Inhibitory synapse loss and accumulation of amyloid beta in inhibitory presynaptic terminals in Alzheimer's disease

Synapse degeneration in Alzheimer's disease (AD) correlates strongly with cognitive decline. There is well‐established excitatory synapse loss in AD with known contributions of pathological amyloid beta (Aβ) to excitatory synapse dysfunction and loss. Despite clear changes in circuit excitability in AD and model systems, relatively little is known about pathology in inhibitory synapses.


INTRODUC TI ON
Alzheimer's disease (AD) is the most common cause of dementia in the elderly, affecting approximately 50 million people worldwide [1].
Brain changes in AD begin years before symptom onset and include accumulation of amyloid beta (Aβ) peptide in extracellular plaques, aggregation of tau protein in neurofibrillary tangles and neuropil threads, gliosis and extensive loss of neurons and synapses [2]. Of the neuropathological changes that occur in AD, synapse loss correlates most strongly with cognitive decline [3][4][5].
To date, most of the work examining synapse degeneration in AD has been focused on excitatory synapses, which clearly degenerate extensively in AD, particularly near amyloid plaques [6,7]. In an unbiased proteomic study of synaptic fractions from human AD and control temporal and occipital cortex, proteins involved with glutamatergic signalling were previously observed to be significantly decreased whilst no changes were detected in inhibitory synaptic proteins [8]. Epileptiform activity and seizures are observed in some people with AD and network hyperactivity is observed in animal models of amyloidopathy, leading to testing of anti-epileptic drugs as potential treatments [9]. Whilst it is unlikely that epilepsy is a causative factor in developing AD, there is ample evidence to suggest that pathological changes in AD brain lead to disrupted excitatory/ inhibitory balance. This could involve disruption to inhibitory synapse structure or function, but there is some controversy about the role of inhibitory synapse damage in AD with some studies finding decreases in inhibitory synapse density or function [10][11][12][13][14], some finding increases [15][16][17], some finding a mix of increases/decreases depending on marker or disease stage [18,19] and some finding no change [20,21] (Table 1).
In model systems including acute rodent brain slices, in vivo injection of Aβ into rodent brain, and transgenic mouse models with plaque pathology, there is abundant evidence showing that oligomeric Aβ causes excitatory synapse dysfunction and loss [6,7,[25][26][27].
Our previous data using high-resolution array tomography imaging suggests that accumulation of oligomeric Aβ within individual presynaptic and postsynaptic terminals is associated with excitatory and total (excitatory and inhibitory) synapse loss around plaques in human temporal cortex of AD patient samples [6,28,29]. There are contradictory results examining inhibitory synapse loss or dysfunction downstream of Aβ in model systems with some studies finding increased inhibitory postsynaptic potentials [15] or γ-aminobutyric acid (GABA) conductance [16] whilst others found decreased inhibitory postsynaptic potentials [13] and impaired inhibitory signalling [11]. Some of these discrepancies could be due to differences in model systems or ages, but discrepancies remain even in very similar studies. For example, in transgenic APP/PS1 mice, one group observed an increase in inhibitory synapses in young mice before pathology developed and a later stage loss of inhibitory synapses [19], whilst in a very similar APP/PS1 model another group found no change in inhibitory synapse density at the same late age (12 months) [20]. These two APP/PS1 studies examined different brain regions (hippocampus and somatosensory cortex, respectively) with slightly different techniques. From these data, the role of Aβ in inhibitory synapse changes remains unclear in animal models and has not been explored yet in detail in human brain. In this study, high-resolution array tomography imaging was used to test the hypothesis that inhibitory synapses degenerate in human AD brain and that pathological forms of Aβ may contribute to this process.

Human cases
In this study, postmortem human brain tissue from the inferior or middle temporal gyrus (BA20/21) and primary visual cortex (BA17) was examined. Experiments were approved by the Edinburgh

Tissue processing and immunohistochemistry
For determining inhibitory neuron density, tissue blocks from each region of interest were fixed in 10% formalin for at least 24 h and then dehydrated and embedded in paraffin wax. Sections were cut at 4 μm thickness on a Leica microtome and collected on glass slides.
Sections were de-waxed and re-hydrated through a series of xylenes and graded ethanol solutions followed by antigen retrieval in 10 mM citrate buffer (pH 6.0) in a pressure cooker for 20 min. Sections Primary antibody solution was washed off with TBS and secondary antibody solution containing biotinylated anti-rabbit antibody at 1:500 dilution was applied for 1 h (antibody details are in Table 3). For examining inhibitory synapses and synaptic Aβ, samples from each region of interest were fixed and embedded for array tomography as described previously [30,31]. and Aβ with Alexa Fluor 488 (antibody details are found in Table 3).

Stereology estimation of inhibitory neuron density
Inhibitory neuron density was estimated in GAD65/67 stained paraffin sections using a Zeiss AxioImager Z2 with StereoInvestigator software (MicroBrightField). Each cortical layer was outlined as a region of interest at 1.5× magnification and tile scans were acquired at

Inhibitory neuron loss in Alzheimer's disease
To determine whether inhibitory neurons are lost in our AD cases, two brain regions were examined: the inferior/middle temporal gyrus (BA20/21) which is heavily affected by pathology at the end stage of disease and the primary visual cortex (BA17) which accumulates tau pathology only at very late stages of disease and has less pathology than the temporal cortex [2]. Sections from these regions were immunostained with antibodies recognizing GAD 65 and 67. GAD is an enzyme required for the synthesis of GABA and is present in

Inhibitory synapses loss and accumulation of synaptic Aβ in AD
High-resolution array tomography imaging was used to examine inhibitory synapse density and the accumulation of oligomeric To determine whether oligomeric Aβ within inhibitory presynaptic terminals may be contributing to synapse loss, colocalization of synaptophysin, GAD and 6E10 staining was examined ( Figure 5).

DISCUSS ION
Synaptic function is the key for healthy cognition, and loss and dysfunction of synapses are associated with cognitive decline in AD [7,33]. with AD and the animal model data indicating altered excitatory/ inhibitory balance highlight the need to understand pathological changes in both excitatory and inhibitory synapses in disease [9,[35][36][37]. Here inhibitory neuron and synapse densities were examined in postmortem brain samples from people with AD and control subjects in two brain regions: inferior/middle temporal gyrus which is affected early in the disease process and primary visual cortex which is affected much later in disease. Both inhibitory neuron loss and inhibitory synapse loss were observed in AD.
Inhibitory neuron loss has been reported in both AD and mouse models of amyloidopathy (e.g., [38]). Previous studies examining inhibitory synapses in human AD brain report conflicting results with brain region and even sub-region differences ( Table 1). The only

F I G U R E 2
Inhibitory neuron density is decreased in AD. Inhibitory neuron densities in each cortical layer were determined and plotted for each case (a). The mean and standard error per group are plotted in (b). To compare groups, a linear mixed effects model was run with fixed effects of disease, brain region, cortical layer, sex and age and an interaction term between disease and brain region. Case was included as a random effect to account for multiple measures per subject. Examination of residual plots revealed that this model did not meet the test assumptions, so the data were Tukey transformed and the same model on Tukey transformed data met test assumptions. The ANOVA test on the model revealed significant effects of disease, brain region and cortical layer, no effects of sex or age, and a trend toward an interaction between disease and brain region (F = 3.14, p = 0.08), trending towards worsened loss of inhibitory neurons in temporal cortex than in visual cortex. known study examining a similar brain region to the two looked at here found no loss of inhibitory synapses even near plaques in AD in the inferior temporal cortex [20]. Both our study and the work by Mitew et al. [20] examined inhibitory presynaptic terminals, theirs stained with vesicular GABA transporter, GAD65 or GAD67 separately and ours co-stained with presynaptic marker synaptophysin and inhibitory marker GAD. Mitew et al. did observe loss of inhibitory synapses within fibrillar plaque cores, but not in the periphery as measured here. One potential reason for the discrepancy between these studies is the more refined method used here with higher axial resolution and the ability to count inhibitory synapses only when they contain both a presynaptic marker and a marker of inhibitory neurons. Our total synapse density in control subject temporal cortex (1.04 × 10 9 synapses/mm 3 ) is comparable to previous data using electron microscopy to calculate synapse density in the same brain region, which found approximately 1 × 10 9 synapses/ mm 3 with notably higher densities in male than female subjects [39].
These data lend credence to the array tomography technique to examine synapse density and point to the importance of including sex in our statistical models despite gender matching our cases and controls. Contrary to these previous findings, sex effects on synapse density were not observed in our study.
Inhibitory synapse loss was also observed in AD brain in the visual cortex, which is more pronounced than the loss in the temporal cortex. This is somewhat surprising as temporal cortex contains more pathology than visual cortex even at the end stages of disease [8]. A 30% higher inhibitory synapse density was observed in visual cortex compared to temporal cortex in control subjects, similar to our previous case study on a control subject from the Lothian Birth Cohort 1936 [40]. In monkey primary visual cortex, there is agerelated loss of both excitatory and inhibitory synapses in layer 3 [41].
Excitatory synapse degeneration in AD has been linked strongly to accumulation of oligomeric Aβ in synaptic terminals [6,7,[25][26][27], but less is known about the effects of Aβ on inhibitory synapses  Table 1). Here a small subset of inhibitory synapses containing Aβ in human visual and temporal cortices was observed with the most accumulation near amyloid plaques. Whilst excitatory synapses were not directly measured in this study, it can be inferred that the synaptophysin stained synapses that do not colocalize with GAD65/67 are excitatory. Under that assumption, an overall 5% loss of excitatory synapses and 20% loss of inhibitory synapses were observed in AD in this study (with more loss near plaques), indicating that inhibitory synapses may be particularly vulnerable to degeneration. This would be expected to alter the excitatory/inhibitory balance in these brain regions. On the other hand, 0.85% of excitatory synapses and 0.07% inhibitory synapses in our study were positive for Aβ. This is proportionally over 10-fold more excitatory than inhibitory synapses positive for Aβ, leading us to speculate that, whilst it is possible that Aβ damages inhibitory synapses, this may not be as important as the damage caused to excitatory synapses. Use of postmortem tissue allows only a snapshot of disease and Aβ could play a role in inhibitory synapse loss at earlier disease stages than were examined here.
Both microglia and astrocytes have been implicated in refinement of synaptic circuits and in excitatory synapse degeneration in AD [42].
There are very few data regarding a potential role of glia in inhibitory synapse loss, which will be an important area of future research.
In conclusion, our current data show that inhibitory synapses degenerate in two brain regions in AD and point to the need for future work to understand the mechanisms of inhibitory synapse loss. The loss of inhibitory synapses may contribute to cognitive decline and could be beneficial to target with therapeutics. where HK presented preliminary data and received useful feedback are thanked. The human brain tissue donors and their families are also thanked for their generous donations.

DATA AVA I L A B I L I T Y S TAT E M E N T
R scripts for statistics, data spreadsheets, and raw images can be downloaded from the Edinburgh DataShare repository (https://doi. org/10.7488/ds/309).