Protective effect of PDE4B subtype-specific inhibition in an App knock-in mouse model for Alzheimer’s disease

Alzheimer’s disease (AD) is the leading cause of dementia and disability in old age, with pathological features including extracellular plaque deposits of the β-amyloid (Aβ) peptide and an estimated heritability of ∼60% [1–3]. Although recent clinical trials of Aβ-targeting monoclonal antibodies resulted in moderately less cognitive decline in people with early AD [4, 5], no therapeutics capable of halting or reversing the progression of the disease have been described.

The risk of developing senile dementia is 34% higher in individuals with gastroesophageal reflux disease (GERD) compared with controls [6]. Meta-analysis of genome-wide association study (GWAS) data indicates that AD and GERD share seven genome-wide significant susceptibility loci [7]. Among the implicated genes, PDE4B, encoding one of four subtypes of cyclic adenosine monophosphate (cAMP)-specific phosphodiesterase-4 (PDE4A–D), was proposed as a plausible therapeutic target that should be investigated further [7].

Across various tissues, PDE4B is expressed in five isoforms (PDE4B1–5) [8]. In adult mammalian brain tissue, Pde4b transcripts have been identified by single-cell RNA sequencing in nearly all subclasses of GABAergic (gamma-aminobutyric acidergic) inhibitory neurons and glutamatergic excitatory neurons, and in some types of glial cell (Figs. S1 and S2) [9, 10]. PDE4B transcription was shown to be upregulated in primary rat microglial cell cultures by exposure to Aβ peptides, resulting in induction of the inflammatory cytokine TNFα, which was markedly decreased by the non-subtype-selective (pan-) PDE4 inhibitor rolipram (targeting all of four subtypes, PDE4A–D) [11].

The super-family of phosphodiesterases (PDE1–11) has long been considered as potential targets for AD therapy [12–19]. Transgenic amyloid precursor protein (APP)-overexpressing mouse models of AD have shown amelioration of cognitive deficits following treatment with the pan-PDE4 inhibitors rolipram [20–22], roflumilast [23, 24] and FFPM [25], and with the PDE4D subtype-selective inhibitors GEBR-7b [26] and GEBR-32a [27]. Rolipram had no effect on Aβ peptide levels or plaque load in the PS/APP and Tg2576 strains [20, 21], although treatment over 24 days was shown to decrease Aβ peptide levels in brain tissue from 11-month-old 3xTg-AD mice [22]. Prior to the genetic validation of PDE4B by the GWAS meta-analysis [7], no work had been published in this context specifically on the PDE4B subtype.

We previously showed that a hypomorphic mutation (Pde4b^Y358C), which affects all PDE4B isoforms and decreases the enzyme’s cAMP hydrolytic activity by 27%, results in increased phosphorylation of CREB (cAMP response element binding protein) and cognitive enhancement in young adult (12-week-old) C57BL/6J mice [28]. To interrogate the involvement of PDE4B in the manifestation of AD-related phenotypes, we evaluated the neurocognitive effects of the Pde4b^Y358C hypomorph in the App^NL-G-F knock-in mouse model [29], which shows Aβ peptide accumulation, neuroinflammation, and cognitive impairment in an age-dependent manner, without the non-physiological overexpression of APP [30].

The murine App^NL-G-F allele has a humanized Aβ peptide sequence, owing to three amino acid substitutions (G676R, F681Y, R684H; exon 16), and harbors three familial AD mutations: Swedish (K670N, M671L; exon 16), Arctic (E693G; exon 17), and Iberian (I716F; exon 17) [29]. The lecanemab monoclonal antibody, developed as an immunotherapy and tested in recent successful clinical trials, was raised against the Arctic Aβ variant [4].

Herein, we report that spatial memory and brain metabolism deficits exhibited by 12-month-old homozygous App^NL-G-F mice are counteracted by the hypomorphic PDE4B, without decreasing Aβ plaque burden. This genetically mediated protective effect identifies PDE4B as a key regulator of disease manifestation in the App^NL-G-F model and a promising therapeutic target for AD.

For more detailed methodology, see the Supplementary Materials and Methods.

Authentic animal models serve as valuable tools for determining the molecular mechanisms of disease progression and testing potential therapeutic approaches. The App^NL-G-F model develops Aβ plaques, neuroinflammation, damaged synapses, and behavioral and cognitive deficits, while accurately recapitulating endogenous APP expression [29, 30]. However, it was unknown whether App^NL-G-F mice also replicate the cerebral hypometabolism that is a hallmark feature of AD [39]. Accumulating studies have shown that this decline in cerebral glucose metabolism occurs before pathology and symptoms manifest, continues as symptoms progress, and is more severe than the gradual decline in metabolic efficiency during normal aging [47]. 18-fluorodeoxyglucose positron emission tomography (PET) brain imaging in AD patients has observed cerebral glucose metabolism deficits in a range of brain regions, including the PFC and medial temporal lobe/hippocampus [48, 49], with this hypometabolism being a potentially useful diagnostic biomarker [50].

Using the translational ¹⁴C-2-DG brain imaging technique, we observed that 13-month-old App^NL-G-F mice exhibit widespread glucose hypometabolism in multiple brain regions, which was prevented by the hypomorphic PDE4B present in App^NL-G-F/Pde4b^Y358C mice. Glucose utilization was highest in the thalamus, consistent with the distribution of PDE4B in the primate brain [51, 52]. Targeted inhibition of PDE4B is thus an intervention that increases cerebral metabolism in App^NL-G-F mice (i.e., improves the neuronal energy state) and therefore has the potential to be disease-modifying in patients. By contrast, acute pretreatment of normal mice and rats with the pan-PDE4 inhibitor rolipram decreases glucose utilization in brain tissue [53–56]. Moreover, pan-PDE4 inhibitors have dose-dependent side effects of nausea and emesis – attributed to the selective inhibition of PDE4D – which limit their tolerability and translational value [57].

The SLM serves as a relay between the entorhinal cortex (EC) and the CA1 and is believed to represent the substrate of distinct aspects of spatial and episodic memory [58, 59]. Neurons of the EC that send projections to the CA1 are the initial degenerating cells in AD [60]. Considering the spatial memory impairment of App^NL-G-F mice, it is noteworthy that their hippocampal hypometabolism was restricted to the SLM. However, App^NL-G-F mice also exhibited hypometabolism in the PFC, which itself has a role in spatial memory formation [61, 62]. The restoration of glucose metabolism in both the SLM and PFC thus likely contributed to the improved spatial memory of App^NL-G-F/Pde4b^Y358C mice.

Cerebral glucose hypometabolism has previously been observed in transgenic APP-overexpressing mouse models of AD. The PDAPP strain showed glucose hypometabolism across multiple brain regions at 10 months of age [63], but this reached statistical significance only in the posterior cingulate cortex in 17-month-old mice following Bonferroni correction for multiple comparisons [64]. The PS/APP strain had localized hypometabolism in several brain regions but an unaltered whole brain average at 16 months [65], whereas the 3xTg-AD strain showed widespread hypometabolism in all measured brain regions at 18 months of age [66].

The area occupied by thioflavin-S-stained Aβ plaques in brain sections from 12-month-old App^NL-G-F mice was similar to that previously reported for 11-month-old App^NL-G-F mice, which showed a mild corticolimbic Aβ pathology relative to age-matched 5xFAD and APPswe/PSEN1dE9 mice that overexpress APP [67]. The similar Aβ plaque burdens of App^NL-G-F and App^NL-G-F/Pde4b^Y358C mice in the present study suggest that a decrease in Aβ plaque load is not responsible for the prevention of spatial memory and brain metabolism deficits by hypomorphic PDE4B. The situation in App^NL-G-F/Pde4b^Y358C mice is thus comparable with that of the ∼30% of older adults without signs of cognitive impairment who exhibit the neuropathological features of AD upon autopsy at the time of death [68]. This is not, however, contradictory to PDE4B being a promising therapeutic target for AD. There is a need for non-Aβ-directed therapies for AD because Aβ-targeting antibodies have shown only modest cognitive benefit in early AD [4, 5], and more than 20% of individuals diagnosed with AD do not have Aβ plaque burden as assessed by PET imaging [69].

Despite the central role of PDE4B in inflammation [70–72], the upregulation of multiple inflammatory markers in both App^NL-G-F/Pde4b^Y358C and App^NL-G-F mice suggests that the protective effect of hypomorphic PDE4B cannot be attributed to a broad anti-inflammatory response. Although TNFα levels were not significantly affected by genotype, the lack of detectable TNFα in four of the five App^NL-G-F/Pde4b^Y358C mice is consistent with the known suppression of TNFα production by PDE4B ablation and selective inhibition [72, 73].

RNA sequencing to identify molecular pathways potentially underlying the protective effect of hypomorphic PDE4B revealed that 531 of 76,076 robustly expressed gene transcripts were DE in cerebral cortical tissue from App^NL-G-F versus WT mice. None of the PDE4 transcripts (Pde4a–d) were DE in App^NL-G-F mice (Table S10), unlike the upregulation of Pde4b in cultured microglia exposed to Aβ peptides [11]. Among APP-overexpressing mouse models, the 3xTg-AD strain exhibited increased PDE4B and PDE4D protein levels in hippocampus and PFC, which were decreased by rolipram [22], and the APPswe/PSEN1dE9 strain exhibited increased PDE4B and PDE4D protein levels in cerebral cortex, which were decreased by roflumilast [24].

Among the transcripts DE in App^NL-G-F mice, 59% were also DE in App^NL-G-F/Pde4b^Y358C versus WT mice, including the upregulation of CCl3, encoding MIP-1α that was also upregulated in the inflammatory marker assay. None of the Pde4b transcripts (0 of 9) were DE in App^NL-G-F/Pde4b^Y358C versus WT mice (Table S10), as expected from prior analyses of Pde4b^Y358C mouse brain tissue [28]. Both increased and decreased expression of Pde4b in the hippocampus have been shown to impair contextual fear memory in mice [74].

Remarkably, the levels of only 13 of the DE transcripts – from four genes: Ide, Btaf1, Padi2, and C1qb – were significantly different between App^NL-G-F/Pde4b^Y358C and App^NL-G-F mice, suggesting that their expression was the most modulated by the hypomorphic PDE4B. The 13 include all Ide (7 of 7), Padi2 (2 of 2), and C1qb (1 of 1) transcripts, but only three of the five Btaf1 transcripts in the RNA sequence dataset. Multiple independent transcripts thus provide support for three of these genes at the conservative adjusted significance threshold employed (p < 0.01), but we cannot exclude the possibility that the single transcript representing C1qb is a false positive.

Padi2, encoding protein-arginine deiminase type-2 (PADI2) that converts arginine residues in proteins into citrullines (citrullination) through deamination [75], and Ide, encoding a ubiquitously expressed metalloprotease (IDE) that cleaves peptides including Aβ [76, 77], are noteworthy because their expression is upregulated in postmortem AD patient brains compared with age-matched controls. Markedly increased levels of PADI2 are detected in hippocampal samples from AD patients, and the immunoreactivity of PADI2 and citrullinated proteins coincides with glial fibrillary acidic protein (GFAP)-positive astrocytes [78]. Increased levels of IDE are detected in postmortem brains from patients with moderate AD pathology (Braak stages III–IV) although decreased levels of IDE are found in severe AD (Braak stages V–VI) [79].

IDE levels in cerebral cortex from 9-month-old APPswe/PSEN1dE9 mice are elevated after the formation of the first Aβ plaques and show a positive correlation with full-length APP levels [44]. In 10- and 18-month-old APPswe/PSEN1dE9 mice, the expression of IDE is inversely correlated with spatial memory in the Morris water maze [45]. In 16-month-old Tg2576 mice, increased IDE expression appears within GFAP-positive astrocytes surrounding Aβ plaques [46]. Like these transgenic APP-overexpressing strains, 12-month-old App^NL-G-F mice, but not App^NL-G-F/Pde4b^Y358C mice, show upregulation of IDE protein in the brain – conceivably representing a compensatory mechanism aimed to decrease Aβ, MIP-1α, and MIP-1β levels. Despite the involvement of GFAP-positive astrocytes in the expression of Padi2 and Ide, this cell type is unlikely to mediate the modulation of these transcripts by PDE4B inhibition because Gfap was upregulated in both App^NL-G-F and App^NL-G-F/Pde4b^Y358C mice. However, the upregulation of some less abundant reactive astrocyte markers, such as Aqp4, was blunted in App^NL-G-F/Pde4b^Y358C mice.

The normalization of IDE expression in App^NL-G-F/Pde4b^Y358C mice suggests that activation of the cAMP signaling pathway negatively regulates IDE expression. This observation is consistent with the finding that IDE expression in streptozotocin-treated APPswe/PSEN1dE9 mice is decreased by administration of the cAMP analog and non-subtype-selective cAMP phosphodiesterase inhibitor bucladesine in a dose-dependent manner [80]. The convergence between our observations in App^NL-G-F mice and published findings from transgenic mice that overexpress APP suggest that cerebral hypometabolism and IDE upregulation are not artifacts related to the non-physiological overproduction of various APP fragments in the APP-overexpressing strains [30].

Although we have identified that the Pde4b^Y358C hypomorph corrected expression changes of specific cortical transcripts but did not decrease Aβ plaque burden in App^NL-G-F mice, a limitation of this study is that it has not elucidated the mechanism underlying the protective effect of hypomorphic PDE4B in the App^NL-G-F model. Since we aimed to evaluate the neurocognitive effects of genetically inhibiting PDE4B in App^NL-G-F mice, the study did not include a group with the Pde4b^Y358C mutation on a WT (App^+/+) background. Previous studies have employed similar experimental designs without a treatment-only App^+/+ group to assess the effects of genetic modifications in App^NL-G-F mice [81–83]. However, the lack of a Pde4b^Y358C-only group precluded statistical analysis of Pde4b genotype as an independent variable. Consequently, an additional limitation is that we could not evaluate the effect of PDE4B inhibition in 12-month-old mice independently of the App^NL-G-F mutant allele. Another limitation is that the App^NL-G-F model does not exhibit tau-containing neurofibrillary tangles, a histopathological hallmark of AD [29, 30]. Rolipram suppresses tau phosphorylation in 11-month-old 3xTg-AD mice [22] and in the rTg4510 mouse model of frontotemporal dementia at 3–4 months of age [84], but no work has been published on the effects of PDE4B subtype-specific inhibition on tau pathology. A further limitation is that the B6.C-Pde4b^enu1H mouse line employed does not permit spatial or temporal control over expression of the Pde4b^Y358C mutation. A Pde4b^Y358C conditional knock-in line would allow us to test whether hypomorphic PDE4B can arrest or reverse progression at different stages of disease in the App^NL-G-F model.

In summation, our data show that App^NL-G-F mice exhibit spatial memory and brain metabolism deficits that are prevented by targeted inhibition of PDE4B, a cAMP hydrolyzing enzyme that has been implicated in the pathogenesis of AD by a recent meta-analysis of GWAS data [7]. To the best of our knowledge, this is the first study demonstrating that PDE4B subtype-specific inhibition has a protective effect in a preclinical model of AD. This novel finding identifies the PDE4B subtype as a key regulator of disease manifestation in the App^NL-G-F model and a promising therapeutic target for AD.