gal80 and UAS-dTrpA1 transgenes are labeled in Fig. 2C as DN1 > TrpA1 for simplicity.
Larvae were entrained in LD cycles at 22°C and then shifted to 30°C for 4hr from ZT8 to ZT12 to activate dTrpA1 in DN1s. The enhancer in the Hr38-Tomato construct also gives expression in DN1s, allowing us to validate that dTrpA1 induced firing in DN1s. The data in Fig. 2C show that Hr38-Tomato levels at ZT12 were 3.2-fold higher in dTrpA1-expressing DN1s than in control DN1s that went through the same temperature shift but lacked a dTrpA1 transgene.
Next we measured Hr38-Tomato in LNvs. The data in Fig. 2C show that Hr38-Tomato levels were 1.4-fold lower in LNvs in larvae in which DN1s were induced to fire than in LNvs from control larvae (Fig. 2C). Thus increased neuronal activity in DN1s induced Hr38 expression in DN1s and repressed Hr38 in LNvs, which is consistent with DN1s inhibiting LNv activity.
Next we measured Pura-Tomato levels in LNvs when DN1s were induced to fire from ZT8 to ZT12. The data in Fig. 2C show that Pura-Tomato levels in LNvs at ZT12 were 2.5-fold higher in larvae with dTrpA1 active in DN1s than in control larvae without a dTrpA1 transgene. Thus inhibiting LNv activity by activating DN1s increases Pura transcription in LNvs. These data support the conclusion that Pura transcription can be activated by transient hyperpolarization.
Expression of Hr38 and Pura is mutually exclusive
The opposite responses of Hr38 and Pura transcription to neuronal activity and hyperpolarization suggest that they are components of mutually exclusive gene expression programs. To test this idea, we simultaneously measured Hr38 and Pura transcription in a single cell. We converted Hr38-Tomato to Hr38-GFP by replacing tdTomato with GFP but retaining the PEST, NLS and other sequences shown in Fig. 1A. Hr38-GFP was inserted to chromosome 3 to generate animals with both Hr38-GFP and Pura-Tomato transgenes. The data in Fig. S2B show that Hr38-GFP levels peaked at ZT2 in LNvs and had similarly timed rhythms as Hr38-Tomato. Thus the temporal pattern of the two Hr38 reporter genes is independent of the fluorescent protein used and of the genomic insertion site.
Next we measured the levels of Hr38-GFP and Pura-Tomato in LNvs at different times of day (Fig. 2D and Fig. S2C). The data show that an individual LNv has one of three patterns of gene expression: (i) high Pura-Tomato and low Hr38-GFP levels – most cells at ZT18 (blue dots); (ii) low levels of both reporter genes – most cells at ZT0 (black dots); or (iii) high Hr38-GFP and low Pura-Tomato levels – many cells at ZT2 (orange dots in graph on right). LNvs at ZT12 have low Tomato expression and either intermediate or low levels of GFP (Fig. S2D) but were excluded from the graph in Fig. 2D for visual simplicity. The inability to detect LNvs with high levels of both reporter genes supports the idea that Hr38 and Pura are parts of mutually exclusive transcriptional programs.
Given that Hr38 and Pura transcription is activated by activity and hyperpolarization respectively (Fig. 1C, Fig. 2A-C), it seems likely that the three patterns of reporter expression in LNvs represent three different states of LNv excitability: hyperpolarized (ZT18), intermediate or resting (ZT0) and excited (ZT2).
Increases in neuronal activity have been measured via immediate early genes such as c-fos and also by Calcium sensors such as GCaMP (Morgan et al., 1987; Wang et al., 2003). However, it has been more challenging to detect hyperpolarization or decreases in neuronal activity. The data presented here suggest that transcriptional reporter genes will help distinguish between neurons at rest with low activity and hyperpolarized neurons. In addition, the Pura reporter gene could be useful to identify neurons undergoing structural plasticity.
Hr38 and Pura respond in opposite directions to activity-dependent transcription factors Creb and Mef2
Mammalian transcription factors such as CREB and Mef2 are post-translationally modified by neuronal firing to activate transcription (Lyons and West, 2011; West and Greenberg, 2011). Indeed, both CREB and Mef2 regulate Nr4a1, the mammalian Hr38 orthologue, in neurons (Fass et al., 2003; Lam et al., 2010). We found three TGACGTCA CREB response elements (CRE) and two AT-enriched Mef2 binding sites clustered within 780bp of the 4kb Hr38 enhancer that we used to construct the Hr38-Tomato reporter gene. Thus CREB and Mef2 were candidates to link neuronal activity to Hr38 transcription.
To test if CREB can regulate Hr38 expression, we used a CrebB repressor isoform [UAS-CrebB repressor (Yin et al., 1994)] since there are concerns with the available UAS-CrebB activator transgenes (Perazzona et al., 2004). We assayed Hr38-Tomato levels in LNvs across the day and found that constitutive expression of CrebB repressor with Pdf-Gal4 reduced the peak of Hr38-Tomato at ZT2 by 2-fold, although there was no statistical difference from control LNvs at other times (Fig. 3A). A synthetic reporter gene with 4 copies of a CRE gave a similar temporal pattern to Hr38-Tomato in LNvs (Fig. S3A-B) and its expression was also reduced by expressing CrebB repressor in LNvs with Pdf-Gal4 (Fig. S3B). To test if Mef2 regulates Hr38 transcription, we used Pdf-Gal4 to overexpress Mef2 in LNvs. The data in Fig. 3A show that Hr38-Tomato levels were 1.5-fold higher in Mef2-overexpressing LNvs than in control LNvs at ZT0, but not significantly different at other times. Thus Mef2 and Creb contribute to the normal profile of Hr38 expression, but there must be additional factors involved.
Performing the same experiments with Pura-Tomato, we found that Pura-Tomato levels in LNvs were slightly higher in CrebB repressor-expressing LNvs than GFP-expressing controls at ZT12, but not different from control LNvs at other times (Fig. 3B). However, Pura-Tomato levels were at least 2-fold lower in Mef2-overexpressing LNvs than in control LNvs at all times of day (Fig. 3B). Repression of Pura by Mef2 is consistent with (Sivachenko et al., 2013) who showed that over-expressing Mef2 blocked retraction of s-LNv projections and left them in a constitutively expanded state. The novel responses of Pura to neuronal activity (Fig. 1 & 2) and to Mef2 (Fig. 3B) led us to study Pura regulation in more detail.
Dissecting the Pura enhancer to identify cis-regulatory elements
We first sought to find the minimal Pura enhancer that responds to changes in neuronal activity in LNvs in order to identify cis-regulatory elements. The 1.5kb Pura-B enhancer in Fig. 1B contains the likely transcription start site for Pura transcript B and our deletion constructs all used 100bp surrounding this transcription start site as a basal promoter with varying lengths of upstream DNA. The different constructs were inserted into the same chromosomal location as Pura-Tomato B and assayed in larval LNvs.
To identify the minimal Pura enhancer that retains the spatial and temporal expression pattern of the full-length enhancer in LNvs, larvae were entrained in LD and Pura-Tomato levels measured at ZT18, the time of peak expression (Fig. 1B). Fig. 4A and Fig. S4A show a series of deletions of the Pura enhancer with purple or blue indicating expression or lack of detectable expression in LNvs, respectively. The smallest transgene that expressed in LNvs was Pura-Tomato P, with only 116bp upstream of regulatory DNA upstream of the basal promoter. Furthermore, Pura-Tomato P has the same temporal pattern in LNvs as Pura-Tomato A (Fig. 4B, Fig. S4B).
We then tested if Pura-Tomato P responds to neuronal activity and hyperpolarization like the larger constructs. We used Pdf-Gal4 to express either UAS-NCB, UAS-Kir2.1 or UAS-GFP and measured Pura-Tomato A and Pura-Tomato P levels in LNvs at ZT12. The data in Fig. 4C show that Pura-Tomato P was induced by Kir2.1 and repressed by NCB just like Pura-Tomato A. Thus we focused on Pura-Tomato P to identify cis-regulatory elements that respond to neuronal activity.
Pura is repressed by activity-regulated transcription factors
The 116bp Pura enhancer is highly conserved across 23 Drosophila species (Siepel et al., 2005), although this is partly because this region overlaps with a protein coding exon from Pura RNA isoforms A and D (Fig. 4A). To predict cis-regulatory elements in the Pura enhancer P, we used MATCH analysis from TRANSFAC (Kel et al., 2003) and focused on transcription factors known to be regulated by neuronal activity. Our analysis identified one potential Mef2 binding site, two potential Egr-1 binding sites and three potential Atf3 binding sites in Pura P (Fig. 4D). These sites are conserved in D. pseudoobscura and D. virilis (Fig. S4C).
While Mef2 is post-translationally modified by neuronal activity (Flavell et al., 2006), expression of Egr-1 and Atf3 are rapidly induced by neuronal activity in mammals (Milbrandt, 1987; Zhang et al., 2011). Expression of the fly Egr-1 homolog – stripe (sr) – is rapidly induced by neuronal activity in LNvs (Chen et al., 2016) and sr mRNA is 3.6-fold higher at CT3, when LNvs are more excitable, than at CT15 (Mizrak et al., 2012). Similarly, Atf3 mRNA levels in LNvs are 3.1-fold higher at CT3 than CT15 (Mizrak et al., 2012).
We performed gain- and loss-of-function experiments to test if these transcription factors regulate the minimal Pura enhancer. We used Pdf-Gal4 to target either overexpression or RNAi transgenes to LNvs and measured Pura-Tomato P levels at ZT12. The data in Fig. 4E show that Pura-Tomato P levels in LNvs were reduced by over-expressing Mef2, Sr or Atf3. Over-expressing Atf3 gave the strongest repression, which is consistent with the number of potential Atf3 binding sites (Fig. 4D). We also found that Pura-Tomato P levels increased when RNAi targeting any one of these transcription factors was expressed in LNvs (Fig. 4E). Thus we conclude that Mef2, Sr and Atf3 all repress Pura transcription and their effects are likely direct given the predicted binding sites in the minimal Pura enhancer.
Mammalian Mef2 is post-translationally modified to rapidly regulate gene expression in response to increased neuronal activity (Flavell et al., 2006). If the same is true in LNvs, neuronal activity could rapidly activate fly Mef2 to repress transcription of Pura and co-regulated genes. With a short delay, the induction of activity-regulated genes such as Sr would cement Pura repression for longer – with the duration of repression presumably depending on the intensity of neuronal activity, the resulting levels of the activity-induced repressors and their intrinsic half-lives.
An E-box in the Pura enhancer is not essential for transcriptional activation by hyperpolarization
One sequence that stood out as a potential binding site for a transcriptional activator in the minimal Pura enhancer was a CACGTG E-box. This sequence can be directly bound by a dimer of Clock (Clk) and Cycle (Cyc), two core clock components that are bHLH transcription factors. Clk and Cyc drive circadian rhythms in the expression of many genes by interacting with the rhythmically produced repressor protein Period that regulates Clk/Cyc DNA-binding (reviewed by (Hardin, 2011).
To test the importance of this E-box in the response to hyperpolarization, we mutated its core to CAGCTG as in (Darlington et al., 1998) and called this reporter gene Pura-Tomato P mE for mutant E-box. The data in Fig. 4F show that Pura-Tomato P mE levels were 10-fold lower than Pura-Tomato P levels in LNvs at ZT12. To test whether the E-box is necessary for the response to hyperpolarization, we used Pdf-Gal4 to constitutively express Kir2.1 and compared levels of Pura-Tomato P mE at ZT12 with control LNvs expressing a GFP transgene. We found that levels of Pura-Tomato P mE were 3.5-fold higher in Kir2.1-expressing LNvs than in control LNvs (Fig. 4G). Thus the E-box in the minimal Pura enhancer affects Pura expression but is not necessary for transcriptional activation by hyperpolarization. We did not test if increased neuronal activity represses Pura-Tomato P mE because expression of this reporter gene was too low in control LNvs to reliably measure potential decreases.
Pura is activated by the transcription factor Toy
As a complementary approach to identifying transcription factors regulating Pura, we performed an RNAi screen in LNvs. We focused on a set of 110 transcription factors we identified that are expressed in fly heads (Adryan and Teichmann, 2006; Celniker et al., 2009; Hughes et al., 2012) and for which there were available short hairpin RNA (shRNA) lines in a common genetic background (Ni et al., 2011). We used Pdf-Gal4 to express an shRNA transgene against each transcription factor and measured Pura-Tomato B reporter levels in LNvs at its peak of ZT18. Only three lines changed Pura-Tomato B levels by more than two standard deviations from the mean of control LNvs (Fig. 5A, Table S1). Two of these lines increased Pura-Tomato B levels by 1.7-fold – these targeted trachealess and pointed. The third line reduced Pura-Tomato B and it stood out since it affected reporter gene levels so dramatically: shRNA targeting twin-of-eyeless (toy) reduced Tomato levels 5.5-fold compared to control LNvs, with the next strongest reduction of Pura-Tomato B only 1.4-fold (Fig. 5A).
toy encodes one of the two Drosophila Pax-6 transcription factors. These proteins have two separate DNA binding domains: a Paired domain and a Homeodomain (Treisman et al., 1991). toy is best known as a master regulator of eye development (Czerny et al., 1999) and is also involved in mushroom body development (Kurusu et al., 2000). However, toy is also expressed in various neurons in adult flies, including LNvs (Glossop et al., 2014) and mushroom body neurons (Henry et al., 2012), both of which are sites of structural plasticity (Fernandez et al., 2008; Heisenberg et al., 1995).
To test if Toy regulates the minimal 116bp Pura enhancer, we used Pdf-Gal4 to express UAS-toy (Czerny et al., 1999) for overexpression or UAS-toy-shRNA or UAS-toy-RNAi transgenes to reduce expression. toy-shRNA and toy-RNAi target different regions of toy. The results in Fig. 5B show that Pura-Tomato P levels at ZT12 in LNvs were 2.8-fold higher in toy-overexpressing LNvs and 4- or 5-fold lower with toy-shRNA and toy-RNAi. Thus we conclude that Toy regulates Pura.
We searched for Pax-6 binding sites in the minimal Pura enhancer and found one Pax-6 binding site that overlaps with two of the Egr-1 sites and the Mef2 site (Fig. 4D). To test its function, we mutated this Pax-6 site in the context of Pura-Tomato P – see STAR methods for details. This mutation also affects the predicted Egr-1 and Mef2 binding sites. Levels of the resulting construct (Pura-Tomato P mT) were almost undetectable in LNvs even at the normal peak of Pura expression at ZT18 (Fig. 5C). Thus this Pax-6 site is required for Pura expression and leads us to propose that Toy directly regulates Pura expression. Pura-Tomato P mT could not be activated with Pdf-Gal4 expressing UAS-Kir2.1 in LNvs (Fig. S5). However, the very low levels of Pura-Tomato P mT expression mean that we cannot make a strong conclusion that this Pax-6 site is required for induction by hyperpolarization.
toy transcription is activated by hyperpolarization
Since the repressors that regulate Pura expression are likely regulated by post-translational modification (Mef2) or by gene expression (Sr, Atf3), we wanted to test if toy also responds to neuronal activity. We first measured Toy protein levels in LNvs at different times of day and found that Toy levels change over 24hr in LNvs (Fig. 5D). Toy protein levels peak at ZT18, in phase with Pura transcription (Fig. 1B). Next we tested if Toy levels are regulated by neuronal activity. We measured Toy protein levels at ZT12 in LNvs with Pdf-Gal4 expressing Kir2.1, NCB or GFP as a control. The data in Fig. 5E show that Toy levels were 1.5-fold higher in Kir2.1-expressing LNvs and 1.2-fold lower in NCB-expressing LNvs than control LNvs. Thus Toy protein levels are slightly altered by neuronal activity.
To test if toy is also transcriptionally regulated in LNvs, we sought to find the relevant enhancer for LNv expression. Two overlapping toy enhancers (R9G08 and R9G09) appear to drive expression in LNvs (Jenett et al., 2012). This 1.1kb overlapping region is highly conserved across 23 Drosophila species. We generated a toy transcriptional reporter by fusing the toy enhancer R9G08 plus 87bp from R9G09 to an Hsp70 basal promoter and destabilized tdTomato. The data in Fig. 5F show that toy-Tomato levels change over 24hr in LNvs. toy-Tomato levels increased 2.5-fold between ZT12 and ZT18, when LNvs become more hyperpolarized (Cao and Nitabach, 2008).
We hypothesized that the increase of toy-Tomato from ZT12 to ZT18 is driven by hyperpolarization. To test this idea, we measured toy-Tomato levels at ZT12 in LNvs expressing Kir2.1. The data in Fig. 5G show that toy-Tomato levels were 5.8-fold higher in Kir2.1-expressing LNvs than in control LNvs. However, there was no difference in toy-Tomato levels between NCB-expressing and control LNvs at ZT12. Thus we conclude that toy transcription is induced by hyperpolarization.
Thus we propose that hyperpolarization induces toy transcription and the resulting Toy protein then activates Pura. This indirect activation of Pura by hyperpolarization presumably adds a time delay between hyperpolarization and Pura transcription. We speculate that this delay is important and only allows Pura expression in response to prolonged hyperpolarization. If a neuron started firing again after a relatively short period of hyperpolarization, then this would induce activity-dependent transcription repressors such as Sr and Mef2, which likely compete with Toy for binding to the Pura enhancer since the binding sites for Sr and Mef2 overlap the Toy binding site (Fig. 4D). Thus we propose that this mechanism allows the Pura enhancer to report the recent electrical history of an individual LNv – as seen in Fig. 2D.
Discussion
Transcription of genes such as c-fos is rapidly induced in mammalian neurons in response to increased neuronal activity. Many of these immediate early genes encode transcription factors that regulate a second set of genes and overall these two waves of gene expression stabilize changes in the excitability and connectivity of neurons in response to firing (West and Greenberg, 2011). Here we describe a complementary gene expression program that is activated in response to hyperpolarization or neuronal inactivity. Using Drosophila LNvs as model neurons, we showed that transcription of the gene encoding the Pax-6 transcription factor Toy is induced by hyperpolarization. Toy then activates transcription of Pura, a “plasticity gene” which encodes a Rho1 GEF that increases Rho1 activity and helps retract s-LNv projections to reduce their signaling potency (Fig. 6) (Petsakou et al., 2015). Although many of our experiments used the mammalian Kir2.1 channel to induce hyperpolarization, the Pura transcriptional reporter normally shows a 24hr rhythm in LNvs and can also be induced via an inhibitory neural circuit.
If one function of the gene expression program induced by neuronal inactivity is to retract projections, then it makes sense that the activity-induced gene expression program would need to block a pre-existing program with an opposite effect on plasticity. Indeed, the rapid induction of repressors such as Egr-1 / Sr in response to neuronal activity in both mammals and flies (Milbrandt, 1987; Chen et al., 2016) suggests that repressing an alternative gene expression program is a conserved feature of how neurons respond to firing. Data supporting this idea comes from studying genes whose expression is altered by singing in the zebra finch brain: singing induced the expression of hundreds of genes and simultaneously repressed hundreds of different genes (Whitney et al., 2014). It would be interesting to know if the genes repressed by singing are induced by hyperpolarization or neuronal inactivity. Intriguingly, blocking neuronal activity in cultured mouse cortical neurons induced the expression of many genes (Schaukowitch et al., 2017), suggesting that a gene expression program that responds to neuronal inactivity exists in mammals.
We found that an individual LNv shifts from expressing high levels of Pura-Tomato to expressing Hr38-GFP and repressing Pura-Tomato within 8 hours and we did not find any LNvs that expressed high levels of both Hr38-GFP and Pura-Tomato. Thus the activity- and inactivity-regulated gene expression programs are coordinated to be mutually exclusive, which fits with their likely biological functions. Mechanistically, the minimal Pura enhancer has overlapping binding sites for Sr and Toy. This offers a potential mechanism for how the Pura enhancer can integrate the recent activity of LNvs and decide whether or not to activate Pura transcription.
There are many genes whose transcription increases in response to firing. So far, we only know of 2 genes – toy and Pura – whose transcription is induced by hyperpolarization. While it is probably premature to call this a gene expression program, the induction of the transcription factor Toy as a putative first responder makes it likely that it will have multiple target genes. Furthermore, we previously identified many genes whose expression was activated by hyperpolarizing LNvs (Mizrak et al., 2012). However, we do not yet know if these genes are regulated transcriptionally and whether they are targets of Toy.
It will be interesting to know if inactivity-driven gene expression is a general property of neurons. Intriguingly, toy expression persists in adult mushroom body neurons, indicative of a function in mature neurons and Pax6 is present in cortical neurons, a major site of plasticity in the mammalian brain (Zhang et al., 2014).
The intracellular signaling pathways that are activated by neuronal firing and lead to induction of gene expression are well-understood: increases in Ca2+ and cyclic AMP activate kinase and phosphatase pathways that post-translationally activate pre-existing transcription factors such as Creb and Mef2 (Flavell et al., 2006; Gonzalez et al., 1989). However, we do not yet know how toy transcription is activated in response to neuronal inactivity or hyperpolarization. One possibility is that a novel signaling pathway is activated by hyperpolarization. However, we consider it more likely that low levels of a second messenger such as Ca2+ and cyclic AMP activate toy transcription. This idea is appealing for two reasons. First, it fits with the observation of mutually exclusive Pura and Hr38 gene expression: if the same signaling molecule triggers both gene expression programs but at very different concentrations, then the two programs would be coordinated since it would be impossible for the signal to simultaneously be at both high and low concentrations. Second, it seems likely to take longer for a signal to decay back to basal levels than for a signal to arise in the first place. While it may be advantageous for a neuron to rapidly respond to firing to strengthen synapses, we believe that a neuron should only start to induce a synaptic retraction or pruning program with slow kinetics.
One situation in which there may be prolonged low levels of neuronal activity in the brain is sleep. Synaptic downscaling has been reported to reduce the overall connectivity of the brain (de Vivo et al., 2017; Diering et al., 2017). We speculate that a gene expression program activated by neuronal inactivity could sweep through parts of the brain during sleep and restore the brain to a similar level of connectivity as on waking the previous day. However, strong memories would persist either due to the intensity of firing in these neurons the previous day that makes it harder to remove the activity-dependent program during sleep or perhaps even because those memories are replayed during sleep – and therefore the neurons reactivated – as seen in rats after learning a spatial navigation task (Lee and Wilson, 2002). This could even explain why sensory inputs are damped during sleep and why sleep typically lasts for so many hours.
Author contribution
ZZ and JB conceived and designed the study. ZZ performed all experiments and data analysis except that TSO stained Toy in wild type LNvs. SM and SK provided the codon-optimized tdTomato plasmid prior to publication. ZZ and JB interpreted the results and wrote the manuscript. JB supervised the project and provided funding.
Methods
1. Cloning strategies
We modified pBPGUw (Pfeiffer et al., 2008), Addgene) to create a vector to make fluorescent reporter genes for different enhancers. pBPGUw contains a Gateway cassette to easily insert enhancers through the LR reaction upstream of the synthetic basal promoter DSCP. pBPGUw also contains a PhiC31 attB sequence for site-specific genome insertion and a mini white+ gene to select transformants. We PCR-amplified DNA for codon-optimized tdTomato, 3 copies of nuclear localization sequence, a PEST sequence and a SV40 3’UTR (Mezan et al., 2016). PCR products were TOPO cloned to the pCRXL-TOPO vector (Life Technologies), digested out with KpnI and SpeI. This insert was ligated to KpnI / SpeI-digested pBPGUw which removes the Gal4 sequence. We called the resulting plasmid pZT.
Since Pfeiffer et al. (Pfeiffer et al., 2010) and our preliminary data showed that the DSCP basal promoter is relatively weak, we replaced DSCP with the Hsp70 basal promoter. We also added an intervening sequence (IVS) and a synthetic 5’UTR (Syn21) to increase translational efficiency as in (Pfeiffer et al., 2010; Pfeiffer et al., 2012). Since there were no restriction sites available for direct cloning, a DNA fragment containing the Hsp70 basal promoter, IVS, Syn21 5’UTR and part of tdTomato cDNA was synthesized by Integrated DNA Technologies using the sequences from (Pfeiffer et al., 2010; Pfeiffer et al., 2012). This DNA fragment was flanked by FseI and SbfI sites. The fragment was amplified by PCR, TOPO cloned to pCR8GWTOPO, and digested by FseI and SbfI to remove it from the vector. This fragment was inserted into pZT digested with FseI and SbfI to create pZT2.0. The DNA sequence was verified by Sanger sequencing (Genewiz).
For Hr38-Tomato, we used PCR to amplify the 4kb of Hr38 regulatory DNA in Hr38GMR36B12-Gal4 (Jenett et al., 2012) from fly genomic DNA. This DNA was TOPO cloned to pCR8GWTOPO vector to create pCR8GWTOPO-Hr38 enhancer. Then we performed an LR reaction (GatewayTM LR ClonaseTM II Enzyme mix, Thermo Scientific) to insert the Hr38 enhancer into pZT2.0 to make Hr38-Tomato.
For Hr38-GFP, we modified pZT2.0 by replacing the tdTomato sequence with a codon optimized GFP (Lee and Luo, 1999). We used PCR to amplify GFP from pJFRC81-10XUAS-IVS-Syn21-GFP-p10 (JFRC, Addgene) and added an AgeI site to the reverse primer. The PCR products were TOPO cloned to pCR8GWTOPO vector. The resulting plasmid was digested with KpnI and AgeI and inserted into pZT2.0 digested with KpnI and AgeI to remove tdTomato sequences. The resulting plasmid (pZT2.5) contains the same elements as pZT2.0 except GFP instead of tdTomato. Then we used an LR reaction to insert 4kb of Hr38 regulatory DNA from pCR8GWTOPO-Hr38 enhancer to pZT2.5 to create Hr38-GFP.
All Pura-Tomato transgenes were driven by an endogenous Pura basal promoter (−50bp to +50bp of the transcription start site of Pura isoform B). We modified pZT2.0 to either remove the HSP70 basal promoter (pZT2.3) or replace it with the Pura basal promoter (pZT2.4). To make pZT2.3, we digested pZT2.0 with FseI and BglII and ligated a 22bp synthetic fragment with FseI and BglII sticky ends. Then we used PCR to amplify the Pura enhancers A, B, B1, B2, C, D and E from fly genomic DNA. These fragments contained the Pura basal promoter. They were cloned into pCR8GWTOPO and LR reactions were used to insert these enhancers into pZT2.3 to generate Pura-Tomato transgenes.
pZT2.4 contains the Pura basal promoter in the backbone. We synthesized an 114bp fragment containing the Pura basal promoter with an FseI site and a BglII site on the 5’end and 3’ end respectively. We cloned the 114bp basal promoter fragment into pCR8GWTOPO and cut with FseI and BglII to make the insertion. We also cut pZT2.0 with FseI and BglII to create the backbone and ligated it with the Pura basal promoter insertion. This generated plasmid pZT2.4. We used PCR to amplify Pura enhancer fragments F, G, H, I, J, K, L, M, N and P from fly genomic DNA, which do not include the Pura basal promoter. Then we used TOPO cloning to insert each one to pCR8GWTOPO and used LR reactions to insert these enhancers into pZT2.4 to generate Pura-Tomato transgenes.
To make Pura-Tomato P mE, we changed the E-box sequence CACGTG in Pura enhancer P to CAGCTG (mutations in bold) according to (Darlington et al., 1998) and synthesized the mutated 116bp DNA sequence (Integrated DNA Technologies). We cloned the synthesized DNA fragment into the pCR8GWTOPO vector to create pCR8GWTOPO-Pura enhancer P mE and then performed an LR reaction to insert Pura enhancer P mE to pZT2.4 to generate Pura-Tomato P mE.
To make Pura-Tomato PmT, we mutated the sequence TCTACTGGGCACGCACGC in Pura enhancer P that contains the Toy binding site to TTTCCGGAGTATGTATGT (mutations in bold) and synthesized the mutated 116bp DNA sequence (Integrated DNA Technologies). We cloned the synthesized DNA fragment into the pCR8GWTOPO vector to create pCR8GWTOPO-Pura enhancer P mT and then performed an LR reaction to insert Pura enhancer P mT to pZT2.4 to generate Pura-Tomato P mT.
For toy-Tomato, we used PCR to amplify a 3.6kb toy enhancer DNA fragment from fly genomic DNA and cloned it into the pCR8GWTOPO vector to create pCR8GWTOPO-toy enhancer. Then we performed an LR reaction to insert the toy enhancer to pZT2.0 to make toy-Tomato.
For 4X CRE-Tomato, a 143bp DNA fragment containing 4 copies of a canonical CRE element (Integrated DNA Technologies) using a sequence provided by Dr. Subhabrata Sanyal was cloned into the pCR8GWTOPO vector to create pCR8GWTOPO-4X CRE. We then performed an LR reaction to insert the synthetic 4X CRE sequence to pZT to make 4X CRE-Tomato.
2. Generating transgenic animals
All plasmids containing reporter transgenes were prepared using Qiagen Plasmid Mini Kit (Qiagen Ref No.12125). The plasmids were sent to Genetivision Corporation for injection. All reporter transgenes were inserted to a specific site using PhiC31 recombinase mediated methods (Groth et al., 2004). All transgenes were inserted to VK37: (2L)22A3 (Venken et al., 2006) except for Hr38-GFP which was inserted to attP2: (3L)68A4 (Groth et al., 2004). Since all of the transgenes contain a mini white+selection marker, plasmids were injected to white loss-of-function mutant flies to make transgenic lines and crossed to white loss-of-function mutant flies to screen for white+ transgenic flies. Red eyed flies were balanced to make stock lines.
3. Immunofluorescence staining
Flies were developed at 25°C in LD cycles unless specifically noted. Brains of Drosophila 3rd instar wandering larvae were dissected at specific times of day in phosphate-buffered saline (PBS) and immediately fixed in fresh 4% formaldehyde in PBS for 20 min at room temperature. Brains were then permeabilized by washing twice in PBS with 1% Triton X-100 for 10 min and then blocked in PBS with 0.5% Triton X-100 and 10% normal horse serum (VECTOR S-2000) for at least 30 min at room temperature. Brains were incubated in PBS with 0.5% Triton X-100 and 10% normal horse serum and primary antibodies overnight at 4°C. Brains were rinsed once and then washed 10 min for 3 times in PBS with 0.5% Triton X-100 before incubating with secondary antibodies in PBS with 0.5% Triton X-100 and 10% normal horse serum either at room temperature for 1 hr or overnight at 4°C. Brains were rinsed once and then washed 10 min for 3 times in PBS with 0.5% Triton X-100. Brains were soaked in 50% glycerol in PBS or SlowFade Gold antifade reagent (Invitrogen S36936) for 20 min before mounting.
Antibody list:
Primary antibodies:
PDF: “PDF C7”, Developmental Studies Hybridoma Bank, mouse, used at 1:100 RFP: “PM005”, MBL, rabbit, used at 1:500
Toy: gift from Claude Desplan, previously unpublished, rabbit, used at 1:500 GFP: “NB100-62622”, Novus Biologicals, sheep, used at 1:500
Secondary antibodies:
Alexa Fluro 488 donkey anti mouse (Life Technologies, A21202)
Alexa Fluro 488 donkey anti sheep (Life Technologies, A11015)
Alexa Fluro 555 donkey anti rabbit (Life Technologies, A31572)
Alexa Fluro 647 donkey anti rabbit (Life Technologies, A31573)
Alexa Fluro 647 donkey anti mouse (Life Technologies, A31571)
All used at 1:500.
4. Confocal Microscopy and Image Analysis
Brains were scanned as single stack images with 1µM in depth (1024 × 1024 or 1024 × 256 resolution) under the 20X objective lens of Leica SP5 confocal microscope with 2X or 4X digital zoom in. For quantification, a stack of images (usually 10-13 images) was projected together on the Z-axis. Mean fluorescence intensity in the soma of single neurons were quantified using FIJI with background subtracted. Background levels were determined by the average of mean fluorescence intensity of 10 regions with the same size randomly selected in the same brain lobe.
5. Bioinformatics analysis to search for transcription factor binding sites
We first used JASPAR CORE Vertebrata [jaspar.genereg.net, (Mathelier et al., 2016)] with default settings to predict CREB and Mef2 binding sites in the Hr38 and Pura enhancers. We chose the database of vertebrate transcription factors binding sites because it includes more transcription factors (1045 for vertebrates vs 266 for insects). Then we used MATCH analysis from TRANSFAC (Kel et al., 2003) with default settings to identify potential CREB, Mef2, Egr1, Atf3 and Pax6 binding sites as well as E-boxes. We used the vertebrate transcription factor PWMs for the same reason as above. In addition, the vertebrate transcription factor PWM database is better annotated for Pax6 as it includes the Paired domain binding sequence as well as the Homeodomain binding sequence – in contrast, the insect PMW only has the Homeodomain binding sequence.
6. Conservation analysis
We used the bioinformatics tool phastCons analysis in the UCSC genome browser (Siepel et al., 2005) to identify the DNA sequences in the Pura enhancer P that are conserved in D. pseudoobscura and D. virilis. Then we used MATCH analysis to predict transcription factor binding sites in these regions in D. pseudoobscura and D. virilis as above.
7. Statistics
All data points were normalized to the mean of the control group unless specified in figure legends. Error bars in all figures show mean ± SEM. Sample size (n) of all experiments is indicated in figure legends. All experiments with LNvs measured at least 15-20 single cells from 5 individual brains were quantified in each of 2 independent experiments. Student’s t test was used to test for significant difference (p < 0.05) vs. control groups in most experiments. One-way ANOVA was performed to determine significant difference (p < 0.05) for rhythmic expression levels.
Acknowledgements
We are indebted to numerous stock centers for sharing flies, antibodies and plasmids: the HHMI Janelia Farm Research Campus, the Vienna Drosophila Resource Center (VDRC), the Bloomington Drosophila Stock Center (supported by NIH P40OD018537), the Developmental Studies Hybridoma Bank (created by the NICHD and maintained at The University of Iowa), the TRiP stock center at Harvard Medical School (supported by NIGMS, NIH) and FlyORF (Bischof et al., 2013). We thank Claude Desplan for Toy antibodies, Gerry Rubin’s lab for plasmids, Jia Ling for advice on basal promoters, Jens Rister for advice on generating mutants, Matthieu Cavey for advice on confocal microscopy and Chris Hackley for advice on cloning. We thank Esteban Mazzoni, Tom Carew, Adam Carter, Claude Desplan, Erich Jarvis and Niels Ringstad for advice and support and Claude Desplan, Simon Kidd, Alison Ehrlich and Kyla Hamling for comments on the manuscript. We are indebted to Dogukan Mizrak and Marc Ruben for deciding to profile LNvs in altered electrical states and to Subhabrata Sanyal for sharing his activity reporter genes. This investigation was conducted in facilities constructed with support from Research Facilities Improvement Grant Number C06 RR-15518-01 from the NIH National Center for Research Resources. Imaging was performed at the NYU Center for Genomics & Systems Biology. ZZ was partly supported by the MacCracken Program and a James Arthur Dissertation Fellowship from NYU’s Graduate School of Arts and Science. Work in the authors’ laboratories is supported by BSF grant 2015506 (SK and JB) and NIH grant GM063911 (JB).
Footnotes
↵5 Lead contact