Attenuation of epigenetic regulator SMARCA4 and ERK-ETS signaling suppresses aging-related dopaminergic degeneration

Parkinson’s disease (PD) is a complex disease with high heterogeneity. How complex interactions of genetic, environmental factors and aging jointly contribute to dopaminergic degeneration in PD is largely unclear. Here, we applied frequent gene co-expression analysis on human patient substantia nigra-specific microarray datasets to identify potential novel disease-related genes. In vivo Drosophila studies validated two of 32 candidate genes, a chromatin remodeling factor SMARCA4 and a biliverdin reductase BLVRA. Inhibition of SMARCA4 was able to prevent dopaminergic degeneration not only caused by overexpression of BLVRA but also in four most common Drosophila PD models. Mechanistically, aberrant SMARCA4 and BLVRA converged on elevated ERK-ETS activity, attenuation of which by either genetic or pharmacological manipulation effectively suppressed dopaminergic degeneration in vivo. Drug inhibition of MEK/ERK also mitigated mitochondrial defects in PD gene-deficient human cells. Our findings underscore the important role of epigenetic regulators and implicate a common signaling axis for therapeutic intervention in a broad range of aging-related disorders including PD.

More challengingly, most postulated novel genetic associations or risk factors await further validation.
Gene co-expression analysis allows identifying genes with similar expression patterns across a set of samples, which has facilitated identifying genes involved in certain disease pathways, new gene functions, and potential biomarkers [5,6]. In this study, we used the known PD genes as "anchors" in order to identify new PD candidate genes that are highly co-expressed with known PD genes with multiple human brain microarray datasets. Among the predicted 32 candidate genes, SWI/SNF related, matrix associated, actin dependent regulator of chromatin, subfamily A, member 4 (SMARCA4) and biliverdin reductase A (BLVRA) were further studied in vivo using Drosophila melanogaster, and their potential involvement in PD pathogenesis were confirmed.
Furthermore, we revealed a potential common aging-related pathogenic signaling pathway consisting of the chromatin-remodeling factor SMARCA4 and the ERK-ETS signaling axis, suggesting new therapeutic targets for PD and other ageing-related disorders, as three ERK-ETS inhibitors were tested for their efficacy in multiple Drosophila PD models. Our work also illustrates the high efficiency of combining bioinformatics analysis of large-scale human transcriptomic data and small-scale genetic screening using model organisms to interrogate highly heterogeneous diseases including aging-related disorders.

Frequent gene co-expression analysis
We chose seven of the most commonly known PD genes as anchor genes. They are ATP13A2, HTR2A, SCNA, LRRK2, PARK2, PARK7, and PINK1. In addition, we identified eleven gene expression datasets from NCBI Gene Expression Omnibus (GEO), which contain samples from human brain tissues, especially the substantia nigra region in which the death of dopamine contain cells leads to PD. They are: GDS2519, GDS2821, GDS3128, GDS3129, GSE19587, GSE20141, GSE20146, GSE20153, GSE20292, GSE20295, and GSE20333.
Our workflow is similar to previously described in Conference papers, with a slight modification as in the following steps: Step 1: For the i-th dataset (i= 1, 2, …, 11), compute the Pearson correlation coefficients (PCC) between every pair of genes within each dataset, and set the top five percentile of all PCC values as threshold Ti. PCC values were converted to absolute values before setting the threshold.
Step 2: For the k-th anchor gene Ak (k = 1, …, 7), denote as the set of genes in the i-th dataset whose PCC values with respect to Ak are higher than Ti. These genes are considered to have high correlation with Ak in the i-th dataset.
Step 3: For a gene Gj, its frequency of having high correlation with Ak is denoted as , where is the indicator function which is 1 if the input is TRUE and 0 otherwise.
Step 4: For each anchor gene Ak, let Pk be the set of genes with high values.
Specifically, which is the collection of all genes that have high PCC values with Ak in at least five datasets for all seven anchor genes.
The count number of each gene Gj can be derived as Finally, the selected gene networks are subjected to gene set enrichment analysis using TOPPGene (http://toppgene.cchmc.org/) and Ingenuity Pathway Analysis (IPA®, http://www.ingenuity.com).
The 32 PD candidate gene list is the result of a very stringent selection process, which involves three thresholds for gene selection. First, in each dataset only genes with top 5-percentile correlation coefficients for a specific anchor gene is selected for the next step. Second, only genes who have strong correlation (within top 5-percentile) with at least three anchor genes (out of total of seven anchor genes) are further selected as the gene list for this dataset. Third, genes that are selected in at least five datasets out of a total of 11 datasets are used in this study. Statistical analysis (based on Fisher's exact test) shows that the chance for a random gene to be selected through this process is 3.4E-10.

Drosophila Stocks and nomenclature
Fly strains were maintained with standard molasses-cornmeal-yeast food and were kept in 25°Cwhereas PD model flies were maintained at 21-25°C routinely and raised at 29°C for the experiments unless mentioned. TH-Gal4 was a gift from S. Birman [7]. Brm::GFP is the

SNP query method
We queried with gene symbol "SMARCA4" and "BLVRA" in PD gene database PDGene (http://www.pdgene.org), which described that the SNP data source as "the data" currently available on PDGene include all results pertaining to the discovery phase of the GWAS metaanalysis by [1]. This includes data on 7,782,514 genetic variants in up to 13 showing genome-wide significant association (p <5x10 -8 ) in the discovery phase with PD risk) as well as for 6 additional, previously reported GWAS signals. Details on the included datasets as well as all genotyping procedures and statistical analyses can be found in our original publication [1].
Experimental flies were sorted into individual vials at a density of 15~20 flies per vial and were transferred to fresh vials three times a week. Experimental flies were raised at 29°C. Images were processed with Adobe Photoshop and subjected to identical post-acquisition brightness/contrast effects.

Western Blot Analysis
Fly heads were homogenized with a pestle, and protein extracts were prepared with lysis buffer

Drug treatment experiment
The experimental flies were collected after eclosion, assorted into 20 flies per via and raised at

Generation of pink1 knockout HeLa cell lines using CRISPR/Cas9 gene editing
The HeLa cell line was sent to GENEWIZ, Inc. (Beijing, China) to perform authentication test.
Firstly, genomic DNA was extracted from the cell pellets. Samples together with positive and negative control were amplified using GenePrint 10 System (Promega). Then, the amplified products were processed using the ABI3730xl Genetic Analyzer. Finally, data were analyzed using GeneMapper software V.4.0 and then compared with the ATCC for reference matching. To generate pink1 knockout cell lines, CRISPR guide RNAs (gRNAs) were chosen to target exon 1 which is common to all splicing variants. Oligo nucleotides containing CRISPR target sequences previously described [10,11]. DMSO was the solvent and equivalent amount was used in parallel as the drug treatment control. More than 30 randomly selected individual cells were analyzed for each data points. Three biological replicates were performed.

Assessment of whole brain Redox state
The CM-H2 DCFDA fluorescein dye (Invitrogen, Cat# C400) and redox-sensitive GFPs (roGFPs) protein were employed to measure the whole brain ROS stress of PD Drosophila.
Measurement methods involved in were introduced from previously reported publications [12,13]. The Drosophila brains with different genetic backgrounds were live dissected and incubated with 10 μ M DCFDA for 5 min at RT if applicable, brain images were captured using a Leica TCS SP5 II confocal microscope with 488 nm excitation and 525 nm emission. Images were analyzed by image J software.
Alternatively, fly lines of tub-mito-roGFP2 or UAS-roGFP2 genotype were crossed with the PD models respectively in order to estimate the ROS level in whole brain or in the PPL1 neurons.
The aged Drosophila brains were dissected in PBS with 20 mM NEM (N-ethyl maleimide) and then were imaged with a 535nm filter, followed with excitations at 405 nm and 488 nm. Image J software were employed to analyzed the 405 nm:488 nm ratios.

Quantification and statistical analysis
Error bars represent standard deviations (S.D.) as indicated in the figure legends. Statistical analyses were performed using GraphPad Prism (GraphPad Software). Statistical analysis of differences between two groups was performed using Mann-whitney test. * indicates P < 0.05; **, P < 0.01; ***, P < 0.001; ns, not significant. Differences in means were considered statistically significant at P < 0.05.

RESUTLS Gene co-expression network analysis identified 32 novel PD-associated candidate genes
We chose seven of the most commonly known PD genes as anchor genes, namely SCNA, LRRK2, PARKIN, DJ1, PINK1, ATP13A2, and HTR2A. Eleven datasets from NCBI Gene Expression Omnibus (GEO) were used, which contained samples from human brain tissues, especially the substantia nigra region. Our workflow is illustrated in Figure 1a. A total of 32 genes were identified to have high Pearson correlation coefficients (PCC) with at least three anchor genes in at least five datasets (Table 1 and Additional file 1: Table S1). According to the gene ontology (GO) enrichment analysis, the identified candidate PD genes were highly enriched with the genes associated with age-dependent metabolic reprogramming and neural disorders [14].

Drosophila
Among the 32 candidate genes, we chose SMARCA4 (Brahma or Brm for Drosophila homologue) and BLVRA (biliverdin reductase A, dBVR for Drosophila homologue, CG9471) for further studies based on the rationales below (Additional file 1: Table S2). First, mutations of SMARCA4/Brm, a subunit of the SWI/SNF chromatin-remodeling complex that regulates higher order chromatin structure and gene expression, have been linked to multiple neurological and psychiatric disorders including autism spectrum disorders and schizophrenia [15,16]. Although expression of SMARCA4/Brm has been reported in both murine and human DA neurons by recent single cell RNA-seq profiling [17,18], there has been no functional reports of its role for DA neurons yet. Biliverdin reductases (BLVRs), together with hemeoxygenases (HOs), constitute the evolutionarily conserved enzymes in the heme metabolism, exert multiple physiological functions and have been considered as a potential biomarker for AD and mild cognitive impairment [19]. We then queried with gene symbol "SMARCA4" and "BLVRA" in PD gene database PDGene (http://www.pdgene.org), which incorporates all available SNP data pertaining to the discovery phase of the GWAS-meta-analysis [1]. We found that both SMARCA4 (Additional file 1: Table S3) and BLVRA (Additional file 1: Table S4) harbor SNPs with metaanalysis P value between 1E-4 and 0.05, which can be regarded as the potential PD risk SNPs albeit not in the top 10,000 most significant GWAS results.
To study the in vivo roles of candidate genes in PD pathogenesis, we took advantage of the Drosophila melanogaster model organism. The age-dependent progressive DA neuronal loss in the lateral protocerebral posterior 1 (PPL1) cluster was used as the neurodegenerative index ( Fig.   2a-b). We then used an available Brm::GFP reporter fly strain to examine whether Brahma is expressed in fly DA neurons [20]. Brm::GFP was first verified by its nuclear localization in Drosophila larval tissues (Additional file 1: Figure S1a-c). Whole-mount immunostaining then confirmed the expression of Brm in the fly DA neurons (Fig. 2c), consistent with a previous report [21].
Drosophila genetic manipulations were then carried out to dissect the roles of Brm and dBVR.
When Brm RNAi-mediated down-regulation [22] or ectopic supply of a dominant negative allele of Brm (Brm DN ) was induced specifically in DA neurons, no change in the number of PPL1 DA neurons was observed, neither was the overexpression of wild type Brm (Fig. 2d-e). However, pan-neuronal (elav-Gal4 driver) overexpression or down-regulation of Brm led to early lethality.
Meanwhile, we generated two UAS-dBVR transgenic fly strains, Bvr II and Bvr III, which enabled overexpression of dBVR (Additional file 1: Figure S2). Specific overexpression of dBVR in DA neurons resulted in DA neuronal loss in the aged fly brains as compared with controls, while an available dBVR RNAi line did not show any effects (Fig. 2d-e, Additional file 1: Figure   S2). Remarkably, ectopic supply of Brm DN suppressed the progressive DA loss caused by BVR overexpression (Fig. 2d-e). The rescuing effect was not owing to titration of UAS-mediated overexpression (Additional file 1: Figure S3).

PD models
We then examined how Brm and dBVR genetically interact with known PD genes. Four previously reported Drosophila PD models were successfully reproduced in our laboratory [23,24]. Given that the homozygous Parkin null alleles were found to be unhealthy in our laboratory, we used Parkin RNA interference (RNAi) flies instead. These four Drosophila PD models were abbreviated as α Syn (A30P), Lrrk2 (I1915T), Parkin, Pink1 (Pink1 mut or Pink1 RNAi) PD models respectively (see Methods & fly genotypes listed in the Additional file 1: appendix).
Consistent with previous findings, age-dependent progressive degeneration was mild but statistically significant, as evidenced by the decreased number of PPL1 DA neurons in 30-day-old flies compared with age-matched controls. In contrast, young 2-day-old PD flies displayed no DA neuronal loss (Additional file 1: Figure S4).
We found that Brm::GFP level exhibited an age-dependent progressive elevation in the PD fly brains compared with controls (Fig. 2f). We next addressed how Brm is up-regulated in the degenerative dopaminergic neurons. One possibility is there could be a link between Brm activity and oxidative stress, which has been widely believed to be a common pathogenic factor in PD. To this end, we monitored the oxidative stress indicated by the ROS fluorescent dye (DCF-DA) (Additional file 1: Figure S5) or the reduction-oxidation-sensitive GFP (roGFP) (Additional file 1: Figure S6-S7). We also evaluated the level of anti-oxidant response in the brains of PD model flies using GstD-GFP as a reporter [25] (Additional file 1: Figure S8). To our surprise, unlike Brm which was pronouncedly elevated at 15d AE during the disease progression in all four PD models, no substantial increase of brain oxidative stress reporting signal was detected until 15-20d AE in the brains of PD flies except for the α Syn detected by roGFP (Additional file 1: Figure   S6-S7). Therefore, it implicates that Brm up-regulation in PD flies might not be largely due to the change of DA neuronal oxidative stress.
Remarkably, when Brm RNAi was introduced into the four different PD model flies, significant suppression of PPL1 DA neuronal loss was detected in the aged flies ( Fig. 3a-d). As a control, an introduction of irrelevant Luc RNAi did not exhibit such inhibitory effects (Fig. 3a-d).
Accordingly, overexpression of Brm DN fully prevented the progressive PPL1 DA neuron degeneration in all four PD fly models, although overexpression of Brm in the PD context did not induce further neuronal loss ( Fig. 3a-d, Fig. 2e). The observed rescuing effects did not occur during the developmental stage, since the inhibition of Brm alone did not increase the number of DA neurons more than normal in the young flies (Fig. 2d, Additional file 1: Figure S9a). Neither was the rescuing effect due to titration of UAS-mediated overexpression (Additional file 1: Figure   S3).
On the other hand, overexpression of dBVR aggravated the progressive DA neuronal loss in aged four PD model flies, while dBVR RNAi did not mitigate the degeneration (Fig. 3e-h, Additional file 1: Figure S9b). Importantly, the aggravation caused by dBVR overexpression could be fully rescued through the addition of Brm DN in all four PD model flies, and in three PD models (except the PINK1 deficiency model) through Brm RNAi (Fig. 3e-h). The rescuing effect was not owing to titration of UAS-mediated overexpression (Additional file 1: Figure S3).
Collectively, these results demonstrated that inactivation of Brm protects DA neurons from agedependent degeneration in a variety of pathogenic genetic backgrounds.

Prolonged over-activated MEK-ERK-ETS signaling in multiple PD fly models
We then investigated the possible mechanisms through which Brm and dBVR could affect DA degeneration. An Ingenuity Pathway Analysis (IPA) revealed the connection of SMARCA4/Brm with ERK signaling pathway (Fig. 1b), as was supported by a previous study [22]. On the other hand, hBVR has been previously suggested as an ERK activator in HEK293A cells [26].
Activator Pointless (Pnt) and repressor Anterior open (Aop) are two downstream antagonizing players of the MEK-ERK signaling and both belong to E-twenty six transcription factors [27]. We used the reporter fly line of Pnt::EGFP [27] to monitor the MEK-ERK activity in the fly brain and found that both Brm RNAi and dBVR RNAi resulted in reduced MEK-ERK activity in fly DA neurons, while overexpressing either Brm or dBVR up-regulated the signaling (Fig. 3i-n).
Conversely, DA-neuron-specific knock-down of Drosophila MEK (Dsor1) by RNAi [28] or overexpression of a wild type allele of the negative regulator of ERK pathway, Aop [wt] , suppressed the aggravation of progressive DA degeneration induced by dBVR overexpression in four PD fly models (Fig. 3e-h); Aop [wt] overexpression also prevented DA degeneration caused by dBVR overexpression alone (Additional file 1: Figure S10). These rescuing effects were not due to titration of UAS-mediated overexpression (Additional file 1: Figure S3).
We then examined whether over-activation of ERK-ETS was prevalent in those common PD fly models. Remarkably, we observed increased phosphorylated ERK (pERK) levels in all four PD model fly brains (Fig. 4a-b), which were concordant with sustained up-regulation of Pnt ( Fig.   4c-e).

Drosophila PD models
We next examined whether directly reducing MEK-ERK-ETS activation was sufficient to prevent those common forms of DA degeneration. MEK RNAi fully rescued PPL1 DA neuronal loss in three 30-day-old PD model flies (Fig. 4f-I, Additional file 1: Figure S11a-b) and the rescue was not owing to titration of UAS-mediated overexpression (Additional file 1: Figure S3). In contrast, Drosophila ERK (rolled or rl) RNAi [28] exerted no apparent rescuing effects, suggesting that it was the activated fraction of ERK (pERK), not the abundance of the ERK protein, that caused the neurotoxicity. In parallel, when MEK RNAi or ERK RNAi was induced alone specifically in DA neurons, no changes in the PPL1 DA neurons were observed when compared with age-matched controls, suggesting that the rescuing effects of MEK RNAi was epistatic, but not due to simple addition (Additional file 1: Figure S11a). Overexpression of constitutively-active ERK led to early larval lethality before eclosion. We further found that the DA-neuron-specific overexpression of a constitutively active form of Pnt (Pnt [P1] ) led to early lethality before eclosion too. Nevertheless, DA-neuron-specific RNAi knock-down of Pnt [28] did not lead to a rescue or mitigation, probably because the manipulation was not potent enough to significantly disrupt the ERK activity. Remarkably, DA-neuron-specific overexpression of the negative regulator, Aop [wt] , completely blocked the aging-related PPL1 DA neuronal loss in all four PD model flies (Fig. 4j-m, Additional file 1: Figure S12a, c), and the rescue was not owing to titration of UAS-mediated overexpression (Additional file 1: Figure S3). No further aggravation was observed upon DAspecific RNAi [28] knock-down of Aop ( Fig. 4j-m, Additional file 1: Figure S11a, S11c). Similar to Pnt, DA-neuron-specific overexpression of a constitutively active form of AOP (AOP [CA] ) led to early lethality before eclosion, and overexpression of Aop [wt] or Aop RNAi alone resulted in mild DA neuronal loss (Additional file 1: Figure S11a). Based on all the above evidences, we conclude that there might be a delicate range of the ERK-ETS signaling strength that is beneficial for the maintenance of DA neurons, while deviation from that range such as prolonged overactivation could be rather detrimental.

Drosophila PD models
To examine the MEK/ERK signaling pathway as a drug target for intervening in DA degeneration, we started with the MEK1 inhibitor, U0126. The effective inhibitory dose was first determined in flies subjected to 7 days of drug feeding (Additional file 1: Figure S12a-b). After continuous oral supplementation of U0126 (10 μ g/ml) in adult flies for 30 days, DA neuronal loss in PPL1 clusters were blocked effectively in all four PD fly models (Fig. 4n). Exposure to another MEK1 inhibitor, PD0325901 (10 μ g/ml), also completely blocked PPL1 DA neuronal loss (Fig. 4o). We further tested Trametinib, another potent and highly specific MEK1 inhibitor, an FDA-approved drug for the treatment of melanoma [29]. With the optimal feeding concentration of 16 μ M of Trametinib (Additional file 1: Figure S12c-d) [28], PPL1 DA neurons were fully protected from degeneration (Fig. 4p). No global brain or motor behavioral abnormalities were detected with all these drug treatments. Taken together, our data demonstrated that the MEK-ERK pathway could be a valid drug target to revert DA degeneration and illustrated that oral administration could be a promising pharmacological intervention.

Drug inhibition of ERK activity ameliorated mitochondrial defects in pink1 -/-HeLa cells
In humans, pink1-deficiency leads to mitochondrial defects [30]. To determine whether our findings were relevant to human pathology, we generated a pink1 -/-HeLa cell line, using the CRISPR/Cas9 technology (Fig. 5a) [31]. Enhanced phosphorylation of ERK and MEK1 was observed in the pink1 -/-HeLa cells, indicating an aberrant MAPK signaling over-activation (Fig.   5b). Treatment with the MEK1 inhibitor, PD0325901, effectively mitigated multiple mitochondrial defects in the pink1 -/cells, such as reduced mitochondrial membrane potential, lowered mitochondrial contents, and abnormal network interconnectivity (Fig. 5c-d). In agreement with our results, MEK1 inhibitor has been shown to reverse PD-associated phenotypes induced by pathological LRRK2 alleles in cultured human iPS derived neurons [32].

DISCUSSION
In summary, we identified a subset of 32 novel PD-associated genes, which were highly enriched in aging and neural disorders. The roles of two candidate genes SMARCA4/Brm and BLVRA/BVR were validated in vivo. The activity of ERK-ETS signaling, as a common effector for SMARCA4/Brm and BVR, was found to be also elevated in different genetic forms of Brm was found here to be progressively up-regulated in the aging brains of PD fly models. One possibility is through a genetic imprinting response to elevated calcium level triggered by aberrant neuronal activities or calcium metabolism [33]. Alternatively, Brm could be activated through NF-kB mediated inflammatory responses that are well recognized in the development of neurodegenerative diseases [34]. Besides, there is a clue that Brm functions downstream of of Hippo pathway and plays an important role in a feedback loop between Crumb and Yorkie in this pathway [35]. On the other hand, in our unpublished data, we did observe the differential expression of other oxidoreductases (e.g., P450, sulfiredoxin, phenoloxidase) at the early-middle stage of PD. Therefore, it remains possible that Brm might directly interact with Keap1/Nrf to elicit a program of sequential cellular responses to oxidative stress in DA neurons.
Inactivation of Brm was shown here to prevent dopaminergic neurons from degeneration. One putative route was through modulating the MAPK/ERK signally activity. Accordingly, Brm was shown to directly interact with Dsor1 (MEK1) and promotes EGFR-Ras-MAPK signaling activity [22,36]. Nevertheless, the possibility could not be ruled out that direct interactions exist between Three compounds that inhibited MEK-ERK signaling also ameliorated DA degeneration in all PD fly models. Interestingly, Trametinib has been shown to extend lifespan [28], while targeted inhibition of ERK signaling prevents spinocerebellar ataxia type 1 in Drosophila and mice models [39]. To our knowledge, this manuscript represents the first report of compounds that are effective in preventing DA degeneration in vivo in the four most common genetic forms of PD, compared to a previous report of Rapamycin [40]. MEK-ERK-ETS and mTOR/4E-BP pathways might converge on common downstream effectors such as mitochondrial activity/quality control, proteostasis, autophagy, oxidative stress response, DNA damage repair, and DNA-chromatin modifications, all of which are important aspects during aging. It is conceivable that cocktail strategies using both lines of inhibitors might eventually balance symptoms mitigation and side effects.

CONCLUSIONS
By combining bioinformatics analysis of large-scale human transcriptomic data and a smallscale genetic screening using Drosophila disease models, we identified two novel PD-associated genes SMARCA4 and BLVRA and disclosed a potential common pathogenic PD pathway, i.e.