Comparison of clinical, genetic, and pathologic features of Limbic and Diffuse Transactive Response DNA-binding Protein 43 pathology in Alzheimer’s disease neuropathologic spectrum

Abstract

Background:

Increasing evidence suggests that TAR DNA-binding protein 43 (TDP-43) pathology in Alzheimer’s Disease (AD), or AD-TDP, can be diffuse or limbic-predominant. Understanding whether diffuse AD-TDP has genetic, clinical, and pathological features that differ from limbic AD-TDP could have clinical and research implications.

Objective:

To better characterize the clinical and pathologic features of diffuse AD-TDP and differentiate it from limbic AD-TDP.

Methods:

363 participants from the Mayo Clinic Study of Aging, Alzheimer’s Disease Research Center, and Neurodegenerative Research Group with autopsy confirmed AD and TDP-43 pathology were included. All underwent genetic, clinical, neuropsychologic, and neuropathologic evaluations. AD-TDP pathology distribution was assessed using the Josephs 6-stage scale. Stages 1–3 were classified as Limbic, those 4–6 as Diffuse. Multivariable logistic regression was used to identify clinicopathologic features that independently predicted Diffuse pathology.

Results:

The cohort was 61% female and old at onset (median:76 years[IQR:70–82]) and death (median:88 years[IQR:82–92]). Fifty-four percent were Limbic and 46% Diffuse. Clinically, ~10–20% increases in odds of being Diffuse associated with 5-year increments in age at onset(P=0.04), 1-year longer disease duration(P=0.02), and higher Neuropsychiatric Inventory scores(P=0.03), while 15-second longer Trailmaking Test-B times(P=0.02) and higher Block Design Test scores(P=0.02) independently decreased the odds by ~10–15%. There was evidence for association of APOE $\mathrm{\epsilon }$4 allele with limbic AD-TDP and of TMEM106B rs3173615 C allele with diffuse AD-TDP. Pathologically, widespread amyloid-$\mathrm{\beta }$ plaques (Thal phases:3–5) decreased the odds of diffuse TDP-43 pathology by 80–90%, while hippocampal sclerosis increased it sixfold(P<0.001).

Conclusions:

Diffuse AD-TDP shows clinicopathologic and genetic features different from Limbic AD-TDP.

INTRODUCTION

Transactive response DNA-binding protein 43 (TDP-43) is a nuclear protein which aggregates aberrantly in the cytoplasm and processes of neurons and glia in many neurodegenerative diseases, whether as the primary alteration or concomitant with other proteinopathies [1]. It was first detected in tau-negative, ubiquitin-positive frontotemporal lobar degeneration and amyotrophic lateral sclerosis [2, 3]. Before long, it was identified in normal aging [4–6] and Alzheimer’s disease (AD) [7, 8]. Co-occurrence of TDP-43 in AD (AD-TDP) more frequently occurs in late-onset AD and is associated with more episodic cognitive impairment (i.e., AD-type dementia), longer disease duration, Apolipoprotein E $\epsilon$4 (APOE4) carrier status, and numerous co-existing pathologies [9–13].

It is believed that AD-TDP follows a 6-stage distribution scheme as proposed by Josephs et al. [14, 15], whereby inclusions first deposit in the amygdala (stage 1), and extend to the entorhinal cortex and/or subiculum (stage 2); hippocampal dentate gyrus and/or occipitotemporal cortex (stage 3); insula, ventral striatum, basal forebrain, and/or inferior temporal cortex (stage 4); substantia nigra, inferior olive, and/or midbrain tectum (stage 5); and finally the basal ganglia and/or middle frontal cortex (stage 6). This staging system has since been validated [16].

A groundbreaking study [17] identified two distinct types of AD-TDP inclusions: type-$\mathrm{\alpha }$, mirroring the ones found in FTLD-TDP; and type-$\mathrm{\beta }$, which colocalized with tau. Type-$\mathrm{\alpha }$ was associated with older age at death, cognitive decline, increased frequency of TMEM106B risk (CC) haplotypes, hippocampal sclerosis (HpScl), and widespread TDP-43 pathology (stages 4–6) [17, 18]. Type-$\mathrm{\beta }$ cases were younger at death, showed cognitive resilience, seldom coexisted with HpScl and followed a limbic distribution (stages 1–3). In a recent study [19], we found that while the frequencies of all 6 stages of AD-TDP increased linearly over time, the frequencies of stages 4–6 increased to a greater extent and at a faster rate in very old participants, particularly those with dementia [19]. Given the distinct clinical, genetic, and pathologic associations observed, we speculated whether “diffuse” AD-TDP would have characteristics that differed from “limbic” AD-TDP.

Our primary aim was to investigate differences in clinical, genetic, and neuropathologic features associated with AD-TDP limited to limbic regions and diffuse AD-TDP extending to the basal forebrain, brainstem, basal ganglia, and neocortex. Our second aim was to assess whether specific and clinicopathologic features could independently predict the possibility of finding diffuse AD-TDP at postmortem. We hypothesized that limbic AD-TDP would have a clinical picture more reminiscent of typical AD, with predominant episodic memory impairment and more severe AD pathology, while diffuse AD-TDP would show more involvement of other cognitive domains and have a higher frequency of hippocampal sclerosis.

METHODS

Design, Setting and Participants

This cross-sectional, clinicopathologic study was conducted at Mayo Clinic in Rochester, MN. Participants were prospectively recruited and followed in the NIH-funded Mayo Clinic Alzheimer’s Disease Research Center, Mayo Clinic Study of Aging, or the Neurodegenerative Research Group and died between May 1999 and August 2021. They underwent standardized genetic, neurologic, neuropsychologic, and neuropathologic evaluations. Only participants that showed AD-TDP inclusions at postmortem and had a known distribution of pathology (Josephs TDP-43 staging[15]) were included. Cases with a pathological diagnosis of FTLD, with evidence of focal frontal and/or temporal neurodegeneration, as observed macroscopically or microscopically with neuronal loss, microvacuolation, and gliosis in either a predominantly-superficial or transcortical pattern, were excluded [20] Hence, all FTLD-TDP and FTLD-tau (PSP, CBD, Pick’s disease) cases were excluded. Patients with a diagnosis of other neurodegenerative diseases like amyotrophic lateral sclerosis or Huntington’s disease were also excluded. However, participants with a concomitant Lewy body disease (LBD)(40% of the final cohort) or multiple system atrophy (n=1) were not excluded. Of the 1150 participants with known AD-TDP status from the Mayo Clinic database, 492 were TDP-43-positive. Of these, 363 had a known TDP-43 stage and were thus included. The remaining 129 cases are currently being staged on a continuous basis.

This study was performed in accordance with the Helsinki Declaration of 1975 and was approved by the Mayo Clinic Institutional Review Board. All participants or their proxies had signed written informed consent.

Clinical Evaluations

The neurological exam included the use of: Mini Mental State Examination (MMSE) [21] to screen for cognitive impairment; Clinical Dementia Rating-sum of boxes (CDR-SOB) [22] to assess functional independence; and the Unified Parkinson’s Disease Rating Scale (UPDRS) [23] to evaluate for parkinsonism and gait dysfunction. The neuropsychological battery included: Trail Making Test-A (TMT-A) [24] to test for visuomotor speed; Trail Making Test-B (TMT-B) for cognitive flexibility [24]; Auditory Verbal Learning Test (AVLT) delayed recall [25] for verbal episodic memory; Weschler Adult Intelligence Scale-Revised block design test (WAIS-BD) [26] for visuospatial and constructional ability; Boston Naming Test (BNT) [27] for confrontational naming; Controlled Oral Word Association Test (COWAT) [28] for verbal fluency; and Neuropsychiatric Inventory (NPI) [29] for behavioral disturbances. Consensus meetings that included a behavioral neurologist and neuropsychologist were held to determine cognitive state (cognitively unimpaired, mild cognitive impairment or dementia) based on all available data [30, 31]. For each variable, the last available scores taken closest to the time of death were used for the analyses.

Genetic Analyses

Participants underwent APOE genotyping as previously described [32, 33]. Genotyping for transmembrane protein 106B gene (TMEM106B) rs3173615 variants (C/G SNPs, C risk allele) and progranulin gene (GRN) rs5848 variants (C/T SNPs, T risk allele) was performed using TaqMan genotyping assays (Applied Biosystems, Foster City, CA), as previously described [34]. The presence of expanded GGGGCC hexanucleotide repeat in C9orf72 gene was also determined, as previously described [35].

Neuropathologic Evaluations

All participants underwent postmortem neuropathologic examination. TDP-43 pathology was determined using sections immunostained with a monoclonal anti-phosphorylated TDP-43 antibody (pS409/410, 1:5,000, Cosmo Bio, Tokyo, Japan). The amygdala, which is the earliest region affected by AD-TDP, was screened for TDP-43-immunoreactive inclusions. Cases showing either type-$\mathrm{\alpha }$ and/or type-$\mathrm{\beta }$ [17] inclusions within the parenchyma of the amygdala were designated as TDP-43-positive and subjected to staging using additional sections from the medial temporal lobe, basal forebrain, brainstem regions, basal ganglia, and neocortex [15].

AD pathology was assessed following the recommendations of the National Institute on Aging (NIA)-Reagan Criteria [36] or the National Institute on Aging-Alzheimer’s Association (NIA-AA) [37] guidelines. Each case was assigned a Braak neurofibrillary tangle (NFT) stage [38] using modified Bielschowsky silver stain and had an assessment of neuritic plaques based on the Consortium to Establish a Registry for Alzheimer’s Disease (CERAD) [39] guidelines. A Thal phase [40] was determined using an antibody to amyloid-$\mathrm{\beta }$ (A$\mathrm{\beta }$) (clone 6F/3D, 1:10; Novocastra Vector Labs, Burlingame, CA). Lewy body pathology was determined on sections immunostained for $\mathrm{\alpha }$-synuclein (clone LB509, 1:200; Zymed, San Francisco, CA) [41]. Hippocampal sclerosis was diagnosed when severe neuronal loss and gliosis were seen in the subiculum and/or CA1 region of the hippocampus with Hematoxylin and Eosin staining [42]. The presence of cerebral amyloid angiopathy and arteriolosclerosis was assessed on sections stained with Hematoxylin and Eosin, and graded using a 4-point scale: for absent, mild, moderate and severe [43]. Microinfarcts and lacunar(<1cm)/large infarcts(>1cm) were graded as absent or present [43].

Statistical analysis

Counts and percentages were used to describe categorical variables, while medians and interquartile ranges (IQR) were used for continuous variables. Participants were divided based on staging: AD-TDP stages 1–3 were categorized as the “Limbic” group, while stages 4–6 were classified under the “Diffuse” group. As preliminary analyses, the demographic, clinical and pathologic features of both groups were compared using Chi-square test for categorical variables and Mann Whitney U test for continuous variables. To address our primary aim, linear regression models were used to predict differences in clinical variables for Limbic versus Diffuse status, after controlling for age at evaluation and time from evaluation to death. Logistic regression was used to estimate the odds ratio (OR) of finding diffuse AD-TDP at postmortem as a function of individual genetic or pathologic variables and age at death, as well as to determine and plot the predicted probabilities of diffuse AD-TDP positivity across the age at death spectrum according to different genotypes or severity of co-pathologies. For our secondary aim, multivariable logistic regression models were utilized to determine the predictive value of each genetic, clinical and non-TDP-43 pathologic feature – again as measured by an (OR) – to predict the chances that participants would show diffuse AD-TDP at postmortem, after controlling for other variables in the model. A “clinical” logistic model was fit to a subset of 85 participants with data on all the following variables: years of education, dementia status, age at onset, disease duration (time from onset to age at last complete evaluation), and scores from UPDRS, TMT-A and B, AVLT delayed recall, WAIS-BD, BNT, COWAT and NPI. All were considered as continuous variables except for dementia status. A “pathology” logistic model was fit to a subset of 223 cases with all the following variables: age at death, Braak NFT stage, Thal A$\mathrm{\beta }$ phase, hippocampal sclerosis, cerebral amyloid angiopathy, and arteriolosclerosis status. Braak NFT stage (I-VI) and Thal A$\mathrm{\beta }$ phase (0–5) were considered ordinal variables. Cerebral amyloid angiopathy and arteriolosclerosis were treated as dichotomous variables. An interaction term between Braak NFT stages and Thal A$\mathrm{\beta }$ phases was initially included in the model but was later removed because the interactions were not significant at P<0.05. All statistical analyses were performed using R version 4.2.2 and plots were generated using ggplot2 package. P-values <0.05 were considered significant.

RESULTS

Of the 363 TDP-43-positives with AD neuropathologic change, 197 (54%) were classified as Limbic and 166 (46%) as Diffuse. Within the Limbic group, 96 (48%) had stage 1, 59 (30%) stage 2, and 42 (21%) stage 3 AD-TDP pathology. From the Diffuse group, 71 (43%) had stage 4, 57 (34%) stage 5, and 38 (23%) had stage 6 distribution. The demographic, clinical, genetic, and neuropathologic features are shown in Table 1. The Diffuse group was older at death (median:88 years versus 87, P=0.01) and had longer disease duration (median:11 versus 9 years, P=0.02) than the Limbic group.

Table 1.

Demographic, genetic, clinical and neuropathologic features of participants stratified by TDP-43 distribution (Limbic vs Diffuse).

Characteristic	Limbic (n=197)	Diffuse (n=166)	P-value
Demographics
Sex, female	117/197 (59%)	105/166 (64%)	0.38
Education, years	15 (12, 16)	14 (12, 16)	0.06
Age at onset, years	75 (67, 82)	77 (71, 83)	0.07
Age at death, years	87 (80, 91)	88 (83, 94)	0.01
Disease duration, years	9 (6, 12)	11 (7, 14)	0.02
Clinical features
Cognitive state			0.05
CU	33/197 (17%)	14/164 (9%)
MCI	25/197 (13%)	18/164 (11%)
Dementia	139/197 (70%)	132/165 (80%)
MMSE (/30), ↑=better	18 (11, 26)	16 (11,22)	0.04
CDR-SOB (/18), ↓=better	8 (3, 14)	12 (5, 17)	0.004
UPDRS (/44), ↓=better	3 (0, 6)	2 (0, 7)	0.75
TMT-A (/180), ↓=better	60 (44, 120)	75 (50, 125)	0.15
TMT-B (/300), ↓=better	240 (132, 300)	248 (134, 300)	0.79
AVLT delayed (/15), ↑=better	0 (0, 3)	0 (0, 0)	<0.001
WAIS-BD (/48), ↑=better	12 (4, 19)	9 (2, 15)	0.02
BNT (/60), ↑=better	41 (29, 51)	34 (23, 46)	<0.001
COWAT, ↑=better	23 (13, 34)	23 (14, 31)	0.66
NPI (/36), ↓=better	4 (1, 8)	4 (1,9)	0.42
Genetic features
APOE $\mathrm{\epsilon }$4			0.06
Non $\mathrm{\epsilon }$4 carrier	81/193 (42%)	84/164 (52%)
One copy of $\mathrm{\epsilon }$4	90/193 (47%)	56/164 (34%)
Two copies of $\mathrm{\epsilon }$4	22/193 (11%)	23/164 (14%)
TMEM106B rs3173615			0.17
GG	20/91 (22%)	11/94 (12%)
CG	38/91 (42%)	44/94 (47%)
CC	33/91 (37%)	39/94 (42%)
GRN rs5848			0.98
CC	25/58 (43%)	34/76 (45%)
CT	29/58 (50%)	37/76 (49%)
TT	4/58 (7%)	5/76 (7%)
c9orf72 mutation			--
negative	34/34 (100%)	45/45 (100%)
positive	0/34 (0%)	0/45 (0%)
Neuropathologic features
AD-TDP type			<0.001
Alpha	33/131 (25%)	124/148 (84%)
Beta	98/131 (75%)	24/148 (16%)
Thal amyloid phase			0.07
0	12/141 (9%)	14/113 (12%)
1	12/141 (9%)	23/113 (20%)
2	5/141 (4%)	5/113 (4%)
3	32/141 (23%)	20/113 (18%)
4	43/141 (31%)	30/113 (27%)
5	37/141 (26%)	21/113 (19%)
Braak NFT stage			0.90
I	3/197 (2%)	4/166 (2%)
II	16/197 (8%)	12/166 (7%)
III	13/197 (7%)	8/166 (5%)
IV	28/197 (14%)	25/166 (15%)
V	52/197 (26%)	39/166 (24%)
VI	85/197 (43%)	78/166 (47%)
CERAD			0.20
Normal	19/197 (10%)	26/164 (16%)
Sparse	23/197 (12%)	12/164 (7%)
Moderate	39/197 (20%)	34/164 (21%)
Frequent	116/197 (59%)	92/164 (56%)
ADNC			0.22
Not	12/141 (9%)	14/111 (13%)
Low	14/141 (10%)	18/111 (16%)
Intermediate	49/141 (35%)	38/111 (34%)
High	66/141 (47%)	41/111 (37%)
Hippocampal sclerosis	22/188 (12%)	63/161 (39%)	<0.001
LBD status	82/196 (42%)	60/166 (36%)	0.27
LBD stage			0.58
Brainstem predominant	3/82 (4%)	4/58 (7%)
Limbic transitional/amygdala predominant	39/82 (48%)	23/58 (40%)
Neocortical	39/82 (48%)	29/58 (50%)
Unspecified/ Olfactory bulb only	1/82 (1%)	2/58 (3%)
CAA			0.40
Absent	24/192 (13%)	27/160 (17%)
Mild	53/192 (28%)	36/160 (23%)
Moderate	71/192 (37%)	65/160 (41%)
Severe	44/192 (23%0	31/160 (19%)
Arteriolosclerosis			0.58
Absent	11/180 (6%)	5/160 (3%)
Mild	26/180 (14%)	21/160 (13%)
Moderate	76/180 (42%)	73/160 (46%)
Severe	67/180 (37%0	61/160 (38%)
Infarcts	62/190 (33%)	64/157 (41%)	0.12
Microinfarcts	32/190 (17%)	26/157 (17%)	0.92

Data are expressed as n (%) or median (Q1, Q3). P-values are from Chi-square test for categorical variables and Mann Whitney U test for continuous variables. Significant P-values <0.05 are considered shown in bold italics.

Abbreviations: ADNC = Alzheimer’s disease neuropathologic change; APOE = apolipoprotein E; CAA = Cerebral Amyloid Angiopathy; AVLT = auditory verbal learning test; C9orf72 = chromosome 9 open reading frame 72 gene; CDR = Clinical Dementia Rating-Sum Of Boxes; CERAD = Consortium to Establish a Registry for Alzheimer’s Disease; COWAT = Controlled Oral Word Association Test; CU = cognitively unimpaired; GRN= progranulin gene; LBD = Lewy body disease; MCI = mild cognitive impairment; NFT = neurofibrillary tangle; NPI = Neuropsychiatric Inventory; TMEM106B = transmembrane protein 106B gene; UPDRS = Unified Parkinson’s Disease Rating Scale; WAIS-BD = Wechsler Adult Intelligence Scale-Block Design Test

Association with clinical variables

Scatter plots showing the distribution of the clinical scores analyzed across the age at evaluation spectrum and by Limbic or Diffuse status are shown in Figure 1. Results from the linear regression models adjusted for age at evaluation and time from evaluation to death revealed that the Diffuse group on average performed worse on MMSE by ~2 points (P=0.01) and had more functional impairment with ~2.5 points higher CDR-SOB (P<0.001). Although verbal episodic memory was severely impaired in both groups, the Diffuse group had 1-point lower mean scores on AVLT-delayed test (P<0.001). Moreover, lower mean scores on BNT (decreased by 5 points, P=0.002) and on WAIS-BD (decreased by 3 points, P=0.008) were seen in the Diffuse group, suggestive of more impaired confrontational naming and poorer ability to reconstruct two-dimensional images using patterned blocks, respectively. There was also a trend for the Diffuse group having more impaired visuomotor speed and requiring longer time (~11.5 seconds) to complete TMT-A (P=0.06), as well as having more behavioral abnormalities with 1-point higher scores on NPI (P=0.05).

Figure 1. Distribution of neurologic and neuropsychologic scores.

The scatterplots show the comparison of distribution of clinical and neuropsychological scores across the age at evaluation spectrum between Limbic (green dots) and Diffuse (purple dots) groups, along with the associated regression lines (green for Limbic, purple for Diffuse).

Abbreviations: AVLT delayed = Auditory Verbal Learning Test; BNT = Boston Naming Test; CDR-SOB = Clinical Dementia Rating scale- Sum Of Boxes; COWAT = Controlled Oral Word Association Test; MMSE = Mini Mental State Examination; NPI = Neuropsychiatric Inventory; TMT-A = Trail Making Test A; TMT-B = Trail Making Test B; UPDRS = Unified Parkinson’s Disease Rating Scale; WAIS-BD = Wechsler Adult Intelligence Scale-Block Design Test

Association with genetic variables

The plots in Figure 2 show the predicted probabilities of having diffuse pathology – given a TDP-43 positive status – across the age at death spectrum in the subgroups of patients with different numbers of APOE, TMEM106B rs3173165, or GRN rs5848 risk alleles. After controlling for age at death, there was no significant associations between odds of being Diffuse and APOE $\mathrm{\epsilon }$4 carrier status (with one or two $\mathrm{\epsilon }$4 copies). However, when the carriers were separated based on the number of $\mathrm{\epsilon }$4 copies, there was indication of decreased odds of being Diffuse seen with one copy of APOE $\mathrm{\epsilon }$4 allele (OR: 0.67 [95% CI: 0.42–1.06], P=0.09). There was also evidence for about twofold increase in odds of having diffuse AD-TDP associated with at least one copy of the TMEM106B rs3173615 risk (C) allele (OR: 2.06 [95% CI: 0.92–4.62], P=0.08). Analysis studying the effect of one C allele and two C alleles separately yielded comparable results. The predicted probability plot also showed roughly 15% higher probabilities of diffuse AD-TDP in both CG and CC subgroups compared to the GG subgroup across the age at death spectrum (Figure 2b). Conversely, no significant findings were found in relation to the GRN rs5848 variants.

Figure 2. Diffuse AD-TDP probabilities and APOE, TMEM106B and GRN genotypes.

The predicted probabilities of Diffuse AD-TDP positivity according to APOE, TMEM106B rs3173615, and GRN rs5848 genotypes are plotted against age at death. Shaded areas represent 95% confidence intervals. The figure legends display the genotypes according to the number of risk alleles, from zero to two copies.

Abbreviations: APOE = Apolipoprotein E gene; GRN = progranulin gene; TMEM106B = transmembrane protein 106B gene

Association with other pathologic variables

The predicted probabilities of diffuse AD-TDP across the age at death spectrum and as a function of various sublevels of co-pathologies are shown in Figure 3. The presence of TDP-43 type-$\mathrm{\alpha }$ inclusions in the amygdala increased the odds of having diffuse AD-TDP (OR: 15.05 [95% CI: 8.28–27.37], P<0.001). Indeed, as can be seen in Figure 3, given the presence of type-$\mathrm{\alpha }$ inclusions, participants were predicted to have 50% higher probabilities of being Diffuse than type-$\mathrm{\beta }$ at all ages. In a model considering Thal phases as an ordinal variable, there was evidence of association between a one-unit increase in Thal phases (more widespread A$\mathrm{\beta }$ pathology) and about 17% reduction in odds of having diffuse AD-TDP (OR: 0.83 [95% CI: 0.70–0.96], P=0.02). This was also mirrored in the probability plots, as the lowest probabilities of diffuse AD-TDP were seen in the subgroups with Thal phases 3–5. In terms of overall AD pathology, the presence of intermediate to high ADNC trended to associate with ~50% decrease in odds of concurrent diffuse AD-TDP (OR: 0.57 [95% CI: 0.32–1.04], P=0.07). The probability plot also shows separation of no-to-low ADNC from intermediate-to-high ADNC at all ages. Finally, the presence of HpScl was associated with more than a four-fold increase in odds of diffuse AD-TDP (OR: 4.67 [95% CI: 2.69–8.11], P<0.001). The panel in Figure 3 also shows ~30% differences in probabilities across the age at death spectrum between HpScl-positive and HpScl-negative subgroups. Logistic regression did not reveal significant findings in relation to LBD or cerebrovascular pathology. However, the subgroup without arteriolosclerosis showed lower predicted probabilities of diffuse AD-TDP pathology at all ages compared to those with arteriolosclerosis, irregardless of severity.

Figure 3. Diffuse AD-TDP probabilities and co-pathologies.

The predicted probabilities of Diffuse AD-TDP positivity according to various levels of different copathologies are plotted against age at death. Shaded areas represent 95% confidence intervals.

Abbreviations: ADNC = Alzheimer’s disease neuropathologic change; AS = arteriolosclerosis; CAA = Cerebral Amyloid Angiopathy; CERAD = Consortium to Establish a Registry for Alzheimer’s Disease; HpScl = hippocampal sclerosis; LBD = Lewy body disease; TDP-43 = transactive response DNA-binding protein 43

Clinical regression model

The “clinical” logistic regression model (Figure 4a) was statistically significant (χ² = 31.396, P=0.002) and correctly identified 80% of cases (sensitivity 70%, specificity 87%). After controlling for the other variables in the model, a 5-year increase in age at onset (OR 1.54 [95% CI: 1.03–2.32], P=0.04) and a 1-year increase in disease duration (OR:1.18 [95%CI: 1.03–1.36], P=0.02) were independently associated with increased odds of having diffuse AD-TDP. Among the neuropsychological tests, a 15-second increase in time to complete TMT-B (i.e., longer time to connect alternating numbers and letters in consecutive order) was independently associated with 15% decrease in odds (OR: 0.84 [95% CI: 0.73–0.97], P=0.02) of being Diffuse. Conversely, a one-unit increase in WAIS-BD scores indicating relatively preserved visuospatial skills were associated with an 8% decreased odds of diffuse AD-TDP (OR 0.92 [95% CI: 0.85–0.99], P=0.02). Finally, a one-unit increase in NPI scores was associated with increased odds of being Diffuse (OR 1.13 [95% CI: 1.01–1.27], P=0.04). AVLT and BNT scores were not independent predictors of Diffuse pathology.

Figure 4. “Clinical” and “Pathology” regression models.

The forest plot shows the odds ratios (OR) and 95% confidence intervals (CI) associated with individual neurologic/ neuropsychological (a) or pathologic (b) predictors. Odds ratios are represented by the black dots and the confidence intervals by the horizontal lines. Confidence intervals not crossing the line of null effect (red dashed vertical line) are considered significant (P<0.05). For the “clinical” model, results are shown after additionally controlling for years of education and presence of dementia. *5-year increments in age at evaluation or age at death; **15-second increments in TMT-A or TMT-B completion time.

Abbreviations: AS = arteriolosclerosis; AVLT delayed = Auditory Verbal Learning Test; BNT = Boston Naming Test; CAA = Cerebral Amyloid Angiopathy; COWAT = Controlled Oral Word Association Test; HpScl = hippocampal sclerosis; NPI = Neuropsychiatric Inventory; TMT-A = Trail Making Test A; TMT-B = Trail Making Test B; UPDRS = Unified Parkinson’s Disease Rating Scale; WAIS-BD = Wechsler Adult Intelligence Scale-Block Design Test

Pathology regression model

The “pathology” model (Figure 4b) was also significant (χ²=55.850, P<0.001), and correctly identified 70% of cases (sensitivity 55%, specificity 83%). A 5-year increment in age at death was independently associated with increased odds of diffuse AD-TDP (OR: 1.25 [95% CI: 1.00–1.57], P=0.05). After adjusting for the other pathologies in the model, more extensive A$\mathrm{\beta }$ deposition independently decreased the odds of having diffuse AD-TDP. Compare to Thal A$\mathrm{\beta }$ phase 1, Thal A$\mathrm{\beta }$ phase 3 was associated with $~$80% decrease (OR 0.22 [95% CI: 0.05–0.94], P=0.04), and Thal A$\mathrm{\beta }$ phases 4–5 with $~$90% decrease in odds (OR 0.12 [95% CI: 0.03–0.52], P=0.005 for Thal 4; and OR 0.11 [95% CI: 0.02–0.55], P=0.007 for Thal 5). Contrarily, HpScl independently increased the odds of having diffuse AD-TDP sixfold (OR 6.18 [95% CI: 2.93–13.03], P<0.001). There were no significant associations of diffuse TDP-43 pathology with Braak NFT stages, cerebral amyloid angiopathy, or arteriolosclerosis.

DISCUSSION

The present study focused solely on neuropathologically defined cases of AD-TDP. We aimed to further elucidate whether widespread AD-TDP extending to the basal forebrain, basal ganglia, brainstem, and neocortex (stages 4–6) correlated with clinicopathologic features that differed from those of AD-TDP limited to the limbic regions (stages 1–3). We found that diffuse AD-TDP correlated with older age at onset and death, longer disease duration and more severe overall cognitive and functional impairment. On the other hand, poorer performance on a measure of mental flexibility was an independent predictor of limbic TDP-43 pathology, while frequent behavioral symptoms and worse performance on visuospatial and constructional testing individually predicted a more diffuse pathology. From a pathological standpoint, extensive A$\mathrm{\beta }$ pathology (higher Thal A$\mathrm{\beta }$ phases) and, to a lesser extent, overall higher ADNC were prognostic of a limbic pathology, while TDP-43 type-$\mathrm{\alpha }$ and both HpScl strongly hinted a more diffuse AD-TDP. Overall, diffuse and limbic pathology are associated with different clinicopathologic features, which can potentially be used to predict the distribution of AD-TDP.

To the best of our knowledge, no previous study has divided AD-TDP cases into Limbic or Diffuse groups. The finding that diffuse AD-TDP was associated with older age at death is in line with existing literature [17]. The older age at onset and longer disease duration compared to limbic AD-TDP could be explained by the less severe AD neuropathologic changes found in this group [17]. Our data showed that episodic memory was severely impaired in both groups, although more frequently in the Diffuse group. However, AVLT delayed scores were not an independent predictor of diffuse pathology, which supports the notion that higher likelihood of Alzheimer’s-type dementia is present with TDP-43 pathology once it has spread beyond the amygdala ($\ge$stage 2) [9].

Executive functions encompass high-order cognitive abilities, including working memory, inhibitory control, and cognitive flexibility among others [44]. TMT-B is a measure of cognitive flexibility and is reliant on the integrity of the bilateral dorsolateral, ventrolateral, and medial frontal regions [45, 46]. We found that impaired TMT-B scores were independently predictive of decreased odds of being Diffuse. Executive dysfunction is common in typical late-onset AD, even in prodromal stages [47, 48]. Given that executive dysfunction is related to frontal lobe damage, one would expect better correlation with diffuse AD-TDP, where there is direct involvement of the frontal cortex in AD-TDP stage 6. One possible explanation for our result, as suggested by other authors [49], is that high tau pathology within the entorhinal cortex could lead to disruptions to its connections with prefrontal areas. This could theoretically explain why limbic AD-TDP, which is usually accompanied by greater neurofibrillary tangle pathology, is related to worse TMT-B scores [17]. While we did not find associations between Braak NFT stages (which only describes the spatial distribution, but not quantitative burden, of tau NFTs), we did find strong associations between AD-TDP type-$\mathrm{\beta }$, the type of TDP-43 inclusions that colocalize with tau, and intermediate-to-high ADNC to be more associated with limbic AD-TDP. COWAT is another measure of executive function and is anatomically related to lateral frontal and parietal regions [50]. In this study, relatively preserved COWAT scores only showed a trend for association with higher odds of being Diffuse. This would complement the TMT-B results and imply preserved executive functions in Diffuse AD-TDP.

We also found that higher scores on WAIS-BD were related to decreased odds of diffuse AD-TDP. The anatomical correlate of WAIS-BD scores has been mapped to the right posterior parietal lobe [51]. In AD, predominant visuospatial impairment is characteristic of a variant called posterior cortical atrophy [52]. We can only speculate if posterior cortical atrophy is associated with more widespread AD-TDP. A previous study from our group with a smaller cohort found that the bulk of TDP-43 positives in non-amnestic AD were between stages 2–5 [53]. Future studies comparing the pathological features of the atypical AD variants and their association with TDP-43 pathology including TDP-43 type, regional burden and distribution should be pursued.

We did not find significant findings in terms of parkinsonism which parallels a previous study from our group which found no association between TDP-43 pathology in the basal ganglia and/or substantia nigra and tremor, rigidity, bradykinesia, or gait/postural instability [54]. Another study, which only screened TDP-43 positivity in the amygdala, reported increased scores in the motor disturbance part of the NPI [55]. These motor stereotypies are present in many degenerative dementias, including AD [56].

Concerning neurobehavioral symptoms, higher total NPI scores correlated with increased odds of being Diffuse. Direct involvement of the frontal lobes could help explain this result. An interesting finding from one study showed that TDP-43 positivity correlated with higher total NPI scores but only in the presence of sparse diffuse A$\mathrm{\beta }$ plaques [55]. Intriguingly, we found similar results in this study, as high Thal phases (3–5) correlated with decreased odds of having diffuse AD-TDP, suggesting that lower phases were more associated with widespread TDP-43 pathology. The “ABC” score for AD neuropathologic change [37] advocates that Thal phases 0–2 could only correspond to, at most, an intermediate probability that the cognitive impairment is caused by AD pathology. Furthermore, the absence of A$\mathrm{\beta }$ plaques in the presence of tau tangles suggest a diagnosis of primary age-related tauopathy [57, 58]. While some argue that primary age-related tauopathy is a separate entity from AD [57], others believe it is part of the AD spectrum [59]. In this study, we found evidence for absent to low ADNC to be more associated with diffuse AD-TPD, a finding that is likely driven by the differences in Thal A$\mathrm{\beta }$ phases. The finding that diffuse AD-TDP is more associated with sparse A$\mathrm{\beta }$ deposition contradicts the popularly believed concept that overexpression of A$\mathrm{\beta }$₄₂ triggers expression, phosphorylation and intracellular accumulation of TDP-43 and that clearance of A$\mathrm{\beta }$ prevents TDP-43 pathology [60]. In the absence of A$\mathrm{\beta }$, TDP-43 aggregation has been otherwise linked to increased kinase activation (tau hyperphosphorylation), inflammation, and mitochondrial dysfunction [61, 62]. Generally, TDP-43 pathology is believed to cause neurodegeneration from loss of normal functions and/or toxic gain-of-functions [63]. Our findings are concordant with a recent study showing how loss of TDP-43 triggers a microglial-specific dysfunction that sequentially causes increased A$\mathrm{\beta }$ clearance and simultaneously promotes synaptic loss [64]. The authors concluded that this could explain the low AD pathology, specifically low A$\mathrm{\beta }$ pathology (Thal phase), found in older people with ALS or FTLD-TDP. It is then feasible to hypothesize whether the mechanism underlying limbic AD-TPD could be a primary excessive A$\mathrm{\beta }$ or tau pathology which then prompts TDP-43 pathology, and whether diffuse AD-TDP might be triggered by a primary loss of nuclear TDP-43 leading to dampened AD pathology and enhanced synaptic loss. Should this be the case, it would be less appropriate to classify diffuse TDP-43 pathology as AD-TDP. Finally, and as expected, we found that HpScl increased the odds of being Diffuse corroborate previous findings [11, 17]. Whether HpScl is simply the result of natural progression of TDP-43 pathology in the hippocampus or is a separate entity that is linked to FTLD is still under debate.

The APOE $\mathrm{\epsilon }$4 has long been noted for being one of the major risk factors for AD, with homozygosity for $\mathrm{\epsilon }$4 known to lower the age of onset of AD[65–67]. The effect of APOE $\mathrm{\epsilon }$4 on AD is strongly linked to its ability to cause earlier and more abundant A$\mathrm{\beta }$ pathology in the brain [68, 69] and to have a direct and indirect effect on TDP-43[10]. On the other hand, TMEM106B is associated with numerous neurodegenerative diseases, particularly FTLD-TDP and hippocampal sclerosis[70–74] and to a lesser extent with AD-TDP[75]. Rs3173615 is the only variant on the risk haplotype encoding an amino acid substitution from the highly conserved threonine to serine at position 185 [72]. From our genetic analyses, we found evidence suggesting associations between APOE $\mathrm{\epsilon }$4 allele and limbic AD-TDP and between TMEM106B rs3173615 C allele and diffuse AD-TDP. These results are concordant with the pathological results, with less A$\mathrm{\beta }$/AD pathology and more TDP-type-$\mathrm{\alpha }$/HpScl in the Diffuse group[17]. However, these findings did not reach statistical significance, possibly due to poor statistical power. This can specially be seen in the relatively low number of participants with two copies of APOE $\mathrm{\epsilon }$4 in our cohort and the overall number of participants with data on TMEM106B rs3173615.

A strength of this study is the large, community-based cohort of older participants with known AD and TDP-43 pathology status. Another strength includes the standardized clinical, neuropsychological and neuropathologic evaluations and the use of the more appropriate 6-stage scheme with greater range. This study is limited by the fact that we only looked for the presence of inclusions. A quantitative assessment of TDP-43 burden could have provided additional information and should therefore be addressed in future studies. Lastly, our cohort was predominantly White, which may somewhat limit the generalizability of the findings.

In summary, our results offer additional proof of different clinicopathologic associations and possible genetic underpinnings between limbic and diffuse AD-TDP. The presence of AD-TDP inclusions in certain anatomical areas can partly explain the worse performance seen on their corresponding neuropsychological tests, such that diffuse AD-TDP is more associated with behavioral symptoms, and limbic AD-TDP correlates with cognitive flexibility. However, in terms of pathology and genetics, diffuse AD-TDP is related to less AD pathology (particularly A$\mathrm{\beta }$) and frequency of APOE $\mathrm{\epsilon }$4 and is instead related to more hippocampal sclerosis and frequent TMEM106B rs3173615 risk (C) allele. Diffuse AD-TDP then apparently and more closely resembles a primary TDP-43 proteinopathy (i.e., young-onset frontotemporal lobar degeneration) than AD. This could have public health implications relating to treatment and research on AD and dementia. Furthermore, the current recommendations of the LATE-NC consensus group require screening for TDP-43 pathology in 3 regions (amygdala, hippocampus, and frontal regions)[76]. This 3-stage scheme is useful for screening or diagnostic purposes. The Josephs 6-stage scheme can be more time consuming and costly but is also more appropriate to use in research aimed at better understanding clinicopathologic relationships and disease mechanisms. AD-TDP stages 4–5, which can be easily missed by the LATE-NC staging, are frequent in older patients with dementia[19]. The combination of clinical, genetic, and pathologic features able to predict diffuse AD-TDP that are described in this study could potentially be used to identify patients requiring more in-depth analyses, including the more extensive staging scheme. Results from this study should however be interpreted with caution and validated by future studies.

DATA AVAILABILITY

Data that supports the findings in this study is available from the corresponding author (K.A.J) upon reasonable request.