Regulatory feedback cycle of the insulin‐degrading enzyme and the amyloid precursor protein intracellular domain: Implications for Alzheimer’s disease

Abstract One of the major pathological hallmarks of Alzheimer´s disease (AD) is an accumulation of amyloid‐β (Aβ) in brain tissue leading to formation of toxic oligomers and senile plaques. Under physiological conditions, a tightly balanced equilibrium between Aβ‐production and ‐degradation is necessary to prevent pathological Aβ‐accumulation. Here, we investigate the molecular mechanism how insulin‐degrading enzyme (IDE), one of the major Aβ‐degrading enzymes, is regulated and how amyloid precursor protein (APP) processing and Aβ‐degradation is linked in a regulatory cycle to achieve this balance. In absence of Aβ‐production caused by APP or Presenilin deficiency, IDE‐mediated Aβ‐degradation was decreased, accompanied by a decreased IDE activity, protein level, and expression. Similar results were obtained in cells only expressing a truncated APP, lacking the APP intracellular domain (AICD) suggesting that AICD promotes IDE expression. In return, APP overexpression mediated an increased IDE expression, comparable results were obtained with cells overexpressing C50, a truncated APP representing AICD. Beside these genetic approaches, also AICD peptide incubation and pharmacological inhibition of the γ‐secretase preventing AICD production regulated IDE expression and promoter activity. By utilizing CRISPR/Cas9 APP and Presenilin knockout SH‐SY5Y cells results were confirmed in a second cell line in addition to mouse embryonic fibroblasts. In vivo, IDE expression was decreased in mouse brains devoid of APP or AICD, which was in line with a significant correlation of APP expression level and IDE expression in human postmortem AD brains. Our results show a tight link between Aβ‐production and Aβ‐degradation forming a regulatory cycle in which AICD promotes Aβ‐degradation via IDE and IDE itself limits its own production by degrading AICD.


| INTRODUC TI ON
Currently, more than 50 million people globally are estimated to suffer from dementia. Alzheimer's disease (AD) is a progressive, irreversible neurodegenerative disease which is the most common cause of dementia in the elderly. The excessive accumulation and aggregation of the amyloid-β (Aβ) peptide in brain tissue leading to the formation of extracellular senile plaques is considered to represent the initial pathological process of the disease characterized by synaptic loss and neuronal injury (Chen et al., 2017). Aβ peptides are products of the sequential amyloidogenic processing of the type I transmembrane amyloid precursor protein (APP), a member of a conserved protein family also including the APP-like proteins 1 and 2 (APLP1 and APLP2), by β-and γ-secretase ( Figure S1). Beside the amyloidogenic APP processing pathway, APP can be cleaved in the predominant α-and γ-secretase dependent non-amyloidogenic cleavage cascade precluding the generation of Aβ peptides.
In both APP processing pathways, cleavage of APP by γ-secretase additionally leads to the release of the C-terminal APP intracellular domain (AICD) into the cytosol. Due to multiple-site cleavages by γ-secretase, Aβ and AICD peptides can vary in length with the main products being Aβ38, Aβ40, Aβ42, and AICD C50, C53, C57, C59, respectively (Chen et al., 2017;Grimm et al., 2013).
Total cerebral Aβ level is not only determined by Aβ-production, but also by Aβ-clearance and degradation mechanisms, which have been reported to be impaired in the predominant late onset form of AD (Mawuenyega et al., 2010). These Aβ-clearance mechanisms include among others the enzymatic elimination of Aβ peptides by proteases like insulin-degrading enzyme (IDE) and neprilysin (NEP) (Nalivaeva & Turner, 2019). IDE is a zinc metallopeptidase most abundant in the cytosol, but also in several other subcellular compartments (Saido & Leissring, 2012) and represents one of the most important Aβ-degrading enzymes in brain tissue. IDE deficient mice show increased cerebral accumulation of Aβ peptides while amyloid plaque formation is reduced in the brain tissue of mice with transgenic overexpression of IDE Leissring et al., 2003;Miller et al., 2003).
Besides Aβ, AICD has also been demonstrated to be degraded by IDE in vitro and in vivo Miller et al., 2003). AICD has been reported to be involved in the transcriptional regulation of several target genes including APP, BACE1, NEP, key enzymes of different lipid pathways and the mitochondrial master transcriptional coactivator PGC-1α (Grimm et al., 2015;Pardossi-Piquard et al., 2005;Robinson et al., 2014;von Rotz et al., 2004). The rapid cytosolic breakdown of AICD peptides by IDE and other enzymes might be precluded by binding to adaptor proteins like Fe65 enabling the translocation of AICD to the nuclear compartment (Kimberly et al., 2001). Within the nucleus, a trimeric protein complex consisting of AICD, Fe65, and the histone acetyltransferase Tip60 (AFT-complex), which functions in transcriptional regulation, is formed (Cao & Sudhof, 2001;von Rotz et al., 2004).
In this study, we identified the Aβ-degrading protease IDE as a target gene of AICD nuclear signaling. Hence, the two major Aβdegrading enzymes IDE and NEP are transcriptionally upregulated by AICD. This indicates the existence of a regulatory cycle in which proteolytic APP processing generates Aβ peptides and concurrently ensures their enzymatic degradation.
Total intracellular Aβ-degradation was measured by the addition of synthetic human Aβ40 peptides to the cell lysates for 1 h and subsequent quantification of the remaining, not degraded human Aβ40. No significant difference in Aβ-degradation was observed between MEF wild type (MEF WT) and MEF PS1res (Figure 1a, production regulated IDE expression and promoter activity. By utilizing CRISPR/Cas9 APP and Presenilin knockout SH-SY5Y cells results were confirmed in a second cell line in addition to mouse embryonic fibroblasts. In vivo, IDE expression was decreased in mouse brains devoid of APP or AICD, which was in line with a significant correlation of APP expression level and IDE expression in human postmortem AD brains. Our results show a tight link between Aβ-production and Aβ-degradation forming a regulatory cycle in which AICD promotes Aβ-degradation via IDE and IDE itself limits its own production by degrading AICD.
These results indicate that the PS-dependent APP cleavage product AICD might also be involved in the regulation of IDE besides the reported influence of AICD on NEP (Grimm et al., 2015).

| IDE enzyme activity and protein level are reduced in MEF cells devoid of PS1/2, APP/ APLP2, and AICD
In order to analyze whether the PS/APP/AICD-dependent effects on total Aβ-degradation are partially based on an altered IDE activ-  Table 1). Importantly, a similar effect was also observed by inhibition of γ-secretase activity in the PS1 retransfected control cells demonstrating IDE protein level to be strongly dependent on γ-secretase activity (MEF PS1res + DAPT: 68.7% ± 5.9%, p ≤ 0.001) ( Figure 2c, Table 1).
Similarly, IDE protein level was found to be significantly decreased in  Table 1). These results further support a mechanism in which IDE might be regulated in an AICD-dependent manner.

| Influence of PS, APP, and AICD on IDE gene expression in MEF and SH-SY5Y cells
AICD has been reported to translocate to the nucleus and to be  PS1 (SH-SY5Y PS1−/−) ( Figure 3c, Table 1, cell line controlled in Figure S2H) or APP (SH-SY5Y APP−/−) ( Figure 3d, Table 1, cell line controlled in Figure S2G). In accordance on the other hand overexpression of APP695, the most common APP isoform in neuronal cells resulted in a significantly increased IDE gene expression in SH-SY5Y cells (SH-SY5Y + APP695) ( Figure 3d, Table 1, cell line controlled in Figure S2E). As the nuclear localization and gene regulatory activity is discussed to be restricted to AICD derived from  Table 1), but did not reach the level of MEF WT cells.

| Impact of AICD on IDE gene expression and IDE protein level
To further strengthen the importance of AICD in the regulation of IDE we transiently transfected MEF APPΔCT15 cells, lacking a functional AICD domain, with an AICD-expressing plasmid corresponding to the last 50 aa of the APP C-terminus (C50) (cell line characterized in Figure   S2I). APPΔCT15 cells expressing C50 showed a significant increase of IDE gene expression to 131.4% compared to cells lacking AICD (MEF APPΔCT15) ( Figure 4a, Table 1). MEF APPΔCT15 short-and long-term incubation with AICD peptides also revealed a significant increase in IDE gene expression to 139.6% and 145.4%, respectively (Figure 4a,

| The effect of a functional AICD domain on IDE promoter activity
Next, we analyzed whether IDE promoter activity is affected in cells lacking the APP protein family or a functional AICD domain. Therefore, cells were transiently transfected with the dual

| In vivo relevance of AICD-dependent IDE gene expression
IDE gene expression was monitored in APP knockout mice (APP−/−) and in heterozygous mice expressing the truncated APP lacking the last 15 aa of the C-terminus (APPΔCT15+/−), to validate our findings in vivo. Brain homogenates of APP−/− mice showed a significant reduction in IDE gene expression ( Figure 6a, Table 1). Similarly, IDE gene expression was significantly reduced in brain homogenates of APPΔCT15 expressing heterozygous transgenic mice ( Figure 6a, Table 1). IDE protein level was also found to be significantly reduced in brain homogenates of APPΔCT15+/− mice ( Figure 6b, Table 1).
To further investigate the in vivo relevance of an AICD-  Figure S5C). Similarly, amyloid burden did not influence IDE and APP gene expression ( Figure S5D). The positive correlation between IDE and APP gene expression is not dependent on the gender as we obtained a significant positive correlation for both, women and men ( Figure S5F). The ApoE status of the patients had no impact on APP and IDE gene expression ( Figure S5E).
Additionally, no significant correlations were obtained for age and postmortem delay ( Figure S5G,H).  Table S3. Error bars represent the standard error of the mean and significance was set at *p ≤ 0.05, **p ≤ 0.01 and ***p ≤ 0.001 degraded among others by IDE or NEP, which play also an important role in intracellular Aβ-degradation (Iwata et al., 2001;Stargardt et al., 2013). NEP levels have been found to be reduced in hippocampus, temporal gyrus, and cortex of human postmortem AD brains (Grimm et al., 2013). However, there are still controversies in regard to the expression and activity of IDE in AD brains, showing reduced (Stargardt et al., 2013;Zhao et al., 2007), unchanged (Miners et al., 2010;Wang et al., 2010) or increased IDE activity (Miners et al., 2009;Morelli et al., 2004). Recently, we and others could show an AICD-dependent regulation of NEP increasing its gene expression,  Table S3. Error bars represent the standard error of the mean and significance was set at * p ≤ 0.05, ** p ≤ 0.01 and *** p ≤ 0.001 protein level, and activity (Belyaev et al., 2009(Belyaev et al., , 2010Grimm et al., 2015;Pardossi-Piquard et al., 2005;Xu et al., 2011). In our previous study, we could show that an AICD-dependent change in Aβdegradation could only be partially rescued by utilizing thiorphan, a specific inhibitor of NEP. These results suggest that beside NEP another Aβ-degrading protease might be also regulated by AICD (Grimm et al., 2015), which is in line with our recent finding that thiorphan could only partially attenuate the difference between

MEF WT and MEF APP/APLP2−/− cells in respect to Aβ degradation.
In the present study, we identified IDE as a further target gene of AICD using both cells devoid of AICD or AICD generation and AICD overexpressing cells. We found IDE gene expression to be consis-  Table S3. Error bars represent the standard error of the mean and significance was set at *p ≤ 0.05, **p ≤ 0.01 and ***p ≤ 0.001 increased IDE gene expression for APP751 and APP770 overex- Further illustrating an AICD-dependent regulation of IDE we found IDE gene expression to be significantly increased in SH-SY5Y F I G U R E 6 (a) IDE gene expression in brain homogenates of APP-deficient mice (APP−/−) and of mice expressing truncated APP lacking the last 15 aa of the APP C-terminus (APPΔCT15) compared to wildtype mice. (b) IDE protein level in brain homogenates of mice expressing truncated APP lacking the last 15 aa of the APP C-terminus (APPΔCT15+/−). Corresponding WBs are shown. No significant differences in β-actin signals exist between the two compared cell lines (APPΔCT15+/− mice: 105.5%, p = 0.406). Statistical significance was calculated as described in Table S3. Error bars represent the standard error of the mean and significance was set at *p ≤ 0.05, **p ≤ 0.01 and ***p ≤ 0.001. APPΔCT15 expressing C50 revealed significantly increased IDE promoter activity. This direct influence of AICD on IDE promoter activity might have an additional impact on the also discussed effect of the histone deacetylases HDAC1 and HDAC3 on the IDE promoter (Nalivaeva et al., 2016) found by the APP751 and APP770 isoforms.
Taken into consideration that we have shown in a previous study that AICD upregulates the expression of the peroxisome prolifera- learning and recognition impairments in these mice (Li et al., 2018).

Additionally, inhibition of PPARγ by injecting the PPARγ antagonist
GW9662 in the fourth ventricle of APP/PS1 transgenic mice markedly decreased cerebellar levels of IDE and significantly induced Aβ levels ). In line with our hypothesis that PGC-1α might be involved in the regulation of IDE gene transcription, Leal et al. (2013) reported a significant increase in cytosolic and mitochondrial levels of IDE in cells transfected with PGC-1α. Besides increasing PPARγ transcriptional activity, PGC-1α induces nuclear respiratory factor 1 (NRF-1) overexpression, which has been found to bind to the IDE promoter region in vivo (Leal et al., 2013). Moreover, a strong positive correlation between PGC-1α or NRF-1 and the long mitochondrial IDE isoform was found in non-demented brains, whereas this correlation was weaker in AD brains (Leal et al., 2013). The postulated AICD/PGC-1α/ PPARγ involvement in IDE transcriptional regulation might also play a role in the transcriptional regulation of NEP as it has been shown, that activation of the nuclear retinoid X receptor (RXR), the heterodimeric partner of PPARγ, upregulates not only IDE but also the Aβ degrading enzyme NEP (Nalivaeva et al., 2016). In the present study, we found some indications that the AICD/PGC-1α/PPARγ pathway might indeed be involved in the regulation of IDE. PGC-1α gene expression is significantly downregulated in SH-SY5Y WT cells incubated with a γ-secretase inhibitor and thus devoid of AICD generation ( Figure S6A) which is in line with our previous finding that PGC-1α mRNA level as well as protein level are decreased in PS-deficient cells (Robinson et al., 2014). Additionally, we found that the magnitude of effect on IDE mRNA level between MEF WT and MEF APPΔCT15 cells in presence of a PPARγ inhibitor is significantly less pronounced than without inhibitor ( Figure S6C However, although the data of human postmortem brain tissue is in line with the other experimental data, it has to be emphasized that this data has to be interpreted carefully. The average postmortem time of 06:07 hours and the short half-life of AICD make it impossible to directly correlate AICD levels with IDE expression (Kimberly et al., 2001). Instead, APP protein or RNA levels were analyzed and correlated with IDE making this approach more indirect.
In summary, we propose a feedback cycle for the AICDdependent regulation of IDE, in which AICD increases its own degradation as IDE has been also found to degrade AICD peptides (Edbauer et al., 2002). AICD upregulates IDE gene expression, either direct or by the above-discussed involvement of the PGC-1α/ PPARγ pathway leading to increased IDE protein level and activity ( Figure 6e). The increased IDE activity, in return, results in elevated degradation of Aβ as well as AICD peptides, resulting in the proposed feedback cycle. This cycle is closely linked to a feedback mechanism proposed for Aβ generation and degradation. AICD decreases APP processing by downregulating the expression of WASP-family verprolin homologous protein 1 (WASF1), resulting in impaired budding of APP containing vesicles from the Golgiapparatus, thereby reducing cell-surface APP and Aβ generation (Ceglia et al., 2015).
For the understanding of the disease mechanism, it should be taken into consideration that APP processing and therefore Aβ production is a continuous ongoing process under physiological conditions. Obviously, to achieve a homeostasis where no accumulation of Aβ takes place, Aβ-degradation and production has to be tightly regulated. Our paper might help to understand that this regulation encompasses AICD as a pivotal element both in regulating Aβ-degradation and Aβ production and importantly also in regulating its own degradation. Under pathological conditions, the disturbance of these complex entangled cycles leads to an accumulation of Aβ and promotes the progression of the disease.

| Chemicals and reagents
All chemicals and reagents were obtained from Merck former Sigma-Aldrich if not stated otherwise. The second one includes 36 female and 31 male brain samples with

| Cell culture, mouse and human brain samples
Braak stages 1-3 and an average postmortem delay of 06:10 hours (see Table S1).

| Generation of SH-SY5Y APP−/− and PS1−/− cells by CRISPR/Cas9
CRISPRdirect was used to design the CRISPR/Cas guide sequences to mediate APP and PS1 KO. Cloning into the pSpCas9(BB)-2A-Puro (PX459) plasmid was performed according to Ran and colleagues (Ran et al., 2013). A detailed description can be found in supporting information.

| Treatment of cells with inhibitors and AICD peptides
Incubation of cells with γ-secretase inhibitor DAPT (2.5 μM) and γ-secretase inhibitor X (2 µM) or the corresponding solvent con-

| Protein concentration
Bicinchoninic acid assay was used for determination of the protein concentrations in samples according to Smith et al. (1985)

| Total Aβ-degradation
Degradation of total Aβ in different MEF cell lines was performed according to Grimm et al. (2016) as described in detail in supporting information.

| Western blot experiments
For examination of IDE protein level, cell lysates were prepared as described above. Lysis buffer was additionally supplemented with Complete protease inhibitor cocktail (Roche Diagnostics). After centrifugation of the lysates for 5 min at 13,000 g and 4°C the supernatants were adjusted to equal protein amounts and loaded on 10-20% tris-tricine-gradient gels (Anamed Elektrophorese) and proteins were transferred onto nitrocellulose membranes afterward

(Whatman). A detailed description of Western blot analysis includ-
ing the used antibodies can be found in supporting information.
Signal detection was performed with the enhanced chemiluminescence (ECL-) method (Perkin Elmer) and for densitometrical quantification of band intensity after subtraction of the background signal; Image Gauge version 3.45 software (Fujifilm) was used.

| IDE activity assay
The enzyme activity of IDE was measured as published by Miners et al. (2008) with minor modifications as described earlier (Grimm et al., 2016). A detailed overview is given in supporting information.

| IDE promoter activity assay
Activity of the IDE promoter was measured by transiently transfecting cells with the dual reporter system vector pEZX-PG04-IDE-GLuc as described before (Grimm et al., 2016). For a detailed description see supporting information.

| RT-PCR experiments
For gene expression analysis quantitative real-time (RT) polymerase chain reaction (PCR) was performed and results were normalized to β-actin and changes in expression were calculated using the 2 −(ΔΔCt) method (Livak & Schmittgen, 2001). A detailed description can be found in supporting information.

| Data analysis
The quantified data represent an average of at least five independent experiments for each cell culture experiment. 223 human brain samples were analyzed. For APP−/− mice four brain samples and for For the statistical analysis of the human brain samples, we assumed that the data were normally distributed, since the sample size was over 200 (Ghasemi & Zahediasl, 2012).
Correlation coefficients were thus calculated via the Pearson method.
Significance was set at *p ≤ 0.05, **p ≤ 0.01 and ***p ≤ 0.001. All calculations were done with IBM SPSS Statistics version 25. Detailed overview of used statistical test can be found in Table S3.

CO N FLI C T O F I NTE R E S T
The authors declare no conflicts of interest.

E TH I C A L A PPROVA L
Treatment of WT, APP−/− and APPΔCT15+/− mice followed the German law for the use of laboratory animals (animal welfare act, TierSchG) and the Directive 2010/63/EU. The German administration approved animal housing, breeding and sacrifice. Human postmortem brain samples were collected from donors for or from whom a written informed consent for a brain autopsy and the use of the material and clinical information for research purposes had been obtained.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available from the corresponding author upon reasonable request.