Increased isoform-specific phosphodiesterase 4D expression is associated with pathology and cognitive impairment in Alzheimer’s disease

Pharmacological phosphodiesterase 4D (PDE4D) inhibition shows therapeutic potential to restore memory function in Alzheimer's disease (AD), but will likely evoke adverse side effects. As PDE4D encodes multiple isoforms, targeting specific isoforms may improve treatment efficacy and safety. Here, we investigated whether PDE4D isoform expression and PDE4D DNA methylation is affected in AD and whether expression changes are associated with severity of pathology and cognitive impairment. In post-mortem temporal lobe brain material from AD patients (n = 42) and age-matched controls (n = 40), we measured PDE4D isoform expression and PDE4D DNA (hydroxy)methylation using quantitative polymerase chain reaction and Illumina 450k Beadarrays, respectively. Linear regression revealed increased PDE4D1, -D3, -D5, and -D8 expression in AD with concurrent (hydroxy)methylation changes in associated promoter regions. Moreover, increased PDE4D1 and -D3 expression was associated with higherplaque and tau pathology levels, higher Braak stages, and progressed cognitive impairment. Future studies should indicate functional roles of specific PDE4D isoforms and the efficacy and safety of their selective inhibition to restore memory function in AD.


Introduction
Impaired memory function in Alzheimer's disease (AD) patients has a debilitating effect on daily activities and, consequently, negatively affects the quality of life of patients, their caretakers, and family (Schneider et al., 1999). The second messenger cyclic adenosine monophosphate (cAMP) is crucial in the molecular signaling cascades underlying memory function (Kandel, 2012). Intracellular cAMP levels are mainly controlled through degradation by the phosphodiesterase 4 (PDE4) enzyme family (Houslay, 2010). Accordingly, previous studies have shown that PDE4 inhibition can improve memory consolidation processes in rodents (Vanmierlo et al., 2016). In humans, the PDE4 inhibitor roflumilast was found to enhance verbal learning in young adults and old healthy volunteers (Blokland et al., 2019;Van Duinen et al., 2018). Thus, PDE4 inhibition seems to hold clinical potential as a treatment to enhance memory functioning. The therapeutic potential in AD is supported by the finding that PDE4 inhibition can restore memory impairments induced by intra-hippocampal injections with amyloid-b in rats (Cheng et al., 2010). Moreover, in a mouse model of tauopathy, the number of tau aggregates was reduced after administration of the nonselective PDE4 inhibitor rolipram (Myeku et al., 2016). Despite these promising results, progression of new PDE4 Tim Vanmierlo, Daniel van den Hove, and Jos Prickaerts contributed equally to this work. inhibitors into the clinic is hampered by the occurrence of doselimiting side effects, mainly nausea and emesis. More selective inhibition of PDE4 subtypes or isoforms may circumvent these adverse effects. In the present study, we investigated whether expression regulation of PDE4D isoforms is altered in AD brain material to identify potential targets for more selective PDE4 inhibition. The PDE4 family consists of 4 genes (PDE4A-D) of which PDE4D seems to be mostly involved in the modulation of cognitive functions . Hence, inhibitors selective for PDE4D have been developed to investigate their cognition-enhancing potential. These new selective PDE4D inhibitors have been found to enhance memory performance in rodents, including models of AD (Ricciarelli et al., 2017;Zhang et al., 2018). The procognitive effects of PDE4D-specific inhibition are supported by PDE4D knock-down studies (Baumgartel et al., 2018). The memory-enhancing potential of PDE4D inhibition has also been tested in Phase I clinical trials (ClinicalTrials.gov Identifier: NCT02648672 and NCT02840279) and currently, preparations are under way to initiate a Phase II trial in patients with early-stage AD (ClinicalTrials.gov Identifier: NCT03817684). Notably, PDE4D inhibition is also hypothesized to induce emetic side effects given its expression in brain areas involved in emesis, that is, the area postrema, and based on gene deletion studies (Mori et al., 2010;Robichaud et al., 2002). The human PDE4D gene gives rise to multiple protein isoforms, that is, PDE4D1-9, through alternative promoters and alternative splicing. These isoforms regulate a multitude of different processes and are localized to specific compartments within the cell owing to specific amino acids in the Nterminal domain (Houslay, 2010;Houslay et al., 2007). Hence, by inhibiting specific PDE4D isoforms, it can be argued that memoryenhancing effects could be retained without evoking side effects (Schepers et al., 2019). Despite the potential of PDE4D isoform-specific inhibition as a safe and efficacious treatment strategy in AD, the role of PDE4D isoforms in the context of specific brain functions or AD has only sporadically been described (McLachlan et al., 2007;Perez-Torres et al., 2000;Ugarte et al., 2015). PDE4D isoforms showing high or increased expression in AD could be interesting targets, as these isoforms may contribute largely to cAMP hydrolysis and consequently impede memory functioning. PDE4D isoform expression is likely controlled by epigenetic mechanisms to regulate, for example, isoform-specific promoter accessibility. Previous research indicated that epigenetic modifications at the histone level influence promoter accessibility in the PDE4D gene and consequently affect isoform-specific expression in vascular smooth muscle cells (Tilley and Maurice, 2005). At the nucleotide level, epigenetic processes can reversibly influence transcription through conversion of unmethylated cytosines (UC) into methylated (5mC) or hydroxymethylated (5hmC) cytosines. Specific changes in DNA methylation and hydroxymethylation have previously been associated with AD (Smith et al., 2019). Recently, PDE4D was found those genes displaying differentially methylated in AD in an epigenome-wide association study (Lardenoije et al., 2019).
To investigate whether PDE4D isoform expression is altered in AD and whether this is associated with epigenetic changes at specific positions in the PDE4D gene, we established PDE4D expression and epigenetic profiles of human post-mortem brain material of AD patients and age-matched controls. Since AD is a progressive disease, we also explored whether changes in expression were associated with region-specific plaque and tau load, Braak stage, and the degree of cognitive impairment.

Brain material and study cohort
Informed consent was obtained from all individual participants included in the study. Human post-mortem brain material from the middle temporal gyrus (MTG) of AD patients and neurologically healthy controls was obtained from the Banner Sun Health Research Institute in Sun City, AZ. For every sample information on Mini-Mental State Examination (MMSE) score (Folstein et al., 1975), gender, age, post-mortem interval (PMI), APOE ε4 possession, amyloid-b plaque load, tau tangle load, Braak stage, and diagnosis were documented and tissue was prepared by standardized procedures (Beach et al., 2015). Braak staging was evaluated as previously described (Braak and Braak, 1991). AD diagnostic criteria followed guidelines for the National Institute on Aging-Reagan Institute criteria (Consensus Recommendations for the Postmortem Diagnosis of Alzheimer's Disease, 1997) combined with clinical diagnosis reports. Samples collected after a PMI of more than 5 hours and samples from patients with comorbidities were excluded. A total of 82 samples were included from AD patients (n ¼ 42) and age-and gender-matched neurologically healthy controls (n ¼ 40) with an average PMI of 2.8 hours. For subsequent analysis investigating associations between expression and MMSE score, cases for which cognitive performance in the MMSE was assessed longer than 36 months before death were excluded, leaving a subset of 41 cases. A detailed description of the cohort demographics can be found in Table 1.

RNA isolation, complementary DNA synthesis, and qPCR
Total RNA was isolated from frozen MTG tissue using the TRIzol Plus RNA Purification Kit (Life Technologies, Carlsbad, CA) following the manufacturer's protocol. Isolated RNA pellets were dissolved in RNAse-free water and assessed for quantity and quality (average RIN: 9.1) using a Nanodrop ND-1000 spectrophotometer (Thermo Scientific, DE) and Bioanalyzer (Agilent). Then, complementary DNA was synthesized using the Revert Aid First Strand Synthesis Kit (Thermo Scientific, Landsmeer, Netherlands) according to the manufacturer's protocol.
TaqMan-based quantitative polymerase chain reaction (qPCR) primers were used for the following PDE4D isoforms: PDE4D2, PDE4D4, PDE4D5, PDE4D6, PDE4D8, and PDE4D9 (Applied Biosystems, Foster City, CA, USA). PDE4D2 could not be detected. Primers were designed for PDE4D1, PDE4D3, and PDE4D7 as no TaqMan-based qPCR primers were available for these isoforms (SIGMA Life Science). Housekeeping gene expression of EIF4A2 and TOP1 showed the most stable expression based on geNorm criteria and was used for normalization (Vandesompele et al., 2002). Taq-Man probe and primer sequences are listed in Table 2. TaqManbased qPCR was performed using TaqMan Assay 20x and TaqMan Universal Master Mix II, with UNG 2x (Applied Biosystems) following the manufacturer's protocol. For qPCR using primers, SensiMix 2X (Bioline) was used according to the manufacturer's protocol. All reactions were performed using 6 ng complementary DNA on a LightCycler 480 (Roche).

Tissue preparation for (hydroxy)methylomic profiling
DNA was isolated from the MTG at the Banner Sun Health Research Institute. From 81 of the 82 subjects, a total of 76 mg of

Oxidative bisulfite conversion and DNA (hydroxy)methylation assay
The (oxidative) bisulfite (oxBS) conversion procedures and HumanMethylation450 BeadChip array were carried out at Serv-iceXS (ServiceXS B.V., Leiden, the Netherlands), in accordance to the manufacturer' protocols and as described previously (Lardenoije et al., 2019). Briefly, a TrueMethylTM 24 kit version 2.0 by CEGXTM (Cambridge Epigenetix Limited, Cambridge, UK) was used to convert the isolated genomic DNA (gDNA) by bisulfite (BS) or oxBS treatments to detect and localize 5mC and 5hmC marks at single-base resolution. Before conversion, gDNA quality was assessed using a PicoGreen assay (Invitrogen, Carlsbad, CA, USA) and high molecular weight gDNA was quantified by gelelectrophoresis. Upon purification and denaturation, high molecular weight gDNA (0.5 mg per treatment) was used for DNA oxidation or mock DNA oxidation treatments followed by BS conversion. DNA yield was assessed by Qubit ssDNA assay (Invitrogen) and qualitative assessment of 5hmC oxidation and BS conversion was performed using a restriction quality control. Then, 8 mL of each DNA sample was used for amplification and hybridization on the HumanMethylation450 BeadChip (Illumina, Inc., San Diego, CA, USA) according to the Illumina "Infinium II Methylation Assay Manual" protocol. Eventually, the Illumina iScan was used to scan the samples.

Processing of DNA (hydroxy)methylation assay data
For (hydroxy)methylomic profiling, data were processed and analyzed using R statistical programming language (version 3.5.3) (R Development Core Team, 2019) and RStudio (version 1.2.1335) (RStudio Team, 2019). The minfi package (version 1.28.4) was used to load raw IDAT files from the Illumina iScan (Aryee et al., 2014). Probes that cross-hybridize, locate to X and Y chromosomes and probes binding within 10 bp of a common SNP location were removed (Chen et al., 2013). Remaining probes were filtered using the "pfilter" function of the wateRmelon package (version 1.26.0) (6414 probes were removed) (Pidsley et al., 2013). Data from 397,160 remaining probes were split into 2 sets of beta values according to the oxBS and BS arrays, representing 5mC and 5mC þ 5hmC, respectively. These data were dasen normalized using the wateRmelon package (Pidsley et al., 2013). The maximum likelihood methylation levels (MLML) method was used, by means of the "MLML" function within the MLML2R package (version 0.3.2), to calculate the proportions of 5mC, 5hmC, and UC by combining the signal from the BS and oxBS arrays as input (Kiihl et al., 2019;Qu et al., 2013). Subsequently, probes showing either a 5hmC value of zero in at least half of the cases or BS beta value lower than 0.1 were filtered out (112,697 5hmC values were excluded). Data distributions of raw and normalized beta values per sample were visualized in boxplots and density plots to identify outliers (3 samples were excluded due to clear deviation from the other samples; data not shown). Data were processed again upon removal of these outliers.
After data processing, 78 samples remained, with 397,160 remaining probes for 5mC and UC, and 284,463 5hmC probes. Data were split into 2 sets of beta values according to the oxBS and BS arrays, representing 5mC and 5mC þ 5hmC, respectively.

Statistical analysis
Raw expression data from the LightCycler 480 were analyzed using LinRegPCR to determine logarithmic fluorescence values at cycle zero (Ruijter et al., 2009). qPCR reactions that did not show fluorescent signal amplification were excluded from further analysis. Logarithmic fluorescence values were normalized against the average expression of reference genes TOP1 and EIF4A2. Normalized data followed a Gaussian distribution and were used for statistical analysis.
For expression data, linear regression analyses were performed using IBM SPSS Statistics for Windows, Version 25.0 (IBM Corp., Armonk, NY) with normalized starting fluorescence values per isoform as dependent variable and diagnosis (control ¼ 0, AD ¼ 1, respectively), plaque load in the temporal lobe (0-3), tangle load in the temporal lobe (0-3), Braak stage level (Braak I-II ¼ 0, Braak III-IV ¼ 1, and Braak V-VI ¼ 2), or MMSE score (0e30) as independent variable. Age, gender, PMI, APOE ε4 possession, and qPCR plate were included as covariates. Unstandardized regression coefficients were exponentially transformed to calculate the fold expression change in AD. p values were Bonferroni-corrected by multiplying p values by the number of dependent variables (i.e., expression of isoforms) tested per model, and values smaller than 0.05 were considered statistically significant.
For (hydroxy)methylomic data, a surrogate variable (SV) analysis was done with the sva package (version 3.30.1) (Leek et al., 2012) on an initial linear regression model with beta values as dependent variable, diagnosis as independent variable, and age, gender, PMI, and APOE ε4 possession as covariates. To adjust for unobserved confounders, the first 5 SVs of the SV analysis were included in the regression model. Linear regression was performed using the limma package (version 3.38.3) (Ritchie et al., 2015). Adjustment for bias and inflation of test statistics was done with the bacon package (version 1.10.1) (van Iterson et al., 2017). To explore whether isoform-specific expression changes could be mechanistically explained by specific epigenetic changes, only probes that locate closest to PDE4D isoform promoters showing significant expression changes (i.e., PDE4D1, -D3, -D5 and -D8) were selected (24 probes). Nominal significance for DNA methylation changes between AD cases and controls was set at p-value <0.05. Diagnosis-associated methylation changes were investigated further by follow-up linear regression analyses using the same model, but instead of diagnosis, using plaque load in the temporal lobe (0e3), tangle load in the temporal lobe (0-3), Braak stage level (Braak I-II ¼ 0, Braak III-IV ¼ 1, and Braak V-VI ¼ 2), or MMSE score (0e30) as predictor variable. False discovery rates for these predictors' p values were determined by the Benjamini-Hochberg procedure.

Increased PDE4D isoform expression in Alzheimer's disease
Linear regression analyses revealed that AD diagnosis was significantly associated with increased expression levels of PDE4D1, -D3, -D5, and -D8, in the range of 2.56-to 3.85-fold increased expression (p < 0.05; Table 3).

Differential epigenetic signatures in PDE4D promoter regions in Alzheimer's disease
As PDE4D was found to be differentially methylated in AD in an epigenome-wide association study using the same tissue (Lardenoije et al., 2019), as an exploratory approach, we investigated whether 5'-cytosine-phosphate-guanine-3' sites within the promoter regions of differentially expressed PDE4D1, -D3, -D5, -D8 were subject to methylation differences. Twenty-four of the epigenome-wide association study assay probes located to these promoters and allowed the assessment of the methylation status (5mC, 5hmC, or UC) of specific 5'-cytosine-phosphate-guanine-3' sites. Linear regression analyses revealed diagnosis-dependent methylation status differences in each of the promoters. The 5'cytosine-phosphate-guanine-3' sites for which methylation status was associated with diagnosis are shown in Table 4. AD-associated effects on methylation status appear to be different per promoter region. For example, in the PDE4D3 promoter, at position cg13112511, AD diagnosis was associated with decreased 5mC (fold change [FC] ¼ 0.9818, p < 0.05) and increased 5hmC (FC ¼ 1.0197, p < 0.05) levels. Contrastingly, increased 5mC (FC ¼ 1.0144, p < 0.01) and decreased 5hmC (FC ¼ 0.9812, p < 0.05) levels were found in the PDE4D8 promoter at cg23987137 in AD cases. Within the PDE4D5 promoter, effects with opposite directionality were found; increased 5mC and decreased UC levels were detected in AD cases at position cg03653541 (5mC: FC ¼ 1.0023, p < 0.05 and UC: FC ¼ 0.9972, p < 0.05), while position cg14784398 indicated decreased 5mC (FC ¼ 0.9958, p < 0.05) and increased UC (FC ¼ 1.0035, p < 0.05) levels. These findings suggest that AD is associated with PDE4D DNA (hydroxy)methylation changes that are promoter-and site-specific, concomitant with expression changes of the corresponding isoforms.

PDE4D isoform expression is associated with AD pathology and cognitive impairment
For isoforms showing different expression between AD and controls (i.e., PDE4D1, -D3, -D5, and -D8), linear regression analyses were performed to investigate whether expression was associated with the degree of pathology or cognitive impairment. Higher levels of plaque pathology in the temporal lobe were associated with increased PDE4D1 and -D3 expression (p < 0.05), while tangle pathology in the temporal lobe was associated with increased expression of PDE4D1 (p < 0.05), PDE4D3 (p < 0.01), and PDE4D5 (p < 0.05; Table 5 and Fig. 1). In addition, Braak staging associated significantly with increased expression of PDE4D1 (p < 0.05) and PDE4D3 (p < 0.001). Additionally, we examined the association between isoform expression and the level of cognitive impairment as assessed by the MMSE. Lower MMSE scores were significantly associated with increased expression of PDE4D1 (p < 0.01), PDE4D3 (p < 0.01), and PDE4D5 (p < 0.05; Table 5 and Fig. 1). These results indicate that increased expression, of mainly the PDE4D1 and PDE4D3 isoforms, correlates with higher pathology levels and more severe cognitive impairment.

PDE4D promoter methylation is associated with AD pathology and cognitive impairment
The diagnosis-dependent methylation changes as determined in Table 4 were investigated further for their potential association  with the degree of pathology or cognitive impairment by means of linear regression analyses. These analyses revealed significant associations between methylation status changes and pathology or cognitive impairment in accordance to the directionality of changes observed between AD cases and controls. In the PDE4D3 promoter, both decreased 5mC (hypomethylation) and increased 5hmC (hyperhydroxymethylation) at CpG site cg13112511 were found to associate with an increased plaque load and a higher Braak stage (Table 6). Contrarily, in the PDE4D8 promoter, increased 5mC and decreased 5hmC at position cg23987137 was associated with increased pathology levels. Methylation status in the PDE4D5 promoter also associated with levels of pathology but the directionality of methylation status change was different per CpG site (e.g., cg03653541 shows increased 5mC, while cg12519013 shows decreased 5mC with higher levels of pathology; Table 6). Finally, lower MMSE scores were associated with increased 5mC levels in the PDE4D5 promoter (cg03653541) and PDE4D8 promoter (cg23987137).

Discussion
Inhibition of PDE4D has been shown to hold promise as an effective strategy to enhance memory in humans (clinical trials NCT02648672 and NCT02840279), but is also assumed to contribute to possible emetic side effects upon non-selective PDE4 inhibition (Giembycz, 2002;Mori et al., 2010). The PDE4D gene encodes multiple isoforms that, for example, by specific intracellular localization, can regulate different compartmentalized pools of cAMP . Hence, targeting specific PDE4D isoforms may provide a safer and more efficacious treatment strategy to alleviate memory deficits in AD without evoking side effects (Schepers et al., 2019). To investigate which specific PDE4D isoform(s) is/are altered in terms of expression in AD, we established PDE4D isoform expression profiles in post-mortem MTG tissue derived from AD patients and healthy non-demented controls by means of isoform-specific qPCR.
Regression analyses revealed that AD was associated with increased expression of PDE4D1, -D3, -D5, and -D8. Expression of these isoforms was increased in the range of 2.56-to 3.85-fold. Previous studies have indicated that 3-to 4-fold increased PDE4D (isoform) mRNA expression results into similar increases in both PDE4D protein expression and PDE4 activity (Levallet et al., 2007;Peter et al., 2007). Therefore, the 2.59-to 3.85-fold expression increases likely increase PDE4 activity. To our knowledge, the number of studies investigating PDE4D expression in light of AD is limited. Ugarte et al. (2015) did not find a difference in overall PDE4D expression in temporal lobe material from AD patients (n ¼ 7) and controls (n ¼ 8). McLachlan et al. (2007) investigated hippocampal PDE4D isoform expression in material from a single patient with advanced AD and 3 control cases and found a 2.62-fold increase in PDE4D1 expression, while the expression of other isoforms was reduced or unaffected. Our study, making use of a much larger sample size, shows, in addition to a similar increase in PDE4D1 expression, increased expression of PDE4D3, -D5, and -D8. The current findings imply that PDE4D expression is affected in AD in an isoform-specific manner.
Because AD is a progressive disease, we also inspected whether PDE4D isoform expression increases with higher degrees of pathology or more severe cognitive impairment. Expression of PDE4D1 and PDE4D3 was associated with temporal plaque load, temporal tangle load, Braak staging, as well as cognitive impairment. PDE4D5 expression was found to be only associated with temporal tangle load and cognitive impairment, while PDE4D8 expression was not associated with any pathology measure or cognitive impairment. These findings suggest that mainly PDE4D1 and PDE4D3 expression changes as the disease progresses and cognitive impairment worsens. The observation that PDE4D3 transcription is increased in presence of higher amyloid-b plaque load corresponds to earlier findings. Hippocampal injections of Ab 1-42 in mice caused memory deficits and increased protein expression of PDE4D3, which was reversed upon non-selective PDE4 inhibition by resveratrol (Wang et al., 2016). Interestingly, Olah et al., (2011) found a direct interaction between oligomeric Ab and PDE4D. Whether this interaction occurs in vivo and how this interaction affects PDE4D enzymatic activity remains to be determined. From a potentially therapeutic perspective, it would be interesting to identify whether inhibition of PDE4D1 and/or PDE4D3 specifically can delay disease progression and improve cognitive functioning.
As the PDE4D gene, with its various isoforms embedded, contains multiple promoter regions, epigenetic mechanisms are likely involved in its isoform-specific expression regulation, for example, by influencing promoter accessibility (Maunakea et al., 2010;Tilley and Maurice, 2005). Here, we found that DNA (hydroxy)methylation signatures in the promoter regions of PDE4D1, -D3, -D5, and -D8 were altered in the same tissue as used for the expression profiling. Results showed differentially methylated CpG sites in these promoters that were associated with AD diagnosis. Moreover, methylation status was also associated with the degree of pathology and cognitive impairment for several CpG sites. Various effects were observed, that is, both increased and decreased 5mC, 5hmC, and UC levels. Differences in methylation and hydroxymethylation can be linked to altered gene expression by affecting the binding of transcription factors and chromatin proteins to the DNA (Suzuki and Bird, 2008;Yin et al., 2017). As such, decreased methylation and increased hydroxymethylation in AD in the PDE4D3 promoter might explain, at least in part, the observed pathology-and ADassociated increased expression of PDE4D3. In contrast, in the PDE4D8 promoter, cg23987137 displayed AD-associated hypermethylation and hypohydroxymethylation. However, cytosine methylation can have differential effects on transcription factor binding and subsequent gene expression. Interestingly, cg23987137 is located 14 nucleotides downstream of an activator protein 1 (AP-1) binding site (Farre et al., 2003). The transcription factor AP-1 is usually described as a transcriptional activator, but can also act as transcriptional repressor (Brellier et al., 2004). Moreover, it has been shown that methylation adjacent to, but not within, the AP-1 binding site can hamper AP-1 binding and consequently induce transcription (Fujimoto et al., 2005). Although purely speculative, a similar mechanism might affect transcription near the PDE4D8 promoter resulting in increased expression upon CpG site hypermethylation. Similarly, epigenetic changes observed in the other promoters can, by influencing binding of transcriptional activators and repressors, induce increased expression of the associated isoforms. Interestingly, increased methylation at cg23987137 PDE4D8 promoter was associated with worse cognitive performance while PDE4D8 expression did not associate with cognitive performance. This discrepancy may suggest that DNA methylation may not solely influence expression via its closest located promoter but may also regulate expression of other isoforms which do show associations with cognitive performance (i.e., PDE4D1, -D3, and -D5).
The current findings show that in AD, expression of specific PDE4D isoforms is altered with concurrent changes (hydroxy) methylation changes in associated promoters of the PDE4D gene. Although PDE4D isoforms locate to specific cellular compartments, the role of individual PDE4D isoforms in specific cellular processes is still largely unknown. The mouse orthologue of PDE4D1 is reported to localize to the nucleus (Chandrasekaran et al., 2008). With 95.6% homology in the N-terminal, human PDE4D1 may share a similar localization signal and hence could be involved in the regulation of nuclear cAMP levels which may directly affect PKA-CREB mediated transcription crucial for memory consolidation (Kandel, 2012). Transgenic overexpression of PDE4D1 in specific  Table 5. Significance levels Bonferroni-corrected for the amount of dependent variables tested; *p < 0.05, **p < 0.01, ***p < 0.001. Abbreviations: AD, Alzheimer's disease; CTRL, control; MMSE, Mini-Mental State Examination; PDE4D, phosphodiesterase 4D; PMI, post-mortem interval; qPCR, quantitative polymerase chain reaction. (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.) neurons was found to decrease Ca 2þ oscillations, which may imply a specific role for this isoform in neuronal firing (Yoshida et al., 2003). Increased PDE4D1 expression in AD may therefore decrease neuronal firing rates that subsequently impacts neuronal plasticity and memory formation processes.
Specific interactions have been reported between PDE4D3 and A-kinase anchoring proteins (AKAPs), for example, AKAP9 and mAKAPa, through which PDE4D3 can locally degrade cAMP pools (Boczek et al., 2019;Terrenoire et al., 2009). Disrupting the perinuclear PDE4D3-mAKAPa signaling complex promoted survival and axon growth in neurons (Boczek et al., 2019). Vice versa, increased PDE4D3 expression as observed in the current study could hamper these effects, thereby likely impairing neuronal plasticity and memory processes in AD.
Similar to PDE4D3-AKAP interactions, the scaffolding protein barrestin was found to preferentially interact with PDE4D5 . b-arrestin is recruited to G-protein coupled receptors (GPCRs) upon activation, which places PDE4D5 in close proximity to sites where cAMP is produced, that is, close to GPCRs and activated adenylyl cyclases that produce cAMP (Baillie et al., 2003;Houslay and Baillie, 2005). Increased PDE4D5 levels may therefore excessively degrade cAMP at the site of synthesis. Vice versa, downregulation of PDE4D5 may stimulate cAMP signaling, which is supported by the interesting finding that knockdown of PDE4D4 or PDE4D5 can improve memory performance in mice (MacKenzie et al., 2011;Wang et al., 2013). In addition to cytosolic interactions, it was found that PDE4D5 also regulates cAMP levels in nuclear signaling complexes, indicating that a specific isoform can regulate cAMP levels in spatially distinct compartments (Clister et al., 2019).
Regarding PDE4D8, most insights have been gathered through cardiovascular research. In cardiomyocytes, PDE4D8 is located near the b1-adrenergic receptor through a SAP97-mediated interaction and dissociates upon receptor stimulation (Fu et al., 2014;Richter et al., 2008). As SAP97 is a synapse-associated protein that scaffolds multiple receptors, including AMPA receptors crucial for memory processes, similar interactions may occur in the brain (Zhang et al., 2015). Moreover, PDE4D8 is found to localize in leading-edge structures of migrating vascular smooth muscle cells and affects actin organization upon overexpression (Raymond et al., 2009). Actin-rich leading edges in neuronal or glial cells may also be affected by increased PDE4D8 levels and consequently induce dysfunctional migration and growth, possibly contributing to impaired cognitive functioning eventually.
Through their specific localizations, the isoforms PDE4D3, -D5, and -D8 have all been reported to modulate downstream signaling of b-adrenergic receptors (Baillie et al., 2003;De Arcangelis et al., 2009;Richter et al., 2008). As adrenergic signaling is involved in cognitive functioning and can be affected in AD (Gannon et al., 2015), changes in expression of PDE4D isoforms may disrupt downstream b-adrenergic signaling leading to cognitive dysfunction. In sum, upcoming studies will have to indicate isoformspecific roles in memory processes, so local cAMP compartments can be targeted and treatment efficacy can be increased.
Finally, the current findings represent PDE4D isoform expression and DNA (hydroxy)methylation in homogenized brain tissue, meaning no conclusions can be drawn yet on potential cell typespecific effects. Hence, follow-up studies will have to indicate whether the current findings can be linked to changes in specific cell types (e.g., neurons, microglia, astrocytes, oligodendrocytes, endothelial cells). Also, whether increased PDE4D expression is causal to or a consequence of worsening pathology remains to be specified. Nevertheless, these findings highlight isoform-specific PDE4D inhibition as a potential strategy to aid in delaying disease progression with associated procognitive effects.

Conclusions
Taken together, regulation of PDE4D transcription is affected in AD as indicated by differences in gene (hydroxy)methylation and expression levels. Moreover, the specific isoforms PDE4D1 and PDE4D3 show a significant association with the degree of pathology and cognitive impairment. In line with the fact that PDE4D isoforms regulate specific intracellular processes, this implies that targeting specific PDE4D isoforms may provide a more efficacious treatment strategy with less adverse effects in different stages of the disease. For this purpose, future studies will have to reveal which PDE4D isoforms regulate mechanisms underlying memory enhancement and do not show adverse effects upon inhibition in the context of brain function in both health and AD.