Contributions of major tau kinase activation and phospho-tau accumulation to cortical and hippocampal tangle formation and cognition in older adults

Abstract

Aberrant activation of tau kinases (tauK) has been proposed as a major step in tau hyperphosphorylation and misfolding, and subsequent formation of neurofibrillary tangles (NFT) in Alzheimer’s disease (AD). However, evidence of tauK hyperactivation in actual AD brains is scarce and inconsistent, and their role in age-related cognitive decline remains undocumented. We evaluated activated/inhibited species of CDK5/p35/p25, GSK3α/β, and ERK1/2 as well as ten tau/phospho-tau (ptau) peptides (mapping Ser²⁰², Thr²¹⁷, Ser²⁶², Ser³⁰⁵, and Ser⁴⁰⁴ phospho-residues) by Western blot or selected reaction monitoring proteomics, respectively, in postmortem dorsolateral prefrontal cortex (DLPFC) and hippocampal samples of 150 participants from the Rush Memory and Aging Project (MAP). Regression models and mediation analyses assessed the contributions of these variables to tau phosphorylation, NFT deposition and antemortem cognitive status of MAP participants. Surprisingly, greater p25 and p35 (indices for CDK5 activation) and lower pSer^21/9-GSK3α/β (inhibited species) immunodensities were associated with lower ptau peptide amounts. Individuals with higher p25 cortical densities displayed better cognitive outcomes, particularly working memory. Statistical mediation analyses indicated that the beneficial effect of CDK5/p25 on cognition was mediated by lower densities of phospho-Thr²¹⁷-tau and NFT deposition in DLPFC, and also identified Thr²¹⁷ and Ser²⁶² as the ptau sites with greatest influence in both NFT accumulation and cognitive impairment. The present data suggest that tau hyperphosphorylation, tangle deposition, and the subsequent cognitive impairment do not rely on aberrant activation of major tauKs. Additionally, novel evidence was provided for the beneficial contribution of cortical CDK5/p25 to the maintenance of working memory.

1. Introduction

Following the discovery of hyperphosphorylated tau species as the primary constituents of neurofibrillary tangles (NFTs), much scientific effort was directed towards identifying kinases capable of phosphorylating tau at specific residues to potentially unveil targets for Alzheimer’s disease (AD) therapeutics (Mazanetz and Fischer, 2007). Notably, tau kinases (tauKs) assume a pivotal role as a putative intermediary linking the processes of amyloid-beta (Aβ) aggregation with the subsequent formation of NFTs (Bloom, 2014; Ferrer et al., 2005; Lee and Tsai, 2003). Many enzymes were reported to mediate tau phosphorylation at more than fifty serine (S) and threonine (T) residues within the tau protein sequence (Hanger et al., 2009; Kimura et al., 2018). However, pathogenic forms of tau phosphorylation yielding protein misfolding and aggregation into paired helical filaments (PHFs) is typically attributed to three major tauKs: cyclin-dependent kinase-5 (CDK5), glycogen synthase kinase-3 (GSK3), and extracellular signal-regulated kinases-1/2 (ERK1/2) (Martin et al., 2013).

Activation of these three tauKs to catalyze tau phosphorylation is tightly regulated by specific mechanisms (Fig. 1A-B). As examples, neuronal CDK5 activation requires the interaction with a cofactor protein, p35 (or the lower expressed p39) (Lew et al., 1994). Under physiological conditions, the rate of CDK5 activity relies on the availability of p35, which is in turn limited due to its short half-life (Patrick et al., 1998). p35 is constitutively attached to the cytoplasmic membrane via N-terminal myristoylation, restricting CDK5 activity to the membrane environment (Patrick et al., 1999). However, p35 is cleaved by sustained stimulation of the Ca²⁺-sensitive protease calpain-1 (Kusakawa et al., 2000). The resulting p35 C-terminal breakdown product, p25, lacks both the membrane-attaching modifications and the ubiquitin-sensitive residues responsible for quick degradation of full-length p35, while the ability to bind and activate CDK5 is preserved (Patrick et al., 1999; Patrick et al., 1998). The CDK5-p25 complex can translocate to different cellular compartments, including the cytoskeleton, where tau is a target for phosphorylation. Preclinical studies showed that accumulation of Aβ can elevate intracellular Ca²⁺ concentrations, contributing to activation of calpain-mediated p35 cleavage (Giese, 2014; Maitra and Vincent, 2022). While many of these studies attributed a neurotoxic role for p25, calpain-mediated p35 cleavage may also be essential in synaptic plasticity and working memory (Angelo et al., 2003; Engmann et al., 2011b; Giese, 2014), which could account for some of the contrasting findings across studies (Giese, 2009).

Fig. 1.

Tau peptides and tau kinases in MAP participants. (A) Schematic representation of human 2N4R tau showing the two N-terminal insertions (N1 and N2) and 4-repeat microtubule-binding domains (MTB) in the C-terminal region, and the approximate positions of the phospho-Ser/Thr residues of interest are indicated with black arrowheads. Red boxes highlight the (phospho)peptides assessed by SRM proteomics within the 441 amino acid sequence of 2N4R tau. Peptides were named after the anti-tau antibody recognizing the overlapping sequence. Colour squares above the labels indicate the tauK(s) implicated in its phosphorylation (purple, CDK5; maroon, GSK3; green, ERK). (B) Illustration summarizing the activation mechanisms of tau kinases relevant to the present study. (C) Heatmap on Pearson’s R-(top panels) and P-(bottom panels) values following multiple pairwise correlations between demographic, age-related neuropathologies, cognitive variables and tauKs amounts of MAP participants (indicated on top), and the amounts of tauKs and tau species (on the right) quantified in the DLPFC (n = 150) and/or hippocampus (n = 62) of MAP participants. Abbreviations: AD, Alzheimer’s disease; Amyloid, area fraction of the amyloid beta staining by immunohistochemistry; AT8 Tangles, tangle density by AT8 antibody immunohistochemistry; Arterioloscl, arteriolosclerosis; Atheroscl, cerebral atherosclerosis; CAA, cerebral amyloid angiopathy; Demo, demographic variables; Diff plaques, diffuse plaques by Bielschowsky silver impregnation; Edu, years of education; Hipp, hippocampus; Hippscl, hippocampal sclerosis; LBD, Lewy body disease; Macroinf, chronic macroinfarcts; Microinf, chronic microinfarts; MMSE, mini-mental state examination; MTB, microtubule-binding domain; NFT, neurofibrillary tangles; Neur plaques, neuritic plaques by Bielschowsky silver impregnation; SRM, liquid chromatography selected reaction monitoring; PMI, postmortem interval; TDP43, TAR DNA-binding protein 43; tauKs, tau kinases. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

The two other major tauKs, GSK3 and ERK1/2 are themselves activated by phosphorylation. Unlike most kinases, GSK3 is self-stimulated by autophosphorylation at tyrosine-216 (Y²¹⁶), and may remain constitutively active in neurons. The key regulatory sites for phosphorylation in GSK3 are S²¹ (in GSK3α) and S⁹ (in GSK3β); phosphorylation at these sites exerts a dominant inhibitory effect on GSK3 enzymatic activity (Giese, 2009; Lauretti et al., 2020; Takashima, 2006). ERK1/2 activation requires dual phosphorylation at T²⁰²/Y²⁰⁵, following the sequential activation of the Raf/MEK/ERK cascade (Sweatt, 2001; Zhu et al., 2003). The mechanisms linking Aβ accumulation with GSK3 and ERK1/2 activation are multifactorial, and may involve neuroinflammation and oxidative stress pathways (Chu et al., 2004; Ising et al., 2019; Kinney et al., 2018). Other studies suggested that tau phosphorylation and Aβ accumulation in brain tissue are independent processes, and NFT formation may actually precede amyloid plaque deposition (Armstrong, 2014; Kepp et al., 2023).

In contrast to these elegant studies in tissue culture and animal models, assessing the contribution of tauK activity to the pathophysiology of AD in human brain is far more complicated. Prior studies addressing tauK activation in postmortem brain samples from AD cases and controls revealed inconsistent results (see Supplementary File, Table S1). Small sample sizes, long postmortem intervals (PMI), and case-control designs may limit the likelihood of demonstrating an association between activation of tauKs and the course of accumulation of pathology. None of these studies reported associations between the activation status of tauKs, the amounts of phospho-tau (ptau) or NFTs in human brain tissues, and/or the impact on cognition (Paquet et al., 2018; Takashima et al., 2001). Here, using a large sample (n = 150) of postmortem brains from participants in a prospective aging study, the Rush Memory and Aging Project (MAP), we aimed to clarify the associations between the three major tauKs, CDK5, GSK3 and ERK1/2, and (1) tau phosphorylation, (2) NFT deposition, and (3) cognitive impairment in older adults. We evaluated tissue amounts of five relevant ptau peptides, reflective of overall tau phosphorylation, as well as the immunodensities of p35 and p25 cofactors, phospho-inhibition of GSK3 at S^21/9 (pGSK3), and phospho-activation of ERK1/2 at T²⁰²/Y²⁰⁵ (pERK1/2) as respective indices of antemortem tauK activation in the dorsolateral prefrontal cortex (DLPFC) of MAP participants. Assay findings were combined with prior cognitive and neuropathologic assessments in the same MAP participants to build statistical models interrogating the possible pathways leading to NFT formation and/or cognitive dysfunction. Finally, to address potential brain region differencies in tauK activation, exploratory immunoblot analyses were also performed in available hippocampal samples (n = 62) of the same MAP participants.

2. Methods

2.1. Participants and cognitive testing

All individuals included in the present work (n = 150) were enrolled in MAP, a prospective aging study with overall follow-up and autopsy rates exceeding 95 % and 80 %, respectively (Bennett et al., 2018). Table 1 summarizes the demographic, clinical and pathological features of the included MAP participants. The study was approved by the Rush Institutional Review Board, and adhered to the ethical guidelines set forth by the Declaration of Helsinki. At enrollment, participants did not display signs of dementia and agreed to perform cognitive tests annually. All participants signed an Anatomic Gift Act for organ donation and consented to data and biospecimen sharing for prospective research. Cognitive evaluations consisted of 19 structured tests aimed at scoring episodic memory, semantic memory, working memory, perceptual speed, and visuospatial ability. The mean across the five domains was used to estimate the global cognitive function. The mini-mental state examination (MMSE) was additionally assessed. A board-certified neurologist blind to participants’ pathological data performed all clinical diagnoses, following standard criteria (McKhann et al., 1984). For the present study, only cognitive scores obtained nearest death were used, allowing stratification of MAP participants into no (NCI, n = 51), mild (MCI; n = 41) or severe (i.e., dementia; n = 58) cognitive impairment groups. While analyses in the DLPFC were performed in all 150 participants, we explored potential brain region differences in tauK activation in available hippocampal tissue homogenates (NCI, n = 22; MCI, n = 13; DEM, n = 27). Further details on clinical approaches were described elsewhere (Bennett et al., 2012; Schneider et al., 2004).

Table 1.

Demographic, cognitive and pathological characteristics^a of MAP participants stratified by clinical diagnoses.

Variable	NCI (n = 51)	MCI (n = 41)	Dementiab (n = 58)	All (n = 150)
Demographic
Female, no. (%)	40 (78 %)	26 (63 %)	39 (67 %)	105 (70 %)
Age at death, years	86.7 ± 6.7	87.6 ± 6.2	91.2 ± 5.1	88.7 ± 6.3
Education, years	13.9 ± 2.5	15.0 ± 2.8	15.0 ± 3.3	14.6 ± 2.9
APOE ε4 carriers, no. (%)	8 (16 %)	12 (29 %)	15 (26 %)	35 (23 %)
PMI, hours	6.3 ± 2.1	7.7 ± 5.6	6.2 ± 3.1	6.6 ± 3.8
Cognitive function proximate to death
Global score	0.12 ± 0.41	−0.55 ± 0.45	−1.93 ± 0.86	−0.85 ± 1.21
Episodic memory	0.26 ± 0.50	−0.73 ± 0.84	−1.91 ± 0.95	−0.83 ± 1.21
Semantic memory	0.07 ± 0.59	−0.45 ± 0.67	−2.20 ± 1.50	−0.94 ± 1.46
Working memory	0.08 ± 0.71	−0.34 ± 0.65	−1.46 ± 0.90	−0.62 ± 1.02
Perceptual speed	−0.27 ± 0.88	−0.69 ± 0.81	−2.14 ± 0.83	−1.08 ± 1.18
Visuospatial ability	0.12 ± 0.62	−0.08 ± 0.66	−1.41 ± 0.99	−0.50 ± 1.05
MMSE	27.7 ± 1.9	25.5 ± 2.4	14.2 ± 8.3	21.9 ± 8.2
Global pathology NIA/Reagan scalec, no. in 1:2:3:4	3:13:33:2	3:20:17:1	14:31:13:0	20:64:63:3
CERAD scaled, no. in 1:2:3:4	8:8:16:19	9:14:8:10	25:20:2:11	42:42:26:40
Braak stagee, no. in 0-II:III-IV:V-VI	16:32:3	6:29:6	4:31:23	26:92:32
Amyloid-β (% stained area)	2.00 ± 2.29	3.85 ± 4.38	5.48 ± 4.91	3.85 ± 4.28
AT8 tangles (count/mm2)	3.04 ± 4.33	6.36 ± 6.44	8.00 ± 6.10	5.87 ± 6.02
Lewy body disease, no. (%)	4 (8 %)	1 (2 %)	21 (36 %)	26 (17 %)
Hippocampal sclerosis, no. (%)	0 (0 %)	4 (10 %)	10 (17 %)	14 (9 %)
Arteriolosclerosisf, no. in 0:1:2:3	10:22:16:3	5:20:12:4	9:17:24:8	26:59:52:15
Atherosclerosisf, no. in 0:1:2:3	10:21:13:7	6:18:10:7	3:21:21:13	19:60:44:27
Amyloid angiopathyf, no. in 0:1:2:3	15:23:9:4	10:21:5:5	9:24:19:6	32:68:33:15
Macroinfarcts, no. (%)	10 (20 %)	14 (34 %)	21 (36 %)	45 (30 %)
Microinfarcts, no. (%)	8 (16 %)	7 (17 %)	16 (28 %)	31 (21 %)
TDP-43 stage, no. in 0:1:2:3	30:10:5:6	15:8:7:11	19:14:7:18	64:32:19:35
Local (DLPFC) AD pathology
Amyloid-β (% stained area)	2.93 ± 3.34	5.56 ± 6.58	7.26 ± 6.48	5.32 ± 5.90
Diffuse plaquesg (count/mm2)	6.17 ± 11.5	11.0 ± 15.2	13.3 ± 13.7	10.2 ± 13.7
Neuritic plaquesg (count/mm2)	5.20 ± 8.47	6.66 ± 8.46	10.4 ± 9.24	7.59 ± 9.01
AT8 tangles (count/mm2)	0.35 ± 1.66	1.12 ± 3.08	1.74 ± 3.60	1.63 ± 4.99
NFTsg (count/mm2)	0.29 ± 1.38	1.27 ± 4.09	3.07 ± 6.93	1.10 ± 2.96

Abbreviations: AD, Alzheimer’s disease; APOE, apolipoprotein E; AT8, anti-phosphotau antibody, clone AT8; CERAD, Consortium to establish a registry for Alzheimer’s disease; DEM, dementia; DLPFC, dorsolateral prefrontal cortex; MAP, Memory and Aging Project; MCI, mild cognitive impairment; MMSE, mini-mental state examination; NCI, no cognitive impairment; NFT, neurofibrillary tangles; NIA, National Institute on Aging; no., number of subjects; PMI, postmortem interval; SD, standard deviation; TDP-43, TAR DNA-binding protein 43.

^aValues are mean ± SD unless noted otherwise.

^bDementia group includes AD diagnoses and other dementias.

^cNIA/Reagan scale: high (1), intermediate (2), low (3), or no (4) likelihood of AD, according to the presence and distribution of neuritic plaques and tangles.

^dCERAD scale: frequent (1), moderate (2), sparse (3), or no (4) neuritic plaques.

^eSummarized Braak stages: no or transentorhinal tau-tangles only (0-II), limbic spread (III-IV), or neocortical spread (V-VI).

^fSeverity of the pathology scaled as follows: absent (0), mild (1), moderate (2), or severe (3).

^gAssessed by Bielschowsky silver impregnation.

2.2. Brain samples and neuropathologic assessments

Postmortem brain specimens were obtained at autopsies of deceased MAP participants. For neurochemical analyses, grey matter samples from the DLPFC (approximating Brodmann’s area 46/9) and the hippocampus were dissected from one hemisphere and immediately frozen at −80 °C. The other hemisphere was cut into 1 mm-thick coronal slabs, fixed in 4 % paraformaldehyde, and sectioned for histopathology and immunohistochemistry assays, as previously described (Schneider et al., 2004). Examination of AD-related pathology included Bielschowsky silver impregnation to quantify the density of diffuse and neuritic plaques, and NFTs, as well as immunohistochemistry assays with anti-Aβ (clone MO0872; Dako, Carpinteria, CA) and anti-ptau (clone AT8; Invitrogen) monoclonal antibodies, to estimate the amyloid-stained area fraction and stereological tangle counts, respectively (Bennett et al., 2004). Routine neuropathologic assessments also examined Lewy bodies, hippocampal sclerosis, cytosolic inclusions of phospho-TDP-43, macro- and microinfarcts, arteriolosclerosis, atherosclerosis, and amyloid angiopathy, following standard protocols (Bennett et al., 2012; Schneider et al., 2004). A board-certified neuropathologist blind to clinical data examined all samples, performed diagnoses, and rated the severity and/or stage of the pathologic lesions, including Braak, CERAD (Consortium to Establish a Registry for Alzheimer’s Disease) and NIA/Reagan assessments.

2.3. Targeted proteomics

To approach overall cortical amounts of phosphorylated tau at relevant residues, frozen DLPFC samples from MAP subjects were processed to quantify five phosphorylated peptides, along with their corresponding nonphosphorylated proteoforms, mapping key phosphoserine (pS) or phosphothreonine (pT) residues along the tau protein sequence (see Fig. 1A) by liquid chromatography selected reaction monitoring (LC-SRM) proteomic assays, as previously described (Kapasi et al., 2022; Yu et al., 2018; Zammit et al., 2022). The selected ptau proteoforms overlap with target epitopes of the tangle-detecting, anti-ptau monoclonal antibodies AT8 (pS²⁰²), AT100 (pT²¹⁷), 12E8 (pS²⁶²), 77G7 (pS³⁰⁵), and PHF1 (pS⁴⁰⁴), which are in turn known substrates of CDK5/p25, GSK3 and/or ERK1/2 (Fig. 1A) (Hanger et al., 2009; Kimura et al., 2018; Martin et al., 2013). Herein, peptides are further referred to using the associated antibody clone name. Briefly, brain samples were first homogenized in denaturation buffer, containing 8 M urea, 50 mM Tris-HCl pH 7.5, 10 mM dithiothreitol, and 1 mM ethylenediaminetetraacetic acid. The solubilized proteins were then alkylated with iodoacetamide and digested with trypsin. All subsequent procedures, including sample transfer to 96-well microplates, were semi-automated in an Epmotion 5075 TMX liquid handling robot (Eppendorf). The abundances of target peptides were estimated in relation to reference peptides added to the brain samples at fixed concentrations (Andreev et al., 2012), and finally log₂-transformed.

2.4. Western blotting

Cortical and hippocampal immunodensities of phosphorylated and total forms of tauKs in MAP samples were estimated by sodium dodecyl sulfate (SDS)-polyacrylamide gel electrophoresis (PAGE), followed by Western blotting (WB), as previously described (Ramos-Miguel et al., 2017a). Frozen tissue samples were first homogenized under standard WB conditions (Ramos-Miguel et al., 2017a), and protein concentrations estimated by DC-Assay (Bio-Rad, Hercules, CA, USA). Before electrophoresis, tissue homogenates were mixed with equal volumes of a 2× loading buffer, containing 100 mM Tris-HCl pH 6.8, 3 % SDS, 20 % glycerol, 5 % beta-mercaptoethanol and 0.1 % of bromophenol blue. Cortical and hippocampal samples (30 μg/well) were respectively resolved in 10 % SDS-PAGE minigels or 4–16 % gradient TGX precast gels (Bio-Rad), and transferred to nitrocellulose membranes. All gels were also loaded at least 3 times with a reference sample to compare results across the different experimental gels. Blockade of membrane nonspecific binding sites was achieved by incubation (1 h, at room temperature) with either Tris-buffered saline (TBS; pH 7.6) supplemented with 5 % bovine serum albumin (BSA) and 0.1 % Tween-20 (TBST; for assays involving phosphospecific primary antibodies) or phosphate-buffered saline (PBS) supplemented with 5 % skim-milk, 1 % BSA, and 0.2 % Tween-20 (PBST; for all other primary antibodies). The membranes were then incubated overnight at 4 °C with appropriate concentrations of the primary antibody (see Supplementary File, Table S2) diluted in TBST (phosphoepitope detection) or PBST (all other assays). The next day, membranes were incubated with anti-rabbit or anti-mouse horseradish peroxidase-conjugated secondary antibodies (1:5000; Cell Signaling), overlaid with commercial enhanced chemiluminescence reagents (ECL; Amersham, Buckinghamshire, UK). In the original experiments in DLPFC, membranes were exposed to autoradiography film (Amersham ECL Hyperfilm) for 1–60 min. Autoradiograms were quantified by densitometric scanning (GS-800 Imaging Calibrated Densitometer, Bio-Rad). Later assays in the hippocampus were imaged in an Amersham Image Reader AI600 (Cytiva, Marlborough, MA USA). Due to the limited amount of sample, membranes were incubated initially with anti-CDK5, and sequentially stripped and reprobed with anti-p35/p25 and finally with anti-β-actin antibody as loading control (see Supplementary File, Fig. S1 for evidence of unchanged protein levels across groups). Membranes originally incubated with phospho-specific antibodies against pS^21/9-GSK3α/β or anti-pT²⁰²/Y²⁰⁴-ERK1/2 (hereafter referred to as pGSK3 and pERK1/2, respectively) were subsequently reprobed with antibodies recognizing the total form of the corresponding kinase. Thus, non-phosphorylated proteins were normalized against β-actin, whereas phospho-protein levels were normalized relative to their corresponding total protein (Hernández-Hernández et al., 2024; Ramos-Miguel and Garcia-Sevilla, 2012) (see Supplementary File, Fig. S2 for complete, uncropped immunoblot images). Total and phosphospecific antibodies to GSK3 and ERK1/2 recognized the two major isozymes of each kinase expressed in brain; GSK3α (51 kDa) and GSK3β (46 kDa), as well as ERK1 (44 kDa) and ERK2 (42 kDa). Both GSK3α and GSK3β isoforms contribute to tau phosphorylation and NFT formation in AD brains (Alon et al., 2011; Giese, 2009), and separate quantification of these isoforms yielded very similar results (see Fig. 1C). For practical reasons, the reported GSK3 and ERK1/2 results refer to the combined quantification of GSK3α + β or ERK1 + 2, respectively. Since SRM proteomic data was not available for the hippocampus, we additionally quantified the overall immunodensities of ptauAT100 and total tau (Tau5) by WB in a subset of 62 MAP participants with available hippocampal postmortem tissue homogenates.

2.5. Statistical analyses

Multivariate Pearson’s correlation analyses surveyed putative associations between tauK immunodensities, relative peptide amounts, and demographic, cognitive, and pathologic variables of MAP participants. Comparisons across diagnosis- or pathology-stratified participants were performed with analyses of covariance (ANCOVA), adjusting by age, sex, education, and PMI. The contributions of tauK activation and/or cortical peptide abundances to tau phosphorylation, NFT deposition, and cognitive performance were explored by adding and removing the predictor variable of interest in a series of linear regression models adjusted by demographics and neuropathologies, as previously described (Ramos-Miguel et al., 2017b). Linear regression analyses were used as indicated elsewhere (Hayes, 2017; Ramos-Miguel et al., 2019) to identify probable mediators within the molecular pathway involving tauK activation, tau phosphorylation, NFT deposition, and their influence on cognitive performance. These analyses were structured in three stages, considering the dependent variable of the model (see Supplementary File, Fig. S3). In stage#1, ptau peptides were set as outcomes in a series of independent models testing tauK activation as predictors. Stage#2 evaluated the effects of tauK activation and tau phosphorylation on the frontal cortical density of NFTs. Finally, stage#3 combined appropriate sets of all the above pathologic and neurochemical variables to evaluate their individual and joint effects on participants’ cognitive performance nearest death. Bielschowsky-based estimates were used as the primary measure of NFTs. Similar results were obtained in models using AT8-positive tangle counts instead of Bielschowsky-assessed NFTs. In the context of a central role in AD pathogenesis, the potential impact of Aβ accumulation was included as a potential contributor to all model outcomes in the three stages described above. Regarding non-AD-related neuropathologies, stage#3 models were only controlled for those statistically significantly associated with cognitive outcomes and/or the neurochemical measures to avoid model overload. Similarly, added-in variables not generating statistically significant findings were removed before including more predictors in the model. For all tests, the level of significance was set to p < 0.05. Of note, linear regression modeling approaches were only performed with DLPFC data, as the sample size of the hippocampal subset (n = 62) did not have sufficient statistical power to support the number of covariates used in the analyses. JMP17 (SAS Institute, Cary, NC, USA) and/or Prism 9 (GraphPad Software, La Jolla, CA, USA) were used to perform all statistical analyses and data plotting.

3. Results

3.1. Direct associations between tauKs, tau peptides, and other demographic and pathologic characteristics of MAP participants

We first evaluated the direct effects of potentially confounding variables on tauK species and tau peptides. Age at death, sex, and years of education had no apparent influence on the cortical amounts of the studied tauK species and tau peptides (Fig. 1C). Some proteins and peptides (e.g., p35, pGSK3, and tau77G7) displayed lower densities in samples with longer PMI (see Supplementary File, Fig. S4 for detailed analyses). Regarding non-AD-related neuropathologies, the presence of arteriolosclerosis, hippocampal sclerosis, amyloid angiopathy and/or phospho-TDP43 significantly modified the cortical amounts of p35, p25, pGSK3 and most of the tau peptides (Fig. 1C). These observations highlight the importance of controlling for demographic and pathologic variables to address specific AD-related molecular alterations in postmortem brain studies.

Direct associations between the AD-related pathology indices and tauK expression levels were unclear. Surprisingly, NFT densities, assessed by Bielschowsky impregnation or AT8 immunostaining, did not correlate with cortical or hippocampal tauK activation indices. Greater plaque deposition was observed in DLPFC samples exhibiting lower p35 or p25 immunoreactivity (r = −0.164 to −0.260, p < 0.05) (Fig. 1C). As expected, greater amounts of ptau12E8, ptauAT100, and ptau77G7 (but not ptauAT8 or ptauPHF1) species were associated with NFT tissue accumulation (r = 0.297–0.610, p < 0.001), regardless of the methodology used for tangle detection (i.e., AT8 immunohistochemistry or Bielschowsky silver impregnation), or the brain area analyzed. Accumulation of Aβ-plaque pathology was also associated with greater ptau12E8 cortical levels (r = 0.257–0.377, p < 0.001) and ptauAT100 hippocampal densities (r = 0.658, p < 0.001). The complex associations between cognitive outcomes and brain densities of tauKs and ptau peptides were addressed in detail using series of linear regression models (see below).

In samples from the DLPFC of MAP participants, the immunodensities of total kinase forms of CDK5, GSK3 and ERK1/2 (i.e., regardless of their phosphorylation status) showed no or marginally significant associations with tau (phospho)peptide amounts (Fig. 1C and Supplementary File, Fig. S5). Thus, greater total CDK5 density was associated with lower ptau77G7 amounts (r = −0.167, p < 0.05), while greater immunoreactivity of total GSK3 correlated with lower concentrations of ptau77G7 (r = −0.166, p < 0.05) and greater ptauPHF1 (r = 0.172, p < 0.05) and nonphosphorylated tau12E8 (r = 0.194, p < 0.05) amounts in the same MAP cortical samples. In contrast to our initial expectations, greater cortical immunodensities of the CDK5 cofactors p35 and p25 were associated with reduced tau phosphorylation and/or increased nonphosphorylated proteoforms (Fig. 1C). Particularly, lower ptauAT100 and ptauAT8 peptide amounts (r = −0.300 to −0.376, p < 0.001), as well as greater tau77G7, tauAT8, and tauPHF1 peptide levels (r = 0.267–0.452, p < 0.001) correlated with greater immunodensities of both p35 and p25 CDK5 activators in DLPFC samples. Similarly, increased pS⁹-GSK3 phosphoinhibition was associated with greater ptauPHF1 (r = 0.244, p < 0.001) and ptauAT8 (r = 0.188, p < 0.05) peptide amounts, and lower densities of nonphosphorylated tauAT100 (r = −0.198, p < 0.05), tauAT8 (r = −0.333, p < 0.001), and tauPHF1 (r = −0.339, p < 0.001) species (Fig. 1C). Other than ptauAT100 (r = −0.197, p < 0.05), ERK1/2 activation was not associated with any of the tau proteoforms quantified in the DLPFC samples of MAP participants.

To address brain region specificity of the above associations in the DLPFC, we assessed similar analyses in available hippocampal tissue homogenates from a subset of MAP participants. For practical reasons, ptau quantitative studies in the hippocampus were performed for one target only. Among the above ptau species, we selected ptauAT100 for its greatest associations with cognitive outcomes and activated tauK densities. Like in the DLPFC, greater indices of tauK activation were not associated with higher immunodentities of monomeric or oligomeric ptauAT100 in hippocampal tissue homogenates.

These data suggest that stimulation of major tauKs (i.e., upregulation of CDK5 cofactors and/or downregulation of pGSK3) may be related to a reduction, rather than an increase of tau phosphorylation levels at key residues associated with NFT formation and AD progression in human brain. These findings may not be restricted to the frontal lobe, as comparable findings were obtained in both DLPFC and hippocampus.

3.2. TauKs and ptau peptides in MAP samples by clinical and pathological diagnoses

Cortical amounts of tauK species and tau peptides were then compared across groups of MAP participants stratified by cognitive- or Braak-based diagnoses, adjusting by demographics (Fig. 2–3). The immunodensities of total CDK5, GSK3 and ERK1/2 forms did not significantly differ across groups. Further analyses focused on activated (p35, p25, pERK1/2) or inhibited (pGSK3) tauK species, as well as tau peptides. Compared to NCI participants, DLPFC samples from participants with dementia had lower densities of the CDK5 cofactor p25 (−20 %, p < 0.05) (Fig. 2A). This modulation may be specific of the DLPFC, as hippocampal densities of p25 did not differ between NCI and cognitively impaired participants (Fig. 2B). Similarly, cortical, but not hippocampal amounts of ptauAT100 (but not any other ptau peptide) were significantly greater in subjects with dementia, compared to NCI subjects (p < 0.05; Fig. 2A). Brain tissue amounts of other tauKs (pGSK3 and pERK1/2) and ptau species showed similar levels of expression across all three groups of clinical cognitive status (Fig. 2A-B).

Fig. 2.

Cortical and hippocampal densities of tau kinases (tauKs) and tau peptides by clinical diagnoses. Regulation of tauKs (upper boxplots) and phospho-tau (ptau) peptides (bottom boxplots) in the DLPFC (A) and the hippocampus (B) of MAP participants ranked by clinical diagnoses with no cognitive impairment (NCI, n = 51/22 [DLPFC/hippocampus], green), mild-cognitive impairment (MCI, n = 42/13 [DLPFC/hippocampus], yellow) or dementia (DEM, n = 58/27 [DLPFC/hippocampus], blue). Representative immunoblots of different tauKs and β-actin (loading control) of several participants and standard (std) samples are shown on the right. Subjectś clinical diagnoses were indicated using the same colour code of the boxplots. The reported tauK immunodensities were all normalized by β-actin or non-phosphorylated protein forms, and expressed as percentage of in-gel standard (std) sample. Between-group differences were assessed by ANCOVA controlling for age, sex, education and PMI, followed by Tukey’s HSD post hoc test. *p < 0.05. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 3.

Braak stages, regional tau pathology, and cortical immunodensities of tau kinases (tauKs) and ptau peptides. (A–C) Representative photomicrographs of AT8-immunostained brain sections, along with cresyl violet counterstaining, from MAP participants classified as Braak stage I (A), III (B), and V (C). Each panel pair illustrates the progression of tau pathology across three brain regions: the middle-frontal gyrus of the dorsolateral prefrontal cortex (DLPFC; top row, panels 1 and 2); the CA1 subfield of the hippocampus (middle row, panels 3 and 4); and the entorhinal cortex (EC; bottom row, panels 5 and 6). For each brain region, the left image (odd-numbered panels) shows a low-power overview, and the corresponding right image (even-numbered panels) presents a high-power magnification of the area indicated by a black arrow. Scale bars in a6/b6/c6 panels correspond to 250 μm for low power images (i.e., left panels) or 100 μm for high power images (i.e., right panels). (D) Regulation of tauKs (upper boxplots) and ptau peptides (right boxplots) in the DLPFC of MAP participants stratified by summarized Braak staging, indicating no or transentorhinal NFT deposition (0-II, n = 26, green), limbic spread (III-IV, n = 92, orange), or neocortical spread (V-VI, n = 32, blue). The scatterplots depict pairwise correlations between cortical amounts of tauKs and ptau peptides in MAP participants stratified by Braak staging. The reported tauK immunodensities were all normalized by β-actin or non-phosphorylated protein forms, and expressed as percentage of in-gel standard (std) sample. Between-group differences were assessed by ANCOVA controlling for age, sex, education and PMI, followed by Tukey’s HSD post hoc test. *p < 0.05; **p < 0.01; ***p < 0.001. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

To gain statistical power, we clustered Braak staged participants into three groups: (1) Braak 0-II (i.e., absence or entorhinal presence of NFT); (2) Braak III-IV (limbic propagation); and (3) Braak V-VI (neocortical spread) (images of representative AT8-tau-stained entorhinal, hippocampal, and DLPFC slices from a MAP participant in each group shown in Fig. 3A-C). Cortical immunodensities of the active forms of the studied tauKs showed no statistically significant differences across Braak stage groups (Fig. 3D). In contrast, hippocampal densities of CDK5 (+22 %, p < 0.05), p25 (+54 %, p < 0.05), and pGSK3 (−38 %, p < 0.05) were significacntly altered in late versus early Braak stages (Supplementary File, Fig. S6). As expected, cortical ptauAT100 and ptau12E8 (but not ptauAT8, ptau77G7 or ptauPHF1) peptide levels, as well as hippocampal immunodensities of ptauAT100 forms, were largely increased in samples of participants at Braak stages V-VI as compared to those at Braak 0-IV (p < 0.001; Fig. 3D and Supplementary File, Fig. S6).

To evaluate possible compensatory feedback mechanisms inhibiting tauKs following excessive ptau accumulation, we reassessed the associations between tauK immunodensities and ptau peptide levels stratified by Braak staging. Similar kinase-substrate associations were observed across Braak stages, notably including those at early Braak stages (Fig. 2C, and Supplementary File, Fig. S6). This observation does not support the existence of an initial tauK hyperactivation in DLPFC followed by a compensatory tauK inhibition in later Braak stages.

3.3. Modeling associations with tau phosphorylation and tangle accumulation in DLPFC of MAP participants

The contribution of tauK activation/inhibition to tau phosphorylation and NFT deposition in brains of MAP participants was estimated from a series of age-, sex- and PMI-controlled regression models using either ptau peptide amounts (Fig. 4A) or NFT cortical densities (Fig. 4B) as model outcomes. We used structured combinations of tauK activation estimates and Aβ accumulation measures as predictors in the models to test their combined contributions (see Supplementary File, Fig. S3, stage#1 and stage#2). To simplify the model construction (and guided by the findings reported in Fig. 2), we used cortical levels of p25 to estimate CDK5 activation in the subsequent analyses. Similar results were obtained when using p35, instead of p25, immunodensities in the model.

Fig. 4.

Plots summarizing regression models predicting tau peptide amounts, NFTs, or cognitive scores. Bar (A) and pie (B, C) plots represent the percent contributions of the predictors (indicated in the corresponding Fig. keys) to explain the variance in cortical ptau peptide levels (A), NFT accumulation (B) and domain-specific or global cognition scores (C) across n = 150 MAP participants. *p < 0.05; **p < 0.01; ***p < 0.001; significance of the estimate in the final model (see Supplementary File, Table S3). Abbreviations: Aβ, amyloid beta; CAA, cerebral amyloid angiopathy; Edu, education; HS, hippocampal sclerosis; LBD, Lewy body disease; NFT, neurofibrillary tangles; Non-AD-path, non-Alzheimer’s disease related neuropathologies; MI, macroinfarts; PMI, postmortem interval.

Differences in cortical tauK activation and Aβ accumulation explained a modest portion (0–16 %) of between-subject variance in ptau peptide levels (see Fig. 4A and Supplementary File, Table S3). Cortical p25 immunodensity was the major factor contributing to the variance in ptauAT100 and ptauAT8 levels (8–11 %, p < 0.001). pGSK3 density also contributed 8–11 % of the variance (p < 0.01) in ptauAT8 and ptauPHF1 cortical amounts. In contrast, variations in cortical ptau12E8 levels were only associated with Aβ accumulation (9 %, p < 0.001). None of the studied variables had a significant effect on ptau77G7 cortical amounts. The direction of these effects did not change from those observed in Fig. 1C, i.e., greater tauK activation was associated with lower tau phosphorylation levels. No statistical interaction or mediation effects between Aβ accumulation and tauK activation were observed, indicating that Aβ and tauKs impact on tau phosphorylation in a relatively independent manner. These modest contributions further suggest that other kinases or reduced phosphatase activity may be involved in tau phosphorylation in the frontal lobe of MAP participants.

Altogether, differences in tauK immunodensities, ptau peptide levels and Aβ accumulation explained 48 % of between-subject variance in cortical NFT deposition (Fig. 4B and Supplementary File, Table S4). Greater densities of ptauAT100 (21 %, p < 0.001), ptau12E8 (9 %, p < 0.001) and Aβ immunoreactivity (15 %, p < 0.001), as well as lower density of pGSK3 (2 %, p < 0.05), were the major contributors to NFT accumulation. Of note, lower pGSK3 density (i.e., greater GSK3 activation) was associated with greater NFT deposition, suggesting that GSK3 activity may contribute to tangle formation independently from the studied tau phosphorylation sites. The effect of ptau77G7 on NFT formation was fully mediated by ptauAT100 (Supplementary File, Table S4). No other significant mediation or interaction effects were observed between the above predictors, indicating that their contributions to NFT formation were relatively independent.

3.4. Cortical alterations contributing to cognitive function in MAP participants

Next, we used mediation analyses to address the hypothesis that alterations in tauK activation may impact on cognitive performance via tau phosphorylation and/or NFT accumulation in the aging brain. Domain-specific or global cognitive scores were regressed as a function of tauK activation, ptau peptide levels, and/or NFT density by sequentially adding the predictors to series of structured linear regression models adjusted by demographics, non-AD-related neuropathologies, and Aβ accumulation (see Supplementary File, Fig. S3, stage#3). A greater immunodensity of the CDK5 activator p25 (but not other studied tauKs) was associated with better cognitive outcomes (Fig. 3C and Table 2). Particularly, differences in p25 DLPFC levels further explained 3 % and 2 % (p < 0.05) of the variance in working memory and semantic memory, respectively, while no effects were observed on other cognitive domains. The beneficial impact of preserving higher p25 expression levels on cognition was fully mediated by ptauAT100 or ptau12E8, depending on the memory domain (Table 2, and Fig. 4C). Combined, differences in ptauAT100 and ptau12E8 contributed 3–8 % of the variance in cognitive scores (Fig. 4C), while the amounts of other phosphopeptides did not show significant associations with cognition. As expected, NFT densities fully mediated all prior effects on cognitive scores, and added another 2–5 % of the variance to the model (Fig. 4C and Table 2).

Table 2.

Regression models estimating the contributions of sequentially added predictors to the variance in cognitive performance in MAP participants.

Modela	Predictor	Working memory		Episodic memory		Global cognition
		Adj. R2 or Stdβ	p-val	Adj. R2 or Stdβ	p-val	Adj. R2 or Stdβ	p-val
Reference	Adj. R2	0.178		0.243		0.247
	Age	−0.186	0.025 *	−0.151	0.059	−0.177	0.025 *
	Sex	0.054	0.528	0.113	0.173	0.035	0.671
	Education	−0.095	0.269	−0.069	0.406	−0.089	0.286
	PMI	−0.074	0.380	−0.032	0.699	−0.043	0.590
	Macroinfarcts	0.226	0.006 *	0.166	0.035 *	0.182	0.019 *
	CAA	−0.015	0.856	−0.187	0.018	−0.103	0.182
	Lewy bodies	−0.218	0.009 *	−0.172	0.032 *	−0.258	0.001 *
	Hipp Scler.	0.073	0.394	−0.147	0.076	−0.128	0.111
	Aβ	−0.280	<0.001 *	−0.308	<0.001 *	−0.294	<0.001 *
Model 1	Adj. R2	0.209		0.238		0.251
	p25	0.195	0.014 *	−0.005	0.944	0.100	0.188
Model 2	Adj. R2	0.238		0.259		0.333
	p25	0.114	0.171‡	−0.054	0.507	0.005	0.952
	ptau12E8	0.067	0.432	−0.128	0.127	−0.167	0.038 *
	ptauAT100	−0.230	0.009 *	−0.108	0.206	−0.222	0.007 *
Model 3	Adj. R2	0.270		0.277	0.206	0.361
	p25	0.141	0.086	−0.033	0.686	0.031	0.687
	ptau12E8	0.151	0.092	−0.062	0.482	−0.084	0.320‡
	ptauAT100	−0.140	0.125‡	−0.037	0.681	−0.133	0.126‡
	NFT	−0.255	0.009 *	−0.201	0.039 *	−0.241	0.010 *

Abbreviations: Aβ, amyloid-β; Adj. R², whole model adjusted R squared; CAA, cerebral amyloid angiopathy; Hipp Scler., hippocampal sclerosis; MAP, Rush Memory and Aging Project; NFT, neurofibrillary tangles; ptau12E8, tau peptide mapping phosphoserine-262 recognized by anti-tau clone 12E8; ptauAT100, tau peptide mapping phosphothreonine-217 recognized by anti-tau clone AT100; P-val, P value; PMI, postmortem interval.

^aModels 1–4 were constructed by adding the indicated predictors to the ‘Reference’ model.

^*Statistically significant effect on the cognitive outcome.

^‡Statistically significant mediation effect compared to the preceding model.

4. Discussion

The present study evaluated the associations between brain activation of three major tauKs (CDK5, GSK3, and ERK1/2), tau phosphorylation, NFT formation, and cognitive function in a large postmortem brain collection from MAP participants, representing a community-based sample. Globally, the results are inconsistent with prior studies reporting that aberrant activation of these tauKs may result in tau hyperphosphorylation and subsequent progression of AD. Instead, our data align with studies that have not observed hyperactivation of tauK in AD postmortem brain tissue. To date, no prior study simultaneously examined the entire sequence from tauK activation to tau hyperphosphorylation, NFT accumulation, and cognitive decline in human brain tissue, making this investigation the first to offer a comprehensive assessment of these presumably interconnected processes in AD pathology.

The strongest evidence supporting for a role of CDK5, GSK3, and/or ERK1/2 in the pathological mechanisms of AD came from animal models in which pharmacological inhibition of these kinases halted tau hyperphosphorylation and NFT propagation (Kitazawa et al., 2005; Noble et al., 2005; Serenó et al., 2009). Preclinical studies also demonstrated that CDK5, GSK3, and/or ERK1/2 can mediate tau phosphorylation in vitro at specific residues associated with tau misfolding and aggregation (Hanger et al., 2009; Kimura et al., 2018). However, direct translation of these findings in the preclinical literature to the AD pathology of the human postmortem brain is challenging, and not well supported by scientific evidence. An overview of previous reports revealed that the number of studies finding statistically significant activation of tauKs in AD brains is balanced with those reporting non-significant stimulation (Supplementary File, Table S1). Thirteen studies report assessing p25 density (as an index of CDK5 hyperactivation) in postmortem brains from AD cases and controls. Six of these studies reported significantly greater p25 tissue amounts in AD brains, while seven found no alterations or reduced protein levels in the pathologic tissues. Similar discrepancies were reported for GSK3 and ERK1/2 across studies (Supplementary File, Table S1). Although debate on the role of major tauKs in AD pathophysiology remains unsettled (Giese, 2014; Pao and Tsai, 2021), clinical trials for tauK inhibitors continue with apparent optimism (Basheer et al., 2023; Jadhav et al., 2019; Xia et al., 2021). To date, no randomized-controlled trial testing tauK inhibitors in AD patients has proved efficacy on any relevant clinical outcome (Basheer et al., 2023).

The present results did not suggest that greater indices of CDK5, GSK3, and/or ERK1/2 activation are associated with increased tau phosphorylation or a greater burden of AD pathology in the DLPFC and the hippocampus of MAP participants. In contrast, cortical activation of major tauKs correlated with lower densities of three ptau peptides and/or better cognitive performance. The analyses of tauK stratified by clinical diagnosis revealed a significant p25 decrease in the DLPFC among subjects with dementia. In contrast, no differences in cortical p25 densities were observed across Braak stages. These findings suggest that the loss of cortical p25 may primarily impact cognitive function, rather than the neuropathologic progression of the tauopathy. Consistently, prior preclinical studies reported dysregulated synaptogenesis and memory formation following genetic disruption of p25 expression (Angelo et al., 2003; Giese, 2014). Postmortem brain studies addressing tauK-ptau associations are scarce in the scientific literature. To our knowledge, only two prior studies addressed this issue, and reported no associations between activated tauK immunodensities (CDK5-p25 or pY²¹⁶ GSK3) and ptau or NFT cortical amounts (Paquet et al., 2018; Takashima et al., 2001). Considering that elevated cellular levels of ptau may stimulate protein phosphatases or alter tauKs activity (Leugers et al., 2013; Mueller et al., 2021), we investigated whether the dissociation of tauK-ptau could arise from the activation of inhibitory feedback mechanisms counterbalancing excessive tau phosphorylation by inhibiting tauK activity (Giese et al., 2005). If true, the inverse tauK-ptau relationship would be observed in the most severe cases, whereas subjects with no or mild accumulation of pathologic tau should display positive associations. However, the pattern of tauK-ptau inverse correlations described above were maintained following the Braak stage stratification of the sample. Consequently, hyperactivation of CDK5, GSK3 and ERK1/2 may not be required for tau phosphorylation and propagation throughout brain tissue, at least in the DLPFC. Although the involvement of other tauKs such as p38α, PKA, JNK, or Fyn cannot be discarded (see (Montalto and Ricciarelli, 2023) and references therein), previous studies reported the additional contribution of decreased phosphatase activity (e.g., protein phosphatase 2 A) in AD brains (Liu et al., 2005; Sontag et al., 2004). An alternative or additional explanation to the tauK-ptau dissociation could also be provided by the prion-like spread hypothesis of AD. According to this hypothesis, pathogenic tau molecules can induce misfolding of native tau through direct protein-protein interaction (Jucker and Walker, 2013). Misfolded tau could potentially serve as a more favorable substrate for major tauKs compared to native species. In this context, the significance of tauK hyperactivation may be confined to the specific brain regions where pathologic ptau seeds originate at the onset of the disease (i.e., transentorhinal cortex), leading to a rapid propagation of tau phosphorylation without an overall increase in tauK activation in regions where tau spreads in the late stages of the illness, including the DLPFC.

Regression analyses, controlling for demographics and predicting ptau peptide densities suggested, that (1) Aβ accumulation is the major contributor to ptau12E8 (pS²⁶²), and (2) activation of tauK may contribute to reducing (rather than increasing) brain levels of ptauAT8 (pS²⁰²), ptauAT100 (pT²¹⁷), and ptauPHF1 (pS⁴⁰⁴). Preclinical in vitro and in vivo studies indicate that frontal lobe accumulation of Aβ may precede and exacerbate NFT formation (Iijima et al., 2010; Pooler et al., 2015), consistent with postmortem plaque- and NFT-based Braak staging in humans (Braak and Braak, 1991). In animal models, overexpression of Aβ₄₂ led to tau phosphorylation at S²⁶² and cortical degeneration (Iijima et al., 2010). These studies align with two findings in MAP samples. First, increased Aβ deposition predicted greater ptau12E8 density. Second, Aβ plaques and ptau12E8 together accounted for up to 25 % of NFT deposition in the DLPFC. However, the inverse relationship between tauK activation and tau phosphorylation in MAP samples is difficult to reconcile with our current understanding of tau pathophysiology. Possibly, in human DLPFC, the examined tauKs could phosphorylate alternative residues (e.g., S²¹⁴) with protective effects against PHF-forming phosphorylations (Xia et al., 2021). Indeed, CDK5 and ERK1/2 activation had no significant impact on NFT accumulation in MAP samples. Only GSK3 showed a marginal but significant contribution to NFT formation, not mediated by the phosphopeptides analyzed in the current study. Other GSK3-targeted phospho-sites (e.g., T²³¹) may contribute to this effect (Montalto and Ricciarelli, 2023). Together with Aβ accumulation, tissue levels of ptauAT100 (pT²¹⁷) and ptau12E8 (pS²⁶²) were the best predictors of NFT formation among the phosphopeptides analyzed, suggesting independent but additive contributions of these modifications to tau aggregation.

Regression models predicting cognitive function found interesting relationships between the studied molecules. The effects of p25 on working and semantic memory were fully mediated by ptauAT100 and ptau12E8, respectively, indicating that greater levels of p25 may be beneficial for cognitive abilities by preventing tau phosphorylation at T²¹⁷ and S²⁶². Working memory (and less so semantic memory) largely relies on prefrontal cortex function, whereas efficient hippocampal-entorhinal connectivity may underlie episodic memory (Finn et al., 2019; Zeidman and Maguire, 2016). The observed alterations affecting the DLPFC are consistent with the affected domains of memory function. Arguably, CDK5 loss of function may represent a common feature in disorders accompanied by cognitive impairment. As examples, CDK5/p35/p25 were downregulated in DLPFC samples from subjects with schizophrenia, a serious mental disorder with abnormalities of frontal neurocognitive functions (Engmann et al., 2011a; Ramos-Miguel et al., 2013). Supportive preclinical studies showed that physiological p25 overexpression is required for spatial memory formation in rodents (Angelo et al., 2003; Engmann et al., 2011b). As expected, all ptau peptide effects on cognitive function were fully mediated by NFT accumulation.

The present data should be carefully interpreted in the context of the general limitations of postmortem studies. Nevertheless, in the absence of validated positron emission tomography ligands tracing tauK activity, well-designed postmortem brain studies remain the most valuable source of information to account for associated neurochemical alterations. However, postmortem tissue immunodensities of CDK5-p25, pGSK3, and pERK1/2 may not entirely reflect the actual kinase activity prior to death. Enzymatic assays using selective tauK substrates may complement the information resulting from the present immunoassays (Pei et al., 1997). Additionally, SRM proteomics detects phosphorylation events resulting from both physiological and pathological tau processing, thereby limiting the ability to infer the origin of pathological insults based on changes in ptau peptide levels. Since perimortem conditions could have a strong impact on protein availability and immunodetection, larger sample sizes may be needed for future studies. The current sample size was underpowered to address the mediation effects in subgroups of subjects stratified by Braak staging. However, it is unlikely that adding more samples to the analyses would change the direction of the overall effects reported.

5. Conclusions

Our results indicate that hyperactivation of CDK5/p25, GSK3, and ERK1/2 may not be strictly required to yield tau hyperphosphorylation and subsequent aggregation into NFTs during AD progression in the DLPFC. This prompts an invitation to critically reevaluate the implications of these kinases for the pathological spread of tau and underscores the pivotal role of postmortem human brain studies. According to our findings, the inhibition of CDK5-p25 activity may contribute to greater cognitive impairment in older adults. Consequently, the present data anticipates that the pharmacological development of inhibitors targeting CDK5, GSK3, and ERK1/2 may result in another setback in the pursuit of effective, disease-modifying treatments for AD.

Supplementary Material

Appendix A. Supplementary data

Supplementary data to this article can be found online at https://doi.org/10.1016/j.nbd.2025.106924.

Data availability

Requests to MAP study data can be applied to https://www.radc.rush.edu.