Histone H3K23A mutation affects courtship learning in Drosophila
To identify the causative roles of individual histone modifications in learning and memory, various Drosophila strains with UAS-histone H3 lysine-to-alanine mutations were constructed. These mutant strains overexpressed histone H3 with specific lysine residues changed to alanine residues. First, to validate whether the overexpressed histone H3 mutants, which were fused with GFP, were incorporated into the chromatin, we examined the expression of GFP on polytene chromosomes from third-instar larvae of the histone mutant lines via immunofluorescence assays. We found that GFP was indeed expressed in the polytene chromosomes (Additional file 1), suggesting that these histone mutants were also incorporated into the chromatin. Next, we performed a courtship-suppressing experiment with an OK107-GAL4 driver, which drives histone mutation expression specifically in the mushroom body (the key tissue for learning and memory in Drosophila). Unsuccessful courtship reduces the subsequent courtship behaviour of male flies.[31] This experiment is one of the major paradigms used to study learning and memory in Drosophila (Fig. 1a-b). According to the calculation method, learning and memory levels are inversely related to the courtship indexes. We therefore tested the learning and memory levels of the strains overexpressing histone H3 mutants. Overexpression of H3K23A significantly impaired the courtship learning of Drosophila after training for 1 hour compared with that observed in the H3WT group (control) (Fig. 1c, Additional file 9, videos 1-2). This result indicated that the modification of the H3K23 residues may influence learning in Drosophila. In contrast, we did not observe significant changes in learning levels in Drosophila carrying other H3 mutants such as H3K4A, H3K18A, H3K37A and H3K122A (Fig. 1. e, g, i, k).
Then we asked whether the memory levels changed in these Drosophila lines. Considering the lower learning abilities of H3K23A group, we decided to extend the training time from 1 hour to 5 hours. We used another courtship learning experiment to value the learning level of H3K23A overexpression flies after a 5-hours-training. No significant difference of learning index was observed between the groups of overexpressing H3WT and H3K23A after training for 5 hours (Additional file 2, p=0.1379). Thus, we used this training method to study the memory levels of Drosophila in this article.
No obvious effect on memory was observed at 2 hours, 1 day and 6 days in the H3K23A, H3K37A, and H3K122A overexpression flies after training for 5 hours (Fig. 1d). To our surprise, some minor changes in memory ability were observed in the H3K4A group (1day, p=0.0297) and H3K18A group (6 days, p=0.0294) (Fig. 1f, h). Considering the slight but not significant changes in the learning index (Fig. 1e, g), modifications to H3K4 and H3K18 may slightly affect learning and memory in Drosophila.
In addition, no significant differences in the total courtship time were observed except for the group overexpressing the H3K37A mutant. (Additional file 3a, b, c, d and e). To determine whether the initial courtship time (initial CI) affected the learning index, we analysed the relationship between the initial CI and the learning index in these Drosophila lines. We combined all H3WT groups and bounded them by the median. No significant difference in the learning index was found between the two subgroups (Additional file 3f). The same analysis was performed with the other groups, and no significant changes were observed, except in the H3K18A mutant (Additional file 3g, h, i and k). It appeared that there was no significant relationship between the initial courtship time and the learning index.
According to the data, even though the H3K4A and H3K18A lines showed similar trends in learning and memory, these phenotypes were weak. As the H3K23A overexpression line showed a large difference in learning but not memory, it was very interesting to study how H3K23A regulates the learning behaviour. Therefore, we focused on H3K23, which might be a major site of modification to regulate learning.
Decreased H3K23ac impairs neuronal gene activation in the H3K23A overexpression line
Before investigating how H3K23A affects learning, we used a western blot experiment to test the ratio of exogenous H3. Exogenous H3 fused with GFP showed a 42-KDa band, which ran higher than the endogenous H3 band of approximately 15-KDa (Fig. 2a). At the same time, we noticed that the overexpression of both H3WT-GFP and H3K23A-GFP accounted for approximately 4 percent of the total H3 protein (Fig. 2a).
As courtship learning was impaired in the H3K23A overexpression line, we next asked whether the expression of neuronal genes related to Drosophila learning were also affected in this line. To test this, an RT-qPCR assay was conducted to determine the expression levels of neuronal genes, including voltage-gated ion channels (Sh, eag and para), ligand-gated channels (SK, Cngl and gfA), phospholipase C (norpA), and phosphodiesterase (dnc) [32,33,34,35,36,37,38,39,40]. The expression of these candidate neuronal genes, except for eag and shakB, significantly decreased in the H3K23A overexpression line. (Fig. 2b). Four housekeeping genes were tested as the additional controls at the same time. Compared to the H3WT groups, all these four genes in the H3K23A overexpression groups showed no significant differences (Additional file 4a). At the same time, compared to the H3WT groups, there were only a little change in the elav-GAL4 control groups (Fig. 2b, Additional file 4a).
Cngl and gfA can regulate the activity of calcium signalling pathways [41, 42], and calcium signalling pathways have emerged as key players in learning and memory[43,44,45]. Therefore, we used calcium imaging to monitor the Ca2+ responses induced by KCl stimulation of larval brains. Fura-red was used as a red Ca2+ indicator dye combined with GFP-tagged brain. Ca2+ signals can be presented as values (ΔR/R0) of the relative change rate in fluorescence intensity (ΔR) normalized to the baseline fluorescence rate (R0). The H3K23A mutant was driven by Elav-GAL4. H3K23A mutant larval brains showed a lower calcium response after KCl stimulation than the control group (Additional file 4b). This result demonstrates that calcium signalling pathways were impaired in the brains of the H3K23A overexpression line.
A previous study reported that acetylation occurs in Drosophila at the H3K23 residue [46], although methylation occurs at the H3 lysine 23 site in Tetrahymena and C. elegans [47]. H3K23ac activates the transcription of genes. However, H3K23 methylation (H3K23me), including H3K23me1, H3K23me2 and H3K23me3, is associated with heterochromatin. Therefore, we examined global H3K23ac and H3K23me1 levels followed the overexpression of H3WT or H3K23A. Both of H3K23ac and H3K23me1 showed no difference in either salivary glands or the brain. (Fig. 2c).
We hypothesized that these neuronal genes were regulated in a gene-specific manner. To test this, we performed a chromatin immunoprecipitation (ChIP) assay with larval brains using antibodies against H3K23ac or H3K23me1 followed by real-time PCR. The PCR primers were designed to amplify the sequences of the coding regions of the affected neuronal genes. The intergenic regions were used as controls[46]. Compared with that in H3WT group, the enrichment of H3K23ac was significantly decreased among most of these neuronal genes in the group overexpressing the H3K23A mutant (Fig. 2d), indicating that H3K23ac is involved in the regulation of gene expression. However, there were no significant differences in H3K23me1 enrichment levels in these neuronal genes (Fig. 2e). Because we lacked the ChIP-grade antibodies against H3K23me2 or H3K23me3, we could not exclude whether the possibility that these two modifications contribute to the regulation of neuronal genes expression. In summary, H3K23ac, instead of H3K23me1, seemed to play a role in regulating learning in Drosophila by activating the expression of neuronal genes.
Knockdown of dCBP expression in the Drosophila nervous system decreased H3K23ac levels and led to a defect in courtship learning ability
To further study the functions of H3K23ac post-translational modification in learning, we determined which histone acetyltransferase acetylates H3K23 in the nervous system in Drosophila. GCN5 can acetylate the H3K23 site in yeast [48] and is the major acyltransferase for two distinct histone residues, H3K9 and H3K14, in Drosophila[49], while dCBP catalyse H3K23 acetylation in flies [46, 50]. Thus, we suspected that dCBP is one of the enzymes related to H3K23ac, that regulates courtship learning. Therefore, we constructed transgenic RNAi flies with an Elav-GAL4 driver to disrupt the expression of dCBP. The RNAi efficiency was validated by RT-qPCR, and the mRNA level of dCBP was greatly reduced in larval brains with the pan-neuronal driver Elav-GAL4 (Fig. 3a). Then, we examined H3K23ac levels in polytene chromosomes from third-instar larvae of the dCBP RNAi lines via immunofluorescence assays. This RNAi line was driven by salivary gland-specific GAL4 (SG-GAL4). Indeed, the levels of H3K23ac were significantly decreased in the dCBP RNAi line (Fig. 3b), and this result was confirmed by western blot (Fig. 3c). These results suggested that dCBP can catalyse H3K23ac in Drosophila.
Furthermore, to determine whether dCBP is responsible for the acetylation of H3K23 in the nervous system, we constructed a neuron-specific dCBP knockdown fly strain with the pan-neuronal driver elav-GAL4. Consistently, dCBP depletion in the larval nervous system resulted in a striking reduction in H3K23ac levels (Fig. 3d). Thus, we concluded that dCBP is likely the histone acetyltransferase for H3K23 in the nervous system of Drosophila. Nevertheless, we could not exclude the possibility that additional epigenetic enzymes may contribute to modifications on the H3K23 site.
Next, we tested whether dCBP RNAi Drosophila exhibited a phenotype like that of the H3K23A overexpression line, which had exhibited defects in learning ability according to the courtship suppression behavioural analysis. The Ok107-GAL4 driver was used (the GFP RNAi line was used as a control). As expected, when tested in the behavioural assay, the dCBP RNAi line exhibited learning defects (Fig. 3e, Additional file 9, videos 3-4). As with overexpression of the H3K23A mutant, we did not observe obvious effects on memory (Fig. 3f). In summary, dCBP may contribute to courtship learning, likely by modulating H3K23 acetylation. Surprisingly, dCBP RNAi flies appeared to be more active at the initial time of courtship (Additional file 5:a). The initial courtship time did not affect the learning index of the GFP RNAi group but did affect that of the dCBP RNAi group. There seems to be some interaction between these two effects (Additional file 5:b-c). It indicating that dCBP RNAi might regulate other functions besides learning.
Impaired learning ability in dCBP RNAi Drosophila may due to decreased neuronal gene expression, which is regulated by the level of H3K23ac
To examine whether dCBP RNAi also affected the expression of neuronal genes related to Drosophila learning, we conducted an RT-qPCR assay to determine the expression levels of neuronal genes. The expression of candidate neuronal genes was significantly decreased in dCBP RNAi flies, except for shakB, dnc and norpA (Fig. 4a). The four housekeep genes were examined at the same time, and we observed a down-regulation in the beta-tubulin but not in gapdh1 (Fig.4b). Consistently, the expression of most neuronal genes we examined were decreased in both the dCBP RNAi and the H3K23A overexpression groups (Fig. 3b, Fig. 4a-b). It seems that dCBP and its target H3K23ac have similar functions in regulating neuronal gene expression. Thus, we hypothesized that the expression levels of these neuronal genes are related to H3K23ac levels, which is catalysed by dCBP. To test this hypothesis, we performed a ChIP assay with control (GFP RNAi) and dCBP RNAi larval brains using an antibody against H3K23ac followed by real time-PCR. Compared with those in the control group, H3K23ac enrichment levels were significantly decreased in most of these neuronal genes, except for eag (Fig. 4c), which confirmed with the global down-regulation of H3K23 level (Fig. 3c-d, Fig. 4c). Combined with the results of the previous ChIP assay, which revealed an obvious reduction in H3K23ac levels in the neuronal genes in larval brains overexpressing the H3K23A mutant (Fig. 2d), these data demonstrated that H3K23ac, catalysed by dCBP, contributes to courtship learning in Drosophila likely by regulating the expression of neuronal genes related to learning. To be noticed, the inconsistency in expression levels and enrichment levels in some genes such as shakB suggested that these genes might be regulated by other modifications beside H3K23ac.
Inhibition of dCBP in the H3K23A overexpressing strain did not aggravate the learning defect
Based on the data above, we then asked whether dCBP is one of the major enzymes that regulates learning by catalysing H3K23ac. To investigate this question, we treated flies with ICG-001, an inhibitor specific to CBP[51]. First, a western blotting assay was used to assess the H3K23ac level after treatment with ICG-001. As expected, the H3K23ac level was slightly down-regulated in larval brains after treatment with ICG-001 (Fig. 5a). Then we tested the expression levels of neuronal genes in larval brains after treatment with ICG-001. Compared to those in the control groups (DMSO), the expression levels of these neuronal genes decreased after treatment with ICG-001 (Fig. 5b). Interestingly, when we treated flies with ICG-001, we observed the down-regulation of dCBP expression (Fig. 5b), which might have an additional effect on the regulation of gene expression.
Furthermore, a courtship suppression experiment was conducted to investigate whether the effect of CBP downregulation and H3K23A mutant overexpression are additive. In this experiment, we observed learning defects in both Elav-GAL4 control group and the H3WT overexpression group treated with ICG-001 (Fig. 5c, lanes 1, 2, 4 and 5). There were no additive effects in the H3K23A overexpression group which treated with dCBP inhibitor (Fig. 5c, lanes 3 and 6). In addition, in the Ok107-GAL4 control and H3WT overexpression groups, there were no significance differences in the total courtship time after treatment with ICG-001 (Additional file 6, lanes 1, 4 and lanes 2, 5). However, there were some changes in the H3K23A overexpression group treated with ICG-001 (Additional file 6, lanes 3 and 6). At the same time, we observed a difference in the initial courtship time levels between the H3WT overexpression group and the OK107-GAL4 control group (Additional file 6, lanes 1 and 2) but not in the groups treated with ICG-001 (Additional file 6, lanes 4 and 5). Overall, the similar learning indexes of the H3K23A overexpression mutant and the groups treated with ICG-001 indicated the important role of H3K23ac, which is catalysed by dCBP.
Because the CBP has a well-established role in neuronal differentiation during development, an immunostaining assay was conducted to study whether the H3K23A mutant plays a role in mushroom body development. Both H3K23A overexpression group and dCBP knockdown group did not affect the brain size (Additional file 7a). To our surprise, the mushroom bodies in H3K23A overexpression group was not significantly different from that in the H3WT group, but knockdown of dCBP impaired the mushroom bodies (Additional file 7b-c). In addition, treatment with ICG-001 also impaired the development of mushroom bodies like knocking down dCBP (Additional file 7d). Because of the mushroom bodies were the key tissue for learning and memory in Drosophila, the defect of learning ability might be due to the impaired mushroom bodies in the dCBP knockdown flies. To further study the direct role of CBP in courtship learning, we included a courtship learning experiment of the WT flies treated by ICG-001 after eclosion. Because the treating time was much shorter, we increased the dosage of ICG-001 to 10 μM. Similar to the dCBP RNAi groups, flies treated with 10 μM ICG-001 after eclosion showed the defect in courtship learning (Additional file 8a), without obvious defect in the structure of the mushroom bodies (Additional file 8b). Above all, these data indicated that dCBP plays a role in courtship learning and mushroom body development, although H3K23ac may have a function in regulating learning instead of in mushroom body development.