Neuronal α‐amylase is important for neuronal activity and glycogenolysis and reduces in presence of amyloid beta pathology

Abstract Recent studies indicate a crucial role for neuronal glycogen storage and degradation in memory formation. We have previously identified alpha‐amylase (α‐amylase), a glycogen degradation enzyme, located within synaptic‐like structures in CA1 pyramidal neurons and shown that individuals with a high copy number variation of α‐amylase perform better on the episodic memory test. We reported that neuronal α‐amylase was absent in patients with Alzheimer's disease (AD) and that this loss corresponded to increased AD pathology. In the current study, we verified these findings in a larger patient cohort and determined a similar reduction in α‐amylase immunoreactivity in the molecular layer of hippocampus in AD patients. Next, we demonstrated reduced α‐amylase concentrations in oligomer amyloid beta 42 (Aβ42) stimulated SH‐SY5Y cells and neurons derived from human‐induced pluripotent stem cells (hiPSC) with PSEN1 mutation. Reduction of α‐amylase production and activity, induced by siRNA and α‐amylase inhibitor Tendamistat, respectively, was further shown to enhance glycogen load in SH‐SY5Y cells. Both oligomer Aβ42 stimulated SH‐SY5Y cells and hiPSC neurons with PSEN1 mutation showed, however, reduced load of glycogen. Finally, we demonstrate the presence of α‐amylase within synapses of isolated primary neurons and show that inhibition of α‐amylase activity with Tendamistat alters neuronal activity measured by calcium imaging. In view of these findings, we hypothesize that α‐amylase has a glycogen degrading function within synapses, potentially important in memory formation. Hence, a loss of α‐amylase, which can be induced by Aβ pathology, may in part underlie the disrupted memory formation seen in AD patients.


| INTRODUC TI ON
Alzheimer's disease (AD) is the most common form of dementia and is neuropathologically characterized by the presence of amyloid beta (Aβ) plaques and neurofibrillary tau-tangles (NFT) (Bloom, 2014). The pathological changes are thought to begin several years before symptoms appear (Jack et al., 2010;Sperling et al., 2011), and one of the earliest changes, besides the accumulation of Aβ and NFT, is reduced brain glucose metabolism (Mosconi, 2005). Since glucose constitutes the main energy source of the brain, impaired glucose metabolism has been suggested to be one of the major events underlying the neuronal loss and synaptic dysfunction seen in AD patients (Harris et al., 2012). Brain glucose metabolism is dependent on a constant uptake of glucose from the periphery, but as a backup mechanism, brain cells store and use glycogen, a multibranched polysaccharide produced by glycogen synthase. The production of glycogen occurs foremost in astrocytes (Ibrahim et al., 1975) and is degraded when energy is needed for the cell itself or, as studies suggest, is further converted into lactate, which is shuttled into neurons where it is used for energy (Bastian et al., 2019), memory formation (Duran et al., 2013;Gibbs, 2015), and neurotransmitter production (Hertz et al., 2015). This hypothesis has in recent years been challenged (Dienel, 2017;Drulis-Fajdasz et al., 2018;Lundgaard et al., 2015) and several studies have shown that neurons can produce and degrade glycogen on their own (Duran et al., 2019;Saez et al., 2014;. This glycogenolysis appears to be crucial for neuronal function, as studies have shown that mice lacking neuronal glycogen synthase have impaired memory and long-term potentiation (LTP) (Duran et al., 2019). The role of glycogen in memory formation, together with the known AD associated reduction of glucose metabolism, makes it tempting to speculate that impairment in glycogenolysis could in part underlie the memory decline and cognitive impairment seen in AD.
Degradation of brain glycogen is known to be executed by two enzymes called glycogen debranching enzyme and glycogenphosphorylase. However, we have recently demonstrated the presence of an additional glycogen degrading enzyme in the brain (Byman et al., 2018). This enzyme, called alpha (α)-amylase, is foremost known to be expressed in the salivary gland (Chopra & Xue-Hu, 1993) and pancreas (Whitcomb & Lowe, 2007), but it can also be found in the intestines (Date et al., 2020), liver (Koyama et al., 2001), muscle, lung, and several other organs (Whitten et al., 1988). By staining a small subset of human postmortem hippocampal tissue, we showed that α-amylase can be found in astrocytes (Byman et al., 2019), pericytes, and neurons (Byman et al., 2018). The neuronal αamylase was specifically located in structures resembling dendritic spines, which was clearly visible in non-demented individuals (NC), but much less or completely lost in AD patients (Byman et al., 2018).
The impact of AD pathology on brain α-amylase was further verified by qPCR gene expression analysis of brain tissue, which showed that α-amylase gene expression was significantly lower in AD brains compared to NC and correlated negatively with amyloid beta and NFT load (ABC and Braak staging). Whether the neuronal α-amylase plays a role in glycogen degradation and whether a loss of the enzyme could lead to failure of glycogen dependent function such as LTP remains to be investigated, but we have shown that individuals with high copy numbers (CN) (>10) of the salivary α-amylase isoform perform significantly better on an episodic memory test and also show significantly lower hazard ratio for developing AD (Byman et al., 2020). Since we also found a correlation between α-amylase CN and brain α-amylase production, we hypothesized that individuals with high α-amylase CN may be protected against AD and have better cognitive abilities due to a higher amount of glycogen degrading αamylase in synapses. This challenging hypothesis needs to be tested, where the first step is to determine the cellular localization and function of neuronal α-amylase. In the current study, we use human AD and NC postmortem hippocampal tissue and three different neuronal cell models; primary mouse neurons, hiPSC-derived neurons with and without presenilin 1 (PSEN1) mutations and SH-SY5Y cells to investigate whether (a) our previous observed reduction of α-amylase in dendritic spine-like structures in AD patients can be verified in a larger cohort; (b) Aβ has a direct impact on neuronal α-amylase; (c) neuronal α-amylase plays a role in glycogenolysis; (d) α-amylase is localized in synaptic spines and associated with glycogen granules; and finally, (e) inhibition of α-amylase activity affects neuronal activity using calcium imaging.

| Lower alpha-amylase immunoreactivity in neuronal processes of AD patients
We initiated our study by scoring α-amylase staining intensity in CA1 and ML of hippocampus in non-demented controls (NC) and AD cases. As reported earlier, the α-amylase immunoreactivity in NC was detected in structures resembling dendritic spines (Figure 1a,b). Also, the ML of NC showed an intense α-amylase staining, yielding a grained appearance (Figure 1a,c). The immunoreactivity in ML and CA1 was weaker in AD patients, and the presence of Hirano bodies (HB) was detected in both areas ( Figure 1d-f). The difference in α-amylase immunoreactivity was confirmed by our semiquantitative scoring, which showed significantly lower scores of dendritic spines in CA1 and ML grains in AD patients compared to NC (Figure 1g,h). The scores of CA1 correlated with ABC and Braak scores (Figure 1i,j). Also scores of ML correlated with both ABC and Braak scores (r = −0.458, p = 0.029 and r = 0.428, p = 0.019, respectively). The mean age of NC did not significantly differ compared to mean age of AD patients (76 ± 13 vs 80 ± 11, p = 0.415), and the gender distribution was similar in the NC and AD groups (54% and 59% females, respectively). Furthermore, the difference in scores of both CA1 and ML between ND and AD as well as the correlation with ABC and Braak scores remained significant after analysis using age or gender as a co-variate (data not shown).

| High presence of amyloid beta is associated with reduced alpha-amylase levels
To investigate whether the loss of neuronal α-amylase that we observed in AD cases is linked to Aβ pathology, we analyzed α-amylase levels in neurons derived from hiPSC cells containing a familial AD mutation in the PSEN1 gene (L150P) and its isogenic control (L150P-GC). Immunostaining of the cells showed that α-amylase was found in processes, nuclei and cytosols of hIPSC cells ( Figure 2a) and that the α-amylase in processes were closely associated with phalloidin-positive protrusions (Figure 2b). Further analysis showed that levels of α-amylase were significantly lower in L150P cells compared to L150P-GC (p = 0.003) (Figure 2c). To also investigate the direct effect of Aβ 42 on α-amylase, we next stimulated SH-SY5Y with either vehicle control or oligomer Aβ 42 . The SH-SY5Y cells showed α-amylase immunoreactivity in the whole cell (Figure 2e), but the concentration of the enzyme was lower in cells stimulated with Aβ 42 O compared to vehicle control (p = 0.032) (Figure 2f). No significant difference in cell death measured by LDH cytotoxicity assay between the two stimulations was found (0.16 (015-0.19) vs 0.14 (0.12-0.18) p = 0.22).

| Reduction of alpha-amylase levels and activity is associated with higher glycogen storage
To understand the significance of a reduction of α-amylase, we silenced the gene expression of α-amylase in SH-SY5Y using siRNA technique and investigated thereafter changes in glycogen load.
Glycogen analysis using ImageJ showed that siAMY transfected cells had significantly higher glycogen area/cell compared to siCtrl F I G U R E 1 Immunohistochemical staining of alpha (α)-amylase of human hippocampal postmortem tissue. Image in (a) shows an α-amylase staining of Cornius ammonis 1 (CA1) and the molecular layer (ML) in a non-demented control (NC). Higher magnification of the CA1 region in (b) shows that the α-amylase immunoreactivity (IR) is found in dendritic spine-like structures (DS). Higher magnification of the ML region in (c) shows that α-amylase IR in ML forms a grained pattern. Image in (d) represent an α-amylase staining of CA1 and ML in a patient with Alzheimer's disease (AD). Higher magnification of the CA1 region in (e) shows α-amylase IR in form of Hirano bodies, which is also seen in the ML region in (f). Graph in (g and h) shows that scores of the DS α-amylase IR in CA1 (g) and scores of grained α-amylase IR in ML (h) are significantly higher in NC compared to AD. Scatterplot in (i-j) demonstrates a negative correlation between DS α-amylase IR in CA1 and amyloid beta (Aβ) ABC scores (i) and neurofibrillary tangles (NFT) Braak scores in (j). Scale bar in (a and d) = 50 μm, in b-c = 20 μm and (e-f) = 10 μm. Graphs in (g-h) is presented as mean ± SD, and statistical analysis was done using t test. The correlation analyses in (i-j) were done using Spearman correlation test. ***indicates p-value < 0.001, **indicates p-value < 0.01 Figure 3a). Cell death measurement with LDH cytotoxicity assay showed no significant difference between siCtrl and siAMY (0.66 ± 0.038 vs 0.65 ± 0.045, p = 0.87). To verify our findings, we also investigate glycogen load in SH-SY5Y after stimulation with α-amylase inhibitor Tendamistat, a polypeptide that strongly binds to α-amylase and inhibits its hydrolytic activity (Vertesy et al., 1984). Stimulation with Tendamistat for 2 h led to a significantly higher glycogen area/cells compared to control (p = 0.036) (Figure 3b). The glycogen area/cell was still higher after 24 h of Tendamistat stimulation compared to control, but the difference did not reach significance (10.1 ± 6.5 vs 17.1 ± 7.7 p = 0.14).

| High presence of amyloid beta is associated with reduced glycogen storage
Since both L150P hiPSC and Aβ 42 stimulated SH-SY5Y cells showed lowered α-amylase levels, we investigated whether glycogen load in these cells was increased. To our surprise, our analysis instead showed significantly lower glycogen area/cell in Aβ 42 O stimulated SH-SY5Y compared to vehicle control (p = 0.0002) (Figure 3c and g).
A similar effect was seen in hiPSC with L150P mutation compared to isogenic control (p = 0.0049) (Figure 3d and h).

F I G U R E 2
Alpha-amylase in hiPSC-derived neurons and oligomer amyloid beta-42 stimulated SHSY5Y cells. Image in (a) shows hiPSCderived neurons without mutation in PSEN1 gene (isogenic control) (L150P-GC) immunoflourescently stained against salivary α-amylase (green) with DAPI marker nuclei (magenta). A hiPSC neuronal process is shown in higher magnification in (b) where the arrow indicates αamylase (green) and phalloidin (magenta) in close association. Graph in (c) shows the significant lower levels of α-amylase in L150P compared to L150P-GC. Image in (e) shows control (ctrl) SH-SY5Y cells immunofluorescent staining of α-amylase (green), and nuclei are visualized with DAPI (magenta). The graph in (f) shows significantly lower α-amylase concentrations in amyloid beta 42 oligomers Aβ42O stimulated cells compared to Ctrl. Scale bar in (a and e) indicates 15 μm, and scale bar in (b) indicates 0.5 μm. Statistical analysis was done with t test, and values in graphs (c and f) are presented as mean ± SD. ***indicates p-value < 0.001, **indicates p-value < 0.0, *p-value < 0.05

| Alpha-amylase co-localize with synaptic markers
To further verify the presence α-amylase in neurons, and determine its subcellular localization, we immunostained isolated primary mouse neurons against different neuronal markers together with αamylase. The staining showed α-amylase immunoreactivity in the nucleus, cytosol, and neuronal processes of MAP2-positive neurons ( Figure 4a). The α-amylase staining associated with the processes appeared dotted and were found scattered along and enclosing the MAP2-positive processes ( Figure 4a). This staining pattern resembled the neuronal α-amylase staining we found in hippocampi of NC and similar staining pattern of α-amylase could be seen when the primary neurons were stained with two other α-amylase antibodies (S1). The dotted α-amylase along the neuronal processes were closely associated with synaptic protrusions stained with the actin F I G U R E 3 Glycogen in siRNA transfected SH-SY5Y cells, Tendamistat or amyloid beta-42 stimulated SH-SY5Y cells and hiPSC-derived neurons. (a-c) show representative images of SHSY5Y cells immunostained against glycogen (magenta) and cell nuclei marker DAPI (green) after being transfected with α-amylase siRNA (siAMY) and non-targeting siRNA (siCtrl) (a), stimulated with Tendamistat for 2 h (Tenda 2 h) and vehicle control (Ctrl) (b), and stimulated with Amyloid beta 42 oligomers (Aβ42O) and vehicle control (Ctrl) (c). Image in (d) shows glycogen (magenta) and DAPI (green) in neurons derived from hiPSC with mutation in PSEN1 (L150P) and isogenic control (L150P-GC). Scale bar indicates 15 μm. Graph in (e) shows significant larger glycogen area/cell in SH-SY5Y cells transfected with siAMY compared to siCtrl). Graph in (b) demonstrates significantly larger glycogen area/cell in SH-SY5Y cells stimulated with Tenda. 2 h compared compared to Ctrl.
Graph in (c) shows significantly smaller glycogen area/cell in Aβ42O stimulated SH-SY5Y cells compared to Ctrl. Graph in (d) demonstrates significantly smaller glycogen area /cell in L150P compared L150P-GC. Statistical analysis was done using t test, and data are presented as mean ± SD, ***indicates p-value < 0.001, **indicated p-value < 0.01, and *indicates p-value < 0.05

| Inhibition of alpha-amylase alters spontaneous calcium oscillations in neurons
As our results indicate α-amylase is implicated in glycogen degradation and since it appears to be localized proximal to synapses, it is F I G U R E 4 Cellular localization of alpha (α)-amylase in primary mouse neurons. Image in (a) shows immunofluorescent staining of α-amylase with antibody directed against salivary α-amylase (green) and neuronal marker microtubule-associated protein 2 (MAP2) (magenta). The α-amylase immunoreactivity is seen in the cell body (arrowhead) and along the dendrites (arrow). Confocal image in (b) shows a primary mouse neuronal process where α-amylase staining (green) and phalloidin (magenta) are in close association. Confocal image in (c) shows immunofluorescent staining of primary mouse neuronal process where α-amylase (green) and synaptotagmin (magenta) are in close association (arrow). Confocal image in (d) shows a close association between α-amylase (green) and CAMKII (magenta) (arrow). Confocal image in (e) shows a primary mouse neuron stained against α-amylase (green) and glycogen (magenta). The white squares indicate magnification of the area seen in (f and g). Arrows in (f-g) indicate an close association between α-amylase and glycogen. Scale bar in (a and e) indicates 5 μm, scale bar in (d and g) indicates 0.5 μm of interest to investigate whether the enzyme is needed for neuronal communication. We therefore analyzed the calcium transients

| DISCUSS ION
Our study showed that α-amylase IR scores are significantly decreased in CA1 and ML of AD patients compared to NC and that the scores correlate negatively with both ABC (Aβ) and Braak (NFT) scores. By staining primary mouse neurons, we also showed that αamylase is present in dendritic spines and the cell body, where it co-localizes with glycogen. The α-amylase production appears to be affected by Aβ 42, as we found significantly reduced α-amylase levels in both hiPSC neurons containing the PSEN1 gene mutation and SH-SY5Y cells stimulated with oligomer Aβ 42 . A reduction of α-amylase activity, either by silencing the gene expression with siRNA or by inhibiting the activity with Tendamistat, led to an increase in glycogen area/cell in SH-SY5Y cells, but both hiPSC with PSEN1 mutation and oligomer Aβ 42 stimulated SH-SY5Y cells showed lowered glycogen area/cell compared to control. Finally, inhibition of α-amylase activity altered neuronal calcium oscillations, specifically lowering amplitudes and shortening inter-spike intervals.
In line with our previous observation (Byman et al., 2018), staining for α-amylase yielded a staining pattern resembling dendritic spines in CA1, and the immunoreactivity of these structures in this region was lowered in AD patients. The α-amylase staining also showed a previous not reported intense α-amylase staining in ML of non-demented controls, which reduced in strength in AD patients.
The grained pattern could reflect the high density of dendrites in ML, which extends from the granular cells of the dentate gyrus (Amaral et al., 2007). These granular cells and the glutamatergic pyramidal neurons located in CA1 are known to be important for learning and memory (Lopez-Rojas & Kreutz, 2016), which raises the question whether α-amylase has a function within synapses of these cells related to learning and memory formation. To further verify that αamylase is truly found in synapses and since it is difficult to clearly visualize synapses in tissue, we analyzed α-amylase in isolated primary mouse neurons, a neuronal model with functional synapses. Our results showed that α-amylase is found associated with dendritic protrusions positive for the F-actin marker phalloidin as well as the synaptic markers synaptogamin-1 and CAMKII. The former synaptic marker is commonly known as a regulator of the SNARE complex in pre-synapses but is also found in CA1 post-synapses where it regulates LTP-dependent exocytosis of AMPA receptor (Wu et al., 2017).
CAMKII, which is foremost found in post-synapses, is also known to be implicated in LTP and memory formation (Byth, 2014;Sacchetto et al., 2007). Of note, the imaging techniques used in this study are not sufficient enough to specify whether α-amylase interacts with these synaptic markers or whether α-amylase is specifically localized in pre-or post-synapses or found in both compartments.
The reduction in α-amylase IR found in our postmortem study could be due to the severe loss of neuronal synapses and dendritic spines seen in AD patients (Pereira et al., 2021). But since the α- it may well be that Aβ has a direct negative impact on α-amylase expression. The lowered α-amylase concentration found in SH-SY5Y cells after stimulation with Aβ 42 oligomers support this idea.
Our finding of close association between α-amylase and glycogen granules in the cell body and processes of primary mouse neurons are interesting given that the main function of α-amylase throughout the body is to degrade polysaccharides (Singh & Guruprasad, 2014).
It is thus not unlikely that α-amylase has a similar role in neurons, degrading glycogen in order to free glucose needed for processes in the different neuronal compartments. Indeed, when we silenced α-amylase in SH-SY5Y cells using siRNA, we noted a significant increase in glycogen compared to control-treated cells. Similar findings where seen when SH-SY5Y cells were incubated with Tendamistat, a polypeptide that tightly binds to α-amylase and inhibits its enzymatic activity (Vertesy et al., 1984). In view of these findings, we speculate that α-amylase has a role in neuronal glycogenolysis, but whether this role is direct (i.e., degrades glycogen) or indirect (i.e., affects enzymes, kinases, and proteins implicated in glycogenolysis) remains to be investigated. Nevertheless, it is important to mention that although our studies point toward a glycogen degrading role for α-amylase, we do not exclude the possibility that the enzyme has other, for us unknown, functions and has its impact on neurons by, for example, interaction with other proteins and ions.
To our surprise, even though oligomer Aβ 42 stimulated SH-SY5Y cells showed lowered α-amylase concentration, glycogen load was equally lowered in these cells. Of note, patients with AD show both reduced glycogen load (Bass et al., 2015) and increased glycogen synthesis-regulating GSK3beta (Reddy, 2013) which is shown to lower production of glycogen (Bass et al., 2015;Beurel et al., 2015). The Aβ-induced reduction of both glycogen and α-amylase in neurons suggests that there is a doubled impact on glycogen metabolism, that is, both a reduction in degradation and formation of glycogen. Such a scenario is of interest given the fact that neuronal glycogen is specifically important in LTP (Duran et al., 2019) and it is therefore tempting to speculate that the α-amylase reduction in combination with lowered access to glycogen plays a role in the cognitive decline seen in AD patients. This idea would be in line with our previous study demonstrating lower hazard ratio for developing AD as well as significantly better results on episodic memory test in individuals with high AMY1A copy number variants (Byman et al., 2020). In view of the hypothesis, we expected our calcium imaging analysis to show an elimination of neuronal activity after inhibiting α-amylase activity with Tendamistat. To our surprise, we instead found a non-significant increase in frequency of calcium oscillations as well as lowered amplitude of intracellular calcium concentrations and inter-spike intervals. It thus appears as if α-amylase is not crucial for neuronal activity per se, but a lack of the enzyme induces an alteration which may be pathological. From this perspective, it is interesting that a recent study demonstrates lowered amplitude of intracellular calcium and higher frequency of calcium oscillations in motor cortex neurons of transgenic JNPL3 tauopathy mice (Wu et al., 2021). The calcium dyshomeostasis in these mice was linked to a selective downregulation of AMPA receptors, which regulates formation of LTP via postsynaptic abundance (Diering & Huganir, 2018). Whether a shortage of free glucose molecules affects AMPA receptor trafficking and calcium oscillation in neurons is not (to our knowledge) known, but oxygen / glucose deprivation (a model for ischemia) has been shown to both increase glutamate at synapses to excitotoxic levels (Sattler et al., 2000) and decrease AMPA receptors (Dennis et al., 2011). It is thus tempting to speculate that the calcium dyshomeostasis seen in the Tendamistat treated neurons is due to a shortage of glucose caused by altered glycogen degradation, which in the long run could affect AMPA-dependent LTP formation.
Our work has some limitations that need to be addressed.
The glycogen antibody (ESG1A9) used in our study has only been evaluated in few studies and given that there is still some debate whether healthy neurons produce and store glycogen, it is important to discuss the antibody's specificity. The ESG1A9 was produced in Professor Hitoshi Ashida laboratory and was evaluated using an ELISA with 15 different glycogen species as baits (6 from

| CON CLUS ION
To conclude, our study suggests that α-amylase can be found in both human and mouse neuronal synapses and that loss of neuronal αamylase activity increases neuronal glycogen load and alters neuronal activity. Our study further shows that neuronal α-amylase is reduced in AD patients and indicates that this loss may be due to a direct impact of Aβ 42 . Based on these results, we hypothesize that α-amylase plays a role in glycogenolysis important for neuronal activity and that loss of α-amylase in AD patients partly underlie the cognitive decline seen in these patients.
Every day the media were changed, and the cells were passaged every third day for about 2 weeks using 0.5 mM EDTA (B52, Thermo and PhosSTOP (Roche, Basel, Switzerland).

| Tendamistat stimulation
Spike detection threshold was set at 10% over baseline. In order to process spike height and inter-spike intervals, silent neurons were omitted as these interfere with amplitude and interval measurements. Data were then exported and analyzed in GraphPad Prism.

| Statistical analysis
Statistical analysis was performed using GraphPad Prism (version

ACK N OWLED G EM ENT
The authors thank Johanna Christensson for technical support, Ulrika

CO N FLI C T O F I NTE R E S T
The authors declare that they have no competing interests.

AUTH O R CO NTR I B UTI O N S
EB co-designed the study, carried out the siRNA studies, RT-qPCR, ELISA immunocytochemical stainings and analyzed the data. IM performed the calcium imaging and data analysis. HH generated the hiPSC. NBB collected the brain tissue samples and performed the neuropathological evaluation. GG and KKF supervised the calcium imaging and hIPSC generation, respectively, and edited the manuscript. MW co-designed and coordinated the study, performed immunohistochemical stainings and edited the manuscript. All authors read and approved the final manuscript.

PA RTI CI PA NT CO N S ENT S TATEM ENT
Written informed consent for the use of brain tissue and clinical and neuropathological data for research purposes was obtained from all donors or their next of kin in accordance with the International Declaration of Helsinki and Europe´s Code of conduct for Brain Banking.

DATA AVA I L A B I L I T Y S TAT E M E N T
All data generated or analyzed during this study are included in this published article.