Moderate decline in select synaptic markers in the prefrontal cortex (BA9) of patients with Alzheimer’s disease at various cognitive stages

Synaptic loss, plaques and neurofibrillary tangles are viewed as hallmarks of Alzheimer’s disease (AD). This study investigated synaptic markers in neocortical Brodmann area 9 (BA9) samples from 171 subjects with and without AD at different levels of cognitive impairment. The expression levels of vesicular glutamate transporters (VGLUT1&2), glutamate uptake site (EAAT2), post-synaptic density protein of 95 kD (PSD95), vesicular GABA/glycine transporter (VIAAT), somatostatin (som), synaptophysin and choline acetyl transferase (ChAT) were evaluated. VGLUT2 and EAAT2 were unaffected by dementia. The VGLUT1, PSD95, VIAAT, som, ChAT and synaptophysin expression levels significantly decreased as dementia progressed. The maximal decrease varied between 12% (synaptophysin) and 42% (som). VGLUT1 was more strongly correlated with dementia than all of the other markers (polyserial correlation = −0.41). Principal component analysis using these markers was unable to differentiate the CDR groups from one another. Therefore, the status of the major synaptic markers in BA9 does not seem to be linked to the cognitive status of AD patients. The findings of this study suggest that the loss of synaptic markers in BA9 is a late event that is only weakly related to AD dementia.


Results
Western blot optimization. In this study, 7 biomarkers were detected by western blot. To optimize the western blot detection of the various biomarkers, we compared ECL to IR revelation (Fig. 1). Various amounts of total BA9 extracts (1-20 µg protein) were loaded on SDS-PAGE gels and blotted for VGLUT1 or VGLUT2 detection by ECL (Fig. 1A,B) or IR detection (Fig. 1C,D). As seen in Fig. 1, for both VGLUT1 and VGLUT2 the dynamic range of the ECL detection was comprised of between 1 and 5 µg of protein. In contrast, the IR detection was linear between 1 and 20 µg of cortical protein. IR detection linearity was also established within this range of protein concentration for the following proteins: α-tubulin, synaptophysin, PSD95, VIAAT and EAAT2 (Fig. 1E-I).
Consequently, the WB studies were conducted with IR detection rather than ECL detection (see examples of blots in Fig. S1A-C). The gender ratio (f/m) varied from 0.8 (CDR0.5) to 4.4 (CDR3) among the groups ( Table 1). The distribution of men and women according to their CDR groups and categories of age is shown in Fig. 2. Some CDR/age groups were composed exclusively of men (for example CDR0 (50-60 years old (yo)), CDR1 (70-80 yo), CDR4 (60-80 yo)) or of women (for example CDR0, CDR2 and CDR 5 (>100 yo)). Consequently, the means of the different neuronal markers for men and women could not be compared among the different CDR/age groups. Therefore, marker expression was analysed in all men vs. all women (Fig. S3). A gender effect was never observed, irrespective of the marker studied. The lowest p-values that were found concerned somatostatin expression (t-test p = 0.12). Hence no differences were shown between women and men in the overall expression of the different markers studied.

Effect of cognitive impairment on glutamatergic markers in BA9.
We then compared the amounts of various presynaptic (VGLUT1 and VGLUT2), postsynaptic (PSD95) and glial (EAAT2) markers of glutamatergic transmission in BA9 areas of the subjects stratified according to their CDR scores. As illustrated in Fig. 3, inter-individual differences were important and precluded the use of these markers to diagnose dementia. VGLUT2 and EAAT2 were not affected by dementia (Kruskal-Wallis test, p > 0.05; Table 2). In contrast, VGLUT1 and PSD95 were significantly reduced (Kruskal-Wallis test, p = 0.0001 and 0.0118, respectively; Table 2). A Kruskal-Wallis post hoc test was performed to identify the CDR groups that were most affected ( Fig. 4): VGLUT1 and PSD95 protein expression levels were decreased in individuals with severe dementia (CDR5), but only VGLUT1 expression was significantly decreased in individuals with a CDR score of 3.
We assessed VGLUT1-positive terminals in BA9 grey matter samples from controls ( Fig. 5A,C) and patients (Fig. 5B,D) using immunohistochemistry. As previously reported in the human cerebral and cerebellar cortices 30,51 , the black precipitate formed by VGLUT1 immuno-positive terminals appeared to be very densely packed. Therefore, VGLUT1-positive puncta in all layers of BA9 were hardly quantifiable on the immunostained sections. Interestingly, and as previously reported 30 , the density of VGLUT1-IR seemed to be uniformly decreased in patients (see Fig. 5). We observed no perinuclear accumulation of VGLUT1-positive immuno-material. Therefore, in 2 individuals with cognitive impairment, the density of VGLUT1 per terminal appeared to be decreased in grey matter areas.
Effect of cognitive impairment on GABAergic markers in BA9. We then inspected two GABAergic markers (Fig. 6). VIAAT, the GABA/glycine vesicular transporter, is present in all GABAergic terminals. In contrast, somatostatin is expressed by a restricted population of GABAergic interneurons that densely innervate pyramidal cells in the frontal cortex 52,53 . VIAAT expression levels were found to be similar across the CDR scores of 0-4 (Figs 4 and 6). Only patients with a CDR score of 5 showed a significant decrease in VIAAT protein expression (−21.5%, Kruskal-Wallis test, p < 0.05). Somatostatin levels remained stable in samples associated with CDR scores below 5. However, confirming previous findings 37 , when dementia severity increased to a CDR score of 5, somatostatin expression significantly decreased in BA9 (−42.4%, Kruskal-Wallis test, p < 0.05, Figs 4 and 6).
Effect of cognitive impairment on synaptophysin and α-tubulin in BA9. Synaptic loss is commonly considered as one of the earliest hallmarks of AD. We, therefore, quantified synaptophysin expression by immunoblotting BA9 extracts. Surprisingly, synaptophysin was minimally affected by the progression of dementia (Fig. 7B). Synaptophysin expression levels significantly decreased only in subjects with CDR scores of 5 (−12.1%, Kruskal-Wallis test, p < 0.05, Fig. 4).  Table: CDR: clinical dementia rating; N, size of the sample; PMI(min): postmortem interval in minutes; AOD: age of death; f/m: females/males. One hundred seventy-one brain samples out of the 182 individuals were used and are presented above. Alpha-tubulin, the major constituent of microtubules that is expressed by all brain cells, is often used as a loading control in western blots. As shown in Fig. 7C, α-tubulin values decreased moderately during the progression of dementia. A maximal decrease was observed in subjects with a CDR score of 3 (−18.3%, Kruskal-Wallis test, p < 0.05, Fig. 4). Therefore, the use of α-tubulin to normalize western blot in AD samples should be considered cautiously.
Correlation Analysis. The large number of subjects in the present study (n = 171) allowed us to perform a polyserial correlation analysis between all the variables and the CDR scores ( Fig. 8, dashed squares on first line). As previously reported with a more limited number of subjects 30 , VGLUT1 had the strongest correlation coefficient with the CDR scores (ρ = −0.41, Fig. 8), followed by somatostatin (ρ = −0.33), VIAAT (ρ = −0.30) and ChAT (ρ = −0.27). The correlation coefficients (ρ) between the CDR scores and VGLUT2 and EAAT2 were very close to 0.
Principal Component Analysis. As described above, the 9 markers assessed in the BA9 samples appeared to be only weakly associated with early stages of cognitive impairment (CDR 0-4). Thus, we explored the possibility of a linear combination of the different markers segregating the different CDR groups. Therefore, we combined 7 markers for which a difference between the groups was detected (VGLUT1, VIAAT, somatostatin, ChAT, α-tubulin, synaptophysin and PSD95 (Fig. 4)) into a Principal Component Analysis (PCA 54 ), (Fig. 9) to further explore our dataset. As shown on the scree plot (Fig S4), the first two principal dimensions explained 56.4% of the total variance of these 7 markers. Dimensions 1 and 2 accounted for only 40.6% and 15.8% of the variance, respectively. VGLUT1 (≈23%), followed by VIAAT and PSD95 (both approximately 18%), were the major contributors to the variance in the first dimension (Fig. 9A). This suggested that these 3 markers are somewhat involved in cognitive impairment, albeit weakly. The variance of the second dimension was mainly influenced by α-tubulin (≈32%), ChAT (≈25%), and synaptophysin (≈22%) (Fig. 9B). As shown on the PCA biplot (Fig. 9C), where the samples are projected onto the first two principal components, the seven CDR groups overlapped widely. This absence of clear clusters indicates that the linear combination of the biomarkers used in this study could not be

Discussion
Synaptic loss has long been considered among the key hallmarks of AD 8,12,13,55 . The aim of the present study was to investigate the status of glutamatergic, GABAergic and cholinergic synaptic markers in a large cohort (171 subjects) of samples from prefrontal cortex (BA9) extract stratified by the CDR scores of the subjects. BA9 is a major integrative cognitive area of the prefrontal cortex, and previous results have suggested that it is strongly affected in dementia 30,56,57 .
Whether synapses are lost in the cortex during normal ageing has been a matter of controversy 8,58 . The group of control subjects (CDR0) included 38 individuals aged 59-102 yo. As illustrated in Fig. S2, normal ageing had no effect on the expression of VGLUT1, VGLUT2, PSD95, EAAT2, VIAAT, somatostatin, ChAT, synaptophysin and α-tubulin. Interestingly, in our control group, most of the biomarkers were insensitive to a PMI ranging between 240 and 1437 minutes. Only PSD95 values were correlated with PMI (Pearson, R 2 = 0.2573, p < 0.001). These observations suggest that, while presynaptic proteins are rather stable, dendritic spines are more labile in post-mortem tissues.
The investigated synaptic markers were either negatively correlated to or not affected by dementia. None of their expression levels were significantly increased during the progression of dementia. These data confirm and extend previous findings showing that AD and dementia result in a loss of synaptic proteins (for review see ref. 12 ). Synaptic proteins are not all equally affected by AD 12,14 or AD-like pathology in mouse models 15,27,28 . However, the extent of the decrease in the biomarkers observed here was less pronounced than previously described (see for example 12,30 ). In the present study, at the most severe stages of dementia, VGLUT1, VIAAT and synaptophysin were reduced by only 26.4%, 21.5% and 11.9%, respectively. Therefore, synaptophysin, a major and widespread synaptic protein, was only minimally altered in the BA9 samples of the most severely demented patients (CDR5). A severe loss of cholinergic neurons has been documented in the temporal lobes of AD patients 42 . In the prefrontal lobes, Kashani and colleagues (2008) described a 30% decrease in ChAT expression levels in BA9. These findings were confirmed in the present study since we observed a 34.6% decrease in the expression levels of ChAT during the latest stage of dementia (CDR5).
GABAergic interneurons control and synchronize the activity of glutamatergic networks and are instrumental in information processing and memory [59][60][61] . In the CDR5 group, GABAergic terminals labelled with VIAAT declined in the same proportion as VGLUT1 (≈−20%). Therefore, the glutamate/GABA balance seems to be globally preserved in AD patients. However, the sharpest decline observed in the present study concerned somatostatin (−42.4%), a finding that was in line with previous investigations showing a reduction ranging between 25 and 66% 38,39,62 . VIAAT is present in all GABAergic terminals, whereas somatostatin is only found in a subpopulation of inhibitory interneurons (also known as Martinotti cells 63 ). Our results, therefore, suggested that different subpopulations of GABAergic interneurons may display differential sensitivity to AD, and therefore, the glutamate/GABA coupling may well be altered in patients with severe dementia, as was already shown in AD mouse models [48][49][50] .
Glutamatergic transmission has been the focus of numerous studies in the field of AD 64,65 . Canonical glutamatergic neurons from the human cortex express either VGLUT1 or VGLUT2 51 . In previous studies, cortical VGLUT2 was either severely affected by AD and dementia (−50% 30 ) or unaffected 26 . In the present study, we confirmed that cortical VGLUT2 is not affected at any stage of dementia (−0.3%). In the cortex, VGLUT2-immunopostive terminals emerge from layer V pyramidal neurons and from thalamo-cortical axons 51 . These neuronal pathways appeared to be minimally affected by dementia and AD.
EAAT2 is essentially expressed in astrocytes that are spared during normal brain senescence but are associated with degenerating neurons and senile plaques 66 . The status of EAAT2 expression in the AD brain has not been  [67][68][69] . We found no modification of EAAT2 in the BA9 samples of AD subjects with various levels of dementia. Therefore, the excitotoxicity hypothesis of AD [23][24][25] , which is related to a decreased clearance and reduced expression of EAAT2, was not supported by the findings of the present study. VGLUT1, the major subtype of vesicular glutamate transporters, supports 80% of vesicular glutamate uptake in the mouse brain 70,71 . In the human brain, this protein is expressed mostly (but not exclusively) by cortical neurons 51 . VGLUT1 was previously found to be severely decreased in AD and to be highly correlated with cognitive status 26,30 . These results were only partially reproduced here since we only observed a 26.4% reduction in VGLUT1 expression levels. As shown in Fig. 3, there was an important interindividual dispersion of the VGLUT1 expression levels. For instance, in the control group, the VGLUT1 expression levels varied between 10 and 25 AU and between 7 and 20 AU in the most demented subjects (CDR5). This dispersion could either reflect an actual interindividual variability or could be related to pre-mortem conditions. The mRNA integrity number (RIN) and the pH level of the tissue are parameters that could influence the density of membrane proteins. Future studies should investigate the impact of RIN and pH on the stability of VGLUT1, VGLUT2, PSD95, VIAAT or synaptophysin expression. Nonetheless, there was a large overlap between controls and patients that clearly precluded the use of VGLUT1 as a reliable biomarker of dementia.
Interestingly, the decrease in VGLUT1 expression was more sustained than the decrease in synaptophysin expression (−11.9%). This result suggests that VGLUT1 and synaptophysin are only partially co-localized. Alternatively, and more likely, this finding could reflect the fact that pathological ageing differentially regulates   2 synaptic markers. Immunohistological analysis of the BA9 sections ( Fig. 5 and 30 ) suggested that in grey matter areas, a decline in VGLUT1 density per terminal occurred rather than a reduction in the number of VGLUT1-immunopositive terminals. These observations suggest that, in AD, synaptic terminals are physically spared but functionally impaired. However, given the dispersion of the VGLUT1 expression levels, these observations should be interpreted with care and deserve further investigation to be confirmed.
PSD95-positive dendritic spines are often apposed to VGLUT1-postive terminals 72 . Interestingly, in the BA9 samples of demented patients (CDR 5), PSD95-IR dendritic spines exhibited the same extent of reduction as the VGLUT1-positive terminals (≈26%). Furthermore, it should be highlighted that these two markers were highly intercorrelated (r = 0.5, p = 0 Fig. 8). Therefore, AD seemed to impact VGLUT1-positive terminals and associated dendritic spines similarly.
Serial correlation and PCA confirmed that, among the synaptic variables assessed in this study, VGLUT1 was the best potential biomarker of dementia 30 . However, the magnitude of its decrease was half of what we initially reported in a smaller sample 30 . The way in which a 26% reduction in VGLUT1 expression impacts glutamatergic transmission is not clear. Data from the literature has suggested that vesicular accumulation of glutamate  71,73,74 . It has been suggested that a single VGLUT1 protein per vesicle could be sufficient to fill synaptic vesicles and maintain a normal quantal size 71,73,74 . However, using VGLUT1 heterozygous mice, Balschun and colleagues found a reduction in hippocampal LTP accompanied by specific spatial memory deficits 33 . Thus, it is difficult to evaluate how a 26% reduction of VGLUT1 expression could alter glutamatergic transmission in the BA9 areas of demented subjects. Our PCA results supported a minimal effect of such a reduction. However, it was recently suggested that VGLUT1 performs additional synaptic functions of the known vesicular glutamate loading. Indeed, VGLUT1 is involved in the intersynaptic exchange of synaptic vesicles 75 . Whether a limited decrease in VGLUT1 expression impacts synaptic vesicle exchanges and whether synaptic vesicle exchange directly impacts neurotransmission are still open questions.
Overall, the results of the present study suggest that the loss of synaptic proteins in crude BA9 extracts is limited. However, our data should be interpreted with care since this study analysed the total homogenate of the BA9 grey matter. This approach allowed us to quantify various biomarkers in a large number of samples. However, it precluded a more specific layer-by-layer synaptic analysis. Therefore, our findings do not rule out the occurrence of subtle lamina-specific changes in synapses. However, our results suggest that there is not a dramatic change in synaptic markers in cortical grey matter.
In future studies, it will be important to assess whether synaptic markers are more profoundly altered in other brain areas, such as the hippocampus or entorhinal or temporal cortices. Interestingly, our data show that, with a few exceptions, the majority of significant changes were observed only in severe dementia (CDR5). This and Pearson (r) correlation coefficients between the CDR scores and VGLUT1, VGLUT2, somatostatin (Som), ChAT, VIAAT, EAAT2, α-tubulin, synaptophysin (syn) and PSD95 were determined. The various correlation coefficients were colour coded from dark blue for a correlation of 1 to a deep red for −1 (and white for 0). Polyserial correlations between the markers and CDR scores are represented by the dashed boxes on the first row. Pearson correlations are presented between the different markers. VGLUT1 shows a negative correlation with the CDR scores, which was higher than that of somatostatin, VIAAT and synaptophysin (r < −0.3). The Pearson correlations highlighted a positive correlation between VGLUT1 vs. α-tubulin, PSD95 and VIAAT (r > 0.46) and between PSD95 vs. VIAAT and synaptophysin (r > 0.5). The p-values of the Pearson correlation analyses are presented in Table 2. observation suggests the following: (i) something dramatic occurs at synapses during the transition from CDR 4 to CDR 5, and (ii) most of the synaptic variables examined in this study are not causal in earlier stages of dementia (CDR 0.5-3). PCA was used to determine the key variables in the present multidimensional dataset. As shown in Figure S4, the first component explained 40.6% of the total variance of the seven biomarkers that varied to some extent among the CDR groups. Along this dimension, VGLUT1, VIAAT and PSD95 seemed to be the most influential. However, as can be visualized on the PCA biplot in which the dataset was projected onto the two first components (Fig. 9C), a linear combination of variables could not explain the CDR score of a patient. Therefore, the various synaptic markers used in this study, whether taken separately or in combination, did not seem to play a pivotal role in the development of dementia.
These observations suggest that AD is a complex disease involving either a large number of limited changes in multiple markers or non-linear interactions among these markers. Our data show a limited effect of AD on markers of terminals, dendritic spines and astrocytes in BA9. Overall, the present investigation only weakly supported the notion that dementia is associated with a sustained age-dependent loss of synaptic markers in BA9.
Immunohistochemistry experiments (IHC) were conducted on the brain samples from 3 controls and 3 patients provided by the Douglas-Bell Canada Brain Bank (DBCBB; Table 3).
Preparation of brain tissue samples. BA9 grey matter was dissected from flash-frozen coronal sections pulverized at −80°, as described previously 3 . Frozen powdered tissues were homogenized with a Potter-Eveljhem and then sonicated (3 × 10 sec Power 60 Vibra-Cell TM 72446, Sonic USA) in phosphate buffered saline buffer (PBS) containing protease inhibitors (complete, Roche, France). Total BA9 extracts were stored in aliquots at −80 °C until usage; aliquots were unfrozen only once. Protein concentrations were measured with the Bio-Rad protein assay kit (Bio-Rad, France). Protein concentrations in post-mortem extracts from the controls (CDR 0) and the subgroups of different CDR scores were not significantly different (data not shown).
Western blots were visualized with either enhanced chemiluminescence (ECL) or infrared detection (Fig. 1). For ECL, bound primary antibodies were detected with horseradish peroxidase (HRP)-conjugated anti-rabbit or anti-mouse IgG antisera (Sigma, 1:20,000) and visualized by chemiluminescent detection (SuperSignal West Dura, Pierce, France) on films (Biomax MR Film Kodak) ( Table 4). For infrared (IR) detection, bound antibodies were detected with secondary anti-mouse or anti-rabbit IgG antiserums conjugated to IRDye 700 (red display, 1:5,000) or IRDye 800 (green display, 1:5,000) (Rockland) with an infrared imager (Odyssey, Li-Cor, France) ( Table 4). ECL films were digitized using a Umax PowerLook 1100 scanner (Umax, Willich, Germany).   Optical densities of the ECL or infrared images were determined in arbitrary units using the MCID software (Imaging Research, St. Catharines, Ont., Canada). Samples for each biomarker and subject were run in 4-5 replicates and the average of the replicates was calculated. Standard proteins commonly used as loading controls, such as α-tubulin, β-actin, and glyceraldehyde-3-phosphate dehydrogenase (GAPDH) were found to decrease in AD (for α-tubulin see Fig. 6; β-actin and GAPDH not shown). Therefore, samples were not normalized to classic standards but rather to one standard sample of a control subject that was loaded on each SDS-PAGE gel. Examples of representative blots for the entire cohort and for each biomarker are shown in Supplementary Figure S1, with the control sample mentioned above identified by *.
Choline acetyltransferase activity. The choline acetyltransferase (ChAT) activity was measured in the prefrontal cortex BA9 extracts 80 as previously described 30 .
Somatostatin radioimmunoassay. Somatostatin-like immunoreactive material was measured by radioimmunoassay as previously described 39,81 . Data availability statement. The datasets generated and analysed during the current study are available from the corresponding authors upon reasonable request.