Selection of MAP participants: cognitive and neuropathological evaluations

Prior reports extensively reported the methodological approaches to perform systematic cognitive, clinical and neuropathological evaluations [9, 36]. Annual cognitive evaluations included a series of 21 standard tests, 19 of which were used for summary measures of episodic, semantic and working memory, perceptual speed, and visuospatial ability, and finally summarized into one single variable to derive a global cognitive function score [9, 37]. The mini mental state examination (MMSE) is also reported for comparison to other studies (see Table 1). A board-certified neuropsychologist blind to all pathological data reviewed test results and rated the level of cognitive impairment. A study clinician evaluated each participant and diagnosed dementia and AD following the National Institute of Neurological and Communicative Disorders and Stroke and the Alzheimer’s Disease and Related Disorders Association criteria [38] implemented as described [39]. Cognitive impairment not meeting the criteria for dementia was diagnosed as mild cognitive impairment (MCI) as described [40]. NCI refers to those without MCI or dementia [41].

The pathological examinations were made by a board-certified neuropathologist, blind to all clinical data. AD pathology (i.e. plaques and tangles) was evaluated in formalin-fixed, paraffin-embedded sections from multiple key brain regions in the frontal, temporal, parietal, and occipital lobes, as previously described [42], although only data from the dorsolateral prefrontal cortex (DLPFC) was used for statistical modeling, unless otherwise specified. Briefly, sections from all subjects and brain areas were assessed using both a modified Bielschowsky silver staining for counts of diffuse and neuritic plaques, and neurofibrillary tangle (NFTs), as described [43]. Immunocytochemistry with amyloid-β (clones 10D5 or 4G8) to quantify the percent area occupied by amyloid-β by image analysis, and phosphotau (clone AT8) antibodies – to quantify the density of tau tangles by stereology [42]. The severity and/or stage of AD in each participant was later addressed following the National Institute on Aging (NIA)-Reagan criteria, which incorporates the Consortium to Establish a Registry of Alzheimer’s Disease (CERAD) scale [44], and Braak staging [45]. Other neuropathologies, including cerebrovascular diseases (macroscopic and microscopic infarcts, arteriolosclerosis and atherosclerosis), Lewy bodies, and hippocampal sclerosis, were also examined as described elsewhere [9, 36]. Stereological approaches to account for resting, activated or total microglial cells in the DLPFC were detailed earlier [46].

Animals

APP23 transgenic mice, overexpressing a variant of human APP carrying the ‘Swedish double mutation’ KM670/671NL [33], and wild-type (WT) littermates were provided by Novartis Pharma (Basel, Switzerland) at different ages (3, 12 and 22 months old; n = 5–6 per genotype and age). Mice were killed by decapitation, and the frontal cortex was dissected and prepared for further Western blot (WB) analysis [47].

Tissue collection, immunoassays and quantification of target proteins

At the time of autopsy of MAP participants, tissue slabs from the middle-frontal gyrus (Brodmann’s area 46/9) of the DLPFC were dissected following a standard human brain atlas [48], and stored at −80 °C. The DLPFC was selected for its central role in complex cognitive tasks and contribution to age-related cognitive decline [49]. For further immunoassays, grey matter tissue was carefully sampled from each of the slabs avoiding thawing. Total homogenates from DLPFC samples (40–80 mg) were prepared in ice-cold PBS pH 7.4 following usual procedures [11]. Then, protein concentrations were determined by DC assay (Bio-Rad, Hercules, CA, USA) and samples were adjusted to equal concentrations with homogenization buffer. A greater amount of a reference MAP cortical sample (~1 g) was homogenized and prepared following the same conditions to be used as an internal control (i.e., standard sample) in the immunoblot assays.

Cortical samples from MAP participants (40 μg) or APP23/WT mice (10 μg) were resolved by electrophoresis on 10% SDS–PAGE minigels (Bio-Rad Laboratories, Hercules, CA, USA). Every gel was run with 14 brain samples, including 11 MAP participants (or 11 APP23/WT mice samples) selected randomly, and the triplicate standard samples (the reference MAP sample or a pool of WT mice) space-loaded across the gel, and a molecular weight ladder (Bio-Rad). Each cortical sample was assessed at least three times in different gels (i.e., with randomly allocated brain samples) on different days. Following electrophoretic separation, proteins were transferred to nitrocellulose membranes, then incubated overnight at 4 °C in blocking solution with anti-FADD (H-181) (1:5000; Santa Cruz Biotechnology, CA, USA) (see [50, 51] for labeling in post-mortem human brain tissue). The secondary antibody was incubated for 1 h at room temperature (1:5000 dilution; Cell Signaling). Immunoreactivity of target proteins was detected with ECL reagents (Amersham, Buckinghamshire, UK) and signal of bound antibody was visualized by exposure to autoradiographic film (Amersham ECL Hyperfilm) for 1 to 60 min, then quantified by densitometric scanning (GS-800 Imaging Calibrated Densitometer, Bio-Rad). Quantification of p-FADD and ß-actin protein contents were performed by sequentially stripping and reprobing all membranes, first for anti-phospho-Ser194 FADD (1:1000; Santa Cruz Biotechnology, CA, USA), and then for anti-ß-actin (clone AC-15) (1:10000; Sigma-Aldrich, MO, USA). For each sample, immunoreactivity of FADD or p-FADD was first divided by that of ß-actin (i.e., protein content data normalization) within the same gel, and then calculated as a percentage of in-gel standards. The mean value for each sample from at least three separate gels was used as a final estimate. During the above procedures, the experimenter was blind to the demographic, cognitive and pathological characteristics of MAP participants.

Quantification of the presynaptic proteins syntaxin-1, synaptosomal-associated protein of 25 kDa (SNAP25), vesicle-associated membrane protein (VAMP) and syntaxin-binding protein-1 (STXBP1) was performed previously by either enzyme-linked immunosorbent assay [11] or Western blotting [35], in the same brain samples. The DLPFC or overall brain immunodensities of the three SNARE proteins (i.e. syntaxin-1, SNAP25, VAMP) were z-scored and averaged to obtain a variable accounting for cortical or global synapse density, directly related to synaptopathy [52].

Immunofluorescence analysis

In an attempt to better characterize the potential relationship between FADD (i.e., cellular and anatomical localization), cognitive decline and AD pathology, 6 MAP subjects (3 pathology-free NCI subjects and 3 dementia cases with confirmed cortical AD pathology) were selected for further immunofluorescence analysis. Tissue blocks from the DLPFC (BA9) were cut coronally with a vibrating microtome (Leica, Nussloch, Germany) to a thickness of 40 μm, and floating sections were cryo-preserved in solution at −20 °C until further use. Antigen retrieval was done in 20 mM citrate buffer (pH 6.0, 80 °C, 20 min) and was followed by blocking sections and overnight incubation at 4 °C with one of the following primary antibodies: anti-FADD (H-181, Santa Cruz, 1:50), anti-NeuN (clone A60) (1:250; Chemicon, CA, USA), anti-syntaxin-1 (clone SP7) (1:1000; [53]), anti-amyloid β (clone 6 F/3D) (1:100; Dako, Glostrup, Denmark), anti-mis-folded pathologic tau (clone Alz-50) (1:500; [54]). The next day, sections were incubated with Alexa Fluor® 488-, 555- or 647-conjugated anti-mouse or anti-rabbit secondary antibodies (1:500; Southern Biotech, AL, USA). Auto-fluorescence was eliminated by incubation in 0.1% Sudan Black B and 70% ethanol solution for 15 min. Sections were mounted in gelatin-coated slides and protected with an anti-fade mounting reagent. A series of orthogonal images were captured at a resolution of 1024×1024 pixels using a LSM 5 Pascal confocal microscope (Zeiss; Jena, Germany) and were visualized using a 63x/1.2 N.A. water-immersion objective. Co-staining of FADD with NeuN and syntaxin-1 was assessed to determine presence in neurons and/or synapses, while co-staining of FADD with amyloid β or mis-folded pathologic tau determined its presence within plaques and/or tangles, respectively.

Statistical analyses

Data were analyzed and plotted with GraphPad Prism, Version 6 (GraphPad Software, CA, USA) and/or JMP. The level of significance was p ≤ 0.05. Initial comparisons of the cortical pathologic burden between clinically diagnosed groups were done by Kruskal-Wallis followed by Dunn’s test. Following WB assays, FADD and p-FADD immunodensities were normalized to corresponding β-actin values, and calculated as a percentage of in-gel standards (see [35]). While these values were used for plots, later statistical modeling procedures required a logarithmic transformation and standardization in order to obtain normally distributed measures, as confirmed by Shapiro-Wilk test. Multivariate analyses were performed to detect potentially confounding factors (e.g. demographic variables, APOE genotype, tobacco or alcohol consumption, psychotropic drug prescription, etc.) influencing cortical FADD/p-FADD levels, as well as other interesting associations of these molecules with multiple clinical, pathologic or neurochemical variables measured along the study. Among the confounding factors, only postmortem interval (PMI) significantly correlated with FADD (r = −0.278; p < 0.001) and p-FADD (r = 0.240; p < 0.003) brain values, and therefore was included as a covariate (among other variables) in all follow-up models. Differences in FADD and p-FADD immunodensities between clinically diagnosed (i.e., NCI, MCI, dementia) or pathologically graded (i.e. CERAD scaled or Braak staged) groups were assessed by analysis of covariance (ANCOVA), controlling for sex, age, years of education and PMI, followed by Tukey’s HSD test. Given the potential effects of cortical FADD levels on cognitive performance, logistic or linear regression models (controlled for sex, age, education, PMI, and APOE genotype) were evaluated with clinical dementia or cognitive function proximal to death as respective outcomes, and pathological and neurochemical variables as predictors (see [35]). Additionally, we constructed univariate random-effects models, controlled for demographics and neuropathologies as above, to study the potential influence of FADD cortical immunodensites (measured postmortem) on the cognitive decline rates of MAP participants, as previously described [12]. Note that these models assume fixed values of cortical FADD levels longitudinally, a limitation that must be considered when interpreting the results.

For WB experiments involving transgenic mice, data was analyzed with two-way ANOVA, in which genotype (WT vs. APP23) and age (3, 12 and 22 months old) were treated as independent variables, followed by multiple t-tests for two-group comparisons at each age.

ImageJ 2.0 (NIH, MA, USA) was used to determine and quantify the extent of colocalization between two immunofluorescent dyes in confocal imaging using an unbiased built-in method [55, 56].

Characteristics of MAP participants

Descriptive statistics for demographic, cognitive and pathological characteristics of MAP participants are summarized in Table 1. Out of the total of 150 MAP participants, 51 subjects presented no cognitive impairment (NCI), 42 displayed mild cognitive impairment (MCI), while 57 were clinically diagnosed with dementia (see Table 1, cognitive function proximate to death). As expected, common AD disease pathology (i.e. amyloid-β load and tau tangle density) in the DLPFC was more abundant in dementia cases, as compared to NCI participants (2.5–5.1 fold, p < 0.001) (see Fig. 1). None of these markers could separate MCI from NCI participants, while dementia cases showed greater density of phosphotau tangles than MCI participants (1.6 fold, p = 0.003) (Fig. 1b). Notably, and as previously reported in larger epidemiologic studies [7, 9] there is ample variability in all clinically diagnosed groups.

Association of cortical FADD with common age-related pathology

The total amounts of FADD protein forms (FADD and p-FADD) were quantified in the DLPFC of MAP participants and normalized by β–actin protein content. Our research group, which has been working with these specific antibodies for over a decade, characterized FADD and p-FADD specific bands with several antibodies and enzymatic dephosphorylation assays in brain tissue [50, 51, 57, 58] (see detailed revisions on these antibodies characterization in [59, 60]). As reviewed in [60], and shown in the present results, total FADD is mainly immunodetected as a 51-kDa dimeric form, while p-FADD is immunodetected as a 116-kDa oligomeric form both in rodent and human brains. The results revealed large variability among MAP participants for FADD immunodensities (interquartile range = 20–83%), and p-FADD (57–102%) as compared to that of β-actin (93–105%).

Protein expression levels of FADD species did not differ in the DLPFC of MAP participants displaying cerebrovascular diseases (including infarcts), Lewy bodies, cortical atrophy, or hippocampal sclerosis from those who did not (data not shown). By contrast, FADD levels (but not those of p-FADD) negatively correlated with cortical amyloid-β load (r = −0.197; p = 0.038), without apparent association with phosphotau density (r = −0.124; p = 0.118) (Fig. 2). The association between FADD and plaque pathology was stronger when using neuritic (r = −0.266; p < 0.001) and diffuse (r = −0.258; p = 0.006) plaque counts on the silver-stained sections, rather than amyloid-β immunodensities, which are more sensitive (Fig. 2). Consequently, subjects graded with definite AD pathology by CERAD had lower cortical FADD (but not p-FADD) immunodensities (−45%, p = 0.004) compared to plaque pathology-free participants (Fig. 3a). These differences in FADD contents were not found in association with Braak stage (Fig. 3b). Of note, the above neuropathological assessments were part of previous studies, where detailed information on image processing and quantitative approaches was reported [36, 41–43].

Additionally, greater FADD cortical immunodensity was associated with lower number of activated (r = −0.224; p = 0.018, n = 110), but not resting or total, microglial cells (Fig. 2), as characterized and quantified previously [46]. Confocal microscopy studies labeling FADD and HLA-DR-positive (activated) microglia revealed significant colocalization between both markers within microglial processes, but not in their nuclei (Additional file 1: Figure S1). This type of overlap might be attributed to the engulfment of neuronal-derived apoptotic material, and therefore activated microglia might mediate pro-apoptotic FADD clearance. By contrast, higher FADD immunodendities, but not p-FADD, were strongly associated with greater amounts of most presynaptic markers quantified in the same cortical samples in prior studies [11, 35], including syntaxin-1 (r = 0.274; p = 0.004), synaptosomal-associated protein of 25 kDa (SNAP-25; r = 0.290; p = 0.002), vesicle-associated membrane protein (VAMP; r = 0.327; p < 0.001), and syntaxin-binding protein-1 (STXBP1; r = 0.520; p < 0.001) (Fig. 2). Of note, loss of these markers is related to synaptic pathology in aging and AD [61]. An index of synapse density was estimated by averaging syntaxin-1, SNAP-25 and VAMP immunodensities (i.e., the so-called SNARE proteins) in order to obtain a variable accounting for cortical synaptopathy in MAP participants in later statistical models. As expected, this index also correlated with FADD values (r = 0.312; p < 0.001).

Association of cortical FADD with clinical dementia and cognitive function

As expected, age and the presence of Lewy bodies were associated with both higher odds of dementia and poorer cognitive performance prior to death. Surprisingly, in the present MAP subset, cerebrovascular diseases and hippocampal sclerosis were related with neither dementia nor cognitive function. However, in previous studies using a much larger MAP sample, these latter associations were indeed observed (see e.g. [12]). Interestingly, the effect of DLPFC amyloid-β load on the likelihood of dementia, which was marginally significant when the FADD data was not included in the model (data not shown), was not significant after addition of FADD suggesting a mediation effect. Thus, participants with lower FADD cortical density displayed a significantly greater likelihood of dementia (odds ratio = 0.433, p = 0.002). For example, a MAP participant with average demographic and pathologic characteristics had a 2.5 fold-higher likelihood of dementia if the cortical density of FADD is in the lower quartile versus the higher quartile. Likewise, higher FADD levels in the DLPFC were associated with better global cognitive function in MAP participants (β = 0.499, p = 0.008), while the DLPFC load of amyloid-β showed a marginal but significant effect on cognitive function (β = −0.119, p = 0.046) after adding FADD to the linear regression model. Cortical FADD immunodensity did not mediate the large effects of the DLPFC tauopathy on cognitive function; and surprisingly had no impact on the risk of clinical dementia in the current MAP sample (Table 2). Importantly, after controlling for the demographic and pathologic indices in Table 2, variations in DLPFC FADD immunodensity explained 3.44% of the total variance in cognition among MAP participants. Of note, the associations between the clinical outcomes and the DLPFC pathologic and neurochemical variables reported in Table 2 were very similar to those observed when using the overall brain measurements of amyloid and tau pathologies (data not shown).

Given that clinical diagnoses were not performed longitudinally, diagnoses of NCI/MCI/dementia were based on cognitive evaluations nearest death. Therefore, establishing an association between the present neurochemical data and the duration of the illness was not possible. Because the onset of cognitive decline was somewhat variable (see left panel in the Fig. 4), we performed random-effects models to evaluate the potential association between longitudinally ascertained cognitive decline rates and postmortem cortical immunodensities of FADD, adjusting for demographics and neuropathologies. Although a trend of faster cognitive decline rate was observed for those subjects within the lower tertile of FADD cortical values (see Fig. 4), in the statistical approaches using univariate random-effects models (controlling for all above confounders and pathologies) this neurochemical variable only showed a marginal, non-significant association with the annually evaluated cognitive function (β = 0.266, p = 0.080), possibly because the current sample size is underpowered for this type of analysis.

Differences in FADD cortical distribution in clinical dementia

In DLPFC sections from subjects (n = 3) with no cognitive impairment (NCI) and free from age-related neuropathology, most FADD intensity appeared in the neuronal cell body, especially within nuclei (see merge FADD-NeuN), and neuropil (synapses) (see Fig. 5a). In particular, about 65% of FADD signal intensity was colocalized with NeuN, while 4% of FADD signal intensity was colocalized with syntaxin-1. By contrast, in samples from MAP participants with definite AD (e.g. AD1 and AD2 subjects in Fig. 5a), there was no obvious colocalization between FADD and NeuN labelings in neuronal bodies. Interestingly, FADD appeared to accumulate in dystrophic neurites and tangle-like structures. Notably, the same results were observed for all 3 AD cases, while none of the NCI subjects presented this anomalous distribution of FADD. As these results suggested a possible role for FADD in the mechanisms of pathologic tau deposition, DLPFC sections from subjects with AD were used to study the possible colocalization of FADD with pathological tau (Alz-50) and/or β-amyloid. The results shown in Fig. 5b demonstrated the presence of FADD in tangles and in dystrophic neurites where FADD colocalized with Alz-50 immunoreactivity. FADD was also present in the dystrophic neurites accumulating around neuritic plaques (see Fig. 5b).

Decreased cortical FADD in APP23 mice with aging

Given the observed decrease in FADD content in MAP participants with clinical dementia, and the potential association with common AD pathology, we utilized APP23 transgenic mice to further explore the possible role of FADD in a common animal model of AD-like syndrome (i.e., amyloid-β plaques accumulation with age). The results showed, in parallel to the human data, decreased immunodensities of cortical FADD (normalized by β-actin content) in APP23 mice as measured by a two-way ANOVA (interaction Genotype x Age: F_2,28 = 4.43, p = 0.021). Post-hoc multiple comparisons via t-tests revealed significant decreases for adult (12 months old, p = 0.027) and aged (22 months old, p = 0.030) APP23 mice as compared to age-matched WT controls (Fig. 6).